AB-INITIO PHASE DETERMINATION OF PROTEINS WITH HIGH RESOLUTION DATA BY DIRECT METHODS. Monika Mukherjee, Department of Solid State Physics, Indian Association for the Cultivation of Science, Calcutta-700032, India

The direct method program SAYTAN is applied to a known protein structure Rubredoxin (space group P2₁, a=15.97, b=41.45, c=24.41Å, [[beta]] =108.3deg., data upto 1Å resolution) containing 393 protein atomic sites, one Fe atom, a sulfate ion and 102 water molecules. By making 1000 trials at different resolutions with sets of initially random phases SAYTAN yielded ab-initio phases for about 800 reflections with mean phase errors (MPE) of 41, 41, 47, 47, 60deg. for 1, 1.25, 1.5, 1.75 & 2Å resolutions respectively. The phases were extended to 2000 reflections with MPE ~ 62deg.. Conventional Multan figures of merit were not useful for macromolecular structures, but modified figures of merit proposed by Mukherjee & Woolfson (Acta Cryst.,1993, D49, 9 -12) and successfully applied in the cases of Avian pancreatic polypeptide (space group C2, data upto 0.89Å,302 protein sites +Zn +80 water sites) and 2-Zn insulin (space group R3, data upto l.5Å, 806 non-hydrogen atoms + 2Zn + solvent atoms; Mukherjee & Woolfson, Acta Cryst., 1995, D51, 626-628) seem capable of selecting better phase sets in Rubredoxin. This modified FOM's were still heavily based on statistical principles i.e. [[integral]] p^3. dv should be maximum, a key property of the map for small structures which is not often true for macromolecular structures. Thus from the high resolution data (upto 2.5Å) of the structures comprising of upto 1000 non-H and some heavy atoms in the asymmetric unit useful sets of phases can be obtained using SAYTAN strategy. The best maps obtained from the direct-methods having map correlation coefficient coefficient 0.48 should benefit from further phase refinement before model fitting is attempted.

The procedure was found to be effective even in the case of known protein without any heavy atom RNAP1(1.17Å, 808 non-H atoms + 83 water molecules). It may be hoped that this approach will allow routine application of direct method to unknown protein structures yielding high resolution data.