G protein-coupled receptor inactivation by an allosteric inverse-agonist antibody

The GPCR structures in an inactive conformation solved recently^3-12 largely advance our understanding of the molecular signalling mechanisms of the receptors. The first details of GPCR activation were provided by the structure of bovine opsin in an active conformation complexed with a G-protein C-terminal peptide (GαCT)¹. Most recently, Kobilka and colleagues obtained the crystal structures of β₂AR in an active state with a camelid antibody fragment (nanobody, Nb80)² and with a heterotrimeric Gs-protein¹³. In these structures, the complementarity-determining region (CDR-3) of Nb80 and C-terminal α-helix of a subunit (Gαs) of Gs-protein were located in the same pocket as for GαCT in the opsin structure. They showed that Nb80 and Gs protein change the conformational equilibrium of β₂AR toward the active state in a similar manner, thereby substantially increase their agonist affinities^2,13.

A_2AAR is responsible for regulating blood flow to the cardiac muscle and is important in the regulation of glutamate and dopamine release in the brain¹⁴. Caffeine is a well-known antagonist of this receptor. Strong epidemiological evidence indicates that coffee drinkers have a lower risk of Parkinson’s disease¹⁵. The structure of A_2AAR has been reported^9,16 as a complex with both an antagonist (ZM241385) and an agonist (UK-432097). These structures reveal the molecular framework of the receptor; however, in both cases the intracellular loop 3 (ICL3), critical for G-protein binding, has been replaced by T4-lysozyme (T4L).

Here, we report the crystal structure of A_2AAR with complete ICL3 in complex with a mouse monoclonal-antibody Fab-fragment, Fab2838. A_2AAR was expressed in Pichia pastoris and the antibody was raised to the purified receptor with antagonist (ZM241385) bound using the conventional mouse-hybridoma system combined with improved immunisation and screening methods (for details, see Methods). Fab2838, a Fab fragment generated from one (IgG2838) of the obtained antibodies completely inhibited binding of the agonist [³H]-NECA but did not affect binding of the antagonist [³H]-ZM241385 (Fig. 1a,d and Supplementary Fig. 2). The results were confirmed by competition binding assays (for details, see Supplementary Discussion and Fig. 1). These findings suggest that Fab2838 induces an inactive conformation, (i.e. to which agonist cannot bind) of the A_2AAR ligand-binding pocket without blocking the ligand-binding site.

We crystallised A_2AAR with Fab2838 in the presence of ZM241385 and solved the structure at a resolution of 2.7 Å (Supplementary Table 2). Since the occupancy of ZM241385 in the structure was low for unknown reasons, we repeated the experiments and obtained a higher occupancy structure at 3.1 Å (Supplementary Table 2 and Supplementary Fig. 3 and 4). Except for the occupancy of the ligand, the two structures are almost identical (RMSD of Cα; 0.57 Å) (Supplementary Table 2). ZM241385 occupies the ligand-binding pocket on the extracellular side by making hydrophobic interactions with F168^5.29 and I274^7.39, and hydrogen-bonds with N253^6.55 as observed in the A_2AAR-T4L structure (Supplementary Fig. 4). While the overall structure of A_2AAR in the A_2AAR-Fab2838 complex is similar to that of the T4L construct (PDB; 3EML) (RMSD of Cα; 0.85 Å), there is a major difference around the intracellular portions of helices V and VI, which are connected by ICL3, where T4L is inserted in A_2AAR-T4L (Supplementary Fig. 5). In our structure, ICL3 forms two regular helices, effectively continuations of helices V and VI respectively, connected by a short turn (Supplementary Fig. 6a).

The A_2AAR–Fab2838 structure has a modified ‘ionic lock’ where E228^6.30 (helix VI) and R102^3.50 of the D(E)RY motif (helix III) interact via a water molecule (W1; Fig. 2c,d). In the inactive bovine rhodopsin structure, the equivalent residues form a direct salt-bridge³ (Supplementary Fig. 7). R102^3.50 of A_2AAR-Fab2838 forms salt-bridges/hydrogen-bonds with D101^3.49, Y112 in ICL2 and T41^2.39 as observed in the A_2AAR-T4L structure (Supplementary Fig. 5b). Because of the insertion of the water molecule, E228^6.30 shifts towards the cytoplasmic space, as compared to the equivalent residue in rhodopsin (E247^6.30), resulting in the formation of a salt bridge with R220 in the short helical turn of ICL3. This interaction may be important in the formation of the helical structure in ICL3. The ‘ionic lock’ has not been observed in the crystal structures of other inactive GPCRs^6-11, including A_2AAR-T4L, except for the D3 dopamine receptor¹². This may be because the ICL3s in the other structures were modified to stabilise the protein. While this manuscript was in review, the crystal structures of thermostabilised A_2AAR mutants with native ICL3 were published^17,18. The antagonist-bound inactive structures have the ‘ionic lock’¹⁸. Thus, the ‘ionic lock’ of A_2AAR seems to stabilise the inactive conformation of the protein, which is why the receptor has a low basal activity.

Fab2838 binds on the intracellular side of the receptor (Fig. 2a). CDR-H3 of Fab2838 is unusually long and penetrates into a pocket formed by helices II, III, VI and VII (Fig. 2b). CDR-H3 interacts with the surrounding helices by forming 6 hydrogen bonds and 8 van der Waals contacts (Fig. 2c,d). The most extensive interactions are with helix II (mainly through hydrogen bonds) and helix VI (mainly through van der Waals contacts). In addition, a hydrogen-bond network including 2 water molecules is observed between CDR-H3 and helices III and VI (Fig. 2c,d). This hydrogen-bond network together with the van der Waals interactions seem to stabilise the modified ‘ionic lock’ interaction between E228^6.30 (helix VI) and R102^3.50 (helix III) discussed above. Other CDRs further stabilise the A_2AAR–Fab2838 complex by forming 14 hydrogen bonds with helices VI and VIII and ICLs 1, 2, and 3 (Fig. 2b). The extensive interactions explain the high affinity of Fab2838 (K_D = 4.4 nM) (Supplementary Fig. 8).

The Fab2838 CDR-H3 binding site in A_2AAR is similar to those for Nb80 CDR-3 in β₂AR² and for GαCT in opsin¹. A critical difference is that Fab2838 stabilises an inactive conformation whereas the others recognise active conformations of the receptors. These structures are compared in Figure 3. In the opsin structure, GαCT, which forms a short α-helix, fits into a large pocket formed by helices II, III, V, VI, and VII interacting with the Arg residue of the D(E)RY motif in helix III (Fig. 3, left panels). CDR-3 of Nb80 in the β₂AR structure binds in a similar position to GαCT although CDR-3 forms a β-hairpin¹ (Fig. 3, middle panels). Interestingly, CDR-H3 of Fab2838 also forms a β-hairpin but induces a differently shaped binding-pocket (Fig. 3c). In the β₂AR structure, CDR-3 of Nb80 is positioned between helices III and VI, whereas in the A_2AAR structure CDR-H3 of Fab2838 is ~ 6 Å closer to helices II and VII (Fig. 3b and Supplementary Fig. 9). This allows the close association of helices III and VI and the formation of the modified ‘ionic lock’ between R102^3.50 in helix III and E228^6.30 in helix VI, consequently stabilising the inactive conformation. In the β₂AR-Gs protein complex structure, the C-terminal α-helix (α5) of Gαs also binds in a similar position to CDR-H3¹³ (supplementary Fig. 10). The conformational changes of α5 together with the Gαs N-terminal region induced by the activated receptor was proposed to lead a nucleotide exchange from GDP to GTP in Gαs and to subsequent dissociation of the subunit from the receptor¹⁹. Thus, the binding pocket formed by helices II, III, VI, and VII seems to be the key site for the signal transfer between GPCR and G-protein.

A possible inactivation mechanism of A_2AAR by Fab2838 is summarised as follows. Agonist binding induces large displacements of the intracellular ends of helices III, VI, and VII^16,17, which are essential to form the G-protein binding-pocket^13,19 (Supplementary Fig. 1). This indicates that the signal from the ligand-binding pocket is transferred through these helices and the conformations of the two pockets are strongly coupled. Our agonist and antagonist binding experiments indicate that this coupling also allows signal transfer in a reverse direction, from the G-protein binding to the ligand-binding pockets (Fig. 1). CDR-H3 of Fab2838 locks the positions of helices III, VI, and VII from the cytoplasmic side, leading to an inactive conformation of the extracellular ligand-binding pocket to which agonists cannot bind probably because of the rearrangement of the side chains at the bottom of the ligand-binding pocket including W246^6.48, the toggle switch for activation (for details, see Supplementary Fig. 1 and 11). A similar conceptual model on the β₂AR activation was reported by Kobilka, Sunahara, and colleagues²⁰. In the case of β adrenergic receptors, the conformational coupling of the ligand and G-protein binding pockets seems less strict as demonstrated in the structures of β₁AR-agonist complexes²¹ and β₂AR-irreversible agonist complex²². This may be because the A_2AAR and β₁/β₂AR agonists interact with different helices in the binding pockets (for details, see Supplementary Discussion).

Antibody fragments (and nanobodies) such as Nb80 and Fab2838 that recognise conformational epitopes of GPCRs have great potential for GPCR studies in vitro and in vivo. Although antibodies recognising the intracellular surface are not suitable for direct therapeutic use, the CDR structures should provide useful information to design peptides or small-molecule compounds against their clearly defined pockets to control the activation states of GPCRs. The antibody-fragments will be also useful tools to study ligand-binding kinetics of GPCRs because they can separate ligand-binding from equilibrium-shifts between different activation states of the receptors. Our approach based on the conventional mouse-hybridoma system allows us to raise antibodies against various receptors in 3-4 months using standard laboratory equipment.