1. INTRODUCTION

The steep increase in the number of Electron Microscopy Data Bank entries over the last decade indicates that cryo electron microscopy (cryo-EM) has found its way into the mainstream of methods used for macromolecular structure determination. Continual advances in every aspect of the pipeline for structure determination using cryo-EM have created significant interest in a wide spectrum of biological disciplines in which the application of these methods could address previously intractable problems. Numerous recent reviews have documented the rise and general application of cryo-EM methods (1–6). In this review, we highlight a selected subset of biological problems that have been particularly challenging to address using existing structural methods, but where the application of cryo-EM is now affording new and unprecedented insights. In Section 2, we provide an overview of some of the critical technical steps involved in the cryo-EM workflow. In Section 3, we highlight three paradigmatic examples (G protein–coupled receptors, spliceosomes, and fibrils) that illustrate what is now possible with cryo-EM in the context of historically challenging problems in structural biology. In Section 4, we briefly outline emerging themes in the use of tomography. Section 5 summarizes.

2. SINGLE-PARTICLE CRYO ELECTRON MICROSCOPY METHODOLOGY

Cryo-EM has evolved into a powerful structural biology method for visualizing a wide range of biological specimens in a near-native state (5, 7–13). Initially, the major advances in the field came from studies of highly symmetrical viral particles and large, stable, multicomponent complexes such as ribosomes (14–17). More recently, barriers such as practical resolution limits (18, 19), size of macromolecular targets (20–22), lack of internal symmetry (23), and sample solubility (24–31) have in many cases been overcome. Although many critical improvements have propelled the application of cryo-EM methods, progress in all aspects of the workflow, ranging from specimen preparation to better detectors and improved image processing, can be expected in the near future (6, 32–34).

2.1. Key to Success: Sample Preparation

The likelihood of successful structure determination depends directly on sample quality, and thus sample preparation. In a broad sense, protocols for sample preparation for cryo-EM methods resemble those for X-ray crystallography and NMR. However, there are both challenges and advantages that are unique to cryo-EM, as reviewed below.

One of the most advantageous features of single-particle cryo-EM over X-ray crystallography is the requirement for only small quantities of pure sample instead of well-diffracting crystals. This can be transformative for a number of macromolecular targets that are difficult to obtain in large quantities for crystallization trials and/or are recalcitrant to crystallization (5). There are important differences in the nature of the sample requirements. Crystallization is essentially a purification technique that captures a particular sample in a specific, low-energy state within an ensemble of conformations. However, the same protein solution that results in diffracting crystals may adopt multiple conformational states in solution. This problem is accentuated for multicomponent macromolecular complexes, where weak association may lead to dissociation of components from a given assembly. The degree of difficulty in structure determination is directly related to the conformational homogeneity of the specimen, although powerful image processing methods are available to classify and resolve discrete conformations that are present in mixtures (35). These methods are especially useful for following time-resolved changes in conformation, when specimens are plunge-frozen at different points in time after the start of a reaction sequence (36).

Classical biochemical and biophysical tools can be useful in addressing some of the challenges posed by sample heterogeneity. Optimization of the buffer composition may stabilize the target sample while minimizing the impact of compounds that contribute to unwanted electron scattering and decrease the signal-to-noise ratio (SNR) (37). Dynamic light scattering and differential scanning fluorimetry devices provide information about sample polydispersity, size distribution, and ultimately sample stability (38). Labile multisubunit macromolecular complexes may require mild cross-linking for structural stabilization using different cross-linking agents [e.g., glutaraldehyde, bis(sulfosuccinimidyl) suberate, bis(sulfosuccinimidyl) glutarate] in conjunction with the gradient-based cross-linking method GraFix, where the sample is subjected to ultracentrifugation separation through simultaneous density and cross-linker gradients (39).

Membrane proteins are notoriously recalcitrant to crystallization and structure determination by X-ray crystallography, but cryo-EM offers an alternative and perhaps superior route to high-resolution structure determination (see Reference 40 for a recent review). Table 1 illustrates the variety of membrane protein targets whose structures were reported in 2018 at a reported resolution of 4 Å or less. Since purification of integral membrane proteins necessitates their extraction from the native membrane environment by detergents, the presence of the detergent may introduce artifacts that pose challenges for structure determination (40, 41). For example, detergent monomers alter the surface tension of the sample, which could lead to a thicker ice layer during grid preparation. Free micelles in the sample may decrease the SNR and could be mistakenly identified as protein particles in the selection of projection images for three-dimensional (3D) reconstruction (i.e., the particle picking step). Also, detergent bound to the membrane protein cores can form belts of inconsistent sizes, causing problems during data processing (40). A case study on detergent background in negative-stain electron microscopy concluded that decyl-maltoside and digitonin were acceptable for structural studies; although reduced, the background was still noticeable below the critical micellar concentration (41). GraDeR, a method that involves gradual removal of the excess of detergent micelles, uses concurrent density gradient and size separation (42). The reader is referred to Table 1 for a list of the detergents used for the purification of various membrane protein targets.

Table 1

Selected membrane protein structures by single-particle analysis with a reported resolution ≤4 Å as of 2018^a

An alternative approach is the preparation of detergent-free membrane protein samples. This approach includes the use of amphiphilic polymers or amphipols that bind to membrane proteins with stronger affinity than detergents (43) and nanodiscs that mimic the native membrane bilayer (Table 1) (44, 45). A recent study used peptidiscs, composed of short, amphipathic, bihelical peptides (46), and engineered nanostructured β-sheet peptides, which form filaments in solution but undergo reconstruction once the membrane protein is added (47). Styrene maleic acid copolymer–lipid particles (SMALPs) are also beneficial to membrane protein studies (48) and have been successfully used for structure determination of proteins such as the Escherichia coli multidrug efflux transporter AcrB (28).

2.2. The Air–Water Interface: Grid Preparation

A central element of the cryo-EM pipeline is the process of vitrification, in which a bulk suspension of the solution of interest is first reduced to a thin film by blotting, followed by rapid plunge-freezing into a liquid cryogen such as ethane cooled by liquid nitrogen (49). Currently, vitrification of EM samples is partially automated in commercially available plunge-freezing stations that offer options for humidity and temperature control (50). The length of time between blotting and freezing can be shortened to be within ∼100 ms, but protein denaturation at the air–water interface remains a major impediment to sample integrity. This is because proteins diffuse from the bulk of an aqueous solution to the air–water interface within milliseconds (51), giving rise to multiple opportunities for the protein to be subject to denaturing forces during specimen preparation.

Numerous attempts have been made to optimize grid preparation so that the interaction between particles and buffer surfaces is altered. The most common modifications include changes in the buffer composition (e.g., pH, additives, detergents, ligands), the use of grids with a support film [e.g., thin carbon layer (52), graphene oxide], or the use of grids with affinity support films (e.g., streptavidin monolayer, nickel–nitrilotriacetic acid, antibody-functionalized film) (5). D'Imprima et al. (51) performed a quantitative analysis of protein denaturation at the air–water interface, comparing unsupported grids and grids coated with a layer of graphene rendered hydrophilic with 1-pyrene-carboxylic acid (53). Only a minority of the fatty acid synthase particles used in their study retained integrity in unsupported vitreous films. However, use of EM grids with hydrophilized graphene resulted in a strong tendency of fatty acid synthase to locate at the graphene–water interface, preserving the integrity of the complex (51). Efforts to discourage preferential particle orientation (and improve the ice thickness distribution of the grid) include saturation of the drop surface with, for example, fluorinated detergents (Table 1) (44, 54, 55).

Most recent approaches include the grid-making methods devoid of the blotting step. One of these methods is the blotless Spotiton method, which uses a piezoelectric inkjet dispenser depositing picoliters of sample onto special nanowire self-blotting grids. The method is optimized to wick away the excess of the liquid as soon as the sample touches the nanowires, leaving only a thin layer of the sample on the grid (56). Glaeser et al. (57) are developing another interesting approach in which, instead of blotting, sample thinning is accomplished by creating the surface tension gradient using a flow of a volatile surfactant under conditions of high humidity.

2.3. Engineering Advances and Programming Enthusiasts: Hardware and Software

Innovations in the design of electron microscopes and related accessories have continued to advance over the last decade. These have included tools for robotic handling of specimens with an autoloader; lens design that permits stable operation over a wide range of voltages; energy filters for imaging of thicker specimens; phase plates for enhanced contrast (58); and most importantly, the switch from the use of photographic film and charge-coupled devices to direct electron detectors, which has resulted in dramatic improvements in image quality and the ability to compensate for beam-induced specimen movement during electron-beam exposure (59). The use of these detectors in conjunction with automated data-collection procedures implemented in many software packages [e.g., SerialEM (60), Latitude (GATAN Inc.), Leginon (61), EPU (Thermo Fisher)] has enabled cryo-EM data collection to become much more streamlined and user friendly. A variety of software packages are available to analyze the data (2, 62–64), empowering expert and nonexpert users alike to convert the collected data into useful 3D structures. The democratization process is being further accelerated with the growth of regional and national cryo-EM facilities, which provide users access to state-of-the-art instrumentation and, in some cases, access to on-the-fly preprocessing routines that incorporate both motion correction and contrast transfer function estimation while data are being collected (65).

3. FRONTIER APPLICATIONS

3.1. G Protein–Coupled Receptor–G Protein Complexes

G protein–coupled receptors (GPCRs) mediate many responses of cells to external stimuli. Upon activation by a ligand, the receptor interacts with heterotrimeric G proteins and/or arrestins, initiating downstream signaling (66). Insights into the allosteric coupling between ligand binding and G protein or arrestin interaction are arising from a limited number of GPCR structures in inactive and active conformations, as well as from spectroscopic and computational analyses. However, understanding of the high-resolution structural changes during GPCR activation, signaling, and regulation is necessary to conceptualize the complex pharmacology of GPCRs.

GPCR–heterotrimeric G protein complexes have been exceedingly difficult targets for 3D crystallization and structure determination by X-ray crystallography owing to the intrinsic conformational flexibility and instability of the complex. The first active-state crystal structure of a GPCR, solved in 2008, was that of opsin complexed with GαCT, a synthetic peptide derived from the C terminus of the transducin α subunit (67). This was followed in 2011 by the structure of the β₂-adrenergic receptor in complex with the heterotrimeric G protein Gs, composed of Gαs, β₁, and γ₂ subunits (68), reflecting a heroic effort in methods development. The stability of the receptor complex was enhanced by use of the then newly available highly stabilizing lauryl maltose neopentyl glycol (LMNG) detergent (69). A modified lipid for lipidic cubic-phase crystallization (68, 70) was designed to accommodate the large hydrophilic components of the β₂-adrenergic receptor–Gs assembly. Lastly, use of a nanobody (Nb35) prevented dissociation of the complex and further facilitated crystallogenesis (68). In 2016, the third crystal structure, that of the adenosine A_2a receptor in complex with an engineered minimal G protein (mini-Gs) composed of a single domain from the Gαs subunit, was published (71), signifying years of determination to engineer a minimal G protein to facilitate crystallization of GPCRs in their active conformation (72). Despite the availability of an array of tools, X-ray structure determination of GPCR–G protein complexes remains intractable, as the growth of well-ordered, 3D crystals continues to present a major obstacle for many membrane protein targets.

Single-particle cryo-EM does not require the formation of crystals for structure determination. Despite the challenges in biochemistry resulting from the intrinsic conformational heterogeneity, there is a growing list of GPCRs (Table 2) whose structures have now been analyzed using cryo-EM, starting with the 150-kDa complex of the calcitonin receptor bound to heterotrimeric Gs (73). As of late 2018, eight additional cryo-EM structures of GPCR–G protein complexes have been reported—the glucagon-like peptide 1 receptor–Gs complex (74, 75), the adenosine A_2a receptor coupled to an engineered heterotrimeric G protein (76), the μ opioid receptor bound to Gi (77), the adenosine A₁ receptor–Gi complex (78), the rhodopsin–Gi complex (79), and the serotonin 5HT1B receptor coupled to Go (80)—all in an astonishing span of only 2 years. Most recently, the structure of the human calcitonin gene–related peptide (CGRP) receptor, a heterodimer of calcitonin receptor–like receptor and receptor activity–modifying protein 1 in complex with Gs, has been published (81).

Table 2

GPCR–G protein structures determined by single-particle cryo-EM

As discussed in Section 2.2, successful structure determination by cryo-EM requires specimen preparation procedures that can withstand potential denaturation from the interfacial forces at the air–water interface during grid preparation. Furthermore, inherent flexibility between the complex components must be minimized to obtain stable 3D classes with detailed features for all subunits. The GPCR–G protein complexes analyzed by cryo-EM to date were prepared under highly stabilizing buffer conditions, using the LMNG detergent or digitonin, with the inclusion of lipid in some cases (Table 2). As a result, the complexes were robust enough to yield good-quality data. Even so, in all cases to date, only a small fraction of the molecular projection images was utilized to obtain the final density map, potentially because a significant proportion of the protein complexes were denatured at the air–water interface (74–80).

The Volta Phase Plate (VPP) (58) has been used in approximately half of the reported GPCR structures (73, 74, 76, 78, 80, 81), and a comparison of the structure with and without use of the phase plate has been reported for the A_2a receptor–Gs complex (76). While the presence of the VPP resulted in better resolution in combination with a given detector, the improvement was found to be sample dependent, and more comparisons will be required to understand the variability in resolution between samples (76).

All receptor–G protein complexes, except A₁ receptor–Gi (78) and 5HT1B receptor–Go (80), were prepared in the presence of antibody fragments. For complexes with the Gs-coupled receptors for calcitonin (73), CGRP (81), glucagon-like peptide 1 (74, 75), and 5′-N-ethylcarboxamidoadenosine (A_2a receptor) (76), the nanobody Nb35 was utilized; Nb35 binds at the Gαs–Gβ interface, stabilizing the respective GPCR complex in detergent solution. Nb35 was first used in the crystal structure determination of the β₂-adrenergic receptor with Gs (68). The μ opioid receptor–Gi complex (77) was stabilized by a single-chain variable fragment (scFv16) that binds to heterotrimeric Gi. In contrast to Nb35, scFv16 binds an epitope containing the αN helix of Gαi and the β propeller of Gβ, more than 20 Å away from the receptor–Gi interface. Subtraction of the scFv16 signal from raw particle images improved the map to an overall global resolution of 3.5 Å (77). The rhodopsin–Gi complex was prepared in the presence of Fab_50, binding far from the rhodopsin–Gi interface (Figure 1) (79). The map at a nominal global resolution of 4.5 Å showed well-defined density for the rhodopsin seven-transmembrane domain, the Gαi Ras-like domain, and the Gβ and Gγ subunits. The position of the α-helical domain (AHD) of Gαi was well defined owing to the direct stabilization of this domain by the Fab fragment, which also interacted with Gβγ (79).

Gs-bound receptor structures provided insight into the general activation mechanism of Gα proteins. Binding of an activated receptor induces displacement of the α5 helix and conformational changes within the Ras domain of the G protein. The rearrangement of the loop between β6 and the α5 helix of Gα, away from the nucleotide binding pocket, and changes in the β1–α5 loop (P-loop) weaken GDP binding. Because GDP helps stabilize the closed-domain conformation of Gα, the release of GDP from Gα allows a marked displacement of the AHD from the Gα Ras domain. The comparison between the Gs-coupled and Gi/o-coupled receptor structures showed how GPCRs selectively activate a particular family of G proteins (Figure 1). For example, there are no interactions between the receptors and the Gβ subunits in the Gi/o-bound receptor structures. Furthermore, the positioning of the Gi/o α5 helices differs from that of the respective helices in the GPCR–Gs complexes, and transmembrane domain TM6 is displaced outward from the receptor core to a lesser extent than in the Gs-occupied receptors. The smaller displacement of TM6 in Gi-coupled GPCRs may preclude binding of Gs, possibly explaining how GPCRs can bind selectively to Gi/o proteins.

To date, observations in GPCR–G protein complexes have shown that sample quality and especially stability determine success or failure of the structure determination, not only in X-ray crystallography but also (perhaps more so) in cryo-EM. Thus, the development of approaches, applicable to all GPCRs, for stabilization of the highly dynamic complexes will be essential to understanding the plethora of ligand-specific signaling responses.

3.2. Spliceosome

Another prominent example of single-particle cryo-EM comes from structural studies of the precursor messenger RNA (mRNA) splicing machinery: the spliceosome. For more than 30 years, the catalytic mechanism and intricate structural choreography of the spliceosome have been under intense scrutiny (82). The problem of mRNA splicing captivates scientists not only because of its fundamental aspects but also because unlocking its mysteries offers a plethora of opportunities for rational interventions in pathological conditions. Multiple efforts have resulted in structural resolution by X-ray crystallography, which, albeit significant, represents snapshots of only relatively small, individual components (83, 84). It is now clear that the sheer complexity of the machinery could not have been, and perhaps cannot be, captured by crystallographic approaches. This is due to both the large size and complexity of the target.

Although the size of the spliceosome is comparable to that of the ribosome, there are clear distinctions between these two supramolecular particles that explain why the latter could have been initially studied in great detail with X-ray crystallography and why a similar approach was inapplicable for the former. The ribosome represents a stable cellular machine in which well-defined RNA scaffolds, organized into two subunits, are arrayed with proteins that play an important role but do not participate directly in catalytic events. Also, aside from conformational movements of stalk segments and subunit rotational movements during protein synthesis, ribosomal RNA (rRNA) is a relatively stable structure (85). In contrast, the spliceosome is a highly dynamic multimegadalton ribonucleoprotein particle that undergoes changes in both stoichiometry and composition during mRNA splicing. Spliceosome proteins are critical not only for the structure of specific spliceosome intermediates but also for actively engaging in the splicing process, in contrast to ribosomal proteins (86). Moreover, spliceosomal RNA is not as ordered as rRNA; it is rather flexible. In fact, it is this flexibility that enables successful mRNA splicing because only in the presence of such an RNA can various intermediates along the catalytic cycle be attained. The inherent transient nature of the spliceosome was the major hurdle for classical X-ray crystallographic studies that relied on large quantities of a well-defined, homogeneous sample. However, in recent years several cryo-EM studies have been completed on the yeast spliceosome, and these have unearthed a wealth of information.

The yeast spliceosome is composed of five ribonucleoprotein complexes [i.e., the U1, U2, U4, U5, and U6 small nuclear ribonucleoproteins (snRNPs)] and a variety of proteins that are able to recognize and assemble around specific intron sequences of precursor mRNA (86). A series of highly organized and regulated steps (Figure 2) starts with the recognition of a 5′ splice site and a branching point by U1 and U2 snRNPs. This is followed by binding of preformed triple U4/U6.U5 snRNPs, which form precatalytic complex B. Activation of complex B (B*) occurs with a significant rearrangement of constituents; U1 and U4 snRNPs dissociate concurrently with the recruitment of NTC (nineteen protein complex) and NTR (nineteen-related complex). The product of the step I reaction is a phosphodiester bond between the branch point adenosine 2′-OH and the 5′ splice site (complex C). Further steps involve additional rearrangements of the spliceosome to align exons (complex C*) for the step II reaction. The final product, spliced mRNA, is released, and intron lariat spliceosome (ILS) is dissembled and reused for the next splicing cycle (87).

Visualization of this process was enabled by cryo-EM methods. Starting with the purified yeast spliceosome, Yan et al. (88) utilized computational advances in cryo-EM to perform in silico purification. This approach revealed structures of three different spliceosome stages: precatalytic complex B, complex C*, and ILS. In addition to the yeast spliceosome, significant efforts have been made to illuminate the analogous process in humans, with sample preparation aided by the use of mild chemically induced protein cross-linking (89–91).

3.3. Fibrils

Another group of macromolecules that attract significant scientific interest consists of proteins that form fibrils (Figure 3). These particular proteins are considered the principal causative agents in many neurodegenerative diseases but are poorly understood at the structural and functional levels (92–94).

It is thought that proteins synthesized in a cell typically adopt the most thermodynamically stable conformation. Native proteins can, however, undergo an abnormal conformational transition, wherein α helices are converted to β sheets or to partially ordered states. Misfolded, and presumably toxic, proteins serve as a substrate for rapid nucleated growth of linear aggregates or amyloid fibrils (95). These fibrils are insoluble and resistant to degradation, proteolysis, and biological clearance. Amyloid deposits of serum amyloid A, light-chain immunoglobulin G, and transthyretin have been strongly implicated in pathologies of various organ systems (reviewed in Reference 96). In addition, it is thought that accumulation of fibrils of β-amyloid, α-synuclein, tau, and prion protein peptide either directly causes neuronal death or is a manifestation of a pathological condition. Thus, a better understanding of fibril structure and the mechanisms of fibril formation may provide new avenues for therapies that target both the development and progression of neurodegeneration of the human brain.

Structural studies of fibrils have been hindered because of their variable nature. Although fibrils exhibit locally ordered stacks of protofilaments, minor changes such as twists, bends, and/or axial flexibilities can result in different fibril morphology at the global level (97). These structural departures hamper the ability to obtain pure sample, and even if one is obtained, it is almost certain that only a fraction of the entire fibril will be visualized. Moreover, most recent findings and observations suggest that fibrils exhibit significant structural polymorphisms both in vitro and in patient- and animal-derived samples. The intrasample polymorphism is not obviously dependent on amino acid sequence or other features, which brings the complexity of fibril studies to another level. Different polymorphs may cause different aggregation kinetics, oligomerization dynamics, and disease phenotypes, which may presumably complicate the development of therapies. Recently, mathematical analyses have been employed to better understand and perhaps predict fibril polymorphisms (98).

Because of improvements in sample preparation and structural methods, a number of different fibril structures have recently been determined using cryo-EM helical image processing. A cryo-EM study of patient-derived tau fibrils, the major hallmark of tauopathies (Alzheimer disease in this case), revealed structural differences in the core of ultrastructural polymorphs, paired-helical and straight tau fibrils, derived from two identical protofilaments (99). This finding indicates that different tau fibril conformers might be connected to different, specific neuropathies within the sample derived from the brain tissue of the same patient. Similar cryo-EM studies were done on a patient-derived sample from an individual afflicted with Pick's disease (100). In this case, additional tau fibrils, narrow and wide Pick's fibrils, were described at near-atomic resolution. The narrow tau fibrils are composed of one protofilament and are conspicuously different from tau fibrils found in Alzheimer disease. Another fibril structure comes from a study completed on almost-full-length α-synuclein fibrils, which are linked to Parkinson disease. Given the observed protofilament interactions, the authors of that study proposed an underlying mechanism for α-synuclein fibril elongation (101).

It is not clear whether the presence and formation of fibril structures are the cause or the product of neurodegenerative diseases. However, it is clear that cryo-EM methods could dramatically improve our understanding of these mysterious structures and perhaps facilitate the development of novel therapies.

4. CRYO ELECTRON TOMOGRAPHY AND SUBTOMOGRAM AVERAGING

Electron tomography at cryogenic temperatures (cryo-ET) is an approach to structure determination of biological entities such as whole cells and viruses that are unique and do not lend themselves to averaging using single-particle cryo-EM methods. An important use of tomography is in combination with subvolume averaging, which allows for 3D averaging of relative homogeneous components within viruses and cells, providing a powerful opportunity to study proteins in their native physiological environment (3, 102–104). Thus, cryo-ET provides a route to macromolecular structures without the need to first express and purify the components (4). This section provides only a brief overview of cryo-ET and subtomogram averaging; we refer the reader to recent reviews and references therein for more details (3, 4, 103, 105).

Methods for imaging of whole viruses and cells typically use the same protocol for plunge-freezing as for single-particle specimens, although the resulting specimens tend to be considerably thicker because the larger dimensions of the viruses or cells dictate this thickness. Thinning the samples with focused ion beams both offers a way to control the specimen thickness and presents a new approach to the examination of structural aspects of the cell interior, such as the nucleus (4, 106, 107).

Methods of tomographic data collection involve rotation of the vitrified specimen to a limited angular range (typically ±70°), and the resulting projection images are combined computationally to obtain a tomogram of the region of interest. A variety of schemes are used to obtain this series of tilted images, ranging from a continuous progression from negative to positive tilts to dose-symmetric tilt series, where data collection starts at zero and progresses to increasingly high negative and positive tilt angles (108–110).

Through the use of specialized tilt schemes and methods, the high-resolution structure of the human immunodeficiency virus 1 (HIV-1) CA–SP1 (capsid–spacer peptide 1) has recently been reported (109, 111). Immature HIV-1 buds from the plasma membrane before proteolytic cleavage of the viral Gag polyprotein induces maturation, which can be blocked by maturation inhibitors, abolishing virus infectivity. This study used plunge-frozen immature virus–like particles, assembled in vitro from the purified CA and SP1 region of the Gag polyprotein, in the presence (3.9-Å resolution) or absence (4.5-Å resolution) of maturation inhibitors; as well as an immature, protease-defective HIV-1 particle (4.2-Å resolution). An atomic model, based on the 3.9-Å-resolution map, provided the template to catalog mutations, deletions, and polymorphisms that confer resistance to maturation inhibitors, showing that the sites of mutations do not coincide with potential drug-binding pockets. This finding implies that maturation inhibitors do not sterically block the proteolysis site, but rather stabilize the immature Gag protein. As proteolytic cleavage requires unfolding of Gag, HIV-1 seems to develop maturation inhibitor resistance by destabilizing its immature form (109). A concurrent X-ray crystallographic study reached similar conclusions (112).

Subnanometer resolution structures have been reported for other viral and bacterial objects. Bharat et al. (108) examined the S-layer of the gram-negative bacterium Caulobacter crescentus by cryo-ET and subtomogram averaging of cell stalks, and compared the resulting 7.4-Å structure with the X-ray structure of the purified S-layer protein RsaA. Wan et al. (110) reported the structure of the Ebola virus nucleocapsid within intact viruses and recombinant nucleocapsid-like assemblies, revealing the identity and arrangement of the nucleocapsid components. This study allowed the authors to draw conclusions concerning nucleoprotein oligomerization and nucleocapsid condensation of mononegaviruses. Further examples of recent subnanometer-resolution structures include the COPI coat budded in vitro from giant unilamellar vesicles using purified coat protein components (COPI proteins mediate trafficking within the Golgi compartment) (113) and an endoplasmic reticulum translocon-associated ribosome protein complex, observed in rough microsomal membranes or purified in the detergent digitonin (114).

While most of the examples discussed above employed in vitro reconstituted specimen, recent structures from within whole cells offer the prospect of visualizing biological molecules in their native environment in unprecedented detail. Bykov et al. (115) applied focused ion beam milling, cryo-ET, and subtomogram averaging to determine the native structure of the COPI coat protein within the Golgi stacks of Chlamydomonas reinhardtii. The native COPI structure resembled that of the in vitro–generated model (113), but additionally uncovered bound cargo. Notably, the unicellular green alga C. reinhardtii is a genetically tractable model organism, with reproducible Golgi architecture that allows comparative analysis across multiple cells (115). Vitrified Chlamydomonas cells have also been used to visualize the algal translocon-associated protein complex as an integral component of the translocon, assisting the Sec61 protein–conducting channel in transport across or insertion into the endoplasmic reticulum of proteins (116). In addition, two classes of proteasomes tethered to two specific nuclear core complex locations were observed through a combination of subtomogram averaging and nanometer-precision localization (117). Lastly, 3D snapshots into HeLa cells provided insight into the native structure and organization of the cytoplasmic translation machinery, the nuclear core complex, and lamina (118).

5. FUTURE PERSPECTIVE

Given the rapid technological advances in automated high-resolution data collection, processing, and 3D reconstruction, it is reasonable to expect that cryo-EM methods will be suitable for structure determination of ever smaller and asymmetric macromolecules. The methodology is already being tested for screening of binding of drug targets (10, 119, 120), an approach that has historically been reserved for X-ray crystallography and, to a lesser extent, for NMR. Moreover, cryo-EM will likely find its utility in studying large, transient, multisubunit and multicomponent systems that are vital for regulation of cellular processes. Preparation, optimization, and stabilization of a sample will be a major aspect of the next wave of advances in cryo-EM and will drive further advances in structural biology. We can also anticipate substantial improvements in the use and application of cryo-ET to solve important problems in cell biology. More widespread access to rapid data collection, which is becoming increasingly available at regional and national facilities, will likely also be a major catalyst in the adoption of cryo-EM and cryo-ET methods as mainstream tools in biology.

disclosure statement

The authors are not aware of any affiliations, memberships, funding, or financial holdings that might be perceived as affecting the objectivity of this review.

acknowledgments

The writing of this review was supported by the Intramural Research Program of the National Institutes of Health, National Cancer Institute.