We used state-of-the-art computational methods to support our work in the lab, proving the feasibility of our project and supplementing our understanding of the interactions between aminotransferases and keto acid levothyroxine fragments. Specifically, we conducted both docking and molecular dynamics simulations, the results of which are discussed below. You can find protocols describing our simulations here.

Why Docking and Molecular Dynamics?

Traditional experimental techniques require brute-force strategies to gather data on whether the experimental system of interest is likely to succeed. However, given resource and time limitations, this approach is increasingly infeasible. In order to aid and speed-up the process of scientific discoveries, computational techniques can be used to predict likely candidates and experiments. In our pursuit of using transaminases to produce levothyroxine precursors, Genehackers sought to utilize both static and dynamic simulations in order to add validity to our experiments. The primary transaminase we analyzed, 4je5, and the levothyroxine fragment with which we conducted our simulations are visualized below.

The primary aminotransferase we analyzed, 4je5, is shown in green with PLP cofactors present.
The keto acid levothyroxine fragment we used within our docking simulations.

Docking Simulations

Methods

We began simulations of our project using protein-ligand docking techniques, which provided useful information on the positions, orientations, and energies of bound ligands in our enzyme-substrate complexes. 3D structures of the transaminase enzymes were obtained from the RCSB Protein Data Bank6 and Swiss-Model3. In order to perform our protein-ligand docking simulations, an open-source GUI known as DockingPie1 was installed into PyMol to streamline our file preparation, docking, analysis and visualization. Herein, Genehackers utilized DckingPie to add missing hydrogen atoms; and clean the receptors of non-standard residues and water molecules. Then, the optimized receptors and ligands were inputted into Vina4 through the plug-in. PyMol itself was additionally utilized to visualize the results of the docking simulations.

The result of our docking simulation with 4JE5 and the keto acid levothyroxine fragment ligand.

Results

For each of the three possible candidates, multiple independent docking trials were performed with moderate simulational exhaustiveness. The simulations generated a list of minimum energy configurations of the ligand within the transaminase enzymes (Figure 1A). Based on the minimum energy configurations, Root-mean-squared-displacements (RMSDs) were calculated by comparing the position of each configuration to the configuration with the minimal energy for that enzyme (Figure 1B). Given that the docking simulations agreed with our predictions that the levothyroxine precursor would bind to the active-site, denoted by ligands near the active site having the lowest affinity energy, we were able to computationally validate that these transaminase enzymes were likely candidates for our system of interest.

Figure 1: Protein-ligand docking simulations on candidate aminotransferases.

(a) Estimated affinities of the docked ligand with the enzyme complexes. (b) Root mean squared displacements of different docked ligand configurations compared to the lowest energy configuration.

Molecular Dynamics Simulations

To improve our understanding of the protein structure/function relation we began running parallel molecular dynamics, which provide a complex visualization of particle movements over a short period of time. We obtained 3D structures of the transaminase enzymes from The Human Metabolome Database and Swiss-Model. In order to obtain the configuration files used for the MD simulations, we used QuickMD through the VMD graphic interface. The PDB file was first modified: we filled and connected the gaps of several chains of the protein and, and the file was cleaned up from cofactors and ligands. The files were prepared from the Advance Run setup settings. In the QuickMD setup, a minimal box of solvent was specified and the buffer was changed to 13, none of the other predetermined parameters were changed. The files were run through NAMD. Finally, we obtained the index file to generate a stride uniformly in a ratio 1:10 the resulting data. VMD was used to visualize the final data, extract the RMSD values for every frame, and create the movie using a New Cartoon drawing method and highlighting the active site. We were unable to complete our molecular dynamics simulations this summer, but gained invaluable experience setting up these simulations and understanding the insight they give us into the mechanisms behind our project.

Next Steps

Based on the results of the protein-ligand docking simulation, we were able to get an idea of where the precursor would bind to the active site. While these simulations helped to shed light on the interaction between the transaminase and the levothyroxine precursor, and computationally confirmed that it is a valid system in-silico; we would ideally like to utilize the strengths of both docking and molecular dynamics simulations to generate better understandings of the dynamic interaction between our enzyme and ligand. In the future, molecular dynamics can be utilized to perform protein-ligand dynamics simulations, enabling us to calculate the affinity energies more accurately and to visualize and analyze the procedural mechanism of the ligand binding to the enzyme active site. Moreover, a molecular dynamics simulation of the protein-ligand docking mechanism would allow for greater predictive power, meaning that candidate systems can be selected for wet-lab experimentation with more confidence.

References

  1. Rosignoli, S.; Paiardini, A. DockingPie: A Consensus Docking Plugin for PyMOL . Bioinformatics 2022, 38 (17), 4233–4234. https://doi.org/https://doi.org/10.1093/bioinformatics/btac452.
  2. Human Metabolome Database. https://hmdb.ca/ (accessed 2022-12-31).
  3. Waterhouse, A.; Bertoni, M.; Bienert, S.; Studer, G.; Tauriello, G.; Gumienny, R.; Heer, F. T.; de Beer, T. A. P.; Rempfer, C.; Bordoli, L.; Lepore, R.; Schwede, T. SWISS-MODEL: Homology Modelling of Protein Structures and Complexes. Nucleic Acids Research 2018, 46 (W1), W296–W303. https://doi.org/10.1093/nar/gky427.
  4. Eberhardt, J.; Santos-Martins, D.; Tillack, A. F.; Forli, S. AutoDock Vina 1.2.0: New Docking Methods, Expanded Force Field, and Python Bindings. J. Chem. Inf. Model. 2021, 61 (8), 3891–3898. https://doi.org/10.1021/acs.jcim.1c00203.
  5. Ribeiro, J. V.; Bernardi, R. C.; Rudack, T.; Stone, J. E.; Phillips, J. C.; Freddolino, P. L.; Schulten, K. QwikMD — Integrative Molecular Dynamics Toolkit for Novices and Experts. Sci Rep 2016, 6 (1), 26536. https://doi.org/10.1038/srep26536.
  6. RCSB Protein Data Bank. https://www.rcsb.org/.
  7. Naqvi AAT, Mohammad T, Hasan GM, Hassan MI. Advancements in Docking and Molecular Dynamics Simulations Towards Ligand-receptor Interactions and Structure-function Relationships. Curr Top Med Chem. 2018;18(20):1755-1768. doi: 10.2174/1568026618666181025114157. PMID: 30360721.