Induced Fit Docking
Schrödinger has developed technology that accounts for receptor flexibility in ligand-receptor docking by iteratively combining rigid receptor docking [using Glide (1,2)] with protein structure prediction and refinement [using Prime (3–5)] (6). While traditional rigid-receptor docking methods work well when the receptor structure does not change substantially upon ligand binding, success is limited when the protein conformation must change in order to accommodate the correct binding conformation of the ligand. Schrödinger's induced fit docking (IFD) protocol accounts for both small backbone relaxations in the receptor structure as well as significant side-chain conformational changes. This IFD protocol has been validated on a large set of pharmaceutically relevant examples with surprisingly good results (6). In a study of 21 cases requiring a wide range of receptor movements to properly accommodate particular ligands, traditional rigid-receptor docking yields an average ligand root-mean-square deviation (RMSD) of 5.5 Å, while the average ligand RMSD for IFD is 1.4 Å, and in 18 cases the RMSD is less than 1.8 Å (6). As seen in Figure 1, over 95% of the cases from IFD have an RMSD less than 2 Å, when compared with less than 20% for rigid-receptor docking.
The IFD protocol described above has recently been extended to allow for full flexibility in loop regions. In a study of the activation loop in p38 MAP kinase, the automated IFD protocol was successfully used to generate the DFG-out conformation starting from a DFG-in structure (1a9u) and the ligand from 1kv1 (BMU). The high degree of similarity between the IFD structure and 1kv1 (ligand RMSD = 1.15 Å) is striking given the significant difference between the DFG-out structure and the starting DFG-in structure (Figure 2).
In a retrospective virtual screening study of 25K decoy ligands and 46 known actives, using an ensemble consisting of the IFD structure (DFG-out) and the 1a9u crystal structure (DFG-in), 14 actives were identified in the top 1% of the database, including BMU and BIRB 796. This is compared to only three actives when 1a9u was used alone. In summary, we have produced, using a fully automated protocol, the DFG-out conformation of p38 MAPK starting from the DFG-in conformation and an inhibitor that binds to DFG-out. Combining the induced-fit structure in an ensemble with the original DFG-in structure dramatically increased enrichment of actives in a database screen. These results demonstrate that it is possible with IFD to produce viable receptor structures that as part of an ensemble with other conformations of the receptor can produce significantly higher enrichment factors in database screening.