Protein NMR Spectroscopy II Reinhard Wimmer – Aalborg University Focus: how NMR data becomes 3D structure.
Key themes:
The Escher artwork symbolizes structural ambiguity — multiple possible interpretations depending on perspective.
Protein NMR structure determination follows this pipeline:
Important message: 📌 Data collection is long. Computation is long. Interpretation is iterative.
Three fundamental types:
→ Define secondary, tertiary, quaternary structure
→ Define local conformation
→ Define global fold and domain orientation
Important distinction:
Solving NMR structures is like:
Meaning:
Structure determination = constraint satisfaction under uncertainty.
Clear methodological niches:
NMR:
X-ray:
X-ray easier once crystal obtained. NMR allows more functional studies.
Use NMR when:
There are three main structural data types:
Distances are often the most important.
Same classification repeated:
Distances define structure extremely well.
Each cross peak = one distance constraint.
But problem: 👉 WHO IS WHO?
This requires:
Without assignment, NOE peaks are meaningless.
Because structure = spatial arrangement.
If you know enough pairwise distances → geometry is constrained.
Example:
Long-range NOEs are most important for defining tertiary structure.
Higher dimensions = better resolution.
Programs:
They:
Requires near-complete resonance assignment.
Key equation:
V = rac{k}{r^6}
Thus:
r = left( rac{k}{V} ight)^{1/6}
But practically:
So instead of exact distance: 👉 Use upper distance limits
Empirically often behaves like:
V le rac{k}{r^4}
Important: NOEs are converted to distance restraints, not exact values.
Characteristic patterns:
These patterns help identify secondary structure.
Combines:
To map:
Sometimes detectable via:
But: ⚠️ H-bonds should NOT be added unless strong evidence exists.
Insert spin label (unpaired electron).
Effect:
Use:
PREs give long-range distance information (up to ~25 Å).
Very powerful for domain orientation.
Three sources:
NOEs:
Angles and orientation add additional constraints.
Cα and Cβ shifts differ in:
Secondary chemical shift: Observed − random coil value
Patterns:
Programs:
Use chemical shifts to predict:
Machine learning + database comparison.
Provides torsion angle restraints.
Electron-mediated, through-bond interaction.
Karplus relationship:
J = Acos^2( heta) + Bcos( heta) + C
J depends on dihedral angle.
Common example: ³J(HN–Hα)
Large J (~8 Hz) → β-sheet Small J (~3–4 Hz) → α-helix
Thus J-couplings provide dihedral angle constraints.
Normally:
If weakly aligned:
How to align?
RDCs give:
Extremely valuable for multi-domain proteins.
Programs:
Input:
Procedure: Iterative:
Structure calculation minimizes deviation between:
Many conformers generated.
Select best-fitting ensemble.
1️⃣ Target function (restraint violations)
2️⃣ RMSD (precision of ensemble)
3️⃣ Ramachandran statistics
RMSD = root mean square deviation between structures.
Lower RMSD → tighter ensemble → higher precision.
But: Depends on:
You can artificially lower RMSD by fitting only secondary structure.
NMR, X-ray, modeling can give different structures.
Solution vs crystal packing effects.
Check:
Use:
Use independent data (RDCs, unusual shifts) for validation.
Allowed φ/ψ regions:
Low percentage in disallowed region = good structure.
Example publication statistics:
Important: RMSD of structured core is more meaningful than full-length RMSD.
Protein NMR structure determination integrates:
→ Define fold
→ Define local conformation
→ Define global orientation
Then:
Structure calculation → Minimize violations → Select best ensemble → Validate geometry → Report RMSD + statistics
NMR structures are ensembles, not single static models.