Computer Folding of RNA Tetraloops: Identification of Key Force Field Deficiencies

Abstract

The computer aided folding of biomolecules, particularly RNAs, is one of the most difficult challenges in computational structural biology. RNA tetraloops are fundamental RNA motifs playing key roles in RNA folding and RNA-RNA and RNA-protein interactions. Although state-of-the-art Molecular Dynamics (MD) force fields correctly describe the native state of these tetraloops as a stable free-energy basin on the microsecond time scale, enhanced sampling techniques reveal that the native state is not the global free energy minimum, suggesting yet unidentified significant imbalances in the force fields. Here we tested our ability to fold the RNA tetraloops in various force fields and simulation settings. We employed three different enhanced sampling techniques, namely temperature replica exchange MD (T-REMD), replica exchange with solute tempering (REST2), and well-tempered metadynamics (WT-MetaD). We aimed to separate problems caused by limited sampling from those due to force-field inaccuracies. We found that none of the contemporary force fields is able to correctly describe folding of the 5’-GAGA-3’ tetraloop over a range of simulation conditions. We thus aimed to identify which terms of the force field are responsible for this poor description of TL folding. We showed that at least two different imbalances contribute to this behavior, namely, overstabilization of base-phosphate and/or sugar-phosphate interactions and underestimated stability of the hydrogen bonding interaction in base pairing. The first artifact stabilizes the unfolded ensemble while the second one destabilizes the folded state. The former problem might be partially alleviated by reparameterization of the Van der Waals parameters of the phosphate oxygens suggested by Case et al., while in order to overcome the latter effect we suggest local potentials to better capture hydrogen bonding interactions.

Free Energy Landscape of GAGA and UUCG RNA Tetraloops

“…2 While simulations initialized in the vicinity of the native state are stable on short time-scales under a variety of simulation conditions, [3][4][5][6][7][8] more recent works strongly suggest that these systems are not correctly modeled by the current Amber force field. [9][10][11][12][13][14] Although different improvements have been proposed, 15 there is growing evidence that none of the available corrections are able to capture the crucial non-canonical interactions present in these tetraloops. 12,13 Despite their small size, an ergodic sampling of these systems requires substantial computational resources, in the order of hundreds of µs using massively parallel simulations.…”

“…[9][10][11][12][13][14] Although different improvements have been proposed, 15 there is growing evidence that none of the available corrections are able to capture the crucial non-canonical interactions present in these tetraloops. 12,13 Despite their small size, an ergodic sampling of these systems requires substantial computational resources, in the order of hundreds of µs using massively parallel simulations. 5,10,12 For this reason, full convergence of MD simulations on RNA has been so far achieved for simple systems such as tetranucleotides.…”

“…The discrepancy with respect to predictions obtained from experimental data is significant for GAGA and critical for UUCG tetraloop. Two effects mainly contribute to this discrepancy: i) overstabilization of highly-stacked, compact structures with no base-pairs 13 This analysis could not be carried out for the GAGA tetraloop, due to the paucity of available NOE…”

Polarizable force field for RNA based on the classical drude oscillator

“…As RNA tetraloop structures are important to stabilizing RNA secondary structure, they serve as an important indicator of force field quality . Four RNA stem‐loop structures were simulated in this work: 1UUU, 2KOC, 1YN2, and 1YN1.…”

“…force field quality. [28,[101][102][103] Four RNA stem-loop structures were simulated in this work: 1UUU, 2KOC, 1YN2, and 1YN1. All experimental structures were determined by solution NMR.…”