Computer-aided modeling of activity and selectivity of quinazolinones as noncompetitive NMDA receptor antagonists

Abstract

NMethylDaspartic acid (NMDA) receptor, belonging to the group of glutamate ionotropic recep� tors, is a very important biological target. It is involved in many significant neurophysiological processes in the central nervous system associated with the transfer of fast synaptic excitation, formation of memory, etc. Hyperactivation of NMDA receptor leads to a number of pathological conditions, including various neuro� degenerative diseases. For this reason, its antagonists and blockers were shown to be effective as neuropro� tectors, in particular, in treatment of Alzheimer's dis� ease. In terms of architecture, NMDA receptor has a heterotetrameric structure. Usually it consists of two glycinebinding subunits GluN1 and two glutamate� binding subunits GluN2. The neurophysiological pro� file of a compound is determined primarily by the selectivity for different GluN2 subtypes, which are located predominantly in different brain structures. In particular, selective antagonists of GluN2D are of considerable interest for the treatment of Parkinson's disease (1). The purpose of this work was to study the rela� tionships of the structure with activity and selectiv� ity for four subtypes of the GluN2 subunit (GluN2A, GluN2B, GluN2C, and GluN2D) in a series of quinazolin�4�one derivatives, noncompet� itive antagonists of the receptor. Within the frame� work of the method used in the study—molecular field topology analysis (MFTA) (2-4)—a model of bioactivity is constructed on the basis of consider� ation of values of local molecular descriptors (prop� erties of atoms and bonds) in related structures and analysis of their effect on activity. A common frame of reference for a meaningful comparison of such local properties is provided by a molecular super� graph, which is constructed by topological superpo� sition of twodimensional structures of compounds (structural formulas) of a training set and allows each structure in the training set to be superim� posed. In addition to the predictive model, which is based on partial least squares regression and corre� lates these properties in all positions of the molecu� lar supergraph with bioactivity, this method makes it possible to obtain a graphic map of their influence on activity.