Traditional methods for the genetic analysis of metabolic syndrome (MetSyn) (left) include identification of genetic variations and mutations underlying single-gene disorders with large effects on metabolic traits through genetic mapping and positional cloning, or, more recently, genome-wide association studies in which polymorphisms in specific genes are statistically linked with clinical traits of interest in large human cohorts. Systems approaches (right) are being adopted to allow the identification of contributions of small effects from genetic variations in many genes on traits, including levels of mRNA transcripts, proteins and metabolites, as well as clinical traits. The systems approach allows the generation of network models that suggest functional interactions between genes within the network. Genes identified with both approaches can be used to formulate hypotheses, and the corresponding mechanisms can be tested through the generation of genetically engineered mouse models, through the manipulation of gene expression levels in cell-based assays using RNAi or overexpression technologies, and through human clinical studies.

a | The examination of molecular phenotypes, such as transcript levels, in genetically randomized mouse or human populations is a powerful strategy to identify molecular networks underlying complex traits. For example, genetic variations perturb both the molecular phenotypes (transcript levels) and clinical traits. Loci controlling transcript levels (termed expression QTLs, eQTLs) can be mapped and used to prioritize positional candidates111. Genes that vary together when subjected to multiple genetic perturbations are likely to share regulatory pathways, and the relationships thus obtained often closely parallel known biology, and have in some cases been validated by experimental perturbations (for example, REFS 83, 107, 111). b | A co-expression network of about 3,000 genes constructed from liver gene expression patterns of a segregating population of mice identifies modules of highly correlated genes133, which are indicated by the different colours. As some modules are highly correlated with clinical traits, they can explain much more of the total variation than any individual locus. Although it is possible that modules are due to the perturbing effects of the clinical traits themselves, this is unlikely because there is little overlap between the loci controlling the clinical traits and the loci perturbing the module genes. c | Correlations of the individual colour-coded liver modules from part b (correlations averaged over their genes) and body weight133. Strong correlation of particular network modules (for example, the module marked with an asterisk) with clinical traits provides a means of identifying new disease pathways. Although co-expression networks do not indicate the direction of interaction between nodes, causal modelling can be performed as discussed in BOX 5 to create ‘directed networks’. d | Sub-section of predicted causal interactions involving sortilin 1 (SORT1) and other genes in the liver30. Part d is reproduced from REF. 30.