From Protein Structure to Function with Bioinformatics.pdf

12 Prediction of Protein Function from Theoretical Models 305in insulin secretion and, consequently, diabetes. The diagnosis of the genetic etiologyof the disease has revolutionized therapy for patients with neonatal diabetesresulting from Kir6.2 mutations, as those channels can still be closed by therapeuticssuch sulfonylureas and glinides and the insulin treatment could be limited or discontinued.The homology model of Kir6.2 subunit allowed for the spatial mapping ofthe residues mutated in the neonatal diabetes and thus illustrated the molecularmechanism underlying reduced K ATPsensitivity (Hattersley and Ashcroft 2005).Patients carrying mutations in Kir6.2 exhibit a spectrum of phenotypes that aredirectly correlated to the nature of mutation. For example, patients with neurologicalsymptoms carry the mutations which do not directly impair ATP binding but markedlybias the channel toward the open state and thus reduce the ability of ATP toblock the channel (ATP stabilizes the closed state of the channel).Recent studies showed that there is a group of patients with permanent neonataldiabetes, carrying the L164P mutation in Kir6.2, who are unresponsive to sulfonylureatherapy (Tammaro et al. 2008). Analysis of the spatial L164 position revealsthat this residue lies deep within the structure, 35 Å away from the ATP-bindingsite. It is therefore unlikely that it acts by reducing ATP binding directly. Instead,L164P probably destabilizes the closed state of the channel, to which sulfonylureaspreferentially bind, and which is rarely reached in channels with enhanced channelopen probability. Taken together, these results show that the drug response isdependent of the nature of particular mutation, but that it can be predicted bydetailed analysis of a protein model.12.4.3 Protein ComplexesA full understanding of the complex networks of protein-protein interactions thatexist in cells is essential if systems biology, whereby these and other large-scaledatasets are integrated into a meaningful whole, is ever to become a success. Thereis therefore much interest in adding predictions from comparative modelling to thebattery of experimental and computational methods for prediction of protein-proteininteraction (Aloy and Russell 2006). The principle is very simple: given aknown structure in which A is complexed to X, does analysis of the putative complexof B-Y (with B homologous to A and Y homologous to X) suggest that theinteraction will occur in vivo (Aloy and Russell 2002). The first methods in the areaassessed the favourability of the interface by borrowing pairwise interaction potentialsdeveloped from threading methods and analysing known contacts from the A:X structure using the sequences of B aligned to A, and Y aligned to X. Servers nowavailable for this type of study include InterPreTS (Aloy and Russell 2003) andMULTIPROSPECTOR (Lu et al. 2002). Later work modelled the protein complexexplicitly and again used interaction potentials to discriminate between predictedtrue and false interactions (Davis et al. 2006).An interesting large-scale application of these predictions has been reported(Davis et al. 2007). In this work prediction of interactions were made for human