IMPROVING THE QUALITY OF NMR AND CRYSTALLOGRAPHIC PROTEIN STRUCTURES BY MEANS OF A CONFORMATIONAL DATABASE POTENTIAL DERIVED FROM STRUCTURE DATABASES. G. Marius Clore, John Kuszewski, Angela M. Gronenborn, Laboratory of Chemical Physics, Building 5, National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, Bethesda, MD 20892-0520

A new conformational database potential involving dihedral angle relationships in databases of high resolution highly refined protein crystal structures is presented as a method for improving the quality of structures generated from NMR data. The rationale for this procedure is based on the observation that uncertainties in the description of the non-bonded contacts present a key limiting factor in the attainable accuracy of protein NMR structures. The idea behind the conformational database potential is to restrict sampling during simulated annealing refinement to conformations that are likely to be energetically possible by effectively limiting the choices of dihedral angles to those that are known to be physically realizable. In this manner, the variability in the structures produced by this method is primarily a function of the experimental restraints, rather than an artifact of a poor non-bonded interaction model. We tested this approach with the experimental NMR data (comprising an average of about 30 restraints per residue and consisting of interproton distances, torsion angles, ³J_HN[[alpha]] coupling constants, and ¹³C chemical shifts) used to previously calculate the solution structure of reduced human thioredoxin. Incorporation of the conformational database potential into the target function used for refinement (which also includes terms for the experimental restraints, covalent geometry, and non-bonded interactions in the form of either a repulsive, repulsive-attractive or 6-12 Lennard-Jones potential) results in a significant improvement in various quantitative measures of quality (Ramachadran plot, sidechain torsion angles, overall packing). This is achieved without compromising the agreement with the experimental restraints and the deviations from idealized covalent geometry which remain within experimental error, and the agreement between calculated and observed ¹H chemical shifts which provides an independent NMR parameter of accuracy. The method is equally applicable to crystallographic refinement, and should be particular useful during the early stages of either an NMR or crystallographic structure determination and in cases where relatively few experimental restraints can be derived from the measured data (due, for example to broad lines in the NMR spectra or to poorly diffracting crystals).