Team:TU-Delft/Modeling/StructuralModeling

Structural Modeling

In order to engineer a yeast strain that is able to detect a tuberculosis (TB) molecule, its receptor should be designed in such a way that the molecule can act as an agonist. By modeling the olfactory receptor in silico, its biophysical and biochemical properties are investigated at the molecular level. The aim is to get a clear understanding of how a ligand binds in the receptor and how to mutate the binding niche to let it bind more specifically.

The main chemical compound that we want to act as an agonist, is methyl nicotinate (figure 1a), which is very close related to the agonist of the niacin receptor, niacin (figure 2).

Figure 1. Chemical compounds present in the breath of a TB-patient.[1] Methyl nicotinate (a), methyl phenylacetate (b), methyl p-anisate (c) and o-phenylanisole (d)

By reprogramming the binding niche of the niacin receptor, also known as GPR109A, an olfactory receptor for methyl nicotinate could be engineered.

To get a more reliable diagnose whether a patient has TB, receptors for the remaining ligands (Figure 1b-d) should be designed, synthesized and integrated in the membrane of the yeast-strain. Therefore, during this project, there has been an effort in designing a receptor for another ligand to improve the diagnostic result.

The structural modeling of this iGEM-project was based on the method of using the crystal structure of bovine rhodopsin receptor as a template for our olfactory receptors, as described earlier.[2] In contrast to the protein modeling software used in this paper, during this project the software YASARA was used for this purpose, mainly because of its many features like Molecular Dynamics (MD) simulations, docking ligands in a specific niche, etc.

Modeling approach

Below is the initial modeling approach described, applied to all olfactory receptors used in our research, unless written otherwise.

1. Firstly, the in silico simulations have to be redone in YASARA in order to produce similar results as reported.[2] This involves the modeling of the olfactory receptor hOR2AG1 by using the crystal structure of the bovine rhodopsin. An alignment executed by BLAST [REF] between the two sequences gives multiple gaps in the helices of the crystal scructure. By shifting the gaps to the nearest extra- or intercellular loop, these helices are preserved. This swapping method is completed by swapping the residues of bovine rhodopsin into the residues of the desired protein. Next, the binding cavities of the receptor must be defined and examined if it is similar. Run a 10 ns MD without any ligand and analyze if the binding niche shrinks in volume.

By docking the ligand amyl butyrate, used in the paper, in the binding niche of the hOR2AG1 olfactory receptor, and providing similar results as described, provides a conformation that YASARA is indeed applicable to the structural modeling. Executing a 10 ns MD on the vertical configuration of amyl butyrate and analyzing how well the hydrogen bonds are preserved, should give more insight in the reliability of the program.

2. The modeling approach for the rGPR109A receptor with its rI7 flanks is equivalent to the one of the hOR2AG1 protein. The 'wildtype' rGPR109A and hGPR109A receptors underwent the same procedure. Also, the hOR2AG1 receptor model is to be used for further simulations. After undergoing step 1, other ligands were used for all proteins.

For the rGPR109A/rI7, rGPR109A and hGPR109A models the molecules in figure ??? represent the ligands for the above described models.

Figure 2. Characterization of used ligands for all models. a) The ligands for GPR109A derivatives include niacin (A1 = C, A2 = N, A3 = OH), methyl nicotinate (A1 = C, A2 = N, A3 = OMe) and methyl isonicotinate (A1 = N, A2 = C, A3 = OMe).

After docking a ligand in the cavity of the receptor, a 10 ns MD simulation is executed and the preservation of the hydrogen bonds between the ligand and related amino acids are analyzed to examine how well the ligand is docked in the receptor. These results are compared with the experimental labwork data.

3. In all cases, depending on the conclusions of the in silico and experimentally results, specific point mutations in silico – and if time is left, also experimentally – can reconfigure the ligand-binding niche in such a way that it only specifically binds to the desired compound, i.e. only methyl nicotinate to the rGPR109A/rI7 receptor. Also, the compound methyl phenylacetate is closely related to isoamylbenzoate, which gained a higher affinity to the hOR2AG1 receptor after reprogramming the selectivity filter.[2] By mutating more amino acids in the binding cavity, a binding niche for methyl phenylacetate could be constructed.

Results

The modeling of the hOR2AG1 receptor

After building the model of the hOR2AG1 receptor by the method of the paper as described earlier, the calculated volume of the binding niche was examined graphically to conclude any similarity. While the cavity was indeed similar, the outcome of the 10 ns MD simulation without any ligand did not provide the correct end configuration of the binding niche.

A first reason why the model did not yield the same result, is that it might be poorly build. Repeated requests by e-mail to the authors of the paper, to provide their model to compare it, proved to be fruitless. Two other models were used during the 10 ns simulation; the model of the Zhang Server [REF] and the model build by the model building macro provided by YASARA. While the latter did not provide the desired result, the first one did. However, the important amino acids that are responsible for the docking of the ligand, were shifted by 8 amino acids towards the C-terminal, providing it to not usable for the simulations.

Another reason why the MD simulations did not yield the expected result, is that the authors of the paper used the protein simulation software GROMACS, in contrast to the software YASARA used within this project. Both programs use different set-ups of their force fields. While the provided information on the adjustable parameters was implemented where ever possible, it did not yield the same result. Also, take in account that YASARA provides already 10 force fields for different kinds of simulation approaches.

The main conclusion of the structural modeling of the hOR2AG1 olfactory receptor was that the results of the paper were not reproducible, at least not in YASARA.

The modeling of the hGPR109A receptor

The modeling of the hGPR109A receptor yielded more result, with a slightly different approach. Instead of 10 ns MD simulation, 1 ns MD simulations were executed because of time shortage. No initial MD simulation without ligand was executed, and ligands were docked directed in the receptor. Experimental data was obtained from research on the protein-ligand interaction between the receptor and niacin. The results described in the paper indicate that the amino acids Ser 178 and Arg 111 play a crucial role in hydrogen bonding, of which the latter plays the most important one.[REF] Both the model build by the YASARA macro and the amino acid-swapping method did not deliver a docking between the amino acids Ser 178 and Arg 111, however the model of the Zhang server did. [REF]

The docking of the ligand niacin in the protein yielded a bonding mainly between the amino acids Arg 111, Ser 178 and Arg 251 (FIGURE ???), of which Arg 111 is bonded the strongest. The amino acids that make up this binding niche differ partially from the ones described in the paper. This ligand-protein docking was the initial conformation of the "wildtype" receptor MD simulation. A Youtube-movie (choose 720p HD for the best visualization) shows this ligand-protein interaction.

Figure ???. Configuration of the hGPR109A binding niche and niacin. a) Detailed representation of the cavity after 1 ns MD simulation. The ligand niacin is colored orange, residues cyan and transmembrane helices gray. b) Experimental results from research on the interaction between niacin and the human niacin receptor 1. [REF] The chart represents the activation by cells containing mutant receptors normalized to cells containing the wildtype (WT) hGPR109A protein. c) An in silico model of the hGPR109A receptor (gray) with the ligand niacin (orange).

The M103V, L107F, S178I, F198L, R251A, C183A, C266A, R111A and C100A mutants of the hGPR109A receptor were investigated computationally in a 1 ns simulation of a thousand iterations. After analyzing the results, a same criteria as described in hOR2AG1 research paper on classifying the hydrogen bonds was applied; Robust bonds are present in 100-50 % of the simulating time, fluctuating in 49-25% and temporary in 24-1%.

Table ???. In silico hydrogen-bond contacts of niacin with hGPR109A variants. The percentages represent frequency of hydrogen-bond contact occurrence.

[a] Criteria for hydrogen bonding; temporary bonding to [QUALIFY] (>0%) and robust bonding to Arg 111 (>49%). [b] Mutated residue variant compared to its relevant residue. [c] Low expression of the receptor.

The simulation times were a crucial part of the structural modeling in YASARA. YASARA has a friendly users interface and has many integrated features. However, the MD simulations are optimized for use on a 8-core computer. This means that a simulation of 10 ns takes about 7-8 days to completely simulate on a 16-core computer, let alone an 8-core computer. Also, by taking in account the error probability for every simulation, this process exhausts a lot of time.

Thankfully, the SARA institute [URL] was willing to help us and assist us in setting up the environment to use their HPC Cloud Server remotely from our office at Delft University. This gave us the opportunity to execute our simulations on a configurable amount of cores. By testing what amount of cores was the fastest to use (but not necessarily the most efficient) short simulations on 4, 8, 12, 16, 18, 20 and 24 cores were performed. The outcome was in favor of the 16-core computer. This meant that a simulation of 10 ns would take 7-8 days.

Future follow-ups

Some additional ideas on structural modeling had to be given a lower priority. An interesting approach would be to engineer an olfactory receptor for one of the other three chemical compounds (Figure 1b-d) which are found in the breath of a tuberculosis patient. In the same manner, the reprogramming of the binding niche could allow different derivatives of a certain compound (methyl isonicotinate instead of methyl nicotinate) allow to bind, of which a kind of universality could be investigated to predict a configuration of the binding niche to any ligand.

Anoother follow-up would be to investigate in silico how large the conformational change of the receptor is, and see how this correlates to the hydrogen bonding and energy. This would require data on simulations done with and without a ligand. This could eventually lead to a value for the dissociation rate between the G-alpha protein and the receptor.

References

[1] Syhre M, Chambers ST (2008) The scent of Mycobacterium tuberculosis. Tuberculosis. 88:317–323

[2] Gelis L, Wolf S, Hatt H, Neuhaus EM, Gerwert K (2012) Prediction of a Ligand-Binding Niche within a Human Olfactory Receptor by Combining Site-Directed Mutagenesis with Dynamic Homology Modeling. Angew. Chem. Int. Ed. 51:1274-1278

[3] Tunaru S, Lättig J, Kero J, Krause G, Offermanns S (2005) Characterization of Determinants of Ligand Binding to the Nicotinic Acid Receptor GPR109A (HM74A/PUMA-G). Mol Pharmacol. 68:1271-1280

[4] Kurtz AJ, Lawless HT, Acree TE (2010) The Cross-Adaptation of Green and Citrus Odorants Chem. Percept. 3:149–155

J Zhang, Y Zhang. GPCR-ITASSER: A new composite algorithm for G protein-coupled receptor structure prediction and the application on human genome. 2011