RNA Folding Nearest Neighbor Parameters Including the Modification 1-Methyl-Pseudouridine

This study introduces new nearest neighbor parameters derived from 208 optical melting experiments to accurately model the stabilizing effects of 1-methyl-pseudouridine on RNA folding, which are now integrated into the RNAstructure software to improve the prediction of secondary structures for both natural and therapeutic mRNA sequences.

The Gist

Imagine you are trying to build a complex origami crane out of a long strip of paper. The way the paper folds depends on how the different parts stick to each other. In the world of biology, this "paper" is RNA, and the "stickiness" is determined by how its chemical building blocks pair up.

For decades, scientists have had a "rulebook" (called the Turner Rules) to predict exactly how RNA will fold. This rulebook tells them how much energy it takes to snap two pieces together or how much energy is lost when a loop forms. Knowing this is crucial because the shape of the RNA determines what it does—whether it helps make proteins, regulates genes, or acts as a vaccine.

However, there was a missing piece in this rulebook.

The Missing Piece: The "Magic Glue"

In recent years, scientists discovered a special chemical tweak called 1-methyl-pseudouridine (1mΨ). Think of this as a "magic glue" added to the RNA.

• Why it matters: This magic glue is the secret ingredient in modern mRNA vaccines (like the ones for COVID-19). It stops the body's immune system from panicking and attacking the vaccine, allowing the vaccine to do its job.
• The Problem: While we knew this glue worked, we didn't know how it changed the folding rules. Does it make the RNA fold tighter? Looser? Does it change the shape of the loops? Without knowing the exact "energy cost" of using this magic glue, scientists were guessing when trying to design better vaccines or RNA therapies.

The Experiment: The "Tug-of-War"

To fix this, the researchers in this paper acted like master architects. They built hundreds of tiny, simple RNA models in the lab.

1. The Setup: They created pairs of RNA strands. Some had the standard building blocks (Uridine), and others had the "magic glue" (1mΨ) in specific spots.
2. The Test: They heated these strands up in a process called optical melting. Imagine slowly heating a piece of paper until the glue holding it together melts and the paper falls apart. By measuring exactly how much heat it took to break the bond, they could calculate the "stickiness" (stability) of that specific pair.
3. The Scale: They did this 208 times with different combinations to get a massive amount of data. They even sent samples to three different labs to make sure their results were consistent, like having three different judges score a gymnastics routine to ensure fairness.

The Discovery: "It's Stronger, But Context Matters"

The results were fascinating. They found that swapping the standard block for the "magic glue" generally makes the RNA more stable (it holds together tighter).

However, it's not a simple "add 10 points of strength" rule. It's more like a social network:

• If the magic glue is standing next to a specific neighbor (like an Adenine), it makes the bond super strong.
• If it's next to a different neighbor (like a Guanine), the boost is different.
• Sometimes, if it's in a "loop" (a bend in the chain), it stabilizes the bend in a unique way.

The team used all this data to write a new, updated rulebook. They calculated exactly how much energy is saved or spent for every possible combination of the magic glue with its neighbors.

The Payoff: Better Vaccines and Therapies

Why does this matter to you?

1. Better Vaccines: Now, when scientists design an mRNA vaccine, they can use a computer program (called RNAstructure) with these new rules. Instead of guessing, the computer can predict the perfect shape for the vaccine RNA. A more stable shape means the vaccine lasts longer in the body and produces more protein, making it more effective.
2. tRNA Fixes: They tested this on tRNA (the workers that build proteins in our cells). Many tRNAs naturally have this magic glue. Using the old rules, the computer predicted the wrong shape for these tRNAs. With the new rules, the predictions became much more accurate, helping us understand how cells work at a deeper level.

The Bottom Line

Think of this paper as the team that finally wrote the instruction manual for the "Magic Glue." Before, we knew the glue existed and that it was useful, but we didn't know the physics behind it. Now, we have the precise numbers to calculate exactly how it works. This allows engineers (scientists) to build stronger, smarter, and more effective RNA-based medicines and vaccines.

Technical Summary

1. Problem Statement

RNA secondary structure prediction relies heavily on thermodynamic nearest neighbor parameters (Turner rules) to estimate the Gibbs free energy change ($\Delta G^\circ_{37}$) of folding. While these parameters are well-established for canonical nucleotides (A, C, G, U), they lack accuracy for sequences containing modified bases.

1-methyl-pseudouridine (1m$\Psi$) is a critical modification:

• Natural Occurrence: Found in 18S rRNA and archaeal tRNA.
• Therapeutic Importance: It is widely used in mRNA vaccines and therapeutics (fully substituting for uridine) to suppress innate immune responses and enhance translation.
• The Gap: The contribution of 1m$\Psi$ to RNA folding stability was previously unclear. Existing models either ignored the modification or used parameters derived from limited data, leading to inaccurate predictions of secondary structures for therapeutic mRNAs and natural tRNAs. Specifically, 1m$\Psi$ can form canonical base pairs with Adenine (A) and Guanine (G), creating a complex parameter space (34 helical terms) that requires extensive experimental data to resolve.

2. Methodology

The authors developed a comprehensive set of nearest neighbor parameters for 1m$\Psi$ using a rigorous experimental and computational approach.

• Experimental Data Generation:

  • Scope: A total of 224 optical melting experiments were performed.
  • Split: 208 experiments were used to fit the parameters, and 16 independent experiments were reserved for validation.
  • Sites: Data was integrated from three laboratories (University of Rochester, DNA Software Inc., and the Institute of Bioorganic Chemistry in Poland) to ensure reproducibility.
  • Calibration: Three calibration duplexes were measured across all sites, showing an average uncertainty of 2.1% in $\Delta G^\circ_{37}$, validating the integration of multi-site data.

• Parameter Fitting Strategy:

  • The study utilized the Turner 2004 functional form as the baseline, updating it with recent G-U stack parameters.
  • Fitting Order: Parameters were fitted sequentially to manage dependencies:
    1. Helical Stacks: 81 fully base-paired duplexes were used to fit 34 nearest neighbor parameters involving 1m$\Psi$-A and 1m$\Psi$-G pairs.
    2. Terminal Effects: Dangling ends and terminal mismatches were fitted using 20 and 40 model systems, respectively.
    3. Loops: Hairpin loops (8 systems), internal loops (47 systems), and bulge loops (4 systems) were analyzed to determine stability contributions of 1m$\Psi$ in unpaired regions.
  • Extrapolation: For unmeasured motifs, the authors calculated the difference ($\Delta\Delta G$) between 1m$\Psi$-containing motifs and their canonical U-analogs, applying these shifts to existing Turner 2004 values.

• Validation:

  • The new parameters were tested against the 16 reserved optical melting experiments.
  • tRNA Modeling: The parameters were applied to predict the folding ensembles of 55 tRNA sequences from the Sprinzl database containing 1m$\Psi$. Metrics included Normalized Ensemble Defect (NED), sensitivity, and positive predictive value (PPV).

3. Key Contributions

• Comprehensive Parameter Set: The first complete set of nearest neighbor parameters for RNA folding that includes 1m$\Psi$, covering helical stacks, dangling ends, terminal mismatches, hairpins, internal loops, and bulge loops.
• Software Integration: The parameters have been integrated into the RNAstructure software package. They are available in two alphabets:
  1. Hybrid: A, C, G, U, and 1m$\Psi$ (for sequences with mixed modifications).
  2. Full1: A, C, G, and 1m$\Psi$ (optimized for mRNA vaccines where U is fully substituted by 1m$\Psi$).
• Refined Thermodynamic Rules: The study established that 1m$\Psi$ generally stabilizes RNA structures compared to U, but the magnitude of stabilization is highly sequence-dependent.

4. Key Results

• Helical Stability:
  • Substitution of U with 1m$\Psi$ generally stabilizes RNA folding.
  • 1m$\Psi$-A pairs: Average stabilization of -0.32 ± 0.17 kcal/mol compared to U-A.
  • 1m$\Psi$-G pairs: Average stabilization of -0.27 ± 0.48 kcal/mol compared to U-G.
  • Terminal Pairs: Unlike canonical U-A pairs (which incur a 0.45 kcal/mol penalty), terminal 1m$\Psi$-A and 1m$\Psi$-G pairs require no stability penalty.
• Loop and Mismatch Stability:
  • Dangling Ends: 3' dangling ends adjacent to 1m$\Psi$-A pairs are significantly more stabilizing (-0.43 kcal/mol on average) than those adjacent to U-A.
  • Internal Loops: 1m$\Psi$-1m$\Psi$ and 1m$\Psi$-U first mismatches provide additional stabilization (-0.92 and -1.41 kcal/mol, respectively) compared to U-U mismatches.
  • Hairpins: The Turner 2004 model for hairpins works well for 1m$\Psi$, with specific bonuses applied for 1m$\Psi$-U first mismatches similar to U-U bonuses.
• Model Accuracy:
  • Helical RMSD: The root mean squared deviation (RMSD) between estimated and measured stability was 0.40 kcal/mol for the fitting set and 0.68 kcal/mol (excluding outliers) for the validation set.
  • tRNA Prediction: Using the new parameters significantly improved tRNA structure prediction:
    • NED: Improved from 0.267 to 0.252 ($P=3.0\times10^{-6}$).
    • Sensitivity: Improved from 86.0% to 87.5% ($P=0.026$).
  • Case Study: For Haloferax volcanii Asp tRNA, using 1m$\Psi$ parameters increased prediction sensitivity from 63.2% to 89.5%.

5. Significance

• mRNA Therapeutics: The ability to accurately model the secondary structure of mRNA fully substituted with 1m$\Psi$ is crucial for vaccine design. Stable structures can protect mRNA from degradation and optimize translation efficiency. These parameters allow for the rational design of mRNA sequences that balance immune evasion with structural stability.
• Fundamental Biology: The parameters improve the understanding of how natural modifications in tRNA and rRNA influence folding and function, providing a more accurate tool for studying epitranscriptomics.
• RNA Nanotechnology: The expanded alphabet (A, C, G, 1m$\Psi$) enables the design of more complex and stable RNA nanostructures by leveraging the specific stabilizing effects of 1m$\Psi$ in loops and mismatches.
• Reproducibility Standard: The study establishes a protocol for cross-site optical melting reproducibility, recommending specific calibration duplexes for future thermodynamic studies.

In summary, this work bridges a critical gap in RNA thermodynamics, providing the community with the necessary tools to accurately predict and engineer the structures of 1m$\Psi$-modified RNAs, which are central to modern mRNA vaccine technology.