From Electron Density to a Full Atom Model Suitable for Drug Discovery Refinement of Existing Crystal Structures

Dramatic progress in GPCR expression, purification, and crystallography techniques has recently resulted in high-resolution structures of p2AR, p1AR, and AA2aR in complex with several partial inverse agonists [13-16, 18]. These receptors are presently the only "druggable" GPCRs to have been crystallized and provide a good testing ground for structure-based drug discovery methods. Nonetheless, further computational analysis is required to convert and extend a set of PDB-deposited coordinates to a fully protonated all-atom model suitable for rational drug design. Such 3D model preparation goes beyond standard crystallographic refinement and validation procedures [ 19, 20] , and is usually based upon conformational energy optimization, which may be restrained by (or combined with) electron density map fitting. As functional understanding of ligand binding depends critically on hydrogen bond satisfaction, generation of a fully protonated model focuses on the correct assignment and optimization of polar and ionic interactions. This analysis takes into account: (1) flipping of glutamine and asparagine side-chain orientations; (2) protonation and tautomerization of histidine; (3) appropriate protonation of asparatate, glutamate, and cysteine as defined by the local environment; and (4) placement of rotatable hydrogens in other polar side chains. When recently evaluated, the error in asparagine/glutamine rotamer assignment within the PDB was on the order of 20%, indicating that errors in hydrogen bonding network structure are fairly widespread [21]. Additionally, in conformationally flexible regions displaying high atomic b-factors, the electron density is often not sufficient for correct assignment of side-chain rotamers or ligand confor-mational state. It should be noted that resolutions as high as 2.2A and 2.4A, as observed for bRho (PDB: 1U19) and P2AR (PDB: 2RH1), respectively, may be rather exceptional for GPCRs. High-resolution crystals were obtained for these two receptors due in part to the extreme stability of the dark-adapted bRho state and the picomolar affinity of p2AR with carazolol. Furthermore, local variations in electron density map quality may result in poorly resolved regions of protein structure. For example, in the antibody-bound complex of p2AR, the entire region from the ligand binding pocket to the extracellular domains is unresolved [13]. It is anticipated that many GPCR complexes relevant to drug discovery will be characterized at lower resolutions and display more poorly defined ligand binding interactions. Ligand-guided refinement of the receptor ligand binding pocket, wherein the binding pocket side chains and ligand conformation are flexibly optimized under an empirical energy potential, can be employed to optimize the ligand/receptor interactions and provide a more thorough description of ligand recognition.

Previously, we conducted ligand -guided refinement of a full atom p2AR model derived from the carazolol-bound p2AR structure (PDB: 2RH1) to further optimize the binding pocket conformation and analyze deviations of the computational predictions from the crystal structure [22]. In this procedure, the binding pocket side chains and carazolol ligand were flexibly optimized using biased probability Monte Carlo sampling as implemented in the Internal Coordinates Mechanics (ICM) software package (Molsoft, La Jolla, CA) [23] . This gave rise to a new model including five additional hydrogen bonds in the ligand binding pocket that are missing or suboptimal in the PDB-deposited coordinates. Importantly, the energy-optimized model calculates a binding pocket conformation and (-)-carazolol geometry nearly identical to the crystal structure, with the exception of the rotameric states of Ser203-42 and Ser204543 (Fig. 15.2a). The new rotamer of Ser203542 allows formation of

Figure 15.2 (a) The p2AR binding pocket in complex with (-)-carazolol. The p2AR backbone is shown in gray cartoon and (-)-carazolol in yellow sticks. Receptor side chains from the PDB-deposited coordinates are depicted as gray sticks, while side -chain conformations from a ligand -guided model of the p2AR pocket are shown in orange. The electron density map is contoured at 1.2 o and displayed in blue mesh. (b) The lowest energy conformation of (-) --soproterenol (yellow sticks) in an agonist-bound p2AR model. This conformation was generated by LGM with backbone flexibility in the TM5 proline--nduced kink (residues 205-210) and a portion of ECL2 loop (191-196). The flexible backbone regions are indicated in green ribbon, and the optimized position of the extracellular portion of TM5 is shown as red ribbon. The original position of TM5 and other static helices are indicated in gray ribbon, and flexible side chains in proximity of the binding site have carbon atoms colored green. Hydrogen bonds are shown with dotted lines. (c) Predicted ligand binding affinities (pKpred) for p2AR models with flexible and rigid TM5 backbones. Docking results are shown for a diverse set of ligands ranging from inverse agonists (e.g., carazolol) and partial agonists (e.g., dopamine, MAPE) to full agonists (e.g., isoproterenol and epinephrine). Arrows highlight the most pronounced increases in predicted ligand binding affinity from the rigid backbone model to the flexible TM5 model for the full agonists. This increase is much smaller for partial agonists and negligible for antagonists/inverse agonists. The accuracy of affinity predictions is estimated as R2 = 0.75, RMSE = 1.3pKd for the rigid backbone model, and R2 = 0.89, RMSE = 0.7 pKd for the flexible TM5 model.

2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.011.0 Measured binding affinity, pKd

a hydrogen bond to the Tyr199538 backbone carbonyl and significantly improves its hydrogen bonding distance to the carbazole moiety of carazolol from 3.3 to 2.7 A . Additionally, the predicted rotamer of Ser2045 43 has an improved geometry of hydrogen bonding with the Ala200. 39 backbone and displays a new hydrogen bond between the Ser2045 43 hydroxyl and Asn2936 55 side-chain nitrogen, which is likely to play a role in the basal activity of p2AR [22]. These optimized conformations of Ser2035 42 and Ser2045 43 are consistent with the electron densities observed in the crystal structure yet provide an enhanced view of stabilizing interactions within the ligand binding pocket.

In addition to side-chain placement and protonation state, analysis of GPCR structural data requires special attention due to the conformational flexibility inherent to these receptors. The crystallization conditions, crystal contacts, specific mutations, and fusions employed for receptor stabilization affect the final conformation and should be considered when constructing and evaluating a model. For example, biophysical experiments have shown that the magnitude of activation-related helix mobility is decreased in membrane-associated GPCRs relative to detergent-solubilized receptors [24, 25]. In contrast, the limited conformational changes observed between a photoactivated deprotonated rhodopsin intermediate and the dark-adapted ground state rhodopsin may possibly be explained by physiologically relevant dimerization of the receptor in the crystal structures [26, 27]. Later in this chapter, we discuss conformational differences among the presently available GPCR crystal structures and the impact upon the ligand binding pocket architecture.

0 0

Post a comment