Structural biology in context

It was 2019, and Josephine Botsch was in a hurry to get back to the lab. In fact, she was in such a hurry that Boston-area police pulled her over for speeding. A ticket was the least of her worries.

“I just thought, ‘OK, please don’t look in the back,’ ” says Botsch, now a PhD student at the Max Planck Institute of Biochemistry. She worried about what the police would make of her cargo: a trunk full of iceboxes holding cow tracheae she’d just picked up at a Boston-area slaughterhouse. “It was just very bloody,” she says.

It was going to take an hour and a half to return to Cambridge, Massachusetts, where Botsch worked in Alan Brown’s structural biology lab at Harvard Medical School. Botsch needed the tracheae as fresh as possible when she got there. She was going to study the molecular makeup of the tiny, hairlike structures called cilia that line the trachea’s surface.

Fortunately for that day’s experiment, the police took one look at Botsch’s driver’s license and sent her on her way—back to one of the many laboratories using cryo-electron microscopy (cryo-EM) to understand biological molecules. In the past decade, the Nobel Prize–winning technique has revolutionized structural biology and become a mainstay of structural studies.

But scientists are continually seeking more detailed and contextualized views of the systems they study. Now researchers are making advances in a related field, cryo-electron tomography (cryo-ET), to study molecular machines in the crowded, complex cellular context where they operate.

Neither technique is without challenges, but researchers are pushing toward a future where they can use the two, often in combination with other techniques, to determine the shape and composition of biological structures within cells and even within tissues.

Cryo-EM becomes a tool

For decades, scientists trying to determine the structures of proteins—knowledge key to understanding how they function—needed to work with purified solutions of single proteins. “Traditionally, structural biology has been a bit reductionist,” says Brown, Botsch’s former boss. “When you’re studying individual proteins, you’re not really seeing them functioning in their wider context.”

Consider, for example, the structure that Brown’s lab studies. Throughout the body, cilia on the surfaces of cells beat like a rowing crew to move fluids about. Inside them are cylindrical structures called axonemes, bundles of structural proteins linked together with a complex internal organization.

For some years, Brown’s lab has studied subunits of the axoneme to try to uncover how cilia can bend dramatically without breaking and how they might differ between tissues. These days, the lab uses cryo-EM. In this subtype of transmission electron microscopy, scientists first capture a protein or complex in a thin layer of glassy, noncrystalline ice and then bombard it with electrons. Researchers are left with a micrograph showing where those electrons encountered resistance on their way through.

Those raw images can be difficult to interpret. Biological samples are delicate compared with other materials, so researchers must use a relatively weak electron beam to avoid damaging them. That gives cryo-EM a very low signal-to-noise ratio. Jaap Brink, a product manager at the scientific instrument company Jeol USA, says that years ago, “we’d call [cryo-EM] ‘sorting potatoes’ because you’re looking at stuff that has almost no distinct features.”

All that changed about a decade ago, during what’s now called the resolution revolution. Now those potato-shaped blobs have sharpened into waffle fries, with crisply defined features. Researchers can define those features at atomic resolution; in a recent paper, a group at Riken showed a proton poised between a hydrogen bond donor and acceptor within a protein (Commun. Chem. 2023, DOI: 10.1038/s42004-023-00900-x).

Improvements that raised the signal-to-noise ratio in instruments were key to this revolution. Microscope developers introduced new cameras with direct electron detectors, boosting the strength of signals.

At the same time, image analysis software improved dramatically. Researchers now produce structural models by aligning electron micrographs of many copies of the same protein or group of proteins. Using algorithms and better statistical methods, researchers can compile an average from thousands of images of proteins in different orientations relative to the microscope.

Those advances have profoundly affected structural biology, says Emmanuel Smith, a senior application specialist at Jeol. “Suddenly all those impossible-to-characterize proteins were getting solved through cryo-EM,” he says. And the technology has continued to improve.

“We can do so much now in comparison to 10 years ago,” says Edward Yu, a structural biologist at Case Western Reserve University. The latest cameras can record a video instead of a static image, and software advances let researchers consider multiple conformations instead of just one for each protein or complex. Meanwhile, machine learning tools have sped up analysis workflows and enabled researchers to identify unknown proteins in a sample.

With faster and more automated data collection, as well as increasingly streamlined data analysis, cryo-EM is “really transitioning from an instrument to a tool used by biologists,” says Jeff Lengyel, a director of life sciences electron microscopy at the instrument manufacturer Thermo Fisher Scientific.

But cryo-EM has not overcome all its hurdles. Challenges in sample preparation can slow data collection and reduce the efficiency of automated EM systems. There’s also a limit to the complexity of particles that can be analyzed—and some problems don’t lend themselves to molecular dissection.

In Brown’s lab, with the cilia backbones, postdoctoral scholar Travis Walton optimized a way to biochemically separate axonemes from algal flagella to study them with cryo-EM. The data he collected became part of an all-atom model of axonemes from algal flagella and human respiratory cilia that Brown’s lab published last year (Nature 2023, DOI: 10.1038/s41586-023-06140-2).

But even in that final structure, some regions remained blurry. One example is a small protein that, during sample preparation, becomes separated from one of two larger bundles that it bridges. Without the stabilizing connection, Walton says, “it’s flapping in the breeze,” too unstable to average into a high-resolution structure. To understand tricky proteins like this, researchers need to look at more-complex samples using different techniques.

Tomography prepares for a leap

The related technique cryo-ET is gaining in popularity because it offers the ability to study biological structures without plucking them out of their cellular context and breaking up weak interactions. “A lot of the projects that we were doing with cryo-EM we’re now doing with tomography—particularly in cells,” Brown says.

For cryo-ET experiments, structural biologists use the same kind of electron microscopes as they do for cryo-EM. But instead of a solution of small particles in a thin liquid layer, the sample can be a thick slice of a cell, densely packed with a mixture of molecules.

By collecting images from different angles, researchers can reconstruct a tomogram—a 3D representation of the structures inside.

For a while, tomography methods lagged behind improvements in cryo-EM, experts say, but they’re now beginning to take off. Some observers say the method is having a resolution revolution of its own. “It’s not an easy technique, but we’re really at the onset of that wave of very substantial improvements,” says Steve Reyntjens, a director of life sciences electron microscopy at Thermo Fisher.

“You can now look at things that cannot be purified, that have to exist in the cell,” Lengyel, Reyntjens’s colleague, says. “These are projects you couldn’t even envision a few years ago.”

Although tomography opens new avenues of study, it brings its own challenges. Data collection is slow compared with that for cryo-EM—partly because of the constraints of sample preparation and partly because of how researchers obtain a 3D view of a sample.

To capture the sample from many points of view, researchers physically tilt it within the microscope. Then programs align and compile information from the series of micrographs into a volume representation of the cell’s features.

The analytical workflows are still not fully settled, but they are improving. “We’re in a fantastic era now where the software is just developing almost weekly,” Brown says. The field’s rapid pace introduces a dizzying array of choices for researchers, with a side of friction when they try to transfer data between software packages not designed to communicate.

Finding what you’re looking for

The resolution that researchers can achieve using tomography is still much lower than with cryo-EM, largely because samples suffer from more severe radiation damage. Training a several-​hundred-kiloelectron-volt electron beam on a biological molecule tends to damage the same delicate features that researchers are interested in. So ideally, the data collection is the first time any given region of the sample encounters the beam.

Kem Sochacki, whose team in Justin Taraska’s lab at the US National Heart, Lung, and Blood Institute is working on new cryo-ET methods to study vesicle trafficking, says that’s a big problem for researchers. “One of the major difficulties with cryo-EM and cryo-ET techniques is we’re all trying to avoid frying the sample,” she says. “We can’t actually look at it at high resolution before we image it—which, it turns out, is problematic.”

One issue is that researchers can feasibly image only a small fraction of a cell. Most cells are too large and densely packed with molecules and so thick that they are almost opaque to transmitted electrons. So researchers have to carve out single slices roughly 200 nm deep.

To cut out such slices, researchers often use a focused beam of gallium ions to blast away material from above and below the area of interest. The protocol, known as focused ion beam milling, leaves a section just thin enough to image.

Obliterating the material above and below without destroying the sample in between is a delicate and very slow process, like nibbling away both cookies from an Oreo without damaging the cream—and recent research shows that the slice that remains does sustain some damage (Proc. Natl. Acad. Sci. U.S.A. 2023, DOI: 10.1073/pnas.2301852120). On a good day, Sochacki says, a researcher might mill just eight samples.

Newer instruments that use argon or xenon instead of gallium ion sources work faster. But the researchers in Taraska’s lab are developing an alternative approach. They use a measured shear stress—essentially a water gun—to isolate cell membranes. Though the sample preparation process can be used to study only phenomena that happen near the cell membrane, Sochacki calls the approach “faster than any cellular tomography pipeline out there.”

But speedier slicing does not solve the more fundamental problem of where a researcher should look to find features of interest within a cell.

According to Bronwyn Lucas, a biochemist and computational biologist at the University of California, Berkeley, many scientists mill out sections almost at random and study structures they can be sure of finding almost anywhere, such as cytoskeleton components or ribosomes. “You run the statistics game” in scanning through slices to identify these structures “and then hope that enough of them will have the thing that you’re looking for,” Lucas says.

But flagging less-common cell features is one of the most important challenges ahead for the field, Taraska says. To assist in finding features of interest, some labs use fluorescent labels to locate objects in a cell. Then they use a technique known as correlative light and electron microscopy, or CLEM, to image the same area at high resolution. Because of limits imposed by the wavelengths of visible light, the researchers can’t currently visualize those labels at the same tiny scale as EM shows. “Basically, fluorescence can only tell us where we should go take a picture,” Sochacki says.

Once they have that picture—or a series of micrographs at different tilt angles—researchers come to the second major problem in tomography: figuring out what they’re looking at.

Knowing what you’re looking at

In a small office space off to the side of Taraska’s lab in a US National Institutes of Health building in Bethesda, Maryland, Sochacki and postdoc Dennis Michalak sit in front of an extralong, concave monitor. On screen, a tomogram of a cell’s plasma membrane appears as a pointillistic bulk against a pale background. As Michalak scrolls through the projection, features take gauzy shape and then fade away.

“The low resolution of the actual tomogram makes it difficult to identify most things,” Lucas says. “The main reason why there’s been a lot of the ribosome structures and a lot of cytoskeletal structures [published] is because those are things that we can find.”

The honeycomb structure of clathrin, for example, is easy to pick out on Michalak’s screen. That’s what the lab usually studies. But today, Michalak is looking for a well-characterized complex so he can test a new analysis workflow. From this hazy representation of the cell he will pluck particles that—through iterative averaging and image analysis—sharpen into a clear, high-resolution model of a ribosome.

The software he is working on involves averaging distinctive features to boost the overall resolution of a model. “The goal is to be able to do something like [single-particle] cryo-EM in a tomogram,” Taraska explains. “You can extract millions of particles from an image of the cell, classify them all into their different types and structures, and then solve atomic structures of those objects.”

Michalak extracts particles that are probably ribosomes and aligns them together to make a composite 3D model with higher resolution than the component images.

“Subtomogram averaging analysis usually gives you multiple structures,” Michalak says. Sometimes the differences are subtle, and sometimes they are large. By looking back at where the particles that make up each structure were located in the broader tomogram, researchers can sometimes gain insight into the context for each structure.

Other techniques are also helping researchers piece together the molecular makeup of the 3D shapes they image.

Docking algorithms iteratively fit known or predicted structures into the contours of a 3D model, trying different proteins in different orientations within each blob until they find a promising match. This approach works best when starting with a list of known component proteins.

If the component proteins first need to be identified, researchers can use mass spectrometry–based proteomics to determine a list of candidates to dock. But researchers can also now use programs to identify the proteins in a structure entirely from scratch.

New atomic modeling software uses electron density to estimate the likelihood that each site in a protein is a given amino acid, says Sjors Scheres, a structural biologist at the MRC Laboratory of Molecular Biology. Then researchers can use combined probabilities to search entire proteomes and “fish out very, very accurately which protein is present,” Scheres says.

Researchers at the University of California, San Francisco, recently used this software to identify new proteins within the axoneme in sperm tails (Cell 2023, DOI: 10.1016/j.cell.2023.09.017). After constructing tomograms of sperm tails from mice, the researchers docked AlphaFold predictions of every protein structure in the mouse proteome into the axoneme. By doing that, the researchers identified flexible cross-linking proteins inside the tube that appear to give the structure extra strength for bending in many directions.

Finding new proteins within a cellular context and finding out how they contribute to biological functions is exactly what the field hopes to do more of.

Toward a structure of the whole cell

Funders appear to find cryo-ET promising and have allocated money to research groups that are helping improve the technique. In 2021, the Chan Zuckerberg Initiative awarded grants totaling $27.5 million for new method development. Some of the funds supported the Taraska lab’s work. More recently, the NIH’s cryo-EM program announced a new national cryo-ET network.

Though there are still problems to solve, structural techniques that can examine proteins and complexes in context are beginning to yield answers to biological questions that have long eluded researchers. Atomic-level structures of large complexes can guide drug design, lend an understanding of how complexes form and what they do after having formed, and give insight into how proteins work together.

In the Brown lab, which studies cilia and flagella, structural biology in context has identified proteins that were not previously known to participate in cilia. Working with geneticists, the team has incorporated those insights into genetic screens for ciliopathies that can cause lung infections and infertility, helping physicians give more patients a definitive molecular diagnosis.

The researchers have also begun to work in the opposite direction, studying organoids from patients with unresolved ciliopathies to try to determine why their cilia don’t work well.

Meanwhile, some researchers are interested in an even larger-scale approach, one that Lucas describes as determining the structural organization of a whole section of a cell.

To understand this organization, researchers must identify all the molecules present in a given slice of cell and determine how they connect and interact—even if there are not enough copies of a protein to average into a high-resolution structure. Lucas’s lab is one of several developing methods to spot molecules in micrographs of a cell by matching them to a reference proteome of structures.

That’s an ability, not yet fully realized, that funders call visual proteomics. It will become possible if hardware and software improve enough that researchers can reliably locate small proteins, says Stefan Raunser, a structural biologist at the Max Planck Institute of Molecular Physiology. Then researchers may be able to identify most or all of the proteins present using information from electron microscopy instead of current techniques.

Researchers imagine using visual proteomics to answer questions beyond those that structural biology has historically been able to answer. For example, they’d like to determine how the overall structure and composition of a synapse change after a neuron is activated or how a membraneless organelle orders its components.

Whether the focus is on parts of the cell or the whole, studying more-complex samples may be difficult, Taraska says, but it is key to the future of biochemistry. “To understand how molecules are behaving—how they’re actually acting inside a living cell inside of an organism—you need to understand how all the parts come together.”