Cryo-EM structure of the human adenosine A_2B receptor–G_s signaling complex

Abstract

The human adenosine A_2B receptor (A_2BR) is a class A G protein–coupled receptor that is involved in several major physiological and pathological processes throughout the body. A_2BR recognizes its ligands adenosine and NECA with relatively low affinity, but the detailed mechanism for its ligand recognition and signaling is still elusive. Here, we present two structures determined by cryo–electron microscopy of A_2BR bound to its agonists NECA and BAY60-6583, each coupled to an engineered G_s protein. The structures reveal conserved orthosteric binding pockets with subtle differences, whereas the selectivity or specificity can mainly be attributed to regions extended from the orthosteric pocket. We also found that BAY60-6583 occupies a secondary pocket, where residues V250^6.51 and N273^7.36 were two key determinants for its selectivity against A_2BR. This study offers a better understanding of ligand selectivity for the adenosine receptor family and provides a structural template for further development of A_2BR ligands for related diseases.

The Cryo-EM structures of A_2BR–G_s complexes provide structural basis for recognition of NECA and selectivity of BAY60-6583.

INTRODUCTION

G protein–coupled receptors (GPCRs) are the largest and most diverse family of membrane proteins in eukaryotes, with the common seven-transmembrane helix bundle (7TM) architecture (1). GPCRs transmit a variety of signals through cell membranes and regulate a wide range of physiological and pathological processes. Adenosine is an endogenous nucleoside ubiquitously distributed in mammalian organisms, which modulates different important physiological functions upon triggering of the downstream signaling of its receptors (2). Of the four adenosine receptors, adenosine A_2A receptor (A_2AR) and A_2BR were originally found to interact with G_s/olf proteins, while A₁R and A₃R function in an inhibitory manner by interacting with the G_i/o subtypes (2). Recently, A_2BR was also found to couple to the G_i subtype; thus, it may trigger downstream events in various cell types through different G proteins (3). The broad distribution and expression of adenosine receptors in various body tissues, as well as their role in controlling essential functions in the body, make them potential drug targets for treating various pathological diseases, such as heart disease, cancer, Parkinson’s disease, inflammation, and glaucoma (4–7). A_2AR received extensive attention over the past few decades, and its crystal structure was determined in 2008 (8). From then, A_2AR was considered a prototypical receptor within the GPCR superfamily, and structures with different type of ligands were determined by crystallography or cryo–electron microscopy (cryo-EM), revealing the inactive, intermediate, and fully active conformations (9–13). In contrast, neither inactive nor active state structures have yet been determined for A_2BR.

Compared to the other three adenosine receptors, A_2BR has a weaker affinity to the adenosine receptors’ common ligands, e.g., adenosine and its derivative, 5′-N-ethylcarboxamidoadenosine (NECA) (14). However, previous research has shown that there is an apparent up-regulation of A_2BR expression along with a marked increase of the extracellular adenosine concentration under pathological conditions (5, 14, 15). Hence, A_2BR was reported to be involved in several critical physiological and pathological processes, including glucose metabolism, angiogenesis induction, tumor growth, intestinal inflammation, myocardial ischemia, and acute lung and kidney injuries. Furthermore, accumulating evidence has demonstrated that A_2BR can activate different intracellular signaling and exert physiological and pathological functions that are distinct from those of A_2AR (3, 16). While several selective agonists or antagonists have been reported for A_2BR (17–19), the leading agonist, BAY60-6583 [2-({6-amino-3,5-dicyano-4-[4-(cyclopropylmethoxy)phenyl]pyridin-2-yl}sulfanyl)acetamide] (20, 21), has been investigated for the treatment of atherosclerosis and cardiac diseases (22). In another example, the potent A_2BR antagonist, PBF-1129, is now being evaluated in clinical trials for tumor immunotherapy. To understand the mechanism for low affinity of NECA and selectivity of BAY60-6583 against A_2BR, we determined two cryo-EM structures of A_2BR bound with NECA and BAY60-6583 with the presence of engineered trimeric G_s proteins. These structures, combined with cell-based assays, provide molecular details of the recognition specificity of ligands for the adenosine receptor family.

RESULTS

Structure determination

In preparing the homogeneous human A_2BR-G_s complexes, we used the NanoBiT tethering strategy to facilitate the formation of stable complexes (fig. S1) (23). To facilitate the binding of a single-chain variable fragment (scFv16) (see Materials and Methods) (24), the Gα_s was modified on the basis of mini-Gα_s (11) with its αN replaced by the equivalent region of Gα_i. Unless otherwise stated, the engineered G protein is henceforth referred to as G_s. NECA or BAY60-6583 was added during large-scale purification, and protein samples from size exclusion chromatography (SEC) were collected for the cryo-EM analysis. Despite the addition of scFv16 at the membrane extraction stage, it appeared to be dissociated partially during sample preparation, and the classification suggested the presence of two populations of particles for both complexes. In each case, the one without scFv16 generated better density overall in the receptor region, although in the NECA-bound complex, the scFv16 apparently stabilized the G proteins by binding to the Gβ–Gγ interface (fig. S2). Thus, we removed the scFv16 in the final A_2BR–NECA–G_s and A_2BR–BAY60-6583–G_s models, and their density maps were reconstructed to resolutions of 3.26 and 2.99 Å, respectively (figs. S2 and S3 and Table 1).

Table 1. Cryo-EM data collection and refinement statistics.

RMSD, root mean square deviation; PDB, Protein Data Bank.

	A2BR–NECA–Gs	A2BR–BAY60-6583–Gs
Data collection and processing
Magnification	130,000	130,000
Voltage (kV)	300	300
Electron exposure (e−/Å2)	60	60
Defocus range (μm)	−1.2 ~ −1.8	−1.2 ~ −1.8
Pixel size (Å)	0.96	0.96
Symmetry imposed	C1	C1
Initial particle images (no.)	1,546,966	1,063,086
Final particle images (no.)	165,307	190,323
Map resolution (Å)	3.26	2.99
FSC threshold	0.143	0.143
Refinement
Initial model used (PDB ID)	6GDG	6GDG
Map sharpening B factor (Å2)	−123.2	−114.8
Model composition
Non-hydrogen atoms	7048	7045
Protein residues	895	895
Ligands	1	1
B factors (Å2)
Protein	81.59	73.82
Ligand	56.51	65.88
RMSD
Bond lengths (Å)	0.003	0.003
Bond angles (°)	0.539	0.538
Validation
MolProbity score	1.31	1.37
Clash score	5.62	5.13
Poor rotamers (%)	0.26	0.27
Ramachandran plot
Favored (%)	98.07	97.5
Allowed (%)	1.93	2.5
Disallowed (%)	0	0
PDB ID	7XY7	7XY6

Overall structure

The A_2BR–NECA–G_s and A_2BR–BAY60-6583–G_s complexes are very similar to each other [Cα root mean square deviation (RMSD), 0.4 Å], and they share an essentially identical GPCR–G_s interface (Fig. 1). The human A_2BR has a 55% sequence identity with A_2AR and 42% with A₁R; hence, it is expected that the active structures of human A_2BR are similar to those of the other two members, with a Cα RMSD of 0.9 and 1.1 Å to A₁R [Protein Data Bank (PDB) ID: 7LD3) (25) and A_2AR (PDB ID: 6GDG) (26), respectively (fig. S4A). Although an inactive structure of A_2BR is not available, the sequence and active structure comparisons of A_2AR and A₁R indicate that the key motifs and common features of class A receptor activation are also preserved in A_2BR (fig. S4B) (27).

Fig. 1. Overall structure of A_2BR–NECA–G_s and A_2BR–BAY60-6583–G_s.

(A and B) Cryo-EM maps showing the disk-shaped micelle (left) and cartoon representation (right) of the A_2BR–NECA–G_s (A) and A_2BR–BAY60-6583–G_s (B) complexes. Densities and structures of NECA and BAY60-6583 are shown in the top-right corner of each complex structure.

The most pronounced difference between A_2BR, A_2AR, and A₁R is in the extracellular loops (ECLs) (Fig. 2 and fig. S5). The ECL1 of A_2BR is similar in both sequence and structure with the other two receptors. The ECL3 of A_2BR is unique as it forms a short helix and misses an intraloop disulfide that is present in A_2AR and A₁R (Fig. 2, A and D). ECL2 is the most diversified region within the adenosine receptor family and that of A_2BR is the longest. Here, at the N-terminal region, K147^ECL2 forms a π-cation interaction with the F75^3.22 [Ballesteros-Weinstein numbering in superscript (28)] of ECL1. This is similar to A₁R, where a hydrophobic interaction is preserved between L149^ECL2 and F77^3.22, but differs from A_2AR, which contains another interloop disulfide bond (C74^3.22–C146^ECL2) (Fig. 2, B to D). The ECL2 of A_2BR is only partially solved in both A_2BR complexes, and a fragment of 153 to 167 is not solved in the A_2BR models. The unmodeled fragment includes three cysteines, and we predict that one of these three cysteines should form a disulfide bond with C72^ECL1 as there is clear density near C72^ECL1 (fig. S6A). A similar disulfide bridge is also present in A_2AR (C71^ECL1–C159^ECL2) to link ECL1 and the C-terminal region of ECL2 (Fig. 2B). Analyzing the sequence of A_2BR species reveals that C72^ECL1 and C166^ECL2 are only partially conserved and present pairwise in six species, including human, while the C154^ECL2 and C167^ECL2 are invariantly conserved in all A_2BR orthologs (fig. S6B). Thus, we speculated that the former and the latter pair of cysteines may be joined via disulfide bonding. Unlike the conserved C78^3.30–C171^ECL2 disulfide bond that is important for GPCR’s function, these two A_2BR-specific disulfide bonds may not play an essential role in its structural folding or ligand binding because individual mutations of these four cysteines did not significantly deteriorate the potency or affinity of its ligands (29) and, at times, even increased the agonist potency (30). Although the literature could not provide direct evidence of specific disulfide linkages, previous functional studies, together with our structures, indicate that these two potential disulfide bonds are partially flexible and may alter the receptor function indirectly by perturbing ligand binding or receptor equilibrium between different states.

Fig. 2. The extracellular loops of A_2BR.

(A) Superposition of A_2BR structures with A_2AR (PDB ID: 6GDG) and A₁R (PDB ID: 7LD3) from the extracellular view. The A_2BR–NECA–G_s, A_2BR–BAY60-6583–G_s, A_2AR, and A₁R structures are shown in cartoon and colored green, dark green, light blue, and brown, respectively. (B and C) Zoomed-in views of the extracellular loop 1 (ECL1) and ECL2 regions reveal differences between the three receptors. Disulfide bonds and key residues are labeled. (D) Sequence alignment of ECLs for adenosine receptors. This alignment is modified from a whole sequence alignment of these receptors (fig. S3). Key cysteines are labeled, and residues that form the helix are colored pink. The disordered ECL2 residues in A_2BR and A_2AR (6GDG) structures are shown with gray background.

The orthosteric pocket

The NECA binding mode in the A_2BR–NECA–G_s structure is similar overall to previous A_2AR-NECA and A₁R-adenosine binding poses (25, 26). The conserved interactions between orthosteric pocket residues and NECA in the A_2BR complex include: π-stacking with F173^ECL2; hydrogen bonding with T89^3.36, N254^6.55, S279^7.42, and H280^7.43; and hydrophobic interactions with V85^3.32, L86^3.33, M182^5.38, V191^5.47, W247^6.48, V250^6.51, and I276^7.39 (Fig. 3A). All these residues are highly conserved within the adenosine receptor family with small variations in A₃R (Fig. 3, B to C, and fig. S5). One notable position in A_2BR is V250^6.51 as its corresponding residue in the other three receptors is leucine. Because of this less bulky side chain, we found that, in the A_2BR complex structure, the NECA rotates slightly, and the ribose moiety moves toward V250^6.51 by ~1 Å to form hydrophobic contacts (Fig. 3B). Furthermore, the V250^6.51 also appears to affect nearby residues H251^6.52 and N254^6.55 as their rotamer conformations are different from their counterparts in the A_2AR and A₁R structures (Fig. 3B). These subtle changes result in preserved hydrogen bond interactions between N254^6.55 and the adenine moiety of NECA, whereas the previously identified NECA-H^6.52 interaction in A_2AR (PDB ID: 6GDG) is slightly compromised in A_2BR. The importance of N254^6.55 for NECA-mediated A_2BR activation is confirmed by the cyclic AMP (cAMP) accumulation assay, showing that the N254^6.55A mutation reduced the potency of NECA >20-fold (Fig. 3D). Meanwhile, the hydrophobic interaction contributed by V250^6.51 was confirmed when the V250^6.51A mutation significantly reduced the potency of NECA. We found that the potency of NECA is only slightly increased by the V250^6.51L mutation, suggesting that the V/L^6.51 is not the key determination for NECA’s selectivity against adenosine receptors (Fig. 3D).

Fig. 3. NECA binding pose.

(A) Detailed binding pose of NECA. NECA and key pocket residues are shown as sticks. Hydrogen bonds are shown as red dashed lines. (B) Comparison of A_2BR–NECA–G_s structure (current study) with A_2AR (PDB ID: 6GDG) and A₁R (PDB ID: 7LD3) in the orthosteric pocket. Key movements of NECA and pocket residues are shown with black arrows. ADN, adenosine. (C) Sequence alignment of the orthosteric pocket residues. (D) Mutagenesis analysis of A_2BR key residues on the potency of NECA by the cAMP assay. The maximum and minimum activation levels of wild-type (WT) A_2BR were set to 100 and 0%, respectively. The X-fold median effective concentration (EC₅₀) was calculated by dividing the EC₅₀ of the mutant by the EC₅₀ of WT. Data are shown as means ± SEM from at least three independent experiments. RLU, relative luminescence unit.

Another residue that gained our attention is E174^ECL2. In several previous A_2AR structures (8, 9, 12), the corresponding residue E169^ECL2 forms a salt bridge with ECL3 residue, H264^ECL3. It was suggested that this linkage forms a portion of the orthosteric binding pocket, specifically, to hold or contact with the moieties that extend from the adenine group, e.g., the phenylethylamino moiety of ZM241385 (8, 9) or the (2-carboxyethyl)phenylethylamino moiety of CGS21680 (12). These structures, together with mutagenesis and molecular dynamic (MD) simulation studies on A_2AR, confirmed that the salt bridge plays a role in the dissociation of its orthosteric ligands (31, 32). In A_2BR, a similar linkage was not observed as the ECL3 holds a different conformation, and a histidine is absent in the corresponding position. Furthermore, the side chain of E174^ECL2 was not modeled in the A_2BR–NECA structure, indicating that, although adjacent to the N⁶ of the adenosine group, the polar interaction of E174^ECL2 may be negligible. In alignment with these structural findings, E174^ECL2A displays comparable potency to wild-type (WT) A_2BR in our cAMP assay; on the contrary, the corresponding E169^ECL2 mutation in A_2AR markedly reduced the potency of its ligands (33). Therefore, the E174^ECL2 along with its counterparts in ECL3 significantly contribute to the pharmacological difference of NECA on A_2BR.

The secondary pocket

In the A_2BR–BAY60-6583–G_s structure, the ligand occupies an overlapped position with NECA but with some unique features (Fig. 4). Similar to NECA, BAY60-6583 also forms hydrogen bonds with T89^3.36 and N254^6.55, as well as van der Waals interactions with previously mentioned residues (Fig. 4A). The cyclopropyl group of BAY60-6583 points deep into the orthosteric pocket and makes close contacts with L86^3.33, V191^5.47, W247^6.48, and H251^6.52. In the middle region, the phenyl and pyridine moieties of BAY60-6583 partially overlapped with the ribose and adenine groups of NECA, respectively. Unlike the ribose group of NECA, BAY60-6583 does not form hydrogen bond with S279^7.42 and H280^7.43, and the hydrophobic phenyl moiety of BAY60-6583 is located 2 to 3 Å further away from these TM7 residues. These structural differences are aligned with the functional data showing that the S^7.42 and H^7.43 mutations greatly decreased the potency and affinity of the ribose-containing compounds on all adenosine family receptors (34–37), in contrast to a slight increase on the potency of BAY60-6583 on A_2BR by the S279^7.42A mutation (35).

Fig. 4. BAY60-6583 binding pose.

(A and B) The detailed binding pose of BAY60-6583 (A) and comparison with NECA (B). The ligands and key pocket residues are shown as sticks. Hydrogen bonds are shown as red dashed lines. Key movements of the secondary pocket residues are shown with black arrows. In (A), the NECA structure is superimposed for comparison (50% transparency). (C) Comparison of A_2BR structure with A_2AR (PDB ID: 6GDG) and A₁R (PDB ID: 7LD3) in the orthosteric pocket and secondary pocket. ADN, adenosine. (D) Mutagenesis analysis of A_2BR orthosteric and secondary pocket residues on the potency of BAY60-6583 by the cAMP assay. (E) The potency of BAY60-6583 on the pocket mutations of A_2AR. In (C) and (D), the maximum and minimum activation levels of WT receptor were set to 100 and 0%, respectively. The X-fold EC₅₀ was calculated as in Fig. 3. Data are shown as means ± SEM from at least three independent experiments.

There are two cyano modifications on the pyridine moiety of BAY60-6583: one cyano group hydrogen bonds to the N254^6.55 of A_2BR, and the other points to an empty void embraced by TM2, TM3, and TM6. Another amino modification on the pyridine moiety points upward and forms a week hydrophilic interaction with E174^ECL2. The cAMP accumulation assay confirmed that the potency of BAY60-6583 is ~2-fold reduced by the E174^ECL2A mutant (Fig. 4D), in contrast to a negligible effect of E174^ECL2A on NECA. These assay results also aligned with the successful modeling of E174^ECL2 in the A_2BR–BAY60-6583–G_s complex and may be a consequence of a stronger polar interaction between E174^ECL2 and BAY60-6583. In contrast to the assay results of NECA, the potency of BAY60-6583 is markedly reduced by the V250^6.51L mutation but only moderately affected by the V250^6.51A mutation. These distinct mutation results are compatible with the precise differences in the binding pose of these two molecules. We further conducted the reverse mutation on A_2AR and found that the L249^6.51V mutant can increase the potency of BAY60-6583 on A_2AR by >714-fold (Fig. 4E). These results thus identify V250^6.51 as a key determinant for the high selectivity of BAY60-6583 toward A_2BR.

BAY60-6583 contains an acetamide group on the C₁ position of the pyridine moiety that points toward a secondary pocket composed of the extracellular tips of TM1, TM2, and TM7 (Fig. 4, B and C). Density of the acetamide group is relatively weak but allows us to model the polar tail toward Y10^1.35 and N273^7.36 of the receptor. Three nearby receptor residues, Y10^1.35, S68^2.65, and N273^7.36, are well modeled. Among them, S68^2.65 points in the opposite direction and did not form a hydrogen bond with BAY60-6583, while Y10^1.35 and N273^7.36 face the acetamide group and within hydrogen bond distance of the polar tail on BAY60-6583. It is also worth mentioning that comparison of the A_2BR–BAY60-6583–G_s and A_2BR–NECA–G_s complex structures reveals interesting subtle main-chain or side-chain movement of these residues in both the orthosteric and secondary pockets (Fig. 4B). Pronounced differences include the three residues in the secondary pocket that move away from the pocket to accommodate the acetamide group of BAY60-6583 (Fig. 4B). Aligned with the structural findings, the cAMP assay revealed that the S68^2.65A mutant does not affect the potency, whereas the Y10^1.35F and N273^7.36A caused an ~8- and ~6-fold decrease in potency, respectively, suggesting that these two residues contribute to the recognition of BAY60-6583 (Fig. 4D). As a control, the mutant Y10^1.35F of the secondary pocket does not affect the activity of NECA (Fig. 3D). Y10^1.35 is absolutely conserved, but N273^7.36 is only present in A_2BR. In the other three adenosine receptors, the corresponding position of 7.36 is a tyrosine. Thus, we mutated the N273^7.36 to tyrosine on A_2BR, and the compromised activity indicated that the bulky side chain of Y^7.36 may interfere with the acetamide group of BAY60-6583 (Fig. 4, C and D). Aligned with these results, the reverse mutant (Y271^7.36N) on A_2AR can significantly increase the potency of BAY60-6583 (Fig. 4E). These results identify N273^7.36 as another determinant for BAY60-6583’s selectivity against A_2BR.

The A_2BR–G interface

Like other GPCRs that featured a larger outward shift of TM6 when coupled to G_s protein, TM6 of A_2BR occupies an orientation similar to the TM6 of A_2AR but is distinct from the counterpart of A₁R, which coupled to G_i (Fig. 5A). Consistently, the A_2BR–G_s interface buries 2610 Ų of the total surface area, which is very close to the A_2AR–G_s interface (2572 Ų), and both are much larger than the A₁R–G_i interface (2086 Ų). Despite its similarity with the A_2AR–G_s interface, there are several interesting features of the A_2BR–G_s interface. The A_2BR–G_s interface is distinctive in that helix8 tilts toward cytosolic and makes extra contacts with the β subunit of G proteins, whereas, in A_2AR and A₁R, it points toward the cell membrane and shows a lack of contact with the G proteins (Fig. 5A). A close-up view of the structures shows that the Y292^7.55 of A_2BR changes its rotamer and points toward TM1, and hydrogen bonds to the backbone of G24^1.49 (Fig. 5B). Consequently, the side chain of Y292^7.55 partially occupies the position of F301^8.54, which likely pushes the downward shift of helix8.

Fig. 5. A_2BR-G_s interface.

(A) Superposition of A_2BR–NECA–G_s (current study) with A_2AR (PDB ID: 6GDG) and A₁R (PDB ID: 7LD3) structures on receptor. (B) Comparison of the TM7-helix8 regions of A_2BR, A_2AR, and A₁R showing the key differences on Y292^7.55 and nearby residues. (C) The A_2BR helix8–Gβ interface. (D) Mutagenesis analysis of A_2BR helix8 residues on the potency of NECA by the cAMP assay. The maximum and minimum activation levels of WT A_2BR were set to 100 and 0%, respectively. The X-fold EC₅₀ was calculated by dividing the EC₅₀ of the mutant by the EC₅₀ of WT. Data are shown as means ± SEM from at least three independent experiments. (E) The receptor–α5 interface in A_2BR. Interacting residues are shown as sticks and are labeled. Color codes are indicated at the top-left corner.

Despite of weak densities, the Y299^8.52, K303^8.56, and R307^8.60 of helix8 point toward the Gβ residues D312, H311, and A309, respectively, in the A_2BR–Gβ interface (Fig. 5C). We thus introduced mutations on helix8 of A_2BR to validate the potential function of the helix8–Gβ interface. The cAMP assay results showed that the mutant R307^8.60A decreased the potency of NECA by 4-fold, while other mutants did not significantly deteriorate the potency (Fig. 5D). These results suggested that the A_2BR–Gβ interface is determined by the very C-terminal region of helix8. Furthermore, the insusceptibility of the potency by the F301^8.54A mutation suggested that the tilted conformation of helix8 is an intrinsic feature and may not relate to the contacts between Y292^7.55 and F301^8.54. Such a tilted helix8 is frequently seen in class B GPCRs (38, 39).

The tilt of helix8 in A_2BR also induces 2 to 4 Å downward movement of G protein subunits. Nevertheless, the receptor–Gα_s interface of A_2BR is quite similar to A_2AR, and most of the key interactions are preserved, e.g., the π-cation interaction of R103^3.50 and Y374^H5.23 [uppercase refers to common G protein numbering (40)], and the hydrophobic network by L376^H5.25 and TM3 and TM5 residues (I107^3.54, I205^5.61, V208^5.64, and A209^5.65) (Fig. 5E). There are also extensive hydrophilic interactions between α5 residues (Q367^H5.16, R368^H5.17, R372^H5.21, and L377^H5.26) and the A_2BR residues from ICL2, ICL3, and helix8 (Fig. 5E). Therefore, these results are consistent with the fact that the receptor–Gα_s interactions are largely conserved between A_2BR and A_2AR despite being disturbed by the tilted helix8, and the tiny vibration of the interface is also frequently seen in other subfamilies of G_s-coupled receptors.

DISCUSSION

Our complex structures provided a precise structural template for previous site-directed mutagenesis and structure-activity relationship studies (14, 35, 41, 42). The A_2BR structures reveal a highly conserved orthosteric binding pocket with diversity mainly coming from the extracellular regions, including the ECL2 and ECL3 residues. We found that the ECL2 and ECL3 of A_2BR may not form a similar salt bridge as in A_2AR (E169^ECL2–H264^ECL3) because of sequence and structural differences. Importance of this salt bridge has previously been demonstrated on A_2AR structurally: The ligand ZM241385 occupied the orthosteric pocket and its phenylethylamino group formed contacts with the salt bridge when crystallized under acidic condition (8), whereas under basic condition (pH > 8), the salt bridge was broken, and the phenylethylamino group of ZM241385 pointed to a reverse direction (43). In addition, it was shown that agonist binding to A_2AR is pH sensitive with higher affinities observed at a lower pH (44, 45). These structural and functional findings suggested that, for A_2AR, the key determinant for the sensitivity is probably the H264^ECL3 considering its pK_a (where K_a is the acid dissociation constant) value of ~7 because, at lower pH, the formed E169^ECL2–H264^ECL3 linkage may function like a lid over the orthosteric pocket and thus enhanced the residence time of its ligand and increased the binding affinity. Therefore, it is our best speculation for A_2BR that, without such kind of linkage, the ligand NECA thus binds with fast-on/fast-off kinetics and weak affinity (46).

Besides the ECLs, the subtle differences of the residues within the orthosteric pocket, including but not limited to the V/L^6.51 and its nearby residues H^6.52 and N^6.55, may further contribute to the selectivity. We found that the potency of NECA is slightly increased by the A_2AR/A₁R/A₃R-mimicing mutation V250^6.51L, reminiscent of a previous study that suggested L^6.51 as a hotspot for A_2AR’s selective antagonist ZM241385 (47). The phenomenon that a ligand adopts slightly different conformations in homologous receptors is not unusual and has been illustrated, for example, in the d-lysergic acid diethylamide binding poses in serotonin receptors HTR2A and HTR2B (48, 49). In this study, we also identified V250^6.51 and N273^7.36 as two key determinants for the selectivity of BAY60-6583 against A_2BR. The distinct effects of V250L^6.51 and V250A^6.51 on BAY60-6583 and NECA aligned well with their binding details (Figs. 3D and 4D). N273^7.36 is uniquely located in the secondary binding pocket of A_2BR. The secondary pocket occupied by BAY60-6583 is similar to previously determined binding modes of DU172-bound A₁R and XAC-bound A_2AR (fig. S8A), whereas most other adenosine receptor ligands occupy only the orthosteric pocket (fig. S8B). Compared to the position of BAY60-6583 in the A_2BR pockets, the two A_2AR selective agonists, UK-432097 and CGS21680, extend upward from the orthosteric pocket and interact with the ECL2–ECL3 salt bridge that is uniquely present in A_2AR. Without such a linkage in A_2BR, the A_2BR agonist BAY60-6583 likely extends horizontally toward TM1/2/7 and occupies the secondary pocket, which contains an aliphatic residue N273^7.36.

Although having a similar level of potency, BAY60-6583 was previously reported as a partial agonist because its efficacy is much lower than that of NECA or adenosine (20). Our cAMP assay confirmed that BAY60-6583 shares quantitatively equipotent potency with NECA (3.67 versus 8.85 nM) but a much lower efficacy (~5%) compared to NECA (fig. S7). Although the NECA- and BAY60-6583–bound structures revealed a nearly identical active conformation of A_2BR, the binding details may shed light on the structural basis of their distinct efficacies. The GPCR activation pathway is featured by shrinking the middle region and expanding the intercellular region of the 7TM upon ligand binding, a process that is mainly facilitated by outward movement of TM6 and inward movement of TM7 (27). Using A_2AR as an example, compared to the antagonist-bound inactive state, the TM7 moves toward the agonist, and the S^7.42 and H^7.43 hydrogen bond to the ligand ribose group in the agonist-bound conformation, thus stabilizing the active state of TM7 to accommodate downstream G proteins (8, 10). For A_2BR, the consensus ligand NECA makes conserved contacts with the corresponding TM7 residues, whereas the selective ligand BAY60-6583 misses these polar contacts. Hence, our structures, together with previous functional results (34–37), suggest that, on A_2BR, the NECA (or adenosine) can efficiently trigger the conformational change toward the active state through these hydrogen bond interactions, whereas, for BAY60-6583, the transition efficacy is relatively low as a result of the absent polar contacts. In contrast, for those tested residues in the secondary pocket, their decreased efficacies are likely because the mutations indirectly affected the conformational transition that involves all helices from inactive to active conformation.

We further hypothesize that it is feasible to enhance the efficacy of BAY60-6583 by modifying the cyano group on C₂ of the pyridine moiety that faces the lower edge void of the secondary pocket (fig. S8, C and D). This proton-accepting cyano group is relatively incompatible with nearby residues A60^2.57, A64^2.61, V85^3.32, and H280^7.43 (distances, 3.5 to 6.8 Å). Thus, it is possible to increase the efficacy by replacing the cyano group with a proton-accepting group to hydrogen bond to the previously mentioned H280^7.43, a residue that is critical for the conformation transition of the adenosine receptors toward active state. Relatedly, an analog of BAY60-6583 that cyclized the C₁ and C₂ derivatives (compound 52) is more rigid, and its activity is almost abolished (50). From our docking model, the phenyl group of compound 52 rotates and becomes incompatible with the nearby nonpolar residues, and its polar acetamide tail interferes with A64^2.57 of TM2 (fig. S8E).

With the A_2BR active structure in hand, we are now able to map the structure of the cancer-related mutations that have been pharmacologically characterized (51). All these mutations are located far from the ligand-binding pocket (fig. S9), so they may function by affecting receptor folding and stability, the equilibrium between inactive and active states, or the coupling efficacy with the downstream adaptors. Of note, our last modeled residue in the receptor structure is Y308. Thus, the dysfunction caused by L310 mutations is possibly through vibrating the preceding helix8 that was involved in Gβ interactions, or the following C311, which is potentially a palmitoylation site.

Together, our structures reveal interesting similarities and diversities among adenosine receptors, thus representing an important complement to the previously determined A_2AR and A₁R structures. The current research suggests that residues of the orthosteric pocket and the ECLs together contribute to the potency and affinity differences between A_2BR and the other adenosine receptors. This study also identified key determinants for the specificity of A_2BR agonist BAY60-6583, illuminating the path forward for future discovery of even more potent A_2BR ligands for related diseases.

MATERIALS AND METHODS

Construct cloning, expression, and purification of A_2BR-G_s complexes

For the receptor, human A_2BR (residues 2 to 332), containing an N-terminal thermostabilized apocytochrome b₅₆₂RIL fusion protein and a C-terminal LgBiT protein, was cloned into pFastBac1 vector, with hemagglutinin signal peptide followed by FLAG and 10× His tags at the very N terminus. For the G protein components, mini-Gα_sIN, equal to mini-G_s but with its αN replaced by the counterpart of G_i (to facilitate binding of scFv16, which binds to the interface between Gβγ subunits and Gα_i subunit) (52), was fused to bovine Gγ2 at its N terminus (Gγ–mini-G_sIN). Human Gβ1, followed by SmBiT at its C terminus, was cloned into pFastBac-Dual together with Gγ–mini-Gα_sIN (fig. S1). Sf9 insect cells were infected with viruses of the receptor (A_2BR-LgBiT) and Gγ–mini-Gα_sIN–Gβ–SmBiT in the ratio of 1:2 for 72 hours at 27°C. The cell pellets were lysed by Dounce homogenization in 20 mM Hepes (pH 7.5), 100 mM NaCl, 5 mM MgCl₂, 5 mM CaCl₂, 10% glycerol, EDTA-free protease inhibitor cocktail (Bimake), apyrase (25 mU/ml; New England Biolabs), 200 μM Tris(2-carboxyethyl)phosphine (TCEP) (Thermo Fisher Scientific), 20 μM agonist NECA (Abcam) or BAY60-6583 (MedChemExpress), and scFv16 (15 μg/ml), incubated at 4°C for 3 hours, and then centrifuged at 30,000 rpm for 45 min to collect the membranes. The washed membranes were solubilized in 20 mM Hepes (pH 7.5), 100 mM NaCl, 5 mM MgCl₂, 5 mM CaCl₂, EDTA-free protease inhibitor cocktail, 1% (w/v) lauryl maltose neopentyl glycol (LMNG; Anatrace), 0.2% (w/v) cholesteryl hemisuccinate (CHS; Sigma-Aldrich), 20 μM agonist (NECA or BAY60-6583), apyrase (25 mU/ml), and 150 μM TCEP and incubated at 4°C for 3 hours. The supernatant was collected by centrifugation at 30,000 rpm for 45 min and then incubated with TALON immobilized metal affinity chromatography resin (Clontech) with an addition of 20 mM imidazole at 4°C overnight. The resin was collected by centrifugation at 1000g for 5 min, packed into a gravity flow column, and washed with 20 column volumes of buffer containing 20 mM Hepes (pH 7.5), 100 mM NaCl, 2 mM MgCl₂, 2 mM CaCl₂, 5 μM agonist (NECA or BAY60-6583), 0.01% (w/v) LMNG, 0.002% (w/v) CHS, and 20 mM imidazole and eluted using buffer containing 20 mM Hepes (pH 7.5), 100 mM NaCl, 2 mM MgCl₂, 2 mM CaCl₂, 100 μM agonist (NECA or BAY60-6583), 0.01% (w/v) LMNG, 0.002% (w/v) CHS, and 300 mM imidazole. The elution was collected and concentrated with an Amicon Ultra Centrifugal Filter [molecular weight cutoff (MWCO), 100 kDa] and subjected to SEC using a Superdex 6 10/300 GL column (GE Healthcare) with running buffer, containing 20 mM Hepes (pH 7.5), 100 mM NaCl, 2 mM MgCl₂, 2 mM CaCl₂, 100 μM TCEP, 5 μM agonist (NECA or BAY60-6583), 0.01% (w/v) LMNG, and 0.002% (w/v) CHS. The fractions of monomeric protein complex were pooled and concentrated with an Amicon Ultra Centrifugal Filter (MWCO, 100 kDa) to 3 to 5 mg/ml for cryo-EM studies.

Cryo-EM sample preparation and image acquisition

Samples (3 μl) of A_2BR–NECA–G_s and A_2BR–BAY60-6583–G_s were applied to Au grids (Quantifoil, 300 mesh AU R1.2/1.3) that were glow-discharged for 40 s using Gatan Solarus (950). Grids were subsequently plunge-frozen by Vitrobot Mark IV (Thermo Fisher Scientific) at 4°C and 100% humidity.

Cryo-EM datasets were collected on the Krios G4 Cryo–Transmission Electron Microscope equipped with the Falcon 4 Direct Electron Detector (Thermo Fisher Scientific) in superresolution mode operating at 300 kV accelerating voltage with 10-eV slit width. Movies were taken under Energy-filtered transmission electron microscopy (EFTEM) nanoprobe mode, with 50-μm C2 aperture and calibrated magnification of ×130,000, corresponding to a pixel size of 0.96 Å. Each movie was taken in electron-event representation (EER) format with a total dose of 60 electrons per Ų. EPU software (Thermo Fisher Scientific, v.1.11.0) was used to collect data with a defocus range of −1.2 to −1.8 μm.

Cryo-EM data processing and three-dimensional reconstruction

A total of 10,274 movies of the A_2BR–NECA–G_s sample and 18,412 movies of the A_2BR–BAY60-6583–G_s sample were imported into RELION 4.0 (53) to perform motion correction. The resulting .mrc files were imported into CryoSPARC 3.3.1 for further processing (54). Contrast transfer function (CTF) parameters were determined by patch CTF estimation (multi). After that, automated particle picking and rounds of two-dimensional classification were performed.

For the A_2BR–NECA–G_s sample, selected subsets with 1,546,966 particles were used to generate the initial model. After rounds of heterogeneous refinement, models of A_2BR–NECA–G_s with 165,307 particles and A_2BR–NECA–G_s-scFv16 with 188,487 particles were generated. Subsequently, homogeneous refinement, nonuniform refinement, and local refinement were performed on these two models. Last, A_2BR–NECA–G_s with a 3.26-Å map and A_2BR–NECA–G_s-scFv16 with a 3.19-Å map were determined by gold standard Fourier shell correlation (FSC) using the 0.143 criterion.

For the A_2BR–BAY60-6583–G_s sample, selected subsets with 1,063,086 particles were used to generate the initial model. After rounds of heterogeneous refinement, a model with 190,323 particles was generated to perform homogeneous refinement, nonuniform refinement, and local refinement. Because the scFv16 only generated better density in the G protein region, but not in the receptor region in the A_2BR–NECA–G_s-scFv16 complex, we thus removed the scFv16 from the data process and refinement for the A_2BR–BAY60-6583–G_s complex. The final 2.99-Å map was determined by gold standard FSC using the 0.143 criterion.

Model building and refinement for A_2BR structures

Structures of A_2BR predicted by AlphaFold (55, 56) and A_2AR-miniG_s (PDB ID: 6GDG) (26) were used as the starting models for model building and refinement against the electron density map of A_2BR–NECA–G_s and A_2BR–BAY60-6583–G_s. Models were docked into the EM density maps using Chimera (57). Subsequently, iterative manual adjustments and rebuilding were performed in Coot (58) and phenix.real_space_refine in Phenix (59). The final refinement statistics are provided in Table 1. The structure figures were prepared by Chimera and PyMOL (www.pymol.org).

cAMP assay

For the cAMP assay, the codon-optimized WT and mutant A_2BR and A_2AR genes were cloned into a customized pLEXm vector with N-terminal signal peptide, Flag tag, and C-terminal His₁₀ tags. Human embryonic kidney 293T cells were cultured and preseeded into 96-well plates in Dulbecco’s modified Eagle’s medium supplemented with 10% (v/v) fetal bovine serum, penicillin (50 IU/ml), and streptomycin (50 μg/ml). After 36 hours, the cells were transiently transfected with WT or mutant receptors in combination with cAMP response element reporter at the ratio of 1:1 (w/w) using polyethylenimine (ratio of 1:3). After 12 hours of culture at 37°C and 5% CO₂, the cells were stimulated with serial-diluted concentrations of agonists (NECA, 10.24 pM to 4 μM; BAY60-6583, 10.24 pM to 4 μM). Stimulated cAMP accumulation was measured by the Steady-Glo Luciferase Assay System (Promega). Briefly, after 4 hours of stimulation, the medium was discarded, and cells were lysed using a passive lysis buffer (Promega) and transferred into 96-well assay plates. Assay substrate was added to each well following the supplier’s instructions. Plates were incubated in the dark at room temperature for 5 min, and the glow intensity is measured by a LUMIstar Omega reader (BMG Labtech). Relative luminescence unit (RLU) was acquired by subtracting intensity of the well with apo receptor. Cell surface expression for WT and each mutant was monitored by flow cytometry. In brief, the expressed cells were treated by trypsin, resuspended in phosphate-buffered saline (PBS), and incubated with mouse anti-Flag (M2–fluorescein isothiocyanate) antibody (Sigma-Aldrich) for 15 min in the dark, and then, a ninefold excess of PBS was added to cells. Last, the surface expression data were measured by detecting the percentage of cells sorted by fluorescence of fluorescein isothiocyanate using BD FACSCanto II (BD Biosciences).

Supplementary Materials

This PDF file includes:

Figs. S1 to S9

View/request a protocol for this paper from Bio-protocol.