Simple models of molecular flexibility and their application to drug design and protein folding

An increasing number of structural studies reveal alternative binding sites in protein receptors that become apparent only when an inhibitor binds. The ability to predict these sites would significantly enhance computer-aided drug design efforts. Here we show that an alternative binding site requiring rearrangement of the DFG motif backbone of p38 MAP kinase can be successfully and repeatedly identified in explicit-solvent molecular dynamics (MD) simulations of the protein that begin from an unliganded crystal structure. Ligand-docking calculations performed on 5000 MD-generated structural snapshots indicate that these alternative conformations are often surprisingly competent to bind the inhibitors designed to target this site in their crystallographically correct positions. Further, two quite different, potentially inhibiting conformations of p38's DFG motif are also sampled for extended periods of time. These results suggest existing computational methods may be of surprising utility in predicting cryptic binding sites prior to their experimental discovery.;The processes of protein folding, misfolding and aggregation all involve the relative diffusion and association of protein structural elements. Since partitioning between folding and aggregation pathways often appears to be determined by kinetic, rather than thermodynamic factors, it is anticipated that successful modeling of such events may require accurate descriptions of proteins' diffusive characteristics. Since the latter are expected to be influenced by hydrodynamic effects we have carried out a comprehensive study of the diffusion and folding of 11 model proteins with an established simulation model that has been extended to include hydrodynamic interactions (HI). We show that simulations that include hydrodynamic interactions are able to simultaneously capture expected experimental values for translation and rotation; simulations that do not include hydrodynamic interactions drastically underestimate both. Simulations with hydrodynamic interactions also correctly reproduce the significant decrease in translational diffusion coefficient that accompanies protein unfolding. When the two models are used to simulate folding, the inclusion of hydrodynamic interactions accelerates folding by a factor of 2--3 but causes no obvious change in the folding mechanism. These results strongly suggest that simulation models of protein folding and aggregation with an implicit representation of the solvent are ultimately likely to require some kind of treatment of hydrodynamic interactions.