(736f) MD Simulations On Two RNA Hairpins: Restrained Consensus and Unrestrained Failures

We are interested in studying molecular dynamics simulations for two broad purposes: first, we aim to understand the limitations of these techniques in order to define guidelines for their appropriate use as well as to develop improved methods which expand their application; second, we hope to provide computational feedback for experimental hypotheses in studies of structure elucidation, dynamics, ligand binding, and structure-function relationships. Both of these goals were pursued in a computational study of two solution NMR structures of Domain V (DV) from the group II intron. These RNA hairpin structures, one from yeast Ai5γ and the other from Pylaiella littoralis (PL), are quite different structurally despite containing similar structural elements: two helices connected by a bulge region, with the lower helix containing a "catalytic triad", and the upper helix capped by a GAAA tetraloop. We performed two series of explicitly solvated simulations on each variant using the AMBER molecular dynamics suite: five unrestrained simulations to 100 ns each (total of 500 ns for each variant); five restrained (distance, dihedral, and RDC) simulations to 10 ns each (50 ns total for each variant). Both the unrestrained and restrained simulations provided evidence for greater consensus between the global structure of the two variants. The overall structure length of the Ai5γ-DV shrank significantly to resemble the more compact experimental and simulation structure of PL-DV. Although PL-DV was reported to bind ions in the catalytic triad, studies of Ai5γ-DV did not reveal ion binding in this region. However, both unrestrained and restrained simulations suggested ion binding in the catalytic triad, confirming the suggestion of the PL-DV authors that a more sensitive technique was required to observe the binding. Restrained simulations showed a significant rearrangement of the bulge region in both variants without developing restraint violations. Follow-up studies using modified restraints provided evidence that just a few distance restraints produce the remaining differences between the two variants. The unrestrained simulations develop a variety of structural pathologies in the bulge and tetraloop region, which are confirmed by increasing restraint violations at these locations. The failure of MD force fields to accurately model these non-canonical RNA elements suggest a deficiency in the underlying parameters. Until improvements for these deficiencies are developed, computational investigators are cautioned to carefully analyze non-canonical nucleic acid motifs and urged to use experimental restraints whenever possible.

2010-->