Comparative molecular field analysis (CoMFA) study of epothilones – tubulin depolymerization inhibitors: Pharmacophore development using 3D QSAR methods

Abstract

A three-dimensional quantitative structure-activity relationship (3D QSAR) study has been carried out on epothilones based on comparative molecular field analyses (CoMFA) using a large data set of epothilone analogs, which are potent inhibitors of tubulin depolymerization. Microtubules, which are polymers of the α/β-tubulin heterodimer, need to dissociate in order to form the mitotic spindle, a structure required for cell division. A rational pharmacophore searching method using 3D QSAR procedures was carried out and the results for the epothilones are described herein. One-hundred and sixty-six epothilone analogs and their depolymerization inhibition properties with tubulin were used as a training set. Over a thousand molecular field energies were generated and applied to generate the descriptors of QSAR equations. Using a genetic function algorithm (GFA) method, combined with a least square approach, multiple QSAR models were considered during the search for pharmacophore elements. Each GFA run resulted in 100 QSAR models, which were ranked according to their lack of fit (LOF) scores, with a total of 40 GFA runs having been performed. The 40 best QSAR equations from each run had adequate fitted correlation coefficients (R from 0.813 to 0.863) and were of sufficient statistical significance (F value from 7.2 to 10.9). The pharmacophore elements for epothilones were studied by investigating the hit frequency of descriptors (i.e. the sampling probabilities of grid points from the GFA studies) from the set of the 4000 top scoring QSAR equations. By comparing the frequency with which each grid point appeared in the QSAR equations, three candidate regions in the epothilones were proposed to be pharmacophore elements. Two of them are completely compatible with the recent model proposed by Ojima et al. [Proc. Natl. Acad. Sci. USA, 96 (1999) 4256], however, one is quite different and is necessary to accurately predict the activities of all 166 epothilone molecules used in our training set. Finally, by visualizing the 35 most probable grid points, it was found that changes related to the C6, C7, C8, C12, S20, and C21 atoms of the epothilones were highly correlated to their activity.