A model for propagation of RNA structural memory through biomolecular condensates

Abstract

We hypothesize that stress-induced RNA structural changes, stabilized by RNA-binding proteins in biomolecular condensates, propagate via conformational catalysis in a prion-like manner across generations. Our model suggests that RNA structure encodes heritable memory, and its roles should be explored in epigenetic inheritance, evolutionary adaptation and disease.

Epigenetic research has long emphasized the role of DNA and histone modifications, yet emerging evidence highlights the capacity of RNA to encode hereditary information across generations¹. Beyond their sequence-based functions, RNAs can adopt complex structures, particularly when interacting with proteins and other molecules to form ribonucleoprotein (RNP) complexes and biomolecular condensates². These interactions can reshape RNA conformation, offering a flexible mechanism to modulate RNA function and potentially encode heritable information if these structural states are maintained or propagated.

In this Comment, we propose a model of RNA structural memory propagation, which integrates three key ideas. First, RNAs switch between stable conformations in response to stress or changes in cell status, overcoming energy barriers through random refolding or catalysis. These alternative conformations, once stabilized, encode biological (heritable) information. Second, certain phase-separated condensates, such as nucleoli and stress granules, may lock in these stress-adapted RNA conformations and act as catalytic hubs to template the new structure onto nascent RNAs in a prion-like manner. Finally, this model also allows active and passive exit mechanisms, whereby stress-associated RNA structures may either gradually revert to their default state (active) or persist, yet the cells bearing them are outcompeted during selection after returning to a normal environment (passive). These processes can operate either independently or together to regulate the propagation and persistence of RNA structural memory, influencing cellular adaptability and evolutionary fitness.

Our model integrates evidence of epigenetic inheritance mediated by RNAs, prions and RNP complexes, proposing that RNA structural information, stabilized in biomolecular condensates, governs heritable phenotypic changes without altering DNA sequence.

RNA folding energy and the multistable landscape

Both large and small RNAs exhibit structural versatility, potentially folding into various conformations based on intrinsic energy and interactions with biomolecules, such as proteins and lipids². This context-dependent folding enables diverse functional adaptations, such as altered binding or catalytic activities. Recent studies highlight the ability of RNA to adopt different structures in response to environmental stressors, with these structures closely associated with or driving their functional diversity³.

Here, we propose that while a single RNA sequence can adopt multiple metastable folding structures³, it can switch between two (or more) stable folding conformations driven by specific cellular environments, with each linked to distinct functional properties. This process resembles cellular differentiation in Waddington’s epigenetic landscape, in which cells settle into stable ‘valleys’ representing unique fates⁴. However, to switch between different stable forms, a substantial energy barrier needs to be overcome, as each stable form represents an energy valley (Fig. 1).

Fig. 1 |. RNA structural switching and energy landscape.

a, RNA structural switching from structure A (blue) to structure B (red) under random or catalysed conditions. Stress triggers the generation of multiple random structures, which can then be subject to selection (top). Alternatively, a structural switch can be mediated by catalysis (bottom), similar to how molecular chaperones assist in protein folding through specific proteins or RNAs. b, Illustration of an energy landscape with a high barrier between RNA structure A (valley 1) and structure B (valley 2). A random transition over the peak (blue arrow) is rare, whereas catalysis (red arrow), illustrated as tunnelling though the bottom of the peak, lowers the energy barrier, enabling the transition with reduced energy cost and higher efficiency. We depict two stable structures (A and B) for simplicity, although in reality, a single RNA sequence can adopt multiple stable conformations depending on its sequence and RNA modifications. Created in BioRender. Yu, J. (2025) https://BioRender.com/0si1dxr.

When an RNA adopts a default conformation under standard cellular conditions, it represents the minimal free energy state and is often stabilized by RNA-binding proteins (RBPs). However, environmental stressors, such as heat, oxidative stress or nutrient deprivation, can alter the cellular environment, triggering RNA refolding even in highly structured RNAs, for example, tRNAs⁵. Under these altered conditions, changes in local factors (for example, pH, ion strength, temperature and binding partners) may produce multiple intermediate RNA forms. Although most intermediate forms are transient, a small percentage may stabilize into new conformations that are adaptive to the changed conditions (Fig. 1). Alternatively, catalytic activities can lower the energy barrier needed for switching between RNA structures (Fig. 1), similar to chaperone-assisted protein folding.

Hypothetically, if an alternative RNA structure formed under stress (or triggered by a cellular state transition such as during development or cancer) is stabilized by protein and/or RNA interactions, the RNA could become locked into this new structure even after the initial stressor fades. This structural ‘locking’ may enable RNAs to retain a molecular memory of ancestral stress, potentially driving stress-adaptive phenotypes through the newly stabilized structure.

Biomolecular condensates as possible propagation hubs of prion-like heritability

At the heart of this RNA structural memory propagation hypothesis lies the key question of how RNA structures persist and replicate across generations. This requires a mechanism in which a structured RNA catalyses the folding of nascent RNAs with an identical sequence into the same conformation, enabling the newly folded RNA to catalyse the folding of additional nascent RNAs. Conceptually, this resembles the idea of protein-based prion propagation, whereby misfolded proteins can induce other proteins to misfold⁶.

However, relying solely on structured RNA for catalysis poses challenges: the catalytic RNA molecule (referred to here as catalyst RNA) risks structural alteration during RNA–RNA base pairing, potentially losing its catalytic activity, and controlling its interaction with nascent RNA in the cellular environment is difficult without additional support. To overcome these limitations, our model incorporates biomolecular condensates and RBPs to address these issues (Fig. 2).

Fig. 2 |. RNA structural replication through RNA- and RNP-mediated conformational catalysis in condensates.

The schematic illustrates the hypothesized structural replication of RNA through conformational catalysis within a biomolecular condensate that is enriched in ions, metabolites and lipids, creating a crowded molecular environment that promotes RNA–RNA and RNA–protein interactions. Within this condensate resides a catalytic RNA, stabilized by RBPs into an RNP. When a nascent RNA — sharing the same sequence and structural potential as the catalytic RNA — enters the condensate, it engages with the catalytic RNA via transient base-pairing. This interaction guides the nascent RNA to fold into a conformation matching that of the catalytic RNA, a process facilitated by local small molecules and ions, similar to riboswitch modulators. In a ‘kiss-and-run’ scenario, the newly folded RNA is released from the condensate and can bind to RBPs to form new functional condensates. Throughout this process, the RBPs stabilize and constrain the mobility of the catalytic RNA, preventing excessive misfolding and structural collapse during its interactions with nascent RNAs and maintaining its functional conformation after the kiss-and-run cycle. Created in BioRender. Yu, J. (2025) https://BioRender.com/vi35tg9.

Condensates are membrane-less organelles, such as stress granules and nucleoli, formed through liquid–liquid phase separation, which provide a high local concentration of RNAs and proteins, increasing interaction frequency and efficiency⁷. Meanwhile, potentially stress-responsive (or sensitive to change of cell states during development or diseases) RBPs, such as those in stress granules, restrict the molecular mobility of the catalyst RNA, stabilizing its catalytic conformation and ensuring functional stability (Fig. 2). Moreover, ions and metabolites enriched within condensates may lower energetic barriers for conformational transitions in nascent RNAs — akin to metabolite effects on riboswitch structures — facilitating the templated replication of the catalyst RNA conformation (Fig. 2). These factors may drive the kinetics of the folding process by stabilizing favourable conformations, with sequence-specific base pairing guiding catalyst RNA interactions with the nascent RNA.

When the newly folded RNAs are released from the condensate, they can again bind to stress-specific RBPs and assemble additional condensates, catalysing more nascent RNAs to amplify the RNA structural memory (Fig. 3). This amplification model also enables propagation during cell division; once daughter cells inherit these condensates⁸ they initiate further cycles of RNA structural replication and condensate formation, thus propagating memory to descendant cells (Fig. 3).

Fig. 3 |. The propagation cycle of RNA structural memory within cells and between generations.

Depicted here is a biomolecular condensate with catalytic RNA conformation (red) and stress-specific (or cell state-specific) RBPs, forming an RNP, which facilitates transformation of a nascent RNA (grey) into the same catalytic conformation (red). The newly folded RNA again binds the stress-associated RBPs to assemble new catalytic RNP-containing condensates, which are inherited by daughter cells during cell division. This recurring cycle in descendant cells enables long-term propagation of the RNA conformation, supporting cross-generational transmission of condensate-associated phenotypes (for example, stress-adapted traits). Created in BioRender. Yu, J. (2025) https://BioRender.com/bm3yqcu.

Our model diverges from the classical ‘RNA-world’ hypothesis, which emphasizes that RNA could copy its own linear sequence independently of proteins. Instead, we propose a prion-like structural replication process, wherein catalytic RNA directly drives conformational changes in nascent RNAs, supported by RBPs and condensates.

Importantly, the RNA structural memory propagation model could be important in influencing evolutionary fitness and disease susceptibility, if the different RNA structures are associated with different functional readouts. Indeed, recent evidence shows that synonymous mutations in RNA can generate diverse RNA structures, which in turn dictate the properties of biomolecular condensates and their associated functions⁹, supporting the idea that RNA structural memory, once established, can drive functional diversity. When the self-perpetuating system enables the persistence of functional RNA structures in descendant cells, it offers a potential new framework for understanding RNA-based heritability.

Potential exit mechanisms that regulate RNA structural memory

A key observation in transgenerational epigenetic inheritance (that is, not involving DNA mutation) is that adaptive phenotypes gradually decline after the removal of the initial environmental stressor¹. This suggests the existence of exit mechanisms during generational propagation, which remains underexplored. In our model, we propose two potential exit mechanisms, active and passive, which regulate RNA structural memory to enhance an organism’s fitness in fluctuating environments.

The active mechanism involves rapid adaptation, allowing RNA structures to switch conformations in response to environmental shifts, such as a return to normal conditions after stress (or change to another cell state). This shift alters the cellular environment, for instance changes in chemical concentrations (for example, pH, ion levels and metabolites) within subcellular compartments and condensates, along with decreased levels of stress-associated RBPs. These changes may destabilize stress-associated biomolecular condensates, allowing RNAs to refold and alter their binding affinity towards RBPs² prevalent under normal conditions, thus losing their catalytic activity but adopting alternative functions optimized for non-stress cellular states (Fig. 4a).

Fig. 4 |. Active and passive exit mechanisms regulating RNA structural memory.

a, Active mechanism. When the stress environment returns to normal, changes in the cellular environment, such as decreased levels of stress-associated RBPs (dashed line) and altered condensate environment (for example, pH or ion levels), destabilize the stress-associated condensate. This allows RNA to refold (grey) and bind with new RBPs (green), forming normal RNP complexes. b, Passive mechanism. After the environment returns to normal, RNA structural memory persists in descendant cells (beige) across generations but renders them less adaptable (for example, keeping the stress-adapted slower growth rate) in the normal environment. These stress-adapted cells are gradually outcompeted by cells with normal RNP complexes (green), leading to their eventual decline over multiple generations. Created in BioRender. Yu, J. (2025) https://BioRender.com/j8a8d5g.

Alternatively, we propose a passive mechanism in which stress-associated RNA structures might persist in a subset of cells even after the environment returns to normal, with these structures stabilized within biomolecular condensates to maintain stress tolerance. However, maintaining these stress-adaptive RNA structures often comes at the expense of slower growth rates, a common trade-off in unicellular organisms, such as bacteria and yeast. Over generations, cells containing these stress-responsive condensates are gradually outcompeted by cells with normal RNA structures in unstressed environments, resulting in a population shift (Fig. 4b).

These two exit mechanisms can co-exist, depending on the species and reproductive cycle, shaping evolutionary strategies and disease outcomes, as discussed next.

Functional implications for evolution and diseases

In unicellular organisms, such as bacteria and yeast, the formation and persistence of stress-adapted RNA structures enable a subset of cells to endure environmental stressors including heat or starvation, albeit at the cost of reduced growth rates compared with normal conditions. When the environment returns to normal, some cells in the population may exit the RNA structural memory state through active and/or passive mechanisms (see above). This results in a heterogeneous population: some cells retain stress-adapted RNA structures, conferring stress tolerance but slower growth, while others revert to a state optimized for normal conditions, enabling faster growth. This constitutes a bet-hedging strategy without altering the genomic sequence, preparing the population for environmental fluctuations — cells retaining stress-adapted RNA structures can rapidly adapt to recurring stress, whereas those that have exited the memory state are primed for rapid growth in normal conditions.

In mammals, development begins with a single fertilized egg (zygote) that develops into an embryo and differentiates into various tissues and organs. If RNA structural memory forms in the zygote and persists, it can pass down to all cells, shaping the entire organismal traits. However, mosaicism — whereby cells within the same individual differ — can arise through at least two ways: some daughter cells may activate an exit mechanism (for example, condensate destabilization and RNA refolding) that removes the RNA memory, or condensates carrying this memory may be unevenly distributed during cell division (for example, one daughter cell inherits two condensates while the other gets none). This leads to distinct RNA structural states in different lineages, resulting in tissue-specific traits and varying disease susceptibility within an individual.

Practically, RNA structural memory in the embryo could be shaped by factors including RNAs from sperm. For example, sperm tRNA-derived small RNAs and other non-coding RNAs, influenced by environmental exposures, may engage with nuclear RNPs and condensates in the embryonic nucleoli where ribosome biogenesis takes place¹⁰. We hypothesize that these sperm-derived RNAs may affect ribosomal biogenesis by generating structurally fine-tuned rRNAs and ribosomes, favouring the translation of disease-related mRNA pools and predisposing offspring to conditions such as metabolic disorders.

Our model may extend beyond epigenetic inheritance, applying to altered cellular states in diseases such as cancer and neurodegeneration, in which abnormal RNA structures may associate with or drive pathological cell behaviour. Analogous to Waddington’s epigenetic landscape, in which cells settle into stable fates represented as valleys, RNA structural memory shapes these fates by locking cells into specific functional states. For instance, in cancer, these structures may promote oncogenic protein translation, fixing cells in proliferative states.

Perspectives and future validation

Our model for RNA structural memory propagation integrates RNA structural dynamics, phase separation and epigenetic inheritance, expanding beyond the traditional sequence-centric view of RNA in heredity, development and disease. Future validation requires experiments to track RNA structural changes, map interactions in biomolecular condensates and confirm heritability across generations. This could involve creating a traceable model to observe epigenetic inheritance under stress, followed by phenotypic fading in subsequent generations, such as using high-throughput sequencing to track transcriptome-wide RNA conformational changes linked to phenotypic shifts. Additionally, live-cell imaging to visualize condensate-stabilized RNA structures and disrupting condensates with chemical or genetic methods will further test their role in RNA memory propagation.

If validated, this framework could complement existing epigenetic inheritance mechanisms^1,10 by emphasizing a structure-centric view, whereby RNA shape, beyond its sequence, drives information propagation and phenotypic memory, extending the current understanding of genetic and epigenetic inheritance¹. Similarly, tracking RNA structural changes during cell differentiation, cancer progression and neurodegenerative diseases could illuminate the roles of RNA structure in triggering and/or sustaining these processes, potentially guiding novel therapeutic approaches by targeting RNA structure.

Intriguingly, there is increasing evidence highlighting the involvement of RNA in guiding protein structures, acting as chaperones for protein folding^11,12, as cofactors in assisting prion and amyloid propagations^13,14, and in mediating transgenerational memory via amyloid-like RNA–protein complexes¹⁵ (Box 1). This evidence indicates that RNAs can actively influence structural outcomes, supporting the plausibility of RNA-mediated RNA structural changes. We now extend this concept by providing a model that RNA can catalyse the self-replication of its own conformation via conformational catalysis, assisted by RBPs within biomolecular condensates, positioning RNA structure itself as the central catalytical entity and information carrier.

BOX 1. RNA-guided protein folding.

Protein folding depends not only on intrinsic amino acid sequences but also on dynamic guidance from external factors. Chaperones are proteins that assist other proteins in folding correctly and maintaining their functional conformations. By contrast, prions are misfolded, infectious proteins that can induce other normal proteins of the same sequence to misfold into the prion form, creating a self-perpetuating cycle. Remarkably, RNA emerges as a key player in both processes, acting as a chaperone for protein folding, and as a cofactor in assisting prion-like mechanisms.

RNA as a chaperone

RNA can directly influence protein folding. For example, a certain domain of bacterial 23S rRNA from Bacillus subtilis can refold denatured human carbonic anhydrase I into its active state in vitro¹¹. In addition, RNA molecules, particularly those with poly(U) sequences, exhibit potent chaperone-like activity by binding to unfolded proteins and preventing aggregation, outperforming traditional protein chaperones such as GroEL by approximately 300-fold¹¹. Moreover, the long noncoding RNA SLERT binds the helicase DDX21, stabilizing a closed conformation through a chaperone-like mechanism. This enables DDX21 to form loose clusters around nucleolar fibrillar centre and dense fibrillar component units, maintaining the environment necessary for efficient RNA polymerase I transcription¹². These chaperone-like functions highlight the capacity of RNA to modulate protein structures beyond traditional protein-based chaperones.

RNA facilitates prion and amyloid propagation

RNA acts as a critical cofactor in prion-like mechanisms. Stoichiometric transformation of PrP^C to protease-resistant PrP^Sc-like protein (PrP^res) in vitro requires specific RNA molecules. Notably, whereas mammalian brain RNAs stimulate in vitro amplification of PrP^res, yeast RNAs do not¹³. Therefore, host-encoded stimulatory RNA molecules may have a role in the pathogenesis of prion disease. Prion-like RBPs, such as TDP43 and FUS, are soluble in the nucleus but tend to aggregate into amyloid-like assemblies in the cytoplasm associated with neurodegenerative diseases. Their phase behaviour is regulated by RNA: low RNA-to-protein ratios promote droplet formation, whereas high ratios inhibit it. These findings suggest that RNA acts as a buffer to maintain RBP solubility¹⁴.

Transgenerational memory through amyloid-like RNA–protein complexes

In Caenorhabditis elegans, the conserved AN1 zinc finger proteins MSTR-1 and MSTR-2, both RNA-binding proteins, form amyloid-like structures in the germline, mediating transgenerational epigenetic inheritance of traits¹⁵. These amyloid assemblies persist across generations and are influenced by environmental factors such as temperature, with MSTR mutants exhibiting increased sterility at elevated temperature. MSTRs interact with the 26S proteasome to maintain amyloid stability, preventing developmental defects. The amyloids may be inherited as RNA–protein complexes, with RNA potentially stabilizing or propagating these structures, offering a noncanonical epigenetic memory that enhances adaptive fitness.

RNA as the central structural entity?

Whereas these abovementioned examples highlight the involvement of RNA in guiding protein structures — whereby protein conformations serve as the primary functional units — we now extend this concept by hypothesizing that RNA can catalyse the self-replication of its own conformations through conformational catalysis within biomolecular condensates, positioning RNA itself as the central structural entity.

Although our model remains a hypothesis, the vast permutations of evolutionary trial and error suggest that what is theoretically possible may well have already occurred. Only the evidence awaits discovery.