(a) The structure-calculation protocol for the substrate D- or substrate D’-loaded sRNPs. Calculations are run in parallel assuming that the fibrillarin corresponding to the loaded side of the RNP is either in the methylation-competent position (on-state) or far from the RNA (off-state); the other fibrillarin copy is in the off-state. Restraint violations larger than 10 Å in either state are classified, to generate two separate restraint lists, representing either the [on,off]- or the [off,off]-state of the sRNP. The procedure is repeated iteratively to eliminate all violations. The lists are then used in a global torsion-angle dynamics (TAD) conformational search. A first selection-step is applied based on a stringent restraint-violation energy filter, combined with cut-offs based on fitness to the 2H-Nop5, 2H-L7Ae, 2H-RNA SANS curves acquired in 42%:58% D2O:H2O. The selected structure is then used as the basis for refinement in Cartesian space. The final ensemble is chosen by a strict filter based on the fitness to the 2H-Nop5, 2H-L7Ae, 2H-RNA and 2H-Nop5/2H-RNA SANS curves acquired in 42%:58% D2O:H2O, combined with a loose energy filter accounting for restraint violations, van der Waals and total energy. (b) Pseudo-genetic algorithm used for selection of structural ensembles with consensus fit to all PRE and SAS data. The number of conformers generated or selected at each step is given in parentheses. Representative mean structures of the four ensembles resulting from Cartesian refinement of [on,off]- and [off,off]-state structure calculations for both substrate D- and substrate D’-loaded sRNPs (Figure 3) are used as templates for a further round of torsion-angle simulated-annealing. For both [on,off]- and [off,off]-states, 300 structures are generated where the conformations of the spin-labels are randomized, and a further 3500 structures with randomized positions of the Nop5-NTD–fibrillarin modules not in contact with the RNA. This generates 4 conformer pools of [on,off]- and [off,off]-state structures, which are then sampled in 250 successive iterations, each containing 20 sub-sampling events. For each sub-ensemble, theoretical ensemble PREs and SAS scattering curves are computed, and fitness to the experimental data is calculated via χ2 (see Methods). Normalized consensus SAS fitness and PRE score are equally weighted in selecting the best sub-ensemble from each iteration. Structural ensembles were selected either exclusively from one pool of conformers ([on,off]- or [off,off]-conformers) or from both the pools of [on,off]- and [off,off]-conformers. In the final refinement step, the conformations of the paramagnetic tags are optimized for the best-fit ensemble selected at the previous step. Here, the conformation of each tag is randomized 3000 times and the best-fit conformations are selected scoring each PRE dataset individually. Both the positions of the fibrillarin modules and the relative populations of the ensemble conformers remain unaltered at this stage.