How to Learn Cryo-Electron Tomography (Cryo-ET): From Tilt Series and the Missing Wedge to Cryo-FIB Lamellae, Subtomogram Averaging and In-Situ Structural Biology

## Wait, What? Cryo-ET Does Not Average Away the Cell Before It Starts Single-particle cryo-EM usually collects huge numbers of isolated copies of a molecule. Cryo-electron tomography can do something different. It takes one unique region: part of a bacterium, a mitochondrion, a viral assembly site, a synapse or a cellular lamella. Then it tilts that same region and images it again and again. The result is a three-dimensional reconstruction of that one native-like cellular neighbourhood. > **Cryo-ET keeps molecular structure inside cellular context, but pays for that context with lower signal, a missing wedge of angles, specimen-thickness limits and a strict radiation-dose budget.** ## The One-Sentence Answer **Learn cryo-ET by tracing vitrification → low-dose tilt series → alignment → CTF/dose correction → tomographic reconstruction → missing-wedge interpretation, then add cryo-FIB thinning, subtomogram averaging, template bias, particle heterogeneity and AI-assisted annotation before turning a 3D density into an in-situ molecular structure or cell-mechanism claim.** # Beginner Layer — One Specimen, Many Tilts ## Stage 1: Vitrify the Sample Water is frozen so rapidly that it forms vitreous rather than crystalline ice. ## Stage 2: Keep the Sample Cryogenic The hydrated cellular structure is preserved without conventional dehydration. ## Stage 3: Place the Sample in the Electron Microscope ## Stage 4: Record a Projection Image An electron micrograph is a projection through the specimen thickness. ## Stage 5: Tilt the Specimen Record another projection. ## Stage 6: Repeat Over Many Angles The collection is a tilt series. # Tomographic Geometry ## Stage 7: Each Projection Contains a Different View of the Same 3D Object ## Stage 8: A Tomographic Reconstruction Combines Those Views ## Stage 9: Unlike Single-Particle Cryo-EM, the Object Is Not Assumed to Exist in Thousands of Identical Copies The region is unique. # Electron Dose ## Stage 10: Biological Samples Are Beam Sensitive Each electron helps imaging but also damages the specimen. ## Stage 11: The Total Tilt Series Shares One Dose Budget If you collect 40–60 projections, each image receives only a fraction of the allowable dose. ## Stage 12: Individual Tilt Images Are Therefore Very Noisy This is expected. # Tilt Scheme ## Stage 13: The Order of Tilts Determines Which Angles Receive the Least-Damaged Sample ## Stage 14: Dose-Symmetric Acquisition Starts Near Zero Tilt Then alternates positive and negative angles while increasing magnitude. ## Stage 15: Early High-Information Views Are Spread Symmetrically in Fourier Space This improves high-resolution subtomogram work. # Why High Tilt Is Hard ## Stage 16: Effective Sample Thickness Increases With Tilt At angle θ, the electron path through a slab increases roughly as: **thickness / cosθ** ## Stage 17: High Tilt Has More Multiple Scattering and Lower Signal ## Stage 18: Microscope Pole-Piece Geometry Also Limits Maximum Tilt Typical biological tilt ranges stop near ±60–70°. # The Missing Wedge ## Stage 19: Angles Beyond the Accessible Tilt Range Are Never Measured In Fourier space this creates a missing wedge of information. ## Stage 20: Missing Information Produces Directional Blur Features can become elongated along the electron-beam direction. ## Stage 21: Nearby Structures May Appear Artificially Connected > **The missing wedge is not an image-processing nuisance. It is missing experimental information.** # Fiducial Alignment ## Stage 22: A Correct Tomogram Requires All Tilt Images to Represent One Fixed Rotating Object But the stage drifts and the sample moves. ## Stage 23: Gold Beads Can Act as Fiducials Their positions are tracked through the tilt series. ## Stage 24: Fit the Projection Geometry Estimate translations, rotation, tilt axis and distortion. # Fiducialless Alignment ## Stage 25: Cryo-FIB Lamellae Often Lack Useful Gold Beads The milling process can remove surface fiducials. ## Stage 26: Fiducialless Methods Align Using Specimen Features ## Stage 27: Alignment Becomes Harder Because Projected Appearance Changes With Angle # Beam-Induced Motion ## Stage 28: Each Tilt Can Be Recorded as a Movie Direct detectors allow frame-wise motion correction. ## Stage 29: Motion Correction Is Needed Before Tilt-Series Alignment Two different scales of motion are being corrected: within an exposure and between tilts. # Contrast Transfer Function ## Stage 30: Cryo-ET Usually Uses Defocus to Create Phase Contrast ## Stage 31: The CTF Modulates Spatial Frequencies ## Stage 32: Tilt Adds a Defocus Gradient Across the Image One side of the field is physically higher than the other. ## Stage 33: A Single Defocus Value Can Therefore Be Wrong Across the Whole Tilted Image Modern corrections divide images into strips or patches or use more complete 3D CTF models. # Reconstruction ## Stage 34: Weighted Backprojection Is a Classic Tomographic Method Each corrected projection is smeared back through the volume with frequency weighting. ## Stage 35: SIRT and Related Iterative Methods Can Produce Smoother Volumes ## Stage 36: A Smoother Tomogram Is Not Automatically a More Accurate Tomogram Iterative methods can trade resolution and appearance. # Tomogram Versus Structure ## Stage 37: The Tomogram Is a 3D Volume of Cellular Density It can show membranes, ribosomes, filaments, vesicles, viral particles and organelles. ## Stage 38: Most Individual Proteins Are Still Low Contrast A tomogram is not automatically an atomic map. # Thick Cells ## Stage 39: Many Eukaryotic Cells Are Too Thick for Useful Transmission Cryo-ET Electrons undergo strong multiple scattering. ## Stage 40: The Central Cellular Region Must Be Thinned # Cryo-Focused Ion Beam Milling ## Stage 41: Use an Ion Beam to Mill Away Vitrified Material A thin lamella is left. ## Stage 42: The Lamella Can Be Approximately 100–200 nm Thick This brings interior cellular regions into the electron-transmission range. ## Stage 43: Cryo-FIB Creates a New Preparation Transfer Function The structure you image has passed through vitrification, ion milling, transfer and electron exposure. # Lamella Damage ## Stage 44: Gallium Ions Can Damage Near-Surface Material ## Stage 45: Damage Depth Depends on Ion Species and Accelerating Voltage A 2025 study directly compared milling damage across conditions. ## Stage 46: Thinner Is Not Always Better Research has shown that ultrathin lamellae can unnecessarily sacrifice cellular volume without improving subtomogram resolution. # Curtaining ## Stage 47: Different Material Densities Mill at Different Rates The lamella develops stripe-like thickness variation. ## Stage 48: Protective Coatings and Polishing Strategies Reduce Curtaining ## Stage 49: A Curtained Lamella Can Create Apparent Density Variation # Crystalline Ice ## Stage 50: Poor Vitrification Creates Crystalline Ice ## Stage 51: Ice Diffraction Can Contaminate the Tilt Series ## Stage 52: Thick Tissue Is Especially Difficult to Vitrify Uniformly # Cryo-Lift-Out ## Stage 53: On-Grid Milling Is Limited for Deep Tissue Regions ## Stage 54: Cryo-Lift-Out Removes a Chunk From a Larger Specimen A micromanipulator transfers it to another grid for thinning. ## Stage 55: This Opens Tissue and Organoid Interiors to In-Situ Structural Biology ## Stage 56: Lift-Out Introduces More Transfer and Contamination Risks # Correlative Cryo-Fluorescence ## Stage 57: Fluorescence Can Identify a Rare Cellular Event Before Electron Tomography Examples include a specific organelle, infection focus, labelled protein or damage site. ## Stage 58: Coordinate Transforms Connect the Fluorescence and Cryo-EM Frames ## Stage 59: Correlation Error Must Be Measured A fluorescent spot is often much larger than a molecular complex. # Cellular Architecture ## Stage 60: Cryo-ET Excels at Membranes and Macromolecular Organization A cell can be interpreted as a 3D molecular landscape rather than isolated biochemical parts. # Subtomogram Averaging ## Stage 61: Repeated Complexes Can Be Extracted From the Tomogram Examples include ribosomes, ATP synthases, viral spikes and membrane complexes. ## Stage 62: Align the Subvolumes in 3D ## Stage 63: Average Them to Improve Signal ## Stage 64: Now Cryo-ET Combines Context With Repetition > **Tomography finds the molecules in their native neighbourhood; subtomogram averaging recovers the repeated structure from that neighbourhood.** # Missing-Wedge Compensation in Averaging ## Stage 65: Every Subtomogram Shares Directional Missing Information ## Stage 66: Different Particle Orientations Can Fill One Another’s Missing Directions ## Stage 67: Orientation Diversity Is Therefore Essential A membrane complex with one fixed orientation may remain anisotropically resolved. # Particle Picking ## Stage 68: Find Candidate Complexes in a Noisy 3D Volume Methods include manual picking, template matching and neural networks. ## Stage 69: Particle Picking Is Not Ground Truth False coordinates propagate to the average. # Template Bias ## Stage 70: Template Matching Asks Whether a Known Structure Resembles Local Density ## Stage 71: A Strong Prior Can Detect the Structure It Expects Even when the data are weak. ## Stage 72: Use Negative Controls and Independent Processing A template match should survive changes in template, frequency range and threshold. # Subtomogram Classification ## Stage 73: In-Situ Complexes Can Occupy Several States Classification can separate conformations, assembly states and ligand-bound states. ## Stage 74: The Cellular Environment Can Create State Distributions Different From Purified Samples This is one of cryo-ET’s deepest scientific advantages. # Resolution Validation ## Stage 75: Independent Half-Set Refinement Is Required ## Stage 76: Fourier Shell Correlation Estimates Reproducible Spatial Frequency ## Stage 77: Map–Model Fit Must Remain Separate An atomic model cannot validate the same data that strongly constrained it without independence. # In-Situ Mitochondrial Frontier ## Stage 78: Cryo-ET Can Reveal Membrane-Protein Organization Inside Intact Mitochondria A 2025 Science study visualized respiratory-chain complexes in native mitochondrial cristae and resolved a native respirasome to around 5 Å. ## Stage 79: Structural Context Changed the Biological Question The study asked not only “what is the complex structure?” but **where is the complex located in the real membrane architecture?** # Mitophagy ## Stage 80: 2025 In-Situ Cryo-ET Captured Mitochondrial Depolarization and Engulfment The study combined cellular ultrastructure, ATP-synthase redistribution and subtomogram averaging. ## Stage 81: One Tomogram Can Connect Organelle-Scale and Molecular-Scale Evidence # Virus–Host Interactions ## Stage 82: Cryo-ET Can Capture Virus Assembly and Entry Inside Cells ## Stage 83: Purified Virions Cannot Show the Same Host-Cell Context # Bacterial Cells ## Stage 84: Many Bacteria Are Thin Enough for Direct Cellular Cryo-ET ## Stage 85: Their Membranes, Flagellar Motors and Macromolecular Machines Can Be Visualized In Situ # Direct Electron Detectors and Energy Filters ## Stage 86: Efficient Electron Detection Improves Low-Dose Signal ## Stage 87: Energy Filters Can Remove Some Inelastically Scattered Electrons This is especially useful for thicker specimens. # Denoising ## Stage 88: Neural Denoisers Can Make Tomograms Dramatically Easier to Interpret ## Stage 89: Denoised Detail Can Be Hallucinated The Topaz-Denoise authors explicitly warned not to use denoised particles as the basis for high-resolution reconstruction. > **Use denoising to see candidates. Return to raw data to prove them.** # 2025 Particle-Picking Frontier ## Stage 90: Template Learning Combines Deep Learning With Domain Randomization The method trains from structural templates without requiring huge manually annotated datasets. ## Stage 91: The Central Risk Becomes Transfer Does the detector recognize the protein in new cellular backgrounds, ice and microscopes? # Public Benchmarks ## Stage 92: The CZII CryoET Object Identification Challenge Released Benchmark Tomograms and Ground Truth This creates a shared test for particle localization and machine-learning generalization. ## Stage 93: Benchmarks Are Useful but Not the Full Biological World A competition dataset can never represent every specimen state. # 2026 Foundation-Model Frontier ## Stage 94: Self-Supervised 3D Representation Models Are Being Developed for Cryo-ET CryoDINO is an example of a large volumetric foundation model trained on hundreds of thousands of cryo-ET patches. ## Stage 95: Foundation Models Reduce Label Requirements ## Stage 96: They Also Import a Broad Prior Into Every Downstream Segmentation A low-label result still requires raw-density validation. # Automated FIB ## Stage 97: Modern Cryo-FIB Workflows Can Produce Multiple Lamellae With Much Less User Intervention ## Stage 98: Machine Learning Is Entering Site Selection ## Stage 99: Automation Can Scale Both Success and Systematic Bias If the targeting model avoids one morphology, the missing biology may never enter the microscope. # Professional Layer ## Stage 100: Separate Eight Objects 1. true native cellular state; 2. vitrified specimen; 3. cryo-FIB/lift-out lamella; 4. electron-projection tilt series; 5. aligned/CTF-corrected projections; 6. tomographic reconstruction with missing wedge; 7. picked/averaged molecular subvolumes; 8. biological structural interpretation. ## Stage 101: Professional Cryo-ET Is a Dose–Geometry–Context Inverse Problem > **Which cellular structure or molecular conformation remains identifiable after missing-angle information, beam damage, lamella preparation, CTF gradients, alignment error, template bias, particle heterogeneity and learned priors are all allowed to explain the same tomographic density?** # Evidence: What Makes a Cryo-ET Claim Strong? Stronger evidence combines verified vitrification, lamella thickness, milling metadata, low-dose tilt scheme, motion correction, fiducial/alignment residuals, CTF estimation, raw tilt-series inspection, missing-wedge-aware interpretation, independent half-set subtomogram refinement, negative template controls, local resolution, repeated cells/lamellae, raw-data deposition and orthogonal fluorescence or biochemistry where appropriate. # Misconceptions Worth Hunting – Cryo-ET is just single-particle cryo-EM with fewer particles. – Tomography records a 3D image directly. – More tilt angles always improve the tomogram. – The missing wedge can be completely filled by software. – SIRT is more accurate because it looks smoother. – A very thin lamella is always better. – Cryo-FIB only removes material and does not damage the sample. – Denoised tomograms are equivalent to raw experimental density. – Template matching proves the target protein is present. – A high-resolution subtomogram average proves every particle has that conformation. – In-situ structure is automatically more physiologically correct than purified structure. – Foundation-model segmentation needs little validation because it was pretrained broadly. # Transfer Check A membrane complex looks stretched along the beam direction in the tomogram. Did the protein necessarily elongate in the cell? **No. Missing-wedge anisotropy is a strong explanation.** A neural denoiser reveals a new filament, but the filament is not visible in independent raw half-data or Fourier-limited reconstructions. Is it proven? **No. Hallucination is possible.** A 100 nm cryo-FIB lamella gives worse subtomogram resolution than a 170 nm lamella prepared more gently. Is that impossible? **No. Milling damage and specimen quality can matter more than extreme thinness.** A template-matching pipeline finds thousands of particles only when one high-resolution template is used. Is the particle population secure? **No. Template bias needs testing.** An in-situ ATP-synthase average differs from a purified single-particle structure. Must one be wrong? **No. The cellular environment can stabilize a different conformational ensemble.** # How We Know the Learning Has Held A learner should be able to explain vitrification; distinguish single-particle cryo-EM from cryo-ET; explain tilt-series geometry; explain dose budget and dose-symmetric acquisition; explain the missing wedge; align tilt series with or without fiducials; explain CTF gradients; distinguish weighted backprojection and iterative reconstruction; explain cryo-FIB lamellae and damage; explain subtomogram averaging and missing-wedge compensation; identify template bias and denoising hallucination; understand public benchmarks and foundation models; and preserve raw-data validation. # Model Limits Cryo-ET is limited by radiation damage, sample thickness, missing angles, low signal per tilt, expensive instrumentation, difficult sample preparation and low throughput. Professional cryo-ET keeps **cell state + vitrification + lamella history + thickness + tilt scheme + dose + alignment + CTF + reconstruction + missing wedge + particle selection + half-map validation + AI prior + raw data** visible together. # Teaching Guide Teach in this order: **vitrification → projection → tilt series → dose budget → dose-symmetric scheme → high-tilt thickness → missing wedge → fiducial alignment → CTF gradient → reconstruction → tomogram → cryo-FIB → lamella damage → cryo-CLEM → subtomogram averaging → template bias → in-situ structures → AI/object identification → foundation models → raw-data validation.** # Connect This to the eduKate Learning Estate – https://edukatesengkang.com/2026/08/30/how-to-learn-cryo-electron-microscopy-single-particle-reconstruction/ – https://edukatesengkang.com/2026/08/28/how-to-learn-microscopy-scientific-imaging-super-resolution-image-evidence/ – https://edukatesengkang.com/2026/08/28/how-to-learn-protein-folding-proteostasis-amino-acid-sequence-cellular-quality-control/ – https://edukatesengkang.com/2026/08/30/how-to-learn-mitochondria-mitochondrial-dynamics/ # Research Foundations and Further Learning – Cryo-EM 101 cryo-ET chapters: vitrification, fiducials, missing wedge, CTF and lamella preparation. – Dose-symmetric tilt-scheme literature. – 2024 IUCr discussion of cryo-ET acquisition strategies and missing-angle geometry. – 2024–2025 cryo-FIB lamella-thickness and damage studies. – 2025 review: cryo-FIB milling for in-situ structural biology. – 2025 Science: in-cell architecture of the mitochondrial respiratory chain. – 2025 Nature Communications: Template Learning for cryo-ET particle picking. – CZII CryoET Object Identification benchmark. – 2026-era self-supervised cryo-ET foundation-model work such as CryoDINO. # The Quiet Ending The beginner asks: “How can a stack of tilted noisy images become a 3D cell?” The developing structural biologist asks: “Which orientations and dose history built this tomogram?” The advanced learner asks: “How much of this structure belongs to the cell, and how much to missing wedge, milling and template bias?” And the professional asks: > **Which molecular feature survives after vitrification, ion milling, missing information and every computational prior are all treated as part of the experiment?**