Gene network information has been used to improve gene selection in microarray-based studies by selecting marker genes based both on their expression and the coordinate expression of genes within their gene network under a given condition. Here we propose a new network-embedded gene selection model. In this model, we first address the limitations of microarray data. Microarray data, although widely used for gene selection, measures only mRNA abundance, which does not always reflect the ultimate gene phenotype, since it does not account for post-transcriptional effects. To overcome this important (critical in certain cases) but ignored-in-almost-all-existing-studies limitation, we design a new strategy to integrate together microarray data with the information of microRNA, the major post-transcriptional regulatory factor. We also handle the challenges led by gene collaboration mechanism. To incorporate the biological facts that genes without direct interactions may work closely due to signal transduction and that two genes may be functionally connected through multi paths, we adopt the concept of diffusion distance. This concept permits us to simulate biological signal propagation and therefore to estimate the collaboration probability for all gene pairs, directly or indirectly-connected, according to multi paths connecting them. We demonstrate, using type 2 diabetes (DM2) as an example, that the proposed strategies can enhance the identification of functional gene partners, which is the key issue in a network-embedded gene selection model. More importantly, we show that our gene selection model outperforms related ones. Genes selected by our model 1) have improved classification capability; 2) agree with biological evidence of DM2-association; and 3) are involved in many well-known DM2-associated pathways.

Thiazolidinediones (TZDs) act through peroxisome proliferator activated receptor (PPAR) γ to increase insulin sensitivity in type 2 diabetes (T2DM), but deleterious effects of these ligands mean that selective modulators with improved clinical profiles are needed. We obtained a crystal structure of PPARγ ligand binding domain (LBD) and found that the ligand binding pocket (LBP) is occupied by bacterial medium chain fatty acids (MCFAs). We verified that MCFAs (C8-C10) bind the PPARγ LBD in vitro and showed that they are low-potency partial agonists that display assay-specific actions relative to TZDs; they act as very weak partial agonists in transfections with PPARγ LBD, stronger partial agonists with full length PPARγ and exhibit full blockade of PPARγ phosphorylation by cyclin-dependent kinase 5 (cdk5), linked to reversal of adipose tissue insulin resistance. MCFAs that bind PPARγ also antagonize TZD-dependent adipogenesis in vitro. X-ray structure B-factor analysis and molecular dynamics (MD) simulations suggest that MCFAs weakly stabilize C-terminal activation helix (H) 12 relative to TZDs and this effect is highly dependent on chain length. By contrast, MCFAs preferentially stabilize the H2-H3/β-sheet region and the helix (H) 11-H12 loop relative to TZDs and we propose that MCFA assay-specific actions are linked to their unique binding mode and suggest that it may be possible to identify selective PPARγ modulators with useful clinical profiles among natural products.

Reduction in insulin clearance plays an important role in the compensatory response to insulin resistance. Given the importance of this trait to the pathogenesis of diabetes, a deeper understanding of its regulation is warranted. Our goal was to identify metabolic and cardiovascular traits that are independently associated with metabolic clearance rate of insulin (MCRI). We conducted a cross-sectional analysis of metabolic and cardiovascular traits in 765 participants from the Mexican-American Coronary Artery Disease (MACAD) project who had undergone blood sampling, oral glucose tolerance test, euglycemic-hyperinsulinemic clamp, dual-energy X-ray absorptiometry, and carotid ultrasound. We assessed correlations of MCRI with traits from seven domains, including anthropometry, biomarkers, cardiovascular, glucose homeostasis, lipase activity, lipid profile, and liver function tests. We found inverse independent correlations between MCRI and hepatic lipase (P = 0.0004), insulin secretion (P = 0.0002), alanine aminotransferase (P = 0.0045), total fat mass (P = 0.014), and diabetes (P = 0.03). MCRI and apolipoprotein A-I exhibited a positive independent correlation (P = 0.035). These results generate a hypothesis that lipid and adiposity associated traits related to liver function may play a role in insulin clearance.