The Current Revolution in Cryo-EM

Main Text

Structural biology is the study of the molecular architecture of proteins and nucleic acids, which are the basis for all life forms. Structural biology came into its own as a field during the 1950s when the atomic structures of DNA (1) and several globular proteins (2) were solved. Knowledge of these structures alone is not enough to understand their functions, but it has become clear that a detailed mechanistic picture of function is not possible without structural information. Studying structure can reveal how molecules have evolved, and this type of insight would otherwise be lost by looking at only the molecule’s sequence.

X-ray crystallography has been the primary technique responsible for determining atomic models of macromolecules and macromolecular complexes during the past 60 years. X-ray crystallography begins with the crystallization of a purified molecule or molecular complex. An x-ray beam is directed at the crystal, and a unique pattern results from the diffraction of x-ray beams in different directions after encountering atoms in the crystal. An atomic model is generated from this diffraction pattern data that satisfies stereochemical constraints and whose diffraction matches the observed pattern within experimental error. This approach has been enormously powerful but is limited by the fact that the molecule or complex of interest must be crystallized, which is not always possible. Polymers, such as protein or nucleoprotein filaments, typically must have exactly two, three, four, or six subunits per turn, or multiples of these numbers, to pack into a crystal.

Nuclear magnetic resonance techniques emerged in the 1980s and allowed people to determine the three-dimensional (3D) structure of macromolecules in solution (3). These solution techniques can be quite laborious and are very difficult to use for large complexes.

In this review, I will discuss how cryo-EM (electron cryo-microscopy to some in the field, cryo-electron microscopy to others) has rapidly (Fig. 1) emerged as one of the main techniques for determining the structures of many macromolecular complexes at near-atomic resolution.

Figure 1.

The number per year of 3D cryo-EM reconstructions deposited to the Electron Microscopy Data Bank (EMDB) at better than 5.0 Å resolution. To see this figure in color, go online.

The “hardware”

The first electron microscope was built in the 1930s by Ernst Ruska, but it was not until ∼1945 that the first applications to biological samples were made by Keith Porter and colleagues. Electron microscopy (EM) has been a valuable imaging tool for biological specimens but is limited by both hardware constraints and the nature of biological samples. In contrast to light, electrons scatter strongly from air, which requires that the column in an electron microscope be under a vacuum. This generates a problem in imaging biological material because a specimen, which may be found naturally in an aqueous environment, must be dried before it can be introduced into the vacuum of the electron microscope. Another major problem of the electron microscope arises from the fact that the contrast generated by thin biological samples is quite weak, much as the contrast generated by a cell in solution is quite weak in the conventional light microscope. Different staining methods have been developed to generate contrast for dried biological samples in the electron microscope, just as histochemists have worked for more than a century to create such stains that can be used in the light microscope. Since the focus of this review is on EM as a tool for macromolecular structure determination, I will not deal with the very rich history of using EM to look at whole cells and tissues, or with the recent impressive advances in electron tomography (4).

The negative staining method for EM emerged in the 1960s and was used extensively in the 1970s and 1980s for looking at isolated macromolecules and macromolecular complexes. In light microscopy, positive stains typically bind to a particular substance and this provides the contrast between what is stained and readily observed and the unstained remainder. The negative stain technique, developed by Hugh Huxley and colleagues (5), involves surrounding a sample with an electron-dense solution, such as uranyl acetate. This solution is excluded by the molecule or complex of interest and can be dried to form a glass that is placed in the vacuum of the electron microscope. Negative staining reveals the molecule due to its stain exclusion properties, and only the shape of the region where stain has been excluded can be distinguished. The negative stain technique might be able to show whether a protein was globular or elongated, but does not elucidate characteristics of a molecule’s secondary structure, such as the presence of α-helices and β-sheets, which determine the overall folding of the protein.

A great advance came in 1975 when Richard Henderson and Nigel Unwin demonstrated that unstained two-dimensional protein crystals of arrays of bacteriorhodopsin, imaged by EM at a resolution of ∼7 Å, could show the presence of α-helices (6). These arrays were embedded in glucose, which preserved the array’s structure even when dry. The contrast was exceedingly weak in glucose-embedded samples, and the use of crystals was still needed to generate a visible structure. At around the same time, Ken Taylor and Bob Glaeser demonstrated that frozen and fully hydrated samples could be imaged by EM (7) using a cryo-stage that kept the specimen near liquid nitrogen temperatures while it was in the vacuum of the electron microscope. This surmounted the fundamental problem plaguing biological EM, which was the previous inability to image hydrated samples in the microscope. But Taylor and Glaeser froze samples conventionally, which meant that the water froze into crystalline ice and caused irreversible damage to biological samples due to changes in water volume. The field of cryo-EM took an enormous leap forward when Jacques Dubochet and colleagues developed a method for the routine vitrification of EM samples (8). When water is frozen extremely quickly, it undergoes vitrification and forms an amorphous solid phase, a glass that is not crystalline (9). Given the relative molecular simplicity of water, this was a surprising observation because such a phase was never predicted theoretically. This transition to a vitreous glass does not disrupt macromolecular structures. Rather, molecules become frozen in whatever state they exist in solution, which is called cryofixation. The thin vitrified film containing the molecules of interest can be maintained at liquid nitrogen temperatures for many days in a vacuum with negligible sublimation. An entertaining personal account of the development of vitrification for EM samples by Jacques Dubochet has appeared recently in Biophysical Journal (10).

The contrast of macromolecules embedded in vitreous ice was much greater than when embedded in glucose, but these are still weakly scattering objects that can best be viewed by a phase-contrast method, similar to phase-contrast in light microscopes. The phase-contrast technique used in cryo-EM involves defocusing the microscope but requires a coherent source. Conventional electron microscopes use a simple filament as an electron source much like that found in an incandescent light bulb, but these sources lack the coherence needed for high-resolution phase-contrast imaging. A field emission gun for the electron microscope had been developed (11), and the combination of this source, commercial cryo-stages, and a vitrification method that could be used reproducibly and reliably meant that cryo-EM started to be used in many laboratories around the world in the 1990s.

EM can be quite labor intensive, as an experienced microscopist might need to spend weeks or even months on the microscope to collect the large amount of images needed for the image analysis and reconstruction discussed below. The fully automated electron microscope was developed by Carragher and colleagues (12), which was a significant advance that allows current microscopes to work in an unattended manner 24 h a day, 7 days a week.

The traditional means of recording an image in the electron microscope, dating back to the time of Ruska, involved photographic film. But using film is very tedious, it must be developed and then scanned for subsequent digital image processing, and it limits how many images can be acquired in a day. For many applications, charge-coupled device (CCD) detectors were used to surmount these problems, but CCD detectors were worse than film in terms of sensitivity and resolution. The current revolution in cryo-EM is due directly to the adaptation of complementary metal oxide semiconductor chips (13), hardened to prevent damage from electrons, which have a resolution and sensitivity greater than film and a readout rate much faster than CCD detectors.

The “software”

The images obtained by cryo-EM are projections of a 3D structure onto a two dimensional film or detector. Like a medical chest x-ray, these projections can be rich in information but can be hard to interpret due to the superposition of all structure onto a single plane. At about the same time that computed tomography was being developed in medical radiology, it was also realized that one could recover the 3D information from EM specimens. David DeRosier and Aaron Klug generated the first 3D EM reconstruction (14). Rather than using multiple images as would be done in medical tomography, they took advantage of the helical symmetry present in the tails of an icosahedral bacteriophage. The helical symmetry means that identical copies of a protein are related to each other by just a rotation and translation in the tail, so a single projection image of the tail provides all of the information to generate a 3D reconstruction. A vast number of assemblies in biology are helical, but many other types of structures exist. Single particle methods in EM began with Joachim Frank and colleagues (15), and took advantage of the fact that when a large ensemble of molecules is imaged by EM, all possible orientations may be found. Thus, a 3D reconstruction can be generated by determining the orientations of individual particles. Single-particle approaches have been applied to helical structures (16), overcoming the need in the original DeRosier and Klug method (14) for long-range helical order in a structure to achieve a reasonable resolution. In a crystal, long-range order is maintained by space group symmetry but a helical polymer does not maintain such long-range order. Deviations from an ideal or average symmetry accumulate, and a segment that is far from a reference segment may show little correlation in terms of the expected positions of subunits. This is similar to the bending of a polymer in solution or a strand of spaghetti in boiling water. The conformation of a polymer having the minimum energy is typically a straight rod, but thermal fluctuations accumulate, thereby knowing the orientation of the polymer at one point does not allow one to predict the orientation of a segment a long distance away. The helical viruses shown in Fig. 2 are rather rigid compared to flexible filamentous plant viruses, such as bamboo mosaic virus (17), but an analysis of the helical symmetry in such viruses shows that long-range helical order does not exist. With single particle methods in cryo-EM, a reconstruction at ∼3.8 Å resolution (Fig. 3) was obtained for Sulfolobus islandicus rod-shaped virus 2 (SIRV2) (18), something that would not have been possible at any reasonable resolution just 3 to 4 years ago, before direct electron detectors were commercially available.

Figure 2.

A cryo-EM showing two Sulfolobus islandicus rod-shaped virus 2 (SIRV2) virions (18). SIRV2 is a virus that infects a hyperthermophile living in nearly boiling acid. Almost all of the long capsid seen is formed by several thousand identical copies of a coat protein that wraps around and protects the DNA. The arrows point to double-stranded DNA that can be seen emerging from the ends of the virions. The scale bar is 500 Å.

Figure 3.

The interior of a 3D reconstruction from SIRV2 (18). Most of the capsid protein in the virus is found in the form of α-helices, and at a resolution of ∼3.8 Å the pitch and right-handed twist of these α-helices are readily visible in the density map (transparent gray surface). Every capsid protein is identical, and the building block of this assembly is a symmetrical dimer formed by two copies of the capsid protein. Thus, the yellow and green helices come from two different subunits and form antiparallel helices because the polypeptide chains are oriented in opposite directions from each. At the available resolution, bulky aromatic side chains can be visualized (arrows). Because the sequence of the capsid protein is known, this allows for an unambiguous threading of the sequence through these helices. To see this figure in color, go online.

The number of single particles used in 3D reconstructions by cryo-EM can easily exceed hundreds of thousands, something that would have been computationally impossible back in 1968 when 3D EM began. Just as progress in DNA sequencing (due to the highly computational nature of current sequencing) has roughly followed Moore’s Law, stating that the number of transistors in an integrated circuit doubles every 2 years, progress in cryo-EM has also followed the advances in computing power. Many images are needed to overcome the inherently poor signal/noise ratio in cryo-EM. Fig. 2 shows an example of a typical micrograph. Most of the power in the image, which is proportional to the variance, or the sum of the squared deviations from the average intensity, in the image is actually due to noise. A significant part of this noise arises from the electron counting statistics or shot noise because a low enough electron dose is used, which can cause random fluctuations in electron counts in neighboring pixels to become significant. This noise is more negligible as many particle images are aligned and averaged together, and a much greater problem becomes the conformational heterogeneity that exists for many molecules and complexes when they are not packed tightly in a crystal.

The approach to this heterogeneity in solution NMR is to generate an ensemble of structures, all consistent with the constraints observed, whereas in cryo-EM one might generate a few structures that clearly correspond to different conformational states or only a single structure, excluding those particles that do not match this structure. For the ribosome, a recent study (19) described starting with >1.6 million particles, which were reduced to a more homogeneous subset of ∼400,000 particles, yielding a resolution of better than 2.9 Å. We simply do not know the limit of resolution obtainable by cryo-EM for macromolecules. A recent structure (20) of a bacterial enzyme, β-galactosidase, reached 2.2 Å (Fig. 4) and the authors suggested that the main limitation on resolution that we now face may be simply the intrinsic flexibility of proteins and not the hardware or software that we use. Nevertheless, it is clear that further improvements in microscopes, detectors, and image processing software will lead to many more macromolecular complexes that can be solved at resolutions approaching 2.0 Å. The growing field of super-resolution light microscopy has been aimed at surmounting the fundamental limitation on resolution in light microscopy, the wavelength of light (∼0.5 μm or 5000 Å). In cryo-EM, the wavelength of the electrons is typically <0.02 Å, therefore the wavelength of the illumination does not set the physical limit on resolution.

Figure 4.

A small region from the 2.2 Å resolution cryo-EM reconstruction of β-galactosidase (20). The resolution is high enough to see an ordered and bound water molecule in the center (yellow). To see this figure in color, go online.

What biological insight does one gain from higher resolution? Consider Fig. 5, which shows a β-sheet from the sheath of a type VI secretion system in Vibrio cholerae. The resolution of this cryo-EM reconstruction (21) was ∼3.2 Å, high enough to allow a complete chain trace of ∼600 amino acids in the asymmetric unit of the sheath. Because the sequences of the two proteins in this asymmetric unit were known, it was possible to thread this sequence through the density placing the large and bulky side chains (as in Fig. 3) into their corresponding density. If one had only 5 Å resolution, portions of this structure might have been built correctly but ambiguities would have existed, such as in the β-sheet shown, and it would be unlikely that the correct connectivity could be established. Or consider Fig. 4, where ordered water molecules are visualized by cryo-EM reconstruction, and suggest that the resolution is high enough to understand enzymatic reactions or design drugs.

Figure 5.

A region from the type VI secretion system sheath of Vibrio cholerae (21), reconstructed by cryo-EM at ∼3.2 Å resolution. The atomic model that was built into the reconstruction (transparent gray surface) has three different molecules shown in cyan, red, and blue. At a resolution worse than ∼4 Å, one might have ambiguities in tracing the individual polypeptide chains present in each molecule. The resolution that was achieved prevented any such ambiguities, and it was clear that the β-sheet shown involved two strands from one molecule (cyan), and one strand from each of two different molecules (red and blue). To see this figure in color, go online.

Conclusions

The rapid advances in cryo-EM over the past several years make it nearly impossible to predict where the field will be in several years. It is reasonable to expect that the exponential growth of near-atomic resolution structures determined by cryo-EM (Fig. 1) will continue, but we still do not know possible limits in resolution or how large a complex must be to reach near-atomic resolution. The future of cryo-EM is certain to be exciting, with new biological insights gained from this powerful technique.