Analysis of natural structures and chemical mapping data reveals local stability compensation in RNA

Abstract

RNA molecules adopt complex structures that perform essential biological functions across all forms of life, making them promising candidates for therapeutic applications. However, our ability to design new RNA structures remains limited by an incomplete understanding of their folding principles. While global metrics such as the minimum free energy are widely used, they are at odds with naturally occurring structures and incompatible with established design rules. Here, we introduce local stability compensation (LSC), a principle that RNA folding is governed by the local balance between destabilizing loops and their stabilizing adjacent stems, challenging the focus on global energetic optimization. Analysis of over 100 000 RNA structures revealed that LSC signatures are particularly pronounced in bulges and their adjacent stems, with distinct patterns across different RNA families that align with their biological functions. To validate LSC experimentally, we systematically analyzed thousands of RNA variants using DMS chemical mapping. Our results demonstrate that stem folding, as measured by reactivity, correlates with LSC (R² = 0.458 for hairpin loops) and that instabilities show no significant effect on folding for distal stems. These findings demonstrate that LSC can be a guiding principle for understanding RNA function and for the rational design of custom RNAs.

Graphical Abstract

Figure 1.

Local stability compensation in bpRNA-1m90. (A) Hairpin, (B) bulge, and (C) internal loop ΔGs from refolded bpRNA-1m90 structures are binned, and the corresponding stem ΔGs are plotted as violins with linear regressions drawn for the median (blue) and 5% quantile (red). bpRNA-1m90 net free energy kernel density estimates for (D) hairpin loops, (E) bulges, and (F) internal loops belonging to different RNA types.

Figure 2.

Net ΔG distributions are distinct for RNA types and particular RNA loops. (A) C4 antisense RNA example secondary structure and (B) hairpin net ΔG distribution (H1 in blue, H2 in red) and other hairpins in bpRNA-1m90 (gray). (C) TwoAYGGAY RNA example secondary structure and (D) net ΔG distribution (H1 in blue, H2 in red) and other hairpins in bpRNA-1m90 (gray). (E) Example tRNA secondary structure (bpRNA_RFAM_1424) and (F) two-dimensional box plots for the D-loop, T-loop, and anticodon loop in red, green, and blue, respectively. (G) pre-miRNA example secondary structure and (H) net ΔG distribution (blue) and other hairpins in bpRNA-1m90 (gray).

Figure 3.

Differences in net ΔG by phylogeny for select RNA types. Plant pre-miRNA hairpin net ΔGs (green) and metazoan (gray) net ΔGs (A) before folding with constraints and (B) after folding with constraints. (C) Plant pre-miRNA bulge ΔGs (green) and metazoan (gray) net ΔGs after folding with constraints. (D) Plant pre-miRNA internal loop ΔGs (green) and metazoan (gray) net ΔGs after folding with constraints. (E) Thermophilic (orange) hairpin net ΔGs and non-thermophilic net ΔGs (blue). (F) Thermophilic (orange) bulge net ΔGs and non-thermophilic net ΔGs (blue). (G) Thermophilic (orange) internal loop net ΔGs and non-thermophilic net ΔGs (blue).

Figure 4.

Library design and local structure AUROC heatmaps. (A) Designed structure examples for hairpins, bulges, internal loops (top, middle, and bottom, respectively); dashed lines indicate the variable region. Average AUROC and structure count heatmaps for cells of (B) hairpin, (C) bulge, and (D) internal loop ΔG by adjacent stem ΔG; yellow cells show low AUROC or poor agreement of DMS data with the designed structure, while dark blue cells show high AUROC with strong agreement, and black cells contain no data. Count heatmaps show the number of RNAs contributing to each cell with darker red coloring indicating a higher number of RNAs.

Figure 5.

Local versus distal stem reactivity correlations. (A) Example structure from the hairpin library with adjacent and distal stems to the hairpin loop labeled and local stem base pair indices labeled. (B) Linear regression between hairpin net ΔG and log average local stem reactivity, compared to (C) average distal stem reactivity. (D) Average net ΔG bin reactivity per position away from the hairpin loop. (E) Example structure from the bulge library with adjacent and distal stems to the bulge labeled and local stem base pair indices labeled. (F) Linear regression between bulge net ΔG and log average local stem reactivity, compared to (G) average distal stem reactivity. (H) Average net ΔG bin reactivity per position away from the bulge, where an index is present on both stems; the average is used. (I) Example structure from the internal loop library with adjacent and distal stems to the internal loop labeled and local stem base pair indices labeled. (J) Linear regression between internal loop net ΔG and log average local stem reactivity, compared to (K) average distal stem reactivity. (L) Average net ΔG bin reactivity per position away from the internal loop, where an index is present on both stems; the average is used.

Figure 6.

Local structure fidelity curve fitting. (A) Hairpin net ΔG versus average local AUROC; datapoints indicated by larger black, magenta, and cyan dots are shown in panel (B); raw reactivity values color-coded by nucleotide for three example designed hairpins; the boxed region is involved in AUROC calculation. (C–E) Hill equation curves fit to net ΔG bins (0.2 kcal/mol) versus the mean AUROC for each bin.