Signaling Heterogeneity is Defined by Pathway Architecture and Intercellular Variability in Protein Expression

Abstract

Insulin's activation of PI3K/Akt signaling, stimulates glucose uptake by enhancing delivery of GLUT4 to the cell surface. Here we examined the origins of intercellular heterogeneity in insulin signaling. Akt activation alone accounted for ~25% of the variance in GLUT4, indicating that additional sources of variance exist. The Akt and GLUT4 responses were highly reproducible within the same cell, suggesting the variance is between cells (extrinsic) and not within cells (intrinsic). Generalized mechanistic models (supported by experimental observations) demonstrated that the correlation between the steady-state levels of two measured signaling processes decreases with increasing distance from each other and that intercellular variation in protein expression (as an example of extrinsic variance) is sufficient to account for the variance in and between Akt and GLUT4. Thus, the response of a population to insulin signaling is underpinned by considerable single-cell heterogeneity that is largely driven by variance in gene/protein expression between cells.

Keywords: Cell Biology; Experimental Models in Systems Biology; Mathematical Biosciences; Systems Biology.

Graphical abstract

Figure 1

Akt2 membrane recruitment is heterogeneous and non-random (A and B) The insulin-stimulated (1 nM) translocation of both GLUT4 (A) and Akt2 (B) to the PM is heterogeneous within a population of cells. (C) Akt2 recruitment to the PM is reproducible following a 2-h washout in response to 1 nM insulin within a single cell. (D) Statistical test comparing reproducibility of Akt2 recruitment (in response to 1 nM insulin before and after 2-h washout) within a cell (red boxes) and between cells (light blue boxes). p values were calculated from a two-sided Wilcoxon rank-sum test. The boxes capture the lower and upper quartiles with the median displayed as a horizontal line in the middle; whiskers are 1.5∗IQR.

Figure 2

Quality control analyses of 5 GLUT4 and Akt2 recruitment datasets reveal no deleterious effects of overexpression (A) Temporal GLUT4-pHluorin trafficking and Akt2-tRFPt recruitment in individual 3T3-L1 adipocytes upon 1-nM insulin stimulation visualized by TIRFM. (B) Principal-component analysis plots summarizing response profiles of GLUT4 and Akt2 in each cell into two principle components. Cells are color coded by experiment. (C) Correlations between expression levels of each construct (upper panel) and Akt2 expression level and maximal GLUT4 response (lower panel). Data are from 206 cells from 5 independent experiments (n = 43, 40, 31, 34, 58 for each experiment, respectively).

Figure 3

The magnitude and shape of insulin-stimulated Akt2 recruitment are predictive of GLUT4 response heterogeneity in single cells (A) Correlations of intracellular GLUT4 and Akt response characteristics. (B) Scatterplot of individual Akt2 response features with respect to GLUT4 maxima. Fitted lines are from linear regression model. The color points in the third panel denote the clusters of each cell based on (C). (C) Akt recruitment clusters as determined by response shape and their corresponding GLUT4 maxima. For the line plot data are mean ± SD. For the boxplot, the boxes capture the lower and upper quartiles with the median displayed as a horizontal line in the middle; whiskers are min and max. See also Figure S4.

Figure 4

Akt AUC predictiveness of GLUT4 max response can be decomposed into Akt max and Akt shape (A) Predictive power of cluster membership scores from the two fuzzy c-means clusters, AUC, and Akt2 max and Akt response shape. Time points before the insulin stimulation (from 1 to 10) are included as negative controls. Variables are sorted by their predictive power based on a random forest regression model. B, C) Correlation and partial correlation analyses on Akt AUC redundancy in GLUT4 max response prediction (B), and independence of Akt max and Akt shape in GLUT4 max response prediction (C).

Figure 5

Individual genes are expressed at highly variable levels between cells (A) Boxplot of relative frequency of gene expression across 18 cell lines from the Tabula Muris Consortium. (B) Boxplot of the difference in gene expression observed between cells. Boxes capture the interquartile range with the median represented by a horizontal bar and whiskers representing the upper and lower 1.5∗IQR values. Points denote outliers. (C) Scatterplot illustrating the expression range of genes defined by GO term transmembrane receptor protein tyrosine kinase signaling pathway. x axis denotes the median gene expression across the 18 cell lines. y axis denotes the median log2 fold change in gene expression. Each point denotes a gene and has been colored by the coefficient of variation (CV). See also Figure S1.

Figure 6

Spatiotemporal correlation between nodes within a generalized multi-tier signaling network (A) A generalized multi-tier signaling cascade with no feedback loop. Reactions (in this case phosphorylation) within each tier are reversible, and the active form of the nodes can be converted to the inactive form by phosphatases (Pi, i = 1 … n). (B) As in (A) but where S1 is under regulation of a negative feedback loop. (C) Representative simulated time course response profiles of each network node in (A) following insulin stimulation at t = 0; each curve represents one of 50 different runs where the expression of network nodes were randomly varied. (D) Scatterplots showing steady-state correlation between the primary upstream node (active S1, aS1) and downstream nodes from the representative simulated time course response profiles from (C). (E and F) (E) As in (C) but for the negative feedback network (B). (F) As in (D) but for (E). (G) Coefficient of variation for each of the nodes across all parameter sets for both models. (H) Correlation coefficients between the simulated steady state of active S1 (primary node) and the downstream nodes showing correlation decreases as their distance increases. The correlation coefficient indicates Pearson's linear correlation coefficient. The error bar indicates standard deviation from randomly generated kinetic parameter sets (n = 1,000). ∗p<0.05 compared with aS4/aS4 node with the same feedback, by by two-way ANOVA with Dunnett correction for multiple comparisons. (I) Correlation between the simulated steady state of active S4 and the up- and downstream nodes for both models where the feedback is strong. ∗∗∗∗p < 0.0001 for change with distance from node1 by two-way ANOVA. #p < 0.01 compared with no feedback within the same node by two-way ANOVA with Dunnett correction for multiple comparisons. See also Figures S2–S5.

Figure 7

Adjacent components of the insulin signaling pathway are highly correlated The magnitude and shape of insulin-stimulated PDK1 and Akt recruitment to the plasma membrane are highly correlated. (A) 3T3-L1 adipocytes were co-electroporated with Akt2-tRFPt and PDK1-eGFP and treated with 1 and 100 nM insulin. Recruitment was assessed using TIRFM. Each line is the response of an individual cell, colored in accordance with the magnitude of recruitment. (B) Comparisons of the PM recruitment maxima (left) and shape (right) observed for Akt and PDK1 in response to 1 (upper) and 100 (lower) nM insulin. (C) PM recruitment of Akt2-tRFPt in single 3T3-L1 adipocytes following a 2-min 1-nM insulin stimulation, as measured by TIRFM. Each cell's response was ranked from dark to light blue, in order of increasing T309 phosphorylation quantified by immunofluorescence. (D) Scatterplot indicating the degree of correlation between T309 and Akt2-tRFPt PM recruitment maxima for the 11 cells in (C).