Ntroduction

Despite a common structural scaffold of seven transmembrane helices, the G protein-coupled receptor (GPCR) superfamily enables recognition of a broad range of extracellular stimuli encompassing odorants, peptides, lipids, biogenic amines, hormones, nucleotides, and even light (Fig. 15.1). Phylogenetic analysis of the human GPCR superfamily has identified 860 distinct receptors, of which approximately 300 are nonolfactory [1, 2]. Given this remarkable diversity in sequence and function, dissecting the relationship between primary amino acid sequence, receptor structure, ligand structure, and ligand activity is nontrivial. In many instances, a single ligand binds multiple receptor types, possibly with differing activities at each. Similarly, a single receptor may recognize a wide array of structurally diverse ligands. Rational GPCR drug discovery, which requires the identification of novel high-affinity GPCR ligands with a predicted specificity and activity, is therefore a particularly complex problem.

As an additional complication, the experimental characterization of GPCR structures is extremely challenging, and the conformational changes associated with ligand binding remain largely unknown. Until recently, the only GPCR to be crystallized was bovine rhodopsin (bRho). The rhodopsin structures supply essential information for constructing three-dimensional

GPCR Molecular Pharmacology and Drug Targeting: Shifting Paradigms and New Directions,

Edited by Annette Gilchrist

Copyright © 2010 John Wiley & Sons, Inc.

Figure 15.1 General GPCR structure. Panel (a) depicts the extracellular view of bRho (PDB: 1U19). The receptor backbone is shown in ribbon and is color coded from the N- to C-terminus (blue to red). The full inverse agonist retinal is shown in yellow sticks. Panel (b) provides a side view of bRho; several Class A conserved microdomains are indicated with yellow circles. These amino acids are thought to play a role in receptor activation. Panel (c) compares the bRho (N- to C-termini color-coded ribbon) and unliganded opsin structures (pale blue ribbon). The unliganded opsin structure corresponds to an activated conformation.

Figure 15.1 General GPCR structure. Panel (a) depicts the extracellular view of bRho (PDB: 1U19). The receptor backbone is shown in ribbon and is color coded from the N- to C-terminus (blue to red). The full inverse agonist retinal is shown in yellow sticks. Panel (b) provides a side view of bRho; several Class A conserved microdomains are indicated with yellow circles. These amino acids are thought to play a role in receptor activation. Panel (c) compares the bRho (N- to C-termini color-coded ribbon) and unliganded opsin structures (pale blue ribbon). The unliganded opsin structure corresponds to an activated conformation.

(3D) GPCR models but are poorly representative of ligand binding pocket structure in some respects. Unlike many GPCRs, rhodopsin does not bind a freely diffusible ligand but rather is covalently linked to the inverse agonist 11-cis-retinal, which undergoes photo conversion to the agonist all-trans-retinal. Moreover, the retinal binding pocket displays very low sequence identity to the ligand binding pocket of other receptors, including those within the Class A GPCR family.

Due to these limitations, ligand-based methods, such as structure-activity relationship (SAR) models, 2D and 3D pharmacophores, and chemogenomics analysis, have prevailed in GPCR drug discovery [3, 4]. These approaches do not rely on a 3D receptor structure but instead select new small molecules based on similarity to existing agonists or antagonists. However, ligand-based methods are limited in the novelty of the chemical leads identified and are restricted to receptors where a number of known ligands have already been characterized. Therefore, over the past decade, significant efforts have been made at developing computational techniques for predicting GPCR structure and receptor/ligand interactions. Practical application of a structure-based approach to ligand screening and optimization has nonetheless been limited by the low accuracy of these models, which were initially built by homology to bacteriorhodopsin [5] and, later, bRho [6, 7], or constructed de novo [8—11].

The recent publication of high-resolution structures for the p2-adrenergic (p2AR) [12-14], pi-adrenergic (piAR) [15], and adenosine A2a (AA2aR) [16] receptors in complex with diffusible small molecule ligands ushers in a new era for GPCR modeling. These structures furnish additional templates for homol-ogy model construction and a set of test cases for computational protocols. Nonetheless, experimental characterization of additional high-resolution GPCR structures remains a challenging task because of the conformational flexibility intrinsic to these proteins. This will likely limit the number of solved GPCR crystal structures, as well as the achievable resolution for some receptors. Importantly, even when the crystallographic coordinates provide insight into the interactions underlying ligand binding and receptor activation, these observations are restricted to the specific receptor type and co-crystallized ligand. This limited set of structures must be extrapolated to describe the conformation of hundreds of essential GPCRs in complex with tens of thousands of small molecule modulators. The protein databank (PDB)-deposited coordinates themselves do not specify the location of hydrogen atoms and may display inaccuracies in conformation arising from the uncertainty inherent in electron density fitting. Such errors include incorrect orientations of asparagine or gluta-mine, suboptimal hydrogen bonding networks, or misplaced side chains. In all cases, these errors lead to diminished interpretation of the existing ligand/recep-tor interactions, as well as poor ligand docking and screening performance. Additional complications arise from the existence of multiple ligand-selective receptor conformational states corresponding to differences in ligand potency and functionality. Without the aid of further computational analysis, a single GPCR structure is unable to address the binding pocket conformational changes necessary for recognition of structurally distinct agonists and antagonists [17].

In the absence of comprehensive structural data, computational approaches can leverage existing structures to more accurately interpret the experimental electron density and receptor coordinates, provide insight into other ligand-bound receptor conformations, and model other receptor types. Such methods maximize the benefit of the crystallographic data to the drug discovery community and will gain importance as they are increasingly validated and a growing number of structures are determined. While approximate, low-resolution models of the full-length receptor are suitable for certain tasks, such as selecting amino acid positions for experimental mutagenesis or fluorescent label placement, atomic level accuracy in a select group of binding pocket side chains is required for ligand pose prediction, lead optimization, and virtual ligand screening (VLS). In each case, the availability of experimental data and

TABLE 15.1

Application of 3D Protein Modeling and Ligand Docking Approaches to GPCR Drug Discovery

Structural Data

Current Targets

What Can Be Done

3D Modeling

Drug Design

Available

Methods

Applications

Crystal structure with

PiAR, AA2aR

Refine ligand contacts

Combined energy-

Comprehensive 3D

a lead compound

and H-bonds,

based/electron

model for lead

1-2 Â

predict SAR

density refinement

optimization

Crystal structure with

• PiAR, p2AR,

Predict ligand

Flexible receptor

3D VLS for new

unrelated

AA2aR

induced fit, binding

docking, MRC,

chemical scaffolds.

ligand + SAR data

• GRM1-3-7

mode, SAR,

SCARE

optimization of

(N-terminal

including agonists

new leads

domain)

Modeling

Structure of a close

• Most adrenergic.

Predict ligand binding

Homology modeling

3D lead discovery

resolution

homolog + SAR

5HT, dopamine

mode, SAR, and

and flexible

and optimization.

data

• Nucleotide like

selectivity profile

docking

with support of

• Most Class C

ligand-based

(N-term domain)

methods

Structure of a distant

• Most Class A

Predict ligand binding

Ligand-guided 3D

Combined receptor-

homolog + SAR

• TM pockets in

and selectivity sites.

modeling of

based and

data

Classes B and C

extend SAR

flexible receptor

ligand-based drug

3-5 Â

discovery

Overall 7TM geometry.

All GPCRs,

Predict low-resolution

Global optimization

Predict low-affinity

little or no ligand

including orphans

model, some

with experimental

nonselective

binding data

ligand-receptor

restraints, de novo

ligands, 3D

contacts

methods

support for ligand-

based methods

computational protocol employed will determine the quality and nature of information that may be reliably obtained by receptor modeling (Table 15.1). In the present chapter, we address how computational modeling can augment experimental structure and discuss the strengths and limitations of a structure-based approach in the context of varying resolutions of experimental data. The discussion is generally arranged to proceed from "high-resolution" modeling, wherein extensive experimental data are available and the corresponding models are atomic level in accuracy, to "low-resolution" modeling, where little to no experimental data exist and the resulting predictions are therefore more approximate.

0 0

Post a comment