Agonist-bound adenosine A2A receptor structures reveal common features of GPCR activation.

Adenosine receptors and β-adrenoceptors are G-protein-coupled receptors (GPCRs) that activate intracellular G proteins on binding the agonists adenosine or noradrenaline, respectively. GPCRs have similar structures consisting of seven transmembrane helices that contain well-conserved sequence motifs, indicating that they are probably activated by a common mechanism. Recent structures of β-adrenoceptors highlight residues in transmembrane region 5 that initially bind specifically to agonists rather than to antagonists, indicating that these residues have an important role in agonist-induced activation of receptors. Here we present two crystal structures of the thermostabilized human adenosine A(2A) receptor (A(2A)R-GL31) bound to its endogenous agonist adenosine and the synthetic agonist NECA. The structures represent an intermediate conformation between the inactive and active states, because they share all the features of GPCRs that are thought to be in a fully activated state, except that the cytoplasmic end of transmembrane helix 6 partially occludes the G-protein-binding site. The adenine substituent of the agonists binds in a similar fashion to the chemically related region of the inverse agonist ZM241385 (ref. 8). Both agonists contain a ribose group, not found in ZM241385, which extends deep into the ligand-binding pocket where it makes polar interactions with conserved residues in H7 (Ser 277(7.42) and His 278(7.43); superscripts refer to Ballesteros-Weinstein numbering) and non-polar interactions with residues in H3. In contrast, the inverse agonist ZM241385 does not interact with any of these residues and comparison with the agonist-bound structures indicates that ZM241385 sterically prevents the conformational change in H5 and therefore it acts as an inverse agonist. Comparison of the agonist-bound structures of A(2A)R with the agonist-bound structures of β-adrenoceptors indicates that the contraction of the ligand-binding pocket caused by the inward motion of helices 3, 5 and 7 may be a common feature in the activation of all GPCRs.

The adenosine receptors and β-adrenoceptors (βARs) are G protein-coupled receptors (GPCRs) that activate intracellular G proteins upon binding agonist such as adenosine[1] or noradrenaline[2], respectively. GPCRs have similar structures consisting of 7 transmembrane helices that contain well-conserved sequence motifs, suggesting that they are probably activated by a common mechanism[3,4]. Recent structures of βARs highlight residues in transmembrane region 5 that initially bind specifically to agonists rather than to antagonists, suggesting an important role in agonist-induced activation of receptors[5-7]. Here we present two crystal structures of the thermostabilised human adenosine A2A receptor (A2AR-GL31) bound to its endogenous agonist adenosine and the synthetic agonist NECA. The structures represent an intermediate conformation between the inactive and active states, because they share all the features of GPCRs that are thought to be in a fully activated state, except that the cytoplasmic end of transmembrane helix 6 partially occludes the G protein binding site. The adenine substituent of the agonists bind in a similar fashion to the chemically-related region of the inverse agonist ZM241385[8]. Both agonists contain a ribose group, not found in ZM241385, which extends deep into the ligand binding pocket where it makes polar interactions with conserved residues in H7 (Ser2777.42 and His2787.43; superscripts refer to Ballesteros-Weinstein numbering[9]) and non-polar interactions with residues in H3. In contrast, the inverse agonist ZM241385 does not interact with any of these residues and comparison with the agonist-bound structures suggests that ZM241385 sterically prevents the conformational change in H5 and therefore it acts as an inverse agonist. Comparison of the agonist-bound structures of A2AR with the agonist-bound structures of β-adrenoceptors suggests that the contraction of the ligand binding pocket caused by the inward motion of helices 3, 5 and 7 may be a common feature in the activation of all GPCRs.

In the simplest model for the conformational dynamics of GPCRs[10] there is an equilibrium between two states, R and R*. The inactive state R preferentially binds inverse agonists and the activated state R* preferentially binds agonists[11]. Only R* can couple and activate G proteins. Although there are far more complex schemes[12] describing intermediates between R and R*, studies on rhodopsin have indicated that there is only one major conformational change that significantly alters the structure of the receptor[3]. Thus the structures of dark-state rhodopsin[13,14] and of opsin[15,16] are considered to be representative structures for the R and R* state, respectively. Structures of 6 different GPCRs[8,13,17-21] in conformations closely approximating to the R state have now been determined and it is clear that they are similar to each other, with RMSDs between any pair of structures in the transmembrane domains being less than 3 Å. As observed in light-activation of rhodopsin, the major structural difference between R and R* is the movement of the cytoplasmic ends of helices 5 and 6 away from the receptor core by 5-6 Å, opening up a cleft in the centre of the helix bundle where the C-terminus of a G protein can bind[16]. Recently, the structure of an agonist-bound β-adrenoceptor (β2AR) was determined in complex with an antibody fragment (nanobody Nb80)[5]. This structure of β2AR is very similar to the structure of opsin, which suggests that the nanobody mimicked the action of a G protein by maintaining the receptor structure in an activated state. Given the structural similarities between opsin and the β2AR-Nb80 complex, it is likely that the structures of the R* states of other GPCRs are also highly similar. This is consistent with the same heterotrimeric G proteins being able to couple to multiple different receptors[22]. However, do the conserved structures of R and R* imply that all agonists activate the receptors in an identical fashion? The recent structures of a thermostabilised β1AR bound to 4 different agonists suggested that a defining feature of agonist binding to this receptor is the formation of a hydrogen bond with Ser5.46 on transmembrane helix 5 that accompanies the contraction of the ligand binding pocket[7]. Here we describe two structures of the adenosine A2A receptor (A2AR) bound to two different agonists, which suggests that the initial action of agonist binding to A2AR has both similarities and differences compared to agonist binding in βARs.

The native human A2AR when bound to its endogenous agonist adenosine or to the high-affinity synthetic agonist NECA is unstable in detergent, so crystallization and structure determination relied on using a thermostabilised construct (A2AR-GL31) that contained four point mutations, which dramatically improved its thermostability. Pharmacological analysis showed that the mutant receptor bound the five antagonists tested with greatly reduced affinity (1.8 - 4.3 log units), whereas four agonists bound with similar affinity to the wild-type receptor (Supplementary Fig. 1). However, A2AR-GL31 is only weakly activated by the agonist CGS21680 (Supplementary Fig. 2), which suggests that the thermostabilising mutations might also decouple high-affinity agonist binding from the formation of R*. The conformation of GL31 is not consistent with it being in the fully-activated G protein coupled state, because we do not observe a 42-fold increase in affinity for NECA binding measured for Gαs-coupled A2AR[23]. These data all suggest that A2AR-GL31 is in an intermediate conformation between R and R*, which is consistent with the structural analysis presented below.

The two structures we have determined are of A2AR-GL31 bound to adenosine and NECA with resolutions of 3.0 Å and 2.6 Å, respectively (Supplementary Table 1). Global alignments of the A2AR-GL31 structures with A2A-T4L (A2AR with T4 lysozyme inserted into inner loop 3) bound to the inverse agonist ZM241385 were performed based on those residues in the region of the ligand binding pocket that show the closest structural homology (Fig. 1 and Supplementary Text). This gave an rmsd in Cα positions of 0.66 Å for the 96 atoms selected, which include all residues involved in binding either adenosine or NECA, with the exception of those in H3. Using this transformation, the adenine-like moiety of the two ligands superimposes almost exactly (rmsd 0.56 Å). The most significant differences between the two structures are seen in a distortion and a 2 Å shift primarily along the helical axis of H3, a bulge in H5 (resulting from non-helical backbone conformation angles of residues Cys185 and Val186) that shifts residues into the binding pocket by up to 2 Å and also a change in conformation of the cytoplasmic ends of H5, H6 and H7 (Fig. 1). Comparison of the A2AR-GL31 structure with the agonist-bound β2AR-Nb80 complex indicates that these differences are similar to the conformational changes in the β2AR that are proposed to be responsible for the formation of the R* state[5]. However, it is unlikely that the structure of A2AR-GL31 represents the fully activated state, because comparison with opsin bound to the C-terminal peptide of the G protein transducin shows that there is insufficient space in A2AR-GL31 for the C-terminus of the G protein to bind (Supplementary Fig. 3). This is based on the assumption that all G proteins bind and activate GPCRs in a similar fashion, but given the highly-conserved structures of both G proteins and GPCRs this seems a reasonable hypothesis.

Figure 1

Structure of the adenosine A2A receptor bound to NECA compared to other GPCR structures. (a) The structure of NECA-bound A2AR is shown as a cartoon (yellow) aligned with the structure of A2A-T4L bound to the inverse agonist ZM241385 (blue; PDB code 3EML[8]). NECA is shown as a space-filling model (C, green; N, blue; O, red). (b, c) Sections through the aligned receptors in (a) to highlight the differences in the intracellular face of the receptors (b) and in the ligand binding pocket (c), with the bulge in H5 shown as an inset. (d, e) Alignment of NECA-bound A2AR (yellow) with agonist-bound β2AR-Nb80 (red; PDB code 3P0G[5]) showing the intracellular face of the receptors (d) and the ligand binding pocket (e). NECA is shown as a space-filling model in c and e. The figures were generated using CCP4mg[31]. Analogous alignments to opsin are depicted in Supplementary Fig. 7.

The fact that the structure of A2AR-GL31 represents an agonist-binding state is consistent with how A2AR-GL31 was engineered. Thermostabilising mutations were selected by heating the NECA-bound detergent-solubilised receptor, so the mutations are anticipated to stabilize the agonist-bound state either by stabilizing helix-helix interactions and/or biasing the conformational equilibrium between the agonist-bound R* state and the agonist bound R-state[24-26]. The two most thermostabilising mutations, L48A and Q89A, are in regions of the receptor that are involved in transitions between R and R*, providing a possible explanation for their thermostabilising effect (Supplementary Fig. 4). The other two mutations, A54L and T65A, are at the receptor-lipid interface and the reason for their thermostabilising effect is unclear. Although the overall shape of the ligand binding pockets of A2AR and β2AR are different, the structural similarities with the β2AR-Nb80[5] and the structural differences to ZM241385-bound A2A-T4L[8] indicate that the structure of the binding pocket in A2AR-GL31 is a good representation of the agonist-bound binding pocket of the wild-type receptor (Fig.1).

Adenosine and NECA bind to A2AR-GL31 in a virtually identical fashion, in addition, the adenine ring in the agonists interacts with A2AR in a similar way to the chemically-related triazolotriazine ring of the inverse agonist ZM241385 (Fig. 2). Thus the hydrogen bonds between exocyclic adenosine N6 (Supplementary Fig. 5) with both Glu169 in extracellular loop 2 (EL2) and Asn2536.55 in H6 are similar, with the significant π-stacking interaction with Phe168 in EL2 also conserved. One of the major structural differences between ZM241385 and the agonists is the presence of a furan substituent on C20 of triazolotriazine in the inverse agonist, whilst agonists contain a ribose substituent linked to N9 of adenine (Fig. 2 and Supplementary Fig. 5). In ZM241385, the furan group forms a hydrogen bond with Asn2536.55 in H6 and van der Waals contacts with other residues in H3, H5 and H6[8]. In contrast, the ribose moiety in agonists forms hydrogen bonds with Ser2777.42 and His2787.43 in H7, in addition to van der Waals interactions with other residues in H3 and H6 (Fig. 2). In particular, Val843.32 has to shift its position upon agonist binding due to a steric clash with the ribose ring, which may contribute to the 2 Å shift observed in H3 (Fig. 3). These differences in binding between ZM241385 and either adenosine or NECA suggest that the residues that bind uniquely to agonists (Ser2777.42 and His2787.43) play a key role in the activation of the receptor, as previously shown by mutagenesis studies[27,28]. This is analogous to the situation in the activation of β1AR, where only full agonists cause the rotamer conformation changes of Ser5.46 in H5, whereas the inverse agonist ICI118551 prevents receptor activation by sterically blocking the rotamer change[7,29]. However, the details of the activation differ in that the critical residues that bind agonists and not antagonists are in H5 in β1AR, but in H7 in A2AR (Fig. 4).

Figure 2

Receptor-ligand interactions compared for the adenosine receptor bound to the inverse agonist ZM241385 and the agonists NECA and adenosine. Structures of the human A2AR in cartoon representation are shown bound to the following ligands: (a) ZM241385 (PDB code 3EML[8]); (b) NECA; (c) adenosine. (d, e) Polar and non-polar interactions involved in agonist binding to A2AR are shown for NECA (d) and adenosine (e). Amino acid residues within 3.9 Å of the ligands are depicted, with residues highlighted in blue making van der Waals contacts (blue rays) and residues highlighted in red making potential hydrogen bonds with favourable geometry (red dashed lines, as identified by HBPLUS, see Methods Online) or hydrogen bonds with unfavourable geometry (blue dashed lines, donor acceptor distance less than 3.6 Å). Where the amino acid residue differs between the human A2AR and the human A1R, A2BR and A3R, the equivalent residue is shown highlighted in orange, purple or green, respectively. Panels a-c were generated using Pymol (DeLano Scientific Ltd). Omit densities for the ligands are shown in Supplementary Fig. 6 and densities for water molecules in Supplementary Fig. 8.

Figure 3

Positions of adenosine and ZM241385 in the adenosine A2A receptor ligand binding pocket. The structures of adenosine-bound A2AR-GL31 and ZM241385-bound A2A-T4L were aligned using only atoms from the protein to allow the ligand positions to be compared, with adenosine in yellow and ZM241385 in pink (N, blue; O, red). The ligands are shown in the context of the binding pocket of A2AR-GL31, with transmembrane helices of A2AR-GL31 shown in yellow and the surfaces of the receptor, including the cavity of the ligand binding pocket, shown in grey. The side chains of Val84 and Leu85 that interact with the ribose moiety of the agonist are shown in green.

Figure 4

Comparison of the positions of agonists in the binding pockets of the adenosine A2A receptor and a β-adrenoceptor. (a) The structures of A2AR bound to adenosine and β1AR bound to isoprenaline (PDB code 2Y03)[7] were aligned by superimposing equivalent atoms in the protein structure and the positions of both ligands shown as stick models with the carbon atoms in blue-green (isoprenaline) or yellow (adenosine) nitrogen in blue and oxygen in red. The A2AR structure is shown, with H5 and H7 as space-filling models (C, grey; N, blue; O, red) and the remainder of the structure as a cartoon (pale green). Some water molecules are shown as red spheres, hydrogen bonds as red dashed lines and the polar contacts as blue dashed lines. The orientation of the figure is identical to that shown in Fig. 2. (b) Structure of A2AR bound to adenosine viewed from the extracellular surface. (c) Structure of β1AR bound to isoprenaline (PDB code 2Y03)[7] viewed from the extracellular surface. In panels b and c, equivalent side chains in the respective structures that make contacts to both isoprenaline and adenosine in their respective receptors are shown as space-filling models and they have the following Ballesteros-Weinstein numbers (amino acid side chains are shown in parentheses for the A2AR and β1AR, respectively): 3.32 (V84, D121); 3.36 (T88, V125); 5.42 (N181*, S211); 6.51 (L249, F306); 6.55 (N253, N310); 6.52 (H250*, F307); 7.39 (I274, N329); 7.43 (H278, Y333). Some residues (*) only make indirect contacts to the agonists via a water molecule.

Adenosine and NECA activate the A2AR through interactions with H3 and H7 that are absent in the interactions between the receptor and the inverse agonist ZM241385 (Fig. 2). The inward shift of H7, the movement of H3 and the consequent formation of a bulge in H5 are all observed in the structure of agonist-bound A2AR-GL31 and β2AR-Nb80 (Fig.1). The formation of the bulge in H5 of the β2AR-Nb80 structure was linked to a series of conformational changes that generates the 60° rotation of H6 about Phe2826.44, resulting in the cytoplasmic end of H6 moving out from the receptor centre and opening the cleft where the C terminus of a G protein is predicted to bind as observed in opsin[5,6]. There are analogous side chain movements in A2AR-GL31 that result in a 40° rotation of H6, but the cytoplasmic end of H6 remains partially occluding the G protein-binding cleft (Supplementary Fig. 3), perhaps because the fully active conformation requires the binding of G proteins to stabilize it. Interestingly, the structure of β2AR[6] with a covalently bound agonist is also not in the fully activated R* conformation, which is only seen after the nanobody Nb80 is bound[5]. The importance of the bulge in H5 in the activation of A2AR is highlighted by how inverse agonists bind. Formation of the H5 bulge results in the inward movement of Cys1855.46 (Cβ moves by 4 Å), which in turn causes the movement of Val186 and ultimately a shift of His2506.52 by 2 Å into the ligand binding pocket thereby sterically blocking the binding of ZM241385 (Supplementary Fig. 4). Hence, when the inverse agonist binds, it is anticipated that the H5 bulge is unlikely to form due to the opposite series of events and hence the formation of the R* state is inhibited.

Thus in both βARs and A2AR, the formation of the H5 bulge seems to be a common action of agonists, whereas inverse agonists seem to prevent its formation. However, the energetic contributions to its formation may be different between the two receptors. In the βARs there is a major contribution from direct interaction between the agonist and Ser5.46, while in the A2AR, the major interaction appears to come from polar interactions involving residues in H7 combined with interactions between the agonist and H3. Despite these differences, agonist binding to both receptors involves strong attractive non-covalent interactions that pull the extracellular ends of H3, H5 and H7 together, which is the necessary prerequisite to receptor activation.

While this manuscript was in review, a related manuscript appeared[30], describing the structure of the A2A-T4L chimera bound to the agonist UK432097, which is identical to NECA except for two large substituents on the adenine ring. The structure of UK432097-bound A2A-T4L is very similar to the structures presented here in the transmembrane regions (rmsd 0.6 Å), although there are differences in the extracellular surface due to the bulky extensions of UK432097 interacting with the extracellular loops and the absence of density for residues 149-157. Xu et al. conclude that the structure of UK432097-bound A2A-T4L is in an “active state configuration”, whereas we conclude that the NECA- and adenosine-bound structures are best defined as representing an intermediate state between R and R*.

METHODS SUMMARY

Expression, purification and crystallization

The thermostabilised A2AR-GL31 construct contains amino acid residues 1-316 of the human A2AR, four thermostabilising point mutations (L48A2.46, A54L2.52, T65A2.63 and Q89A3.37) and the mutation N154A to remove a potential N-glycosylation site. A2AR-GL31 was expressed in insect cells using the baculovirus expression system and purified in the detergent octylthioglucoside using Ni2+-NTA affinity chromatography and size exclusion chromatography (see Online Methods). The purified receptor was crystallized in the presence of cholesteryl hemisuccinate by vapour diffusion, with the conditions described in Online Methods.

Data collection, structure solution and refinement

Diffraction data were collected in multiple wedges (20° per wedge) from a single cryo-cooled crystal (100 K) for the GL31-NECA complex at beamline ID23-2 at ESRF, Grenoble, France and from 4 crystals for the GL31-adenosine complex, at beamline I24 at Diamond, Harwell, UK. The structures were solved by molecular replacement using the ZM241385-bound A2A-T4L structure (PDB code 3EML)[8] as a model (see Online Methods). Data collection and refinement statistics are presented in Supplementary Table 1 and omit densities for the ligands are shown in Supplementary Fig. 6.