Machine Learning-Based Modeling of Olfactory Receptors in Their Inactive State: Human OR51E2 as a Case Study

Abstract

Atomistic-level investigation of olfactory receptors (ORs) is a challenging task due to the experimental/computational difficulties in the structural determination/prediction for members of this family of G-protein coupled receptors. Here, we have developed a protocol that performs a series of molecular dynamics simulations from a set of structures predicted de novo by recent machine learning algorithms and apply it to a well-studied receptor, the human OR51E2. Our study demonstrates the need for simulations to refine and validate such models. Furthermore, we demonstrate the need for the sodium ion at a binding site near D 2.50 and E 3.39 to stabilize the inactive state of the receptor. Considering the conservation of these two acidic residues across human ORs, we surmise this requirement also applies to the other ∼400 members of this family. Given the almost concurrent publication of a CryoEM structure of the same receptor in the active state, we propose this protocol as an in silico complement to the growing field of ORs structure determination.

Calcium-Driven In Silico Inactivation of a Human Olfactory Receptor

“…We employed the CHARMM-GUI web server , for receptor setup and its embedding into a 3:1 POPC/cholesterol bilayer (see Methods section and Supporting Information for further details). Following system preparation, we adopted a multistep equilibration protocol, applying restraints to preserve the initial fold while relaxing the system (refer to Methods), as done in our previous work . This was followed by five independent unrestrained MD simulations, each lasting 5 μs.…”

“…Despite their potential as drug targets, the study of ORs has been hindered by the lack of structural data: the first experimental structure was published in March 2023, consisting of human OR51E2 in its active state, bound to an agonist (propionate) and a mini-G protein. In light of this limitation, the modeling community has approached the OR family with different techniques, from homology modeling − to de novo structure prediction. , Such computational models have provided useful insights into the mechanism of OR-ligand recognition at the atomistic level as well as pinpointed crucial residues for OR function, in combination with experimental mutagenesis data.…”

“…Our primary objective is to simulate the inactivation of the receptor in silico over a microsecond timescale, drawing upon the pioneering work of Dror et al on the β 2 -adrenergic receptor . Moreover, OR51E2, along with 90% of the OR family members, bears two acidic residues in the ion binding site (2.50 and 3.39 using the Ballesteros–Weinstein class A GPCR generic numbering), indicating the potential involvement of these charged residues in the ion binding mechanism and conformational dynamics of ORs. This hypothesis is supported by our previous observations of sodium-bound OR51E2 in the inactive state .…”

“…Moreover, OR51E2, along with 90% of the OR family members, bears two acidic residues in the ion binding site (2.50 and 3.39 using the Ballesteros–Weinstein class A GPCR generic numbering), indicating the potential involvement of these charged residues in the ion binding mechanism and conformational dynamics of ORs. This hypothesis is supported by our previous observations of sodium-bound OR51E2 in the inactive state . Accordingly, we aim at elucidating the role of the ion binding site for OR conformational plasticity, also building on previous work on sodium binding to other non-olfactory class A GPCRs. , …”

AlphaFold2 in Molecular Discovery

“…Odorant receptors are G protein-coupled receptors (GPCRs) with poor structural characterization due to the low number of experimental structures available and low sequence identity to other GPCRs with experimental structures determined, making them unsuitable for homology modeling . Two publications , applied AF2 to build models of odorant receptors, paving the way to the investigation of recognition of odorant molecules at the molecular level. Enriching a training set of experimental structures with AF2 structures improved the performance of a machine learning model to predict protein function .…”