Lecture 7/8 Ex Paper 2 Cheng

Protein structure

🧊 What Is Single-Particle Cryo-EM?

Single-particle cryo-electron microscopy (cryo-EM) determines 3D structures of proteins and complexes without crystallization.

Instead of growing crystals (like in X-ray crystallography), we:

  1. Freeze proteins in vitreous ice
  2. Image thousands–millions of particles
  3. Average them computationally
  4. Reconstruct a 3D map

Recent advances (especially direct electron detectors) pushed cryo-EM to near-atomic resolution (<4 Å).

But — and this is important — it is not plug-and-play. There are many pitfalls.


🧪 Step 1: Protein Purification – Garbage In = Garbage Out

Cryo-EM depends on averaging thousands of identical particles.

If your sample is heterogeneous, your reconstruction becomes blurry or misleading.

Two Types of Heterogeneity

1️⃣ Compositional heterogeneity

  • Missing subunits
  • Sub-stoichiometric binding
  • Partial dissociation

Solutions:

  • Optimize buffer conditions (Thermofluor screening)
  • Affinity-tag weak subunits
  • Mild crosslinking (e.g., GraFix)
  • “On-column” crosslinking

⚠️ BUT: Crosslinking can introduce artifacts or trap only one conformational state.


2️⃣ Conformational heterogeneity

  • Flexible domains
  • Multiple functional states

Harder to fix. Possible approaches:

  • Add ligands, substrates, inhibitors
  • Lock protein into defined functional state
  • Carefully use crosslinking

🖼 Always Check with Negative Stain EM

The paper strongly emphasizes:

Every new project should start with negative-stain EM.

Why?

  • Quick
  • High contrast
  • Reveals aggregation
  • Reveals orientation bias
  • Reveals heterogeneity

This is your low-resolution quality control phase before expensive cryo work.


❄️ Step 2: Specimen Preparation – The Ice Matters

You must:

  • Protect sample from vacuum
  • Prevent radiation damage
  • Preserve native structure

Vitrification

Rapid freezing creates amorphous (vitreous) ice.

A perfect grid has:

  • Thin ice (but thick enough to fit particle)
  • Even particle distribution
  • Many orientations
  • No crystalline ice (no “bend contours”)

On page 3 (Figure 2), you see:

  • Raw movie frame
  • Motion trace
  • Thon rings before/after correction
  • Final motion-corrected image

This visually demonstrates how motion correction restores high-resolution information.


Common Problems

🧊 Ice too thick

→ Low contrast → Defocus spread

🧊 Ice too thin

→ Large particles excluded

🧲 Particles stick to carbon

→ Few particles in holes

Fix:

  • Apply sample twice
  • Use graphene
  • Modify glow discharge
  • Add mild detergent

⚠️ Preferred Orientation

If particles lie in one orientation: → Missing views → Bad 3D reconstruction

Solutions:

  • Thicker ice
  • Low detergent
  • Support film
  • Tilt data (very hard, lowers resolution)

📸 Step 3: Image Acquisition – Physics Matters

Cryo samples are extremely low contrast.

Images are taken in underfocus to create phase contrast.


Contrast Transfer Function (CTF)

The CTF:

  • Modulates image in reciprocal space
  • Produces oscillating Thon rings
  • Has zero crossings (information loss)

Because of this:

  • Images must be taken at different defocus values
  • CTF must be estimated and corrected

Trade-Off: Defocus

More underfocusLess underfocus
More contrastLess contrast
Worse high resolutionBetter high resolution

Small proteins (<200 kDa) often need high defocus → resolution limited.


Electron Dose

Too high: → Radiation damage

Typical:

  • ~20 e⁻/Ų for single image

With movies:

  • Dose fractionated
  • Early frames = high resolution
  • Later frames = damaged

🎥 Movies & Motion Correction

Major breakthrough.

Direct detectors allow:

  • Dose fractionation
  • Frame alignment
  • Beam-induced motion correction

Page 3 Figure 2 shows:

  • Motion trace
  • Improved Thon rings
  • Near-atomic resolution restoration

This is one of the main reasons cryo-EM became powerful in the 2010s.


💻 Step 4: Image Processing – Where Most Work Happens

This is the computational heart.


1️⃣ CTF Estimation

Parameters needed:

  • Voltage
  • Spherical aberration
  • Defocus
  • Astigmatism
  • Amplitude contrast

Fit theoretical CTF to observed Thon rings.

Correction:

  • Phase flipping
  • Full amplitude + phase correction

2️⃣ Particle Picking

Critical step.

Options:

  • Manual
  • Semi-automated
  • Template-based automated

⚠️ Template bias risk: Noise can match template → fake structures.

The HIV envelope example shows how template bias can mislead.

Rule: Only use template picking if particles are clearly visible.


3️⃣ 2D Classification

Goal:

  • Remove junk
  • Assess angular distribution
  • Generate high-SNR class averages

Based on K-means clustering (Figure 3 page 6).

Issue: "Group collapse" — dominant views attract more particles.

Advanced method: ISAC (iterative stable alignment and clustering)


🧱 Step 5: Initial 3D Model

Two approaches:


🧲 Tilt-based (RCT – Random Conical Tilt)

  • Collect tilted and untilted pairs
  • One angle known experimentally
  • Reliable but limited by missing cone
  • Often done in negative stain

🧮 Computational (Common lines)

Based on central section theorem: 2D Fourier transforms intersect along common lines.

Requires high-quality class averages.


🔄 Step 6: Refinement

Main method: Projection matching (Figure 3B)

Process:

  1. Generate projections of current 3D map
  2. Compare to particles
  3. Update Euler angles
  4. Reconstruct
  5. Iterate

📏 Resolution & FSC – A Major Caveat

Resolution measured via:

Fourier Shell Correlation (FSC)

Split dataset in half:

  • Reconstruct two maps
  • Correlate shells in Fourier space

Common thresholds:

  • 0.5
  • 0.143 (“gold standard”)

⚠️ Overfitting Problem

Noise can align. This inflates FSC.

Solution: Refine two halves independently ("gold standard refinement").

BUT authors emphasize:

There is no true gold standard yet.

Masking, filtering, processing tricks can artificially improve FSC.

Resolution number ≠ map quality.


🧩 Structural Heterogeneity (3D MRA)

If multiple conformations exist:

Use 3D multi-reference alignment.

Limitations:

  • Depends heavily on initial models
  • Depends on number of classes
  • K-means bias

Validation: Check local variability maps.


🔍 Validation & Interpretation

Resolution regimes:

ResolutionWhat You See
>10 ÅOverall shape
4–10 ÅSecondary structure
<4 ÅSide chains

Page 8 (Figure 4) shows:

  • Two independently refined TRPV1 maps
  • Similar overall structure
  • Only 60% particle overlap
  • Local resolution variation map

Important insight:

Local resolution is not uniform. Membrane region better resolved than extracellular region.


🧠 Interpretation Guidelines

Low resolution (>10 Å)

  • Architecture only
  • Docking risky

Intermediate (4–10 Å)

  • Helices visible
  • Docking reliable
  • Detect conformational changes

High resolution (<4 Å)

  • Side chains visible
  • De novo model building possible
  • Atomic interpretation

Always:

  • Avoid over-interpretation of poorly resolved regions
  • Check local resolution
  • Validate with tilt tests if possible

🧭 Big Conceptual Lessons from the Paper

  1. Cryo-EM is powerful but fragile.
  2. Sample quality determines everything.
  3. Motion correction changed the field.
  4. Resolution is not a single number.
  5. Overfitting is real.
  6. Validation must be rigorous.
  7. Interpretation depends strongly on resolution regime.

🏁 Final Takeaway

Single-particle cryo-EM workflow:

1️⃣ Optimize sample (negative stain) 2️⃣ Prepare good vitrified grids 3️⃣ Collect high-quality movies 4️⃣ Correct motion 5️⃣ Estimate and correct CTF 6️⃣ Pick particles carefully 7️⃣ Perform 2D classification 8️⃣ Generate initial model 9️⃣ Refine via projection matching 🔟 Validate with FSC + local metrics 11️⃣ Interpret based on resolution regime

Quiz

Score: 0/30 (0%)