Pathway and mechanism of drug binding to chemokine receptors revealed by accelerated molecular simulations

Abstract

Background:

Chemokine GPCRs play key roles in biology and medicine. Particularly, CXCR4 promotes cancer metastasis and facilitate HIV entry into host cells. Plerixafor (PLX) is a CXCR4 drug, but the pathway and binding site of PLX in CXCR4 remain unknown.

Results & methodology:

We have performed molecular docking and all-atom simulations using Gaussian accelerated molecular dynamics (GaMD), which are consistent with previous mutation experiments, suggesting that PLX binds to the orthosteric site of CXCR4 as an antagonist. The GaMD simulations further revealed an intermediate allosteric binding site at the extracellular mouth of CXCR4.

Conclusion:

The newly identified allosteric site can be targeted for novel drug design targeting CXCR4 and other chemokine receptors.

Chemokine receptors are key GPCRs, which control cell migration during immune system responses and development of cardiovascular and central nervous systems and in numerous diseases (including inflammation and cancer) [1,2]. Importantly, the CCR5 and CXCR4 chemokine receptors also function as co-receptors that facilitate HIV entry into host immune cells [3–5]. Maraviroc and vicriviroc, two clinic drugs of HIV entry inhibitors, are antagonists of the CCR5 receptor [6]. These drugs could block HIV replication, but ultimately resistance develops, due to emergence of viruses that can utilize the CXCR4 co-receptor.

Unfortunately, the development of the CXCR4 antagonists as effective drugs of HIV infection has been greatly hindered. This is due to the facts that CXCR4 is widely expressed in many different human tissues and the primary endogenous chemokine-binding (orthosteric) site is conserved across different subtypes of the chemokine receptors. Off-target binding often occurs for the prototypical CXCR4 antagonists [7]. Alternatively, it is appealing to develop allosteric modulators, which selectively bind to a topographically distant (allosteric) site with divergent sequences. They are promising to regulate the responsiveness of CXCR4 to endogenous chemokine with reduced side effects [8–10].

X-ray crystal structures have been determined for the CXCR4 receptor, in complex with an antagonist compound IT1t [4], a cyclic peptide CVX15 [4] and vMIP-II that is a viral chemokine encoded by Kaposi’s Sarcoma-associated herpesvirus [3]. These ligands are antagonists that stabilize the CXCR4 in the inactive state. Homodimers are formed for the CXCR4 in the x-ray structures with interface involving the transmembrane (TM) helices 5 and 6. While the CVX15 cyclic peptide binds to a major subpocket, the IT1t compound and vMIP-II viral chemokine occupies a minor subpocket at the receptor orthosteric site. In addition, the vMIP-II viral chemokine forms extensive interactions with the N-terminus of the CXCR4 receptor. These structures provide important insights into antagonist binding of the CXCR4. They also serve as an excellent starting point for computational modeling and structure-based drug design.

Plerixafor (PLX) or AMD3100 is an organic compound with two positively charged cyclam rings that are connected by a 1,4-phenylenebis(methylene) linker [11]. It has been identified as the first selective small-molecule inhibitor of the CXCR4 receptor. Previous studies suggested multiple binding modes of PLX in the CXCR4, raising the question of whether it is an antagonist or a negative allosteric modulator [11–13]. The target binding site of known drugs in the CXCR4 such as PLX [11–13] remains elusive. Furthermore, the pathways and mechanism of drug binding to the CXCR4 receptor are still poorly understood, which has greatly hindered effective drug design of this important target.

Gaussian accelerated molecular dynamics (GaMD) is a computational technique that provides simultaneous unconstrained enhanced sampling and free energy calculations of large biomolecules [14]. GaMD works by adding a harmonic boost potential to smooth the potential energy surface of biomolecules to reduce the system energy barriers [14]. A Gaussian distribution is followed by the added harmonic boost potential. Proper energetic reweighting of GaMD simulations is achieved through cumulant expansion to the second order (‘Gaussian approximation’). While accurate free energy profiles have been obtained from GaMD simulations for small biomolecules such as alanine dipeptide compared with long-timescale conventional MD simulations [14,15], it is exceedingly difficult to calculate convergent free energy profiles for large systems like GPCRs. Nevertheless, relatively low-energy conformational states of biomolecules can be identified from the GaMD reweighted free energy profiles [16,17]. Without the requirement of carefully predefined collective variables, GaMD is advantageous to study complex biological processes such as drug binding to proteins. GaMD simulations have been successfully demonstrated on ligand binding [14–18], protein folding [14,18], GPCR activation [19] and the protein-membrane [20], protein–protein [21,22] and protein–nucleic acid [23,24] interactions.

In this study, we have combined molecular docking and all-atom GaMD simulations to determine the binding mode of the PLX drug in the CXCR4 receptor. The docking and GaMD simulations are consistent with previous mutation experimental data [13,25], suggesting that PLX binds to the orthosteric site of CXCR4 as an antagonist. The GaMD simulations have also identified important intermediate states of the drug and an intermediate allosteric site of the CXCR4 receptor. This opens a new platform to design novel allosteric modulators as selective drugs of CXCR4 for the treatment against HIV and cancers associated with this receptor.

Materials & methods

Gaussian accelerated molecular dynamics

GaMD is an unconstrained enhanced sampling approach that works by adding a harmonic boost potential to smooth the potential energy surface of biomolecules to reduce energy barriers [14]. Brief description of the method is provided here.

Consider a system with N atoms at positions $\stackrel{\rightharpoonup }{r}= \left\{\stackrel{\rightharpoonup }{r_1} ,\dots ,\stackrel{\rightharpoonup }{r_N}\right\}$. When potential energy of the system $V\left(\stackrel{\rightharpoonup }{r}\right)$ is less than a threshold energy E, a boost potential $\Delta V\left(\stackrel{\rightharpoonup }{r}\right)$ is added to the system as follows:

$$V^{\mathrm{*}}\left(\stackrel{\rightharpoonup }{r}\right)=V\left(\stackrel{\rightharpoonup }{r}\right)+\mathrm{ }\Delta V\left(\stackrel{\rightharpoonup }{r}\right),\mathrm{ }V\left(\stackrel{\rightharpoonup }{r}\right)<E$$ (Eq. 1)

$$\Delta V\left(\stackrel{\rightharpoonup }{r}\right)= \frac{1}{2}k{(E-V(\stackrel{\rightharpoonup }{r}\left)\right)}^2, V\left(\stackrel{\rightharpoonup }{r}\right)<E$$ (Eq. 2)

where k is the harmonic force constant. The two adjustable parameters E and k can be determined by application of three enhanced sampling principles. First, for any two arbitrary potential values $V_1\left(\stackrel{\rightharpoonup }{r}\right)$ and $V_2\left(\stackrel{\rightharpoonup }{r}\right)$ found on the original energy surface, if $V_1\left(\stackrel{\rightharpoonup }{r}\right)<$ $V_2\left(\stackrel{\rightharpoonup }{r}\right)$, $\Delta V$ should be a monotonic function that does not change the relative order of the biased potential values, in other words, $V_1^{*}\left(\stackrel{\rightharpoonup }{r}\right)<$ $V_2^{*}\left(\stackrel{\rightharpoonup }{r}\right)$. Second, if $V_1\left(\stackrel{\rightharpoonup }{r}\right)<$ $V_2\left(\stackrel{\rightharpoonup }{r}\right)$, the potential difference observed on the smoothed energy surface should be smaller than that of the original, in other words, $V_2^{*}\left(\stackrel{\rightharpoonup }{r}\right)-V_1^{*}\left(\stackrel{\rightharpoonup }{r}\right)< V_2\left(\stackrel{\rightharpoonup }{r}\right)-V_1\left(\stackrel{\rightharpoonup }{r}\right)$. By combining the first two criteria and plugging in the formula of $V^{*}\left(\stackrel{\rightharpoonup }{r}\right)$ and $\Delta V$, we obtain:

$$V_{max}\le E\le V_{min}+\frac{1}{k}$$ (Eq. 3)

where $V_{min}$ and $V_{max}$ are the system minimum and maximum potential energies. To ensure that (Equation 3) is valid, k has to satisfy: $k\le \frac{1}{V_{max}-V_{min}}$. Let us define $k\equiv \frac{k_0}{V_{max}-V_{min}}$, then $0<k_0\le 1$. Third, the standard deviation (SD) of ΔV needs to be small enough (i.e., narrow distribution) to ensure accurate reweighting using cumulant expansion to the second order: ${\sigma }_{\Delta V}=k(E-V_{avg}){\sigma }_V\le {\sigma }_0$, where $V_{avg}$ and ${\sigma }_V$ are the average and SD of ΔV with ${\sigma }_0$ as a user-specified upper limit (e.g., $10k_{B}T$) for accurate reweighting. When E is set to the lower bound $E=V_{max}$ according to equation 3, $k_0$ can be calculated as:

$$k_0=\min \left(1.0,k_0^{‘}\right)=min\left(1.0,\frac{{\sigma }_0}{{\sigma }_V}.\frac{V_{max}-V_{min}}{V_{max}-V_{avg}}\right)$$ (Eq. 4)

Alternatively, when the threshold energy E is set to its upper bound $E=V_{min}+\frac{1}{k}$, $k_0$ is set to:

$$k_0=k_0^{“}\equiv \left(1-\frac{{\sigma }_0}{{\sigma }_V}\right).\frac{V_{max}-V_{min}}{V_{avg}-V_{min}}$$ (Eq. 5)

if $k_0^{”}$ is calculated between 0 and 1. Otherwise, $k_0$ is calculated using (Equation 4).

The original GaMD method provides schemes to add only the total potential boost ${\Delta V}_P$, only dihedral potential boost ${\Delta V}_D$ or the dual potential boost (both ${\Delta V}_P$ and ${\Delta V}_D$) [14]. The dual-boost GaMD (GaMD_Dual) simulation generally provides higher acceleration than the other two types of simulations [26]. The simulation parameters comprise the threshold energy E for applying boost potential and the effective harmonic force constants, $k_{0P}$ and $k_{0D}$ for the total and dihedral potential boost, respectively. Since ligand binding mainly involves nonbonded interactions, another scheme of nonbonded dual-boost GaMD (GaMD_Dual_NB) is introduced in this study of drug binding to the CXCR4 chemokine receptor, for which boost potential is applied to the system dihedral energy and nonbonded potential energy terms.

Energetic reweighting of GaMD simulations

For energetic reweighting of GaMD simulations to calculate potential mean force (PMF), the probability distribution along a reaction coordinate is written as $p^{*}\left(A\right)$. Given the boost potential $\Delta V\left(r\right)$ of each frame, $p^{*}\left(A\right)$ can be reweighted to recover the canonical ensemble distribution $p\left(A\right)$, as:

$$p\left(A_j\right)=p^{*}\left(A_j\right)\frac{{⟨e^{\beta \Delta V\left(r\right)}⟩}_j}{{\sum }_{i=1}^M{⟨p^{*}\left(A_i\right)e^{\beta \Delta V\left(r\right)}⟩}_i}, j=1,\dots ,\mathrm{ }M$$ (Eq. 6)

where M is the number of bins, $\beta =k_{B}T$ and ${\left(e^{\beta \Delta V\left(r\right)}\right)}_j$ is the ensemble-averaged Boltzmann factor of $\Delta V\left(r\right)$ for simulation frames found in the j^th bin. The ensemble-averaged reweighting factor can be approximated using cumulant expansion:

$${\left(e^{\beta \Delta V\left(r\right)}\right)}_j=\mathrm{e}\mathrm{x}\mathrm{p}\left\{\sum _{k=1}^{\infty }\frac{{\beta }^k}{k!}{C}_k\right\}$$ (Eq. 7)

where first two cumulants are given by

$$C_1=\left(\Delta V\right),$$

$$C_2=\left({\Delta V}^2\right)-{\left(\Delta V\right)}^2={\sigma }_V^2$$ (Eq. 8)

The boost potential obtained from GaMD simulations usually follows near-Gaussian distribution [27]. Cumulant expansion to the second order thus provides a good approximation for computing the reweighting factor [14,28]. The reweighted free energy $F\left(A\right)=-k_{B}T\mathrm{l}\mathrm{n}p\left(A\right)$ is calculated as

$$F\left(A\right)=F^{*}\left(A\right)-\sum _{k=1}^2\frac{{\beta }^k}{k!}{C}_k+F_c$$ (Eq. 9)

where $F^{*}\left(A\right)=-k_{B}T\mathrm{l}\mathrm{n}{p}^{*}\left(A\right)$ is the modified free energy obtained from GaMD simulation and $F_c$ is a constant.

System setup

The highest-resolution (2.5 Å) x-ray structure of CXCR4 co-crystalized with antagonist IT1t (Protein Data Bank [PDB]: 3ODU) [4] was used for setting up the simulation system. The structure included 293 residues (27–319) of the 352 residues of the CXCR4. The T4-lysozyme residues that were added to facilitate crystallization and the IT1t antagonist were removed. Structure of the PLX drug that was refined with molecular dynamics was downloaded from the Automated Topology Builder and Repository (http://compbio.biosci.uq.edu.au/atb).

PLX (AMD3100; Mozobil^®, Genzyme, MA USA) is an organic small molecule having a 1,4-phenylenebis(methylene) linker that connects two cyclam rings [11]. With its symmetric structure, it has been suggested to exhibit a unique binding mode to CXCR4 unlike other ligands [11]. At physiological pH, each cyclam ring (1,4,8,11-tetrazacyclotetradecane) has two positive charges, giving an overall charge of +4 to the entire bicyclam PLX molecule [11]. Maestro in Schrodinger [29] was used to prepare the ligand at physiological pH 7.4, which generated only one configuration of the ligand with four positive charges on atoms N1, N3, N4 and N7 of the bicyclam rings. This configuration was used to carry out GaMD simulations.

Simulation systems were prepared for the unbound and the bound state of the ligand. The unbound state was prepared by placing a total of ten PLX molecules at a distance >15 Å from the receptor (Figure 1A). The bound state was prepared by rigid-body docking of PLX to the receptor using AutoDock [30]. The lowest energy conformation (-7.48 kcal/mol) was selected from docking for performing GaMD simulations. The simulation systems were prepared with the CHARMM-GUI [31–33] web server for using the membrane protein input generator. All chain termini were capped with neutral patches (acetyl and methylamide). The systems were solvated in 0.15 M NaCl solution at temperature 310 K. The CHARMM36m [34] parameter set was used for the receptor and lipids and the CGenFF 2.2.0 parameters [35,36] for the ligand. The output files from CHARMM-GUI were used to perform GaMD simulations with AMBER as summarized in Table 1.

Figure 1. . Gaussian accelerated molecular dynamics simulation (Sim2 in Table 1) successfully captured spontaneous drug binding to the CXCR4 chemokine receptor.

(A) Computational model used for simulation of the CXCR4 receptor (blue ribbons) with ten PLX drug molecules (red spheres) placed away in the solvent. The receptor was inserted in a POPC lipid bilayer (cyan sticks) and solvated in an aqueous solution (cyan) of 0.15 M NaCl, (B) structure of PLX with numbered nitrogen atoms and the symmetry-corrected RMSD of PLX relative to its bound conformation plotted as a function of simulation time, (C) the distance between the C_γ atom of receptor residue D262^6.58 of the binding pocket and the N7 atom of PLX, (D) the distance between the Cγ atom of receptor residue D97^2.63 of the binding pocket and the N3 atom of PLX, (E) the distance between the C_δ atom of receptor residue E288^7.39 of the binding pocket and the N4 atom of PLX and (F) GaMD predicted binding pose of PLX (red sticks) at the orthosteric site of CXCR4 (blue ribbons) with their interacting residues labeled and highlighted in green sticks. Antagonist IT1t (orange) and docking conformation of PLX (yellow) are shown for reference. The seven TM helices I–VII and three extracellular loops (ECL) 1–3 are labeled in the CXCR4 receptor.

GaMD: Gaussian accelerated molecular dynamics; PLX: Plerixafor; RMSD: Root-mean-square deviation; TM: Transmembrane.

Table 1.  . Summary of Gaussian accelerated molecular dynamics simulations performed on the CXCR4 chemokine receptor in the presence of the plerixafor drug.

System	Natoms†	Method‡	ID	Length (ns)	$\Delta V_{avg}$§ (kcal/mol)	${\sigma }_{\Delta V}$¶ (kcal/mol)
CXCR4 + PLX (unbound)	120275	GaMD_Dual_NB	Sim1	800	19.81	4.49
			Sim2	1000	19.85	4.49
			Sim3	800	19.76	4.48
			Sim4	800	19.83	4.49
			Sim5	800	19.77	4.48
CXCR4 + PLX (unbound)	120275	GaMD_Dual	Sim6	800	14.93	4.32
			Sim7	800	14.91	4.31
			Sim8	800	14.92	4.32
			Sim9	800	14.96	4.32
			Sim10	800	15.01	4.33
CXCR4 + PLX (bound)	76625	GaMD_Dual	Sim11	300	15.22	4.37
			Sim12	300	15.46	4.40
			Sim13	300	15.19	4.37

^†N_atoms: number of atoms in the system.

^‡GaMD_Dual is dual boost GaMD and GaMD_Fual_NB is nonbonded dual-boost GaMD.

^§$\Delta {\mathit{V}}_{\mathit{a}\mathit{v}\mathit{g}}$: average of the GaMD boost potential.

^¶${\mathit{\sigma }}_{\Delta \mathit{V}}$: SD of the GaMD boost potential.

GaMD: Gaussian accelerated molecular dynamic; PLX: Plerixafor; SD: Standard deviation.

Simulation protocol

The output files of CHARMM-GUI web server were used for initial energy minimization, equilibration and conventional molecular dynamics (cMD) to prepare the systems for GaMD simulations. The systems were energy minimized and equilibrated at 310 K using default parameters given by CHARMM-GUI. Energy minimization of the systems was done for 5000 steps using steepest-descent algorithm at constant number, volume and temperature at 310 K. The systems were further equilibrated for 375 ps at 310 K at constant number, pressure and temperature. The systems were simulated using cMD for 10 ns at 1 atm pressure and 310 K temperature. GaMD implemented in GPU version of AMBER 18 [14,37] was applied to simulate the CXCR4 systems. The simulations involved an initial short cMD of 2.5 ns to calculate GaMD acceleration parameters and GaMD equilibration of added boost potential for 40 ns.

Five independent 800–1000 ns production GaMD simulations with boost potential applied to the system nonbonded potential and dihedral energy terms (GaMD_Dual_NB) were performed on the CXCR4 receptor with PLX unbound. In this case, the GaMD_Dual_NB algorithm was developed for more efficient sampling of the nonbonded ligand binding process. For comparison, the same system was also simulated using the previous dual-boost GaMD (GaMD_Dual) in five independent 800 ns production runs, for which boost potential was applied to the dihedral energy and the system total potential energy. In the case of the drug bound state, simulations were performed to refine the docked conformation of PLX inside the receptor, for which GaMD_Dual was generally considered to be better for biomolecular conformational sampling and thus applied for the structural refinement. Therefore, three independent 300 ns GaMD_Dual simulations were performed on the CXCR4 receptor with PLX docked obtained with AutoDock rigid-body docking. The GaMD simulations are summarized in Table 1.

Simulation analysis

Simulation analysis was carried out using VMD [38] and CPPTRAJ [39]. The hierarchical agglomerative clustering algorithm was used to cluster snapshots of the diffusing PLX using its center ring with all production GaMD simulations combined for each system. The root-mean-square deviation (RMSD) cutoff was set to 2.0 Å. The combined GaMD simulations of PLX docked to CXCR4 were clustered to obtain five clusters for which the simulation trajectories showed stable binding of PLX at the orthosteric site of CXCR4. The top ranked cluster was selected as a reference, termed the ‘bound conformation’, to calculate the RMSD of PLX during simulations of its binding to the receptor from the solvent.

Since PLX is symmetric with two cyclam rings, its RMSD relative to the bound conformation was calculated twice, once with the original configuration and another with the two cyclam rings flipped. The minimum of the two RMSD values of each frame (i.e., symmetry-corrected RMSD) was used for further analysis.

The RMSD of PLX relative to the bound conformation and distances of important interactions between the drug and receptor residues were identified to calculate PMF free energy profiles using the PyReweighting toolkit [28]. A bin size of 2 Å was used for distances and RMSD. Free energy values were also reweighted for each of the ligand structural clusters. The cutoff was set to 500 frames in a bin or cluster for reweighting.

Results

Three independent 300 ns dual-boost GaMD (GaMD_Dual) simulations on the PLX docked CXCR4 obtained from AutoDock rigid-body docking showed that the drug maintained stable binding at the orthosteric site. Five structural clusters of PLX were generated from clustering snapshots of the drug with all three GaMD simulations combined (Supplementary Figure 1A). The top ranked cluster, which exhibited a similar conformation compared with the docking pose, was defined the ‘bound conformation’ of PLX (Supplementary Figure 1B). In one of the nonbonded dual-boost GaMD simulations (GaMD_Dual_NB) of CXCR4, spontaneous binding of PLX from free diffusion in the solvent to the orthosteric site of the CXCR4 receptor was captured (Figure 1 & ‘Sim2’ in Table 1). The minimum RMSD of PLX relative to the bound conformation was 2.76 Å. The other four GaMD simulations revealed important interactions of the drug with the receptor surface, although no complete binding of PLX to the CXCR4 orthosteric site was observed (Supplementary Figure 2). The five GaMD simulations showed similar average and SD of the added boost potential as 19.8 kcal/mol and 4.48 kcal/mol, respectively.

The GaMD_Dual simulations of CXCR4 with unbound PLX showed lower boost potential of 14.9 ± 4.32 kcal/mol. No complete binding of PLX to the CXCR4 was captured in any of the five simulations (Supplementary Figure 3). Moreover, GaMD_Dual simulations of CXCR4 with PLX bound showed stable binding of PLX at the CXCR4 orthosteric site with similar GaMD boost potential of 15.2 ± 4.38 kcal/mol.

GaMD simulations revealed PLX binding to the CXCR4 receptor as an antagonist

In Sim2 of the GaMD_Dual_NB simulation trajectory, out of the ten PLX drug molecules freely diffusing in the solvent, PLX1 approached the extracellular mouth of the CXCR4 receptor and bound to the orthosteric site (Figure 1). Stable binding was observed after approximately 480 ns with an RMSD of approximately 2.8 Å relative to the GaMD refined docking conformation of PLX (Figure 1B).

Within approximately 180 ns of simulation time, a salt bridge was formed between the receptor residue D262^6.58 and N7 of PLX (Figure 1C). The protein residue D97^2.63 formed a salt bridge with atom N3 of PLX at approximately 240 ns despite fluctuations (Figure 1D). Residue E288^7.39 that was located deep inside the receptor-binding pocket formed another salt bridge with atom N4 of PLX much later (∼480ns) during the GaMD simulation (Figure 1E). In the final bound conformation, the positively charged PLX formed stable salt bridges with residues D97^2.63, D262^6.58 and E288^7.39 as shown in Figure 1F. While the IT1t antagonist bound to only the minor subpocket of CXCR4 (PDB: 3ODU) [4], our GaMD simulations revealed binding of PLX to both the minor and major subpockets of the receptor (Figure 1F).

Free energy profiles of PLX binding to the CXCR4 receptor

We combined all five GaMD_Dual_NB simulations to calculate reweighted free energy profiles of the drug binding process to characterize the binding of PLX to the CXCR4 receptor. The RMSD of PLX relative to the bound conformation and the salt bridge distances formed between the protein and ligand were selected as reaction coordinates. From the free energy profiles as shown in Figure 2, we identified four low-energy conformational states: unbound, intermediate 1 (I1), intermediate 2 (I2) and bound. In the unbound state, the distances between the charge center of D262^6.58 (the CG atom) and PLX atom N7 (Figure 2A), D97^2.63 (the CG atom) and PLX atom N3 (Figure 2B); E288^7.39 (the CD atom) and PLX atom N4 (Figure 2C) exhibited a broad energy well centered around approximately 60 Å, suggesting that the drug diffused in bulk solution. In the Intermediate I1 state, the distances between the charge center CG atom of D262^6.58 and PLX atom N7 (Figure 2A) and D97^2.63 (the CG atom) and PLX atom N3 (Figure 2B) were 12.65 and 15.06 Å, respectively. The RMSD of PLX relative to the bound conformation was 11.42 Å. In the intermediate I2 state, the distances between the charge center CG atom of D262^6.58 and PLX atom N7 (Figure 2A) and E288^7.39 (the CD atom) and PLX atom N4 (Figure 2C) were 7.38 and 21.49 Å, respectively. The RMSD of PLX relative to the bound conformation was 11.14 Å. In the bound state, the distances between the charge center CG atom of D262^6.58 and PLX atom N7 (Figure 2A), D97^2.63 (the CG atom) and PLX atom N3 (Figure 2B) and E288^7.39 (the CD atom) and PLX atom N4 (Figure 2C) were approximately 3.0 Å and the RMSD of PLX relative to the bound conformation was approximately 2.8 Å. Further PMF calculations showed that the distances between the C_α atoms of receptor residues R134^3.50-T241^6.37 and S131^3.47-I245^6.41 maintained the IT1t antagonist-bound CXCR4 x-ray structural values in the free energy minima despite PLX movements in the GaMD simulations (Supplementary Figure 4). Therefore, the CXCR4 receptor remained in the inactive state during the PLX drug binding to the orthosteric site.

Figure 2. . 2D potential of mean force free energy profiles of the CXCR4–plerixafor interactions.

(A) The distance between the C_γ atom of CXCR4 residue D262^6.58 of the binding pocket and the N7 atom of PLX as plotted in Figure 1D, (B) the distance between the C_γ atom of CXCR4 residue D97^2.63 of the binding pocket and the N3 atom of PLX as plotted in Figure 1E, (C) the distance between the C_δ atom of CXCR4 residue E288^7.39 of the binding pocket and the N4 atom of PLX as plotted in Figure 1F. (D) A 2D PMF of the distance between receptor residue D262^6.58 and PLX atom N7 versus the distance between receptor residue E288^7.39 and PLX atom N4. The I1 and I2 states at energy minima were identified at (D262:CG-PLX:N7, E288:CD-PLX:N4) distances of (13.01 Å, 8.60 Å) and (7.38 Å, 21.49 Å), respectively. (E) A 2D PMF of the distance between receptor residue D262^6.58 and PLX atom N7 versus the distance between receptor residue D97^2.63 and PLX atom N3. (F) Zoomed view of the 2D PMF as plotted in Figure 2E. The I1 and I2 states at energy minima (∼0.47 kcal/mol and 0.0 kcal/mol) were identified at (D262:CG-PLX:N7 and D97:CG-PLX:N3) distances of (13.01 Å and 16.40 Å) and (7.38 Å and 19.54 Å), respectively. Low-energy unbound, intermediate I1 and I2 and bound conformational states are labeled in the PMF profiles.

GaMD: Gaussian accelerated molecular dynamic; PLX: Plerixafor; PMF: Potential of mean force; RMSD: Root-mean-square deviation.

In addition, we calculated 2D PMF profiles of the D262^6.58:CG-PLX:N7 distance versus the D97^2.63:CG-PLX:N3 distance (Figure 2E & F) and the E288^7.39:CD-PLX:N4 distance (Figure 2D) to further characterize the intermediate states I1 and I2 conformational states. In the intermediate I1 state, the D262^6.58:CG-PLX:N7, D97^2.63:CG-PLX:N3 and E288^7.39:CD-PLX:N4 distances were 13.01, 16.40 and 8.60 Å, respectively (Figure 2D–F). In the intermediate I2 state, the D262^6.58:CG-PLX:N7, D97^2.63:CG-PLX:N3 and E288^7.39:CD-PLX:N4 distances were 7.38, 19.54 and 21.49 Å, respectively (Figure 2D–F).

A novel intermediate drug binding site was identified in the CXCR4

The intermediate conformational states I1 and I2 identified from the free energy profiles of GaMD simulations were shown in Figures 3A & B, respectively. Remarkably, polar and charged residues in the ECL2-TM6-TM6 region of the CXCR4 (including residues D187^ECL2, D193^5.32 and D262^6.58) formed favorable interactions with the positively charged nitrogen atoms of PLX. These interactions played a significant role in the recognition and binding of PLX to the CXCR4 receptor.

Figure 3. . Novel intermediate drug binding site in the CXCR4 receptor. Intermediate I1 and I2 conformational states of PLX (red sticks) bound to the CXCR4 receptor (blue ribbons).

The charged nitrogen atoms of PLX are numbered as 1, 3, 4 and 7. (A) Intermediate I1 state with interacting residues highlighted in green sticks, including D187^ECL2 and D262^6.58 that formed ionic interactions with PLX atoms N4 and N3, respectively. (B) Intermediate I2 state with interacting residues highlighted in green sticks, including D187^ECL2, D193^5.32 and D262^6.58 that formed salt bridges with positively charged N7, N4 and N1 of PLX, respectively.

GaMD: Gaussian accelerated molecular dynamic; PLX: Plerixafor.

In the intermediate I1 conformation, residues D187^ECL2 and D262^6.58 formed ionic interactions with PLX atoms N4 and N3, respectively (Figure 3A). In the intermediate I2 conformation, residues D187^ECL2, D193^5.32 and D262^6.58 formed salt bridges with positively charged N7, N4 and N1 atoms of PLX, respectively (Figure 3B). Therefore, the two intermediate states of drug–receptor complex share the common recognizing determinants that involve key residues D187^ECL2, D193^5.32 and D262^6.58 of the receptor. Two distinct low-energy states were obtained in the free energy profiles as the symmetric ligand flips its conformation at the same site of the receptor in the ECL2–TM5–TM6, which therefore was considered as a single intermediate drug binding site of the CXCR4.

Pathway of drug binding to the CXCR4 receptor

Next, structural clustering was performed on GaMD simulation frames of the PLX drug to identify the representative binding pathway of PLX to the CXCR4 receptor (see ‘Materials & methods’ section). The structural clusters were also reweighted to obtain their original free energy values which ranged from 0 to approximately 5.5 kcal/mol. The top 35 reweighted clusters were selected to represent the binding pathway of PLX to the CXCR4 receptor as shown in Figure 4. Starting from diffusion in the solvent, PLX binds to an intermediate site between ECL2–TM5–TM6 and then the final orthosteric site in the TM domain. In the top ranked structural cluster, PLX bound to the target orthosteric site. The second lowest-energy cluster was located at the opening of the receptor in the region between ECL2 and the TM5 and TM6 helices, whereas few ligand clusters were observed on the receptor surface near ECL2.

Figure 4. . Binding pathway of the plerixafor drug to the CXCR4 chemokine receptor revealed from Gaussian accelerated molecular dynamic simulations.

Starting from diffusion in the solvent, PLX bound to the target site of the CXCR4 receptor via an intermediate site located between ECL2 and TM V–VI helices. The CXCR4 is shown in blue ribbons. The PLX structural clusters (sticks) are colored by the reweighted PMF free energy values in a green (0 kcal/mol)-white-red (2.5 kcal/mol) color scale.

PLX: Plerixafor; PMF: Potential of mean force; TM: Transmembrane.

Discussion

In this study, molecular docking and all-atom GaMD simulations were applied to determine the target site and mechanism of the PLX drug binding in the CXCR4 receptor. The GaMD simulations successfully captured spontaneous binding of the PLX drug to the receptor orthosteric site. However, it is important to note that the free energy profiles presented here were not converged since the binding of PLX to the receptor orthosteric site was observed only once in one of the five GaMD simulations. It is exceedingly difficult to calculate accurate free energy profiles for such large system of this study even with enhanced sampling. Nevertheless, we observed multiple events of the PLX drug binding to the CXCR4 in the extracellular mouth region (Figure 1B; Supplementary Figure 2 & 3), which has been identified as an allosteric site of many class A GPCRs [40–43]. It has proven challenging to simulate complete binding of GPCR antagonists, especially with a relatively large size like the PLX drug molecule. Previous tens-of-microsecond molecular dynamics (MD) simulations using the Anton supercomputer [44] and accelerated MD simulations [45] were able to capture binding of a similar sized antagonist tiotropium to only the extracellular mouth of the M₃ muscarinic receptor, but not to the final ‘orthosteric’ target site deeply buried in the receptor TM domain. Therefore, the GaMD simulations in this study presented an advance for us to capture complete binding of PLX to the orthosteric site of CXCR4. Furthermore, energetic reweighting of the GaMD simulations allowed us to obtain ‘semi-quantitative’ PMF profiles to identify relatively low-energy conformational states of the PLX drug binding to CXCR4. Consistent results were obtained from molecular docking and GaMD simulations, showing that the PLX drug bound to the same IT1t antagonist-binding orthosteric site of CXCR4 (being an antagonist). The docking and simulation findings were further supported by previous mutation experiments, which suggested PLX interacts particularly with residues D262^6.58 and E288^7.39 in the orthosteric site of CXCR4 [13,25]. Taken together, the computational modeling and experimental data strongly supported our finding that the PLX drug molecule bound to the deeply buried orthosteric site of CXCR4 via an intermediate allosteric site in the receptor extracellular mouth.

The orthosteric binding pocket of the CXCR4 receptor largely consists of negatively charged residues [46] that form salt bridges with the positively charged PLX (Figure 1F). The negatively charged binding pocket is known to have two distinguished major and minor subpockets [46]. The crystal structures of CXCR4 with small molecule antagonist IT1t (PDB: 3ODU) and viral chemokine antagonist vMIP-II (PDB: 4RWS) revealed similar binding sites in the minor subpocket [3,4]. The IT1t and the N-terminus of vMIP-II make polar contacts with residues D97^2.63 and E288^7.39 in the minor subpocket [3]. In contrast, a moderate spatial overlap observed between the N-terminus of vMIP-II and small cyclic peptide antagonist CVX15 (PDB: 3OE0), however, showed polar contacts with the same residues such as D187 from ECL2 and D262^6.58 [4]. Our GaMD simulations revealed simultaneous occupancy of PLX in both the major and minor subpockets of the receptor with its two cyclam rings.

PLX is clinically used for hematopoietoc stem cells mobilization to treat cancer patients, however, being a selective inhibitor of CXCR4, has not yet been approved to treat HIV infection due to its side effects [47,48]. In this study, we have explored atomistic drug interactions with the CXCR4 through GaMD simulations. Detailed analysis of the intermediate states I1 and I2 revealed a significant role of the receptor ECL2 and TM5–TM6 region in the recognition and binding of PLX to the CXCR4. Residues D187^ECL2, D193^5.32 and D262^6.58 that formed salt bridge interactions with the PLX were suggested to be the recognizing determinants in the intermediate states of PLX binding.

The bound conformation of PLX captured in this study involved interactions with important negatively charged residues D97^2.63, D262^6.58 and E288^7.39 in the orthosteric pocket of the CXCR4 as also seen in the receptor crystal structures bound by the small molecule (IT1t) [4], viral chemokine (vMIP-II) [3] and small cyclic peptide (CVX15) [4] antagonists. Therefore, through the GaMD simulations, we have elucidated the role of PLX to be an antagonist of the CXCR4 receptor.

Conclusion

We have captured spontaneous binding of the PLX drug to the CXCR4 receptor through nonbonded dual-boost GaMD simulations. The GaMD simulations have revealed the mechanism and pathway of drug binding to the CXCR4. The newly identified intermediate binding site involving residues D187^ECL2, D193^5.32 and D262^6.58 in the ECL2–TM5–TM6 region will serve as a novel target site for designing allosteric modulators of the CXCR4. The simulation findings have provided a molecular basis for rational computer-aided drug design against HIV and other diseases associated with the CXCR4 chemokine receptor.

Future perspective

Chemokine receptors are key GPCRs involved in controlling cell migration during immune response, cardiovascular and central nervous system development and are associated with inflammation and cancer. The CXCR4 chemokine receptors facilitate HIV entry into host immune cells and serve as an important therapeutic target against HIV infection. Development of antagonists against the CXCR4 has often caused off-target side effects. Discovery of novel allosteric sites of the receptor has become promising for structure-based design of selective drugs. Together with previous mutation experiments, the molecular docking and GaMD enhanced simulations in this study strongly suggest that the PLX drug molecule is bound to the deeply buried orthosteric site of CXCR4 via an intermediate allosteric site at the receptor extracellular mouth.

While GaMD simulations have successfully captured complete binding of PLX to the deeply buried orthosteric site of CXCR4, the calculated free energy profile is not converged. Further studies are still needed to simulate multiple events of drug binding and even slower unbinding in the future, which is expected to enable accurate free energy calculations. In addition, the newly identified allosteric site will be targeted for computer-aided structure-based design of potent allosteric modulators of the CXCR4. Virtual screening of large compound libraries to identify novel allosteric modulators, followed by in vitro and in vivo experimental validation, will serve as a promising approach for developing effective therapeutics against the CXCR4.

Summary points.

• Molecular docking and all-atom simulations using Gaussian accelerated molecular dynamics were combined to determine the mechanism and target site of the plerixafor drug binding in the CXCR4 receptor.

• The docking and simulation findings were consistent with previous mutation experiments, suggesting that plerixafor binds to the orthosteric site of CXCR4 as an antagonist.

• The Gaussian accelerated molecular dynamics; simulations further revealed an intermediate allosteric binding site at the extracellular mouth involving residues from the ECL2–TM5–TM6 of CXCR4.

• The newly identified allosteric site can be targeted for novel drug design targeting CXCR4 and other chemokine receptors.