Cryo-electron tomography, or cryo-ET, produces three-dimensional images of biological molecules inside cells that have been flash-frozen to preserve their natural state, and recent advances have pushed the technique’s resolution past the point where individual atoms become distinguishable. By tilting a frozen sample under an electron beam and collecting a series of images from different angles, researchers reconstruct a volume that reveals how proteins, membranes, and other molecular machinery are arranged in their actual cellular surroundings. What makes recent progress so striking is that improvements have arrived simultaneously across sample preparation, detector hardware, computational processing, and automation, compounding one another in ways that have transformed cryo-ET from a niche method into a central tool of structural biology.
How Cryo-ET Differs from Single-Particle Cryo-EM
Both cryo-ET and single-particle cryo-EM use electrons to image frozen biological samples, but they answer different kinds of questions. Single-particle cryo-EM has traditionally focused on highly purified molecules spread across a thin layer of ice: tens of thousands of identical copies are imaged, and a computer averages them together to produce a high-resolution structure. Cryo-ET, by contrast, images molecules inside intact or minimally processed cells, capturing them in their native surroundings rather than in isolation.1PubMed Central. Combining per-particle cryo-ET and cryo-EM single particle analysis to elucidate heterogeneous DNA-protein organization The trade-off has historically been resolution: single-particle methods routinely reach 2–3 Å, while cryo-ET tomograms on their own are noisier and lower-resolution because the sample can only tolerate a limited electron dose before radiation damages it. A processing method called subtomogram averaging narrows that gap by computationally aligning and averaging many copies of the same molecule extracted from tomograms, now reaching beyond 3 Å in favorable cases.2PubMed Central. Applications and prospects of cryo-electron tomography in drug discovery and understanding disease
The real power of cryo-ET is context. A purified protein tells you what a molecule looks like on its own. A tomogram shows you where it sits relative to a membrane, how it clusters with partners, and what shape the surrounding organelles take. That contextual information is often the part that matters for understanding how diseases progress or how drugs act.
Preparing the Sample Without Destroying It
The biggest practical bottleneck in cryo-ET has long been sample preparation. Cells are thick, and electrons cannot penetrate more than a few hundred nanometers of ice without losing the information needed for a clear image. The standard solution is to use a focused ion beam, or FIB, to shave frozen cells down into ultra-thin slabs called lamellae, typically 100–250 nm thick.3Nature Protocols. Preparing samples from whole cells using focused-ion-beam milling for cryo-electron tomography This cryo-FIB approach opened a window into previously inaccessible cellular interiors without the chemical fixation or staining that distorts structures in traditional electron microscopy.4PubMed Central. Site-Specific Cryo-focused Ion Beam Sample Preparation Guided by 3D Correlative Microscopy
A newer generation of instruments uses plasma ion sources instead of the gallium beams found in conventional FIB systems. Plasma FIB, or PFIB, can mill faster and handle thicker starting material, which is especially useful when working with high-pressure frozen tissue samples rather than thin cell monolayers. One study demonstrated that an argon plasma source produced lamellae with an 85% success rate and kept ice contamination below 2 nm per hour for weeks, enabling datasets of hundreds of tomograms from a single preparation session. The resulting data were good enough to resolve the human 80S ribosome at about 4.9 Ã…, with well-ordered regions approaching 3.8 Ã….5Nature Communications. Plasma FIB milling for the determination of structures in situ A parallel effort using xenon plasma achieved a 4.0 Ã… structure of the bacterial ribosome on lamellae milled from high-pressure frozen samples, showing that the plasma approach works across different organisms and freezing methods.6Nature Communications. Xenon plasma focused ion beam lamella fabrication on high-pressure frozen specimens for structural cell biology
These throughput gains matter beyond raw speed. In cryo-ET, statistical power comes from the number of molecules you can image and average. Producing hundreds of lamellae instead of a handful means researchers can attempt subtomogram averaging on structures that are rare inside cells, not just abundant ones like ribosomes.
Better Detectors and Phase Plates
The electron detectors used in cryo-ET have undergone a quiet revolution. Direct electron detectors, which register electrons without the intermediate scintillator layer of older cameras, dramatically improved the signal captured in each image. When combined with energy filters that remove inelastically scattered electrons before they reach the detector, these systems extract more useful information from every dose of electrons hitting the sample.7PubMed Central. Exploring high-resolution cryo-ET and subtomogram averaging capabilities of contemporary DEDs Because dose is the fundamental currency of cryo-ET (you cannot give the sample more electrons without damaging it), squeezing more signal from each electron is one of the most effective ways to improve resolution.
Another hardware innovation is the Volta phase plate, a thin carbon film placed in the back focal plane of the microscope. It shifts the phase of scattered electrons relative to unscattered ones, boosting contrast in the resulting images, particularly for the large-scale features that are essential for accurately aligning tilt series.8PubMed Central. Volta potential phase plate for in-focus phase contrast transmission electron microscopy Recent systematic evaluations show that while the Volta phase plate does cause some loss of high-frequency detail and moderate signal dampening, it enhances exactly the low-frequency information bands that matter most for tilt-series alignment.9PubMed Central. Evaluating the Volta phase plate for improved tomogram alignment in cryo-electron tomography For small or low-contrast targets that are difficult to see in traditional defocus-based imaging, the phase plate can make the difference between being able to find the molecule of interest and not seeing it at all.10eLife. Using the Volta phase plate with defocus for cryo-EM single particle analysis
Taming the Stage and the Tilt Series
A tilt series in cryo-ET means physically rotating the sample stage while collecting images at each angle. Every rotation introduces mechanical instabilities: the stage jolts, drifts, and settles. Careful characterization of stage behavior has shown that each tilt produces a brief jolt of about 10 Ã… perpendicular to the tilt axis, lasting roughly 0.04 seconds. After that initial jolt, stage movement drops below 2 Ã… per frame and falls to hundredths of an Ã… per frame within about a second.11PubMed Central. Rapid Tilt-Series Method for Cryo-Electron Tomography: Characterizing Stage Behavior During FISE Acquisition Understanding this behavior precisely allows acquisition software to time exposures to avoid the worst drift, recovering image quality that would otherwise be lost.
New acquisition strategies exploit this knowledge. Continuous-tilt methods, where the stage rotates smoothly rather than stopping at discrete angles, can reduce the total time per tilt series and the number of mechanical settling events. Beam-image-shift techniques allow multiple nearby areas to be imaged at each tilt angle without physically moving the stage between them, effectively collecting several tilt series in parallel.12PubMed Central. Advances in automation for cryo-electron tomography data collection Together, these approaches have turned what used to be a multi-day data collection effort into something that can yield hundreds of tomograms per session.
Computational Processing and AI Denoising
Raw cryo-ET tomograms are extremely noisy. Each image in the tilt series receives only a fraction of the total electron dose, and the tilting geometry means that information from high angles is missing entirely, creating what is known as the missing wedge, a gap in the reconstructed volume that elongates features along one axis and limits resolution.13PubMed Central. MBIR: A cryo-ET 3D reconstruction method that effectively minimizes missing wedge artifacts and restores missing information Reconstruction algorithms that account for this geometry can partially recover the lost information, and GPU-accelerated iterative methods have made these computationally intensive approaches practical on lab-scale hardware.14PubMed Central. High-performance iterative electron tomography reconstruction with long-object compensation using graphics processing units (GPUs)
Deep learning has become an increasingly important part of the processing pipeline. Topaz-Denoise, for example, uses a neural network trained on paired even/odd frame tomograms to suppress noise without relying on any prior structural model. The developers trained general-purpose 3D denoising models on 32 tilt series collected from diverse biological samples, producing pre-trained networks that work across different sample types and imaging conditions.15Nature Communications. Topaz-Denoise: general deep denoising models for cryoEM and cryoET The practical impact is substantial: denoised tomograms let researchers identify and pick particles that would be invisible in the raw data, feeding better inputs into downstream subtomogram averaging.
Subtomogram averaging itself has matured considerably. By extracting thousands of copies of a given molecule from many tomograms, aligning them computationally, and averaging, researchers can break through the noise barrier of individual tomograms. When combined with FIB milling, this approach can resolve protein complexes at near-atomic resolution inside cells that have never been disrupted.16PubMed. Multishot tomography for high-resolution in situ subtomogram averaging The best published in-situ subtomogram averages now push past 4 Ã…, a resolution at which amino acid side chains begin to become visible and where drug-binding pockets can be meaningfully interpreted.17PubMed. Toward high-resolution in situ structural biology with cryo-electron tomography and subtomogram averaging
Finding the Needle with Correlative Microscopy
A recurring challenge in cryo-ET is targeting. A frozen cell is a vast landscape at the molecular scale, and the structure you want to image might occupy only a tiny fraction of it. Correlative light and electron microscopy, or CLEM, solves this by first locating a feature using fluorescence microscopy, then guiding FIB milling and tomography to exactly that spot. One team used fluorescently tagged proteins to find hexagonal lattice structures formed by the antiviral protein TRIM5α inside intact mammalian cells, structures that would have been virtually impossible to locate by electron microscopy alone.18PubMed Central. Correlated cryogenic fluorescence microscopy and electron cryo-tomography shows that exogenous TRIM5α can form hexagonal lattices or autophagy aggregates in vivo
Cryo-CLEM workflows have expanded beyond cultured cells. A recent method for tissue samples uses a physical pattern called a FinderTOP, imprinted into the sample surface during high-pressure freezing, to register fluorescence images with FIB/SEM volume imaging at cryogenic temperatures.19Communications Biology. Precise targeting for 3D cryo-correlative light and electron microscopy volume imaging of tissues using a FinderTOP In plant biology, a cryo-CLEM approach was used to image specialized cell wall structures called Casparian strips in root tissue, features confined to a single cell layer only 2–3 µm thick within a root cylinder over 80 µm across. Fluorescence markers guided cryo-lift-out to reliably produce lamellae containing these rare structures.20Nature Communications. Targeted imaging of specialized plant cell walls by improved cryo-CLEM and cryo-electron tomography
Watching Viruses from the Inside
Virology has been one of the most productive application areas for cryo-ET, especially for studying how viruses hijack cellular machinery. Instead of examining purified virus particles in isolation, researchers can now see the entire replication cycle playing out within infected cells.21PubMed Central. Visualizing the virus world inside the cell by cryo-electron tomography During the COVID-19 pandemic, cryo-ET was used to characterize SARS-CoV-2 replication compartments in situ. Researchers directly visualized RNA filaments inside double-membrane vesicles, the compartments where the virus copies its genome, and observed that the filament diameters were consistent with double-stranded RNA. They also showed that assembled spike trimers in internal membrane compartments do not by themselves bend the membrane but instead reorganize laterally during virion assembly, with the viral ribonucleoprotein complexes accumulating preferentially at curved membrane sites.22Nature Communications. SARS-CoV-2 structure and replication characterized by in situ cryo-electron tomography
Work on flaviviruses has similarly benefited. Cryo-ET of infected cells, plunge-frozen and FIB-milled at 24 hours post infection, revealed dilated endoplasmic reticulum containing clustered replication organelles, identifiable as near-spherical membrane invaginations not present in uninfected cells.23Nature Communications. Cryo-electron tomography reveals coupled flavivirus replication, budding and maturation These studies move virology past static snapshots of purified particles toward dynamic, contextual views of infection.
Protein Aggregates and Neurodegeneration
One of the more striking cryo-ET applications has been in neurodegenerative disease, where protein aggregation is a hallmark. Cryo-ET of inclusion bodies in cellular models of Huntington’s disease revealed that both cytoplasmic and nuclear inclusions consist of radially arranged amyloid-like fibrils, each about 7–8 nm in diameter and 100–200 nm long. The fibrils made direct contact with the membranes of surrounding organelles, particularly the endoplasmic reticulum, causing extreme membrane curvature at the contact sites. Ribosome-free ER tubes protruded into the inclusions, interacting extensively with the fibril network, and numerous irregularly shaped vesicles were embedded at sites where fibrils appeared to have disrupted organelle membranes.24Trends in Cell Biology. Seeing the Inside: Cryo-Electron Tomography of Protein Aggregates in Neurodegeneration This kind of finding, showing that protein fibrils physically damage surrounding organelles, would be invisible to any technique that requires breaking cells open first.
Capturing Molecular Flexibility in Native Settings
A distinctive advantage of cryo-ET is that each particle captured in a tomogram is a unique snapshot of whatever conformation the molecule happened to be in at the moment of freezing. Traditional subtomogram averaging blurs this out by aligning everything to a single structure, but newer computational methods are learning to preserve the conformational diversity. MDTOMO, a method that combines molecular dynamics simulations with subtomogram data, was used to map the continuous conformational landscape of SARS-CoV-2 spike glycoproteins directly on virus surfaces, detecting independent motions of individual spike domains that averaging-based approaches miss entirely.25PubMed Central. MDTOMO method for continuous conformational variability analysis in cryo electron subtomograms based on molecular dynamics simulations26Current Opinion in Structural Biology. Conformational landscapes in cryo-electron tomography data based on molecular dynamics simulations Separately, cryo-ET of active SARS-CoV-2 virions at a biosafety level 3 facility revealed transient open-trimer prefusion states of the spike protein, a hidden flexibility not previously observed, because the virus had never been imaged intact and infectious before.27bioRxiv. Native spike flexibility revealed by BSL3 Cryo-ET of active SARS-CoV-2 virions
This shift from static structures to conformational landscapes is consequential for drug design. A drug that targets a single rigid conformation of a protein might miss the conformations that actually dominate in vivo. Cryo-ET, when paired with methods like MDTOMO, gives researchers a way to see which conformations are common and which are rare inside living systems.
Integrating AI-Predicted Structures with Experimental Maps
AlphaFold and similar deep-learning tools predict protein structures from amino acid sequences, but predictions alone do not reveal how those proteins are arranged in larger complexes or inside cells. Cryo-ET and cryo-EM maps provide the experimental envelope, and predicted structures can be fit into those envelopes to produce atomic models of assemblies that are too large or too heterogeneous for any single technique. A landmark demonstration of this approach produced a nearly complete atomic model of the cytoplasmic ring of the nuclear pore complex from the frog Xenopus laevis. The team reconstructed the ring at about 7 Ã… resolution using cryo-EM, predicted the structures of individual nucleoporin subunits with AlphaFold, and placed them into the density to build a model containing multiple copies of several distinct protein complexes.28PubMed Central. Structure of cytoplasmic ring of nuclear pore complex by integrative cryo-EM and AlphaFold
Deep-learning tools are also being developed to enhance the maps themselves. CryoFEM, for instance, uses neural networks to improve the features in cryo-EM density maps before AlphaFold models are fit into them, improving model accuracy in cases where the initial predictions are less precise.29Briefings in Bioinformatics. Integrating AlphaFold and deep learning for atomistic interpretation of cryo-EM maps As cryo-ET resolution continues to climb and AI predictions improve, the gap between a raw tomogram and an atomic-level understanding of a cellular scene keeps shrinking.
Drug Discovery and Seeing Drug Action in Context
Pharmaceutical applications of cryo-ET are beginning to move from speculative to practical. The ability to visualize macromolecules inside cells at resolutions beyond 3 Ã… makes it possible, in principle, to see how a drug interacts with its target in a physiologically relevant setting rather than in a crystal lattice or a purified sample.2PubMed Central. Applications and prospects of cryo-electron tomography in drug discovery and understanding disease A recent proof-of-concept study used protein nanocages as labels to visualize how a membrane receptor, EGFR, assembles and activates on native cell membranes, demonstrating that cryo-ET can capture the structural consequences of receptor clustering induced by an external stimulus.30PubMed. Visualization of EGFR Assembly and Activation Induced by a Protein Nanocage Using Cryo-Electron Tomography For drug developers, the appeal is obvious: you could potentially watch a drug candidate engage its target inside a cell, see whether it changes the target’s conformation or its interactions with neighbors, and identify off-target binding, all without the artifacts introduced by purification.
Automation and the Push Toward High Throughput
Historically, cryo-ET was slow enough that collecting a single dataset was a significant undertaking. Current automation efforts aim to change that by reducing human intervention at every stage. Machine learning-driven targeting selects areas of interest on grids, real-time predictive adjustments correct for drift during acquisition, and web-based interfaces allow microscopes to be operated remotely. Montage tomography stitches images from adjacent areas to increase the observable cellular area per tilt series, while innovations like rectangular condenser apertures improve dose efficiency by matching the illuminated area more closely to the detector’s field of view.12PubMed Central. Advances in automation for cryo-electron tomography data collection
Infrastructure for data sharing is growing alongside the instruments. CryoCRAB, a curated dataset designed for training deep-learning models on cryo-EM data, contains raw movie frames from over 700 proteins, totaling more than 116 terabytes, with each movie split into odd and even frames to provide natural training pairs for denoising networks.31PubMed Central. A large-scale curated and filterable dataset for cryo-EM foundation model pre-training The availability of large, standardized training datasets is important because the AI tools now embedded throughout the cryo-ET pipeline, from denoising to particle picking to map interpretation, all improve as the data they train on becomes more diverse and better curated.
Remaining Limitations
For all the progress, cryo-ET still operates under hard physical constraints. Radiation damage caps the total electron dose a sample can receive, and the limited tilt range (typically ±60–70° rather than a full 180°) creates the missing wedge, which distorts reconstructions along one axis.13PubMed Central. MBIR: A cryo-ET 3D reconstruction method that effectively minimizes missing wedge artifacts and restores missing information FIB milling, while dramatically improved, still fails a fraction of the time and requires expensive, specialized instruments. The best subtomogram averaging resolutions are achieved only for abundant, well-ordered complexes like ribosomes; flexible or rare targets remain much harder. And throughput, though increasing, still lags far behind single-particle cryo-EM, where thousands of micrographs can be collected in hours.
The computational demands are also steep. Processing hundreds of tomograms, each a three-dimensional volume, requires substantial GPU resources and storage infrastructure. As datasets grow into the hundreds-of-terabytes range, the logistics of data management and analysis become challenges in their own right, ones that are being addressed in part by standardized formats and cloud-based processing pipelines but remain far from fully solved.
Mapping Molecular Landscapes Across Cell Types
Beyond individual molecules, cryo-ET has the unique ability to map how large numbers of different molecular species are organized relative to one another inside a cell, producing what some researchers call a molecular landscape. Early applications of this concept examined the cytoskeleton, revealing the three-dimensional organization of filament networks in both bacteria and eukaryotic cells at a resolution sufficient to distinguish different filament types and their spatial relationships.32PubMed. New insights into the structural organization of eukaryotic and prokaryotic cytoskeletons using cryo-electron tomography More recently, studies of membrane contact sites, regions where two organelles come close enough for lipid transfer, have used cryo-ET to measure that the endoplasmic reticulum and plasma membrane are separated by roughly 18–20 nm at contacts mediated by extended synaptotagmin proteins, and that this distance shrinks when calcium levels rise.33PubMed Central. Lipid transfer proteins: From molecular mechanisms to functional validation These nanometer-scale distance measurements carry functional meaning, dictating which protein bridges can span the gap and how efficiently lipids move between compartments, and they are essentially inaccessible to any other structural technique.