Improvements and Extensions in the Conformational Database Potential for the Refinement of NMR and X-ray Structures of Proteins and Nucleic Acids

Abstract

ment by simulated annealing is to circumvent the poor diswhere Pi is the probability of occurence of a particular concriminatory power of conventional nonbonded interaction formation, and kDB a scale factor. Several features of the terms (either attractive–repulsive or purely repulsive) benew conformational database potentials are noteworthy. In tween high and low probability local conformations. This is the original version (1) , the f /c distributions were partiachieved by biasing the sampling during simulated annealing tioned into three groups: Gly, Pro, and all other residues. refinement to conformations that are energetically possible The higher resolution structure database, however, permits by limiting the choices of dihedral angles to those that are a further partitioning of the non-Gly and non-Pro residues known to be physically realizable. In the previous work (1) , into four groups: residues with a hydrogen bond donor or the conformational database potentials were generated from acceptor in the g or d position, residues preceeding a proline, two protein crystal structure databases: the PROCHECK datab-branched residues, and the remainder. Subtle but consisbase (2 ) and a backbone-dependent rotamer database (3) tent differences in the f /c distributions are observed bederived from 160 and 170 X-ray structures, respectively, retween these groups (see Figs. 1A and 1B). In the original fined at a resolution of 2.0 A or better. In the present Commuapplication (1 ) , the side-chain torsion angle correlations nication, we extend and improve the protein conformational were limited to one-dimensional potential energy surfaces database potential by employing a new database of 70 diverse for the x1 , x3, and x4 angles, two-dimensional potential X-ray structures refined at 1.75 A or better (4) . Because of energy surfaces of x1 /x2, and three-dimensional potential the higher quality of the structures present in this database, energy surfaces of f /c/x1 for each residue. In the current we are able to incorporate the dependence of the peptide conformational database potentials, all one-dimensional pocovalent geometry on the f and c backbone torsion angles, tential energy surfaces have been eliminated with the excepand to partition thef /c, f/c/x1,x1 /x2, andx1 /x2 /x3 distrition of x4 for Lys and Arg, and have been replaced by twobutions into different groups according to amino acid types. In dimensional x1/x2 and x2/x3 potential energy surfaces and addition, the same concept is used to generate conformational three-dimensional f /c/x1 (with distinct groupings; see Tadatabase potentials for nucleic acids. ble 1 ) and x1 /x2/x3 potential energy surfaces. Examples of Protein backbone and side-chain torsion angles, as well as some twoand three-dimensional potential energy surfaces backbone bond angles, were obtained from the database of are shown in Figs. 1 and 2, respectively. 70 highly refined, high-resolution (o1.75 A) protein crystal In the case of the Leu x1/x2 distributions, a number of structures assembled by Karplus (4). Residues with temperapotentially false rotamers may be present in the database ture factors u25 A and those bordering a cis peptide bond owing to experimental ambiguity for the x1/x2 values of were eliminated from the analysis. The resulting torsion angle Leu residues in structures derived from medium resolution correlations (see Table 1 ) were processed into potential of X-ray analyses. In particular, the well-staggered rotamer with x1 /x2 values near 060 /180 occupies approximately the same space as a poorly staggered rotamer with x1 /x2 * To whom correspondence should be addressed.