The nuclear magnetic resonance of CCCC RNA reveals a right-handed helix, and revised parameters for AMBER force field torsions improve structural predictions from molecular dynamics

Abstract

The sequence dependence of RNA energetics is important for predicting RNA structure. Hairpins with C(n) loops are consistently less stable than hairpins with other loops, which suggests the structure of C(n) regions could be unusual in the "unfolded" state. For example, previous nuclear magnetic resonance (NMR) evidence suggested that polycytidylic acid forms a left-handed helix. In this study, UV melting experiments show that the hairpin formed by r(5'GGACCCCCGUCC) is less stable than r(5'GGACUUUUGUCC). NMR spectra for single-stranded C(4) oligonucleotide, mimicking the unfolded hairpin loop, are consistent with a right-handed A-form-like helix. Comparisons between NMR spectra and molecular dynamics (MD) simulations suggest that recent reparametrizations, parm99χ_YIL and parm99TOR, of the AMBER parm99 force field improve the agreement between structural features for C(4) determined by NMR and predicted by MD. Evidently, the force field revisions to parm99 improve the modeling of RNA energetics and therefore structure.

Figure 1

(A) Single-stranded r(CCCC) with the β, γ, ε, and χ torsion angles labeled. (B) Atom notation used in cytidine and d-ribose. (C) Hairpin formed by r(5′GGACCCCCGUCC), termed HPC₄.

Figure 2

NOESY walk region from an 800 ms mixing time ³¹P-decoupled spectrum. Arrows indicate the direction of the walk from the 5′ to 3′ end. The C3 H6–C2 H1′ cross-peak overlaps with the C2 H6–C2 H1′ resonance.

Figure 3

200 ms NOESY spectrum of r(CCCC) (top) at 5 °C showing the cross-peaks of H5 and H1′ protons to sugar protons and 200 ms NOESY spectrum of r(CCCC) (bottom) at 5 °C showing the cross-peaks from the H6 protons to the sugar protons. Intense cross-peaks between C2 H6 and C1 H2′, C3 H6 and C2 H2′, and C4 H6 and C3 H2′ indicate base–base stacking. The weak cross-peak between C3 H6 and C4 H2′ indicates a population of r(CCCC) where the 3′ sugar is inverted.

Figure 4

Time evolution (in nanoseconds) for the minimized A-form starting structure of r(CCCC) with the parm99TOR force field. The top two plots show the rmsds of the heavy atoms for the whole structure and of the backbone, respectively, relative to A-form r(CCCC). After 770 ns, C1 intercalates between C3 and C4. The remaining plots correspond to the χ and δ dihedral angles for each residue. δ dihedral angles of 78–90° and 140–152° correspond to C3′-endo and C2′-endo sugar puckers, respectively. Anti, high anti, and syn conformations were defined by χ dihedral angles of 180–239°, 240–300°, and 0–120°, respectively.

Figure 5

(A) Three-dimensional representation of r(CCCC) when C1 intercalates between C3 and C4 after 770 ns in the MD simulation with A-form (C3′-endo/anti) starting structure with the parm99TOR force field. (B) C1 intercalated between C3 and C4. The distances shown correspond to C1 H5–C4 H3′ (2.5 Å), C1 H5–C3 H3′ (2.1 Å), and C1 H5–C3 H2′ (3.3 Å) distance. (C) 200 ms NOESY spectrum of r(CCCC) showing the absence of the hypothetical H–H cross-peaks (red boxed labels) predicted by the parm99TOR simulation after 770 ns. (D) Typical A-form base stacking between C1 and C2 (from the nucgen structure). (E) Base stacking between C1 and C3 after C1 intercalates between C3 and C4 observed after 770 ns. (F) Base stacking between C1 and C4 after C1 intercalates between C3 and C4 observed after 770 ns. Residues C1–C4 are colored green, pink, orange, and cyan, respectively.

Figure 6

Time evolution (in nanoseconds) of the heavy atom rmsd of the simulations of non-A-form starting structures relative to A-form r(CCCC) for each force field. In both the C3′-endo/syn and C2′-endo/syn starting structures in the parm99 force field, C1 stacks on C4 forming a looplike structure after 656 and 931 ns, respectively. Around 600 ns for the C3′-endo/syn starting structure with the parm99χ_YIL force field, C1 intercalates between C3 and C4.

Figure 7

Comparisons between MD-predicted and NMR-measured ³J couplings (top) and distances (bottom). The top bar graph in each panel summarizes the results from simulations starting from an A-form structure (Tables 3 and 4). The bottom bar graph (Extended MD) in each panel adds to A-form starting structure simulations the results obtained from simulations starting with non-A-form structures in the time range after reaching an A-form-like structure and before any base intercalation. Table 5 details the portion of each simulation used for comparison. For parm99, parm99χ_YIL, and parm99TOR, the total Extended MD time is 1500, 4975, and 2896 ns, respectively. The Extended MD simulation data for parm99 are the same as the C3′-endo/anti simulation data because of the non-A-form starting structure simulations not forming an A-form-like structure. parm99 (green), parm99χ_YIL (red), and parm99TOR (blue) force fields were tested. (A) Percentage of MD ³J couplings correctly predicted within ±0.5 Hz. (B) Average of the absolute values of differences between MD-predicted and NMR-measured ³J couplings. (C) Percentage of MD distances correctly predicted between the error limits of the measured NOEs. (D) Average of the absolute values of differences between MD-predicted distances and NMR-measured NOEs. Tables S18–S23 of the Supporting Information show the MD-predicted and NMR-measured values used in these plots.

Figure 8

Structures of r(CCCC) representative of those with typical rmsds relative to A-form and aligned with nucgen A-form structure (black) in panels A and B by aligning residue C1 in both structures: (A) parm99χ_YIL (green), (B) parm99TOR (red), and (C) parm99χ_YIL aligned with parm99TOR. These alignments were generated with PyMOL using the align function.