Thermodynamic, Kinetic, and Structural Determinants of Ligand Selectivity in A2A and A2B Adenosine Receptors

G protein-coupled receptors are major pharmacological targets, yet achieving subtype selectivity remains challenging when closely related receptors share highly conserved orthosteric binding sites. Here, we investigate the molecular determinants governing ligand recognition and unbinding at the adenosine A2A and A2B receptors, two closely related class A GPCRs with markedly different pharmacological profiles. We combine Funnel Metadynamics and adaptive infrequent metadynamics to characterize the thermodynamics and kinetics of three representative ligands: the non-selective antagonist theophylline (TEP), the A2A-selective inverse agonist ZM-241385 (ZMA), and the non-selective full agonist NECA. Across six ligand-receptor complexes, our simulations reproduce experimentally resolved binding modes, predict the unresolved binding poses of TEP and ZMA at A2B, and provide binding free energies consistent with experimental trends. Kinetic simulations further resolve ligand-specific unbinding pathways, metastable intermediates, residence times, rate-determining transitions, and their associated transition-state configurations. Comparison of A2A and A2B reveals how subtle differences within and around their highly conserved orthosteric sites are amplified into distinct thermodynamic and kinetic behaviors. In particular, we identify three major receptor-specific features: differences in hydration and polarity near TM1/TM2/TM7, differences in steric packing and pocket volume at the TM3/TM5/TM6 floor, and a more dynamic network of charged extracellular residues and lipids in A2B that modulates ligand egress. These features rationalize ligand-dependent differences in affinity, residence time, and subtype selectivity, including the preferential stabilization of ZMA-like antagonists at A2A. Overall, our results provide a dynamic atomistic map of the A2A and A2B orthosteric regions and demonstrate how thermodynamic and kinetic information can reveal pharmacologically relevant differences that are not apparent from static structures alone. This framework may support the rational design and repurposing of subtype-selective adenosine receptor ligands.