The overall structure of CysB(88-324) is strikingly similar to those of the periplasmic substrate-binding proteins. A similar fold has also been observed in the cofactor-binding domain of Lac repressor, implying a structural relationship between the Lac repressor and LysR families of proteins. In contrast to Lac repressor, in CysB the twofold axis of symmetry that relates the monomers in the dimer is perpendicular rather than parallel to the long axis of the cofactor-binding domain. This seems likely to place the DNA-binding domains at opposite extremes of the molecule possibly accounting for CysB's extended DNA footprints.

The cysB gene of Klebsiella aerogenes has been cloned, sequenced and shown to complement the cysteine auxotrophic phenotype of Escherichia coli cysB mutants. The K. aerogenes cysB gene is predicted to encode a protein of 324 amino acid residues that shares approx. 95% sequence similarity with the Salmonella typhimurium and E. coli CysB proteins. Gel-retardation assays demonstrate that the purified protein binds to DNA fragments containing either the K. aerogenes cysb promoter or the S. typhimurium cysJIH promoter. Acetylserine enhances CysB binding to the cysJIH promoter fragment while diminishing its binding to the cysB promoter fragment. Fluorescence-emission-spectroscopy measurements suggest strongly that N-acetylserine binds to CysB apoprotein but that O-acetylserine does not, and support the notion that N-acetylserine is the physiological inducer of cysteine biosynthesis.

The elucidation of the structure of CysB(88-324) by multiple isomorphous replacement (MIR) techniques was seriously delayed by problems encountered at every stage of the analysis. There was extensive non-isomorphism both between different native crystals and between native and heavy-atom-soaked crystals. The heavy-atom substitution was invariably weak and different soaking experiments frequently led to substitution at common sites. These correlated heavy-atom binding sites resulted in an overestimation of the phase information. Missing low-resolution reflections in the native data set, constituting only 2% of the total observations, reduced the power of density modification and phase refinement. Finally, the extensive dimer interface made it difficult to isolate a single molecule in the course of model building into the MIR maps. The power of maximum likelihood refinement (REFMAC) was exploited in solving the structure by means of iterative cycles of refinement of a partial model, initially comprising only 30% of the protein atoms in the final coordinate set. This technique, which uses experimental phases, can automatically discriminate the correct and incorrect parts of electron-density maps and give properly weighted combined phases which are better than the experimental or calculated ones. This allowed the model to be gradually extended by manual building into improved electron-density maps. A model generated in this way, containing just 50% of the protein atoms, proved good enough to find the transformations needed for multi-crystal averaging between different crystal forms. The averaging regime im-proved the phasing dramatically such that the complete model could be built. The problems, final solutions and some possible causes for the observed lack of isomorphism are discussed.

No abstract