Crystal structure of adenosine A_2A receptor in complex with clinical candidate Etrumadenant reveals unprecedented antagonist interaction

Adenosine GPCRs (A1, A2A, A2B, A3) are widely expressed and modulate inflammation and immune responses. A2A and A2B subtypes, generally Gs-coupled, mediate immunosuppressive signaling and are attractive targets in cancer immunotherapy. Antagonizing A2AAR has therapeutic precedent (e.g., istradefylline for Parkinson's disease) and preclinical promise in neurodegeneration. Etrumadenant (AB928) is a dual A2A/A2B antagonist in clinical trials but its detailed binding mode was unknown. The study aims to determine high-resolution A2AAR crystal structures with Etrumadenant using constructs that avoid binding-site mutations, to reveal the ligand–receptor interactions and to understand the compound’s subtype selectivity profile.

Prior structural work on A2AAR largely used thermostabilized StaR2 constructs that include binding pocket mutations (notably T88^3.36A and S277^7.42A), potentially obscuring native interactions. A2AAR antagonists and agonists have been structurally characterized (e.g., ZM241385, xanthines, NECA) showing key roles for residues like N253^6.55 and E169^ECL2. Clinically, A2A antagonists have been developed (istradefylline), and dual A2A/A2B antagonism is of interest for overcoming adenosine-mediated tumor immunosuppression. However, Etrumadenant’s precise binding pose and selectivity determinants had not been elucidated.

Two A2AAR constructs were engineered and used for crystallography. An optimized construct, A2A-PSB2-BRIL, contains S91^3.39K (stabilizes inactive state via sodium pocket) and N154^ECL2A (removes a putative glycosylation site) and lacks mutations in the orthosteric site. Glycosylation state differences between constructs were confirmed by SDS-PAGE and modulated using tunicamycin and PNGase F. A2A-PSB2-BRIL was expressed in Sf9 cells, purified in the presence of Etrumadenant, and crystallized in lipidic cubic phase (monoolein/cholesterol) with PEG400/Tacsimate/HEPES/hexanediol precipitant. Diffraction data (2.1 Å) were collected at DESY P14 and processed with autoPROC/XDS/CCP4; refinement used PHENIX and model building in Coot. The structure was deposited as PDB 8C9W. A second construct, A2A-StaR2-bRIL-A277S, reverts S277^7.42 to wild type but retains T88^3.36A; it was expressed in insect cells, purified (DM/CHS) via Ni-NTA and SEC, crystallized by LCP with PEG400/MPD/thiocyanate, and crystals were soaked with Etrumadenant. Data (2.1 Å) were collected at SLS PXII and processed with XDS/AIMLESS; structure solved by MR with PHASER and refined in PHENIX (PDB 8CIC). Pharmacology: Radioligand binding assays determined Ki values at human A1, A2A, A2B, A3 using CHO or Sf9 membranes and subtype-selective radioligands ([3H]DPCPX or [3H]CCPA for A1; [3H]MSX-2 for A2A; [3H]PSB-603 for A2B; [3H]PSB-11 for A3). Binding at crystallization constructs (A2A-PSB2-BRIL, A2A-StaR2-bRIL) was also measured. Functional antagonism was assessed by TRUPATH BRET G protein dissociation assays in HEK cells co-expressing ARs with Gα-RLuc8/Gβ/Gγ-GFP2 biosensors. NECA at EC80 per subtype stimulated responses; concentration–response curves for inhibitors were fit to derive pIC50 or IC50.

- Two 2.1 Å A2AAR structures with Etrumadenant were solved: A2A-PSB2-BRIL (PDB 8C9W) and A2A-StaR2-bRIL-A277S (PDB 8CIC). - An unprecedented interaction was discovered: the cyano group of Etrumadenant forms a direct hydrogen bond with T88^3.36 (N–O distance ~2.8 Å) in the native-pocket construct. This interaction has not been observed previously for A2AAR antagonists. - Etrumadenant engages canonical anchors: hydrogen bonds from its 2-aminopyrimidine (N3 and exocyclic NH2) to N253^6.55; interaction with E169^ECL2; π–π contacts with F168^ECL2, W246^6.48, and H250^6.52; hydrophobic contacts to V84^3.32, L85^3.33, F168^ECL2; water-mediated network to H278^7.43 and backbone atoms. - Side-chain Y271^7.36 exhibits conformational adaptability to accommodate the relatively large ligand; differences in rotamers correlate with ligand orientation and pocket occupancy. - Despite the T88^3.36A mutation in A2A-StaR2-bRIL-A277S, the overall binding pose of Etrumadenant is similar, but the specific T88^3.36 hydrogen bond is absent; the cyano group is planar in the mutant. - Affinity measurements: Etrumadenant shows high affinity for A2AAR (Ki 0.851 nM), A2BAR (Ki 3.16 nM), and A1AR (Ki 7.59 nM vs [3H]DPCPX; 7.08 nM vs [3H]CCPA), but lower at A3AR. Selectivity over A1 is only ~9-fold (A2A vs A1). - Construct-dependent affinity: For A2A-StaR2-bRIL (with T88^3.36A), Ki is ~39.8 nM, a ~47-fold reduction versus wild type; affinity for A2A-PSB2-BRIL remains essentially unaltered (Ki 1.12 nM). - Functional antagonism (BRET): Etrumadenant potently blocks NECA-induced G protein activation at A2A and A2B, and also inhibits A1 and A3, consistent with binding data. Comparative compounds (Preladenant, ZM241385, PSB-603) show expected subtype selectivity trends. - Sequence analysis indicates strong conservation of Etrumadenant-contact residues between A2A and A2B (differing mainly at L249^6.51/V250^6.51 and extracellular positions), rationalizing dual potency; differences in A3 at several contact residues explain lower affinity.

Determining A2AAR structures with a construct lacking ligand-binding site mutations uncovered a previously unrecognized direct hydrogen bond between T88^3.36 and the cyano group of Etrumadenant, explaining high antagonist affinity and inactive-state stabilization. The frequent use of T88^3.36A-mutated constructs likely obscured this interaction in past antagonist structures. Pharmacological data align with the structural insights: high A2A and A2B affinity and only modest selectivity versus A1AR reflect conservation of key pocket residues, while divergence in A3 explains reduced affinity. The substantial loss of affinity in the StaR2 T88^3.36A construct underscores the functional importance of this residue for Etrumadenant and potentially other antagonists. These findings inform rational design of AR antagonists, suggesting that leveraging the conserved T3.36 interaction can enhance potency while variations at less conserved extracellular positions may be exploited to tune selectivity.

This work presents the first A2AAR co-crystal structures with Etrumadenant and reveals a novel antagonist interaction: a direct hydrogen bond between T88^3.36 and the ligand’s cyano group. Structural and pharmacological data explain Etrumadenant’s potent A2A/A2B antagonism and its limited selectivity versus A1. The study highlights the value of crystallization constructs without binding-pocket mutations to uncover native interactions and shows that commonly used mutations (e.g., T88^3.36A) can mask critical contacts and alter affinity. The structural insights provide a basis for designing next-generation A2A-, dual A2A/A2B-, or pan-AR antagonists with improved selectivity profiles; future work could include solving structures at other AR subtypes (A1, A2B, A3) with Etrumadenant or analogs, and exploring chemotypes that engage T3.36 while discriminating at divergent extracellular residues.

Functional G protein dissociation assays rely on overexpression of receptors and G proteins, which can influence apparent potencies. Structural data were obtained only for A2AAR (two constructs); binding modes at other subtypes were inferred from sequence conservation and are not structurally validated. The StaR2 construct retains a T88^3.36A mutation that prevents observing the native hydrogen bond, and only subtle conformational differences were compared between constructs.