GPCR Model Validation and Evaluation

Validation of GPCR models by direct comparison to crystallographic data is hindered by a limited number of structures. Nonetheless, the currently available set of GPCR coordinates provides a basic benchmark for testing modeling protocols, and indirect experimental evidence of ligand/receptor interactions can be used to evaluate predictions for uncrystallized receptors. For receptors with several characterized ligands, VLS is an additional metric for assessing binding pocket conformation. Here, we address procedures for validating predicted GPCR conformations in light of the available experimental data.

New modeling protocols should first be tested with several benchmark cases based upon crystallographic data before application to other ligand/receptor complexes. For example, an accurate docking algorithm should reproduce ("self-dock") the conformations of retinal within the bRho ligand binding pocket, carazolol and timolol with the corresponding p2AR structures, cyanopindolol with PiAR, and ZM241385 in complex with AA-aR. ICM has been demonstrated to dock retinal to the bRho pocket within an RMSD of 0.2Á to the crystallographic coordinates [77]. Similarly, (-)-carazolol is docked to p2AR with an RMSD of 0.3 Á, and calculation of the relative binding affinities of (-)-carazolol and (+)-carazolol with ICM provides a rationalization for p2AR stereoselectivity [22] . A stringent second test would then be to cross-dock several known ligands to a single structure to check for consistency in ligand binding pose and ligand/receptor interactions. Additionally, if optimization of the receptor side chains is to be employed, it should be verified that the refinement procedure converges on a minimum energy conformation resembling that of the starting crystal structure. Once this has been achieved, the docking and refinement procedure may be confidently applied to other receptor types.

In the absence of a high-resolution structure for comparison, indirect measures can be used to assess model accuracy. One metric of correct binding pocket conformation and docked ligand pose is the presence of intermolecular ligand/receptor hydrogen bonds. Unsatisfied ligand donor and acceptor atoms result in decreased binding affinity and may indicate an incorrectly docked conformation. Additionally, predicted binding pocket conformation should be globally consistent with experimental mutagenesis results. In the ideal case, experimental binding affinities are available for several ligand analogs with a panel of receptor mutants. Such data provide evidence for specific contacts between a particular receptor side chain and ligand functional group, and can even be used to impose distance restraints during LGM. However, mutagenesis data are often subject to alternate interpretations. A receptor mutation that decreases the apparent agonist binding affinity may act directly by removing an important agonist/receptor interaction or indirectly by affecting the receptor activation mechanism. Further, muta-genesis data alone are often insufficient for evaluating ligand pose. Limited experimental results may indicate an approximate binding location without specifying the ligand orientation. The conformation of long and conforma-tionally flexible ligands is particularly difficult to validate solely via mutagen-esis experiments. Fentanyl, a powerful analgesic that selectively targets the ^-opioid receptor, contains seven rotatable bonds. Despite numerous modeling studies and the availability of extensive mutagenesis and SAR data, the binding mode and molecular determinants of fentanyl recognition remain poorly defined [79]. In such situations, the comparison of binding modes for multiple docked ligands of varying chemotypes and complementarity of the binding pocket conformation with pharmacophore models can improve confidence in a predicted conformation. For example, when modeling the complex between arylpiperazine ligands and the 5HT1A receptor, the known ligand/receptor interaction points are symmetric, preventing unambiguous assignment of correct ligand binding orientation [80]. By docking several conformationally constrained arylpiperazine derivatives and comparing binding conformation with several other diverse ligand chemo-types, Nowak et al. proposed a consistent binding mode for a 5HT1A homology model.

Finally, an accurate GPCR model should provide high enrichment factors and hit rates in VLS trials. This demonstrates that the binding pocket conformation allows recognition for a set of diverse antagonists (or agonists) and energetic partition of these ligands from a set of decoys. For many GPCRs, little mutagenesis or SAR data exist to specify probable ligand/receptor interactions, and VLS may be the primary criteria used for model selection and validation (see Section 15.2.2). Several online databases provide compilations of known GPCR ligands that can be used in constructing a VLS test set, including the KiDB, GPCR ligand database (GLIDA), and DrugBank [81-83]. VLS enrichment for known ligands from a random test set has been demonstrated for several GPCR models, including the dopamine D3, cannabinoid CB2, alpha1a adrenergic, and neurokinin I receptors [84-87]. When available, ligand/ receptor interactions or pharmacophore information may be combined with VLS as a prescreening filter, ligand-receptor restraints, or a component of the docked ligand scoring function. This integrated approach facilitates correct ligand docking and enhances VLS enrichment factors, but requires prior experimental knowledge.

0 0

Post a comment