Scanning Microscopy

Helical structures are more disordered and diffract more weakly than two-dimensional crystals, and images of them are inherently noisier. Automated techniques allowed us to correct, align and merge data from hundreds of images. To extend from ~1.0 nm resolution to the near atomic level (~0.4 nm), we needed to know whether our handling of images could be improved. For example, curved particles after correction may or may not produce data as good as particles that need no correction. We also wanted to evaluate the accuracy with which we predict the locations of layer lines hidden by noise and determine if all images contribute equally to higher resolution data. For this purpose, we developed an algorithm (the sniffer) which evaluates regions of Fourier transforms that contain signal buried in noise. We examined both the image handling procedures and the sniffer using images of bacterial flagellar filaments embedded in glucose and phosphotungstic acid. Although there was a correlation between phase residual and such factors as tilt, curvature and disorder, these defects explained only ~5% of the variance. This suggested that helical particles following correction contribute almost as much as particles requiring no correction. In addition, positions of high resolution layer lines appeared accurately predicted from lower resolution layer lines. We also found that the signal-to-noise ratio in low resolution layer lines only weakly correlated with that of high resolution layer lines.