Noncoding RNAs in Nuclear Condensates and RNA Processing in Skeletal Muscle

1. Introduction

Skeletal muscle differentiation requires coordinated post-transcriptional regulation [1], while nuclear architecture spatially organizes gene-regulatory and RNA-processing activities [2]. A proliferating myoblast and a post-mitotic myotube therefore differ not only in the abundance of individual RNAs and proteins, but also in where these molecules are concentrated and which interactions become possible. Nuclear bodies can retain selected transcripts, partition RNA-binding proteins, and support RNA processing [2]. These spatial effects are especially relevant to muscle because alternative splicing, transcriptional competence, and ribosome production must be remodeled over a short developmental interval and then maintained in a large, multinucleated cell.

Long noncoding RNAs can participate in this organization in several distinct ways. An RNA may be necessary for assembly of a nuclear body, regulate the activity of proteins within an existing compartment, transiently enter a compartment without determining its structure, or accumulate because a pathological transcript cannot be exported or degraded. These roles are not interchangeable. The distinction is visible in three muscle-relevant systems. NEAT1 marks paraspeckles and has a defined architectural long isoform [3,4]. MALAT1 is concentrated at nuclear speckles but is dispensable for their gross structural integrity [5,6]. Expanded DMPK or CNBP repeat RNAs form nuclear foci in myotonic dystrophy and reduce the available pool of splicing regulators [7,8]. The similarities are spatial; the drivers, dynamics, and biological consequences differ.

The term biomolecular condensate adds a further evidentiary requirement. Microscopic concentration, round morphology, or RNA fluorescence in situ hybridization (FISH) puncta do not by themselves establish phase separation. A condensate claim is stronger when assembly depends on defined multivalent interactions and when the structure shows dynamic exchange, concentration buffering, reversibility, fusion, or another material-property measurement [9]. Paraspeckles satisfy several of these criteria in non-muscle experimental systems, including domain-resolved assembly and phase-separation assays [10,11,12]. Whether the same properties determine a particular muscle phenotype must still be tested in the relevant muscle state.

Muscle also poses a spatial problem that ordinary mononuclear cell culture does not capture. A mature myofiber contains many nuclei distributed across a shared cytoplasm, and individual myonuclei can specialize near the neuromuscular junction, myotendinous junction, or sites of repair [13,14]. Nuclear RNA abundance and compartment state may therefore vary along one fiber. Averaging across nuclei can conceal a restricted but functionally important population. The most informative studies will preserve nuclear position, cell identity, and differentiation state while measuring the proposed RNA assembly.

This review asks four connected questions. Which nuclear noncoding RNAs are linked to defined compartments or processing reactions in skeletal muscle? Which experiments distinguish architectural causality from expression change or colocalization? How do physiological compartments differ from repeat-RNA assemblies that drive spliceopathy? Finally, which therapeutic interventions have demonstrated target engagement at the relevant causal layer? Throughout this review, compartment, nuclear body, or RNA focus is used whenever condensate behavior has not been measured.

The scope is intentionally selective. Cytoplasmic competing-endogenous-RNA networks, extracellular-vesicle RNAs, and expression-only biomarker studies are outside the central argument. Chromatin-associated lncRNAs are included only when physical localization or interaction is connected to a defined nuclear process. Non-muscle studies establish experimental principles but are labeled as extrapolations. This boundary keeps the review focused on spatial nuclear mechanisms and prevents a broad catalog of muscle lncRNAs from obscuring the evidence needed to assign causality.

2. Evidence Review Approach

2.1. Literature Identification, Scope, and Source Verification

This article presents a structured critical narrative review. PubMed was searched through 29 September 2026 using combinations of terms related to skeletal muscle and myogenesis, nuclear bodies and RNA assemblies, noncoding and repeat RNAs, and RNA processing. Citation chaining was used to identify additional primary studies. For ongoing clinical programs, trial registries and dated sponsor disclosures were checked through the same date. Studies using myoblasts, myotubes, animal muscle, or patient muscle were prioritized. Foundational non-muscle studies were included only when they defined an architectural mechanism or material property not yet examined in skeletal muscle. Statements about evidence gaps are therefore limited to the literature identified by this search.

Central and disputed claims were evaluated according to species, model, perturbation, comparator, measured outcome, and relevant negative findings. Publication status was checked in PubMed and publisher records before the reference list was finalized.

2.2. Terminology Tied to Measurements and Evidence

A nuclear compartment is a spatially organized domain demonstrated by imaging or proximity-based measurements. A nuclear body is a recognized structure with accepted components, such as a paraspeckle, nuclear speckle, or nucleolus. An RNA focus is an operational imaging term: RNA signal is concentrated in a discrete region, without a claim about material state. A biomolecular condensate is a mechanistic interpretation that requires evidence for assembly through multivalent molecular interactions and for relevant material behavior. An architectural RNA is more restrictive still. Its depletion should impair assembly or maintenance, and an isoform- or domain-specific rescue should restore the structure when feasible.

These definitions separate the measured observable from the inferred biological quantity. RNA FISH measures spatial signal and, depending on calibration, molecule number. It does not establish direct binding or phase behavior. Biochemical fractionation measures recovery in an operational fraction, not immobility in the intact nucleus. Co-immunoprecipitation supports membership in the same biochemical complex, but not necessarily direct contact. Changes in exon inclusion after RNA depletion establish a functional relationship; they do not identify whether altered speckle organization, transcription kinetics, splicing-factor phosphorylation, or an indirect stress response is causal. Each inference therefore requires its own control.

Paraspeckles illustrate why isoform identity matters. Alternative 3′-end processing produces short NEAT1_1 and long NEAT1_2 transcripts, and HNRNPK-dependent production of NEAT1_2 initiates paraspeckle construction [3]. Genome editing showed that a substantial fraction of NEAT1_1 can localize outside paraspeckles, while NEAT1_1-specific loss produced limited phenotypes in the systems examined [15,16]. Semi-extractability is another useful property of several architectural nuclear RNAs, but it is a biochemical signature and is not sufficient proof of an architectural role [17]. A study reporting “NEAT1” without resolving isoform or assembly state may therefore establish a NEAT1-dependent phenotype without establishing a paraspeckle-dependent mechanism.

Nuclear speckles impose a different interpretive boundary. Super-resolution imaging supports a multilayer organization of speckle proteins and RNA [18]. MALAT1 is enriched around these domains and can alter the distribution and phosphorylation of serine/arginine-rich splicing factors in cultured human cells [5]. Yet three independent mouse loss-of-function studies found normal viability and no requirement for MALAT1 in gross speckle formation, with context-dependent effects on neighboring transcription or molecular phenotypes [6,19,20]. The evidence supports a regulatory, concentration-dependent role more strongly than an obligatory architectural role.

The evidence ladder used here assigns the permitted interpretation and does not rank a paper globally. E0 denotes expression association. E1 adds nuclear enrichment or colocalization. E2 establishes physical interaction or maps components. E3 shows that perturbation changes assembly or RNA processing. E4 adds rescue, domain mapping, epistasis, or an orthogonal perturbation. E5 links the mechanism to an in vivo or human phenotype. E6 requires target engagement together with molecular and phenotypic rescue. A single study can occupy different tiers for different claims. For example, a study may provide E5 evidence that an RNA affects muscle regeneration but only E1 evidence that the effect occurs through a named nuclear body.

The modules that meet the inclusion threshold and the interpretation supported by each evidence tier are summarized in Table 1.

The ladder also constrains language. E1 supports “localizes to” or “is associated with,” not “scaffolds.” E2 supports participation in a molecular complex, not necessity. E3 supports a functional contribution within the tested perturbation and model. E4 strengthens causal assignment by showing that the phenotype tracks with a mapped RNA element, partner, or rescue construct. E5 is required before a cellular mechanism is presented as muscle physiology or disease biology. E6 is reserved for interventions that reach the intended target and correct a downstream molecular defect together with a relevant phenotype.

2.3. Physiological Compartments and Pathological Assemblies are not Equivalent

Physiological nuclear bodies are maintained by regulated synthesis, turnover, and exchange of their components. Their assembly can be coupled to differentiation or stress without being pathogenic. Repeat-expansion RNAs are produced from mutant alleles, retained for abnormal durations, and present repeated binding motifs that distort the availability of RNA-binding proteins. Both categories can concentrate RNA and protein, but the selective pressures and consequences are different. The comparison is useful only when it preserves these differences.

Table 2 compares the assembly evidence and processing consequences of the principal physiological and pathological systems.

This framework avoids two common errors. First, loss of a visible focus after treatment does not necessarily prove that dissolution of a condensate caused the biological rescue; the treatment may simply reduce RNA abundance. Second, a phenotype after depletion of a compartment-associated lncRNA does not prove that changing compartment architecture caused the phenotype. The decisive experiment measures RNA abundance, assembly state, molecular processing, and muscle output in the same perturbation series and includes a rescue that separates these variables.

Assembly and function should therefore be represented as separable causal nodes. RNA abundance can affect assembly, but it can also alter transcription, chromatin occupancy, or protein binding without changing a visible body. Assembly can affect processing, yet processing can also change because the same RNA or protein acts in the surrounding nucleoplasm. A muscle phenotype may arise from either route or from a change in cell-state composition. Factorial perturbations are more informative than a single knockdown: one intervention changes RNA abundance, a second changes assembly competence while preserving abundance, and a third restores the proposed downstream processing event. Concordant rescue across these levels supports the complete path. Divergent rescue identifies where the proposed mechanism fails.

OpenAI ChatGPT was used to generate conceptual artwork for Figure 1, Figure 2 and Figure 3. The authors subsequently reviewed and modified each figure in Microsoft PowerPoint and verified all scientific labels, molecular relationships, and interpretations against the cited evidence.

Figure 1 integrates the compartment-centered modules across the myoblast-to-myotube transition while marking direct muscle evidence and disease-specific inference separately.

3. NEAT1 and Paraspeckles in Myogenic State Transitions

Among the systems reviewed here, NEAT1 is the strongest muscle-linked candidate for an architectural nuclear RNA, but direct muscle studies currently separate into evidence for paraspeckle association and evidence for NEAT1-dependent myogenic regulation; few experiments connect both in the same system.

3.1. Differentiation-Dependent Paraspeckle Association in C2C12 Cells

The foundational muscle study identified the then-named Men ε/β transcripts as differentiation-regulated, nuclear-retained RNAs and later recognized them as NEAT1 isoforms [4]. During C2C12 differentiation, transcript abundance increased approximately 3.2- to 4.9-fold, and RNA FISH showed larger and more numerous nuclear foci in myotubes. The RNA signal colocalized with PSPC1, and biochemical experiments linked the transcripts to NONO and other DBHS-family paraspeckle proteins. These observations provide direct muscle evidence for differentiation-dependent spatial association and molecular participation, corresponding to E2 for the C2C12 paraspeckle claim.

The same paper also provided strong architectural evidence, but the model boundary is important. Antisense-mediated depletion disrupted paraspeckles in most depleted cells, and transcriptional reactivation experiments placed the RNA early in paraspeckle reassembly. Those perturbation experiments were performed in HeLa cells, not in C2C12 myoblasts or myotubes. They therefore support the general architectural function of NEAT1, not direct proof that paraspeckle assembly is necessary for myogenic differentiation. A later study in human embryonic stem-cell differentiation similarly linked NEAT1 expression, paraspeckle formation, and nuclear retention of selected inverted-repeat-containing transcripts, but again outside skeletal muscle [39].

Domain-resolved work strengthened the non-muscle architectural model. NEAT1_2 contains separable regions that recruit NONO and SFPQ, promote assembly, and generate a core–shell organization; perturbations of these regions and in vitro assays support phase separation [10]. Block-copolymer and triblock-micelle models further explain how an RNA scaffold with distinct interaction domains can impose paraspeckle geometry [11,12]. These studies justify condensate language for paraspeckles as a general structure. They do not establish that a change in condensate material state mediates a specific muscle phenotype.

3.2. NEAT1-Dependent Myogenesis is not Yet a Paraspeckle Mechanism

Neat1 perturbation in C2C12 cells and mouse muscle provides direct evidence for myogenic function [21]. Knockdown reduced proliferation and promoted differentiation and fusion, whereas gain-of-function produced the opposite cellular tendency. In vivo depletion increased myofiber cross-sectional area but delayed cardiotoxin-induced regeneration, consistent with reduced expansion of PAX7-positive progenitors before differentiation. Mechanistically, Neat1 interacted with EZH2, and perturbation altered EZH2 or H3K27me3 occupancy at Cdkn1a, Myog, Myh4, and Tnni2. A mapped RNA segment and epistasis with EZH2 or P21 strengthen the chromatin-regulatory claim to E4, while the regeneration result reaches E5 for a Neat1-dependent muscle phenotype.

The study did not demonstrate that paraspeckle assembly mediated these effects. It did not quantify NEAT1_2 separately, measure paraspeckle number or dynamics after perturbation, or rescue the phenotype with isoform-specific constructs. The mapped 1001–1540 nucleotide region lies within sequence shared by the shorter and longer transcripts. Consequently, the supported conclusion is that Neat1 can regulate myogenic proliferation and differentiation through an EZH2-linked mechanism. Calling this a paraspeckle mechanism would silently add an untested causal step.

The apparent tension between larger fibers and delayed regeneration also illustrates the need to define the observation window. Accelerating cell-cycle exit can increase differentiation in a short in vitro assay while depleting the proliferative pool needed for efficient repair in vivo. Fiber cross-sectional area, progenitor expansion, and regeneration kinetics are distinct outputs. A time-resolved experiment that quantifies satellite-cell activation, clonal expansion, fusion, and mature force is needed before these outcomes are collapsed into a single “pro-” or “anti-myogenic” label.

The physiological interpretation must also account for myonuclear heterogeneity. Adult skeletal muscle contains nuclei from myofibers, satellite cells, endothelial cells, fibro-adipogenic progenitors, immune cells, and other populations. A change measured in whole muscle may reflect altered cellular composition and not a within-nucleus change in NEAT1 or paraspeckles. Single-nucleus RNA sequencing can localize expression, but it does not by itself measure nuclear-body assembly and may under-recover long, semi-extractable RNAs. Combining isoform-resolved in situ hybridization with cell-type markers and quantitative three-dimensional imaging would distinguish cell-composition effects from remodeling within a defined lineage.

Stress is an additional confounder. NEAT1_2 and paraspeckles respond to several stresses in non-muscle systems, while muscle injury simultaneously changes inflammation, oxygenation, mechanical load, and progenitor state. Increased paraspeckles after injury could represent an adaptive retention program, a consequence of stress signaling, or a prerequisite for regeneration. These alternatives predict different timing. An adaptive response should follow the initiating stress and may be dispensable for lineage progression; an architectural requirement should precede or coincide with the processing transition and should be rescued by assembly-competent NEAT1_2.

3.3. PABPN1, MATR3, and Paraspeckle-Linked RNA Processing in Muscle Disease

Work on poly(A)-binding protein nuclear 1 (PABPN1) and MATR3 provides a more direct bridge from muscle disease proteins to NEAT1-centered RNA processing [22]. In primary mouse myoblasts, depletion of PABPN1 or MATR3 increased Neat1 stability and nuclear foci, redistributed SFPQ, and altered A-to-I editing of the paraspeckle-retained Ctn transcript. Co-depletion of Neat1 moved the editing phenotype toward control, supporting functional placement of Neat1 downstream of these proteins. The study also examined altered SFPQ distribution in an oculopharyngeal muscular dystrophy mouse model and nuclear–cytoplasmic redistribution of selected transcripts in PABPN1-deficient human myoblasts.

This module is stronger than colocalization because it combines muscle cells, perturbation, RNA-processing output, and genetic interaction. It nevertheless does not prove that altered material properties of a paraspeckle are causal. Increased foci could reflect greater transcript abundance, greater stability, altered protein recruitment, or a change in assembly kinetics. Quantitative molecule counting, isoform-specific depletion, and rescue with assembly-competent versus assembly-deficient NEAT1_2 would distinguish these possibilities.

Muscle studies establish differentiation-regulated NEAT1 localization, interaction with paraspeckle proteins, Neat1-dependent chromatin regulation, and a PABPN1/MATR3-linked processing phenotype. Architectural necessity, domain-resolved rescue, and material behavior have been established mainly in non-muscle systems. The central unresolved question is whether NEAT1_2-dependent paraspeckle assembly is itself required for a defined myogenic transition or disease phenotype.

4. MALAT1, Nuclear Speckles, and Context-Dependent RNA Processing

MALAT1 is best treated as a speckle-associated regulator whose effect depends on cell state and perturbation history, not as an obligatory speckle scaffold or a uniformly pro- or anti-myogenic RNA.

4.1. A Regulatory Role at Speckles

In human cultured cells, MALAT1 associates with serine/arginine-rich splicing factors, changes their distribution between nucleoplasm and speckles, alters phosphorylation-associated pools, and affects a subset of alternative-splicing events [5]. These experiments provide a mechanistic route from a nuclear-retained lncRNA to splice-site choice. The route is not equivalent to structural maintenance of the speckle. Mouse knockout studies found preserved speckle morphology and generally normal development, although local transcriptional effects and context-specific phenotypes remained possible [6,19,20].

Recent biophysical studies refine this distinction. Speckle proteins can form compositionally distinct microphases, and MALAT1 can preferentially associate with particular serine/arginine-rich protein phases [36]. Under hypoxia in cancer cells, MALAT1 was reported to promote SRSF1 condensation near speckles and to alter pre-mRNA processing [40]. These findings make assembly state a plausible regulatory variable. They are non-muscle results obtained under specific conditions and should guide tests in myoblasts and myotubes.

4.2. Opposing Myogenesis Phenotypes

Three frequently cited muscle studies do not support a single directional model. In C2C12 and primary human myoblasts, MALAT1 increased during differentiation, and myostatin reduced its abundance; acute silencing impaired proliferation and modestly altered a differentiation endpoint [23]. A second C2C12 study reported that MALAT1 promoted differentiation through a proposed miR-133–SRF competing-endogenous-RNA pathway [24]. These data support a pro-myogenic interpretation under the tested schedules, although the stoichiometric requirements of the ceRNA mechanism were not established.

By contrast, an independent study observed a pronounced early decline in Malat1 after differentiation induction and found that two loss-of-function approaches increased myogenic markers, Myogenin- or myosin-heavy-chain-positive cells, and MyoD-responsive reporters [25]. Malat1 recruited Suv39h1, HP1β, and HDAC1 to MyoD target genes, while miR-181a/Ago2 contributed to nuclear Malat1 degradation. Constitutive Malat1 loss did not disrupt baseline mouse muscle but enhanced regeneration after cardiotoxin injury and in the mdx background. These data support an inhibitory chromatin mechanism during an early differentiation window and E5 evidence for a regeneration phenotype.

The divergent results remain a genuine contradiction. They differ in confluence, differentiation timing, acute versus constitutive loss, knockdown magnitude, proliferation state, and chosen endpoint. A late increase in MALAT1 does not exclude an early decline, and an effect on proliferation can alter the number of cells available to differentiate. None of these differences was isolated experimentally across studies. They are candidate moderators, not an established reconciliation.

The proposed miRNA-sponge mechanism requires an additional quantitative test. A competing endogenous RNA can measurably alter a miRNA target only when binding-site abundance and occupancy are sufficient relative to the miRNA and competing targets. Reporter rescue and reciprocal expression are supportive, but they do not establish this stoichiometry. For a predominantly nuclear transcript such as MALAT1, the subcellular availability of the relevant miRNA and Argonaute complex is also critical. Absolute molecule counting, site mutation in endogenous MALAT1, and rescue at physiological abundance would determine whether miR-133 or miR-181a interactions explain the muscle phenotype or accompany another nuclear mechanism.

Constitutive knockout adds a different uncertainty. The absence of an overt baseline muscle phenotype can mean that MALAT1 is dispensable under homeostasis, that related pathways compensate during development, or that the chosen endpoints are insensitive. Enhanced regeneration after injury identifies a conditional phenotype but does not distinguish these alternatives. An inducible adult knockout, paired with acute RNA degradation in isolated satellite cells, would reveal whether the same effect appears without developmental adaptation.

A decisive experiment would use the same myoblast preparation at matched density and quantify MALAT1 every few hours during cell-cycle exit and fusion. It would compare rapid degradation, CRISPR interference, and constitutive loss, then rescue with expression calibrated to the endogenous range. Readouts would include cell-cycle state, MyoD occupancy, chromatin marks, SR-protein localization and phosphorylation, transcriptome-wide splicing, fusion, and force after maturation. If the direction of the phenotype changes with timing while the perturbation and endpoint are held constant, the temporal-state model would be supported. If it does not, differences in reagent specificity or genetic adaptation would become more likely.

4.3. The Unclosed Link Between MALAT1 and Muscle Splicing

The strongest MALAT1–splicing mechanism was established outside muscle, whereas the direct muscle studies emphasized proliferation, miRNA-linked regulation, chromatin occupancy, and differentiation. Among the primary muscle studies evaluated here, none combined MALAT1 perturbation with transcriptome-wide alternative-splicing analysis and quantitative speckle or SR-protein measurements. This is a bounded statement about the source set searched through 29 September 2026, not proof that no relevant study exists.

That gap matters because changes in myogenic markers can arise through several routes. MALAT1 may influence chromatin at MyoD-responsive genes, local concentrations of splicing factors, transcription elongation, RNA stability, or cell-cycle composition. A speckle-mediated mechanism should predict that an assembly-selective perturbation changes a defined splicing program even when total MALAT1 abundance and proliferation are controlled. A chromatin-mediated mechanism should instead track with local histone marks and promoter occupancy. Both mechanisms could operate at different stages.

MALAT1 has direct muscle phenotypes, including an in vivo regeneration phenotype, but its structural role at speckles is not supported. The direction of its myogenic effect remains context-dependent, and a direct causal chain from MALAT1 through speckle state to muscle alternative splicing has not yet been demonstrated in the source set examined.

5. Other Nuclear Muscle ncRNAs that Meet a Mechanistic Threshold

Additional muscle lncRNAs should enter this framework only when spatial localization is connected to a defined nuclear process; most chromatin-associated myogenic lncRNAs are regulatory candidates, not established nuclear-body scaffolds.

5.1. FRG2A/B Transcripts Define a Nucleolar Disease Module

Nucleolar FRG2 transcripts provide the strongest additional compartment-centered module. In primary human myoblasts from individuals with facioscapulohumeral muscular dystrophy (FSHD), FRG2A/B transcripts were elevated and localized to a defined nucleolar region [26]. Chromatin isolation by RNA purification enriched rDNA and centromeric sequences, and FSHD cells showed increased associations between centromeres and nucleoli. These measurements support spatial participation in nucleolar and perinucleolar organization, while the high sequence identity of FRG2A and FRG2B limits paralog-specific assignment.

The functional chain extends beyond localization. FSHD myoblasts had reduced 45S pre-rRNA abundance, lower nascent nucleolar RNA synthesis, altered nucleolar number and area, and reduced nascent protein synthesis. Depletion of FRG2A/B transcripts increased 45S pre-rRNA and nascent RNA synthesis and reversed the protein-synthesis defect. Proteomics identified incomplete compensation during myotube differentiation, with persistent underrepresentation of several muscle-related proteins. These results support E4 evidence for a nucleolar rDNA-transcription mechanism in patient-derived cells.

The study does not establish that FRG2 transcripts create a separate condensate or that this mechanism explains the full clinical phenotype of FSHD. rDNA transcription, nucleolar morphology, translation, and disease severity were not all linked by an in vivo rescue. The supported model is narrower: D4Z4-associated derepression increases nucleolar FRG2 RNA, which contributes to impaired rRNA synthesis and translational capacity in the tested FSHD cells.

The measured quantities further constrain the model. Lower 45S pre-rRNA and reduced nucleolar 5-ethynyl uridine incorporation support impaired nascent rRNA production, whereas unchanged processing intermediates argue against a primary maturation defect in the assays used. Reduced homopropargylglycine incorporation reports lower nascent protein synthesis, not selective degradation of mature proteins. Proteomic underrepresentation of contractile proteins after differentiation is consistent with an insufficient translational response, but it could also reflect transcriptional or differentiation differences. Ribosome profiling, pulse–chase proteomics, and rescue of FRG2 at calibrated abundance would separate initiation, elongation, protein stability, and cell-state effects.

5.2. Chromatin-Associated Regulators Belong at the Boundary

Ppp1r1b-lncRNA illustrates a chromatin-centered mechanism that reaches the inclusion threshold without defining a nuclear body. In mouse and human muscle models, the RNA interacted with polycomb repressive complex 2 (PRC2) and promoted myogenic gene expression; genetic studies linked it to cardiac and skeletal muscle development [27]. More recent work showed that hnRNPA1 stabilizes a Ppp1r1b-lncRNA–PRC2 complex that helps maintain muscle-progenitor identity [28]. The evidence supports a nuclear RNP and chromatin-regulatory mechanism, but not condensate or compartment terminology.

Myoparr arises near the Myog locus and supports myogenic differentiation through promoter-associated transcriptional regulation [29]. Linc-RAM interacts with MyoD and promotes assembly of a transcriptional complex at myogenic genes [30]. Dum recruits DNA methyltransferases to repress Dppa2 during differentiation and regeneration [31]. MYH1G-AS provides an avian example of a chromatin-associated lncRNA that regulates skeletal muscle development, but its species context and absence of a named nuclear-body mechanism limit extrapolation to mammalian muscle [41]. These RNAs are most useful as comparators that demonstrate the range of nuclear lncRNA mechanisms, not as evidence that every chromatin-associated RNA forms a compartment.

FRG2A/B transcripts connect nucleolar localization to rDNA transcription and translation in patient-derived muscle cells. Other selected lncRNAs connect physical interaction or chromatin occupancy to myogenic transcription. None yet has the domain-resolved assembly and material-property evidence required to define a new muscle nuclear condensate.

6. Pathological Repeat-RNA Assemblies in DM1 and DM2

Expanded CUG and CCUG RNAs cause disease through abnormal nuclear retention and loss of RNA-binding-protein function, but the visible structures range from single retained transcripts to multimeric assemblies; “RNA focus” is therefore the correct default term.

6.1. Genetic Lesions and Sequestration of Processing Factors

Myotonic dystrophy type 1 (DM1) is caused by a CTG expansion in the 3′ untranslated region of DMPK. The mutant transcript contains expanded CUG repeats. Early work showed repeat-length-dependent binding of muscleblind proteins and their accumulation with expanded RNA in nuclear foci from affected cells [7]. DM2 is caused by a CCTG expansion in intron 1 of CNBP (formerly ZNF9), producing expanded CCUG RNA and establishing a parallel toxic-RNA mechanism [34]. In both disorders, the mutation changes RNA behavior without requiring loss of the encoded protein to explain the core spliceopathy.

The most reproducible molecular consequence is reduced functional availability of muscleblind-like (MBNL) proteins. MBNL and CELF families normally regulate developmentally programmed RNA processing, often in opposing directions [42,43]. Expanded repeats recruit MBNL, while CELF1 activity can increase through separate signaling and post-translational mechanisms. Increased nuclear CELF1 activity is sufficient to produce muscle wasting in mice, demonstrating that redistribution or activation of a processing factor can drive tissue pathology without being a simple readout of an RNA focus [44].

Patient and model studies place the expanded RNAs near processing domains but do not make them physiological speckles. DM1 RNA accumulated at the periphery of nuclear speckles in cultured cells [45]. DM2 patient myoblasts and tissues contain CCUG RNA/MBNL1-positive foci [8,35]. These observations support spatial proximity to RNA-processing machinery and sequestration. They do not show that the foci share the composition, topology, or reversibility of a normal nuclear speckle.

6.2. From Phase Transition to Single Retained RNA

Purified repeat RNAs can undergo length- and concentration-dependent sol–gel transitions, and repeat-RNA foci in transfected cells were sensitive to interventions that disrupted intermolecular base pairing [37]. This work established biophysical plausibility for RNA-driven phase transitions and identified repeat length as a determinant of assembly. It is appropriate evidence for condensate-like behavior under the tested in vitro and cellular conditions.

Quantitative imaging in patient-derived DM1 myotubes produced a critical refinement. Most apparent foci corresponded to individual nuclear-retained expanded transcripts, while larger multimeric foci were less common, arose from transcriptional bursts, and dissipated over time [32]. Individual transcripts were sufficient to bind MBNL1, whereas multimeric focus formation depended on MBNL. Thus, sequestration and toxicity need not require a bulk RNA condensate. An image-limited punctum can represent one unusually large RNA–protein complex.

The two results are compatible only at a defined level. Repeat RNA has an intrinsic capacity for intermolecular assembly, but endogenous molecule number, transcriptional bursting, protein occupancy, and nuclear clearance determine whether that capacity is realized in a patient cell. The material state of one experimental construct cannot be assigned to every endogenous focus. Future studies should combine calibrated molecule counting, single-particle tracking, fluorescence recovery, perturbation of repeat–repeat pairing, and measures of MBNL availability in the same patient-derived muscle model.

This distinction changes how focus burden should be interpreted. Counting puncta can underestimate dispersed individual expanded transcripts, while integrated fluorescence can overestimate molecule number when probe accessibility varies with repeat length or protein occupancy. A treatment that converts a multimeric focus into several individual RNAs may reduce focus size without restoring MBNL availability. Conversely, degradation of individual transcripts can improve splicing without a dramatic change in the number of large foci if a transcription site remains active. Molecular imaging should therefore be calibrated against transcript abundance and paired with a functional sequestration readout.

6.3. Spliceopathy, Developmental State, and Tissue Specificity

Transcriptome studies of DM skeletal muscle and heart show extensive but tissue-dependent changes in alternative splicing and gene expression [46]. Congenital DM1 disrupts prenatal RNA processing and myogenesis, placing the mechanism within a developmental window, not only in degenerating adult muscle [47]. DM1 and DM2 are not molecular duplicates: competition between rbFOX1 and MBNL1 for CCUG RNA and comparisons of human cellular models identify subtype-specific processing signatures [38,48].

Specific splicing events connect loss of processing-factor activity to physiological output. Mis-splicing of CACNA1S changes CaV1.1 channel gating and contributes to weakness [49]. Combined loss of MBNL1 and MBNL2 in mice produces severe cardiac electrical and processing phenotypes, while zebrafish mbnl mutants reproduce molecular and physical features across tissues [50,51]. These models support MBNL loss of function as a causal layer, although no single model reproduces the repeat mutation, multisystem distribution, and temporal progression of human disease.

Human data increasingly connect molecular biomarkers to function. Splicing changes in children with DM were associated with physical performance, and hybrid myotube systems combining patient and unaffected nuclei modified several cellular hallmarks [52,53]. Differentiation can itself aggravate the RNA phenotype in DM1 models, showing that the myogenic state changes the relationship among transcript abundance, nuclear retention, and processing-factor demand [54]. Such results caution against comparing proliferating patient myoblasts with mature control muscle as though cell state were a nuisance variable.

Splicing signatures also require causal discipline. An exon that correlates with disease severity can be a sensitive biomarker without mediating weakness. Conversely, a physiologically important event may be difficult to measure reproducibly in a small biopsy. Mechanistic prioritization should combine direct MBNL or CELF binding, perturbation of the candidate exon, protein-level confirmation, and a relevant physiological readout. This standard is met most closely for selected ion-channel events; it remains incomplete for many transcriptome-wide changes.

Regeneration is another causal layer. Experimental reduction of MBNL activity impairs aspects of skeletal muscle regeneration, while a repeat-RNA mouse model showed loss of PAX7-positive satellite cells, delayed maturation, fibrosis, and partial correction by a DMPK-targeting oligonucleotide [55,56]. Congenital DM models also show altered myokine signaling [57]. These studies extend spliceopathy beyond contraction toward progenitor maintenance and repair, but they do not imply that every regenerative defect results from the same exon-level event.

Spatial distribution varies within muscle. Ribonuclear foci occur at neuromuscular junctions, and sense and antisense DMPK foci show developmental and tissue-specific patterns [58,59]. Abnormal nuclear aggregation has been linked to myotube degeneration in DM1 models [60]. In DM2, multi-level profiling points to mitochondrial dysfunction [61]. Separate CCUG-repeat studies implicate chloride-channel splicing [62], reduced global protein synthesis [63], repeat-associated non-AUG translation [64], and apoptosis [65]. These findings support a shared toxic-RNA core with disease- and tissue-specific downstream branches.

DM2 requires particular restraint because its average expansion can be much longer than the DM1 repeat, yet clinical anticipation and tissue patterns differ. The CCUG repeat is excised from an intron, and the fate of the host transcript, excised repeat RNA, and associated proteins need not match that of the DMPK 3′ untranslated region. Evidence for MBNL sequestration is strong, but contributions from CNBP dosage, RAN translation, and other RNA-binding proteins remain incompletely separated. A therapy effective against CUG RNA cannot be assumed to bind, traffic, or degrade CCUG RNA with the same selectivity.

Human genetics, patient tissue, and multiple animal models establish repeat-RNA toxicity and MBNL-dependent spliceopathy. The number, composition, and material state of individual RNA foci remain variable. Condensate terminology is justified for experiments that directly measure phase or gel behavior, not as a default description of all DM1 or DM2 puncta.

Figure 2 separates physiological RNA compartments from pathological repeat-RNA assemblies and indicates where evidence supports architectural, regulatory, or pathogenic roles.

7. Matching Intervention to Causal Layer

Therapeutic evidence is strongest when an intervention reaches the intended molecular layer, restores a processing readout, and improves a muscle phenotype; disappearance of foci or change in RNA abundance alone is insufficient.

7.1. Reducing Toxic RNA Abundance

Systemic antisense oligonucleotides (ASOs) targeting nuclear-retained CUG-expanded RNA produced a rapid and durable reduction of toxic RNA in the human skeletal actin long-repeat (HSA^LR) mouse model, correcting physiological, histopathological, and transcriptomic abnormalities [33]. This study is an E6 preclinical anchor because it measured target reduction, processing rescue, and tissue phenotypes together. It also demonstrated efficient knockdown of the nuclear-retained lncRNA Malat1, supporting nuclear residence as a determinant of ASO susceptibility.

Clinical translation exposed delivery as the limiting variable. In a phase 1/2a trial, subcutaneous baliforsen was generally tolerated and accumulated in muscle, but concentrations were below those predicted to produce substantial DMPK reduction [66]. A transferrin-receptor-1-targeted antibody–siRNA conjugate improved muscle delivery in a subsequent phase 1/2 trial. Delpacibart etedesiran reduced DMPK RNA in muscle biopsies and improved a composite missplicing score in some dose groups, while serious adverse events occurred in two treated participants [67]. Mechanistic analyses linked DMPK reduction to increased estimated functional MBNL and dose-dependent splicing correction [68]. However, a sponsor-reported topline analysis of the phase III HARBOR trial announced on 8 September 2026 did not show a statistically significant improvement over placebo in the primary endpoint, video hand opening time, although activity was reported for secondary and exploratory endpoints [69]. Detailed peer-reviewed phase III results were not available at the review cutoff. These findings establish muscle delivery and molecular activity, but not clinical efficacy.

The contrast between these trials is informative but not a head-to-head comparison. The molecules differ in chemistry, delivery route, dosing, tissue exposure, sampling time, and pharmacodynamic assays. Higher muscle concentration does not guarantee uniform delivery across fiber types or to the heart and other affected organs. Biopsy measurements are also local and can be influenced by fat replacement, fibrosis, and the fraction of nuclei carrying high repeat-RNA burden. Future trials should prespecify how tissue exposure, DMPK reduction, splicing correction, and functional outcomes will be linked within participants.

Clinical endpoints operate on different time scales. Splicing can respond within days, myotonia may improve after downstream protein turnover, and strength or endurance may require structural remodeling and rehabilitation. Irreversible fibrosis, motor-unit loss, or central nervous system involvement may limit the response even when muscle target engagement is adequate. Trial duration and enrollment stage should therefore match the predicted reversibility of the selected endpoint. A molecularly successful early study can justify continued development without proving clinical efficacy, whereas a clinical claim requires a controlled and sufficiently powered functional outcome.

Splicing can serve as a pharmacodynamic bridge if it is measured rigorously. Targeted sequencing of a panel of MBNL-responsive exons showed graded responses to CUG-expanded RNA, MBNL dosage, and ASO treatment in mouse models, with molecular correction preceding loss of myotonia [70]. Such composite biomarkers reduce dependence on one exon and can compare interventions that act at different upstream layers. They do not substitute for functional outcomes, especially when fatigue, regeneration, cognition, or cardiac disease may follow different kinetics.

7.2. Editing or Blocking Production of Repeat RNA

Programmable nucleases can act on RNA or DNA. RNA-targeting Cas9 eliminated repeat RNA in patient cells and reversed foci, protein relocalization, and DM1 splicing defects [71]. Catalytically inactive Cas9 targeted to expanded DNA impeded transcription in a repeat-length-, strand-, and PAM-dependent manner and improved molecular and physiological outcomes in models [72]. Genome editing that excises or contracts the CTG tract offers a more durable strategy, and time-controlled, muscle-specific approaches have improved molecular phenotypes in mice [73,74]. Each approach must be evaluated for delivery to distributed skeletal muscle, allele selectivity, DNA or RNA off-targets, immune responses, and durability.

7.3. Small Molecules and Engineered Binders

Repeat structure creates binding sites for multivalent small molecules and targeted degraders. A transcriptome-wide study developed a photoaffinity-informed degrader that bound a limited set of cellular RNAs and selectively reduced r(CUG)^exp, improving broad RNA-sequencing defects in DM1 myotubes [75]. Cugamycin and a related structure-selective dimer cleaved expanded CUG RNA in patient-derived myotubes or mouse models while sparing shorter repeats more effectively than sequence-only recognition [76,77]. These experiments demonstrate that structure can provide selectivity, but metabolism, exposure, and off-target binding remain separate translational questions.

Repurposed or optimized ligands have reached varying levels of evidence. Erythromycin disrupted CUG–MBNL interactions and improved splicing and myotonia in models [78]. Furamidine acted through several mechanisms, including repeat binding and increased MBNL abundance, complicating assignment of a single causal target [79]. Quercetin and enzymatically modified isoquercitrin reduced repeat-RNA toxicity in preclinical systems, and newer polycyclic compounds rescued missplicing in disease models [80,81]. These compounds should be compared using free drug concentration, repeat-length range, transcriptome-wide specificity, and exposure in affected muscle.

Engineered proteins and endogenous RNA-remodeling factors provide additional routes. A designed pentatricopeptide-repeat protein targeted CUG RNA and improved toxicity in a DM1 mouse model [82]. Increasing DDX5 reduced toxic RNAs in DM1 and DM2 models [83]. These strategies can alter RNA occupancy or turnover without necessarily dissolving a condensate. Their mechanism should be judged by changes in RNA number, MBNL availability, processing, and phenotype, not by focus morphology alone.

7.4. Oligonucleotide Architectures and Restoration of Processing Capacity

Gapmer ASOs [84], splice-modulating oligonucleotides [85], repeat-directed RNA interference [86], and AAV-delivered U7 small nuclear RNAs [87] represent distinct pharmacological architectures. A gapmer recruits RNase H to reduce RNA abundance; a steric blocker can displace proteins or alter processing without degradation; RNA interference depends on access to a compatible compartment; and an AAV cassette provides persistent expression but introduces vector-specific risk. These modalities should not be grouped simply as “antisense therapy.”

Increasing functional MBNL is an orthogonal strategy. Initial antagomiR-23b and antagomiR-218 studies increased MBNL and corrected molecular phenotypes in preclinical models [88,89]. Peptide-conjugated and primary-patient-cell studies extended the approach to splicing, myogenic defects, and, in some systems, reduced DMPK RNA or foci [90,91]. Exercise combined with ASO treatment improved fatigue more completely than molecular therapy alone in aged DM1 mice, indicating that correction of the initiating RNA lesion may not immediately reverse every systems-level phenotype [92]. Histone-deacetylase inhibition and modulation of the GSK3β–CELF1 axis target downstream regulatory states, not the expanded RNA itself [93,94].

Direct restoration of RNA-binding-protein function is also possible. Engineered RNA-binding proteins improved splicing regulation in experimental systems, and MBNL1 overexpression showed therapeutic effects in skeletal muscle or cardiac models [95,96,97]. The safety concern is dosage: MBNL proteins regulate broad developmental programs, and excessive or ectopic activity can create new processing defects. A successful strategy must restore an adult processing range without maximizing one factor.

Representative interventions are mapped to their causal targets and evidence limits in Table 3.

Figure 3 places current therapeutic approaches along the causal path from repeat-locus transcription to muscle dysfunction and separates molecular from clinical validation.

Therapeutic claims should therefore be stated at the level directly demonstrated. Target engagement is RNA reduction, repeat occupancy, protein release, or editing at the intended locus. Molecular rescue is restoration of MBNL availability or a validated splicing signature. Tissue rescue includes myotonia, force, histology, regeneration, or another relevant outcome. Clinical benefit requires a controlled patient endpoint. Conflating these levels obscures why an intervention can be molecularly active yet clinically limited.

Normal transcript function is another design constraint. DMPK reduction has been tolerated in several preclinical and early clinical settings, but an intervention aimed at the expanded allele should still measure wild-type transcript and protein, local genomic effects, and longer-term cardiac or metabolic consequences. Repeat-selective strategies exploit structure or length, whereas non-allele-selective strategies rely on a therapeutic window that reduces both alleles. The appropriate balance may differ across organs and disease stages.

Finally, compartment-directed therapy is less mature than toxic-repeat targeting. NEAT1 or MALAT1 depletion may simultaneously alter progenitor proliferation, stress adaptation, chromatin regulation, and RNA processing. A compound that changes the size or number of paraspeckles or speckles is not therapeutic evidence unless it also shows the intended processing correction and a favorable muscle phenotype without disrupting normal nuclear function. Physiological compartments should therefore be modulated only after identifying the causal assembly feature.

8. Unresolved Questions and Experimental Standards

8.1. What Would Prove an Architectural RNA Role in Muscle?

Box 1 summarizes the experimental combination needed to assign an architectural RNA role in muscle.

The key experimental principle is separation of RNA abundance from assembly state. For NEAT1, an ideal experiment would deplete endogenous Neat1, rescue NEAT1_1 or NEAT1_2 at matched abundance, and compare wild-type NEAT1_2 with a domain mutant that retains stability but cannot organize paraspeckles. Quantitative imaging would measure molecule number, paraspeckle number, size, and exchange. RNA sequencing and retention assays would define processing outputs, while differentiation, regeneration, and force would provide muscle endpoints. A phenotype rescued only by assembly-competent NEAT1_2 would support architectural causality.

For MALAT1, the decisive design must be temporal. Rapid degradation and reversible rescue would minimize developmental compensation. Density-matched myoblasts would be sampled before cell-cycle exit, during commitment, and during fusion. Speckle morphology, SR-protein microphase partitioning, phosphorylation, nascent transcription, chromatin occupancy, and splicing would be measured in the same cells or matched aliquots. If a processing phenotype follows altered protein partitioning while chromatin occupancy remains stable, the speckle route gains support; the converse result favors a chromatin mechanism.

For repeat RNA, counting endogenous molecules is essential. RNA FISH protocols and expanded-repeat reverse transcription require calibration because probe accessibility and reverse-transcriptase processivity can bias the inferred abundance [98,99,100]. Live-cell imaging should distinguish individual RNAs, transcription-site bursts, and multimeric foci. Material-property perturbations must be separated from global RNA depletion or stress. The most informative endpoint is functional MBNL availability linked to a processing signature and muscle phenotype.

Model selection should follow the claim. C2C12 cells are useful for controlled differentiation and imaging but do not reproduce the multinucleated architecture, loading, innervation, or aging of adult human muscle. Primary patient myoblasts preserve disease alleles but vary in repeat length, passage, epigenetic history, and differentiation capacity. Mouse models enable regeneration and force measurements but can differ in repeat context, transcript level, and tissue distribution. Human biopsy establishes clinical relevance yet offers limited perturbation and spatial sampling. Strong conclusions use these models as complementary tests without treating one as a universal standard.

Negative results should be equally explicit. Failure to detect a compartment change is informative only if the assay could resolve the expected magnitude, measured the relevant isoform, and included the proposed transition in the observation window. Failure to rescue can reflect incorrect mechanism, inadequate expression, or a rescue construct that lacks proper processing or localization. Predefined positive controls, blinded imaging, and reporting of detection limits are necessary before absence is interpreted as biological dispensability.

8.2. Contested Claims and Decisive Experiments

Table 4 links each contested claim to a discriminating experiment and a result that would weaken the proposed mechanism.

These experiments are falsifiable. If NEAT1_2-specific assembly rescue does not restore the muscle phenotype, paraspeckle architecture is unlikely to be the required causal layer. If MALAT1 timing does not alter the sign of the differentiation phenotype under matched conditions, temporal state cannot explain the published contradiction. If endogenous DM1 RNAs remain predominantly single molecules across repeat lengths and transcription rates, a bulk condensation model would have limited relevance to sequestration. If molecular correction consistently fails to improve a mature muscle endpoint despite adequate exposure and duration, researchers must address downstream tissue damage or parallel mechanisms.

The broader repeat-expansion literature identifies common principles involving repeat structure, RNA-binding-protein redistribution, and multiple downstream processing defects, while emphasizing the disease-specific effects of sequence, genomic locus, tissue, and developmental context [101,102]. Therapeutic reviews similarly document a wide portfolio of modalities without resolving which combination will provide durable multisystem benefit [103]. DM1 and DM2 remain clinically and molecularly heterogeneous diseases and cannot be reduced to a single focus-formation disorder.

9. Conclusions

Nuclear noncoding RNAs influence skeletal muscle through several spatial mechanisms that should not be reduced to a single condensate model. NEAT1 links differentiation to paraspeckle association, chromatin regulation, and PABPN1/MATR3-dependent RNA processing, but isoform-specific architectural causality in muscle remains to be shown. MALAT1 regulates nuclear processes and muscle state without being required for gross speckle formation, and its opposing myogenesis phenotypes require a controlled temporal comparison. FRG2 transcripts connect a nucleolar RNA to rDNA transcription and translational capacity in FSHD cells. Expanded DMPK and CNBP RNAs produce a well-supported toxic-RNA and spliceopathy axis, yet individual retained transcripts can be pathogenic without forming bulk condensates.

The most useful unifying principle is therefore causal layering. Spatial localization identifies where a mechanism may occur. Perturbation and rescue determine whether an RNA or assembly is required. In vivo and patient evidence establish physiological relevance. Target engagement plus molecular and tissue rescue define translational progress. Applying these layers consistently will clarify which nuclear structures organize normal muscle RNA processing, which assemblies reflect pathology, and which intervention points are sufficiently supported for therapy.