Supercomputers Catch GPCRs in Motion, Reshaping Drug Discovery

G-protein-coupled receptors, or GPCRs, are the largest family of membrane proteins involved in cellular signaling, with more than 800 members in the human genome and roughly one-third of all approved drugs acting on them. For decades, scientists have relied on X-ray crystallography and cryo-electron microscopy to capture these molecular machines, but those techniques deliver essentially static snapshots. They freeze the receptor at a single point in time, leaving the dynamic transitions that underlie activation, allostery, ligand recognition, and signaling largely invisible. A new perspective published in Frontiers in Chemical Biology by Keya Joshi, Victor Ayo Adediwura, and Yinglong Miao of the University of North Carolina at Chapel Hill argues that enhanced sampling molecular dynamics simulations have now matured into a powerful framework for filling in those missing frames, revealing transient intermediate states that experiments alone cannot reach and accelerating the rational design of selective therapeutics with improved efficacy and reduced side effects.

The core problem is one of timescales. GPCR activation typically occurs over milliseconds and longer, far beyond the reach of brute-force molecular dynamics on commodity hardware, which can simulate nanoseconds to microseconds of a receptor’s life. Enhanced sampling methods close this gap by facilitating barrier crossing and improving the characterization of free-energy landscapes. The perspective surveys a diverse toolkit, including Metadynamics, which deposits a history-dependent bias along selected collective variables to push the system over energy barriers; Gaussian-accelerated MD, or GaMD, which adds a harmonic boost potential to smooth the potential energy surface without requiring predefined collective variables; supervised MD and its multiple-walker variant; string-of-swarms approaches that refine transition pathways using swarms of short unbiased trajectories; weighted ensemble simulations that split and merge trajectories while preserving statistical weights; and the recently introduced On-the-fly Probability Enhanced Sampling, or OPES, which reformulates adaptive biasing as the on-the-fly estimation of a target probability distribution rather than the accumulation of a history-dependent bias.

Among the newest arrivals is OneOPES, a multi-replica extension of OPES that combines a convergence-focused replica with exploratory replicas biased along auxiliary collective variables. Its impact is most visible in the quantification of receptor activation free energies. In earlier work on the beta-1 adrenergic receptor, researchers had to construct path collective variables laboriously, tying every intermediate milestone to a curated reference structure in the original RMSD-based formulation known as RPATH. The newer Euclidean PATH, or EPATH, formulation eliminates that curation entirely by placing milestones within a two-dimensional RMSD space spanned by the inactive and active endpoints. When combined with OneOPES, EPATH reproduced the beta-1 adrenergic receptor activation free energy obtained with the tailored RPATH approach, and on the mu-opioid receptor both methods agreed on an activation free energy of 18.0 plus or minus 1.0 kilocalories per mole, but EPATH reached that value in roughly 200 nanoseconds versus the approximately 800 nanoseconds required by RPATH, a fourfold reduction in convergence time with no manual milestone curation.

The same OneOPES framework, deployed with biologically motivated collective variables drawn from the PIF, DRY, and NPxxY microswitches together with a hydration descriptor, mapped multidimensional free-energy landscapes of the beta-1 adrenergic receptor in a POPC-cholesterol bilayer. Three independent multi-replica simulations converged on an activation free energy of 13 plus or minus 1.0 kilocalories per mole for the apo receptor and minus 3 plus or minus 1.0 kilocalories per mole for the adrenaline-bound form, with adrenaline preferentially stabilizing an intermediate pre-active basin in striking agreement with high-resolution backbone NMR experiments. The simulations also quantified how chemistry tunes the barrier: protonation of the conserved D2.50 residue lowered the activation barrier to 3 plus or minus 1.0 kilocalories per mole, whereas sodium ion binding raised it to 15 plus or minus 1.0 kilocalories per mole, rationalizing the inverse-agonist effect of sodium and the activation-coupled protonation switch. A companion study mapped the YY-motif water network in the receptor, showing that adrenaline binding loosens a rigid asparagine-tyrosine scaffold and favors longer water chains of three to four molecules bridging the two tyrosines, while sodium entry is kinetically gated by a hydrophobic cluster.

Not every advance requires an explicit biasing potential. Multiple-walker supervised MD, or mwSuMD, uses short conventional MD windows and a supervised selection metric without adding any biasing energy. With 100-picosecond walker windows, mwSuMD reconstructed the vasopressin V2 receptor complex with its peptide hormone to a resolution of 3.85 angstroms RMSD and dissociated the peptide about ten times faster than the parent SuMD method. Crucially, it captured the full sequence from the inactive glucagon-like peptide-1 receptor, through agonist binding and TM6 outward rearrangement, to G-protein engagement and GDP release. The caveat is important: because progression is guided by supervised selection of short segments, the resulting trajectory characterizes a plausible activation and nucleotide-release cycle rather than unbiased ensemble statistics or physical kinetics, so equilibrium populations and rate constants are not directly recoverable.

Enhanced sampling has also bridged simulation and structural biology in striking ways. Researchers trapped a near fully-activated intermediate of the A2A adenosine receptor in complex with mini-G protein at 2.6 angstrom resolution using a mutation to block the transition to the fully active state. Quantitative fluorine-19 NMR uncovered the conformational states, and nucleotide-exchange assays showed that this intermediate initiates rate-limited nucleotide exchange while the G-protein alpha helical domain remains only partially open. GaMD simulations, validated against the newly resolved cryo-EM structure, revealed that the transition to the fully active complex requires a 2-angstrom lift-up of the G-protein alpha5 helix, an 8-degree clockwise rotation of receptor transmembrane helix 6, and reorganization of the PIF, CWxP, and NPxxY microswitches, providing the missing dynamic bridge between pre-coupled and fully active complexes. In a separate study, metadynamics overcame free-energy barriers that 2-microsecond conventional simulations could not, revealing two major minima corresponding to alternative beta-arrestin orientations in the ghrelin receptor system and showing that the lipid PIP2 modulates the exchange between those states by anchoring the arrestin C-edge loop to the membrane.

Allostery, the regulation of receptor activity through sites distant from the main binding pocket, has emerged as a particularly fertile ground for these methods. Because the orthosteric site is highly conserved across receptor subtypes and often activates multiple receptors simultaneously with poor selectivity, allosteric modulators that bind to more divergent sites offer a promising route to selective therapeutics. Cryo-EM combined with GaMD simulations identified a previously unknown extrahelical allosteric binding site at the M5 muscarinic receptor, located between transmembrane helices 3 and 4 above intracellular loop 2, where the positive allosteric modulator VU6007678 stabilizes the active conformation through interactions consistent with the structure and mutagenesis data. A workflow called GLOW, integrating GaMD with deep learning, ran a total of 66 microseconds of simulations across 44 GPCR systems and found that allosteric modulation primarily acts through conformational selection, with modulator binding reducing conformational flexibility and restricting receptors to fewer low-energy states. Coarse-grained simulations, meanwhile, revealed that PIP2 preferentially stabilizes the active state of the adenosine A2A receptor by binding at the intracellular TM6-TM7 site, enhancing G-protein coupling through lipid-mediated interactions.

Ligand binding and unbinding kinetics, long the domain of surface plasmon resonance and stopped-flow experiments, are now accessible in silico. Weighted ensemble simulations combined with machine learning-based dimensionality reduction characterized how norepinephrine and epinephrine dissociate from beta-1 and beta-2 adrenergic receptors, identifying receptor-specific exit routes: the ligands leave beta-1 primarily through the canonical front-side extracellular loop 2 pathway, whereas both preferentially exit beta-2 through a previously unreported back-side pathway, and the predicted dissociation rate for the beta-1 norepinephrine complex matched experimental values. Adiabatic-bias MD captured the rate-limiting translocation of the drug tiotropium from the muscarinic M3 receptor orthosteric site to the extracellular vestibule, with calculated translocation times correlating well with experimental dissociation rates. Machine learning combined with infrequent metadynamics predicted residence times for morphine and buprenorphine at the mu-opioid receptor that agreed with experiments, revealing that morphine dissociation proceeds through a vestibular intermediate while buprenorphine faces a distinct kinetic bottleneck at the orthosteric-to-vestibular transition.

Peptides and G-proteins, the large partners that small-molecule studies often cannot address, have yielded to specialized variants. Peptide GaMD simulations of the adhesion GPCR ADGRD1 captured the spontaneous binding of a tethered agonist from an extended beta-strand conformation into the orthosteric pocket, where it folded into the U-shaped alpha-helical active conformation seen in cryo-EM structures, and identified two previously unknown intermediate states along the way. Without the G-protein present, only partial binding and receptor deactivation were observed, demonstrating that G-protein coupling is critical for stabilizing the fully active conformation. Protein-protein interaction GaMD then captured spontaneous dissociation of peptide agonists through a conserved extracellular pathway in two adhesion receptors, revealing subtype-specific partially dissociated states and the structural hallmarks of deactivation, including relaxation of the TM6 kink and closure of the binding pocket.

Perhaps most consequentially, the first detailed characterization of G-protein dissociation kinetics from GPCRs combined PPI-GaMD simulations with fluorescence- and BRET-based kinetic assays across five experimentally characterized complexes. The simulations captured spontaneous dissociation events and predicted rate constants that agreed well with experimental data, resolving three low-energy states along the dissociation pathway and two distinct routes: one dominated by interactions between the G-protein alpha5 helix and receptor intracellular loop 2, and another mediated primarily through transmembrane helix contacts. Extending the approach to the adenosine A1 receptor showed that positive allosteric modulators significantly reduced G-protein dissociation rates and stabilized receptor-G-protein interactions. The authors conclude that GPCR activation is intrinsically multidimensional and intermediate-rich, that structured water networks and conserved microswitches act as rate-limiting, ligand-tunable allosteric relays, and that complete signaling cycles can now be characterized computationally. Challenges remain in convergence assessment, quantitative kinetics, and extending coverage beyond class A receptors, but the integration of enhanced sampling with deep learning, cryo-EM, and NMR is positioning these simulations as routine tools for GPCR pharmacology and drug discovery.

Subject of Research: Enhanced sampling molecular dynamics simulations of G-protein-coupled receptor activation, allostery, ligand binding, and G-protein dissociation

Article Title: Advances in enhanced sampling simulations of G-protein-coupled receptors

Article References: Joshi, K., Adediwura, V. A., & Miao, Y. (2026). Advances in enhanced sampling simulations of G-protein-coupled receptors. Frontiers in Chemical Biology, 5, Article 1943336. https://doi.org/10.3389/fchbi.2026.1943336

Image Credits: AI Generated

DOI: 10.3389/fchbi.2026.1943336

Keywords: GPCRs, enhanced sampling, molecular dynamics, GaMD, metadynamics, OneOPES, allostery, ligand binding kinetics, G-protein dissociation, drug discovery, cryo-EM, microswitches