Structure determination of aldose reductase: joys and traps of local symmetry averaging

The structure of aldose reductase, a monomeric enzyme of 314 amino acids which crystallizes in space group P1 with four monomers per asymmetric unit, has been solved using a combination of single isomorphous replacement (SIR), solvent flattening and local symmetry averaging. The self rotation showed evidence of 222 local symmetry. The map calculated from the original single isomorphous replacement phases showed a clear solvent envelope but was uninterpretable. A first averaging attempt failed because the molecular envelope obtained from the SIR map weighted with monomer correlation was too small and the averaging was biased by low-resolution truncation. A second attempt with an enlarged envelope and including low-resolution reflections succeeded in refining phases at 3.5 A resolution but failed to extend them correctly. Rigid-body refinement of a partial model based on the 3.5 A map calculated from refined phases showed significant departures from the 222 symmetry. A third averaging attempt using the improved symmetry succeeded in producing a clear map with phases extended to 3.07 A resolution. This map revealed a (beta/alpha)(8) fold, not previously found in NADPH-dependent enzymes. This work shows the importance of mask definition and local symmetry elements accuracy for averaging, and describes a method for improving these parameters.