G-protein-coupled receptors, or GPCRs, sit in cell membranes and translate external signals into cellular responses, and their function depends strongly on the conformational states they sample. Insights into these dynamics have remained limited, partly because the two main techniques for probing them, molecular dynamics simulations and NMR spectroscopy, operate on very different scales. Simulations deliver atom-by-atom structural movies but struggle to reach the millisecond-to-second timescales relevant to many NMR experiments, while NMR yields highly sensitive but hard-to-interpret spectroscopic readouts. The new platform is designed to bridge this gap.

At the heart of the approach is the chemical shift, a site-specific NMR parameter that is highly sensitive to protein structure, molecular interactions and the local environment. Rather than treating chemical shifts only as experimental outputs, the team computed them forward from protein structures using established machine-learning predictors, including SHIFTX2, SPARTA+ and UCBShift version 2.0, which were trained on experimental protein NMR data. These predictors were integrated with molecular dynamics simulations across what the authors call the three-dimensional GPCRome.

The scale of the resulting dataset is considerable. Built on the GPCRmd database, a large-scale collection of unified GPCR molecular dynamics data, the team computed chemical shift evolution over 1,000 GPCR simulation systems belonging to more than 400 experimentally solved receptor structures captured in different functional states, such as inactive, intermediate and active. The chemical shift exploration covered over 2.3 milliseconds of accumulated simulation time and produced 7.2 million conformational snapshots with assigned chemical shift profiles, amounting to 38 billion chemical shift values at a resolution of 0.2 nanoseconds. All of these data are accessible through the GPCRmd-NMR web tool, which provides streaming, visualization and analysis features, including simulated one-dimensional and two-dimensional NMR spectra that users can explore residue by residue.

To test whether the computed values hold up against reality, the researchers benchmarked the platform against experimental NMR data. In one case, they simulated the human bradykinin 1 receptor, a peptide receptor involved in chronic pain and inflammation, bound to a peptide agonist, and reconstructed the two-dimensional carbon-13 double-quantum single-quantum spectral pattern for the P8 residue of the ligand. The predicted signals overlapped substantially with experimental spectra previously recorded by DNP-enhanced solid-state NMR at around 100 kelvin, a temperature at which the frozen sample preserves the conformational distribution of the bound ligand. Additional replicate simulations reproduced the same spectrum.

The team then turned to solution NMR data collected at more physiological temperatures. Using backbone chemical shift assignments for the rat neurotensin receptor 1 bound to its natural agonist, retrieved from the Biological Magnetic Resonance Data Bank, they ran ten replicate simulations of one microsecond each mirroring the experimental conditions. SHIFTX2-based predictions showed strong correlations with the experimental chemical shifts for heavy atoms, with correlation coefficients ranging from 0.80 to 0.99, while predictions for amide hydrogen atoms were weaker, at 0.42 to 0.46, which the authors attribute to the greater sensitivity of those atoms to subtle inaccuracies in structural and electronic modeling. SPARTA+ and UCBShift 2.0 performed comparably for heavy atoms. An extended benchmark covering four further datasets, including the beta-1 and beta-2 adrenergic receptors and the adenosine A2A receptor, confirmed robust performance across receptors and heavy atom types, with correlations of 0.82 to 0.99. These five datasets represent all GPCR chemical shift entries currently deposited in the BMRB, comprising 679 experimental measurements in total.

Beyond validation, the framework exposed dynamics that conventional analysis would miss. Monitoring the beta chemical shift of residue F331 in the ligand-binding pocket of the 5-hydroxytryptamine 2B receptor, a drug target in pulmonary arterial hypertension and valvular heart disease, the simulations revealed four distinct conformational microstates exchanging on the nanosecond scale, a process that would typically appear as a single averaged signal in an NMR experiment. In a separate case study on the neurotensin receptor, the platform showed that an experimentally observed average chemical shift of about 1.66 parts per million for the methyl group of residue M250, located next to the P5.50 kink in transmembrane helix 5, actually emerges from three distinct chemical shift populations corresponding to different side-chain orientations, some pointing toward the membrane and others toward the receptor interior. The predicted average of 1.71 parts per million closely matched the experimental value, supporting the relevance of the sampled conformational space and uncovering a previously unrecognized rapid conformational switch near a key receptor activation site.

The platform also distinguished intrinsically disordered regions from ordered ones. Analyzing published simulations of the beta-2 adrenergic receptor carboxyl terminus, the team computed correlations between the temporal evolution of predicted amide nitrogen chemical shifts and the coil versus non-coil state of each residue. Residues in the helix-forming segment 368 to 378 showed the strongest correlations, indicating that structural ordering within this disordered tail can be captured by combining chemical shift predictions with simulations. The same analysis revealed that the carboxyl terminus transiently contacts the intracellular side of the receptor core, with chemical shift changes plausibly reflecting shielding effects from those contacts.

The authors also outline how the platform could streamline the experimental NMR pipeline itself. Simulated one- and two-dimensional chemical shift plots can simplify signal assignment, as illustrated for two aspartate residues in the beta-2 adrenergic receptor, one at the allosteric sodium binding site and one in extracellular loop 2, whose different environments produce clearly distinguishable signals. For reporter selection, traditionally a costly trial-and-error process, the platform can systematically evaluate candidate reporter positions: analysis of adaptive-sampling simulations of the mu-opioid receptor identified residues with significant chemical shift differences between active-like and inactive-like states, concentrated in functionally known regions such as the NPxxY motif and transmembrane helix 6, but also revealing additional interior regions not apparent in static structures. In rhodopsin, the receptor responsible for dim-light vision, the platform tracked the spontaneous insertion of a membrane lipid tail into a transient pocket between transmembrane helices 5 and 6, and the proton chemical shift of residue V204 correlated strongly with the cavity opening distance, with a correlation coefficient of 0.83, shifting from about 1.2 to 0.4 parts per million as the lipid inserted. The authors suggest this could guide future NMR experiments to detect and localize such entrance gateways.

To address the timescale mismatch between simulation and spectroscopy, the authors propose a stepwise protocol covering sampling of state transitions, translation of computed chemical shifts into NMR observables, and reconstruction of observables for distinct conformational ensembles. When simulations reach equilibrium between states, ensemble averages can directly approximate fast-exchange NMR signals; when they do not, a population-weighted combination of state-specific chemical shifts can reconstruct the observable. The authors note limitations, including the relatively poor proton chemical shift performance of current predictors and the scarcity of experimental validation data, with only five GPCR backbone chemical shift datasets publicly available in the BMRB. They position GPCRmd-NMR as a template for integrating simulations with other spectroscopic and biophysical methods, and potentially for other protein families such as kinases, ion channels and nuclear receptors.

Subject of Research: An integrative computational-experimental platform linking molecular dynamics simulations with NMR chemical shifts to study GPCR dynamics

Article Title: Integrative biophysical illumination of the 3D GPCRome dynamics

Article References: Medel-Lacruz, B., García-Recio, A., Aranda-García, D., Patra, A. P., Stepniewski, T. M., Torrens-Fontanals, M., Hagn, F., Bonomi, M., Mao, J., & Selent, J. (2026). Integrative biophysical illumination of the 3D GPCRome dynamics. Nature Structural & Molecular Biology. https://doi.org/10.1038/s41594-026-01891-8

Image Credits: AI Generated

DOI: 10.1038/s41594-026-01891-8

Keywords: GPCRmd-NMR, G-protein-coupled receptors, molecular dynamics simulations, NMR spectroscopy, chemical shifts, GPCRome, conformational dynamics, intrinsically disordered regions, open-access database, receptor activation, rhodopsin, neurotensin receptor