SHAPE-directed RNA secondary structure prediction

Abstract

The diverse functional roles of RNA are determined by its underlying structure. Accurate and comprehensive knowledge of RNA structure would inform a broader understanding of RNA biology and facilitate exploiting RNA as a biotechnological tool and therapeutic target. Determining the pattern of base pairing, or secondary structure, of RNA is a first step in these endeavors. Advances in experimental, computational, and comparative analysis approaches for analyzing secondary structure have yielded accurate structures for many small RNAs, but only a few large (>500 nts) RNAs. In addition, most current methods for determining a secondary structure require considerable effort, analytical expertise, and technical ingenuity. In this review, we outline an efficient strategy for developing accurate secondary structure models for RNAs of arbitrary length. This approach melds structural information obtained using SHAPE chemistry with structure prediction using nearest-neighbor rules and the dynamic programming algorithm implemented in the RNAstructure program. Prediction accuracies reach >or=95% for RNAs on the kilobase scale. This approach facilitates both development of new models and refinement of existing RNA structure models, which we illustrate using the Gag-Pol frameshift element in an HIV-1 M-group genome. Most promisingly, integrated experimental and computational refinement brings closer the ultimate goal of efficiently and accurately establishing the secondary structure for any RNA sequence.

Figure 1

Overview of the SHAPE experimental and data analysis steps. Adapted from ref. [60].

Figure 2

Overview of the steps involved in processing capillary electrophoresis data, obtaining normalized SHAPE reactivities, and calculating experimentally-informed RNA secondary structure models [11, 30, 39, 60].

Figure 3

Base pair prediction sensitivities for E. coli 23S rRNA for a range of slope (m) and intercept (b) values (Eqn. 1). Optimal values of m = 2.6 kcal/mol and b = −0.6 kcal/mol are depicted by the white box. Adapted from ref. [30].

Figure 4

Summary of thermodynamic and SHAPE-derived free energy change contributions for a simple HIV-1 hairpin (NL4-3 nucleotides 594 – 626) [41]. Favorable nearest-neighbor stacking and unfavorable loop thermodynamic terms are shown in green and red, respectively. The total nearest neighbor free energy change ΔG_NN is the sum over all these contributions. ΔG_SHAPE pseudo-free energy change terms are shown for base paired (black) and non-base paired (gray) nucleotides; only base paired values are included in the net free energy change. The ΔG_SHAPE term is added once for each nucleotide at the ends of helices and twice for interior nucleotides (blue symbols). The ΔG_SHAPE calculations used m = 3.0 kcal/mol and b = −0.6 kcal/mol. The total folding free energy change, ΔG_total, is the sum of nearest neighbor and SHAPE-derived contributions.

Figure 5

RNA secondary structure models for the HIV-1 M-group Gag-Pol frameshift element. All models are colored by their SHAPE reactivities as reported in [41] using the scale shown in Figure 4. The “slippery sequence” where frameshifting occurs is enclosed in a blue box. The numbering is for the NL4-3 reference sequence. (A) Classical model [69]. (B) Two-stem model [80]. (C) SHAPE-supported model [41].