Enhancing Structure Prediction and Design of Soluble and Membrane Proteins with Explicit Solvent-Protein Interactions.

Solvent molecules interact intimately with proteins and can profoundly regulate their structure and function. However, accurately and efficiently modeling protein solvation effects at the molecular level has been challenging. Here, we present a method that improves the atomic-level modeling of soluble and membrane protein structures and binding by efficiently predicting de novo protein-solvent molecule interactions. The method predicted with unprecedented accuracy buried water molecule positions, solvated protein conformations, and challenging mutational effects on protein binding. When applied to homology modeling, solvent-bound membrane protein structures, pockets, and cavities were recapitulated with near-atomic precision even from distant homologs. Blindly refined atomic-level structures of evolutionary distant G protein-coupled receptors imply strikingly different functional roles of buried solvent between receptor classes. The method should prove useful for refining low-resolution protein structures, accurately modeling drug-binding sites in structurally uncharacterized receptors, and designing solvent-mediated protein catalysis, recognition, ligand binding, and membrane protein signaling.

Computational design of ligand-binding membrane receptors with high selectivity.

Accurate modeling and design of protein-ligand interactions have broad applications in cell biology, synthetic biology and drug discovery but remain challenging without experimental protein structures. Here we developed an integrated protein-homology-modeling, ligand-docking protein-design approach that reconstructs protein-ligand binding sites from homolog protein structures in the presence of protein-bound ligand poses to capture conformational selection and induced-fit modes of ligand binding. In structure modeling tests, we blindly predicted, with near-atomic accuracy, ligand conformations bound to G-protein-coupled receptors (GPCRs) that have rarely been identified using traditional approaches. We also quantitatively predicted the binding selectivity of diverse ligands to structurally uncharacterized GPCRs. We then applied this technique to design functional human dopamine receptors with novel ligand-binding selectivity. Most blindly predicted ligand-binding specificities closely agreed with experimental validations. Our method should prove useful in ligand discovery approaches and in reprogramming the ligand-binding profile of membrane receptors that remain difficult to crystallize.