III.H. INTERPRETATION AND DISPLAY OF IMAGE ANALYSIS RESULTS

Once one carefully establishes that all processing manipulations have been correctly applied, the most important step of any structural study is to objectively assess and interpret the observed structural features. The effects of specimen preparation, imaging, and processing conditions (including graphical display methods) all directly influence the appearance of specimen features. The objectivity by which structural features revealed in image reconstructions are interpreted is often questioned. What is the functional significance of structural details revealed by reconstructions? The problem of interpretation is said to be really one of pattern recognition (Crowther (1976) in Structure-Function Relationships of Proteins, R. Markham & R. W. Horne, eds., North Holland Pub., Amsterdam, pp. 15-25). Crowther advocates the combination and correlation of structural information from micrographs with other types of experiments (i.e. biochemical, genetic, immunological, or other complementary structural information such as x-ray or neutron diffraction and model building) in order to improve our understanding of biological structure-function relationships. The work of Crowther, et al. (J. Mol. Biol. (1977) 116:489-523) on the molecular reorganization in the baseplate of bacteriophage T4 is an excellent example which shows how different types of data are combined and interpreted. In addition, correlative information can be quite beneficial especially when the obtained results are unexpected and controversial.

Combination and correlation of electron microscopy data with measurements obtained by other physical techniques such as x-ray and electron diffraction have become more common. The structure of the purple membrane was determined with structure factor phases measured from transforms computed from micrographs and structure factor amplitudes measured from electron diffraction patterns (Henderson & Unwin (1975) Nature 257:28-32). Electron diffraction amplitudes are more accurate than amplitudes derived from Fourier transformed images since ³image² amplitudes are modulated by the contrast transfer function of the electron microscope. The effects of the transfer function on the calculated amplitudes can only be partially corrected by digital methods.

In any structural study, it is important to consider whether any bias is introduced by the image selection process. Are the chosen images representative of the sample, or, if not, what justifies the selections made? Also, what assumptions, if any, were made that influence the manner of processing or the interpretation of features observed? Have specimen distortions been identified, measured and corrected or have they merely been averaged away by the processing procedures? When possible, several representative specimen images should be analyzed and the results compared before averaging them. If samples are stained, it is important to realize that the most prominent features observed arise from the high contrast produced at the specimen surfaces most accessible to stain. Surfaces accessible to solvent but not to stain, are not revealed visible in the images. Consider, for example, how TMV appears if it is not stained properly (i.e. the sample is not stained for a long enough period or the stain is washed away): stain may not sufficiently penetrate or may not remain in the central (axial) hole, thus rendering the hole invisible.

The manner in which processing results are displayed strongly influences both the visibility and interpretation of specimen features. There are numerous procedures and styles by which different kinds of structural data can be displayed. Two-dimensional projections are often displayed as continuous-tone black and white or pseudo color-coded images or as maps with contour lines depicting the separation between different density levels. Three-dimensional density maps are displayed by solid model, surface-shading techniques, or in serial “sections”, etc. There are countless ways to display the various kinds of data analyzed (including the unprocessed, digitized images and their Fourier transforms, etc., to the reconstructed density maps).

The statement by DeRosier (in Topics in the Study of Life: The BIO Source Book (1971) A. Kramer, ed., Harper & Row, Pub., Inc., N.Y., p. 24) seems an appropriate thought on which to end: