Addressing Protein Flexibility and Ligand Selectivity by “in situ Cross‐Docking”

To overcome the “single-structure paradigm“ in current methods for computational protein–ligand docking, we have recently introduced the “in situ cross-docking“ (ISCD) approach to simultaneously address multiple targets, using the grid-based AutoDock program as search engine. Whereas the feasibility of ISCD for dealing with dissimilar binding sites and non-crossreactive, tight-binding ligands had been demonstrated, it remained to be shown whether it could also be applied to different conformations of the same target (to address protein flexibility) or to closely related targets binding the same ligand with varying affinity (to address selectivity). Here, we investigate the first issue using aldose reductase as a test case, and the second using a recently introduced series of thrombin and trypsin inhibitors. Aldose reductase (AR), a target against late-onset diabetic complications, catalyzes the reduction of aldoses and other aldehydes to the corresponding alcohols. Its substrate promiscuity is in part due to an “induced-fit”-like mechanism of ligand binding, whereby a specificity pocket can be closed or opened in different conformations, depending on the ligand being bound. Crystal structures of AR–inhibitor complexes have revealed three major binding-pocket conformations, best represented by the complexes with sorbinil (PDB 1AH0), tolrestat (1AH3), and IDD594 (1US0) as shown in Figure 1. For docking and structure-based ligand design AR poses the obvious problem that a single conformation of the protein is not sufficiently representative as a target structure; instead, at least three major conformations need to be addressed. In standard docking, this would be done sequentially, using each protein conformer for separate docking simulations, thus requiring for each ligand as many separate simulations as there are protein conformers to investigate. With ISCD, instead, the conformers can be combined to a single search space such that only one simulation must be run per ligand. Using 1AH0, 1AH3, and 1US0 as structures for the three AR binding-site conformers, separate AutoDock grids were first calculated for each of them (further details about the methods are provided as Supporting Information A) and AutoDock runs were carried out on the separate single grids, proving that the experimental binding mode can indeed be reproduced by standard flexible docking (sorbinil to 1AH0: docking result on rank 1 shows a root-mean-square deviation (RMSD) of 0.26 @ with respect to the crystal structure; tolrestat to 1AH3: 0.94 @; IDD594 to 1US0: 1.39 @ for rank 1, 0.84 @ for rank 2. A detailed tabular report of these results is provided as Supporting Information B). This is obviously a prerequisite for testing the ISCD approach, which can only be successful if single-grid docking with the applied scoring function is able to provide the correct binding mode. Clearly, an appropriate scoring function is essential for any (cross-) docking procedure. To setup ISCD, the single grids were then combined to a joined grid with repulsive layers between them. The joined grid representing all three binding-site conformers was used to test whether ISCD is able to identify the native binding pocket of a given ligand (sorbinil, tolrestat, and IDD594) in a single docking calculation. For this purpose the same standard docking parameters were used as applied before in the singlebinding-site docking. Table 1 illustrates that with ISCD the