Improving The Performance Of The Amber Rna Force Field By Tuning The Hydrogen-Bonding Interactions

Abstract

# authors contributed equally * corresponding authors email: sponer@ncbr.muni.cz and pavel.banas@upol.cz KEYWORDS RNA, force field, MD simulation, enhanced sampling, folding, tetraloop.Recent studies generated large conformational ensembles of TNs 14-15, 17, 28-29, 31, 33-34 and TLs 14-15, 17, 28-29, 31, 35, 53-56 in order to assess the performance of RNA ffs. They showed that the available RNA ffs have persisting deficiencies causing, e.g., (i) shifts of the backbone dihedral angles to non-native values, (ii) problems with the c-dihedral angle distribution, and (iii) over-populated SPh and BPh hydrogen bond interactions. 15,25,55 Some of those studies also suggested potential directions for ff improvement, which included modifications of backbone dihedral terms, van der Waals (vdW) radii and charges, RNA interaction with solvent/ions (adjustments of Lennard-Jones parameters typically accompanied by modifications of the Lennard-Jones combining rules via nonbonded fix, NBfix, to balance the RNA-solvent interaction) and enforced distributions from solution experiments. 14, 17, 29-31, 53, 57-60 Computer folding of UNCG TLs appears to be much more challenging than folding of GNRA TLs and description of TNs, 15,35,53,55,61 as for the latter two systems some partial successes have been reported. Note that there have been repeated past claims in the literature about successful folding of RNA TLs in simulations. However, the claims were not confirmed by independent research groups, as extensively reviewed in Ref. 4 . The performance and possible shortcomings of another recently released ff 60 are discussed as part of this work.Previously, we have shown that at least two different imbalances likely contribute to the (in)correct folded/unfolded free-energy balance of TNs and TLs. The first problem was excessive stabilization of the unfolded ssRNA structure by intramolecular BPh and SPh interactions. 62 The excessive binding of 2'-hydroxyl groups (2'-OH) towards phosphate nonbridging oxygens (nbO) was reported earlier 27, 58 and could be partially reduced by using alternative phosphate oxygen parameters developed by Case et al. 63 in combination with the OPC 64 explicit solvent water model. 14 However, the OPC water model was shown to destabilize

Fine-tuning of the AMBER RNA Force Field with a New Term Adjusting Interactions of Terminal Nucleotides

“…11 Subsequently, we have suggested its generalized interaction-specific variant (gHBfix) which represents a true extension of the parent ff. 12 We have demonstrated that boosting of all -NH…N-base-base interactions by 1.0 kcal/mol and destabilizing all sugarphosphate (SPh) H-bonds by 0.5 kcal/mol significantly improves simulations of RNA tetranucleotides and GNRA tetraloop without deteriorating simulations of other systems ( Figure 1). These parameters complement the common OL3 (ff99bsc0χOL3) [13][14][15][16] AMBER RNA ff version used together with modified phosphate van der Waals (vdW) parameters 17 and OPC water model.…”

“…12 is tuning H-bond interactions in order to: (i) support base -base interactions (left panel, blue dashed lines for -NH…N-interactions) and, simultaneously, (ii) destabilize SPh interactions (red dashed lines on the right panel); SPh interactions are those between 2'-OH groups and bridging (bO)/non-bridging (nbO) phosphate oxygens. The gHBfix potential is affecting all interactions of the same type while an analogous term can be used to modulate only structure-specific (individual) H-bonds (abbreviated as HBfix 10,12 ). In the present work, we suggest a variant separately targeting interactions involving terminal nucleotides (tHBfix term).…”

“…Nevertheless, contemporary simulation methods and hardware already allow to obtain sufficiently converged simulation ensembles for TNs. 12,20,30 TN simulations can specifically monitor performance of ffs for several salient energy contributions, namely, (i) SPh and BPh interactions, (ii) base stacking interactions, (iii) backbone conformations and (iv) balance of these contributions with solvation. Experimental data shows that TNs mostly populate A-form conformations while MD simulations tend to significantly sample also non-native intercalated structures (or some other non-native structures, Figure 2) that are considered to be ff artifacts.…”

“…Experimental data shows that TNs mostly populate A-form conformations while MD simulations tend to significantly sample also non-native intercalated structures (or some other non-native structures, Figure 2) that are considered to be ff artifacts. 12,20,24,30 Obviously, when using TNs as a benchmark for ff refinement, one has to be concerned about a possible over-fitting of the ff towards the canonical A-RNA conformation (a potential problem of some recently published ffs 32 ), which may lead to side-effects for simulations of folded RNAs. 12,33 Here, we applied enhanced-sampling simulations to obtain conformational ensembles of five RNA tetranucleotides, i.e., r(AAAA), r(CAAU), r(CCCC), r(GACC), and r(UUUU).…”

Stability of Two-quartet G-quadruplexes and Their Dimers in Atomistic Simulations

“…60 This problem can be partly corrected, for example, by adding a specific term to stabilize the H-bonds. 98 Note that it does not mean that the H-bonding is uniformly underestimated in MD simulations of nucleic acids; there are other H-bonds which are over-stabilized by the force field, especially in RNA. 60,98 No means to enhance stability of the GG base pairing was used in this study.…”