Companies hope the Nobel Prize-winning imaging methodology will reveal biomolecule characteristics that can guide drug discovery projects.
This year's Nobel Prize in Chemistry celebrated the remarkable rise of cryo-electron microscopy (cryo-EM), which uses a beam of electrons to map the structures of proteins and other biomolecules.
Once confined to academic labs, the technique is now being embraced by pharmaceutical and biotech companies, who are racing to set up cryo-EM collaborations and facilities. They hope that cryo-EM will reveal detailed structural characteristics of membrane proteins and a host of other biological targets that have resisted the attentions of established methods like X-ray crystallography and NMR. Such structural information is invaluable for drug designers, because it shows how small molecules bind to active sites. “There's definitely a rush to evaluate the technology,” says Harren Jhoti, chief executive of Astex Pharmaceuticals.
Ten years ago, cryo-EM was disparagingly nicknamed 'blobology', because it offered relatively poor resolution images that blurred the atomic details of target molecules. As recently as 2011, the Protein Data Bank (PDB) was logging a mere 50 EM structures per year.
Times have changed. Improvements in hardware, software and sample preparation have spurred a resolution revolution, and more than 460 EM structures have already been added to the PDB this year. The PDB now lists 1,800 cryo-EM structures. Although these are still dwarfed by the more than 120,000 X-ray structures that have been solved, researchers say that the increased interest in cryo-EM is a clear sign that the technique has come of age. “In the past 5 years, we've seen a real step forwards in cryo-EM,” says Fiona Marshall, chief scientific officer at Heptares Therapeutics.
One of the key attractions of cryo-EM is that researchers do not need to coax obstinate proteins into crystalline form. Instead, they flash freeze a solution of the molecule into tiny discs of ice. Inside the microscope, the molecules deflect electrons to project a 2D pattern onto a detector. Each molecule is in a slightly different orientation, so the researchers take multiple snapshots from different angles and combine these into a 3D model of the target. Image acquisition takes at least 24 hours, usually more, and generates terabytes of data.
The microscope's high-energy electrons can also rip apart the target molecule, so the electrons must be used sparingly. This produces blurry images, making it difficult to assemble the jigsaw of projections. Until recently, cryo-EM had struggled to match the performance of crystallography, which typically offers a resolution of 2 Å or better, enough to show the atomic detail of a small molecule in an active site.
In 2013, however, a new generation of digital electron detectors became commercially available. These can catch more electrons, speed up image capture and streamline data processing, triggering a slew of higher-resolution cryo-EM structures. Last year, a team led by Sriram Subramaniam at the US National Cancer Institute unveiled the highest resolution cryo-EM structure to date, a 1.8 Å snapshot of glutamate dehydrogenase (Cell 165, 1698–1707; 2016).
Cryo-EM generally works better with larger proteins, because they contain more atoms that can scatter incoming electrons. “The smaller they get, the harder it gets,” says Subramaniam. But he is also pushing this limit: his team's structure of cancer target isocitrate dehydrogenase, bound to a preclinical drug candidate, is the smallest protein–ligand complex to be imaged by cryo-EM, weighing in at just 93 kDa (Cell 165, 1698–1707; 2016).
In the past few years, researchers have also proved that cryo-EM is more than just a fallback method for proteins that won't crystallize. “Many of these targets are molecular machines — a ribosome does its ratcheting, an ion channel is open or closed — so you often have a mixture of 3D structures,” says Sjors Scheres at the Medical Research Council Laboratory of Molecular Biology (MRC LMB).
Whereas crystals lock biomolecules into a single conformation, cryo-EM can produce an album of freeze-frame images, showing a molecule in different stages of a mechanical cycle. In 2012, Scheres developed a software package called RELION to disentangle these conformations, which has become the most popular statistical processing tool for cryo-EM data.
Cryo-EM can also image biomolecules in forms that are closer to their natural, physiological states — unlike crystallography, where researchers often modify structures to make them crystallize.
In July, for example, Scheres unveiled the first cryo-EM structure of tau filaments, which are implicated in Alzheimer disease (Fig. 1). Whereas most previous studies have generated tau filaments in vitro, Scheres sourced his from the brain of a single Alzheimer disease patient. The study revealed that one stretch of amino acids might act as a template for growing aggregates, suggesting that a drug that targets this sequence could halt aggregation (Nature 547, 185–190; 2017).
Cryo-EM doesn't need much starting protein to work with, either. This enabled Jake Baum, of Imperial College London, and Scheres to recently figure out the previously unknown mechanism of action of the malaria drug mefloquine. “One of the gifts of cryo-EM is that we could never grow enough material from the parasite for crystallography,” says Baum.
Using cryo-EM they showed that mefloquine binds to a ribosome of Plasmodium falciparum, the parasite that causes malaria (Nat. Microbiol. 2, 17031; 2017). The team then used their cryo-EM structure to redesign mefloquine so that it bound more strongly to the ribosome, making a derivative that was twice as effective at killing the parasite. This is an important proof of principle that cryo-EM can aid structure-based drug design, says Baum.
Industry zooms in
Despite the growing number of cryo-EM structures, Subramaniam thinks that the technology is still years away from becoming a routine drug development tool. One major problem is that it is finicky work to prepare the samples, requiring as much art as science. It also has a steep learning curve. Nevertheless, Scheres thinks that drug developers are wise to get into cryo-EM now, so that they are up to speed when the methods and technology have matured. “There's still better optics coming in, better detectors, better software — so give it 5 to 10 years,” he says.
Pharma and biotech companies are already heeding that call. Genentech is busy assembling an in-house cryo-EM team. Last year, Pfizer started using a £5 million Titan Krios, the Rolls-Royce of cryo-EM. Novartis followed suit in November 2016 in collaboration with the Friedrich Miescher Institute. “For us, it's already been of use,” says Christian Wiesmann, director of protein science at the Novartis Institutes for BioMedical Research. The structure of a protein bound to a small molecule has helped to guide drug development chemistry, he says, work that he hopes to publish soon.
Other companies have embraced a consortium model — partly to manage costs, which remain a major barrier, but also to share expertise as they get to grips with the technique. In the UK, the Cambridge Pharmaceutical Cryo-EM Consortium launched in 2016, with more than £3 million in funding and in kind contributions from five pharmaceutical companies, from the MRC LMB, the University of Cambridge and FEI, the company that makes the Titan Krios. “There are huge advantages to doing this in a collaboration,” says Chris Phillips, associate director of structural biology at AstraZeneca in Cambridge, a consortium member. Because the technology is improving so rapidly, the hardware and software they use are continually being updated to deliver state of the art performance, he says.
In May, AstraZeneca unveiled the first fruits of its foray into cryo-EM: the structure of human ataxia telangiectasia mutated (ATM) protein, a large kinase that triggers DNA repair and is implicated in cancer (Sci. Adv. 3, e1700933; 2017). The resolution of 4.4 Å was enough to show ATM in two conformations — one open and ready to bind to a substrate, and the other closed — proving that the protein acts as a molecular switch. “This is the first time that we've been able to build a pretty high-resolution structure of the protein,” says Phillips.
Another Cambridge consortium member, Heptares, is using cryo-EM to study G protein-coupled receptors (GPCRs), which carry signals across cell membranes and are linked to a wide range of diseases. It is notoriously difficult to solve the structure of GPCRs by crystallography. Removed from their cell membrane, they lose their structure and activity; and when bound to their G protein partners, they are huge and floppy, making crystallization challenging.
These chunky complexes make good targets for cryo-EM, though. Not only can the new technique image structures that would not be possible with crystallography, but it can also take snapshots of subdomains in different positions so that researchers can select the relevant conformation for computational docking studies. For now, though, the best resolution cryo-EM GPCR structure clocks in at 3.8 Å, good enough to assess interactions with larger peptide ligands but missing the detail needed for small molecules. “It's not yet at the resolution where we can do proper drug discovery,” says Marshall.
Even as resolutions improve, the long-term place of cryo-EM in the drug discovery tool kit is still up for debate. The methodology is on track to become a useful complementary approach to crystallography and NMR, says Wiesmann. “But will cryo-EM replace crystallography? I don't think it will,” he adds. To optimize a lead compound, research labs need to routinely and rapidly solve dozens of structures of a protein bound to different ligands, to find which offers the most favourable interactions. For now, cryo-EM sample preparation and structure acquisition are still achingly slow.
Yet Astex, a partner in the Cambridge consortium, is betting that researchers can speed up the process. The company has built its business on high-throughput X-ray structure analysis for fragment-based drug design, using innovations such as robotic crystallization and automated crystal analysis to increase its output.
If cryo-EM sample preparation can become more automated, and next-generation electron detectors deliver the expected performance, Jhoti reckons cryo-EM could become indispensable. “I wouldn't be surprised if it superseded most other techniques for visualizing proteins,” he says. “But there's a lot of work to do first”.