Molecular Determinants of CGS21680 Binding to the Human Adenosine A_2A Receptor

Receptor Constructs.

Pharmacology of wild-type A_2AR was determined using the full-length human A_2AR cDNA cloned into pcDNA3.1 (pcDNA3.1_WT-A_2AR; Bennett et al., 2013) provided by Kirstie A. Bennett from Heptares Therapeutics (Welwyn Garden City, England). A_2AR mutants S67A, E169A, H264A, and L267A were produced by introducing single-point mutations into pcDNA3.1_WT-A_2AR using the QuikChange II site-directed mutagenesis kit (Agilent Technologies, Stockport, England). Receptor expression at the cell surface was assessed using A_2AR with a HaloTag at the N-terminus of the receptor, HALO-A_2A, provided by Cisbio Bioassays. Construction of the thermostabilized agonist-bound conformation of the human adenosine receptor, A_2AR-GL31, was described previously (Lebon et al., 2011a,b).

Cell Culture, Transient Transfections, and cAMP Assays.

Chinese hamster ovary (CHO) cells were grown to 80% confluence in Glutamax Dulbecco’s modified Eagle’s medium Ham’s F12 medium (Invitrogen, Paisley, UK) complemented with 10% fetal bovine serum (Labtech International, Uckfield, UK). For transfection, cells were seeded at a density of 12,500, 25,000, 50,000, or 100,000 cells/well in a volume of 100 μl in a 96-well plate and transfected with 50 ng of either wild-type A_2AR or the mutants S67A, E169A, H264A, or L267A using Lipofectamine 2000 (Life Technologies, Paisley, UK). Dose-response curves were obtained by transfecting 25,000 cells/well with 50 ng of plasmid DNA. In brief, 50 ng of plasmid DNA was mixed with Lipofectamine 2000 in a total volume of 50 μl and incubated 30 minutes at room temperature, according to the manufacturer’s guidelines, and then added to the required number of cells. After 48 hours of receptor expression, cAMP production was measured using the Dynamic 2 cAMP HTRF kit (Cisbio Bioassays), according to the manufacturer’s protocol. In brief, the cells were washed in phosphate-buffered saline, then stimulated using different compound concentrations in a total volume of 50 μl, complemented with the phosphodiesterase inhibitor Ro20-1724 (50 μM) and adenosine deaminase (1 U/ml). Compounds were diluted in Dulbecco’s modified Eagle’s medium Ham’s F12 medium, without serum. After 30-minute incubation, 25 μl of cAMP-d2 conjugate and 25 μl of anti–cAMP-cryptate diluted in lysis buffer were added per well. The plates were incubated at room temperature for 1 hour, and then the fluorescence per well was recorded using the Pherastar 96-well time-resolved fluorescence plate reader (BMG Labtech, Champigny sur Marnee, France). Data were analyzed using GraphPad Prism version 5.0 (GraphPad Software, San Diego, CA). Functional concentration response data were fitted to a four-parameter logistic equation. The pEC₅₀ values are expressed as the mean ± S.E.M. Statistical analyses were performed using a one-way analysis of variance with Dunnett’s post test. Values were stated as significantly different at P < 0.05.

Expression, Purification, and Crystallization.

Cell-surface receptor expression levels in human embryonic kidney 293 cells were measured using the HALO-A_2AR construct, which were labeled according to the manufacturer’s protocol (Cisbio Bioassays). In brief, cells were washed with 150 µl of Tag-Lite buffer (Cisbio Bioassays), incubated for 1 hour (37°C) with 100 nM Chloroalkane-Lumi4Tb (Cisbio Bioassays) diluted in Tag-Lite buffer (50 µl/well), washed three times with 100 µl of Tag-Lite buffer, and the fluorescence measured using a Pherastar 96-well time-resolved fluorescence plate reader.

The A_2AR construct, A_2AR-GL31, is an engineered version of human A_2AR. A_2AR-GL31 contains four thermostabilizing mutations, L48A, A54L, T65A, and Q89A, that improve thermostability of the agonist-bound conformation of human A_2AR (Lebon et al., 2011a). In addition, to facilitate crystallization, the C-terminus was deleted after Ala316, and the point mutation N154A in EL2 was introduced to remove the N-linked glycosylation site. A proteolytically cleavable His₁₀ tag was introduced at the C-terminus to facilitate purification. Baculovirus expression and purification were performed exactly as previously described, except that the detergent DM was used throughout at a final concentration of 0.15%. Purified A_2AR-GL31 used for crystallization was in a buffer containing 25 mM Hepes (pH 7.4), 0.1 M NaCl, 100 μM CGS21680, and 0.15% DM. Purified receptor fractions were concentrated to 15–25 mg/ml in a final volume of 50–60 μl. Determination of the receptor concentration was performed using the amido black assay (Schaffner and Weissmann, 1973).

The A_2AR-GL31–CGS21680 complex was crystallized using the lipidic cubic phase (LCP) crystallization method (Cherezov et al., 2006; Caffrey and Cherezov, 2009). LCP crystallization setups contained a 2:3 (v/v) receptor-to-monoolein ratio, which was dispensed in 25-nl aliquots using a lipid-handling instrument, the mosquito-LCP from TTP Labtech (Melbourn, England). High-quality diffraction crystals were obtained in two different crystallization conditions, leading to two different crystal forms, monoclinic and orthorhombic. Monoclinic crystals were grown in 0.05 M acetate-HCl (pH 4.8), 6% sucrose, 21.9% pentaethylene glycol with the monoolein supplemented with 10% (w/w) cholesterol before mixing with the purified receptor. Orthorhombic crystals were grown in 0.1 M ADA (pH 7.0) and 21.6% PEG 600, with cholesterol hemisuccinate added to the purified receptor [10% (w/w) cholesteryl hemisuccinate: A_2AR] before mixing with the monoolein. All LCP crystallization trials contained 800 μl of precipitant per well (lipidic cubic phase kit; Molecular Dimensions Ltd., Newmarket, UK). Crystals were grown at 22°C, harvested with cryo-loops, and cryo-cooled in liquid nitrogen.

Data Collection, Structure Solution, and Refinement.

Diffraction data for the CGS21680 complex were collected at the European Synchrotron Radiation Facility (Grenoble, France) on the microfocus beamline ID23-2 (wavelength 0.8726 Å) using a 10-μm focused beam and a Mar 225CCD detector. The microfocus beam was essential for the location of the best diffracting parts of single crystals and the collection of several wedges of data from different positions on the crystal. Images were processed with MOSFLM (Leslie, 2006) and SCALA (Evans, 2006). The CGS21680 complex was solved by molecular replacement with PHASER (McCoy et al., 2007) using the adenosine-bound A_2AR-GL31 crystal structure (PDB ID 2YDO) as a search model after removal of all solvent molecules and the ligand. Two protein chains were found per asymmetric unit in the orthorhombic crystal form and one protein chain per asymmetric unit in the monoclinic crystal form. Refinement and rebuilding were carried out using REFMAC (Murshudov et al., 1997) and COOT (Emsley and Cowtan, 2004), respectively. Validation of the final refined models was performed using Molprobity. All figures in the manuscript were generated using either PyMOL or CCPmg (Potterton et al., 2004).

Crystallization of the Selective Agonist CGS21680 Bound to Thermostabilized A_2AR-GL31.

A_2AR-GL31 is an engineered version of the human A_2AR with improved thermostability in the agonist-bound conformation through the introduction of four point mutations, L48A, A54L, T65A, and Q89A (Lebon et al., 2011a). In addition, to facilitate crystallization, the C-terminus was deleted after Ala316, and the point mutation N154A in EL2 was introduced to remove the N-linked glycosylation site. A_2AR-GL31 has been successfully cocrystallized with the endogenous agonist adenosine and the synthetic agonist NECA (Lebon et al., 2011b) (Fig. 1). Pharmacological characterization was previously performed showing that the affinity of A_2AR-GL31 for agonist is identical to wild-type A_2AR, although the signaling properties of the receptor are impaired due to the orthosteric binding pocket being uncoupled from the G protein binding site (Lebon et al., 2011b).

CGS21680-bound A_2AR-GL31 was purified as described previously for NECA-bound A_2AR-GL31, and the purified receptor was of high quality with a monodisperse gel filtration profile. No crystals were obtained using vapor diffusion, even when crystallization conditions were used similar to those previously identified for the crystallization of A_2AR-GL31 bound to either NECA or adenosine. However, crystallization trials in LCP were successful and yielded two different crystal forms, with space groups P2₁ (monoclinic) and P2₁2₁2₁ (orthorhombic). Complete data sets to 2.6 Å resolution were obtained for each crystal form at European Synchrotron Radiation Facility using the microfocus beamline ID23-2. The structures were solved by molecular replacement using the adenosine-bound conformation of A_2AR-GL31 (PDB ID 2YDO) (Table 1).

The monoclinic crystal form contained one molecule of A_2AR-GL31 per asymmetric unit, whereas the orthorhombic crystal form contained two molecules of A_2AR-GL31 (referred to as molecule A and molecule B) arranged as a nonphysiologic antiparallel dimer. Both structures are very similar to the adenosine-bound and NECA-bound structures with overall RMSDs of 0.4 Å for all residues. Overall, the electron density was well defined for receptors crystallized in both space groups, with the exception of the extracellular surface. In the orthorhombic form, electron density for EL2, EL3, and CGS21680 was clearly resolved (Fig. 2), although density for residues 148–164 in EL2 of molecule A was very weak, and this region was not modeled. In the monoclinic form, residues 155–157 in EL2 and 263 in EL3 could not be modeled, and no density was observed for the (2-carboxyethyl)phenylethylamino group of CGS21680 (Supplemental Fig. 1). In addition, in the monoclinic space group, the electron density for the Glu169^EL2 side chain was weak, and TM2 appears to be in a conformation similar to the NECA-bound A_2AR-GL31 structure. As the model for molecule A of the orthorhombic form was the most complete, and as the structures from the two different space groups were very similar (overall RMSD 0.25 Å), further discussion will focus on molecule A of the CGS21680-bound A_2AR-GL31 structure determined from the orthorhombic space group, as this was considered to be the most representative structure of A_2AR-GL31 bound to CGS21680.

The overall conformation of CGS21680-bound A_2AR-GL31 is the same, within experimental error, as the structures of A_2AR-GL31 bound to either adenosine or NECA, which, with the exception of the extracellular loops, are virtually identical to the conformation of the wild-type receptor bound to UK432097. This conformation of the receptor is best described as “active-like” (Lebon et al., 2011b). The major internal rearrangements of side chains and α-helices characteristic of agonist binding are observed in the receptor core within all of these structures, but the opening of the cytoplasmic face to allow G protein coupling has not occurred to the same extent as observed in the β₂-adrenergic receptor bound to a heterotrimeric G protein (Kobilka, 2011; Rasmussen et al., 2011b; Lebon et al., 2012). Thus, the structures of agonist-bound A_2AR represent a thermodynamically stable conformation primed to bind G proteins upon movement of the cytoplasmic end of TM6, presumably through either thermal fluctuations or by induced fit.

The Binding Mode of CGS21680.

The synthetic agonist CGS21680 is an analog of the natural agonist adenosine with a (2-carboxyethyl)phenylethylamino group on the C2 position of adenine and an N-ethylcarboxyamido group on C5′ of the ribose (Fig. 1). CGS21680 is a selective agonist of human A_2AR that shows weak binding and activation of the human A₁ and A_2B receptors (Jacobson and Gao, 2006). The N-ethylcarboxyamido adenosine moiety of CGS21680 occupies the same binding site as described for NECA, with hydrogen bonds between Ser277^7.42 and His278^7.43 and the 3′ and 2′ hydroxyls of the ribose moiety, respectively. N6 of the adenosine moiety makes hydrogen bonds with Asn253^6.55 and Glu169^EL2, whereas Phe168^EL2 makes strong π-stacking interactions with the adenine ring. Further hydrophobic interactions involving Leu85^3.33, Met177^5.38, Trp246^6.48, Leu249^6.51, Met270^7.35, and Ile274^7.39 complete the orthosteric ligand binding pocket (Fig. 3; Supplemental Table 1). The binding mode of the adenosine moiety in CGS21680 is thus identical to the mode of binding of the analogous regions of the agonists adenosine, NECA (Lebon et al., 2011b), and UK432097 (Xu et al., 2011).

Despite the overall similarity in the binding mode of the adenosine moieties of A_2AR agonists, the shape of the binding pocket at the extracellular surface differs significantly between the various ligand-bound structures (Fig. 4). In the NECA-bound A_2AR structure, seven ordered water molecules form a hydrogen bond network in a wide opening at the extracellular entrance to the ligand binding pocket (Lebon et al., 2011b). In contrast, only a single water molecule resides in the equivalent position of the CGS21680 binding site, which, along with significant van der Waals interactions between the receptor and the (2-carboxyethyl)phenylethylamino moiety, causes the entrance of the ligand pocket to narrow. This is the opposite of that observed in the UK432097 structure, where the bulky extensions to the adenosine moiety cause the entrance to the ligand binding pocket to widen appreciably compared with the NECA-bound structure (Xu et al., 2011). The entrance to the ligand binding pocket is also very open when the inverse agonist ZM241385 is bound, particularly in the structure determined of A_2AR in detergent (3PWH; Dore et al., 2011) in comparison with the structure determined in LCP (3EML; Jaakola et al., 2008), which may reflect the different poses of the phenylethylamino moiety observed between the two structures (Fig. 4).

The (2-carboxyethyl)phenylethylamino group of CGS21680 makes van der Waals interactions with mainly three side chains—namely, Leu267^7.32 at the top of TM7, Glu169^EL2, and His264^EL3 in EL3 (Fig. 3). His264^EL3 also makes a hydrogen bond with Glu169^EL2, thus potentially stabilizing the extracellular structure of the receptor. It is interesting to observe in the structure of A_2AR-GL31 in the monoclinic crystal form that the geometry of the His264^EL3- Glu169^EL2 hydrogen bond is much less favorable, and there is only weak density for His264^EL3, Glu169^EL2, and the (2-carboxyethyl)phenylethylamino group of CGS21680 (Supplemental Fig. 1). It is clear that the extracellular region can adopt a number of different conformations in different crystal structures (Fig. 5). It is not unknown for structures of GPCRs to exhibit slightly different conformations, even in the same crystal (Moukhametzianov et al., 2011), and it is also known that the extracellular region of many GPCRs is the site at which allosteric modulators of receptors can bind (Wheatley et al., 2012). Thus, the subtle differences in structure observed between CGS21680-bound A_2AR in the two different crystal forms may be of some biologic relevance in affecting the equilibrium between the inactive and active states of the receptor.

There are three water molecules (W1, W2, and W3) in the CGS21680-bound A_2AR-GL31 structure that have clear electron density and low B factors (Figs. 2 and 3; Supplemental Fig. 3). The water molecule W1 forms hydrogen bonds to the side chain of Ser67^2.64 and the ribose and adenine moieties of CGS21680 (Fig. 3), and forms a weaker polar interaction with the backbone carbonyl of Ala63, thus forming a bridge between the receptor and the ligand. The top of TM2 is tilted toward the ligand in comparison with the adenosine-bound A_2AR structure, with the Cα of Ser67^2.65 shifted by 2 Å. In addition, the rotamer of Ser67^2.65 has changed with respect to the adenosine-bound structure so that the hydroxyl group is oriented toward the ligand. The water molecule W1 is not resolved in the structure solved from the monoclinic crystal form, in which Ser67^2.65 adopts a similar position seen in the adenosine-bound structure. The water molecule W2 forms hydrogen bonds with the side chains of amino acid residues Glu169^EL2, Asn253^6.55, and Thr256^6.58, whereas the water molecule W3 forms hydrogen bonds with Met177^5.38, His250^6.52, and Asn253^6.55. W2 and W3 both form hydrogen bonds with Asn253^6.55, which probably helps to orient the side chain and position it optimally to form hydrogen bonds with the adenine moiety of the ligand. This is supported by mutagenesis data, which suggest that Asn253^6.55 is one of the most important amino acid residues for the binding mode of ligands to human A_2AR (Kim et al., 2006; Lane et al., 2012). The positions of these three water molecules are also conserved between the structures of A_2AR bound to either adenosine, NECA, or ZM241385 (Jaakola et al., 2008; Lebon et al., 2011b) (Supplemental Fig. 3), supporting their importance in ligand binding.

Functional Characterization of the CGS21680 Ligand Binding Site.

The structure of CGS21680-bound A_2AR-GL31 identified the residues that make direct interactions with the ligand (Fig. 3), which is consistent with the mutagenesis of amino acid residues Leu85^3.33, Thr88^3.36, Phe168^EL2, Leu249^6.51, His250^6.52, Asn253^6.55, Ile274^7.39, Ser277^7.42, and His278^7.43, which significantly reduced the affinity of CGS21680 and/or impaired G protein coupling (Kim et al., 1995; Jiang et al., 1997; Gao et al., 2000; Jaakola et al., 2010). However, the binding mode of the (2-carboxyethyl)phenylethylamino group of CGS21680 was previously unknown. The structure reported here identified contacts between this region of the selective agonist CGS21680 and Ser67^2.65, Glu169^EL2, His264^EL3, and Leu267^7.32. These residues were subsequently mutated and the signaling properties of A_2AR analyzed to define the contribution of each residue to the activity of CGS21680 (Fig. 6). Mutants of the full-length, wild-type A_2AR (S67A^2.65, E169A^EL2, H264A^EL3, and L267A^7.32) were transiently transfected into CHO cells, and the level of basal activity was determined in the absence of ligand and the level of G protein coupling was determined in the presence of the full agonists NECA and CGS21680 and the inverse agonist ZM241385. Of the four mutations tested, the mutant L267A^7.32 showed the biggest reduction (24-fold) in CGS21680 potency, with a nearly 6-fold reduction in potency observed in the mutant H264A^EL3 (Table 2). There was no significant effect on CGS21680 potency in the mutants S67A^2.65 or E169A^EL2, which was particularly surprising given the reduction in potency observed upon NECA binding (3-fold and 15-fold, respectively). Only the mutant H264A^EL3 gave similar reductions in potency for both NECA and CGS21680.

A_2AR expressed in CHO cells displayed basal activity that was totally inhibited by binding of the potent inverse agonist ZM241385 (Fig. 6). In contrast, the mutants S67A^2.65, E169A^EL2, and H264A^EL3 all showed impaired basal activity similar to the levels observed for ZM241385-bound A_2AR (Fig. 6). The basal activity of A_2AR observed in CHO cells is expected to be directly proportional to the expression levels of cell surface receptors; thus, increasing the number of receptors per cell increased the basal activity measured (Fig. 6; Supplemental Figs. 4 and 5). However, when the expression levels of the mutants S67A^2.65, E169A^EL2, and H264A^EL3 were increased, there was no detectable increase in basal activity, and it remained at levels similar to that observed for ZM241385-bound A_2AR (Fig. 6; Supplemental Fig. 5). The effect of the mutation L267A^7.32 was less pronounced than that observed for the other three mutations, but was still significant. These data suggest that residues in the extracellular region of the A_2AR play an important role in dictating the basal activity of the receptor.