A thematic review series: systems biology approaches to metabolic and cardiovascular disorders

Abstract

The next series of the JLR’s thematic review articles concerns systems-level approaches to cardiovascular and metabolic traits. “Systems-level” means a kind of biologic analysis that looks beyond individual genes or proteins or lipids to the ensemble of multiple elements of a system. Some examples of biologic systems are the transcripts in a cell (a “transcriptome”), the proteins of an organelle (a “proteome”), and the metabolites in a liver (a “metabolome”). Systems-level approaches have been stimulated by the genome project, by the development of experimental techniques that can simultaneously interrogate many elements of a system (such as expression microarrays), and by advances in computational methods for analyzing large data sets (1). A key concept in systems biology is that of “emergent properties,” important features of biologic systems that can best be identified by examining the system as a whole. A simple example of an emergent property is illustrated in Fig. 1. In this experiment, the levels of the transcripts in livers of mice from an intercross between two common inbred strains were determined using expression microarrays. The entire data set was then tested to identify transcripts whose levels correlate with one another. One of many such sets of highly correlated transcripts from this experiment is the set of cholesterol biosynthetic enzymes, as shown. In this case, the relationships of these enzymes were already known through the painstaking, decadeslong studies of many biochemists. However, even if this information was not known, our global experiment would immediately group all of the enzymes together, implying a functional relationship. Jim Weiss’s review in this series will discuss a particularly striking emergent property in metabolism. Thus, both glycolysis and oxidative phosphorylation are capable, under the right conditions, of developing self-sustained oscillations. Such oscillations are observed only when the individual enzymes are coupled into a “network” with other metabolic enzymes to create positive and negative feedback loops. Clearly, in this example, studies of the individual components of the system would not provide mechanistic understanding of the overall dynamics of the system. A particularly important goal of systems biology is to construct “networks,” sets of genes or proteins or metabolites that act in concert in a common biologic process (2). These networks can be experimentally identified by classical methods (e.g., the pathways of cholesterol homeostasis), but systems-level approaches such as expression arrays, chIP-chip studies, and whole-genome yeast two-hybrid experiments can efficiently test millions of possible interactions. Topologically, networks consist of elements (termed “nodes” in network nomenclature) that exhibit functional connections (“edges”) (Fig. 2). The connections can be identified using coregulation (as in Fig. 1), physical interactions (as in protein complexes), or metabolic relationships (e.g., the intermediates in glycolysis). “Undirected” networks are simply nodes connected by edges, with no causal direction, whereas “directed” networks have edges with a given direction (Fig. 2). An important emergent property observed for most biologic networks is scale-free topology, in which some nodes have many edges (these nodes are termed “hubs”) but most nodes have few edges. A convenient way of organizing the data sets generated using various global platforms is as a series of orthogonal stages, ordered according to the sequential stages of gene expression (genome, transcriptome, proteome) followed by the metabolome (the set of metabolites) and the phenome (the set of physiologic or disease parameters of interest) (Fig. 3). Each of these data sets can be used to construct networks or derive other useful information, but none alone tells the whole story. For example, analyses of the transcriptome will miss crucial aspects of protein realization and cellular signaling. Thus, the intersection of these orthogonal data sets is an important challenge. One particularly useful application of such genomic integration is for the identification of genes and pathways con-