Entropy drives the ligand recognition in G-protein-coupled receptor subtypes

Significance

In G-protein-coupled receptor (GPCR) drug discovery, a key challenge is achieving ligand subtype selectivity. From a thermodynamic perspective, this selectivity is driven by differences in binding free energies which stem from both static (enthalpic) and dynamic (entropic) interactions. While much attention has been given to the role of specific protein–ligand enthalpic interactions in selectivity, the origins and role of entropy in ligand-specific recognition of GPCR subtypes remain elusive. Through simulations and experimental verification, we have revealed the unusual molecular mechanism that highly subtype selective recognition of binding pockets by ligands is primarily governed by entropy. These results can advance the understanding of GPCRs and have the potential to reshape the future of innovative drug discovery targeting pharmacologically significant GPCR subtypes.

Abstract

Achieving ligand subtype selectivity within highly homologous subtypes of G-protein-coupled receptor (GPCR) is critical yet challenging for GPCR drug discovery, primarily due to the unclear mechanism underlying ligand subtype selectivity, which hampers the rational design of subtype-selective ligands. Herein, we disclose an unusual molecular mechanism of entropy-driven ligand recognition in cannabinoid (CB) receptor subtypes, revealed through atomic-level molecular dynamics simulations, cryoelectron microscopy structure, and mutagenesis experiments. This mechanism is attributed to the distinct conformational dynamics of the receptor’s orthosteric pocket, leading to variations in ligand binding entropy and consequently, differential binding affinities, which culminate in specific ligand recognition. We experimentally validated this mechanism and leveraged it to design ligands with enhanced or ablated subtype selectivity. One such ligand demonstrated favorable pharmacokinetic properties and significant efficacy in rodent inflammatory analgesic models. More importantly, it is precisely due to the high subtype selectivity obtained based on this mechanism that this ligand does not show addictive properties in animal models. Our findings elucidate the unconventional role of entropy in CB receptor subtype selectivity and suggest a strategy for rational design of ligands to achieve entropy-driven subtype selectivity for many pharmaceutically important GPCRs.

G-protein-coupled receptors (GPCRs), the largest superfamily of membrane proteins, play a crucial role in various physiological processes (1, 2) and are also closely related to numerous diseases. GPCRs are the most important targets for drug discovery with approximately one-third of all marketed drugs acting on these receptors (3–5). Despite the remarkable success of GPCR drug discovery, one of the major hurdles remains ligand selectivity, particularly subtype selectivity; (6, 7) each subfamily typically has multiple isoforms with high sequence homology that often have distinct pharmacological functions but can all be activated by the same/similar endogenous ligand(s). In recent years, significant efforts have been devoted to achieving subtype selectivity, and a number of highly selective GPCR ligands have been discovered (8). Surprisingly, however, recently solved high-resolution GPCR structures have shown that the binding pockets across receptor subtypes are structurally similar or identical, raising the question how selective ligands achieve their subtype selectivity (9–13).

From a thermodynamic perspective, ligand selectivity is often driven by differences in their binding free energies that depend on the contributions from both static (enthalpic) and dynamic (entropic) interactions (14). Although substantial attention has been focused on the role of specific protein–ligand enthalpic interactions in ligand selectivity, strong evidence is emerging that changes in entropy can control whether a protein–ligand interaction occurs, even among protein complexes with identical binding interfaces (15–17). However, the molecular origin and the role of entropy in ligand-specific recognition of GPCR subtypes remain elusive. Given the fact that GPCRs are highly flexible and dynamics (18), we hypothesize that entropy might play a hitherto underappreciated role in subtype-specific ligand selectivity of GPCR.

To explore the role of entropy in subtype-specific ligand selectivity of GPCR, we take the cannabinoid (CB) receptor subfamily as a model system. The CB receptor subfamily contains two highly homologous subtypes (CB1 and CB2). CB1, as the main mediator of the psychoactive effects of Cannabis sativa and its derivatives (19), is mainly expressed in the brain, and its targeted drugs are prone to produce neuropsychiatric adverse effects. For example, rimonabant, the first inverse agonist of CB1, has been withdrawn from the market due to a significantly increased incidence of psychiatric side effects, including increased anxiety and depression or suicidal intentions (20). In contrast, CB2 is primarily found in immune cells and is involved in the regulation of immunosuppression, apoptosis, cytokine release, proliferation, and migration (21, 22). Precisely because of the different expression patterns of CB1 and CB2, selectively targeting the CB2 receptor provides a promising avenue for the development of drugs to treat a variety of diseases while avoiding the severe psychiatric side effects associated with CB1 (23).

Designing CB2-selective agonists with promising clinical applications is difficult due to the high similarity in both the sequences and three-dimensional structures between CB1 and CB2, especially in the orthosteric binding pockets. Indeed, the recently determined structures of the CB receptors show that their orthosteric binding pockets consist of nearly identical amino acid residues with approximately the same geometric arrangements (Fig. 1A) (24), adding to their selectivity mechanism a layer of mystery. How can a ligand selectively bind to the corresponding receptor when the binding pockets of isoforms are nearly the same?

Fig. 1.

Orthosteric pockets in the CB receptor structures are similar in shape. (A) The orthosteric pockets observed in the cryo-electron microscopy structures of the active CB1 and CB2 CB receptors are similar in shape (pocket surface displayed in gray with selected side chains shown as sticks). (B) The chemical structure of YL025 and dose–response curves of YL025 in CB2 (or CB1)-expressing CHO cells using the GloSensor cAMP (cyclic adenosine monophosphate) assay. EC₅₀ value represents the mean value ± SEM of five independent experiments. The data were normalized to the amount of cAMP in cells stimulated with CP55,940 at a concentration of 1 μM, which was considered as 100%. (C and D) Dose–response curve of CP55,940 in various concentrations of YL025 in CB2 (or CB1)-expressing CHO cells was measured using the GloSensor cAMP assay. The plots in the curve are presented as mean ± SEM of four biological replicates.

This investigation uses a combined computational and experimental approaches to reveal the role of entropy in subtype recognition of CB2-selective agonists. By solving the high-resolution cryoelectron microscopy (cryo-EM) structure of CB2 with highly selective ligand, we find that enthalpic interactions cannot explain how the ligand achieves selectivity for CB2 over CB1. Further large-scale unbiased and metadynamics all-atom molecular simulations reveal that highly selective recognition of the ligand by binding pocket is primarily controlled by favorable binding entropy. To validate our computational results, we use them to design mutations predicted to alter receptor dynamics. We then confirm experimentally that these mutations significantly affect the activity of the CB2-selective ligand but not of the non-CB2-selective ligand. Based on the above findings, we further exploit this entropy-driven recognition to design ligands with higher selectivity, which exhibit favorable pharmacokinetic properties as well as significant efficacy in rodent inflammatory analgesic models. Furthermore, it exhibits none of the side effects associated with CB1 activation, including hypothermia and catalepsy. Our results provide important insight into the molecular forces responsible for ligand-GPCR-specific recognition and should aid in the design of alternative strategies for GPCR drugs.

Results

To investigate the molecular origin and the role of entropy in GPCR subtype ligand-specific recognition, we selected our previously discovered novel CB2-selective agonist, YL025 (previously compound (R)-25r in ref. 22), a 4-(1,2,4-oxadiazol-5-yl)azepan-2-one derivative, as our first model ligand (Fig. 1B). In CB2-transfected cells, YL025 inhibits forskolin-stimulated cyclic AMP (cAMP) production in a concentration-dependent manner, with an average EC₅₀ value of 26.5 ± 0.3 nM (n = 5) and a mean E_max value of 97.8 ± 1.2% (Fig. 1B). YL025 has no effect on forskolin-stimulated cAMP production in cells transfected with CB1 at concentrations up to 30 μM (n = 3) (Fig. 1B). Competitive experiments between YL025 and the CB1/CB2 nonselective agonist CP55,940 (24, 25) showed that YL025 significantly affects the potency of CP55,940 on CB2 (Fig. 1C). In contrast, YL025 has no effect on the potency of CP55,940 on CB1, even at concentrations as high as 10 μM (Fig. 1D). These data provide strong evidence that YL025 exhibits higher selectivity for the CB2 orthosteric site than for the CB1 orthosteric site.

High-Resolution Cryo-EM Structure Reveals Nearly Identical CB2 Pocket Bound to Selective and Nonselective Ligands.

As the first step toward understanding the molecular basis of subtype-specific binding of YL025, we solved the cryo-EM structure of YL025-CB2-G_i1-scFv16 at 3.13 Å resolution (Fig. 2 A and B and SI Appendix, Fig. S1 and Table S1). High-quality cryo-EM density in the CB2 orthosteric site allowed unambiguous modeling of YL025 (Fig. 2C). The ligand only has hydrophobic interactions with the residues from the orthosteric site residues (Fig. 2D), and no polar interactions are observed. Specifically, the 1-(4-fluorophenyl)azepan-2-one fragment is located above F117^3.36 and W258^6.48 and forms stable (T-shaped) π–π interactions with W194^5.43 and F183^ECL2 (Fig. 2D). The indazole ring of YL025 faces the TM1-TM2-TM7 region and has hydrophobic interactions with multiple aromatic residues on TM2 and TM7, including F87^2.57, F91^2.61, F94^2.64, H95^2.65, and F281^7.35. Mutating any of these residues to alanine significantly reduces the potency of the ligand induced the receptor activation (Fig. 2E and SI Appendix, Fig. S2 and Table S2). Considering that all these residues are also involved in the recognition of the CB2 antagonist AM10257 (26), the nonselective agonist CP55,940 (24) and AM12033 (27), and the CB2 slightly selective agonist WIN55,212-2 (28), no special interaction responsible for ligand subtype selectivity is found in the binding mode of YL025. Indeed, superposition of the YL025-bound and CP55,940-bound [Protein Data Bank (PDB) entry: 8GUR] CB2 structures reveals that all orthosteric pocket residues are in similar positions (Fig. 2F). Therefore, the static structure, although providing key information on ligand binding, cannot explain the molecular details of YL025's high subtype selectivity for CB2.

Fig. 2.

Cryo-EM structure and interactions between YL025 and CB2 receptor. (A) Cryo-EM map of the YL025-CB2–G_i1 complex. (B) Cryo-EM structure model of the YL025-CB2–G_i1 complex. (C) Map of YL025. (D) Contact residues around YL025 in the CB2 structure. The receptor side chains are shown as teal sticks, the backbone is a transparent cartoon, and YL025 is in orange. (E) Dose–response studies of YL025 activity for each mutant compared with wild-type CB2. Data shown are mean ± SEM of a representative experiment conducted in triplicate. (F) Binding pose comparison of YL025 in CB2 and CP55,940 in CB2. The residues involved in the ligand binding are shown as sticks.

Residue Differences in the Orthosteric Pocket Do Not Contribute to the Subtype Selectivity of the CB2-Selective Agonist.

Given the high degree of homology between CB receptors, all residues forming the orthosteric pocket (defined as within 4 Å of the ligand) of YL025 are nearly conserved between CB1 and CB2 (Fig. 3A). Only five residues differ, one in TM6 (CB2-V261^6.51/CB1-L359^6.51), one in TM3 (CB2-I110^3.29/CB1-L193^3.29), and three residues in ECL2 loop (CB2-L182^ECL2/CB1-I267^ECL2, CB2-L185^ECL2/CB1-H270^ECL2, and CB2-E181^ECL2/CB1-D266^ECL2). Our functional assays revealed that replacement of these residues in CB2 with the corresponding CB1 residues, however, did not reduce YL025’s potency (Fig. 3B and SI Appendix, Fig. S3). Likewise, no gain in potency for the swap mutants in CB1 was found with YL025 compared to the wild-type CB1 (SI Appendix, Fig. S3). In addition to the three different residues mentioned above that are closest to the ligand, ECL2, which participates in the formation of the orthosteric pocket, also possesses other different residues between the two receptors. Considering several studies have suggested that its shape may contribute to subtype selectivity of orthosteric ligands (13), we further generated two chimeric constructs: one containing the ECL2 sequence from CB1 and the remaining sequence from CB2 (hereby named CB1^ECL2/CB2^other) (SI Appendix, Fig. S4A) and the other containing the ECL2 loop of CB2 and other parts of CB1 (hereby named CB2^ECL2/CB1^other) (SI Appendix, Fig. S4B). Forskolin-stimulated cAMP assay studies showed that replacing the entire ECL2 loop of CB2 with the CB1 ECL2 loop had no effect on the EC₅₀ of the selective-agonist YL025 or nonselective agonist CP55,940 (Fig. 3C and SI Appendix, Fig. S4C). Furthermore, the CB1 receptor containing the CB2 ECL2 loop was expressed and recognized CP55,940 with the same potency as wild-type CB1 (SI Appendix, Fig. S4D). This receptor, however, still cannot be activated by YL025 (SI Appendix, Fig. S4E). All these data suggest that amino acid differences in the orthosteric site of the CB receptors are not determinants of the ligand subtype selectivity.

Fig. 3.

The binding mode of YL025 in CB1. (A) Residue differences within 4 Å of the ligand YL025. (B) Dose–response studies of YL025 activity for each mutant compared with wild-type CB2. Data shown are mean ± SEM of a representative experiment conducted in triplicate. (C) Dose–response curves of YL025 in CB1^ECL2/CB2^other-expressing and wild-type CB2-expressing CHO cells using the GloSensor cAMP assay. Data shown are mean ± SEM of a representative experiment conducted in triplicate. (D) Binding mode of YL025 in the CB1 orthosteric pocket in MD simulation trajectory. (E) Superposition of the CB2-YL025 and CB1-CP55,940 ligand-binding pockets.

To further confirm that residue differences in the orthosteric pocket do not play a role in achieving ligand subtype selectivity against CB receptors, we docked YL025 into the orthosteric binding site of active (PDB entry: 6KPG) CB1 in the same pose as in the CB2 cryo-EM structure and did not find any receptor–ligand steric conflicts (Fig. 3D). Unbiased molecular dynamics (MD) simulations based on this structure also did not observe any enthalpic interactions detrimental to ligand binding in the CB1 orthosteric pocket (SI Appendix, Table S3, Condition A). In fact, across all 15 independent trajectories, YL025 was found to stably bind to the orthosteric pocket of CB1 (Fig. 3D and SI Appendix, Fig. S5). In the majority of these trajectories, the binding pose of YL025 closely matched its conformation observed in the cryo-EM structure of CB2 (SI Appendix, Fig. S5B), while in the other seven trajectories, the indazole ring is rotated 180° to point toward TM7 and can form a hydrogen bond with the hydroxyl group on the side chain of CB1-S383^7.39 (Fig. 3D and SI Appendix, Fig. S5). In this binding mode, the geometric structure of the residues in the CB1 orthosteric site is basically the same as that in our previously resolved CB1-CP55,940 structure (29), in which the agonist CP55,940 can also form a hydrogen bond with S383^7.39 to stabilize the ligand binding (Fig. 3E).Taken together, our cryo-EM structure, MD simulations, and functional studies show that ligand subtype selectivity is not determined by residue differences in the orthosteric pockets of the CB receptors.

Favorable Entropy Drives Binding of YL025 and CB2.

Understanding the origin of the high subtype selectivity of YL025 required detailed thermodynamic analysis. We therefore computed the full free energy landscape associated with ligand binding using parallel tempering metadynamics simulations (PT-MetaD); PT-MetaD is a reliable enhanced MD sampling technique and has been successfully used to simulate rare events and reconstruct free energy landscapes of complex conformational rearrangements in various proteins (30–34), including GPCR (29, 35, 36). To aid simulation convergence, a funnel-shaped restraint was used to limit the ligand's exploration of the bulk water region (SI Appendix, Fig. S6A) (37, 38). The inactive-state conformations of both receptors were used because the extracellular site of the inactive-state receptor is expected to be more similar to the unoccupied, non-G-protein-coupled form of the receptors (39). The chosen collective variables (CVs) were the projections of the vector connecting the binding pocket and the ligand onto the Z-axis (CV_Z-projection) and onto the XY-plane (CV_{XY-projection}) parallel to the membrane (SI Appendix, Fig. S6A). In addition, the projection of the vector from the N terminus to the center of the receptor onto the Z-axis (CV_N-term) (SI Appendix, Fig. S6B) was also included in the CV, since our extensive preliminary experiments showed that increasing the swing of the N terminus can accelerate the convergence of the free energy. For comparison, we present analogous simulations in which we biased without CV_N-term (SI Appendix, Table S3, Condition H). In this case, the sampling quality is much worse and convergence cannot be not reached (SI Appendix, Fig. S7).

Metadynamics simulations were terminated when the relevant CV space was thoroughly explored and the estimation of the binding free energy adopted asymptotic behavior (SI Appendix, Figs. S8–S11). It is clear from the free energy landscape that YL025 undergoes a stepwise binding process. Initially, it fluctuates in the extracellular solvent phase (unbound state, U, Fig. 4A and SI Appendix, Fig. S12). In this state, the N terminus of CB1 lies above the orthosteric pocket and is partially inserted into the pocket, while the N terminus of CB2 swings freely above the receptor pocket. After overcoming the desolvation barrier, YL025 binds to the extracellular region of the receptor and forms a metastable intermediate state (Intermediate state, I, Fig. 4A and SI Appendix, Fig. S12). Specifically, YL025 is located between the extracellular vestibule and the N terminus of CB2 and can form polar interactions with the N terminus. This is consistent with the signaling assays, that is, the activity of YL025 is reduced when the N-terminal of CB2 is replaced by the N-terminal of CB1 (SI Appendix, Fig. S13). In addition, it can be found that the F atom of YL025 can form hydrogen bonds with the extracellular residues of CB2, which facilitates the binding of ligands. This is also consistent with our previous finding that removing the F atom from the ligand significantly affects its activity on CB2 (22).

Fig. 4.

Different ligands have different thermodynamic bases for binding to CB receptors. (A) Free energy surface of CB2 in complex with YL025 at 300 K. Representative conformations associated with the different states indicated in FES are shown in small boxes. (B) Shown are the Helmholtz free energy (ΔF), enthalpy (ΔU), and entropy (−TΔS) for the binding of different ligands onto the CB receptor at 300 K. Values are tabulated in SI Appendix, Fig. S14A. (C) Schematic representation of CB1-YL025 and CB2-YL025, colored by rms fluctuation (RMSF) differences. The color scale used is −0.4 Å (blue) to 0.4 Å (red). (D) Superimposition of YL025 conformations from the last 500 ns of the CB1-YL025 simulation. (E) Distributions (histograms) of alpha carbon atom distances of residues Q116^1.32 and P251^4.60 located in the extracellular region of the CB1 orthosteric pocket under different simulation conditions: green, with CP55,940; blue, with YL025. (F) Distance between the alpha carbon atoms of residues Q116^1.32 and P251^4.60 located in the extracellular region of the CB1 orthosteric pocket. Representative frames are shown in the box below, with corresponding distances indicated in red. (G) Superimposition of CP55,940 conformations from the last 500 ns of the CB1-CP55,940 simulation.

After passing through the intermediate state, the ligand enters the orthosteric pocket and adopts a global minimum pose similar to the cryo-EM pose. Although the binding process and the most stable binding poses of YL025 are very similar in CB1 and CB2, the binding free energies are significantly different, about −9.6 kcal mol⁻¹ in CB2 and −5.8 kcal mol⁻¹ in CB1 after correction for the loss of translational and rotational freedom of the unbound ligand due to the funnel-like boundaries (SI Appendix, Fig. S14) (38), consistent with YL025 having a high selectivity on CB2.

To further quantify the role of the underlying thermodynamics in ligand binding, we undertook a detailed thermodynamic analysis and estimated separately enthalpy ΔU and entropy ΔS of binding, thanks to the high level of accuracy that the combination of PT-metaD and good quality CVs (SI Appendix, Figs. S15–S18). The results clearly show that not only does the binding strength of YL025 to the two receptors differ, but also the thermodynamic basis of the interaction with the receptor is fundamentally different (Fig. 4B and SI Appendix, Fig. S14). Specifically, the binding of YL025 to CB2 is primarily entropy driven (−TΔS = −39.2 kcal mol⁻¹, Fig. 4B), with a binding enthalpy penalty (ΔH = 29.6 kcal mol⁻¹, Fig. 4B). On the contrary, the binding of YL025 to CB1 has a favorable enthalpy contribution (ΔH = −26.5 kcal mol⁻¹), which is consistent with the above finding that YL025 can form a stable hydrogen bond with S383^7.39 of CB1, but the entropy loss is relatively large (−TΔS = 20.7 kcal mol⁻¹). Thus, the entropy difference Δ(−TΔS), of YL025 binding to CB1 and CB2 is enormous. Generally, entropy changes upon binding may be due to reduced translational and rotational degrees of freedom, changes in the conformational flexibility of the binding protein, and reorganization of their solvation shells upon binding (40). The desolvation effects of CB1 and CB2 during ligand binding should be comparable because their binding pockets are highly conserved. Furthermore, the ligand binding mode of YL025 is highly similar between the isoforms, ruling out differences in ligand strain or ligand desolvation as primary contributors to the observed selectivity. This switch in the thermodynamic profile is therefore directly related to the conformational dynamic nature of the orthosteric pocket of the receptor.

To gain a deeper understanding of the conformational dynamics of the orthosteric pocket, we further carried out unbiased MD simulations for a total of 5 μs to quantify the amplitude of movement of the residues during ligand binding (SI Appendix, Table S3, Conditions E and F). The results show that when no ligand is bound, the extracellular region of CB1, especially that of TM1 and TM2, is significantly more flexible than that of CB2, as reflected by the larger rms fluctuation (RMSF) (SI Appendix, Fig. S19). When YL025 binds to CB2, the residues surrounding the orthosteric pocket show little overall change (Fig. 4C). However, the majority of residues constituting the G-protein-binding site exhibit significantly higher RMSF values, indicating an enhancement in the magnitude of motion that is beneficial to receptor activation and is also consistent with an increase in the entropy of YL025 binding to CB2.

In contrast, binding of YL025 to CB1 results in a widespread decrease in RMSF, indicating a global rigidification of the protein (Fig. 4C). In particular, conformational changes in the extracellular regions of TM1, TM2, and TM7 are significantly restricted, which is agreement with YL025 having unfavorable binding entropy in CB1. At the same time, YL025 is also relatively restricted in CB1 due to its rigidity, and its rmsd amplitude oscillates around 0.5 Å (Fig. 4D). These data appear to suggest that due to YL025's limited ability to change conformations, its binding at the CB1 orthosteric site limits the flexibility of CB1 pocket, thereby detrimental to binding entropy and, thus, binding of YL025.

To confirm that agonists of CB1 need to be flexible enough to adapt to changes in receptor pocket shape, we simulated the binding thermodynamics of the agonist CP55,940 on CB1 (SI Appendix, Figs. S20–S22). Metadynamics simulations clearly show that unlike the binding of YL025 on CB1, the binding of CP55,940 on CB1 is an entropy-increasing process (Fig. 4B and SI Appendix, Fig. S14D). Unbiased MD simulations (Fig. 4 E–G and SI Appendix, Fig. S23) further confirmed that the extracellular region of the orthosteric pocket of CP55,940-bound CB1 exhibits great flexibility, with pocket distance varying by up to 7 Å (Fig. 4 E and F and SI Appendix, Fig. S23A). More importantly, although the binding of CP55,940 is overall very stable (SI Appendix, Fig. S23B), its flexible propanol chain can adjust its position according to the conformational changes of the CB1 pocket, thereby exhibiting multiple binding modes (SI Appendix, Fig. S23 C and D). These data provide strong evidence that the orthosteric pocket of CB1, which is inherently more flexible than that of CB2, prefers to bind agonists that are intrinsically more flexible and can thus adjust their conformations as the pocket shape changes. In other words, when a ligand matches the pocket shape of a CB receptor, the greater its rigidity, the less favorable the binding entropy for CB1 and the higher the subtype selectivity for CB2.

Mutagenesis Confirms the Importance of Entropy in Ligand Subtype Selectivity.

In order to evaluate the mechanism of entropy-driven subtype selectivity, we set out to investigate the conformational plasticity of CB receptors by mutagenesis studies.

Structural comparison clearly shows that the orthosteric pocket shape of CB2 remains essentially unchanged during receptor activation, but the active-state binding site of CB1 shows a significant reduction in size (53%) due to the inward bending of the extracellular parts of TM1 and TM2 toward the orthosteric binding site (SI Appendix, Fig. S24) (27). This is also consistent with the above results, that is, the TM1 and TM2 extracellular regions of apo CB1 are more dynamic than those of CB2. Previous studies have shown that the conformational changes of TM2 play a crucial role in the activation of CB1, which gives CB1 a unique activation mechanism compared with CB2 (29, 41). Therefore, we altered receptor plasticity by constructing mutants that alter TM2 dynamics.

The kink of TM2 in CB1 is located near S167^2.54, a helix above the highly conserved D163^2.50 residue that is critical for signal transduction of multiple GPCRs (4, 42), including CB1 and CB2 (SI Appendix, Fig. S24) (43). In CB receptors, D^2.50 forms a polar hydrogen bonding network with surrounding TM1, TM7, and TM3 to maintain the helical stacking structure and transmit signals (SI Appendix, Fig. S24). S84^2.54 in CB2 is located just above D80^2.50 and forms a stable hydrogen bond with S47^1.46 on TM1 (Fig. 5A), which enhances the interaction between TM1 and TM2 and helps the receptor maintain rigid during activation. In contrast, S167^2.54 in CB1 does not participate in any transmembrane helix interactions except that its side chain forms a hydrogen bond with the main chain of D163^2.50 on the same transmembrane helix (Fig. 5A). MD simulations show that this is mainly due to the fact that T130^1.46 forms a hydrogen bond with T391^7.47 on TM7 (Fig. 5 A and B), resulting in the inability to form a hydrogen bond with S167^2.54. Such a result leads to a weakening of the interhelical constraints between TM2 and TM1 in CB1 and the increase of the dynamics of TM2. Simulations further show that during receptor activation, the S47^1.46-S84^2.54 hydrogen bond of CB2 exists very stably (Fig. 5B), while the T130^1.46-T391^7.47hydrogen bond in CB1 needs to be released (Fig. 5B), so that TM1 can move toward the center of the helix bundle) and drive TM2 conformational rearrangement, ultimately promoting receptor activation. Therefore, if M293^7.47 and S47^1.46 in CB2 are mutated to threonines at the corresponding positions on CB1, it should increase the probability of TM1-TM7 hydrogen bonds and reduce the probability of TM1-TM2 hydrogen bonds, which should increase the flexibility of CB2. We predicted CB2-M293^7.47T&S47^1.46T mutation would not have much impact on the binding of CP55,940, which itself has a certain degree of flexibility, but it should have an impact on the binding of YL025, which is relatively rigid. Indeed, this mutation resulted in a substantial reduction in the activity of YL025, while the activity of CP55,940 remained essentially unchanged (Fig. 5 C and D). On the contrary, if T391^7.47 and T130^1.46 in CB1 are mutated to methionine and serine, respectively, at the corresponding positions on CB2, then theoretically the hydrogen bond between TM1 and TM7 in CB1 will be destroyed, and the hydrogen bond similar to that in CB2 will be formed between TM1 and TM2, which will weaken the flexibility of CB1 and should be beneficial to the binding of YL025 on CB1. Mutation experiment showed that the CB1-T391^7.47M&T130^1.46S double mutation indeed improved the activity of YL025 on CB1 (SI Appendix, Fig. S25). In addition, precisely because of the high dynamics of TM1-TM2 in CB1, the presence of the D176^2.63-K192^3.28 salt bridge between TM2 and TM3 is crucial for maintaining the receptor function of CB1 (Fig. 5E). Simulations showed that the salt bridge interacts more strongly when the CB1 pocket opens and becomes weaker when the pocket tightens (Fig. 5F). Differently, the corresponding position in CB2 forms a hydrogen bond interaction (D93^2.63-K109^3.28) that is weaker than the salt bridge interaction. The strength of this interaction basically does not change during receptor activation (Fig. 5G). These results imply that the D176^2.63-K192^3.28 salt bridge in CB1 is important for constraining the conformational dynamics of TM2. Therefore, if D176^2.63 in CB1 is mutated to the asparagine, TM2 would become more flexible, which would be not conducive to ligand binding. Indeed, mutation experiments showed that the D176^2.63N mutation of CB1 significantly reduces the activity of CP55,940 (Fig. 5H).

Fig. 5.

Different plasticity of CB1 and CB2 orthosteric pockets. (A) CB1 and CB2 have different transmembrane interhelical interactions around the highly conserved D2.50. (B) Probability of hydrogen bond interactions between different pairs of amino acid residues. (C) Dose–response studies of YL025 activity for M293^7.47T&S47^1.46T double mutants compared with wild-type CB2. Data shown are mean ± SEM of a representative experiment conducted in triplicate. (D) Dose–response studies of CP55,940 activity for M293^7.47T&S47^1.46T double mutants compared with wild-type CB2. Data shown are mean ± SEM of a representative experiment conducted in triplicate. (E) CB1 and CB2 have different polar interactions at the extracellular ends of TM2 and TM3. (F) Distributions (histograms) of the distance between atoms with polar interactions in the side chains of residues D176^2.63 and K192^3.28 under different simulation conditions: green, inactive; blue, active. (G) Distributions (histograms) of the distance between atoms with polar interactions in the side chains of residues N93^2.63 and K109^3.28 under different simulation conditions: orange, inactive; red, active. (H) Dose–response studies of CP55,940 activity for D176^2.63N mutant compared with wild-type CB1. Data shown are mean ± SEM of a representative experiment conducted in triplicate.

Ligand Modification Further Confirms the Important Role of Entropy in Ligand Subtype Selectivity.

To further validate the computationally revealed entropy-driven selectivity mechanism, we sought to enhance or weaken subtype selectivity by modifying YL025's ability to adapt to dynamic changes in the conformation of the orthosteric pocket of the CB receptors.

First, since the orthosteric pocket of CB1 was found to be intrinsically more flexible than CB2, a more rigid ligand should be more favorable to its binding entropy on CB2 and less favorable to its binding entropy on CB1. Therefore, we sought to make YL025 more rigid, which would further facilitate the realization of CB2 high activity and subtype selectivity. Specifically, we designed YL035 by replacing the seven-membered lactam of YL025 with a six-membered amide fragment. To further increase the rigidity of the compound, a -CH₃ group was added to the chiral center position, coverting the chiral carbon into a quaternary center to maintain YL035 as (R)-configuration (consistent with the configuration of YL025) (Fig. 6A). Quantum chemical calculations clearly show that these modifications significantly reduce the flexibility of the small molecule, i.e., increase its rigidity (Fig. 6A). The computationally derived selectivity mechanism would predict that, due to the increased rigidity of the ligand, its ability to undergo conformational changes as the orthosteric pocket shape of CB1 changes would decrease, which would be detrimental to the binding entropy of the ligand on CB1. Meanwhile, the increase in ligand rigidity would be more conducive to matching the relatively rigid binding pocket of CB2. Therefore, YL035 would further have higher potency on CB2 and CB2 subtype selectivity than YL025.

Fig. 6.

Rational design of ligands with altered entropy-driven subtype selectivity. (A) To test our prediction, we synthesized two YL025 analogs with different flexibility. (B) Dose–response curve of YL035 in CB2-expressing CHO cells using the GloSensor cAMP assay. EC₅₀ value represents the mean value ± SEM of four independent experiments. The data were normalized to the amount of cAMP in cells stimulated with CP55,940 at a concentration of 1 μM, which was considered as 100%. (C) Dose–response curves of YL065 in CB2(or CB1)-expressing CHO cells using the GloSensor cAMP assay. EC₅₀ value represents the mean value ± SEM of three independent experiments. The data were normalized to the amount of cAMP in cells stimulated with CP55,940 at a concentration of 1 μM, which was considered as 100%. (D) YL035 was able to reduce inflammatory hyperalgesia by activating CB2 in vivo. Carrageenan (1%) was administered into the foot paws of mice to induce inflammatory hyperalgesia. Mechanical allodynia was assessed at the 5 h following the injection of carrageenan, a time at which hyperalgesia would be at its highest point. The YL035 solution (20 mg/kg, 10 mg/kg, 5 mg/kg, i.p.) (n = 8) or vehicle (n = 8) was administered 1 h prior to the measurement. AM630 (10 mg/kg) was injected intraperitoneally to mice 10 min before administration of vehicle (n = 8) or YL035 (20 mg/kg, i.p.) (n = 8). The data, presented as mean ± SEM, were subjected to analysis using Student's t test. ^****P < 0.0001, ^###P < 0.001, ns: no significance. (E) The changes of body temperature after administration of WIN55212-2 and YL035. The body temperature of mice was measured using a forehead thermometer at the time of 0, 0.5 h, 1 h, 2 h, and 4 h subsequent to the administration of WIN55212-2 (20 mg/kg, i.p.) (n = 6), YL035 (20 mg/kg, i.p.) (n = 6), or vehicle (10 μL/g, i.p.) (n = 6). Data (mean ± SEM) were analyzed by Student’s t test between two groups. ^**P < 0.01; ns: no significance. (F) The flow diagram of the experiment of YL035 in the Conditioned Place Preference (CPP) test. (G) Effects of vehicle (n = 9), CP55,940 (0.3 mg/kg, i.p.) (n = 8), and YL035 (20 mg/kg, i.p.) (n = 10) in the conditioned place preference paradigm. The CPP score, denoted as T1 − T2, represents the difference between the time spent in the drug-paired compartment on the test day and the pretest day, as indicated by T1 and T2, respectively. This calculation serves as a measure of the change in preference for the drug-associated environment over time. Data (mean ± SEM) were analyzed by Students t test between two groups. ^*P < 0.05 for the CP55,940 group relative to the vehicle group; ^#P < 0.05 for the YL035 group relative to the CP55,940 group.

Indeed, YL035 behaves as predicted, i.e., it activates CB2 more effectively than YL025, with the mean EC₅₀ value decreasing from the original 26.5 ± 0.3 nM to 8.1 ± 0.2 nM (Fig. 6B). In the meantime, similar to YL025, YL035 has no obvious effect on forskolin-stimulated cAMP production in cells transfected with CB1 at concentrations up to 30 μM (SI Appendix, Fig. S26). Furthermore, YL035 also displays favorable pharmacokinetic properties with sufficient exposure when administered intraperitoneally at a dose of 20 mg/kg (SI Appendix, Fig. S27). In addition, YL035 can produce dose-dependent analgesic effects in a mouse inflammatory pain model and can eliminate pain when mice were pretreated with the specific CB2 antagonist AM630 (Fig. 6D and see SI Appendix for details). YL035 also does not produce side effects associated with CB1 activation, including significant hypothermia (Fig. 6E), catalepsy (Movies S1 and S2), and the development of conditional place aversion following an intraperitoneal injection (Fig. 6 F and G and see SI Appendix for details). Additionally, YL035 demonstrated low toxicity both in vitro and in vivo. The MTT(3-[4,5-dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide) assay revealed that YL035 did not exhibit significant cytotoxic effects on various cell lines, including HK-2, LX-2, Müller, HUVEC, COS-7, and BEAS-2B, with IC₅₀ values over 60 μM (SI Appendix, Fig. S28). Furthermore, in a repeated dose toxicity study, mice subjected to daily intraperitoneal administration of YL035 at a dose of 60 mg/kg for 14 consecutive days exhibited no noticeable changes in body weight (SI Appendix, Fig. S29) or any evident damage to various organs (SI Appendix, Fig. S30). Moreover, the high degree of conservation of sequences of CB receptors in human and mouse sources supports the feasibility of correlating cell-based data with human receptors and validating data derived from animal models (SI Appendix, Fig. S31).

To further verify this mechanism, we tried to make ligand more flexible, which is expected to be detrimental to its binding entropy on CB2 but beneficial to its binding entropy on CB1, and thus not expected to contribute to its high activity/subtype selectivity on CB2. Therefore, we designed YL065 by removing the -CH₃ group in the chiral center and the carbonyl group on the basis of YL035. Quantum chemical calculations show that this enhances the ligand's flexibility (Fig. 6A). Signaling assays demonstrate that activation of CB2 by YL065 is significantly attenuated, with the average EC₅₀ value increasing from the original 26.5 ± 0.3 nM to 96.1 ± 4.6 nM. It also shows an obvious activation effect on CB1, with EC₅₀ of 109.7 ± 7.6 nM (Fig. 6A).

Taken together, the ligand modification results agree with our computational predictions and provide additional support that binding entropy is critical for achieving receptor subtype selectivity. Subtype selectivity tuning can be achieved by modulating ligand flexibility to alter the binding entropy.

Discussion

The impact of entropic effects on selectivity has received less attention in GPCRs due to the difficulty in experimentally estimating changes in membrane protein binding entropy. Using large-scale atomic MD simulations, coupled with experimental validation, we highlight the important role of entropic effects in GPCR subtype selectivity. We focus here on CB receptors not only because they represent a class of classic GPCR targets for studying receptor regulation and signaling transduction but also because achieving their subtype selectivity is crucial for the development of small molecules for the treatment of various diseases.

We found that within the CB receptor family, the orthotopic pocket of CB1 is naturally highly dynamic compared to CB2. This requires that CB1 agonists need to be flexible enough to dynamically adapt to the highly dynamic nature of its pocket. When ligands are rigid, their structures are less capable of conforming to the varying shapes of the CB1 pocket across different states. Moreover, such rigidity can restrict the conformational adjustments of the CB1 pocket, which are crucial for its activation. Consequently, rigid ligands are disadvantageous for binding entropy at CB1. In contrast, the CB2 pocket undergoes less variation during activation and is not as dynamic as that of CB1. Therefore, the more rigid a ligand that can match the shape and electrostatic properties of the CB2 pocket, the smaller the entropy loss during binding. Additionally, because it induces conformational changes in the CB2 receptor structure, particularly in the intracellular region, the binding of such a ligand results in an increase in binding entropy. Therefore, similar to a recent result based on a Markov state model (44), plasticity in the pocket itself between receptor subtypes enables subtype selectivity through differences in entropic effects.

Our study supplements existing experimental results. Recent high-resolution crystallographic and cryo-EM structures have demonstrated a high degree of consistency in the orthosteric pockets of CB1 and CB2 receptors (25–28, 41, 45–49). Notably, a structural biology research has shown that even when CB2 is bound with selective agonists, the consistency with nonselective agonist-bound pocket residues remains high (24). While these findings are crucial for understanding the mechanisms of ligand binding, they do not fully explain the subtype selectivity of CB receptors, adding complexity to the understanding of the underlying reasons for this selectivity. Our simulation and experimental results indicate the presence of significant differences in entropy effects associated with ligand binding, highlighting the critical role of entropy in subtype-specific binding.

There are several key points worth noting. First, the force field used in our MD simulations is not perfect and may introduce artifacts; however, simulations have been validated as highly accurate in characterizing fundamental thermodynamic properties such as free energy, entropy, and enthalpy (50). Additionally, we employed a replica exchange strategy in our binding free energy calculations, allowing us to obtain binding free energies at different temperatures within the same simulation, significantly offsetting systematic errors. Given the substantial computational resources required by PT-MetaD, future studies might consider more advanced enhanced sampling methods, such as OneOPES (51). Furthermore, while our analysis primarily focuses on Class A GPCR CB receptors, future research should explore the role of entropy in driving subtype selectivity across other GPCR classes.

The mechanism we identified may be generalized to other GPCRs. Despite the high degree of homology between GPCR isoforms, there are differences in their dynamic properties. At several GPCRs, small adjustments to ligand flexibility can confer or reverse selectivity, suggesting that entropic effects play an important role in selectivity (52, 53). Although further computational and experimental work is required to determine how entropic effects can be exploited to achieve selectivity for other receptors, this strategy of exploiting inherent differences in dynamics between highly homologous receptors to design entropy-driven subtype-selective ligands may be broadly applicable to other receptor GPCRs and other protein superfamilies and will become more common in future therapeutic design.

Materials and Methods

We performed MD simulations of CB1 and CB2, with lipids and water represented explicitly, using the CHARMM force field (54). MD and metadynamics simulations were performed using GROMACS (55) and PLUMED (56). General MD simulations setup and metadynamics setup are provided in SI Appendix, Methods 2.1. The binding enthalpy and entropy are estimated by a linear fit procedure, using the relationship ΔF = ΔU − TΔS. Details of the pharmacological experiments, including cell culture, GloSensor cAMP assay, MTT assay, toxicity study and H&E (hematoxylin-eosin) staining, enzyme-linked immunosorbent assay, plasmids construction, carrageenan induced inflammatory pain in mice, central nervous system-mediated side effects, pharmacokinetic analysis in vivo, competition binding assay of CP55,940 and YL025 to orthosteric binding sites, conditioned place preference or aversion are provided in SI Appendix, Methods 2.2. Details of the cryo-EM structure determination (including construct construction, expression and purification of antibody scFv16 and YL025-CB2-G_i1-scFv16 complex, cryo-EM grid preparation and data collection, and data processing) and synthesis procedures for CB2 agonists are provided in SI Appendix, Methods 2.3 and 2.4, respectively.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (Z01)

Dataset S03 (Z02)

Dataset S04 (Z03)

Dataset S05 (Z04)

Dataset S06 (Z05)

Dataset S07 (Z06)

Dataset S08 (Z07)

Dataset S09 (Z08)

Dataset S10 (Z09)

Dataset S11 (Z10)

Dataset S12 (Z11)

Dataset S13 (Z12)

Dataset S14 (Z13)

Dataset S15 (Z14)

Dataset S16 (Z15)

Dataset S17 (Z16)

Dataset S18 (Z17)

Dataset S19 (Z18)

Dataset S20 (Z19)

Dataset S21 (Z20)

Dataset S22 (Z21)

Dataset S23 (Z22)

Dataset S24 (Z23)

Dataset S25 (Z24)

Dataset S26 (Z25)

Dataset S27 (Z26)

Dataset S28 (Z27)

Dataset S29 (Z28)

Dataset S30 (Z29)

Dataset S31 (Z30)

Dataset S32 (Z31)

Dataset S33 (Z32)

Dataset S34 (Z33)

Dataset S35 (Z34)

Dataset S36 (Z35)

Dataset S37 (Z36)

Dataset S38 (Z37)

Dataset S39 (Z38)

Dataset S40 (Z39)

Dataset S41 (Z40)

Dataset S42 (Z41)

Dataset S43 (Z42)

Dataset S44 (Z43)

Dataset S45 (Z44)

Dataset S46 (Z45)

Dataset S47 (Z46)

Dataset S48 (ZIP)

Dataset S49 (ZIP)

Movie S1.

The positive control compound caused significant symptoms of catalepsy. After intraperitoneal injection of WIN55,212-2 (20 mg/kg), mice exhibited symptoms of catalepsy.

Movie S2.

YL035 did not cause obvious symptoms of catalepsy. After intraperitoneal injection of YL035 (20 mg/kg), mice did exhibit catalepsy symptoms.

Data, Materials, and Software Availability

Cryo-EM structure and MD simulations' input files data have been deposited in PDB BANK (PDB CODE: 8X3L) (57). Cryo-EM map data have been deposited in Electron Microscopy Data Bank (EMDB), EMDB ID: EMD-38039 (58).