D0102

REFINEMENT OF PROTEINS AT ATOMIC RESOLUTION. Victor S. Lamzin¹, Thomas R. Schneider¹, Zbigniew Dauter^1,2 Keith S. Wilson^1,2, ¹[[opthyphen]]EMBL Hamburg Outstation, c/o DESY, Notkestraße 85, 22603 Hamburg, Germany; ²[[opthyphen]]Department of Chemistry, University of York, Heslington, York YO1 5DD, UK

For small molecules X-ray data can be recorded to atomic resolution and positions of ordered atoms identified with an error of about 0.002[[opthyphen]]Å. Particular problems for proteins involve their bigger size and disorder. Lack of data causes difficulties at all stages of structure analysis. Advances in recent years, area detectors, synchrotron sources and cryogenic freezing, allow recording of atomic resolution data for at least a subset of protein crystals. Currently data, extending to at least 1.2 Å, have been collected by visitors and in-house for about 40 proteins at EMBL Hamburg alone. No longer are these only small tightly packed systems such as rubredoxin: the list includes alcohol dehydrogenase with 80 kDa in the asymmetric unit.

There has generally been a model available giving an initial R factor about 30 %. The model is refined with stereochemical restraints and isotropic temperature factors. Subjective inspection and building of water structure becomes increasingly time consuming as more potential sites emerge. Semi-objective criteria for water selection on the basis of distance and electron density have been adopted. Introduction of hydrogen atoms riding on their parent atoms reduces the R factor by about 1 %. The isotropic models typically have R factors of 14 to 18 %. Anisotropic atomic thermal parameters are then refined leading to final values of R factors of 8 to 12 %. A final cycle of block-matrix minimisation provides a reliable estimate of coordinate error from inversion of the normal matrix.

At atomic resolution anisotropic refinement of thermal motion is clearly valid. The improvement in the maps allows easier identification of solvent and disordered residues. The main chain atoms in the ordered parts have a coordinate error of about 0.03[[opthyphen]]Å, the average for the whole structure is 0.05[[opthyphen]]Å. On introduction of anisotropy Rfree falls by almost as much as the R factor. Having established the protocol it is unnecessary to assess anisotropy with Rfree for each subsequent refinement. The last cycles must include all data, even those previously omitted for the Rfree.

The number of atomic resolution protein structures will increase and within the next years provide a phenomenal data base for detailed analysis. Preliminary comparison of protein stereochemistry has already showed significant deviations from parameters derived from small molecules.