An Integrated System For Transmission Electron Microscopy

Macromolecular microscopy is becoming an increasingly important tool for structural biology. The development of improved capabilities for three-dimensional electron microscopy is critical for optimal progress in emerging integrative research in molecular cell biology. These techniques currently suffer from several severe disadvantages related to the tremendous time and effort required to acquire and analyze the data.

For several years we have been developing software for automated and intelligent acquisition of transmission electron micrographs^{1,2, 3}. Our overall goal is to develop a system for rapid and routine structure determination of macromolecular assemblies from specimens preserved in vitreous ice. Ultimately we plan to develop an integrated system that can produce an electron density map of a structure within a few hours of inserting a specimen into the electron microscope. With this goal in mind it is essential that the images be acquired using a digital camera rather than film. This requires the development of imaging strategies that account for the characteristics of the CCD, in particular the relationship between the available field of view and attainable resolution. In order to ensure that we maximize the limited field of view on the CCD we must employ feature recognition techniques that place the structure of interest at the center of the digital image during high magnification acquisition.

Methods: We are using specimens of microtubules embedded in vitreous ice and suspended over holey carbon foils supplied by Quantifoil⁴. The automated acquisition system is based around a Philips CM200 TEM and a Gatan MSC CCD camera (1Kx1K) and controlled by the emScope software library¹. The automated system systematically scans the grid starting with the collection of a low magnification image of each grid square. The low magnification image is analyzed to identify target holes that contain ice of suitable thickness and filaments of appropriate density and distribution. An image of each target hole is acquired at an intermediate magnification to assess the quality of the helical filaments within the hole and to define an appropriate high magnification imaging strategy. Procedures for focusing and adjusting astigmatism, as well as ensuring that the specimen is not drifting, are then performed under low dose conditions. Finally a high magnification image is acquired.

Results: We can identify holes containing ice of suitable thickness with almost 100% accuracy. The automated hole identification technique is extremely robust even for grids where the carbon foil has been damaged and the geometrical lattice distorted. We can also correctly identify those holes that contain helical filaments with about 95% accuracy. The system for automated focus and astigmatism correction ⁵ allows us to accurately set focus to within +/-200 nm even when the flatness of the grid requires focus shifts of many microns between target positions. The algorithm works satisfactorily through a layer of vitreous ice and it has not proved necessary to burn the ice off the underlying carbon film in order to set the focus accurately.

The system as described is currently being used to automatically acquire high magnification images to film. We are now concentrating our efforts on developing strategies that will allow us to acquire the final high magnification images to the CCD camera. The restricted field of view available on the CCD imposes stringent requirements for precisely locating suitable targets within each hole so that the target will be centered in the digital image. We have recently developed an algorithm to identify approximately straight segments along each filament identified in the intermediate magnification image. The algorithm uses cross correlation and a variation of the Hough transform to map the length and density along the filament. This information is then used to select a target for high magnification image acquisition. In a typical experimental run we automatically acquire approximately 300 digital images of decorated microtubules at a magnification of 38000x, and of these, over 90% contain acceptable high magnification images of filaments. We are currently assessing the quality of these images against those acquired to film.

Page 2 of 2 -