# Chapter 54. RNA Folding in Cells, Chaperones, Helicases, and Kinetic Traps

## Scope Note

This chapter explains how RNA molecules fold in living cells, why cellular folding differs from purified in vitro folding, and how proteins, ions, ligands, transcription, crowding, and energy-consuming remodelers reshape the RNA folding landscape. The central theme is kinetic control: many RNA molecules do not simply relax to a single minimum-free-energy structure, because nascent synthesis, binding partners, local concentration, compartment, and remodeling enzymes determine which structures are sampled, stabilized, disrupted, or avoided.

## Executive Summary

RNA folding in cells is a kinetic and environmental process, not a simple calculation of the lowest-energy structure from sequence. Purified RNA can reveal intrinsic base-pairing, tertiary contacts, ion dependence, and misfolding pathways, but living cells add transcriptional directionality, RNA-binding proteins, chaperones, helicases, ligand binding, modification, degradation, compartmentalization, and macromolecular crowding. A cellular RNA molecule therefore samples a landscape whose routes and endpoints depend on when each segment emerges, which partners bind first, and whether ATP-dependent remodeling can reset an unfavorable structure.

Co-transcriptional folding is especially important because RNA is synthesized from the 5′ end toward the 3′ end. A base pair that forms early can either nucleate the correct fold or sequester sequence that later needs to pair elsewhere. This creates kinetic partitioning: one transcript population may enter the functional pathway, another may enter a metastable nonproductive pathway, and still another may be degraded or held for remodeling. Riboswitches, HIV-1 TAR, bacterial terminators, splice-site regions, and preribosomal RNAs illustrate how local folding decisions can change biological output.

RNA chaperones and RNA helicases solve different but overlapping problems. RNA chaperones lower the barrier between alternative RNA structures or promote annealing and strand exchange without necessarily consuming ATP. RNA helicases, especially DEAD-box and related DExD/H-box proteins, bind RNA and nucleotides to unwind, clamp, destabilize, anneal, or remodel RNA-protein assemblies. Many helicases are not processive motors on long duplexes; instead, they act locally, often in defined RNP assembly pathways such as ribosome biogenesis, splicing, translation initiation, RNA export, decay, antiviral defense, and stress responses. Reviews of RNA helicase biology emphasize that disease phenotypes often reflect pathway-specific remodeling defects rather than a generic loss of RNA unwinding.

Cellular folding also depends on proteins, ligands, ions, chemical modifications, and crowding. Magnesium ions and monovalent salts screen RNA charge and stabilize compact structures, but cellular magnesium availability is buffered and locally heterogeneous. Proteins can stabilize exposed single-stranded regions, block mispairing, recruit enzymes, or act as architectural scaffolds. Small ligands can lock riboswitch states, metabolites can shift conformational ensembles, and RNA modifications can alter base pairing or protein recognition. Crowding and phase-separated environments can favor compaction or assembly while also changing diffusion and reaction rates. These effects are context-dependent, so "in cell" is not a single condition.

Metastability is not always a defect. Some functional RNAs use alternative structures to encode timing, regulatory switching, or proofreading. Other metastable states are harmful because they block processing, translation, ribosome assembly, or viral replication. Cells manage this risk with co-transcriptional checkpoints, surveillance nucleases, remodeling factors, and quality-control pathways. Current in-cell probing and single-molecule methods show that RNA structures are dynamic and heterogeneous, but they also require caution: chemical reactivity is an indirect readout, ensemble averages can hide subpopulations, and perturbations used to probe structure can change the structure being measured.

## Concept Inventory

- **RNA folding landscape:** the set of RNA conformations accessible to a sequence, together with the energy barriers and transition routes among those conformations. In cells, the landscape is conditional on transcription, ions, proteins, ligands, modifications, compartment, temperature, and time.
- **Native state:** the structure or ensemble that supports a specified biological function under a specified condition. For many RNAs, the native state is not a single rigid conformation.
- **Kinetic trap:** a locally stable RNA conformation that is slow to leave because the transition to a more functional conformation requires breaking several interactions or passing through a high-energy intermediate.
- **Metastability:** persistence of a non-equilibrium or locally stable state for a biologically meaningful time. A metastable RNA state can be functional, regulatory, neutral, or deleterious.
- **Co-transcriptional folding:** folding that occurs while RNA polymerase is still synthesizing the transcript. This process depends on transcription rate, pausing, nascent RNA length, protein recruitment, and the order in which sequence elements become available.
- **Kinetic partitioning:** division of an RNA population into alternative folding or assembly pathways before the molecules equilibrate. Partitioning can create distinct functional outputs from the same sequence.
- **RNA chaperone:** a molecule, usually a protein but sometimes another RNA or small molecule, that accelerates RNA annealing, strand exchange, or escape from misfolded states without being part of the final structure in a stoichiometric structural role.
- **RNA helicase:** an enzyme that couples nucleotide binding and hydrolysis to RNA duplex unwinding, RNA-protein remodeling, RNP assembly, RNA annealing, or conformational switching. The word "helicase" should not be overread as processive duplex unwinding in every biological context.
- **In-cell probing:** chemical, enzymatic, crosslinking, sequencing, imaging, or single-molecule approaches that infer RNA structure or interactions inside living cells or minimally disrupted cellular environments.

## What to Know Before Reading This Chapter

RNA is a negatively charged polymer with a ribose-phosphate backbone and bases that can pair, stack, and form noncanonical contacts. The same sequence can often form multiple secondary structures, especially when repeated, G-rich, U-rich, or complementary segments occur in different registers. A hairpin forms when nearby complementary sequences pair; a long-range interaction forms when distant regions pair; tertiary folding adds coaxial stacking, junction packing, metal-ion coordination, pseudoknots, ribose zippers, kissing loops, and protein-assisted contacts.

Thermodynamic folding asks which structure or ensemble is most stable at equilibrium. Kinetic folding asks how rapidly the molecule reaches each structure and whether barriers prevent equilibration. A predicted minimum-free-energy secondary structure is therefore not the same as a cellular structure. Minimum-free-energy algorithms can be useful approximations, but they usually omit transcriptional order, protein binding, local ion activity, chemical modification, degradation, and active remodeling.

The central examples in this chapter are nascent mRNAs, riboswitches, HIV-1 TAR RNA, preribosomal RNA, and regulatory RNA-protein complexes. These examples illustrate a general rule: the biological effect of an RNA structure depends on when and where the structure forms, not only on whether the base pairs are possible.

## 54.1. Folding in vitro versus in cells

In vitro RNA folding is the foundation for much of RNA biophysics. A purified RNA can be transcribed or synthesized, denatured, cooled, titrated with magnesium, mixed with protein, and observed by native gels, chemical probing, optical tweezers, fluorescence resonance energy transfer, small-angle scattering, cryo-EM, NMR, or activity assays. These experiments identify intrinsic structural preferences and allow kinetic models to be fitted under controlled conditions. They also expose how easily RNA can misfold. Even a molecule with a well-defined functional state may populate alternative helices or inactive tertiary arrangements when refolded from denaturant or shifted into magnesium-containing buffer.

![Figure 54.1. Three RNA Folding Landscapes](../assets/figures/chapter1050_figure1.png)

**Figure 54.1. Three RNA Folding Landscapes.** RNA sequence defines possible structures, but initial condition and cellular context determine which folding routes are accessible. Purified refolding begins with a complete denatured molecule and can resolve native and trapped states under controlled ion and protein conditions. Nascent cotranscriptional folding restricts the pathway by exposing the 5′ end before downstream regions are available. Protein binding, local ion activity, and helicase-mediated remodeling in the cellular RNP context can therefore populate different intermediates even when the underlying sequence is identical.

![Figure 54.2. Cotranscriptional Kinetic Partitioning](../assets/figures/chapter1050_figure2.png)

**Figure 54.2. Cotranscriptional Kinetic Partitioning.** Cotranscriptional folding converts transcriptional timing into structural choice. As a nascent transcript emerges from RNA polymerase, an early hairpin forms and the molecule can partition into a functional fold, a regulatory state, or a nonproductive trap depending on polymerase pausing, ligand binding, protein recruitment, and helicase action. The same RNA sequence can thus produce distinct biological outputs from a single transcriptional event.

**Table 54.1. Evidence Types for Cellular RNA Folding Claims.** Comparison of experimental approaches used to infer RNA structure in cellular contexts, highlighting what each measures, its key strengths, common sources of artifact, and primary use cases.

| Evidence type | What it measures | Strengths | Common artifacts | Best use |
| --- | --- | --- | --- | --- |
| **In vitro refolding** | Intrinsic folding pathways and ion dependence | Controlled conditions; quantitative kinetics | Artificial initial conditions; absent cellular partners | Mechanistic biophysics; kinetic model fitting |
| **In-cell SHAPE or DMS probing** | Chemical accessibility of nucleotides in living cells | Transcriptome-scale; captures RNP context | Indirect readout; protein protection mimics base pairing | Mapping cellular structural constraints |
| **Cotranscriptional probing** | Nascent-length-dependent structural signals | Detects early folding intermediates | Mixed transcript ages; ambiguous length assignment | Pathway timing and early intermediate detection |
| **Compensatory mutagenesis** | Functional effect of base-pair disruption and restoration | Causal evidence for specific helices | Mutations can alter protein binding, codons, or splicing signals | Validating predicted base pairs |
| **Single-molecule probing or FRET** | Subpopulation distributions and transition kinetics | Resolves heterogeneity hidden by ensemble averages | Labeling or surface artifacts; limited throughput | Kinetic pathway analysis and minority-state detection |
| **Helicase or RBP perturbation** | Factor dependence of RNA structure or function | Links remodeling to pathway outcome | Indirect pleiotropic effects; compensation by related factors | Assigning specific remodeler roles in cellular folding |

The advantage of purified systems is interpretability. If a ribozyme becomes active only above a magnesium threshold, the experiment links ion-dependent compaction to catalysis. If a DEAD-box protein accelerates strand exchange in a defined duplex substrate, the experiment isolates remodeling chemistry from downstream cellular outcomes. If a hairpin folds with a measured rate constant, kinetic modeling can compare nucleation, zipping, fraying, and loop closure. Such experiments established the basic vocabulary of folding intermediates, energy barriers, kinetic traps, and chaperone activity.

![Figure 54.3. Local DEAD-Box Helicase Remodeling Cycle](../assets/figures/chapter1050_figure3.png)

**Figure 54.3. Local DEAD-Box Helicase Remodeling Cycle.** DEAD-box helicases often remodel RNA locally rather than translocating processively along a duplex. ATP binding promotes domain closure and local strand separation or RNP rearrangement; hydrolysis and product release reset the enzyme for another round. Accessory factors and the broader RNP context provide biological specificity that determines which substrate is acted upon and when.

The limitation is that in vitro folding often starts from an artificial initial condition. Many protocols heat-denature the complete RNA and then cool it in the presence of salts. A cellular transcript is not born as a full-length denatured molecule. RNA polymerase exposes the 5′ end first, may pause, may recruit processing machinery, and may keep the 3′ region physically unavailable until later. In bacteria, translation can begin while transcription continues. In eukaryotes, the polymerase II carboxy-terminal domain, capping enzymes, splicing factors, cleavage factors, export factors, and RNA-binding proteins produce a crowded local assembly around the nascent pre-mRNA. The relevant initial condition is therefore a moving, partially synthesized, partially protein-bound polymer.

Cells also make folding an actively maintained state. RNA structures that form in vitro may be opened by helicases in vivo, masked by proteins, cut by nucleases, methylated, edited, exported, localized, or translated. Conversely, structures that look unstable in dilute buffer may be stabilized in a cellular RNP because proteins clamp a junction, a ligand occupies a pocket, or neighboring molecules increase effective concentration. This is why in-cell probing can disagree with deproteinized probing. The disagreement does not automatically mean that one assay is wrong; it may reveal that the RNA's cellular structure is a property of an RNP state rather than the naked RNA sequence.

The phrase "RNA folds in cells" therefore includes several distinct claims. One claim is that an RNA sequence has intrinsic base-pairing potential. A second claim is that a structure is populated in living cells. A third claim is that a structure changes function. A fourth claim is that the structure is causally required for that function. These claims require different evidence. Chemical probing may support population of a local helix. Mutational covariation or compensatory mutagenesis may support sequence-specific pairing. Biochemical reconstitution may show mechanism. Genetics, rescue, and time-resolved perturbation are needed to connect folding to biological output.

Do not overgeneralize from either purified or cellular assays. Purified RNA can reveal real physical states that cells exploit, but it may overrepresent traps created by nonphysiological refolding. In-cell assays measure more native contexts, but they average across cell-cycle stage, transcript age, isoform, compartment, and protein occupancy unless designed to resolve those variables. The best-supported cellular folding models combine purified biophysics, cellular perturbation, comparative sequence analysis, structural probing, and functional rescue.

## 54.2. Cotranscriptional folding and kinetic partitioning

Co-transcriptional folding is the process by which RNA structure forms while transcription is still underway. The mechanism begins with a simple asymmetry: nucleotide 1 is available before nucleotide 100, and nucleotide 100 before nucleotide 1000. Early segments can fold locally before downstream pairing partners exist. If those early structures are compatible with the final functional arrangement, co-transcriptional folding can guide the RNA down the correct pathway. If they are incompatible, early structures can trap the transcript in a state that must later be resolved.

Kinetic partitioning describes the split between alternative pathways. Imagine a nascent RNA containing region A near the 5′ end, region B in the middle, and region C downstream. Region A may first pair with B because C has not yet been transcribed. Later, when C emerges, the functional structure may require A-C pairing. Some molecules will exchange B for C quickly. Others will keep the A-B helix long enough that a protein binds, a nuclease cuts, a spliceosome assembles differently, or a terminator forms. The final biological outcome depends on the timing of transcription, folding, binding, and remodeling.

Riboswitches provide a clear example. A metabolite-sensing aptamer can fold before the expression platform has fully formed. Ligand binding during a limited time window stabilizes one structure and biases whether downstream sequences form a terminator, antiterminator, sequestered ribosome-binding site, or exposed ribosome-binding site. The same RNA sequence can therefore encode a kinetic decision rather than a simple equilibrium binding curve. The ligand concentration, transcription rate, pausing, temperature, and protein cofactors determine whether the transcript commits to one output before it can sample another.

HIV-1 TAR RNA illustrates how viral RNAs can use nascent folding pathways. TAR forms near the 5′ end of HIV-1 transcripts and participates in transcriptional regulation through protein interactions. Recent work modeling and probing co-transcriptional TAR folding emphasizes that pathway order matters: structures available during synthesis may differ from structures predicted for a completed RNA allowed to equilibrate. This does not make thermodynamics irrelevant. Instead, thermodynamics defines the relative stability of states, while transcriptional order and barriers determine which states are reached on the biological time scale.

Eukaryotic pre-mRNAs add another layer because folding competes and cooperates with processing. Splice sites, branch points, polypyrimidine tracts, exonic splicing enhancers, intronic splicing silencers, polyadenylation signals, and RNA modifications can be exposed or hidden by local structure. Co-transcriptional splicing is not simply splicing before release; it is a coupled process in which polymerase speed, chromatin state, nascent RNP assembly, and RNA folding influence spliceosome recognition. Plant and metazoan systems show that co-transcriptional processing and modification can be integrated with RNA folding decisions rather than appended after synthesis.

Preribosomal RNA is a particularly demanding substrate. The eukaryotic rRNA precursor contains large domains that must fold, be modified, be cleaved, and assemble with ribosomal proteins and assembly factors in a specific sequence. Folding errors are not merely local; an incorrectly packed region can block later processing or subunit maturation. Ribosome biogenesis therefore uses many RNA helicases and assembly factors as timing devices, local remodelers, and checkpoints. The word "folding" in this setting includes RNA secondary structure, RNA tertiary packing, protein recruitment, snoRNA-guided modification, and ordered removal of assembly factors.

Co-transcriptional chemical probing gives experimental access to nascent folding, but interpretation is difficult. A chemical signal at one nucleotide may reflect base-pairing, protein protection, RNA polymerase proximity, transcript age, or a mixture of nascent species. Methods that capture time or transcript length are therefore essential for separating early intermediates from mature states. A robust model should specify which nascent length, which transcriptional condition, and which cellular compartment is being discussed.

## 54.3. RNA chaperones and helicases

RNA chaperones are factors that help RNA molecules find productive structures without necessarily becoming permanent structural components. Many chaperones bind single-stranded or weakly structured RNA, destabilize incorrect contacts, promote annealing between complementary strands, or accelerate strand exchange. Some are proteins with broad RNA-binding surfaces; others are viral or bacterial factors adapted for a particular life cycle; some structured RNAs can also facilitate pairing reactions. The defining functional feature is not a single fold or domain but an effect on the kinetic barriers between RNA conformations.

RNA chaperone activity is easiest to understand in two limiting cases. In the first, an RNA is stuck in a mispaired state. A chaperone binds transiently, weakens the incorrect helix, and increases the probability that the RNA refolds. In the second, two complementary strands fail to find each other efficiently because their pairing regions are transiently hidden. A chaperone binds, presents, or destabilizes the strands so that annealing occurs. In both cases, the chaperone changes the rate of reaching a product more than it changes the ultimate equilibrium product.

RNA helicases are enzymes, but their cellular roles go beyond the textbook image of a protein walking along a duplex and separating strands. DEAD-box helicases often bind a short duplex region, use ATP binding and hydrolysis to produce local strand separation or bending, and then release. Some DExD/H-box helicases remodel RNA-protein contacts, displace proteins, stabilize assembly intermediates, or serve as checkpoints. The same family can show unwinding, annealing, clamping, or RNP-remodeling behavior depending on substrate and cofactors.

The ATP cycle gives helicases directionality in time even when they are not processive motors in space. ATP binding can increase RNA affinity or change domain closure. Hydrolysis and product release can reset the enzyme. Accessory proteins and RNA motifs can recruit a helicase to a defined substrate, stimulate its ATPase activity, or prevent unproductive action. This logic allows a broadly RNA-binding enzyme to act specifically in ribosome assembly, spliceosome activation, translation initiation, RNA export, decay, antiviral defense, or chromatin-associated RNA remodeling.

Ribosome biogenesis demonstrates why cells need many helicases rather than one general unwinder. Distinct helicases act at different maturation stages, on different preribosomal particles, and in different compartments. Some promote snoRNA release after modification, some resolve local RNA structures before cleavage, and some remodel protein-bound intermediates. A helicase defect can therefore produce a specific maturation block even if the enzyme has generic unwinding activity in vitro. Disease associations of eukaryotic helicases similarly reflect pathway-specific requirements, tissue context, and network buffering rather than a single biochemical failure mode.

Viral systems show additional variation. Viral helicase or chaperone activities can assist genome replication, subgenomic RNA synthesis, packaging, or evasion of host restriction. A viral protein with helicase and chaperoning activity may be tuned to viral RNA structures rather than to host RNP assembly pathways. These cases are useful reminders that "RNA helicase" is a biochemical label, while the biological substrate and purpose must be established separately.

Recent studies also broaden where RNA chaperone activity matters. DEAD-box helicases eIF4A1 and eIF4A2, known for roles in translation initiation, have been implicated in RNA-chaperone activity during mitotic exit and chromatin decondensation. The key lesson is not that every helicase has every function, but that RNA remodeling can connect RNA structure to nuclear organization, cell-cycle transitions, and chromatin-associated RNP states.

## 54.4. Protein, ligand, ion, and crowding effects

RNA folding in cells occurs in a chemically dense environment. The backbone carries one negative charge per phosphate, so electrostatic repulsion is a major barrier to compaction. Monovalent cations screen charge; magnesium ions and other divalent metals can stabilize close phosphate packing, tertiary contacts, and catalytic sites. However, the magnesium concentration added to an in vitro tube is not equivalent to free magnesium activity in a cell. Much cellular magnesium is bound by ATP, nucleic acids, metabolites, membranes, and proteins. Local ion availability may differ between cytoplasm, nucleolus, mitochondrion, viral replication compartment, stress granule, and chromatin-associated condensate.

Proteins change folding in several ways. A protein can bind single-stranded RNA and prevent an alternative helix from forming. A protein can stabilize a hairpin loop, kink-turn, internal loop, or junction. A protein can bridge distant RNA segments, creating an effective concentration that favors tertiary assembly. A protein can also block chemical probing reagents, so a protected nucleotide is not automatically base-paired. For this reason, in-cell probing is most powerful when combined with protein perturbation, crosslinking, structural data, or compensatory mutation.

Ligands add specificity. Riboswitch metabolites bind aptamer pockets and stabilize structures that control transcription, translation, splicing, or RNA stability. Small molecules, antibiotics, and therapeutic ligands may stabilize or disrupt structured RNAs. Protein ligands such as ribosomal proteins and spliceosomal components can act with similar timing logic: binding to one intermediate can prevent misfolding and promote later assembly, while premature binding can trap an off-pathway state.

RNA modifications are another folding variable. Methylation, pseudouridylation, editing, acetylation, and other chemical changes can alter base-pair strength, stacking, hydration, protein recognition, and innate immune sensing. Some modifications are installed co-transcriptionally or during RNP assembly, so they can influence folding pathways rather than merely decorate final structures. In rRNA and tRNA, modifications often stabilize local geometry or support decoding and catalysis. In mRNA, modification effects can be more context-dependent, because a single modification may influence structure, reader-protein binding, decay, translation, or immune detection.

Macromolecular crowding is often described as favoring compact states, but that shorthand is incomplete. Crowding changes excluded volume, diffusion, viscosity, association rates, and effective concentrations. It may favor compact RNA, promote phase separation, slow large-scale rearrangements, or increase nonspecific contacts. A crowded lysate, a bacterial nucleoid, a nucleolus, and a membrane-associated viral replication organelle are not interchangeable environments. Cellular crowding should therefore be treated as a set of physical constraints rather than a single parameter.

Protein, ligand, ion, and crowding effects also create apparent contradictions between assays. A helix can appear stable in deproteinized RNA but reactive in cells because an RBP opens it. A region can appear single-stranded in chemical probing but be functionally inaccessible because a protein covers it. A ligand-responsive structure can be missed if the ligand is absent during extraction. A magnesium-stabilized tertiary fold can collapse during purification if the ion or protein environment changes. These are not minor technicalities; they determine whether a structural claim is biologically meaningful.

## 54.5. Metastability, traps, and refolding

A kinetic trap is a structure that persists because escape is slow, not because the structure is globally most stable. For RNA, escape often requires breaking multiple base pairs before new base pairs can form. If the intermediate with broken pairs is too unstable, the RNA remains trapped. Long RNAs are especially vulnerable because distant regions can form incorrect helices before all correct partners are present. Repeats, palindromic segments, G-quadruplex-forming sequences, pseudoknot alternatives, and multihelix junctions can create rugged landscapes.

Metastability is a time-scale concept. A structure that persists for milliseconds may matter in a riboswitch decision if transcription passes the decision point quickly. A structure that persists for minutes may matter for mRNA localization or translation. A structure that persists for hours may define a stable RNP state. Conversely, a trap observed after heat-denaturation in vitro may be irrelevant if a cell never exposes the RNA to that pathway.

Cells use several strategies to avoid or resolve harmful traps. First, co-transcriptional folding can restrict the search space by allowing local domains to form in order. Second, RNA-binding proteins can bind nascent regions and prevent inappropriate contacts. Third, RNA helicases and chaperones can destabilize incorrect structures. Fourth, processing and degradation pathways can remove failed substrates. Fifth, compartmentalization can separate RNAs from inappropriate partners or concentrate needed factors.

Refolding is not always restoration of a final native state. Some RNAs deliberately switch between states. Riboswitches, thermosensors, frameshift signals, viral regulatory elements, and some untranslated-region structures use conformational alternatives as functional states. In these cases, a "trap" from one viewpoint can be a regulatory memory state from another. The key question is whether the state has a defined biological role, whether transitions are controlled, and whether perturbing the state changes output in a predictable way.

For large RNPs, traps often include proteins. A misassembled preribosomal particle may contain correct RNA segments but wrong protein order, unresolved snoRNA interactions, or missing modification marks. A spliceosomal complex may require rearrangement of both RNA-RNA and RNA-protein contacts before catalysis. An mRNP may need remodeling during export, localization, translation initiation, or decay. Calling these events "RNA folding" is acceptable only if the claim specifies the RNP components and remodeling step.

Therapeutic and synthetic RNAs make kinetic traps a design problem. An mRNA vaccine or synthetic mRNA must be transcribed, capped, polyadenylated, purified, formulated, delivered, released from endosomes, translated, and degraded. Secondary structures can protect RNA or impede translation; modified nucleotides can tune innate sensing and stability; codon choice can alter elongation and decay; delivery formulation can expose RNA to ions, lipids, and proteins. A design that looks optimal by static folding prediction may perform poorly if it forms persistent structures near the 5′ untranslated region or engages cellular sensors. This paragraph is a conceptual bridge to [Chapter 134](chapter1122.md); product-specific therapeutic RNA folding evidence should be added during that chapter's therapeutic-RNA reference expansion.

## 54.6. In-cell probing and single-molecule evidence

In-cell RNA structure probing attempts to measure structure where the RNA normally functions. Chemical reagents such as dimethyl sulfate or SHAPE reagents modify nucleotides according to accessibility, flexibility, or base-pairing context; reverse transcription or mutational profiling then maps modification sites. In-cell SHAPE-MaP, DMS-MaPseq, icSHAPE, structure-seq, and related methods have made it possible to survey transcriptome-scale RNA structure and compare living cells with extracted RNA.

These methods are powerful because they can reveal the difference between naked RNA and cellular RNPs. A transcript region may be less structured in cells because ribosomes, helicases, or RBPs open it. Another region may be more protected in cells because proteins or ligand-bound structures stabilize it. Transcriptome-wide studies can identify trends in untranslated regions, coding sequences, splice regions, viral genomes, stress responses, and developmental transitions. They also generate hypotheses about regulatory structures that can be tested by mutation.

The evidence is indirect. A SHAPE-reactive nucleotide is flexible or conformationally permissive under the reagent's chemistry, not simply "unpaired" in every context. A DMS-protected adenine or cytosine may be base-paired, protein-bound, inaccessible, modified, or present in a subpopulation. Reverse transcriptase stops and mutational signatures can be biased by sequence, modification, and library preparation. Cell permeability, reagent half-life, dose, temperature, and quench conditions matter. Therefore, a probing profile is a structural constraint, not a complete structure.

Single-molecule methods address ensemble averaging. Optical tweezers, single-molecule FRET, nanopore measurements, and correlated chemical probing can distinguish subpopulations and transition paths that bulk assays merge. Single-molecule correlated chemical probing, for example, can identify whether modifications at distant positions occur on the same molecule, providing evidence for mutually exclusive structures or heterogeneous ensembles. Such data are especially valuable for kinetic partitioning because the mean signal may hide a minority state that has strong biological consequences.

In-cell single-molecule evidence remains technically challenging. The cellular environment is hard to control, labels can perturb folding, and observation windows can bias which RNAs are measured. Many studies therefore combine in vitro single-molecule experiments with in-cell or in vivo ensemble probing. This combination can be rigorous if the model explicitly states which observations come from purified RNA, which from living cells, and which are inferred by synthesis.

The strongest causal tests perturb structure without simply destroying the RNA. Compensatory mutations are a classic standard: mutation 1 disrupts a predicted pair and changes function; mutation 2 restores pairing by changing the partner base and rescues function. For large cellular RNAs, this is often difficult because mutations can alter protein binding, codon usage, splicing motifs, modification sites, or RNA abundance. Still, the logic remains essential. A structural model should ideally predict the effect of targeted mutations, protein depletion, ligand addition, transcription-rate change, or helicase perturbation.

> **Box 54.1. How Not to Overinterpret Chemical Probing**
>
> - High reactivity does not always mean a nucleotide is permanently unpaired; it reflects flexibility or chemical accessibility under the specific probing conditions.
> - Low reactivity does not always mean a nucleotide is base-paired; protein binding, chemical modification, local reagent inaccessibility, and compartment effects can all reduce signal.
> - Transcript abundance differences, cell-state heterogeneity, and reverse-transcriptase bias can create artifactual patterns that mimic structural signals.
> - Strong structure-function claims require orthogonal evidence: compensatory mutagenesis, purified reconstitution, time-resolved probing, protein perturbation, or functional rescue.

> **Box 54.2. Kinetic Trap or Functional Switch?**
>
> - Ask whether the metastable state occurs on a physiological pathway rather than only after nonphysiological refolding from denaturant.
> - Ask whether the timing of structural transitions affects biological output, such as a riboswitch ligand-binding decision or a terminator-antiterminator choice.
> - Ask whether cells deploy helicases, RNA chaperones, or surveillance nucleases to actively manage the state.
> - Ask whether targeted mutations or ligand binding shift the structure in a predictable and functionally correlated way.
> - Illustrative examples include riboswitch decision states, HIV-1 TAR transcriptional regulatory elements, preribosomal misassembly intermediates, and synthetic mRNA 5′ untranslated region structures.

## Recent Consensus

The current consensus is that cellular RNA folding is a co-transcriptional, RNP-coupled, and actively remodeled process. Sequence and thermodynamics matter, but they do not determine cellular structure alone. Folding pathways are shaped by transcription rate, RNA polymerase pausing, nascent RNA length, processing factors, RNA modifications, ligand binding, ions, RBPs, chaperones, helicases, degradation, and compartment-specific physical chemistry.

RNA helicases are now viewed as RNA and RNP remodelers rather than generic duplex-unwinding machines. DEAD-box proteins often act locally; related helicase families can have distinct processivity, directionality, substrate preferences, and accessory-factor dependence. Their biological specificity usually comes from recruitment, timing, cofactors, and pathway context.

In-cell probing has shifted the field from asking whether RNA can form structures to asking which structures form in which cells, at which transcript ages, and in which RNP states. The consensus is also cautious: chemical probing and computational folding need orthogonal validation before a structure-function claim is secure.

## Open Questions, Controversies, Deprecated Models, and Common Misconceptions

Open questions:

- How often do transcriptome-wide probing differences correspond to functionally important structures rather than protein occupancy, RNA abundance differences, or indirect stress responses? This question requires targeted perturbations and rescue assays, not only structure maps.
- Which cellular RNAs are under selection for kinetic folding pathways rather than only final structures? Riboswitches and some viral RNAs are clear examples, but many mRNA and lncRNA claims remain less certain.
- How should models integrate transcriptional speed, polymerase pausing, cofactor recruitment, ribosome movement, RNA modification, and helicase action into predictive folding simulations? Recent computational work has improved co-transcriptional modeling, but cellular parameterization remains incomplete.

Controversies:

- Controversy: Some proposed regulatory RNA structures are supported mainly by computational prediction or population-level probing. Without compensatory mutations, binding perturbations, or kinetic evidence, such structures should be treated as hypotheses.

Common misconceptions:

- "The minimum-free-energy structure is the cellular structure." This is false. The minimum-free-energy structure is an equilibrium-model output under simplified assumptions.
- "A helicase always unwinds RNA like a processive motor." Many RNA helicases act locally, remodel RNPs, anneal strands, clamp substrates, or act through accessory-factor-controlled cycles.
- "Protein protection and base pairing are interchangeable explanations for low chemical reactivity." Low reactivity can reflect base pairing, protein binding, inaccessible compartments, chemical modification, or reagent limitations.