J Mol Med (2000) 78:191–202 Digital Object Identifier (DOI) 10.1007/s001090000101
Three-dimensional electron cryo-microscopy as a powerful structural tool in molecular medicine
Received: 1 February 2000 / Accepted: 28 March 2000 / Published online: 28 April 2000 © Springer-Verlag 2000
Abstract Electron cryo-microscopy has established itself as a valuable method for the structure determination of protein molecules, protein complexes, and cell organelles. This contribution presents an introduction to the various aspects of three-dimensional electron cryomicroscopy. This includes the need for sample preservation in the microscope vacuum, strategies for minimizing radiation damage, methods of improving the poor signalto-noise ratio in electron micrographs of unstained specimens, and the various methods of three-dimensional image reconstruction from projections. The various specimen types (e.g., flat and tubular two-dimensional crystals, protein filaments, individual protein molecules, and large complexes) require different means of three-dimensional reconstruction, and we review the five major reconstruction techniques (electron crystallography, helitute of Biophysics in Frankfurt. He is currently Postdoctoral Research Associate at the Skirball Institute of Biomolecular Medicine, New York University, and Guest Investigator at the Laboratory of Sensory Neuroscience, Rockefeller University, New York, USA. His research interests include the structure of membrane proteins and large complexes using X-ray crystallography and three-dimensional electron microscopy. His focus lies on the mechanoelectrical transducMANFRED AUER received his Ph.D. at the Euro- tion apparatus (hearing mapean Molecular Biology Labo- chinery) of the inner ear hair cell sterocilia. ratory in Heidelberg, Germany, and the Max Planck InstiM. Auer (✉) Skirball Institute of Biomolecular Medicine, New York University Medical Center, 540 First Avenue, New York, NY 10016, USA e-mail: [email protected]
Tel.: +1-212-2638635, Fax: +1-212-2638951
cal reconstruction, icosahedral reconstruction, singleparticle reconstruction, and electron tomography), with an emphasis on electron crystallography. Several medically relevant three-dimensional protein structures are chosen to illustrate the potential of electron cryo-microscopy and image reconstruction techniques. Among the structural methods, electron cryo-microscopy is the only tool for studying objects that range in size from small proteins over macromolecular complexes to cell organelles or even cells. Key words Electron cryo-microscopy · Three-dimensional electron microscopy · Two-dimensional crystallization · Electron crystallography · Image reconstruction Abbreviations 2-D: Two-dimensional · 3-D: Three-dimensional · EM: Electron microscopy
The role of structural biology in modern medicine Structural biology is having a tremendous impact in modern medicine as we enter the twenty-first century. Together with advances in genetics, genomics, and cell biology, the elucidation of macromolecular structures has revolutionized the field of molecular medicine and lead to specifically designed drugs and therapies. With the foreseeable completion of the Human Genome Project we soon will be able to identify the entirety of proteins responsible for the cell’s function and behavior in various organs. Malfunction of a protein, caused by a single mutation, can lead to the onset of a disease. Structural biology helps to understand the function of such a protein by determining its structure in atomic detail. This structure can then serve as a basis for rational drug design . In larger macromolecular complexes structural biology reveals the molecular interactions between proteins and provides a framework for developing new therapeutic concepts.
Techniques for structural biology Despite their importance in modern cell biology, conventional light microscopy and confocal fluorescence scanning light microscopy cannot be used for the study of individual proteins. These techniques cannot resolve distances smaller than 100–200 nm, which is about 1000 times the length of a typical chemical bond. Atomic force microscopy, although a source of high-resolution information , allows the visualization only of the surfaces and therefore provides limited insight into the internal three-dimensional (3-D) structure of macromolecules. Magnetic resonance imaging is a very powerful diagnostic tool in modern medicine, especially at the level of multicellular assemblies and tissues . However, this technique currently reaches a resolution of only a few micrometers, which is about the size of an eukaryotic cell. Three different approaches have proven successful for reaching beyond the wavelength limits of visible light and into a resolution range in which individual molecules and their internal structure can be visualized: nuclear magnetic resonance spectroscopy (NMR), X-ray crystallography, and high-resolution electron microscopy (EM). Each of these three methods have their strengths and weaknesses and can complement each other.
Nuclear magnetic resonance spectroscopy NMR is particularly well suited for small molecules and for individual domains of proteins. It allows the study of the macromolecule in solution, and, unlike X-ray analysis, NMR avoids the time-consuming search for optimal crystallization conditions. However, large amounts of protein (several milligrams) and a high solubility in water are required to obtain an interpretable signal. In addition, there is a theoretical size limit for the proteins to be studied, which makes the structure elucidation of typical membrane proteins and filamentous proteins unlikely. In principle this size limitation can be overcome by solidstate NMR [4, 5]; however, to date no protein structure determined by solid-state NMR has been reported. Since this method has been developed only recently and has not yet reached its full potential, no definitive judgement can be made at this point whether solid-state NMR can become an alternative to X-ray crystallography and EM.
X-ray crystallography X-ray analysis is currently the predominant source of structural information. It was also the first method for structure determination of macromolecules. By irradiating a 3-D crystal and recording the spatial distribution and the intensities of the diffracted X-ray beams, it allows the investigation of various biological specimens, ranging from peptides to viruses. From such diffraction patterns the structure can be calculated by Fourier methods. Data collection and processing are automated to a
large extent, and therefore it is relatively straightforward to obtain an electron density map to which an atomic model can be fitted. The keys for a successful X-ray analysis are to grow large, well-ordered 3-D crystals and to obtain isomorphous heavy atom derivatives for the initial phasing of the diffraction data. Phase information can also be obtained by multiple wavelength anomalous dispersion experiments or by molecular replacement. The main rationale of any crystallization experiment is to slowly decrease the solubility of the protein by mixing the protein solution with a precipitant solution. If amorphous precipitation and/or denaturation of the protein can be avoided, it will most likely arrange into a periodic array: a 3-D crystal. Finding the suitable conditions can be a labor-intensive process as the conditions for crystal growth cannot be predicted . Integral membrane proteins and filamentous proteins, such as actin and tubulin, often form aggregates rather than well-ordered 3-D crystals. In these notoriously difficult cases the proteins must be prevented from aggregation and stabilized by either adding detergents or by other means, such as proteolytic or genetic alteration of the amino acid sequence. Detergents, necessary for keeping membrane proteins soluble in aqueous solutions, extend the already tedious screening procedure significantly. It comes as no surprise that only a small number of membrane protein structures have been determined by X-ray analysis. Most of these have been referenced in a recent review ; the referenced proteins are: photosynthetic reaction centers, bacteriorhodopsin; respiratory enzymes such as eukaryotic/prokaryotic cytochrome bc1complexes and cytochrome oxidases; a number of bacterial outer membrane porins; aerolysin and α-hemolysin; and a prokaryotic potassium channel. Since this review appeared, several other membrane protein structures have been solved by X-ray crystallography, namely several Escherichia coli outer membrane proteins, a mechanosensitive bacterial channel, two respiratory enzymes and a yeast mitochondrial ATP synthase subcomplex: ● Escherichia coli outer membrane protein A (OmpA)  ● Escherichia coli outer membrane phospholipase A (OMPLA)  ● Escherichia coli outer membrane active transporter FepA  ● Escherichia coli outer membrane ferrichrome-iron receptor FhuA [11, 12] ● Escherichia coli outer membrane protein X (OmpX)  ● Mycobacterium tuberculosis MscL homolog  ● Escherichia coli fumarate reductase  ● Wolinella succinogenes fumarate reductase  ● Yeast mitochondrial ATP synthase subcomplex 
Transmission electron microscopy We owe a large part of our knowledge about cell and tissue organization to EM. Until recently EM was considered a low-resolution tool, mainly suited for morphology and pathology of cells and tissues. Over the past decade it has become evident that EM can also reveal high-resolution structural information on biological macromolecules. Four problems must be overcome to obtain highresolution information: First, the object of interest must be investigated in a vacuum. As water evaporates at the low atmospheric pressure in EM, special care must be taken to avoid sample drying. Second, the object is heavily bombarded with high-energy electrons. Radiation damage must be minimized in order to record intelligible information. Third, since biological objects are radiation sensitive, they can tolerate only a small electron dose. It is the weak scattering of biological samples together with the requirement of low-dose imaging that leads to a poor signal-to-noise ratio. Therefore filtering and averaging techniques need to be applied to enhance the signal. Finally, electron micrographs represent twodimensional (2-D) projections of a 3-D object. Hence, a 3-D structure must be reconstructed from 2-D projections of the molecule in various orientations [18, 19]. Biological sample in a vacuum Since matter interacts with high-energy electrons about 100,000 times more strongly than with X-rays, an electron beam would be stopped in air after a few millimeters. Therefore the electron beam must be kept in a high vacuum, which also means that the sample must be investigated in the vacuum and thus needs to be stable under low pressure. This creates a challenge in investigating biological molecules as the evaporation of water in vacuum leads to major structural alterations. Various ways have been explored to address this problem. Conventional electron microscopy The traditional ways for stabilizing the sample in the microscope vacuum are fixation, staining, dehydration, and embedding of cells in plastic, followed by cutting thin sections. This method is still widely used in cell biology. Other investigators have explored the method of critical point drying followed by metal shadowing, which, again, is a useful approach if one is not interested in highresolution information. Yet another widely used approach is to allow electron-dense material such as uranyl acetate and tungsten phosphate to penetrate buffer-accessible cavities of the molecule. Upon drying, a negative imprint of the protein in the stain is imaged in the EM. The areas that are not penetrated by the stain are usually interpreted as protein. However, the macromolecule of interest can collapse during this drying procedure, and further shrinkage of the specimen occurs upon electron bom-
bardment in the microscope. Other artifacts can arise from the strong scattering properties of negative stain. The data can no longer be interpreted quantitatively as the weak scattering approximation, which forms the heart of bright-field EM, is no longer tenable. In other words, what is seen in the electron micrographs might not be a faithful representation of the macromolecular object . The limited value of stain is evident as the protein’s structure is observed only by indirect means and, in addition, could be obscured by imaging artifacts. Hence for high-resolution work another approach must be taken to preserve the macromolecule in the vacuum of the microscope. Electron cryo-microscopy It was the pioneering work of Taylor and Glaeser  and of Dubochet and colleagues (reviewed in ) that paved the way for electron cryo-microscopy. These were the first researchers to take advantage of the fact that the vapor pressure of water becomes negligible at temperatures below –100°C. Sample drying can thus be avoided by keeping the protein at liquid nitrogen temperature. However, the sample must be frozen, while at the same time avoiding the formation of crystalline ice. Ice crystals can destroy the delicate ultrastructure of proteins. For some proteins small organic molecules such as glucose  and tannin  can substitute for water upon drying of the sample; however, a rapid freezing technique such as plunge-freezing  is preferable as the molecules are frozen in their native environment. Vitrification of the sample using the plunge-freezing method works particularly well for individual proteins and multiprotein complexes whereas for whole cells and tissues vitrification can be achieved only by physical fixation such as in slam-freezing or high-pressure freezing [25, 26]. The latter technique allows a vitrification depth of about 300 µm. Vitrification can then be followed by either cryo-sectioning  or freeze-substitution . Cutting sections of the vitrified tissue at liquid nitrogen temperatures (cryo-sectioning) avoids artifacts that arise from a chemical treatment of the sample. However, cryosectioning is technically very challenging and is still in its early days. More pioneering work is needed to add this method to the standard repertoire of electron cryomicroscopists. An alternative to cryo-sectioning that works surprisingly well is freeze-substitution, in which the physically fixed sample is stained, dehydrated, and embedded at temperatures well below the freezing point of water. After UV light induced hardening of the plastic, thin sections for transmission EM are cut at room temperature, such as in conventional sectioning. Freezesubstitution is much simpler than cryo-sectioning; however, the limitations of stain and potential artifacts introduced by the chemical treatment should be kept in mind. Although the various cryo-techniques are as yet being used by only a small number of investigators, electron cryo-microscopy has grown out of its infancy and is no
longer considered an exotic method for structural analysis. As the advantages of unstained specimen are obvious, we now address the problems that one encounters. Radiation damage While most of the accelerated electrons pass the sample without interaction with the potential field of the atoms, a fraction of the incident electron beam is deflected from the original trajectory. Upon focusing of the elastically scattered electrons by the electromagnetic lens system, a magnified image of the object is obtained by the socalled phase contrast. Phase contrast forms the heart of quantitative bright-field EM and means that contrast is achieved by interference of the elastically scattered electrons with the nonscattered electrons. Ideally the scattering event is elastic, which means that the kinetic energy of the electrons remains unchanged and no electrons are absorbed. In reality, however, there are also inelastic scattering events, which means that part of the kinetic energy of the electron is transferred to chemical bonds of the macromolecule. This process creates radicals and ultimately destroys the internal structure in the macromolecule. In other words, the elastically scattered fraction of the electron beam contains the information about the scattering object (the protein), whereas the inelastically scattered electrons are responsible for the radiation damage. Local radiation damage (ionization events) cannot be avoided, but its effect, namely the diffusive damage of free radicals, can be minimized by trapping the radicals and keeping them from spreading into neighboring areas where they cause further destruction. Investigating the sample at liquid nitrogen temperatures, necessary to prevent the biological sample from drying in the microscope vacuum, proves to be fortunate as diffusion of radicals is marginal at such low temperature. It should be said that radiation damage is not unique to unstained specimens and also occurs in negatively stained samples. Upon irradiation, the negatively stained sample experiences substantial shrinkage in the direction of the electron beam. In addition, the protein molecules suffer from their interaction with the electron beam. However, this usually goes unnoticed, as in negative stain imaging one looks mainly at the stain distribution, in other words, at the negative imprint of the protein. Radiation damage is not restricted to EM but also occurs in X-ray crystallography. Radiation damage proves even more severe in X-ray crystallography, as for X-rays the ratio of elastic to inelastic scattering events is several hundred times worse than for electrons interacting with matter . 3-D crystals contain about 108–1010 times more molecules than 2-D crystals. Therefore the effects of radiation damage are less obvious. Usually cryo-preservation of the sample in transmission EM is combined with the so-called low-dose imaging technique. In this mode, focusing at a magnification of approx. 200,000 and optimization of the microscope’s performance is carried out in an area adjacent to the area that subsequently
will be photographed at a magnification between 10,000 and 60,000. Therefore the area of interest does not experience unnecessary irradiation and is exposed to an electron dose of only 10–20e–A2, which is low enough to be tolerated by the biological macromolecules. Therefore the negatives are underexposed and undergo an enhanced photographic development procedure. Using this lowdose imaging technique minimizes damage to the protein and also lowers the signal-to-noise ratio. The presence of substantial noise in data is a consequence of the inherent low contrast of unstained specimen and the small number of elastically scattered electrons. Images recorded under low-dose conditions are usually subjected to an extensive computational image-processing procedure, which includes image restoration and image enhancement. The former addresses the effect of the instrument on the image, for example, contrast-transfer-function and aberrations, whereas the latter aims at improving the signal-to-noise ratio, for example, by applying a number of computational techniques such as thresholding, contouring, averaging, and filtering. Improving the signal-to noise ratio Since radiation sensitivity of the biological samples requires that images be recorded with a low electron dose, the resulting micrographs show a poor signal-to-noise ratio. The signal can be enhanced by averaging identical particles that are in identical orientations. The averaging is particularly easy for particles in periodical arrangements, and this is why crystalline arrays are particularly well suited for structure determination at high resolution. In the case of individual particles, alignment of similar projections is relatively straightforward for improving the signal-to-noise ratio. However, 3-D reconstruction requires the alignment of different projections and is usually more difficult. The number of particles that can be averaged, their size and homogeneity, and the accuracy of alignment decide how many details can be resolved in a density map. In theory, alignment of only approx. 10,000 macromolecular particles should be sufficient to reach 3 Å resolution , at this point, however, a much larger number of particles (up to several millions) are needed to obtain atomic resolution. For very large complexes, for example, cell organelles, the structure is unlikely to be preserved identically between two given objects. Therefore direct averaging techniques cannot be applied. In such cases 3-D reconstructions must be obtained from one specimen, which inevitably leads to a lower resolution. At a resolution of 3–4 Å, densities of individual amino acid side chains can be discriminated, and the protein sequence can be fitted to the density to obtain an atomic model. A density map at 5 Å allows β-strands to be detected, and at 8–9 Å α-helices become obvious. This resolution range, which reveals the arrangement of secondary structure elements, is often referred to as intermediate resolution. At 20–30 Å resolution, the so-called low
resolution, individual proteins can be detected in a multiprotein complex, and the shape of the molecules, including protrusions and indentations, can be determined. 3-D reconstruction from 2-D projections Transmission EM projects a 3-D object along the axis of the EM lenses onto the plane of the recording medium. Due to the large focal depth of transmission EM, every part of the 3-D object, whether in or out of focus, contributes to the recorded image. Hence an image is a superposition of all atoms of the 3-D object along the electron beam trajectory. Obviously one needs to recover the spatial information in the direction of the electron beam in order to retrieve the 3-D structure. This can be done by imaging the molecules from various angles, not unlike in a computed axial tomography, a procedure well known to the medically interested reader. In EM, however, the specimen rather than the imaging device, is tilted. Projections of certain types of specimens, such as those with helical or icosahedral symmetry and randomly oriented single particles, contain all the views of the macromolecule that are needed for the 3-D reconstruction. All other specimens, for example 2-D crystals, require specimen tilting to obtain the different views of the molecules. Specimen tilting, however, often introduces technical problems and results in a poor yield of good images. The effective thickness of the specimen increases upon tilting, which sets a limit for the maximum tilt angle, as electrons must be able to penetrate the object. Moreover, the specimen is supported with a metal grid, whose grid bars may get into the way of the electrons. For these two reasons the specimen is usually tilted up only to about 70°, and as a consequence certain views are missing and do not contribute to the 3-D reconstruction. Thus the resolution is anisotropic, meaning that the resolution in the direction of the incident beam is lower than that in the specimen plane. Five techniques have been developed to deal with the 3-D reconstruction of a variety of specimens: electron tomography , single-particle analysis , icosahedral reconstruction , helical reconstruction , and electron crystallography : Electron tomography is particularly well suited for very large particles such as cell organelles and large multiprotein complexes. If the exact architecture is unlikely to occur many times, such as in the case of cell organelles, electron tomography is the optimal method for structural analysis. Here images of the same specimen are recorded at tilt angle increments of about 1°. Since the tolerable electron dose must be divided over some 120–140 images , and since averaging techniques cannot be applied, the resolution of such tomograms is fairly low (20–70 Å). However, in comparison with the usually large dimensions of such objects, the information content of tomograms can be enormous. The specimen can be either large individual particles  or tissue sections . The maximal thickness of the sections that
can be tolerated is dependent on the acceleration voltage and can be as much as 300 nm. Using fast-frozen, freezesubstituted tissue sections in combination with electron tomography allows one to investigate cells in three dimensions and in fine detail. Therefore this method might be especially appealing to the molecular medicine oriented community [38, 39]. Single-particle analysis deals with large proteins or protein complexes (usually exhibiting a molecular weight higher than 500 kDa) that can be obtained in sufficient quantities and homogeneity. Upon classification and alignment of the particles, a 3-D structure can be calculated. Averaging of approx. 73,000 particles yielded a resolution of 11.5 Å for the 70S ribosome , whereas a resolution of 7.5 Å has been claimed for a reconstruction of the ribosome 50S subunit, with only approx. 16,000 particles included . This paradox may reflect the fact that there are at least two ways for calculating the resolution of a single particle reconstruction. However, it is undisputed that both the accuracy of alignment and the homogeneity of a large number of particles are critical for reaching atomic resolution . Icosahedral reconstruction is a special case of singleparticle analysis. It successfully exploits the internal icosahedral symmetry of many viruses. This internal symmetry is caused by the small size of a viral genome and the need to build a protein capsid from only a few gene products. 3-D reconstructions of viral capsids have been obtained at an impressively high resolution (7–9 Å), allowing the arrangement of secondary structure elements to be discriminated [43, 44]. Helical reconstruction methods rely upon the helical symmetry of filaments and upon tubular crystals as in the case of the acetylcholine receptor or of the sarcoplasmic Ca-ATPase pump. Unlike with flat 2-D crystals, tilting of the specimen is not necessary as all views are present in a single micrograph. Isotropic intermediate resolutions (4.6–11 Å) have been obtained [45, 46, 47], and the structure of the nicotinic acetylcholine receptor determined at high resolution by this reconstruction method is expected soon. In terms of resolution, the most powerful EM technique to date is electron crystallography of flat 2-D crystals, in which a large number of protein molecules can be averaged in a convenient way, resulting in a high signalto-noise ratio.
Electron crystallography: electron microscopy of 2-D crystals Of all current EM approaches only electron crystallography has the potential to provide atomic coordinates. Among the milestones rank the atomic 3-D structures of bacteriorhodopsin [48, 49], light-harvesting complex , and more recently tubulin , all derived from flat sheetlike 2-D crystals. The studies of flat 2-D crystals of bacteriorhodopsin and of tubular crystals of the acetylcholine receptor are particularly noteworthy because 3-D
structures of reaction cycle intermediates have been revealed by electron cryo-microscopy [52, 53]. Over recent years a number of proteins have been crystallized in 2-D and studied at various levels of resolution. Most of these proteins are abundant and have been easily purified in large quantities. However, the natural abundance of most medically relevant membrane proteins, such as channels, transporters, and G protein coupled receptors, is too low to purify the proteins from their native source. These must be expressed in large quantities either in a bacterial or eukaryotic expression system. This can be a real challenge for some of the most wanted membrane proteins . Overexpression and purification is followed by attempts to grow 2-D crystals. In some fortunate cases 2-D crystals occur naturally in the membrane. This has been seen to be the case with bacteriorhodopsin  and gap junction . In other cases small manipulations of membrane patches containing protein in a dense packing have led to well-ordered 2-D crystals, for example, the acetylcholine receptor  and the sarcoplasmic Ca-ATPase . More often, however, membrane proteins first need to be solubilized in detergents and then purified to homogeneity. They can then be reconstituted into lipid bilayers by removal of the detergent, for example, by dialysis [58, 59] or by adsorption to polystyrene beads . Alternatively, 2-D crystals from detergent-solubilized proteins can be induced by addition of precipitants analogous to a crystallization experiment for 3-D crystals. Such crystals form at either the air-water interface  or more conveniently on the surface of a carbon-coated support grid . After freezing, 2-D crystals are then imaged at liquid nitrogen temperatures, and projections recorded at various tilt angles are processed using well-established image processing protocols . The data are processed and combined most conveniently using Fourier methods, and includes filtering of the data to reduce the noise component, correction for crystal faults (so-called “unbending”), correction for defocus and image distortions caused by the imperfect lens properties of EM. Given that the 2-D crystals are well ordered, a 3-D structure can easily be achieved at 6 Å. To reach atomic resolution, special care must be taken to ensure specimen flatness and to overcome various technical problems such as beam-induced specimen movement and charging. In addition, phase data from images must be combined with amplitude data from electron diffraction to achieve atomic resolution. However, this requires 2-D crystals of a size of at least 1–2 µm, a requirement that is not necessary met by all proteins.
Medically relevant examples of proteins solved by electron cryo-microscopy The following 3-D structures have been selected for their medical relevance, either for their role in physiology or as therapeutic targets, namely hepatitis B virus capsid, ribosome, tubulin, P-type ATPases, rhodopsin, connexin,
aquaporin, nicotinic acetylcholine receptor, and actinmyosin complexes (see Table 1). These structures were determined at various resolutions, ranging from highresolution (