SHAPE-directed discovery of potent shRNA inhibitors of HIV-1

Abstract

The RNA interference (RNAi) pathway can be exploited using short hairpin RNAs (shRNAs) to durably inactivate pathogenic genes. Prediction of optimal target sites is notoriously inaccurate and current approaches applied to HIV-1 show weak correlations with virus inhibition. In contrast, using a high-content model for disrupting pre-existing intramolecular structure in the HIV-1 RNA, as achievable using high-resolution SHAPE (selective 2'-hydroxyl acylation analyzed by primer extension) chemical probing information, we discovered strong correlations between inhibition of HIV-1 production in a quantitative cell-based assay and very simple thermodynamic features in the target RNA. Strongest inhibition occurs at RNA target sites that both have an accessible "seed region" and, unexpectedly, are structurally accessible in a newly identified downstream flanking sequence. We then used these simple rules to create a new set of shRNAs and achieved inhibition of HIV-1 production of 90% or greater for up to 82% of designed shRNAs. These shRNAs inhibit HIV-1 replication in therapy-relevant T cells and show no or low cytotoxicity. The remarkable success of this straightforward SHAPE-based approach emphasizes that RNAi is governed, in significant part, by very simple, predictable rules reflecting the underlying RNA structure and illustrates principles likely to prove broadly useful in understanding transcriptome-scale biological recognition and therapeutics involving RNA.

Figure 1

The guide strand-target RNA interaction equilibrium. RISC Argonaute protein is represented by an orange oval, although only RNA–RNA thermodynamics were considered in this study. RISC, RNA-induced silencing complex.

Figure 2

Relative levels of virus production for five shRNAs. shRNA-encoding plasmids were transfected into 293T cells in amounts ranging from 1 to 125 ng per well. Each shRNA is labeled by the first nucleotide position of its binding site on the NL4-3 HIV-1 messenger RNA (see Supplementary Table S1).

Figure 3

HIV-1 genome locations of the 84 target sequences for the shRNAs in the ter Brake et al. dataset used to derive design rules (top) and of the 26 sequences, designed by the rules defined in this work (bottom). Numbers indicate the position of the 5′ nucleotide of each shRNA target site in the HIV-1 NL4-3 messenger RNA.

Figure 4

Correlation coefficients (r) between calculated target folding energies, ΔG_target, and experimental activity values for the 84 shRNAs in the training dataset. (a) Linear r-values as a function of target unfolding window size (vertical axis) and position (horizontal axis). Colors denote the relative strength of the correlation. A white box highlights the strongest correlation. (b) Optimal accessible target window. Window positions are numbered relative to the 5′ end of the guide strand binding site on the target RNA. The window size denotes the length of the window extending downstream of the window position value. The accessible target window that yielded the strongest correlation is shown by a gray rectangle.

Figure 5

Relative viral production versus (a) target folding energies, ΔG_target, and (b) total binding energies, ΔG_total, for the 84 shRNAs in the training dataset. The optimal target unfolding window, identified in Figure 4, was used for ΔG_target calculations. The SHAPE-directed HIV-1 secondary structure model was used to calculate ΔG_target. The same correlations for (c) ΔG_target and (d) ΔG_total calculated without using experimental constraints to estimate the secondary structure.

Figure 6

Prediction success rates of designed shRNAs. (a) Inhibition of HIV-1 production by shRNAs in the test set. Target sequences were chosen to have accessible 13-nucleotide windows (ΔG_target ≥0 kcal/mol) and strong overall binding energy ΔG_total. Prediction success rates are shown for ΔG_total criteria of -25, -27, and -29 kcal/mol; n, the numbers of shRNAs meeting a given threshold. (b) Inhibition of HIV-1 production by shRNAs in the 84-member training set. (c) Prediction success rates for 16 shRNAs that failed the target accessibility criterion. ΔG_target <0 corresponds to RNA target sites with some pre-existing structure.