Abstract
Water molecules inside G-protein coupled receptor have recently been spotlighted in a series of crystal structures. To decipher the dynamics and functional roles of internal waters in GPCR activity, we studied A2A adenosine receptor using μsec-molecular dynamics simulations. Our study finds that the amount of water flux across the transmembrane (TM) domain varies depending on the receptor state, and that the water molecules of the TM channel in the active state flow three times slower than those in the inactive state. Depending on the location in solvent-protein interface as well as the receptor state, the average residence time of water in each residue varies from psec to nsec. Especially, water molecules, exhibiting ultraslow relaxation ( nsec) in the active state, are found around the microswitch residues that are considered activity hotspots for GPCR function. A continuous allosteric network spanning the TM domain, arising from water-mediated contacts, is unique in the active state, underscoring the importance of slow waters in the GPCR activation.
INTRODUCTION
The activation of GPCRs is associated with a conformational change of the receptor modulated by an agonist binding. As suggested by the X-ray crystal structure of β2-adrenergic receptor, the outward tilting of TM5 and TM6 helices that accompanies a helix rotation of the cytoplasmic part of TM5-ICL3-TM6 is one of the key structural changes that facilitate G-protein recruitment [1–3]. More microscopically, orchestrated dynamics among a common set of highly conserved fingerprint residues, dubbed as microswitches, of the class A GPCR family is critical for the regulation of GPCR activity [4, 5]. These microswitches, including three ubiquitous sequence motifs, DRY, CWxP, and NPxxY (where ‘x’ denotes any amino acid residue), adopt distinct rotameric configurations in their side chains, which are sensitive to the type of a cognate ligand bound to the orthosteric site [6]. In the agonist-bound state, the dynamic correlations among microswitches allow the receptor to adopt and maintain the active conformation, which facilitate accommodation of G-protein [6].
Besides the allosteric network defined by the interresidue contacts [6, 7], water molecules can also contribute to the formation of the intra-receptor signaling network. Given that GPCRs are the key therapeutic targets that sense and process external stimuli with an exquisite precision, it is critical to understand the roles played by water as well as the residue dynamics in GPCR activation. As is well appreciated by many studies, water is essential for structure, dynamics, and function of biomolecules [8–10]. Water molecules that form hydration shell on protein surfaces [8, 11–13] or a lipid-mediated effect of water via hydrophobic matching to the protein surface [14] have long been studied. Especially, possible functional roles of internal water molecules found inside TM domain in GPCR activation have recently been discussed [15–19]. Recently, MD simulations studies have analyzed water in various type of GPCRs, including A2AAR [15, 20–22], β2-adrenergic receptor [23], rhodopsin [17, 19, 24, 25], dopaminergic receptor [26], and μ-opioid receptor [16]; yet the foci of these studies were mostly on mapping the locations of internal waters and metal ions inside TM domain, or on the local dynamics of water specific to the ligand binding cleft in the process of ligand-receptor recognition [22]. Dynamical properties of internal water over the entire architecture of GPCR have not fully been discussed. While X-ray crystal structures provide a glimpse of ordered water molecules interacting with the interior of GPCRs [16, 18, 27, 28], these structural waters are not completely static but have finite lifetimes. Furthermore, the roles of mobile water with a relatively faster relaxation kinetics remain unknown. Thus, it is timely to examine the dynamics of water in GPCR and ask the functional roles of water in conjunction with the local dynamics of amino acid residues.
Here, we explored water dynamics in A2AAR by conducting μsec all-atom MD simulations. In order to assess the role of water dynamics in GPCR function, we strategically evaluated various quantities. We first calculated the water capacity in the TM region, and mapped the probability density of water for each receptor state. Explicit calculation of water flux through the TM region has revealed that the water flux in the agonist (UK432097)-bound active form is three times smaller than that in the apo or antagonist (ZM241385)-bound form. Water streams are divided into multiple pathways, some of which form one-dimensional “water wire” [29–31]. The receptor surfaces, mapped with the water relaxation time, indicate that the water dynamics is heterogeneous in space. Waters in the extra- and intra-cellular domains move fast (≲ 1 nsec). Especially in the agonist-bound form, waters trapped around the microswitches display ultra-slow relaxation (≳ 100 nsec), coating the polar and charged surface of the TM channel. Our study finally shows that water-mediated inter-residue interaction network, formed along microswitches in TM7, reinforces and extends the range of allosteric interface in the active state, aiding robust activation of GPCRs.
MATERIALS AND METHODS
Preparation of the receptor structures complexed with ligands
The X-ray crystal structures of A2aAR bound with an agonist or antagonist were modeled based on the structures from the Protein Data Bank (PDB) [32, 33]. We used the X-ray crystal structures with PDB entries of 3QAK [32] for agonist-bound and 3EML [33] for antagonist-bound model, and used UK432097 and ZM241385 for agonist and antagonist ligands, respectively. Since some of the loop regions are not resolved in the crystal structures, we performed homology modeling using the MODELLER program implemented in Discovery Studio v.3.1 to prepare the full-length A2AAR models including all the loop regions. The X-ray crystal structures with PDB entries of 2YDV [3] and 3PWH [34] were used to model the loop regions missing in 3QAK and 3EML, and the conserved disulfide bridges connecting the loops, i.e., C71-C159, C74-C146, C77-C166, and C259-C262, were retained. These models were optimized by simulated annealing and selected based on the DOPE score. The final structures were obtained by energy-minimization using the Conjugate Gradient method and the Generalized Born with simple Switching (GBSW) implicit solvent model.
Although an antagonist-bound engineered crystal structure of A2AAR complexed with apocytochrome b562RIL replacing intracellular loop 3 (ICL3) (PDB entry: 4EIY [27]), whose structure was fully determined with no missing region including extracellular (EC) and intracellular (IC) loops, became available after we started this research, the extent of structural overlap between the crystal structure and our modeled structure is as high as Cα-RMSD = 0.407 Å. Furthermore, as the memory of initial configuration, especially the initial configuration of “flexible” loops, would be erased after some time (at most a few tens of nsec), and an “identical force field” is used for simulation, the conclusion of this study would not be altered even when a more precise and accurate structure is used as a starting conformation for the simulation run. For the apo form, we used the protein structure, minimized after removing the antagonist ligand from the original structure (3EML) as a starting conformation of our MD simulations.
Molecular dynamics simulations
To construct the explicit membrane system, the transmembrane region of A2AAR was predicted based on the Orientations of Proteins in Membranes (OPM) database and the hydropho-bicity of the protein. Using SOLVATE 1.0 (http://www.mpibpc.mpg.de/home/grubmueller/solvate) we first solvated the receptor structure with TIP3P water molecules and removed the water molecules outside the receptor in the TM region. Then, the 1-Palmitoyl-2-oleoylphosphatidylcholine (POPC) lipid bilayer (88 × 91 A2 wide) was placed around the TM region of A2AAR. After aligning the initially solvated receptor structure and the bilayer system, we removed the lipid and water molecules overlapping within the distance of 0.5 A from the receptor structure. The prepared system was solvated with the water box, covering all the previous output molecules, and neutralized with K+/Cl− ions to make 150 mM salt concentration. Our system contains 68,799 (apo, 13,549 water + 173 lipid), 70,052 (antagonist-bound, 13.551 water + 182 lipid), and 69,449 (agonist-bound, 13.552 water + 177 lipid) atoms in 85 × 88 × 99 A3 periodic box.
MD simulations were performed using NAMD package v.2.8 [35] with CHARMM22/CMAP force field [36]. The topology and parameter files for the ligands were generated by SwissParam web server which provide topologies and parameters for ligand, based on Merck Molecular Force Field (MMFF), compatible with the CHARMM force field [37]. The simulations were then conducted with the following steps: (i) energy-minimization for 2,000 steps using conjugate gradient method in the order of membrane, water molecules, protein structure, and the whole system; (ii) gradual heating from 0 K to 300 K using a 0.01 K interval at each step; (iii) 50 nsec pre-equilibration with NVT ensemble before the production runs; and (iv) 1.2 μsec production runs with NPT ensemble (T = 300 K, P = 1 atm). Pressure was maintained at P = 1 atm using the Nosé-Hoover Langevin piston method with an oscillation period of 200 fsec and a decay time of 50 fsec. The temperature was controlled by applying Langevin forces to all heavy atoms with a damping constant of 1 psec−1. Non-bonded interactions were truncated with a 12.0 Å cutoff and a 10.0 A switching distance, and the Particle Mesh Ewald summation with a grid spacing 0.91 Å was used to handle electrostatic interactions. We used a uniform integration time step of 1 fsec, and during the whole production runs, no specific constraints were applied to the system except for restraining the hydrogen atoms. SETTLE algorithm was used to keep water molecules rigid, and SHAKE to restrain the bond lengths of hydrogens other than water molecules. The atomic coordinates of the simulated system were used every 50 psec for water flux analysis and 400 psec for other calculations.
Importantly, in the production runs it takes more than 200 – 300 nsec to fully equilibrate the (recep-tor+lipid bilayer) system in NPT ensemble. Water molecules fill the TM channel gradually (Figure 1). While the protein RMSD gets to its steady state at t ≲ 200 nsec (Figure S1A), the area per lipid or the bilayer thickness is still under relaxation in the first 200 – 300 nsec (see Figure S1B and S1C). In the steady state, the area per lipid (APL) for POPC molecule is ~ 61 ± 2 Å2 and bilayer thickness is ~ 40 A, which compare well with the experimental values at T = 27°C (APL ~ 63 ± 2 Å2, thickness ~ 39 ± 1 Å) as well as with other simulation studies [38, 39]. After the first t = 0 – 300 nsec simulation, we restrained the xy-plane area of the simulation box by using the option of ‘useConstantArea’ in NAMD package. As a result, APL and thickness of lipid bilayer are stably maintained for t ≥ 300 nsec. In our study, the production runs of the first 0 – 400 nsec, which could usually be regarded a long simulation time, are still in the process of nonequilibrium relaxation. Thus, we conducted various analyses in this study using the data from t ≥ 400 nsec after the system had reached the steady state.
Time correlation function to probe water unbinding kinetics
To quantitatively probe the dynamics of water near the receptor structure, we calculated the auto-correlation function [40, 41]:
The function h(t) describes whether a contact is formed between an atom in the receptor and a water molecule at time t. h(t) = 1 for a water molecule within 4 Å from any heavy atom of a residue α at time t; otherwise, h(t) = 0. 〈…〉 denotes a time average along the trajectory and an ensemble average over the number of water molecules (Nα) associated with the residue α. Since h(0) = 1, C(t) is equivalent to the survival probability of a bound water at time t. Thus, the average lifetime of water is obtained using .
Betweenness centrality
Simplifying a structure of complex system into a graph, represented with vertices and edges, can be used as powerful means to extract key structural features of the system [42–45]. The graph theoretical analysis can be performed for protein structures [46–51], so as to judge the importance of each vertex that defines the graph. For instance, a residue in contact with many other surrounding residues, which in the language of graph theory is called a vertex with high degree centrality, could be deemed important for characterizing a main feature of the given structure. However, in order to address the issue of allostery, the betweenness centrality can be used for quantifying the extent to which a node (v) has control over the transmission of information between nodes in the network [45]. The betweenness centrality is defined as with s,t ≠ ν. σst is the number of paths with the shortest distance linking the nodes s and t, and σst (v) is the number of minimal paths linking the nodes s and t via the node ν.
To build a network (graph), the residue-residue contact was defined (i) for water-free contact using the threshold distance of 4 Å between any two heavy atoms of two residues, and (ii) for water-mediated contacts between two residues if a water oxygen is shared within 3.5 Å by any two heavy atoms of two residues or if the two residues are within 4 Å threshold distance as in (i). To quantify the importance of the ν-th residue in mediating signal transmission, we calculated two distinct betweenness centralities; (i) for water-free and (ii) for water-mediated inter-residue network of GPCR. and calculated for each snapshot of molecular structure were averaged over the ensemble of structures obtained along the MD trajectories. To calculate CB (ν), we employed Brandes algorithm [52], which substantially reduces the computational cost of Eq. 2.
RESULTS
Water capacity in the TM region
Our simulations show that it takes more than 200 – 300 nsec for water molecules to fill the empty TM channel and to reach the steady state (Figure 1A, see also Figure S1 for the time evolutions of the receptor RMSD, area per POPC lipid, and membrane thickness). In the steady state, on average 121±20, 69±18, and 63±16 water molecules are found in the interior (between the phosphate atoms of upper and lower leaflets) of the apo, agonist-bound active, and antagonist-bound inactive states, respectively (Figure 1A). As depicted in the probability density map of water (Figure 1B), the apo form, due to the water-filled ligand binding cleft, contains twice the volume of the water compared with the other ligand-bound states. In the ligand-bound states, a large volume of water has to be discharged from the vestibule. The expanded G-protein binding cleft is also seen in the agonist-bound active state (Figure 1B, middle). A water cluster, whose functional role will be further discussed in the Discussion, is identified in the midst of TM domain (see the white arrow in Figure 1B, middle).
Water flux through the TM pore
Although the average number of water in the TM channel remains constant in the steady state (Figure 1A, t ≳ 400 ns), this does not imply that water is static inside the channel. The number of waters in and out of the channel are balanced in the steady state. In order to compare the water flux between the different receptor states, we traced the coordinate of every water molecule along the axis (Z-axis) perpendicular to the lipid bilayer plane and counted the number of water molecules that traverse through the channel from EC to IC or from IC to EC domain (Figure 2). In the early stage of simulations, non-steady state fluxes are observed in the apo and antagonist-bound forms. Thus, we excluded the first 400 nsec from the analysis. For t > 400 ns, the water fluxes in the two directions satisfy jEC→IC ≈ jIC→EC in all the receptor states. Overall, the water flux of the antagonist-bound state is three fold greater (jantago ~ 30 μsec−1) than that of the agonist-bound state (jago ~ 10 μsec−1). The water flux of the apo state lies in between (japo ~ 20 μsec−1). In fact, the lowest water flux in the agonist-bound form has its molecular origin. Below we will map each receptor state by probing the relaxation kinetics of water around each residue, which will help elucidate the role of water in receptor activation as well as the molecular origin of differential water fluxes.
Relaxation kinetics of water
We mapped the water dynamics on the receptor surfaces by computing time correlation functions of water from each residue (C(t), see Materials and Methods) (Figure 3A-B). Water in the vicinity of the IC or EC loops, exposed to the bulk, is expected to have a shorter lifetime. In the TM region, in contrast, dynamics of water is much slower (Figure 3C), displaying a broad spectrum of relaxation times depending on (i) the receptor state and (ii) their locations (Figure 3D-F). The average lifetimes of water are 14.5 nsec, 26.3 nsec, and 7.1 nsec for the apo, agonist, and antagonist-bound states, respectively. The water molecules, hydrating the TM1, TM6, and TM7 helices in the agonist-bound form, reside longer than 100 nsec especially around the microswitches (marked with the cyan circles at the top of Figure 3D-F). This long residence time of water in the interior of TM domain ( nsec) is noteworthy, given that a typical water lifetime on biomolecular surface probed by spin-label measurement is psec [53, 54].
Similar conclusions as the above analysis calculating relaxation times were drawn by calculating the Fano factor involving water number fluctuation around each residue (see SI text and Figure S2).
DISCUSSION
Detailed look into water flux across the TM domain
In order to glean the molecular origin of receptor state-dependent water flux, we first visualize the geometry of water channel. The radius of water channel, calculated using the program HOLE [55] (see Figure 4A), reveals that the geometrical bottleneck (r ≈ 0) in the midst of TM domain (Z ≈ 0) (Figure 4A, see the black arrow) is formed around W2466.48 and Y2887.53 in the agonist-bound active form. This is compatible with our finding that the water flux is substantially reduced in the agonist-bound form (Figure 2). W2466.48, a key microswitch that senses an agonist and relays its signal to other microswitches [6, 56], is located deep inside the orthosteric binding vestibule, regulates the entry of water from the EC domain; whereas Y2887.53 is located at the lower part of TM channel, regulating the entry of water from the IC domain.
Although the geometry of channel in each receptor state provides a glimpse of pipeline across the TM region, the actual water flux through the channel is not fully explained by the radii of the channel alone. For example, the apo state, overall, has greater radii along the channel and does not have a particularly more restrictive geometrical bottleneck than those in the antagonist-bound state (Figure 4A); yet the flux is smaller than the one observed in the antagonist-bound state. Depending on the extent of hydropho-bicity or electrostatic nature of residues comprising each region of the channel surface, stochastic wetting-dewetting transition [30, 57, 58] can occur along the channel. Furthermore, a stable water cluster in the channel, not exchanging water molecules with the surroundings, would impede the water dynamics through the channel. Explicit calculations of water occupancy along the Z-axis [58] visualize how water molecules actually fill the TM channel at time t (Figure 4B). As expected, in the apo state (Figure 4B, top) the EC region (Z > 0 Å) is always filled with a high density of water since the empty ligand binding pocket is accessible from the bulk; however, IC region (−20 ≲ Z ≲ −10 Å) remains “dry”, indicating that the entry of water through the IC region is blocked. This dry zone corresponds to the second hydrophobic layer (HL2) around NPxxY motif above Y2887.53, illustrated by Yuan et al. [15]. In the agonist-bound form, another water-free layer is observed right above the HL2 (−10 < Z < −5 Å) (Figure 4B, middle). This is the region below which the water cluster is formed. On the other hand, the first hydrophobic layer (HL1) [15], corresponding to another dry zone between the orthosteric and the al-losteric sites [15](Z ≈ 0 Å) is observed in the inactive state. Notably, despite the HL1, the water occupancy map of the antagonist-bound form (Figure 4B, bot tom) finds multiple instances (t = 600 – 800 ns) that both HL1 and HL2 are filled with water bridging across the entire TM region. This accounts for the greater water flux in the inactive state.
The location of the hydrophobic layer and the receptor state-dependent water flux are affected by the alignment of polar residues along the channel. The TM domain is mostly composed of non-polar hydrophobic residues, but there are polar/charged amino acids buried inside the TM domain as well. The streams of water molecules are found along an “Y” shaped array of these polar and charged residues that bridge through the TM domain (Figure S3). This array of polar/charged residues corresponds to the buried ionizable networks in GPCRs which have recently been underscored [59]. Our study reveals that these networks shape the passages of water molecules through the TM domain. A misalignment of polar residues in the agonist-bound state is led to dehydration of the IC zone around −20 < Z < −10 Å, which gives rise to the HL2 (Figure S3 left). The rotameric state of Y288753 sidechain in the antagonist-bound form enables a single file of water molecules constituting a water wire to flow across the HL2 (Figure S3 right, SI Movie M3. See also Figure S4). The formation of the Y-shaped bridge made of polar residues including the NPxxY motif is the molecular origin underlying the “hydrophobic gating” [60, 61] that regulates the water flux through the IC region of A2AAR.
Microscopic origin of receptor state-dependent water flux
To further glean the microscopic underpinnings of the receptor state-dependent water flux (Figure 2), configurations of microswitches at three key locations in the TM domain are probed (Figure 5): (i) The ionic lock (R1023.50-E2286.30) (Figure 5A, top panel), the hallmark of the inactive state of GPCRs, is intact in the antagonist-bound form, maintaining the inter-residue distance dR102-E228 ≈ 2.5 Å. In the apo form, the ionic-lock repeatedly disrupts and rebinds, suggestive of the receptor’s basal activity [62]. In the agonist-bound form, the ionic lock is completely disrupted; (ii) The distance between R1023’50 and Y2887.53 depends on the receptor state (Figure 5A, the second panel from the top), and importantly Y2887.53 gates the entry of water from the IC region. In the active form, R1023.50 released from the influence of E2286.30 can interact with and stabilize the side chain orientation of Y2887.53, resulting in blocking the passage of water stream as well as misaligning the bridge made of NPxxY motif (see Supporting Movie M2, Figure 5B, Figure S3); (iii) The side chain of W2466.48 residue gates the entry of water from the EC domain. In the active form, W2466.48 blocks the water passage from the EC region, but it allows water to flow more freely in the inactive form. In the apo form, the rotamer angle of W2466.48 undergoes sharp transitions ( in Figure 5A, black trace) multiple times during the simulation (500 ≲ t ≲ 1000 ns), displaying correlations with the ionic-lock (dR102-E228 in Figure 5A, black line) and with the increased level of water flux (notice the sudden increase of the flux at t ≈ 500 nsec from EC to IC (black trace in solid line in Figure 2)). For example, when the ionic-lock was stabilized at ≈ 1000 nsec in the apo form (Figure 5A, top panel, black trace), the rotameric angle of W2466.48 also displayed a sharp change from −90° to +90° (cyan arrow in the plot of Figure 5A). This is the moment when jIC→EC in the apo form has increased (t > 1000 nsec in Figure 2).
The movies (SI Movies M1, M2, and M3) from simulations of the water dynamics across TM region provide nice visualizations of receptor state-dependent water flux, which is recapitulated in the cartoons in Figure 6. In the apo form, the water molecules freely navigate the wide volume of the empty ligand binding cleft in the EC domain, but a further penetration across the TM region is regulated by the narrow channel gated by W2466.48. When the agonist or antagonist occupies the binding cleft, however, water flux is divided into the major (solid lines in Figure 6) and minor streams (dashed lines in Figure 6); the major stream is formed between TM1, 2, and 7, and the minor stream is formed between TM3, 5, and 6. In the agonist-bound form, the W2466.48 blocks the minor stream and the water flow along the major stream is tightly regulated by the several other microswitch residues (N241.50, D522.50, N2807.45, S2817.46, and N2847.49), which creates the stable water cluster. In case of the antagonist-bound form, no water cluster is observed; W2466.48 gate is open and lets water molecules flow in and out of the TM channel. The water flux from the IC region is regulated by the side chain configuration of Y2887.53. In the active state whose ionic-lock is disrupted, R1023.50 interacts with Y2887.53, which in turn blocks the passage of water flux in the IC domain.
Allosteric interface reinforced by water-mediated interactions
In order to underscore the contribution of water-mediated contacts between TM residues to the receptor’s allosteric signaling and function we conducted a graph theoretical analysis on the ensemble of GPCR structures. As shown by our previous study [7] the microswitches of GPCRs in general are identified by graph theoretical analysis using betweenness centrality to be the key sites for intra-molecular orthosteric (allosteric) signal transmission (see Methods). Allosteric interface can be visualized by highlighting those allosteric hotspots [7]. While waters inside channel are generally dynamic, some water molecules, especially around microswitches in the active state, display slow relaxation kinetics and even can be trapped for an extended amount of time ( ns) (see SI text, Figure S5 and Figure S6). As long as water dynamics is sufficiently slow, stable water-mediated contacts can be made between two residues that are not in direct contact. Defining a water-mediated contact when two residues share a water oxygen within 3.5 Å from any heavy atom in each residue, we constructed a water-mediated residue interaction network for a given structure. Using an ensemble of structure obtained from simulations, we calculated an average betweenness centrality (CB(ν)) at the ν-th residue (see SI text, Figure 7A). At present, there are many ways to consider the protein allostery; some of them consider thermodynamic aspect of protein allostery [63, 64], and others focus more on identification of allosteric hotspot of a given structure [65–67]. The graph theoretical method [7, 68<70] can also be employed to identify an allosteric hotspot of a given network structure, and in this method a residue with high CB value corresponds to a site important for allosteric signal transmission [7].
In the agonist-bound active state, the CB (ν) values calculated with and without water-mediated contacts (see Figure S7) show clear differences (ΔCB(ν)) along the microswitches in TM7 (Figure 7B, C), which is in accord with our findings that a number of slow water molecules stably coordinating with microswitches along the water channels are present in the active state. Highlighted in Figure 7C with magenta surface is the allosteric interface in the active state reinforced by water-mediated interactions (ΔCB(ν) > 0.03). The interface mainly formed from along the TM7 helix, reaches R1023.50 in the TM3 helix via Y2887.53 and A2316.33, spanning the whole TM domain. We surmise that this wide-spread interface across the TM domain enables a robust long-range “signal transmission” (compare the map of agonist-bound active state with those for apo and antagonist-bound inactive states in Figure 7C). Notably, recent calculation of energy tranport in homodimeric hemoglobin also underscores the importance of interface water cluster, substantiating our proposal of ultra-slow water mediated allosteric signaling [71].
CONCLUSION
Systematic analyses of GPCR structures reveal the network of inter-TM contacts mediated by microswitches [28]. Network analysis also put forward that these microswitches act as the hub of the intrareceptor signaling network of A2AAR [7], and analyses of MD simulation trajectories confirmed that dynamics of microswitches occurs in concert [6]. Here, we extended our analysis to the dynamics of water molecules traversing through the TM channel and investigated their role in allosteric (orthosteric) signaling.
As is well appreciated, water, an essential component of living systems, provides driving force for selfassemblies of biomolecules and enhances their conformational fluctuations [11, 41, 72–74]. Without water, biomolecules cannot function [10]. Guided by the array of polar residues, water permeates inside the mostly dry and hydrophobic pore of GPCRs. In the antagonist-bound inactive state, this array of polar residues is connected continuously from the EC to IC domain. In accord with Yuan et al. [15, 21], our simulations confirm the presence of the two hydrophobic layers, HL1 and HL2; however, these hydrophobic layers are not static, but highly dynamic. Our study also confirms the Yuan et al.’s finding [21] that the water flux along the TM channel is regulated by the rotameric states of several microswitches (Figure 6). Among them, two key microswitches, W2466 48 and Y2887 53 act together to form an “AND-gate” controlling the water flow.
There are also other MD simulation studies on GPCRs (e.g. rhodopsin [19]) which propose that a continuous stream of internal water is important for GPCR activation. By counting the number of water molecules inside TM domain, Leioatts et al. [19] showed that upon an elongation of retinal an influx of water increases inside the TM domain rhodopsin. They found that the increase of hydration level was significant in the complex-counterion simulation of rhodopsin, but not in the dark-state. Their simulation results on water hydration in the dark-state of rhodopsin differ from A2AAR in the inactive state in that only 20 – 30 water molecules are allowed inside the TM domain during the 1.6 μsec simulation time [19]. However, the hydration level in the complex-counterion form is consistent with our study on the active form of A2AAR in terms of the number of water molecules that fills the TM channel. In our study, both active and inactive states of A2A AR could accommodate approximately an equivalent amount of water molecules, 60 – 100, at steady state (Figure 1A), but the water flux in the inactive form was found greater than that in the inactive form by three times (Figure 2). Here, it is crucial to distinguish “the flux of water across TM domain” (Figure 2) from “the number of water inside TM domain” (Figure 1). As described in Results, the water flux across TM domain was calculated by tracing the individual water molecule and counting the number of waters that enter and exit from one side of the membrane to the other. It is not merely the increase of TM water with time. While the number of water molecule inside the dark-state of rhodopsin is smaller than that in the inactive state of A2AAR, the water flux in the dark-state of rhodopsin could also be large. Our study puts more emphasis on the dynamic aspect of water molecules across TM domain in the steady state, which leads us to further ask the questions of which residues are hydrated by slow water and how those slow waters contribute to the water-contact mediated allosteric network.
Our explicit calculations of water fluxes indicate that the continuous water stream can be formed in all three states (i.e., j = 0 in Figure 2), but with differing degrees. Thus, we want to argue that what is more relevant for GPCR activation is the water-mediated contacts among the key allosteric residues than the existence of continuous water stream. It is easier to make a water-mediated contact if a water molecule is slower. In the agonist-bound active state, water molecules inside the TM channel are almost stagnant, displaying minimal flux (Figure 2); they can stably hydrate the microswitches mainly along the TM7 helix. We show that water-mediated residue network extends from extracellular domain to the intracellular part of TM6 helix via TM3 helix (Figure 3 and Figure 7).
The waters around TM microswitches, some of which constitute the water cluster, stabilize the relative orientation and distance between TM helices by bridging them together (Figure 7 and Figure S7). Our study highlights the interactions of internal water with microswitches, which contribute to extending and reinforcing the allosteric interface of GPCRs (Figure 7). Our study puts forward that these interactions are especially critical for the functional fidelity of the GPCR activity.
SUPPORTING MOVIES
Supporting Movie M1
Water dynamics in the apo form during the time interval t = 1100 – 1150 ns. All the water oxygens are shown with small spheres in different colors. The key micro-switch residues are depicted using the stick representation marked with their residue numbers.
Supporting Movie M2
Water dynamics in the agonist-bound form during the time interval t = 700 – 750 ns. The details of the representation are identical with Movie M1.
Supporting Movie M3
Water dynamics in the antagonist-bound form during the time interval t = 700 – 750 ns. The details of the representation are identical with Movie M1.
Supporting Movie M4
Dynamics of two water molecules in the apo form around the for the time interval t = 821 – 835 ns. The receptor structure is represented as white ribbon, and two water molecules that compete around are depicted using spheres in cyan and green.
Acknowledgments
This work was supported by the grant from the National Leading Research Laboratory (NLRL) program (2011-0028885) funded by the Ministry of Science, ICT & Future Planning and the National Research Foundation of Korea (to S.C.), and by the RP-Grant 2015 funded by Ewha Womans University (to S.C. and Y.L.). We thank KIAS and KISTI Supercomputing Center for providing computing resources.
References
- [1].↵
- [2].
- [3].↵
- [4].↵
- [5].↵
- [6].↵
- [7].↵
- [8].↵
- [9].
- [10].↵
- [11].↵
- [12].
- [13].↵
- [14].↵
- [15].↵
- [16].↵
- [17].↵
- [18].↵
- [19].↵
- [20].↵
- [21].↵
- [22].↵
- [23].↵
- [24].↵
- [25].↵
- [26].↵
- [27].↵
- [28].↵
- [29].↵
- [30].↵
- [31].↵
- [32].↵
- [33].↵
- [34].↵
- [35].↵
- [36].↵
- [37].↵
- [38].↵
- [39].↵
- [40].↵
- [41].↵
- [42].↵
- [43].
- [44].
- [45].↵
- [46].↵
- [47].
- [48].
- [49].
- [50].
- [51].↵
- [52].↵
- [53].↵
- [54].↵
- [55].↵
- [56].↵
- [57].↵
- [58].↵
- [59].↵
- [60].↵
- [61].↵
- [62].↵
- [63].↵
- [64].↵
- [65].↵
- [66].
- [67].↵
- [68].↵
- [69].
- [70].↵
- [71].↵
- [72].↵
- [73].
- [74].↵
- [75].↵
- [76].
- [77].↵