🧬 3.1.2 NMR Bundle – Why NMR Structures Come as “Ensembles”

🔁 1. The Iterative Process of NMR Structure Determination

Determining a protein structure by solution NMR is not a one-step calculation. It is an iterative cycle (illustrated in Figure 3.1.4 on page 37):

Core Steps:

1. Acquire experimental NMR data
2. Assign resonances (match peaks to specific atoms)
3. Assign interactions (e.g., NOEs between atom pairs)
4. Translate interactions into structural restraints
5. Calculate structures
6. Analyze violations and errors
7. Refine and repeat

This loop is repeated multiple times until:

• A consistent set of restraints is achieved
• No major contradictions remain

Only then does the process branch into validation and quality checks.

🔎 Important: Early preliminary structures are sometimes used to refine assignments in later cycles — a feedback mechanism.

🎯 2. What Is an “NMR Bundle”?

Unlike X-ray crystallography (which gives a single structure), NMR produces:

🧩 A bundle of conformers (typically 20–40 structures)

Each conformer:

• Is a full atomic model
• Satisfies the experimental restraints
• Is equally consistent with the data

See Figure 3.1.5 (page 38):

• The backbone structures are superimposed
• Helices (red) and β-strands (blue) are shown
• Structural uncertainty is visualized as tube thickness

🧠 3. Why Not Just One Structure?

Because NMR restraints are not exact values — they are ranges.

Example:

Distance restraints are input as:

• Upper limit
• Lower limit (often van der Waals contact distance)

Why ranges?

🌀 Because NMR observables are averages.

Proteins in solution:

• Continuously sample many conformations
• On timescales up to hundreds of microseconds
• NMR measurements reflect an average over these states

⚠️ The averaged value may not correspond to any single real conformation.

So computationally: → Using intervals (ranges) is the most practical approach.

📏 4. Types of Structural Restraints

Structural restraints can include:

• 📐 Distance restraints (mostly from NOEs)
• 🧭 Orientational restraints (e.g., RDCs – residual dipolar couplings)
• 🔄 Dihedral angle restraints
• 🧲 Paramagnetic restraints (in metal-containing proteins)
• 💡 Chemical shift-derived restraints

Special note on RDCs:

Residual dipolar couplings can, in principle, be sufficient alone for structure determination. They provide orientational information, which can sometimes be more powerful than distances.

However:

• Extracting precise limits from NMR data is complex
• Requires careful error estimation
• Often prevents full automation
• May demand additional experimental time

📊 5. How Do We Describe Precision? (RMSD)

Since we have a bundle, we need to quantify:

How similar are the conformers?

This is done using RMSD (Root Mean Square Deviation).

Equation (3.1.1) (page 39):

RMSD compares corresponding atomic coordinates after superposition.

RMSD = sqrt{rac{1}{N} sum (x_A - x_B)^2 + (y_A - y_B)^2 + (z_A - z_B)^2}

Where:

• N = number of atoms compared
• Usually heavy atoms or backbone atoms only

📌 Global vs Local RMSD

🔹 Global RMSD

Measures overall structural precision of the bundle.

Standard reporting:

• RMSD of each conformer to the mean conformer

Important:

• The mean conformer is a geometric average
• It is often not physically realistic
• Sometimes energy-minimized before deposition in PDB

🔹 Per-residue RMSD (Local Precision)

Shown in Figure 3.1.6 (page 40).

Common pattern:

• High RMSD at termini and loops
• Low RMSD in helices and β-sheets

Why?

Three main reasons:

1. Loops sample more conformations
2. Core residues have more restraints
3. Motion can suppress NOEs (no restraints → more variability)

⚠️ Crucial clarification: Per-residue RMSD ≠ measure of local dynamics. It reflects:

• Restraint density
• Structural definition
• Sampling variability

True dynamics must be measured separately.

❗ 6. Common Misconceptions About NMR Bundles

❌ Misconception 1: The bundle shows the real conformational space.

Wrong because:

1. If restraints are sparse → variability is random sampling
2. Force fields are simplified → distribution not energetically meaningful
3. Protocols minimize RMSD → may artificially increase apparent precision

Therefore: The bundle often represents only a subset of conformations consistent with data.

❌ Misconception 2: RMSD reflects accuracy.

No.

• RMSD measures precision (reproducibility within bundle)
• Accuracy must be evaluated by:
  • Restraint agreement
  • Stereochemical quality
  • Validation parameters

🧩 7. Selecting Regions for RMSD Calculation

Since RMSD depends on superposition:

We should:

• Include well-defined regions
• Exclude poorly restrained/disordered regions

Methods:

• Manual selection (secondary structure only)
• Algorithms (e.g., structural order parameter by Snyder & Montelione)

Goal: Include as many residues as possible Exclude regions dominated by computational artifacts

🔍 8. Comparing Two NMR Structures

Comparing bundles is tricky.

To assess if a structural difference is meaningful:

Check whether:

Backbone RMSD between the two structures

is larger than the sum of their internal RMSDs

If yes → difference likely significant.

⚠️ But result depends on:

• Superposition method
• Which residues were included

More advanced statistical methods exist but are rarely used.

📏 9. RMSD Depends on Protein Length

Longer proteins → larger RMSD values.

A normalized metric exists:

• RMSD100

But it is not widely used because NMR proteins typically fall within a narrow size range.

🧠 Final Conceptual Takeaways

🧩 An NMR structure is:

A bundle of 20–40 conformers All consistent with experimental restraints

📐 RMSD measures:

Precision — not accuracy, not dynamics

🌀 The bundle does NOT represent:

True conformational ensemble in solution

🧪 Structural variability depends on:

• Restraint density
• Protein flexibility
• Superposition choices
• Computational protocols

🔬 Big Picture

NMR structure determination reflects a fundamental truth:

Proteins in solution are dynamic, flexible systems — not rigid objects.

The bundle representation is both:

• A limitation of experimental averaging
• A realistic reflection of structural uncertainty

It is a computational compromise between: Experimental data ⟷ Physical flexibility ⟷ Practical modeling constraints