Molecular Modelling of Odorant/Olfactory Receptor Complexes
Providing a rationale that associates a chemical structure of an odorant to its induced perception has been sought for a long time. To achieve this, a detailed atomic structure of both the odorant and the olfactory receptor must be known. State-of-the-art techniques to model the 3D structure of an olfactory receptor in complex with various odorants are presented here. These range from sequence alignment with known structures to molecular dynamics simulations in a realistic environment.
Key wordsMolecular modelling Homology 3D-structure Docking GPCR Olfactory receptor MM-GBSA
The connection between the structure of an odorant and its role in the perception of odors has been long sought. At the molecular level, the perception of smell is rooted in the activation of olfactory neurons, each of them housing an olfactory receptor (see (1) for a review). The large number of olfactory receptors (ORs) gives rise to the idea of olfaction being associated with a combinatorial signal (2) which is, for now, virtually impossible to mechanistically elucidate in terms of the odorant’s chemical structure. Prior to the discovery of olfactory receptors, Structure–Property relationship studies of odors projected promising results (3). Unfortunately, they suffered from a major limitation since the role of the ORs was not considered.
With the recent advances in both computing power and bioinformatics methodologies, molecular modelling has evolved as a force that allows us a glimpse into the nature of OR–odor interactions at a molecular level. Although one of the major quests of molecular modelling, i.e., the description of a protein structure on the basis of its amino acid sequence is within reach (4), ORs structural description represents a particularly challenging task, since these proteins belong to the family of G-protein-Coupled-Receptors (GPCRs)-membrane proteins, whose structures are notoriously difficult to structurally characterize.
Two distinct approaches, ab initio or homology modelling methods, can be used to overcome the lack of X-ray structures of ORs. Ab initio protocols use classical laws of physics to construct 3D structures on the basis of the amino acid sequence (see (5) for an example on GPCRs). Homology modeling methods aim at predicting the structure of a protein of interest from a set of experimentally known structures. The latter is the major methodology used to build theoretical 3D models. Here, we describe the materials and methods to build a full, atomistic 3D structure of an OR, both free or complexed with an odorant. Additionally, embedding the OR or its complex with an odorant within a solvated phospholipid bilayer and system relaxation by molecular dynamics simulations is presented. The present protocol has notably been used in ref. 6.
2.1 Sequence Comparison and Alignment
The protein sequences can be downloaded on servers like the Protein Information Resource (PIR: http://pir.georgetown.edu/) or the Human Olfactory Receptor Data Explorer (HORDE: http://genome.weizmann.ac.il/horde/) (seeChapter 2).
The alignment can be carried out by freeware such as Jalview (http://www.jalview.org/) or directly on servers like PIR.
2.2 3D Structure Building
Modeller (a homology or comparative modelling of protein three-dimensional structures freeware).
Any 3D-visualization software (VMD, Chimera, etc.).
2.3 Ligand Docking
Any docking software (AUTODOCK VINA, GOLD, etc.).
Files containing the 3D structure of the odorants (in pdb or mol2 file format).
2.4 Membrane Embedding
GROMACS molecular modelling software.
A file containing the 3D structure of the phospholipid membrane (DPPC or POPC for example can be found on P. Tieleman’s Web page: http://moose.bio.ucalgary.ca/index.php?page=Structures_and_Topologies).
InflateGro script to build the membrane environment around the OR (http://moose.bio.ucalgary.ca/files/inflategro).
Alternatively, the Maestro from Schödinger, inc. (http://schrodinger.com). The software has a membrane building protocol.
2.5 Molecular Dynamics
Any molecular dynamics software (AMBER, GROMACS, NAMD, etc.).
Choose a force field developed for both proteins and lipids.
3.1 Sequence Alignment
LHXPMYFFL in the beginning of helix 2.
MAYDRYVAICXPLXY in the end of helix 3.
SY in helix 5.
KAFSTCXSH in helix 6.
PMLNPFIYSLRN in helix 7.
- 1.Paste both the OR and experimentally known GPCR sequences in FASTA format in the PIR Web site (http://pir.georgetown.edu/pirwww/search/multialn.shtml). Several sequences are particularly useful for ORs studies:
The multiple alignment published by Man et al. (16) aligned five ORs with five others GPCRs (including the rhodopsin sequence). It helps to correctly align conserved regions.
The sequences of the other experimental GPCR structures (notably the human β1 and β2 adrenergic receptors).
- 2.Retrieve the alignment and open it in the Jalview software. Eventually, manually assess the alignment to ensure that the conserved regions discussed above are correctly aligned. Figure 1 shows an alignment performed to produce a structure of the human OR, OR1G1 (also named hOR17-209).
3.2 From Sequence to 3D Structure
On the Protein Data Bank Web site, obtain the 3D structure of: the bovine Rhodospin (PDB id: 1U19), the human β2 adrenergic receptor (PDB id: 2RH1), the turkey β1 adrenergic receptor (PDB id: 2VT4), and the human A2a adenosine receptor (PDB id: 2YDV). Modify each PDB file by removing water molecules, β-factor, etc., keeping only the receptor’s residues and the natural ligand.
Prepare the MODELLER input file. Specify the cysteine residues forming S-S bridges. Note that two sulfur bridges are highly probable in ORs based on the alignment with known class A GPCRs such as β2-adrenergic receptor. In our case these two bridges involve cysteine residues 97 and 179 and also 169 and 189. Be sure to consider that the ligand (retinal in 1U19, cyanopindolol in 2VT4, carazolol in 2RH1, or ZM241385 in 3EML) located within the reference protein is inserted in the OR during the building process. It will build the OR structure with a quite large internal cavity, in which the odorants will later be docked.
Generate a large number of putative structures (say 50–100) to allow maximum flexibility among these structures. The template structure considered for the building procedure can be either one of the solved X-ray structures of GPCR (rhodopsin, β1, β2, or adenosine receptors, CXCR4 chemokine receptor, dopamine D3 receptor, histamine receptor, lipid GPCR, muscarinic M2 receptor, nociceptin receptor, delta opioid receptor, kappa opioid receptor, mu opioid receptor), or one obtained by using all as templates for the OR target. Typically, currently, most OR models have been built with rhodopsin as a template. One of the major differences between the crystallized GPCRs lies in the conformation of the extracellular loop 2 (EL2). This EL2 was shown to be important for ligand recognition in class A GPCRs (26) and its structure has to be considered with care. The several families of structures will be analyzed in the next steps (see Note 2).
3.3 3D Model Analysis and Validation
View each model using the visualization software.
Eliminate the apparently badly folded structures (e.g., those with entwined loops).
Eliminate models where too many hydrophobic residues are in the extracellular loops.
Check the residues which are in a sphere of five Angstrom units around the ligand of the reference protein and compare with already studied ORs (6, 17, 18, 19, 20, 21, 22, 23, 24, 25). Eliminate models for which either none or too few correspondences are found with ORs built with experimental constraints (site-directed mutagenesis) (see Note 3).
- 6.One or two structures should be chosen for each family, as shown in Fig. 3.
Figure 3 highlights small deviations between the models built from either the β2-adrenergic receptor or the rhodopsin structures. These differences arise from various kinks or shifts in the secondary structure of some helices. Moreover, variations in the sequence of the extracellular domain among the GPCRs contribute to the diversity of secondary structures. In this case, however, the residues that constitute the binding cavity are similar, rationalizing the use of these models for ligand–receptor interaction analysis purposes.
3.4 Building the Complexes
At this step, the ligand of the template protein is always present in the binding site. Remove it manually to enable the docking of the models of odorant molecules.
Dock the ligand into the binding site, either with a docking protocol or manually if sufficient evidence exists of a preferred docking configuration (see Note 4).
Check the environment of the odorant by comparing the ligand/receptor interactions with those described in studies involving site-directed mutagenesis or other pure in silico studies. Select the docking conformations involving residues found to be important in site-directed mutagenesis experiments.
- 4.This stage can be considered the final step, resulting in a 3D-structure of an OR bound to an odorant. Much information is already present in this model, such as the binding affinity estimation, the nature of residues forming the main contacts with the odorants and the orientation of the odorant within the binding site, as shown in Fig. 4. We describe nonetheless, additional and more technical procedures, to refine these data. The relaxation of the structure is likely to slightly change conclusions drawn from this first model. Indeed, the docking protocol has proposed several positions of the ligand within the binding site. Generally, the scoring functions lead to very weak energy differences between the different poses. This suggests that at least two or three conformations may be considered for further refinement. A rescoring function, based on statistics accrued during Molecular Dynamics (MD) simulations should help to get a more accurate model.
3.5 Membrane Embedding
To relax the system in a realistic environment, the phospholipidic membrane, as well as the intra and extracellular medium should be modelled. These steps are rather technical since they require a good knowledge of the use of molecular simulation software.
Check the width of the initial membrane you have built or downloaded. Eventually duplicate it with the genconf command of GROMACS in directions specified by the option -nbox.
Open both the membrane and the OR files using visualization software. Check the position of the membrane with respect to the residues belonging to the OR (hydrophobic residues in the membrane, etc.). A tryptophan residue is for example a good indicator of the membrane position. Its pyrrole functional group can form hydrogen bonds with the polar heads of the lipids while its aromatic cycle interacts with the hydrophobic part of the lipids. If the membrane is badly positioned use the editconf command of Gromacs with the option -center specifying the coordinates of the center of the membrane. Use the option -rotate to rotate the protein around x, y or z and check that the OR principal axis of inertia is orthogonal to the water/lipid interface.
Paste the coordinates of the membrane PDB file in the PDB file of the rotated OR. At this stage, you have probably created a structure where many phospholipids have steric clashes with the receptor. Further refinement is warranted.
Generate the topology (top) and the coordinates (gro) files of the system OR/membrane. Use the pdb2gmx command of Gromacs with the -ignh command (ignore hydrogen). Choose the force field (for example “Gromos 54a7”). During the process, the cysteine bridges are recognized and created. Nonetheless, check that the appropriate cysteine residues are taken into account (if this is not the case, it indicates that the alignment prior to modelling is incorrect).
Inflate the lipids to eliminate steric clashes. Use the inflategro script. This script requires a scaling factor and a cutoff, to scale the phospholipids coordinates and deletes those that are too close to the receptor.
- 6.Slowly deflate the lipids.
Create a minimization input file restraining the coordinates of the protein and the phosphor atom of the lipids in the three dimensions and in the z-dimension respectively.
Begin minimizing (with Gromacs command grompp and mdrun).
Use the inflategro script.
Solvate your system with water molecules and add ions to neutralize the system using a default protocol.
Maestro (Schrödinger inc.) has a membrane building protocol that allows inserting your OR in a box containing the membrane, the water phase and the ions. It is very user-friendly but requires longer equilibration time by means of molecular dynamics.
3.6 Molecular Dynamic Simulations
Minimize the energy of the system in two steps. First, minimize the energy of the solvent by freezing the OR and the lipids. Second, minimize the energy of the whole system.
Heat the system gradually up to the desired temperature (generally 310 K), keeping the volume of the simulation box fixed.
Equilibrate the system at one atm. with a semi-isotropic pressure scaling.
- 4.The molecular dynamics simulation production phase is continued from the last equilibration step, to collect sampling for further analysis. For most odorants, the cavity created during the homology modelling is too large and tightens during the equilibration steps. The molecular dynamics production phase is then performed to evaluate the residues in contact with the odorant from a statistical point of view. These residues can be different from those found in the starting structure obtained with the docking procedure. Figure 6 illustrates the structure of hOR1G1 just after the docking step and after a 20 ns molecular dynamics simulation. Additionally, binding free energy estimation protocols can be considered to compare the computed affinity with experimental data, such as calcium imaging or dissociation constant measurements.
3.7 Docking Rescoring
Since the interaction between an odorant and an olfactory receptor are mainly hydrophobic, one can observe a reorganization of the ligand in the binding cavity during MD simulations.
To decipher the binding mode of a ligand, multiple molecular dynamics protocols can be used. According to available computational resources, several poses obtained from the docking protocol are chosen as starting points for MD simulations. Each trajectory is then analyzed using for example a MM-GBSA protocol. It has the advantage of providing a decomposition of the free energy of binding on a per-residue basis.
This method enhances the sampling of the binding cavity by the ligand, and thus allows finding the main amino acids involved in the interaction. It appears to be particularly efficient in the case of nondirectional interaction between the ligand and the cavity.
For a detailed protocol of using MM-GBSA with AMBER, the reader should refer to http://ambermd.org/tutorials/advanced/tutorial3/.
Even if no OR structure is known, it is important to put several other ORs in the alignment, to be sure that the crucial conserved sequences or residues are correctly accounted for during the alignment process.
At this stage, one can eventually perform a geometry optimization of the residues side-chains (and only the side-chains) with molecular modelling software. Indeed, it is difficult to estimate side-chain orientations.
If many models fulfil only a part of the criteria (discussed in Subheadings 3.3, steps 3 and 4), this can point to a bad alignment of the sequences. This may necessitate a modification of the alignment (Subheading 3.1).
You can use a standard docking protocol for this step (not described here). Generally, the olfactory receptor is considered rigid and the only the odorants can undergo conformational changes. One can consider that the binding site will be identical to those found in rhodopsin and adrenergic receptors.
Membrane embedding can be done by several methods. Reference 28 summarizes them.
It is very useful to use a script at this stage. Indeed, the deflation of the system takes lot of simulation steps (depending on your scaling factor during the inflation).
The CINES is acknowledged for providing computer time for the project cmi1024.
- 10.Cherezov V, Rosenbaum Daniel M, Hanson Michael A, Rasmussen Soren GF, Thian Foon S, Kobilka Tong S, Choi H-J, Kuhn P, Weis William I, Kobilka Brian K, Stevens Raymond C (2007) High-resolution crystal structure of an engineered human beta2-adrenergic G protein-coupled receptor. Science 318(5854):1258–1265PubMedCrossRefGoogle Scholar