Efficient Characterization of GPCRs Allosteric Modulation: Application to the Rational Design of De Novo S1PR1 Allosteric Modulators

Abstract

G-protein-coupled receptors (GPCRs) constitute the largest superfamily of integral membrane proteins in the human genome that mediate most transmembrane signaling processes. Malfunction of these signaling processes is related to many human pathologies (Parkinson’s, heart diseases, etc.), causing GPCRs to be the largest family of druggable proteins. Traditionally, GPCRs were targeted by orthosteric ligands. However, this regulation usually causes side effects, provoking many GPCRs-associated pathologies to remain without an effective treatment. Allosteric regulation offers a promising alternative to circumvent this problem, and consequently, comprehending its details is of utmost importance. For this reason, we developed in the present work a methodology to study the allosteric modulation in a comprehensive way. Specifically, this methodology allows calculating the free energy profiles $\left(\text{Δ}{G}_{\text{total}}\right)$ for activation processes of GPCRs and their derived complexes combining the usage of targeted molecular dynamics (TMD) simulations to generate the intermediate structures of a given activation process, with the protein-dipole Langevin-dipole (PDLD) method within the linear response approximation (LRA) framework (PDLD/S-LRA-2000) and our refined coarse-grained $\left(\text{CG}\right)$ model for GPCRs to calculate the binding and conformational free energy contributions ($\text{Δ}{G}_{\text{conf}}$, $\text{Δ}{G}_{\text{bind}}$), respectively, which take into account the cellular membrane effects by an implicit membrane. Sphingosine 1-phosphate receptor 1 (S1PR1) has been chosen as a case study due to its available data for benchmark purposes. Apart from validating our developed methodology, the conducted S1PR1 study has partially filled its knowledge gap regarding allosteric modulation and has allowed rational design of a de novo pure positive allosteric modulator for one of its prospective allosteric cavities according to our calculations. The methodology presented in this paper provides a very useful tool to study the GPCRs allosteric modulation, and the GPCRs activation processes in general, which will hopefully encourage a more thorough exploration of the topic.

Graphical Abstract

1. INTRODUCTION

G-protein-coupled receptors (GPCRs) constitute the largest superfamily of integral membrane proteins in the human genome that exhibit a characteristic seven transmembrane α helices (TMs) structure.^1,2 Thanks to their presence in most cell types, GPCRs play a key role in a vast number of physiological processes by mediation of the majority of intracellular responses to external stimuli, such as light and hormones, through signal transduction processes that canonically involve G proteins.^2–4 All this causes GPCRs to be one of the most outstanding receptor families in the human body whose malfunction of their signal transduction processes is associated with many human pathologies that are not totally solved medically, pointing out neurological/neurodegenerative and heart diseases.^5,6 Consequently, given the importance of proper signal transduction, GPCRs represent one of the most important pharmacological targets; indeed, around 35% of FDA (Food and Drug Administration)-approved drugs currently target these receptors.^7,8 For all this, it is of utmost importance to fully comprehend how GPCRs take part in these signal transduction processes in order to be able to tackle their associated pathologies.

GPCRs are not simple on/off switches since there is a preexisting equilibrium between their inactive and active states which interconvert through a series of intermediate states.^3,9 Distinct conformational ensembles and functions are associated with these different states whose relative populations can be modified by ligand binding.^10,11 Active states deserve special mention given that they initiate signal transduction processes triggering G-proteins downstream (see Figure 1).

Figure 1.

Schematic representation of GPCRs preexisting equilibrium between their inactive (I), intermediate (Int) and active (A) states where it has been depicted how orthosteric and allosteric regulation are able to shift this equilibrium changing the relative stabilities and populations of the aforementioned states. In this figure, an activation process, TM6 (in green) undergoes an outward movement from the GPCR transmembrane bundle (in navy blue), induced by ligand binding (orthosteric and allosteric ligands in orange and magenta, respectively) has been represented whose intermediate and active states trigger signal transduction processes.

GPCRs possess three different types of cavities in which ligand binding takes place, particularly, the orthosteric, G-protein and the allosteric sites, though its regulation mainly takes place by the orthosteric and allosteric sites.^12,13 According to the binding site, two types of GPCR regulation can be distinguished. On the one hand, there is the orthosteric regulation that occurs in the orthosteric site. This cavity, found on the extracellular side, is generally employed by exogenous ligands to trigger signal transduction processes. It is important to emphasize that the orthosteric site shows high sequence similarity among members of the same GPCR subfamily, which has resulted in most druggable GPCRs not having been exploited pharmacologically yet due to side effects provoked by the multiple targets of homologous receptor subtypes, causing many GPCRs-associated pathologies to remain without an effective treatment.^12,14 On the other hand, there is the allosteric regulation that occurs in the allosteric cavities located at the surface of GPCR transmembrane bundle. This regulation offers a promising alternative approach to circumvent the aforementioned problematic since the degree of conservation of allosteric sites between homologous receptor subtypes is generally much less.^13,15,16 From a mechanistic point of view, allosteric ligands modulate the orthosteric-ligand response, potentiating/attenuating their affinity and/or efficacy, i.e., acting as positive/negative allosteric modulators (PAMs/NAMs). In addition, allosteric ligands can also exhibit intrinsic efficacy by themselves, which can take place together with their orthosteric-ligands modulation.^17,18 However, neither the allosteric cavities in most GPCRs (if they have any) nor their possible ligands are currently known. Then, in order to exploit the enormous potential of allosteric regulation, it is necessary to tackle the following issues: (1) Identification of allosteric cavities/sites; (2) Rational design of ligands for given cavities; (3) Activity characterization of the designed allosteric modulators. Although there are different robust methods and methodologies to deal with these two first points,^19–21 determining the allosteric-modulators character is still a challenge since available alternatives, both experimental and computational, are quite costly. Despite recent progress in terms of efficiency for computational methodologies,²² high-performance computing and major computing time are still necessary for reliable simulation of biological systems with the size and complexity of GPCRs.

For this reason, the present work tackles the development and benchmark of a methodology that should allow one to study allosteric modulation in a comprehensive way, adapting the existing methods that have been used by us in investigating GPCRs. Specifically, our computational methodology allows one to calculate the free energy profiles $\left(\text{Δ}{G}_{\text{total}}\right)$ of the activation processes of GPCRs and their derived complexes. This methodology combines the usage of targeted molecular dynamics (TMD) simulations^23,24 to generate the intermediate structures for activation processes from their initial and final structures (inactive and active states), with the protein-dipole Langevin-dipole (PDLD) method within the linear response approximation (LRA) framework (PDLD/S-LRA-2000)²⁵ and our refined coarse-grained $\left(\text{CG}\right)$ model^26,27 for GPCRs to calculate the binding and conformational energy contributions ($\text{Δ}{G}_{\text{conf}}$, $\text{Δ}{G}_{\text{bind}}$), respectively, which take into account the cellular membrane effects by an implicit membrane.²⁸ Given the abundance of GPCR structures of inactive and active forms thanks to the consolidation of cryo-electron microscopy (cryo-EM) as a characterization technique²⁹ and the tremendous progress in in-silico structure prediction, whose most prominent representative is AlphaFold 3,³⁰ we do not have a significant limitation on having structural information for our studies.

We have chosen Sphingosine 1-phosphate receptor 1 (S1PR1) as a case study to test our developed methodology given that there are available all necessary data to check it. Specifically, there are previous computational studies of this system that can be employed as a benchmark,²² its inactive and active states have been experimentally solved,^31,32 and there is available a repertoire of binding free energy experimental values $\left(\text{Δ}{G}_{\text{bind}}\right)$, between which it is found the S1P value that is the S1PR1 natural bioactive signaling molecule $\left(\text{Δ}{G}_{\text{bind}}^{\text{S1P}}\right)$.³³ Additionally, the corresponding structures of the 4 main states exhibited by the S1P:S1PR1 complex throughout its activation process have been provided (see Sections 2.1 and 3.1 for further details), which were obtained in a previous state-of-the-art computational study.²² Because the S1P–S1PR signaling system is an effective therapeutic target for multiple diseases, substantial drug discovery efforts have been devoted to developing distinct compounds with cross-subfamily and individual S1PR specificity.³⁴ Although numerous effective compounds have been characterized, as illustrated by the FDA-approved S1PR1 agonists fingolimod,³⁵ ozanimod,^36,37 and ponesimod^38,39 that treat multiple sclerosis, most are orthosteric ligands whose selectivity and side effects are determined to a large extent by the high sequence homology among S1PRs. This points out the allosteric modulation importance; however, there is limited knowledge about this GPCRs regulation for the S1PR family, especially being notorious the S1PR1 case. Thus, the obtained results will not only serve to develop the methodology in question, but could also fill this knowledge gap and be useful to improve existing treatments, or even develop new ones, for the S1PR1-associated pathologies, such as cancer,⁴⁰ multiple sclerosis,⁴¹ or inflammatory bowel disease.⁴² The aforementioned lack of knowledge made it crucial to apply methodologies that can predict prospective allosteric cavities. Considering our previous experience and questions of software compatibility, the CavityPlus web server^19,20 has been used to determine the S1PR1 cavities, while the KeyAlloSite computational method²¹ has been employed to predict their allosteric character. Although the employed methodology for cavity analysis has been proved to be robust, it was also applied to the Cannabinoid receptor 1 (CB1R), which was explored in our previous study.⁴³ Our main aim in that case was to determine the performance of the cavity search approach in GPCRs because the CB1R allosteric cavities are experimentally determined, and it is known what types of ligands bind to them. Additionally, the CB1R results turned out to be crucial for the rational design of a de novo S1PR1 allosteric modulator given the existing similarity between the CB1R allosteric cavities and the S1PR1 predicted ones.

From the point of view of computational resources, our methodology allows studying the GPCRs allosteric modulation in an efficient way given that the number of atoms in the GPCR system is significantly reduced by using an implicit membrane. Additionally, both binding and conformational energy calculations and molecular dynamics (MD) simulations to generate the equilibrated and relaxed GPCR complexes for those calculations are optimized to use minimum necessary resources and to continue capturing the system properties correctly. Specifically, the relaxation and equilibration MD simulations and binding energy calculations employ system regions carefully selected to consider only the GPCR part of interest, whereas conformational energy calculations use our refined $\text{CG}$ model. Indeed, the robustness and efficiency of the developed methodology are reflected in short simulation times that can be easily performed on conventional central processing units (CPUs), which is an advantage since specialized computers, such as graphics processing units (GPUs), are not always available.

2. METHODS

In Figure 2 appears a flowchart that summarizes the methodological framework used for determining the activity character of potential allosteric modulators. The methodologies that constitute this framework are developed in the following subsections.

Figure 2.

Flowchart of the methodological framework used for assessing potential allosteric modulators. Once the GPCR of interest has been selected (step 0), the allosteric cavity to be targeted and its potential allosteric modulators to be evaluated have to be chosen (step 1). In the case of no previous knowledge, CavityPlus, KeyAlloSite and LigBuilder can be used for this purpose. Then, intermediate conformations of the activation processes needed to characterize potential ligands (systems GPCR, Allo:GPCR, Ortho:GPCR and Ortho:Allo:GPCR) are generated by targeted MD simulations and docking calculations (step 2). Next, the equilibrated complexes of such conformations are obtained by relaxation and equilibration MD simulations (step 3). After, the activation free energy profiles are calculated using the PDLD/S-LRA-2000 method and our refined coarse-grained model to determine $\text{Δ}{G}_{\text{bind}}$ and $\text{Δ}{G}_{\text{conf}}$, respectively (step 4). At this point, activity character and properties (affinity and potency) of potential allosteric modulators can be assessed, which allow deciding whether these compounds are good enough or need further refining (step 5). Allo and ortho stand for allosteric and orthosteric modulator, respectively.

2.1. System Assembling and Setup.

Regarding cavity analysis, in the case of S1PR1, the provided structures corresponding to the 4 main states exhibited by the S1P:S1PR1 complex along its activation process, specifically, inactive state (I), transition state (TS), intermediate state (Int) and active state (A) that were obtained in a previous state-of-the-art computational study,²² were analyzed in order to determine how its cavities evolve throughout its activation. By contrast, in the case of CB1R, which was employed as a benchmark because its cavities are identified and characterized, our process and relaxed structure of the ZCZ011-(S):CP55940:CB1R complex was used (PDB code 7WV9).^43,44

The in-between conformations of different activation processes considered in this work were generated by targeted molecular dynamics (TMD) simulations^23,24 from the corresponding endpoint states of those processes. For the S1P:S1PR1 complex, its provided structures for the inactive and intermediate states were used. In contrast, for S1PR1, the structure corresponding to the fully inactive state of the ML056:S1PR1 complex (PDB Code 3V2Y)³² was employed as an initial point removing the antagonist present at the orthosteric site, whereas the active state of the previous complex was used again as a final point emptying its orthosteric site. This initial point change was in order not to bias the S1PR1 inactive state because S1P, due to its agonist character, induces by itself a destabilized inactive state which, in principle, is not reached in its absence. Similarly, these same endpoint states were considered for the C1:S1PR1 complex, though C1 was added inside the S1PR1 allosteric cavity for which was rationally designed (the S1PR1 PAM-equivalent cavity; see Section 3.3), determining its binding modes by docking calculations. However, the generation of the intermediate structures for the S1P:C1:S1PR1 complex was simpler since these were assembled from the corresponding ones of the S1P:S1PR1 complex adding C1 in the same way as in the previous case.

In relation to the selectivity of the considered compounds (see panel D of Figure 7) toward the S1PR1 cavities (see Figure 6), binary and ternary complexes must be distinguished. Binary complexes were generated combining the empty S1PR1 structures of the inactive and intermediate states corresponding to the ML056:S1PR1 and S1P:S1PR1 complexes, respectively, with different compounds located inside the orthosteric site and the S1PR1 PAM-equivalent cavity. For such compounds, only the most favorable binding modes inside these sites, which were determined by docking calculations, were considered. In contrast, ternary complexes were generated combining the provided structures of the S1P:S1PR1 complex for the inactive and intermediate states with different compounds whose binding modes in the considered S1PR1 allosteric cavity were determined similarly.

Figure 7.

Representation of the CB1R PAM and the S1PR1 PAM-equivalent cavities (Panels A and B, respectively) in which the cavity surface (in sheer purple), main residues that define its shape (in green), and principal pharmacophore features (hydrogen bond donor centers as blue spheres and hydrophobic centers as gray spheres) have been depicted. Additionally, a schematic comparison of these main residues (Panel C) has been included in order to show that both considered cavities exhibit the same degree of hydrophobicity. In this scheme, position 3.35 has been highlighted, whose residue performs a key role in anchoring the NO₂ group at the bottom of considered cavities. Finally, ZCZ011(S), the compound employed as a template, along with analogs with an increased size and hydrophobicity, which should favor better interaction, have been depicted (Panel D).

Figure 6.

Representation of the S1PR1 preserved cavities throughout its activation process. Along with the orthosteric (in orange) and G-protein (in navy blue) sites, two prospective allosteric cavities (S1PR1 PAM-equivalent in purple and NAM-equivalent cavities in green) were identified, which are equivalent to the CB1R homologous ones. Additionally, other three cavities without allosteric properties were determined, which have been identified by numbers (cavities 1, 2, and 3 in khaki, magenta, and dark green, respectively).

Regarding the general setup, it is worth noting that the provided S1PR1 structures had already assigned a protonation state for their amino acids, which was preserved in subsequent calculations. Ballesteros–Weinstein nomenclature⁴⁵ was employed with the aim of facilitating comparison with other GPCRs. For incorporating cellular membrane effects, once the corresponding complex was assembled, a 90 × 90 × 16 Å membrane particle grid was added around the Z-axis of the S1PR1 transmembrane bundle, whose center of mass shifted 1.6 Å throughout this axis is the origin of that grid. As for considered ligands, S1P was only contemplated as an orthosteric ligand because its binding site is well-known, while ZCZ011(S) and candidates 1–3 (see panel D of Figure 7) were inspected as both orthosteric and allosteric ligands in order to check their selectivity toward the S1PR1 cavities. The partial charges of all previous compounds were calculated at the CAM-B3LYP/6–31G* level of theory using Gaussian 16.⁴⁶ Those resulting charges were fitted to generate the restrained electrostatic potential (RESP)-fitted charges⁴⁷ by Ambertools.⁴⁸ However, the remaining system was described using the ENZYMIX force field.²⁵ From now on, all calculations were carried out using the Molaris-XG package⁴⁹ unless otherwise mentioned. All visualizations and representations have been made with the VMD⁵⁰ and UCSF Chimera⁵¹ programs.

2.2. Cavity Analysis.

The CavityPlus web server^19,20 was used to determine the S1PR1 and CB1R cavities employing the specific setting for shallow cavities, that is, the parameters SEPARATE_MIN_DEPTH, MIN_ABSTRACT_LIMIT, SEPARATE_MAX_LIMIT, and MIN_ABSTRACT_DEPTH were set to values of 4, 187.5 ų, 750 ų, and 4, respectively. Additionally, the option “Apply soft separation” was used in order to recover fragments of big cavities that were considered as independent small cavities erroneously.

Regarding allosterism, the KeyAlloSite computational method²¹ was used to predict allosteric sites. This method is based on the existing evolutionary coupling between the orthosteric and allosteric sites. For both considered receptors, their homologous sequences were determined applying the BLAST algorithm⁵² over the curated UniProtKB/Swiss-Prot database⁵³ with a sequence identity threshold of 80%. Additionally, in this search the Auto-BLOSUM62 matrix⁵⁴ was used, and the number of hits was increased to 1000. Regarding evolutionary coupling calculations, the previously-calculated cavities by CavityPlus were used to assign which amino acids constitute them, and the orthosteric and G-protein sites were assigned as orthosteric pockets since both cavities are where natural GPCRs ligands bind. Only the S1P:S1PR1 cavities for the intermediate state were considered for these calculations because this state is the predominant one (see Section 3.1).

Finally, CavPharmer,^55,56 a module inside the CavityPlus program, was employed to build a pharmacophore model for some of the identified allosteric cavities, specifically, for the S1PR1 PAM-equivalent cavity and its CB1R counterpart (the CB1R PAM cavity).

2.3. Docking Calculations.

The docking calculations of the considered compounds inside the S1PR1 orthosteric site and the S1PR1 PAM-equivalent cavity were carried out by a two-step process: (1) Determination of main binding modes restraining the center of mass of the corresponding compound to be at a distance of 4.0 Å from the C_α atom of Leu275^6.54 when the former cavity is targeted, because this residue is placed at the center of given cavity, or the oxygen backbone atom of Leu119^3.27 when the latter cavity is targeted, since according to our previous knowledge the considered compound must be attached to this atom forming a hydrogen bond with its NH group; (2) Optimization of the resulting main binding modes performing 1000 Monte Carlo energy minimizations. The receptor remained rigid, but total flexibility was given to allosteric modulators during the conformational exploration.

2.4. Targeted Molecular Dynamics Simulations.

TMD simulations^23,24 were employed to generate intermediate conformations for the activation processes of the systems S1P:S1PR1, C1:S1PR1, and S1PR1. These systems were dragged from their initial states into their final ones through 251 successive mapping relaxation simulations of 50,000 steps applying a dragging harmonic force over all heavy atoms with a force constant of 100 kcal mol⁻¹ Å⁻². The temperature and step size considered for those simulations were 300 K and 0.1 fs, respectively.

2.5. Relaxation and Equilibration Molecular Dynamics Simulations.

All relaxation and equilibration MD simulations followed the same protocol regardless of whether they were applied to the intermediate conformations of different activation processes inspected or to binary and ternary complexes employed to determine the selectivity of considered compounds toward the S1PR1 cavities. The only existing differences were the starting structure and the unconstrained system region to be relaxed and equilibrated. The starting structures have been taken as the assembled complexes mentioned in Section 2.1. The center of this unconstrained region was conveniently chosen as an atom of the ligand (S1P or allosteric modulator) whose binding free energy $\left(\text{Δ}{G}_{\text{bind}}\right)$ was assessed subsequently. In particular, this center was defined as the substituted indole/pyrrol ${\text{C}}_{\beta }$ of the allosteric modulator and the central carbon atom of the S1P hydrophobic tail when allosteric modulators and S1P $\text{Δ}{G}_{\text{bind}}$ were calculated, respectively. This system-center choice allows considering orthosteric and allosteric sites simultaneously for ternary complexes without employing an excessively large unconstrained region of the system. For binary complexes, the unconstrained region consists in a 20 Å radius sphere around the chosen center, which was solvated with a surface-constrained all-atoms solvent (SCAAS) model⁵⁷ water sphere of equal radius and center. In order to consider both binding sites simultaneously, this radius was enlarged to 30 Å for ternary complexes. Those resulting spheres were surrounded by a 2 Å spherical shell of Langevin dipoles and then a bulk continuum. Water molecules were restrained to be at least at 4 Å of membrane particles to control their penetration into the implicit cellular membrane (see Figure 3). The local reaction field (LRF) method⁵⁸ was employed to treat the long-range electrostatic interactions, though those coming from the outside of the unconstrained region were not included in the energy calculations. Once the system was assembled and solvated, it was submitted to 2000 energy minimization steps using the steep-descent method to avoid close contacts. Then, the system was gradually heated from 0 to 300 K for 200 ps. Throughout this heating, harmonic restraints were applied to the protein backbone and ligand/water heavy atoms with a force constant of 10 kcal mol⁻¹ Å⁻² that were stepwise released during the first 100 ps of this period. It is worth noting that in the case of intermediate conformations, those restraints applied to the protein backbone atoms were loosened and kept at 5 kcal mol⁻¹ Å⁻² to preserve the corresponding conformation. Next, a 200 ps MD equilibration simulation was calculated to generate starting geometries for subsequent calculations. Specifically, the last geometry was employed for binding energy calculations, whereas 10 different geometries uniformly extracted throughout the last 100 ps were used for conformational energy calculations. It is worth mentioning that, taking into account how the S1P:C1:S1PR1 activation process is calculated, only the last geometry of this MD simulation was needed since conformational and S1P binding energies come from the initial S1P:S1PR1 activation process. Finally, the equilibration of such systems was considerably fast thanks to the usage of spherical boundary conditions, ensuring that the system energy is converged.

Figure 3.

Schematic representation of the S1P:C1:S1PR1 ternary complex (the orthosteric and allosteric ligands have been indicated by CPK depiction in green and purple, respectively) embedded into the membrane and its water sphere solvation. The structural formula of the considered ligand has been included, indicating with an asterisk their atoms that can be considered as the system center in subsequent calculations. As a ternary complex, the unconstrained region consists in a 30 Å radius sphere, which is centered around the substituted pyrrol ${\text{C}}_{\beta }$ of C1 since it has been considered that $\text{Δ}{G}_{\text{bind}}$ is being calculated for this compound, which constitutes Region 1 (black circle). The remaining unconstrained region defines Region 2 (dark green circle). The boundary of the water sphere, which corresponds to the Region 2 end, is surrounded by a 2 Å spherical shell of Langevin dipoles (blue circle) and then a bulk continuum.

2.6. PDLD/S-LRA Simulations.

The protein-dipole Langevin-dipole (PDLD) method within the linear response approximation (LRA) framework (PDLD/S-LRA-2000)²⁵ was employed to calculate the binding free energy $\left(\text{Δ}{G}_{\text{bind}}\right)$ for ligands bound to S1PR1. This approach evaluates the non-electrostatic term explicitly by a special cycle, while the electrostatic term is evaluated by the LRA approach. Therefore, $\text{Δ}{G}_{\text{bind}}$ is given the following expression:

$$\text{Δ}{G}_{\text{bind}}=\text{Δ}{G}_{\text{bind}}^{\text{elec}}+C\text{Δ}{G}_{\text{bind}}^{\text{non}-\text{elec}}$$ (1)

Where $\text{Δ}{G}_{\text{bind}}$ refers to the total binding energy of ligand, $\text{Δ}{G}_{\text{bind}}^{\text{elec}}$ and $\text{Δ}{G}_{\text{bind}}^{\text{non}-\text{elec}}$ correspond to the electrostatic and non-electrostatic contributions of $\text{Δ}{G}_{\text{bind}}$, respectively, and $C$ is the scaling factor of this last contribution. The scaling factor $C$ is not a universal constant but is specific to the studied system since it is influenced by the interplay between different thermodynamic contributions such as van der Waals interactions, hydrophobic effects, water penetration, and configurational entropy changes.⁵⁹ Therefore, this factor is cavity-dependent. For the current work, an optimal $C$ value of 0.5 has been determined for the S1PR1 orthosteric site, which allows reproducing the experimental value of the S1P binding free energy $\left(\text{Δ}{G}_{\text{bind}}^{\text{S}1\text{P}}\right)$ in this cavity. In contrast, due to the lack of experimental $\text{Δ}{G}_{\text{bind}}$ for the S1PR1 PAM-equivalent cavity, the optimal $C$ value of 0.9 obtained for the CB1R homologous cavity⁴³ regarding the $\text{Δ}{G}_{\text{bind}}$ reproduction for a diverse series of CB1R PAMs belonging to the 2-phenylindole structural class was assigned to the S1PR1 PAM-equivalent cavity because both exhibit similar pharmacophore features (see Section 3.3).

The simulation setup for $\text{Δ}{G}_{\text{bind}}$ calculations was the same as that for relaxation and equilibration MD simulations, with the exception that in the PDLD/S-LRA simulations, the unconstrainted system region was divided into two regions. Region 1, for which $\text{Δ}{G}_{\text{bind}}$ was calculated, was defined by the whole S1P or allosteric modulator, while region 2, which comprises the system region that contributes to $\text{Δ}{G}_{\text{bind}}$, was defined by the remaining surrounding system. For each previously equilibrated complex, a PDLD/S-LRA simulation was carried out on four different replicas whose distinct initial configurations were obtained through four successive equilibration MD simulations of 10 ps. For binding free energy in protein, 50,000 different configurations were considered, while for the reference in water 5000 different configurations were assessed. It is important to note that the employed length for the PDLD/S-LRA simulations is more than enough to achieve a converged $\text{Δ}{G}_{\text{bind}}$ according to previous benchmarks.⁶⁰

2.7. Coarse-Grained Model.

Our refined $\text{CG}$ model^27,28 has been employed with the aim of determining the S1PR1 conformational free energy $\left(\text{Δ}{G}_{\text{conf}}\right)$. The total energy of our $\text{CG}$ model is given by

$$\begin{gathered} \text{Δ}{G}_{\text{conf}}=\text{Δ}{G}_{\text{main}}+\text{Δ}{G}_{\text{side}}+\text{Δ}{G}_{\text{main}-\text{side}} \\ =c_1\text{Δ}{G}_{\text{side}}^{\text{VdW}}+c_2\text{Δ}{G}_{\text{solv}}^{\text{CG}}+c_3\text{Δ}{G}_{\text{HB}}^{\text{CG}}+c_4\text{Δ}{G}_{\text{mem}}^{\text{hyd}} \\ +\text{Δ}{G}_{\text{side}}^{\text{hyd}}+\text{Δ}{G}_{\text{main}-\text{side}}^{\text{VdW}}+\text{Δ}{G}_{\text{side}}^{\text{elec}}+\text{Δ}{G}_{\text{side}}^{\text{polar}} \end{gathered}$$ (2)

Here, the total conformational free energy $\text{Δ}{G}_{\text{conf}}$ consists of three terms, the main chain term $\text{Δ}{G}_{\text{main}}$, the side chain term $\text{Δ}{G}_{\text{side}}$, and the interaction term between main and side chain $\text{Δ}{G}_{\text{main}-\text{side}}$, which can be broken down as specified in the second raw on eq 2. The terms on the right of this raw are the side chain van der Waals energy, the main chain solvation energy, the main chain hydrogen bond energy, the cellular membrane hydrophobic energy, the side chain hydrophobic energy, the main chain/side chain van der Waals energy, the side chain electrostatic energy, and the side chain polar energy, respectively. The scaling coefficients $c_1$, $c_2$, $c_3$_, and $c_4$ take values of 0.10, 0.25, 0.15, and 1/3.6, respectively. It is worth noting that eq 2 has successfully been employed to determine $\text{Δ}{G}_{\text{conf}}$ in other GPCRs.^61,62 As stated in Section 2.5, in order to evaluate $\text{Δ}{G}_{\text{conf}}$ in a reliable way for a given equilibrated complex, 10 different geometries uniformly extracted throughout the last 100 ps of its equilibration MD simulation were assessed. These geometries were trimmed into the $\text{CG}$ representation, in which the main chain of amino acids was still in all-atom form, but each of their side chains were reduced into a simplified united atom. Next, prior to evaluating their $\text{Δ}{G}_{\text{conf}}$, the resulting structures were submitted to a small relaxation in the gas phase and the present ligands were removed.

3. RESULTS AND DISCUSSION

3.1. Benchmarking Process of GPCRs Conformational Changes.

In order to test whether our methodology is adequate to describe the energetics and to locate the main states of the GPCRs conformational changes, the activation process of the S1P:S1PR1 complex was tried to be reproduced as a benchmark, since this process has been studied using state-of-the-art computational methodologies²² and, in addition, the experimental binding free energy value of S1P in this receptor $\left(\text{Δ}{G}_{\text{bind}}^{\text{S}1\text{P}}\right)$ is available.³³

For a given GPCR, its total free energy $\left(\text{Δ}{G}_{\text{total}}\right)$ will be given by the sum of its conformational contribution $\left(\text{Δ}{G}_{\text{conf}}\right)$ with the binding contribution for each of the present ligands $\left(\text{Δ}{G}_{\text{bind},i}\right)$:

$$\text{Δ}{G}_{\text{total}}=\text{Δ}{G}_{\text{conf}}+\sum _{i=1}^n\text{Δ}{G}_{\text{bind},i}$$ (3)

In our methodology, the conformational contribution is determined using our refined coarse-grained $\left(\text{CG}\right)$ model for GPCRs, while the binding contribution for each of the present ligands is calculated using the PDLD/S-LRA-2000 method appropriately fine-tuned to the cavity in which they are located. In both contributions, the cell membrane effects were included using an implicit membrane. On the other hand, with the aim of building the $\text{Δ}{G}_{\text{total}}$ profile, the in-between structures of a given GPCR conformational change have been generated employing a TMD simulation, whose end points corresponded to the beginning and end of the considered conformational change, respectively (see Section 2.1 for further details).

Regarding the benchmarking process, we initially examined whether the experimental value of $\text{Δ}{G}_{\text{bind}}^{\text{S1P}}$ can be reproduced. This was done in order to verify that $\text{Δ}{G}_{\text{bind}}$ is properly described, for testing subsequently that both the energetics and the main states of the activation process of the S1P:S1PR1 complex are characterized correctly by the calculation of its $\text{Δ}{G}_{\text{total}}$ profile, proving indirectly that $\text{Δ}{G}_{\text{conf}}$ is also well-described.

For the sake of favoring the comprehension of the benchmark carried out, the main characteristics of the considered activation process in terms of energy and exhibited states are briefly presented. According to the previously-obtained results,²² the S1P:S1PR1 complex exhibits four main states throughout its activation process, specifically, inactive state (I), transition state (TS), intermediate state (Int), and active state (A), whose structures have been provided. The inactive and intermediate states correspond to energy minima, which are connected by the transition state that involves a free energy barrier $\left(\text{Δ}{G}^{\ast }\right)$ of 6.60 kcal/mol, which corresponds to the activation process barrier $\left(\text{Δ}{G}^{‡}\right)$. Once the transition state is overcome, the system passes through a series of local minima, separated by significantly small barriers and with energies that monotonously decrease very softly, until the intermediate state is reached. It is worth noting that the intermediate state resembles the active state in almost all its structural features, with the outward displacement of the transmembrane helix 6 (TM6) being the most prominent one. However, the situation is completely opposite from an energetic point of view since the intermediate state corresponds to the global minimum of the activation process of the S1P:S1PR1 complex, being 11.78 kcal/mol below the inactive state; that is, the overall free energy change of the activation process $\left(\text{Δ}{G}_{\text{act}}\right)$ is −11.78 kcal/mol. By contrast, the active state does not correspond to an energy minimum because it exhibits a remarkable instability due to the absence of the corresponding G protein (see panel A of Figure 4 as an example). Therefore, the only significantly populated state of the S1P:S1PR1 complex will be the intermediate state, determining the experimentally-observed $\text{Δ}{G}_{\text{bind}}^{\text{S1P}}$.

Figure 4.

$\text{Δ}{G}_{\text{total}}$ profile obtained for the activation process of the S1P:S1PR1 complex (Panel A) together with the representation of main conformational changes induced by the S1P hydrophobic tail that trigger the S1PR1 activation process (Panel B). Regarding panel A, the position of main states identified throughout the S1P:S1PR1 activation process has been indicated. Error bars indicate the $\text{Δ}{G}_{\text{total}}$ uncertainty, which has been calculated propagating the standard error of its contributions adequately. Additionally, the $\text{Δ}{G}^{‡}$ and $\text{Δ}G$ values have been specified. Regarding panel B, an overlap of the inactive (in green with S1P in dark green) and intermediate (in blue with S1P in purple) states has been depicted.

Regarding the first step of the benchmark, Table 1 contains $\text{Δ}{G}_{\text{bind}}^{\text{S1P}}$ together with its electrostatic and nonelectrostatic contributions for the inactive and intermediate states (global energy minima on both sides of the transition state). It should be mentioned that $\text{Δ}{G}_{\text{bind}}$ has no associated uncertainty because a single calculation has been performed per equilibrated complex. As can be noticed, the intermediate state allows satisfactorily reproducing the experimental value of $\text{Δ}{G}_{\text{bind}}^{\text{S1P}}$ (−10.10 kcal/mol vs −10.14 kcal/mol), with the electrostatic contribution being substantially more intense (−8.22 kcal/mol vs −1.88 kcal/mol), which is in accordance with the strong fixation of the S1P polar head in S1PR1. On the other hand, the obtained results for the inactive state show that $\text{Δ}{G}_{\text{bind}}^{\text{S1P}}$ now differs significantly from the experimental value, being substantially smaller. However, considering $\text{Δ}{G}^{‡}$ of the activation process, $\text{Δ}{G}_{\text{bind}}^{\text{S1P}}$ seems to be one of the main contributions to the activation process driving force because it provides almost completely the energy needed to reach the transition state (4.99 kcal/mol vs 6.60 kcal/mol). All this shows that our methodology is adequate to determine the binding contribution of the complexes derived from GPCRs, which is in line with the previously-obtained CB1R results⁴³ given that these demonstrated that the considered methodology is able to reproduce dissociation constants (linked to $\text{Δ}{G}_{\text{bind}}$) and cooperativity factors (linked to $\text{Δ}{G}_{\text{bind}}$ difference in the presence and absence of orthosteric modulator) for a diverse assortment of CB1R PAMs belonging to the 2-phenylindole structural class considering CP55940 as the orthosteric agonist.

Table 1.

$\text{Δ}{G}_{\text{bind}}^{\text{S1P}}$ Along with its Electrostatic and Non-electrostatic Contributions for the Global Energy Minimums at Both Sides of the Transition State Identified in the Activation Process of the S1P:S1PR1 Complex

state	$\text{Δ}{G}_{\text{bind}}^{\text{elec}}\left(\text{kcal}/\text{mol}\right)$	$\text{Δ}{G}_{\text{bind}}^{\text{non}-\text{elec}}\left(\text{kcal}/\text{mol}\right)$	$\text{Δ}{G}_{\text{bind}}^{\text{S1P}}\left(\text{kcal}/\text{mol}\right)$
Inactive	−3.67	−1.32	−4.99
Intermediate	−8.22	−1.88	−10.10

In relation to the second step of the benchmark, the $\text{Δ}{G}_{\text{total}}$ profile obtained for the activation process of the S1P:S1PR1 complex (see panel A of Figure 4) shows that it has been possible to reproduce both its main exhibited states and the energetics thereof. In our case, four energy minima have been obtained, corresponding the first and last ones to the inactive and intermediate states, respectively, due to their location within the activation process, which is corroborated by analyzing these two minima from a structural point of view. Specifically, throughout the transition from the inactive state to the intermediate one, apart from the TM6 outward movement, it can be observed how the S1P hydrophobic tail induces the rotation of the Trp269^6.48, Phe265^6.44, and Phe210^5.47 side chains. This conformational change is known to destabilize the interactions between TM3 and TM6 and trigger the S1PR1 activation process (see panel B of Figure 4). This result emphasizes the potential of our methodology to capture the key interactions that provoke the GPCRs activation, being able to be used to both study the GPCRs activation process in a comprehensive way and rationally design orthosteric ligands with improved properties, such as better potency or selectivity. However, these issues have not been explored in detail because they are outside the scope of the present work. On the other hand, these states are connected by several maximum energy structures being the first one the transition state, and once the intermediate state is overcome (which corresponds to the global energy minimum again), the energy begins to increase in a continuous way, being congruent with the apparent instability of the active state due to the G-protein absence. It is important to note that it has not been possible to reproduce the series of local minima observed once the transition state is overcome, since only two minima separated by relatively small barriers have been obtained. According to their position inside the activation process, these minima correspond to a partially activated inactive state (I*) and a partially inactivated intermediate state (Int*), respectively. However, this limitation describing the interconversion between local minimums with very similar energies does not pose a problem since the global minimum is correctly identified, and this determines the system’s observable magnitudes. In relation to the process energetics, the calculated values for $\text{Δ}{G}^{‡}$ and $\text{Δ}{G}_{\text{act}}$ are compatible with the previously-obtained values (7.78 ± 1.10 vs 6.60 kcal/mol and −11.94 ± 0.96 vs −11.78 kcal/mol, respectively). These results show that the activation process of the S1P:S1PR1 complex has been satisfactorily reproduced in terms of both the exhibited states and energy, proving that our methodology is adequate to describe GPCRs conformational changes.

3.2. Benchmarking Process of Cavity Analysis.

In order to test the performance of the considered methodology for locating allosteric cavities with capacity to be pharmacological targets in GPCRs, we took as a benchmark the CB1R system. This GPCR has its allosteric cavities experimentally characterized and a wide variety of its ligands are known. Additionally, we already have experience with this receptor.

Using the relaxed structure of the ZCZ011-(S):CP55940:CB1R complex (PDB code 7WV9),^43,44 19 different cavities were obtained, identifying the CB1R main cavities correctly (see Figure 5). Specifically, the orthosteric and G-protein sites, found in any GPCR, along with its two experimentally-determined allosteric cavities placed at the surface of TM2, TM3, and TM4, one associated with positive allosteric modulation (PAM cavity), while the other is linked to negative allosteric modulation (NAM cavity). According to this analysis (see Table 2), all these cavities have high dissociation constants $\left(K_{\text{d}}\right)$ and adequate druggabilities (protein pocket ability to bind drug-like molecules with high affinity), indicating that they are really binding sites, as known experimentally. In relation to allosterism, the PAM and NAM cavities were identified as allosteric (Z-score > 0.8), in addition to a cavity formed by the TM4 and TM5 central part that would be located inside the cell membrane (membrane allosteric cavity). On the other hand, the binding and druggability properties of this predicted allosteric cavity are also appropriate to be exploited pharmacologically, suggesting that, in principle, it could be used in the development of new CB1R allosteric modulators. These results suggest that the proposed methodology based on the identification of cavities by CavityPlus^19,20 together with its subsequent screening in terms of allosteric properties employing KeyAlloSite²¹ is effective in determining allosteric cavities that can be exploited pharmacologically.

Figure 5.

Representation of the CB1R main cavities (orthosteric site in orange, G-protein site in navy blue, PAM cavity in purple, and NAM cavity in green) along with the additional predicted allosteric cavity (membrane allosteric cavity in dark green). The TMs that constitute the allosteric cavities, both experimental and predicted, have been indicated.

Table 2.

Examination of Binding, Druggability, Size, and Allosterism for CB1R Orthosteric and G-protein Sites Along with its Experimentally-determined and Predicted Allosteric Cavities

cavity	$\text{pred}.\text{max}.\text{p}{K}_{\text{d}}$	drugscore	druggability	surface area (Å2)	volume (Å3)	Z-score
Orthosteric Site	11.06	4322.00	Strong	1159.25	1108.62	n/a
G-protein Site	10.15	444.00	Medium	772.75	1104.25	n/a
PAM Cavity	9.82	177.00	Medium	416.50	470.88	2.42
NAM Cavity	10.69	825.00	Strong	399.75	553.12	1.31
Membrane Allosteric Cavity	8.32	651.00	Strong	309.75	298.00	1.17

3.3. S1PR1 Cavity Analysis and Rational Design of an Allosteric Modulator.

Applying the same analysis methodology to the provided structures associated with the main states identified in the activation process of the S1P:S1PR1 complex,²² it was determined that a series of cavities were preserved in a state-independent way (see Figure 6 and Tables S1–S5). Among these, there were obviously the orthosteric and G-protein sites of S1PR1, which again have high $K_{\text{d}}$ and adequate druggabilities, identifying them as binding sites. However, cavities which are spatially equivalent to the CB1R allosteric ones were also identified, which are going to be referred as S1PR1 PAM-equivalent cavity and S1PR1 NAM-equivalent cavity according to their exhibited spatial equivalence. This equivalence also extends to the parameters of binding, druggability, and size (compare Tables 3 and 4 with rows 3 and 4 of Table 2, respectively), although the S1PR1 PAM-equivalent cavity exhibits a negative deviation in terms of binding and druggability for the inactive, transition, and active states. Furthermore, regarding allosterism, only these S1PR1 spatially equivalent cavities were identified as allosteric. For all this, it can be concluded that the S1PR1 PAM-equivalent and NAM-equivalent cavities are the only allosteric cavities of this receptor with binding and druggability properties suitable for pharmacological exploitation.

Table 3.

Examination of Binding, Druggability, Size, and Allosterism for the S1PR1 Cavity Equivalent to the PAM CB1R Cavity in the Main States of the S1P:S1PR1 Complex

state	$\text{pred}.\text{max}.\text{p}{K}_{\text{d}}$	drugscore	druggability	surface area (Å2)	volume (Å3)	Z-scorea
Inactive	8.15	−138.00	Weak	390.00	386.00
Transition State	8.61	−75.00	Weak	384.75	389.00
Intermediate	10.07	669.00	Strong	566.50	647.25	0.81
Active	8.55	−79.00	Weak	375.75	413.62

^aZ-score has only been calculated for the intermediate state of the activation process of the S1P:S1PR1 complex since this state is the only one significantly populated and this parameter should not depend on the considered state.

Table 4.

Examination of Binding, Druggability, Size and Allosterism for the S1PR1 Cavity Equivalent to the NAM CB1R Cavity in the Main States of the S1P:S1PR1 Complex

state	$\text{pred}.\text{max}.\text{p}{K}_{\text{d}}$	drugscore	druggability	surface area (Å2)	volume (Å3)	Z-scorea
Inactive	10.55	509.00	Medium	378.50	549.62
Transition State	10.51	701.00	Strong	417.00	609.00
Intermediate	10.00	735.00	Strong	420.25	624.62	2.13
Active	11.11	738.00	Strong	408.25	605.88

^aZ-score has only been calculated for the intermediate state of the activation process of the S1P:S1PR1 complex since this state is the only one significantly populated and this parameter should not depend on the considered state.

In relation to the rational design of a S1PR1 allosteric modulator, it can be observed that both predicted allosteric cavities expand throughout the activation process, reaching their maximum expansion at the intermediate state (see columns 5 and 6 of Tables 3 and 4 and Figure S1). This result, along with the fact that the S1PR1 PAM-equivalent cavity has much better binding and druggability properties for the intermediate state, can be used to develop a selective ligand for this cavity and state that favors the activation process, that is, to develop an S1PR1 positive allosteric modulator. It is worth noting that, apart from the aforementioned resemblance in terms of spatial position inside GPCR, binding, druggability and size, this S1PR1 allosteric cavity bears a reasonable resemblance to its CB1R counterpart in terms of topology and pharmacophore properties. Specifically, these cavities exhibit a similar shape presenting an accessible hydrogen bond donor center, which would serve as an anchor of the allosteric modulator, surrounded by four hydrophobic centers, whose relative positions are largely preserved for three of them placed at the bottom of the cavity (see Panels A and B of Figure 7). Therefore, an allosteric modulator of the CB1R PAM cavity could be used as a template to rationally design a positive allosteric modulator for the S1PR1 PAM-equivalent cavity.

Particularly, ZCZ011(S) was employed for this purpose since it is the only allosteric modulator of the CB1R PAM cavity for which its binding mode is experimentally resolved.⁴⁴ This compound was modified so that its analogues (see Panel D of Figure 7) adapt to the following characteristics of the S1PR1 PAM-equivalent cavity in relation to its CB1R counterpart: (1) Larger size (see Panels A and B of Figure 7 and compare third rows of Tables 3 and 4); (2) Equal degree of hydrophobicity (see Panel C of Figure 7); (3) Absence of a serine at the bottom end of the cavity (position 3.35; see Panels A, B and C of Figure 7). As can be noticed, these analogues have an increased size and hydrophobicity, which should favor better interaction with both the considered cavity and the cell membrane because this cavity and therefore its ligands are partially introduced into it. It is worth noting that the thiophene ring replacement for a ring with greater hydrophobicity (benzene or naphthalene) is motivated by the absence of the aforementioned serine. According to our previous studies about the binding modes of this type of compounds in the CB1R PAM cavity,⁴³ Ser199^3.35 is key for the nitro group to be able to be accommodated in the highly hydrophobic bottom end of this cavity which is located within the cell membrane. Therefore, in the case of the S1PR1 PAM-equivalent cavity in which this serine is missing, this region of the cavity will be occupied by the substitute ring, whose greater hydrophobicity will lead to a better interaction with its surrounding area. Finally, in the case of lack of knowledge about what type of ligands can target the identified allosteric cavity, different computational tools can be employed to find suitable candidates. Taking into account compatibility with the software used to perform cavity analysis, the CavityMatch module,²⁰ belonging to CavityPlus, and the LigBuilder program⁶³ are the best alternatives. On the one hand, CavityMatch together with its CavSim tool allows obtaining similar cavities to the considered one in pharmacophoric terms, whose ligands, if they exist and are known, could be used as a template. On the other hand, LigBuilder allows the design of de novo multitarget drugs, which can be easily applied to the determined cavities with CavityPlus thanks to their format compatibility.

3.4. Selectivity of ZCZ011(S) Analogues toward the S1PR1 PAM-Equivalent Cavity for the Intermediate State.

In order to determine whether the designed ZCZ011-(S) analogues exhibit the desired selectivity toward S1PR1, $\text{Δ}{G}_{\text{bind}}$ of this compound and its derived analogues were calculated in the orthosteric site and in the S1PR1 PAM-equivalent cavity for the inactive and intermediate states, taking into account for the latter cavity the effect of the S1P presence.

Given that as a rule of thumb the orthosteric site tends to be the GPCR cavity with the best binding and druggability properties, such as the performed S1PR1 cavity analyses suggest (see Tables 3 and 4 along with Tables S1–S5), the $\text{Δ}{G}_{\text{bind}}$ comparison of these compounds in the aforementioned cavities and states allows determining whether they possess the intended selectivity. It is important to note that, similarly to the CB1R homologous cavity, this type of compound exhibits different binding modes in the S1PR1 PAM-equivalent cavity. Specifically, 4 different poses that result from flipping the groups that are around the C–C bond formed by the carbon atoms in $\beta$ position of the NO₂ group and the pyrrole or 1H-indole ring depending on the compound considered (see Figure 8). Then, with the aim of making this comparison correctly, $\text{Δ}{G}_{\text{bind}}$ of these compounds has been calculated in their most favorable binding modes identified by docking calculations for the considered cavities and states. Tables 5–8 contain these $\text{Δ}{G}_{\text{bind}}$s for ZCZ011(S) and candidates 1, 2, and 3 (C1, C2 and C3), respectively.

Figure 8.

Representation of the four different binding modes that the considered compounds can adopt inside the S1PR1 PAM-equivalent cavity (Panels A, B, C, and D correspond to binding modes 1, 2, 3, and 4, respectively), which have been illustrated by C1. All binding modes exhibit a hydrogen bond between the NH group of the pyrrole or 1H-indole ring and the oxygen backbone atom of Leu119^3.27. It is worth noting that binding modes 2 and 4 are destabilized, regarding those for the CB1R homologous cavity, because no residue at the bottom of the considered S1PR1 cavity can attach the NO₂ group.

Table 5.

ZCZ011(S) $\text{Δ}{G}_{\text{bind}}$ for its Most Favorable Binding Modes Inside the Orthosteric Site and the S1PR1 PAM-Equivalent Cavity for the Inactive and Intermediate States^a

cavity	state	pose	presence of S1P	$\text{Δ}{G}_{\text{bind}}^{\text{elec}}\left(\text{kcal}/\text{mol}\right)$	$\text{Δ}{G}_{\text{bind}}^{\text{non}-\text{elec}}\left(\text{kcal}/\text{mol}\right)$	$\text{Δ}{G}_{\text{bind}}\left(\text{kcal}/\text{mol}\right)$
Ortho	Inactive	n/a	n/a	2.43	−2.77	−0.34
Ortho	Intermediate	n/a	n/a	3.00	−3.47	−0.47
Allo	Inactive	1	No
Allo	Intermediate	1	No	0.91	−3.14	−2.23
Allo	Intermediate	3	No	1.27	−2.39	−1.12
Allo	Inactive	1	Yes	0.91	−4.51	−3.57
Allo	Inactive	4	Yes	2.02	−3.72	−1.70
Allo	Intermediate	1	Yes	0.95	−3.99	−3.04
Allo	Intermediate	2	Yes	1.72	−3.02	−1.30
Allo	Intermediate	3	Yes	0.71	−4.23	−3.52
Allo	Intermediate	4	Yes	1.84	−3.25	−1.41

^aIn the case of columns of $\text{Δ}{G}_{\text{bind}}$ and its contributions (columns 5 to 7), the hyphen symbol indicates that the considered compound for the specified state is instable and unbinds during the equilibration and relaxation MDs. In the first column, ortho and allo stands for orthosteric and allosteric, respectively.

Table 8.

Candidate 3 $\text{Δ}{G}_{\text{bind}}$ for its Most Favorable Binding Modes Inside the Orthosteric Site and the S1PR1 PAM-Equivalent Cavity for the Inactive and Intermediate States^a

cavity	state	pose	presence of S1P	$\text{Δ}{G}_{\text{bind}}^{\text{elec}}\left(\text{kcal}/\text{mol}\right)$	$\text{Δ}{G}_{\text{bind}}^{\text{non}-\text{elec}}\left(\text{kcal}/\text{mol}\right)$	$\text{Δ}{G}_{\text{bind}}\left(\text{kcal}/\text{mol}\right)$
Ortho	Inactive	n/a	n/a	−1.16	0.42	−0.74
Ortho	Intermediate	n/a	n/a	−1.27	−1.34	−2.60
Allo	Inactive	1	No
Allo	Intermediate	1	No
Allo	Inactive	1	Yes
Allo	Inactive	3	Yes
Allo	Intermediate	1	Yes

^aIn the case of columns of $\text{Δ}{G}_{\text{bind}}$ and its contributions (columns 5 to 7), the hyphen symbol indicates that the considered compound for the specified state is instable and unbinds during the equilibration and relaxation MDs. In the first column, ortho and allo stand for orthosteric and allosteric, respectively.

According to the reasoning of the previous section, ZCZ011(S) would have to exhibit a preference for the S1PR1 PAM-equivalent cavity regardless of the considered state, although its binding in both cavities should be weak since it is not conceived for targeting either of them. Our calculations point in this direction; however, the ZCZ011(S) binding is quite poor in both cavities, which may have caused this compound to leave the allosteric cavity for the inactive state. On the other hand, it can be noticed how the S1P presence inside the orthosteric side improves the ZCZ011(S) binding inside the S1PR1 PAM-equivalent cavity independent of the considered state, which would be in accordance with the postulated modulation between both cavities. According to the law of microscopic reversibility of thermodynamics,⁶⁴ this binding improvement implies positive modulation, as had already been experimentally determined for the CB1R homologous cavity. As for the binding modes, as predicted, the absence of serine at the bottom of the S1PR1 PAM-equivalent cavity destabilizes those in which the NO₂ group is placed in this region (compare poses 2 and 4, in which the NO₂ group is found in this location, with poses 1 and 3, in which this group is not there).

As for candidates 2 and 3, their design has been unsuccessful because none exhibits the desired selectivity. Specifically, candidate 2 (C2) exhibits slightly stable binding modes in both cavities for the considered states, although its effective binding in the allosteric cavity requires the S1P presence, but it does show no selectivity for any cavity. In the case of candidate 3 (C3), its selectivity is opposite to the desired one because this compound is capable of binding weakly in the orthosteric site for both states, but it is unable to bind effectively in the allosteric cavity.

In contrast, candidate 1 (C1) does exhibit the desired selectivity in the S1P presence. Unlike the rest of the compounds, candidate 1 shows a clear preference for the allosteric cavity thanks to the introduced modifications, since these substantially improve its binding in the S1PR1 PAM-equivalent cavity without altering it significantly in the orthosteric site. Similarly to ZCZ011(S), the C1 binding in the allosteric cavity is significantly improved by orthosteric-ligand presence. However, this improvement now depends on the S1PR1 state, being much better for the intermediate state, and thus allowing C1 to act as a positive allosteric modulator.

3.5. Characterization of the C1 Positive Allosteric Modulation.

In order to characterize the positive allosteric modulation exerted by compound 1 (C1), the $\text{Δ}{G}_{\text{total}}$ profiles of the activation processes for the systems S1PR1, C1:S1PR1 and S1P:C1:S1PR1 have been calculated (see Figure 9), in addition to taking into account the previously-calculated profile for the S1P:S1PR1 complex (see panel A of Figure 4).

Figure 9.

$\text{Δ}{G}_{\text{total}}$ profile of the activation processes for the systems S1PR1, C1:S1PR1 and S1P:C1:S1PR1 (Panel A, B and C, respectively). The position of the main states identified throughout different activation processes has been indicated. Error bars indicate the $\text{Δ}{G}_{\text{total}}$ uncertainty, which has been calculated propagating the standard error of its contributions adequately. Key energy differences ($\text{Δ}{G}^{‡}$ and $\text{Δ}{G}_{\text{act}}$) have been specified. Additionally, for comparative purposes, the S1PR1 activation profile (Panel A) has been depicted along with the S1P:S1PR1 one (green dashed line), enabling to observe how the S1PR1 activation process is altered by the presence of the S1P ligand.

Regarding the S1PR1 activation process (see panel A of Figure 9), our results indicate that the interconversion of the inactive and intermediate states is feasible without orthosteric and/or allosteric regulation $\left(\text{Δ}{G}^{‡}=15.58\pm 1.82\text{kcal}/\text{mol}\right)$, being the inactive state the global energy minimum of the system $\left(\text{Δ}{G}_{\text{act}}=9.08\pm 1.19\text{kcal}/\text{mol}\right)$. On the other hand, a partially-activated inactive state (I*) is identified with a similar stability to the original inactive state $\left(\text{Δ}G=1.16\pm 1.18\text{kcal}/\text{mol}\right)$, which could act as a precursor of the intermediate state $\left(\text{Δ}{G}^{\ast }=12.84\pm 1.34\text{kcal}/\text{mol}\right)$.

In contrast, for the S1P:S1PR1 complex (see panel A of Figure 4 and/or green dashed line in Panel A of Figure 9), there is a totally different situation in relation to the minimum energy states. Although energy minima equivalent to those obtained for the S1PR1 activation can be identified, with the exception that an additional one now appears between the partially-activated inactive state and the intermediate state, in particular the partially-inactivated intermediate state (Int*), their relative stabilities change completely. Specifically, the intermediate state is the global energy minimum, exhibiting much greater stability $\left(\text{Δ}{G}_{\text{act}}=-11.94\pm 0.96\text{kcal}/\text{mol}\right)$, while now the partially-activated inactive state is slightly more stable than the original one $\left(\text{Δ}G=-4.23\pm 1.35\text{kcal}/\text{mol}\right)$. Additionally, the partially-inactivated intermediate state shows an in-between stability in relation to the two previous states $\left(\text{Δ}G=-7.27\pm 1.36\text{kcal}/\text{mol}\right)$. These results indicate that the S1P agonistic behavior is due not only to a stabilization of the intermediate state with a better binding (see Table 1), but also to a stabilization of the conformation of this state (see Table S6), resulting in a considerably lower activation barrier ($\text{Δ}{G}^{‡}=7.78\pm 1.10$ and 15.58 ± 1.82 kcal/mol in the presence and absence of S1P, respectively).

As for the effect of candidate 1 (see panel B of Figure 9), this compound is not able to reduce the activation process barrier by itself $\left(\text{Δ}{G}^{‡}=15.26\pm 1.65\text{kcal}/\text{mol}\right)$, although it induces an additional stability of the intermediate state $\left(\text{Δ}{G}_{\text{act}}=7.11\pm 1.01\text{kcal}/\text{mol}\right)$. In addition, C1 also causes the appearance of the partially-activated inactive state, which again exhibits a slightly higher stability than the original inactive state $\left(\text{Δ}G=-4.25\pm 1.08\text{kcal}/\text{mol}\right)$, and the partially-inactivated intermediate state, although this is now substantially more unstable than when it was induced by S1P $\left(\text{Δ}G=7.30\pm 1.43\text{kcal}/\text{mol}\right)$. By contrast, when C1 acts together with S1P (see panel C of Figure 9), this compound induces an additional stabilization of all energy minima that appear after the transition state, being the corresponding one to the intermediate state the most relevant from a point of view of allosteric modulation $\left(\text{Δ}{G}_{\text{act}}=-12.90\pm 0.96\text{kcal}/\text{mol}\right)$. It is worth noting that the joint action of C1 and S1P preserves the decrease in the activation barrier caused by S1P $\left(\text{Δ}{G}^{‡}=7.96\pm 1.10\text{kcal}/\text{mol}\right)$. Therefore, C1 favors the S1PR1 activation process, inducing an additional stabilization of the intermediate state without altering the activation barrier decrease caused by the orthosteric agonist.

For all this, it can be drawn that the C1 behaves as a pure positive allosteric modulator since it is not capable of triggering the S1PR1 activation process by itself (the additional stability induced in the intermediate state is not enough to drag the system toward this state), but this compound does favor the activation process when it acts together with an orthosteric agonist inducing an additional stabilization of the intermediate state.

4. CONCLUSIONS

The present work has proven that our innovative developed methodology is suitable for studying the GPCRs allosteric modulation through free energy profiles for activation processes of GPCRs and their derived complexes from modulators binding.

As has been illustrated with S1PR1, this methodology allows the allosteric-modulator activity characterization in an efficient way, one of the most challenging issues regarding allosteric modulation that was still to be solved, which can be used for the rational design of this type of compounds. Additionally, both its ability to correctly describe the orthosteric-ligand binding and its potential to identify the key interactions that trigger the GPCRs activation processes have been appreciated. Therefore, the applicability of this methodology could be easily extended to rationally design selective orthosteric ligands and to study the GPCRs activation processes in a comprehensive way.

In relation to the results obtained for the case study, a de novo S1PR1 pure positive allosteric modulator (C1) has been rationally designed once the S1PR1 prospective allosteric cavities were identified (the S1PR1 PAM-equivalent and NAM-equivalent cavities), which were unknown and have also been determined in the present work. This rational design has taken advantage of the resemblance of these S1PR1 cavities to the CB1R allosteric cavities and the existing knowledge about the CB1R allosteric ligands. The C1 activity characterization has been carried out from the study of the activation processes of the systems S1PR1, C1:S1PR1, S1P:S1PR1 and S1P:C1:S1PR1. According to these activation processes, C1 is not capable of triggering the S1PR1 activation by itself, though this compound does favor the activation process when it acts together with an orthosteric agonist inducing an additional stabilization of the intermediate state. Consequently, this work also represents a step forward regarding the S1PR1 allosteric modulation, since its knowledge gap has been completed partially. Specifically, its prospective allosteric cavities have been identified and, according to our calculations, a de novo pure PAM for the S1PR1 PAM-equivalent cavity has been rationally designed, which could help improve the treatment of the S1PR1-associated pathologies.

For all this, the developed methodology provides a very useful tool to study the GPCRs allosteric modulation, though its applicability is not limited only to this topic, as has been shown, and we hope that this will contribute to extracting the full potential of given regulation.

Supplementary Material

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jcim.5c01163.

Tables with properties of binding, druggability, size and allosterism for the S1PR1 orthosteric and G-protein sites and the S1PR1 cavities 1–3 in the main states of the S1P:S1PR1 complex; representation of the expansion process of the S1PR1 PAM-equivalent and NAM-equivalent cavities throughout the S1PR1 activation process; table with $\text{Δ}{G}_{\text{conf}}$ for the inactive and intermediate states of the systems S1PR1 and S1P:S1PR1 (PDF)

ZIP file with model input files of Molaris for TMD simulations, relaxation and equilibration MD simulations, PDLD/S-LRA simulations and CG modeling, and the developed force field parameters of S1P, ZCZ011-(S), C1, C2, and C3 (ZIP)

Table 6.

Candidate 1 $\text{Δ}{G}_{\text{bind}}$ for its Most Favorable Binding Modes Inside the Orthosteric Site and the S1PR1 PAM-Equivalent Cavity for the Inactive and Intermediate States^a

cavity	state	pose	presence of S1P	$\text{Δ}{G}_{\text{bind}}^{\text{elec}}\left(\text{kcal}/\text{mol}\right)$	$\text{Δ}{G}_{\text{bind}}^{\text{non}-\text{elec}}\left(\text{kcal}/\text{mol}\right)$	$\text{Δ}{G}_{\text{bind}}\left(\text{kcal}/\text{mol}\right)$
Ortho	Inactive	n/a	n/a	−2.99	0.73	−2.26
Ortho	Intermediate	n/a	n/a	−2.20	−1.87	−4.07
Allo	Inactive	1	No	0.92	−12.96	−12.04
Allo	Intermediate	1	No	0.29	−9.94	−9.65
Allo	Inactive	1	Yes	1.05	−15.66	−14.61
Allo	Intermediate	1	Yes	0.22	−16.38	−16.16

^aIn the case of columns of $\text{Δ}{G}_{\text{bind}}$ and its contributions (columns 5 to 7), the hyphen symbol indicates that the considered compound for the specified state is unstable and unbinds during the equilibration and relaxation MDs. In the first column, ortho and allo stand for orthosteric and allosteric, respectively.

Table 7.

Candidate 2 $\text{Δ}{G}_{\text{bind}}$ for its Most Favorable Binding Modes Inside the Orthosteric Site and the S1PR1 PAM-Equivalent Cavity for the Inactive and Intermediate States^a

cavity	state	pose	presence of S1P	$\text{Δ}{G}_{\text{bind}}^{\text{elec}}\left(\text{kcal}/\text{mol}\right)$	$\text{Δ}{G}_{\text{bind}}^{\text{non}-\text{elec}}\left(\text{kcal}/\text{mol}\right)$	$\text{Δ}{G}_{\text{bind}}\left(\text{kcal}/\text{mol}\right)$
Ortho	Inactive	n/a	n/a	−3.08	0.73	−2.35
Ortho	Intermediate	n/a	n/a	1.66	−4.40	−2.74
Allo	Inactive	1	No
Allo	Intermediate	1	No
Allo	Inactive	1	Yes	0.72	−2.53	−1.81
Allo	Intermediate	1	Yes	1.51	−4.40	−2.89

^aIn the case of columns of $\text{Δ}{G}_{\text{bind}}$ and its contributions (columns 5 to 7), the hyphen symbol indicates that the considered compound for the specified state is instable and unbinds during the equilibration and relaxation MDs. In the first column, ortho and allo stand for orthosteric and allosteric, respectively.

ABBREVIATIONS

GPCRs

G-protein-coupled receptors

TMs

transmembrane α helices

S1PR1

sphingosine 1-phosphate receptor 1

S1P

sphingosine 1-phosphate

PAM

positive allosteric modulator

NAM

negative allosteric modulator

CB1R

cannabinoid receptor 1

MD

molecular dynamics

TMD

targeted molecular dynamics

PDLD

protein-dipole Langevin-dipole

LRA

linear response approximation

CG

coarse-grained

SCAAS

surface-constrainted all-atoms solvent

LRF

local reaction field

Data Availability Statement

The data underlying this study are available in the published article and its Supporting Information with the exception of the employed S1PR1 and CB1R structures, which were already published in previous works. The S1PR1 structures were provided by the corresponding authors of the research article at 10.1073/pnas.2317893121 by permission, therefore they are available upon reasonable request. By contrast, the CB1R ones are openly available in the repository called Proteins2024 at https://github.com/acruzsaez/Proteins2024.