Cryo-EM and Structural Determination

A Tier 2 specialty covering the cryogenic-electron-microscopy revolution that displaced X-ray crystallography as the dominant method for macromolecular structure determination. Three innovations between 2012 and 2017 — Faraz Mahmood Bai-Scheres direct electron detectors, Sjors Scheres’s Relion maximum-likelihood reconstruction, and the Henderson-Glaeser-Frank-Dubochet methodological lineage — collapsed achievable resolution from ~10 Å in 2010 to ~1.2 Å (atomic) by 2024. The 2017 Chemistry Nobel went to Dubochet-Frank-Henderson; by 2026 cryo-EM accounts for >50% of new PDB entries among complexes >150 kDa, and EMDB deposition crossed 35,000 maps. This note covers the single-particle workflow end-to-end — vitrification, grid prep, detectors, microscopes, software stack (CryoSPARC, Relion-5, Topaz, Warp, CTFFind4.1, MotionCor3), resolution / B-factor calculus, heterogeneous reconstruction (3DVA, 3D flex, cryoDRGN, RECOVAR), helical reconstruction, sub-tomogram averaging (cryo-ET), micro-ED, AlphaFold integration (ModelAngelo, ModelCraft, fit-into-density), commercial services, and the cost / access economics that govern who actually gets to a 2.5 Å structure.

Sample preparation and vitrification

Sample quality requirements

Purity. Ideally >95% by SDS-PAGE silver stain or analytical SEC; HPLC trace single peak. Heterogeneity ruins 2D classification.
Concentration. Typically 0.5-5 mg/mL for ~150 kDa-500 kDa complexes; nano-cube concentrators (Vivaspin, Amicon Ultra). Higher MW → lower concentration acceptable.
Monodispersity. Aggregation visible by negative-stain EM screening; dynamic light scattering (Malvern Zetasizer) shows PDI <0.2.
Buffer. Avoid glycerol (radiation-sensitive, low contrast), sucrose; salt 50-300 mM acceptable; detergent at minimum required for membrane proteins (typically below CMC: 0.5-1× CMC of DDM, GDN, LMNG, or nanodiscs / amphipols / SMA / styrene-maleic acid copolymer / saposin / lipid-MSP nanodiscs).
Stability. Differential scanning fluorimetry (Tycho NT.6 NanoTemper, Prometheus) to confirm Tm >30 °C above experimental temperature.

Negative-stain pre-screen

Drop 3 µL sample onto carbon-coated grid (Ted Pella 01753); blot; stain with 2% uranyl acetate (or uranyl formate, Nano-W, or methylamine vanadate for biosafe alternative); image at room T on FEI/Thermo Tecnai T12 or T20 (120 kV) or Talos L120C. ~10⁵× cheaper than vitreous prep; quickly diagnoses aggregation, preferred orientation, particle distribution before committing to cryo.

Vitrification

Plunge-freezing into liquid ethane (90 K) cooled by liquid nitrogen. Vitrifies water in <1 ms → amorphous ice (no crystalline lattice → no electron diffraction interference).

Devices:

FEI / Thermo Vitrobot Mark IV. Workhorse; controllable humidity (95-100%), blot force, time (2-5 s typical), temperature (4-22 °C). Catalog VTB-MK4.
Leica EM GP / EM GP2. Single-side blot; capacitive blot detection.
Spotiton (Carragher-Potter NYSBC; Chameleon commercial via SPT Labtech, 2019). Piezoelectric inkjet-spotted nL droplets on grid; ~50 ms between spot and plunge → reduces air-water-interface (AWI) damage. Particularly useful for fragile complexes that partially denature at AWI.
Cryosol VitroJet. Pin-printing analogous to Spotiton.
Manual plunger. Pre-Vitrobot; still used in some labs for unusual samples.

Grid choice

Quantifoil holey carbon. Standard. R1.2/1.3, R2/1, R2/2, R3.5/1 spacing; Cu, Au, or Ni mesh. Quantifoil 1.2/1.3 200-mesh Cu most common.
C-flat (Protochips). Smoother carbon; alternative; less ice-thickness drift.
Au-flat / UltrAuFoil (Quantifoil). Gold film instead of carbon — reduces beam-induced motion ~5×; expensive (~10× cost of holey carbon).
Lacey carbon. Random holes; useful for tomography of cells, organelles.
Graphene-oxide (GO) coated. Increased particle density at AWI; reduces preferred orientation; thin support adds minimal background.
Graphene. Single-layer carbon; very low background; in-house preparation tricky.
Continuous carbon thin film (~5 nm). Adsorbs particles; reduces motion; useful for very small particles (<200 kDa); adds background that hurts resolution.

Glow discharge / plasma cleaning

Render grid surface hydrophilic before sample application. PELCO easiGlow (~20-40 s, 15 mA, air), Solarus II plasma cleaner (H₂/O₂ plasma). Without it, sample beads up; particles cluster at hole edges.

Microscopes

High-end (300 kV)

Thermo Titan Krios G4 (2022) / G5 (2024). 300 kV; field-emission gun (XFEG, cold-FEG on G5); autoloader (~6-12 grids); spherical aberration Cs ~2.7 mm or 1.4 mm corrected (Cs-corrected Cryo-EM Spectra); typical pixel size 0.4-1.5 Å. ~$8-12M install + ~$300-500k/yr service. Most national facilities house Krios G3/G4.
JEOL CryoARM-300. Competing 300 kV; Ω energy filter integrated; ~$5-8M.
JEM-3200FSC. Older JEOL 300 kV; in-column filter.
FEI/Thermo Polara, F30 (legacy). 200-300 kV; manual stage; pre-autoloader era; still operating in some labs.

Mid-range (200 kV)

Thermo Glacios / Glacios 2 (2023). 200 kV; smaller footprint; ~$2-3M; autoloader. Routine ~2.5-3.5 Å for well-behaved samples; appropriate for screening + production at smaller facilities.
JEOL CryoARM-200.

Low-cost / screening

Talos Arctica (predecessor of Glacios). 200 kV.
TFS Tecnai F20. 200 kV; manual; widely deployed.

Detectors

The detector revolution (~2012-2014) was the inflection point.

Gatan K3 (2018+). Successor to K2 Summit. 5760 × 4092 px; 4.6k frames/s in correlated double sampling (CDS); counted mode. DQE ~0.65 at Nyquist. Most-installed direct detector in 2024.
Gatan K3 IS (2022). Faster electronics; super-resolution counting.
Falcon 4i / Falcon 4 (Thermo). 4096 × 4096 px; event-based “EER” data format (saves only individual electron events) → tiny files vs K3.
Direct Electron Apollo / DE-64. Newer entrant; comparable specs to K3.
Direct Electron Mavor. Just-released 2024-2025.
CMOS (Gatan UltraScan, FEI Eagle, Tietz). Legacy; replaced for high-resolution work; useful for screening.
Selectris X / Selectris energy filter. Post-column Ω filter for in-column-filter Krios; ~5-10 eV slit width; improves contrast on thick / membrane samples.
Sigma post-column EFTEM (CEOS) — competitor.

Direct counting

Frame-based recording → identify single-electron events → eliminate readout noise → DQE ~0.65-0.85 at Nyquist vs ~0.3 for CMOS. Critical for small particles (<200 kDa).

Phase plate

Volta phase plate (Danev-Baumeister 2014). Heated amorphous-carbon film with central charge spot → adds ~π/2 phase shift → boosts low-frequency contrast. Useful for tomography of cells and very small particles; goes out of focus over hours.
Laser phase plate (von Loeffelholz-Glaeser). Stable laser standing-wave; pre-commercial 2023-2025.

Data acquisition

Microscope control software

SerialEM (Mastronarde UC-Boulder). Most-used open-source. Multi-shot, image-shift, dose-fractionated movies.
EPU (Thermo). Commercial; integrates with Krios + Falcon detector. AFIS (aberration-free image shift) for fast multi-hole acquisition.
Latitude (Gatan). For K3.
TFS Smart EPU 3 (2023). ML-based hole selection, ice-quality filtering, on-the-fly QC.

Acquisition strategy

Magnification / pixel size. Choose Nyquist limit < target resolution / 2; typical 0.5-1.4 Å/pixel. Smaller pixel = higher Nyquist but more pixels = more storage.
Defocus range. -0.5 to -3.0 µm typical; spread across range for CTF estimation robustness.
Total dose. 40-60 e⁻/Å² typical for proteins; fractionated into 30-60 frames of 0.7-1.5 e⁻/Å²/frame.
Exposure time. 2-5 s per movie.
Throughput. ~300-600 movies/h on Krios G4 with AFIS; 1500-3000 per day after stage-stabilization burn-in.

Aberration-free image shift (AFIS) and beam-image-shift (BIS)

Acquire multiple targets per stage move using fast beam/image shift; corrects induced beam-tilt aberrations during reconstruction. ~5× throughput vs stage shift alone.

Image processing pipeline

Motion correction

Beam-induced motion + stage drift moves particles between frames. Per-particle motion correction sharpens map.

MotionCor2 / MotionCor3 (Zheng-Cheng-Agard 2017 Nat Methods, 2024 v3). Patch-based; very fast on GPU; default in most pipelines.
Warp (Tegunov-Cramer 2019 Nat Methods). Per-pixel deformation field; integrates motion + CTF + particle picking.
alignframes (IMOD). TILT series motion correction.

CTF estimation

Contrast transfer function modulates image as function of defocus + aberrations. Each micrograph fit to extract defocus + astigmatism + phase shift.

CTFFind4 / CTFFind5 (Rohou-Grigorieff 2015, 2024 J Struct Biol). Reference open-source.
Gctf (Zhang 2016 J Struct Biol). GPU; slightly faster.
CTF refinement per particle. Run after initial reconstruction (Relion CtfRefine, CryoSPARC Local CTF Refinement) to allow per-particle defocus, beam tilt, anisotropic magnification.

Particle picking

Template-based. Cross-correlation against templates from 2D classes. Most-used at scale.
Topaz (Bepler-Berger 2019 Nat Methods). CNN-based; positive-unlabeled learning; ~10× recall improvement on hard datasets. Standard for difficult samples.
crYOLO (Wagner-Stabrin-Raunser 2019 Commun Biol). YOLOv2-based; very fast.
DoG / Laplacian of Gaussian. Classic; needs no training.
Warp BoxNet, EMAN2 NeuralNet. Older NN-based pickers.
CryoSPARC Live blob picker + 2D-class refinement loop. Real-time interactive.

2D classification

Group similar particle views (rotational + translational alignment); discard junk classes. 50-200 classes typical. Reveals preferred orientation, junk, contaminants.

Ab initio 3D reconstruction

Build initial 3D model from particles without prior reference.

CryoSPARC Ab Initio (Punjani-Brubaker 2017 Nat Methods). Stochastic gradient descent on SGD-VI loss; default in CryoSPARC.
Relion 3D initial model (Scheres lab). Maximum-likelihood; slower but more interpretable.
EMAN2 e2initialmodel. Older heuristic.

Relion Refine3D / Refine3D-helical. Maximum-likelihood, Bayesian regularization; gold-standard FSC convention.
CryoSPARC Homogeneous / Non-Uniform Refinement. Non-uniform refinement (Punjani-Fleet 2020 Nat Methods) handles flexible regions with adaptive masking; superior for membrane proteins.
3D Classification. Sort particles by conformation; Relion 3D-Class.

Heterogeneous reconstruction

Most complexes have continuous flexibility — not discrete states.

CryoSPARC 3D Variability Analysis (3DVA; Punjani 2021 J Struct Biol). PCA on per-particle latent components.
CryoSPARC 3D Flex (Punjani 2023 Nat Methods). Deformation field; smooth heterogeneity.
cryoDRGN (Zhong-Bepler-Davis-Berger 2021 Nat Methods). Variational autoencoder; continuous heterogeneity manifold.
cryoDRGN2, cryoDRGN-AI (2023-2024). Improved.
RECOVAR (Gilles-Singer 2024). PCA-based; provides uncertainty.
Multi-body refinement (Relion). Discrete rigid bodies with flexible linkers.
OPUS-DSD, e2gmm. Alternatives.

Resolution and FSC

Fourier Shell Correlation between two half-maps (independent particle halves processed separately = “gold standard”). FSC = 0.143 cutoff defines reported resolution. B-factor sharpening (Rosenthal-Henderson 2003) sharpens high-frequency components — bring out side-chain density at expense of noise amplification.

Map post-processing

Relion PostProcess. Auto-B-factor; mask correction.
DeepEMhancer (Sanchez-Garcia-Vargas 2021 Nat Commun). ML map sharpening; reduces noise overshoot at high res.
EM-Ringer (Barad-Echols 2015 Nat Methods). Quality assessment via side-chain dihedral fit to density.
Q-score (Pintilie-Chiu 2020 Nat Methods). Per-residue resolvability metric.

Software ecosystems

CryoSPARC (Structura Biotechnology)

Punjani-Fleet-Brubaker UToronto; commercial (free academic license). Web-UI, GPU-accelerated, integrated end-to-end pipeline. 2D, 3D, refinement, heterogeneity, helical, sub-tomogram averaging. Dominant in industry and many academic labs since ~2018. CryoSPARC Live for real-time on-the-fly processing during data collection.

Relion (Sjors Scheres, MRC-LMB)

Open-source; long the academic reference. Relion 5 (2024) — major rewrite; native multi-class refinement, blush regularization (NN-based prior), Tomography pipeline overhaul. Slower than CryoSPARC; more transparent for method development.

Warp / WarpTools / M / mowarp (Tegunov-Cramer)

Pre-processing-focused; per-particle motion + CTF + denoising integrated. M for tomography. Warp 2 (mowarp, 2024) — ML-based denoising for low-dose data.

EMAN2 (Steve Ludtke Baylor)

Open-source; older but feature-rich; mostly used for tomography and special applications now.

SPHIRE / SPARX

Bremen-based; SP fork.

IMOD (Mastronarde UC-Boulder)

Tomography reconstruction toolkit. Stack alignment, weighted back-projection, SIRT, dose-weighted tomogram. AreTomo (Zheng-Agard 2022) — automated alignment alternative.

Phenix / Coot / ChimeraX

Model building + refinement on top of cryo-EM maps. Phenix.real_space_refine, Coot interactive building, ChimeraX visualization (Pettersen-Goddard 2021 Protein Sci).

ModelAngelo (Jamali-Scheres 2024 Nature)

ML-based automatic atomic model building into cryo-EM maps; sequence-aware (uses HMM + amino-acid type prediction from density). Builds 95% of residues correctly at 2.5-3.5 Å maps without manual chain-tracing.

ModelCraft (Bond-Wilson 2023 Acta Cryst D)

Automated model building; comparable to ModelAngelo, integrates with Buccaneer.

Buccaneer, ARP/wARP

Crystallographic model-building tools also useful at high-res cryo-EM.

Tomography and sub-tomogram averaging (STA)

Tomography captures 3D volumes of cells, organelles, viruses by tilting the stage from -60° to +60° (5° steps typically) and reconstructing the 3D tomogram.

Acquisition

Dose-symmetric scheme (Hagen-Wan-Briggs 2017 J Struct Biol) — alternate +θ, -θ to keep highest-dose information at zero-tilt where features are most informative. Total dose 100-180 e⁻/Å² across tilt series.

Tomogram reconstruction

IMOD eTomo (Mastronarde). Standard. Fiducial-based or patch-tracking alignment; weighted back-projection or SIRT.
AreTomo (Zheng-Agard 2022 J Struct Biol X). GPU-accelerated; fiducial-free.
Tomo3D, ICON. Iterative.
TomoBEAR, nextPYP. Pipeline orchestration.

Sub-tomogram averaging

Pick particle positions in tomograms → extract sub-volumes → align + average. Achieves sub-nm to ~3-Å resolution.

Dynamo (Castaño-Díez 2017 J Struct Biol). GUI-based.
EMAN2 e2spt.
PEET (UC-Boulder). Particle Estimation for Electron Tomography.
Relion 5 tomography. Native STA via subtomogram representations.
Warp + M. Tegunov pipeline; per-tilt motion + CTF + multi-particle refinement.
emClarity.
STOPGAP.

In situ structural biology

Cryo-ET of intact cells / cell lamellae prepared by cryo-FIB-SEM milling (FEI Aquilos, Aquilos 2, Helios Hydra, Hitachi NX5000-Sapphire, Zeiss Crossbeam 350 Cryo, JEOL JEM-1400 cryo-FIB).

Major successes: ribosome structures in situ at 3.5 Å (Tegunov-Cramer 2021 Nature); nuclear pore complex at 9-12 Å (Beck-Hummer 2022 Nature); cilium / flagellum (Bui-Pigino); HIV maturation (Briggs 2015 Cell originals, ongoing).

Helical reconstruction

For helical assemblies (actin, microtubules, amyloid fibrils, tobacco mosaic virus, type IV pili, intermediate filaments). Treat each segment as particle of helical symmetry; refine helical parameters (rise, twist) jointly with structure.

Relion 3D-helical (He-Scheres 2017).
CryoSPARC Helical Refinement.
SPRING (Desfosses-Sachse 2014).
IHRSR (Iterative Helical Real Space Reconstruction; Egelman).

Amyloid structure determination has been transformed — Aβ, α-synuclein, tau (multiple polymorphs), TDP-43 fibrils all solved by helical cryo-EM 2017-2024 (Goedert-Crowther MRC-LMB; Eisenberg-UCLA). Tau structures distinguish Alzheimer paired-helical filament from chronic-traumatic-encephalopathy fold from corticobasal degeneration fold.

Micro-ED (microcrystal electron diffraction)

Tilt-series electron diffraction from nano-crystals (~100 nm-2 µm). Bridges X-ray crystallography and cryo-EM — works on small molecules + peptides + small proteins.

Pioneered by Tamir Gonen (HHMI-UCLA). Achieves sub-Å resolution on small molecule crystals.

Applications: pharmaceutical small-molecule structure (Pfizer, GSK in-house; pharmaceutically important alternative to X-ray when only nano-crystals are obtainable); structure determination of peptide nanocrystals (insulin nano-crystal); natural product structure elucidation.

Software: MicroED.org pipeline; XDS, DIALS (standard X-ray crystallography software re-purposed); SHELXC/D/E for direct methods.

AlphaFold integration

AlphaFold2 (Jumper-Hassabis 2021 Nature; Nobel Chemistry 2024) and successors (ESMFold, OpenFold, RoseTTAFold, AF-Multimer, AlphaFold3) provide accurate atomic-resolution structure predictions that complement cryo-EM workflow at several stages.

Building from AF model into density

At moderate cryo-EM resolution (3-5 Å), side chains are ambiguous; rigid-body docking of AF model gives starting model. ChimeraX Fit-in-Map, Phenix.dock_in_map.

Hybrid AF + cryo-EM modeling

ModelAngelo + AlphaFold. Use ModelAngelo for backbone tracing, AF sidechain conformations as restraint.
AlphaFold-Multimer + cryo-EM. Predict complex; use density to validate / discriminate among predicted models.
DiffMap. Generative model conditioned on density.

Validation

AF-predicted models with PAE / pLDDT scores can be over-interpreted. Density is the empirical evidence; AF is a strong prior but not ground truth. Particularly for novel folds, mutations, ligand-bound states — AF may extrapolate wrongly.

Map and model deposition

EMDB (Electron Microscopy Data Bank)

Affiliated with PDB (PDBe, RCSB, PDBj). All published maps deposited as MRC or CCP4 format. EMDB ID prefix EMD- (e.g., EMD-12345). Metadata: imaging conditions, software, resolution, B-factor.

PDB deposition

Atomic models deposited at RCSB / PDBe / PDBj. mmCIF format mandatory for high-resolution (~deprecated PDB-format depositions).

Validation reports

PDB validation server runs MolProbity, EMRinger, Q-score, FSC, side-chain rotamer, Ramachandran, atom clash; published as part of structure paper expectations.

OneDep (wwPDB)

Single deposition portal for X-ray + cryo-EM + NMR + cryo-ET; runs validation automatically.

Commercial cryo-EM services

Outsourced data collection has matured into a multi-vendor industry:

NanoImaging Services (San Diego). Pharma-focused; data collection on Krios + processing. ~$15-30k per dataset.
Eaglecraft Cryo-EM Services (acquired by Curia 2023).
Wibu / cryo-EM consulting. Boutique.
Polara, Mentora Genomics.

Academic national centers

NCCAT (Simons Electron Microscopy Center; NYSBC New York). NIH-supported; competitive access; free for funded users.
PNCC (Pacific Northwest Center for Cryo-EM; OHSU). NIH-supported.
NCEF (National Cancer Institute Cryo-EM Facility, Frederick MD).
SEMC at NYSBC.
EMBL Cryo-EM Heidelberg.
eBIC (Diamond Light Source, UK).
CM01 ESRF Grenoble (France).
CryoNet (Sweden).
MRC-LMB internal (Cambridge UK; Scheres + others).

Access models: peer-reviewed proposal (most national centers; 1-3 month wait); fee-for-service academic / commercial; in-house at well-funded labs and major pharma.

Cost economics

Capital

Titan Krios G4 + Falcon 4i / K3 + Selectris: $8-12M install + $300-500k/yr service.
Glacios + Falcon 4i: $2-4M install + $200k/yr.
FIB-SEM (Aquilos 2): $2-4M.
Vitrobot Mark IV: $60-80k.
Cryo storage dewars + grid handling: $50-150k.
GPU compute cluster (for processing 100-500 TB/yr): $200-500k.

Per-structure cost

National center via NIH proposal: ~free (taxpayer funded) but 3-12 month wait.
Commercial service: $15-50k per dataset; $30-100k including processing + report.
In-house data collection + processing labor: $3-10k consumables + 1-2 month FTE.
Sample prep + biochem: $5-50k depending on construct optimization.

Comparison to X-ray crystallography

Cryo-EM total cost per structure (~$30-100k industry, ~$5-20k academic with subsidized facility access) is comparable to X-ray crystallography when crystallization succeeds, but cryo-EM bypasses the often-multi-year crystallization bottleneck for large flexible complexes.

Resolution-vs-particle-count relationship

Empirical B-factor relation (Rosenthal-Henderson 2003 J Mol Biol): resolution ~ B-factor × log(N_particles)^(-1/2). Halving B (better grid, less motion) doubles effective N. Typical:

3.5 Å with 50-100k particles, B ~150 Å².
3.0 Å with 100-300k particles, B ~100 Å².
2.5 Å with 300k-1M particles, B ~70 Å².
2.0 Å with 1-3M particles, B ~50 Å².
<2.0 Å (atomic) requires very rigid sample, ultra-stable microscope, cold-FEG, Cs-corrected optics, gold grids, often >5M particles.

Records: apoferritin 1.15 Å (Yip-Stabrin 2020 Nature) and 1.04 Å (Nakane-Kimanius-Lindahl-Scheres 2020 Nature) — but apoferritin is unusually rigid.

Preferred orientation problem

Particles tend to adopt limited orientations at the air-water interface — Fourier-space anisotropy in reconstruction. Diagnostics: Cone-shaped angular distribution histogram, asymmetric FSC, 3DFSC (Tan-Subramaniam 2017 Nat Methods) showing direction-resolved resolution.

Mitigations:

Detergent (low concentration). 0.005-0.01% n-octyl-β-D-glucoside (β-OG; Anatrace O311) at last step before vitrification.
Spotiton / Chameleon. Reduce AWI exposure time → reduce orientation bias.
Continuous carbon, graphene oxide. Particles adhere to support rather than partitioning at AWI.
Tilt collection. ~30° stage tilt fills missing-cone (Tan-Subramaniam 2017).
Crosslinking + alternative buffer. GraFix gradient + glutaraldehyde (Kastner-Stark 2008) — also helps for complex stabilization.
Engineering scaffold / fiducial. DARPin-rigidified, antibody/Fab-bound, megabody-bound (Steyaert; nanobody-derived scaffold) — adds mass, breaks orientation bias, improves alignment.

Membrane protein workflows

Membrane proteins remain harder than soluble proteins. Detergent micelle adds background, often destabilizes. Reconstitution approaches:

Detergents. DDM (n-dodecyl-β-D-maltoside; Anatrace D310), LMNG (lauryl maltose neopentyl glycol; Anatrace NG310), GDN (glyco-diosgenin; Anatrace GDN101), DM, OG. LMNG and GDN preferred for cryo-EM; lower CMC, smaller micelle.
MSP-nanodiscs (Sligar). Membrane scaffold protein wraps lipid bilayer disc; reconstitution with detergent → lipid+MSP → remove detergent. Cubic nanodiscs (Sligar 1D1, 1E3, etc.) or saposin A nanoparticles (Frauenfeld-Carlsson-Nordlund 2016 Nat Methods).
Saposin A (Salipro). Lipid + saposin nanodisc; commercially via Salipro Biotech.
Styrene-maleic acid (SMA) polymers / DIBMA. “Polymer nanodiscs”; extract directly from membrane.
Amphipols (A8-35). Anatrace; substitute detergent post-purification.
Peptide-bicelles, bilayer-mimetics.

GPCR-G-protein cryo-EM has been the great success — Brian Kobilka, David Julius, Georgios Skiniotis platforms — hundreds of GPCR structures since 2016 in complex with agonists, antagonists, allosteric modulators (e.g., GLP-1R + semaglutide, GIP-R + tirzepatide; Liang-Eric Xu 2018-present Nature, Cell series).

Recent benchmark structures

Apoferritin 1.04-1.15 Å. Atomic resolution benchmark.
Nuclear pore complex 9-12 Å in situ. Largest macromolecular assembly fully resolved (Beck-Hummer 2022 Nature).
Spliceosome at multiple intermediate states. Shi lab Tsinghua (2015-present) — ~10 PDB structures of spliceosome conformational cycle.
Ribosome / 70S, 80S, mitoribosome. Cryo-EM has replaced X-ray as primary method for ribosomal structures.
SARS-CoV-2 spike in pre-fusion and post-fusion (Wrapp-McLellan 2020 Science; Cai-Chen 2020 Science) — informed mRNA vaccine design within weeks of pandemic onset.
AlphaFold2 + cryo-EM of mu opioid receptor and other GPCRs.
Tau and α-synuclein amyloid filaments — multiple disease-specific polymorphs (Goedert-Crowther; Eisenberg).

Practical workflow — single particle from scratch

Sample QC. SDS-PAGE silver stain, analytical SEC (Superose 6 Increase 10/300, Cytiva 29-0915-96; or Superdex 200 Increase), DLS, mass photometry (Refeyn TwoMP). Mass photometry confirms oligomeric state at nM concentrations; transformative for low-yield samples.
Negative-stain screen on 2-3 buffer conditions to confirm particle distribution and absence of aggregation.
Cryo screening grids on Glacios at NCCAT or in-house — 3-5 grids, varying concentration ×2, blot time ×2, hole spacing.
Pilot dataset (~500-1000 movies) → CryoSPARC Live → 2D classes → judge orientation and resolution potential.
Production data collection ~ 3-7 days continuous on Krios; 3-15k movies depending on particle size.
Processing pipeline in CryoSPARC: motion correction (MotionCor2) → CTF estimation (Patch CTF) → blob picking → 2D classify → re-pick with Topaz trained on good classes → 2D classify → ab initio → non-uniform refinement → CTF + global refine → 3D class to isolate states → final non-uniform refinement.
Map post-process: B-factor sharpen, mask, DeepEMhancer; resolve at FSC=0.143.
Model build: dock AF prediction → ModelAngelo / ChimeraX → refine in Phenix.real_space_refine + Coot manual.
Deposit: PDB / EMDB via OneDep; FSC, validation report, model coordinates.

Total wall-clock from purified sample to deposited structure: ~4-12 weeks for well-behaved samples; 6-24 months for hard ones.

Open challenges

Smaller particles

Below ~80 kDa, signal-to-noise approaches limit. Phase plate, graphene support, fiducial scaffolds extend lower limit; sub-50-kDa structures still rare.

In situ atomic resolution

Cryo-ET STA at <3 Å in cells remains hard — radiation dose budget limits SNR. FIB-milling lamellae + better STA software pushing the limit (3.5 Å ribosome in cells, Tegunov-Cramer 2021).

Dynamics

Steady-state populations capture energy minima; intermediates require time-resolved methods. Time-resolved cryo-EM via fast mixing (Frank-Ourmazd) → spray-freezing 5-100 ms after substrate addition → captures sub-second intermediates.

Conformational continuum

Many machines (e.g., ribosome translocation) sample continuous conformations. cryoDRGN, 3D flex extract these but require careful regularization.

Throughput vs depth

A Krios costs $8-12M and produces ~1 structure per FTE-month. Throughput improvements: parallel detectors, dual-axis cassette, faster autoloaders, distributed-data-collection sites.

Compendium

Explorer

Cryo-EM and Structural Determination

Cryo-EM and Structural Determination

See also

Sample preparation and vitrification

Sample quality requirements

Negative-stain pre-screen

Vitrification

Grid choice

Glow discharge / plasma cleaning

Microscopes

High-end (300 kV)

Mid-range (200 kV)

Low-cost / screening

Detectors

Direct counting

Phase plate

Data acquisition

Microscope control software

Acquisition strategy

Aberration-free image shift (AFIS) and beam-image-shift (BIS)

Image processing pipeline

Motion correction

CTF estimation

Particle picking

2D classification

Ab initio 3D reconstruction

3D refinement

Heterogeneous reconstruction

Resolution and FSC

Map post-processing

Software ecosystems

CryoSPARC (Structura Biotechnology)

Relion (Sjors Scheres, MRC-LMB)

Warp / WarpTools / M / mowarp (Tegunov-Cramer)

EMAN2 (Steve Ludtke Baylor)

SPHIRE / SPARX

IMOD (Mastronarde UC-Boulder)

Phenix / Coot / ChimeraX

ModelAngelo (Jamali-Scheres 2024 Nature)

ModelCraft (Bond-Wilson 2023 Acta Cryst D)

Buccaneer, ARP/wARP

Tomography and sub-tomogram averaging (STA)

Acquisition

Tomogram reconstruction

Sub-tomogram averaging

In situ structural biology

Helical reconstruction

Micro-ED (microcrystal electron diffraction)

AlphaFold integration

Building from AF model into density

Hybrid AF + cryo-EM modeling

Validation

Map and model deposition

EMDB (Electron Microscopy Data Bank)

PDB deposition

Validation reports

OneDep (wwPDB)

Commercial cryo-EM services

Academic national centers

Cost economics

Capital

Per-structure cost

Comparison to X-ray crystallography

Resolution-vs-particle-count relationship

Preferred orientation problem

Membrane protein workflows

Recent benchmark structures

Practical workflow — single particle from scratch

Open challenges

Smaller particles

In situ atomic resolution

Dynamics

Conformational continuum

Throughput vs depth

Adjacent

Further reading

Graph View

Table of Contents