|Home | About | Journals | Submit | Contact Us | Français|
The function of G-protein coupled receptors is tightly modulated by the lipid environment. Long timescale molecular dynamics simulations (totaling ~3 microsec) of the A2A receptor in cholesterol-free bilayers, with and without the antagonist ZM241385 bound, demonstrate an instability of helix II in the apo receptor in cholesterol-poor membrane regions. We directly observe that the effect of cholesterol binding is to stabilize helix II against a buckling type deformation, perhaps rationalizing the observation that the A2A receptor couples to G-protein only in the presence of cholesterol (Zezula and Freissmuth, 2008). The results suggest a mechanism by which the A2A receptor may function as a coincidence detector, activating only in the presence of both cholesterol and agonist. We also observed a previously hypothesized conformation of the tryptophan “rotameric switch” on helix VI in which a phenylalanine on helix V positions the tryptophan out of the ligand binding pocket.
G-protein coupled receptors (GPCRs) are responsible for transducing chemical signals across cell membranes in many diverse physiological contexts, comprising the largest family of proteins in the human genome. The exquisitely tuned specificity of GPCRs makes them ideal targets for drugs, and indeed they are the most frequent pharmacological target in signaling contexts (Schwartz and Hubbell, 2008). The entire family shares a structurally homologous seven transmembrane helix topology, with a high degree of sequence homology within GPCR subtypes. Until recently however, the only GPCR that had been diffracted to high resolution was bovine rhodopsin (Palczewski et al., 2000), and so much of what was inferred about the function of GPCRs was based on this one structure. Recently however several other class A GPCRs have been diffracted to high resolution, including the human A2A adenosine receptor (Jaakola et al., 2008), a human β2 adrenergic receptor (Cherezov et al., 2007; Rasmussen et al., 2007; Rosenbaum et al., 2007), a turkey β1 adrenergic receptor (Warne et al., 2008), a higher resolution structure of bovine rhodopsin (Okada et al., 2004), and bovine opsin (Scheerer et al., 2008). The focus of the present work is the A2A adenosine receptor, a pharmacologically important receptor that modulates the dopaminergic response (Dunwiddie and Masino, 2001; Jacobson and Gao, 2006). As activation of the A2A receptor generally results in inhibition, misfunction of this GPCR is implicated in neurological disorders such Parkinson's disease (Schapira et al., 2006) and epilepsy (During and Spencer, 1992).
The sudden profusion of structural information on class A GPCRs opens a window into the function of this exceptionally important class of receptors. Of particular interest are questions such as, “How does a particular receptor attain specificity, both for G-proteins and ligands?” And, “What is the process by which ligand binding leads to activation and binding of a G-protein to the intracellular loops?” It is known that GPCR signaling (and nonsignaling) is characterized by ligands with a spectrum of activities, and consequently GPCR activation is hypothesized to involve a number of conformational substates and intermediates (Kobilka and Deupi, 2007). These substates have been observed experimentally in rhodopsin (Altenbach et al., 2008; Ruprecht et al., 2004; Scheerer et al., 2008) and recently by Dror et. al. in simulations of the β2 adrenergic receptor (Dror et al., 2009). The observations made by Dror et. al. were made possible by recent algorithmic advances that allow MD simulations of modest sized proteins on microsecond timescales (Bowers et. al., 2006; Klepeis et al., 2009).
In tandem with the recent progress in obtaining high-resolution structures of several class A GPCRs, work on membrane proteins in general has made clear that specific protein-lipid and protein-sterol interactions are often of critical importance to the function of such proteins (Burger et al., 2000; Lee, 2004). This is true for GPCRs, as activities of some class A GPCRs have been demonstrated to be sensitive to cholesterol concentration (Gimpl et al., 1997; Pucadyil and Chattopadhyay, 2004), including the A2A receptor (Zezula and Freissmuth, 2008). In addition, structural (Hanson et al., 2008) and simulation (Grossfield et al., 2006; Pitman et al., 2005) studies have demonstrated specific lipid and sterol binding. It is therefore of great current interest to investigate the specific functional role that is played by lipids and sterols in GPCR signaling.
In the present work, microsecond timescale simulations of the A2A adenosine receptor yield insight into the structural rearrangements that occur upon binding of an antagonist. The simulations predict that helix II is destabilized in the absence of cholesterol, perhaps contributing to inactivity of the receptor. The fact that the site where the lipid gains entry to the binding pocket is in the same location as a recently identified specific cholesterol binding site on the β2 adrenergic receptor (Hanson et al., 2008) suggests a hypothesis about the role of competition between lipid and cholesterol in determining if the receptor is agonist-ready and therefore primed for activation. We therefore performed simulations with two cholesterol molecules bound at the cholesterol binding site. In this simulations, helix II is markedly stabilized. These results suggest a structural hypothesis about the role of cholesterol in stabilizing the receptor, and lead us to suggest several experiments to connect the structural hypothesis to A2A activity as a function of cholesterol.
Before examining the results for the apo A2A receptor, we present data based on the receptor as it was crystallized, with the antagonist ZM241385 bound (Jaakola et al., 2008). This data serves as a control for the apo study, as well as providing insight into the conformation of the ionic lock under physiological conditions. The ionic lock is the name given to a highly conserved D/ERY motif at the intracellular ends of helices III and VI. When closed, a salt bridge is formed that constrains the ends of the two helices, maintaining the receptor in an inactive conformation (Ballesteros et al., 2001). It was therefore surprising that the ionic lock was broken in the crystal structures of both the A2A (Jaakola et al., 2008) and β2 (Rasmussen et al., 2007) receptors bound to inactivating ligands. In both the β2 (Cherezov et al., 2007) and the A2A receptor structures, the highly flexible third intracellular loop (ICL3) was stabilized by replacing it with T4 lysozyme, and in another structure of the β2 receptor by binding it to a specifically engineered antibody (Rasmussen et al., 2007). Based on microsecond timescale simulations of the β2 receptor, Dror et al. rationalized the ionic lock conformation, demonstrating that it closes when the T4 lysozyme is removed from ICL3 (Dror et al., 2009).
The observation that the A2A receptor possesses much lower constitutive activity in the absence of agonist than β2 leads naturally to consideration of the ionic lock conformation in the T4L-free A2A receptor. We therefore removed the T4L and constructed the missing loop using Prime, developed by Schrödinger, L.L.C., and performed molecular dynamics simulations of the T4L-free A2A receptor in a palmitoyloleoylphosphatidylcholine bilayer with explicit water (details in Methods). We first studied the receptor as it was crystallized, with the antagonist ZM241385 bound; results for the apo receptor are presented in the next section. The data in Figure 1 demonstrate stable closing of the ionic lock after about 375 nsec. (Note that absolute time scales must be interpreted with care, as discussed in Methods.) The lock that forms in the A2A appears to be tighter than that observed in the simulations of the β2 receptor — the Cα distance fluctuates around 7 Å, while the same distance in the β2 system never samples distances less than 8 Å — as well as in crystal structures of inactive rhodopsin. Also, the initial behavior is different from that observed for the β2 system; the lock samples many nearly closed conformations before forming stably, and takes about twice as long to form. It is not clear at this point whether these differences are significant, as the physical process is stochastic and is not likely to behave precisely the same way from one closing event to the next. However, it is noteworthy that in an independent trajectory (Fig. S1) the lock fluctuates around the closed state, emphasizing the stochastic nature of the process.
In order to gain insight into the mechanism of inactivation upon antagonist binding, we performed simulations of the A2A receptor in which the ZM241385 ligand was removed. A great deal of data has been collected on the movement of the seven transmembrane helices during activation, especially at the cytoplasmic face, where exhaustive spin-labeling techniques have mapped out in detail conformational changes associated with activation (Altenbach et al., 1999; Altenbach et al., 2008). Here however, we are interested in movements within the binding pocket, as we observed little change at the cytoplasmic face upon the transition from holo to apo, consistent with the observation that the apo A2A receptor has no constitutive activity (Jaakola et al., 2008). Information on critical sites for activation within the ligand binding pocket, and their relative distances in the active and inactive conformations, has been gathered by engineering metal ion binding sites in the ligand pocket of the β2, neurokinin-1, and κ-opioid receptors (Schwartz et al., 2006), so that metal ions can serve as either agonists (Elling et al., 1999), or antagonists (Elling et al., 1995). In both cases the data suggest that binding of either type of ligand draws the pocket inward, though clearly the details determine whether such motion results in activation.
Figure 2 demonstrates a very rapid and dramatic opening of the binding pocket upon removal of the ZM241385, suggesting that the ligand exerts considerable inward force via its contacts with helices III, V, VI, and VII. The opening of the binding pocket was also measured by computing its volume using SiteMap, a tool available in the Schrödinger suite which estimates volumes of binding pockets. In the holo simulation, the volume of the binding pocket fluctuates around 376 Å3, while in the apo simulation the volume increases to 500 Å3 and extends much deeper into the protein. Most of the motion observed upon removal of the ligand is localized to the binding pocket area, which is roughly bounded on the intracellular side by the three conserved prolines in helices V, VI, and VII. As far as we are aware, no distance data is available for the extracellular ends of the TM helices for the apo receptor. However, it is noteworthy that engineered metal ion binding sites that activate the receptor draw helix III together with either or both helices VI and VII, resulting in a corresponding opening of the intracellular portions of the TM helices (Schwartz et al., 2006; Schwartz and Hubbell, 2008). This process has been conceptualized by Schwartz and coworkers as a rigid body motion of the helices around the conserved proline kinks, the so-called “global toggle switch model (Elling et al., 2006; Schwartz et al., 2006).” Our data only consider the conformational change associated with binding and unbinding of an antagonist, and therefore cannot directly address the global toggle switch, which is a model of activation. However, we do observe that the conserved prolines in helices V, VI, and VII soften the helices enough that the agonist can draw the pocket significantly inward compared to the apo receptor without activating the receptor. Presumably a small molecule agonist that binds in the space between the TM helices draws them in more tightly than the antagonist ZM241385, a hypothesis that is supported by the distance data from metal ion activation studies (Elling et al., 2006; Elling et al., 1995; Elling et al., 1999) and the observations that agonists tend to be smaller than antagonists for this class (Jacobson and Gao, 2006).
The highly conserved tryptophan residue on helix VI (position 246, numbering of amino acids is based on the human A2A sequence, accession number P29274) has been proposed to function as a rotameric switch (Shi et al., 2002), the position of which affects the properties of the kink induced by the neighboring proline. In the inactive conformation, Trp2466.48 is situated between helices III and VI (Figure 3, panel A), but rotates out of the binding pocket upon activation. Superscript X.YY denotes a residue on helix X at position YY relative to the most highly conserved residue on helix X, which is arbitrarily designated as position 50, with numbers decreasing/increasing toward the N/C terminus. This convention follows that developed by Ballesteros et. al. (Ballesteros and Weinstein, 1995). Figure 4 demonstrates that in the ZM241385 bound receptor, Trp2466.48 never changes rotameric state, as the furan ring of the ligand constrains its mobility. In the absence of the antagonist, however, Trp2466.48 samples both gauche+ and trans configurations, though it never returns to its original conformation. At 250 nsec, interaction with Phe1825.47 (described shortly) pulls Trp2466.48 about 40 degrees beyond the gauche+ conformer, maintaining this strained state for the duration of the simulation. Rotation of Trp2466.48 is accompanied by the breaking of a hydrophobic contact with Val552.44 on helix II (Figure 3 panels A and B), which facilitates the opening of the receptor. Interestingly, the combination of Trp2466.48 and Phe1825.47 on helix V samples conformations that are hypothesized to be associated with the active conformation (Figure 3, panel B and C). It has been suggested that the highly conserved Phe1825.47 is positioned to form and edge-to-face aromatic interaction with Trp2466.48, keeping it in its “active” rotameric state (Schwartz et al., 2006). Indeed, mutation of the conserved Phe1825.47 to Ala has been observed to eliminate the signaling of the otherwise highly constitutively active ghrelin receptor (Schwartz et al., 2006). Figure 3C clearly demonstrates that Trp2466.48 does in fact spend the majority of the time in this configuration. This begs the question, why does the apo A2A receptor possess no constitutive activity?
A close look at helix II in Figure 2C and D reveals that it has undergone a significant change in conformation after the binding pocket has opened and several hundred nanoseconds have elapsed. As a result of the loss of the stabilizing interaction with Trp2466.48, the region of helix II between Gly562.54 and a Pro612.59 becomes quite distorted, as several hydrophobic side chains are pulled into the core of the bilayer. A closer inspection reveals a striking development, shown in Figure 5 — a gap has opened between helices I and II, and a lipid head group has inserted through the gap into the empty, open binding site. The lipid remains there for the duration of the simulation, prying apart the binding pocket and stabilizing the open conformation. (Similar results were obtained in independent simulations, Fig. S2.) The data in Fig. 6 show that binding of the lipid holds the intracellular ends of helices III and VI within 8 Å of each other, consistent with an inactive receptor. However, the ionic lock forms only transiently, as indicated by the N – O distance, which fluctuates considerably more than in the ZM241385 bound receptor (Figure 1). This suggests that if the receptor is inactivated by binding a lipid headgroup, stable formation of the ionic lock is not required, instead the ends of helices III and VI are held in close proximity, allowing transient formation of the lock. Considering that a wide variety of ligands, from metal ions (Schwartz et al., 2006) to small molecules (Jacobson and Gao, 2006) to peptides (Schwartz, 1994), are able to inactivate GPCRs by binding to various sites within the binding pocket and on the extracellular loops (Schwartz, 1994), it seems reasonable that an inactive conformation could be observed without the ionic lock formed. This notion is also corroborated by recent simulations of the β2 receptor (Dror et al., 2009) and by an independent simulation of the ligand bound A2A receptor, in which the ionic lock is also transiently formed (Fig. S1). It ought to be stressed however that the events observed are stochastic, and that the timescales simulated are not sufficient to observe activation of the receptor, and so these data are by themselves not conclusive evidence for an inactivated state.
The lipid that inserts into the ligand binding site through the gap between helices I and II sits in a pocket on the membrane-exposed face of the receptor, between helices I, II, III, and IV — a location which was predicted to be a cholesterol binding site based on msec simulations of rhodopsin (Grossfield et al., 2006), and recently observed to bind cholesterol on the β2 receptor (Hanson et al., 2008). The thermal stability of the β2 receptor was observed to increase in the presence of cholesterol due to improved packing in the cholesterol binding pocket. More than 40% of class A GPCRs are likely to specifically bind cholesterol at this site, as evaluated by scanning sequences for a cholesterol consensus motif that demands three residues to make specific interactions with two cholesterols in the cholesterol pocket. The A2A receptor in fact falls into an even more restrictive set of GPCRs that possess an additional residue that makes favorable interactions with the two cholesterols in the binding pocket (Hanson et al., 2008). Also in this class are two other receptors for which cholesterol greatly increases the affinity of agonists (Gimpl et al., 1997; Pucadyil and Chattopadhyay, 2004). This shines a quite interesting light on the observation that A2A does not couple to G proteins in the absence of cholesterol (Zezula and Freissmuth, 2008).
Within the cell membrane the local concentration of cholesterol is, of course, not constant. For those class A receptors that possess the cholesterol binding motif, it may have evolved as a coincidence detector (Zezula and Freissmuth, 2008), so that the receptor is only active when both cholesterol and agonist are present in sufficient quantity. The specific binding of a lipid headgroup in the ligand pocket observed in the absence of cholesterol in the present work would serve as a reversible means of deactivating the receptor — then, upon moving from a cholesterol poor to a cholesterol rich region, cholesterol would compete for the binding site between helices I II and III, displace the lipid, and prepare the receptor to receive an agonist. Indeed, the presence of agonist would compete with the lipid headgroup for binding in the ligand binding pocket, perhaps simultaneously. We therefore performed calculations with cholesterol bound in the site identified experimentally by Hanson et. al. (Hanson et al., 2008). While a direct calculation of the free energy difference between the cholesterol/lipid bound conformations is not computationally feasible, we can investigate the stability of Helix II in the presence of cholesterol.
We performed two independent simulations of apo A2A with cholesterol bound in the binding site between helices I, II, and III, in order to test the hypothesis that the cholesterol “protects” helix II from buckling, and prevents the entry of lipid headgroups into the binding pocket through a gap between helices I and II. The first observation is that the two cholesterols remain in the cholesterol binding site during both simulations, and show no indication that they will be displaced by lipid (Fig. 7). The two cholesterols make many contacts with helices I, II, and III, and overall have the effect of stabilizing the receptor, as measured by the average deviations from the initial structure. The most dramatic increase in stability is observed in helix II (Fig. 8), for which much smaller positional fluctuations are observed with cholesterol bound, as compared to the apo simulation without cholesterol. Based on these data, our hypothesis is that the specifically bound cholesterol stabilizes the receptor by preventing the buckling of helix II. This hypothesis is consistent with data recently published for the β2 receptor, where it was observed that stability against denaturation is increased by the presence of cholesterol.
With molecular dynamics simulation timescales of microseconds now within the reach of many research groups (Klepeis et al., 2009), predictions about features like the rotameric states of sidechains become reliable, and the conformational rearrangements of small loops, turns, and helices become observable. The present work follows recent successful predictions by μsec simulations of details of the activation of rhodopsin (Grossfield et al., 2008; Martinez-Mayorga et al., 2006) and of the dynamics of its ligand (Lau et al., 2007). The predictions of any simulation must always be viewed with the understanding that empirical force field development is an active area of research, and so should be considered provisional until corroborated by experiment and further simulation. We feel, however, that the softening of helix II and insertion of a lipid at that location, along with the observation that binding of cholesterol stiffens this same helix and prevents the entry of a lipid headgroup into the binding pocket merits consideration. One series of experiments might be performed to look more carefully at the cholesterol dependence of A2A activation. Zezula and Freissmuth have mentioned that coupling of A2A to Gs requires cholesterol, but it is not known to the authors whether this cholesterol dependence has been explored in any detail. A first experiment would examine the stability of the receptor as a function of cholesterol — since we propose that cholesterol binding stabilizes helix II, the receptor ought to be stabilized against chemical denaturation by cholesterol. A second experiment would examine the activity of the receptor as measured by G-protein binding as a function of cholesterol and agonist concentration. Perhaps at sufficiently high cholesterol concentration, some small constitutive activity of the receptor would be observed. Finally, a test of the underlying structural hypothesis might be performed by stiffening helix II, making it less likely to buckle and invite a lipid to bind in cholesterol poor membranes. If the bound lipid stabilizes the inactive form while making it more challenging for an agonist to bind, stiffening helix II should increase agonist binding efficiency in the absence of cholesterol and perhaps even introduce some constitutive activity. Stiffening helix II might be accomplished by mutating Gly562.54 to an Ala, greatly reducing the volume of backbone conformational space at this key position. Preliminary simulation data on this system are so far inconclusive, and will be continued in future work.
The present paper maintains that complementing GPCR structural biology with high performance computing deepens our understanding of the stochastic events along the pathways between active and inactive receptors. These results, other simulations suggesting the importance of specific interactions between rhodopsin and polyunsaturated fatty acids, and the recent structure of the β2 receptor bound to two cholesterol molecules all point to a rich future for studies focused on modulation of GPCR activity by the lipid environment. Further biochemical investigations and simulation studies will no doubt add to the story of how the cell uses the lipid environment to control signaling.
Initial structures were prepared from the 2.6 Å crystal structure, PDB code 3EML (Jaakola et al., 2008). The T4 lysozyme was removed and residues 149 to 155 and 311 to 326 were inserted using the Prime loop building module from Schrödinger, L.L.C. The protein was embedded in a palmitoyloleyolphosphatidylcholine bilayer by aligning the structure to the orientations of proteins in membranes database (Lomize et al., 2006), with at least 10 Å between the protein and its closest periodic image. The protein-membrane system was solvated by approximately 16,500 SPC water molecules in a simulation box of approximately 60 Å × 80 Å × 80 Å, one Na+, and 11 Cl−. The amber99 force field (Wang et al., 2000) was used to model the protein, while the ligand, cholesterol, and lipids were parameterized using the GAFF, with partial charges computed by the antechamber module of the Amber simulation package (Case et. al., 2008). All histidines were singly protonated, all arginines and lysines were fully protonated. The area per lipid was found to be in agreement with recent experimental results — without application a surface tension. Details are in the supplemental information. All the systems were relaxed using a 22 stage relaxation protocol developed specifically for membrane protein systems, consisting of initial minimization stages followed by slow heating and pressure coupling via Berendsen thermostat and barostat, while releasing restraints on the membrane and solute (Membrane protein relaxation protocol from the Schrödinger 2009 suite developed by Dmitry Lupyan, unpublished work). Minimizations and molecular dynamics calculations were performed with Desmond v20109 (Bowers et. al., 2006), on 256 cores of the Ranger Sun Constellation Linux Cluster, achieving approximately 50 nsec/day of simulation time. The systems were set up and relaxed using tools available in Maestro 8.5. All of the following simulations were performed under constant pressure of 1 atm and a temperature of 310 K, thermostatted and barostatted using the Martyna-Tobias-Klein method (Martyna et al., 1994). Long-ranged electrostatics were computed every third timestep by the particle mesh Ewald method. The reader should note that the pressure/temperature control and the use of multiple time-stepping renders timescales only informative in a relative sense; the determination of experimental timescales directly from the simulation is not easily accomplished. For 1.3 nsec, a Gaussian shaped barrier potential was placed at the membrane/solvent interface, in order to prevent solvent from entering the interior of the bilayer while the lipid relaxed around the protein. At this point the barrier potential was removed and the simulation was conducted free of restraints. The data presented here are for continuous simulations of 930 nsec (holo) and 720 nsec (apo). In addition, 2 shorter simulations of the holo system were performed totaling 780 nsec, and 2 shorter simulations of the apo system were performed totaling almost 1 microsecond.
All molecular images were created and analyses were performed with the Visual Molecular Dynamics package, version 1.8.7alpha (Humphrey et al., 1996). The smoothed distance data in Figures 1, ,2,2, and and66 were calculated as described in the supplementary info in Dror et. al. (Dror et al., 2009), by computing a weighted average using one-half of a period of a sine function as the weighting function, with a half-width of 1.2 nsec.
The work was supported by the NIH through NIGMS grant R43GM080726. Computer time was provided through NSF's TeraGrid program, all calculations were performed on facilities at the Texas Advanced Computing Center.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.