|Home | About | Journals | Submit | Contact Us | Français|
G protein-coupled receptors play a major role in transmembrane signalling in higher organisms and many are important drug targets. We report the 2.7 Å resolution crystal structure of a β1-adrenergic receptor in complex with the high-affinity antagonist cyanopindolol. The modified turkey receptor had been selected to be in its antagonist conformation and its thermostability improved by earlier limited mutagenesis. The ligand-binding pocket comprises 15 side chains from amino acid residues in 4 transmembrane α-helices and extracellular loop 2. This loop defines the entrance of the ligand-binding pocket and is stabilised by two disulphide bonds and a sodium ion. Cyanopindolol binding to the β1-adrenergic receptor and carazolol binding to the β2-adrenergic receptor involve similar interactions. A short well-defined helix in cytoplasmic loop 2, not observed in either rhodopsin or the β2-adrenergic receptor, directly interacts via a tyrosine with the highly conserved DRY motif at the end of helix 3 that is essential for receptor activation.
G protein-coupled receptors (GPCRs) are a large family of integral membrane proteins that are prevalent in eukaryotes from yeast to man, and which function as key intermediaries in the transduction of signals from outside to inside the cell1. Activating molecules (agonists), such as hormones and neurotransmitters, bind to GPCRs from the extracellular side of the cell membrane and induce a large conformational change which propagates to the cytoplasmic surface2,3, resulting in activation of G proteins and a consequent change in the level of intracellular messengers such as cAMP, Ca2+ or signalling lipids. There are over 800 different human GPCRs4, all sharing the characteristic arrangement of 7 transmembrane α-helices with the polypeptide N-terminus on the extracellular side of the plasma membrane5. Analysis of their primary amino acid sequences has resulted in the definition of a number of families6, the largest of which, family A, includes the archetypal GPCR, rhodopsin. The three human β-adrenergic receptor (βAR) subtypes, β1, β2 and β3, belong to family A and share 51% sequence identity between Trp1.31 - Asp5.73 and Glu6.30-CysH8-Cterm i.e. excluding the N- and C-termini and most of cytoplasmic loop 3 (Supplementary Fig 1; superscripts refer to Ballesteros-Weinstein numbering7). Drugs that inhibit β1 and β2 receptor signalling (antagonists and inverse agonists) are used to modulate heart function and are known as β-blockers8, but selective β1-antagonists are preferred since they have fewer side effects due to bronchial constriction via β2 receptors in the lung. In contrast to the β1 and β2 receptors, the β3-adrenergic receptor (β3AR) is found in adipose tissue, where adrenaline stimulates metabolism, and is a potential target to treat obesity. Elucidation of the specificity determinants for drug affinity of the different βAR subtypes will allow the development of better subtype-specific β-blockers, with fewer side effects.
A milestone in the study of βARs was recently reached with the publication of a β2-adrenergic receptor (β2AR) structure in complex with an antibody fragment, β2AR:Fab9, followed by the higher resolution structure of an engineered β2AR fused in the middle of the third cytoplasmic loop (CL3) to T4 lysozyme, β2AR-T410. These structures, both containing the high affinity antagonist carazolol, defined the overall architecture of β2AR, and the structure of the ligand-binding pocket. However, the structures also raised questions of how a range of compounds can bind to the different but closely related βAR subtypes with different affinities. For example, the human β1 and β2 receptors are 67% identical within their transmembrane regions, but the residues that directly surround the ligand binding pocket appear to be identical. Despite these similarities, larger antagonists such as CGP 20712A (see Supplementary Fig. 2) bind 500 times more strongly to β1AR than to β2AR, whilst ICI 118551 shows a 550 fold specificity for β2AR over β1AR11. There are also β1 and β2 specific agonists12. As an important step towards understanding subtype specificity, we have determined the structure of a β1-adrenergic receptor (β1AR).
GPCR crystallisation is challenging, since GPCRs are usually unstable in detergent, contain unstructured regions, and spontaneously cycle between an inactive antagonist state (R) and an active agonist state (R*), which may further decrease the stability13. The human β1AR is more difficult to purify than β2AR, because it is very unstable in detergent. We therefore used turkey (Meleagris gallopavo) β1AR, which is more stable than human β1AR14 though less stable than human β2AR (MJSV and CGT, unpublished observation). A mutated receptor, β1AR-m23, was constructed with enhanced thermostability over the wild-type receptor of 21°C and an altered equilibrium between R and R* so that the mutant receptor was preferentially in the antagonist (R) state15. The receptor construct, β1AR36-m23 (Fig. 1) purified in octylthioglucoside and in the presence of cyanopindolol gave good crystals showing isotropic diffraction beyond 2.7Å.
The mutant receptor β1AR-m23 bound the antagonists dihydroalprenolol and cyanopindolol with similar affinities to the wild-type receptor, but the agonists noradrenaline and isoprenaline bound more weakly by a factor of 2470 and 650 respectively15. This reflects a change in the R to R* equilibrium of the receptor towards the antagonist R state. From this we predicted that, in a G protein-coupling assay, the receptor would show no basal activity and that the concentration of agonist required for signalling would be orders of magnitude higher. Signalling assays were performed on stable cell lines expressing the wild type β1AR truncated at the N- and C- termini (β1ARtrunc) and also containing the six thermostabilising mutations (m23) (Supplementary Fig. 3). β1ARtrunc-m23 coupled efficiently to G proteins and elicited a robust stimulation of cAMP-responsive reporter gene, although the agonist concentration response curve, as expected was shifted to the right16. The drug ICI 118551, an inverse agonist for both β1AR17 and β2AR18, showed no reduction in the basal level of cAMP when added to cells containing β1ARtrunc-m23, at a concentration 100-fold above its Ki, implying there is negligible basal constitutive activity. The structure we have determined contains the very high affinity antagonist cyanopindolol in the binding pocket and represents closely the inactive conformation with respect to G protein coupling.
The structure was solved by molecular replacement to 2.7 Å resolution with an Rwork of 0.212 and Rfree of 0.268 (Supplementary Table 1). The four receptor molecules in the unit cell, labelled A-D (Supplementary Figs. 4-6), were all very similar except that molecules A and D both had a 60° kink in helix 1 (H1). Also modelled are 31 water molecules, 4 Na+ ions and 14 detergent molecules (see Supplementary Information). Unless otherwise stated, all further discussion refers to molecule B, as this molecule has an unkinked H1 and a relatively well-ordered H8. The helix boundaries, disordered regions and overall structural motifs are presented in Fig. 1.
The amino acid sequence of turkey β1AR19 is 82% and 67% identical to human β1AR and human β2AR, respectively, over residues Trp401.31-Asp2425.73 and Glu2856.30-Cys358H8-Cterm (i.e. excluding the N- and C- termini and most of CL3) and it is therefore expected that the structure of the transmembrane regions of β1AR and β2AR should be very similar. Our superposition of β2AR (PDB code 2RH1) and β1AR (chain B) is based on selected residues in H3, H5, H6 and H7 since we were particularly interested in comparing the ligand binding pockets; 78 alpha carbons can be superimposed with an rmsd of 0.25 Å. The rmsd over all transmembrane helices is 0.7 Å (269 Cα atoms; Supplementary Fig 7). Comparison of the β1AR and β2AR structures reveals no evidence for any significant changes in backbone conformation at the sites of the six point mutants introduced15 to stabilise β1AR. This is consistent with the observation that β1AR-m23 binds antagonists with similar affinities to the wild type receptor15 and that it can couple efficiently to G proteins, though at higher agonist concentration (Supplementary Fig. S3). The basis for the thermostabilisation by the six mutations R681.59S, M902.53V, Y2275.58A, A2826.27L, F3277.37A and F3387.48M is not immediately apparent from the structure.
The structures of the three extracellular loops (EL1-3) in β1AR are very similar to those of β2AR (Cα rmsd 0.8 Å), consistent with the high sequence conservation of these regions in the βAR family (Supplementary Fig. 1). On the extracellular surface, a clear peak in the electron density is present at a position co-ordinated by the carbonyl groups of residues Cys192, Asp195, Cys198 and a water molecule (Supplementary Fig 8). This density was assigned to a sodium ion based on its coordination geometry20. Its role, bound at the negative end of the EL2 α-helix dipole, may be to stabilise the helical conformation of EL2 and thus the structure of the entrance to the ligand-binding pocket. The large difference in EL2 conformation between the α-helix found in β2AR and the β-hairpin that closes off the retinal binding site in rhodopsin is confirmed in the structure of β1AR, suggesting that the α-helix may be a common feature in GPCRs that bind their ligands rapidly and reversibly.
In all GPCRs, CL2 and CL3 are believed to play an important role in the binding, selectivity and activation of G proteins, CL2 being important for the strength of the interaction and CL3 for specificity21,22,23,24,25. The β1AR and β2AR structures, along with rhodopsin26, have similar conformations for CL1, but there are major differences in CL2 and CL3. The CL3 differences are not of physiological relevance because they arise from deletions (β1AR), deletion and insertion of T4 lysozyme (β2AR-T4) or formation of an antibody complex (β2AR:Fab), with only the rhodopsin structure having a native CL326. However, differences in the conformation of CL2 (Fig. 3) are important, because this region is very highly conserved between β1AR and β2AR, though poorly conserved with rhodopsin. In β1AR, CL2 forms a short α-helix (residues Pro1463.57-Leu1523.63; Supplementary Fig. 9) parallel to the membrane surface whereas in both β2AR structures and in rhodopsin this loop is in an extended conformation (Fig. 2). This short α-helix cannot be accommodated in either the β2AR:Fab complex9 or the β2AR-T4 fusion10 crystal structures perhaps because of lattice contacts with adjacent molecules. In β1AR, CL2 also makes lattice contacts, but these are different between each of the four molecules and it is therefore likely that the helical conformation found here represents the physiologically relevant structure for all βARs in the inactive conformation.
The CL2 loop has been proposed to function as the switch enabling G protein activation21 and it is clear from the β1AR structure that this short α-helix interacts directly with the highly conserved Asp1383.49Arg1393.50Tyr1403.51 (DRY) motif in H3. Tyr149 in CL2 is located sufficiently close to Asp1383.49 to allow the formation of a hydrogen bond (Fig. 2) between the tyrosine hydroxyl and the aspartate side chain. Supporting evidence for this structural role of Tyr149 comes from the observation that the Y149A mutation makes β1AR less thermally stable (Supplementary Table 2). The equivalent Tyr141 in both β2AR structures is in a cavity between H3, H4 and H6, but the biological relevance of this is unclear, due to the perturbations in this region caused by either the T4 lysozyme fusion or by the bound antibody. Interestingly, a pattern of mutations consistent with an α-helical conformation for CL2 was found in the muscarinic M5 receptor and the equivalent M5 mutation Y138A led to increased constitutive activity21. Thus it is likely that both the tyrosine residue and the CL2 α-helix play key roles in G protein coupling.
A salt bridge between Arg3.50 and Glu6.30, termed the “ionic lock”, (Fig. 2) was proposed to play an essential role in maintaining GPCRs in an inactive state27 but to break upon receptor activation. Since the β1AR structure represents a receptor lacking basal activity and complexed with an antagonist, it is highly likely to represent the R conformation. Yet this salt bridge is not present either in the β1AR or the β2AR structures (Fig. 2). This suggests that the ionic lock is not an essential feature of the inactive state. Even in dark-state rhodopsin, where these two charged residues are within hydrogen bonding distance 26,28,29, the side chain B-factors of the two residues differ greatly (20-40 Å2)26 so there is no direct experimental evidence for any “lock”.
Cyanopindolol is similar in structure to carazolol (Supplementary Fig. 2) that is present in the ligand binding pocket of both β2AR structures; both ligands bind with very high affinity to all βARs. In the β1AR structure there are 15 amino acid residues, using a 3.9Å distance criterion, whose side chains make contacts with cyanopindolol in the ligand binding pocket; 4 side chains are from H3, 3 from H5, 4 from H6, 2 from H7 and 2 from EL2 (Fig. 3). All these residues are identical to those in human β2AR and the mode of binding of cyanopindolol to β1AR is, therefore, similar to that of carazolol in β2AR. However, the extra ring in the carazolol heterocyclic ring, due to van der Waals contact with Tyr1995.38 in β2AR, pushes the ligand more deeply into the binding site. The nitrogen in the cyano-moiety of cyanopindolol makes a weak hydrogen bond with the hydroxyl of Thr2035.34, which is located together with Phe2015.32 on EL2 (Fig. 3). The same hydrogen bonds between the ligand and Asp1213.32, Asn3297.39 and Ser2115.42 are present in both β1AR and β2AR structures, but the side chain rotamer conformation of Ser2115.42 is different (Fig 4 and Supplementary Data).
To explain why some ligands preferentially bind to either β1AR or β2AR, which is important in understanding the sub-type specificity of the human receptors 11, there must be differences in amino acid residues close to the ligand binding pocket that directly or indirectly affect binding. A comparison of residues within 8Å of the binding pocket identified only two residues that are different between human β1AR and β2AR subtypes. The respective residues are Val1724.56 and Phe3257.35 in β1AR, equivalent to Thr1644.56 and Tyr3087.35 in β2AR. These differences introduce polar residues near the binding pocket of β2AR relative to β1AR (Fig. 4), which could affect ligand selectivity. Tyr3087.35 has also been implied by mutagenesis studies30,31 to be important for agonist selectivity in β2AR. In β2AR, Tyr3087.35 is positioned close to the binding pocket and can form a hydrogen bond to Asn2936.55. In β1AR the side chain of Asn3106.55 is closer to the cyano group of cyanopindolol and the equivalent residue, Phe3257.35, is further from the binding pocket (Fig. 4). As a result, there is no contact between Phe3257.35 in β1AR and cyanopindolol.
Part of the ligand binding site is formed by EL2. Although the backbone positions within this highly structured region of β1AR differ from β2AR by an rmsd of only 0.84Å, compared with 0.63Å between the same residues in molA and molB, there are significant amino acid side chain differences in this region. These change the shape and charge distribution around the entrance to the ligand binding pocket (Supplementary Fig. 10) with an ion pair formed between Asp1925.31 and Lys3057.32 in β2AR that is absent in β1AR because the respective residues are both aspartate (Asp2005.31 and Asp3227.32). Differences between β1AR and β2AR in this region could affect ligand binding especially for larger ligands with extensions that make direct interactions with non-conserved side chains. Recent mutational studies show that EL2 influences the specificity of ligand binding to both the normal (orthosteric) site32,33 and the sites of allosteric modulators34, and that the loop flexibility is important to the binding kinetics35.
The structure of β1AR, when compared to β2AR, provides a sound basis for studying selectivity differences between βAR antagonists structurally similar to cyanopindolol and carazolol. However, many ligands, such as the inverse agonist CGP 20712A (Supplementary Fig. 2), show very high selectivities11 but are physically larger and structurally distinct from either cyanopindolol or carazolol. These ligands could well make contact with residues other than those described here.
The β1AR crystal structure shows the inactive state of the receptor but it is notable that many agonists, including the natural ligands adrenaline and noradrenaline are smaller than many of the best antagonists, including cyanopindolol. Agonists have a shorter distance, by two carbon-carbon bonds or 2-3 Å, between the catechol hydroxyl groups or their equivalent and the obligatory amine nitrogen. We superimposed (Fig. 4B) a model of adrenaline with that of cyanopindolol and examined its relationship to the side chains of Asp1213.32and Asn3297.39, which make hydrogen bonds with the amine, and those of Ser2115.42, Ser2125.43 and Ser2155.46, which are expected to hydrogen bond with the meta- and para-hydroxyl groups on the catechol ring36,37,38. As noticed by Rosenbaum et al.39, the catechol hydroxyl groups are well spaced and well oriented to interact with the side chain hydroxyl groups of Ser2115.42/Ser2125.43 and Ser2155.46 on H5, but cannot reach far enough to make good hydrogen bonds if the amine occupies the same position as it does adjacent to Asp1213.32 in the cyanopindolol complex, without a substantial structural change in the receptor. It seems very reasonable that the ligand binding site in β1AR will contract by 2-3 Å on activation so that both ends of adrenaline can make good interactions with the residues on H3/H7 and H5. This view is also supported by engineered zinc binding sites that activate the receptor40,41. How this tightening around the ligand binding site could propagate to the cytoplasmic surface and cause an outward 5-6 Å movement of H62,3 is difficult to predict, since all the transmembrane helices except H1 and H3 have pronounced kinks at conserved proline residues, which means they could easily bend. However, one speculation is that the pulling of H5 towards the centre of the receptor on activation could force H3 and H6 apart, causing cytoplasmic loops CL2 and CL3 to move apart as observed in photoactivated rhodopsin3 and trigger recruitment of G protein complex.
The β1AR construct T34-424/His642 was the starting point for the generation of the β1AR36-m23 construct that crystallized. The C-terminus was further truncated after Leu367, and 6 histidines were added. Two segments, comprising residues 244-271 and 277-278 of CL3, were also deleted. The construct included the following 8 point mutations: C1163.27L increased expression; C358A at the C-terminus of H8 removed palmitoylation and helped crystallisation; R681.59S, M902.53V, Y2275.58A, A2826.27L, F3277.37A and F3387.48M thermostabilised the receptor in the antagonist conformation15. Baculovirus expression and receptor purification42 were performed in decylmaltoside, with the detergent exchanged to octylthioglucoside on the alprenolol sepharose column. Crystals were obtained by vapour diffusion at 18°C with hanging drops after addition of an equal volume of reservoir solution, 0.1M N-(2-acetamido)iminodiacetic acid:NaOH pH 6.9-7.3 and 29-32% PEG600 to receptor concentrated to 6.0 mg/ml.
Diffraction data were collected from many crystals on beamlines ID13 and ID23-2 at ESRF, Grenoble43,44; the data used for structure determination were collected at ID23-2 with a 10 μm beam using three positions on a single cryo-cooled crystal (100 K). Images were processed with MOSFLM and SCALA45. The structure was solved by molecular replacement with PHASER46, using the structure of human β2AR10 as an initial model. All four copies of the molecule in the triclinic unit cell were located (Supplementary Figs. 4 and 5). The amino acid sequence was corrected, the model refined with PHENIX47 and rebuilt with O48 (see Full Methods online for further details). An overview of the B-factor distribution for β1AR molecules A and B is shown in Supplementary Fig. 6. Figures were produced using Pymol (DeLano Scientific LLC).
Supplementary information includes Figures 1-10 and Tables 1-3.
Supplementary Table 1 presents data on the structure refinement and crystal parameters. Supplementary Table 2 presents data on the stability of β1 receptor mutants containing alanine residues throughout cytoplasmic loop 2. Supplementary Table 3 presents the amino acid residues involved in crystal contacts. Supplementary Fig. 1 is an alignment of the primary sequences of turkey β1AR and the human β-adrenergic receptors and Supplementary Fig. 2 shows the structures of agonists and antagonists discussed in the main text. Supplementary Fig. 3 shows G protein coupling data and basal activity for β1AR-m23 in a cell-based assay. Supplementary Figs. 4 and 5 describe the crystallographic packing of receptor molecules and show the location of the ordered detergent molecules. Supplementary Fig. 6 shows the B-factor variation for the structures of rhodopsin, β1AR and β2AR. Supplementary Fig. 7 describes the rmsd according to sequence for the comparison between the β1 and β2 receptors and between molecules A and B of β1AR-m23. Supplementary Fig. 8 depicts electron density maps for the Na+ ion and the ordered water molecule in H6. Supplementary Fig. 9 is an omit map of cytoplasmic loop 2 and Supplementary Fig. 10 shows the extracellular surface charge distribution for the β1 and β2 receptors.
This work was supported by a joint grant from Pfizer Global Research and Development and from the MRCT Development Gap Fund to CGT and RH, in addition to core funding from MRC. GFXS was financially supported by a Human Frontier Science Project (HFSP) programme grant (RG/0052), a European Commission FP6 specific targeted research project (LSH-2003-1.1.0-1) and an ESRF long-term proposal. JGB is funded by a Wellcome Trust Clinician Scientist Fellowship. We thank Elliott Ross (UTSW) for his support in the initial stages of the β1AR project at the LMB and for his comments on the manuscript. In addition, we would also like to thank Reinhard Grisshammer (NIH), Ed Hulme (NIMR), Fiona Marshall (Heptares), Malcolm Weir (Heptares) and our colleagues at LMB for their comments on the manuscript. We also thank beamline staff at the European Synchrotron Radiation Facility, particularly C. Riekel and M. Burghammer at ID13 and D. Flot and S. McSweeney at ID 23-2.
Conflicts of interest GFXS, CGT and RH are all on the Scientific Advisory Board of Heptares Therapeutics Ltd.
Author information. Co-ordinates and structure factors have been submitted to the PDB database under accession code 2VT4.