Characterizing co-expression networks in liver, adipose, and brain tissues in segregating mouse populations

Recent advances in gene expression microarray technology have provided the ability to monitor transcriptional changes on a genome-wide scale, allowing for the more general characterization of interaction networks among gene expression traits. Here I discuss the correlation structure among gene expression traits for a comprehensive set of genes (>20k) in mouse liver, adipose and brain samples using oligonucleotide microarrays. This analysis reveals interesting properties of the gene-gene correlation structures that obtain in specific tissues in segregating mouse populations, as well as highlighting differences among the tissues and between the sexes. Specifically, there are a moderate number of gene expression traits (hubs) that have a high degree of connectivity to other gene expression traits, whereas the majority of gene expression traits are found to have limited connectivity to other gene expression traits. The degree of connectivity in the context of expression quantitative trait loci (eQTL) is also characterized, highlighting associations between highly interconnected groups of gene and common genetic control, where the shared genetic control enhances the ability to infer causal associations among gene expression traits and between gene expression and clinically relevant traits. In defining links (edges) between any two genes if they are found to be strongly interacting (i.e., above some absolute correlation threshold or p-value cutoff), the resulting gene-gene interaction network is shown to exhibit scale-free network properties and a hierarchical structure that tends to reflect common functional properties of genes. The degree distributions of these networks are shown to follow an inverse power law with exponents between -1 to -1.3, while the cluster coefficient of some of the networks decrease as an exponential function instead of a power law. Both properties are distinct from most published biological networks. Furthermore, we show that strongly correlated genes appear to group into distinct sub-clusters enriched for a diversity of functional pathways that are further enriched for eQTL hot spot regions in the genome. Our analysis provides a simple and systematic approach to study the functional organization of gene expression interaction networks that can be used to functionally characterize gene groups and identify key control points in the network associated with complex traits such as common human diseases.

Fred Hutchinson Cancer Research Center