Range and sensitivity as descriptors of molecular property spaces in dynamic QSAR analyses

In this paper, we report the first study aimed at correlating pharmacological properties with molecular parameters derived from the physicochemical property space of bioactive molecules. A dataset of 36 ligands of the alpha(1a)-, alpha(1b)-, and alpha(1d)-adrenoceptors as published by Bremner et al. (Bioorg. Med. Chem. 2000, 8, 201-214) was used. One thousand conformers were generated for each ligand by Monte Carlo conformational analysis, and four 3D-dependent physicochemical properties were computed for each conformer of each ligand, namely virtual lipophilicity (log P), dipole moment, polar surface area (PSA), and solvent-accessible surface area (SAS). Thus, a space of four physicochemical properties was obtained for each ligand. These spaces were assessed by two descriptors, namely their range and their sensitivity (i.e., the variation amplitude of a given physicochemical property for a given variation in molecular geometric properties). Little or no correlation was found to exist between the physicochemical properties and their range or sensitivity, indicating that the latter descriptors do not encode the same molecular information as the former properties. As expected, neither the range nor the sensitivity of any of the four physicochemical properties correlated with receptor affinities. In contrast, range and sensitivity showed promising correlations with deltapK(a-b) (i.e., the alpha(1a)/alpha(1b) selectivity) for the complete dataset. The correlations were lower for deltapK(a-d) (i.e., the alpha(1a)/alpha(1d) selectivity), whereas there was no correlation at all with deltapK(b-d). These results are consistent with the results of Bremner et al., which indicate that the alpha(1a)-AR ligands bind in an extended geometry, whereas the alpha(1b)-AR and alpha(1d)-AR ligands assume more folded conformations. Since the property space descriptors presented here take structural variability into account, their correlation with deltapK(a-b) and deltapK(a-d) indicates that these selectivities are indeed driven by differences in conformational behavior and hence in property spaces.