Recognition and separation of single particles with size variation by statistical analysis of their images

Macromolecules may occupy conformations with structural differences that cannot be resolved biochemically. The separation of mixed molecular populations is a pressing problem in single-particle analysis. Until recently, the task of distinguishing small structural variations was intractable, but developments in cryo-electron microscopy hardware and software now make it possible to address this problem. We have developed a general strategy for recognizing and separating structures of variable size from cryo-electron micrographs of single particles. The method uses a combination of statistical analysis and projection matching to multiple models. Identification of size variations by multivariate statistical analysis was used to do an initial separation of the data and generate starting models by angular reconstitution. Refinement was performed using alternate projection matching to models and angular reconstitution of the separated subsets. The approach has been successful at intermediate resolution, taking it within range of resolving secondary structure elements of proteins. Analysis of simulated and real data sets is used to illustrate the problems encountered and possible solutions. The strategy developed was used to resolve the structures of two forms of a small heat shock protein (Hsp26) that vary slightly in diameter and subunit packing.