Cryo-EM as a powerful tool for drug discovery

Graphical abstract

Abstract

The recent revolution in cryo-EM has produced an explosion of structures at near-atomic or better resolution. This has allowed cryo-EM structures to provide visualization of bound small-molecule ligands in the macromolecules, and these new structures have provided unprecedented insights into the molecular mechanisms of complex biochemical processes. They have also had a profound impact on drug discovery, defining the binding modes and mechanisms of action of well-known drugs as well as driving the design and development of new compounds. This review will summarize and highlight some of these structures. Most excitingly, the latest cryo-EM technology has produced structures at 1.2 Å resolution, further solidifying cryo-EM as a powerful tool for drug discovery. Therefore, cryo-EM will play an ever-increasing role in drug discovery in the coming years.

These are exciting times for our ability in drug discovery to exploit the 3D structure of the macromolecular target. Structure-based drug discovery (SBDD) began in the 1980s, when both X-ray crystallography and molecular modelling were first introduced into drug discovery; by the 1990s SBDD had matured into a routinely used approach. The first major success of SBDD was the discovery of HIV protease inhibitors against AIDS; this pandemic began in 1981 and became a huge threat by the mid-90s, when the introduction of both HIV reverse-transcriptase (RT) inhibitors and HIV protease inhibitors began to turn AIDS into a manageable disease (Fig. 1 )¹. Indeed, this is a success story for the pharmaceutical R&D enterprise as a whole, in terms of our ability to respond to a pandemic.

Fig. 1.

Deaths due to AIDS in the US were climbing dramatically, until the introduction of drug therapies in the mid-1990s. (adapted from original graph from cdc.gov).

A revolution in structural biology has occurred in the past few years, due to dramatic advances in the technology of electron microscopy (EM) at cryogenic temperatures - cryo-EM. Not that long ago, EM was often disparaged as ‘blob-ology’, due to the low resolution of the structures; these were useless for drug discovery. Fig. 2 shows how structural information has improved over the years for the ribosome, in parallel with advances in X-ray crystallography and EM.

• •Fig. 2A shows the first EM structure of the ribosome from 1984².
• •Fig. 2B shows the EM structure in 1995^3,
• •Fig. 2C the X-ray density in 1998^4,
• •Fig. 2D the X-ray structure from 2000^5,
• •Fig. 2E a 2008 X-ray structure showing atomic detail how the antibiotic linezolid binds to the ribosome⁶ and finally
• •Fig. 2F shows a recent cryo-EM structure⁷, which now displays the various components of the complex (mRNA, tRNA, the emerging polypeptide chain).

Fig. 2.

Progression of resolution of ribosome structures 1984–2019. (A). the first EM structure of the large subunit of ribosome²; (B). 20 Å EM structure of the large subunit³; (C). 9 Å X-ray structure of the large subunit⁴; (D). 2.4 Å X-ray structure of the large subunit⁵; (E). a 3.5 Å X-ray structure of the antibiotic linezolid bound to the 50S subunit⁶; (F). a cryo-EM structure of the full ribosome at 2.4–3.5 Å.⁷ The nascent polypeptide chain is shown in magenta, and mRNA density is shown in red (it protrudes in the direction of the viewer). The P site tRNA is green, the small subunit in yellow, and large subunit in light blue. (composite figure, from individual figures reprinted with permission).

Improvements on multiple fronts in EM have brought EM structures to the resolution where they are now useful for drug discovery. The 2017 Nobel Prize in Chemistry was awarded to Dubochet, Frank and Henderson, in recognition of their contributions to these improvements that transformed the low-resolution EM into the high-resolution cryo-EM⁸. Some of these improvements were: the ability to rapidly capture the undistorted macromolecule in vitrified ice, the ability to deconvolute the multiple conformations and orientations of the molecule in that vitrified state, and improvements in the instrumentation, especially the sensitivity of the detectors.

Cryo-EM and X-ray crystallography are complementary techniques; each facilitates the other. Compared to X-ray crystallography, cryo-EM provides a number of advantages:

• (1)capturing large complexes containing multiple proteins is routine, allowing the capture of a protein in its native milieu vs. the study of a protein in isolation in crystallography,
• (2)the protein is usually captured in a near-native state, something especially important in transmembrane proteins, like ion channels and G-protein-coupled receptors (GPCRs);
• (3)conformational changes are routinely obtained, making clear the manner in which the target is a ‘molecular machine’. This often exposes unanticipated drug binding sites, or leads to transformational insights into mechanism, or opportunities for selectivity;
• (4)some important off-targets of drugs can now be incorporated into the SBDD approach, e.g. ion channels and transporters; and
• (5)the timeline for determining these structures can be more predictable, because crystallography requires a crystal, which is a serious bottleneck, while cryo-EM can work with the sample directly.

The primary disadvantage is that today the resolution of many cryo-EM structures is still only comparable to a low-resolution X-ray structure. However, this is changing rapidly (Fig. 3 ), and some recent structures have 1.2 Å resolution[9], [10], comparable to the best macromolecular X-ray structures. Fig. 3 shows the growth in the number of publicly available cryo-EM structures and their improving resolution in the past decade. (We are fortunate that most funding agencies and journal policies insist that published structures are made publicly available).

Fig. 3.

Histogram of cryo-EM structures deposited in the PDB over the past 10 years (2010–2019). The bar shows the number of structures deposited in each year, divided into four categories based on the reported resolution of the structures, lower than 4 Å (dark blue), 4 to 3.5 Å (pink), 3.5 to 3 Å (green), 3 to 2.5 Å (yellow), and above 2.5 Å (red).

There are many reviews on cryo-EM structural biology[11], [12], [13], [14], [15], [16], [17], [18], [19]. What differentiates this review is our focus on examples of how cryo-EM structures have impacted drug discovery, in terms of both driving the discovery of new compounds and defining the mechanism of action of known drugs. Some examples are described in detail; others are summarized briefly in a later section. There are two major challenges to compiling such a review: the literature is growing at an astonishing pace, and many examples of uses of cryo-EM for SBDD are still proprietary, but are expected soon. In each case, we will highlight many of the points made above, in terms of the special advantages of cryo-EM, and will primarily highlight examples where crystallography has fallen short of what is required.

Brief description of cryo-EM technology: Structure determination by cryo-EM typically proceeds through the following steps:

• 1.Isolation and purification of the sample, from natural and/or recombinant sources. This step is similar to that needed for crystallography, but the amount of sample required for EM is much smaller. Typically, several microliters at ~1 mg/ml concentration would be sufficient.
• 2.Screening of the quality of the sample by negative stain and/or cryo-EM. A sample displaying good conformational homogeneity can proceed to data collection, while a poor-quality sample will require optimization of the sample and/or the cryo-EM grid preparation. This is often the time-consuming part of a structure determination by cryo-EM.
• 3.Data collection, image analysis and reconstruction. With good quality grids, cryo-EM images (movies) can be acquired and analyzed, which ultimately could lead to a three-dimensional reconstruction revealing the structure of the sample. The image analysis and reconstruction require significant computing power, but the software is becoming more and more powerful and automated[20], [21], [22], [23]. This step typically can be completed within a few days to a week or two.
• 4.Model building, refinement and analysis. The powerful tools that have been developed for model building and refinement for crystallography can be used for building and refining atomic models into cryo-EM reconstructions[24], [25], [26]. Additional tools have been developed to deal with the lower resolution cryo-EM (and crystallography) maps²⁷ as well as for visualization and model fitting²⁸.

Most importantly for drug discovery, recent improvements in both hardware and software have allowed cryo-EM structures to provide visualization of bound small-molecule ligands in the macromolecules (Fig. 4 ). In fact, nearly half of the ~1200 cryo-EM structures deposited in the Protein Data Bank (PDB) in 2019 are at better than 3.5 Å resolution, and the number of structures at better than 2.5 Å resolution has been increasing over the years (Fig. 3). Experience has shown that EM structures between 4 and 3.5 Å resolution can already have well-defined density for the bound ligands. Most excitingly, cryo-EM reconstructions at 1.2 Å resolution have been obtained using the latest technology[9], [10]. This demonstrates that there is no inherent limitation to achieving atomic resolution by cryo-EM, further solidifying cryo-EM as a powerful tool for drug discovery.

Fig. 4.

Observed cryo-EM density for selected drug-like ligands. (A). NDI-091143 with human ACLY. (B). GLPG1837 with human CFTR. (C). Emetine with P. falciparum 80S ribosome. (D). Granisetron with mouse 5-HT_3A serotonin receptor. (E). ZK200775 with rat GluA2 AMPA receptor. (F). Alprazolam (Xanax) with human GABA_A receptor. (G). Flecainide with rat Na_v 1.5 voltage-gated sodium channel. (H). Glibenclamide with rat pancreatic K_ATP potassium channel. (I). Bay K 8644 with rabbit Ca_v1.1 voltage-gated calcium channel. (J). Bictegravir with SIV intasome. The cryo-EM density is shown as a cage (in gray), and the compounds are shown as stick models; black (C), blue (N), red (O), yellow (P or S), green (Cl or F). The resolution of the cryo-EM reconstruction is indicated. Produced with PyMOL (www.pymol.org).

Detailed description of selected cryo-EM structures: ATP-citrate lyase (ACLY). ACLY catalyzes the conversion of citrate and CoA to oxaloacetate and acetyl-CoA, which also requires the hydrolysis of ATP. It is a major source of acetyl-CoA in the cytoplasm, and a potential target for drug discovery against dyslipidemia and cancer[29], [30]. Bempedoic acid (ETC-1002) has been approved for the treatment of hypercholesterolemia³¹, validating ACLY as a drug discovery target. This compound has a linear C15 di-acid backbone and needs to be converted to the CoA ester in the liver for its inhibitory activity against ACLY, which likely occurs through binding to the active site.

A series of low nanomolar ACLY inhibitors has recently been disclosed by Nimbus Therapeutics, and they are competitive against the citrate substrate of the enzyme³². The structure of one of these inhibitors (NDI-091143) in complex with human ACLY was determined by cryo-EM at 3.67 Å resolution in the Tong lab (Figs. 4A, 5A)³². Remarkably, the compound is bound in an allosteric cavity, in the hydrophobic core of the citrate binding domain of the enzyme and has no overlap with citrate (Fig. 5 B). A large conformational change in the domain is required for the compound to bind, which indirectly competes against citrate. Especially, Arg379 interacts with the citrate but clashes with the inhibitor. It moves by 12 Å upon inhibitor binding (Fig. 5B). Ile344 moves by 6 Å upon inhibitor binding, and its new position clashes with citrate. The binding cavity does not exist without this conformational change. Free-energy perturbation calculations based on the binding mode of this compound can explain the binding affinities of other compounds in the series³², supporting the structural observations and indicating that a structure at this resolution could be useful for explaining the SAR of the compounds and drug design. In fact, extensive efforts at determining the binding mode of this compound by crystallography were not successful, and therefore cryo-EM enabled the breakthrough on this compound.

Fig. 5.

Structure of human ACLY in complex with NDI-091143. (A). Schematic drawing of the structure of human ACLY tetramer in complex with NDI-091143. Monomer 1 is colored by its domains and labeled, monomer 2 in yellow, monomer 3 in orange and monomer 4 in magenta. NDI-091143 is shown as a sphere model (carbon atoms in black). (B). Extensive conformational changes in the citrate domain of ACLY is required for inhibitor binding. Regions of large conformational changes are indicated with the red arrow heads, and those for Arg379 and Ile344 are indicated with the red and blue arrows, respectively.

GPCRs. GPCRs (G-protein-coupled receptors) are one of the oldest targets for drug discovery. Structural data on GPCRs have exploded in the past 20 years, starting with the X-ray structures of rhodopsin³³ and the beta-2 adrenergic receptor³⁴. Crystallography, cryo-EM, and NMR have all contributed to a wealth of knowledge on the details of structure and mechanism of existing drugs, which have led to the design of improved molecules³⁵.

There are many examples of cryo-EM structures of GPCRs that we could describe in this review. We will discuss one here in detail, GLP-1R; others will be described in brief in the following section.

GLP-1R. GLP-1R is the receptor for the gastrointestinal peptide hormone GLP-1, a key component of insulin regulation. Peptides based on GLP-1 have long been on the market for treatment of diabetes³⁶. Most other drugs have been discovered via peptidomimetic approaches, targeting the orthosteric site where GLP-1 binds. Activity in the past 10 years has focused on non-peptidic molecules originating from HTS hits, functioning as positive allosteric modulators (Fig. 6 ), but these molecules are sub-optimal in many ways[37], [38], [39]. Upstream of GLP-1R, the “gliptin” drugs inhibit DPP-IV, the peptidase responsible for the cleavage and inactivation of GLP-1.

Fig. 6.

Some GLP-1R allosteric modulators (numbering as in³⁷.

What makes this example of the cryo-EM structure of GLP-1R especially compelling is that it exemplifies all the advantages of cryo-EM: (1) the full structure highlights aspects that were not apparent in the X-ray structures of the individual components; (2) the key conformational change upon ligand binding and activation is now apparent; (3) this structure is significantly different than what had been anticipated via homology models, and (4) based on the state of the art, it is poised to drive the discovery of much improved drug candidates. The first cryo-EM structure of GLP-1R was published in 2017^{40 ,} Fig. 7 .

Fig. 7.

Cryo-EM structure of the hGLP-1:rGLP-1R:Gs complex. a, Schematic of the activation of a family B GPCR by extracellular peptide agonist via a ‘two–domain’ binding mechanism. b, Views of the GLP-1R:Gs complex cryo-EM density map, colored by subunit (TM in light green, NTD in dark green, GLP-1 peptide in orange, GαsRas in gold, Gβ in light blue, Gγ in dark blue and Nb35 in grey). c, Structure of the activated GLP-1R-Gs complex in the same view and color scheme as shown in b. (from⁴⁰, reprinted with permission).

In contrast to earlier work, this new cryo-EM structure shows the full seven transmembrane (TM) domain portion of the receptor, the N-terminal domain, the peptide ligand, and the G protein which transduces the signal. The structure, coupled to other biophysical and enzymatic data, clarifies how peptide binding at the extracellular surface is communicated to the intracellular receptor side: conformational changes in the transmembrane domain result in a sharp kink in the middle of TM helix 6, which pivots its intracellular half outward to accommodate the α5 helix of its G-protein. In that publication, the binding site of the known allosteric modulators could only be modelled, by homology to other GPCRs with modulators bound. In addition, it is thought that this structure may be a paradigm for the structure of other family B GPCRs, highlighting a general mechanism whereby a flexible N-terminal domain outside of the membrane folds to form the binding site for the endogenous ligand. This structure also allowed them to propose a mechanism of activation. Finally, the interaction with the G-protein suggests a molecular basis for the “biased signaling” that is seen in the allosteric modulators, namely that the ratio of signaling in multiple downstream pathways differs from molecule to molecule.

A series of new GLP-1R allosteric modulators has appeared from vtv Therapeutics (the company formerly known as Trans Tech Pharma). The first of these compounds TTP054 was tested in Phase 2 clinical trials, but there is no evidence that it continued on phase 3[38], [41]. However, newer analogs of these molecules are now in Phase 2, e.g. TTP054-201, and the cryo-EM structure of an analog has been published this year^{42 ,} Fig. 8 , This structure shows the same overall conformation as the earlier cryo-EM structure, but now shows the location of the ligand. Of course, others are in this race as well, e.g. Pfizer’s PF-06882961, which has been disclosed briefly, but apparently not in conventional scientific literature.

Fig. 8.

The TT-OAD2–GLP-1R–Gs cryo-EM structure reveals non-peptide binding site. a. TT-OAD2. b. orthogonal views of the TT-OAD2–GLP-1R–Gs complex cryo-EM map (left) and the structure after refinement in the cryo-EM map (right), color-coded to protein chains; GLP-1R (blue), TT-OAD2 (red), heterotrimeric Gs (α: gold, β: dark cyan, γ: purple, Nb35: salmon). c. Detail of TT-OAD2 in the binding site. Interacting residues of GLP-1R (blue) with TT-OAD2 (red). (from⁴² reprinted with permission).

In addition, GLP-1R has been implicated in some CNS disorders, like AD and PD³⁶. Additional work on GPL-1R allosteric modulators driven by this cryo-EM structure may allow further optimization for these indications.

Ion channels. Ion channels control the conduction of signals in nerves and are one of the oldest drug targets. They have been a rich source for structural biologists as well, starting with the first major breakthrough in 1998 with the publication of the X-ray structure of the potassium channel by Rod MacKinnon and colleagues⁴³. Three examples which most strongly exemplify the strengths of cryo-EM will be discussed in detail here: TRP channels, CFTR, and hERG. In each case, novel insights into mechanism have arisen by understanding the conformational changes in these molecular machines. In addition, they are the targets of important new drugs, and are being actively pursued by many drug discovery groups. Unfortunately, in each case there is little in the public literature that describes how these structures are being exploited in drug design.

TRP channels. The TRP channels (Transient Receptor Potential channels) mediate pain and temperature signals, and have been pursued recently as a target for neuropathic pain, migraine, and inflammatory pain; multiple drug candidates are being evaluated in clinical studies⁴⁴ TRPV1 is the receptor to which capsaicin binds (the active substance in hot chili peppers), and has been described as the “‘poster child’ for elucidating basic principles underlying TRP channel function and structure”. Its structure was first determined via cryo-EM by the Julius lab at UCSF (Fig. 9 )[45], [46]. Cryo-EM is spurring an on-going revolution in this area⁴⁷. It has enabled the determination of a large number of TRP channel structures (Table S1).

Fig. 9.

Structure of TRPV1 in complex with vanilloid ligand and spider toxin. a. TRPV1 channel, bottom view. Note that S1–S4 domains flank and interact with S5-P-S6 pore modules from adjacent subunit, reminiscent of voltage-gated ion channels. b. Resiniferatoxin (RTX). c. Vanillotixin (DkTx, spider toxin). d. TRPV1 channel, bound to RTX (red sphere) and vanillotoxin (magenta). (from⁴⁶, reprinted with permission).

This initial cryo-EM structure was noteworthy for multiple reasons. It gave the first, approximate picture how capsaicin and related molecules bind; the second publication gave the atomic-level detail of the ligand-protein complex. These publications also showed multiple functional states, allowing hypotheses to be formed on the mechanism of activation (Fig. 10 A). While the overall architecture resembled other channels like the potassium channel, the mechanism of activation was startlingly different from what was known from other channels, pointing to an unusual allosteric coupling from the ligand-binding S1-S4 helices (homologous to the voltage-sensing domain in voltage-gated ion channels) to the pore-forming domains. These differences are critical to achieving selectivity between and among these families of channels and contributed to the ability to design out the side effects of the first TRPV1 antagonists discovered⁴⁸. These experiences exemplify the power of cryo-EM in drug discovery: (1) multiple conformational states can readily be determined, often in one experiment, (2) these states lead to mechanistic hypotheses, which in turn lead to (3) insights into how to achieve selectivity for the channel of interest.

Fig. 10.

a. Model for TRPV1 activation (from⁴⁶ reprinted with permission)). b. TRPV1 antagonist designed from this structure (from⁴⁹.

One example of a medicinal chemistry effort exploiting this structure is from a group in Seoul, S Korea, who have published extensively on a series of analogs designed using this cryo-EM structure (Fig. 10B)[49], [50] They achieved sub-nM activity and excellent activity in a neuropathic pain model at 10 mg/kg.

The Cystic Fibrosis Transmembrane Conductance Regulator (CFTR). Cystic Fibrosis is a fatal lung disease, which was found in 1989 to originate from a single mutation in one gene, expressing a protein which was later found to regulate chloride levels, given the name CFTR[51], [52], [53]. One mutation, F508del, accounts for the majority of patients, though disease mutations occur in many positions in CFTR. The first CF drugs directly targeting CFTR were discovered by Vertex Pharmaceuticals[54], [55]: ivacaftor, approved in 2012, followed by lumacaftor, and tezacaftor, which now form the standard-of-care as a triple combo, Trikafta, approved in 2019. The Vertex drugs were all discovered empirically, in heroic medicinal chemistry efforts⁵⁶, exploiting a high-throughput cell-based assay. No knowledge of the receptor structure was exploited, except to select specific cell lines with certain mutations. Despite their great commercial success, Vertex has thus far attracted little competition, suggesting that this is a challenging area for drug discovery.

All efforts to crystallize CFTR, the first step to determining an X-ray structure, have failed. However, the lab of J. Chen at Rockefeller has used cryo-EM to determine increasingly useful structures of CFTR: first from zebrafish⁵⁷; then human⁵⁸ with some domains unresolved; then human, better-resolved with indications of the conformational changes upon channel activation⁵⁹; and, finally, in 2019, CFTR in complex with ivacaftor, and a drug candidate discovered by Galapagos (Fig. 11 )⁶⁰. These two structurally distinct compounds bind in the same pocket, at the interface between the CFTR and the lipid membrane. The binding pocket is part of the hinge for channel gating, and the presence of the compound may stabilize the channel pore in an open configuration, explaining its potentiating effect.

Fig. 11.

Structure of CFTR. Ivacaftor (green) bound to CFTR, viewed from the membrane. The first nucleotide binding domain (NBD1) is in yellow. NBD2 in dark blue. The transmembrane helices are light blue, except for the kinked helix TM8 to which ivacaftor binds is shown in orange. F508 is shown in purple. The two ATP molecules, not shown, would be sandwiched in between the NBD’s. (from⁶⁰, PDB code 6O2P).

Persuasive homology models of human CFTR had been built [e.g.⁶¹but these did not prove useful. It is difficult to convey how astonishing this final structure of ligand-bound CFTR was to those active in CF, as all prior hints about possible ligand-binding sites pointed to regions in the nucleotide-binding domains near F508. Instead, both ivacaftor and GLPG-1837 bind to a kink in a transmembrane helix embedded in the membrane. None of the earlier structures showed either a clear kink in that helix, or a binding pocket there, and generally the notion of a binding pocket contacting the lipid bilayer is exceptional and poses special challenges to compound design.

It is in the early days, but it is reasonable to expect that this new structure will now permit structure-based optimization of such molecules. Further structures clarifying conformational changes, mechanism, etc. will undoubtedly be helpful.

As an aside, from the perspective of genomics-based small-molecule drug discovery, it is worth noting the length of time required: the mutation was identified in 1989, and not until 2012 was the first drug approved to treat the disease. And, of course, there are many other targets where disease-causing mutations have been known for a long time but thus far no treatment is available.

hERG channel. Unwanted inhibition of the hERG channel has been a bête noire of drug hunters since the discovery in the mid-1990s that this channel mediates the long-QT syndrome which often leads to fatal arrhythmias[62], [63] Homology modelling⁶⁴ has not shown to be especially useful in designing out the hERG liability. In 2017, a cryo-EM structure was published from the MacKinnon lab (Fig. 12 )⁶⁵, which broadly showed the same architecture as that homology model - a central volume surrounded by four hydrophobic pockets - but this new detail made it clear why this channel should be so promiscuous, why a basic amine is essential, and highlights how this channel’s selectivity filter is subtly different from other known K channel structures. It is not yet evident in the medicinal chemistry literature that this structure has permitted structure-based optimization away from hERG affinity, or whether a structure with a ligand bound will be needed. It has, however, played a role in the steadily increasing ability of various software tools to estimate the long QT liability[66], [67].

Fig. 12.

Model of hERG activation and hERG blockade. (A) hERG channel, overall density, color-coded by resolution. (B) hERG overall architecture viewed down the 4-fold axis from the extracellular side. The hERG channel is represented as semi-transparent molecular surface with ribbon representation for one subunit. (D) The central cavity of hERG, shown as a translucent surface colored by eletrostatic potential. Residues related to drug binding are shown as sticks. (E) Model of hERG activation. (Figures from⁶⁵, reprinted with permission).

Viruses. Here, we will focus on one single virus, SARS-CoV-2, the cause of COVID-19, the scourge of this annus horribilis. The SARS-CoV-2 genome sequence was first published in January 2020⁶⁸. Since that time, over 200 X-ray structures and 20 cryo-EM structures have been published. From the simplest perspective, the key viral proteins are

• •the spike protein, which resides on the exterior of the virus;
• •the RNA-dependent RNA polymerase (RdRp);
• •the protease, which cleaves the polypeptide chain translated from the viral genome, as in HIV protease and HCV protease, which has primarily been the domain of crystallography

Drug discovery driven by these structures is still in its infancy, but the structures do highlight multiple opportunities to design inhibitors of the key steps of viral infection and propagation. Furthermore, one structure suggests the binding conformation of remdesivir, the only drug which has been authorized by the FDA for emergency use against the virus.

The spike protein. The virus gains entry into the cell via interactions between the spike proteins on the virus surface and a transmembrane protein which sits on the surface of the host cell, ACE2. The spike protein functions as a trimer and undergoes a conformational change and a series of steps to infiltrate into the cell. Many X-ray structures have appeared of the isolated spike protein, but especially compelling are the larger complexes which have been determined via cryo-EM, and ones which demonstrate the large conformational changes that occur upon activation (Fig. 13A, B)⁶⁹

Fig. 13.

(A) and (B) show the conformational changes of the spike protein trimer that occur to expose the binding domain for ACE2, as the virus encounters the host cell (from Fig. 3^,69 reprinted with permission) (C) A high-level view of the RNA-dependent RNA polymerase (RdRp), highlighting the thumb/palm/fingers domain, surrounded by accessory proteins (from Fig1B,⁷⁰), reprinted with permission (D) A detailed view of RdRp, with a model of how the bioactive forms of remdesivir and sofosbuvir may bind. (from Fig 4⁷⁰, reprinted with permission (E) A structure of the replicase complex, of which RdRp is one part. The template RNA is shown as a blue helix; the newly synthesized strand in red. Note: the color-coding of thumb/palm/finger is different from (C). from Fig 2^,71

The RNA-dependent RNA polymerase (RdRp). SARS-CoV-2 is an RNA virus, encoding its genome in strands of RNA, unlike humans whose genome is encoded in DNA. Hence to replicate its genome, it requires an enzyme that is not present in the host cell, an RNA-dependent RNA polymerase (RdRp). Some known antivirals are weak RdRp inhibitors, such as the nucleoside mimics remdesivir and sofusbuvir. A cryo-EM structure has been published which allows one to form reasonable hypotheses how these molecules may bind to RdRp (Fig. 13C, D)⁷⁰. This structure probably provides the greatest opportunity for structure-based design of COVID-19 antivirals.

The RdRp is only one component of a much larger complex, the replicase complex. A recent preprint describes a cryo-EM structure of a large section of that replicase complex⁷¹ (Fig. 13E). Understanding the motions of this molecular machine, as RdRp ‘walks’ along the template RNA, may give insights into alternative ways of inhibiting the polymerase action. One new revelation from the structure is that there are extensions of the viral protein NSP-8 which interact with the paired RNA.

Another great opportunity lies ahead for cryo-EM in SARS-CoV-2 antivirals by looking at other mega-complexes involving host proteins, for example the enhanceosome, the target of interferon, a protein which is well-known for its antiviral activity against other viruses, and has even been demonstrated to show clinical efficacy against COVID-19⁷².

Brief descriptions of additional cryo-EM structures with drug-like ligands: Overview.

Here, we describe several additional structures with drug-like small molecules bound (Table P-1 ). Unlike the previous section, only brief descriptions will be given due to space considerations; the readers are referred to the original publications for detail. The cryo-EM density for some of the small molecules are also shown (Fig. 4), as an illustration of the quality of the cryo-EM reconstructions. A more complete listing of the published structures with bound drug-like small molecules is given in Table S1. Only those structures with explicitly modeled small molecules are considered here. Structures with bound biochemical ligands (substrates, substrate analogs, cofactors, prosthetic groups) are generally not included in the consideration here, although good-quality EM density can be observed for them as well. In addition, cryo-EM structures have been obtained for other drug discovery targets, such as inflammasome, but no structures of complex with drug-like molecules are currently reported for them. Of course, there are likely a substantial number of structures of ligand complexes determined at pharmaceutical and biotech companies that have not yet been published or deposited in the PDB.

Table P-1.

Cryo-EM structures with drug-like small-molecule ligands discussed in this review.

Drug-like ligand	Target	Resolution (Å)	PDB IDs and references
Emetine	Ribosome	3.2	3J7A78
Granisetron	Serotonin receptor	2.92	6NP082
ZK200775 (Fanapanel)	AMPA receptor	2.97	6PEQ83
Alprazolam (Xanax), diazepam (Valium)	GABA receptor	3.26, 3.58	6HUO, 6HUP87
Resiniferatoxin	TRP channel	2.95	3J5Q, 3J5R46
Flecainide	Sodium channel	3.24	6UZ088
Glibenclamide	Potassium channel	3.63	6BAA90
Bay K 8644, verapamil, nifedipine, diltiazem,	Calcium channel	2.7, 2.9, 2.9, 2.6	6JP8, 6JP5, 6JPB 6JPA92
Ivacaftor, GLPG1837	CFTR	3.3, 3.2	6O2P, 6O1V60
Bictegravir	Intasome	2.81, 2.9	6RWM98, 6PUW99
NDI-091143	ACLY	3.67	6O0H32

The analysis of deposited cryo-EM structures with bound drug-like ligands also indicates that the majority of them are for large complexes (ribosome, proteasome, spliceosome, microtubule, actin, virus, and others) and membrane proteins (receptor, channel, transporter, and others). This is a reflection of the fact that cryo-EM has made it possible (and/or more readily) to determine such structures, and the drug-like ligands probably also helped to stabilize some of the complexes for structural studies. In comparison, the number of structures of soluble enzymes is smaller, although several of them have been used to demonstrate the possibility of pushing cryo-EM to high resolution[73], [74], [75]. Nonetheless, small-molecule ligands can be observed for these targets⁷⁶, and fragment-based drug discovery by cryo-EM has recently been reported⁷⁷. Some of these proteins are too small for the current EM technology, which favors larger molecules and complexes. Structures of smaller proteins can be determined by EM for example by complexing it with an antibody. At the same time, constant improvement in EM technology has made it possible to determine the structures of ever smaller proteins. It is also likely that many of these structures are still being determined by X-ray crystallography, which is faster and easier if good quality crystals and compound soaking protocols are available. Routine use of high-throughput structure determination by cryo-EM, for example for lead optimization, will benefit from additional hardware and software improvement. However, if good conditions for preparing frozen grids of the sample are known, it may be envisioned that faster data collection and processing can be achieved to substantially increase the throughput of cryo-EM.

Ribosome. A large number of cryo-EM structures have been reported of ribosomes from various organisms bound to antibiotics, anti-protozoals, and other compounds (Table S1). These compounds include natural products (such as erythromycin, avilamycin, paromomycin, emetine, mefloquine) as well as synthetic molecules (such as linezolid and cadazolid). The binding modes of some of these compounds were originally determined by X-ray crystallography, but cryo-EM is now the dominant method for determining ribosome structures. For example, the binding mode of emetine is clearly defined by the cryo-EM map at 3.2 Å resolution (Fig. 4C)⁷⁸.

Serotonin receptor. The 5-HT3 serotonin receptors are pentameric ligand-gated ion channels (pLGIC) and they are targets for therapeutic agents against nausea, vomiting, and depression⁷⁹. The cryo-EM structures of these receptors in complex with serotonin (5-hydroxytryptamine, 5-HT) were reported recently[80], [81]. Serotonin binding causes extensive conformational changes, promoting a conductive, open state. The anti-emetic drug granisetron is bound in the same site as serotonin (Figs. 4D, 14A, B)⁸², but it stabilizes a closed state of the channel, distinct from the apo, resting state.

Fig. 14.

The binding site of the anti-emetic drug granisetron. (a) The density map of granisetron shows the binding site located at the intersubunit interface. The residue labels on the principal subunit are marked in black and those on the complementary subunit are marked in brown. (b) A comparison of the serotonin 3A receptor-apo (5-HT3AR-apo), 5-HT3AR-granisetron, and 5-HT3AR-serotonin structures shows that residues involved in ligand-binding undergo reorientation, due to granisetron’s larger size. (from^82, their Fig 2, reprinted with permission).

AMPA receptor. The ligand-gated ion channels AMPA-type ionotropic glutamate receptors (AMPARs) are associated with serious neurological diseases such as depression and Alzheimer’s. A cryo-EM structure of the GluA2 receptor bound to the competitive antagonist ZK200775 (also known as fanapanel) was reported recently (Fig 4E)⁸³. A crystal structure of this compound bound to the ligand binding domain (LBD) of the receptor is also available, at 1.25 Å resolution⁸⁴.

GABA receptor. Type A GABA (γ-aminobutyric acid) receptors (GABA_A receptors) are ligand-gated chloride channels in the central nervous system and are associated with mental illnesses such as epilepsy, chronic pain, anxiety and insomnia⁸⁵. These receptors contain many allosteric sites, and they are targets for benzodiazepines (such as diazepam/Valium and alprazolam/Xanax) and other therapeutic agents as well as for drug abuse. Competitive antagonists (such as flumazenil) are used clinically to reverse the effects of benzodiazepines, especially their overdose.

The binding modes of flumazenil⁸⁶, Valium and Xanax (Figs. 4F,15 )⁸⁷ have been revealed by recent cryo-EM studies. Xanax and Valium share essentially the same binding modes, at the interface between the extracellular domains of two of its subunits. Binding of these compounds may facilitate channel activation by stabilizing the interactions between the subunits. The antagonist flumazenil binds deeper into this pocket and may not be able to stabilize the interactions between the subunits.

Fig. 15.

The binding modes of diazepam (DZP) and alprazolam (ALP) from cryo-EM studies of Masiulis et al (from⁸⁷, their Fig 4 reprinted with permission). a. Chemical structures of DZP and ALP. b, c, Viewed parallel to the membrane plane, the cryo-EM map of the α1β3γ2 GABA_A R in complex with ALP (cyan) (b) and DZP (teal) (c). d. Detailed view of the benzodiazepine binding site at the α1⁺/γ2^– interface showing ALP bound.

Sodium channel. Voltage-gated sodium channel 1.5 (Nav 1.5) produces the cardiac action potential for heartbeat, and is the target of many drugs, e.g. antiarrythmic drugs such as flecainide. The cryo-EM structure of flecainide in complex with Nav 1.5 (Fig 4G)⁸⁸ reveals the detailed interactions between the compound and the channel pore, next to the selectivity filter. Other voltage-gated sodium channels are sensitive to neurotoxins such as tetrodotoxin (TTX, from the fugu fish) and saxitoxin (a paralytic shellfish toxin). The molecular basis of their pore-blocking activity has also been revealed by cryo-EM studies⁸⁹.

Potassium channel. ATP-sensitive potassium (KATP) channels in pancreatic β-cells control insulin secretion in response to blood glucose levels. They are the targets of the type 2 diabetes drug sulfonylureas, such as glibenclamide (GBC). It is bound in a central cavity of the transmembrane domain of the peripheral sulfonylurea receptor (SUR) subunit (Figs. 4H,16 )⁹⁰. The structure suggests a mechanism how binding of glibenclamide to the peripheral subunit can inhibit the central potassium channel. The carbamoylbenzoic acid drug repaglinide shares the same binding site[90], [91].

Fig. 16.

The cryo-EM structure shows that glibenclamide (red) is bound in a central cavity of the transmembrane domain of the K-channel, the peripheral sulfonylurea receptor (SUR) subunit (left viewed in membrane; right from extracellular orientation). ATP is in green. (from⁹⁰, their Fig 1 reprinted with permission).

Voltage-gated calcium channels. Voltage-gated calcium (Cav) channels are targets for widely used therapeutic agents. Cav1 antagonists such as dihydropyridines (DHP, with nifedipine as an example), benzothiazepines (BTZ, diltiazem), and phenylalkylamines (PAA, verapamil) are used in the clinic to treat hypertension, angina pectoris, and cardiac arrhythmia. BTZ and PAA compounds block the pore of the channel, while DHP compounds bind in an allosteric site, and some DHP compounds (Bay K 8644) are agonists. The binding modes of these compounds have been defined by cryo-EM studies (Figs. 4I,17)[92], [93].

Fig. 17.

Structure of CaV-1.1. A voltage-gated calcium channel CaV-1.1, with nifedipine, Bay K8644, diltiazem, and verapamil bound (from⁹², reprinted with permission).

Ryanodine receptor. Ryanodine receptors (RyRs) are ligand-gated calcium channels important in muscle function; they are a therapeutic target for arrhythmia, muscular dystrophy, neurodegenerative disease, and other myopathies⁹⁴. They are exceptionally large, and hence have been resistant to crystallography. The first cryo-EM structures were determined in 2015⁹⁵, with multiple conformational states associated with gating and binding sites of modulators (e.g. caffeine) determined one year later (Fig. 18)⁹⁶. Ca²⁺ converts the original rigid apo state to a 'primed' (non-conducting) state, poised for activation, which then is activated for pore conduction by binding of ATP. The natural product ryanodine locks the channel in an ion-conducting state. Intriguingly, dilation of the pore occurs by bowing and distortion of the S6 helix from a point midway through the membrane. The binding site and mechanism of competitive partial agonists nucleosides and nucleotides is now clear. A small molecule, ARM210, discovered based on these structures, is in clinical trials for RyR-mediated myopathies⁹⁷.

Fig. 18.

Ca²⁺, ATP, and Caffeine Bind to the C-Terminal Domain of RyR1. (A) The architecture of RyR1/Cs2 is depicted in ribbon representation, different domains color-coded. Putative binding sites for Ca2+, ATP, and caffeine are labeled. (B) Enlarged view with ligands and interacting residues on RyR1 depicted in stick representation (from Fig 2^,96, reprinted with permission).

Intasome. The HIV intasome mediates the integration of the viral genetic material into the host genome and is the target of integrase strand-transfer inhibitors (INSTIs), such as the second-generation drugs bictegravir and dolutegravir. Recent cryo-EM studies have revealed how these compounds, as well as other related molecules, are bound to integrase (Fig. 4J)[98], [99], coordinating two Mg²⁺ ions in the active site. The structure of a resistant mutant in complex with bictegravir provides insights into the mechanism of resistance.

Prospects and summary: The recent revolution in cryo-EM has produced an explosion of structures at near-atomic or better resolution. Further developments in applying cryo-EM to smaller proteins, in preparing cryo-grids for data collection[100], [101], and in automation of data collection and image analysis will further enhance the power and applicability of this technology. The establishment of national EM facilities should make it easier to access the top-end instruments for data collection and reconstruction at high resolution. Moreover, recent developments suggest that electron microscopes at 100 kV could be sufficient to produce high quality structures, which would reduce the cost of instruments and make them more widely available¹⁰².

In terms of impact on drug discovery, we expect cryo-EM has the greatest future potential in cases where the advantages of cryo-EM over other structural approaches are most needed:

• Where ligand binding site is unknown, and where ligand-induced conformational changes are critical for understanding mechanism, like ion channels

• In cytoplasmic proteins where the target may be one or two proteins in a large oligomeric structure. This includes not only the viral proteins mentioned above, but also cases like the inflammasome¹⁰³, and the actin-myosin complex¹⁰⁴, microtubules¹⁰⁵ and the ribosome¹⁰⁶.

• Where the SAR has seemed intractable and/or where optimization of molecules has not achieved the target profile,

• where the targets are resistant to crystallization.

In focusing here on cryo-EM, we are not diminishing the impact of X-ray crystallography. The methods are complementary in many ways. For example, the first X-ray structure of the ribosome relied on information from earlier cryo-EM structures.[4], [5]. Fitting cryo-EM density of large complexes often relies on X-ray structures of individual proteins, for example[70], [78]. A recurring question is “in those cases where the same protein has been studied via X-ray crystallography and cryo-EM, how do those structures compare?”. We are not aware of any systematic studies of this fascinating question. However, for those instances where a comparison can be made, in general the significant differences are in flexible loops on the surface, e.g.[69], [70]. These loops in solution may not have a well-defined structure anyway.

Another question is whether cryo-EM will be able to support the rapid iterative cycles that crystallography is often capable of in structure-based design. At this point, there is little in the published literature that allows us to properly address that. Private communications from those performing such work in pharma R&D organizations indicates that a wave of publications is imminent, which will hopefully shed light on this issue.

The new cryo-EM structures have provided unprecedented insights into the molecular mechanisms of complex biochemical processes. They have also had a profound impact on drug discovery, defining the binding modes and mechanisms of action of well-known drugs as well as driving the design and development of new compounds; they provide compelling opportunities for further optimization of existing molecules. Most excitingly, the latest cryo-EM technology has produced structures at 1.2 Å resolution, which demonstrates the power of this technology and should lead to resolution improvements in most cryo-EM structures. Therefore, we are convinced that cryo-EM will play an ever-increasing role in drug discovery in the coming years.

Funding

Research in the Tong laboratory is funded by the NIH (R35GM118093), Nimbus Therapeutics and Kintor Pharmaceuticals. JVD did not receive any specific grants from funding agencies in the public, commercial, or not-for-profit sectors.

Declaration of Competing Interest

LT is a consultant for Nimbus Therapeutics and Kintor Pharmaceuticals. JVD has no conflicts, but is President of Van Drie Research, which focuses on structure-based drug discovery for small biotech firms, and whose policy is not to disclose names of clients.

Appendix A. Supplementary data

The following are the Supplementary data to this article: