Limits of ligand selectivity from docking to models: in silico screening for A(1) adenosine receptor antagonists

G protein-coupled receptors (GPCRs) are attractive targets for pharmaceutical research. With the recent determination of several GPCR X-ray structures, the applicability of structure-based computational methods for ligand identification, such as docking, has increased. Yet, as only about 1% of GPCRs have a known structure, receptor homology modeling remains necessary. In order to investigate the usability of homology models and the inherent selectivity of a particular model in relation to close homologs, we constructed multiple homology models for the A(1) adenosine receptor (A(1)AR) and docked ∼2.2 M lead-like compounds. High-ranking molecules were tested on the A(1)AR as well as the close homologs A(2A)AR and A(3)AR. While the screen yielded numerous potent and novel ligands (hit rate 21% and highest affinity of 400 nM), it delivered few selective compounds. Moreover, most compounds appeared in the top ranks of only one model. These findings have implications for future screens.