INTRODUCTION

Top Previous Next References

EM (abbreviations are listed at the end of this section) has long been a primary tool for classifying viruses and exploring their structures. The last decade has also seen a burst of activity in the use of EM for the elucidation of virus structures. This has resulted from two advances in technique. Firstly, cryo-EM has allowed the preservation of fragile specimens in the EM (1, 109). Secondly, the development of efficient algorithms for processing micrographs to produce 3D structures of icosahedral particles has allowed this higher-quality data to be used (9, 14, 42, 80, 86, 89, 91, 93, 130, 133, 210). These two developments have made 3D structural information accessible for a broad range of viruses at the same time that high-resolution X-ray diffraction studies have revealed atomic detail about a more limited range. These two approaches are complementary and together are bringing a new excitement to the field of virus structure.

The most widely used approach for the reconstruction of icosahedral structures begins with the method of common lines, which was developed by Crowther in the early 1970s (86, 89). The original application of the method relied upon estimating the orientation of each particle by eye and then using the symmetry relationships which are present in the projection of any icosahedral object to refine these orientations. The data from the resulting set of views are then combined to produce a 3D reconstruction by means of the projection theorem (43). Several 3D structures of negatively stained viruses were solved, and these results helped clarify the principles of quasi-equivalence that were being explored at the time (88, 89, 94, 95, 119, 120, 170, 216). The use of data from negatively stained specimens limited the efficacy of the method. The distortions of the structure caused by drying, flattening, nonuniform staining, and radiation damage resulted in a loss of the icosahedral symmetry upon which the reconstruction method depended. The applicability of the method was further constrained because many interesting structures such as enveloped viruses were destroyed by interaction with the stain. Finally, even ideal conditions of negative staining revealed only the distribution of the heavy metal stain embedding the specimen rather than the density of the specimen itself. The development of cryo-EM changed this situation (1). By maintaining a layer of vitrified water around the specimen, relying on defocus rather than heavy metal stains to generate contrast, and performing microscopy under low-dose conditions at near-liquid-nitrogen temperatures, this method was able to produce data of unprecedented quality. In particular, the distortions and artifacts which had limited the use of 3D reconstruction of icosahedral particles previously were eliminated in data collected by cryo-EM. The limitation then became the processing of the data. Relatively few characteristic views were recognizable by eye because a cryo-electron micrograph shows the entire density of the particle in projection. This was not the case for negatively stained samples, where uneven staining and the fact that only the outline of the specimen is observed resulted in a simpler image. Cryo-electron micrographs revealed higher-resolution data, but they did so with relatively low contrast (1). This further complicated the task of recognizing and refining views. A reformulation of the common lines method was necessary to allow this to be done reliably and automatically for these noisy, low-contrast, and complex images. The past decade has seen the successful development of such methods (9, 14, 42, 80, 93, 130, 133, 210), a resulting flowering of the use of the approach, and an explosion in the number and quality of the results (see Table 1).

The value of the results produced by 3D reconstruction of viruses from cryo-EM must be considered in terms of its contribution to our understanding of viral structural biology. The morphology of the virus is only the first result gleaned from viewing a 3D reconstruction (Fig. 1). For a highly symmetric particle such as a virus, morphology reveals not only the shape of the virion but also the organization and shape of the components. As discussed below, the combination of an understanding of icosahedral symmetry with even a low-resolution density map can often allow one to infer the stoichiometry of the polypeptides comprising the virion. At a somewhat higher resolution (20 to 30 Å), one can visualize systematic changes in subunit conformation that allow the formation of the capsid. This may reveal fundamental relationships among the structures of members of a viral family such as the papovaviruses (14, 15, 19) or unexpected relationships such as that between a bacteriophage and the reoviruses (48). Interactions between the components of the virus which provide a glimpse into the process of virus assembly (e.g., see references 48, 101, 108, 130, 131, 245, 249, and 353) may also be revealed and generate ideas which can be tested by observations of intermediates in the assembly process (48, 100, 166, 213, 245, 294, 311). Together, the information provided by these methods is generating a deeper understanding of capsid assembly (294).

View larger version (131128K): [in this window] [in a new window]   FIG. 1.   Gallery of representative icosahedral viruses studied by us using cryo-EM and 3D image reconstruction methods. The monomer of bacteriorhodopsin, a 26-kDa membrane protein which contains seven  helices oriented perpendicular to the membrane plane, is shown for comparison at the lower right of the right-hand page (extracellular surface faces upward). All shaded-surface virus structures are viewed along a twofold axis of symmetry. Table 1 presents more-detailed information about these and other 3D reconstructions of icosahedral viruses. TBE, tick-borne encephalitis recombinant subviral particle; Nv, Nudaurelia capensis  virus; Nv, Nudaurelia capensis  virus; Ty Retro, yeast retrotransposon Ty1 VLP; SpV4, Spiroplasma virus type 4; DHBc, duck hepatitis B capsid; B19, human parvovirus B19.

An important tool that provides a link between the reconstruction and the biochemistry and that facilitates its interpretation is the use of specific labels. Antibodies are an important example of such labels that have been used to identify components in a reconstruction (e.g., see references 242, 286, and 313). Localization of antibody binding sites also gives important functional information concerning the mechanism of neutralization (285, 287-289, 297) and receptor binding (242, 286). The conclusions reached in these studies can be supported by difference imaging, which takes advantage of the tools of expression and reconstitution to create particles of defined composition (312, 362, 365). The resolution of antibody labeling techniques is higher than expected from the resolution of the reconstruction because the complex can be modeled by fitting in the known structure of an Fab (73, 156, 161, 162, 235, 240, 243, 288, 289, 297, 336, 337). The promise of higher-resolution localization is realized in recent papers in which a combination of site-directed mutagenesis and undecagold labeling has been used to localize a particular residue with high precision in HepBc (364) and in which antibody Fab labeling has been used to localize the six-residue putative immunodominant loop of HepBc (78). Finally, the use of peptide-based difference mapping has been successfully exploited in the hepatitis virus system to locate the N terminus of HepBc (79) and to locate the binding site of peptides that block hepatitis virion assembly (41).

An increasingly important approach in structural biology is that of "divide and conquer" (16). Complex biological systems which are intractable when studied by a single technique yield to a combined approach. The structures of a number of isolated viral capsid proteins have been solved to high resolution by X-ray diffraction (e.g., see references 74, 254, 255, 324, and 347). Image reconstruction from electron micrographs cannot equal the resolution attained by X-ray diffraction (yet see references 42 and 80), but it provides a context for diffraction results by being able to orient the atomic structure of the subunit into the complex 3D structure of the virion (64, 145, 162, 166, 233, 261). Finally, cryo-EM followed by image reconstruction can be performed with heterogeneous populations of particles. This allows one to explore biochemical treatments which have known effects on virus composition or infectivity but which do not cause a transition of the entire population of particles to a defined state (132, 325). The ability to accommodate flexibility makes cryo-EM and 3D image reconstruction (cryoreconstruction) an ideal tool for exploring changes in virus structure.

This review focuses on the results of 3D reconstruction of icosahedrally symmetric viruses from cryo-electron micrographs. We begin with a brief introduction to the standard terminology so that the reader will share our language for describing the icosahedral structures presented. We then discuss the preparation and microscopy of vitrified specimens and describe 3D reconstruction by the common lines technique as it has been recently implemented for viral structures. More-detailed discussions of the processing methods have been presented elsewhere (9, 133, 210). We then address the question of interpretation of the reconstruction, describing the tools and pointing out some of the pitfalls to be avoided. We continue with a description of some of the portions of virus life cycles illustrated by these results. This section owes much to our colleagues who have very generously allowed us to present their work here. Their willingness to share results and ideas has made the development of this field rapid and enjoyable. We conclude with a comprehensive list of relevant details and literature citations for the bulk of image reconstruction work on icosahedral viruses and a brief discussion of the prospects for extending the technique to higher resolution and the examination of more-complex viral and nonviral systems.

Throughout this review, icosahedral structures are described in language which assumes their regularity and symmetry. This description is based on the theory of quasi-equivalence introduced by Caspar and Klug (53). There are several excellent reviews of icosahedral organization and its expression at the atomic level as determined by X-ray diffraction techniques (148, 149, 176, 202, 260). We will not enter into such a detailed discussion of these results but rather confine ourselves to their description and relevance for the interpretation of 3D reconstructions.

The basic problem addressed by Caspar and Klug (53) in their elegant paper was the understanding of virus structure as the consequence of the self-directed interaction of a large number of chemically identical units. Self-assembly requires specificity of bonding between the units of the structure. One way to accomplish this is to allow all subunits to form identical bonds with their neighbors. A crystal of water or sucrose is formed in this way, as are protein assemblies such as the octahedral assembly of pyruvate dehydrogenase (98). Since a virus structure is optimized for the propagation of its genome, it is advantageous to make a large shell with a small amount of information devoted to structural components. One can imagine that a virus with octahedral symmetry could enclose its genome only if it were formed from very large subunits. This would require that it devote a large fraction of its genome to coding for these structural proteins. Forming a shell from a larger number of small, identically bonded subunits would provide a more parsimonious solution both to the problem of encapsidating a large amount of genetic information and to that of self-assembly.

An icosahedron (Fig. 2, left) is an isometric structure with 12 pentagonal vertices and 20 triangular faces. Any icosahedron has a defined set of exact symmetry elements: 6 fivefold axes through the 12 vertices, 10 threefold axes through the 20 triangular faces, and 15 twofold axes through the edges. The related structure, the pentagonal dodecahedron (Fig. 2, right), shares the same symmetry elements but has a complementary morphology: 6 fivefold axes through the pentagonal faces, 10 threefold axes through the vertices, and 15 twofold axes through the edges. Both of these ideal geometric structures have icosahedral symmetry, as do spherical viruses, whose shapes lie between these two extremes.

View larger version (65K): [in this window] [in a new window]   FIG. 2.   An icosahedron (left) and dodecahedron (right) with symmetry axes and the asymmetric unit used by microscopists. The numbers (2, 3, and 5) indicate the positions of some of the symmetry axes. The white triangle defines the asymmetric unit which is bounded by the lines joining adjacent fivefold and threefold positions.

The positions of the symmetry elements are the landmarks used to describe any icosahedral structure. Application of the icosahedral symmetry elements to a subunit which does not lie on a symmetry axis causes it to be repeated 60 times in the complete structure. This means that the complete structure can be generated by taking 1/60th, called the asymmetric unit, and operating on it with the symmetry elements. The choice of the unit cell is arbitrary. In icosahedral reconstruction, the asymmetric unit is usually defined as being the wedge-shaped volume which extends from the icosahedron's center along edges formed by a threefold axis and two adjacent fivefold axes (14, 130, 133). In most high-resolution X-ray structures of viruses, the conventional asymmetric unit is bounded by a fivefold axis and two adjacent threefold axes (350). Another way of understanding the effect of the symmetry elements which is particularly appropriate to image reconstruction is to realize that a single, nonaxial view of an icosahedral object is equivalent to 59 other views generated by the symmetry axes.

The simplest icosahedral structure is one in which 60 identical subunits interact identically. Caspar and Klug (53) pointed out that, when more than 60 subunits interact to form a closed shell, as appeared to be true for many spherical viruses, all subunits cannot have identical environments. This lack of equivalence between the subunits reopened the question of self-assembly of the subunits. Quasi-equivalence is a solution to this problem (53). Their hypothesis was that shells with more than 60 subunits would be formed from chemically identical subunits with slight but regular changes in the bonding. Hence, a virus comprising 180 subunits would contain three types of bonding and three distinct environments for the subunits. The three types of subunits would no longer be equivalent but rather quasi-equivalent because their environments were similar but not identical.

Triangulation number is a geometric and abstract concept that does not necessarily correspond to the structural components of an individual virus. It refers to the organization of the geometric figure (Fig. 2). One can visualize the formation of an isometric shell by beginning with a flat, hexagonal net. In the original net, all internal bonds are identical in environment. To curve the net and generate a closed structure, one converts some of the hexagons to pentagons. This concept of quasiequivalence can be naturally expressed by the triangulation number for an ideal icosahedron. This can be seen from a consideration of the building of larger icosahedra from a hexagonal net (Fig. 3). The simplest icosahedron has only equivalent units: all bonds are identical and all vertices have identical bonding. The larger icosahedra have hexagons between the pentagons and can be visualized as replacing the original triangular faces with larger ones formed from equilateral triangles. The number of triangles replacing the original one is the triangulation number. More precisely, the triangulation number is given by the following relationship, T = h² + hk + k², where h and k are positive integers which define the position of the fivefold vertex on the original hexagonal net. The different triangulation numbers have very different organizations of bonding (Fig. 4).

View larger version (45K): [in this window] [in a new window]   FIG. 3.   Geometric principles of constructing icosahedral lattices of defined T (triangulation) number. (A) An array of hexamers, represented as a flat sheet of hexagons, is the basis for generating icosahedra (178). A closed icosahedral shell that conforms to the principles of quasi-equivalent symmetry contains 60T subunits organized as hexamer and pentamer units (53). Hexamers are initially considered planar (hexagons in the flat sheet), and pentamers are considered convex and introduce curvature in the sheet of hexamers when they are inserted in place of specific hexamers. A closed shell is generated by inserting 12 pentamers at appropriate positions in the hexamer net as specified by (h,k) lattice points that mark the centers of the original hexagons in the sheet. The model of a particular quasi-equivalent lattice is constructed as follows. Generate one face of an icosahedron by defining an equilateral triangle in the net. The first side of the triangle is a line connecting the origin point of the net (h,k = 0,0) to any (h,k) point. This process will lead to a lattice of T number given by the relation T=h2 + hk + k2. The remaining two sides of the triangle are formed by connecting the (h,k) point to the appropriate points needed to form an equilateral triangle, as illustrated for the nonenantiomeric T=3 (B) and T=4 (C) lattices and the enantiomeric T=7l,d (D) and T=13l,d (E) lattices. A planar sheet of 20 such triangles is formed, and the sheet is then folded up to form a closed icosahedron as depicted for T=3 and T=4 lattices (B and C). In this way, each of the hexamers at the (h,k) lattice points that define the corners of the triangles are replaced by pentamers and each triangular face contains 3T subunits. (F) Examples of viruses with different T-lattice symmetries. Not all viruses conform to the simple rules of quasi-symmetry as stated above. For example, all T=7 papovaviruses such as polyomavirus have capsids built of 360 subunits arranged as 72 pentameric capsomers (29, 252, 293). A further useful relation is that the number of capsomers is given by 10T+2. Adapted from reference 178 with the permission of the author and the publisher.

View larger version (36K): [in this window] [in a new window]   FIG. 4.   The environments of subunits in three different triangulation numbers shown by an arrangement of the heads of the members of one of our groups. Only a single environment is required in the T=1 arrangement, while three and four environments are present in T=3 and T=4, respectively. Notice that a larger head (subunit) is necessary to fill the same-sized asymmetric unit for the lower triangulation numbers. The positions of the icosahedral twofold (2), fivefold (5), and threefold (3) and the quasi-sixfold (6) axes are indicated.

The remnants of the hexagons, now rendered nonequivalent by their construction into an icosahedron, remain hexavalent and are seen as the positions of local or quasi-sixfold axes in the larger icosahedra. In general, the nonfivefold lattice points of the icosahedral net must correspond to the positions of hexavalent structural units. These positions often correspond to the positions of capsomers, which are apparent structural units on the surface of some viruses.

The number of different environments occupied by a subunit increases as the triangulation number increases. The original theory of quasi-equivalence stipulated that the number of different environments should equal the triangulation number; therefore, a T=4 virus would have four different subunit environments and 60T or 240 subunits (120, 130, 230). The correspondence between T number and the number of different subunit environments is not always strictly maintained. In agreement with the theory of quasi-equivalence, a T=1 (h=1,k=0) icosahedral virus such as satellite tobacco necrosis virus has only one type of environment for its subunit and a total of 60 subunits (203). Similarly, the T=3 (h=1,k=1) tomato bushy stunt virus has three separate environments for a total of 180 subunits (150). In contrast to this, the T=7 papovaviruses are composed of 72 pentamers and have six subunit environments and yet only 360 subunits (h=1,k=2 for the T=7dextro or right-handed organization as in SV40) (4, 201). The same arrangement is seen in three papillomaviruses (29, 188). However, not all T=7 viruses violate quasi-equivalence, since bacteriophage P2 (101) and CaMV (65) also display triangulation number T=7 but have 12 pentameric and 60 hexameric capsomers for a total of 420 subunits with seven distinct environments. When the correspondence is direct, determination of the triangulation number provides a count of the number of subunits; otherwise, it remains a descriptive tool. The nominal "T=2" arrangement of the fungal virus capsids (63) and the intermediates of the dsRNA viruses 6 and BTV (48, 144) can be viewed as a T=1 arrangement of pairs of chemically identical subunits.

Cryo-EM in combination with image reconstruction provides a direct, objective way to determine triangulation numbers of spherical viruses and also allows direct determination of the number of subunits in the virion when individual units are resolved. For example, the trimeric nature of the capsomers that lie on the T=25 net of adenovirus is apparent in the reconstruction (296), and the number of subunits and their environments are clearly identified. Reconstructions also provide information concerning the alteration of subunit conformations and their interactions in the structure. In adenovirus, these changes appear to occur at the sites of interaction with the minor structural proteins of the virus (135, 296, 298) and so provide a view of the quasi-equivalence in practice.

Classically, the projected images of negatively stained viruses have been used for the determination of triangulation numbers (53). Two approaches have been employed. The first is the inference of capsomer positions and their placement on an icosahedral net, often by comparison with a model of the triangulated structure (121, 188, 189). This method is not only technically difficult but fraught with difficulties due to the vagaries of negative staining and the mental gymnastics associated with inferring a 3D structure from two-dimensional views of particles of unknown relative orientation. It has, however, produced the correct triangulation number when applied carefully to structures that respond well to negative staining (188). The second approach is to combine an estimate of the subunit size with the radius of the virus to infer the number of subunits and a triangulation number (i.e., 60 [T=1], 180 [T=3], 240 [T=4], ... ). This approach is very straightforward, but it often produces the wrong answer. Microscopists who are tempted to utilize this second method should realize that their conclusions may be challenged by the more objective 3D reconstruction methods described here. Naturally, either method depends upon but does not validate the assumption of icosahedral symmetry (134).

The methods of preparing viruses and many other types of biological macromolecules for cryo-EM studies are well established (1, 71, 109, 218). A vitrified specimen of a spherical virus is typically prepared in the following way (Fig. 5). A 2- to 5-µl aliquot of a purified, 0.05- to 5-mg/ml suspension of virus is applied to an EM grid coated with a holey carbon support film. The grid is secured with a pair of forceps in a guillotine-like device and suspended over a bath of cryogen slush (usually ethane or propane) maintained near its freezing point by a reservoir of liquid nitrogen. The grid is then blotted nearly dry by pressing a piece of filter paper directly against it, the guillotine is released, and the grid is rapidly plunged into the cryogen.

View larger version (56K): [in this window] [in a new window]   FIG. 5.   The steps in a typical preparation of a specimen for cryo-EM are shown. A holey carbon film (A) is prepared by the evaporation of carbon onto a grid bearing a holey plastic film and the removal of the plastic by exposure of the grid to the fumes of ethyl acetate. This film contains holes with diameters between 1 and 5 µm. The specimen is applied to the film at concentrations between 50 µg/ml and 5 mg/ml (B). The grid may then be floated on a drop of water or low-ionic-strength buffer to remove excess salt. The grid is then placed in a pair of forceps which are locked into a guillotine-like device (C) and blotted with filter paper to produce a very thin aqueous film (1,000 to 2,000 Å) across the holes of the grid (D). Immediately after blotting (and before the aqueous film has had time to dry), the plunger is released to allow the forceps to drop into a bath of ethane slush held in a container of liquid nitrogen. The efficient cooling afforded by the ethane slush causes vitrification of the sample. It is then either stored under liquid nitrogen or placed in a liquid-nitrogen-cooled specimen holder for viewing in the microscope.

Vitrification of the water rather than crystallization (i.e., ice formation) occurs if the sample is thin enough (~0.2 µm or less), so that cooling occurs very rapidly. Estimates vary, but there is agreement that the rate of cooling exceeds 10⁴ °C/s (see discussion in reference 31). An efficient cryogen such as propane or ethane slush rather than liquid nitrogen must be used because the Leidenfrost effect (the creation of gas upon contact with the specimen) would otherwise slow heat transfer so that crystalline ice would be formed (110).

The concentration of the sample is critical and is usually higher than that needed for negative staining. Application of the sample to a continuous carbon film followed by a brief period of absorption (1 to 2 s) before negative staining can be used to determine the appropriate range of concentration. Prolonged absorption before staining will concentrate the sample on the grid so that it appears more concentrated than it will by cryo-EM.

The thickness of the water layer primarily depends on blotting time, wetting properties of the support, and the humidity near the sample. Excessive drying rapidly thins the specimen, often causes particles to migrate toward the periphery of the holes in the substrate, and can lead to drastically increased solute concentrations as well as dehydration, all of which alters the specimen's environment and perhaps structure (211, 212, 330). Alternatively, inadequate drying leaves a sample in which particles are superimposed or embedded in a water layer that is too thick for the electron beam to penetrate. Some workers prepare their samples in cold rooms (174) or in 37°C rooms at 100% humidity (211, 212) to avoid the effects of drying on sensitive specimens. However, the use of humidity-controlled glove boxes (174), double-blotting techniques (310, 334), or a simple temperature-controlled mist device (75) may preserve the state of the specimen with less stress on the investigator.

All subsequent steps, including the recording of images in the microscope, are carried out with the sample below 160°C to avoid the devitrification which occurs at ~140°C (113, 198, 200). These include transfer of the grid from the cryogen into liquid nitrogen (where it may be stored indefinitely) and then into a cooled cryospecimen holder which is rapidly (and carefully) inserted into the EM. All of the transfer steps must be performed rapidly to avoid warming of the specimen and contamination by the condensation of water vapor. Such condensation can be minimized by continuously bathing the cold sample in dry nitrogen gas during transfer to the microscope vacuum, which is an option on some cryotransfer systems. Excess water vapor that gets into the microscope severely overloads the high-vacuum system and reduces its useful lifetime. This has become increasingly critical with the use of sensitive FEG sources. The beginner should become comfortable with all steps of transfer with cryoholder and specimen at room temperature before preparing a vitrified specimen.

As with other techniques in microscopy, various alternative procedures have been developed for each of the steps described above. Usually the specimen is prepared on a holey carbon film, which sometimes is glow discharged to enhance the spreading of the specimen. Alternatively, continuous carbon films, carbon-coated plastic films, or even bare grids (1) have been successfully used as supports for different viruses. The age of the carbon film often determines the quality of the specimen spreading: old support films can be recarboned to produce usable grids (335a). Sometimes, specimens prepared on holey films do not appear in the holes because they strongly adhere to the support film. If this occurs, the properties of the support must be altered. If hydrophobic support films are required, grids may be glow discharged in an atmosphere of amyl amine (112). However, support films that are too hydrophobic usually exhibit regions of thin "ice" surrounded by very thick "ice." An alternative is to apply a dilute solution of lipids (0.1% [wt/vol] phosphatidylcholine or phosphatidylethanolamine in H₂O) to the grid before the specimen. This results in a vitrified layer of water sandwiched between lipid monolayers (325). A number of different techniques and apparatuses that are capable of producing uniformly thin, vitrified samples have been developed. These procedures include the double blot (310), carbon sandwich (302, 303), jet spray (111, 114), pressurized liquid nitrogen (109), and many others.

Another concern of the microscopist is that many aqueous samples of biological specimens contain high concentrations of buffer salts (>100 mM) or solutes such as glycerol (5 to 10%). Their presence can lead to phase separation upon cooling and diminish the contrast in the image. This alteration of the specimen environment can lead to structural changes. Such solutes may sometimes be removed without disrupting the virus on the grid by floating the grid on a drop of distilled water or an appropriate, low-concentration buffer for a fraction of a minute, in much the same way as is common with the preparation of negatively stained samples. This washing technique works even when holey carbon films are used, since it appears that the sample adheres to the air-water interface (325). A somewhat more concentrated sample (3 to 5×) should be used when this washing approach is employed.

Success in obtaining a vitrified sample suitable for cryo-EM thus depends on many factors, the most important of which are dictated by the properties of the virus (pI, enveloped or not, etc.), solution (pH, ionic strength, etc.), and support film and certainly the experience (and persistence) of the microscopist. A number of studies have demonstrated that the dimensions and integrity of most viruses and other fragile macromolecular assemblies are well preserved by cryopreparation techniques (1, 26, 130, 211, 228). This provides strong evidence that the specimens are fully embedded in a vitreous layer; those that are not become distorted and compressed (197).

Aqueous specimens examined by cryo-EM must be maintained at or below the devitrification temperature (~140°C) to prevent conversion of water in the sample to a crystalline state. A number of different designs for liquid nitrogen-cooled, cryospecimen holders have been described elsewhere (105, 109, 151, 152, 155, 301). Side-entry cryoholders have improved enormously over the past few years, but they remain less stable than conventional, room temperature holders.

Cryoholders are subject to greater instabilities due to the temperature gradient between the microscope at room temperature and the specimen and as a result of boiling of the coolant which transmits vibrations to the specimen. Current stage designs have minimized these problems with a satisfying improvement in resolution. Nevertheless, the maximum instrumental resolving power of most modern microscopes (~0.7 to 2 Å) cannot yet be achieved with currently available cold holders which promise stability in the 2- to 4-Å range. The development of a microscope which incorporates a liquid-helium-cooled, specimen stage and a superconducting objective lens has provided a means to achieve even better resolving power (127, 128, 196, 354, 355), as has the development of practical top- and side-entry stages for liquid helium work. These stages have shown their worth in the increased resolution obtained with crystalline specimens (52, 153, 154, 190) and should provide similar advantages for single particle specimens such as viruses.

Even when the specimen holder is rapidly transferred to the microscope, water vapor is carried in by the rod of the holder. The low temperature of the specimen unfortunately makes it an efficient sink for contaminants. The consequent high rate of specimen contamination forces the cryomicroscopist to work quickly or use an auxiliary anticontamination device. Several devices such as the blade-type anticontaminators, which closely sandwich the specimen, have been constructed and are commercially available (113, 164, 173). These devices significantly reduce the level of contamination so that cryo-EM can now routinely be carried out for a period of several hours with an individual specimen grid. One drawback to the use of the auxiliary anticontaminator is that it often restricts the range of tilt normally allowed on a goniometer stage. Fortunately, this does not pose severe constraints in most work with icosahedral viruses because they tend to be randomly oriented in the vitreous sample layer, and it is usually not necessary to record different views by tilting in order to obtain 3D information. When virions show a strong preference for one or a few orientations as occurs with reovirus and adenovirus due to their surface projections, a tilt of 10°, which can be accommodated by the anticontaminator, is usually enough to give an adequate sampling of orientations (108, 296). Should higher tilts be necessary, the anticontaminator can be retracted after about an hour when most of the water vapor from the specimen holder has been trapped. The structure of adenovirus was determined to 35-Å resolution by combining images of 29 particles (296). The fibers which project from the fivefold axes cause the virus to tend to lie in orientations which are away from the surface of the water layer. This orientation preference was overcome by tilting the stage by 10° during microscopy. This was sufficient in combination with the rotational disorder of the virions in the water layer to allow the collection of a complete data set (296).

After allowing some time (>15 min) for the specimen stage to stabilize after insertion into the microscope, the EM grid is searched at very low magnification (<×2,000 to ×3,000) and at a very low irradiation level (<0.05 e/Ų/s) to locate a suitable specimen area for photography. At low magnification it is possible to assess the relative thickness of the vitrified sample (Fig. 6). The switch from high- to low-magnification operation is conveniently selected in many microscopes by the push of a single button. Some cryomicroscopists prefer to view the "false," low-magnification image that is formed by highly defocusing the pattern formed in diffraction mode. The advantages of this technique are that the "diffraction" image exhibits very high contrast, and realignment of the microscope is usually not required. Many newer microscopes have a three-state system which allows switching among low magnification for searching, high magnification for focusing, and intermediate magnification for recording images. Areas on the grid that appear black are too thick for the electron beam to penetrate; areas that appear bright are likely to be dry. Sometimes, regions of the specimen appear to be vitrified but are actually dehydrated regions in which the buffer salt has crystallized and appears like an "ice" layer. Truly vitrified specimens of optimal thickness are usually characterized by a cloudy appearance. One can verify the vitrified nature of a particular specimen area at high magnification by the occurrence of bubbling at the sites of particles within the area during the first few seconds of irradiation.

View larger version (133K): [in this window] [in a new window]   FIG. 6.   Low-magnification views of vitrified samples of icosahedral viruses, including Nudaurelia capensis  virus (upper left and upper right), SV40 (lower left), and a mixture of polyomavirus and bromegrass mosaic virus (lower right). The grid square with the N. capensis  virus specimen (upper left) is a particularly poor candidate for cryo-EM, but it nicely exhibits a complete spectrum of ice thicknesses, from much too thick at the top (black region) to completely absent at the bottom (holes with no specimen). Only a few holes in the carbon film at the border of the very thick ice have a suitable thickness and distribution of virus particles. The strongly mottled appearance of the N. capensis  virus sample (top right) is caused by ice contamination that quickly builds up on specimens if a blade-type anticontaminator is not used during microscopy. The SV40 sample is a nearly perfect monodispersed distribution of particles in a uniform layer of ice. The bare circular region presumably resulted when surface tension effects in the thinning water layer excluded the ~500-Å-diameter particles just before vitrification occurred. The mixed polyomavirus and bromegrass mosaic virus sample shows a graded change in specimen thickness, from a thin region near the center of the hole where the small (~300-Å-diameter) bromegrass mosaic virus particles congregate in a single layer to progressively thicker regions marked by a ring of 500-Å-diameter polyomaviruses surrounded by more polyomaviruses and multiple layers of bromegrass mosaic virus. All magnification bars represent 5,000 Å.

Because most viruses smaller than ~1,000 Å are difficult to see directly at very low magnification, a convenient way to assess the particle concentration is to view one area (grid square) at a higher magnification and at a higher level of defocus, thereby sacrificing that specimen region. One then assumes that the particle concentration remains relatively constant in other areas of the grid for vitrified samples of similar thickness. The use of a video-rate TV camera system (e.g., see reference 104) provides a very effective means for scanning cryospecimens at a very low magnification and dose and simultaneously allows the user to easily assess particle distributions for viruses as small as 300 Å in diameter. Alternatively, a charge-coupled device camera can be used to visualize particles at higher magnification (232) and to check imaging conditions by performing a Fourier transform online.

Unstained, vitrified biological specimens are very sensitive to the effects of electron irradiation (Fig. 7). At 170°C, features in the 20- to 40-Å resolution range are unaffected by a total electron dose of ~10 to 20 e/Ų, a value similar to that found for negatively stained specimens (7, 318) but 5 to 10 times greater than that for unstained specimens at room temperature (317). Thus, microscopy must be performed in a way which minimizes exposure of the specimen to the electron beam. This requires the use of minimal irradiation procedures (e.g., see references 8, 317, and 345) which act to limit specimen exposure to the minimum necessary to record a micrograph with sufficient optical density and signal. Most modern microscopes offer some convenient mode of low-dose operation. Dose levels may be measured directly by the use of a Faraday cage (215, 316) or they may be estimated from the standard microscope exposure meter after calibration. This latter procedure is done by measuring the optical densities on an EM film of known sensitivity which has been exposed for different times and developed under carefully controlled conditions (e.g., see reference 8). A careful study of the effect of radiation damage on the cryoreconstruction of the HSV-1 capsid showed that the reconstructions of particles which had received 30 to 40 e/Ų had recognizable but blurred features compared to particles which had received a more standard 7 to 10 e/Ų (84). The high-resolution reconstructions of HepBc which yielded resolutions of better than 10 Å used single images taken at 10 to 16 e/Ų or pairs of images taken at 7 to 10 e/Ų each (42, 80).

View larger version (129K): [in this window] [in a new window]   FIG. 7.   Radiation damage in vitrified SV40 samples. Fine structural details are progressively lost in the SV40 particles as the electron dose increases from 10 to 40 e/Å2 (left series). The most prominent damage first appears as bubbling that occurs in the ice over the carbon support film (right series). Bubbling also occurs within particles suspended over the holes of the support film but usually only after doses of about 50 e/Å2 or more (data not shown).

The relationship between the electron image of the specimen and the specimen itself is, unfortunately, not straightforward. This relationship is described by the contrast transfer function, or CTF, which involves both phase- and amplitude-contrast components. These are characteristic of the particular microscope used, the specimen, and the conditions of imaging. The contrast-enhancing effect of the CTF arises from the spherical aberration present in all electromagnetic lenses and varies as a function of the objective lens focal setting and accelerating voltage (Fig. 8). The function includes both a phase-contrast and an amplitude-contrast component:
where (v) =  ·  · v²(f  0.5 · C_s · ² · ²), v is the spatial frequency, (per angstrom), F_amp is the fraction amplitude contrast,  is the electron wavelength (angstroms), where (0.042 Å for 80-kV electrons and 0.02527 Å for 200-kV electrons), V is the voltage (volts), f is the underfocus (micrometers), and C_s is the spherical aberration of the objective lens of the microscope in millimeters. The CTF is usually attenuated by an envelope function which depends upon the coherence of the beam, drift, and other factors (116, 329, 331). Thus, at any particular defocus setting of the objective lens, phase contrast in the electron image is positive and maximal at only a few specific spatial frequencies; at all other frequencies, contrast is either lower than expected, completely absent, or opposite (inverted or reversed) from what it should be. This allows the microscopist to selectively accentuate image details of a particular size much as an optical-phase microscope accentuates some features at the expense of others. This discussion ignores the effects of inelastic scattering, which can be severe for vitrified samples embedded in thick layers of water (192, 283). Inelastic scattering effects can be modeled by including an envelope function which imposes an exponential decay in the transfer function similar to that resulting from the lack of coherence in the beam (124). The effect is dramatically revealed in thick regions of the specimen; images of particles in these areas appear unsharp due to significant chromatic aberration effects.

View larger version (16K): [in this window] [in a new window]   FIG. 8.   The CTF for a Philips CM200 FEG at 200 kV is plotted as a function of resolution in angstroms for an underfocus of 2 µm and an underfocus of 1,000 Å and a magnification of ×36,000. The decrease in the amplitude of the function with increased resolution reflects the measured attenuation due to the lack of coherence in the source, specimen movement, and other optical effects. The value of 0.1 at the origin is the amplitude contrast portion of the function.

In practice, cryomicroscopists typically underfocus images by 8,000 to 20,000 Å (0.8 to 2 µm) to enhance specimen features in the 20- to 40-Å size range (Fig. 9). This amount of defocus is quite foreign to conventional biological microscopists who are used to imaging shadowed or stained specimens much closer to focus (3,000- to 5,000-Å underfocus). Sometimes it is advantageous to record two or more images at different focal settings to selectively enhance different features of the cryospecimen. For example, the overall organization of the glycoprotein spikes in the enveloped SFV is best enhanced at an ~3-µm or larger underfocus (Fig. 9), whereas the lipid bilayer is best visualized in images recorded much closer to focus (<1.5 µm). Images from a focal series can also be combined to compute 3D reconstructions as described in "Image analysis and 3D reconstruction" or can be useful for analyzing very noisy data (e.g., see reference 65).

View larger version (100K): [in this window] [in a new window]   FIG. 9.   The effect of defocus is shown for cryo-EM of SFV. The top four images show a defocus series taken on a Philips 400 microscope with a tungsten filament at 80 kV for illumination and underfocus values of 1.5, 3, 6, and 12 µm. Notice that the overall contrast is lower in the closer-to-focus image; however, the very fine details, such as the membrane, are recorded. The further-from-focus images do not show these details but do show better contrast for the large features of the specimen such as the spikes and the outline of the particle. This reflects the fact that there is high attenuation of the transfer function with a tungsten filament and hence only the information in the first peak of the CTF is transmitted efficiently. The bottom two panels show cryo-electron micrographs of the same sample taken on a Philips CM200 FEG under the same defocus conditions for which the CTF is plotted in Fig. 8. Both images show fine details because the illumination from a field emission source is more coherent than the tungsten filament and hence a larger range in resolution is transmitted. Images at different defocus must still be combined because information is lost at the transfer function nodes and attenuated near them. Identical particles in each of the focal series are indicated (arrows).

The first work on cryo-EM studies of viruses was performed with conventional transmission EMs operated at accelerating voltages of 80 or 100 kV. Recent experience shows that the use of FEG illumination is an enormous advantage for cryo-EM (72, 210). The major source of contrast in an electron micrograph of an unstained specimen is phase contrast. The higher coherence of a FEG source makes the phase contrast in the image much stronger. Practically, focusing and aberration correction become much simpler, and the coherence-dominated falloff of intensity with resolution is greatly reduced. The drive toward higher resolution has required the correction of the contrast transfer effects in the data. This is best done by combining images at different defocus levels (see below). This can be done much more reliably with FEG data because a greater number of Thon rings are visible in optical or computed diffraction patterns of images (116, 304). Higher-voltage instruments are also being employed. The potential advantages of the use of much higher voltages include potentially higher resolution and beam penetration and reduced problems with specimen charging. Higher voltage is particularly important for obtaining higher resolution for larger viruses (>1,000-Å diameter) (360) due to the curvature of the Ewald sphere at lower voltages (210). Finally, several groups have used spot scan imaging for the collection of virus data. This method was originally developed for two-dimensional crystalline specimens, for which it has been shown to decrease the apparent beam-induced drift seen in conventional flood illumination (102, 103). Although a carefully controlled study has not yet been published, several groups have used spot scan images for reconstructions (214, 234, 306, 358, 361-363).

The icosahedral image reconstruction method produces a 3D structure by interpreting the separate images of different particles as distinct views of the same structure. Once the cryo-electron micrographs are obtained, the determination of the structure requires three fundamental steps: (i) determining an initial orientation and center for each particle, (ii) refining these parameters by comparison of common data among different views, and (iii) combining the data from a sufficient number of unique views with their relative orientations to produce the final 3D structure (Fig. 10). This section gives an overview of the processing methods and factors that affect the quality of the reconstruction; the details of this procedure are discussed more extensively elsewhere (9, 133, 210). The methods are implemented in a number of continually evolving software packages (9, 92, 93, 194) which are available from the individual authors. One hallmark of the field is the willingness to share code and ideas among different groups, and consequently the different packages use similar approaches. They all had their origin in the original implementation of common lines (86).

View larger version (65K): [in this window] [in a new window]   FIG. 10.   Schematic diagram of the 3D image reconstruction process from digitization of the micrograph to the dissemination of the structural results. Though some steps such as the boxing out of individual particle images lend themselves to automation (e.g., see references 33 and 306), many steps including the determination of particle origins and orientations must be repeated and involve some trial-and-error decision making (e.g., to determine which data should or should not be included). The scheme depicted here, which uses both cross-common lines (133) and model-based (9) approaches to determine and refine particle origin and orientation parameters, is but one of many suitable schemes.

Other reconstruction methods have been used for cryo-electron micrographs of icosahedral particles, including the ROSE (reconstruction by optimized series expansion) method (326, 327), the COMET (constrained maximum entropy tomography) method (282), and angular reconstitution (297) as implemented in the IMAGIC software package (322). In the latter two papers, the reconstruction was performed by using orientations derived from the common lines approach (282, 297).

The bulk of the work in the reconstruction process rests in the determination and refinement of the orientations and origins of the views. Five parameters must be determined for each particle (Fig. 11): , the inclination of the view vector from a selected twofold axis (typically the z axis); , the azimuthal angle relative to the line connecting the adjacent fivefold axes; , the rotation of the projected view relative to the x axis in the scanned image; and x and y, the coordinates of the center of symmetry of the particle where all the symmetry axes cross. Icosahedral symmetry allows one to refer any view to one in the asymmetric unit. Hence, the unique views comprise a spherical triangle bounded by neighboring fivefold axes ( = 90.0°,  = ±31.72°) and an adjacent threefold axis ( = 69.09°,  = 0.0°). Typically, the view parameters must be determined to within about 1° for a low-resolution reconstruction and about 0.25° for a high-resolution reconstruction. This is possible because the projected image of a large object such as a virus changes dramatically with changes in its orientation (Fig. 12).

View larger version (25K): [in this window] [in a new window]   FIG. 11.   Definition of particle view orientation angles. By convention (189), , , and  angles define the orientation in which each icosahedral particle is viewed. The standard setting of the icosahedron places three mutually perpendicular twofold axes of the icosahedron coincident with the x, y, and z axes of a Cartesian coordinate system. The direction of the view vector (thick arrow) is given by  and  (the view direction 80°,15° is illustrated in this diagram), where  is the angle of the vector projected onto the xz plane measured positive from the z axis, and  is the angle of the vector projected onto the xy plane measured positive from the x axis. The rotational orientation of the icosahedron about the vector is given by the angle . Because the symmetry of an icosahedron makes it 60-fold redundant, any view vector can be referenced to a single asymmetric unit (shaded region), which is a spherical triangle bounded by neighboring fivefold axes ( = 90.0°,  = ±31.72°) and an adjacent threefold axis ( = 69.09°,  = 0.0°).

View larger version (118K): [in this window] [in a new window]   FIG. 12.   Change in projected structure with change in view orientation. (Upper panel) Images of a 3D reconstruction of BPV (19), projected on a regular grid at 3° intervals of  and  ( always = 0°) within a half of the icosahedral asymmetric unit (represented by dots in the right half of the inset). This gallery demonstrates how small changes in viewing angle can produce dramatic differences in the projected views of a 600-Å-diameter particle. The magnitude of these differences is correlated with the change in view direction and with the size of the particle. Hence, a 3° change in view direction leads to more-pronounced differences in projection images for a larger particle (>600 Å) or less-pronounced differences for smaller particles (<600 Å). (Lower panel [corresponds to boxed region of inset]). Demonstration that projected views at , and , are enantiomers related by a vertical line of mirror symmetry. Note that when  = 0°, the particle itself is mirror symmetric about a central vertical line (true for all images in leftmost column of upper panel). All equatorial views give rise to mirror-symmetric images. An icosahedron has 15 equators, each of which encircles the icosahedron along a direction that follows adjacent symmetry axes. For example, the equator in the xy plane ( always = 90°;  varies between 180° and +180°) crosses, in order, the following symmetry axes: two-, five-, three-, two-, three-, five-, two-, five-, three-, two-, three-, and fivefold. Any view corresponding to a combination of  and  which lies on this equator will be a mirror-symmetric image with the mirror line parallel to the equator. For example, the bottom row of projected images in the upper panel (which represent some of the views along the xy equator) all exhibit horizontal lines of mirror symmetry. Projection views along strict symmetry axes are additionally unique because they exhibit n mirror lines, where n (= 2, 3, or 5) is the symmetry of the axis in view.

When a symmetry axis is along the direction of view or projection, the image obeys the symmetry of the axis. An axis which is not along the direction of view gives rise to common lines in the transform of the projection. Application of the symmetry operation to the transform of the projection yields a second, identical plane which intersects the first along a line. Application of the inverse of the symmetry element to the original transform generates a third, identical plane with a second line of intersection. These lines in the original transform are designated "common lines" since they are common to the original and symmetry-related planes. The values of the transform in the original plane along them must be identical since they are related by symmetry operations on the original plane (89, 133). Each pair of a symmetry element with its inverse gives rise to such a pair of common lines which lie to either side of the projection of the symmetry axis.

The 60 symmetry elements of the 532 point group symmetry of an icosahedral particle generate 37 pairs of common lines in the Fourier transform of an image of its projection. These 37 pairs include two pairs from each of the 6 fivefold axes, one pair from each of the 10 threefold axes, and one pair from each of the 15 twofold axes. Sixty pairs of common lines exist between the transforms of any two projected particle views. The positions of these lines, along which the values of the transform should be identical for objects with perfect icosahedral symmetry, are determined solely by the direction of the view. The orientation angles corresponding to an observed view can be found by stepping  and  through the asymmetric unit and selecting the values which give the lowest sum of residuals over all 37 pairs of common lines in the transf