Ruben Abagyan, Ph.D., professor at the Skaggs School of Pharmacy & Pharmacology at the University of California, San Diego, described efforts to expand the x-ray crystallographic structures of the dozen GPCRs determined so far to predictive models of hundreds of GPCRs in different conformational states.
Dr. Abagyan noted that with recent work the community now has the crystal structures for nine GPCRs and will soon have the crystal structures for two or three more. This represents an “incredible advance,” he said, over a decade ago when we had the crystal structure of rhodopsin, but of no other GPCR. Nevertheless, there are estimated to be approximately 1,000 different GPCRs in the human genome, and so the current structural knowledge of a handful of GPCRs represents only 1% of the total.
Dr. Abagyan emphasized that improved modeling techniques promise to expand the scope of GPCR proteins whose structures are understood, as well as increase the understanding of how a given GPCR binds to chemically different agonists, antagonists, inverse or partial agonists, and allosteric modulators.
In particular, he described three different levels of structure-ligand binding computational predictions that can be carried out today. The easiest approach is if the crystal structure of the GPCR is known. In such cases, Dr. Abagyan showed that a virtual ligand screening approach using the latest docking tools and ICM (Internal Coordinate Mechanics) software demonstrated an over 90% success rate in predicting the correct ligand binding pose in a docking exercise to a single cognate pocket determined by crystallography, and 90% reliability in a more realistic cross-docking exercise if multiple experimental pocket conformations were considered.
This represents “an outstanding result,” Dr. Abagyan said, when compared with the typical cross-docking success rates of 30% to 70% obtained with previous approaches. Moreover, in addition to the docking pose of active molecules, he showed for two better-characterized GPCRs that the docking scores are now capable of separating thousands of actives from inactive molecules.
The second, more challenging approach applies when the GPCR crystal structure is not known, but that of a homologue is. In such cases, knowledge of previously discovered receptor ligands provides key information that can be used for helping the modeling process and improving docking and screening performance of the receptor.
Dr. Abagyan demonstrated how the so-called “ligand-guided optimization of homology models” improves the initially inaccurate models of the ligand-binding pockets into an ensemble of highly predictive models. The new implementation, abbreviated as Alibero or LiBERO, improves the outcome of the procedure, Dr. Abagyan said.
The third and most difficult approach is that for the overall model of a GPCR or an allosteric pocket that forms upon binding. For these cases, Dr. Abagyan’s group and his colleagues at Molsoft have developed an improved homology modeling approach that includes several essential elements: (i) a new force field that was tested for a better ability to predict loop conformations, (ii) a new membrane implicit solvation model, and, (iii) an adequate atom-level transfer of the distance restraints from a template to the model.
The approach can be further assisted with what Dr. Abagyan termed as “fumigation” because it offers the potential of opening up the binding site for interaction with a potential small molecule modulator.