{
  "id": "cryo-electron-tomography",
  "title": "Cryo-electron Tomography",
  "lang": "en",
  "source_lang": "en",
  "summary": "Cryo-electron tomography reconstructs three-dimensional views of vitrified specimens, revealing cellular architecture and molecular assemblies in their spatial context.",
  "category": "science",
  "subcategory": "biology",
  "type": "concept",
  "aliases": [
    "Cryo-ET",
    "cryogenic electron tomography",
    "electron cryotomography",
    "ECT"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/cryo-electron-tomography/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/cryo-electron-tomography/",
      "json": "https://aiwiki.page/zh/cryo-electron-tomography.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Cryo-electron tomography (cryo-ET) is a three-dimensional imaging method in which a frozen, hydrated specimen is examined in an electron microscope from a series of viewing angles. Computational reconstruction combines these projections into a volume called a tomogram. As a branch of cryo-electron microscopy, it can visualize membranes, filaments, organelles, viruses, and large molecular complexes without the dehydration and heavy-metal staining used in many conventional electron-microscopy preparations.\n\nIts distinctive contribution is the observation of biological structure in spatial context. Rather than requiring many isolated copies of a purified protein complex, cryo-ET can examine individual assemblies inside a cell, including their positions, orientations, interactions, and structural variation. Repeated complexes can subsequently be aligned and averaged to recover finer molecular detail. The method therefore connects cellular ultrastructure with structural biology, although its capabilities are constrained by radiation damage, specimen thickness, incomplete angular sampling, and the difficulty of identifying molecules in crowded environments."
    },
    {
      "heading": "Historical development and relationship to other structural methods",
      "text": "Electron tomography developed from the mathematical principle that a three-dimensional object can be reconstructed from its projections. In 1968, David De Rosier and Aaron Klug demonstrated three-dimensional reconstruction from electron micrographs of biological structures, helping establish the computational foundations of three-dimensional electron microscopy. Subsequent developments made it possible to reconstruct nonsymmetric objects from images acquired while physically tilting a specimen. Early biological tomography commonly used chemically fixed, dehydrated, resin-embedded material, which tolerated preparation and imaging conditions quite different from those used for hydrated specimens.\n\nThe decisive preparation advance was the development of practical vitrification methods. In the early 1980s, work associated with Jacques Dubochet and colleagues showed how thin aqueous specimens could be frozen rapidly enough to suppress the formation of damaging ice crystals. A 1984 study by Marc Adrian, Dubochet, Jean Lepault, and Alasdair McDowall demonstrated electron microscopy of viruses in vitrified water. The resulting preparations retained hydration without requiring a crystalline ice matrix. Vitrification became foundational to both single-particle cryo-electron microscopy and cryo-electron tomography. Dubochet, Joachim Frank, and Richard Henderson received the 2017 Nobel Prize in Chemistry for developments in cryo-electron microscopy, although the award was not specifically for tomography.\n\nCryo-ET became increasingly practical through the combined development of stable cryogenic specimen holders, computer-controlled microscopes, automated tilt-series acquisition, and reconstruction software. During the 1990s and 2000s, improvements in automation reduced the labor required to keep the same specimen region centered and focused during repeated tilting. Direct electron detectors subsequently improved the recording of weak signals and enabled individual exposures to be divided into movie frames. Cryogenic focused-ion-beam preparation extended the technique from naturally thin specimens to selected regions inside larger cells.\n\nThe distinction from single-particle analysis concerns both sampling and inference. Single-particle analysis usually combines images of many nominally equivalent particles, each recorded in an initially unknown orientation. Cryo-ET records a known sequence of specimen tilts and reconstructs each field of view, retaining the arrangement of objects relative to one another. Subtomogram averaging then introduces an ensemble-analysis stage within that spatially resolved dataset.\n\nOther structural methods provide complementary information. X-ray crystallography derives molecular structures from diffraction by crystals, while nuclear magnetic resonance can characterize molecular structure and dynamics under suitable conditions. Conventional fluorescence microscopy can identify labeled components and follow living processes but does not directly reveal comparable molecular ultrastructure. Cryo-ET occupies an intermediate position: it resolves cellular organization at nanometre scales and can, for sufficiently abundant and well-behaved assemblies, support molecular reconstructions approaching atomic detail."
    },
    {
      "heading": "Image formation, specimen thickness, and radiation damage",
      "text": "Cryo-ET generally uses a transmission electron microscope, in which accelerated electrons pass through a specimen and electromagnetic lenses form an image. Instruments operating at accelerating voltages of 200 or 300 kilovolts are common in biological tomography. Electrons interact strongly with matter, making very thin biological material visible, but this interaction also limits the thickness that can be interpreted reliably. As thickness increases, multiple scattering and inelastic scattering reduce useful contrast and complicate the relationship between the recorded image and the underlying structure.\n\nBiological material embedded in vitreous water consists mainly of light elements and is a weak amplitude object under typical imaging conditions. Much of its useful signal is recorded through phase contrast: electrons scattered by the specimen interfere with the unscattered beam. Intentional defocus converts part of the specimen-induced phase modulation into measurable intensity variation. The microscope's contrast transfer function describes how different spatial frequencies are transmitted, attenuated, or inverted. Its oscillations introduce frequency-dependent distortions that must be estimated and corrected when high-resolution information is required.\n\nSpecimen tilt increases the path length through a flat slab. For a slab of thickness \\(t\\) tilted by angle \\(\\theta\\), an idealized path length is\n\n\\[t_{\\mathrm{eff}}=\\frac{t}{\\cos\\theta}.\\]\n\nAt \\(60^\\circ\\), the path is twice the untilted thickness. Consequently, high-tilt images usually have lower signal quality than low-tilt images. Energy filters can remove many inelastically scattered electrons, improving image contrast, but they cannot recover structural information already obscured by scattering or damage. Thickness limits are therefore practical rather than universal: acceptable thickness depends on accelerating voltage, composition, imaging conditions, and the resolution sought.\n\nElectron exposure also changes the specimen. Ionization and subsequent chemical reactions break bonds, alter sensitive molecular groups, and eventually produce larger-scale mass loss or bubbling. Cryogenic temperatures suppress many secondary movements and reactions but do not eliminate radiation damage. High-resolution information generally deteriorates before broad structural features disappear. A tilt series must therefore distribute a limited cumulative exposure over many images; total exposures of roughly 80–150 electrons per square ångström are common in biological workflows, although protocols vary substantially.\n\nThis creates a central trade-off. More electrons improve counting statistics, but excessive exposure destroys the structure being measured. More projections improve angular sampling, but each projection must receive a smaller fraction of the exposure budget. An individual tomogram consequently contains much less high-resolution signal than might be expected from its nominal voxel spacing. Sampling, visibility, and demonstrated structural resolution are distinct quantities."
    },
    {
      "heading": "Vitrification and preparation of biological specimens",
      "text": "The purpose of cryogenic preparation is to immobilize a hydrated specimen before substantial structural rearrangement occurs. In vitrification, cooling is sufficiently rapid that water solidifies into an amorphous state rather than forming ordinary crystalline ice. Ice crystals can displace biological material and introduce strong diffraction features. Vitrification does not make a specimen literally identical to its living state: it preserves a particular frozen configuration after handling, environmental changes, and cooling, all of which may influence the observed structure.\n\nFor thin suspensions, a small volume is applied to an electron-microscopy grid carrying a perforated support film. Excess liquid is removed, commonly by blotting, and the grid is rapidly plunged into a cryogen, often liquid ethane cooled near its freezing point. The remaining aqueous layer spans holes or lies over the support. Thin bacteria, viruses, isolated organelles, membrane vesicles, and purified assemblies may be imaged directly if their dimensions and surrounding ice permit sufficient electron transmission. Gold grids and other mechanically stable supports can reduce some forms of specimen movement.\n\nPreparation can introduce strong selection effects. Blotting changes local liquid thickness and solute concentrations, and particles may accumulate at the air–water interface. Proteins can denature, dissociate, or adopt preferred orientations at that interface. Cells may be compressed into thin layers or grow differently on a grid than on a conventional culture substrate. An apparently well-preserved field can therefore represent only a subset of the original population. Changes in support chemistry, application method, incubation time, and freezing conditions can alter these effects, but no preparation method eliminates every source of perturbation.\n\nLarger cells and tissues require different strategies. High-pressure freezing permits vitrification of thicker specimens than ordinary plunge freezing by modifying water's freezing behavior and facilitating rapid preservation in suitable carriers. Such specimens remain far too thick for direct transmission imaging and must subsequently be thinned. Cryogenic ultramicrotomy cuts thin sections from frozen material, but cutting can cause compression, crevasses, and other deformations. Ion-beam milling avoids mechanical sectioning and has become a major alternative.\n\nCryogenic handling is essential after freezing. Grids and specimens must remain cold enough to prevent devitrification and must be protected from atmospheric moisture. Frost deposited during transfer adds crystalline contamination that can obscure a region or interfere with milling and imaging. Preparation records therefore include not only biological treatment but also grid type, freezing conditions, ice quality, transfer history, and relevant quality-control observations. These details affect whether differences between tomograms reflect biology, preservation, or sampling."
    },
    {
      "heading": "Cryogenic milling and correlative targeting",
      "text": "Most eukaryotic cells are too thick for direct cryo-ET, except at thin peripheral regions. Cryogenic focused-ion-beam milling addresses this limitation by removing material above and below a selected region, leaving an electron-transparent lamella. The workflow typically uses an instrument combining a focused ion beam with a scanning electron microscope. Gallium ions are widely used for milling; plasma ion sources provide alternative milling characteristics and can remove larger volumes of material.\n\nA typical cellular lamella is roughly 100–200 nanometres thick, although the appropriate thickness depends on the specimen and the intended analysis. Coarse milling removes bulk material, followed by lower-current polishing that approaches the final surfaces more gently. Protective surface layers can reduce uneven erosion. Important artifacts include curtaining, in which variations in sputtering produce grooves, redeposition of removed material, bending, and ion-induced damage near the lamella surfaces. These effects are particularly relevant when molecular structures lie close to a milled boundary.\n\nFinding the desired cellular region is often harder than producing a thin lamella. A rare structure may occupy only a small fraction of the cell, while the final lamella preserves only a narrow slice. Correlative light and electron microscopy uses optical information to guide electron-microscopy preparation and interpretation. Fluorescent labels can identify a protein, organelle, infection site, or other target before or after freezing. Coordinates are then transferred between the light microscope, milling instrument, and transmission electron microscope.\n\nThe accuracy of this targeting depends on registration between imaging modalities. Optical resolution, specimen deformation, marker localization, stage positioning, and differences in viewing geometry all contribute to uncertainty. Fiducial markers visible in more than one modality can improve registration. Fluorescence in a lamella can confirm that a target remains present after milling, but a fluorescent signal does not by itself identify every nearby density in the tomogram. Labels also require biological validation because tagging or overexpression may change the target's behavior.\n\nFor thick tissue or high-pressure-frozen material, cryogenic lift-out methods extract a small block and attach it to a support before final thinning. This permits access to regions that cannot be reached by milling a cell directly on a grid. The preparation is technically demanding because the extracted material must remain vitrified, mechanically stable, and accurately positioned through several transfers.\n\nTargeting determines the kinds of questions that can be answered. Untargeted lamellae can support surveys of abundant cellular structures, whereas localized targeting is important for rare events such as a particular membrane-remodeling intermediate. Because the sampled volume is small, failure to observe a structure cannot establish its absence from the cell."
    },
    {
      "heading": "Tilt-series acquisition and microscope control",
      "text": "A tilt series records repeated images of the same region as the specimen rotates, usually around a single axis. A common angular range is approximately \\(-60^\\circ\\) to \\(+60^\\circ\\), with increments of about \\(2^\\circ\\)–\\(3^\\circ\\), although wider ranges and different sampling intervals are used. Mechanical obstruction by the holder, grid geometry, increasing effective thickness, and deteriorating high-angle contrast prevent complete angular coverage. The resulting acquisition must balance angular range, exposure budget, field of view, and the spatial detail required.\n\nSpecimen tilting does not leave a region perfectly stationary. Small inaccuracies in the stage's rotation axis produce lateral motion and changes in height. Before recording, the operator or acquisition software establishes an approximately eucentric position, in which the target lies near the tilt axis. Automated tracking and focusing then compensate for residual shifts. These operations are often carried out in neighboring regions to avoid exposing the final imaging area unnecessarily. Drift after stage movement must also be controlled because it can blur images or move the target outside the recorded field.\n\nThe order of projection angles matters because radiation damage accumulates throughout the series. A sequential scheme proceeding from one extreme angle to the other records some of the most informative low-tilt images only after substantial exposure. A dose-symmetric tilt scheme begins near zero tilt and alternates between positive and negative angles of increasing magnitude. The method described by Wim Hagen, William Wan, and John Briggs in 2017 formalized an influential approach to preserving high-resolution information in the early, low-tilt views. Other acquisition orders remain useful for particular instruments and specimen geometries.\n\nModern direct electron detectors commonly record each projection as a short movie. Aligning its frames can compensate for some beam-induced movement and stage drift. Exposure weighting can reduce the contribution of frames or projections whose high-frequency information has deteriorated. Neither procedure reverses radiation damage, and inaccurate movement estimates can blur rather than improve weak data. Detector gain correction, defective-pixel treatment, magnification calibration, and image-distortion correction also affect the quality of subsequent reconstruction.\n\nDefocus is another acquisition variable. Greater underfocus increases the visibility of broad cellular features but complicates high-resolution interpretation and introduces additional transfer-function oscillations. A tilted specimen also has a defocus gradient across the image, while objects at different depths have different local defocus values. Accurate molecular analysis therefore requires more than assigning one focus value to an entire series.\n\nThroughput has improved through automated targeting, beam-image-shift acquisition, and coordinated microscope control. Nevertheless, rapid collection is useful only when the acquired regions remain suitable for the biological question. A large dataset of poorly targeted, excessively thick, or damaged specimens does not compensate for inadequate preparation."
    },
    {
      "heading": "Alignment and three-dimensional reconstruction",
      "text": "Before reconstruction, the projections must be placed into a consistent geometric frame. Alignment estimates image translations, rotations, the tilt-axis direction, and, where needed, additional geometric distortions. Gold beads added to a specimen can serve as fiducial markers because their positions can be tracked through the series. In cellular lamellae, markers may be absent or inconveniently distributed, so patch-based correlation and feature-tracking methods are widely used. Alignment residuals provide useful diagnostics, but a small residual does not guarantee that every part of a deforming specimen is correctly registered.\n\nThe reconstruction problem is commonly understood through the Fourier slice theorem. Under an ideal projection model, the two-dimensional Fourier transform of a projection corresponds to a central plane through the three-dimensional Fourier transform of the object. Different tilt angles sample different planes. A complete, noise-free set of projections would therefore supply the information needed to reconstruct the object. Real cryo-ET departs from this ideal because images have noise, incomplete angular coverage, transfer-function modulation, and thickness-dependent scattering.\n\nWeighted back-projection is a standard reconstruction method. Each image is filtered to compensate for uneven frequency sampling and then projected back through the reconstruction volume at its recorded angle. Iterative approaches instead repeatedly compare calculated projections of a provisional volume with the measured images and update the volume to improve agreement. Such methods can produce visually different results because they handle noise and impose constraints differently. Greater apparent smoothness or sharper boundaries do not necessarily indicate a more accurate recovery of fine structure.\n\nLimited tilt produces a missing wedge of unsampled information in Fourier space. Its consequences include directional resolution differences, elongation of features approximately along the poorly sampled direction, and orientation-dependent visibility. A membrane may be clear in one orientation and weak in another even when its composition is unchanged. Dual-axis tomography can reduce the unsampled region by collecting a second series after rotating the specimen in the grid plane, but it adds exposure and registration complexity and still does not provide complete information.\n\nHigh-resolution workflows also correct contrast transfer and exposure-dependent loss of signal. Because a tilted specimen spans different defocus values, correction may be performed locally or with three-dimensional models that account for particle depth. Modern refinement methods can estimate specimen motion and geometric errors jointly across many particles and projections.\n\nThe output is a three-dimensional array of voxels, not a direct map of molecular identity. Contrast reflects electron–specimen interactions modified by imaging and processing. Distances, volumes, and membrane separations require appropriate calibration, and interpretation must account for the reconstruction method, orientation-dependent artifacts, and effective resolution rather than relying solely on voxel size."
    },
    {
      "heading": "Subtomogram averaging and molecular identification",
      "text": "An individual cellular tomogram usually has insufficient signal to reveal the finest details of a macromolecular complex. Subtomogram averaging improves the signal by extracting many volumes containing copies of the same object, estimating their orientations and positions, and combining them. Random noise partly cancels, while reproducible structural features reinforce one another. The method is especially powerful when a complex is abundant, sufficiently large, and present in multiple orientations.\n\nThe idealized gain in signal-to-noise ratio scales approximately as \\(\\sqrt{N}\\) for \\(N\\) independent, equivalent observations. Actual performance is limited by alignment errors, structural variability, unequal angular coverage, and correlations introduced by processing. Ribosomes are frequent targets because they are large, abundant, and recognizable in many cells. Ordered arrays, viral surface proteins, filament-associated complexes, and membrane-bound assemblies can also be suitable, although their orientation distributions may leave persistent gaps in the recovered information.\n\nParticle selection can use manual annotation, geometric constraints, reference-based searches, or machine-learning methods. Template matching searches a tomogram for locations and orientations that resemble a supplied structural reference. It is useful for known complexes but risks false positives and reference bias. A high correlation score is not proof of molecular identity, particularly in crowded regions containing other objects of similar size and shape. Independent biological evidence, expected cellular localization, controls, and reproducibility strengthen an assignment.\n\nAlignment and classification address related but distinct problems. Alignment seeks a common coordinate frame; classification separates different molecular species, assembly states, or conformations. A mixed population can otherwise produce a blurred average that corresponds to no actual state. Conversely, excessively flexible classification can divide noise into apparently meaningful groups. Imposing symmetry can improve signal when the biological object genuinely has that symmetry, but it can conceal asymmetric features or create misleading regularity when the assumption is inappropriate.\n\nMany high-resolution pipelines refine particle parameters directly against the original tilt images rather than relying only on previously reconstructed volumes. This approach can model local defocus, per-tilt motion, orientation, and other imaging parameters more accurately. Under favorable conditions, subtomogram averaging reaches subnanometre and, for selected specimens, near-atomic resolution. Such performance describes particular averaged complexes, not the entire cellular tomogram.\n\nA crucial advantage is that refined particles retain their original coordinates. Molecular state can therefore be related to membrane curvature, distance from another assembly, position along a filament, or local crowding. This links structure to organization rather than replacing the cellular scene with a single isolated average."
    },
    {
      "heading": "Biological applications across molecular and cellular scales",
      "text": "Cryo-ET is particularly informative where biological function depends on three-dimensional arrangement. In thin bacterial cells, tomography can reveal envelope organization, intracellular filaments, storage bodies, flagellar motors, and chemoreceptor arrays. Repeated components of these systems can be averaged while preserving their arrangement within the cell. For example, chemoreceptor arrays illustrate how individually repeated signaling units assemble into an extended lattice whose collective organization is central to its activity.\n\nMembranes are prominent targets because their surfaces define compartments and provide a framework for molecular interactions. Tomograms resolve the topology of the cell membrane, internal vesicles, contact sites, and the double-membrane organization of the mitochondrion. In mitochondrial cristae, rows of ATP synthase dimers occur along strongly curved membrane regions. Tomography connects the placement of these complexes with the geometry of the surrounding membrane, although static images alone cannot establish every step in the causal relationship between assembly and curvature.\n\nThe cytoskeleton presents another set of multiscale questions. Cryo-ET can trace filament trajectories and examine cross-links, motors, and attachments within the same volume. In cilia and flagella, the repeated organization of microtubule doublets supports averaging of axonemal units, revealing arrangements of dynein motors and associated regulatory complexes. In neuronal processes and synaptic regions, tomography can visualize vesicles, membranes, filament networks, and large molecular assemblies. Molecular assignments remain dependent on resolution and supporting evidence.\n\nFor viruses, cryo-ET can examine pleomorphic particles that do not fit a single, highly symmetric reconstruction. It can reveal the distribution of envelope proteins, connections between membrane and internal layers, and structural variation among particles. Studies of infected cells extend this analysis to assembly, budding, membrane remodeling, and replication-associated compartments. A series of differently configured particles may suggest an assembly pathway, but an inferred sequence from static images is not equivalent to directly observing the same particle change over time.\n\nCryo-ET also informs translation by locating ribosomes and examining their association with membranes, neighboring ribosomes, and other cellular components. Averaging and classification can distinguish structural states in favorable datasets. Spatial statistics can then test whether those states correlate with cellular location or local organization.\n\nThese applications depend on integrating morphology with other evidence. Genetic perturbations, biochemical analysis, fluorescence localization, and independently determined molecular structures help distinguish plausible interpretation from direct observation. Cryo-ET provides unusually detailed spatial evidence, but it does not directly measure catalytic rates, molecular concentrations throughout an entire organism, or the temporal order of every observed event."
    },
    {
      "heading": "Validation, quantitative interpretation, and computational frontiers",
      "text": "Validation begins by separating what was recorded from what was inferred. Raw tilt images, aligned projections, reconstructed volumes, segmented structures, averaged particles, and fitted atomic models represent progressively more processed forms of evidence. An attractive rendering may conceal weak boundaries or uncertain molecular assignments. Quantitative analysis therefore needs access to intermediate data and documentation of acquisition, reconstruction, filtering, particle selection, and model-building choices.\n\nResolution assessment for averaged structures commonly uses Fourier shell correlation, which compares independently reconstructed half-datasets as a function of spatial frequency. A correlation threshold of 0.143 is widely used for nominal resolution reporting under appropriate half-map procedures. However, masking, shared references, overlapping particle information, and other correlations can inflate apparent agreement. In tomography, careful data splitting must consider the fact that particles may share projections and local imaging errors. Local and directional resolution can be more informative than one global number.\n\nSegmentation converts a tomogram into labeled objects such as membranes, filaments, or vesicles. Manual segmentation is labor-intensive and can vary between annotators; automated segmentation depends on training data and may fail on unfamiliar structures or imaging conditions. Measurements of curvature, volume, spacing, and connectivity inherit these uncertainties. Statistical comparisons must also recognize the hierarchy of the dataset: many particles from one tomogram are not equivalent to the same number of independent cells or biological preparations.\n\nModern neural networks can denoise tomograms, detect particles, segment organelles, and estimate information missing from limited-angle data. These methods can markedly improve usability, but visual restoration is not independent experimental evidence. A network may suppress unusual genuine features or introduce patterns favored by its training examples. Claims about newly revealed structures therefore require checks against the measured projections, independent datasets, or other experimental evidence. Uncertainty is especially important when computational processing produces detail in directions that were not directly sampled.\n\nPublic data resources support scrutiny and reuse. The Electron Microscopy Data Bank archives reconstructed maps, while the Electron Microscopy Public Image Archive hosts raw imaging data, including tomography datasets. Atomic models derived from suitable reconstructions can be deposited in the Protein Data Bank. Reproducibility also depends on specimen metadata, particle coordinates, processing parameters, and software versions, which permit others to distinguish biological findings from pipeline-specific effects.\n\nMajor technical challenges remain connected: thicker specimens degrade image formation, thinner lamellae restrict the sampled volume, high-resolution averaging favors abundant assemblies, and targeted preparation can bias which cellular regions are observed. Current development consequently includes automated milling, integrated cryogenic fluorescence targeting, faster acquisition, improved physical models of image formation, and methods for analyzing heterogeneous molecular populations without erasing their spatial relationships."
    }
  ],
  "links": [
    "cryo-electron-microscopy",
    "protein",
    "cell",
    "nobel-prize-in-chemistry",
    "x-ray-crystallography",
    "transmission-electron-microscope",
    "electron",
    "water",
    "ionization",
    "vitrification",
    "calibration",
    "machine-learning",
    "bacteria",
    "cell-membrane",
    "mitochondrion",
    "atp-synthase",
    "cytoskeleton",
    "microtubule",
    "virus",
    "translation-biology",
    "ribosome",
    "artificial-neural-network"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}