High-throughput dissection of the thermodynamic and conformational properties of a ubiquitous class of RNA tertiary contact motifs

Abstract

Despite RNA's diverse secondary and tertiary structures and its complex conformational changes, nature utilizes a limited set of structural "motifs"-helices, junctions, and tertiary contact modules-to build diverse functional RNAs. Thus, in-depth descriptions of a relatively small universe of RNA motifs may lead to predictive models of RNA tertiary conformational landscapes. Motifs may have different properties depending on sequence and secondary structure, giving rise to subclasses that expand the universe of RNA building blocks. Yet we know very little about motif subclasses, given the challenges in mapping conformational properties in high throughput. Previously, we used "RNA on a massively parallel array" (RNA-MaP), a quantitative, high-throughput technique, to study thousands of helices and two-way junctions. Here, we adapt RNA-MaP to study the thermodynamic and conformational properties of tetraloop/tetraloop receptor (TL/TLR) tertiary contact motifs, analyzing 1,493 TLR sequences from different classes. Clustering analyses revealed variability in TL specificity, stability, and conformational behavior. Nevertheless, natural GAAA/11ntR TL/TLRs, while varying in tertiary stability by ∼2.5 kcal/mol, exhibited conserved TL specificity and conformational properties. Thus, RNAs may tune stability without altering the overall structure of these TL/TLRs. Furthermore, their stability correlated with natural frequency, suggesting thermodynamics as the dominant selection pressure. In contrast, other TL/TLRs displayed heterogenous conformational behavior and appear to not be under strong thermodynamic selection. Our results build toward a generalizable model of RNA-folding thermodynamics based on the properties of isolated motifs, and our characterized TL/TLR library can be used to engineer RNAs with predictable thermodynamic and conformational behavior.

Keywords: RNA folding; RNA nanotechnology; RNA structure; high-throughput biochemistry; tertiary motifs.

Fig. 1.

Modular RNA structure and energetics. (A) Tertiary structure of the canonical GAAA/11ntR TL/TLR. Crystal structures of GAAA/11ntR from RNase P RNA (Top Left; Protein Data Bank [PDB] 1NBS), P4-P6 domain (Top Right; PDB 1GID), and Azoarcus group I intron (Bottom; PDB 1ZZN), superimposed using the PyMOL molecular graphics system (version 2.0 Schrödinger, LLC). (B) Reconstitution Model of RNA folding (13). Folding free energy (∆G_fold) is decomposed into an internal conformational search (∆G_align) and the formation of stabilizing tertiary interactions (∆G_tert). ∆G_align can be further decomposed into a contribution from electrostatics and ion interactions (∆G_+/-) and a contribution from the conformational properties of helices, junctions, and tertiary contact motifs comprising the RNA (∆G_HJH; HJH stands for helix-junction-helix). Thus, in this model, the conformational ensemble of the motifs is dictated by properties intrinsic to the motif such as sequence and topology, while environmental factors such as ions affect the relative stabilities of the conformations within the ensemble.

Fig. 2.

High-throughput studies of RNA structural motifs. (A) Tertiary structure of tectoRNA dimer. Structure shown is that of homodimer with two identical GAAA/11ntR_wt TL/TLRs (49). One of the GAAA/11ntR_wt is replaced by an orthogonal GGAA/R1 TL/TLR in our studies to ensure heterodimer formation (28). (B) Pathways for the formation of bound tectoRNA dimer. After the formation of one of the tertiary contacts, an internal conformational search aligns the second TL to its cognate TLR. (C) High-throughput RNA-MaP platform used to characterize RNA structural motifs. Tens of thousands of “chip” piece sequences are transcribed from sequenced DNA on an Illumina chip (37). Stars represent fluorescent probes. Junctions and/or short helical elements are inserted into the helical domain of the chip piece (26). (D) Sequence, secondary, and tertiary structure of representative TL/TLRs belonging to different types. Tertiary structures are from GCGA TL docking into CUG-CAG (Protein Data Bank [PDB] 3IGI), GGAA/11ntR_wt (PDB 1GID), and GAAA/Vc2_wt (PDB 3IRW). The GAAA/C7.2_wt structure was modeled computationally (50). (E) Sequence of IC3_wt TLR motif. (F) Schematic for characterization of the TL/TLR library via RNA-MaP. Varying TLR sequences were inserted in place of the 11ntR_wt sequence in the chip piece. Scaffold variants for the insertion of each TLR sequence were generated by altering the length, sequence, and/or secondary structure of the helical domain of the chip piece. (G) Examples of scaffolds of different lengths. The length of the chip piece is defined as the number of bp between the GGAA TL and the TLR, including canonical and noncanonical (i.e., mismatches) bp. Residues without opposing residues in complementary strand (e.g., bulges) do not contribute to the reported length. (H–J) Distribution of TLR sequences in the library. Natural 11ntR and 12ntR TLR variants were obtained from databases and published sequence alignments of functional RNAs (–42).

Fig. 3.

Thermodynamic fingerprints report on TL/TLR 3D conformational properties. (A) Structural changes to the scaffold affect ∆G_bind by altering the internal conformational search to align the TL/TLRs. The figure shows the schematic comparing the internal conformational search undergone by a short scaffold (9 bp, Top) and a long scaffold (11 bp, Bottom). The large gray circle represents the conformational space explored by a particular scaffold during the internal conformational search, and the smaller darker gray area represents the part of this space where the TL and TLR are “correctly” aligned for the formation of the TL/TLR. In this simplified schematic (with equal probabilities throughout the 2D gray space), the fraction of conformational space, leading to productive formation of the TL/TLR, is greater for the 9 bp than for the 11-bp scaffold. Thus, the 9-bp scaffold has a greater probability of aligning the TL/TLR, resulting in higher affinity compared to the 11-bp scaffold. In actual cases, the relative occupancy across the available conformations also affects the probability of binding with each scaffold. (B) Measurements of ∆G_bind across distinct scaffolds produce thermodynamic fingerprints that report on the alignment properties of the TL/TLRs. Each datapoint corresponds to a distinct scaffold varying in length, secondary structure, and/or sequence. Depending on its conformational properties, each scaffold has a distinct probability of aligning the TL/TLRs, resulting in the distinct ∆G_bind values. (C) Differences between TL/TLR variants can affect ∆G_bind by affecting the strength of the tertiary interactions and/or the alignment preferences. Cartoons represent three distinct TL/TLR sequences embedded within the same scaffold. TL/TLR1 and TL/TLR2 have the same alignment preferences but differ in the strength of their tertiary interactions. In this case, the difference in stability between the TL/TLRs is independent of the structural context in which the comparison is made. TL/TLR3 has different alignment preferences compared to the other two TL/TLRs. In this case, the difference in stability between TL/TL3 and the other TL/TLRs depends on the structural context in which the comparison is made. (D) TL/TLRs with the same alignment preferences produce thermodynamic fingerprints with the same shape. The strength differences between TL/TLR1 and TL/TLR2 are expressed by a constant offset in their stability (∆∆G_bind). (E) TL/TLRs that differ in alignment preferences produce thermodynamic fingerprints with different shapes, such that ∆∆G_bind depends on the scaffold identity.

Fig. 4.

Thermodynamic and conformational properties of representative TL/TLRs across scaffolds. (A) ∆G_bind for TLRs binding to GAAA TL across scaffolds. Dashed gray line indicates the threshold of –7.1 kcal/mol. Open symbols are ∆G_bind above this threshold. Dashed colored lines serve as guides for the thermodynamic fingerprints. To compare the fingerprints, they were superimposed to that of GAAA/11ntR_wt by shifting them by a constant ∆G_bind while minimizing rmsd. This minimal rmsd is shown as is the correlation constant, r. Values above threshold were not used to calculate the correlation coefficient (r). ∆G_avg is the median. Values of r for TLRs with ∆G_avg > –7.1 kcal/mol are not reliable and therefore were not considered. Solution conditions: 89 mM Tris-Borate, pH 8.0, 30 mM MgCl₂, 0.01 mg/mL yeast tRNA, and 0.01% Tween 20. Measurements at different ionic conditions are provided in SI Appendix, Fig. S3. (B) ∆∆G_bind relative to GAAA/11ntR_wt, calculated as the differences in ∆G_bind showed in A. Open symbols correspond values below stability threshold and are lower limits. (C) ∆G_bind for GAAA versus GUAA TLs for each representative TLR across its scaffolds. Dashed gray lines indicate a threshold of –7.1 kcal/mol, as above. ∆∆G_avg is the average difference in stability between binding to GUAA and GAAA in units of kcal/mol (e.g., on average, constructs with GAAA/11ntR_wt are 3.55 kcal/mol more stable than constructs with GUAA/11ntR_wt).

Fig. 5.

Overview of TLR thermodynamic behavior. (A) Secondary structure and sequence of five scaffolds common to the entire TLR library. (B) Distribution of ∆G_avg for TLR sequences across five common scaffolds in A and two TL sequences. TLRs with ∆G_avg > –7.1 kcal/mol (n_weak = 387) were not included in the clustering analysis. (C) Fraction of the variance represented by each of the PCs. (D) Hierarchical clustering of the first two PCs using Ward’s method implemented in Python. (E) Hierarchical clustering of TLRs across five common scaffolds and TL sequences GAAA and GUAA. Prior to clustering, ∆G_bind values for each of the TLRs was scaled by ∆G_mean to produce ∆∆G_rel values. The ∆∆G_rel values were used in a PCA (C above), and the first two PCs were hierarchically clustered to generate the clustergrams shown. (F) Distribution of TLR mutants according to type (i.e., 11ntR, IC3, and C7.2) across clusters. (G and H) Sequence motifs representing variability among 11ntR (G) and IC3 (H) variants in the individual subgroups revealed by hierarchical clustering. Sequence motifs were generated using Weblogo (51). Subgroups 1 and 2 were mostly composed of 11ntR variants and subgroup 5 mostly of IC3 variants. (I) PCA loading plot showing the influence of the sequence of the binding TL and the secondary structure of the scaffold on the first two PCs. (J) Pseudofingerprints showing the average behavior of the TLR subgroups across TL sequence and scaffold secondary structure.

Fig. 6.

Thermodynamic analysis of natural GAAA/11ntR variants. (A) Average relative stability and Pearson correlation coefficients of the fingerprints. ∆∆G_avg values are relative to GAAA/11ntR_wt. Larger symbols are for variants that were compared across 50 scaffolds and smaller symbols for those compared across five scaffolds. The red dashed line represents the decrease in correlation expected for theoretically identical fingerprints from error as stability decreases (SI Appendix, Fig. S2). (B and C) Correlation between relative stability in 30 mM MgCl₂ (B) or in 5 mM MgCl₂/150 mM KCl (C) and the frequency of TLR variants in natural RNA sequences.