How to escape from model bias with a high-resolution native data set - structure determination of the PcpA-S6 subunit III.

The structure of procarboxypeptidase A-S6 subunit III, a truncated zymogen E, has been determined by molecular replacement using as search model porcine elastase 1 which, as revealed by crystallographic analysis, contained about 20% of the amino acids in a radically different orientation. Two monoclinic crystal forms were used: the first one diffracts to 2.3 A resolution and contains one molecule per asymmetric unit; the second diffracts to 1.7 A resolution and contains two molecules per asymmetric unit. Molecular replacement and conventional X-PLOR refinement led to a model for which 20% of the chain was ill defined in both crystal forms. To remove the bias introduced by the initial model, an automated refinement procedure [Lamzin & Wilson (1993). Acta Cryst. D49, 129-147] was applied successfully to the second crystal form, which diffracts to high resolution. The resulting dramatic improvement of the electron-density map led to extensive rebuilding of some surface loops. The reliability of the modified model was confirmed by refinement of the first crystal form. For the two forms, the final R factor is 18.8% for data between 8.0 and 2.0 A resolution, and 18.4% for data between 8.0 and 1.7 A, respectively.