Automated and assisted RNA resonance assignment using NMR chemical shift statistics

Abstract

The three-dimensional structure determination of RNAs by NMR spectroscopy relies on chemical shift assignment, which still constitutes a bottleneck. In order to develop more efficient assignment strategies, we analysed relationships between sequence and (1)H and (13)C chemical shifts. Statistics of resonances from regularly Watson-Crick base-paired RNA revealed highly characteristic chemical shift clusters. We developed two approaches using these statistics for chemical shift assignment of double-stranded RNA (dsRNA): a manual approach that yields starting points for resonance assignment and simplifies decision trees and an automated approach based on the recently introduced automated resonance assignment algorithm FLYA. Both strategies require only unlabeled RNAs and three 2D spectra for assigning the H2/C2, H5/C5, H6/C6, H8/C8 and H1'/C1' chemical shifts. The manual approach proved to be efficient and robust when applied to the experimental data of RNAs with a size between 20 nt and 42 nt. The more advanced automated assignment approach was successfully applied to four stem-loop RNAs and a 42 nt siRNA, assigning 92-100% of the resonances from dsRNA regions correctly. This is the first automated approach for chemical shift assignment of non-exchangeable protons of RNA and their corresponding (13)C resonances, which provides an important step toward automated structure determination of RNAs.

Figure 1.

¹H and ¹³C chemical shift statistics for central nucleotides of Watson–Crick base-paired triplets in dependence of the RNA sequence displayed in form of box plots. Chemical shifts of residue i are given for a trinucleotide sequences consisting of residues i – 1, i and i + 1. Residue i is underlined. In addition, categories for nucleotides at the 5′ and 3′ terminus are displayed. The number of data points is given next to each box plot. (a) ¹H chemical shifts of H6 and H8. (b) ¹H chemical shifts of H5 and H1′. (c) ¹³C chemical shifts of base carbons C6 and C8. (d) ¹³C chemical shifts of C5 of uracils. (e) ¹³C chemical shifts of C5 of cytosines. (f) ¹³C chemical shifts of C1′. (g) Definition of the box plots.

Figure 2.

Two-dimensional chemical shift statistics for the central nucleotide of Watson–Crick base-paired triplets. Scatter plots display the clusters of chemical shift correlations for the different categories. Bivariate normal distributions are represented by ellipses at 86% probability. (a) H5–H6 chemical shift correlations that are found in a 2D ¹H–¹H TOCSY spectrum. (b) C8–H8 chemical shift correlations that are found in a 2D ¹H–¹³C HSQC spectrum.

Figure 3.

Application of 2D chemical shift statistics for proposing starting points for the assignment of the 22 nt RNA TASL1. (a) Secondary structure of TASL1. The region of regular dsRNA for which the statistics were made is highlighted in green. The chemical shifts of the last nucleotide of the stem (closing base-pair before the loop) typically fall within the clusters of regular dsRNA and thus the nucleotide is boxed in green (unfilled). In addition, the terminal nucleotides for which separate statistics were generated are boxed in pink. (b) Region of the ¹H–¹³C HSQC of TASL1 showing C8–H8 correlations. The corresponding bivariate normal distributions are displayed in form of ellipses. (c) H5–H6 correlations in a 2D TOCSY spectrum of TASL1 with uracil signals highlighted by orange crosses. (d) H5–H6 correlations in a 2D TOCSY spectrum of TASL1 with cytosine signals highlighted by orange crosses. Only one cytosine cross-peak is located outside the ellipses, which originates from the cytosine in the loop. (e) Scheme of the expected NOE correlations between aromatic and H1′ protons. The 12 unambiguous starting points for resonance assignment obtained from 2D H5–H6 and H8–C8 chemical shift statistics are indicated in green. Ambiguous predictions are shown in orange.

Figure 4.

Application of 1D statistics during assignment walks. (a) Part of a 2D NOESY spectrum of TASL1 showing cross-peaks of the G3 H8 resonance that was already assigned based on 2D statistics. As illustrated schematically (bottom) the expected NOE correlations include two cross-peaks to H1′ resonances, one to nucleotide G2 and one to G3. The 1D statistics (right) help here to differentiate between the two cross-peaks assigning them to G2 H1′–G3 H8 and G3 H1′–H8. Only one of the ¹H chemical shifts lies within the borders of the GGX statistic and is therefore assigned to G3 H1′. The other ¹H chemical shift lies within the borders of the 5′GGX statistic and is assigned to G2 H1′. (b) Cross-peaks of two adenosine H2 resonances in the 2D NOESY spectrum of TASL1. At the bottom and on the right 1D statistics of the relevant triplets are displayed. (c) Scheme of TASL1 showing expected NOE correlations. The two adenosine H2 resonances and their correlations visible in the NOESY spectrum are colored in magenta. The involved H1′ resonances are color-coded as the chemical shifts statistics in panel b.

Figure 5.

Workflow of automated RNA assignment with the programs Chess2FLYA and FLYA. RNA secondary structure information in the connectivity table format (.ct file format) is used as an input for the program Chess2FLYA. RNA chemical shift statistics are supplied with the program, which generates .prot, .aco, .seq and .wc input files that are then used as an input of the FLYA automated resonance assignment algorithm (17). In addition, FLYA is supplied with peak lists of a 2D TOCSY, a 2D NOESY and a natural abundance ¹H–¹³C HSQC spectrum in order to obtain the RNA assignment.

Figure 6.

Assignment of five RNAs by the fully automated assignment approach using Chess2FLYA and FLYA. Regions with triplet-based statistics (central nucleotide of a Watson–Crick base-paired triplet) and of Watson–Crick base-paired termini are colored in cyan. Nucleotides before irregularities with statistics that only consider the base type are shown in cyan boxes. For unmarked nucleotides only general statistics were available. This is the case for nucleotides without Watson–Crick base-pairing in loops. Adenosines with a smaller cyan box lacked ¹³C statistics of C2 due to insufficient chemical shift data. The results of the automated chemical shift assignment by FLYA are color-coded: green and magenta boxes indicate correct and incorrect assignments, respectively. Full boxes correspond to strong (self-consistent) FLYA assignments, empty boxes to weak (tentative) assignments.