Revealing static and dynamic modular architecture of the eukaryotic protein interaction network

Abstract

In an effort to understand the dynamic organization of the protein interaction network and its role in the regulation of cell behavior, positioning of proteins into specific network localities was studied with respect to their expression dynamics. First, we find that constitutively expressed and dynamically co-regulated proteins cluster in distinct functionally specialized network neighborhoods to form static and dynamic functional modules, respectively. Then, we show that whereas dynamic modules are mainly responsible for condition-dependent regulation of cell behavior, static modules provide robustness to the cell against genetic perturbations or protein expression noise, and therefore may act as buffers of evolutionary as well as population variations in cell behavior. Observations in this study refine the previously proposed model of dynamic modularity in the protein interaction network, and propose a link between the evolution of gene expression regulation and biological robustness.

Figure 1

Interaction pattern of proteins according to their EV. (A) Boxplot of proteins in each of 50 bins with the given EV versus their neighborhood EV. (B) Interaction preference matrix of the yeast network. Each square represents the number of interactions between corresponding bins. Left panel: Interaction preference matrix of the actual network, right panel: Interaction preference matrix of randomized network achieved by randomly shuffling the positions of proteins in the network (right panel). (C) Interaction preference matrix of proteins with different node degrees corresponding to the four quartiles of the node degree distribution. Proteins were binned according to their EVs and node degrees. Each square represents the normalized number of interactions between proteins with given node degree (k) and EV. Normalization of a square (i, j) in the matrix was carried out by calculating the number of interactions between proteins in the bins i and j, and dividing that number with the total number of interactions that proteins in bins i and j have. Color key shows the normalized number of interactions between bins.

Figure 2

Static and dynamic neighborhoods resemble functional modules. (A) Plot of the sub-network formed by hubs. Proteins are colored according to their EV. (B) Plot of hubs' nPCC values against their neighborhood density. Neighborhood density (see Materials and methods) is a measure of how densely the neighbors of a protein are connected to each other, and ranges from 0, for the least dense neighborhood, to 1, for a maximally densely connected neighborhood. Proteins within dense clusters are expected to have high neighborhood densities. Dots (hubs) are colored according to the hub EVs (left panel) and neighborhood EVs (right panel).

Figure 3

Functional specialization in the static and dynamic networks. Comparison of network modularity in the static and dynamic networks with that of 100 networks formed by random draws of interactions from the original network. The plot shows the distributions of network modularity values for random draws of 897 (left panel, for comparison with the static network) and 777 (right panel, for comparison with the dynamic network) protein–protein interactions out of the original network. Arrows show the actual network modularity values of the static and dynamic networks (P<0.01 in both cases).

Figure 4

Distinct neighborhood EV and avPCC characteristics of central and modular hubs. (A) Plots of hubs' avPCC versus neighborhood EV; each dot (hub) is colored according to the value of the corresponding measure. (B) Hubs are colored according to the neighborhood EV variation, which is the standard deviation of the neighbors' EV values of a protein and shows how diverse the EVs of proteins in the neighborhood of a protein are.

Figure 5

Evolutionary rate and expression noise of the static and dynamic modules. Evolutionary rates of yeast proteins derived by Hirsh et al (2005) were used. (A) Boxplot of evolutionary rates of family, party and date hubs. Family hubs are static hubs with neighborhood EVs of <0.3, party hubs are hubs with avPCC>0.45 and date hubs are those with neighborhood EV>0.3 and avPCC<0.45. (B) Fractions of proteins in the static and dynamic networks whose gene deletion is lethal to yeast. (C) Boxplot of protein expression noise in the different hub classes.