1H-15N NMR spin relaxation and relaxation dispersion experiments can reveal the timescale and extent of protein motions across the ps–ms range, where the ps–ns dynamics revealed by fundamental quantities R1, R2, and heteronuclear NOE can be well-sampled by molecular dynamics simulations (MD). Although the principles of relaxation prediction from simulation are well-established, numerous NMR–MD comparisons have hitherto focused upon the aspect of order parameters S2 due to common artefacts in the prediction of transient dynamics, arising due to issues such as insufficient convergence, water-model limitations, and subtle nuances in both fields. We therefore summarize here all necessary components and highlight existing and proposed solutions, such as inclusion of quantum mechanical zero-point vibrational corrections, and separate MD–convergence of global and local motions in coarse-grained and all-atom forcefields, respectively.
To evaluate the current state of affairs in MD prediction, two model proteins GB3 and Ubiquitin are used to validate five atomistic forcefields against published NMR data, supplemented by the coarse-grained forcefield MARTINI+EN. In Amber and CHARMM-type forcefields, quantitative agreement was achieved for structured elements with minimum adjustment of global parameters. Deviations from experiment occur in flexible loops and termini, indicating differences in both the extent and time-scale of backbone motions. Lack of systematic patterns and water-model dependence suggest that modelling of the local environment limits prediction accuracy. Nevertheless, qualitative accuracy in a 2 μs-CHARMM36m Stam2 VHS-domain simulation demonstrates the potential of MD-based interpretation in combination with NMR-measured dynamics, increasing the utility of spin-relaxation in integrative structural biology.