RNA Secondary and Tertiary Structure

A single strand of RNA, released from its complementary partner, does not flop about as a random coil. Instead, it folds back on itself, pairing with its own sequence. This builds hairpins, bulges, and elaborate three-dimensional scaffolds that can catalyze chemistry. The 76-nucleotide tRNA molecule, drawn flat, looks like a cloverleaf. Folded in three dimensions, it snaps into a compact L shape. That shape is held together by base pairs its two-dimensional diagram never shows. This is the central surprise of RNA biochemistry. Because RNA is single-stranded, its own sequence is both message and mate. Evolution has exploited that fact to make RNA a molecule that stores information and performs mechanical and catalytic work. The reason this matters extends far beyond textbook curiosity. Ribosomes are ribozymes — RNA machines that catalyze peptide bond formation using a folded rRNA active site. Riboswitches fold differently depending on whether a small metabolite is bound. This switches gene expression on or off without any protein intermediary. The catalytic core of the spliceosome depends on precise secondary and tertiary folds. So do telomerase's RNA template and CRISPR guide RNAs. None of this would be possible if RNA behaved like double-stranded DNA. DNA is a molecule whose job is to stay boring and uniform. RNA's structural versatility is a direct chemical consequence of two features. These are absent, or nearly absent, from DNA. One is the reactive 2′-OH group on every ribose. The other is the tolerance of RNA helices for a wide non-canonical pairing vocabulary. This goes beyond Watson–Crick A–U and G–C. The big picture is a hierarchy familiar from protein folding but with its own chemical grammar. Primary structure is the nucleotide sequence. Secondary structure is the set of base pairs. These are Watson–Crick pairs organized into helices, interrupted by loops, bulges, and junctions. They can be predicted largely from sequence complementarity and thermodynamic stability. Tertiary structure is the actual three-dimensional arrangement of those helices in space. It is stabilized by long-range interactions: coaxial stacking, tetraloop–receptor docking, ribose zippers, and metal ion bridges. Most critical among these is Mg²⁺. It screens the intense negative charge of the phosphodiester backbone folded into close-packed helices. Non-canonical base pairs are not chemical accidents. These include G•U wobble pairs, Hoogsteen pairs, and the elaborate pairing catalog described by the Leontis–Westhof classification. They are structural elements as essential to RNA architecture as α-helices and β-sheets are to protein architecture. This article works through RNA folding from the bottom up. It covers the chemistry that makes RNA foldable in the first place. This includes the 2′-OH and its consequences for both flexibility and instability. It covers the rules of secondary structure prediction via base-pairing thermodynamics. It covers the catalog of loop and junction motifs. It covers the tertiary interactions that pack these motifs into functional three-dimensional machines. It connects directly to Base Pairing and the Chemistry of Complementarity for the pairing chemistry itself. It connects to The Phosphodiester Backbone for the covalent chemistry of the chain. It sets up the discussion in Types and Functions of Cellular RNA of how these folds become tRNAs, rRNAs, and ribozymes.

Key takeaways

RNA folds by intramolecular Watson–Crick and non-canonical base pairing into helices separated by loops, bulges, and junctions — this is secondary structure.
The 2′-OH group makes the ribose backbone flexible and catalytically useful but also chemically labile, since it can attack the adjacent phosphate in-line.
Free energy of folding is approximately additive over stacked base pairs (nearest-neighbor model); loops and bulges cost free energy, driving hairpin formation.
G•U wobble pairs and other non-Watson–Crick pairs (Hoogsteen, sheared G•A) expand RNA's structural vocabulary well beyond DNA's.
Tertiary structure packs secondary-structure helices via coaxial stacking, tetraloop–receptor contacts, and Mg²⁺-mediated electrostatic screening of the backbone.
Ribozymes and the ribosome show that a correctly folded RNA tertiary structure can position substrates and stabilize transition states like a protein enzyme.
Key idea: RNA structure is a hierarchy — sequence dictates pairing (secondary structure) via nearest-neighbor thermodynamics, and pairing patterns then pack into 3D tertiary folds via a small, reusable catalog of non-canonical contacts.

Where this lives today

RNA secondary and tertiary structure prediction is now a working tool in molecular medicine and biotechnology, not merely a structural curiosity. Messenger RNA vaccines (the COVID-19 mRNA vaccines among them) are engineered with modified secondary structures — optimized 5′ and 3′ untranslated regions and codon-optimized coding sequences chosen partly to avoid stable secondary structures that would impede ribosome scanning and translation efficiency, while stabilizing structures elsewhere protect the mRNA from nuclease attack via exactly the 2′-OH chemistry described above. Riboswitches, folded RNA elements found in bacterial mRNA 5′ leaders that change secondary structure upon binding a metabolite (thiamine pyrophosphate, adenosylcobalamin, guanine), are an active target for a new class of antibiotics designed to lock the riboswitch in the wrong conformation and shut down essential bacterial gene expression without a protein target at all. CRISPR-Cas9 guide RNAs rely on a precisely folded secondary structure (a hairpin scaffold binding the Cas9 protein plus a single-stranded spacer for target recognition) engineered by researchers using exactly the nearest-neighbor free-energy calculations derived here. Structural biologists use cryo-EM and X-ray crystallography to solve ribosomal, spliceosomal, and viral RNA structures at atomic resolution, routinely finding new instances of the same recurring tertiary motifs — GNRA tetraloops docking into tetraloop receptors, kink-turns, ribose zippers — first cataloged in tRNA and group I intron structures decades ago. Computational RNA structure prediction tools (Mfold, RNAfold, and their successors) implement the nearest-neighbor dynamic-programming algorithm sketched here directly, now extended with machine-learning corrections trained on the ever-growing database of solved RNA crystal and cryo-EM structures.

Ask the tutor: How does SHAPE chemical probing experimentally determine which nucleotides in an RNA are base-paired versus flexible/single-stranded?

The physical picture

Consider a single strand of RNA synthesized 5′ to 3′, released from the polymerase, floating in solution with Mg²⁺ and monovalent ions present. Every phosphate carries a full negative charge at physiological pH; the strand is a linear polyanion. If the sequence contains a self-complementary stretch — say 5′-GGGAAAACCC-3′, where the 5′ GGG can Watson–Crick pair with the 3′ CCC — the strand can fold back on itself, forming three base pairs and leaving the AAAA as an unpaired loop connecting the two paired strands. This is a hairpin, the single most common element of RNA secondary structure. The paired region is a short A-form double helix (RNA cannot form B-form helices; the 2′-OH group sterically forces the sugar pucker and helix geometry into the A-family, wider and shorter-pitched than B-DNA). The unpaired region is the loop. Real RNAs contain many such hairpins connected by unpaired single strands, bulges (unpaired nucleotides on one side of an otherwise continuous helix), internal loops (unpaired nucleotides on both sides), and multi-helix junctions (three or more helices radiating from one point, as at the center of a tRNA cloverleaf). The full set of base pairs, drawn as a two-dimensional diagram, is the secondary structure. Two nucleotides that are hundreds of positions apart in the linear sequence can still be adjacent in the folded structure and pair directly — a pseudoknot — which breaks the simple nested-parentheses picture of secondary structure and is often classified as an early step toward tertiary structure. Two chemical features distinguish RNA folding from DNA folding. First, the 2′-OH group of ribose sits in the minor groove of the A-form helix, available for hydrogen bonding to bases, other riboses, or bound metal ions — the chemical basis of nearly all RNA tertiary contacts. Second, that same 2′-OH is a latent nucleophile: positioned correctly, its oxygen can attack the adjacent 3′-phosphate in an intramolecular in-line SN2-like transesterification, cleaving the backbone and leaving a 2′,3′-cyclic phosphate. This reaction is the chemical basis of both spontaneous RNA degradation (RNA is far less hydrolytically stable than DNA, which lacks the 2′-OH) and of deliberate ribozyme catalysis, as in the hammerhead and hairpin ribozymes, which fold their 2′-OH into a precisely tuned in-line geometry. ```svg ``` ```svg ```

The core ideas

RNA secondary structure prediction rests on treating the free energy of a folded structure as a sum of local, sequence-dependent contributions — an approximation that works remarkably well because base-pair stacking energies are dominated by nearest-neighbor interactions.

\Delta G^{\circ}_{\text{fold}} = \sum_{i} \Delta G^{\circ}_{\text{stack}}(i) + \sum_{j} \Delta G^{\circ}_{\text{loop}}(j)

where:

$\Delta G^{\circ}_{\text{stack}}(i)$ is the free energy of the $i$ -th adjacent base-pair stack (a nearest-neighbor parameter, roughly $-1$ to $-3.5$ kcal/mol, more negative for GC-rich stacks),
$\Delta G^{\circ}_{\text{loop}}(j)$ is the (positive, destabilizing) free energy penalty for the $j$ -th loop, bulge, or junction, dominated by the conformational entropy lost when a single strand is constrained into a closed loop,
the sum runs over every stacked pair and every loop in the proposed structure.

Reading: folding is favorable when helix stacking (favorable, enthalpically driven by stacking and hydrogen bonding) outweighs the entropic and steric cost of the loops needed to connect the helices. This is why hairpin loops have a minimum size (loops shorter than three nucleotides cannot geometrically close a helix without severe strain) and why GC-rich stems are dramatically more stable than AU-rich stems of the same length. The second governing relationship converts these free energies into an equilibrium constant for helix formation, exactly as in any bimolecular (or here, intramolecular) association:

K_{\text{eq}} = \frac{[\text{folded}]}{[\text{unfolded}]} = \exp\!\left(-\frac{\Delta G^{\circ}_{\text{fold}}}{RT}\right)

where:

$R = 1.987 \times 10^{-3}$ kcal mol⁻¹ K⁻¹,
$T$ is absolute temperature (K),
$\Delta G^{\circ}_{\text{fold}}$ is the total free energy of forming the structure relative to the unfolded single strand, from the nearest-neighbor sum above.

Reading: because $\Delta G^{\circ}_{\text{fold}}$ is a sum of many small stacking terms, folding is highly cooperative — adding one more base pair to a growing helix both extends the pairing and adds a new favorable stack, so helices tend to zip up or unravel as a unit (a two-state-like transition for short hairpins), while whole secondary structures compete against each other by the Boltzmann weight of their total free energies.

Derivation ANearest-neighbor free energy of a hairpin and its melting temperature

Goal. Derive an expression for the melting temperature of an RNA hairpin from its nearest-neighbor stacking free energies and loop penalty, then evaluate it for a concrete asymmetric hairpin.

Step 1 — Decompose the structure. A hairpin with an $n$ -base-pair stem has $n-1$ stacked nearest-neighbor pairs (the terminal pair does not stack onto anything further) plus one hairpin-loop closing penalty. Write the total standard folding free energy as

\Delta G^{\circ}_{37} = \sum_{k=1}^{n-1} \Delta G^{\circ}_{\text{NN}}(k) + \Delta G^{\circ}_{\text{loop}}(m)

where $\Delta G^{\circ}_{\text{NN}}(k)$ is the tabulated nearest-neighbor free energy (kcal/mol, 37 °C, 1 M NaCl reference conditions) for the $k$ -th stacked pair of base pairs, and $\Delta G^{\circ}_{\text{loop}}(m)$ is the loop-initiation penalty for a loop of $m$ unpaired nucleotides.

Step 2 — Enthalpy–entropy split. Each nearest-neighbor free energy is itself measured calorimetrically as $\Delta G^{\circ}_{\text{NN}} = \Delta H^{\circ} - T\Delta S^{\circ}$ at a reference $T = 310.15$ K (37 °C). Summing over the whole structure,

\Delta G^{\circ}_{\text{fold}}(T) = \Delta H^{\circ}_{\text{tot}} - T\,\Delta S^{\circ}_{\text{tot}}

with $\Delta H^{\circ}_{\text{tot}}$ and $\Delta S^{\circ}_{\text{tot}}$ obtained by summing the tabulated enthalpies and entropies of every stack and the loop term.

Step 3 — Melting condition. The hairpin is a unimolecular (intramolecular) equilibrium, so at the melting temperature $T_m$ , defined as where folded and unfolded populations are equal, $K_{\text{eq}} = 1$ , hence $\Delta G^{\circ}_{\text{fold}}(T_m) = 0$ :

0 = \Delta H^{\circ}_{\text{tot}} - T_m\,\Delta S^{\circ}_{\text{tot}} \quad\Longrightarrow\quad T_m = \frac{\Delta H^{\circ}_{\text{tot}}}{\Delta S^{\circ}_{\text{tot}}}

Step 4 — Worked numerical evaluation. Take an asymmetric 4-base-pair stem with sequence 5′-GCAU…AUGC-3′ closing a tetraloop, giving three stacked nearest-neighbor steps: GC/CG, CA/UG, AU/AU (illustrative composite values chosen to be non-degenerate, in the range of real Turner-rule parameters):

Stack 1 (GC on CG): $\Delta H_1 = -3.3$ kcal/mol, $\Delta S_1 = -8.5$ cal mol⁻¹K⁻¹
Stack 2 (CA on UG, i.e. a CA/GU-type step): $\Delta H_2 = -2.1$ kcal/mol, $\Delta S_2 = -6.1$ cal mol⁻¹K⁻¹
Stack 3 (AU on AU): $\Delta H_3 = -1.1$ kcal/mol, $\Delta S_3 = -2.7$ cal mol⁻¹K⁻¹
Tetraloop closing penalty: $\Delta H_{\text{loop}} = +5.4$ kcal/mol, $\Delta S_{\text{loop}} = +13.0$ cal mol⁻¹K⁻¹ (destabilizing enthalpically and favorable-looking in isolated entropy sign only because it is dwarfed by the loss of configurational freedom already counted in the reference state; net loop $\Delta G$ at 310 K works out positive, as required)

Sum the enthalpies:

\Delta H^{\circ}_{\text{tot}} = (-3.3) + (-2.1) + (-1.1) + 5.4 = -1.1\ \text{kcal/mol}

Sum the entropies:

\Delta S^{\circ}_{\text{tot}} = (-8.5) + (-6.1) + (-2.7) + 13.0 = -4.3\ \text{cal mol}^{-1}\text{K}^{-1}

Then:

T_m = \frac{\Delta H^{\circ}_{\text{tot}}}{\Delta S^{\circ}_{\text{tot}}} = \frac{-1100\ \text{cal/mol}}{-4.3\ \text{cal mol}^{-1}\text{K}^{-1}} = 256\ \text{K}

This value (below room temperature) signals correctly that this particular short, AU-rich four-base-pair hairpin is only marginally stable — consistent with the chemical intuition that four base pairs dominated by weak AU and mixed steps, closed by an entropically costly loop, will not form a robust structure at 37 °C. A longer or more GC-rich stem pushes $\Delta H^{\circ}_{\text{tot}}$ strongly negative while $\Delta S^{\circ}_{\text{tot}}$ grows only linearly with stack number, raising $T_m$ well above physiological temperature — exactly the trend seen in real GC-rich terminator hairpins, whose $T_m$ values comfortably exceed 350 K.

Limiting check. As the number of stacks $n-1 \to 0$ (no stem at all, an unstructured loop only), $\Delta G^{\circ}_{\text{fold}} \to \Delta G^{\circ}_{\text{loop}} > 0$ for all $T$ , so $K_{\text{eq}} < 1$ always — no stable fold forms, matching the physical expectation that a "hairpin" with zero base pairs is simply a floppy single strand.

Go deeper · Derivation B · 2 minEffect of a single G•U wobble substitution on helix stability

Derivation BEffect of a single G•U wobble substitution on helix stability

Goal. Show how replacing one Watson–Crick pair with a G•U wobble pair changes the free energy of a helix, using the additive nearest-neighbor framework, and connect this to why G•U pairs are tolerated in RNA helices but essentially forbidden as steady structural elements in duplex DNA.

Step 1 — Geometry of the wobble pair. A G•U wobble pair forms between the O6/N1-H face of guanine and the O2/N3-H face of uracil, related to Watson–Crick G–C pairing by a small (~2.2 Å) lateral shift of one base relative to the other. The pair is still roughly isomorphic with a Watson–Crick pair — it fits into an A-form helix with only minor backbone distortion — which is why RNA polymerases and ribosomes tolerate it, whereas the DNA double helix (constrained by proofreading polymerases discriminating on exactly this geometry) essentially never retains G•U pairs in the genome.

Step 2 — Free energy penalty relative to Watson–Crick. A G•U pair contributes only two hydrogen bonds arranged asymmetrically (compared to three for G–C), and the nearest-neighbor stacking terms for steps containing G•U are measured to be less favorable (less negative $\Delta G^{\circ}$ ) than the corresponding Watson–Crick steps. Represent the substitution's effect as a free-energy difference:

\Delta \Delta G^{\circ}_{\text{sub}} = \Delta G^{\circ}_{\text{helix, with G•U}} - \Delta G^{\circ}_{\text{helix, all WC}}

Step 3 — Numeric estimate using representative nearest-neighbor increments. Suppose an 8-base-pair helix has $\Delta G^{\circ}_{37,\text{all-WC}} = -14.6$ kcal/mol (all Watson–Crick stacks). Replacing the central G–C pair with a G•U wobble changes the two flanking stacking steps from Watson–Crick-type stacks (average $-2.4$ kcal/mol each) to G•U-containing stacks (average $-1.3$ kcal/mol each, a representative destabilization drawn from measured Turner parameters for G•U-containing steps):

\Delta \Delta G^{\circ}_{\text{sub}} = 2 \times \big[(-1.3) - (-2.4)\big] = 2 \times (1.1) = +2.2\ \text{kcal/mol}

\Delta G^{\circ}_{37,\text{with G•U}} = -14.6 + 2.2 = -12.4\ \text{kcal/mol}

Step 4 — Convert to a stability ratio. The ratio of folded-to-unfolded populations changes by

\frac{K_{\text{eq}}(\text{with G•U})}{K_{\text{eq}}(\text{all-WC})} = \exp\!\left(-\frac{\Delta\Delta G^{\circ}_{\text{sub}}}{RT}\right) = \exp\!\left(-\frac{2.2}{(1.987\times10^{-3})(310)}\right)

= \exp(-3.57) = 0.028

A single wobble substitution lowers the equilibrium constant for the folded state by roughly 35-fold at 37 °C — destabilizing, but far from prohibitive; the helix still folds robustly if the remaining seven pairs are strong. This is exactly the functional compromise exploited in the genetic code: position 34 of the tRNA anticodon uses G•U wobble pairing with the third codon position, allowing one tRNA to decode two synonymous codons, at the cost of a slightly less stable (but still perfectly functional) codon–anticodon minihelix.

Limiting check. As the helix becomes very long ( $n \to \infty$ base pairs, all else Watson–Crick), a fixed $+2.2$ kcal/mol penalty from one internal wobble becomes a vanishingly small fraction of the total $\Delta G^{\circ}_{\text{fold}}$ , and $K_{\text{eq}}(\text{with G•U})/K_{\text{eq}}(\text{all-WC}) \to$ a constant set only by that local penalty, not by helix length — confirming that the nearest-neighbor model correctly localizes the wobble's destabilizing effect to its immediate stacking neighbors rather than propagating it indefinitely.

Worked examples

Example 1 (foundational): Predicting whether a short palindromic RNA sequence forms a hairpin. Consider the 14-nucleotide RNA 5′-GGCAC-UUUU-GUGCC-3′ (dashes added only for readability). Check the first 5 nucleotides against the last 5, reading the 3′ end backward: GGCAC pairs with GUGCC read 3′→5′ as CCGUG, i.e. compare 5′-GGCAC-3′ to 3′-CCGUG-5′. Pairing position by position:

G (pos 1) with C (pos 14): Watson–Crick G–C ✓
G (pos 2) with C (pos 13): Watson–Crick G–C ✓
C (pos 3) with G (pos 12): Watson–Crick C–G ✓
A (pos 4) with U (pos 11): Watson–Crick A–U ✓
C (pos 5) with G (pos 10): Watson–Crick C–G ✓

All five positions pair, leaving UUUU (positions 6–9) as a four-nucleotide loop — a stable tetraloop-sized closure. This is a textbook hairpin: 5 bp stem, 4-nt loop. Using representative nearest-neighbor stacking values (GC-rich stem, four stacks: GG/CC, GC/GC, CA/UG, AC/GU, each roughly $-2.5$ to $-3.3$ kcal/mol) and a UUUU tetraloop penalty of roughly $+4.5$ kcal/mol (tetraloop penalties are smaller than generic loop penalties because UUUU and related loops make stabilizing loop-closing contacts):

\Delta G^{\circ}_{37} \approx (-3.3) + (-2.2) + (-2.1) + (-2.6) + 4.5 = -5.7\ \text{kcal/mol}

A negative $\Delta G^{\circ}_{37}$ of several kcal/mol confirms this hairpin folds robustly at physiological temperature, consistent with $K_{\text{eq}} = \exp(5.7/0.6156) \approx \exp(9.26) \approx 1.0\times10^{4}$ , i.e. the folded state outnumbers the unfolded state roughly ten-thousand to one at equilibrium.

Example 2 (applied): Why the 2′-OH makes RNA far more susceptible to alkaline hydrolysis than DNA, quantitatively. DNA lacks a 2′-OH and is stable for years in mild alkaline conditions; RNA under the same conditions (0.1 M NaOH, room temperature) is fully degraded to a mixture of 2′- and 3′-mononucleotides within minutes to hours. The mechanism: base removes the proton from the 2′-OH, generating a 2′-alkoxide that is perfectly positioned — roughly 3 Å from, and in-line with, the adjacent phosphorus — to perform an intramolecular in-line nucleophilic attack on the phosphodiester phosphorus, displacing the 5′-oxygen of the next nucleotide as the leaving group and forming a pentacovalent phosphorane transition state that collapses to a 2′,3′-cyclic phosphate plus a free 5′-OH on the downstream fragment. This is chemically an intramolecular transesterification, not a hydrolysis in the strict sense (water only enters in a second, slower step that opens the cyclic phosphate to a mixture of 2′- and 3′-monophosphates). Because the reaction is intramolecular and geometrically pre-organized, it is enormously accelerated relative to an equivalent intermolecular attack — the same "effective molarity" argument used throughout enzymology, where a favorable ring-closure geometry can be worth many kcal/mol of rate enhancement. Because DNA's 2′ position instead carries only H (no OH, no lone pair, no acidic proton to deprotonate), this pathway is chemically unavailable, and DNA's backbone survives base treatment that would obliterate RNA within minutes. This single hydroxyl group is the entire chemical reason RNA is used as a disposable messenger while DNA serves as the permanent, hydrolytically robust genetic archive — and it is the same chemistry harnessed catalytically by RNase A, the hammerhead ribozyme, and the self-cleaving elements this article's chemistry connects directly to in The Chemistry of DNA Hydrolysis and Strand Cleavage by contrast.

Example 3 (exam/competition level): Predicting the minimum free energy secondary structure of a pseudoknot-forming sequence and evaluating a competing alternative fold. Consider a 20-nucleotide RNA capable of forming either of two mutually exclusive secondary structures:

Structure A (simple hairpin): a 6-bp stem (three GC pairs, three AU pairs, arranged GC/GC/AU stacking then AU/AU/GC) closing a 8-nucleotide loop.

Structure B (pseudoknot-like alternative, treated here as two shorter competing hairpins sharing sequence): two 3-bp stems (one all-GC, one all-AU) each closing smaller loops, using an overlapping but different pairing pattern of the same 20 nucleotides. Estimate representative nearest-neighbor sums: Structure A: five stacking steps in a mixed 6-bp stem, sum $\Delta H^{\circ}_{\text{stack}} \approx -3.4-3.1-2.9-2.0-1.8 = -13.2$ kcal/mol (illustrative composite values), entropy sum $\Delta S^{\circ}_{\text{stack}} \approx -9.5-8.9-8.3-6.0-5.4 = -38.1$ cal mol⁻¹K⁻¹; loop penalty for an 8-nt hairpin loop, $\Delta H_{\text{loop}} \approx +6.4$ kcal/mol, $\Delta S_{\text{loop}} \approx +14.0$ cal mol⁻¹K⁻¹.

\Delta H^{\circ}_{A} = -13.2 + 6.4 = -6.8\ \text{kcal/mol}, \qquad \Delta S^{\circ}_{A} = -38.1 + 14.0 = -24.1\ \text{cal mol}^{-1}\text{K}^{-1}

\Delta G^{\circ}_{A}(310\ \text{K}) = -6.8 - (310)(-0.0241) = -6.8 + 7.47 = +0.67\ \text{kcal/mol}

Structure B: two independent short stems, all-GC stem (2 stacks, $\Delta H \approx -3.3-3.3=-6.6$ , $\Delta S\approx -8.5-8.5=-17.0$ ) plus all-AU stem (2 stacks, $\Delta H\approx-1.1-1.1=-2.2$ , $\Delta S\approx-2.7-2.7=-5.4$ ), plus two separate loop penalties (each smaller, say $\Delta H_{\text{loop}}\approx+4.5$ , $\Delta S_{\text{loop}}\approx+10.0$ each, so doubled: $+9.0$ kcal/mol and $+20.0$ cal/mol·K):

\Delta H^{\circ}_{B} = (-6.6) + (-2.2) + 9.0 = 0.2\ \text{kcal/mol}

\Delta S^{\circ}_{B} = (-17.0)+(-5.4)+20.0 = -2.4\ \text{cal mol}^{-1}\text{K}^{-1}

\Delta G^{\circ}_{B}(310\ \text{K}) = 0.2 - (310)(-0.0024) = 0.2 + 0.74 = +0.94\ \text{kcal/mol}

Both structures come out only marginally stable (positive $\Delta G^{\circ}$ , meaning the fully unfolded strand is actually slightly favored for these particular illustrative parameters) — but Structure A is more favorable than Structure B by

\Delta G^{\circ}_{B} - \Delta G^{\circ}_{A} = 0.94 - 0.67 = +0.27\ \text{kcal/mol}

so the minimum-free-energy algorithm used by real structure-prediction programs (Zuker-style dynamic programming, an extension of exactly this nearest-neighbor additivity) would report Structure A as the predicted fold, with Structure B populated as a minor competing conformer at equilibrium in a ratio

\frac{[B]}{[A]} = \exp\!\left(-\frac{0.27}{0.6156}\right) = \exp(-0.44) = 0.64

meaning roughly 39% of molecules would actually populate Structure B at equilibrium ( $0.64/(1+0.64) \approx 0.39$ ) — a realistic illustration of the conformational heterogeneity real RNAs display, and exactly why some regulatory RNAs (riboswitches, and the pseudoknots that trigger ribosomal frameshifting) are functionally poised near such near-degenerate alternative folds, allowing a small perturbation (metabolite binding, temperature, an interacting protein) to shift the populated ensemble.

Limiting cases, scope & common traps

Situation	What happens	Why
Very short hairpin loop (< 3 nt)	Cannot form — geometrically forbidden	The backbone cannot span the helix diameter with fewer than 3 unpaired residues without prohibitive strain
GC-rich stem, long	High $T_m$ , very robust fold, dominates ensemble	Each GC stack contributes strongly negative $\Delta G^{\circ}$ ; sum grows roughly linearly with stem length
Single internal G•U wobble	Modest destabilization (few kcal/mol), structure still forms	Only 2 H-bonds and weaker stacking, but near-isomorphic geometry preserves helix
Two competing folds of similar $\Delta G^{\circ}$	Population splits between both; sensitive to small perturbations	Boltzmann weighting; near-degenerate states are populated proportionally to $\exp(-\Delta G/RT)$
High Mg²⁺ concentration	Tertiary folds stabilize, compact 3D structure forms	Divalent cations screen phosphate backbone charge, permitting close helix packing
No Mg²⁺, high monovalent salt only	Secondary structure intact, tertiary structure collapses or fails to form	Monovalent ions screen charge less effectively at close range than divalent Mg²⁺
RNA vs. DNA under alkaline conditions	RNA degrades rapidly; DNA is stable	2′-OH provides in-line nucleophile for backbone self-cleavage; DNA lacks this group

Common trapMisconception: secondary structure diagrams show the "real shape" of RNA

A cloverleaf or hairpin diagram is a topological map of base pairs, not a picture of three-dimensional geometry. The tRNA cloverleaf, drawn flat with four arms splayed out, actually folds into a compact L-shaped tertiary structure in which the acceptor stem and the anticodon stem are roughly perpendicular and stacked coaxially with other helices. Two helices adjacent in the secondary-structure diagram are not necessarily close in space, and two nucleotides far apart in the linear diagram (as in a pseudoknot, or in tertiary contacts like tetraloop–receptor docking) can be immediately adjacent in the folded molecule. Secondary structure is necessary information for predicting tertiary structure, but it substantially underdetermines it.

Common trapMisconception: non-Watson–Crick pairs are folding "errors"

G•U wobble pairs, sheared G•A pairs, and Hoogsteen A•U pairs are not mistakes tolerated by a sloppy molecule — they are functional structural elements with their own consensus geometries, cataloged systematically (the Leontis–Westhof nomenclature classifies base pairs by which edge of each base — Watson–Crick, Hoogsteen, or sugar edge — participates, and by strand orientation). These non-canonical pairs recur in conserved positions across evolutionarily unrelated RNAs (the GNRA tetraloop, the sarcin-ricin loop, the kink-turn motif) precisely because they provide reproducible three-dimensional geometries — sharp turns, splayed conformations, or ion-binding pockets — unavailable to Watson–Crick pairing alone.

The chemist behind it

Alexander Rich, working at MIT from the 1960s through the 2000s, was central to establishing that RNA molecules fold into precise, ordered three-dimensional structures rather than remaining floppy single strands. His laboratory solved the first crystal structure of a transfer RNA (yeast phenylalanine tRNA) in 1974, revealing the L-shaped tertiary fold built from stacked helical domains, non-canonical base pairs, and a compact tertiary core stabilized by Mg²⁺ ions and specific base-triple interactions between the D-loop and T-loop — the first direct visualization that a "cloverleaf" secondary structure diagram was merely a schematic, and that the actual molecule packed itself into a shape as ordered as any globular protein. Rich's structural work, alongside the parallel contributions of Sung-Hou Kim's laboratory on the same tRNA structure, established the vocabulary — tertiary base triples, coaxial stacking, the L-shaped fold — still used to describe every RNA structure solved since. His later work on Z-DNA and on RNA editing enzymes extended the same conviction that nucleic acid structure, not just sequence, carries essential biological information — a principle that became foundational once the discovery of catalytic RNA (ribozymes, recognized by Cech and Altman in the early 1980s) showed that RNA tertiary structure could do the work previously thought to require protein enzymes entirely.

Origins · 1 minAlexander Rich and how this was discovered

Alexander Rich

1924–2015 · Structural biology of nucleic acids, MIT, 1950s–2000s

Rich trained as a physician before turning to structural biology, working early with Linus Pauling before establishing his own laboratory focused on nucleic acid crystallography. Beyond the landmark 1974 tRNA structure, he characterized left-handed Z-DNA, triple-helical nucleic acid structures, and RNA–protein and RNA editing complexes, consistently pursuing the idea that nucleic acids are structurally dynamic molecules whose three-dimensional shapes are as functionally important as their sequences.

History

- 1965 — Robert Holley determines the complete sequence of yeast alanine tRNA and proposes the cloverleaf secondary structure.

- 1974 — Alexander Rich's and Sung-Hou Kim's groups independently solve the crystal structure of yeast phenylalanine tRNA, revealing the L-shaped tertiary fold.

- 1981–1982 — Thomas Cech discovers self-splicing group I introns; Sidney Altman shows RNase P's RNA subunit is catalytic — RNA can be an enzyme (a "ribozyme").

- 1996–2000 — Crystal structures of the hammerhead and hairpin ribozymes reveal the in-line 2′-OH attack geometry directly.

- 2000 — The Ramakrishnan, Steitz, and Yonath laboratories solve ribosomal subunit structures, showing the peptidyl transferase center is built entirely of RNA — the ribosome itself is a ribozyme.

What is the difference between RNA secondary structure and tertiary structure?

Secondary structure is the complete set of base pairs — which nucleotide pairs with which — organized into helices, loops, bulges, and junctions, and it can largely be predicted from sequence using nearest-neighbor thermodynamics. Tertiary structure is the actual three-dimensional arrangement of those helices and loops in space: how helices pack against each other, which loops dock into which receptors, and where metal ions and long-range non-canonical contacts hold the overall fold together. A useful analogy is protein structure — secondary structure is like α-helices and β-strands, tertiary structure is how those elements pack into the final globular fold. Secondary structure diagrams are two-dimensional topological maps; tertiary structure is the real molecule.

Why does the 2'-OH group make RNA less chemically stable than DNA?

The 2′-hydroxyl sits close to and in-line with the adjacent 3′-phosphodiester bond. Under basic conditions (or with the help of general acid–base catalysis by a ribozyme or ribonuclease active site), the 2′-oxygen is deprotonated and attacks the adjacent phosphorus intramolecularly, forming a pentacoordinate transition state that collapses to cleave the backbone, releasing a 2′,3′-cyclic phosphate. Because this pathway is available on every single ribonucleotide in the chain and is intramolecular (hence fast, favored by good geometry), RNA is intrinsically far more prone to spontaneous strand scission than DNA, which lacks the 2′-OH entirely and therefore lacks this pathway.

What is a G•U wobble pair and why does RNA tolerate it but DNA does not?

A G•U wobble pair forms when guanine's O6/N1-H face hydrogen-bonds to uracil's O2/N3-H face, held together by two hydrogen bonds in a geometry shifted slightly from ideal Watson–Crick pairing but still close enough to fit into an A-form helix with only minor distortion. RNA polymerases, ribosomes, and folded RNA structures readily accommodate G•U pairs — they appear throughout tRNA anticodon-codon wobble pairing and rRNA helices. DNA, by contrast, is under constant proofreading surveillance by high-fidelity polymerases that discriminate against non-Watson–Crick geometries, and G•U mismatches (which would arise from a G•T-equivalent in DNA) are actively repaired, so they essentially never persist as stable structural elements in the genome.

What are hairpins, bulges, internal loops, and junctions?

A hairpin (or stem-loop) is a helical stem formed when a single strand folds back and pairs with itself, closed by an unpaired loop. A bulge is a small number of unpaired nucleotides on only one strand of an otherwise continuous helix, causing a kink in the helical axis. An internal loop has unpaired nucleotides on both strands facing each other. A multi-helix junction (three-way, four-way, etc.) occurs where three or more helices radiate from a single unpaired or minimally paired region, as at the center of the tRNA cloverleaf where four helical arms converge. Each of these motifs has characteristic thermodynamic penalties and, in real folded RNAs, characteristic preferred geometries stabilized by tertiary contacts.

How does Mg²⁺ stabilize RNA tertiary structure?

The phosphodiester backbone carries one full negative charge per nucleotide, so packing multiple helices close together in a compact tertiary fold requires overcoming enormous electrostatic repulsion between backbones. Mg²⁺ ions, carrying two positive charges and able to approach the backbone more closely than monovalent cations (which are less efficient at screening due to weaker charge density and entropic penalties of ion condensation), neutralize this repulsion far more effectively per ion than Na⁺ or K⁺. Many folded RNAs also have specific, structurally resolved Mg²⁺ binding sites — often at helix junctions or deep pockets — where an inner-sphere or outer-sphere coordinated Mg²⁺ ion bridges phosphate oxygens from different helices, directly gluing the tertiary fold together rather than merely screening charge diffusely.

Can RNA structures be predicted purely from sequence, and how reliable is this?

Secondary structure prediction from sequence alone, using nearest-neighbor thermodynamic parameters and dynamic-programming algorithms (as in Mfold and RNAfold), is quite reliable for simple, single-domain structures without pseudoknots, typically recovering the majority of true base pairs correctly for well-behaved RNAs. It becomes substantially harder for RNAs with pseudoknots (which violate the simple nested-pairing assumption most algorithms rely on), for RNAs whose folding is co-transcriptional and kinetically trapped rather than at true thermodynamic equilibrium, and for tertiary structure, which nearest-neighbor secondary-structure thermodynamics alone cannot predict — tertiary structure prediction generally requires additional information from covariation analysis, chemical probing (SHAPE), or direct structural methods (X-ray crystallography, cryo-EM, NMR).

What is a pseudoknot and why is it treated separately from ordinary secondary structure?

A pseudoknot occurs when nucleotides within a loop of one hairpin pair with nucleotides outside that hairpin, elsewhere in the sequence, creating a base-pairing pattern that cannot be drawn as simple non-crossing nested arcs. Standard secondary-structure prediction algorithms are built on the assumption that base pairs are non-crossing (nested, like balanced parentheses), which makes efficient dynamic programming possible; pseudoknots violate this assumption and require more computationally expensive algorithms or specialized heuristics to predict. Biologically, pseudoknots are functionally important — they occur in ribosomal frameshifting signals in many viral RNAs, in telomerase RNA, and in some riboswitches — and are often considered an intermediate step between pure secondary structure and full tertiary structure, since they introduce genuine three-dimensional constraints.