Network analysis of cell adhesion

Abstract

Complex systems consisting of diverse interlinked elements are often represented as networks and are described according to the principles of network analysis. Among the networks of biological interest, several protein-protein interactomes have been reported in recent years, mostly in conjunction with high-throughput assays and extensive efforts of literature mining. The resulting global networks display well-defined topological properties and provide a comprehensive view of all the biological contexts in which a given interactome is involved. Global networks, however, do not provide enough information about the specific contexts, such as biological processes and subcellular compartments in which the individual interactions occur. Thus, to glean additional insights, it is often advantageous to extract context-defined local subnetworks from the global networks. Our recently published network analysis of the cell-cell adhesome, i.e. the protein-protein interaction subnetwork that underlies both the biological process of cell-cell adhesion and the subcellular compartment of the apical junctions in human epithelial cells, is an example of such context-defined approaches.

This article refers to:

In spite of qualitative differences, common patterns emerge from the analysis of unrelated complex systems that are the object of disciplines as disparate as biology, sociology and information science. Thus, it is often convenient to display these systems as networks, where the elements and the interactions between elements are represented as nodes and edges, respectively. Biological systems that are often displayed as networks include protein-DNA and protein-protein interactions (PPI), as well as transcript co-regulations. The rooting of network analysis in graph theory and the basic topological properties of the biological networks are reviewed elsewhere.1

The increasing use of PPI networks has paralleled the increasing availability of PPI data that are derived from high-throughput assays, such as yeast-two-hybrid analysis (in the model organisms S. cerevisiae,2,3 C. elegans4 and D. melanogaster5 and in human cells6,7) and protein complex analysis (by co-affinity purification and mass spectrometry8,9). Additional PPI data come from extensive searches of the scientific literature on model organisms10 and human cells.11

Network-level approaches offer the immediate advantage of unveiling general features of the complex systems that would remain otherwise unknown by molecule-level approaches. A primary feature is node connectivity (i.e., the number of edges per node), which allows not only to identify the ‘hubs’ (i.e., nodes with an exceedingly high number of edges) but also to define the scale-free distribution of connectivity (with very few hubs amid numerous poorly-connected nodes) as organizing principle of many real networks.12 Other features include the clustering coefficient of a node and the shortest path between pairs of nodes, which instead define the local density of interactions and the ‘small world’ characteristic of a system,13 respectively. Network analysis also assists in subdividing networks into modules, which are groups of densely interconnected nodes often performing a common function.14 Finally, network analysis is expected to shed light on the evolution of cellular circuitry15 and to help identify novel drug targets.16

Nevertheless, global (i.e., large-scale) PPI networks fall short of providing predictive models of cell dynamics, which is the final aim of systems biology. Typically, dynamic modeling applies mathematical equations to quantitative descriptions of directional pathways that consist of a relatively small number of molecules. In contrast, global PPI networks are qualitative descriptions of non-directional interactomes that consist of thousands of proteins. In addition, global networks do not provide dynamic information about the specific conditions under which each PPI occurs (e.g., changes in the subcellular location of a protein during different biological processes or in response to different stimuli). Thus, the introduction of context-defined subnetworks is often perceived as an intermediate step towards the ultimate goal of distilling dynamic information from the global networks.

To extract local (i.e., context-defined) subnetworks from global PPI networks, most methods deploy available information (mostly in the form of DNA micro-array data) on changes in gene expression levels17 or profiles18 that occur in specific contexts. These methods are based on the assumption that the gene products (i.e., proteins) that belong to the same subnetwork tend to be expressed in a coordinated manner. Other methods instead deploy contextual information (mostly in the form of Gene Ontology annotations) to select groups of interacting proteins that share the same annotation (i.e., biological process or sub-cellular compartment). These methods have led to a view of the global interactome as ‘mosaic’ of overlapping (but distinct) context-defined subnetworks.19

We have recently developed a novel strategy to outline a context-defined fraction of the global PPI network that is restricted to a biological process (i.e., cell-cell adhesion) and to a cellular component (i.e., the apical complex) in human epithelial cells.20 The apical complex comprises tight junctions,21 adherens junctions22 and desmosomes.23 In addition to ensuring cell-cell cohesion between adjacent epithelial (or endothelial) cells, the complex regulates important responses, including fluid permeability, membrane polarity, leukocyte transmigration and tissue strength. Like other subnetworks, also the cell-cell adhesome has been assembled from published data. However, at variance with other subnetworks, we have tried to build the cell-cell adhesome on the ‘functional units’ of the system. These units correspond to the intrinsic core proteins of the apical complex. To be defined as intrinsic, proteins must fulfill functional, subcellular and topological criteria, such as being annotated as structural (i.e., membrane, adaptor, cytoskeletal adaptor or cytoskeletal) proteins, being localized to the epithelial junctions and being components of linear 3-node (or 4-node) motifs, which are sequentially composed of a membrane, (an adaptor,) a cytoskeletal adaptor and a cytoskeletal protein. In addition, the intrinsic proteins may be regarded to as the ‘seed crystals’ that serve to recruit the accessory (mostly regulatory) proteins to the subnetwork. Actually, to be defined as accessory, proteins must interact directly with the intrinsic components. As a corollary, the accessory periphery is separated from the intrinsic core just by one degree of separation.20

Network analysis has highlighted that the apical complex is not characterized by a scale-free, but by an exponential (and, therefore, hub-less) distribution of connectivity. Nevertheless, the highly connected (albeit non-hub) proteins in the network behave essentially like hubs, as they are vulnerable points to simulated targeted attack in silico and to lethal mutations during embryonic development. In addition, the subdivision of the network into modules has highlighted that, unlike the morphological partition of the apical complex into three organelles, the corresponding subnetwork comprises four major modules. The first module corresponds to the tight junctions, the second to the adherens junctions/desmosomes, the third to a structural unit, and the fourth to a signaling platform that integrates regulatory PPI between accessory and structural proteins from different organelles.20

Finally, it is worth mentioning that, in addition to the cell-cell adhesion subnetwork of the apical complex, network analysis has been used to study the cell-matrix adhesion subnetwork of the focal adhesions.24 Remarkably, both cell-cell and cell-matrix adhesomes display similar topology and architecture. Thus, we foresee that these initial efforts of analyzing context (i.e., adhesion)-defined subnetworks may lay the foundations for additional studies aimed at defining (among others) the evolution of adhesion in metazoans and identifying protein targets for novel pharmacological therapies.

Abbreviation

PPI

=

protein-protein interaction

Acknowledgements

Support from Association International for Cancer Research, UK (Grant 04-095) and the Fondazione Musarra is gratefully acknowledged.

Addendum to: