A comparison of the CHARMM, AMBER and ECEPP potentials for peptides. I. Conformational predictions for the tandemly repeated peptide (Asn-Ala-Asn-Pro)9.

A search for low-energy helical and near-helical conformations of the tandemly repeated peptide (Asn-Ala-Asn-Pro)9 was undertaken by minimization of the CHARMM potential energy function from eight starting conformations; the latter were obtained from the two low-energy conformations of this repeated peptide found by Gibson & Scheraga, Proc. Natl. Acad. Sci. USA 83, 5649-5653 (1986), and the single conformation found by Brooks et al., Proc. Natl. Acad. Sci. USA 84, 4470-4474 (1987), and from modifications of these three conformations. The same eight starting conformations, as determined by dihedral angles, were used for minimizations of the AMBER and ECEPP potentials. Comparison of the final conformations by least-squares superposition of their C alpha atoms, and by inspection of the parameters of the ideal helix or coiled coil that most closely matched the coordinates of their C alpha atoms in a least-squares sense, showed that: (1) energy minimization, starting from the same conformation but using any two different potentials, could lead to final conformations whose resemblance to each other varied from acceptable to highly unsatisfactory; (2) the ordering of the final energy-minimized conformations, and the energy differences between them, were quite different for all three potentials; (3) the extent of agreement or disagreement between pairs of conformations generated using CHARMM and AMBER, CHARMM and ECEPP, or AMBER and ECEPP, respectively, was not significantly different. The lowest-energy conformation generated using each of the potentials was a left-handed helix, whose pitch and number of residues per turn were similar to those of the left-handed helix found by Gibson & Scheraga. Although the starting conformation which led to the lowest-energy conformation was different for all three potentials, pairwise superposition of the C alpha atoms in the final conformations showed root-mean-square deviations of only 1.0-1.3 A. It is concluded that energy minimizations starting from a large enough sample of initial conformations might on occasion lead to essentially the same conformational prediction whichever potential is used; however, if the sample of starting points is small, predictions based on the three potentials will usually diverge.