Molecular Docking

Molecular docking is a method to investigate the various binding modes of two macromolecules (like proteins) in forming a stable complex.

Docking methods use various sampling algorithms to generate all sorts of possibilities to dock the two input protein structures, put them through a scoring function, and output the best scored complexes. Different docking softwares differ in their sampling methods and scoring functions, and such softwares can be used for different purposes.

We used molecular docking to investigate two questions in dry lab:

1. What are the sensitive residues on the human FcRn receptor that bind to this peptide?
2. What is the best position on the scFv for the FcRn binding peptide to be added (peptide at either N-terminus, C-terminus or Linker)?

We used AutoDock Vina and HADDOCK 2.4 web server for these purposes

AutoDock Vina

• What are we trying to achieve?
  We are trying to identify the sensitive residues of the FcRn receptor when bound to our NeoFv on the 16 amino acid peptide region. We are trying to dock the predicted structure of the 16 amino acid peptide as a ligand with the human FcRn receptor structure found from the protein data bank (PDB: 3M17)
• Why is this the best method?
  Docking the FcRn binding peptide only, instead of the entire NeoFv, can sample a larger conformational space of the peptide. Otherwise, we would have had to rely on other protein-protein docking softwares which may not be as exhaustive in sampling.
  AutoDock is the leading protein-ligand docking software according to benchmarks. It is the most accurate and much faster than other softwares.
• What does this tell us?
  
  • Sensitive residues on the FcRn receptor when complexed with different variants of our FcRn binding peptide.
  • These sensitive residues can further be used as starting points to run HADDOCK 2.4 simulations on NeoFv variants, to investigate the best position of our peptide (N-terminus, C-terminus or Linker)
  • Input complexes for molecular dynamic simulations to investigate complex stability.

Technicals

AutoDock Vina is a protein-ligand docking software. Our main purpose with AutoDock was to identify the sensitive residues on the FcRn receptor. These were identified after analysis of amino acid interactions in the best scored docking complexes.

Preparing the ligand

In our case, the ligands were different variants of our FcRn binding peptide (the 16 amino acid extension that turns an scFv into a NeoFv).
The structures of these FcRn binding peptide variants were generated through AlphaFold 2.0.
Using AutoDockTools software, we prepared the ligand for docking input by editing the PDB structure to a PDBQT file, adding polar hydrogen atoms and calculating charges on atoms.

Preparing the protein

The protein here is the human FcRn receptor. We used PDB 3M17, a crystal structure of a similar inhibitory peptide in complex with the human FcRn receptor. We removed the inhibitory peptide and heteroatoms from the crystal structure to arrive at a PDB structure which only had the human FcRn receptor.
This FcRn receptor PDB structure was converted to a PDBQT file by adding polar hydrogens and charges.
We defined a grid box on the FcRn receptor in which the ligand will sample through its conformational poses to search for a site with the best binding energy.

Our peptide, according to literature, was generated to be binding competitively against the human IgG to the FcRn receptor, and a complex structure of a related peptide variant was also available in the protein data bank which bound at a nearby but not an exactly similar site. We therefore decided to define our box such that it would incorporate both these sites.

Running the Dock

We had 4 different FcRn binding peptide variants to be docked with the FcRn receptor.
Taking advantage of the computational resources we had access to through our institute’s PARAM Brahma supercomputer, we set the exhaustiveness parameter of the docking configuration file to 100, while the default was 8.
We generated slurm files to instruct the cluster to use the AutoDock Vina package and take our inputs. We received the results in a few hours.

Analysis of the results

All our peptide variants generated negative binding energies in the top docked structures, as we expected.
The complexes with the lowest energies for all the variants were then analyzed in AutoDockTools to identify the interacting residues on the FcRn receptor with this peptide. These residues were later input into HADDOCK 2.4 as active residues for generating complexes of all the position variants.

For visualization of how AutoDock generated these complexes, we superimposed all the best predicted structures with the crystal structure of the similar peptide

We found out that our cyclic peptide bound at a similar site as the crystal structure did, especially the position of the consensus sequence was very similar, which is likely to be involved in binding.

We also noticed that the linear variants of our peptides had a more negative score than the cyclic ones. When the complexes were analyzed in AutoDock tools, we found that linear variants had a higher number of interacting residues than the cyclic ones, which may help explain the more negative energies.

However, as we were led to the conclusion through our AlphaFold prediction and structural analysis that linear variants should also maintain their antiparallel beta structure, we realized that AutoDock essentially unfolded the linear variant, giving us unrealistic binding sites. We expect the linear variant to also maintain a cyclic-like structure with the consensus sequence involved in major binding.
We later confirmed this through molecular dynamic simulations on cyclic-like structures of the linear variant, that the linear variant seems to be stable in a cyclic-like structure in solution.

We predict that calling this variant ‘linear’ is a gap in current scientific literature about this topic, and they could instead be called disulfide bonded and non-disulfide bonded.

Analysis of the results

Through our AutoDock simulations, we were able to arrive at the following conclusions:

• Binding site of our peptide is likely the same as the crystal structure of a similar peptide (PDB: 3M17)
• The neighboring interacting residues on the FcRn receptor in the cyclic variant complexes can be inputted as active residues in HADDOCK 2.4.

HADDOCK 2.4

• What are we trying to achieve?
  We are trying to screen for the best NeoFv variant in terms of position or conformation of the FcRn binding peptide (either at the C-terminus, N-terminus or the linker). We used HADDOCK 2.4 to calculate binding scores of the variants to investigate whether there is any influence of the particular position and conformation of the peptide on the binding to the receptor.
• Why is this the best method?
  HADDOCK is one of the most preferred flexible docking softwares. It not only has the capability to generate protein-protein complexes with relaxation of the protein backbone, but it can also relax the side chains of the molecules.
• What does this tell us?
  Based on the output of the best scored complexes in each docking run, we can screen for the best binding variant of our NeoFv. This is helpful in deciding which variant to synthesize in the lab.

Technicals

High Ambiguity Driven Protein-Protein Docking, or HADDOCK, is a docking software for modeling protein complexes and scoring their interactions. It is a flexible docking approach, which means it can model changes in conformation of the protein backbones and side chains, and it is superior to other protein-protein docking methods which treat the two entities as rigid bodies.

We used HADDOCK 2.4 web server to input the structures of our NeoFv variants to dock with the FcRn receptor through the FcRn binding peptide.
We predicted the structures of our NeoFv variants through AlphaFold.
We used PDB 3M17 and cleaned it to remove water, heteroatoms and other complexes to arrive at the right structures to use as inputs.
The amino acid residues in the consensus sequence of the FcRn binding peptide were used as active residues, and the active residues on the FcRn receptor were the ones identified via AutoDock.

However, to be more conclusive about our results, we took all residues in the 5Å radius of the complexes generated by AutoDock and the similar crystal structure as active residues. Here is an example of one of the docked complexes generated through HADDOCK.

We obtained the results from the HADDOCK jobs submitted to the server. We found all energies to be negative, as we had expected.

We plotted the HADDOCK scores of these complexes to compare our NeoFv variants. We did not find any significant enough results to conclusively choose one variant over the other based solely on HADDOCK score. We finally chose to synthesize our gene as the VH-LinY12H-VL variant based on preliminary results in literature.

Conclusion

• We plotted the HADDOCK scores of these complexes to compare our NeoFv variants.
• We did not find any significant enough results to conclusively choose one variant over the other based solely on HADDOCK score.
• We finally chose to synthesize our gene as the VH-LinY12H-VL variant based on preliminary results in literature.

Molecular Dynamics

Future Improvements

To arrive at valid conclusions about the comparison of NeoFv position variants, metadynamics approaches in molecular dynamics simulations may be used to properly estimate the energy of dissociation.