ACCURATE HIGH-RESOLUTION PHASES FROM MAD ANALYSIS: EXPERIMENT AND COMPARISON OF PHASING METHODS. William I. Weis⁺, Kevin M. Flaherty⁺, F. Temple Burling*, Axel T. Brunger*, ⁺Dept. of Structural Biology, Stanford University School of Medicine, Stanford, CA, USA, *Howard Hughes Medical Institute and Dept. of Molecular Biophysics and Biochemistry, Yale University, New Haven, CT, USA

A Yb³⁺-substituted fragment of rat mannose-binding protein A was used in a multiwavelength anomalous dispersion (MAD) experiment to generate a highly accurate and complete set of experimental phases to 1.8 Å resolution (Burling, Weis, Flaherty and Brunger (1996) Science 271: 72-77). Data were collected at four wavelengths on the X4A beamline at NSLS from a single frozen crystal, using inverse-beam geometry to collect Bijvoet mates. Phases for 99% of the possible reflections to 1.8 Å Bragg spacings were obtained using the least-squares method (MADLSQ) of Hendrickson ((1991) Science 254: 51-58), applied to the scaled but unmerged data. An MIR-type probabalistic approach, employing a maximum-likelihood refinement of the heavy atom parameters to provide more realistic figures-of-merit, was also used to phase the data. The combination of the large anomalous scattering effects at the Yb LIII edge and a large partial structure contribution from 4 Yb³⁺ in the 230-residue asymmetric unit produces extremely large anomalous differences that cause a breakdown of the assumptions commonly used when treating MAD data as an MIR problem. A comparison of phasing by MADLSQ, standard MIR programs, and our modified probabalistic method (see accompanying abstract by A.T.B, F.T.B., K.M.F. W.I.W., session 02.04) will be presented. This set of experimental phases will be used for testing improved refinement methods, including bulk solvent and multi-conformer models.