Why Science? | Cryo-Electron Tomography, Tilt Series and 3D Cellular-Ultrastructure Evidence

Section 1 of 36

1. Begin with vitrified context

Cryo-ET aims to preserve a hydrated specimen in vitreous, non-crystalline ice and image its structures in context. The specimen stays cryogenic during transfer and acquisition. Ice quality is part of the measurement, not simply a preparation detail.

Contents · Next section

Section 2 of 36

2. Acquire projections across tilt

The microscope records two-dimensional projections while the grid tilts through a limited angular range. Each projection contains overlapping structure. Together, after alignment, they constrain a three-dimensional volume.

Contents · Previous section · Next section

Section 3 of 36

3. Divide a finite electron dose

Electrons damage frozen specimens. The total tolerable dose must be distributed across many tilts, so every projection is noisy. More angles or exposure can improve sampling while consuming dose. State the dose scheme and order.

Contents · Previous section · Next section

Section 4 of 36

4. Recognise the missing wedge

The holder and specimen geometry prevent a full 180-degree tilt range. Unmeasured angles create anisotropic information and elongation in the reconstruction. Dual-axis acquisition can reduce but not erase the problem.

Contents · Previous section · Next section

Section 5 of 36

5. Connect thickness to feasibility

Electrons scatter strongly in thick material. Small cells, edges or isolated particles may be imaged directly; thicker cellular regions often require cryogenic focused-ion-beam milling into lamellae. Thickness affects contrast, multiple scattering and survival.

Contents · Previous section · Next section

Section 6 of 36

6. Understand contrast transfer

Defocus helps create phase contrast but introduces an oscillating contrast transfer function. Tilt changes effective defocus across the field. CTF estimation and correction should match the geometry rather than borrowing a single micrograph value.

Contents · Previous section · Next section

Section 7 of 36

7. Define the structural claim

Decide whether the study tests membrane spacing, organelle contact, filament organisation, particle state or molecular placement. Predefine the required voxel scale, reconstruction quality, annotation rule and independent biological replicate.

Contents · Previous section · Next section

Section 8 of 36

8. Optimise vitrification

Blotting, humidity, temperature, sample concentration and freezing geometry influence ice thickness and distribution. Screen multiple grid squares and report unusable fractions. Beautiful survivors should not conceal systematic preparation loss.

Contents · Previous section · Next section

Section 9 of 36

9. Choose supports and fiducials

Grid type, film, holes and gold fiducials affect tracking and alignment. Fiducials should be well distributed in three dimensions when possible. Fiducial-free methods still need residual and stability checks.

Contents · Previous section · Next section

Section 10 of 36

10. Prepare lamellae transparently

Cryo-FIB milling can expose interior cellular regions but may create curtaining, redeposition or surface damage. Record target selection, thickness, milling current and protective coating. The lamella is a selected slice, not an unbiased survey.

Contents · Previous section · Next section

Section 11 of 36

11. Maintain the cryogenic chain

Warming can crystallise ice and change structure. Track transfer, contamination, frost and vacuum events. Exclude or flag compromised regions by predeclared criteria instead of rescuing them with aggressive filtering.

Contents · Previous section · Next section

Section 12 of 36

12. Map low-magnification context

Atlas images help relocate cells and lamellae. Dose from mapping counts toward the budget. Preserve coordinate systems and acquisition history so high-resolution tomograms can be traced back to grid and specimen context.

Contents · Previous section · Next section

Section 13 of 36

13. Set tilt geometry and focus

Eucentric height, tilt axis, magnification, pixel size, defocus and energy filtering influence acquisition. Errors in nominal tilt angles can propagate into alignment and CTF. Current protocols therefore measure rather than assume key geometry.

Contents · Previous section · Next section

Section 14 of 36

14. Use dose-symmetric scheduling thoughtfully

A dose-symmetric scheme acquires low tilts early and alternates directions, preserving high-resolution information near zero tilt. It changes motion, accumulated damage and time order. Report the exact scheme and per-tilt dose.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with an invented cryo-ET table

The fictional series below shows why a tomogram needs acquisition context.

A smaller residual does not automatically compensate for dose or angular loss.

Contents · Previous section · Next section

Section 16 of 36

16. Correct motion per projection

Beam-induced motion and stage drift blur individual tilts. Preserve movie frames and correction trajectories. Rejecting high-motion tilts changes angular coverage, so the reconstruction and exclusion record must remain linked.

Contents · Previous section · Next section

Section 17 of 36

17. Estimate CTF across the tilt series

Fit defocus and astigmatism with a method that accounts for tilted geometry. Inspect confidence and failures by angle. The 2025 CTFMeasure protocol illustrates the value of measuring tilt-angle offset and CTF together.

Contents · Previous section · Next section

Section 18 of 36

18. Align with accountable residuals

Fiducial or patch tracking estimates translations, rotations, magnification and tilt-axis parameters. Plot residuals versus angle and position. A good average can hide one region that drives local misregistration.

Contents · Previous section · Next section

Section 19 of 36

19. Reconstruct with declared assumptions

Weighted back-projection, iterative algorithms and regularised methods distribute noise and missing information differently. Report filters, iterations and software versions. A denoised tomogram should never replace the conventional reconstruction as the sole evidence.

Contents · Previous section · Next section

Section 20 of 36

20. Annotate structures reproducibly

Membranes and particles may be traced manually or by algorithms. Use blinded rules, inter-annotator checks and uncertainty labels. Ambiguous density should remain ambiguous rather than being forced into a complete biological story.

Contents · Previous section · Next section

Section 21 of 36

21. Average repeated particles carefully

Subtomogram averaging can improve signal for repeated complexes, but particle selection, orientations and missing-wedge compensation matter. Split data and prevent reference bias. The average supports a population claim, not the exact state of every particle.

Contents · Previous section · Next section

Section 22 of 36

22. Challenge ice selection

Thin ice often gives better contrast but may exclude large complexes or compressed cells. Report grid-square selection and sample distribution. Preparation preference can change the population under study.

Contents · Previous section · Next section

Section 23 of 36

23. Challenge milling artefacts

Curtaining, charging and lamella bending can mimic repeated cellular texture. Compare orientations and milling conditions, inspect surfaces and use controls. A lamella’s accessibility is not proof of native architecture.

Contents · Previous section · Next section

Section 24 of 36

24. Challenge reconstruction hallucination

Iterative and learned denoisers can create continuous-looking membranes from incomplete noisy data. Validate on simulations and held-out raw series, show conventional views and test whether the claimed structure survives parameter changes.

Contents · Previous section · Next section

Section 25 of 36

25. Challenge missing-wedge interpretations

Elongation and directional resolution can make filaments or gaps appear orientation dependent. Rotate specimens where possible, use dual-axis evidence or quantify directional uncertainty. Never call absent density proof of absence without coverage analysis.

Contents · Previous section · Next section

Section 26 of 36

26. Challenge molecular identification

Shape similarity alone rarely identifies a protein in crowded cellular context. Genetic, labelling, biochemical or correlative evidence may be needed. State whether an assignment is direct, supported or tentative.

Contents · Previous section · Next section

Section 27 of 36

27. Compare orthogonal methods

Single-particle cryo-EM can provide higher-resolution averages, fluorescence can identify labelled targets, and biochemistry can test interaction. Cryo-ET contributes spatial context. The methods strengthen one another when their sample differences are acknowledged.

Contents · Previous section · Next section

Section 28 of 36

28. Learn with projection models

Students can photograph a simple object from limited angles and reconstruct a rough volume or use simulations. Missing views create directional ambiguity. No electron microscope, cryogens or ion beam is needed for the reasoning activity.

Contents · Previous section · Next section

Section 29 of 36

29. Connect Primary Science to unseen interiors

Shadow and cross-section models show that one view cannot reveal a complete interior. Teachers can contrast a physical slice with a projection so learners see why tomography needs several angles.

Contents · Previous section · Next section

Section 30 of 36

30. Build PSLE Science process skills

Students identify tilt angle as the changed condition, projection as observation and three-dimensional structure as inference. They explain why repeated angles, stable alignment and a known object are controls.

Contents · Previous section · Next section

Section 31 of 36

31. Extend into Secondary and O-Level Science

Physics contributes electrons, waves and imaging; Biology contributes cells and membranes; Chemistry contributes water and vitrification; Mathematics and Computing contribute geometry and reconstruction. The method is an enrichment context, not a fabricated exam requirement.

Contents · Previous section · Next section

Section 32 of 36

32. Use the topic for school choices

Look for verified teaching in foundational sciences, computing and research reasoning. Do not infer access to cryo-ET hardware, university partnerships or admission benefits from general STEM language. Check current official sources.

Contents · Previous section · Next section

Section 33 of 36

33. See the career ecosystem

Cryo-ET connects structural biology, microscopy, cryogenic engineering, ion-beam preparation, detector science, scientific software and data curation. Roles require different qualifications and facility safety training.

Contents · Previous section · Next section

Section 34 of 36

34. Did You Know? A tomogram is made from shadows

Each tilt image is a projection in which structures overlap. Alignment and reconstruction use many such projections to estimate where density lies in three dimensions. The final volume is calculated evidence, not a tiny camera flight through the cell.

Contents · Previous section · Next section

Section 35 of 36

35. Did You Know? The missing wedge has a direction

Angles the microscope cannot collect leave information missing along particular directions. That is why resolution can differ with orientation and why some structures appear stretched even when voxel dimensions are equal.

Contents · Previous section · Next section

Section 36 of 36

36. Keep tilt-series-to-structure reasoning visible

Define the claim, validate vitrification and thickness, record selection and milling, allocate dose, measure tilt and CTF, preserve movies, inspect alignment residuals and angular coverage, compare reconstructions, label ambiguity, respect replicate hierarchy, deposit traceable data and seek orthogonal identification before making cellular-ultrastructure claims.

A rigorous cryo-ET project begins with a specimen-to-claim plan. Define the cellular context, structure size, useful resolution, number of tomograms and independent specimens. Decide whether direct imaging, plunge-frozen small cells or cryo-FIB lamellae are needed. Acquisition time is expensive, so quality gates should be set before high-dose tilt series begin.

Grid screening should be systematic. Record every grid square considered, ice thickness, contamination, cell distribution and rejection reason. Map where selected regions lie. A montage of only the best lamellae can conceal that preparation enriched one cell state or excluded thick structures.

Vitrification should be verified by diffraction or another accepted ice-quality check. Crystalline ice changes both structure and contrast. Track humidity, blot time, temperature and sample concentration across grids. Protocol transfer to a new buffer or cell type requires fresh optimisation rather than borrowing the old conditions.

Cryo-FIB preparation needs its own controls. Record milling currents from coarse to polish, stage angles, protective deposition, lamella thickness and failure modes. Inspect curtaining, bending and redeposition. Correlative targeting can help find rare regions, but it also introduces selection and registration uncertainty that belongs in the dataset.

The dose budget should include atlas, focus, tracking and high-resolution acquisition, not only the tilt-series setting. Dose-symmetric schemes protect early low-tilt images, yet cumulative damage still changes later projections. Plot signal or high-frequency information against accumulated dose and state whether damaged tilts were excluded.

Tilt angle and eucentricity should be measured rather than assumed. Stage readouts may have offsets or backlash. The current CTFMeasure protocol is valuable because it jointly examines contrast transfer and tilt-angle offset. Small geometric errors can propagate into defocus, alignment and final resolution.

Alignment quality should be mapped locally. Plot fiducial trajectories, residuals by angle and patch consistency across the field. Fiducials that cluster on one surface may not describe deformation through a thick specimen. For fiducial-free methods, use simulations and known structures to establish expected bias.

CTF correction should consider defocus gradients at tilt and confidence at high angles. Report failures and whether phase flipping or more complete correction was used. If the study makes only lower-resolution morphological claims, avoid overstating the role of a high-frequency correction that the data do not support.

Reconstruction comparisons are an important control. Generate a conventional weighted-back-projection volume alongside iterative or learned results. Inspect whether membranes, gaps and particles persist across reasonable parameters. Denoising can aid annotation, but measurement should remain anchored to raw projections and an auditable reconstruction.

The missing wedge should be quantified for the actual range and geometry. Simulate how representative objects appear with the same angular coverage. Directional Fourier analysis or rotated standards can reveal anisotropy. A claimed preferred filament orientation may instead align with information that the microscope measured best.

Segmentation should carry uncertainty. Train annotators with explicit membrane and organelle rules, blind them to condition where possible and compare agreement. Algorithms need validation across contrast, thickness and specimen types. Report unassigned density and ambiguous boundaries rather than forcing a complete label map.

Subtomogram averaging adds a selection pipeline. Preserve coordinates, extraction boxes, alignment references, orientation coverage and classification decisions. Separate training or template generation from evaluation to reduce reference bias. Independent half-sets and withheld particles can reveal overfitting, but biological heterogeneity still needs interpretation.

Resolution should match the claim. Global Fourier shell correlation from an average does not describe every tomogram or region. Use local resolution, directional measures and visible features. A high-resolution average cannot retroactively make the surrounding cellular volume equally resolved.

Molecular identification should combine evidence. Genetic depletion, tagged targets, immunolabelling, known geometry or biochemical composition may support assignments. State the confidence level. Crowded cellular density makes shape matching tempting, but a plausible resemblance is not an identification test.

Biological replication requires several independently prepared specimens and, often, more than one grid or milling session. Tomograms from one lamella share preparation and imaging effects. Model nested data and show per-specimen summaries. Report the yield from starting cells to usable tomograms.

Pre-registration can name the primary structural metric, acceptable ice and alignment, dose range, reconstruction, annotation, missing-wedge analysis, exclusion rules and independent unit. Exploratory observations remain scientifically useful when separated from confirmatory tests. Rare events need clear denominator reporting.

Quality-control charts can track ice thickness, lamella yield, contamination, stage stability, alignment residual, CTF fit, dose and reconstruction metrics by session. Facility upgrades and software changes should be recorded as batches. Long projects are vulnerable to improvements that coincide with experimental groups.

Data stewardship should preserve movies, gain references, metadata, tilt angles, accumulated dose, motion trajectories, CTF fits, fiducial models, transforms, aligned stacks, reconstructions, masks, particle coordinates, code and versions. Public deposition in established archives, when appropriate, lets others inspect the chain from projection to claim.

Figures should pair low-magnification context, representative raw tilts, alignment and CTF diagnostics, orthogonal tomogram slices, missing-wedge orientation and all specimen-level measurements. Three-dimensional renderings are summaries, not evidence substitutes. Transparent surfaces and colours can hide or exaggerate density.

Negative results need a structural detection bound. Combine voxel size, directional resolution, contrast, missing angles, annotation error and biological variability. Absence of a small bridge in one orientation may be uninformative, while absence of a large membrane contact across well-sampled directions can constrain a model.

The next experiment should address the dominant limitation: thinner lamellae for multiple scattering, different tilt geometry for anisotropy, correlative labels for identity, subtomogram averaging for repeated complexes or biochemical perturbation for mechanism. Cryo-ET is at its best when contextual structure becomes a testable hypothesis.

Safety is specialised. Liquid nitrogen, cryogenic burns, oxygen displacement, high voltage, vacuum systems, electron beams, ion beams and biological specimens require facility procedures and trained staff. Classroom learning should use models and public datasets. No student needs direct instrument exposure to understand projections, reconstruction and uncertainty.

Reproducibility benefits from a specimen lineage record. Link culture or organism, grid, grid square, lamella, atlas, tilt series, reconstruction and annotation with stable identifiers. This prevents a striking tomogram from becoming detached from its biological source and lets exclusions be audited. The chain is especially important when different facility specialists perform preparation, milling, acquisition and analysis over several days.

Contents · Previous section · Continue to the Science Learning Hub