Non-canonical G_q activation by orexin receptor type 2 and lemborexant observed in microsecond molecular dynamics simulations

Abstract

Orexins are hypothalamic neuropeptides primarily involved in regulating the sleep/wakefulness cycle and circadian rhythm. They bind to the orexin receptor type 1 (OX₁) and type 2 (OX₂), well-known drug targets in the treatment of sleep disorders, that have recently been shown to play a significant role in different cancers. Lemborexant is one of a few orexin receptor antagonists that have been approved for the treatment of insomnia. Despite being classified as an antagonist, lemborexant may display agonist-like behavior in the non-canonical signaling pathway of the orexin receptors, as confirmed recently in cancer cell models. Here, we generated a model of OX₂ in complex with the full-length G_q protein and used it in the molecular dynamics (MD) study. We compared the impact of lemborexant and the OX₂-selective, potent agonist compound 1 on OX₂ activation and subsequent guanosine diphosphate (GDP) to guanosine triphosphate (GTP) exchange in the Gα_q subunit. These 2 µs MD simulations showed that both ligands evoke similar, activation-like conformational changes in OX₂ and explained the observed lemborexant-mediated apoptosis of cancer cells. In addition, MD simulations of the active-state OX₂-G_q complexes allowed us to uncover a sequence of micro- and macroscale events during the activation of G_q and to detect important micro- and macroswitches in the Gα subunit.

Introduction

Orexins¹ were discovered in 1998 by two independent research groups, de Lecea et al.² and Sakurai et al.³ These hypothalamic neuropeptides exist in two forms: orexin-A (OxA, 33 amino acids) and orexin-B (OxB, 28 amino acids)⁴, and are involved in regulating the central nervous system⁵. Both are formed from the same precursor via proteolytic processing³ but share only 46% of the same sequence⁴. Best known for regulating the sleep/wakefulness cycle and circadian rhythm, orexin deficiencies can cause sleep disorders⁶. They also influence blood pressure, locomotion, hormone secretion⁷, drug addiction, reward seeking, energy homeostasis⁵, and peripheral organ function⁸.

Orexins bind to two G protein-coupled receptors (GPCRs), OX₁ and OX₂⁹, which recognize both OxA and OxB with poor selectivity¹⁰. However, other known ligands demonstrate varied subtype selectivity, e.g., seltorexant (OX₂)¹¹, ACT-335827 (OX₁)¹², with some binding to both receptors¹³. Agonist binding activates a canonical signal cascade involving the G_q protein, triggering intracellular Ca²⁺ release. While orexin receptors additionally couple with Gi/o and Gs¹⁴, our recent data indicated that OX₁ is coupled exclusively to G_q in recombinant human embryonic kidney-OX₁ (HEK-OX₁) cells¹⁵. A second non-canonical signaling pathway of orexin receptors leads to apoptosis. Upon activation by orexins, two immunoreceptor tyrosine-based inhibitory motifs (ITIMs) of OX₁ or OX₂ are phosphorylated, leading to the recruitment of Src homology region 2 domain-containing phosphatase-2 SHP2. This triggers a signaling cascade involving the activation of the p38 mitogen-activated protein (MAP) kinase, the translocation of the apoptosis regulator Bax in mitochondria, cytochrome C release, and the activation of caspases-3 and -7, ultimately causing cell death⁵.

Orexin receptors are promising drug targets for sleep disorders like narcolepsy or insomnia¹⁶, with small-molecule antagonists such as suvorexant, lemborexant, or daridorexant already on the market. While orexins as agonists promote wakefulness, antagonists treat insomnia^17,18. Additionally, the orexin system plays a role in different cancers, e.g., OX₁ in glioblastoma, colon, pancreas, and prostate cancers, and OX₂ in rat pancreas cancer, endometrial carcinomas, or cortical adenomas^19–21. Both orexins can induce apoptosis, thereby displaying antitumoral properties^20,22; similar results have been reported for almorexant²³. Recent studies have shown that suvorexant, lemborexant, and daridorexant induce apoptosis in human colon cancer cells expressing OX₁, noticeably inhibiting OxA-induced intracellular calcium release¹⁵. They also prompted apoptosis in recombinant HEK cells expressing OX₂¹⁵. Additionally, lemborexant induced a partial dissociation of G_q, leading to its quasi-agonistic properties on apoptosis induction in cancer cells¹⁵.

As the structure of the active-state OX₁ has not yet been solved, we used OX₂ in complex with G_q and either a potent, OX₂-selective, small molecule agonist (compound 1)²⁴ or an antagonist (lemborexant)²⁵ to study the activation of the orexin system by graphics processing unit (GPU)-accelerated all-atom MD simulations. Notably, the activation of OX₂ by orexins induced pro-apoptotic effects in cancer cells similar to the one induced by OX₁^15,21. MD is commonly used to observe the evolution of molecular systems in time²⁶, and can be used to study both micro- and macroscale cellular processes²⁷, also in the context of GPCRs^28–32. In many such cases MD is complemented with other computational methods, especially those involving the use of artificial intelligence³³ that have found applications in cancer diagnosis, treatment, and drug discovery, thereby accelerating the development of new treatment strategies³⁴.

The orexin system has already been included in several in-silico studies. In 2016, Yin et al. performed molecular dynamics simulations to elucidate the basis for receptor subtype selectivity in regard to small-molecule agonists³⁵. They determined which residues affected the binding affinity of the antagonist SB-674042 and confirmed it with radioligand binding experiments³⁵. Computational approaches such as quantitative structure–activity relationship (QSAR) models; molecular docking; and absorption, distribution, metabolism, and excretion (ADME) predictions have also been used in the design of novel OX₁ antagonists²⁸, while molecular mechanics have been used to compute the binding free energies of dual orexin receptor antagonists such as lemborexant and suvorexant using previously obtained crystal structures³⁶. Additionally, homology modeling and MD simulations were used to predict the binding mode of OxA to OX₂ before the corresponding Protein Data Bank (PDB) structure was released³⁷.

In our MD study, accompanied by the experimental study of Gratio et al.¹⁵, we observed similarities between the classic OX₂ agonist compound 1 and lemborexant-including systems, suggesting the agonist-like behaviour of lemborexant in the recombinant HEK293T cell line expressing OX₂¹⁵ despite its well-known role as a dual OX₁/OX₂ antagonist³⁶. In principle, an antagonist tightly binds to the orthosteric binding site of the GPCR receptor, preventing an endogenous agonist binding and the subsequent receptor activation. Since the lemborexant orthosteric binding site in OX₁/OX₂ is known and well-described³⁶, and no other binding sites, e.g., allosteric, were detected¹⁵, its surprising pro-apoptotic effect induced by the receptor activation could not be explained other than as evoking activation as an agonist. Additionally, the preceding experimental study¹⁵ showed that the orexin receptor was activated by lemborexant in the absence of orexin, which excludes the possibility of an allosteric mode of action of lemborexant in tested cell lines. This is further supported by the fact that the pre-treatment of HEK cells expressing OX₁ with antagonists such as almorexant, lemborexant, suvorexant, and daridorexant completely abolished the induction of calcium release by OxA¹⁵. The ligand-dependent preferential activation of selected pathways in different tissues (biased agonism or functional selectivity) was previously observed for many GPCR receptors, e.g., for opioid receptors and β-adrenergic receptors³⁸.

Here, we explained the agonist-like behaviour of lemborexant on the molecular level using MD. The simulation systems included OX₂ bound to either lemborexant or the OX₂-selective, classic agonist compound 1, and the full-length Gα_q, which substituted the mini-Gsqi construct present in the template (PDB: 7L1V)^15,24,39 and used in our previous study. Performed MD simulations additionally provided insights into the molecular basis of the OX₂ activation, allowing us to propose a sequence of conformational changes that the receptor and Gα_q undergo during activation, e.g., the opening of the Gα_q helical domain and subsequent release of GDP. Due to a high sequence identity (67% according to Clustal2.1) between two orexin receptors, our results regarding receptor activation could be extrapolated to OX₁, for which even more detailed experimental evidence has recently been provided¹⁵.

Methods

Preparation of simulation systems

Models of the OX₂–G_q systems were prepared using Modeller v. 10.4⁴⁰, with the 7L1V²⁴ PDB entry used as a template for the receptor and G protein complex, and 7SQ2⁴¹ used as the template for the Gα subunit. The positions of conserved residues in the orexin receptor with respect to other GPCRs were confirmed (see Supplementary Fig. S1). Two separate sets of models were generated—one containing compound 1 in the orthosteric binding site, and another containing lemborexant. Compound 1 was present in the active-state 7L1V structure of OX₂. Lemborexant was positioned in the orthosteric binding site by superposing the inactive-state 7XRR PDB structure³⁶ onto the apo 7L1V structure and transferring the lemborexant molecule from the inactive-state structure to the active-state structure in order to retain its binding mode observed in 7XRR. Both simulation systems contained GDP bound to the Gα subunit. The crystal water molecules in the 7SQ2 structure were retained during the model building by Modeller but removed before building the membrane with CHARMM-GUI. The target sequence consisted of the UNIPROT⁴² entries for OX₂ (O43614), Gα_q (P50148), Gβ₁ (P62873), and Gγ₂ (P59768). The UNIPROT entry for OX₂ was compared in terms of sequence identity (67%) with OX₁ (O43613) using ClustalOmega (https://www.ebi.ac.uk/jdispatcher/msa/clustalo). In order to make sure the N- and C-termini of the OX₂ models were comparable to the 7L1V structure, all of the residues before Y^1.31 and after F^7.68 were removed. Additionally, residues G^5.72 to V^6.20 of intracellular loop 3 (ICL3) were removed (33 residues) to simplify the model. Although ICL3 is involved in signaling through the G protein⁴³, this region is highly flexible and predicted to be intrinsically disordered^44,45, and is absent in most GPCR structures in PDB⁴³. It is assumed to affect the receptor specificity for the G protein type⁴³, and then cAMP accumulation after GDP release⁴⁶, neither of which were considered in this study.

For each set, 500 preliminary models were generated. Ten iterations of loop refinements were then performed for each model, resulting in a total of 5000 models per set. Their quality was evaluated by comparing DOPE scores⁴⁷, and the 30 lowest-energy models were further assessed in PyMOL version 2.5⁴⁸. Three lemborexant-OX₂-G_q and three compound 1-OX₂-G_q models were selected for the generation of MD simulation systems, resulting in a total number of six replicas that were simulated in parallel. The protein preparation wizard in Maestro (Schrödinger Suite release 2022-3)⁴⁹ was used to preprocess the models, i.e., add hydrogens and optimize bonds.

Molecular dynamics simulations

Inputs for the molecular dynamics simulations were prepared using CHARMM-GUI’s⁵⁰ Membrane Builder⁵¹ (specifically, the Bilayer Builder). Each model was exported from Maestro in the PDB file format and uploaded into CHARMM-GUI. Information about the locations of disulphide bonds in the proteins was obtained from the templates and all crystal water molecules were removed. The ligands were exported from Maestro in sdf format and uploaded into CHARMM-GUI, which then automatically generated CHARMM topology and parameter files using the CHARMM General Force Field (CGenFF). Using the replacement method, the complexes were placed in a lipid bilayer consisting of a 3:1 ratio of POPC (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine) to cholesterol. Such a membrane composition has been used in other MD studies performed for GPCRs⁵². The periodic rectangular water box (TIP3P) was fitted to the complex, and each system was neutralized by adding Na⁺ and Cl^- ions at a concentration of 0.15 M. The number of atoms in each system was around 240,000–250,000, and the initial size of the simulation box was ca. 122 Å × 122 Å × 179 Å.

The MD simulations were performed using the GPU-accelerated version of NAMD (NAMD 3.0 Alpha)⁵³ with the Charmm36m force field. The equilibration step included 10,000 steps of the steepest descent minimization, then 25,000 steps of the conjugated gradients minimization. The equilibration simulation was performed in NVT (constant particle number, volume, and temperature) using Langevin dynamics (303.15 K). The time integration step in the equilibration and production runs was set to 2 fs. The production run in NPT (constant particle number, pressure, and temperature) was performed using the Langevin piston Nose–Hoover method (1 bar, 303.15 K) and lasted for 2 µs. Every tenth frame of the computed trajectories was combined using CatDCD⁵⁴, then wrapped using PBCTools⁵⁵ and analyzed in VMD v.1.9.3⁵⁶. Further analyses were performed using PyMOL and Maestro.

Results

Similarities in binding modes between the classic OX₂ agonist compound 1 and lemborexant

MD simulations of both the classic agonist compound 1²⁴ and lemborexant-including complexes of the active-state OX₂ with the full-length Gα_q were analyzed in terms of variance in the position of ligands and complex components (Fig. 1, Supplementary Fig. S2). As a dual OX₁/OX₂ antagonist^57,58, lemborexant was expected to display significant RMSD fluctuations in the binding site of the active-state OX₂ structure, especially with additional intracellular stabilization by a G protein. However, its position remained stable, with insignificant RMSD fluctuations comparable to compound 1. This suggests there could be conditions under which lemborexant may induce an active-like conformation of OX₂ and behave similarly to the classic agonist. The maintained active state of OX₂ during simulations was validated by analyzing DRW motif residues (corresponding to the DRY ‘ionic lock’) (Supplementary Figures S3, S4). In class A GPCRs, R^3.50 interacts with T^2.39 (inactive state, e.g., 7XRR³³) or Y^5.58 (active state, e.g., 7L1U, 7L1V²⁴)⁵⁹. The conformations of the DRW motif residues in the compound 1-including complexes remained ‘on’ in replicas 1 and 3 but shifted into an intermediate-like position in replica 2 (Supplementary Fig. S3). In the lemborexant-including complexes, active-state interactions were also formed, with fewer inactive-state interactions than in compound 1-including replicas (Supplementary Fig. S4). Additional microswitches were analyzed: PVF, CYLP, and NPIIY (corresponding to the PIF, CWxP, and NPxxY motifs, respectively; Supplementary Figure S5). In the majority of systems, V^3.40 in the PVF motif changed from an active-like conformation to an intermediate conformation resembling neither the active nor inactive state. The Y^6.48 toggle switch of the CYLP motif changed conformations in the final frame of all of the lemborexant replicas but only one compound 1 replica. Although the inactive-state 7XRR and active-state 7L1V structures include the same Y^6.48 conformation, the changes of this microswitch are reported to be temporary⁶⁰. The Y^7.53 toggle switch of the NPIIY motif remained in an active state in all but one system, where it adopted an intermediate conformation. Additionally, transmembrane helix 6—recognized as a key macroswitch involved in GPCR activation—did not undergo global rearrangements in any of the simulation replicas. This was assessed by measuring the distance between the Cɑ atoms of R^3.50 and A^6.34 in the models, 7L1V (11.8 Å), and 7XRR (8.3 Å)⁶¹. In all models, these distances remained greater than 11.4 Å, thus comparable to 7L1V. These results further confirmed that OX₂ remained active throughout the simulations.

Fig. 1.

Similarities in binding modes between the classic OX₂ agonist compound 1 and lemborexant. A comparison of the binding modes of compound 1 (left, replica 1) and lemborexant (right, replica 1). The first frame of the simulation is shown in gray with a yellow ligand, and the final frame in green with a red ligand. The final frame is meant to represent the ligand conformation at the end of the simulations and includes the key receptor residues described in Supplementary Figures S4 and S5.

A detailed analysis of compound 1 and lemborexant binding modes (Fig. 1) aligns with previous studies^24,36,57,58. In the 7L1V structure, the phenyl rings of compound 1 were surrounded by a lipophilic patch consisting of P^3.29, M^4.64, T^2.61, and V^2.6424. Lemborexant was similarly surrounded by P^3.29, T^2.61, and V^2.64, though Asada et al.³⁶ proposed interactions with T^2.61 to be rather irrelevant for lemborexant binding to the inactive-state receptor, suggesting that it demonstrated a slightly different binding mode for the active-state. H^7.39 interacted with compound 1 (Supplementary Figs. S6, S7), as indicated by Hong et al.²⁴ and also stabilized lemborexant in 7XRR³³ but through π–π stacking instead of a hydrogen bond. Both ligands closely interacted with F^5.42 through π–π stacking and formed hydrogen bonds with Q^3.32. However, extracellular loop 2 (ECL2) closure, observed previously for OX₂–mini-G_sqi complexes¹⁵, caused both ligands to move further inside the receptor. Consequently, the Q^3.32 side chain also moved towards the inside of the receptor (see Fig. 1) in contrast to what can be observed in 7L1V²⁴. Hong et al. showed that Q^3.32 remained stably oriented outward during a 1 μs MD simulation with OxB²⁴. In our 2 μs MD simulations with the small-molecule agonist compound 1, we observed the closing of ECL2 after 1 μs, shifting Q^3.32 towards the inside of the receptor after 2 μs. Perhaps the Q^3.32 conformation was influenced by ligand bulkiness (small-molecule vs. peptide) and correlated with the closing of ECL2, rather than strictly by ligand type (agonist vs. antagonist) as suggested by Hong et al.²⁴ Since ECL2 remained open for peptide agonists, Q^3.32 could be oriented outwards²⁴.

Activation-like changes of microswitches in Gα_q

Ham et al. described residue F341^G.H5.08 (labeled there as F336) in helix H5 of Gα to be a microswitch affecting the GDP binding energy via switch I and the P loop (Supplementary Fig. S8), as shown in mutagenesis studies⁶². The F341^G.H5.08A mutation or a change of F341^G.H5.08 to the ‘on’ position caused fast GDP/GTP exchange rates⁶². In our homology models, this microswitch was initially in the ‘off’ position, as expected for an inactive G protein (Fig. 2A)⁶². During the simulations, its change to the ‘on’ position required a slight counterclockwise rotation. This was observed mostly for the lemborexant-including replicas (1 and 2) representing the furthest stage of Gα activation. In previous 2 µs simulations involving mini-G_sqi¹⁵, F341^G.H5.08 was also in the ‘on’ position at the end of the simulation. This suggested its transition may not be directly related to the opening of Gα, as these systems lacked the alpha helical domain (AHD) of Gα. Instead, this change may have been induced by receptor activation alone.

Fig. 2.

Changes of microswitches D277^G.HG.02 and F341^G.H5.08 and macroswitches I–III in OX₂ in comparison to the rhodopsin-mediated activation of Gα_i (magenta). (A) The D277^G.HG.02 and F341^G.H5.08 microswitches described by Ham et al.⁶² in the ‘off’ position in the first frame of the compound 1 replica 2 simulation (top, left) and 1GP2 PDB structure (bottom, left). The D277^G.HG.02 and F341^G.H5.08 microswitches described by Ham et al.⁶² in the ‘on’ position in the last frame of the compound 1 replica 1 simulation (top, center), the last frame of the lemborexant replica 1 simulation (top, right), the 6CMO PDB structure (bottom, center), and the 6OS9 PDB structure (bottom, right). (B) A comparison between the positions of macroswitches I–III, the P loop, and HC and HD between the active state 6CMO PDB structure⁷⁰ (magenta), the first frame of compound 1 replica 1 (gray, yellow switches), and the final 2 μs frame of the compound 1 replica 1 (green, red switches) (Left). A comparison between the positions of macroswitches I–III, the P loop, and HC and HD in the active state 6CMO PDB structure⁷⁰ (magenta), the first frame of the lemborexant replica 1 (gray, yellow switches), and in the final 2 μs frame of the lemborexant replica 1 (green, red switches) (Right).

Ham et al.⁶² also described the D277^G.HG.02 microswitch (labeled there as D272) as important for GDP/GTP exchange. In our case, most replicas demonstrated a breaking of the GDP–D277^G.HG.02 interactions regardless of ligand type (Fig. 2, Supplementary Figs. S9, S10). D277^G.HG.02 stabilized GDP binding by interacting with the guanine ring⁶². The breaking of this interaction was required for GDP dissociation, which changed the D277^G.HG.02 microswitch into the ‘on’ position—seen as the χ1 dihedral angle (N–C_α–C_β–C_γ) change from ca. 60° to − 60° (Supplementary Figs. S7, S8). We observed the movement of the phosphoryl group, not the guanine ring, which imposed only slight changes on D277^G.HG.02. Notably, GDP dissociated from Gα via its phosphate side⁶³. D277^G.HG.02 was indeed in the ‘on’ position at the end of the lemborexant-including simulations. In the compound 1-including replicas, D277^G.HG.02 fluctuated, changing to the ‘on’ position only at around 1.70 μs for replica 3, while in the other two replicas, it switched to the ‘off’ position at around 1.00 μs.

Activation-like changes of macroswitches in Gα_q inducing the GDP/GTP exchange

The microsecond timescale used for MD simulations allow to observe first stages of the GPCR signal transduction involving both changes in the ligand–receptor complex and the interacting G protein or arrestin³¹. This required the full length Gα_q to be present. To our knowledge, none of the active-state structures of GPCRs in PDB include the full-length Gα_q, though the full-length Gα_i is present in e.g., the rhodopsin PDB structure 6CMO (Fig. 2). Only the P loop and switch II are present in their entirety in mini-G_sqi, as observed when compared to the Gα topology described by Calebiro et al.⁶⁴ (Supplementary Fig. S8). For this reason, we constructed models with the full-length Gα_q to obtain a fully functional OX₂-G protein complex.

Though switch II remained rather unchanged in the previous mini-G_sqi simulations¹⁵, in most of the full-length Gα_q simulations changes in that region were more visible, e.g., a subsequent movement of helix H2. This altered Gα–Gβγ interactions, as Gα macroswitches I and II, and the Gα N-terminal helix form a hydrophobic cleft fitting Gβγ^64–66. These differences between results for the full-length Gα_q and mini-G_sqi simulations suggest that switch II movement might be connected specifically to the presence of the AHD in the former.

In our MD simulations of OX₂ with a full-length Gα_q, a noticeable difference was observed for compound 1 vs. lemborexant. At the end of the simulations, the most open Gα conformations were observed for the lemborexant-including simulations (Supplementary Figs. S11, S12). These opening movements of the Gα subunit, required for the GDP/GTP exchange and confirmed experimentally for other GPCR receptors⁶⁷, were hardly observed for compound 1 (Fig. 3). This may suggest that Gα opens faster in the lemborexant-including systems.

Fig. 3.

Changes in Gα macroswitches inducing the opening of the GDP/GTP exchange tunnel upon compound 1 binding to OX₂. The opening of a tunnel was observed near the phosphate side of GDP. From left to right: interactions between nearby residues, a surface view of the Gα subunit and GDP, hydrogen bonds determined by VMD to form between R183^G.hfs2.20 (switch I) and E49^G.s1h1.04 (P loop), and R183^G.hfs2.20 and E241^G.s4h3.10 (switch III) at a maximum 4 Å distance at an angle cutoff of 20º.

The GDP position changed during the simulations regardless of the OX₂ ligand (Supplementary Fig. S13). Notably, the terminal phosphoryl group of GDP lost its interaction with R183^G.hfs2.02 (GproteinDb⁶⁸ numbering scheme) in switch I on behalf of the second phosphoryl group. The formation of hydrogen bonds between the oxygen atoms in the second phosphoryl group and R183^G.hfs2.02 was analyzed (Supplementary Fig. S14). In the compound 1-including complexes, hydrogen bonds between R183^G.hfs2.02 and GDP were more retained than in the case of lemborexant. However, the average RMSD changes for the switch I (ca. 4 Å) were comparable for both systems (Supplementary Fig. S15).

The beginning of the opening of the ‘GDP/GTP exchange tunnel’ (Supplementary Files S3, S4) near the phosphate side of GDP was observed in both simulation systems, and the same way of GDP dissociation was described by Louet et al.⁶³ This was most visible for compound 1 replica 1 and lemborexant replica 1 (Figs. 3, 4). This opening of the exchange tunnel, along with the opening of Gα, suggested that these replicas represent the furthest stage of G protein activation regarding GDP release. In the remaining replicas, this tunnel was blocked by residues from the P loop and switch I. Namely, the formation of a salt bridge between R183^G.hfs2.02 (switch I) and E49^G.s1h1.04 (P loop) blocked the tunnel and prevented GDP from leaving the binding site (lemborexant replica 3, compound 1 replicas 2 and 3; Figs. 3, 4). The disruption of this salt bridge was necessary for Gα to open⁶⁹. In our simulations, during Gα activation, R183^G.hfs2.02 in switch I moved to the outer side to interact with E241^G.s4h3.10 in switch III instead (lemborexant replica 2). Presumably, switch III was responsible for stabilizing R183^G.hfs2.02 at this stage, preventing it from interfering with GDP leaving the binding site. Interestingly, at the beginning of every simulation, E241^G.s4h3.10 in switch III interacted with two residues of switch II: R210^G.H2.01 and Q209^G.s3h2.03 (to a lesser extent). Thus, the release of E241^G.s4h3.10 from interactions with switch II might also be required⁶⁹.

Fig. 4.

Changes in Gα macroswitches inducing the opening of the GDP/GTP exchange tunnel upon lemborexant binding to OX₂. The opening of a tunnel was observed near the phosphate side of GDP. From left to right: interactions in nearby residues, a surface view of the Gα subunit and GDP, hydrogen bonds determined by VMD to form between R183^G.hfs2.20 (switch I) and E49^G.s1h1.04 (P loop), and R183^G.hfs2.20 and E241^G.s4h3.10 (switch III) at a maximum 4 Å distance at an angle cutoff of 20º.

To verify if the described above direction of the observed changes in Gα is activation-like, we compared our results with the open conformation of Gα bound to rhodopsin (6CMO)⁷⁰ (Fig. 2) in terms of Gα local and global conformational changes. A further comparison of 6CMO with 1GP2⁶⁶ (the inactive-state G protein with bound GDP) showed a similar opening of D277^G.HG.02 and F341^G.H5.08 as observed in our simulations (Fig. 2A). Furthermore, there was a relocation of H63^G.H1.12 near M59^G.H1.08 (mentioned by Ham et al.⁶²) from helix H1 to an adjacent loop resulting from the change in F341^G.H5.08. The extension of the adjacent loop was followed by the opening of Gα (observed in full in 6CMO). This movement of H63^G.H1.12 was not seen in our simulations as it would require much longer simulation timescale to move AHD for such long distance observed in 6CMO (Fig. 2B). However, the beginning of this H63^G.H1.12 translocation was observed in most simulations (Fig. 2A).

The release of GDP is certainly associated with the opening of Gα (6CMO vs. 1GP2). It requires the translation of the AHD to the right and bottom, in accordance with Fig. 2B, which can be seen in the movement of helices HA and HD of 6CMO. Before Gα opening, residues in switches I–III (especially switch I) had to start moving in the same direction as in 6CMO. Both simulation systems showed macroswitch changes resembling those in 6CMO (Fig. 2B), confirming the beginning of the G protein activation.

The average RMSD values of switch III were higher for the lemborexant-including simulations than for the compound 1-including simulations (Supplementary Fig. S16). These larger conformational fluctuations correlated with the downward movement of HD and HA, causing the whole AHD to move downward, as seen in 6CMO. This movement appeared crucial for the opening of Gα (Fig. 2B, Supplementary Fig. S11).

Global conformational changes in G_q

Noticeable fluctuations of the AHD in both simulation systems suggest the first stage of the opening of Gα (Fig. 2). Helix HD of Gα displayed fluctuations to the left or right (Fig. 2B) with a simultaneous HA movement. In some replicas, HD moved right preceding Gα opening, like in 6CMO. HD movement to the left seemed to represent fluctuations of the relative position of the RD and AHD domains that did not lead to the GDP dissociation. After 1 μs, two lemborexant-including replicas showed HD moving right, but this was observed only in one compound 1-including replica. By 2 μs, only one lemborexant-including replica but two compound 1-including replicas displayed this movement. This observation indicated that fluctuations of the AHD, including HD or HA movements, likely precede the opening of Gα as this seems to be a highly dynamic process involving the simultaneous exchange of other molecules in its binding site⁷⁰. Nevertheless, the simulation replicas that demonstrated rightward HD and HA fluctuations like in 6CMO also demonstrated higher fluctuations of GDP RMSD (Supplementary Figure S13), while those with leftward movement had low fluctuations of GDP RMSD. This confirmed that the global rearrangements of HD and HA observed in our simulations were indeed associated with the GDP dissociation.

As mentioned above, the lemborexant systems seemed to demonstrate a faster opening of the Gα subunit, associated with the GDP dissociation, than the compound 1 systems (Supplementary Fig. 11). The way this could occur, i.e., via a hinge involving helix HF, seemed to be consistent with previously described results⁷¹. In the lemborexant-including simulations, a surface view of Gα_q showed AHD gradually separating from RD, creating the GDP/GTP exchange tunnel. This was less noticeable in the compound 1 replica 1 (Fig. 3, Supplementary Fig. S12). AHD remained closed in other replicas. Interestingly, the opening of Gα in the lemborexant-including simulations was accompanied by a slight rotation of the N-terminal helix of Gα, like described by Ahn et al.⁷² on the example of two cryo-EM structures of active-state NTSR1 (Supplementary Fig. S17).

Conclusions and discussion

The results of the all-atom MD simulations, including both mini-G_sqi¹⁵ and full-length Gα_q, suggest that lemborexant could behave as a non-typical agonist under certain conditions, rather than solely as an antagonist^57,58. Minimal fluctuations in its position in the orthosteric binding site of OX₂ indicated that it could stabilize the active-state receptor conformation, similarly to the known OX2R agonist compound 1. The maintained active state of OX₂ during 2 μs of simulations was confirmed through a conformational analysis of selected microswitches. Recently, Tyson et al.⁷³ suggested a new molecular mechanism for inverse agonists, through which they can bind to and act through GPCR-G protein complexes. This raises the question of whether lemborexant could instead be an inverse agonist. The authors observed that the microswitches present in the structures of inverse agonists bound to the κ-opioid receptor were in a mix of active (the PIF and DRW motifs), inactive (the orthosteric and Na + binding pockets, and the CWxP motif), and intermediate states (the NPxxY motif)⁷³. The microswitch conformations in the final frames of our simulations (Supplementary Fig. S5) were not in agreement with these observations but rather indicated the maintained active state of the receptor. On this basis, we concluded that lemborexant is unlikely to be an inverse agonist. However, there may be differences between the κ-opioid and orexin receptors that contribute to variations in inverse agonist behavior, and our simulations may not fully capture all possible receptor conformations, especially under different physiological conditions.

We analyzed and summarized the observed conformational changes (Table 1). Larger conformational changes of switch III correlated with the opening of Gα. This caused higher GDP position fluctuations in the lemborexant-including simulation systems in comparison to the compound 1 systems. Notably, during the first microsecond of the simulations, fluctuations of the GDP binding region were observed mostly for the lemborexant-OX₂-G protein multicomplexes and only slight fluctuations were noticed for the compound 1 complexes. This suggests differences between the assumed agonist-like behavior of lemborexant and the canonical mode of action of compound 1 in G_q signaling. It cannot be undoubtedly stated whether these differences refer only to the activation-like conformational changes of the studied multicomplex or also to the rate of the GDP/GTP exchange. However, we could suggest that GDP dissociation is faster for the lemborexant complexes whereas compound 1 seems to better stabilize the Gα_q-GDP complex. The impact of this on the rate of the GDP/GTP exchange and subsequent G_q-mediated signaling is unclear.

Table 1.

A comparison of the conformational changes observed in microsecond MD simulations including lemborexant and compound 1.

Conformational change	Compound 1 replica 1/2/3*	Lemborexant replica 1/2/3
Ligand position in the receptor binding site	No change	No change/no change/slight change
ECL2 closing	Yes/yes/yes	Yes/yes/yes
Changes of macroswitch III	Yes/no/no	Yes/yes/no
Changes of AHD position (opening of Gα)	Yes/no/yes	Yes/yes/yes
Formation of the GDP/GTP exchange tunnel	Yes/no/no	Yes/yes/no
Active-like direction of rotation of N-terminal helix of Gα like in NTSR1	Yes/no/no	No/no/no
Changes of macroswitch I to ‘on’ with R183G.hfs2.02 movement away from GDP	Yes/yes/yes	Yes/yes/yes
Changes of F341G.H5.08 microswitch to ‘on’	Yes/no/no	Yes/yes/no
Changes of D277G.HG.02 microswitch to ‘on’	Yes/yes/yes	Yes/yes/yes

*If differences between replicas were observed.

In addition, there is the question of whether the simulations performed with an active-state apo receptor would yield similar Gα activation observations³⁰. Though no such study has been performed for the orexin system, de Lima et al.⁷⁴ performed 1 µs MD simulations for an active-state apo µ-opioid receptor in complex with Gi-GDP. They observed that when the receptor was bound to Gi in a closed conformation, it switched to an inactive state and remained that way until the end of the simulation⁷⁴. The authors therefore concluded that GDP kept Gi in a closed conformation, and its activation and subsequent opening could be triggered by either the presence of an agonist or a larger simulation timescale⁷⁴. Previously, we also observed that the presense of an agonist is required to rebuild the receptor conformation into the fully active state, not intermediate, confirmed later with the cryo-EM structure of C5aR1³⁰. Considering the above studies, we first observed changes in G protein microswitch conformations within 1 μs of our simulations, leading to the conclusion that it was indeed the presence of an agonist that led to G_q activation. This was further confirmed by the experimental results obtained by Gratio et al.¹⁵.

The results presented here show only the first stages of the GDP dissociation and G_q activation accessible to MD timescales. In addition to the known role of switches I and II shown recently in metadynamics simulations of the receptor-free spontaneous G protein activation⁶⁹, we described the mechanism of the beginning of the opening of Gα_q and GDP dissociation. The opening of Gα_q involves two major movements of AHD rightward and downward (consistent with Fig. 2), in the direction known previously from the structure of the rhodopsin–Gi multicomplex (6CMO), wherein Gα is in its fully open conformation⁷⁰. As we observed, one movement of AHD (rightward) could require changes of the F341^G.H5.08 microswitch, while the second movement (downward) could require changes in switch III. Slight changes in switch I (Fig. 2), as shown in our simulations and in 6CMO, induced the translation of AHD to both sides. In addition, we analyzed the GDP binding site to determine which residues and interactions may contribute to its dissociation.

The GDP dissociation required the opening of Gα and it seemed to be associated mostly with the movement of AHD to the right (Fig. 4). The GDP release was mainly triggered by conformational changes near its phosphoryl end, i.e., a loss of interactions of GDP with R183^G.hfs2.02 located in switch I. Thus, no clear and definite changes in the D277^G.HG.02 microswitch were observed. Based on the above, we conclude that the sequence of conformational changes during Gα_q activation mediated by an OX₂ agonist could be as follows (Fig. 5):

• changes in the F341^G.H5.08 microswitch,

• a loss of interactions between the phosphoryl end of GDP and R183^G.hfs2.02 (switch I) caused by changes in switch I,

• the formation of temporary interactions between R183^G.hfs2.02 and E49^G.s1h1.04 (P loop),

• a loss of interactions between R183^G.hfs2.02 and E49^G.s1h1.04,

• a loss of interactions between E241^G.s4h3.10 (switch III) and two residues of switch II: R210^G.H2.01 and Q209^G.s3h2.03 (to a lesser extent), releasing E241^G.s4h3.10,

• the formation of interactions between R183^G.hfs2.02 and E241^G.s4h3.10, stabilizing the large distance between R183^G.hfs2.02 and GDP,

• the opening of the GDP/GTP exchange tunnel near the phosphoryl end of GDP,

• a loss of interactions between R183^G.hfs2.02 and E241^G.s4h3.10 and E49^G.s1h1.04

• changes in the D277^G.HG.02 microswitch,

• a loss of interactions between the guanine end of GDP and D277^G.HG.02 caused by changes in the D277^G.HG.02 rotamer,

• changes in helices HD and HA accompanied by the opening of Gα,

• GDP release accompanied by a further opening of Gα.

Fig. 5.

The suggested sequence of conformational changes that take place during the activation of Gα_q. The initial conformation is presented in gray, the suggested conformational change that takes place is presented in green, and the yellow dashed lines represent polar contacts.

In the canonical G_q signaling pathway, the GDP/GTP exchange initiates Gα_q protein activation, followed by the breaking of most Gα–Gβγ interactions. Gα activates phospholipase C (PLC)⁷⁵, which in turn converts phosphatidylinositol 4,5-bisphosphate (PIP₂) into inositol 1,4,5-trisphosphate (IP₃) and diacyl glycerol (DAG). IP₃, acting in the cytosol, induces Ca²⁺ release from the endoplasmic reticulum, while DAG, attached to the cell membrane, activates protein kinase C (PKC). Recent experimental results¹⁵ have proven that lemborexant induces cancer cell apoptosis through a non-canonical, Ca²⁺-independent pathway. This suggests that lemborexant does not activate the OX₁/OX₂-G_q complex in a classic way to release Gα for the subsequent PLC activation. Instead, Gβγ could be released or partially released to activate the Src kinase, which in turn phosphorylates the ITIM site in OX₁/OX₂, leading to SHP2 phosphatase recruitment, p38 MAP kinase activation, and subsequently caspase 3/7 activation. Other examples involving Gβγ have already been observed, e.g., for the Gi-coupled M2 muscarinic acetylcholine, α2-adrenergic, and somatostatin 2 receptors, and for Gs-coupled β-adrenergic receptors, which induced the ERK pathway through Gβγ-mediated stimulation of tyrosine kinases such as Src^76,77. In contrast, the G_q/11-coupled muscarinic acetylcholine receptor M1 and α1-adrenergic receptor activated the ERK-mediated apoptosis signaling by G_q/11^76,77. In the case of G_q-coupled P2Y₁ receptor, the stress-activated protein kinase (SAPK) activation (without involving p38) also evoked caspase activity, leading to apoptosis⁷⁶.

Here, we observed a clear difference in the GDP/GTP exchange rate leading to the opening of Gα_q. The faster activation of G_q by the lemborexant-OX2R complexes could be an impulse to evoke the Gβγ-mediated activation of Src. Presumably, these differences in the rate of the GDP/GTP exchange between lemborexant and compound 1 complexes affect the activation of G_q, leading to the activation of signaling pathways mediated by different subunits (Gβγ vs. Gα, respectively). However, the timescale currently accessible to MD simulations is not enough to observe the evolution of the second messenger system activated by orexin receptor effectors.

Abbreviations

GPCR

G protein-coupled receptor

OX₁

Orexin receptor type 1

OX₂

Orexin receptor type 2

OxA

Orexin-A

OxB

Orexin-B

GDP

Guanosine diphosphate

GTP

Guanosine triphosphate

MD

Molecular dynamics

AHD

Alpha helical domain

RD

Ras-like domain

MD

Molecular dynamics

Author contributions

This computational study was designed by D.L. Protein modeling, molecular dynamics simulations, and the analysis of simulation trajectories were performed by P.D. and D.L. The first draft of the manuscript was written by P.D. and D.L. Figures were designed and prepared by P.D. and D.L. Movies were prepared by D.L. The manuscript was edited and reviewed by all of the authors. All authors have given approval to the final version of the manuscript.

Data availability

PDB structures used in the current study are available at: https://www.rcsb.org (ID: 7L1V, 7SQ2, 6OSA, 6OS9, 6CMO). Protein sequences used in the current study are available at: https://www.uniprot.org (ID: P50148, O43613, O43614). Protein models and simulation trajectories generated during the current study are available at: 10.5281/zenodo.14035283 and 10.5281/zenodo.14035739.

Competing interests

The authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-03857-0.