RNA Structure and Why It Matters

Before you start

You know what RNA is and the central dogma. New? See the biology a coder needs.
This is the first lesson of the advanced RNA track. No coding required here, but later lessons assume comfort with the command line and RNA-seq.
Any term new? The Glossary has it.

Learning objectives

By the end of this lesson you will be able to: describe how single-stranded RNA folds into secondary and tertiary structure, explain why structure is a functional regulatory layer rather than decoration, outline how tools predict structure from sequence, and recognize how experiments measure structure in living cells.

RNA is a shape, not just a sequence

DNA is famously a stable double helix held by two separate strands. RNA is usually single-stranded, which changes everything: a single RNA molecule folds back on itself, and its own bases pair up internally. Adenine pairs with uracil, guanine with cytosine, and RNA also tolerates the weaker G–U "wobble" pair, which DNA does not use. Those internal pairings build the molecule's secondary structure: stems (paired regions), hairpin loops, internal loops, bulges, and junctions.

Stack the secondary structure into three dimensions and you get the tertiary structure, the actual folded shape, sometimes locked together by long-range contacts called pseudoknots. The classic mental image is the cloverleaf of a tRNA folding into an L shape, but every mRNA has structure too, especially in its untranslated regions (UTRs).

Decode the jargonSecondary structure

The set of base pairs an RNA forms with itself: which nucleotide pairs with which. It is usually drawn in 2D (stems and loops) and is the level most computational tools predict. The full 3D fold is the tertiary structure.

Why structure is functional, not decorative

Structure is a regulatory layer that sits on top of sequence. A few of the places it decides biology:

Non-coding RNAs are structure. tRNAs, rRNAs, and the spliceosome's snRNAs do their jobs because of precise folds, not coding content.
Riboswitches are structured segments in the UTRs of some mRNAs that change shape when they bind a small metabolite, switching the gene on or off without any protein involved.
Translation control. A stable hairpin in the 5' UTR can block the ribosome from scanning, lowering translation; an internal ribosome entry site (IRES) is a structure that recruits the ribosome directly.
Stability and localization. Structured elements in the 3' UTR are recognized by RNA-binding proteins that set how long the message survives and where in the cell it goes.
Splicing. Local structure around splice sites can hide or expose them, tuning which exons are included.

So if you only read an mRNA as a sequence, you are missing a whole control system. That is exactly why this track exists.

Predicting structure from sequence (the computational part)

Given just the sequence, can we predict how it folds? The dominant approach is thermodynamic: every possible base pair and stacked helix has a known free-energy contribution, and the algorithm searches for the structure with the lowest total free energy, the minimum free energy (MFE) structure, because the most stable fold is the one most likely to form. This search is solved efficiently with dynamic programming (the Zuker algorithm, 1981).

The standard, free, open-source toolkit is the ViennaRNA package (its RNAfold program); Mfold/UNAFold is the other classic. Output is often written in dot-bracket notation, where a dot is an unpaired base and matching brackets are a base pair:

# sequence
GGGAAACCC
# predicted secondary structure (dot-bracket): a 3-bp stem closing a loop
(((...)))

Decode the jargonMinimum free energy (MFE)

The single predicted structure with the lowest folding free energy, taken as the most thermodynamically stable, and therefore the most probable, fold. Reported in kcal/mol: more negative means more stable.

⚠️ The MFE structure is a prediction, not the truth

Real RNA does not always sit in its single lowest-energy state. It samples an ensemble of structures, folds while it is still being made (co-transcriptionally), is shaped by bound proteins and the cellular environment, and can form pseudoknots that most fast predictors ignore. Treat a predicted MFE structure as a hypothesis, then test it.

Measuring structure in real cells (the experimental part)

Because prediction is imperfect, the field measures structure directly using chemical probing followed by sequencing. The idea: use a chemical that reacts only with unpaired, flexible nucleotides, then read out where it reacted.

SHAPE reagents acylate the 2'-hydroxyl of flexible (single-stranded) nucleotides regardless of base, marking them.
DMS methylates unpaired adenines and cytosines specifically.

Sequencing then reports a per-nucleotide "reactivity": high reactivity = likely unpaired, low = likely paired. Transcriptome-wide, in-cell versions such as SHAPE-seq and icSHAPE let you map structure across thousands of RNAs in living cells, and those reactivities can be fed back in as constraints to make the computational prediction far more accurate.

The bridge to the rest of this track

Structure connects directly to the two big themes ahead. RNA-binding proteins often recognize a shape, not just a motif, which is what CLIP-seq maps (next lesson). And chemical modifications change structure: the modification m6A weakens base pairing and can act as a structural switch that exposes or hides protein-binding sites (Liu et al., 2015). So the shape you just learned about is itself regulated by the epitranscriptomic marks you will study later in this track.

Check your understanding

An RNAfold prediction gives a single MFE structure with ΔG = −22 kcal/mol. A colleague treats this as "the" structure of the RNA in the cell. What is the most accurate critique?

Correct. MFE prediction returns the single lowest-energy fold, but real RNA occupies a structural ensemble and is influenced by co-transcriptional folding, bound proteins, the cellular milieu, and modifications. The prediction is a strong hypothesis to test (e.g., with SHAPE), not ground truth.

In a SHAPE experiment, a stretch of nucleotides shows consistently HIGH reactivity. What does that most likely indicate about those positions?

Right. SHAPE reagents react with the 2'-OH of flexible, unpaired nucleotides. High reactivity marks single-stranded, accessible regions; paired bases in helices are constrained and react weakly. These reactivities can constrain and improve structure prediction.

A 5' UTR contains a very stable hairpin just upstream of the start codon. Based on this lesson, what functional consequence is most plausible?

Exactly. Stable secondary structure in the 5' UTR can block the scanning ribosome and lower translation efficiency. Structured UTR elements (hairpins, IRESs, riboswitches) are a major post-transcriptional control layer, which is the whole point of studying RNA structure.

In dot-bracket notation for an RNA secondary structure, what do matching parentheses such as ( and ) represent?

Correct. In dot-bracket notation a dot is an unpaired base, and a matched pair of parentheses marks two nucleotides that pair together (for example in a stem). It is just a compact text encoding of the secondary structure.

Why is a G-U wobble pair worth knowing about when you reason about RNA folding?

Right. RNA tolerates the non-Watson-Crick G-U wobble pair, which is common and reasonably stable inside helices. Folding algorithms include it, so real stems often contain G-U pairs alongside A-U and G-C pairs.

Sources & further reading

Zuker M, Stiegler P. Optimal computer folding of large RNA sequences using thermodynamics and auxiliary information. Nucleic Acids Res, 1981. (The dynamic-programming MFE algorithm.)
Lorenz R, et al. ViennaRNA Package 2.0. Algorithms Mol Biol, 2011. ViennaRNA website
Spitale RC, et al. Structural imprints in vivo decode RNA regulatory mechanisms (icSHAPE). Nature, 2015.
Liu N, et al. N6-methyladenosine-dependent RNA structural switches regulate RNA–protein interactions. Nature, 2015. (m6A as a structural switch.)

Last reviewed: June 2026.

Next in Track 4

RNA Structure and Why It Matters Computationally