Mechanism of allosteric regulation of β2-adrenergic receptor by cholesterol

1G protein-coupled receptors (GPCRs) are versatile signaling proteins that mediate diverse cellular responses. With over 800 members, GPCRs constitute the largest family of integral membrane proteins in human genome and represent roughly half of all drug targets in modern medicine.

2The human β2-adrenergic receptor (β2AR) is one of the best-characterized GPCRs. It is expressed in pulmonary and cardiac myocyte tissues and is a therapeutic target for asthma and heart failure. The functional diversity of β2AR is associated with its structural dynamics. Recently found structures of β2AR in the inactive and active states have provided valuable insights into the structure-function relationship of β2AR. Subsequent biophysical and biochemical studies have provided direct evidences of multiple distinct conformational states for specific GPCRs, such as β2AR. Meanwhile, molecular dynamics (MD) simulations have depicted the dynamic behavior of β2AR and have significantly enhanced our understanding of the activation mechanism of GPCRs. Intriguingly, it is now evident that the activation of GPCRs is modulated by lipids.

3The lipid raft concept essentially states that cell membranes include functional nanoscale domains where the function emerges from proteins whose structure and activation are modulated by lipids. However, despite a large body of research data, direct substantiation of lipid-induced protein modulation remains limited. Contreras et al. showed that the COPI machinery protein p24 is recognized by a specific sphingomyelin. Coskun et al. showed that monosialoganglioside GM3 influences the activation of the epidermal growth factor receptor, however the mechanism is not known. Lipid modulation also holds to GPCRs in particular through cholesterol, which changes the physical properties of cellular membranes and supports the dynamic assembly of nanoscale membrane domains.

4The best known case is β2AR, which is a prototype of cholesterol-interacting GPCRs. β2AR belongs to the family of class A GPCRs. GPCRs belonging to this class show a high structural similarity and functional diversity. The literature reporting on the specific functional role of cholesterol and other lipids is extensive. It has been experimentally shown that cholesterol affects the conformation and function of many GPCRs. Based on X-ray crystal structures cholesterol has specific contacts with β2AR, suggesting that β2AR has binding sites for cholesterol. Spectroscopic and MD simulation studies have reported direct interactions between cholesterol and GPCRs, including β2AR. Experimental data show that cholesterol binding to β2AR changes its structural properties. Cholesterol is also necessary in crystallizing β2AR, and cholesterol and its analogue cholesteryl hemisuccinate (CHS) have been exhibited to improve β2AR stability. Since the structure and function of GPCRs are closely related, cholesterol binding specifically to β2AR is also expected to change the functional properties of the receptor. Indeed experimental studies indicate that cholesterol has a functional role in β2AR. Further, inhibition of β2AR-associated signaling has been observed with increasing membrane cholesterol content. However, as with GPCRs in general, the atomic-scale mechanism cholesterol uses to regulate β2AR is not known. Does cholesterol modulate β2AR activity through membrane-mediated effects by altering the physical properties of the membrane? Alternatively if regulation takes place through specific direct interactions, then what is the atom-scale mechanism?

5We performed extensive atomistic MD simulations (totaling >100 μs, Table 1) to clarify the mechanism responsible for the modulatory role of cholesterol on β2AR. In essence, we show that as cholesterol concentration reaches ~10 mol%, the conformational distribution of β2AR is drastically altered. The mechanism of action is based on the binding of cholesterol at specific high-affinity sites of the receptor.

6We first studied the impact of cholesterol on the conformational distribution of β2AR by systematically increasing the cholesterol concentration from 0 to 40 mol% in a DOPC (1,2-dioleoyl-sn-glycero-3-phosphocholine) bilayer. Crystallographic studies and previous biophysical and biochemical studies have shown that helices 5–6 (H5-H6) (Figure 1A) constitute a highly dynamic region of β2AR. Upon activation, the most dramatic conformational change, which is conserved among many GPCRs, is a 7–14 Å outward movement of the intracellular end of H6 from the heptahelical core of the receptor. The large rearrangement in the G protein-coupling interface is accompanied by a comparatively subtle change in the ligand-binding pocket. In a conformational change from the inactive to the active state β2AR, H5 (around S2075.46) has been found to move inward by 2 Å to establish an optimal interaction between the agonist and the two anchor sites (D1133.32/N3127.39 and S2035.42/S2045.43/S2075.46) on the receptor.

7In the present work where we started from the inactive structure of β2AR, we calculated the distance between the Cα atoms of D1133.32 and S2075.46 (referred to as LL) to measure the displacement of H5 in the ligand-binding site, and the distance between the Cα atoms of R1313.50 and E2686.30 (referred to as LG) to determine the displacement of H6 in the G protein-binding site (Figure 1A); the position of H3 does not change noticeably (RMSD < 0.8 Å) during the simulations. These two parameters (LL and LG) have been used in many previous studies to monitor changes in β2AR conformation, thus here we discuss the conformational distribution of the receptor as a function of LL and LG (Figure 1B,C and Figure 1—figure supplement 1). In the inactive crystal structure, the LL and LG values are 12.07 and 11 Å, respectively.

8In a cholesterol-free DOPC bilayer, we find β2AR to adopt a wide range of conformations with LL varying between ~11.5–17.5 Å and LG ranging between ~7.5–12.5 Å (Figure 1B). The receptor populates two major conformational states. One of them has a relatively open G protein site (LG being 10–12 Å) and a smaller ligand-binding site (LL ~ 13 ± 1 Å). The other conformation is characterized by a shift of ~3–4 Å from the intracellular end of H6 towards the receptor core that blocks the G protein interface (LG ~ 8.5 Å). At the same time, the ligand-binding pocket expands as the extracellular part of H5 moves ~ 4–5 Å away from H3 (LLnow ~16 ± 1 Å). This conformation represents an alternative inactive structure of the receptor, as both changes occur in the opposite direction compared to the case of agonist binding; we do not observe any transition to the active state of β2AR. Figure 1D shows the receptor oscillating between the different inactive conformations during 2.5 µs. The closing of the intracellular G protein-binding surface is found to correlate with the opening of the extracellular ligand-binding pocket, and vice-versa (Figure 1D). The conformational correlation between the two distal sites supports the view of allosteric regulation in GPCRs.

9In the presence of cholesterol, the picture changes quite dramatically. With a cholesterol concentration of 10 mol%, the conformational flexibility of β2AR reduces significantly (Figure 1C). The receptor stays predominantly in one conformation and no further opening of the ligand-binding site or the opening/closing of the G protein-binding site is observed, unlike in a cholesterol-free membrane. As shown in Figure 1E, LL and LG fluctuate around ~13 and~9.5 Å, respectively. The slowing down of the movements of H5 and H6 correlates with the observed high-density spots of cholesterol at these helices (IC2 and EC1 in Figure 2 discussed in detail below). To further quantify this, Figure 1F depicts the standard deviation for the fluctuations of the intracellular and extracellular ends of H5 and H6, when these ends are bound or unbound to cholesterol. The data show that the deviations of these helices from their respective average positions are much smaller when they are bound to cholesterol. The effect is particularly strong for the extracellular end of H5 at the ligand-binding site and for the intracellular end of H6 at the G protein-binding site.

10The restricted dynamics of β2AR is also observed at higher cholesterol concentrations (25 and 40 mol%; Figure 1—figure supplement 1D,E). In these cases, the receptor samples a similar conformational space as observed with 10 mol% cholesterol. At lower concentrations (2 and 5 mol%), the distribution of the receptor’s conformation is much wider (Figure 1—figure supplement 1A,B). Particularly when the membrane contains a very small percentage of cholesterol (2 mol%), the range of conformations accessible to β2AR is almost comparable to that of a cholesterol-free membrane.

11A broad conformational distribution (Figure 1—figure supplement 1F) is also observed in control simulations, where eight cholesterol molecules were initially placed at the cholesterol-binding sites of β2AR predicted by our simulations (see below), and this receptor-cholesterol complex was then embedded in a cholesterol-free membrane. Here (Figure 1—figure supplement 1F) the concentration of cholesterol in the annular region is therefore high in the beginning of the simulation, while it is zero elsewhere. Cholesterols dissociate from β2AR during the course of the simulation (discussed in detail below) and at long times the system corresponds to a dilute (cholesterol-poor) system, where the total average cholesterol concentration is low (1.9 mol%). One finds that as the data are averaged over the simulation period, the conformational behavior (Figure 1—figure supplement 1F) translates from cholesterol-rich (Figure 1—figure supplement 1E) to cholesterol-poor behavior (Figure 1—figure supplement 1A,B).

12Further, we studied the effect of cholesterol on the active conformation of β2AR in its apo form in the absence of the G protein. In the active state, the intracellular end of H6 is splayed outward from the helical bundle, providing room for the G protein (Figure 3A). We observe inward swinging of H6 towards H3 in the absence of cholesterol (which occurred in two out of three replica simulations). As shown in Figure 3B,E, the intracellular end of H6 spontaneously approaches H3 with LG dropping from 18.97 Å in the starting active conformation to ~11.5 Å that is comparable to the crystallographically observed inactive conformation of β2AR (LG ~ 11 Å). Such spontaneous deactivation of the receptor in the absence of the intracellular binding partner and cholesterol is in agreement with recent simulations and experimental studies. Meanwhile, with 40 mol% cholesterol, we observe that the active-like open conformation is stable during the simulations (Figure 3—figure supplement 1). As shown in Figure 3C,E, the LG value remains stable around 16.5 Å and no deactivation is observed unlike in cholesterol-free systems. Interestingly, here again we found a high cholesterol density at the intracellular segments of H5-H6 (IC2 in Figure 3C,D,F as discussed in detail below).

13These results show that cholesterol restricts the intrinsic conformation dynamics of β2AR and governs changes between different conformational states, thereby modulating its function.

14In all of the simulations (Table 1), cholesterol is observed to diffuse spontaneously to the receptor’s surface. Time-averaged two-dimensional (2D) number density maps demonstrate that there are preferred cholesterol positions around β2AR (Figure 2A,B).

15Localized cholesterol hot spots are often used as an indicator of potential cholesterol binding sites. We identify three such cholesterol interaction sites – two on the intracellular side (IC1 and IC2) and one on the extracellular side (EC1) (Figure 2, Figure 2—figure supplement 1A,B). Here we call them high-affinity sites since they reproducibly exhibit high cholesterol densities (normalized number density above 0.7) at different cholesterol concentrations (Figure 2A,B) and also have large lifetimes as the below discussion shows. IC1 is a shallow groove formed by the intracellular parts of transmembrane helices H1-H4 and coincides well with the location of cholesterol observed in the crystal structure of β2AR. In IC2 cholesterol penetrates deep into the cleft between H5 and H6 on the intracellular side. A high density of cholesterol is observed at IC2 not only in the inactive but also in the active β2AR conformation (Figure 3C,D,F), which suggests that this site is biologically important.

16EC1 is comprised of two closely spaced cholesterol hot spots located in the extracellular part of H5-H6 and H6-ECL3-H7 (where ECL stands for the extracellular loop). The occupancy of two cholesterol molecules at EC1 is in good agreement with the crystal structure of the adenosine receptor A2AAR, while IC2 is so far unidentified among the experimentally determined structures. Notably, the cholesterol binding residues of the three interaction sites are conserved to a large degree among β2AR orthologues (Figure 2—figure supplement 2, Figure 2—figure supplement 3, Figure 2—figure supplement 4), indicating that these sites have conserved during the evolution of the receptor. In addition, a few comparatively low-affinity cholesterol binding sites (IC3-4, EC2-3) with 10 and 40 mol% cholesterol are observed (Figure 2—figure supplement 1). When cholesterol concentration is lowered below 10 mol%, many of the interaction sites, particularly IC1 and EC1, are occupied by cholesterol at concentrations as low as 5 mol% (Figure 2—figure supplement 5). A few sites (IC2 and EC1) are visited, though transiently, by cholesterol even at 2 mol% (Figure 2—figure supplement 5).

17In addition to the above-discussed cholesterol hot spots, we observed two sites with comparatively weak cholesterol occupancies (reproducible at both 10 and 40 mol% cholesterol concentrations): IC3 between H3 and H5, and IC4 between H1 and H8, both on the intracellular side (Figure 2A,B and Figure 2—figure supplement 1). IC4 recaptures the predicted cholesterol position at the dimerization interface of β2AR found by X-ray crystallography. Besides these, another site with a low cholesterol density was observed near the extracellular part of H3-H4 (EC2) in the 10 mol% cholesterol system, and a high-density site was observed on the extracellular side of H1-H2-EC1 (EC3) in the 40 mol% cholesterol system (Figure 2A,B).

18Concluding, we find cholesterol to bind to β2AR in specific binding sites. These sites are in agreement with those found in the crystallographic structures of GPCRs.

19Is it possible that the effects we observed on β2AR conformation could be due to cholesterol-induced changes in membrane properties, and the changes in β2AR would hence not be due to the specific direct binding of cholesterol in the hot spots? To unlock this issue, we study the conformational properties of β2AR in cholesterol-free membranes whose physical properties (thickness, order, diffusion) match those of membranes with a large concentration of cholesterol.

20Effect of increased bilayer thickness. We studied β2AR embedded in a bilayer composed of long-chain mono-unsaturated phosphatidylcholine (PC) lipids PC-20:0/22:1 c13. The thickness of this membrane is larger than the thickness of a DOPC bilayer with 40 mol% cholesterol, while its lipid chain order is comparable to a DOPC bilayer with 5% cholesterol (Figure 4—figure supplement 1A,B). Figure 4A depicts that the increased bilayer thickness is unable to restrict the conformational dynamics of β2AR. The receptor just adjusts itself to the hydrophobic mismatch by inducing bilayer thinning (4–8 Å) in its vicinity (Figure 4B).

21Effect of increased bilayer order. We then studied β2AR placed in a DOPC bilayer with 20 mol% pyrene, which is known to induce similar (ordering and condensing) effects as cholesterol. Figure 4D highlights that pyrene does not show any preference for specific binding on the β2AR surface except for the slowed-down diffusion of pyrene near the receptor surface. β2AR exhibits a very broad conformational distribution, with LL and LG fluctuating between ~9–17.5 and ~7–13.5 Å, respectively (Figure 4C). This conformational behavior of the receptor is distinctly different from the one induced by ≥10 mol% cholesterol, although the order of the pyrene-containing bilayer is similar to a DOPC bilayer with 10 mol% of cholesterol (Figure 4—figure supplement 1D).

22Summarizing, the changes in physical membrane properties, similar to those induced by cholesterol, do not restrict the conformational dynamics of β2AR. We conclude that the cause of the observed changes in β2AR conformation and dynamics is the specific binding of cholesterol to β2AR.

23When cholesterol is specifically bound to β2AR, how stable is the binding? Figure 5 depicts the time-correlation function of cholesterol binding in the three main binding sites (IC1, IC2, EC1) on β2AR and shows that at low cholesterol concentrations (2–5 mol%) the binding lifetime is short, of the order of 100 ns or less. However, at ~10 mol% there is a clear transition to longer lifetimes (see Video 1 and Video 2) given that the lifetime of binding increases to the microsecond time scale for 10 and 40 mol% cholesterol.

24In three control simulations where cholesterols were initially bound at the eight cholesterol-binding sites identified in our simulations and no further cholesterol was in the bilayer (Figure 5—figure supplement 1), cholesterols underwent rapid unbinding from the majority of the binding sites in a timescale of tens to hundreds of nanoseconds (Figure 5—figure supplement 1), similarly to the short binding lifetime observed for cholesterol-poor systems (2 mol%, Figure 5). However, at a few sites cholesterol stayed for the entire simulation time (IC1 and IC2 in two out of three simulations) or dissociated in the μs timescale (IC3 and EC3 in one simulation).

25The results show that the lifetime of cholesterol is of the order of microseconds in the high-affinity binding sites, where the lifetime at large cholesterol concentrations is largely independent of cholesterol concentration.

26We next explore how cholesterol analogues, in comparison to cholesterol, interact with β2AR. We focus on four different analogues (Table 1): (i-ii) cholesteryl hemisuccinate (CHS) and its deprotonated form (CHSA), and (iii-iv) two oxysterols, 4β-hydroxycholesterol (4β-OH-Chol) and 27-hydroxycholesterol (27-OH-Chol), oxidized at the cholesterol ring and tail, respectively (Figure 2—figure supplement 6A). As compared to cholesterol, CHS is a more water-soluble cholesterol ester and is widely used in structural biology and biophysical studies as a cholesterol analogue. Oxysterols, on the other hand, are derivatives of cholesterol with additional oxygen-containing substitutions at different positions of cholesterol. Due to the structural similarities with cholesterol, these analogues mimic cholesterol as to the effects on membrane properties (e.g., increasing bilayer order and thickness), although to different extents (Figure 2—figure supplement 6).

27CHSA is found to interact strongly with β2AR due to the enhanced electrostatic coupling resulting from its negatively charged head-group (Figure 2—figure supplement 7), however it favors to reside around the receptor at locations different from those of cholesterol (Figure 2—figure supplement 8A,B). Meanwhile, CHS closely mimics the behavior of cholesterol (Figure 2—figure supplement 7). Among the three major cholesterol interaction sites observed in our simulations, we find a very high CHS density at IC2 (Figure 2—figure supplement 8C–F). High occupancy of CHS is also observed near IC1 (at 40 mol% CHS concentration) but not at all at EC1. Occupancy of CHS at IC1 is consistent with the crystal structure of β1AR.

284β-OH-Chol interacts only weakly with β2AR (Figure 2—figure supplement 7). Almost all of the interaction sites on the receptor surface are occupied by cholesterol rather than 4β-OH-Chol (Figure 2—figure supplement 8G–J). As a result, the average density maps, showing the lateral arrangement of these sterols around β2AR, are similar to those of 10 and 40 mol% cholesterol systems (Figure 2A), and reproducible. Unlike 4β-OH-Chol, 27-OH-Chol prefers to interact with the receptor directly (Figure 2—figure supplement 7). For the IC1 site, 27-OH-Chol competes, though weakly, with cholesterol, while at EC1 and IC3, 27-OH-Chol exhibits preference over cholesterol (Figure 2—figure supplement 8K–N).

29Altogether, our results show that also other cholesterol-like molecules interact with β2AR and may occupy the same binding sites on the receptor surface as cholesterol. However, the effects of cholesterol-analogues on β2AR are weaker compared to those induced by cholesterol (Figure 2—figure supplement 9). All the cholesterol analogues studied here have a rigid ring structure, yet their slightly different chemical compositions influence their occupancy as well as the strength of binding to the cholesterol-binding sites (Table 2). This is assessed here in terms of the van der Waals energy, which as a short-range interaction reflects how strongly two molecules are in contact and therefore serves as an appropriate measure for the gravity of lipid-protein binding in the binding site.

30The results in Table 2 show that among the three major interaction sites, the binding of CHS at IC1 is much weaker than that of cholesterol. At IC2 the strength of interaction of CHS and cholesterol is comparable. Meanwhile, the extracellular EC1 site remains unoccupied by CHS indicating the binding energy to be low. As to the two oxysterols, 4β-OH-Chol interacts with β2AR only at EC1 and the interaction is weak, while 27-OH-Chol binds at EC1 as tightly as cholesterol, but its interaction at the two other binding sites (IC1 and IC2) is much weaker than in the case of cholesterol. Concluding, CHS interacts at IC2 as strongly as cholesterol but its interactions at IC1 and EC1 are negligible compared to those of cholesterol. The oxysterol 27-OH-Chol interacts at EC1 as strongly as cholesterol but its interactions at IC1 and IC2 are negligible compared to those of cholesterol. The oxysterol 4β-OH-Chol does not interact with β2AR to a significant degree.

31These data can be considered in the context of molecular structures. In CHS, the difference compared to cholesterol is the additional chain bridged to the cholesterol structure via an ester bond (Figure 2—figure supplement 6A). This additional chain does not interfere binding at IC2, but it does alter the binding at IC1 and EC1. In 27-OH-Chol, the oxidation has taken place in the short acyl chain that is the terminal subunit of the molecule. This does not interfere the binding at EC1 but does alter the binding at IC1 and IC2. Finally, in 4β-OH-Chol, the oxidation has occurred in the rigid steroid moiety, making the α-side of the molecule rougher. In cholesterol, the α-side is exceptionally flat. Given this change in surface roughness, and the importance of the surface-surface contact in lipid-β2AR binding in the binding site, it is quite obvious why this oxysterol does not bind to any of the cholesterol binding sites (IC1, IC2, EC1).

32The results support the view that the restriction of β2AR dynamics arises from specific lipid binding to the receptor binding sites: the tighter the binding, the more is the receptor dynamics suppressed, and cholesterol induces the strongest effect.

33Our results show that cholesterol has a preference to bind to β2AR at specific locations on its surface. We identified three high-affinity cholesterol interaction sites in β2AR (Figure 2C,D): IC1 (at the cleft of H1-H4 on the intracellular side), IC2 (H5-H6 on the intracellular side), and EC1 (the H5-H6-ECL3-H7 region on the extracellular side). IC1 and EC1 are in agreement with the locations of cholesterol found in GPCR crystal structures. IC1 contains a cholesterol consensus motif that predicts cholesterol binding for 44% of human class A receptors. Moreover, these binding sites appear to be evolutionarily conserved in β2AR, which suggests their possible allosteric role in receptor function. A recent simulation study reported a correlation between cholesterol occupancy at IC1 and β2AR dimerization. However, not much is known about the functional relevance of cholesterol binding to the other sites of β2AR.

34The present work for the inactive conformation of β2AR shows that cholesterol binding at IC2 and EC1 (Figure 2C,D) strongly influences the conformational dynamics of β2AR (Figure 1). In a cholesterol-free membrane the receptor samples multiple conformational states (Figure 1B) accounting for the high basal activity of β2AR. Our results show that the presence of cholesterol in high densities around H5-H6-H7 impedes the dynamic nature of the receptor. In cholesterol-containing (≥10 mol% cholesterol) membranes (Figure 1C and Figure 1—figure supplement 1D,E), the overall structural flexibility of the receptor is significantly reduced to one predominant conformation. We observed that in the presence of strongly bound cholesterol, H5 and H6 undergo much smaller displacements from their average positions as compared to the situation without cholesterol (Figure 1F). Cholesterol analogues that occupy the same interaction sites also restrict the β2AR conformation (Figure 2—figure supplement 9), although their effects are weaker compared to those of cholesterol. Cholesterol or cholesterol-like molecules bound at these inter-helical clefts can thus confine the movement of the respective helices to a substantial degree, thus dampening the overall conformational dynamics of the receptor. At IC2 of inactive β2AR, cholesterol pushes the intracellular end of H6 more towards the core of the helical bundle and prevents the outward movement of H6 required for G protein binding. The restriction of H6 movement by cholesterol is a potentially important allosteric effect, which can be used to modulate the receptor activity.

35Interestingly, our study on the active-state β2AR also exhibits a high cholesterol density at IC2 (Figure 3D,F). Here cholesterol bound at IC2 acts as a spacer between H5-H6 and restricts the movement of H6, thereby stabilizing the open active-like conformation of the receptor (Figure 3D), while in the absence of cholesterol the receptor is more prone to undergoing spontaneous deactivation (Figure 3E; Figure 3—figure supplement 1). This result supports the postulate that cholesterol restricts the conformational dynamics of the receptor by binding at specific interaction sites and governs changes between different receptor states, therefore modulating its function. Moreover, cholesterol binding at IC2 in both inactive and active states of β2AR as found in our simulations highlights the biological relevance of this interaction site in allosteric regulation of the receptor conformation.

36The highly conserved IC1 site shows no major influence on the mobility of H5-H6. On the other hand, IC1 exerts a stabilizing effect on H4 (Figure 2—figure supplement 10), in agreement with experiments. As H4 is one of the weakest points of the β2AR fold, its decreased mobility may account for the enhanced stability of the receptor.

37Cholesterol modulates the physical properties of membranes by increasing the bilayer thickness and order, and slowing down the dynamics. These general membrane effects can also influence the dynamic nature of a membrane protein. However, here we found that membrane-mediated interactions do not affect β2AR conformation to a significant degree (Figure 4).

38GPCRs are signaling machines that function by toggling between multiple conformers. The dynamic nature of GPCRs has made their crystallization process extremely challenging. Besides techniques like protein engineering and use of detergents to increase the intrinsic stability of the receptor, cholesterol/CHS has emerged as a necessary component for crystallization of many GPCRs, including β2AR. Our work shows that in the presence of more than ~10 mol% cholesterol, inactive β2AR partly loses conformational variability and populates just one major conformation. Achieving conformational homogeneity is the key to crystallize membrane proteins. In agreement with our results, a recent experimental study showed that CHS impacts the conformational dynamics of a GPCR leading to a restricted conformational space. Earlier it was experimentally reported that cholesterol induces a more compact conformational state of the oxytocin receptor. Our results are also in agreement with a recent dynamic single-molecule force spectroscopic study, which showed that CHS strengthens interactions that stabilize the structural segments in β2AR and thereby considerably increase the kinetic, energetic, as well as the mechanical stability of the receptor. In addition, the function of adrenergic receptors is known to be modulated by cholesterol: cholesterol depletion enhances β2AR-associated signaling, while increased cholesterol content inhibits signaling.

39To our knowledge, the results presented in this work provide the first atomic-scale picture of how lipids can govern the conformation of membrane receptors through direct lipid-protein interactions in specific lipid binding sites, and hence dictate the state of a receptor. The receptor-cholesterol interactions, such as those observed in our simulations for β2AR, can conceivably govern the signaling of many GPCRs in the given protein family.

40We performed all-atom molecular dynamics simulations of β2AR embedded in lipid bilayers with various lipid compositions (Table 1) using the GROMACS 4.6.x software package.

41All simulations were performed using the GROMACS 4.6.x package. The all-atom OPLS-AA (optimized potentials for liquid simulations) force field was used to parameterize the protein, ions, and pyrene. Force field parameters for cholesterol, cholesteryl hemisuccinate, and oxysterols were taken from previously published papers. For the studied phosphatidylcholines (DOPC and PC-20:0/22:1 c13), we used new torsional and Lennard-Jones parameters derived for saturated and unsaturated hydrocarbons and the torsional potential developed for the glycerol backbone and the phosphatidylcholine head group. The TIP3P model, which is compatible with the OPLS parameterization, was used for water molecules.

42All simulations of the systems considered in this work (Table 1) were performed under the isobaric-isothermal (NpT) ensemble. A time step of 2 fs was used for integrating the equations of motion. Periodic boundary conditions were applied in all three directions of the system. The temperature of the system was maintained at 310 K by employing the v-rescale (stochastic velocity rescaling) thermostat with a time constant of 0.1 ps. The temperatures of the receptor, lipids, and solvent molecules were controlled independently. The pressure of the system (1 bar) was maintained semi-isotropically using the Parrinello–Rahman barostat with a 1 ps time constant. The LINCS algorithm was applied to preserve hydrogen covalent bond lengths. Lennard-Jones interactions were cutoff at 1.0 nm. The particle mesh Ewald (PME) method was employed for long-range electrostatic interactions using a real space cutoff of 1.0 nm, β-spline interpolation (order of 6), and a direct sum tolerance of 10−6.

43The initial coordinates of β2AR were taken from our recently published work, in which the structural modifications made for crystallization of the inactive β2AR structure [PDB id: 3D4S] were reverted back to its original sequence. This inactive crystal structure of β2AR bound to the partially inverse agonist timolol was heavily engineered to facilitate crystallization. We reverted all the structural modifications from the experimentally determined structure, i.e., we removed mutations (E1223.41W on the transmembrane helix H3 and the N1875.26E mutation on the extracellular loop 2), removed the T4-lysozyme attached between the transmembrane helices 5 and 6, and replaced it with the missing intracellular loop 3. We did not attempt to model the unresolved N-terminal (32 residues) and C-terminal (71 residues) parts. The details of the procedure used to prepare the receptor model for our simulations are described elsewhere. In the present work, we considered the apo-receptor (without a ligand), as we were interested in the intrinsic dynamics of β2AR.

44For simulations with the active-state β2AR conformation, the starting structure was taken from the crystal structure of the receptor bound to an agonist and a Gs protein. Here again we considered the apo-form of the receptor without the ligand and the G protein. Additionally, we removed the lysozyme and modeled the missing loop regions (A176-H178 and F240-F264), but the mutations were kept as such.

45We simulated β2AR embedded in a number of lipid bilayers (Table 1) with varying lipid composition. The lipid contents used in the studies were as follows:

46DOPC bilayers with different cholesterol (Chol) concentrations: 0, 2, 5, 10, 25, and 40 mol%.

47DOPC bilayers with a cholesterol analogue cholesteryl hemisuccinate (CHS; 10 and 40 mol%) or its deprotonated form CHSA (10 and 40 mol%). CHS is known to enhance the stability of GPCRs. It is frequently used for GPCR characterization.

48DOPC bilayers mixed with several sterols: 21 mol% cholesterol and 4 mol% oxidized sterol (4β-hydroxy-cholesterol (4β-OH-Chol) or 27-hydroxy-cholesterol (27-OH-Chol)). Oxysterols used in this study are among the most common oxysterols found in human serum.

49A single-component bilayer composed of the long-tail monounsaturated phospholipid PC-20:0/22:1 c13.

50DOPC bilayers with 20 mol% pyrene.

51The lipid bilayers (without β2AR) were constructed using in-house scripts, and they were subsequently solvated with water. These lipid bilayers were then equilibrated for 100–200 ns.

52Next, β2AR was placed into the above-mentioned pre-equilibrated bilayers in such a manner that the lipid arrangement around the receptor was completely random and that there was no cholesterol or cholesterol analogue initially bound to β2AR. For incorporating the receptor into a pre-equilibrated lipid bilayer, we followed our recently published method, where the receptor was pushed into a lipid membrane from its side by applying a high lateral pressure on the system.

53Each system contained one β2AR and 256–512 lipids. Each of the systems was explicitly solvated by water. In all cases, counterions (8 Cl– ions for β2AR, and additional Na+ counter ions for bilayers containing the anionic CHSA) were added to maintain electroneutrality of the systems. NaCl salt was added to achieve the physiological salt concentration of 150 mM. Subsequently each system was energy minimized and then equilibrated in two stages with position restraints first on protein heavy atoms and then on the backbone. Following equilibration (100 ns), all restraints were released and the equilibrated systems were subjected to microsecond length (1–2.5 μs) production simulations. Multiple independent simulations were performed for each lipid composition, either by starting from a different lipid arrangement around β2AR (for systems with no sterols initially bound to the receptor) or starting with different initial velocities (for systems with sterols initially bound to the receptor).

54Additional simulations were performed where cholesterol or its analogues were initially attached to certain locations on the surface of the receptor, and this receptor-lipid complex was then embedded to a cholesterol-free DOPC bilayer. Here we performed two sets of control simulations. In one set of simulations, two cholesterol or CHS (neutral or anionic) molecules were bound at the cleft formed by the intracellular side of the transmembrane helices H1-4 as predicted from the crystal structure. In another set of control simulations, cholesterol molecules were initially bound at the eight interaction sites of β2AR predicted by our simulations (see discussion in the main article). The simulation conditions (as to counterions and salt, release of restrains, simulation times, etc.) were as described above.

55The systems investigated in this study are summarized in Table 1. The total simulation time for the atomistic systems studied in this work covers a period of more than 100 μs.

56For calculation of deviations of helix ends, we first calculated their time series of X, Y, and Z coordinates. The coordinates were then divided into two groups based on whether the upper and lower halves of the helixes (backbone atoms) were in contact ( ≤ 0.5 nm) with cholesterol (heavy atom) or not. Separately in each group, the distance from the average point of the group at each time frame (say ith frame) was calculated by:

57where,, were the coordinates of the ith frame, and,, were the average values. The standard deviation of each group was then calculated by:

58The average standard deviation of different simulations was calculated as a weighted average depending on the number of frames (Ng) of the group in each simulation.

59The 2D number density maps were calculated using the g_densmap tool of GROMACS. The two bilayer leaflets were calculated separately. The output was then processed (using an in-house script) to normalize the maximum number density to one. We calculated the 2D number densities of cholesterol (non-hydrogen atoms) and β2AR (backbone atoms of transmembrane region) separately.

60A residue of β2AR was considered to be in contact with cholesterol, when any of its non-hydrogen atoms was within ≤0.5 nm of any heavy atom of cholesterol. The total occupancy time was then normalized over the entire length of a simulation, i.e., an occupancy time of one means that the particular residue of β2AR was in contact with cholesterol throughout the simulation, whereas a value of zero means no contact. The calculated total occupancy time per residue of β2AR was mapped onto the receptor’s surface to highlight the regions of β2AR involved in cholesterol binding.

61We analyzed amino acid sequences of β2AR orthologues from the available databases. We used the PhylomeDB server (http://phylomedb.org/) for finding orthologues and Clustal Omega (http://www.ebi.ac.uk/Tools/msa/clustalo/) for sequence alignment. The amino acid residues of β2AR segments constituting the cholesterol binding sites as obtained from our simulations were used for the set of sequences obtained (Figure 2—figure supplement 2, Figure 2—figure supplement 3, Figure 2—figure supplement 4). The sequences in question belong to diverse species, such as insects, fish, birds, reptile, mammals, etc.

62The order parameter of lipid acyl chains was calculated using:

63where θi is the angle between a C-D bond (C-H in simulations) of the ith carbon atom and the bilayer normal. The angular brackets denote averaging over time and molecules in a bilayer.

64Bilayer thickness was defined as the distance between the average planes formed by phosphorous atoms in the two bilayer leaflets. We used the g_lomepro tool to generate the 2D distribution of bilayer thickness.

65For the calculation of the lifetime of cholesterol bound to the cholesterol interaction sites on the receptor surface, we first monitored the binding/unbinding events of each individual cholesterol molecule along the simulation trajectory. A cholesterol molecule was considered bound when any of its heavy atoms came within ≤0.5 nm from an interaction site. To define the three major interaction sites on the β2AR surface, we used the amino acid residues (with contact fraction ≥ 0.4) as shown in Figure 2—figure supplement 2. The time series was then additionally smoothed (over one ns time windows) to discard very rapid ‘leave and return’ motions of cholesterol that take place due to thermal fluctuations. Given that lateral diffusion of lipids at the protein surface is very slow, and the lipids essentially do not move at all during a 1-ns time window, these fluctuations were then taken care of by the smoothing procedure. We then calculated the normalized time correlation function (to describe the time-dependent probability of cholesterol that is next to the receptor to stay in contact with the receptor) over all individual cholesterol binding/unbinding events occurred in all independent simulation trajectories for all cholesterol molecules present in a system at a given cholesterol concentration.

66For all analysis to measure time-averaged properties, the first 100 ns of production simulations were excluded from the calculation. Error bars were estimated through standard error, calculated by dividing the standard deviation of a given data set with the square root of its sample size. We used the g_analyze tool of GROMACS for error estimation.