RNA Is Not a Flat Message. It Is a Shape-Shifting Machine

 

Graphical Abstract

For decades, biology students have been taught a clean story: DNA stores information, RNA carries the message, and proteins do the work.

That story is useful. It is also incomplete.

RNA is not simply a courier moving genetic instructions from one place to another. It folds. It bends. It hides some regions and exposes others. It can adopt more than one structure, sometimes within the same population of molecules. These alternative shapes can influence whether an RNA is translated, degraded, stabilized, or ignored.

A new study in Nature Methods pushes this idea further by showing how individual RNA molecules can be read not only as sequences, but as structural objects. The authors developed a method called sm-PORE-cupine, which combines chemical RNA structure probing with nanopore direct RNA sequencing to detect RNA structure ensembles in single molecules. In simpler terms, they built a way to ask: what shapes are different copies of the same RNA molecule actually taking inside a cell?

Why RNA Structure Is Hard to See

RNA structure is usually measured as an average. Scientists treat many copies of an RNA molecule with a chemical probe, sequence the result, and infer which bases are paired or unpaired. This is powerful, but it hides variation.

Imagine taking a photograph of a crowd and averaging all the faces into one image. You would get a blurry “average person,” but you would lose the actual individuals.

RNA has the same problem.

One transcript may not exist as a single structure. Some molecules may fold one way, others another way. These different structural states are called RNA structure ensembles. The biological meaning may lie not in the average structure, but in the minority conformation that appears only under certain conditions.

That is the central challenge this study addresses.

The Core Idea: Read RNA Directly, Then Recover Its Shape

The method builds on nanopore direct RNA sequencing. Unlike many sequencing methods that first convert RNA into cDNA, direct RNA sequencing pulls native RNA molecules through a nanopore and measures current changes as the molecule passes through.

The authors combined this with SHAPE chemical probing using NAI-N3, a reagent that preferentially modifies flexible, single-stranded RNA regions. Modified bases alter the nanopore signal. By detecting those altered signals along each molecule, the researchers could infer which parts of that individual RNA molecule were structurally exposed.

This sounds straightforward, but there was a technical trap. Higher chemical modification rates improve structural information, but heavily modified RNA reads become harder to basecall and map. Many reads that contain valuable structure information are lost because standard alignment struggles with them.

The clever solution was to stop relying only on basecalled sequence alignment. The authors used direct signal alignment with dynamic time warping, allowing them to recover reads that conventional mapping would miss. In benchmark RNAs, this rescued a substantial fraction of otherwise failed reads and increased the usable data for downstream structure analysis.

That detail matters. The reads most likely to be thrown away are often the ones carrying rich modification signals. Recovering them improves the ability to distinguish structural populations.

Sorting RNA Molecules Into Structural Populations

After detecting modification patterns on individual molecules, the next problem was clustering: how do you separate one RNA shape from another?

The authors tested several clustering approaches and found that a Bernoulli mixture model performed well for separating RNA structural populations. They validated this using known riboswitches, including the adenosine riboswitch.

Riboswitches are useful test cases because they change structure when bound to specific ligands. The method could distinguish ligand-bound and unbound populations and even detect intermediate or minority conformations. Importantly, it could identify alternative structure populations even when one state represented only about 10% of the molecules.

This is the biological payoff: not merely “RNA has this structure,” but “this RNA population contains multiple structural states, and their proportions change.”

SARS-CoV-2: One Genome, Many Structural Possibilities

The authors then applied sm-PORE-cupine to SARS-CoV-2 RNA. Viral RNAs are especially interesting because structure can regulate replication, translation, packaging, and immune evasion.

The study found that the 3′ end of the SARS-CoV-2 genome is highly structurally heterogeneous. This region contains several subgenomic RNAs, and the authors showed that different subgenomic RNAs, including nucleocapsid, ORF7a, and ORF8, display different levels of structural heterogeneity. The nucleocapsid RNA was especially heterogeneous among the tested subgenomic RNAs.

This suggests that viral RNA structure is not a fixed map. It is more like a set of competing layouts, with different viral transcripts folding into distinct structural populations.

That has major implications. If RNA structure affects viral gene expression, then drugs or antisense strategies targeting viral RNA may need to account for structural diversity, not just sequence.

Candida albicans: RNA Structure During a Cellular Identity Shift

The most biologically interesting part of the study may be its work in Candida albicans, a fungal pathogen that can shift from yeast-like growth at 30 °C to hyphal growth at 37 °C.

This transition matters because the hyphal form is associated with pathogenicity. The authors asked whether RNA structural ensembles change during this temperature-dependent transition.

They performed structure probing in vivo and in vitro at both temperatures and found several important patterns.

First, RNA structures were generally more homogeneous in vitro than in vivo. That means the cellular environment introduces structural complexity that purified RNA does not fully capture.

Second, RNA structures became modestly more homogeneous at higher temperature.

Third, coding sequences were more structurally heterogeneous than 3′ untranslated regions, while highly translated transcripts tended to have more homogeneous 3′ UTR structures at 37 °C.

This points toward a regulatory role for 3′ UTR structure. The 3′ UTR is often treated as a control panel for RNA stability, localization, and translation. This study adds another layer: the structure of that control panel may shift with temperature.

RNA Thermometers Beyond Bacteria?

The authors identified 95 regions in C. albicans 3′ UTRs that changed structural heterogeneity between 30 °C and 37 °C. They focused on two transcripts, RPS19A and RPL29, and showed that their 3′ UTR structural changes were linked to changes in translation using luciferase reporter assays.

This is a striking result because it suggests that some fungal mRNAs may behave like RNA thermometers. Their structures respond to temperature, and those structural changes affect protein production.

The phrase “RNA thermometer” is familiar in bacterial gene regulation, but this study suggests a broader principle: eukaryotic mRNAs may also use temperature-sensitive structure ensembles to tune expression.

Why This Study Matters

The real advance here is not just another RNA probing method. It is a change in resolution.

Older approaches often asked:

What is the average structure of this RNA?

This study asks:

How many structural states does this RNA population contain, and how do those states change across conditions?

That distinction matters for RNA biology, virology, fungal pathogenesis, and therapeutic targeting. If an RNA exists in multiple structural states, then the biologically relevant state may not be the dominant one. A low-abundance conformation could control translation, expose a regulatory motif, recruit a protein, or create a druggable structural pocket.

The study also highlights a broader lesson for transcriptomics. RNA sequencing has become extremely good at counting molecules and identifying isoforms. But RNA molecules are not linear strings floating passively in the cell. Their folding creates another layer of information—one that may explain why two RNAs with similar abundance can behave differently.

The Bigger Picture

Biology is moving from sequence to structure, from averages to single molecules, and from static models to ensembles.

sm-PORE-cupine fits directly into that transition. It gives researchers a way to observe RNA structural diversity molecule by molecule, transcript by transcript, and condition by condition.

The work also reminds us that the cell is not a test tube. RNA folding in vivo is shaped by temperature, proteins, translation, decay machinery, molecular crowding, and local cellular context. A structure predicted on a computer or measured in purified RNA may capture only part of the story.

RNA is not just a message.

It is a molecule with memory, movement, and choice. It can fold into different futures. This study gives us a sharper way to watch those futures form.

CRISPR’s New Kill Switch: How Cas12a2 Turns a Cell’s Own RNA Against It

 

Graphical Abstract

CRISPR has usually been described as a molecular scalpel. That metaphor is useful, but it is also a little too polite. A scalpel cuts where it is told. It edits. It repairs. It leaves behind a changed genome and, in many cases, a living cell.

The new study “RNA-triggered cell killing with CRISPR–Cas12a2” pushes CRISPR into a different role. Here, CRISPR is not merely an editor. It becomes a programmable execution system. The enzyme Cas12a2 can be guided to recognize a specific RNA molecule inside a cell. Once it finds that RNA, it unleashes widespread DNA damage, pushing the cell toward death. In simple terms: the cell’s own transcript becomes the trigger for its destruction.

That is why this work feels important. It does not just ask whether CRISPR can change a cell. It asks whether CRISPR can decide which cells should survive.

The old CRISPR problem: editing is easier than killing

In bacteria, CRISPR-based killing is relatively straightforward. If a CRISPR nuclease cuts an essential DNA sequence, the bacterium often cannot recover. But eukaryotic cells, including human cells, are much better at surviving DNA damage. They have repair systems such as non-homologous end joining and homology-directed repair. A conventional Cas9 or Cas12a cut may produce an edit rather than death.

That is useful for genome engineering, but frustrating if the goal is to eliminate a dangerous cell.

Cas13, another CRISPR system, targets RNA. It can degrade RNA transcripts, but in mammalian cells this does not always translate into robust cell death. The cell may lose a transcript, slow down, or compensate. The authors of this study frame Cas12a2 as a different kind of tool: an RNA-sensing nuclease that responds to transcript recognition by shredding DNA in trans.

This distinction matters. Cas12a2 is not simply cutting the target RNA. It is using the RNA as a molecular tripwire.

How Cas12a2 works

Cas12a2 is guided by a small RNA sequence. When the guide finds a matching target RNA, especially near an adenine-rich protospacer-flanking sequence, the enzyme becomes activated. Once activated, Cas12a2 does not politely cut one defined locus. It begins collateral cleavage of nucleic acids, including double-stranded DNA.

That sounds dangerous, and biologically it is. But the danger is also the point.

The researchers tested two related versions, SuCas12a2 and GeCas12a2, and showed that they can be programmed against specific transcripts. In yeast, targeting the ADE2 transcript with GeCas12a2 caused a dramatic reduction in surviving transformants. The system worked even when cells were given a repair template, suggesting that Cas12a2 killing was not easily escaped by standard local DNA repair.

Then came the more important test: human cells.

Killing human cells by recognizing a transcript

The team first used HeLa cells engineered to express GFP. GFP is useful because it gives researchers a clean, visible target. When they delivered GeCas12a2 with a guide aimed at the GFP transcript, the GFP-expressing cells failed to grow and were strongly depleted. The study reports about 86% cell depletion after targeting GFP in HeLa-GFP cells, while non-targeting controls continued to proliferate.

The result was not limited to an artificial GFP transcript. The authors also targeted endogenous transcripts across several cancer-derived cell lines, including transcripts with different abundance levels. Cas12a2 could deplete cells even when some target transcripts were relatively poorly expressed, although transcript abundance still mattered. The tool also worked when delivered as Cas12a2 mRNA and guide RNA packaged in lipid nanoparticles, an important delivery format for future therapeutic development.

The central idea is straightforward but powerful: if a cell expresses the RNA that Cas12a2 has been programmed to recognize, it can be eliminated. If it does not express that RNA, it should be spared.

What actually kills the cell?

The authors did not stop at showing cell loss. They asked what was happening inside the cell.

Activated Cas12a2 produced extensive double-stranded DNA breaks. The team measured this using 53BP1 foci, a marker of DNA double-strand break repair. Cas12a2 targeting GFP or GAPDH caused at least a 5.2-fold increase in 53BP1 foci compared with non-targeting or vehicle controls. The DNA damage level was comparable to that caused by established DNA-damaging anti-cancer drugs such as cisplatin and etoposide.

But there is a crucial difference. Cisplatin and etoposide damage DNA broadly. Cas12a2 is activated only when the chosen transcript is present.

The downstream consequences looked like a cell in serious trouble: abnormal DNA-content profiles, reduced G1 cell population, signs of mitotic catastrophe, apoptosis markers such as annexin V and caspase-3/7 activity, and inflammatory gene-expression signatures. The authors conclude that RNA-triggered Cas12a2 eliminates human cells mainly through extensive DNA damage followed principally by apoptosis, with other death pathways also contributing.

The specificity question

A programmable cell-killing system is only useful if it does not kill the wrong cells.

This is the most obvious concern. If Cas12a2 tolerates mismatches too easily, a guide intended for one RNA might accidentally recognize a related transcript and kill healthy cells. The authors therefore tested guides against transcripts absent from human cells, looked for DNA damage, examined barcode integration as a readout of double-strand breaks, and tested predicted off-target RNA candidates.

Under the tested conditions, they found no measurable off-target activation in human cells. Mismatched guides generally failed to trigger depletion, and non-targeting guides did not induce the transcriptomic disruption seen with true on-target activation.

This is encouraging, but it should not be overread. The system is still early. Specificity will need to be tested across more cell types, transcriptomes, delivery contexts, disease models, and guide designs. For a cell-killing technology, “mostly specific” is not enough. The safety threshold will be much higher than for ordinary gene perturbation.

Application 1: killing HPV-positive cells

The first major application was viral infection.

Cells infected with high-risk human papillomavirus express viral transcripts that are absent from normal human cells. That makes HPV an attractive test case. The researchers designed guides against HPV E6 and E7 transcripts, two viral oncogenes central to HPV-driven cancers.

In HPV18-positive HeLa-GFP cells, Cas12a2 guides targeting E6 or E7 produced about 94% cell reduction. The same guides did not significantly deplete HPV-negative HEK293-GFP cells.

The team also moved into an in vivo model. In a patient-derived xenograft model of HPV16-positive head and neck squamous cell carcinoma, intratumoral administration of lipid nanoparticle-packaged GeCas12a2 mRNA plus an HPV16 E6 guide significantly reduced tumour growth compared with buffer control. Histology showed Cas12a2 expression and apoptotic markers after treatment.

This is not yet a therapy. It is a proof of concept. But it demonstrates why RNA-triggered killing is interesting: viral transcripts can act as highly specific molecular flags.

Application 2: enriching successfully edited cells

Genome editing often produces mixed populations. Some cells receive the intended edit; others remain unedited. Researchers usually need selection markers, sorting, cloning, or laborious screening to enrich the edited cells.

Cas12a2 offers a clever alternative. Program it to recognize the unedited transcript. Cells that failed editing still express the original RNA and are killed. Cells carrying the desired edit disrupt the guide-recognition site and survive.

The authors first showed that Cas12a2 could remove one cell type from a mixed culture. In a co-culture of GFP-expressing and RFP-expressing HeLa cells, GFP-targeting GeCas12a2 caused about 93% reduction of GFP-positive cells while RFP-positive cells continued growing.

They then used the method to enrich genome edits. After FnCas12a editing of a GFP locus, GeCas12a2 targeting the unedited GFP transcript increased indel frequency by 3.1-fold. For prime editing of GAPDH, Cas12a2 counterselection enriched precise edits by up to 4.3-fold compared with non-targeting controls.

This could become valuable in editing workflows, especially where edited cells are rare and difficult to isolate.

Application 3: targeting a cancer mutation

The most dramatic part of the study involves KRASG12C, a clinically important oncogenic mutation. The challenge is severe: the mutant RNA differs from wild-type KRAS by a single nucleotide. A useful system must kill cells expressing mutant KRAS while sparing cells expressing wild-type KRAS.

The researchers empirically selected a guide that activated Cas12a2 with the KRASG12C transcript but not the wild-type transcript. In engineered U2OS cells overexpressing either wild-type KRAS or KRASG12C, the KRASG12C-targeting Cas12a2 RNP depleted mutant cells by about 62% without measurably depleting wild-type KRAS-overexpressing cells.

They then tested NCI-H23 cells, which naturally carry heterozygous KRASG12C. Cas12a2 targeting KRASG12C caused about 50% depletion and increased DNA damage markers. When combined with sotorasib, an FDA-approved KRASG12C inhibitor, cell depletion exceeded 85% in the tested setting. Importantly, sotorasib-resistant cells were still depleted by Cas12a2, suggesting a possible complementary strategy for resistant cancer cells.

Again, this is early-stage biology, not a ready clinical intervention. But the concept is striking: a single mutated RNA base can potentially become the trigger for selective cell destruction.

Why this paper matters

The larger significance is not simply that Cas12a2 kills cells. Many things kill cells. The significance is that Cas12a2 can connect cell identity to cell death through RNA recognition.

That opens a broad design space. In principle, one could imagine targeting cells based on viral RNAs, fusion transcripts, cancer mutations, aberrant splice junctions, circular RNAs, edited RNAs, or other disease-associated transcript signatures. The authors explicitly suggest applications across basic research, medicine, biotechnology, biomanufacturing, and agriculture.

It also changes how we think about CRISPR. Traditional CRISPR editing asks: “What sequence do we want to change?” Cas12a2 asks a harsher question: “Which transcript marks a cell that should not remain alive?”

The hard problems ahead

Several barriers remain.

Delivery is the largest one. Getting Cas12a2 and its guide into the right cells, at the right dose, without unacceptable toxicity, will be difficult. Local injection into a tumour is very different from systemic delivery in a patient.

Guide design also needs maturation. The enzyme requires suitable target features, including the appropriate sequence context. High-throughput screens and machine-learning models may be needed to predict which guides kill efficiently and which are safest.

Another issue is survival. Some cells may escape Cas12a2-triggered damage. Understanding what those surviving cells look like genetically, epigenetically, and functionally will be essential. A system that damages DNA but fails to kill every target cell could create complicated risks.

Finally, immune effects may be double-edged. Cas12a2-induced damage and inflammatory signalling might help expose tumours to the immune system, but uncontrolled inflammation could also create toxicity.

A new kind of programmable biology

The study presents Cas12a2 as a programmable RNA-triggered cell-killing platform. It is not merely another CRISPR editor. It is closer to a molecular trap: quiet until it hears the right RNA, destructive once activated.

That makes it both exciting and dangerous in the productive scientific sense. It is exciting because it gives researchers a way to remove cells based on transcriptional identity. It is dangerous because any technology built to kill cells must earn trust through rigorous specificity, delivery, and safety testing.

Still, the conceptual advance is clear. If Cas9 taught us to rewrite genomes, and Cas13 taught us to manipulate RNA, Cas12a2 may teach us how to eliminate cells by listening to what they express.

For cancer biology, virology, genome engineering, and synthetic biology, that is a serious new possibility.

When a Fungus Uses RNA as a Weapon: The Hidden RNA Battle Inside Rice Cells

 

Thernablog.blogspot.com 

How a fungal long non-coding RNA hijacks a rice microRNA and opens a new chapter in cross-kingdom RNA warfare

For years, plant immunity was explained mostly through proteins.

Plants detect pathogens using receptor proteins. Pathogens fight back using effector proteins. The battle was imagined as a molecular arms race between host immune receptors and pathogen-secreted protein weapons.

That picture is still true.

But it is no longer complete.

A new Nature study, “A pathogen lncRNA secreted into rice sequesters a host miRNA for virulence,” shows that a fungal pathogen can use a long non-coding RNA as an effector. Not a protein. Not a toxin. Not an enzyme. An RNA molecule.

The pathogen is Magnaporthe oryzae, the devastating fungus that causes rice blast disease. The host is rice. The weapon is a fungal long non-coding RNA called lnc117761. The target is a rice microRNA called miR5827. The outcome is suppressed rice immunity and enhanced fungal infection.

This is not just another plant pathology paper.

It changes how we think about host–pathogen communication.

The fungus is not only sending proteins into the plant. It is sending regulatory RNA.

The Old Model: Pathogens Attack With Proteins

Most classical models of plant immunity begin with recognition.

Plants detect conserved pathogen signals using pattern-recognition receptors at the cell surface. They also use intracellular immune receptors to recognize pathogen effectors. Pathogens, in response, secrete effectors to suppress immunity and create a more favorable environment for infection.

This protein-centered view has shaped decades of plant pathology.

But RNA biology has been quietly complicating the story.

Small RNAs are already known to move across kingdoms. Plants can send small RNAs into pathogens to silence virulence genes. Pathogens can send small RNAs into plants to suppress host immunity. This cross-kingdom RNA interference has become one of the most exciting areas in plant–microbe interaction research.

The new study goes further.

It suggests that long non-coding RNAs can also act as cross-kingdom pathogenic effectors.

That is a much bigger conceptual jump.

The Fungal RNA: lnc117761

The researchers searched for fungal long non-coding RNAs associated with rice infection by M. oryzae. They examined different fungal stages, including mycelium, conidiophore, appressorium, and invasive hyphae.

One RNA stood out.

lnc117761, a 1,589-nucleotide long non-coding RNA, was highly expressed during infection-related stages. When the researchers deleted lnc117761 from the fungus, pathogenicity dropped sharply. The mutant fungus could still grow and form appressoria, but it was defective in invasive hyphal development inside rice tissue.

That distinction matters.

The RNA was not simply helping the fungus grow. It was helping the fungus infect.

When rice plants were engineered to express lnc117761, they became more susceptible to blast disease, showing longer lesions and higher fungal biomass. This strongly suggested that lnc117761 behaves like an effector — except its active form is RNA.

The Rice Defense RNA: miR5827

The next question was obvious: what does lnc117761 do inside the host?

Long non-coding RNAs often act through base-pairing with other RNAs. So the researchers searched for rice RNAs that could pair with lnc117761. The strongest candidate was a rice microRNA called miR5827.

This rice miRNA normally promotes disease resistance.

Its job is to repress PKR1, a gene encoding a serine/threonine-protein kinase receptor that functions as a negative regulator of rice immunity. In simpler terms, PKR1 holds back defense. miR5827 suppresses PKR1, allowing rice immunity to work more effectively.

When miR5827 was knocked out, rice became more susceptible to blast disease. When miR5827 was overexpressed, rice became more resistant. Conversely, PKR1 knockout increased resistance, while PKR1 overexpression reduced resistance.

So the host pathway is clear:

miR5827 suppresses PKR1 → PKR1 repression strengthens immunity → rice resists fungal invasion.

The fungus has evolved a countermeasure.

The Trick: RNA Sponging

lnc117761 contains a binding site that pairs with rice miR5827. The researchers showed that miR5827 binds to the miR5827-binding site in lnc117761, but not to mutated versions of that site. They validated this interaction through multiple assays, including EMSA, rice protoplast assays, RNA pull-down, LIGR coupled with RT–qPCR, and ITC. The reported dissociation constant for miR5827–lnc117761 binding was 3.44 μM.

The biological meaning is elegant and unpleasant.

The fungal lncRNA acts as a sponge.

It enters the rice cell and sequesters miR5827. Once miR5827 is trapped, it can no longer efficiently repress PKR1. PKR1 expression rises. Rice immunity is weakened. The fungus gains ground.

The pathway becomes:

fungal lnc117761 binds rice miR5827 → miR5827 is neutralized → PKR1 is released → immunity is suppressed → fungal infection increases.

That is RNA warfare at the level of molecular decoys.

The fungus does not need to destroy the host RNA. It only needs to distract it.

The RNA Crosses Into Rice Cells

One of the most important parts of the study is that lnc117761 does not simply act inside the fungus. It is transported into rice cells during infection.

The researchers detected lnc117761 in infected and nearby uninfected rice cells. They used in situ hybridization to show lnc117761 signals in rice tissue, while control fungal mRNAs remained restricted to fungal cells. They also found lnc117761 enriched in extracellular vesicles from infection-related fungal hyphae, supporting an extracellular-vesicle-mediated secretion route.

They went further by tagging lnc117761 with a Pepper RNA fluorescence system. Red fluorescence from lnc117761–4×Pepper appeared in rice cells distinct from fungal GFP signal, supporting direct visualization of RNA delivery into host tissue.

This is where the paper becomes especially important.

It suggests that fungal extracellular vesicles may carry long non-coding RNAs into plant cells, where they manipulate host RNA regulatory circuits.

That is a remarkable form of infection biology.

A Tiny Binding Site With Big Consequences

The interaction depends on a short complementary region.

The lnc117761 sequence contains a 21-nucleotide miR5827-binding site. Within that region, the authors identified a 9-nucleotide core motif, GUUGCAACA, that is essential for binding and virulence. Mutations in this motif weakened or abolished miR5827 binding and reduced the ability of lnc117761 to promote disease.

This is biologically satisfying because it connects sequence, binding, and pathogenicity.

The paper does not merely show that the RNA is present. It shows that a defined RNA–RNA interaction matters.

Remove the binding site, and the effect is lost.

That is the difference between correlation and mechanism.

The Bigger Evolutionary Clue

The authors also examined whether the miR5827–lnc117761 binding motif is a one-off curiosity or part of a broader regulatory pattern.

They found related sequence features across microorganisms and plants. The study reports that portions of the 21-nucleotide binding site are commonly found in hundreds of microbial species and dozens of plant species. Similar miR5827-like RNAs were detected in several plants, including barley, Arabidopsis, potato, Brachypodium, and wheat.

This raises an intriguing possibility.

Some conserved non-coding DNA regions may encode regulatory RNAs that mediate biological interactions across species. In other words, parts of the genome once dismissed as “dark matter” may participate in host–pathogen recognition, defense, and counter-defense.

That does not mean every similar motif is functional. But it does mean the field should look more carefully.

The next effector may not be a protein.

It may be an RNA hiding in plain sight.

Can This Be Used for Disease Control?

The practical implications are significant.

The study extended the concept beyond rice blast. The researchers identified a related non-coding RNA in Rhizoctonia solani, the causal agent of rice sheath blight. Knocking down this RNA by exogenous small interfering RNA reduced pathogenicity. They also showed that miR5827 contributes to resistance against sheath blight.

The idea was also tested in wheat against Fusarium graminearum, the causal agent of Fusarium head blight. A synthetic wheat miR5827 mimic, called TamiR5827, improved resistance in wheat seedlings and spikes.

This makes the study especially relevant for RNA-based crop protection.

A few possible applications emerge:

Synthetic miRNA mimics could be developed as protective molecules.
Host miRNA alleles with stronger expression could be used in breeding.
Pathogen lncRNA sequences could become targets for spray-induced gene silencing.
Genome editing could be used to reduce negative immune regulators such as PKR1.
RNA delivery systems could be designed to reinforce host defense pathways.

For researchers working on dsRNA sprays, SIGS, RNAi biopesticides, and RNA-guided plant immunity, this paper opens a new direction. It suggests that we should not only target pathogen protein-coding genes. We should also look for pathogen non-coding RNAs that act as virulence factors.

Why This Study Matters for RNA Biology

This paper matters because it expands the definition of an effector.

In classical pathology, an effector is usually a pathogen-secreted protein that manipulates the host. Here, the effector is a long non-coding RNA that manipulates a host microRNA.

That is a different layer of biology.

It means host–pathogen interactions can be governed by RNA–RNA recognition, not only protein–protein or protein–DNA interactions. It also means that non-coding genomic regions can encode active molecules that directly influence disease outcomes.

For RNA biology, the lesson is even broader.

RNA is not merely a message.
RNA is not merely a target.
RNA can be a weapon.
RNA can be a decoy.
RNA can be a mobile effector.
RNA can decide whether immunity holds or collapses.

This is exactly why the RNA field is becoming so important.

The functional genome is larger than the protein-coding genome. And many of its most interesting instructions may be written in regulatory RNAs that we have not yet learned how to read.

The Necessary Caution

This is a powerful study, but it should not be overgeneralized too quickly.

The authors themselves note that the miR5827–PKR1 axis is a major mechanism but probably not the only mechanism by which lnc117761 promotes virulence. lnc117761 may sponge other plant miRNAs, interact with plant proteins, or regulate additional fungal processes.

It is also not yet clear how common this kind of long non-coding RNA effector mechanism is across pathogen–host systems. The paper provides an important example and a strong framework, but future comparative genomics, infection biology, RNA interactome mapping, and functional validation will be needed.

That is the correct way to read this study.

Not as the final answer.

As a door opening.

A New View of Plant–Pathogen Warfare

The image is striking.

A fungal pathogen enters rice tissue. It does not only push in with mechanical force. It does not only secrete enzymes or protein effectors. It sends an RNA molecule ahead, packaged through a secretion route, delivered into host cells, where it binds and neutralizes a host microRNA.

The host miRNA normally keeps an immune brake under control. The fungal RNA releases that brake. Disease follows.

This is not random molecular noise.

This is regulatory combat.

And it reminds us that the RNA world did not end when proteins evolved. It is still operating inside modern cells, inside pathogens, inside crops, and inside the invisible negotiations that decide whether a plant survives infection.

The next generation of plant disease control may come from understanding these RNA conversations.

Not only which genes are expressed.
Not only which proteins are secreted.
But which RNAs move, which RNAs bind, and which RNAs win.

---

Reference

He, M., et al., “A pathogen lncRNA secreted into rice sequesters a host miRNA for virulence.” Nature, 2026. https://www.nature.com/articles/s41586-026-10572-x

RNA Was Never Just a Messenger. Now It Looks Like an Architect.

 

Graphical Abstract

For years, RNA has lived in the public imagination as biology’s courier: DNA writes the instructions, RNA carries them, proteins do the work. That story was always too clean. Ribozymes, riboswitches, long non-coding RNAs, guide RNAs and viral RNA genomes have already shown that RNA can sense, catalyse, regulate and remember molecular states. But a new Nature study pushes the argument further. Some natural RNAs do not merely fold into useful shapes. They assemble into large, symmetrical, protein-free molecular architectures that look closer to engineered nanostructures than ordinary cellular transcripts.

The study, titled “Naturally ornate RNA-only complexes revealed by cryo-EM,” reports high-resolution structures of three unusually large bacterial RNAs: OLE, ROOL and GOLLD. These RNAs were already known from earlier comparative genomics as strange, elaborate non-coding RNAs with unusually ornate predicted secondary structures. Their biological functions, however, remained mostly mysterious. What Kretsch and colleagues found is striking: each RNA can assemble with copies of itself into a defined higher-order structure without requiring proteins as construction scaffolds.

The surprise hidden inside “non-coding” RNA

The phrase “non-coding RNA” often sounds like a negative definition: RNA that does not encode protein. But that label hides a more interesting possibility. If an RNA is not translated, it may instead act through its own shape. Its loops, stems, bulges, pockets and junctions may become the functional surface.

The problem is that most natural RNA structures are still unknown. The paper notes that although the RFAM database contains more than 4,000 RNA families, only a small fraction have experimentally solved tertiary structures. This gap matters because sequence alone rarely tells us what a large RNA can physically do. A long RNA may look like a tangled transcript on paper, yet in three dimensions it can become a pocket, a scaffold, a sensor or a cage.

That is where cryo-electron microscopy changed the story. Instead of treating these RNAs as abstract secondary-structure diagrams, the authors visualized them as physical objects.

OLE RNA: a dimer shaped like bundled pipes

The first RNA, OLE — short for Ornate Large Extremophilic RNA — comes from extremophilic bacteria and has been linked to stress adaptation, metal ion homeostasis, energy availability and drug-related responses. Earlier work suggested that OLE interacts with several proteins, which made it reasonable to suspect that proteins might be needed to stabilize its structure.

But the cryo-EM data showed that OLE can form a defined RNA-only dimer. Two RNA molecules come together into a compact structure shaped like parallel co-axial pipes. The interface is not casual. It is held by multiple RNA–RNA contacts, including unusual base-pairing and loop interactions. One particularly notable feature is an unusual symmetric A–A interaction between the two RNA chains.

This is important because it changes how we think about OLE-associated proteins. Instead of proteins forcing the RNA into shape, the RNA may already form a structured platform that proteins bind afterward. The paper also discusses possible binding sites for OapA, OapC and RpsU, suggesting that OLE may organize protein partners rather than simply being organized by them.

ROOL RNA: an eight-piece RNA nanocage

The second RNA, ROOL, is even more visually surprising. ROOL RNAs are found in bacterial prophages and phages, often near tRNA islands. Their function remains unknown, but they had been predicted to contain complex secondary structures and pseudoknots.

In the cryo-EM reconstruction, ROOL forms an octameric nanocage: eight copies of the RNA assemble into a closed, hollow structure with dihedral symmetry. Its diameter is about 280 Å, larger than the maximal dimension of a bacterial ribosome. Inside, the cage is largely empty except for a disordered linker region.

The ROOL structure is not just eight RNAs stuck together. Each RNA chain forms multiple contacts with neighboring chains. These contacts include A-minor interactions, kissing loops and other non-canonical RNA motifs. In other words, ROOL behaves like a self-assembling RNA object with repeated architectural rules. It looks less like a transcript and more like a molecular container.

GOLLD RNA: a larger, fourteen-part cage

The third RNA, GOLLD, goes further. GOLLD RNAs are among the largest members of the set studied here, with many examples exceeding 800 nucleotides. Like ROOL, they are associated with bacterial phages and prophages, but their sequences and predicted secondary structures are distinct.

The solved GOLLD structure forms a 14-subunit nanocage with D7 symmetry. Its diameter is about 380 Å, making it even larger than the ROOL cage. The structure resembles a closed shell assembled from RNA alone, with an empty interior except for a disordered linker. The authors describe the 5′ and 3′ regions as forming separate structural domains, helping explain why different regions of the RNA may evolve with different constraints.

This is not a minor structural curiosity. Natural RNA molecules forming large, symmetric, hollow cages without proteins is a remarkable biological design principle. Protein cages are familiar in biology: viral capsids, bacterial microcompartments, ferritin-like assemblies and other proteinaceous shells. RNA-only cages, especially natural ones of this scale and order, are much less expected.

Why this is probably not a laboratory artefact

Whenever a molecule forms an impressive structure in vitro, the skeptical question is obvious: does this happen in real biological contexts, or only under experimental conditions?

The authors address that concern directly. They solved another large RNA, the raiA motif, under similar cryo-EM conditions and found it as a monomer, arguing against the idea that any large RNA would automatically multimerize in their setup. They also tested another RNA, HEARO, which was disordered without its protein partner. These controls suggest that the OLE, ROOL and GOLLD assemblies are specific, not generic cryo-EM artefacts.

Additional biophysical evidence strengthens the case. Mass photometry confirmed the expected stoichiometries: OLE as a dimer, ROOL as an octamer and GOLLD as a 14-mer, even at RNA concentrations as low as 12.5 nM. Dynamic light scattering further showed that ROOL and GOLLD form thermostable multimers up to 55 °C, with no detectable monomer fraction under tested conditions.

The evolutionary evidence matters too. Sequence covariation analysis showed conservation not only of internal stems but also of sites involved in intermolecular interactions. That means evolution appears to preserve the very contacts that allow these RNAs to assemble with one another.

The old view of RNA is getting too small

This study does not claim that all large non-coding RNAs form nanocages. Nor does it fully solve what OLE, ROOL and GOLLD do in cells. The authors are careful on that point. Function remains the next hard question.

But the structural message is already clear: RNA can form natural quaternary architectures that rival proteins in symmetry and scale. OLE creates a defined dimeric platform. ROOL and GOLLD form hollow cages. Their architectures are stabilized by RNA-specific structural logic: kissing loops, A-minor motifs, pseudoknots, non-canonical base pairs and conserved intermolecular bridges.

The most exciting possibility is that these cages are not decorative. They may encapsulate or organize other molecules. The authors suggest that ROOL and GOLLD nanocages could potentially contain linkers, metabolites, proteins, tRNAs, ribosomal components or other macromolecules, although these ideas still need experimental proof.

Why RNA nanotechnology should pay attention

Synthetic biologists and RNA nanotechnologists have long tried to design RNA cages, rings, tiles and scaffolds. This paper shows that nature may already have evolved such objects, using motifs that can now be studied at near-atomic detail. These natural RNAs offer a library of structural strategies for building large RNA assemblies without proteins.

That matters for design. If researchers want RNA molecules that self-assemble, carry cargo, organize enzymes, sense conditions or build intracellular compartments, natural RNA cages may provide design rules that are more robust than purely artificial constructs. The same study also provides useful data for RNA structure prediction, because large RNA quaternary structures remain difficult for computational models.

The unfinished question

The most honest ending is not that “RNA has rewritten biology.” That would be too easy. Biology is not rewritten by one structure paper. But this study does something valuable: it widens the imaginable.

RNA is not only a messenger, regulator or catalytic relic from an ancient RNA world. In some organisms, it may be an architect. It can fold, pair, brace, dock, cage and assemble. It can build large molecular objects from nothing more than copies of itself.

The next question is no longer whether RNA can make ornate structures. It clearly can. The harder question is what these structures are doing inside the cell — and how many more are still hiding in genomic databases, mislabelled by our limited imagination as merely “non-coding.”

How experiments and AI are finally beginning to reveal the hidden architecture of RNA

  

Graphical Abstract

RNA Structure Is Entering Its Integration Era

RNA is not just a messenger. It is a negotiator, a sensor, a scaffold, a switch, a catalyst, and sometimes the quiet organizer of entire molecular conversations inside the cell.

For decades, biology treated RNA mainly as a carrier of genetic information: DNA made the instructions, RNA delivered them, and proteins did the work. That older view now feels almost naïve. Many RNAs do not merely transmit information. They perform it. They fold into shapes, bind proteins, sense metabolites, control gene expression, regulate translation, catalyze reactions, and help determine whether a cell responds, adapts, survives, or fails.

The catch is simple: RNA function depends heavily on RNA structure.

And RNA structure is difficult.

A recent Nature Biotechnology Perspective, “Integrated experimental and AI innovations for RNA structure determination,” captures where the field is now heading. Published in 2026, the article argues that RNA structure determination is being reshaped by two forces moving together: improved experimental technologies and rapidly advancing artificial intelligence-based modeling tools.  

The important word is together.

Experiments alone are powerful but limited. AI alone is fast but not always biologically grounded. The real breakthrough is emerging where the two meet.

RNA Is a Shape-Shifting Molecule

RNA structure is not a decorative detail. It is often the mechanism.

A messenger RNA may contain structural elements that influence stability or translation. A riboswitch changes shape when it binds a metabolite. A viral RNA genome can use folded regions to regulate replication or translation. A long noncoding RNA may act through structural modules that organize protein complexes. A guide RNA must adopt the correct form to work efficiently with CRISPR-associated proteins.

In each case, sequence matters, but sequence is not enough.

The same RNA sequence can fold into different conformations depending on magnesium ions, proteins, temperature, chemical modifications, ligand binding, cellular crowding, or stress conditions. RNA is also highly charged and flexible, which makes it harder to capture structurally than many more rigid biomolecules.

This is why RNA structure determination is so demanding. Researchers are not just trying to find a single beautiful 3D model. They are trying to understand a moving structural ensemble.

That movement is not noise.

It is biology.

Why Classical Experimental Methods Still Matter

Before AI can predict RNA structure confidently, experimental data must continue to define what is real.

Classical structural biology has given RNA researchers several powerful methods. X-ray crystallography can reveal atomic detail, but many RNAs resist crystallization because they are flexible and heterogeneous. NMR spectroscopy can capture local structure and dynamics, especially for smaller RNAs, but it becomes harder with large and complex molecules. Cryo-electron microscopy, or cryo-EM, has become increasingly important for large RNA-containing complexes and is now pushing into RNA-only structures with improved scaffolding and modeling approaches. The Perspective highlights multiple recent cryo-EM-related advances for RNA structure determination, including scaffold-enabled and high-resolution approaches. (Nature)

Cryo-EM has changed the mood of structural biology. Molecules that were once too large, too flexible, or too difficult to crystallize can now be visualized in new ways. For RNA, this is especially important because many functional RNAs do not behave like rigid objects. They fold, flex, bind, release, and remodel.

Still, cryo-EM is not magic. Small RNAs remain difficult. Flexible regions may blur. Model building can be challenging. Interpreting density maps requires skill, prior knowledge, and increasingly, computational assistance.

That is where AI begins to enter the story.

AI Is Not Replacing Experiments. It Is Extending Them

AI-based structure prediction has transformed protein biology, and RNA researchers are now asking whether RNA can have a similar moment. The answer is complicated.

Protein structure prediction benefited from massive structural datasets and strong evolutionary signals. RNA has less experimentally solved 3D structural data, fewer large training datasets, and more troublesome conformational behavior. RNA also depends strongly on noncanonical base pairs, ion-mediated interactions, electrostatics, and alternative folding states.

So RNA cannot simply borrow protein AI and call the problem solved.

But the field is moving fast.

Recent RNA modeling tools are using geometric deep learning, language-model-inspired strategies, transformer networks, end-to-end learning, and hybrid physics-plus-machine-learning methods. The Nature Biotechnology Perspective lists several recent computational advances, including trRosettaRNA, RhoFold, NuFold, RoseTTAFoldNA, AlphaFold 3, and other RNA or biomolecular modeling frameworks. (Nature)

These tools are not all doing the same thing. Some predict RNA tertiary structures from sequence. Some work better with secondary structure inputs. Some model protein–RNA complexes. Some help rank predicted structures. Some assist with cryo-EM map interpretation. Together, they show a field in rapid expansion.

The key lesson is this: AI is becoming useful not because it replaces structural biology, but because it can generate hypotheses faster than experiments alone.

A predicted structure can guide mutagenesis.
A predicted fold can help interpret chemical probing data.
A predicted RNA–protein interface can direct biochemical validation.
A predicted conformer can suggest what state a molecule might adopt under specific conditions.

The model is not the answer. It is the beginning of a better experiment.

The Power of Integrative RNA Structure Determination

The strongest future for RNA structure biology is not “experiment versus computation.” It is integrative modeling.

In an integrative workflow, researchers may combine:

RNA sequence
secondary structure prediction
chemical probing data
evolutionary conservation
cryo-EM density maps
NMR constraints
known RNA motifs
deep learning models
physics-based refinement
molecular dynamics simulations

Each method contributes a different kind of truth.

Chemical probing can reveal which nucleotides are exposed or constrained. Cryo-EM can provide global architecture. AI can propose plausible folds. Evolutionary information can identify conserved interactions. Physics-based refinement can improve stereochemistry. Functional assays can test whether the proposed structure actually matters.

This integration is especially valuable because RNA is dynamic. A single method may capture one state, while another reveals a different one. A computational model may predict a possible conformer, while experimental probing can test whether that conformer exists under biological conditions.

For RNA, structure determination must increasingly become ensemble-aware.

The question is not only:

What does this RNA look like?

It is also:

What shapes can this RNA adopt, and which shape controls function?

Cryo-EM and AI Are Becoming Partners

One of the most exciting developments is the pairing of cryo-EM with AI-assisted model building.

Cryo-EM can generate density maps, but interpreting those maps for RNA is not always straightforward. RNA bases, sugar-phosphate backbones, modified nucleotides, and flexible regions can be difficult to assign correctly, especially when resolution varies across the molecule.

AI tools can help automate parts of this process. The Perspective cites recent methods such as CryoREAD for de novo nucleic acid modeling from cryo-EM maps and other automated approaches for detecting and modeling DNA/RNA in cryo-EM density. (Nature)

This matters because scalability has been a bottleneck.

If every RNA structure requires slow, highly specialized manual interpretation, progress remains limited. If AI-assisted pipelines can accelerate model building while maintaining accuracy, RNA structural biology becomes more accessible to more laboratories.

That could change the field.

Not every lab can spend years solving one RNA structure. But more labs may be able to combine probing, cryo-EM, prediction, and validation into a practical workflow.

Why This Matters for Biotechnology and Medicine

RNA structure is not only an academic puzzle. It is becoming a design problem.

RNA therapeutics are expanding rapidly: mRNA vaccines, siRNAs, antisense oligonucleotides, guide RNAs, aptamers, ribozymes, circular RNAs, self-amplifying RNAs, and RNA-based sensors all depend on folding behavior. A therapeutic RNA may fail not because its sequence is wrong, but because its structure is unstable, inaccessible, immunogenic, or incompatible with delivery.

For RNA drug discovery, structure matters even more.

Small molecules do not bind “RNA” in the abstract. They bind pockets, grooves, motifs, exposed bases, dynamic states, and structured regions. If we cannot see or predict those structural features, rational RNA-targeted drug design remains limited.

Better RNA structure determination could help researchers:

design more stable therapeutic RNAs
optimize guide RNA folding
identify druggable RNA motifs
predict RNA–protein interactions
engineer riboswitches and aptamers
design RNA nanostructures
understand viral RNA regulatory elements
improve RNA delivery and expression systems

This is where RNA structure becomes translational.

The future of RNA medicine will not depend only on writing better sequences. It will depend on designing molecules that fold into the right structures at the right time in the right cellular environment.

The Remaining Challenges Are Serious

There is reason for excitement, but not for hype.

RNA structure prediction still faces major limitations. Experimental RNA 3D structure datasets remain smaller than protein datasets. Many RNAs are flexible and context-dependent. Chemical modifications can alter folding. Cellular environments are difficult to reproduce. Protein binding can remodel RNA architecture. Magnesium and other ions can determine whether a predicted fold is realistic.

AI models may also produce confident-looking structures that are wrong.

That danger is not trivial. A beautiful model can mislead if it is not tested. RNA structural biology cannot afford blind trust in prediction tools, especially when the results are used for therapeutic design or mechanistic claims.

The correct attitude is not skepticism for its own sake.

It is disciplined validation.

Predictions should be challenged by experiments. Experiments should be interpreted with computational help. Models should be treated as hypotheses, not final truth.

The Field Is Moving Toward Structure-Guided RNA Biology

The deepest shift is conceptual.

RNA biology is moving from sequence annotation toward structure-guided understanding. Researchers increasingly want to know not only where an RNA is expressed, but how it folds, what it binds, how it changes shape, and whether that shape can be engineered.

This shift will influence many fields: virology, cancer biology, neuroscience, plant biology, synthetic biology, diagnostics, and therapeutic design.

For RNA researchers, the opportunity is enormous.

We can now imagine a future where an RNA sequence is analyzed not only for open reading frames, motifs, or target sites, but also for structural states, ligandable pockets, conformational switches, protein-binding surfaces, and engineering potential.

That future will not come from AI alone.

It will come from the integration of experimental structure determination, biochemical probing, deep learning, physical modeling, and functional validation.

RNA is a social molecule inside the cell because it does not act alone. It interacts, responds, recruits, folds, and refolds.

Perhaps the tools we use to study it must behave the same way.

Not one method.
Not one model.
Not one structure.

A network of evidence.

That is where RNA structure determination is going.

And that is why this field now feels less like a technical subdiscipline and more like one of the central engines of modern RNA biotechnology.

---

References / Sources

1. Wang, W., Su, B., Peng, Z., & Yang, J. “Integrated experimental and AI innovations for RNA structure determination.” Nature Biotechnology, 44, 205–214, 2026. Published 5 January 2026. DOI: 10.1038/s41587-025-02974-5. (Nature)

Why AlphaFold transformed protein biology, while RNA structure prediction remains one of biology’s most stubborn frontiers

 

RNA Function Follows Form — But RNA Refuses to Sit Still

At a virtual conference in November 2020, structural biology changed.

The winner of the CASP14 protein-structure-prediction challenge was announced: AlphaFold, developed by Google DeepMind. The result was not merely better than previous tools. It was dramatically better. AlphaFold showed that artificial intelligence could predict many protein structures with near-experimental accuracy, solving a problem researchers had been chasing for decades.

Protein biology had its revolution.

RNA biology is still waiting for its equivalent moment.

That is the central tension behind Diana Kwon’s Nature Technology Feature, “RNA function follows form – why is it so hard to predict?” The article captures a frustrating truth: RNA is biologically essential, structurally fascinating, and increasingly important for medicine — but predicting its shape remains far harder than many people expected. 

RNA Is Not Just a Messenger

For decades, RNA was introduced in textbooks as a middleman: DNA stores genetic information, RNA carries the message, and proteins do the real work.

That explanation is now painfully incomplete.

RNA can regulate genes, catalyze reactions, guide protein complexes, control splicing, sense metabolites, organize cellular machinery, and influence disease. Ribozymes, riboswitches, long noncoding RNAs, microRNAs, guide RNAs, viral RNAs, circular RNAs, and therapeutic RNAs all remind us that RNA is not passive.

RNA does things.

But RNA does those things because it folds.

Its biological function depends on stems, loops, bulges, pseudoknots, junctions, long-range contacts, base stacking, ion coordination, and interactions with proteins or small molecules. In RNA biology, structure is not decoration. Structure is often the mechanism.

That is why the phrase “function follows form” matters.

An RNA molecule’s sequence tells us what it could become. Its structure tells us what it is actually capable of doing.

Why Proteins Were Easier for AI

AlphaFold’s success in protein structure prediction was built on several advantages.

Proteins have been studied structurally for a long time. Thousands of high-quality protein structures were already available in the Protein Data Bank. Protein sequences also carry rich evolutionary information: if two residues change together across evolution, they may physically interact in the folded protein. AlphaFold and related tools learned from these patterns at massive scale.

RNA does not offer the same easy path.

There are far fewer experimentally solved RNA structures than protein structures. Many RNAs are small, flexible, chemically sensitive, and structurally heterogeneous. RNA often depends on magnesium ions, cellular proteins, modifications, ligand binding, and environmental conditions to fold correctly. A sequence may not point to one stable structure. It may point to a shifting ensemble.

That makes RNA a much harder target for machine learning.

A protein often behaves like a molecule trying to reach a stable folded state.

RNA often behaves like a molecule negotiating among several states.

RNA’s Flexibility Is the Problem — and the Biology

The biggest mistake is to think RNA structure prediction is simply protein structure prediction with different building blocks.

RNA has its own grammar.

It has only four standard bases, but those bases can form canonical and noncanonical interactions. Its phosphate backbone is highly charged. Its folding can depend strongly on ions. It can form alternative secondary structures. It can switch conformations when binding a metabolite or protein. It can expose or hide regulatory regions depending on context.

This is not just a computational nuisance.

RNA flexibility is often exactly how RNA works.

A riboswitch must change shape to regulate gene expression. A viral RNA may remodel itself during infection. A guide RNA must fold into a form compatible with its protein partner. An mRNA may contain structural elements that affect translation, degradation, or immune recognition.

So the goal is not always to predict one “correct” RNA structure.

The goal may be to predict a population of possible structures, then understand which one matters under a specific biological condition.

That is much harder.

AlphaFold 3 Helps, But It Does Not End the RNA Problem

AlphaFold 3 expanded the modeling landscape by predicting biomolecular complexes involving proteins, DNA, RNA, small molecules, ions, and modified residues. That is a major advance because biology rarely happens molecule by molecule in isolation. Cells are crowded with interacting systems. 

But RNA structure prediction is still not solved.

AlphaFold 3 can model RNA-containing complexes, but RNA-only folding and RNA conformational ensembles remain difficult. Many RNA structures depend on experimental constraints, secondary structure priors, or specialized RNA-focused modeling approaches. The RNA field is therefore not simply waiting for one universal model to solve everything.

It is building a different toolkit.

The New RNA AI Toolkit

Several AI-based methods are now pushing RNA structure prediction forward.

Tools such as RhoFold+ use RNA language models and deep learning to predict RNA 3D structures from sequence. RhoFold+ was trained with large-scale RNA sequence information and designed to address the data scarcity that limits RNA modeling. 

Other methods, including trRosettaRNA and trRosettaRNA2, use RNA-specific structural logic, secondary structure information, and deep learning to improve 3D prediction. Recent work on trRosettaRNA2 emphasizes the value of secondary-structure-aware modeling and conformer prediction — a crucial feature because RNA often exists in multiple structural states rather than one fixed architecture. 

These methods suggest an important principle:

RNA prediction will probably not be solved by sequence alone.

It will require secondary structure priors, evolutionary signals, experimental probing data, cryo-EM maps, chemical constraints, molecular simulations, and AI models working together.

Experiments Still Matter

The rise of AI does not make RNA experiments obsolete.

It makes them more important.

Chemical probing methods such as SHAPE and DMS-based approaches can reveal which nucleotides are flexible, paired, exposed, or protected. Cryo-electron microscopy can capture larger RNA-containing complexes. NMR can reveal dynamics and local structure. X-ray crystallography can still provide atomic detail when crystals are available.

Each method sees a different part of the RNA story.

AI can propose models quickly. Experiments can test whether those models are real.

This is where RNA structure biology is heading: not toward purely computational prediction, but toward integrative structure determination.

A model becomes more trustworthy when it agrees with probing data, mutational analysis, cryo-EM density, biochemical function, and evolutionary conservation.

For RNA, evidence must converge.

Why This Matters for Medicine

RNA structure prediction is not just a technical problem for structural biologists. It is becoming central to biotechnology and medicine.

mRNA vaccines, siRNA drugs, antisense oligonucleotides, CRISPR guide RNAs, aptamers, ribozymes, circular RNAs, and self-amplifying RNAs all depend on folding behavior. A therapeutic RNA may fail because it folds incorrectly, exposes the wrong region, activates unwanted immune sensors, degrades too quickly, or binds inefficiently.

RNA-targeted small-molecule drugs are another major frontier. For a drug to bind RNA selectively, the RNA must present a recognizable structural pocket or motif. Without structural knowledge, RNA drug discovery becomes guesswork.

Better RNA structure prediction could help researchers design more stable RNAs, improve guide RNA performance, identify druggable RNA motifs, engineer riboswitches, and understand viral RNA elements.

In short, RNA structure prediction is not only about seeing molecules.

It is about designing them.

The Real Lesson from AlphaFold

AlphaFold changed protein biology because it made high-quality structural models widely accessible. It did not eliminate experiments, but it changed where experiments begin.

RNA needs a similar shift.

But the RNA version of that revolution may look different. It may not be a single model that predicts one final structure from sequence. It may be a network of tools that predicts secondary structures, tertiary folds, alternative conformers, RNA–protein complexes, RNA–ligand interactions, and experimentally testable structural hypotheses.

RNA does not sit still.

So RNA structure prediction must become comfortable with motion.

The next breakthrough will not simply tell us, “Here is the RNA structure.”

It will tell us:

Here are the structures this RNA can adopt. Here is when they appear. Here is what they bind. Here is how they regulate biology. Here is how we can redesign them.

That is the future RNA biology is moving toward.

Protein structure prediction had its AlphaFold moment.

RNA’s moment may be harder, slower, and messier.

But it may also be more interesting — because RNA is not merely a molecule with a shape.

It is a molecule with possibilities.

---

References / Sources

1. Kwon, D. “RNA function follows form – why is it so hard to predict?” Nature, 2025. (Nature)

2. Jumper, J. et al. “Highly accurate protein structure prediction with AlphaFold.” Nature, 2021. (Nature)

3. Abramson, J. et al. “Accurate structure prediction of biomolecular interactions with AlphaFold 3.” Nature, 2024. (Nature)

4. Shen, T. et al. “Accurate RNA 3D structure prediction using a language model-based deep learning approach.” Nature Methods, 2024. (Nature)

5. Wang, W. et al. “The trRosettaRNA server for RNA structure prediction.” Nature Protocols, 2026. (Yang Lab)

6. CASP16 RNA structure prediction assessment and Yang-Server/trRosettaRNA2 reporting. (Wiley Online Library)

RNA Folding Is Not Just Shape: The Principles That Make RNA Predictable

 

RNA is often introduced as DNA's messenger, a disposable copy of genetic instructions. That picture is far too small. RNA can switch genes on and off, guide enzymes to genomic targets, catalyze reactions, scaffold protein assemblies, sense metabolites, and carry vaccine instructions into cells. It does these jobs not only through its sequence, but through the structures that sequence folds into.

That makes RNA folding one of biology's most useful prediction problems. If we can predict how an RNA molecule folds, we can begin to predict how it behaves. If we can design a sequence that folds into a chosen structure, we can build RNA tools for medicine, diagnostics, synthetic biology, and nanotechnology. The challenge is that RNA is not a rigid object. It is a restless molecule moving across an energy landscape, with useful structures competing against near-misses.

The Basic Rule: Pairing Creates Structure, But Energy Chooses The Fold

RNA is built from four bases: A, U, G, and C. The familiar base-pairing rules, A with U and G with C, allow a single RNA strand to fold back on itself. Stems form where complementary regions pair. Loops, bulges, internal loops, and junctions form where pairing is interrupted.

But the final fold is not chosen by base-pairing alone. It is chosen by the balance of free energy across all possible structures. A predicted "minimum free energy" structure is the one a model estimates to be most stable. Stacked base pairs usually stabilize RNA. Large loops, unstable junctions, weak stems, or awkward local motifs can destabilize it. Magnesium ions, temperature, proteins, ligands, chemical modifications, and the cellular environment can all shift the balance.

So the first principle is simple but powerful: RNA folding is competitive. The target fold must be more favorable than the alternative folds the same sequence can make.

The Second Rule: Local Motifs Can Make Or Break A Design

The paper attached to this prompt, Anderson-Lee et al.'s "Principles for Predicting RNA Secondary Structure Design Difficulty," focused on inverse folding: given a desired RNA secondary structure, can we find a sequence that folds into it? The study drew on Eterna, a citizen-science RNA design platform, where tens of thousands of players and multiple algorithms tested what makes RNA designs easy or difficult.

Their results show why folding prediction is also a design problem. Some target structures are easy to specify on paper but hard to realize in a real sequence. Short stems are a classic example. A two-base-pair stem may look harmless in a diagram, but it offers only a small number of stable sequence choices. If many short stems appear in the same design, the sequence often needs repeated mini-patterns, and repeated patterns can mispair with one another.

Bulges and internal loops create another problem. They interrupt stacking interactions, weakening the stem and making nearby alternative folds more competitive. Multiloops, where several stems meet, require careful tuning of closing base pairs and nearby loop energies. Zigzag-like arrangements of opposing bulges are especially difficult: they can make an otherwise straightforward RNA hard for algorithms to design.

This leads to a practical design rule from the Eterna community: the "principle of least elements." The fewer destabilizing or difficult motifs a target structure contains, the more likely it is to be designable.

The Third Rule: Symmetry Is Beautiful, But Dangerous

Human designers like symmetry. RNA often does not.

In RNA design, repeated stems, repeated loops, and exact visual symmetry can be traps. Repetition narrows the usable sequence space and increases the chance that one part of the molecule will pair with the wrong partner. A symmetric diagram may invite misfolded alternatives that are nearly as stable as, or more stable than, the intended fold.

This is one reason natural RNAs often show broken symmetry. They may contain repeated domains, but the repeated parts are not usually exact copies at the secondary-structure level. Small asymmetries can help prevent incorrect pairing while preserving the broader biological function.

For real-world design, this is a quiet but important lesson: do not confuse structural elegance with molecular reliability. A slightly irregular RNA may be easier to make, easier to predict, and more robust in cells.

The Fourth Rule: Prediction Needs Ensembles, Not Just One Fold

Many beginner explanations of RNA folding focus on one predicted structure. In real biology, that is rarely enough. RNA molecules occupy ensembles: collections of structures with different probabilities. Some RNAs need one dominant structure. Others need to switch between states, as riboswitches do when they bind metabolites. Still others need to keep a region unpaired so a protein, ribosome, guide RNA, or reverse transcriptase can access it.

That means useful prediction asks several questions:

-          What is the most likely fold?

-          What alternative folds are close in energy?

-          Which nucleotides are likely to be paired or unpaired?

-          How often does the molecule expose a functional site?

-          How stable is the RNA against chemical degradation?

-          How does the fold change when proteins, ligands, ions, or modifications are present?

High-throughput experiments have become essential here. Chemical probing methods such as SHAPE and DMS can measure which nucleotides are flexible or accessible across thousands of RNA molecules. These datasets can reveal where thermodynamic models succeed, where they fail, and how machine-learning models can improve prediction.

Why This Matters For Biological Applications

RNA folding prediction is not an academic exercise. It affects whether RNA technologies work outside a diagram.

In gene silencing, siRNAs and shRNAs must present the right guide strand and avoid structures that block loading into cellular machinery. In CRISPR genome editing, guide RNAs must preserve the scaffold structures needed for Cas protein binding while keeping the targeting region accessible. In riboswitch and biosensor engineering, the RNA must change structure reliably when it binds a molecule. In RNA nanotechnology, repeated tiles, junctions, and short stems must assemble without generating unwanted mispaired products.

For mRNA therapeutics and vaccines, folding affects translation, immune recognition, and degradation. RNA is chemically fragile; unpaired and flexible regions can be more vulnerable to hydrolysis. Models that predict local structure and degradation patterns can help design mRNAs that last longer while still being translated efficiently.

The most promising real-world strategy is therefore not "predict the perfect fold once." It is an iterative loop:

1. Choose a target function.
2. Propose structures that obey known designability rules.
3. Use computational tools to predict folds, ensembles, accessibility, and degradation risk.
4. Test many candidates experimentally.
5. Feed the results back into improved models.

This is already happening. Eterna-derived work has used community-designed RNA datasets to benchmark and improve folding packages. OpenVaccine-style efforts have combined RNA design and machine learning competitions to predict RNA degradation. The future of RNA engineering will likely come from this blend of physical modeling, high-throughput measurement, human intuition, and machine learning. 

The principles governing RNA folding are not just chemical rules; they are design rules. Stable stems help. Awkward loops, short repeated stems, dense difficult motifs, and exact symmetry can hurt. The best RNA designs respect the whole folding landscape, not just the desired final picture.

That is why RNA prediction is becoming so valuable for biology. It lets scientists ask, before entering the lab, whether a proposed RNA is likely to fold, switch, expose, bind, silence, guide, translate, or survive as intended. The more accurately we can answer those questions, the more RNA becomes a programmable material for living systems.

Sources

Anderson-Lee, J. et al. "Principles for Predicting RNA Secondary Structure Design Difficulty." Journal of Molecular Biology 428, 748-757 (2016). https://doi.org/10.1016/j.jmb.2015.11.013

Wayment-Steele, H. K. et al. "RNA secondary structure packages evaluated and improved by high-throughput experiments." Nature Methods 19, 1234-1242 (2022). https://doi.org/10.1038/s41592-022-01605-0

Wayment-Steele, H. K. et al. "Deep learning models for predicting RNA degradation via dual crowdsourcing." Nature Machine Intelligence 4, 1174-1184 (2022). https://doi.org/10.1038/s42256-022-00571-8

RNA Structure Is Becoming the Next Big Frontier in Biology

 How experiments, AI, and integrative modeling are changing the way we understand RNA folding, function, and therapeutics

RNA Structure

For a long time, RNA was treated as biology’s messenger: DNA made the plan, RNA carried the instructions, and proteins did the real work.

That view is now far too small.

RNA is not just a passive courier of genetic information. It can fold into complex structures, catalyze biochemical reactions, regulate gene expression, control splicing, bind proteins, sense metabolites, and participate in disease processes. In many cases, RNA’s function depends not only on its sequence but on its shape.

That is why RNA structure has become one of the most important frontiers in molecular biology.

A review published in Nature Methods titled “Advances and opportunities in RNA structure experimental determination and computational modeling” captures this shift clearly: if we want to understand what RNA does, we must understand how it folds. The authors summarize modern experimental and computational tools for RNA secondary and tertiary structure analysis and argue that the growing volume of RNA structural data is opening the door to more powerful integrative modeling approaches. (Nature)

The message is simple but profound:

RNA biology is becoming structural biology.

Why RNA Structure Matters

RNA structure begins with base pairing.

Adenine pairs with uracil. Guanine pairs with cytosine. Sometimes guanine pairs with uracil. These interactions create stems, loops, bulges, junctions, pseudoknots, and long-range contacts. Together, they form what we call RNA secondary structure.

But RNA does not stop there.

Secondary structure folds into three-dimensional structure, where distant regions of the molecule come together in space. This 3D architecture determines whether an RNA can bind a protein, recognize a small molecule, act as an enzyme, regulate translation, or switch between active and inactive states.

This is where RNA becomes difficult.

Unlike many proteins, RNA is often highly flexible. It may not have one fixed structure. Instead, the same RNA molecule can exist as a population of conformers—different structural states that shift depending on ions, temperature, binding partners, chemical modifications, or cellular context.

That flexibility is not a technical nuisance. It is often the biology itself.

A riboswitch functions because it changes shape. Viral RNA genomes rely on structural elements for replication and translation. Long noncoding RNAs may act through modular structural domains. Therapeutic RNAs must fold correctly to remain stable and functional.

So, asking “What is the RNA sequence?” is no longer enough.

The better question is:

What structures can this RNA form, and which of those structures matter?

The Experimental Toolbox Is Expanding

RNA structure determination has traditionally relied on powerful but demanding experimental methods.

X-ray crystallography can provide high-resolution structures, but RNA molecules are often hard to crystallize. Their flexibility, charge, and conformational heterogeneity make crystal formation difficult.

Nuclear magnetic resonance spectroscopy, or NMR, is useful for studying RNA dynamics and smaller RNA structures, but it becomes more challenging as RNA size and complexity increase.

Cryo-electron microscopy, or cryo-EM, has become increasingly valuable for larger RNA-containing complexes, especially ribonucleoproteins. It allows researchers to visualize molecular assemblies that were once extremely difficult to capture structurally.

Alongside these classical approaches, high-throughput structure-probing methods have transformed the field. Techniques based on chemical probing, such as SHAPE and DMS-based approaches, can reveal which nucleotides are flexible, paired, exposed, or protected. These methods do not always give a full atomic structure, but they provide crucial experimental constraints for modeling RNA folding.

This is one of the most important changes in the field.

RNA structure determination is no longer limited to solving one purified molecule at a time. Researchers can now collect structural information across many RNAs, under different conditions, sometimes even inside cells.

That matters because RNA structure is context-dependent.

An RNA may fold differently in vitro and in vivo. It may change structure when bound to proteins. It may respond to stress, metabolites, ions, or chemical modifications. Experimental probing helps reveal these dynamic states rather than forcing RNA into a single static model.

Computational Modeling Is Catching Up

For decades, RNA secondary structure prediction relied heavily on thermodynamic models. These approaches estimate the most favorable base-pairing arrangement based on energy rules. They remain useful, especially for shorter RNAs and well-behaved structures.

But RNA folding is not always clean.

Long-range interactions, pseudoknots, tertiary contacts, alternative conformers, and cellular constraints can make structure prediction difficult. That is where computational modeling has had to evolve.

Modern RNA modeling now combines several layers of information:

Sequence conservation
Thermodynamic folding
Experimental probing data
Known structural motifs
Machine learning
Deep learning
3D geometry prediction
Molecular dynamics simulations

The strongest approaches are often integrative. They do not rely on only one source of information. Instead, they combine experimental constraints with computational prediction to generate more biologically realistic models.

This is exactly where the field is moving.

A major theme of the Nature Methods review is that experimental data and computational modeling are becoming increasingly connected. More experimental structural data improves computational tools. Better computational tools help interpret experimental data. Together, they create a feedback loop that can accelerate RNA structure discovery. 

AI Is Changing the RNA Structure Landscape

Protein structure prediction had its AlphaFold moment. RNA structure prediction is still more difficult, but it is rapidly advancing.

Recent AI-based methods are beginning to predict RNA secondary structure, tertiary structure, RNA–protein interactions, RNA–small molecule binding, and conformational states. These tools are helped by the rise of RNA language models, deep learning architectures, and larger biological datasets.

A 2023 review in Briefings in Bioinformatics highlighted how machine learning and deep learning are being applied to RNA secondary structure prediction, RNA aptamers, RNA–protein interactions, and RNA drug discovery. 

More recently, models such as trRosettaRNA2 have shown how secondary structure priors can guide RNA 3D structure and conformer prediction. The 2026 Nature Machine Intelligence article describing trRosettaRNA2 emphasizes that RNA 3D prediction remains difficult because experimental RNA structure data are scarce and RNA molecules are flexible, but structure-aware deep learning can improve prediction performance. 

This is important because RNA models cannot simply copy protein-prediction logic.

RNA has its own grammar.

Its folding depends heavily on base pairing, electrostatics, magnesium ions, noncanonical interactions, and alternative conformations. AI tools that understand or incorporate these features may be better suited for RNA than generic structure predictors.

Why This Matters for RNA Therapeutics

RNA structure is not just an academic problem.

It is becoming central to biomedical research.

mRNA vaccines, siRNAs, antisense oligonucleotides, guide RNAs, ribozymes, aptamers, circular RNAs, and self-amplifying RNAs all depend on RNA behavior. Stability, translation efficiency, immune recognition, intracellular delivery, target binding, and degradation can all be influenced by structure.

For therapeutic RNA design, structure prediction can help answer practical questions:

Will this RNA fold into the intended form?
Will it expose the right binding region?
Will it hide important motifs?
Will it form unwanted secondary structures?
Will it remain stable long enough to function?
Will chemical modifications disrupt or improve folding?

RNA structure also matters for RNA-targeted small-molecule drugs.

For many years, drug discovery focused mainly on proteins. RNA was often considered too flexible, too charged, or too difficult to target selectively. That view is changing. RNA can form structured pockets, motifs, and conformational states that small molecules may recognize.

A 2026 Nature Biotechnology article notes that RNA has emerged as an attractive target for drug discovery because its structures can be modulated by small molecules to influence processes such as splicing, translation, RNA–protein interactions, noncoding RNA processing, and RNA virus replication. 

This is where RNA structure becomes directly translational.

If we can map RNA structures accurately, we may be able to design drugs that bind RNA with the same strategic precision once reserved for proteins.

The Challenge: RNA Is Not Still

The biggest challenge is that RNA is dynamic.

Many methods still try to report one dominant structure. But biological RNA may exist as an ensemble. The “minor” conformer may be the functional one. A transient structure may control splicing. A rare state may expose a druggable pocket. A ligand may shift the entire folding landscape.

This creates a major challenge for both experiments and AI.

The future of RNA structure biology will not be only about predicting one beautiful model. It will be about predicting structural ensembles.

That means researchers will need tools that can capture alternative folds, kinetic traps, folding intermediates, ligand-bound states, protein-bound states, and cellular context.

This is also why experimental validation remains essential.

AI can generate models. Chemical probing can test accessibility. Cryo-EM can visualize large complexes. NMR can capture dynamics. Mutational analysis can test function. No single method is enough.

The best RNA structural biology will be hybrid.

The Future: Structure-Guided RNA Biology

The next phase of RNA research will likely be defined by integration.

Experimental probing will feed computational models. AI will generate structural hypotheses. High-throughput datasets will refine prediction tools. Molecular simulations will explore conformational flexibility. Drug discovery platforms will search for ligandable RNA motifs. Synthetic biologists will design RNA switches, sensors, scaffolds, and therapeutic constructs with structural logic built in from the beginning.

For plant biology, virology, cancer biology, neuroscience, and RNA therapeutics, this shift is significant.

RNA structure is no longer a specialist detail.

It is becoming a design principle.

We are entering a period where researchers will not only ask what an RNA sequence encodes, but what it folds into, how it moves, what it binds, and how that structure can be engineered.

That is the real opportunity.

The future of RNA biology will belong not only to those who can read RNA sequences, but to those who can understand RNA shapes.

And increasingly, those shapes will be discovered by experiments and predicted by machines working together.

---

References / Sources

1. Zhang, J., Fei, Y., Sun, L., & Zhang, Q. C. “Advances and opportunities in RNA structure experimental determination and computational modeling.” Nature Methods, 19, 1193–1207, 2022.  

2. PubMed entry for Zhang et al., “Advances and opportunities in RNA structure experimental determination and computational modeling.”  

3. Sato, K. “Recent trends in RNA informatics: a review of machine learning and deep learning for RNA secondary structure, aptamers and drug discovery.” Briefings in Bioinformatics, 2023. 

4. Wang, W., Peng, Z., & Yang, J. “Predicting RNA 3D structure and conformers using a pre-trained secondary structure model and structure-aware attention.” Nature Machine Intelligence, 2026.  

5. Fei, Y. et al. “Predicting small molecule–RNA interactions without experimentally resolved RNA structures.” Nature Biotechnology, 2026.