Introduction

G protein-coupled receptors (GPCRs) activated by the nucleoside adenosine are widely distributed and play important roles in transcellular signaling1,2. Four subtypes of adenosine receptors (ARs) exist, the preferentially Gi protein-coupled A1- and A3ARs, and the Gs-coupled A2A- and A2BARs3. Coupling to additional G proteins has been described, e.g., to Gq proteins4,5,6. Adenosine acts as a “stop signal” resulting in strong anti-inflammatory and immunosuppressive effects, mediated by the A2A- and A2BAR subtypes7. Blockade of the A2AAR is beneficial for several pathological conditions, in which adenosine-A2AAR signaling is increased8,9. For example, the A2AAR antagonist Istradefylline has been approved in Japan and the USA for the treatment of Parkinson’s Disease10. Preclinical studies suggest major effects of A2AAR antagonists against Alzheimer’s Disease11,12. The A2AAR, and later on also the related A2BAR, both of which are expressed by immune cells and may be upregulated in cancer cells, have recently emerged as drug targets for the immunotherapy of cancer, constituting purinergic immune checkpoints13.

Etrumadenant (3-[2-amino-6-[1-[[6-(2-hydroxypropan-2-yl)pyridin-2-yl]methyl]-4-yl]pyrimidin-4-yl]-2-methylbenzonitrile, also known as AB928) was developed as one of the first dual-acting adenosine A2A/A2B receptor antagonists14. It constitutes a unique chemotype featuring a poly-substituted 2-amino-4-phenyl-6-triazolylpyrimidine core structure. The drug has entered clinical development and has been evaluated in several clinical phase I and phase II trials for the treatment of cancer15. Despite its advanced stage in drug development, the characterization of Etrumadenant is limited, and the exact drug–receptor binding mode is unknown.

Here, we determined the high-resolution crystal structure of Etrumadenant in complex with a thermostabilized A2AAR construct comprising only two point mutations that do not interfere with ligand binding. The structure reveals unique binding pocket interactions of Etrumadenant including an interaction of its cyano group with T883.36; to the best of our knowledge, this type of interaction has not been previously observed. For comparison, we also determined the high-resolution crystal structure of Etrumadenant in complex with a widely used A2AAR construct that contains a T883.36A mutation in the binding pocket (A2A-StaR2-bRIL-A277S). The structural findings were complemented with an in-vitro pharmacological characterization of Etrumadenant at all AR subtypes. The compound was found to display high affinity in the low nanomolar range for A1-, A2A-, and A2BARs, and it potently blocked G protein activation by these subtypes. Structural data provided an explanation for the compound’s lack of selectivity.

Results and discussion

Exploring the A2AAR binding pocket of Etrumadenant using optimized crystallization constructs

We previously developed an optimized A2AAR crystallization construct designated A2A-PSB1-bRIL, that contains a single point mutation (S913.39K) inside the allosteric sodium binding pocket to stabilize the inactive conformation which significantly enhanced protein thermostability16. For the co-crystallization of Etrumadenant, we used the same modification but inserted an additional point mutation (N154ECL2A) to remove a putative glycosylation site on extracellular loop (ECL) 2 of the receptor. This construct is designated A2A-PSB2-bRIL (PSB: Pharmaceutical Sciences Bonn, bRIL refers to thermostabilized apocytochrome b562RIL17). Mutation of the asparagine in position 154 to either alanine or glutamine had previously been utilized to eliminate post-translational N-linked glycosylation of the A2AAR, as protein glycosylation is expected to inhibit crystal growth due to microheterogeneity18,19,20. Evidence of N-linked glycosylation is missing, and N154ECL2 is not surface-exposed in available A2AAR crystal structures21, indicating that the non-glycosylated form of the A2AAR crystallizes predominantly. Here, we additionally employed sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) to demonstrate that A2A-PSB1-bRIL (bearing the wild type (wt) N154ECL2) is still partially glycosylated, whereas A2A-PSB2-bRIL (bearing an N154ECL2A mutation) had lost N-linked glycosylation (Fig. 1). Glycosylated proteins typically migrate more slowly in SDS-PAGE and generate higher molecular weight smearing22. Despite the fact that only a single glycosylation site is present, smearing could be observed for A2A-PSB1-bRIL, whereas a sharper band was detected for the N154ECL2A mutant (A2A-PSB2-bRIL) as well as for N154ECL2Q and N154ECL2D mutants, indicating the loss of glycosylation (Fig. 1). Alternatively, the glycosyl residues of A2A-PSB1-bRIL could be cleaved off by the enzyme peptide-N-glycosidase F (PNGase F)23 when added to the purified protein. Glycosylation could also be prevented during receptor expression by addition of the glycosylation inhibitor Tunicamycin24.

Fig. 1: SDS-PAGE analysis of the A2AAR glycosylation state.
figure 1

a SDS-PAGE of A2A-PSB1-bRIL compared to different N154 mutations in the same protein background to remove N-linked glycosylation. The red arrow points to the characteristic glycosylation smear. A2A-PSB1-bRIL plus N154A corresponds to the crystallization construct A2A-PSB2-bRIL. The protein marker originates from the same SDS-PAGE gel. b The effect of Tunicamycin and PNGase F on the SDS-PAGE mobility of A2A-PSB1-bRIL, compared to N154Q in A2A-PSB1-bRIL. Equal protein amounts were loaded onto the gel. The addition of Tunicamycin during A2A-PSB1-bRIL expression or PNGase F treatment of the purified protein resulted in the removal of the characteristic glycosylation smear. The band for PNGase F (≈36 kDa) is visible directly below the A2AAR band (observed molecular weight ≈40 kDa, theoretical molecular weight ≈49 kDa). See Supplementary Fig. 1 for uncropped SDS-PAGE images.

Full size image

Utilizing the optimized construct A2A-PSB2-bRIL, we obtained the crystal structure of the A2AAR in complex with Etrumadenant at 2.1 Å resolution (see Table 1 for detailed data collection and refinement statistics). Etrumadenant was well resolved within the orthosteric ligand binding pocket (Fig. 2a, b). Its scaffold shows unique interactions within the A2AAR’s orthosteric binding pocket. Importantly, the cyano group forms a direct hydrogen bond to T883.36 (N-O distance 2.8 Å) (Fig. 2a, b) representing an interaction that has so far not been observed in A2AAR co-crystal structures with various antagonists. T883.36 is conserved within the AR family and was shown to be directly involved in A2AAR agonist binding (illustrated for 5’-N-ethylcarboxamideadenosine (NECA) in Fig. 2c)20. It undergoes significant conformational changes during receptor activation25. The interaction of Etrumadenant with T883.36 by direct hydrogen bonding stabilizes the A2AAR in its inactive state. Notably, the A2A-StaR2-bRIL construct that is extensively used to determine inactive state A2AAR crystal structures16 harbors a T883.36A mutation (see Fig. 3a), that can be expected to affect the affinity of Etrumadenant and possibly other antagonists. In fact, the affinity of Etrumadenant is ~47-fold lower for A2A-StaR2-bRIL as compared to the wt A2AAR (Ki values of 39.8 nm compared to 0.85 nm, see Table 2). In contrast, the affinity of Etrumadenant for our optimized crystallization construct A2A-PSB2-bRIL remained unaltered (Ki 1.12 nm).

Table 1 Data collection and refinement statistics.
Full size table
Fig. 2: The A2AAR binding pocket of Etrumadenant.
figure 2

a Binding pocket of Etrumadenant with residues L167 and F168 clipped for enhanced visualization. The 2Fo–Fc electron density for Etrumadenant is shown in orange mesh (contoured at 1.0 σ). b Binding pocket of Etrumadenant rotated by 180° compared to (a) with parts of the ECL2 and residues A265 to M270 clipped. The 2Fo–Fc electron density for Etrumadenant is shown in orange mesh (contoured at 1.0 σ). Black dashed lines represent hydrogen bonds whereas cyan-colored dashed lines show π-π interactions. c Structural alignment of the A2A-PSB2-bRIL-Etrumadenant (blue/yellow) binding pocket with that of NECA (PDB: 2YDV, represented in green/purple). Hydrogen bonds are shown in yellow and purple, respectively. d Chemical structures of Etrumadenant and NECA.

Full size image
Fig. 3: Comparison of the Etrumadenant binding pocket with that of selected A2AAR antagonists.
figure 3

a The binding pose of Etrumadenant (blue) is compared to the binding pockets of ZM241385 (purple, PDB ID 4EIY21), Vipadenant (green, PDB ID 5OLH27) and Imaradenant (orange, PDB ID 6GT326). The red-colored dashed arrow represents the conformational movement of Y2717.36 in the A2A-PSB2-bRIL-Etrumadenant structure. Of note: The structures of the Imaradenant- and the Vipadenant-complex have been obtained with the A2A-StaR2 construct that among other mutations contains a T883.36A point mutation. The structure of the ZM241385-complex showed two different conformations for T883.36. b Chemical structures of the depicted antagonists. The dotted blue rectangles highlight structural moieties that form the key hydrogen bonding anchor to N2536.55.

Full size image
Table 2 Binding affinities of Etrumadenant and selected antagonists for the human adenosine receptors and for crystallization constructsa.
Full size table

Besides the hydrogen bond to T883.36, Etrumadenant shows multiple additional receptor-ligand interactions. The phenyl ring of Etrumadenant is stabilized by π-π interactions to H2506.52 (T-shaped) and W2466.48 (stacked) (Fig. 2b). Its 2-methyl group comes in contact to V843.32, L853.33 and F168ECL2. It is additionally exposed to a water network connecting the ligand to helices II and III (Fig. 2a). The 2-aminopyrimidine core is stabilized by π-π stacking interactions to F168ECL2 (Fig. 2a) and forms key anchoring interactions by hydrogen bonding of the N3 and the exocyclic NH2-group to N2536.55 (Fig. 2a, b). Hydrogen bonding interactions of the side-chain of N253 are also observed for other ligands including agonists (Fig. 2c) and antagonists (see blue rectangles in Fig. 3). The hydrogen bond network is extended by a direct interaction of the exocyclic NH2-group of Etrumadenant with E169ECL2 (Fig. 2b). E169ECL2 forms a salt bridge to H264ECL3 that is frequently observed in A2AAR crystal structures, but was found to be dependent on the structure of the antagonist and the pH value during crystallization16.

The triazolyl ring of Etrumadenant (Fig. 2d), connected to the 6-position of the core aminopyrimidine, and bearing a substituted pyridylmethylene residue, forms π-π stacking interactions with F168ECL2 and water-mediated hydrogen bonding to H2787.43 and to the backbones of A592.57, I803.28 and A813.29 (Fig. 2a). The pyridine ring is located in close proximity to the entrance of the orthosteric ligand binding pocket at the extracellular ends of helices I and II with direct contacts to S672.65 and Y2717.36. The 2-hydroxyisopropyl residue that is attached to the pyridine of Etrumadenant shows three ambiguous rotamers. We chose to model the rotamer conformation with the hydroxy group in close proximity to a nearby water molecule thereby forming an intramolecular water-mediated hydrogen bond to the pyrimidine N1-nitrogen (Fig. 2a, b).

The sidechain of Y2717.36 was observed to be highly flexible when comparing different A2AAR co-crystal structures19,26. It adapts the hydrophobic pocket to the size of the ligand (as depicted for a selection of ligands in Fig. 3). The relatively large Etrumadenant molecule requires a significant sidechain movement of Y2717.36. This sidechain is located much closer to the orthosteric binding pocket in the ZM241385-bound A2AAR crystal structure, where it is hydrogen-bonded to the water network around the ligand (Fig. 3)21. In that structure, an additional oleic acid molecule occupies the space which Y2717.36 adopts in the current Etrumadenant structure, where the hydrophilic head group of the oleate is displaced by the rotation of Y2717.36 (also compare structures of Imaradenant26 and Vipadenant27) (Fig. 3).

Next, we additionally obtained the crystal structure of Etrumadenant in complex with a modified A2A-StaR2-bRIL receptor construct in which the S2777.42A mutation had been reverted to wt (designated A2A-StaR2-bRIL-A277S), but which still harbored the T883.36A mutation in the binding pocket. A co-crystal structure could be obtained at the same high resolution of 2.1 Å (see Table 2 for detailed refinement statistics). Surprisingly, even though a major interaction partner of the ligand was mutated, the binding pockets of A2A-PSB2-bRIL-Etrumadenant and A2A-StaR2-bRIL-A277S-Etrumadenant are largely similar with only subtle differences (Fig. 4). Notably, the cyano group of Etrumadenant in the A2A-PSB2-bRIL structure is slightly tilted, relative to the plane of the phenyl ring, towards the hydroxy group of T883.36 and deviates from the ideal planar orientation by ~8° (Fig. 4b). The same cyano moiety is planar in the T883.36A mutated structure, but is unable to form the same hydrogen bond interaction due to the mutation. Another difference between both structures can be identified in the rotamers of the 2-hydroxyisopropyl residue and the adjacent sidechain of Y2717.36 (Fig. 4a) which confirms the initially observed flexibility of these moieties.

Fig. 4: Comparison of the binding pockets of Etrumadenant in the A2A-PSB2-bRIL and A2A-StaR2-bRIL-A277S structures.
figure 4

a The binding pose of A2A-PSB2-bRIL-Etrumadenant (blue/yellow) is compared to the pose of A2A-StaR2-bRIL-A277S-Etrumadenant (blue/green). Green- and yellow-colored dashes represent hydrogen bond interactions. b Zoomed panel highlighting the interaction of the cyano group with T883.36 at an N-O distance of 2.8 Å. c 2Fo–Fc electron density for Etrumadenant in the A2A-StaR2-bRIL-A277S structure, shown as gray mesh and contoured at 1 σ.

Full size image

Pharmacological characterization of Etrumadenant

In the original patent describing Etrumadenant, affinity ranges were reported, but no specific Ki or half-maximal inhibitory concentration (IC50) values were provided14. In order to complement the pharmacological characterization of Etrumadenant, we determined its affinities for all human AR subtypes as well as for the crystallization constructs by radioligand binding assays (Table 2). To this end, we employed membrane preparations of Chinese hamster ovary (CHO) cells or Spodoptera frugiperda (Sf9) insect cells recombinantly expressing the respective AR subtype, or crystallization construct, respectively. Additionally, we investigated the inhibitory effects of Etrumadenant in G protein dissociation assays (Fig. 5).

Fig. 5: Inhibition of NECA-induced G protein dissociation by adenosine receptor antagonists.
figure 5

pIC50 values were determined as means from at least three independent experiments ± SEM using a BRET G protein dissociation assay32. Human embryonic kidney (HEK) cells were transfected with the respective AR and Gα-RLuc8, Gβ3, and Gγ9-GFP2 subunits. In the case of A1- and A3ARs, Gαi1 was used whereas Gαs (short isoform GNAS-2) was used for the A2A- and the A2BARs. The receptors were activated by NECA at its EC80 for each receptor subtype (A1AR 20 nm, A2AAR 1 µm, A2BAR 5 µm, A3AR 20 nm), and concentration-dependent inhibition of the signal by Etrumadenant (or standard antagonists) was observed. EC80 values depend on receptor expression56 and probably also on G protein expression levels.

Full size image

In addition to its high affinity for the A2A- and A2BAR subtypes confirmed in the present study (Ki values: A2A, 0.851 nm; A2B, 3.16 nm), we found that Etrumadenant also exhibits high affinity for the A1AR (Ki value: 7.59 nm versus the antagonist radioligand [3H]DPCPX, and 7.08 nm versus the agonist radioligand [3H]CCPA) (Table 2). Thus, the compound showed only about ninefold selectivity comparing A2A- with A1AR affinity, and only 2-fold selectivity for the A2B- versus A1AR subtype. In contrast, we approved that Etrumadenant exhibits high selectivity versus the A3AR (>100-fold), as determined in radioligand binding studies. For comparison, we determined the affinities of standard AR antagonists using the same assays (Table 2). While ZM241385 showed a moderate preference for the A2AAR (Ki values: A2A, 2.04 nm; A2B, 29.5 nm; 12-fold difference), the A2AAR antagonist Preladenant displayed similarly high A2A affinity as Etrumadenant (Ki: A2A, 0.884 nm28) but showed high A2A-selectivity. The A2BAR antagonist PSB-603 was somewhat more potent than Etrumadenant (Ki: A2B, 0.553 nm29) showing high selectivity for the A2BAR subtype.

Subsequently, functional assays were performed to determine concentration-dependent antagonistic effects of Etrumadenant on receptor activation. To this end, we performed bioluminescence resonance energy transfer (BRET) based G protein dissociation assays employing Renilla Luciferase 8 (RLuc8) fused to Gα subunits and green fluorescent protein (GFP) attached to the Gγ subunit30,31,32. AR activation was induced with the non-selective agonist NECA at a concentration where it shows 80% of its maximal effect (EC80). The preferentially Gi protein-coupled A1- and A3AR subtypes were co-expressed with Gαi1-RLuc8, Gβ3, and Gγ9-GFP proteins, whereas the Gs protein-coupled A2A- and A2BAR subtypes were co-expressed with Gαs-RLuc8, Gβ3, and Gγ9-GFP proteins. Etrumadenant was able to block the activation of all four AR subtypes in a concentration-dependent manner. The antagonist was found to be most potent at the A2AAR followed by the A2BAR, but also showed significant antagonistic activity at the other AR subtypes, A1 and A3 (see Fig. 5). Blockade of the Gi protein-coupled A1- and A3ARs will lead to an increase in intracellular cyclic adenosine monophosphate (cAMP) levels thereby counteracting the effects of antagonists at the Gs protein-coupled A2A- and A2BARs33. For this reason, A1- and A3ARs can be regarded as anti-targets in the development of AR antagonists for cancer therapy, and the lack of selectivity may contribute to side-effects3.

For comparison, we additionally investigated the prototypical non-selective A2A/A2BAR antagonist ZM241385, the A2A-selective antagonist Preladenant, and the A2B-selective antagonist PSB-603. Preladenant inhibited the A2AAR with similar potency as Etrumadenant in this assay (IC50 values 85.1 nm, 53.7 nm), whereas the potency of ZM241385 (IC50: 178 nm) was lower than that of Etrumadenant (IC50: 4.57 nm) at the A2BAR, but similar at the A2AAR (IC50 values: 100 nm; 53.7 nm). PSB-603 showed similarly high potency at the A2BAR as Etrumadenant (IC50 values: 3.02 nm; 4.57 nm). It should be kept in mind that the employed functional G protein activation assays require overexpression of receptors and G proteins32. Nevertheless, these data confirm that Etrumadenant is a potent antagonist of A2A- and A2BARs, but its selectivity versus the Gi protein-coupled ARs is low.

To explain this observation, we performed a sequence alignment of all AR subtypes and analyzed the conservation of amino acids that interact with Etrumadenant as observed in the A2AAR co-crystal structures (Fig. 6). In fact, these amino acid residues are largely conserved in the A1-, A2A-, and A2BAR subtypes, which is consistent with the high affinity of Etrumadenant for all three subtypes. The orthosteric binding pockets of the A2A- and the A2BAR differ only by one homologous amino acid exchange (L2496.51 in the A2AAR, V2506.51 in the A2BAR). The recently determined cryogenic electron microscopy structures of the A2BAR in the active state confirmed a similar binding mode of the agonists adenosine and NECA in both receptor subtypes34,35. The extracellular ends of the A2BAR are less conserved, and among the residues that are in contact with Etrumadenant in the A2AAR two major differences can be observed: L2677.32 and Y2717.36 of the A2AAR are exchanged for K2697.32 and N2737.36 present in the A2BAR. Therefore, we hypothesize that Etrumadenant’s aminopyridine core exhibits a comparable binding mode in the A2BAR as in the A2AAR, whereas the substituted pyridylmethylene residue, that extends towards the extracellular space and is relatively flexible, may show differences in binding at both A2AR subtypes.

Fig. 6: Adenosine receptor sequence alignment.
figure 6

Colored residues indicate amino acids with direct contacts to Etrumadenant or interactions via one structural water molecule as observed in the A2A-PSB2-bRIL-Etrumadenant structure. Green colored residues are conserved whereas blue colored residues highlight significant subtype differences. Orange residues indicate exchanges for amino acids with similar properties. Sequences were aligned with Clustal Omega. aThe long C-terminal tail of the A2AAR (residues 361–412) was omitted from the alignment.

Full size image

One notable difference between A1- and A2AARs is S672.65 of the A2AAR that is exchanged for N702.65 in the A1AR. S672.65 forms direct contacts with the pyridine core of Etrumadenant at the extracellular ends of the ligand binding pocket. The overall large conservation between the A1- and A2AAR ligand binding pockets substantiates our observation that Etrumadenant exhibits significant A1AR affinity. However, the exchange of S672.65 to N702.65 in the A1AR might affect the binding of Etrumadenant and explain the slightly reduced affinity of Etrumadenant for the A1AR.

The A3AR, on the other hand, shows significant differences being the least conserved AR subtype regarding Etrumadenant’s binding pocket residues. Three hydrophobic amino acid residues that form direct Etrumadenant contacts in the A2AAR (I662.64, V843.32, and A2737.38) are exchanged in the A3AR for different hydrophobic amino acids or glycine (V722.64, L903.32, and G2677.38). Moreover, we observed direct interactions of Etrumadenant to the side chains of E169ECL2 and H2506.52 as well as a water-bridged hydrogen bond to T2566.58. These residues are conserved among the A1-, A2A-, and A2BARs, whereas the A3AR contains V169ELC2, S2476.52, and I2536.58 in the analogous positions (Fig. 4, red boxes). Variation of these interacting residues in the A3AR provides an explanation for the decreased affinity of Etrumadenant for the A3AR.

Conclusions

The adenosine receptor antagonist Etrumadenant represents a promising clinical candidate for the treatment of cancer, with high affinity for A2A and A2BARs. Both AR subtypes represent purinergic immune checkpoints inhibiting the immune system and showing ancillary direct cancer proliferation-enhancing and pro-angiogenic effects7,33. Blockade of A2A- and A2BARs consequently is expected to exert anti-cancer activity. We show that the affinity of Etrumadenant for the A1AR, which has been discussed as an anti-target in cancer therapy, is similarly high as for the A2A- and A2BAR, with a Ki value in the single-digit nanomolar range. The first A2AAR co-crystal structures of Etrumadenant in complex with the A2AAR provide insights into the A2AAR ligand binding pocket. They revealed that Etrumadenant stabilizes the inactive state of the A2AAR by hydrogen bond interaction to T883.36 through its cyano group. A direct hydrogen bond to T883.36 has thus far not been observed in antagonist-bound A2AAR crystal structures. Importantly, the A2A-StaR2 construct that has been used for the vast majority of A2AAR co-crystal structures16 contains two mutations inside the ligand binding pocket (T883.36A and S2777.42A) preventing the discovery of this hydrogen bond. In fact, out of the 24 different A2AAR antagonists for which co-crystal structures have been solved to date (see Supplementary Table 1) only five have been determined with A2AAR constructs harboring the native T883.36 (ZM24138421, “cmpd-1”36, PSB-211316, PSB-211516, and most recently KW-635637). Nevertheless, the structure of a modified A2A-StaR2-bRIL construct (that has the S2777.42A mutation reverted to the wt residue) in complex with Etrumadenant revealed nearly identical binding poses of Etrumadenant, despite the T883.36A mutation. However, we show that the affinity of Etrumadenant for the A2A-StaR2-bRIL crystallization construct is reduced by 47-fold compared to the wt A2AAR (Ki 39.8 nm vs. 0.851 nm) whereas the affinity to the employed optimized crystallization construct A2A-PSB2-bRIL is unaltered (Ki 1.12 nm). The discovered interaction with T883.36 will be relevant for the design and optimization of future A2AAR antagonists as well as for dual A2A/A2BAR antagonists and pan-AR antagonists, in particular, since T3.36 is conserved among the AR subtypes, not only in humans, but also in rats38. Its conservation likely also contributes to Etrumadenant’s high affinity for the A1AR. The A3AR contains several non-conserved residues that are involved in A2AAR binding, which could explain the selectivity of Etrumadenant versus the A3AR subtype as observed in radioligand binding experiments (more than two orders of magnitude). The present X-ray structure will serve as a basis for the future design of tailored AR antagonists, which have great potential for the treatment of cancer as well as neurodegenerative diseases.

Methods

Expression, purification and crystallization of the A2A-PSB2-bRIL-Etrumadenant complex

The crystallization construct A2A-PSB2-bRIL was cloned using site-directed mutagenesis in order to add the glycosylation mutation N154ECL2A to the previously reported crystallization construct A2A-PSB1-bRIL16. The N154ECL2Q and N154ECL2D mutations were cloned analogously. A2A-PSB2-bRIL was expressed and purified in analogy to the procedure described for A2A-PSB1-bRIL16. Briefly, A2A-PSB2-bRIL was expressed in Sf9 insect cells using GP64-pFastBac1 as baculoviral expression vector. Cells were disrupted by osmotic shock and membranes were repeatedly washed using a washing buffer that contained a high amount of NaCl. The resuspended cell membranes were subsequently incubated with 50 µm Etrumadenant (obtained from MedChemExpress, cat. #HY-129393) and 2 mg mL−1 iodoacetamide for 1 h prior to solubilization. A2A-PSB2-bRIL was solubilized and purified from Sf9 membranes similarly as described for A2A-PSB1-bRIL16. Etrumadenant was added to wash buffer I and to wash buffer II at 50 µm and 25 µm concentration, respectively. The protein was eluted with four column volumes of elution buffer containing 25 µm Etrumadenant, 25 mm HEPES pH 7.5, 800 mm NaCl, 10% (v/v) glycerol, 220 mm imidazole, 0.025% (w/v) dodecyl-β-d-maltoside (DDM), and 0.005% (w/v) cholesteryl hemisuccinate (CHS).

The A2A-PSB2-bRIL-Etrumadenant complex was concentrated to a volume of 20–30 µL using 100 kDa cut-off Vivaspin concentrators (Sartorius), and immediately used for crystallization experiments. The complex was reconstituted into lipidic cubic phase using the two-syringe method39 by mixing the protein with a molten lipid mixture [90% (w/w) 1-oleoyl-rac-glycerol (Sigma), 10% (w/w) cholesterol (Sigma)] in a 2 to 3 ratio. Crystallization experiments were performed using an automatic crystallization robot (Formulatrix NT8) by overlaying 50 nL of mesophase with 800 nL of precipitant solution. The A2A-PSB2-bRIL-Etrumadenant complex crystallized in 30% (w/v) PEG400, 7% (w/v) Tacsimate pH 7.0 (Hampton Research, cat. #HR2-755)40, 100 mm HEPES-Na pH 7.4, 1.8% (w/v) 2,5-hexandiol (Molecular Dimensions, cat. #MD2-100-226). Crystals were harvested with micromounts (MiTeGen) and flash-frozen in liquid nitrogen without further cryo-protection.

X-ray diffraction data collection and structure elucidation of the A2A-PSB2-bRIL-Etrumadenant structure

X-ray data collection was carried out at 100 K on EMBL beamline P14 of the DESY synchrotron (Hamburg, Germany). The x-ray wavelength used for the experiment was 0.97625 Å. An Eiger2 16 M detector was placed at a distance of 340 mm behind the crystal, which was rotated for 360° while diffraction images were recorded at 0.15° steps with exposure for 0.01 s. All datasets were indexed, integrated, scaled, and converted to structure factor amplitudes using ISPyB software: autoPROC41, XDS42, CCP443, POINTLESS44, AIMLESS45, STARANISO46. Crystallographic statistics are presented in Table 1. PDB ID 5IU447 was used as the starting model for refinement with phenix48. Coot49 was used for model building. The stereochemical restraints for the ligand were generated with the GRADE web server50. The Ramachandran plot statistics were determined to 97.65% (favored), 2.09% (allowed), and 0.26% (disallowed).

Expression, purification and crystallization of the A2A-StaR2-bRIL-A277S-Etrumadenant complex

A2A construct (A2A-StaR2-bRIL-A277S) containing the same thermostabilizing and deglycosylation site mutations as PDB ID 5IU427 (except for the S2777.42A mutation, which is reverted to wt) was codon-optimized and cloned between the BamHI and HindIII sites of pFastBac1 (Trenzyme). The bacmid was generated by transforming this plasmid into E. coli strain DH10EMBacY (MultiBac, Geneva Biotech). Isolated bacmid DNA was transfected into Sf9 insect cell using Cellfectin (Invitrogen) to generate baculovirus. For large-scale expression, High Five insect cells growing at 27 °C in Sf900 II medium (Invitrogen) at 2.5 ∙ 106 ∙ mL−1 were infected with P2 baculovirus and harvested at 48 h post infection. Cells were harvested by centrifugation and the pellet was stored at −80 °C. Cells were thawed at room temperature and resuspended in 40 mm Tris-HCl pH 7.6, 1 mm ethylenediaminetetraacetic acid (EDTA), and cOmplete EDTA-free protease inhibitors tablets (Roche). Membranes were fractionated by passing the cell once using microfluidizer (Microfluidics) operated at 8000 psi. Membranes were centrifuged at 42000 rpm using a Ti45 rotor (Beckman) and washed once with 40 mm Tris-HCl pH 7.6, 1 M NaCl, and cOmplete EDTA-free protease inhibitors tablets. Membranes were centrifuged again and resuspended in 40 mm Tris−HCl pH 7.6, cOmplete EDTA-free protease inhibitor cocktail tablets and frozen at −80 °C. Unless otherwise stated, all purification procedures were carried out at 4 °C. Membranes were solubilized with 1.5% (w/v) decylmaltoside (DM) and 0.1% (w/v) CHS for 1 h. Insoluble fractions were pelleted by centrifugation at 42,000 rpm using a Ti45 rotor (Beckman) for 30 min. A2A-StaR2-bRIL-A277S was purified by loading the supernatant (supplemented with 8 mm imidazole) into a 5 mL cartridge containing Ni-NTA super flow resin (Qiagen) at 2 mL min−1 using an ÄKTA pure system (Cytiva). The resin was first washed with 25 mL of 40 mm Tris-HCl pH 7.5, 400 mm NaCl, 0.15% (w/v) DM, 0.002% (w/v) CHS, and 8 mm imidazole and then washed once more with 25 mL of 40 mm Tris-HCl pH 7.5, 400 mm NaCl, 0.% (w/v) DM, 0.002% (w/v) CHS, and 72 mm imidazole. The protein was eluted in the same buffer containing 272 mm imidazole, concentrated using Vivaspin turbo 15 mL with a molecular weight cut-off of 50 kDa (Sartorius) and loaded onto a Superdex 200 10/300 GL increase column equilibrated in 40 mm Tris-HCl pH 7.4, 200 mm NaCl, 0.15% (w/v) DM, and 0.002% (w/v) CHS. Fractions containing A2A-StaR2-bRIL-A277S were pooled, concentrated to 22.5 mg ml−1 aliquots and frozen at −80 °C. Protein purity and homogeneity were controlled by SDS-PAGE and fluorescence size exclusion chromatography (FSEC).

For crystallization, frozen A2A-StaR2-bRIL-A277S aliquots were thawed on ice and centrifuged at 18,213×g for 10 min. Lipidic cubic phase (LCP) crystallization was performed by mixing the protein into monoolein (containing 10% (w/w) cholesterol) at 2:3 (w/w) ratio. The resulting LCP was dispensed using the mosquito LCP (SPT Labtech) using a bolus/precipitant solution ratio of 40:800 nL. Crystals were obtained using precipitant solution containing 0.1 M sodium citrate pH 5.0, 50 mm sodium thiocyanate, 3% (v/v) 2-methyl-2,4-pentanediol (MPD), 21–32% (w/v) PEG400, and 2 mm theophylline. Crystals appeared overnight and grew to full size (50–60 µm in the longest dimension) over a week. To prepare the A2A-StaR2-bRIL-A277S-Etrumadenant complex, crystals were soaked overnight in the same precipitant solution by replacing theophylline with 1 mm Etrumadenant27. After soaking overnight, crystals were harvested with MicroLoops LD (mitogen) and frozen in liquid nitrogen.

X-ray diffraction data collection and structure elucidation of the A2A-StaR2-bRIL-A277S-Etrumadenant structure

Diffraction data were collected at the Swiss Light Source (SLS) beamline PXII. Crystal was exposed to a 25 × 13 µm X-ray beam (wavelength 0.99997 Å) at 25% transmission. A total of 180° of rotational data were collected using 0.1° oscillation and 0.05 s exposure per image. Dataset was processed and scaled to 2.1 Å using XDS42 (built 20161205), then merged and converted to mtz file format with AIMLESS45 (version 0.7.3 from CCP443 distribution 7.0.066). The structure was solved by molecular replacement with PHASER51 (version 2.8.2 from CCP4 distribution 7.0.066), using PDB ID 5IU427 as the search model. The model was rebuilt and refined using COOT49 and PHENIX52 (version 1.14) using TLS and optimizing 8CIC and ADP weight. After structure refinement, the model was validated using MolProbity53 (from PHENIX version 1.14). The Ramachandran plot statistics were determined to 99.48% (favored), 0.52% (allowed), and 0% (disallowed).

Radioligand binding studies

Radioligand binding assays were performed on CHO cell membranes or Sf9 insect cell membranes expressing the respective human wt adenosine receptor or A2AAR crystallization constructs (A2A-PSB2-bRIL or A2A-StaR2-bRIL)16,38. The agonist [3H]CCPA or the antagonist [3H]DPCPX were employed as radioligands for the A1AR (at 1 nm and 0.4 nm final concentration, respectively), the antagonist [3H]MSX-2 was used for the A2AAR (at 1 nm final concentration), the antagonist [3H]PSB-603 for the A2BAR (at 0.3 nm final concentration) and the antagonist [3H]PSB-11 for the A3AR (at 0.5 nm final concentration). All assays were performed in 50 mm Tris buffer (pH 7.4 at room temperature) at a final volume of 400 µL (A1-, A2A-, and A3ARs) or 1000 µL (A2BAR). Test compounds were dissolved in dimethyl sulfoxide and incubated at room temperature with the respective membranes and radioligand for 90 min (A1AR, [3H]CCPA), 60 min (A1AR, [3H]DPCPX), 30 min (A2AAR), 75 min (A2BAR) and 45 min (A3AR). The final dimethyl sulfoxide concentration was 1%. Then, the mixture was filtered through GF/B glass fiber filters using a cell harvester (Brandel). For the A2AAR assay, filters were pre-soaked in an aqueous solution of 0.3% (w/v) of polyethyleneimine for at least 30 min to reduce non-specific binding. Filters were then washed three times with 2 mL ice-cold Tris buffer (50 mm, pH 7.4 at room temperature). Filters containing the A2BAR were washed with the same ice-cold Tris buffer but with the addition of 0.1% BSA. The remaining radioactivity was quantified after incubation for 9 h with scintillation cocktail (Beckmann Coulter) using a scintillation counter (Tricarb 2700TR).

G protein dissociation assays

TRUPATH BRET² assays32 were performed as previously described54 (TRUPATH was a gift from Bryan Roth (Addgene kit #1000000163)). In case of Gαi/o-coupled adenosine receptors (A1 and A3ARs), a biosensor composed of Gαi1-RLuc8, Gβ3, and Gγ9-GFP2 was used. In case of the Gαs-coupled A2A- and A2BARs, the biosensor consisted of Gαs-RLuc8, Gβ3, and Gγ9-GFP2. Antagonist solution was incubated with the cells for 20 min before the addition of luciferase substrate solution (50 µm Deep Blue C, Biomol). Agonist solution (NECA) was added 5 min after the addition of the substrate solution at its EC80 concentrations (A1AR 20 nm, A2AAR 1 µm, A2BAR 5 µm, A3AR 20 nm) and incubated for 5 min until measurement. GFP2 fluorescence (515 nm emission filter) was divided by RLuc8 luminescence (395 nm emission filter) to obtain BRET ratios. Data was normalized to controls (100% activation = NECA without antagonist, 0% activation = no agonist), and IC50 values were obtained by a four-parameter sigmoidal curve fit in GraphPad PRISM v8.0 (GraphPad, San Diego, CA). The Gαs biosensor appeared to display a markedly lower expression level than the Gαi biosensor.

SDS-PAGE analysis

Proteins for SDS-PAGE were expressed and purified from 40 mL of Sf9 insect cell culture for thermostability assessment16. Proteins were analyzed on homemade 10% SDS-PAGE gels casted using bis-2-amino-2-(hydroxymethyl)propane-1,3-diol (bis-Tris) buffer. Protein samples were prepared using NuPAGE loading dye (ThermoFisher, cat. #NP0007) supplemented with a final concentration of 50 mm dithiothreitol (DTT). Samples were incubated for 30 min at 37 °C prior to SDS-PAGE analysis using 3-(N-morpholino)propanesulfonic acid (MOPS) running buffer without addition of sodium hydrogen sulfite. SDS-PAGE gels were stained with Coomassie brilliant blue R-250 and destained using hot water. In order to investigate the effect of Tunicamycin on A2AAR glycosylation, the respective insect cell culture was treated with 1 µg mL−1 of Tunicamycin (CaymanChemical, cat. #11445) during infection. PNGase F (New England Biolabs, cat. #P0704S) was used to cleave the glycosylation in the purified protein prior to SDS-PAGE analysis using 375 units in a total reaction volume of 22.5 µL followed by incubation at 16 °C for 16 h.