Three-dimensional quantitative structure–activity relationship for several bioactive peptides searched by a convex hull–comparative molecular field analysis approach

Abstract

Three-dimensional (3D) convex hulls are computed for theoretically generated structures of a group of 18 bioactive tachykinin peptides. The number of peptides treated as a training set is 14, whereas that treated as a test set is four. The frequency of atoms of the same atomic type lying at the vertices of all the hulls computed for all the structures in a structural set is counted. Vertex atoms with non-zero frequency counted are collected together as a set of commonly exposed groups. These commonly exposed atoms are then treated as a set of correspondences for aligning all the other 13 structures in a structural set against a common template, which is the structure of the most potent peptide in the set using the FIT module of the sybyl 6.6 program. Each aligned structural set is then analyzed by the comparative molecular field analysis (CoMFA) module using the C.3 probe having a charge of +1.0. The corresponding cross-validated r² values range from −0.99 to 0.57 for a number of 73 structural sets studied. The comparative molecular similarity indices analysis (CoMSIA) module within the sybyl 6.6 package is also used to analyze some of these aligned structural sets. Although the CoMSIA results are in accord with those of CoMFA, it is also found that the CoMFA results of several structural sets can be improved somewhat for conformations of the structures in the sets that are adjusted by constraint energy minimization and then constraint molecular dynamics simulation runs using distance constraints derived from some commonly exposed groups determined for them. This result further implies that the convex hull–CoMFA is a feasible approach to screen the bioactive conformations for molecules of this class.

Introduction

Three-dimensional (3D) quantitative structure–activity relationship (QSAR) methods are valuable tools that one can use to improve the quality of some preliminary drug designing models. It is known that molecules of different structures but presenting similar biological activity are difficult to superimpose on each other using the conventional atom-by-atom fit (Dammkoehler et al., 1989; Folkers et al., 1993; Dunn et al., 1998; Klebe, 1998). In such cases, 3D QSAR methods can help to find conformations that can be fitted to a certain pattern known as important for affinity (Folkers et al., 1993; Klebe, 1998). The comparative molecular field analysis (CoMFA) method (Cramer et al., 1988a, Cramer et al., 1988b) is emerging as a popular 3D QSAR method. The interactions of molecules at lattice points outside the molecules are computed and analyzed in the method (Cramer et al., 1988a, Cramer et al., 1988b). The success of a CoMFA model is strongly dependent on how the molecules are aligned in the lattice on which both the steric and electrostatic fields are sampled. Structures are usually aligned against a common template using a fitting algorithm that minimizes the root-mean-square (RMS) distances between specified pairs of atoms. The CoMFA method can also be used to screen bioactive conformations for highly flexible molecules (Waller and Marshall, 1993). This approach requires deriving as many CoMFA models as there are plausible conformations that exist and picking the one that has the best predictive power. In deriving a CoMFA model, it is believed that obtaining the self-consistency of an alignment is much more important than knowledge on active site geometry (Cramer et al., 1988a, Cramer et al., 1988b; Martin et al., 1996).

In this report, we have used a 3D convex hull computation algorithm developed in-house (Lin et al., 1998) to detect the commonly exposed groups for structures of some bioactive tachykinin peptides generated from molecular dynamics (MD) simulations. We treated the commonly exposed groups determined as sets of correspondences for aligning the structures generated using the sybyl 6.6 program (Tripos, 1999). The number of peptide compounds studied is 18. The peptides are divided into a training set and a test set; the former contains 14 peptides and the latter contains four peptides. The peptide structures generated in each training set are also randomly permuted and analyzed by the CoMFA program (Cramer et al., 1988a, Cramer et al., 1988b) to show that the CoMFA statistics obtained for structures aligned by the convex hull alignment rules are not chance statistics. Several of these aligned training sets were also analyzed with the comparative molecular similarity indices analysis (CoMSIA) module (Klebe, 1998) within the sybyl 6.6 package (Tripos, 1999). It is found that while the CoMSIA statistics computed for indices of steric, electrostatic, and hydrophobic effects are correlated well with those of the corresponding CoMFA result, those computed for indices of donor and acceptor are not. The CoMSIA result provides the detailed contribution of each of the correlated fields to the total CoMFA statistics after the structural alignment. We also demonstrate that the CoMFA results can be improved using distance constraints derived from vertices of 3D convex hulls computed for structures of some of the aligned training sets. Since the most important task in CoMFA is to align all the structures in a structural set against a common template, we attempt to adjust the conformation of each structure in the set using some distance constraints so that it will be better correlated with the structure of the common template. Each of the adjusted conformations is generated through a constraint energy minimization and then a constraint MD simulation run using the distance constraints derived from the convex hull vertices determined. We find that the CoMFA result is improved when a second CoMFA is performed on a set of structures with adjusted conformations generated and aligned using the same set of commonly exposed groups on which the distance constraints are derived as the set of correspondence.