Single-particle cryo-EM—How did it get here and where will it go

It is actually quite difficult to “look” at biological macromolecules in an electron microscope. Determining their atomic structures from electron micrographs is even more complicated. First, EM images are 2D projections of biological macromolecules, but not their 3D structures. This was resolved by De Rosier and Klug, who demonstrated that a 3D structure can be reconstructed by combining 2D projection images of the same object along different directions (7).

Second, because of strong scattering, the electron beam has to be confined in a high vacuum, and all EM samples need to be placed inside said vacuum. This is not a problem for inorganic materials. But if one simply places a biological sample inside an electron microscope, vacuum-caused dehydration would destroy the sample’s structural integrity. The seemingly impossible task of keeping protein samples hydrated in a high vacuum was accomplished by Taylor and Glaeser. They recorded better than 3-Å resolution electron diffraction patterns from frozen hydrated catalase crystals, demonstrating that the structural integrity of biological macromolecules in a high vacuum can be maintained through frozen hydration (Fig. 1A) (8, 9). The practical implication of this approach, however, was not easy until a plunge freezing technique was developed by Dubochet and colleagues in the 1980s (10, 11). They applied purified protein samples in solution to an EM grid covered with a thin layer of carbon holey film and blotted the grid with a filter paper, which removed most of the solution, and surface tension drove the remaining solution into a thin liquid film across holes in the carbon film. Plunging the grid rapidly into liquid ethane cooled by liquid nitrogen froze it into a thin layer of amorphous ice with the protein sample embedded within it in random orientations. After that, the frozen grid was transferred into an electron microscope and kept at near–liquid nitrogen temperature for imaging (Fig. 1B). This method is still used routinely without major changes, except that we now use a machine to blot and plunge grids with tunable parameters.

Third, radiation damage by the high-energy electron beam limits the total electron dose that can be used to image biological samples. The consequence is that images recorded with such a low electron dose have very poor signal-to-noise ratios (SNRs). Cooling the sample to liquid nitrogen or even liquid helium temperature can reduce the radiation damage and allow the sample to tolerate a higher electron dose (12, 13) but still far from being able to directly visualize high-resolution details from raw micrographs. Henderson and Unwin overcame this problem by use of a crystallographic approach, averaging images of many identical proteins packed as 2D crystals (14). Images of glucose-embedded 2D crystals recorded with very low electron doses show no visible features, but their Fourier transforms show clear reflections. Similarly, electron diffraction from such 2D crystals at very low electron doses produced good-quality diffraction patterns. Combining the phases calculated from the Fourier transformations of images and the amplitudes obtained from diffractions produced a high-resolution projection map of the specimen (14). Combining data collected from specimens tilted at different angles produced a 3D reconstruction similar to the density map of an x-ray crystal structure. This approach produced the first structure of an integral membrane protein, initially at ~7-Å resolution (Fig. 1C) (15) and finally at atomic resolution (16). The method became known as electron crystallography, which relies on well-ordered 2D crystals. This method has produced atomic structures of several integral membrane proteins (17, 18) and one soluble protein (19). The highest resolution achieved was 1.9 Å, resolving a lipid bilayer surrounding an aquaporin-0 (AQP0) water channel (20). But the difficulty of growing well-ordered 2D crystals hinders the broad application of the method.

In parallel, Frank proposed an idea to determine protein structures without crystallization: computationally combining images of many individual protein particles of the same type (5). This conceptually novel idea was first tested by using protein samples that were negatively stained for EM observation (Fig. 1D) (21, 22). The later combination of this approach with the plunge freezing sample preparation became what we now call “single-particle cryo-EM.” It does not require growing proteins into crystals of any form. Instead, it determines structure by computationally aligning and combining cryo-EM images of many biological molecules randomly oriented within a thin layer of vitreous ice. A large number of images is needed to both enhance SNR and to provide all different views needed for 3D reconstruction. More detailed technical descriptions of this method can be found in many recent reviews, such as (23, 24).

Furthermore, what electron micrographs record are projections of the specimen convoluted by a contrast transfer function (CTF). CTF is a sine function oscillating in a frequency-dependent manner that modulates both phase and amplitude of an image in frequency space (25). In addition to a number of microscope-dependent parameters, CTF is determined by how much off the focus an image is recorded—the so-called “defocus.” To retain the highest resolution, images must be recorded very close to focus. Such images, however, have very limited contrast. This is not a problem for radiation-insensitive inorganic materials, which can be imaged with very high electron doses so as to generate sufficient contrast while retaining a high-resolution signal. However, images of radiation-sensitive frozen hydrated biological samples have to be recorded with a large defocus in order to generate sufficient contrast, which in turn dampens the high-resolution signal.

The resolution of a single-particle cryo-EM structure depends on many factors, including the resolution and contrast of individual particle images, accuracy of aligning these images with each other, obtaining a sufficient number of images from all necessary views of the macromolecule within a reasonable time frame, the conformational and compositional homogeneity of these particles, and access to powerful enough computers with which to process images efficiently. Less than optimal conditions for any of these factors—for example, instability of the electron microscope, relatively poor performance of the image-recording device and beam-induced sample motions, accuracy of classifying and aligning particle images limited by image quality and computational algorithms, and limited computer power to process very large numbers of particle images—can limit the resolution. Because of these many technical challenges, the resolution achieved was, for a long time, limited to levels far from sufficient for deriving de novo atomic models.

At a time when the resolution of some best reconstructions was at the 30- to ~50-Å range [for example, the mammalian 40S ribosome (26) and ryanodine receptor (27)], microscopists were nevertheless encouraged by the prediction that single-particle cryo-EM could, theoretically, achieve atomic resolution (28). The prediction was made by considering how much electron scattering a biological sample could tolerate, how much structural information or SNR such scattering can produce, and—in a perfect situation—how many images would be needed to produce a reconstruction at a given resolution. Although the claim was bold, the theories behind are solid, and this motivated the cryo-EM community to push the methodology toward its perfection. At last, 20 years after Henderson made the prediction, the necessary breakthrough came when direct electron detection cameras were developed and became commercially available (29). The expanded capabilities of the new cameras coupled with unprecedented and ever-increasing computational power fueled the development of new computational algorithms so that cryo-EM images could be reliably produced with sufficient quality for atomic structure determination (30, 31). The resolution potential of single-particle cryo-EM at long last became a reality (32).

Thanks to these technological breakthroughs in single-particle cryo-EM, structural biology has entered a new era. Structures of many difficult crystallization targets are now within reach. One such area is integral membrane proteins. A specific example is the transient receptor potential (TRP) ion channel. The TRP channel superfamily contains seven subfamilies with a total of 27 members in humans (45, 46). Each of these channels plays different physiological roles; some of them are drug targets for treatment of various human diseases (47). With the exception of some small domains, attempts to crystallize any member of the TRP channel superfamily had failed (48). This lack of structural information ended when atomic structures of the TRPV1 ion channel—a capsaicin receptor that plays a physiological role in sensing heat and activating pain pathways—were determined by means of single-particle cryo-EM in three different functional states (49, 50). This discovery demonstrated the power of single-particle cryo-EM and showed that it rivaled x-ray crystallography in determining atomic structures of challenging protein complexes that resist crystallization. The TRPV1 structures prompted many crystallographers to think seriously about cryo-EM, and many quickly seized the opportunity to apply it to their favorite difficult targets. With the roadblock of crystallization removed, atomic structures of integral membrane proteins are now being determined at a rapid pace. In less than 5 years, atomic structures of at least one member of each of the seven TRP channel subfamilies have now been determined (Fig. 3).

Single-particle cryo-EM is also game-changing for structural studies of many large and dynamic complexes and machineries that were impossible for crystallization. A traditional approach was to fit crystal structures of individual components or domains into lower-resolution cryo-EM density maps of the whole complex (51). Now, atomic structures of many large complexes are determined directly by using single-particle cryo-EM. An excellent example is the spliceosome complex. Past efforts produced atomic structures of some small components (52), as well as low-resolution cryo-EM structures of the whole complex (53). Now, in only a few short years, atomic structures of the spliceosome in different functional states have been determined (54, 55).

One can easily call forth many more examples to illustrate how the recent technological breakthroughs in single-particle cryo-EM have changed how we tackle complex biological problems. The rapid pace of advancement in structural biology is unprecedented. It has also attracted major pharmaceutical companies, with many hoping to implement cryo-EM into structure-based drug discovery and optimization.

Single-particle cryo-EM continues to move forward fast, with many new technologies being developed. One such example is the Volta phase plate, which is a thin carbon film placed in the back focal plane of the microscope’s objective lens. It adds a phase shift to the CTF so that an image recorded at near focus has good contrast (56), facilitating the study of very small proteins (57). Another example is the development of a new sample preparation technology that is fundamentally different from the current blotting method (58). The new method promises many benefits, including reducing the total amount of sample needed from microliters to nanoliters.

As impressive as these innovations have been, there are still many technical details that can be improved. Regularly achieving resolutions close to 2 Å and beyond is one goal; improving robustness and throughput is another. Once very high resolutions can be achieved reliably, pharmaceutical companies will be able to routinely use EM to speed up structure-based drug discovery. The range of what can be studied by means of single-particle cryo-EM can also be expanded to include smaller or bigger targets with higher resolution, as well as more dynamic complexes or assemblies with irregular shapes.

There is still more that single-particle cryo-EM can offer to further biological discoveries beyond structure determination if we continue to push the boundary of this technology. For example, in theory, image classification can sort a cryo-EM dataset that contains particles of heterogeneous conformation or composition into multiple classes, each corresponding to a different functional state. By freezing cryo-EM grids at specific time points, it could be possible to derive structural information in a time-dependent manner (59). It may therefore be possible to understand and break down the complex cycles, movements, and processes of biological macromolecular complexes and machineries step by step, in complete atomic detail. In the future, it may also be possible to study both physiological and pathological protein states through affinity-purifying specific proteins directly from cells onto EM grids for single-particle cryo-EM studies.

The past 40 years have seen the evolution of single-particle cryo-EM from blobology to a routine and powerful method for atomic structure determination. The technique’s new popularity is attracting not only more users but also talents from different fields, ranging from physics, material science, and mathematics, to machine learning. This infusion of new ideas and a larger community heralds an even brighter future for single-particle cryo-EM full of method development, collaboration, and most importantly, wonderful new biological discoveries.

This is a brief story of single-particle cryo-EM but not a comprehensive historical review. I apologize to many colleagues whose important works are not cited here. I thank many of my colleagues and all past and present members of my laboratory for sharing their thoughts about single-particle cryo-EM with me, and L. Wang for editing. The cryo-EM work in my laboratory is funded by various NIH grants: R01GM082893, R01GM098672, R01HL134183, P50GM082250 (Nevan Krogan), P01GM111126 (Robert Stroud), and S10OD020054 and S10OD021741. I am a Howard Hughes Medical Institute investigator.