Evolutionary pressure on the topology of protein interface interaction networks

Abstract

The densely connected structure of protein-protein interaction (PPI) networks reflects the functional need of proteins to cooperate in cellular processes. However, PPI networks do not adequately capture the competition in protein binding. By contrast, the interface interaction network (IIN) studied here resolves the modular character of protein-protein binding and distinguishes between simultaneous and exclusive interactions that underlie both cooperation and competition. We show that the topology of the IIN is under evolutionary pressure, and we connect topological features of the IIN to specific biological functions. To reveal the forces shaping the network topology, we use a sequence-based computational model of interface binding along with network analysis. We find that the more fragmented structure of IINs, in contrast to the dense PPI networks, arises in large part from the competition between specific and nonspecific binding. The need to minimize nonspecific binding favors specific network motifs, including a minimal number of cliques (i.e., fully connected subgraphs) and many disconnected fragments. Validating the model, we find that these network characteristics are closely mirrored in the IIN of clathrin-mediated endocytosis. Features unexpected on the basis of our motif analysis are found to indicate either exceptional binding selectivity or important regulatory functions.

Figure 1

Schematic illustrating (left) the binding of three proteins, where matching features indicate binding interfaces, (center) the corresponding interface-interaction network with orange and green indicating shared and non-shared binding interfaces, respectively, and (right) the protein-protein interaction network.

Figure 2

Network evolution and validation. A,B) IINs with a fixed number of M=150 edges and N=90 (top) and 200 interfaces (bottom), degree distributions α=0 (A) and 0.8 (B) before and after network evolution (left and right of arrows, respectively). Green interfaces have a single partner, orange interfaces are shared and have multiple partners, and purple indicates closed triangles in the network. The optimized networks have the same degree distribution but have been evolved to reduce the number of partners each node’s partners have (see Methods). C) IIN of yeast clathrin-mediated endocytosis, closely resembling the optimized, sparse, α=0.8 network in the bottom right corner of panel B). Larger, labeled version shown in Figure S7B. Network images created using Cytoscape.

Figure 3

Properties of IINs. A) Specificity of the interface interaction networks with 150 edges as a function of interface numbers (N), degree distribution (α), and local structure altered by the optimization scheme. Each data point represents an average over 10 networks with error bars indicating the standard deviation. The gray dashed lines correspond to the original, non-optimized networks, with degree distributions varying from α=0 (light gray) to α=0.8 (black). The same symbols are used for the optimized networks shown in solid colored lines. The degree distributions vary from α=0 (light yellow) to α=0.8 (red). The blue hexagram denotes the minimum energy gap (in units of the thermal energy kT) for the endocytic IIN, placed on the x-axis according to its average degree <k>=2.06 (upper x-axis). The black and red hexagrams denote the minimum energy gap for the randomized and optimized versions of the endocytic IIN, preserving its degree distribution including self-loops. B) Corresponding degree distributions for the sparse N=200 interface networks averaged over 50 networks. C) Corresponding degree distributions for the dense N=90 interface networks.

Figure 4

Yeast clathrin-mediated endocytosis interaction sub-network. A) Protein-protein interaction network for the 56 proteins selected as part of the endocytic functional module showing all the 186 edges where interfaces were successfully assigned. B) The same protein interaction network as in A) with interfaces assigned to each protein interaction. The IIN has 206 edges because some of the protein interactions have two modes of binding to one another. Green interfaces have a single partner and orange interfaces are shared with multiple partners. The interface numbers range from one up to eleven (for LAS17). The interface interaction network of the endocytic network, shown separated from the proteins in Figure 1C, emphasizes the disconnected topology of the IIN.