Crystal Structures of a Stabilized β₁-Adrenoceptor Bound to the Biased Agonists Bucindolol and Carvedilol

Structures of β₁AR Bound to Bucindolol and Carvedilol

The avian β₁AR was modified to allow crystallization by the removal of flexible regions at the N terminus, C terminus, and in cytoplasmic loop 3, and the introduction of 8 point mutations (see Experimental Procedures), to give the construct β₁AR44-m23. Six of the point mutations result in thermostabilization of β₁AR (Serrano-Vega et al., 2008), which allows crystallization in short-chain detergents (Warne et al., 2008). A consequence of thermostabilization is that the receptor is preferentially in the antagonist-bound state (R) (Serrano-Vega et al., 2008), although β₁AR-m23 is capable of binding agonists with a similar rank order of potency to the wild-type receptor and can couple efficiently to G proteins (Baker et al., 2011). It is not feasible to perform crystallography on human β₁AR due to its extreme instability (Serrano-Vega and Tate, 2009), but the high sequence identity between the receptors (82%) in the transmembrane domains and loop regions (except most of cytoplasmic loop 3) shows β₁AR-m23 is an excellent model for studying ligand-receptor interactions. Indeed, crystal structures of β₁AR-m23 (Warne et al., 2008) and β₂AR-T4L (Cherezov et al., 2007) show high similarity in the transmembrane domains (rmsd 0.7 Å) and especially in the region of the ligand binding pocket (rmsd of 0.25 Å for 78 Cα atoms). None of the mutations in β₁AR-m23 are in the ligand binding pocket and none of the sites of the mutations show a conformational change when comparing the structures of β₁AR-m23 and β₂AR-T4L.

Receptors were expressed, purified and crystallized as previously described. The cocrystal structures of β₁AR44-m23 bound to either bucindolol or carvedilol were solved at resolutions of 3.2 and 2.3 Å, respectively (Figure 2 and Table 1). Overall the two structures are very similar (RMSD of Cα positions <0.5 Å) to the inactive state structure of β₁AR (Protein Data Bank [PDB] 2VT4) with bound cyanopindolol (Warne et al., 2008). The lack of any significant conformational change at the ends of helices 5 and 6 is consistent with previous R-state structures of β₁AR-m23 with bound antagonists and agonists and of β₂AR complexed with either antagonists or with a covalently bound agonist. Thus, the structures presented here of β₁AR bound to either bucindolol or carvedilol show in detail the ligand-receptor interactions (see below), but they do not define a new conformation of the receptor involved in G protein-independent signaling. It is likely that a complex of a phosphorylated βAR with arrestin will be required to fully understand the conformational change induced by the binding of biased agonists, because the signaling conformation of the receptor that allows arrestin binding is likely to be transient. This is consistent with the observation that the binding of a G protein or analog was required to obtain the structure of an R^∗ state of β₂AR (Rasmussen et al., 2011a, 2011b), while the structure of β₂AR bound to a covalent agonist is in the R state (Rosenbaum et al., 2011). Therefore, it seems unlikely that any key conformational changes are missing due to the presence of the thermostabilizing mutations. In contrast to the βARs, some receptors evince considerable conformational changes when crystallized in the presence of an agonist. For example, the structure of the adenosine A_2A receptor bound to an agonist is clearly in an R^∗-like state, without the requirement for binding a G protein or G protein mimetic (Lebon et al., 2011; Xu et al., 2011).

Ligand Binding in the Catecholamine Binding Pocket

The structures of βAR ligands are often very similar, particularly in the region of the secondary amine and β-hydroxyl groups and these are also conserved in bucindolol and carvedilol (Figure 1). The cocrystal structures show that their secondary amine and β-hydroxyl groups form potential hydrogen bonds with Asp121^3.32 and Asn329^7.39 and their “head groups” (equivalent to the catechol moiety in adrenaline) occupy a position adjacent to H5 (Figure 3). Thus, both ligands exhibit the same general mode of binding observed for other βAR antagonists cocrystallized with either β₁AR or β₂AR (Cherezov et al., 2007; Hanson et al., 2008; Moukhametzianov et al., 2011; Wacker et al., 2010; Warne et al., 2008).

Previously, we showed that binding of a full agonist to β₁AR resulted in three differences in receptor conformation compared to when an antagonist was bound, namely the rotamer conformation changes of Ser212^5.43 and Ser215^5.46, and a contraction of the ligand binding pocket. Comparison of the structures of β₁AR bound to bucindolol and carvedilol with the previously determined structures shows that the rotamer configuration of Ser215^5.46 allows the formation of an interhelical hydrogen bond with H3, as seen in the structures with bound partial agonists, but not with bound full agonists where Ser215^5.46 forms a hydrogen bond directly to the ligand (Figure 4). However, the configuration of Ser212^5.43 is similar to that seen in the agonist-bound structures where it makes a potential hydrogen bond to Asn310^6.55. A recently determined structure of β₁AR with cyanopindolol bound in lipidic meso phase at a resolution of 2.1 Å (J.L. Miller and C.G.T., unpublished data) also shows this alternative rotamer conformation of Ser212^5.43 compared to the previous structure of cyanopindolol-bound β₁AR determined in detergent (Warne et al., 2008); these data suggest that Ser212^5.43 can also be in this alternative conformation with an antagonist bound, and the configuration of Ser212^5.43 is therefore unlikely to represent an agonist-specific conformation. Both the bucindolol and carvedilol bound structures also do not exhibit the contraction of the binding pocket observed in the structures with full and partial agonists bound. It is therefore clear that the ligands do not induce the initial conformational changes in the receptor that are characteristic of agonists that activate G proteins. Bucindolol has been variously reported as being either an inverse agonist or partial agonist of the β₁AR, depending on the system or tissue studied (Andreka et al., 2002; Engelhardt et al., 2001; Galandrin et al., 2008; Maack et al., 2000, 2003). However, in both bucindolol-bound and carvedilol-bound β₁AR structures, the characteristic rotamer conformation change of Ser215^5.46 observed in structures with a full agonist bound is sterically blocked by the ligand, which is a characteristic of βAR inverse agonists (Warne et al., 2011).

Ligand Binding in the Extended Ligand Binding Pocket

Both bucindolol and carvedilol, unlike all other antagonists cocrystallized with βARs, have bulky aromatic substituents at their amine ends that make additional contacts in the extended ligand binding pocket composed of residues in helices 2, 3, and 7 and extracellular loop 2 (Figure 4). The additional contacts are detailed along with all other ligand-receptor contacts in Table 2. It therefore seems logical to propose that the ability of bucindolol and carvedilol to stimulate G protein-independent signaling resides in the extensions in the tail region of the ligand that are absent from all other antagonists. However, it is difficult to say how these additional contacts might promote G protein-independent signaling, because the conformations that might finally promote the binding of arrestin or other signaling proteins are currently unknown. However, it seems plausible that the additional contacts may result in an increased probability of subtle conformational changes that might be transmitted to the receptor's C terminus, where phosphorylation by GPCR-specific kinases promotes binding of arrestin and signaling (Granier et al., 2007; Nobles et al., 2011).

Bias of conventional G protein agonists of βARs toward the arrestin-signaling pathway has been investigated and methods to discern levels of G protein-independent activity that are relatively low compared to the dominant G protein-signaling activity are being developed. This has led to the categorization of a number of ligands that can activate both G protein-coupled and G protein-independent pathways, but to different extents (Rajagopal et al., 2011). However, currently there is not a comprehensive list of the propensity of all βAR ligands for inducing signaling via G protein-independent pathways. So far, bias toward arrestin signaling among conventional G protein agonists has only been detected in ligands with either ethyl substitutions at the Cα, or amine-end substituents, such as those present in bucindolol and carvedilol (Drake et al., 2008; Rajagopal et al., 2011). Of the conventional G protein agonists that have been cocrystallized previously with either β₁AR or β₂AR, dobutamine, isoprenaline and salbutamol have been identified as nonbiased agonists of the β₂AR (Rajagopal et al., 2011). The structure of β₁AR bound to carmoterol has been determined (Warne et al., 2011), but this ligand has not been tested for signaling bias. However, formoterol, which is weakly arrestin biased, is structurally identical to carmoterol apart from a minor difference in its head group; both ligands have the same methoxyphenyl amine end extension (see Figure 1 for the structures) (Rajagopal et al., 2010). The cocrystal structure of carmoterol with the β₁AR indicates additional interactions of the methoxyphenyl group with extracellular loop 2 (EL2) as well as H7, and NMR data suggest that formoterol's methoxyphenyl group also interacts with residues on EL2 of the β₂AR (Bokoch et al., 2010). Thus, of the three ligands with amine-end extensions that bind in the extended ligand binding pocket for which there are cocrystal structures and that have been examined for bias, bucindolol and carvedilol are biased agonists, whereas dobutamine is not. As in all probability, formoterol, which also shows weak arrestin bias also binds to EL2, one structural feature that may correlate with agonist bias is that bucindolol, carvedilol, and most likely formoterol all interact with EL2, whereas dobutamine does not. Whether this observation extends to other biased agonists will require further detailed characterization of more βAR ligands with both β₁AR and β₂AR rather than just a limited few.

While this manuscript was in review, a related manuscript appeared (Liu et al., 2012), which detailed ¹⁹F-NMR studies on β₂AR bound to various ligands. Specific Cys residues in detergent-solubilized, purified β₂AR were covalently modified with trifluoroethanethiol and then ¹⁹F spectra were collected in the presence of inverse agonists, agonists, or biased agonists. The most significant difference observed when an inverse agonist was bound compared to when a biased agonist was bound was a change in the spectrum of ¹⁹F-labeled Cys327; this residue is in the short linker between H7 and H8. These data indicate that the environment around Cys327 is different when a biased agonist is bound compared to when an inverse agonist is bound, which is consistent with the interpretation of the structural data presented here and previous biochemical data (Granier et al., 2007; Nobles et al., 2011).

Understanding the Effects of the Arg389Gly Polymorphism in Human β₁AR

The high-resolution of the β₁AR-carvedilol complex fortuitously also allows us to suggest a mechanism for the difference in pharmacology in the frequently occurring Arg389Gly polymorphism of the human β₁AR. The more common β₁AR-Arg389 variant has a slightly higher basal G protein activity and a 3-fold increase in agonist response compared to the β₁AR-Gly389 variant (Mason et al., 1999). It has been suggested that the affected residue is in an area important for G protein coupling (Mason et al., 1999), but it is now clear from the structure of the β₂AR-Gs complex that this is not the case (Rasmussen et al., 2011b). Therefore, the differences in pharmacology between the Arg/Gly variants must lie within the receptor itself.

The human β₁AR residue affected by the polymorphism, Arg389^8.56, is equivalent to Arg355^8.56 in the turkey β₁AR and it is located on H8, with its guanidinium group close to the end of H7. In the 2.3 Å resolution structure of carvedilol-bound β₁AR-m23, the side chain of Arg355^8.56 is well resolved for the first time in any structure (Figure 5) and it forms hydrogen bonds with Ser68^1.59 and Thr69^1.60 at the intracellular end of H1. The hydrogen bonds between H8 and H1 would be expected to stabilize the receptor, which is substantiated by the fact that Ser68^1.59 is one of the six thermostabilizing mutations that facilitated crystallization of β₁AR-m23 (Serrano-Vega et al., 2008; Serrano-Vega and Tate, 2009). In the wild-type turkey β₁AR, the equivalent residue to Ser68^1.59 is Arg68^1.59 and when this is mutated to an uncharged residue (Ala) the receptor becomes more thermostable, presumably because the proximity of Arg68^1.59 and Arg355^8.56 is electrostatically unfavorable. Mutation of Arg68^1.59 to Ser increases the thermostability further, which is consistent with the formation of the hydrogen bond to Arg355^8.56 observed in the structure with bound carvedilol. Increased thermostability of β₁AR is likely to reflect a decrease in the global flexibility of the whole receptor. Thus, changes in thermostability due to mutations at the H1-H8 interface suggests that analogous changes in homologous receptors may also change the global dynamics of these receptors. These observations in the avian β₁AR probably apply to human β₁AR and β₂AR as the equivalent residues to Arg68^1.59 are Lys85^1.59 and Lys60^1.59. Thus, the Arg/Gly389 polymorphism in human β₁AR will similarly alter the packing between H1 and H8, which could result in changes in the dynamics of the receptor, with the barrier to formation of the R^∗ state perhaps being lower in the β₁AR-Arg389 variant. This is indeed what has been observed pharmacologically and biochemically in a number of studies (Liggett et al., 2006; Rochais et al., 2007; Swift et al., 2008). It has also been observed that carvedilol is a more efficient inverse agonist of the human β₁AR-Arg389 isoform than are metoprolol and bisoprolol (Rochais et al., 2007), two beta blockers that do not have extensions at their amine ends (see Figure 1). These findings have led to interest in the potential of genetically targeted therapies for heart failure with bucindolol or carvedilol (DeGeorge and Koch, 2007; Pleger and Koch, 2006). A similar polymorphism at Trp64^1.59 in the human β₃AR has also been observed, where substitution of Trp for Arg has been associated with compromised activity of the receptor and, consequently, increased obesity (Kimura et al., 2000; Kurokawa et al., 2008; Piétri-Rouxel et al., 1997).

Conclusions

The structures of β₁AR bound to either bucindolol or carvedilol show that both of these ligands make additional contacts to helices 2, 3, and 7 and extracellular loop 2 compared to other structurally characterized βAR inverse agonists. Overall the structures show no conformational change when compared to other β₁AR antagonist structures, but it is probable that the additional interactions in the extended ligand binding pocket might increase the likelihood of subtle conformational changes that result in enhanced arrestin binding and G protein-independent signaling. The fact that both bucindolol and carvedilol bind to β₁AR in a similar manner to other βAR G protein antagonists, yet they can stimulate signaling via G protein-independent pathways while apparently inhibiting G protein coupling, strongly supports the contention that arrestin can bind to a different conformation of the receptor to that bound by G proteins.

Expression, Purification, and Crystallization

The turkey (Meleagris gallopavo) β₁AR construct, β44-m23, contains six thermostabilizing point mutations, two point mutations which improve receptor expression and homogeneity, and truncations at the N terminus, inner loop 3 and C terminus (Warne et al., 2011). Baculovirus expression and purification were all performed as described previously, with the detergent exchanged to Hega-10 (0.35%) on the alprenolol Sepharose affinity column (Warne et al., 2003, 2008, 2009, 2011). Purified receptor was competitively eluted from the alprenolol Sepharose column with either bucindolol or carvedilol, but this was difficult because of the poor solubility of the ligands. Ligands were added to saturation from 20 mg/ml DMSO stock solutions to the elution buffer (10 mM Tris-HCl [pH 7.4], 100 mM NaCl, 0.1 mM EDTA, 0.35% Hega-10) with rapid stirring. The approximate final concentrations of the ligands in the elution buffer were 10 μM for carvedilol and 50 μM for bucindolol. Receptor was concentrated to 20 mg/ml in 10 mM Tris-HCl (pH 7.6), 100 mM NaCl, 0.1 mM EDTA, 0.35% Hega-10. Before crystallization, Hega-10 was added to 0.5%. Crystals were grown at 4°C by vapor diffusion in sitting drops with 150 nl receptor + 150 nl precipitant (0.1 M bicine [pH 9.0], 25% PEG 600 in both cases) and cryoprotected by addition of 60% PEG 600 for 1 min before mounting on Hampton CrystalCap HT loops and cryocooling in liquid nitrogen.

Data Collection, Structure Solution, and Refinement

For both complexes, diffraction data were collected from a single cryocooled crystal (100 K) at the European Synchrotron Radiation Facility, Grenoble, France, with a Mar 225 CCD detector on beamline ID23-2 (wavelength, 0.8726 Å) using a 10 μm focused beam. The microfocus beam was required for the location of the best diffracting parts of crystals, as well as allowing wedges of data (20–80°) to be collected from different positions on the crystal. For the β₁AR crystals grown in the presence of bucindolol or carvedilol, 9 or 16 wedges of data from single crystals were merged, respectively. Images were processed with MOSFLM (Leslie, 2006) and SCALA (Evans, 2006). Both structures were solved by molecular replacement with PHASER using the β₁AR44-m23 structure with the agonist carmoterol bound (PDB code 2Y02) as a starting model (McCoy et al., 2007). Refinement, rebuilding and validation were carried out with REFMAC5 (Murshudov et al., 1997), COOT (Emsley and Cowtan, 2004) and MOLPROBITY (Davis et al., 2007). Noncrystallographic symmetry restraints were applied as appropriate between the two monomers in the asymmetric unit for both structures, using the electron density maps and R_free values to judge which residues should be excluded. The two independent copies of the receptor in the asymmetric unit are very similar for the bucindolol complex, although the ligand density was better defined for monomer A. In the carvedilol complex, there is a distortion of the ligand binding pocket in monomer A due to lattice contacts and monomer B represents the more physiologically relevant conformation.