Exploring RNA conformational space under sparse distance restraints

Russell S Hamilton,William R Taylor

doi:10.1038/srep44074

Russell S Hamilton, William R Taylor

Open Access

https://doi.org/10.1038/srep44074

Copy DOI

Abstract

We show that the application of a small number of restraints predicted by coevolution analysis can provide a powerful restriction on the conformational freedom of an RNA molecule. The greatest degree of restriction occurs when a contact is predicted between the distal ends of a pair of adjacent stemloops but even with this location additional flexibilities in the molecule can mask the contribution. Multiple cross-links, especially those including a pseudoknot provided the strongest restraint on conformational freedom with the effect being most apparent in topologically simple folds and less so if the fold is more topologically entwined. Little was expected for large structures (over 300 bases) and although a few strong localised restrictions were observed, they contributed little to the restraint of the overall fold. Although contacts predicted using a correlated mutation analysis can provide some powerful restrictions on the conformational freedom of RNA molecules, they are too erratic in their occurrence and distribution to provide a general approach to the problem of RNA 3D structure prediction from sequence.

Highlights

Molecules are considered in turn in order of the number of predicted non-local constraints and for each molecule, a mean RMSD from the native structure is calculated over 10 models for ea
Non-local interactions mediated through non-canonical base pairing interactions are much more difficult to predict using conventional sequence analysis but sometimes can be strongly predicted using a correlated mutation analysis
Contacts predicted using a correlated mutation analysis can provide some powerful restrictions on the conformational freedom of RNA molecules, they are too erratic in their occurrence and distribution to provide a general approach to the problem of RNA 3D structure prediction from sequence on their own

Summary

Results

Rfam families containing a known structure and enough sequences for a clear correlation signal were further reduced by removing those that had little tertiary structure or were too large. (See Methods section for details of the selection criteria and Supplementary material for a description of the molecules and their restraints). The selected molecules were divided into three groups that had 1–2, 3–4 and 5–6 predicted non-local interactions (Table 1). Each RNA molecule was simulated in a coarse-grained representation using two distinct simulation methods that operate at different granularities. SimRNA represents each nucleotide with five pseudo-atoms whereas SimGen uses a single point per nucleotide but these are tightly restrained in their secondary structures with random displacements and rotations applied at the secondary structure (stem-loop) level. Varying degrees of random perturbation were tested with each method using different restraint sets

Discussion

Conclusions

Methods