Advances in cryo-electron microscopy (cryoEM) for structure-based drug discovery

Abstract

Introduction:

Macromolecular X-ray crystallography (XRC), nuclear magnetic resonance (NMR), and cryo-electron microscopy (cryoEM) are the primary techniques for determining atomic-level, three-dimensional structures of macromolecules essential for drug discovery. With advancements in artificial intelligence (AI) and cryoEM, the Protein Data Bank (PDB) is solidifying its role as a key resource for 3D macromolecular structures. These developments underscore the growing need for enhanced quality metrics and robust validation standards for experimental structures.

Areas covered:

This review examines recent advancements in cryoEM for drug discovery, analyzing structure quality metrics, resolution improvements, metal-ligand and water molecule identification, and refinement software. It compares cryoEM with other techniques like XRC and NMR, emphasizing the global expansion of cryoEM facilities and its increasing significance in drug discovery.

Expert opinion:

CryoEM is revolutionizing structural biology and drug discovery, particularly for large, complex structures in induced proximity and antibody-antigen interactions. It supports vaccine design, CAR T-cell optimization, gene editing, and gene therapy. Combined with AI, cryoEM enhances particle identification and 3D structure determination. With recent breakthroughs, cryoEM is emerging as a crucial tool in drug discovery, driving the development of new, effective therapies.

1. Introduction

Structural biology, a cornerstone of drug discovery, offers profound insights into the three-dimensional structures of biological macromolecules, including proteins, nucleic acids, and even viruses. By unveiling the precise shape, active sites, and interaction surfaces of target molecules at the atomic level, scientists understand how small molecule agents interact with macromolecules of interest, leading to the design of more potent drugs. Techniques such as X-ray crystallography (XRC), nuclear magnetic resonance spectroscopy (NMR), and cryo-electron microscopy (cryoEM) are instrumental in visualizing these structures, enabling the identification of binding sites for potential drug molecules and the study of the competition of various small molecule agents for active sites. Such structural information guides the refinement of lead compounds enhancing their affinity, specificity, and efficacy while minimizing off-target effects, thereby expediting the drug development process.

XRC, NMR, and now cryoEM are the most widely used experimental techniques for drug design. Table I summarizes their advantages and limitations. XRC still plays a crucial role in drug discovery by providing high-resolution, three-dimensional structures of biological molecules. Structural information from diffraction experiments used to be the main approach for understanding how these molecules function and interact with potential drug candidates. Recently, cryoEM became a method of choice for studies of large and very large macromolecules and their complexes. It may be counterintuitive for non-structural biologists that the high-quality analysis of cryoEM data for large macromolecules is fast and relatively more straightforward than for small molecules. An example is provided by investigators working on structural studies of ribosomes (Ada Yonath, Venki Ramakrishnan, and Nenand Ban, among others) who abandoned XRC and moved entirely to cryoEM. While many cryoEM structures today still fall within the lower resolution range and smaller molecules remain technically challenging, the technology is advancing rapidly. With ongoing improvements in hardware, software, and innovative methods such as using larger protein scaffolds (1,2), even smaller proteins are becoming more accessible for high-resolution studies.

Table 1.

CryoEM structures used in drug design (full list is available in Table S1) with unusually high differences between the structure quality measures: CC_MASK and Q-score.

PDB ID	Release Date	Length	Resolution [Å]	CCMASK	Average Q-score
8WLL	2024-06-12	389	3.74	0.76	0.353
8BVT	2023-07-19	623	3.94	0.78	0.364
7Y1K	2023-06-14	1164	3.80	0.82	0.413
7Y1L	2023-06-14	1189	3.73	0.85	0.424
7EJA	2022-04-13	1097	3.60	0.80	0.378

CryoEM has experienced significant advances in recent years due to the development of high-speed direct electron detectors, new sample preparation methods and, equally important, improved algorithms for image processing and reconstruction. Machine learning approaches have been revolutionary for enhancing the speed and accuracy of automated high-throughput grid screening (Smart Leginon: (3)), micrograph processing and particle picking (TOPAZ: (4), CrYOLO: (5)), 3D reconstruction (CryoDRGN: (6), Blush: (7)), automatic post-processing of cryoEM maps (deepEMhancer: (8)), model building (ModelAngelo: (9)), and recently addressing continuous conformational heterogeneity (3DFlex: (10)). New sample preparation and handling techniques, such as lipid nanodiscs that preserve the native structure of biomolecules in solution and reduce damage from ice formation during freezing, as well as the use of various sample freezing apparatus that improves reproducibility of grid preparation (11-13).

The advancements mentioned above have significantly increased the number of annual depositions of cryoEM structures to the Protein Data Bank (PDB) (14-16) and Electron Microscopy Data Bank (EMDB) (17-19). This growth is particularly significant in the context of the increasing demand for PDB downloads, which is growing even faster (20), far exceeding the rate at which new structures are being experimentally determined. Figure 1 presents the number of structures determined by XRC, NMR, and cryoEM and deposited each year in the PDB. The number of structures determined by cryoEM is growing rapidly, while the deposits of structures determined by XRC have stabilized, and by NMR is decreasing. However, what grows extremely fast are not depositions but the number of downloads from the PDB, which is probably the world's most frequently accessed scientific resource. The vertical scale represents the number of deposits per year or, alternatively, the average number of downloads every two minutes. As shown by the rapid growth of the number of such deposits, the importance of cryoEM in meeting this demand and its potential impact on many areas of research, including drug discovery, cannot be overstated.

Figure 1.

The annual number of new deposits in the PDB, categorized by the method of structure determination: XRC (blue), NMR (green), and cryoEM (red). The solid line represents the average number of deposits downloads every two minutes.

The trends in PDB download data reveal a steady increase from 2009 to 2019, followed by a significant surge from 2019 onward, suggesting an underlying shift in usage patterns. The initial period likely reflects a time when downloads were more targeted, with researchers manually selecting relevant structures, resulting in lower volumes. The sharp increase post-2019 aligns with the emergence of the "AI era" in structural biology, marked by widespread adoption of computational tools like AlphaFold2 (21), which encouraged many research groups to download the entire PDB for local use to facilitate large-scale analysis and prediction. Regular updates to maintain current data further contribute to this growth. This period also highlights the interplay between AI and cryoEM; while AI-driven models benefit from structural diversity to improve predictive capabilities, cryoEM supplies unique structures to enhance the training datasets. The continuous development of AI tools, such as AlphaFold3 (22) with ligand docking features, supports hypothesis generation in drug discovery, where cryoEM offers a means for experimental validation. This synergy underscores the co-evolution of AI and cryoEM, reinforcing their collective impact on advancing structural biology and drug discovery.

In drug discovery, examining macromolecules (typically proteins) in complex with small-molecule ligands, such as inhibitors or active pharmaceutical compounds, is essential for understanding ligand-macromolecule interactions. This approach enables rational modifications to ligand structures to improve binding strength and specificity. Additionally, water molecules and metal ions often play significant supporting roles, as they can reveal alternative binding sites crucial for drug design. The success of these studies depends on access to high-resolution, high-quality structural models; however, such models represent only a subset of all entries in the PDB. While quality assessment metrics for atomic models derived from X-ray crystallography (XRC) and nuclear magnetic resonance (NMR) are well-established, the evaluation methods for cryo-electron microscopy (cryoEM) structures are still maturing. The following section explores the ongoing development and application of quality metrics for cryoEM structures in greater detail.

2. Quality of cryoEM structures

Every structural model requires refinement and specific metrics to assess its alignment with experimental data, such as the electron density map in XRC and Coulomb potential map in cryoEM. Non-specialists may not always recognize that the map represents the experimental outcome, while the structural model presents the experiment's interpretation. In 2020 a new quality metric for cryoEM structures, the Q-score, has been proposed by Pintilie (23). The Q-score measures how similar the map profile of an atom is to a Gaussian-like function that would be seen if the atom were well resolved. The validation reports available in PDB contain Q-scores averaged for each chain of a given structure. This score represents an "average resolvability measure" (23). Since September 2022 all validation reports for cryoEM structures include Q-scores (24), as well as another metric called Atom Inclusions. The latter measures the fraction of atoms in the model that reside inside the EM volume (25). The dependence of Q-score on resolution for all cryoEM structures is shown in Figure 2. It should be noted that there are structures with negative Q-scores, even some of those with claimed resolution better than 3.0 Å. A negative Q-score of structures determined with cryoEM suggests that the atomic density in the Coulomb potential map deviates significantly from a well-resolved Gaussian-like peak. This typically means that the atom or region being measured is poorly resolved or does not fit the map well. Possible reasons for a negative Q-score include low map resolution, inaccurate atomic placement, noise or artifacts in the data, and regions of the structure that are flexible or disordered.

Figure 2.

Q-score and CC_MASK values plotted against resolution for all cryoEM structures in the PDB with resolution better than 5 Å. Q-scores for several high-resolution structures (represented as blue dots), not originally reported, were calculated by the authors of this manuscript. Negative and positive values of quality metrics for individual structures are shown as, respectively, blue and green dots. An interactive version of this plot can be explored at https://bioreproducibility.org/figures/cryoem_sbdd/fig2 . The CC_MASK data is limited to structures cataloged by the CERES service(26), accessible at (27).

Evaluating the quality of cryoEM structures has sparked debate over the most effective metrics. While atom inclusions and Q-scores have been commonly used, some argue that they fall short as comprehensive quality measures. The Phenix group has advocated for the model-to-map correlation coefficient (CC_MASK) as a superior alternative, particularly for high-resolution structures (28-30). Unlike atom inclusions or Q-scores, CC_MASK considers essential atomic model parameters such as atomic displacement parameters (ADPs), occupancy, and atom type. Moreover, atom inclusions and Q-scores do not address density shape, missing atoms, or alternative conformations. Historically, identifying multiple conformations in cryoEM structures was challenging due to the difficulty of obtaining maps with both sufficient resolution and the ability to capture fine details. However, recent advances in cryoEM technology appear to have resolved these issues. For example, the atomic-level resolution (1.09 Å) structure 8RQB clearly illustrates double conformations within a significant region of the map (Figure S1). These improvements suggest that more nuanced quality metrics, such as CC_MASK, may offer a more comprehensive assessment of cryoEM structure quality in the future.

In 2021 the Phenix group developed an online re-refinement system for cryoEM named CERES (26). This system, which is regularly updated, automates the re-refinement of structures deposited in the PDB. In addition to re-refinement, CERES computes the CC_MASK for both the original and re-refined cryoEM structures; however, this information is not available for all cryoEM structures in the PDB. As of November 15, 2024, we successfully retrieved data for 15,275 of the 23,212 cryoEM structures listed in the PDB. A comparison of two structure quality measures, presented in Figure 2, demonstrates that CC_MASK appears to be less resolution-dependent than the Q-score. This observation is consistent with the findings of Liebschner et al. (26), who note that CC_MASK behaves differently from the R-factors used in X-ray crystallography quality assessments, which are known to be strongly correlated with resolution (31).

To test the hypothesis proposed by the Phenix group regarding the potential superiority of CC_MASK over the Q-score, we analyzed the relationship between these two measures, as shown in Figure 3. Our analysis reveals that, particularly for newer and higher-resolution structures (represented by blue dots), CC_MASK tends to exhibit more consistent values, whereas the Q-score shows greater variability. However, both measures present interesting outliers. Notable examples include:

Figure 3.

Comparison of Q-score and CC_MASK across different resolutions. The "new and high-resolution" category (blue dots) comprises structures with a resolution of at least 2.5 Å, released by the PDB after January 1, 2022. Red dots represent structures with negative values for either CC_MASK or Q-score. All remaining structures are represented as green dots. For detailed identification of each structure, an interactive version of this plot is available at https://bioreproducibility.org/figures/cryoem_sbdd/fig3.

1. PDB ID: 6RBE: This structure was derived from five different electron density maps, all of which have low Q-scores: EMD-4792 (0.16), EMD-4793 (0.1465), EMD-4794 (0.0285), EMD-4795 (0.032), and EMD-4796 (−0.0005). The structure’s average Q-score is exceptionally low at 0.019. However, surprisingly, the CC_MASK values provided by CERES show a significantly higher value of 0.8 for the EMD-4793 map, while the other two maps (EMD-4794 and EMD-4795) have low CC_MASK values of 0.02 and 0.09, respectively.

2. PDB ID: 7YKS: For this structure, both the EMD-33896 map and the structure itself display a decent Q-score of 0.545. CERES reports a CC_MASK value of only 0.11 for the deposited structure. However, following re-refinement in CERES, the CC_MASK value improves significantly to 0.8.

Additional cases of discrepancies between the Q-score and CC_MASK are summarized in Table 1. This table also highlights examples where the structures, according to the PDB, serve as the structural basis for already approved drugs.

2.1. Resolution of cryoEM maps

While resolution is not a parameter directly addressed in structures obtained by NMR, the concept of resolution is inherently relevant to XRC and cryoEM. Both latter methods use resolution to quantify the level of detail, but the specific metrics used differ, making a direct comparison difficult (32,33).

In crystallography, resolution is primarily determined by the maximum diffraction angles of the X-rays scattered by the crystal that still provide measurable data. Several factors affect resolution in XRC, including the wavelength of the X-rays used for the experiment, crystal quality, both inherent and affected by the conditions of data collection, and computational methods used for processing diffraction data. The expected resolution limits can be roughly estimated by looking at the first diffraction image, so that weakly diffracting crystals could be bypassed in full data collection.

In cryoEM, resolution is primarily determined by the quality of the electron micrographs, the number of particles analyzed, and the intrinsic heterogeneity of the sample (34). In this case, the resolution depends on the quality of the images (determined by the quality of the electron source, detector, and microscope optics), quality and features of the particles (including quality of sample preparation and particle orientation in the sample), as well as the quality of data processing (including not only the quality of image processing, but also the particle picking and classification, which can be significantly affected by the lack of experience). Fourier Shell Correlation (FSC) is usually used to assess the overall resolution of a cryoEM structure. The 'gold standard' method (35), now the most prevalent in the EM community, is often used to determine the resolution of structures deposited in the PDB or EMDB. Similarly to how resolution is presented in XRC, global FSC provides a single value representing the average resolution across the entire structure. Local FSC can also be calculated to assess the resolution variation within a structure, providing more realistic information about the local quality of the structure. Various algorithms have been developed for local validation, using FSC, Q-score, or a combination of both (36,37).

Until about 15 years ago, in the “blobology” era (38), only low-resolution structures obtained by electron microscopy, primarily using negatively stained samples, were available in the PDB. All that changed with the introduction of modern direct electron-counting detectors and new image captions and processing software. We present the rapid increase in the resolution of cryoEM deposits in PDB in Figure 4. The current highest-resolution structure is that of mouse heavy-chain apoferritin (PDB ID 8RQB). It was resolved at an impressive 1.09 Å resolution. However, the paper describing this structure and the process of obtaining it (39) had to wait 6 months for publication (we are indeed slower than we were during the Gutenberg times). This remarkable achievement highlights the rapid advancements in cryoEM and makes it difficult to predict how far the technology will advance in the coming years (40). However, it is important to note that a recent study (41) highlighted a potential trade-off in pushing the limits of resolution in cryoEM. In particular, authors mention the three main resolution limiting problems: higher-order lens aberration, chromatic aberration, and axial coma. The latter one is especially important since in the case of images affected by coma the FSC will show an increase in correlation which may lead to misestimations of the resolution (42). Char and Stark (41) suggest that solvent densities may serve as more reliable indicators of map quality than the protein regions themselves and demonstrate that even among three high-resolution (1.5 Å or better) apoferritin structures, the number of resolved water molecules varies by a factor of three. The authors conclude that the number of resolved solvent molecules depends not only on the microscope but also on the software used in the processing of the structure. These results demonstrate the challenges in achieving overall model quality at extremely high resolution, suggesting that a balance must be maintained between resolution and the accuracy of the structural model.

Figure 4.

Resolution (≥ 5 Å) of cryoEM structures as reported by their authors, plotted against release date. The boxes represent the interquartile range (50% of the data), with whiskers indicating the range of 1.5 interquartile ranges from the first and third quartiles. An interactive version of this figure, which identifies outliers, is available online at: https://bioreproducibility.org/figures/cryoem_sbdd/fig4.

2.2. Metal ligands

Metal ions play a significant role in drug discovery and development, contributing to both therapeutic and diagnostic applications. Their unique chemical properties that include coordination geometry, catalytic abilities, or variable oxidation states make them useful for metal-based or metal-associated drugs. Some metals, for example silver or copper, directly exhibit antibacterial properties by disrupting bacterial membranes or affecting bacterial enzymes. Metal-containing compounds can be developed to inhibit metalloenzymes in infectious diseases and in cancer. CryoEM is particularly well-suited for studying large, complex structures, such as protein-nucleic acid assemblies or membrane proteins, which often involve inclusion of metal ions. Capturing molecules in near-native state helps to preserve important interactions with metal ions. Since cryoEM is also well suited for analysis of conformational changes in proteins that affect their function, metal-induced changes can be studied.

We analyzed the proportion of structures containing the most commonly occurring metals in all macromolecular complexes present in the PDB. The results, shown in Figure 5, are categorized by the technique used for structure determination. The high fraction of structures containing magnesium and zinc is the result of using almost exclusively cryoEM for ribosome studies. However, it is not clear whether all metal ions are correctly identified. The CheckMyMetal (CMM) server (43,44), the primary metal identification resource, was developed for XRC and was recently expanded for structures determined by cryoEM (45).

Figure 5.

Fraction of PDB deposits containing metal ions, categorized by the method used for their determination (XRC in blue, cryoEM in red, and NMR in green).

The use of the CMM server for validating metal binding sites in high-resolution structures has highlighted some challenges. In an initial test, a significant number of metal ions were incorrectly identified (45). Additional caution is needed when using an X-ray crystallography (XRC) structure as a template for solving a cryoEM structure. Metals bound in the XRC structure, which may have been obtained under different conditions, might be incorrectly retained in the final cryoEM model. Another issue with metal identification arises from the cryoEM CIF file, which lacks information on sample preparation (46). Only general details about the data acquisition process are available through the PDBe API (47). For example, the 8RQB deposit, which had minimal or no refinement data in its CIF file, still contained an important iron metal site (39). Similarly, the unusually high number of zinc sites observed in NMR metal dataset studies (Figure 5, green bar) is somewhat artificial. This is primarily due to zinc ions being retained during experiments, as they play a crucial role in protein structure and function. Without zinc, the protein may misfold or become unstable, which explains why zinc ions are frequently reported in these studies.

2.3. Water molecules in cryoEM maps

All biological molecules exist in the aqueous environment and interactions with individual water molecules play an important role in stabilizing biological material. Samples used in cryoEM experiments are rapidly frozen in a thin layer of vitreous ice (non-crystalline water), preserving the native structure of biological molecules without forming damaging ice crystals. When visible, water molecules can be modeled in high-resolution cryoEM structures. Similarly to XRC, density modification algorithms are usually used in cryoEM to improve the maps (48). It has been shown that different methods will differentially affect ligand densities, including enhancing as well as removing densities for waters and selected metals, depending on the method used (49). However, it seems that many cryoEM deposits do not contain water molecules even if the corresponding map density is present and even though solvent interactions are important for understanding intermolecular interactions. A comparison between cryoEM and XRC of the number of water molecules per residue as a function of resolution (Figure 6) shows that the number of water molecules identified by cryoEM is typically lower than that identified by XRC. However, this difference is likely influenced by the fact that the concept of "resolution" is defined differently for the two techniques, which makes direct comparisons less straightforward (33). Additionally, adding water molecules to crystal structures may lead to improvement of the parameters that indicate structure quality, whereas no such effect is present in cryoEM, since in that case a more complete model does not influence the map itself.

Figure 6.

Ratio of water molecules to amino acid residues plotted against the structure resolution determined by XRC (blue) or cryoEM (red) methods. The boxes represent the interquartile range (middle 50% of the data), with whiskers extending up to 1.5 interquartile ranges from the first and third quartiles. The horizontal line inside each box marks the median. Individual dots represent outliers. An interactive version of this plot, enabling the identification of individual outliers, is available at: https://bioreproducibility.org/figures/cryoem_sbdd/fig6.

3. Refinement

Several programs, including Phenix (29), Refmac/Servalcat (50), Isolde (51), and Coot (52) or various combinations of these packages (53) are being used for structure refinement. Differences in how these programs approach the refinement process may lead to slightly different results. However, looking into the deposit, it is often difficult to find which program was used, and sometimes it is even difficult to find whether the model was refined at all. There are three possible places to find such information, namely the validation file, PDB file, and CIF file. On top of that in the CIF file this information can be found in two independent places: 1. In the “software” list as the “refinement” classification. 2. In the “em_software” list in the category “Model Refinement”. These different sources show, in fact, different results. Figure 7 shows this data according to the “EM_software” table specific for cryoEM structures in CIF files, however, different results may be obtained from the PDBe API using the “refinement_software” field or PDB files.

Figure 7.

Software used in the refinement of cryoEM structures, as reported in PDB deposits. The plot was generated by analyzing 22,646 deposits and extracting data from the CIF file using the EM_software table. An alternative analysis, combining information from the CIF files, validation reports, and older PDB files, yields slightly different results. For further details, see the interactive version at https://bioreproducibility.org/figures/cryoem_sbdd/fig7.

In many cases, neither the validation files nor the CIF and PDB files provide the required detail. Our correspondence with the authors of deposit 8RQB revealed that the lack of information about refinement was due to a bug in one of the deposition programs. This fact identifies serious limitations to the dissemination of know-how for this very fast-developing technique.

Additional details were gathered directly from the deposit authors via phone and email. We have consistently notified the PDB about various errors and have made efforts to inform them of each new class of issues that we discovered.

4. Comparison of structure determination methods

Table 2 summarizes the pros and cons of each of the structure determination methods emphasizing the differences of cryoEM with regard to X-ray and NMR based on the aspects mentioned above.

Table 2.

Summary of advantages and disadvantages of different structure determination methods in obtaining quality information for ligand complexes.

X-ray crystallography	CryoEM	NMR
Pros: Fo-Fc maps as a tool for metal ion and other ligands validation Anomalous scattering at different wavelengths for atom identification High-resolution data is frequently obtained and provides detailed information on ligand coordination geometry and distances The rapid acquisition of structures and efficient validation of experimental results facilitate high-speed fragment screening.	Pros: Sample close to native conditions Relatively easier sample preparation If high-resolution data are obtained, it provides detailed information on ligand coordination geometry and distances Can be used not only for structure determination but also for other experiments, such as tracking crystallization phenomena Difficulty of interpretation decreases with the increase in the size of protein Ability to observe hydrogen atoms (54)	Pros: NMR provides ensembles of structures, where multiple conformations satisfy the experimental data. The experimental environment is closer to physiological conditions Provides information about dynamic states of flexible parts of the structure, including the environment of ligands. Works with low affinity ligands.
Cons: Obtaining quality diffraction crystals is frequently a rate-limiting step Challenging interpretation at low resolution Crystallization may potentially alter the protein structure and ligand binding Possibility of introducing model bias in model-phased electron density	Cons: Coulomb potential maps contain information about ion charge, but currently available methods do not use them (55-57) Challenging interpretation due to usually low resolution. Ligand refinement, as well as estimation of resolution and model quality, are difficult.	Cons: The lack of explicit structure resolution makes it difficult to directly visualize atomic details. Quality of the ligand complexes is assessed through various metrics that evaluate geometric and structural consistency

5. CryoEM around the world

5.1. CryoEM in academia and national laboratories

Several synchrotron facilities have recognized that adding capabilities to conduct cryoEM experiments on-site can significantly boost scientific output by enabling researchers to work in concert with both cryoEM and XRC. Such integration allows for more efficient and complementary structural analysis. The most successful facilities are Diamond Light Source synchrotron in the United Kingdom, Stanford Synchrotron Research Laboratory (SSRL) in USA, Paul Sherer Institute in Switzerland, and Photon Factory in Japan, although other synchrotron facilities have also been going in this direction. Also, several university-built facilities are open to external users. New York Structural Biology Center, HHMI Janelia Research Campus, The National CryoEM Facility (NCEF) at the National Cancer Institute, The Max Planck Institute of Biophysics in Frankfurt, University of Texas Southwestern in Dallas, The Scripps Research Institute in La Jolla, Rutgers University, and the University of Virginia in Charlottesville are among the best known and best equipped academic facilities. Some of these centers provide computational resources, software, and expertise that allow users to leave facilities with partially refined structures. Unfortunately, it is difficult to measure the performance of these facilities as, contrary to XRC data collected at synchrotron beamlines, the use of the academic resource is not reported in the PDB deposit.

5.2. CryoEM meets the industry

Several leading pharmaceutical and biotech companies have invested in their own cryoEM facilities to advance structural biology studies, providing critical insights that guide drug design and development. Notable examples include Roche, Novartis, Pfizer, Astra Zeneca, Merck, Amgen, and Bristol-Myers Squibb. These facilities are used for commercial studies and results are rarely published or deposited.

Several companies provide full service to industry and academia. They prepare samples, perform measurements, perform 3-D reconstruction, refine the resulting structural models, and even perform partial analysis of results. The best known are NanoImaging Services (NIS) in San Diego, Structural Biology Center at WuXi in Shanghai, SPT Labtech in Melbourne and MiTeGen in Ithaca. It is interesting that while Thermo Fisher Scientific is primarily known for manufacturing cryoEM instruments, they also provide full cryoEM services. The cutting-edge research conducted in primarily manufacturing companies is beneficial for both instrument development and, presumably, the use of prototypes of the new, better system.

5.3. Utilization of cryoEM in drug discovery

The utility of cryoEM in the process of drug discovery and design has been alluded to for almost a decade (Table S1). The first publication that explicitly showed results of such studies was by Merk et al. (58). The authors presented 3.8 Å resolution cryoEM structures of the cancer target isocitrate dehydrogenase and identified the nature of conformational changes induced by binding of the allosteric small-molecule inhibitor ML309. They also pushed the resolution of the structure of lactate dehydrogenase to 2.8 Å and of glutamate dehydrogenase to 1.8 Å. It should be stressed that all these enzymes were comparatively large (93-334kDa), emphasizing the fact that cryoEM works better for larger macromolecules than for smaller ones. Much more recently Cushing et al. (59) presented a series of structures of CDK-activating kinase determined at the resolution around 2 Å with an explicit purpose of assisting in structure-based drug design. The mass of these heterotrimeric complexes is ~87 kDa, emphasizing that current cryoEM methods could be useful for these comparatively small protein complexes. However, it is not clear if much smaller proteins could provide similar type of information and, for now, XRC is still the major technique in drug design and development studies.

CryoEM has revolutionized structural studies of G protein-coupled receptors (GPCRs) and ribosomes (60-63). GPCRs are critical targets in drug discovery due to their involvement in numerous physiological processes (64,65). Application of cryoEM in GPCR research enables the determination of high-resolution structures, particularly for challenging targets, such as large protein complexes and membrane proteins, which are very difficult to crystallize. CryoEM has allowed for insights into the active and inactive states of GPCRs, their conformational changes, and their interactions with ligands (66,67). The research is conducted entirely within public/private companies and sometimes conducted collaboratively between private companies and academic institutions. These collaborations frequently result in publications with a large number of co-authors, some of whom hold multiple affiliations. For example, two GPCR-related papers, published in joint efforts by the iHuman Institute (ShanghaiTech University), Structure Therapeutics (South San Francisco), and several other organizations (68,69), featured 12 and 19 co-authors, respectively, representing 10 and 9 institutions. Several authors listed multiple affiliations. These examples highlight the extensive collaboration behind such work but also raise questions about whether every contributor fully grasps the broader context of the research—the proverbial "forest"—rather than focusing solely on their specific "trees."

6. Challenges and the future directions

CryoEM is a relatively new, rapidly evolving technique that holds immense promise. One might assume that with the availability of structures boasting resolutions better than 1.25 Å (although it must be noticed that 22 out of the 40 cryoEM structures with resolution better than 1.8 Å represent just 2 proteins: Human (P02794) and Mouse Ferritin (P09528)), the major challenges have been overcome, and the development of life-improving drugs is just around the corner. However, the real challenge lies in the dissemination of know-how. Surprisingly, even at similar resolution, the quality indicators of structures—as measured by the Q-score—vary significantly, even for identical samples (7VZ3 Q-score -0.015; 7VYV Q-score 0.684; 7LZJ Q-score 0.632). Despite numerous courses and workshops, we are still left questioning why, for example, three highly similar structures could exhibit such different Q-scores. We suspect that the low score for 7VZ3 is due to the misalignment between the map and the coordinates not addressed during deposition, and not to any intrinsic problems of the structure itself. The PDB has now adopted the Q-score, and regardless of how we feel about it, biologists will inevitably use it as a benchmark for selecting structures. We wish the authors would do the same.

CryoEM is a very powerful technique, but it still faces several challenges. The most critical is probably sample preparation which may ultimately strongly affect map reconstruction. The cryoEM sample does not need to be crystallized, but sample preparation is still not trivial (70)). Samples must be flash-frozen in a thin layer of vitreous ice, and the procedure is not always fully reproducible. The optimal ice thickness is critical for obtaining micrographs for high resolution reconstructions, with the upper limit of ice thickness considered to be under 300Å for electron transparency (71,72). Thick ice does limit resolution, while thin ice can force out particles, induce a preferred orientation, or even damage biological structures (73-75). Microfluidics has enabled several “blot free” systems to be developed for preparing TEM grids – notably the SPT Chameleon^®, CryoSol Vitrojet^™, cryoWriter, and custom time resolved systems (76-79). Although these systems differ, each offers rapid vitrification on the millisecond timescale, which helps to avoid lengthy interaction with the air-water interface, as well as improves consistency over conventional blotting. A broader strategy may be to devise a range of affinity grids that can immobilize particles in ways that maintain their structural integrity (70). Detergents and other additives could always be screened, but new methods such as incorporation of Late Embryogenesis Abundant (LEA) proteins to samples could overcome environmental concerns of the commonly used polyfluoroalkyl substances (PFAS) based detergents (e.g. fluorinated octyl-maltoside) (80,81). Grid technology has paved the way for high-quality, ultra-stable gold supports with minimal beam-induced motion, enabling some of the highest reported resolutions in structural studies (82,83).

Improvements in the new generation of energy filters continue to reduce noise from inelastically scattered electrons, pushing the resolution limit and overall quality of micrographs (84). The current development of 100 kV TEM systems minimizes radiation damage, while reporting resolution at the same level of existing 200 kV and 300 kV systems (85-87).

Regardless of all these problems, there is no question that cryoEM will provide a powerful tool for drug discovery and complement XRC well as cryoEM overcomes the limits of XRC for large, heterogeneous, and flexible macromolecular complexes. Streamlining computational workflows for automated data acquisition coupled to processing pipelines, along with automated model building will make cryoEM a primary method for drug discovery in the future.

7. Expert Opinion

CryoEM has emerged as a transformative technology, revolutionizing the field of structural biology and, by extension, drug discovery. There are numerous approaches gaining importance in modern drug discovery, and cryoEM has the potential to be particularly useful among them. Since cryoEM is particularly well-suited for studying large, complex structures, one of the avenues where it can be advantageous over other methods is the quickly evolving field of induced proximity, including targeted protein degradation, that carries a promise of drugging undruggable (88). Capturing induced-proximity events is challenging, as complexes containing such interactions can be difficult to crystallize due to their transient or dynamic nature, as well as their size. Antibodies are widely used as therapeutic agents, also a rapidly evolving field. CryoEM can help to visualize the interactions between antibodies and their target antigens, providing rapid epitope mapping. Such information can be used to explain their mode of action, as well as to improve antibody design and enhance their therapeutic efficacy. Combined with other experimental techniques cryoEM can be used for approaches such as iterative vaccine design (89). In that respect cryoEM already has played a significant role in fighting COVID-19 (90), as well as contributed to a rational design of effective respiratory syncytial virus (RSV) vaccine candidates (91). CryoEM can assist in rational design for chimeric antigen receptor (CAR) T-cell receptors, where detailed understanding of their interactions with cancer antigens can facilitate maximizing their therapeutic potential by fine-tuning CAR receptors affinity and specificity (92). CryoEM has also been used to determine structures of gene editing enzymes in action. This knowledge can be used to modify the properties of gene editors and improve their speed and accuracy (93). What is more, cryoEM can be used for characterization and design of viral and non-viral gene delivery vehicles in gene therapy (94,95), as well as vaccine delivery vehicles, like lipid nanoparticles.

Even though the single particle analysis is currently the main branch of cryoEM utilized in drug discovery, cryo-electron tomography (cryoET) is gaining momentum, providing the scientists with the ability to study the molecular architecture of cells and the processes within in unprecedented detail. The key aspect in any drug discovery pipeline is to mechanistically rationalize the choice of novel targets, which requires a deep understanding of cellular processes and pathways involved in the pathogenesis. Use of cryoET allows researchers to gain a detailed understanding of how a target plays a role in disease biology, which is a fundament for all the following steps of the drug discovery process.

Artificial intelligence is playing a significant role in cryoEM data processing and analysis (96). Particle picking and identification of 2D classes are critical steps in cryoEM structure determination. AI algorithms can automatically identify and pick individual particles from cryoEM micrographs, reducing the time and effort required for manual selection. In many cases, it is extremely difficult for an untrained eye to identify quality particles and differentiate between slightly different conformational states of the particles. AI algorithms, being able to differentiate between subtle conformational changes, can facilitate the classification of particles. AI-driven methods can also be used to determine the 3D structure of macromolecules from cryoEM data without prior knowledge of the structure. Even though we are probably not there yet, especially for more challenging cases, eventually AI can automate the entire cryoEM workflow, from data acquisition to structure determination, reducing manual labor and increasing efficiency.

The 2024 Nobel Prize in Chemistry, awarded to David Baker, Demis Hassabis, and John Jumper, highlights the transformative impact of AI in biomedical research. AI not only predicts molecular structures in the absence of experimental data but also improves the accuracy of structures obtained through various techniques, including cryoEM (21,97-100). As these approaches continue to converge, cryoEM is poised to benefit even more from breakthroughs like AlphaFold3 and its future developments.

CryoEM is a cutting-edge, rapidly advancing technique with great potential. In the past two months, two groundbreaking papers have made a significant impact on the research community. The first demonstrates that cryoEM can clearly resolve hydrogen atom positions and water networks (54), while the second reports an atomic resolution (1.09 Å) protein structure (8RQB) that reveals double conformations (39).

The application of cryoEM to structure-based drug discovery is still evolving, but it is rapidly becoming a routine method of choice. As the technology continues to advance, we can expect to see more examples of drugs and therapeutic approaches that will be discovered, designed or optimized using cryoEM-derived structural information, leading to the development of new and effective therapies for numerous pressing health issues.

Supplementary Material

Supplementary material

Figure S1 Double conformations observed in the cryoEM structure of mouse heavy-chain apoferritin (PDB ID: 8RQB). The figure highlights the iron channel, showcasing four histidine residues coordinating the iron atom. The histidine residues exhibit double conformations, demonstrating the ability of modern high-resolution cryoEM to resolve alternative conformations in critical regions of macromolecular structures.

Table S1 This table showcases a set of approved drugs that have been structurally characterized exclusively by cryo-electron microscopy (cryoEM) according to current entries in the Protein Data Bank (PDB). At the time of this study, these drugs lack structural data from alternative methods such as X-ray crystallography or NMR, underscoring the unique role of cryoEM in providing the initial structural insights for these drug-bound complexes. Although future depositions may include structures resolved by other techniques, cryoEM will remain the pioneering method that first captured these specific drug interactions. This compilation highlights the increasing impact of cryoEM in advancing drug discovery, particularly for targets and complexes challenging to study by other structural approaches.

Article Highlights.

• CryoEM is a rapidly evolving and cutting-edge technique with immense potential.

• Its role in structure-based drug discovery is expanding as the field continues to evolve.

• CryoEM has revolutionized the study of large, complex structures, especially in drug discovery, offering unique advantages for investigating transient and dynamic interactions, including applications in induced proximity and targeted protein degradation.

• When combined with cryoET, cryoEM provides unparalleled insights into cellular molecular architecture and disease biology, making it invaluable for rational drug target identification.

• Recent breakthroughs in cryoEM have significantly enhanced resolution, although challenges such as lens aberrations and solvent density variations highlight the delicate balance between achieving high resolution and maintaining model accuracy.

• CryoEM structures, like other structural models, require refinement and robust quality metrics. The Q-score is widely used, but the model-to-map correlation coefficient (CC_MASK) has emerged as an alternative. It addresses the limitations of Q-scores by accounting for atomic displacement, density shape, and recent advances in resolving multiple conformations.

• Artificial intelligence is revolutionizing cryoEM data analysis by automating particle identification, improving classification, and enhancing 3D structure determination, significantly reducing the need for manual intervention.

• Metal ions play a pivotal role in drug discovery, both therapeutically and diagnostically. CryoEM excels in studying large, complex structures involving metal ions, preserving critical metal interactions, and enabling the analysis of metal-induced conformational changes in proteins.