Most structural techniques examine the average properties of a very large number of molecules. For example, a .1 mm cubic crystal of a 10nm particle would contain approximately 1 trillion individual protein molecules aligned in a crystalline lattice. When x-rays diffract from this crystal, the positions and intensities of the resulting spots can be used to determine the structure of an individual protein molecule, often to atomic resolution (although the difficult 'phase problem' still exists). In single particle analysis, the molecules to be studied are prepared in a solution which is then frozen in a very thin (~1um) layer of vitreous ice. This ice is frozen so rapidly it does not crystallize, it remains in a fluid-like conformation at liquid nitrogen or liquid helium temperatures. This preserves the protein in its native fluid conformation. This sample is then placed in a transmission electron microscope where images of the individual molecules in random orientations are collected. These images represent projections through the molecule. The individual molecule images are then processed in a computer to determine the 3D structure of the molecule. A typical reconstruction uses 10,000 - 100,000 particle images. Note that this is an extremely small amount of protein, much less than is required to prepare a crystal. Currently this technique is limited to ~7 angstrom resolution, but microscopes and analysis techniques are rapidly advancing. It is likely that near atomic resolution will eventually be achieved.
This techniques has several advantages over x-ray crystallography: 1) No phase problem. Electron microscopy measures phases directly since images are collected in real-space. 2) In x-ray crystallography, smaller molecules are easier to crystallize and process, in cryo-EM, larger molecules are generally easier. Particles as large as 2000 angstroms have been processed using this technique. X-ray crystallography generally becomes more difficult for particles larger than 100 angstroms or so. 3) Very often a protein can exist in one of several different functional states. Usually it is very difficult to crystallize a protein in each of these states to determine the structural changes involved. However, using single particle analysis, it is often possible to statistically analyze the particles and separate the functional states from the overall population. If this proves difficult, the protein (in solution) can be biochemically driven into a nearly homogenious population of a single functional state. This solution is then frozen and imaged.
The Software
EMAN consists of a set of image processing tools which are specifically
designed to make high resolution single particle reconstructions relatively easy to perform.
It also contains a variety of useful routines for general image processing,
as well as a complete C++ library of routines useful in analyzing electron
micrographs. The core of EMAN is a C++ image processing library. This library is
also wrapped for use from Python (an easy to learn scripting language), so very complicated
tasks can be performed with a simple script. On top of the library and Python
bindings sits a set of command-line programs. These programs perform various
low-level tasks, such as classification, filtration, etc. These programs are in
turn wrapped by a set of high-level command-line programs for performing refinements
and pre/post processing of the images. Finally on top of all of this infrastructure
is a set of graphical programs to make the entire process as user-friendly as possible.
EMAN was designed to automate as much of the single particle reconstruction process as possible. In general, once a set of boxed out particle images have been prepared, the entire reconstruction proceeds with almost no human intervention. When the run is done, a complete set of tools for analyzing the results are provided. An initial, low resolution reconstruction of a new protein can typically be processed overnight on a single processor Athlon class computer. For example, ~2000 particles of Ca++ release channel at 80x80 pixels/particle were processed to produce a ~28 angstrom 3d model in about 2 hours on a single processor Athlon (2 ghz).
IMPORTANT NOTE: EMAN does incorporate features for performing reconstructions of particles with icosahedral symmetry, but it was originally designed with less symmetric particles in mind. The routines for doing icosahedral reconstruction are fairly robust now, but not well covered in the documentation beyond the generic details applicable to all structures. Contact Wen Jiang (wjiang@blake.bcm.tmc.edu) for more information on icosahedral reconstruction in EMAN.
Last Modified: 06/11/03