We will use our joint/combined expertise in cheminformatics, molecular dynamics (MD), structure-based drug design (SBDD), pharmacophore modeling, and medicinal chemistry to generate hits for multiple subcavities of the histone binding groove of the SETDB1 triple Tudor domain (TTD). In a first step, we will use the crystal structures of the SETDB1 TTD provided by the CACHE committee (PDB IDs: 7CJT, 8UWP, 6AU3) and available ligand data to benchmark our state-of-the art docking program (published in scientific literature, proprietary code).[1] Whereas our benchmarking study will act primarily as a sanity check, it will also inform 1) which key amino acids are engaged by the co-crystallized ligands and 2) the development of multiple pharmacophoric models (one or more for each subcavity considered). These models will incorporate both protein and ligand data (i.e., co-crystallized ligand conformation, where available). We will then curate a library of commercially available compounds to screen against our pharmacophoric models. We will start from the Enamine REAL database (either full database or Diversity Set). Irrespective of the chosen starting point, we will filter our library according to Lipinski Ro5 and Veber Ro3 rules using our drug discovery platform (published in scientific literature, proprietary code).[1] To further tailor the library for screening, we will include additional filters like 1) removal of PAINS and reactive groups (i.e., covalent warheads), 2) inclusion of only high QED compounds (> 0.75), 3) removal of any insoluble compounds (predictions made with Simulation-Plus), 4) the Lilly med chem rules, 5) synthetic feasibility constraints (compounds in-stock or easy/medium synthesis), 6) principles outlined in the CACHE white paper, especially fraction of sp3 carbons and logD (pKa computed with our proprietary software) values.[2] For this tailored library, we will generate 25 conformers/compound for pharmacophore screening (based on prior successful in-house benchmark). We will then screen the ligand conformers against the pharmacophore models developed in the first step with our in-house pharmacophore screening tool. Only compounds with a root mean square deviation (RMSD) under a certain threshold (i.e., < 1.0Å) to the models will be kept for further analysis. The compounds that pass the initial pharmacophoric filtering will then be virtually screened with our docking program against the crystal structures of each subcavity we consider. We envision docking at most 250,000 compounds/subcavity. If the number of compounds passing the pharmacophoric filtering is > 250,000, we will cluster them to ensure diversity. Once the docking is completed, we will filter out those binding poses that do not interact strongly with the key amino acids identified during the benchmarking using our in-house protein analysis platform. Then, we will rank the remaining compounds by docking score and select up to 1000 compounds/subcavity for visual analysis. Using our combined/diverse expertise, we will form "hit picking parties" where we will visually assess the binding poses using the following criteria: 1) overall fit of the ligands in the subcavities, 2) relative ligand strain, and 3) number and type of fulfilled and unfulfilled interactions (i.e., hydrogen bonding, π-stacking, hydrophobic contacts, etc.). We will then assign visual scores to these poses, and up to 100 compounds/subcavity with the best visual scores will be chosen for the final refinement step. In this final step, we will use the λ-dynamics method [3] integrated in ATOMFORGE (proprietary software) to calculate the relative binding free energy of these compounds. λ-dynamics is an alchemical method that can simulate the relative binding free energy of multiple ligands in one single simulated system. The λ-dynamics will, on top of the docking and visual results, collect entropic and free energy landscape information from an MD simulation that dynamically samples an ensemble of SETDB1 TTD conformations and different chimera states of the ligands. The ligands will then be ranked based on their relative binding free energy obtained from the λ-dynamics simulation. The top ranked 150 compounds from this refinement step will be proposed for purchase and testing.
References:
1. Moitessier, N. et al., Acc. Chem. Res., 2016, 49(9), 1646-1657.
2. Ackloo, S. et al., Nat. Rev. Chem., 2022, 6, 287-295.
3. Knight, J.L., and Brooks, C.L. 3rd, J. Comput. Chem., 2009, 30, 1692-1700.
