NMR Winterschool 2025: Protein-ligand structure determination using NOEs

In this tutorial we will

1. learn how to calibrate NOE data to obtain distance restraints,
2. derive a protein structure with the software CYANA,
3. try how the exchange of the ligand between free and bound state influence the NOESY spectra,
4. derive the structure of protein-ligand complex,
5. try how the incomplete assignment due to chemically identical atoms affect the NOESY simulation.

We will use the software CYANA. At wolf.ncbr.muni.cz

• For viewing the molecular structures, we will use VMD. 
  

  • module add vmd

• For NOESY simulations, we need a python environment.

  • module add anaconda3:2024.02

  • conda init

  • source .bashrc

  • conda activate base

Instructions:

• download exercices.tgz to and untar (tar xzvf exercises.tgz) them in some accessible folder.
• download  the pythonScripts.tgz, if you wish to do the calculations yourself (many are time consuming). 
  • place the python directory, e.g., to the bin folder, and ensure that the bin/python is in the $PYTHONPATH:
  • export PYTHONPATH=$PYTHONPATH:$HOME/bin/python
• lecture slides presentation.pdf

Part 1

Introduction for task 1:

In real  experimental situation, we have never complete set of distances between each pair of protons. The NOE crosspeaks are detectable for distances up to around 5 Angstrom. From these, many signal would share the
same frequency in the spectrum, and thus, assignment between signal and atom (atom pair) can be done only within some group, or not at all. Furthermore, the experimentally-derived distances contain various sources of error.
Partly, it is due to random noise, but partly due to incompletely resolved relayed transfer  and partly due to different (local) dynamics influencing the cross-relaxation rate.
Let's start anyway with the unrealistic situation, where we know all the distances within 5.5 A, accurately.

They are written Upper distance Limit files (here PxP.upl) file, which is used by CYANA.

Intermezzo: CYANA

There is no closed-form formula to calculate the conformation (structure) from a set of distances. The setup starts with defining an energy penalty for every experimental distance not fulfilled by the molecular conformation. These are also called distance restraints. Starting from one chosen conformation, and trying to minimize the structure (using steepest descent or other local method) to fulfill the distances measured by NOE (or any other means) would fail: the structure would end-up in a local minimum. Instead we have to search for a global minimum. A commonly used algorithm for a global minimum is called simulated annealing, where the molecule is heated up such that high energy barriers (due to van der Waals clashes) can be surpassed. By a subsequent cooling, the imposed distance restraints will drive the molecule towards the conformation with minimal violation of the distance restraints. Many attempts will nevertheless end up in different local minima, and hence, only a subset of resulting conformers, the lowest-energy conformers will be likely to represent the global minimum.

In practice, we have to input the knowledge about the covalent (bonding) structure of the molecule, and the distance restraints. The bonding structure can be as simple as the chain of amino acids, as the standard programs would have libraries of the actual atomic bonding (topology) for those. For an unknown molecule, we have to supply a full topology ourselves. These would be different for different programs.

We will use a program specialized in structure calculation from NMR restraints: CYANA by Prof. Dr. Peter Günter.

CYANA can obtain the bonding topology from a .mol2 molecular structure file, converting it into its own (library) format, a .lib file. This library file will contain information about one molecule, but since biopolymers - proteins contain chain (sequence) of building blocks like amino acid residues of nucleotides, there has to be also information about the sequence of those building blocks. In our case, it contains only one line: the name and the index of our ligand molecule.

Hint: More information on the CYANA commands etc. is in the CYANA 3.0 Reference Manual.

Remark: CYANA is a proprietary software. For any installation problem, contact Peter Güntert, the author of CYANA.

Task 1 (030)

• In 000_cyana subfolder, look into the configuration file, CALC.cya, using a text editor/browser.
  The set of structures will be written in demo.pdb
• Start the calculation by

  cyana CALC.cya

• After the calculation is ready, look at the .owv file.
  See the target functions, its variation.
  See the RMSD - root mean square displacement.
• To view the structure, use

  vmd demo.pdb

  In VMD, the default view shows the interatomic bonds as lines.
• Go to Graphics->Representations
  and change the Drawing Method to CPK.
  To see the common representation for proteins, choose NewCartoon as the Drawing method.
  See how alpha-helices, beta-sheets and loops are identified.
• To simulate the NOESY spectrum, in the 030 folder,

  run ./proteinLigandCalcNOESY.py final.pdb (can take 20 min on some architectures)

Introduction to Task 2

It used to be common to only simply classify the experimental NOE intensities to strong, medium and weak, and assign corresponding distance ranges of around 2 Angsrom for the strongest and 5.5 to the weakest.

031

Here we use a technique, which would be a starting point for more accurate methods. It uses spectra recorded for different mixing times. As the NOESY spectrum of a protein can take days to record, recording series of them is a large investment. In that the NOE crosspeak volume (intensity) is divided by a geometric mean of the corresponding diagonal volumes. The slope is a better approximation of the crossrelaxation rate than if this "linearization" or "normalization" was not done. In the case of two isolated nuclei, such buildup curve is approximately linear over longer mixing times.

• Check some of the simulated buildups, the buildupsLinearized in the supplied folders.
• Without further effort, what are the chances to get accurate distances from these?
• Check how many distance restraints we have: wc -l PxPapp.upl
• Start the structure calculation, or see the ready demo.pdb and demo.ovw file.

032

Here we probe three effortless options to obtain a better set of distances. The first comes from an idea, that short distances are much less likely to be affected by relayed transfer (spin diffusion), so we try to keep only those, within 2.5 Angstrom. Furthermore, to be safer, the approximate distances are multiplied by 1.5

• Check how many distances are left.
• Visualize the structure in VMD, instead of using only one Drawing Method, press Create Rep in Graphical representation. Keep one as Lines and other as NewCartoon.
• (Doublecklick on one of the copy would hide its visibility.)
• What can be problem when using only this short distances?
• Is the result satisfactory?

034

In the next simple attempt, we multiply all the distances by a constant factor (1.75) and use again those within 5.5A.  Commonly we would expect up to around 10 constraints per aminoacid residue.  In the previous example, it was still around 12800/97 > 100 restraints per residue. Here we try to use, only every 10th restraints.

• Check how many are left, what is the statistics of the structure calculation. And
• visualize the structure.
• Switch again to NewCartoon drawing method. What happens?
• What do you think, were these attempts good enough?

Task 3: Adding the exchange

.
We will investigate the effect of exchange of the ligand between the free state in the solution and bound to the protein.

The NOESY spectrum of the ligand  (the intramolecular NOEs) would be formed as a population  average of the bound and free form. In the free form, cross-relaxation rate is commonly negative, whereas in the bound form, it is positive, the same as the protein. Moreover, it is commonly much larger (absolute value).

• For the case of 1:10, 1:1 and 1:200 of protein:ligand fractions, in 035, 035...LessOfLigand and 035...MoreOfLigand.
• Check the simulated NOESY spectra (!note that the axes are not chemical shifts but simply atomic indices!)
  • see the file: NOESYZoomLigand.pdf
• What more is not realistic about this simulation?

038 

• We check the effect of the exchange rate on the resulting cross-relaxation rates.
  See e.g. like: tail -n 15 outputKex5e*
• What would be the consequence of "not so fast" exchange?
• run ./proteinLigandCalcNOESY.py triPepMod.pdb
  • experiment with settings in parameters*.json files.

What is even more important is the population weighting of the intermolecular cross-rates (hence also NOEs)
The "Re weighted" matrix contains includes the correction, whereas the "weight averaged over P, L, PL" does not.

041 

We have exact distances (assume we are able to obtain them), but we ignore the populations, so the intermolecular calibration is wrong.

• Check the PxL.upl
• What are the effects when calculating the protein-ligand complex?
• When viewing the complex, in graphical representation, you can select the atoms of proteins by "proteins" and the ligand as "not protein"

046 

Here the populations are take into account correctly.

• What is the difference WRT 041 ?

solutions

Assembled by Dr. Jiří Mareš, shaped by discussion with Prof. Julien Orts and other members of the research group  (https://bionmr.univie.ac.at/people/)