CELF family of RNA-binding proteins: roles in disease biology and potential for therapeutic intervention

Abstract

The CUG-BP and Elav-like (CELF) family of RNA-binding proteins are key regulators of post-transcriptional gene expression, coordinating alternative splicing, mRNA stability, and translation. Although individual members, particularly CELF1 and CELF2, have been extensively characterized, a systematic, paralog-resolved integration of structural determinants, regulatory mechanisms, and disease relevance across all six CELF proteins remains limited. Here, we establish an integrative framework linking conserved RNA recognition motifs and divergent linker domains to context-dependent regulatory outputs, mediated by phosphorylation, nucleocytoplasmic dynamics, and RNA network interactions. We further highlight the neuron-enriched CELF3–CELF6 subfamily, consolidating emerging evidence that extends their roles beyond neural splicing into cancer-associated regulatory programs. Notably, we delineate functional divergence within the family, with CELF1 frequently acting as an oncogenic driver in contrast to the tumor-suppressive role of CELF2, while positioning less-characterized paralogs within this regulatory spectrum. Together, this work defines a unified structure–function–disease axis for CELF proteins and provides a conceptual framework for their prognostic and therapeutic exploitation. However, current CELF-targeted strategies remain largely preclinical and face key translational challenges, including paralog selectivity, off-target effects, and delivery barriers such as limited blood–brain barrier penetration. Accordingly, the most immediate clinical utility of CELF biology is likely to lie in biomarker development and patient stratification, rather than direct therapeutic intervention.

Introduction

The CELF (CUG-BP and Elav-like) family comprises six evolutionarily conserved RNA-binding proteins that are key regulators of post-transcriptional gene expression across metazoans and plants [1, 2]. Initially named for two prototypical members, CUG-BP (CELF1), which binds expanded CUG trinucleotide repeats implicated in myotonic dystrophy, and ETR-3 (CELF2), first characterized in fetal heart for its role in alternative splicing, the family now encompasses six paralogs: CELF1 (BRUNOL2/CUGBP1/NAPOR/EDEN-BP), CELF2 (BRUNOL3/ETR3/CUGBP2/NAPOR2), CELF3 (BRUNOL1/TNRC4/ETR1/CAGH4/ERDA4), CELF4 (BRUNOL4), CELF5 (BRUNOL5), and CELF6 (BRUNOL6) [1–3]. CELF1 and CELF2 exhibit broad expression in heart, skeletal muscle and brain, whereas CELF3 to CELF6 are largely restricted to neural tissues [4, 5]. Transcriptomic and proteomic atlases such as the Human Protein Atlas indicate particularly high expression of CELF2 and CELF4 in cortex, hippocampus and amygdala, while CELF3 and CELF5 are enriched across forebrain, midbrain and hindbrain but show relatively low expression in the corpus callosum and pons in mammals [2, 6–9]. These spatial expression patterns are evolutionarily conserved, from Caenorhabditis elegans, where the muscle isoform ETR1 contrasts with the neuron-specific UNC-75, to mammalian systems [5, 10].

At the structural level, all CELF proteins share a tripartite RNA-recognition architecture comprising two N-terminal RRMs and one C-terminal RRM separated by a serine/threonine-rich divergent domain of ~ 160–230 amino acids [3, 11–17]. The canonical RNP1 and RNP2 motifs confer affinity for UGU/CUG-rich RNA, while structural and NMR studies reveal tandem RRM1–RRM2 recognition of UGUU motifs and a noncanonical RRM3 pocket that enhances sequence selectivity [14, 18–20]. This modular arrangement contributes to the family’s RNA-binding specificity and splicing regulation, exemplified by CELF1-mediated inclusion of cardiac troponin T exon 5 [21]. The divergent linker acts as a phosphorylation hub that tunes RNA affinity and couples recognition to localization [1, 22, 23]. Through localization signals embedded within this region, CELF1/2 dynamically shuttle between nucleus and cytoplasm, whereas neuronal members CELF3–5 are predominantly nuclear [23, 24]. Post-translational modification, RNA binding, and developmental context further modulate compartmental distribution and functional output [22–25], integrating structure, localization, and regulation into a unified post-transcriptional control system.

Building on these localization patterns, the compartmental distribution of CELF proteins dictates their regulatory scope and biological consequences [22]. The shuttling paralogs CELF1 and CELF2 operate across both nuclear and cytoplasmic compartments, coordinating pre-mRNA splicing, mRNA turnover, and translation, whereas the primarily nuclear CELF3–5 act within neuronal spliceosomal networks in a context- and isoform-dependent manner [10, 22, 26]. Specifically, their nuclear-cytoplasmic distribution varies across developmental stages and is further modulated by alternative splicing of the CELF transcripts themselves [10, 22, 26]. Nuclear CELFs promote alternative splicing, RNA editing, and alternative polyadenylation through binding to UG- or CUG-rich intronic and 3′-end elements, while cytoplasmic CELFs target mature mRNAs at GU-rich 3′ UTRs to modulate deadenylation, stability, and translational efficiency [26, 27]. These compartment-specific programs integrate into broader developmental circuits: in Drosophila and C. elegans, CELF homologs direct muscle and neuronal splicing; in vertebrates, they shape cardiac and neural maturation, with dysregulation leading to hypertrophy and synaptic dysfunction [28–31]. CELF1 commonly acts as an oncogenic driver, promoting proliferation, epithelial–mesenchymal transition, and therapy resistance through GU-rich mRNA stabilization and MAPK activation, whereas CELF2 functions predominantly as a tumor suppressor, silenced by DNA methylation and associated with aberrant splicing and ferroptosis sensitivity. Other neuronal paralogs, including CELF3 to CELF6, are increasingly linked to tumor microenvironment remodeling and neuroendocrine aggressiveness [5, 32–43]. Together, these data establish CELFs as compartment-defined regulators that couple localization and post-transcriptional control to development, homeostasis, and disease.

This review synthesizes genomic, structural and functional data to provide a unified framework for CELF biology and its translational implications. We reconcile nomenclature, integrate atlas-level expression data, emphasize underexplored neuronal paralogs CELF3 to CELF6, and examine emerging therapeutic strategies and biomarker opportunities [2, 3, 12, 27, 44]. The subsequent sections detail genomic and structural features, mechanisms of RNA regulation, disease associations, and prospects for therapeutic targeting and clinical translation [2, 3, 5, 27–29, 45].

To complement the schematic overview shown in Fig. 1, a structured summary of CELF family members, including expression patterns, subcellular localization, and core functions, is provided in Table 1.

Fig. 1.

CELFs share a conserved tripartite architecture of tandem RRM1/RRM2, a divergent linker, and RRM3. Phylogenetically, CELF1/2 are broadly expressed (heart, skeletal muscle, brain), whereas CELF3–6 are neuronally restricted, with CELF6 also present in kidney and testis. During embryogenesis, CELF1/2 are abundant in myogenic tissues, while CELF3–6 localize to the nervous system. The C. elegans homologs ETR-1 and UNC-75 mirror these tissue specializations. Collectively, CELFs act as modular post-transcriptional regulators coordinating splicing, mRNA stability, and translation. The figure was generated with BioRender (https://biorender.com)

Table 1.

Overview of CELF family members: expression, localization and core functions

Paralog	Alternative names	Chromosomal location	Tissue expression	Subcellular localization	Key RNA targets/motifs	Major functions	Disease relevance	Refs
CELF1	CUGBP1, BRUNOL2, NAPOR, EDEN-BP	11p11.2	Ubiquitous; enriched in heart, skeletal muscle, brain	Nucleus ↔ Cytoplasm (shuttling)	GU-rich elements, UGU(U/G) motifs	Alternative splicing, mRNA decay (GRE-mediated), translation regulation	Myotonic dystrophy type 1, cardiac hypertrophy, multiple cancers (oncogenic context-dependent)	[4, 11, 27, 43, 46]
CELF2	ETR-3, BRUNOL3, CUGBP2	10p14	Broad; enriched in immune and neural tissues	Predominantly nuclear; stress-dependent cytoplasmic relocalization	UG-rich motifs, UGUU elements	Alternative splicing, APA regulation, mRNA stability, translation repression	Tumor suppressor (breast, lung, pancreatic), neurodevelopmental disorders	[1, 24, 34, 35, 47]
CELF3	BRUNOL1, TNRC4, ERDA4	1q21.3	Neuron-enriched (brain, pituitary); testis	Mainly nuclear; phosphorylation-dependent shuttling	UG-rich elements	Neuronal alternative splicing (e.g., MAPT, TNNT2), RNA–lncRNA interaction	Neurodegeneration (Tau-related), cancers (GBM, CRC), viral response	[48–52]
CELF4	BRUNOL4	18q12.2	CNS-specific (cortex, hippocampus, cerebellum)	Nuclear → cytoplasmic (developmentally regulated)	(U)GU-rich 3′UTR motifs	Splicing regulation + mRNA localization, stability, translation	Epilepsy, ASD, neurodevelopmental disorders, cardiotoxicity	[10, 18, 25, 53]
CELF5	BRUNOL5	19p13.3	Neuron-enriched (cortex, hippocampus, hypothalamus)	Likely cytoplasmic-biased (context-dependent)	UG/GU-rich motifs	Alternative splicing, potential mRNA regulation (less defined)	Glioblastoma prognosis, motor neuron disease (associative evidence)	[3, 6, 27, 54, 55]
CELF6	BRUNOL6	15q26	Brain and kidney (high); low elsewhere	Nucleus + cytoplasm + neuronal projections	UGU-rich 3′UTR motifs	Splicing, RNA editing, translational repression	ASD, neurobehavioral disorders, tumor suppressor (CRC, lung, TNBC)	[3, 12, 56–59]

Genomic, structural features and basic functions of CELFs

The CELF gene family comprises six paralogous, evolutionarily conserved RNA-binding proteins that together orchestrate post-transcriptional gene regulation across metazoans and plants [3, 5]. Although this modular framework is conserved, CELF paralogs differ markedly in gene organization, expression patterns, and subcellular localization, and these differences underlie their distinct physiological and pathological functions [12]. A comparative account of their genomic context, domain architecture and representative functions is therefore essential to distinguish conserved mechanisms from paralog-specific adaptations and to link structural features with regulatory outcomes [60]. Detailed structural descriptions and RNA-binding mechanisms form the foundation for understanding how CELFs achieve functional specialization [11, 19].

At the domain level, each CELF paralog contains three RNA recognition motifs, RRM1, RRM2 and RRM3, that flank a serine/threonine-rich divergent linker of approximately 160–230 amino acids (Fig. 1) [1]. The linkers vary among family members and contribute to the functional distinction between the ubiquitously expressed CELF1/2 and the tissue-restricted CELF3–6; in some vertebrates CELF3 further carries a 15–18-residue polyglutamine insertion [19, 61]. Each RRM adopts the canonical βαββαβ fold and contains conserved RNP1 and RNP2 motifs, with juxtaposed aromatic residues that stack on RNA bases [62, 63]. C-terminal extensions and inter-RRM sequences further modulate RNA-binding affinity and sequence specificity [64]. Nuclear magnetic resonance studies indicate that RRM1 binds UGU and CUG motifs promiscuously while RRM2 preferentially recognizes UGUU motifs [19, 61]. Mechanistically, these structural motifs enable CELFs to recognize GU-rich elements (GREs) and recruit specific co-factors to execute their regulatory roles.

These conserved structural principles are reflected in functional specialization. CELF paralogs exhibit both overlapping and distinct regulatory repertoires: for example, CELF1 primarily directs alternative splicing of transcripts implicated in myotonic dystrophy type 1 (e.g., cardiac troponin T and the insulin receptor) by binding to intronic GREs and modulating the recruitment of core spliceosomal components [5, 65]. CELF2 additionally influences cytoplasmic mRNA stability and translation, typically by recruiting deadenylases to the 3′ UTR to promote transcript decay or by remodeling ribonucleoprotein (RNP) complexes to control translation initiation [66]. Although CELFs share the tripartite RRM scaffold with the Elav/Hu family, phylogenetic analyses place them in a separate lineage in which divergent linker domains drive subgroup-specific regulatory programs [5, 67]. Consistent with these mechanistic distinctions, the temporal and spatial expression patterns of CELF paralogs further underscore paralog specialization: CELF1 was first identified as a nuclear CUG RNA-binding factor, CELF2 emerged from fetal-heart and other cDNA libraries, and CELF3 was cloned from neural tissue [1, 17, 68, 69]; CELF1 and CELF2 are broadly expressed with enrichment in heart, skeletal muscle and brain, whereas CELF3–CELF6 are largely neuron-enriched, with CELF3 and CELF5 predominantly neuronal, CELF6 extending to kidney and testis, and CELF4 showing probe-dependent patterns [12, 28, 67, 70]. Developmental profiling reinforces this division: CELF1/2 are abundant in myogenic tissues while CELF3–6 are mainly restricted to the nervous system, a pattern conserved across species (Fig. 1) [3, 67, 71, 72].

Collectively, the modular RNA-binding architecture, combinatorial motif usage and spatially patterned expression render CELFs versatile post-transcriptional regulators. At the same time, the full spectrum of their RNA-recognition modalities and the context-dependent rules that couple structure to cellular function remain incompletely characterized, providing the rationale for the subsequent, paralog-resolved analyses presented in this section [11, 73, 74].

CELF1

CELF1 is a ubiquitously expressed RNA-binding protein encoded at 11p11.2. Its ~ 99.6—kb, 15-exon locus produces one canonical transcript and multiple isoforms, with enriched protein expression in brain, heart, and testis [4, 43, 46]. Structural analyses (PDB 2N1I, 6N5N) reveal three RNA recognition motifs (RRMs) that mediate sequence-specific binding to GU-rich elements (GREs) [11]. RRM1–RRM2 cooperatively recognize a UUGUU core via stacking and hydrogen bonding, whereas RRM3 preferentially binds UG repeats; a C-terminal nuclear localization signal (NLS) directs nuclear import (Fig. 2) [1, 43, 46]. Consistent with this architecture, CELF1 binds GU-rich and UGU(U/G) motifs with high affinity and can engage additional sequence contexts, including CCG repeats in C/EBPβ, G/C-rich elements in the cyclin D1 5′UTR, and U(A/G) repeats in Eg and c-mos 3′UTRs [1, 14, 75–78]. Transcriptome-wide analyses further show that CELF1 binding is enriched in introns and 3′UTRs, supporting its dual role in nuclear and cytoplasmic RNA regulation [74, 79].

Fig. 2.

Mechanistic framework of CELF-mediated regulation of alternative splicing, mRNA stability, and translation. CELF proteins bind UG-rich elements in target RNAs to coordinate post-transcriptional regulation in a compartment- and position-dependent manner. In the nucleus, CELF1/2 regulate alternative splicing by binding intronic elements flanking cassette exons: upstream binding inhibits U2AF65 recruitment and promotes exon skipping, whereas downstream binding enhances U2 snRNP recruitment and promotes exon inclusion. CELF3–6 contribute to tissue-specific splicing through similar positioning logic and interactions with other RBPs. Following mRNA export, CELF proteins bind 3′UTR GU-rich elements to control cytoplasmic mRNA fate. CELF1 promotes deadenylation-dependent mRNA decay, whereas CELF2 can stabilize transcripts while repressing translation, indicating functional uncoupling of mRNA stability and translational output. In neurons, CELF4/6 regulate mRNA localization and translation via polysome association and RNA granules. Phosphorylation-dependent shuttling of CELF2 further links nuclear and cytoplasmic functions. The conserved RRM1–RRM2–linker–RRM3 architecture underlies RNA recognition and subcellular distribution of CELF proteins

CELF1 acts as a multi-layered post-transcriptional regulator linking RNA binding to splicing, mRNA stability, and translation. In the nucleus, CELF1 regulates alternative splicing in a position-dependent manner, where upstream intronic binding promotes exon skipping and downstream binding favors exon inclusion, consistent with its predominant intronic occupancy and control of cassette exon usage (Fig. 2) [27, 80]. In the cytoplasm, CELF1 promotes GRE-dependent mRNA decay by binding 3′UTR GU-rich elements and facilitating deadenylation, thereby driving transcript destabilization and coordinated turnover of target mRNAs [74, 79]. In parallel, CELF1 modulates translation, often coupled to poly(A) tail shortening, linking mRNA stability with translational output [1, 74]. CLIP-seq and RIP analyses identify targets including LMO4, BAG1, PKM, SIX5, and DMPK, and transcriptomic profiling reveals dysregulation of proliferation- and angiogenesis-associated mRNAs in cancer [27]. In HeLa cells, CELF1 associates with hundreds of GRE-containing transcripts, whereas in disease models such as diabetic mouse heart it regulates extensive alternative-splicing programs, predominantly affecting cassette exons [74, 79, 80].

While CELF1 is predominantly characterized as an oncogenic regulator, its activity is dynamically modulated by post-translational modifications, particularly phosphorylation, which influences its stability, RNA-binding affinity, and target selection, thereby reshaping its regulatory output in a context-dependent manner [27]. It is recurrently dysregulated in melanoma, colorectal carcinoma, and malignant T-cell leukemogenesis, with frequent copy-number and expression alterations in tumors [43, 81], and is further modulated by noncoding RNAs in contexts such as glioma and cardiac fibrosis [82].

During development, CELF1 cooperates with CELF2 to regulate cardiac, muscle, and neural differentiation [83, 84], whereas its misexpression contributes to myotonic dystrophy type 1 and myocardial hypertrophy through disruption of the CELF–MBNL autoregulatory loop [23, 43, 85, 86].

CELF2

CELF2 is a 484-amino-acid RNA-binding protein encoded at 10p14 (NCBI Gene ID 10659; UniProt Q9BZQ8), comprising three RNA recognition motifs (RRM1–3) flanking a divergent intrinsically disordered region (IDR). This modular architecture enables sequence-specific recognition of single-stranded UG-rich elements, including UG repeats and UGUU motifs, which serve as its primary cis-regulatory targets across pre-mRNAs and mature transcripts [26, 87]. While the tandem RRMs mediate RNA binding, the divergent IDR confers functional specificity: its deletion abolishes splicing regulatory activity without substantially impairing RNA binding, and domain swaps reprogram target selection [1, 88].

Consistent with this architecture, CELF2 functions as a position-dependent regulator of RNA metabolism that integrates alternative splicing, mRNA stability, and 3′ end processing within a unified post-transcriptional framework. Binding to intronic UG-rich elements flanking cassette exons enables CELF2 to modulate spliceosome assembly in a spatially resolved manner: downstream binding enhances U2 snRNP recruitment to the upstream branchpoint and promotes exon inclusion, whereas upstream or branch site–proximal binding interferes with U2AF65 recognition, thereby inhibiting splice-site definition and favoring exon skipping (Fig. 2) [1, 26, 88]. This establishes a context-dependent logic of exon selection analogous to other splicing regulators. Beyond splicing, CELF2 extensively associates with 3′ untranslated regions, where interaction with GU-rich or AU-rich elements modulates transcript fate, including stabilization of specific targets (e.g., FAM198B) or facilitation of decay depending on transcript context [25]. CELF2 further contributes to alternative polyadenylation and intron retention by competing with core 3′ end processing factors, thereby reshaping 3′UTR architecture and post-transcriptional output [26].

At the post-transcriptional output level, CELF2 exerts context-dependent control over mRNA fate by functionally uncoupling transcript stability from translational efficiency. Cytoplasmic CELF2 can repress translation despite stabilizing bound transcripts, as exemplified by COX-2 and Mcl-1 regulation, where CELF2/CUGBP2 binding enhances mRNA stability while simultaneously inhibiting translation. Under stress conditions such as UV or γ irradiation, cytoplasmic relocalization of CUGBP2 reinforces this effect, promoting COX-2 mRNA stabilization while maintaining translational repression, thereby limiting proliferation and inducing apoptosis through Mcl-1 downregulation (Fig. 2) [22, 45, 89–91].

At the systems level, CELF2 regulatory capacity is further expanded by IDR-mediated liquid–liquid phase separation, enabling assembly of ribonucleoprotein condensates that concentrate RNA substrates and regulatory cofactors [92]. Subcellular distribution is dynamically controlled by two nuclear localization signals and a CRM1-dependent nuclear export signal, with predominant nucleoplasmic localization but context-dependent redistribution to cytoplasmic granules under stress or signaling cues [1, 24]. Developmental regulation further refines this compartmentalization, with distinct nuclear and cytoplasmic isoforms observed during cardiogenesis and differentiation [4, 93].

Transcriptome-wide analyses underscore the breadth of CELF2-dependent regulation. In MCF7 cells, CELF2 re-expression remodels at least 82 alternative splicing events linked to proliferation and signaling pathways, including ULK1, CARD10, RHBDF2, FBXL2, and NPTN [32, 94]. In T cells, inducible CELF2 controls a substantial fraction of activation-responsive splicing programs, coordinating exon selection in targets such as MAP2K7/MKK7 and TRAF3 through cooperation or competition with other RBPs, including hnRNPC and RBFOX family members [95–97]. CELF2 also contributes to widespread alternative polyadenylation and autoregulation via its own 3′UTR [26].

CELF2 predominantly acts as a tumor suppressor across multiple cancers. Its expression is reduced in pancreatic cancer, node-positive triple-negative breast cancer, lung squamous carcinoma [34, 35, 47], and colorectal cancer, correlating with advanced stage and poor prognosis [35, 98]. Beyond oncology, CELF2 dysregulation contributes to heart failure, diabetes, and myotonic dystrophy [96], while variants affecting nuclear localization are associated with neurodevelopmental disorders, including epileptic encephalopathy and autism spectrum disorder [99, 100].

CELF3

CELF3 (TNRC4/Brunol1/CAGH4/ERDA4; HGNC 11967; NCBI Gene 11189) is located at 1q21.3 and encodes multiple alternatively spliced isoforms (465–505 aa) characterized by tandem RRM1–2, a low-complexity polyglutamine (polyQ) linker, and a C-terminal RRM3 [1]. Like other CELF family members, CELF3 preferentially binds single-stranded UG-rich elements within target RNAs, providing the molecular basis for its sequence-specific regulatory activity [25]. Residues 192–231 are sufficient for targeting nuclear Cajal body–like structures and share weak similarity with DUF630, suggesting a role in subnuclear RNA processing domains [3, 101]. Its nucleocytoplasmic distribution is dynamically regulated by a C-terminal Lys/Arg-rich nuclear localization signal and a CRM1-dependent nuclear export signal, and can be further modulated by phosphorylation, enabling context-dependent partitioning between nuclear and cytoplasmic compartments [1, 3]. Consistently, CELF3 localizes predominantly to the nucleoplasm with minor cytoplasmic pools and exhibits enriched expression in brain and testis, supporting roles in tissue-restricted RNA processing programs [51, 102].

CELF3 regulates alternative splicing by binding to intronic UG-rich elements flanking cassette exons, thereby modulating exon inclusion. The regulatory outcome is determined by binding position relative to the regulated exon: interactions upstream generally favor exon skipping, whereas downstream binding promotes exon inclusion, reflecting position-dependent control of splice-site recognition. Functional studies demonstrate that CELF3 promotes inclusion of MAPT (Tau) exon 10 and TNNT2 exon 5 by modulating spliceosome assembly and coordinating competition with other RNA-binding proteins, including PTBP1 and multiple SR proteins (SRSF family), while cooperating with auxiliary factors such as Tra2β and U1 snRNP-associated complexes [3, 48–50, 103, 104]. Through these interactions, CELF3 contributes to tissue-specific splicing programs, particularly in neuronal and cardiac systems.

Beyond splicing, direct evidence for CELF3-mediated regulation of mRNA stability and translation remains limited. At the family level, CELF proteins regulate mRNA turnover and translational efficiency through binding to GU-rich elements within 3′ UTR and interaction with decay or translational machinery [25, 60]. Consistent with this framework, CELF3 has been proposed to participate in post-transcriptional regulations beyond splicing. Supporting evidence from vertebrate systems indicates that CELF3 homologs can associate with 3′UTRs and modulate translational output, for example by enhancing Cyclin A translation [105]. However, direct mechanistic evidence for CELF3-dependent control of mRNA stability or decay in mammalian cells is currently lacking, and whether CELF3 exerts transcript-specific stabilizing or destabilizing effects analogous to CELF1 or CELF2 remains to be determined.

At the systems level, CELF3 further participates in RNA–protein network organization through interactions with long noncoding RNAs (lncRNA) such as Gomafu (MIAT), forming spatially distinct ribonucleoprotein assemblies that may contribute to compartmentalized post-transcriptional regulation [101]. Collectively, CELF3 functions primarily as a position-dependent regulator of alternative splicing, with additional roles in mRNA stability and translation that are suggested but not yet mechanistically defined.

From a clinical perspective, CELF3 is implicated in diverse biological processes and disease contexts, including spermatogenesis, osteogenic differentiation, cancer, and viral infection [51, 52]. Its expression shows copy-number variation and is enriched in glioblastoma, pancreatic adenocarcinoma, and metastatic colorectal cancer, where CELF3 clusters with APOBEC3G, EEF1A2, and EIF5AL1 [1, 27, 106]. Human papillomavirus 16 E6 upregulates CELF3, implicating virus-driven regulation [106]. Collectively, CELF3 integrates structural plasticity, phosphorylation-dependent shuttling, and UG-element recognition to coordinate tissue-specific splicing programs [3].

CELF4

CELF4 (GeneID 56853; HGNC 14015) is located on chromosome 18q12.2 and encodes a ~ 486—aa (~ 52 kDa) RNA-binding protein (UniProt Q9BZC1). Similar to other CELF family members, CELF4 exhibits a conserved domain architecture consisting of two N-terminal RNA recognition motifs (RRM1–2), a divergent linker region, and a C-terminal RRM3 [18, 25]. The protein contains multiple phosphorylation sites and a C-terminal nuclear localization signal, indicating that its activity and subcellular distribution are subject to post-translational regulation. Alternative exon 9 usage further generates isoform diversity by modifying the RRM3 region [18, 25].

In contrast to ubiquitously expressed CELF1/2, CELF4 displays a highly restricted and developmentally regulated expression pattern. It is broadly expressed during early stages but becomes predominantly confined to the central nervous system in adulthood, with enrichment in excitatory neurons of the cortex, hippocampus, amygdala, and cerebellum [10]. Concomitant with neuronal maturation, CELF4 shifts from a predominantly nuclear localization to a cytoplasmic distribution, indicating a corresponding transition in its functional role [10].

Consistent with its conserved RRM structure, CELF4 retains the ability to regulate alternative splicing and modulates exon selection in a limited set of transcripts, including TNNT2, MAPT/Tau, ACTN1, and CLCN1, in cooperation or competition with CELF1/2, PTB, and MBNL proteins [16, 107, 108]. However, transcriptome-wide analyses demonstrate that its impact on global splicing is modest, affecting only a small subset of events in vivo [10].

Instead, CELF4 primarily functions as a cytoplasmic regulator of mRNA fate. High-resolution iCLIP studies show that CELF4 preferentially binds (U)GU-rich elements within 3′ untranslated regions, targeting approximately 15–20% of neuronal transcripts [10]. This binding pattern, together with its enrichment in RNA granules and polysome-associated fractions, indicates that CELF4 regulates post-transcriptional processes downstream of splicing, including mRNA stability, subcellular localization, and translational efficiency [10]. Consistently, loss of CELF4 results in redistribution of target mRNAs between soma and neuropil and alters their association with polysomes, thereby affecting the local translation of synaptic proteins and neuronal excitability [10, 109].

CELF4 exhibits dosage-sensitive disease associations. In mice, excitatory neuron–specific deletion causes age-dependent seizures [110]. In humans, variants at the CELF4 locus associate with epilepsy, autism spectrum disorder, colorectal cancer risk, and anthracycline-induced cardiomyopathy. In pancreatic neuroendocrine tumors, CELF4 upregulation enhances proliferation and xenograft growth, whereas reduced expression shows opposite effects [53]. A de novo 18q12.2 translocation truncating CELF4 recapitulates neurodevelopmental defects, highlighting its importance in neuronal and systemic pathophysiology [3].

CELF5

CELF5 (HGNC:14058; Gene ID:60680) is located on chromosome 19p13.3 and encodes a 485-amino-acid (~ 52 kDa) RNA-binding protein. It adopts the conserved CELF domain architecture, comprising two N-terminal RNA recognition motifs (RRM1–2), a divergent linker (~ 160–230 aa), and a C-terminal RRM3 [3]. The linker contains predicted PKC and CK2 phosphorylation sites, and a Lys/Arg-rich C-terminal nuclear localization signal confers nuclear import capability [3]. Alternative splicing generates isoforms differing in the C terminus and 3′UTR, suggesting isoform-specific post-transcriptional regulation [6].

CELF5 belongs to the neuron-enriched CELF3–6 subgroup and is predominantly expressed in the central nervous system, with highest levels in the cerebral cortex, hippocampus, hypothalamus, and amygdala [1]. Its expression arises early during neurulation and shows a broader spatial distribution than other neuronal CELF members [111]. Subcellular localization appears context-dependent, with evidence for predominant cytoplasmic distribution in heterologous systems [54].

CELF5 retains the conserved RNA-binding specificity of the CELF family, preferentially recognizing single-stranded GU-rich elements in both introns and 3′UTRs [13]. This binding enables position-dependent regulation of alternative splicing in a subset of transcripts, although its global impact on splicing appears limited compared with CELF1/2 [13]. In parallel, binding of CELF5 to 3′UTR GU-rich elements suggests a role in cytoplasmic post-transcriptional regulation, including potential effects on mRNA stability and translation [112]. However, direct mechanistic evidence supporting these functions for CELF5 remains limited. Thus, CELF5 is best characterized as a context-dependent regulator of RNA processing rather than a dominant driver of specific post-transcriptional pathways.

Consistent with this functional profile, CELF5 shows association rather than established causality. Expression correlates with motor neuron disease–related pathways [113, 114], alternative splicing signatures linked to overall survival in glioblastoma [27, 55], and responses to human cytomegalovirus infection [27, 115].

CELF6

CELF6 (HGNC:14059; Gene ID:60677) is located on chromosome 15q26 and encodes an ~ 50.5 kDa RNA-binding protein (UniProt Q96J87) containing the canonical CELF architecture of three RRMs separated by a divergent linker [3, 12]. Multiple transcripts and two isoforms are annotated; the longer isoform retains all three RRMs and possesses an unusually extended linker (~ 268 amino acids), exceeding the family-wide range of 160–230 residues. CELF6 expression is enriched in the brain and kidney, with minimal expression in other tissues [1, 12], and within the central nervous system shows region-specific enrichment in monoaminergic-associated areas such as the basal forebrain, substantia nigra, ventral tegmental area, raphe nuclei, locus coeruleus, and medial hypothalamus, while remaining low in the cerebellum [12, 56–58]. Its expression spans embryonic to postnatal development and peaks during early postnatal stages, and the protein localizes to both nuclear and cytoplasmic compartments as well as neuronal processes, indicating the capacity to regulate RNA metabolism across multiple cellular contexts.

Consistent with its conserved RNA-binding domains, CELF6 recognizes UGU-rich elements within target transcripts and participates in post-transcriptional gene regulation [1]. In the nucleus, it modulates alternative splicing in a position-dependent manner, as demonstrated by regulation of TNNT2 exon inclusion and INSR exon 11 skipping, although its global contribution to splicing remains limited [56]. In contrast, CELF6 exhibits a more prominent role in cytoplasmic regulation, where binding to UGU-rich motifs within 3′UTRs enables it to function as a translational repressor of specific neuronal mRNAs [116]. Loss of CELF6 leads to derepression of these targets, including Reln and Fgf13, and increased protein expression (e.g., FOS and FGF13), indicating selective control of 3′UTR-defined transcript pools at the level of translation rather than transcript abundance [116].

Consistent with this functional profile, CELF6 has been implicated in neurodevelopmental and oncologic contexts. Genetic and behavioral studies associate CELF6 with circuits governing social and emotional behavior, with polymorphisms near the gene linked to autism spectrum disorder and Celf6-null mice exhibiting ASD-like phenotypes [56, 59]. In cancer, CELF6 functions predominantly as a tumor suppressor, with reduced expression observed in colorectal carcinoma, lung adenocarcinoma, and triple-negative breast cancer, and re-expression inhibiting proliferation, stemness, and tumor growth [117–122]. These effects are associated with CELF6-dependent remodeling of RNA processing programs, including alternative splicing networks related to apoptosis, cell-cycle control, and extracellular matrix signaling.

Expression of CELF proteins in human cancers and outcome

Altered expression of CELF family proteins is a recurrent feature across multiple human cancers and is frequently associated with clinical outcomes [27, 123, 124]. However, expression changes alone do not fully explain their functional impact on tumor progression [27, 125]. Instead, accumulating evidence indicates that CELF proteins exert their effects through coordinated regulation of post-transcriptional gene expression, linking differential expression to context-dependent control of RNA processing [125].

Rather than functioning as isolated regulators, CELF family members exhibit convergent yet functionally divergent roles in cancer [27]. Across malignancies, CELF1 is commonly upregulated and predominantly exhibits oncogenic functions [122, 123], whereas CELF2 is frequently downregulated and generally acts as a tumor suppressor [34, 36, 126]. In contrast, evidence for CELF3, CELF4, and CELF5 remains inconclusive, whereas CELF6 has been reported to act in a tumor-suppressive manner, although outcome data are limited [5, 119].

In this section, we integrate expression patterns with mechanistic insights to delineate how CELF proteins influence tumor progression through alternative splicing, mRNA stability, and translational regulation, and we further discuss the functional divergence among paralogs within a unified regulatory framework [125].

Colorectal cancer

Colorectal cancer (CRC) is a highly prevalent and deadly malignancy characterized by extensive genetic and epigenetic heterogeneity [127, 128]. Growing evidence demonstrates that its progression and therapeutic resistance are strongly influenced by post-transcriptional dysregulation [125, 129, 130]. Within this regulatory layer, the CELF family of RNA-binding proteins has emerged as a critical regulator of intestinal homeostasis and tumorigenesis [22, 131]. CELF members coordinate RNA splicing, stability, and translation programs that collectively influence key cancer hallmarks, including proliferation, apoptosis, invasion, and chemoresistance, while exhibiting member-specific and sometimes opposing functional effects (Fig. 3) [22, 122, 125, 132]. In CRC, CELF1 generally acts as an oncogenic driver [123, 133], whereas CELF2 displays tumor-suppressive properties [134–136], while CELF3–CELF6 participate in context-dependent regulatory networks that remain incompletely characterized [135–138].

Fig. 3.

Schematic summary of CELF family functions in colorectal cancer (CRC). CELF proteins exhibit distinct yet interconnected regulatory roles in colorectal carcinogenesis. CELF1 acts oncogenically by binding the ETS2 3′UTR to upregulate ETS2 expression, promoting proliferation, invasion, and oxaliplatin resistance. CELF2 exerts tumor-suppressive effects by binding AU-rich elements (AREs) in COX-2 mRNA to repress translation and reduce PGE₂, thereby limiting proliferation and enhancing radiosensitivity; its loss is linked to microRNA-mediated repression (e.g., miR-210—3p). CELF3 shows copy-number alterations and is associated with metastasis and putative carcinogenic RNA targets. CELF4 participates in prognostic RBP networks, though its prognostic directionality remains context-dependent. CELF5 has limited evidence but may harbor rare germline variants linked to CRC risk. CELF6 functions as a tumor suppressor through p53/p21-dependent stabilization of p21 and FBP1 transcripts, upregulating HOXA5, and being epigenetically repressed by LINC01594 via promoter methylation and p53 competition. Collectively, CELF1/2 exert opposing effects, while CELF3–6 contribute context-dependent regulation influencing CRC progression, therapy response, and prognosis. The figure was generated with BioRender (https://biorender.com)

CELF1 promotes tumor progression by binding the 3′ UTR of ETS2 mRNA to enhance ETS2 expression, thereby driving proliferation, invasion, and oxaliplatin resistance [123, 125, 139, 140]. This is further supported by its association with ERBB2 expression and activation of the ErbB signaling pathway through increased AKT and ERK phosphorylation, contributing to metastatic dissemination [43, 133, 141]. Analysis of TCGA datasets indicates that CELF2 is downregulated relative to normal tissue, and higher expression among radiotherapy recipients predicts improved five-year survival [142, 143]. CELF2 exerts tumor-suppressive effects by binding AU-rich elements in the COX2 3′ UTR, stabilizing the transcript while repressing its translation, ultimately reducing prostaglandin E2 (PGE2) production and limiting proliferative signaling [125, 134, 144–146]. CELF2 is further suppressed by oncogenic microRNAs such as miR-210—3p and miR-95, which enhance proliferation and invasion [147–149].

Other CELF family members contribute more variably to CRC progression. CELF3 has been identified through multi-omics analyses as a metastasis-associated RNA-binding protein with distinct copy-number alterations and putative RNA targets linked to tumor progression [137]. CELF4 is primarily implicated in prognostic RBP networks, with recurrent identification in multiple predictive models despite context-dependent expression patterns [138, 150–152]. CELF6 functions as a tumor suppressor by enforcing p53- and p21-dependent G1 arrest through stabilization of p21 and FBP1 transcripts, as well as upregulation of HOXA5 [27, 121, 122]. Its expression is repressed by the long noncoding RNA LINC01594 via promoter methylation and competition with p53, forming the LINC01594–CELF6–CD44 regulatory axis that contributes to CRC progression [135].

These findings indicate that CELF proteins regulate CRC progression through coordinated control of alternative splicing, mRNA stability, and translational efficiency. Rather than functioning as isolated biomarkers, they form a dynamic post-transcriptional regulatory network in which CELF1 and CELF2 exert opposing effects, while CELF3–CELF6 contribute context-dependent regulatory inputs that remain to be fully defined.

Non-small cell lung cancer

Non-small cell lung cancer (NSCLC), which accounts for approximately 85% of lung cancers, remains a major therapeutic challenge because acquired resistance to targeted agents and immunotherapies limits long-term benefit [153–155]. Dysregulation of RNA-binding proteins constitutes an important layer of post-transcriptional control influencing NSCLC pathogenesis and treatment response [156]. Within this context, the CELF family exhibits distinct and sometimes opposing roles [22, 122, 126, 157]. CELF1 is frequently upregulated and correlates with poorer postsurgical survival, supporting malignant phenotypes, while the circular RNA circ_CELF1 contributes to anti-PD-1 resistance [37, 157, 158]. In contrast, CELF2 displays tumor-suppressive activity with in vivo efficacy in restraining tumor growth [159, 160]. CELF6 modulates transcriptional and alternative splicing programs relevant to tumorigenesis [119], whereas functional evidence for CELF3–CELF5 remains limited.

CELF1 promotes NSCLC progression by enhancing tumor cell proliferation and clonogenicity, with its suppression reducing cell viability and growth [37, 129, 158, 161]. Elevated CELF1 downregulates the CCAAT/enhancer-binding protein alpha (C/EBPα) pathway, thereby enhancing proliferation and limiting apoptosis [43]. In addition, circ_CELF1 is increased in primary NSCLC and drives tumor progression and immunotherapy resistance by sponging miR-491—5p to upregulate epidermal growth factor receptor (EGFR) [157]. Supporting evidence from other tumor types shows that CELF1 binds GRE-containing mRNAs involved in apoptosis, proliferation, and motility, reinforcing its oncogenic role [158, 161].

CELF2 is downregulated in NSCLC, particularly in lung adenocarcinoma (LUAD), where reduced expression associates with poorer prognosis and shorter survival [34, 126]. Its tumor-suppressive function is mediated through multiple noncoding RNA–dependent axes. HHIP-AS1 stabilizes HHIP via CELF2 to suppress proliferation, migration, and invasion, whereas reduced HHIP-AS1 or HHIP predicts worse outcomes [159]. In lung squamous cell carcinoma (LUSC), miR-210—3p represses CELF2, activating PI3K/AKT signaling and promoting malignant phenotypes, while CELF2 restoration reverses these effects [160]. Similarly, circLIFR sponges miR-429 to de-repress CELF2, inactivate the PTEN/AKT pathway, and induce apoptosis and cell-cycle arrest in vivo [162].

CELF6 regulates post-transcriptional programs in NSCLC, although the cause of its reduced expression remains unclear [119]. Transcriptomic profiling in A549 cells following CELF6 overexpression reveals widespread gene expression changes, with downregulated genes enriched in immune and inflammatory pathways and cell adhesion processes [119]. CELF6 also remodels alternative splicing of genes involved in p53 signaling and apoptosis, including TP53 and CD44, indicating activation of p53 signaling through splice regulation [117, 119]. These findings suggest that CELF6 coordinates immune and apoptotic pathways through combined transcriptional and splicing control.

CELF3, CELF4, and CELF5 are detectable in lung cancer transcriptomic and proteomic datasets (HPA; TCGA), but their functional and prognostic roles in NSCLC remain undefined.

Collectively, CELF proteins in NSCLC operate through coordinated regulation of alternative splicing, mRNA stability, and translation. CELF1 and CELF2 exhibit opposing oncogenic and tumor-suppressive roles, respectively, while CELF6 contributes to splicing-dependent tumor regulation and CELF3–CELF5 remain insufficiently characterized, highlighting the need for systematic investigation of CELF-centered post-transcriptional networks in NSCLC.

Glioblastoma

Glioblastoma (GBM), the most aggressive form of glioma, exhibits extensive molecular heterogeneity and profound dysregulation of post-transcriptional gene regulation [163–165]. Within this framework, CELF family proteins contribute to tumor behavior through RNA processing mechanisms, with evidence indicating both oncogenic and context-dependent roles [27]. At the expression level, distinct patterns are observed across paralogs: CELF1 is upregulated in GBM and correlates with higher WHO grade and shorter survival, supporting a pro-tumorigenic role [124, 166], whereas CELF2 displays a more complex profile, with higher expression associated with improved overall survival in some cohorts but also linked to proliferative programs in specific tumor cell populations [167]. In contrast, CELF3 expression decreases with glioma grade, suggesting a role in maintaining normal neural RNA programs [168]. Evidence for CELF4 remains limited but points to dysregulation of CELF4-associated regulatory networks in high-grade gliomas [53, 169]. CELF5 has been identified in survival-associated alternative splicing signatures, whereas CELF6 shows tumor-suppressive associations across cancers, although GBM-specific functional data remain limited [27, 55, 122].

CELF1 promotes GBM progression by facilitating cell-cycle progression, partly through repression of cyclin-dependent kinase inhibitor 1B (CDKN1B/p27^Kip1) at the protein level [27, 124]. Its upregulation is reinforced by loss of miR-330-3p, which directly targets the CELF1 3′UTR; restoration of miR-330-3p reduces CELF1 expression and suppresses glioma proliferation and migration, indicating a miRNA–CELF1 axis driving tumor aggressiveness [170]. These findings position CELF1 as a central effector linking post-transcriptional regulation to proliferative signaling in GBM.

CELF2 exhibits context-dependent functions shaped by intratumoral heterogeneity. On one hand, miR-363—3p–mediated repression of CELF2 enhances proliferation, invasion, epithelial–mesenchymal transition (EMT), and Wnt/β-catenin signaling, while its inhibition reduces tumor growth in vivo [171]. On the other hand, CELF2 is enriched in OLIG2-positive, mitotic glioma stem-like cells, where it sustains proliferative capacity by regulating the expression of epigenetic modifiers, including TRIM28 and G9a, thereby promoting H3K9me3 deposition and transcriptional repression of SOX3 [167]. These dual observations indicate that CELF2 can either restrain or support tumor progression depending on cellular context and regulatory inputs.

Other CELF members remain less defined in GBM. CELF3 shows reduced expression with increasing glioma grade and lacks clear survival association, leaving its functional role unresolved [168]. CELF4 exhibits comparable expression between tumor and normal tissues and lacks consistent prognostic association, although network-level analyses suggest involvement in dysregulated RNA programs [53, 169]. CELF5 is detectable and has been implicated in alternative splicing networks, including regulation of GSG1L and pathways such as primary bile acid synthesis, but lacks consistent clinical validation [55, 122]. CELF6 is expressed at lower levels, and higher expression may predict poorer survival, although GBM-specific functional evidence remains lacking [55, 168].

Taken together, CELF proteins in GBM function within a heterogeneous post-transcriptional regulatory landscape. CELF1 consistently promotes tumor progression, whereas CELF2 exhibits context-dependent duality linked to cellular state and microRNA regulation. The remaining paralogs contribute to RNA regulatory networks but lack definitive functional characterization, underscoring the need for integrative analyses combining transcriptomic, epigenetic, and single-cell approaches to resolve their roles in GBM.

Breast cancer

Breast cancer is the most commonly diagnosed cancer in women [172], with triple-negative breast cancer (TNBC) representing ~ 20% of cases and exhibiting poorer survival [173]. CELF RNA-binding proteins influence breast cancer progression in a member-specific manner [22, 32, 119]. CELF1 promotes oncogenic programs [39, 174], whereas CELF2 and CELF6 predominantly exert tumor-suppressive effects [32, 34, 120]. CELF3–CELF5 remain poorly characterized in breast cancer and show limited evidence for functional or prognostic relevance [27, 28].

CELF1 drives epithelial–mesenchymal transition (EMT) and metastatic progression through a translational regulatory program targeting GU-rich elements in 3′UTRs of EMT-associated transcripts [39, 174, 175]. Its protein expression, rather than mRNA, is markedly increased in breast cancer and is necessary and sufficient for mesenchymal transition and metastatic colonization, establishing CELF1 as a central post-transcriptional regulator of tumor aggressiveness [39, 122, 175].

CELF2 exhibits tumor-suppressive activity supported by both clinical and experimental evidence. In TCGA breast cancer cohorts, higher CELF2 expression is associated with improved overall survival, suggesting a potential role in treatment response [142]. At the mechanistic level, CELF2 suppresses proliferation, invasion, and angiogenesis, partly through regulation of alternative splicing programs and downregulation of NFATc1, as well as inhibition of N-cadherin and CD34 expression [27, 176].

CELF2 is epigenetically silenced by promoter hypermethylation in breast cancer, and its restoration suppresses tumor growth while reversing widespread alternative splicing alterations, particularly intron retention events affecting ULK1, CARD10, RHBDF2, FBXL2, and NPTN [32]. Upstream, the tumor-suppressive HR pathway activates CELF2, linking chromatin regulation to post-transcriptional control and supporting the therapeutic relevance of H3K9 methylation inhibitors [177–179]

CELF6 similarly functions as a tumor suppressor, particularly in TNBC, where it stabilizes FBP1 mRNA to inhibit proliferation, migration, and invasion while enhancing paclitaxel sensitivity [120]. Regulatory inputs include microRNA-mediated repression, notably by miR-375, which is elevated in ERα-positive breast cancer and associated with increased proliferation and risk of local recurrence [22, 180–183].

Overall, CELF proteins in breast cancer form a post-transcriptional regulatory axis in which RNA-binding–mediated control of splicing and translation shapes EMT, immune interactions, and treatment response. The opposing roles of CELF1 versus CELF2/6 suggest a functional dichotomy with potential relevance for therapeutic stratification, whereas the limited evidence for CELF3–5 highlights the need for further investigation.

Gastric cancer

Gastric cancer (GC) remains a leading cause of cancer-related mortality despite declining incidence, reflecting aggressive biology and limited therapeutic responsiveness [184, 185]. CELF proteins contribute to GC progression through post-transcriptional regulation, with divergent roles among family members. CELF1 is frequently upregulated and supports tumor growth [186], whereas CELF2 acts as a tumor suppressor whose loss is associated with advanced stage and poor survival [147, 187]. Multi-omics analyses, including TCGA datasets, further position CELF2 within survival-associated regulatory networks, with higher expression linked to favorable outcomes [34, 188]. Evidence for CELF3–CELF6 remains limited.

CELF1 is consistently overexpressed in GC tissues and promotes proliferation by sustaining cell-cycle progression. Its depletion reduces viability and clonogenic growth, accompanied by downregulation of cyclin B1 and cyclin D1, indicating a direct role in maintaining proliferative capacity [43, 186].

CELF2 suppresses proliferation and migration while promoting apoptosis [27]. Upstream regulation is mediated by noncoding RNAs: miR-615—3p directly represses CELF2 enhancing tumor progression [147], whereas circPTK2 restores CELF2 activity via miR-134—5p sequestration and activation of the CELF2/PTEN axis [187]. These data define a convergent regulatory framework in which multiple RNA networks modulate CELF2-dependent tumor suppression.

CELF4 shows limited but notable alterations, including copy-number reduction and potential involvement in oncogenic signaling downstream of H. pylori–induced STAT3 activation, where it is co-upregulated with genes such as FGFR1 and JAK3 [189, 190].

CELF3 and CELF5 exhibit minimal expression and lack prognostic relevance in GC, consistent with their neuronal enrichment. CELF6 has established tumor-suppressive roles in other cancers via p21 regulation, but its function in GC remains undefined [122].

Overall, CELF proteins in GC follow a pattern of oncogenic CELF1 versus tumor-suppressive CELF2, with regulatory effects mediated through RNA-binding–dependent control of cell-cycle progression and noncoding RNA networks, while other family members remain insufficiently characterized.

Pancreatic cancer

Pancreatic cancer, predominantly pancreatic ductal adenocarcinoma (PDAC), remains highly lethal due to pronounced transcriptomic heterogeneity and therapeutic resistance [191, 192]. Dysregulation of RNA-binding proteins contributes to this phenotype, with CELF family members emerging as modulators of tumor behavior and treatment response [193–195]. Among them, CELF2 is consistently downregulated and associated with advanced stage, larger tumors, and poor survival, whereas its restoration suppresses tumor growth and enhances gemcitabine sensitivity [47, 196, 197]. CELF1 evidence in PDAC remains limited, although its oncogenic role in other cancers suggests potential relevance [27, 198]. CELF4, in contrast, is specifically upregulated in pancreatic neuroendocrine tumors (PanNETs), where it correlates with aggressive features and therapeutic response [53].

CELF2 functions as a central tumor suppressor in PDAC by coordinating RNA processing and signaling pathways. Its loss promotes proliferation, invasion, stemness, and resistance to therapy, whereas restoration reverses these phenotypes [35, 199, 200]. Mechanistically, ALKBH5-mediated m6A modification drives YTHDF2-dependent decay of CELF2 mRNA. Downstream, CELF2 regulates CD44 alternative splicing, shifting isoform balance toward CD44v variants and modulating endoplasmic reticulum stress through ERAD pathways [35, 47, 200, 201]. In parallel, CELF2 suppresses NFATc1 and disrupts a feed-forward loop involving ERK1/2 and p38 MAPK, linking RNA processing to oncogenic signaling [35, 176, 202, 203]. Noncoding RNAs further converge on CELF2, while pharmacologic agents such as curcumin restore its expression, enhancing chemosensitivity and inducing apoptosis through COX2/VEGF and TIA1-associated pathways [35, 98, 196, 197].

CELF4 exhibits a context-specific oncogenic role in PanNETs. It is upregulated in tumors, correlates with adverse clinicopathological features, and promotes proliferation, while its depletion restricts tumor growth [53, 204]. Mechanistically, CELF4 modulates mTOR signaling, and its loss sensitizes tumors to everolimus, indicating therapeutic relevance [53, 204]. In contrast, CELF4 exhibits low cancer specificity and no evident PDAC-specific expression pattern based on HPA data. To date, there is no evidence supporting its prognostic relevance in PDAC, as it has not been included in established survival-associated gene sets or multi-cohort prognostic analyses [205, 206].

Other CELF members show limited evidence in pancreatic cancer. CELF1 is moderately expressed in PDAC but lacks consistent prognostic or mechanistic support [207]. CELF3 displays low-to-intermediate expression without defined function [208, 209], whereas CELF5 and CELF6 are minimally expressed and lack validated roles. Current data therefore indicate that CELF2 represents the primary functionally relevant family member in PDAC, with other paralogs remaining insufficiently characterized.

Overall, CELF proteins in pancreatic cancer are dominated by CELF2-dependent tumor suppression mediated through m6A regulation, alternative splicing, and signaling network modulation, while CELF4 contributes to PanNET progression. The limited evidence for other family members highlights a need for further mechanistic investigation.

Other malignancies (HCC, cervical cancer, and additional tumor types)

Compared with colorectal, lung, glioblastoma, breast, gastric, and pancreatic cancers, in which CELF proteins have been more extensively characterized, evidence in other malignancies remains relatively limited and fragmented. Nevertheless, available studies consistently indicate that CELF-mediated post-transcriptional regulation contributes to tumor progression in a context-dependent manner across these cancers.

Primary liver cancer is predominantly hepatocellular carcinoma (HCC), a heterogeneous malignancy with poor prognosis [210, 211]. Across HCC, CELF proteins regulate alternative splicing, mRNA stability, and translation, thereby influencing proliferation, metastasis, and therapeutic response [25, 27]. CELF1 is frequently upregulated and associated with poor survival in TCGA Liver Hepatocellular Carcinoma datasets, supporting a pro-tumorigenic role [212], whereas CELF2 is typically downregulated and suppresses proliferation, migration, and treatment resistance [213, 214]. These effects are further shaped by noncoding RNA networks, including the BACE1-AS/miR-377—3p/CELF1 and miR-95–CELF2 axes, which modulate EMT, tumor growth, and drug response [27, 43, 215–217]. Evidence linking CELF3–CELF6 to HCC remains limited.

In cervical cancer, CELF-mediated RNA regulation contributes to tumor progression beyond HPV-driven oncogenesis [22, 119].CELF1 regulates alternative splicing and promotes anti-apoptotic programs through destabilization of pro-apoptotic transcripts [218–220], whereas CELF2 is associated with improved survival and participates in TCGA-derived splicing networks that stratify patient risk [221, 222]. Additional members such as CELF5 and CELF6 have been implicated through proteomic and genetic analyses, including associations with signaling pathway dysregulation and cancer susceptibility, although mechanistic validation remains incomplete [136, 223, 224].

A cross other malignancies such as oral squamous cell carcinoma [129], bladder [225], ovarian [98], melanoma [226], endometrial cancers [36], and acute leukemia [89], CELF proteins exhibit conserved yet context-dependent functions [22]. CELF2 consistently emerges as a tumor suppressor, with reduced expression linked to enhanced proliferation, invasion, and therapy resistance, as well as poorer clinical outcomes [33, 98, 227]. In contrast, CELF1 frequently promotes tumor growth by destabilizing pro-apoptotic transcripts and supporting proliferative gene expression programs, although its prognostic significance varies by tumor type [140, 198, 226]. Evidence for CELF3–CELF6 across these cancers remains sparse, with most studies limited to expression profiling or indirect association analyses.

Overall, these data indicate that CELF proteins function as context-dependent regulators within post-transcriptional networks rather than uniform oncogenic or tumor-suppressive factors. A recurrent pattern nevertheless emerges in which CELF1 and CELF2 exert opposing effects on tumor progression, while other family members contribute to tissue-specific modulation of RNA regulatory programs. This framework emphasizes that the impact of CELF dysregulation is best interpreted at the level of coordinated RNA processing rather than isolated expression changes, highlighting the need for integrative analyses combining transcriptomics, splicing, and clinical outcome data.

To systematically integrate these findings, a pan-cancer summary of CELF family expression patterns and their associations with clinical outcomes is presented in Figure 4.

Fig. 4.

Pan-cancer landscape of CELF family proteins: expression patterns and outcome associations. A heatmap-style summary matrix summarizing the expression status and clinical outcome associations of CELF family members (CELF1–CELF6) across major human cancers, based strictly on curated evidence described in the main text. Columns represent cancer types, and rows represent individual CELF proteins. Color coding indicates differential expressions: red, upregulation; blue, downregulation; gray, insufficient or unavailable data. Symbols indicate clinical outcome associations: yellow ▲, association with adverse prognosis or aggressive tumor behavior (e.g., increased proliferation, metastasis, therapy resistance, or reduced survival); purple ▼, association with favorable prognosis (e.g., tumor suppression, improved survival, or enhanced treatment response); open circle (○), unclear, inconsistent, or insufficient outcome data. Overall, the heatmap highlights a recurrent pattern across cancers in which CELF1 upregulation is associated with adverse tumor behavior, whereas CELF2 downregulation corresponds to loss of tumor-suppressive function, with retained or higher CELF2 expression generally linked to favorable clinical outcomes. Other family members (CELF3–CELF6) display context-dependent or insufficiently characterized roles. The Fig. was generated with BioRender (https://biorender.com)

Expression and function of CELF proteins in other human diseases

The CELF family of RNA-binding proteins regulates pre-mRNA alternative splicing, C-to-U editing, deadenylation, mRNA decay, and translation, thereby maintaining normal gametogenic, embryonic, muscular, and neural functions [1]. Their dysregulation disrupts tissue-specific RNA programs, contributing to diverse human diseases. In the nervous system, all six CELF members are expressed and implicated in myotonic dystrophy, spinal muscular atrophy, spinocerebellar ataxia, fragile X syndrome, and related disorders [28]. Two disease-related RBP axes exemplify these mechanisms: CELF/MBNL-mediated reactivation of fetal isoforms in myotonic dystrophy, and FUS/TDP-43 nuclear loss with cytoplasmic aggregation in amyotrophic lateral sclerosis [60]. Beyond neurodegeneration, aberrant CELF activity has been linked to Alzheimer’s disease and autism spectrum disorder, suggesting potential therapeutic relevance [2]. In skeletal muscle, CELF and MBNL coordinate splicing programs governing myoblast differentiation and stem cell dynamics [228]. Similar RBP pathologies, including those involving Nova (paraneoplastic opsoclonus–myoclonus–ataxia) and αCP (α-thalassemia) proteins, further illustrate tissue-restricted RNA control in disease [229]. These principles provide a framework for examining CELF dysregulation across neuromuscular, neurodegenerative, and cardiometabolic disorders, emphasizing its role in splicing, mRNA stability, and translational regulation.

Myotonic dystrophy

Myotonic dystrophies 1 (DM1) and myotonic dystrophies 2 (DM2) are autosomal dominant, multisystem disorders caused by toxic RNA repeat expansions (DMPK CTG repeats in DM1; CNBP CCTG repeats in DM2), resulting in widespread splicing dysregulation and systemic functional deficits in muscle, heart, and neural tissues [230, 231]. CELF family proteins, particularly CELF1, act as central mediators of DM-associated splicing dysregulation, shifting pre-mRNA processing toward fetal isoforms and affecting transcripts critical for muscle contraction, insulin signaling, and cardiac electrophysiology, including Chloride Voltage-Gated Channel 1 (CLCN1), Bridging Integrator 1 (BIN1), Insulin Receptor (INSR), and Pyruvate Kinase M (PKM) [48, 232–236]. Dysregulated CELF1, in combination with MBNL loss, underlies the molecular hallmarks of myotonia, progressive weakness, arrhythmia, and insulin resistance, while additional mechanisms such as repeat-associated translation, aberrant polyadenylation, and microRNA perturbation reinforce fetal-type isoforms [23, 129, 237]. Experimental models confirm that CELF1 overexpression recapitulates DM phenotypes and that targeted reduction of CELF1 can partially restore normal splicing and improve histopathological and functional outcomes [91, 235, 238–240]. Collectively, these findings highlight CELF1 as a pivotal post-transcriptional regulator driving DM pathogenesis and provide a framework for therapeutic targeting (Fig. 5).

Fig. 5.

Pathogenic mechanism of Myotonic Dystrophy (DM). The molecular etiology of DM involves a cascade from genetic mutation to multisystemic disease. Mutation: Toxic RNA expansions—(CTG)n in DMPK (DM1) and (CCTG)n in CNBP (DM2)—accumulate as nuclear foci. RBP Imbalance: These toxic foci drive a "dual insult" to RNA-binding proteins (RBPs) by sequestering MBNL1/2 (Loss of Function) and concurrently increasing CELF1 stability and activity via PKC-mediated phosphorylation. Splicing: This RBP imbalance leads to widespread mis-splicing of key transcripts and the aberrant persistence of fetal-type isoforms in adult tissues. Disease: Specific spliceopathy events underlie the clinical hallmarks: CLCN1 (Exon 7a inclusion) induces myotonia; BIN1 (Exon 11 skipping) leads to progressive muscle weakness; INSR (Exon 11 skipping) causes insulin resistance; and TNNT2/PKM promotes cardiac arrhythmia. The Fig. was generated with BioRender (https://biorender.com)

Spinal muscular atrophy

Spinal muscular atrophy (SMA) is an autosomal-recessive α-motor-neuron disorder characterized by progressive proximal muscle weakness and paralysis, caused by homozygous loss of the Survival Motor Neuron 1 (SMN1) gene, with disease severity modified by SMN2 copy number [241]. CELF family proteins, particularly CELF2, are implicated in SMA pathogenesis through post-transcriptional regulation [28]. CELF2 expression is elevated in motor neurons of SMA mouse models and in human patients, and it co-localizes with SMN in neuronal nuclei, with SMN co-immunoprecipitating with CELF2 from mouse brain, indicating a functional interaction [5]. CELF2 is also upregulated in patient-derived muscle cultures and multiple SMA mouse models [242]. Collectively, these findings suggest that CELF2 interacts with SMN and, together with its high expression in the neuromuscular system, likely contributes to the selective vulnerability of motor neurons in SMA [5, 28, 242].

Alzheimer’s disease

Alzheimer’s disease (AD) is the leading cause of dementia in older adults, characterized by progressive cognitive decline associated with amyloid β accumulation and tau pathology [243, 244]. Accumulating evidence implicates CELF RNA-binding proteins in AD by converging genetic, transcriptomic, and mechanistic lines of evidence. Specifically, CELF1 and CELF2 regulate inclusion of exon 3 in Triggering Receptor Expressed on Myeloid Cells 2 (TREM2), a microglial receptor important for amyloid clearance [245, 246]; loss of this splicing event reduces full-length TREM2 and impairs microglial response to amyloid. Genome-wide and eQTL analyses further nominate the CELF1 locus as an AD susceptibility region, with several variants modulating risk particularly in APOE ε4 carriers [247, 248]. At the neuronal level, reduced CELF1 correlates with increased expression of the pathogenic kinesin isoform KLC1_vE, and CLIP-seq demonstrates direct CELF1–KLC1 binding, supporting a mechanism whereby CELF1 loss promotes axonal transport defects relevant to tau propagation [249, 250]. Mechanistically, CELF2 contributes to condensate-based splicing control: its intrinsically disordered region interacts with NOVA2 and SFPQ to influence MAPT exon 10 inclusion and thereby tau isoform balance [48, 92, 251, 252]. At the systems level, CELF2 is broadly downregulated across AD-affected cell types, paralleling a global shift in alternative splicing patterns (notably CLU, SYT1, and FTL) and concomitant DDX5 loss, indicating that CELF2 dysfunction, both dependent on and independent of APOE ε4, acts as a convergent post-transcriptional regulator linking amyloid, tau, and synaptic pathway dysregulation in AD [245].

Epilepsy

Epilepsy affects about 50 million people worldwide, with four major seizure categories and 21 subtypes defined by the International League Against Epilepsy (ILAE, 2025) [253–255]. Dysregulated RNA-binding proteins have emerged as contributors to epileptogenesis, among which CELF4 is a key regulator of neuronal excitability [10]. CELF4 deficiency in mice lowers seizure thresholds, producing handling-evoked and absence-like seizures with strain- and age-dependent penetrance [5, 28, 256]. Loss restricted to excitatory cortical or hippocampal neurons is sufficient to induce convulsions, consistent with CELF4 enrichment in these regions, while inhibitory neuron deletion has no effect [110]. At the cellular level, CELF4 loss enhances intrinsic excitability through ion channel dysregulation at the axon initial segment [257]. Beyond CELF4, human and zebrafish studies implicate CELF2: pathogenic variants cause infantile spasms, and CELF2 loss elicits seizure-like behaviors reversible by wild-type mRNA, reflecting disrupted neuronal and metabolic gene networks [258]. Collectively, CELF4 and CELF2 maintain neuronal excitability homeostasis through post-transcriptional RNA regulation.

Autism spectrum disorder

Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by social communication deficits and repetitive behaviors with heterogeneous genetic origins [259, 260]. Among CELF family proteins, CELF6 shows the strongest link to ASD: its loss in mice reduces ultrasonic vocalizations, impairs behavioral flexibility, and decreases brain serotonin, consistent with disruption of neuromodulatory circuits [59, 261]. Human studies also identify CELF6 variants conferring subset-specific ASD risk, supported by its early and widespread expression in serotonergic and diencephalic regions [56]. CELF2 contributes to synaptic maturation, as its partial loss in mice causes social and anxiety-like behaviors with altered dendritic and synaptic gene programs [100]. CELF4 has been proposed as an additional candidate based on expression in cortical and limbic regions and preliminary human genetic evidence [28, 262]. Collectively, these findings implicate CELF6 as a principal regulator of serotonergic and social communication pathways, with CELF2 and CELF4 exerting secondary effects on cortical development and behavior.

Dilated cardiomyopathy

Dilated cardiomyopathy (DCM) is characterized by left ventricular dilation with systolic impairment and frequent diastolic and right-sided dysfunction [263, 264]. Post-transcriptional regulation by CELF family RNA-binding proteins has been strongly implicated in DCM pathogenesis. Elevated CELF1 expression, observed in mouse models of DCM, myotonic dystrophy type 1 hearts, and after myocardial infarction, correlates with reduced Connexin 43 (Cx43) abundance, whereas CELF1 depletion preserves Cx43 and improves cardiac performance, suggesting that CELF1 overactivation drives adverse remodeling [41, 265, 266]. Cumulative evidence indicates that both gain and loss of CELF1 disrupt cardiac transcriptome homeostasis, leading to pathological remodeling and impaired contractility, highlighting the requirement for finely tuned CELF1 activity in maintaining cardiac structure and function [30, 40, 267]. Under hemodynamic stress, CELF1-mediated regulation of phosphatidylethanolamine-binding protein 1 (PEBP1) and mitogen-activated protein kinase (MAPK) signaling contributes to hypertrophy, fibrosis, oxidative stress, and apoptosis, supporting CELF1 inhibition as a potential therapeutic avenue [86]. Beyond CELF1, genetic and experimental evidence implicates CELF4 as an additional regulator of cardiac stress responses; its variation associates with anthracycline-related cardiomyopathy, and loss of CELF4 confers protection against pressure-overload-induced dilation and fibrosis, indicating a conserved role for CELF proteins in stress-induced cardiac remodeling [268, 269].

Post-infarction cardiac dysfunction

Myocardial infarction (MI) remains a major cause of heart failure, with substantial morbidity despite therapeutic advances [270, 271, 272]. Among regulatory factors, CELF family RNA-binding proteins, particularly CELF1, play a key pathogenic role. In mouse models, cardiomyocyte-specific CELF1 depletion preserves ventricular function, reduces infarct size and fibrosis, and maintains Tnnt2, Tbx5, and Connexin 43 expression and localization, indicating that sustained CELF1 activation drives adverse remodeling and conduction disturbances after infarction [41]. Collectively, these data position CELF1 as a central post-transcriptional regulator of post-MI cardiac dysfunction, linking RNA splicing dysregulation to progressive ventricular failure.

Insulin resistance

Insulin resistance (IR), a key metabolic abnormality spanning obesity, prediabetes, and type 2 diabetes, affects a large proportion of at-risk individuals [273, 274, 275]. Growing evidence implicates CELF family RNA-binding proteins, particularly CELF1, as post-transcriptional regulators of insulin signaling and glucose metabolism. CELF1 modulates INSR splicing toward the mitogenic IR-A isoform and reprograms metabolic RNA networks governing glycolysis, adipogenesis, and insulin secretion, thereby promoting systemic insulin resistance [276, 277, 278]. In diabetes and myotonic dystrophy type 1, CELF1 overexpression parallels ERK/PKC activation and widespread mis-splicing in skeletal and cardiac muscle, linking CELF1-driven transcriptomic remodeling to peripheral IR and diabetic cardiomyopathy [279, 280, 281, 282]. Collectively, these findings position CELF1 as a nodal regulator of metabolic homeostasis and a potential therapeutic target in insulin-resistant states.

CELF protein as therapeutic targets in diseases

The CELF family of RNA-binding proteins governs alternative splicing, mRNA stability, and translation, positioning these factors as mechanistically compelling but clinically constrained therapeutic targets across multiple diseases. In myotonic dystrophy type 1, aberrant protein kinase C signaling enhances CELF1 activity, and PKC inhibition restores CELF-dependent RNA programs and cardiac function in preclinical models [283, 284, 285]. In hepatic fibrosis and cancer, disruption of CELF1 binding to GU- or UG-rich 3′ untranslated region elements by small molecules or repurposed thiopurines suppresses fibrotic activation [43]. In neurodegeneration, altered CELF expression intersects with Alzheimer’s disease pathways through regulation of TREM2 and MAPT, implicating CELF-mediated RNA control in microglial activation and tau dysfunction [92, 245]. Multiple molecular strategies have been explored to therapeutically modulate CELF activity, including RNA interference or CRISPR interference to silence CELF genes or upstream drivers, antisense oligonucleotides to correct CELF-dependent splicing defects in DM1, and small-molecule inhibitors targeting CELF1–RNA interactions, combined with advances in epigenetic and delivery systems [286]. However, clinical translation remains constrained by several factors, including limited target selectivity commonly observed among structurally related RNA-binding protein families, as well as off-target effects arising from nonspecific distribution [287, 288]. Additional challenges include inefficient intracellular delivery of RNA-targeting modalities and, for central nervous system indications, restricted blood–brain barrier (BBB) penetration [289]. Diagnostic applications, such as CDO1/CELF4 methylation panels for endometrial cancer triage, further support the clinical relevance of CELF4-associated epigenetic alterations, although their translational potential remains at an early stage [290, 291]. Collectively, evidence from animal and cellular models demonstrates that targeting CELF1 activity, its regulatory pathways, or downstream splicing networks can restore aspects of RNA homeostasis and improve disease-associated phenotypes; however, these findings currently support CELF-directed interventions primarily as mechanistically validated but translationally constrained approaches rather than clinically established therapies [43, 285, 292].

RNA interference (RNAi) and CRISPRi-mediated gene silencing

Across CELF paralogs, loss-of-function interventions using RNA interference (RNAi) or CRISPR-based silencing have validated these RNA-binding proteins as regulators of oncogenic transcriptomes, cell-cycle control, autophagy, and immune signaling. CELF1 behaves as a pro-tumorigenic effector: its depletion consistently downregulates transcripts enriched for 3′UTR CELF1 motifs and suppresses proliferation, adhesion, angiogenesis, and survival pathways across oral, lung, and glioma models; furthermore, CRISPR-Cas9 deletion in B cells established direct post-transcriptional control of CD40 and downstream NF-κB signaling [37, 124, 129, 293]. By contrast, CELF2 predominantly exerts tumor-suppressive and therapy-sensitizing effects, with loss reducing autophagy-dependent radiosensitivity in colorectal models and restoration limiting proliferation, invasion, and chemoresistance in ovarian and hepatic systems [98, 142, 214]. Other paralogs act in a context-dependent manner: CELF4 supports tumor cell survival in selected neuroendocrine settings, and CELF6 enforces p53–p21 checkpoints such that its loss accelerates S-phase entry and proliferation in p53-competent backgrounds (Figure 6) [53, 122].

Fig. 6.

Gene-silencing approaches targeting CELF family members and their functional outcomes in cancer models. Loss-of-function studies using RNA interference (RNAi) or CRISPR–Cas9 knockout confirm that CELF paralogs exert distinct and context-dependent effects in tumor biology. CELF1 depletion reprograms hundreds of transcripts containing 3′UTR CELF motifs and suppresses proliferation, adhesion, angiogenesis, and survival pathways in oral squamous cell carcinoma (OSCC), lung cancer, and glioblastoma (GBM). CELF2 loss blocks radiation-induced autophagy and attenuates radiosensitivity in colorectal models, whereas restoration reduces proliferation and chemoresistance in other systems. CELF4 downregulation impairs neuroendocrine tumor growth, while CELF6 deficiency diminishes p21 expression, accelerates S-phase entry, and promotes proliferation in a p53/p21-dependent manner. These findings validate CELF paralogs as regulators of cell-cycle control, autophagy, and oncogenic transcriptomes, highlighting RNAi and CRISPR-based editing as mechanistically sound but technically constrained therapeutic avenues

Despite robust preclinical validation, the translational trajectory of RNAi- or CRISPR-based CELF targeting remains early and non-clinical. No CELF-family-specific RNAi or CRISPRi therapeutic has entered late-phase clinical evaluation, and CELF1 in particular remains difficult to pharmacologically or genetically target due to its essential physiological roles and broad transcriptomic footprint [43]. A major technical limitation of RNAi- and CRISPR-based strategies is the risk of off-target effects arising from sequence similarity and partial complementarity, which is particularly relevant for highly conserved gene families [294–296]. Such structural and functional homology increases the likelihood of paralogue cross-reactivity, thereby complicating isoform- or family member-specific targeting. For central nervous system applications, efficient delivery of RNAi and CRISPR-based systems remains a major challenge due to the restrictive nature of the BBB, which prevents systemic nucleic acids and genome-editing complexes from reaching brain parenchyma [296, 297]. As a result, current CNS delivery strategies rely on invasive administration routes or engineered viral and non-viral vectors, both of which face limitations in distribution efficiency, safety, and translational feasibility.

Early siRNA oncology trials demonstrated tolerability but limited tumor delivery, motivating the development of targeted nanoparticle carriers and refined editing platforms [298]; similarly, CRISPR interference approaches that avoid nuclease-induced DNA breaks are promising but require substantial advances in in vivo delivery systems to navigate the BBB and ensure cell-type specificity [298]. Complementary strategies include small-molecule inhibitors of CELF1–RNA binding and pharmacologic upstream modulation, as exemplified by GSK3β inhibition to lower pathological CELF1 levels in myotonic dystrophy models, offering alternate routes to therapeutic modulation [43, 298, 299]. Collectively, these data position gene-silencing approaches as mechanistically validated but technically constrained strategies for CELF-directed therapy; progress will hinge on improved delivery systems, exquisite molecular specificity, and context-aware target selection.

Antisense oligonucleotide (ASO) and gapmer-mediated mRNA targeting

Antisense oligonucleotides (ASOs) and RNase H–recruiting gapmers targeting the expanded DMPK 3′ UTR have demonstrated that selective reduction of toxic CUG repeats can reverse CELF1-associated molecular pathology in DM1. In vitro, MOE and cEt ASOs effectively reduce mutant DMPK RNA, disperse nuclear foci, and redistribute MBNL1, thereby normalizing downstream CELF1 activity [300]. In vivo, systemically delivered gapmers achieve substantial skeletal muscle knockdown and improve electrophysiological measures; however, limited clinical efficacy has been observed to date due to suboptimal biodistribution and muscle uptake [252, 301].

The clinical trajectory of these agents underscores the formidable "delivery-to-efficacy" gap. The failure of the first human gapmer, baliforsen (IONIS DMPKRx), established inadequate delivery as the primary translational barrier [302]. This challenge is amplified in multi-systemic CELF-associated pathologies where, beyond skeletal muscle, the requirement for BBB penetration and the risk of off-target effects on homologous motifs remain critical constraints [298, 302].

To address these limitations, next-generation delivery platforms are producing encouraging clinical signals. An antibody–oligonucleotide conjugate (AOC-1001) recently demonstrated improvements in myotonia and muscle strength, while a transferrin-receptor–targeted gapmer (DYNE-101) showed early functional gains without serious adverse events [303, 304]. Together, these data indicate that while ASO and gapmer modalities can effectively suppress pathogenic RNA, clinical success will depend on solving tissue-specific delivery constraints and enhancing molecular precision across both peripheral and central compartments.

Small-molecule inhibitors of RNA-binding protein–RNA interactions

Small-molecule inhibition of the CELF1–RNA interface is experimentally feasible and provides a direct route to modulate CELF1-driven pathology, although current evidence remains limited to preclinical systems. A virtual-screening hit, compound 27 (IC50 ≈ 23 μM), competitively disrupted CELF1–RNA binding, restored IFN-γ signaling in activated hepatic stellate cells, and suppressed fibrogenesis in vivo while sparing non-target hepatic populations. Subsequent structure–activity optimization produced compound 841, which displays improved affinity and selectivity for CELF1 relative to paralogs such as CELF2, establishing a chemically tractable scaffold for selective CELF1 blockade (Figure 7) [305, 306]. An orthogonal chemotype approach repurposed thiopurines: 6-mercaptopurine and 6-thioguanine covalently modify a cysteine residue in CELF1 to abrogate RNA binding in a cysteine-dependent manner; potency was further enhanced by a 1,2,4-triazole-3-thione analogue and by an optimized derivative (compound 9) that disrupts CELF1–RNA interactions in cells and rescues CELF1-linked myogenesis defects (Figure 7) [307]. However, several translational constraints remain. Achieving sufficient potency for intracellular RBP targets remains intrinsically challenging due to the flat and highly conserved RNA-binding interfaces (e.g., RRMs), which lack well-defined ligandable pockets [295]. Selectivity is further limited by the structural conservation of RNA-binding domains across RBP families, increasing the risk of transcriptome-wide off-target effects. In addition, covalent strategies raise concerns regarding specificity, as electrophilic warheads may react promiscuously with nucleophilic residues across the proteome, leading to potential toxicity [295, 308, 309]. Finally, pharmacokinetic liabilities including poor cell permeability, limited bioavailability, and suboptimal tissue distribution continue to hinder the clinical translation of RBP-targeting small molecules [310]. Although no CELF1-directed small molecule has yet entered clinical trials, these studies provide proof of concept that direct pharmacologic targeting of RBP–RNA contacts is possible; ongoing efforts focus on high-throughput screening, structure-guided optimization, and covalent inhibitor strategies to develop more selective and bioavailable CELF1 inhibitors [43, 307].

Fig. 7.

Small-molecule inhibition of the CELF1–RNA interface and pharmacologic modulation of CELF1 activity. Mechanistic models illustrate two distinct strategies for targeting CELF1: Competitive inhibition (left), where compounds such as 27 and its optimized derivative 841 bind the RNA-binding site of CELF1, disrupting CELF1–RNA complexes, restoring IFN-γ signaling in hepatic stellate cells, and suppressing fibrogenesis in vivo; and Covalent modification (right), where thiopurine analogs (e.g., 6-thioguanine) and triazole-thione derivatives (e.g., compound 9) covalently modify a cysteine residue in CELF1, leading to loss of RNA binding and rescue of CELF1-linked myogenesis defects. These approaches validate the feasibility of pharmacologically targeting RNA–protein interactions and highlight ongoing efforts to optimize selectivity, potency, and bioavailability of CELF1 inhibitors. The figure was generated with BioRender (https://biorender.com)

Epigenetic and transcriptional reactivation

Epigenetic repression of CELF genes is reversible and therefore therapeutically exploitable to restore tumor-suppressive RNA programs. In breast and pancreatic cancer cell lines, promoter CpG hypermethylation silences CELF2, and treatment with the DNA demethylating agent 5-aza-2′-deoxycytidine reinstates CELF2 expression [32]. Complementary histone-targeted approaches produce similar reactivation: the H3K9 demethylase HR restores CELF2 and restrains growth in HR-deficient breast cancer cells, and inhibition of the H3K9 methyltransferase G9a with UNC0642 likewise reactivates CELF2 and reduces proliferation, implicating H3K9 methylation as a tractable target [179]. Long noncoding RNA–mediated silencing exemplifies locus-specific control: in colorectal cancer, LINC01594 recruits DNA methyltransferase 1 (DNMT1) to the CELF6 CpG island, and 5-azacytidine restores CELF6 expression, corrects aberrant CD44 splicing, represses metastasis-associated CD44v4–v7 isoforms, and limits invasion and metastasis, supporting therapeutic targeting of the LINC01594–DNMT1–CELF6–CD44 axis [135].

Clinically, broad epigenetic drugs such as DNMT inhibitors (5-azacytidine, decitabine) are approved for hematologic malignancies and are under evaluation in solid tumors, where they can reactivate silenced tumor suppressors and ameliorate splicing defects, but their nonselective action and modest single-agent activity in solid tumors raise toxicity and efficacy concerns [311, 312]. Moreover, reactivation of CELF genes may not uniformly produce tumor-suppressive outcomes across contexts, given the context-dependent roles of CELF family members in different cancer types. To address these limitations, current strategies emphasize rational combination regimens, development of more selective epigenetic modulators including isoform-selective HDAC inhibitors, and targeted epigenetic editing approaches. CRISPR/dCas9-based epigenome editors enable locus-specific modulation without altering DNA sequence and have entered early clinical evaluation, representing a potential route toward precise reactivation of silenced CELF loci [313, 314].

Epigenetic and transcriptional reactivation strategies provide a mechanistically grounded but currently nonselective approach to restore CELF tumor-suppressor function, with translational success dependent on improvements in specificity, delivery, and context-aware patient stratification.

RNA decoys and aptamer-based sequestration

RNA decoy and and aptamer-based sequestration provide a complementary means to attenuate pathological CELF1 activity by competitively engaging its RNA-recognition surface, although current evidence remains indirect and largely preclinical. In non–small cell lung cancer, miR-574-5p acts as a noncanonical decoy that prevents CELF1 binding to the mPGES-1 3′UTR, increases mPGES-1 expression, elevates PGE2 production, and promotes tumor growth, effects reversible by pharmacologic inhibition of mPGES-1 [315]. Beyond microRNAs, GU-rich EDEN motifs serve as high-affinity binding elements for CELF1, and RNAs bearing these motifs can buffer excessive CELF1 activity [11]. SELEX studies have identified U/G-rich aptamers with multiple UGU motifs that effectively occupy CELF1 RNA recognition motifs, providing structural templates for rational inhibitor design [14].

No therapeutic RNA decoy or aptamer targeting CELF proteins has entered clinical testing [43]. Major translational barriers include inefficient delivery, limited intracellular accessibility, and rapid degradation in vivo. In addition, competitive sequestration strategies may require high intracellular concentrations to outcompete endogenous RNA substrates, raising feasibility concerns [306, 310, 316]. Chemical stabilization approaches, including cyclic or Spiegelmer backbones, improve nuclease resistance and systemic exposure, as demonstrated by cyclic STAT3 decoys that achieved in vivo activity after intravenous administration in mice [317]. Viral or nanoparticle delivery systems can enhance cellular uptake; for example, in a myotonic dystrophy type 1 model, adeno-associated viral delivery of a decoy protein targeting CUG-repeat RNA released sequestered MBNL, corrected splicing defects, and improved muscle pathology [318]. Indirect modulation strategies, such as GSK3β inhibition that reduces aberrant CELF1 activity and improves muscle function, further support the relevance of targeting CELF-dependent pathways [319].

RNA decoy and aptamer platforms remain mechanistically attractive but are currently limited by delivery constraints, stoichiometric requirements, and uncertain selectivity at the transcriptome level, with further development required to establish their translational viability in CELF-associated diseases [317–319].

The diagnostic value of CELFs

Across oncology and selected nonmalignant disorders, convergent human and preclinical evidence indicates that aberrant CELF family expression carries diagnostic and prognostic information [184, 314]. recurrent tumor pattern is CELF1 elevation coupled with CELF2 reduction. CELF1 is frequently upregulated and associates with advanced stage and poorer survival, including higher grade and reduced overall survival in gliomas, and with aggressive clinicopathologic features in non–small-cell lung, colorectal, and oral carcinomas [27, 43]. By contrast, CELF2 typically shows tumor-suppressive associations: reduced expression is common in malignancy, whereas higher CELF2 correlates with improved survival, stronger antitumor immune infiltration, and enhanced predictive power for therapy response in pan-cancer analyses, notably in triple-negative breast and lung squamous cell carcinomas, where CELF2 outperformed several standard immune biomarkers [34]. Additional paralogs show emerging, context-dependent signals: CELF4 downregulation appears in colorectal prognostic RBP networks, and early reports describe metastasis-associated expression shifts of CELF3, CELF5, and CELF6 in selected tumors [22, 27, 34]. These recurrent patterns support inclusion of CELF measurements in integrated diagnostic and prognostic panels.

Outside oncology, altered CELF expression also maps to clinically relevant states. Pathological CELF1 upregulation is a defining molecular feature of myotonic dystrophy type 1 affecting muscle and heart, and CELF1 and CELF2 are implicated in Duchenne and Becker muscular dystrophies [40, 43]. Experimental perturbation of CELF activity in mouse myocardium produces dilated cardiomyopathy, fibrosis, and heart failure, consistent with markedly increased CELF1 observed in failing human hearts and in pressure-overload hypertrophy [40, 43, 86], consistent with markedly increased CELF1 observed in failing human hearts, including pressure overload hypertrophy. Reduced CELF expression in mice accompanies myocardial hypertrophy and cardiomyopathy, whereas CELF1 elevation is observed during pathological remodeling [43, 86]. In the nervous system, haploinsufficiency of the neuron enriched CELF4 gene causes a human neurodevelopmental syndrome with intellectual disability and seizures, mirroring Celf4 deficient mouse phenotypes [320].

Taken together, convergent human and animal evidence indicates that abnormal CELF levels track disease burden and can inform clinical biomarker panels. Examples include CELF1 gain in dystrophic muscle or failing myocardium and CELF4 loss in epileptic encephalopathy. Assays that detect increased CELF1 or decreased CELF2 or CELF4 in patient tissue or blood may aid risk stratification and disease monitoring across cancer and selected neuromuscular and cardiac conditions, consistent with the established prognostic value of these markers in tumors (Fig. 8) [34, 43].

Fig. 8.

Diagnostic and prognostic associations of CELF family expression across cancer and non-cancer diseases. In cancer, elevated CELF1 often correlates with advanced stage and poor survival, while reduced CELF2 is linked to tumor progression and immune suppression; in contrast, higher CELF2 predicts improved survival and therapy response. Altered expression of CELF3–CELF6 is also associated with metastasis and outcome in specific contexts. In non-cancer disorders, CELF1 upregulation is observed in myotonic dystrophy and failing myocardium, whereas CELF4 deficiency causes a neurodevelopmental syndrome. Detection of a CELF1↑/CELF2↓/CELF4↓ expression pattern in patient tissues or blood may support risk stratification and disease monitoring, though clinical translation remains pending further validation

Despite these strong links, translation into clinical diagnostics remains limited [27]. Most oncology associations derive from retrospective analyses or preclinical data, and no prospective clinical trials have yet incorporated CELF measurements as diagnostic tools [43]. Key challenges include the need for invasive tissue sampling because CELF proteins are intracellular RNA binding proteins, a lack of standardized validated assays and cut offs, and limited clinical data on certain family members including CELF3 through CELF6 [27]. Key challenges include the need for invasive tissue sampling (since CELF proteins are intracellular RBPs), a lack of standardized, validated assays and cut-offs, and limited clinical data on certain family members (CELF3–6) [27]. Beyond oncology, aberrant CELF expression is evident in other diseases such as myotonic dystrophy and heart failure, but biomarker development is under developed and not yet applied in practice [319]. Potential solutions involve integrating CELF markers into multi analyte panels and validating them in prospective patient studies. For example, the immunotherapy response signal of CELF2 could be tested in clinical trial cohorts to confirm its added predictive value [34]. Overall, CELF proteins show significant diagnostic and prognostic promise across malignancies and selected non-malignant disorders, but rigorous clinical validation is needed to overcome current limitations and realize their utility.

Conclusion and future perspectives

Dysregulated CELF activity represents a convergent, targetable pathogenic axis across cancer, neurodevelopmental, and cardiometabolic disorders [2, 27, 43]. CELF proteins couple alternative splicing, alternative polyadenylation, mRNA stability/decay and translational control to upstream signaling, thereby reprogramming gene-expression programs that govern proliferation, apoptosis, metabolism, neuronal excitability and tissue remodeling [23, 39, 43, 321]. In disease contexts, CELF functions are highly context dependent: CELF1 commonly serves as a pro-tumorigenic hub that promotes epithelial–mesenchymal transition, proliferation and therapy resistance, whereas CELF2 more often acts as a tumor suppressor by constraining oncogenic signaling and regulating alternative splicing [27, 32, 39, 322]. Neuron-enriched paralogs (CELF3–CELF6) shape brain-specific RNA networks and are emerging contributors to both cancer biology and neurodevelopmental disorders [2, 119, 120, 323]. These contrasts underscore the need for cell-state-resolved atlases of CELF activity that integrate noncoding RNA circuitry and functional readouts of alternative splicing to refine biomarker development and inform therapeutic stratification [27, 32, 34, 43].

Despite substantial progress, key gaps remain. Hematological malignancies and the tumor microenvironment are undercharacterized, although initial data implicate CELF involvement in T cell signaling and transformation [89, 93]. Host–pathogen interactions likewise exploit CELF-related regulation, exemplified by HPV16 E6-mediated induction of CELF3, HCMV-associated splicing networks, and infection-driven STAT3 signaling that modulates CELF4 expression [16, 324]. CELF factors also intersect stress responses, such as DNA damage and endoplasmic reticulum stress, and engage additional regulatory layers that include m6A-dependent epitranscriptomic control of CELF2, ferroptosis-related gene networks, and phase-separated ribonucleoprotein condensates, each representing a distinct regulatory axis with translational potential [35, 92, 322, 325].

Therapeutic strategies targeting CELF proteins are conceptually attractive but remain largely preclinical and face nontrivial translational constraints. RNA-based modalities, including splice-switching oligonucleotides, antisense oligonucleotides, siRNA, programmable RNA editors, and decoy aptamers that target GU-rich motifs, offer theoretical paralog- and isoform-level specificity [326]. However, achieving selective modulation within a highly conserved RNA-recognition framework is inherently challenging, raising the risk of off-target effects across closely related CELF paralogs, particularly within the neuron-enriched CELF3–CELF6 subgroup. In addition, as intracellular RNA-binding proteins, CELFs are not directly accessible to conventional drug modalities, and efficient intracellular delivery remains a major barrier. For central nervous system applications, limited blood–brain barrier penetration further constrains therapeutic feasibility, whereas systemic approaches must contend with tissue-specific uptake, stability, and potential dose-limiting toxicity. These limitations indicate that CELF-directed therapies should currently be regarded as mechanistically promising but not yet clinically validated.

Rational combination regimens and context-specific interventions warrant systematic evaluation in advanced preclinical systems, including organoids, patient-derived xenografts and genetically engineered mouse models [37, 120, 142].

To accelerate translation, five priorities are paramount. (1) Build single-cell and spatially resolved CLIP-seq and long-read datasets mapping CELF RNA interactomes across human tissues to resolve cell-state-specific activities [305, 327]. (2) Standardize assays that quantify CELF proteins with isoform resolution and account for post-translational modifications [4, 174]. (3) Define mechanistic links among CELF-regulated programs, immune infiltration and therapeutic response to enable biomarker-guided combination therapies [34]. (4) Develop drug-like modulators that selectively target CELF–RNA interactions and condensate properties, accompanied by rigorous specificity profiling [326]. (5) Initiate prospective multicenter studies to evaluate CELF-based diagnostics and interventions in clinical settings.

From a translational perspective, we propose that the most realistic near-term clinical impact of CELF biology lies in biomarker development and patient stratification, whereas direct therapeutic targeting remains largely preclinical and constrained by specificity and delivery challenges. CELF1 and CELF2 currently represent the most actionable members based on available functional and clinical evidence, whereas CELF3–CELF6 require deeper isoform-resolved and tissue-specific validation. Future advances will depend on integrating mechanistic resolution with delivery innovation and specificity control, thereby enabling context-aware and clinically viable targeting strategies. Collectively, while CELF proteins provide a compelling conceptual framework for post-transcriptional intervention, their successful clinical translation will require rigorous validation, improved targeting precision, and disease-context prioritization [43, 83, 323].

Abbreviations

AREs

AU-rich elements

ASEs

Alternative splicing events

ASOs

Antisense oligonucleotides

BIN1

Bridging Integrator 1

CELF

CUG-BP and Elav-like family

CK2

Casein kinase 2

CRC

Colorectal cancer

COAD

Colon adenocarcinoma

CESC

Cervical squamous cell carcinoma

C/EBPα

CCAAT/enhancer-binding protein alpha

CNS

Central nervous system

CESC

Carcinoma and endocervical adenocarcinoma

CLCN1

Chloride Voltage-Gated Channel 1

DEGs

Differentially expressed genes

DNMT1

DNA methyltransferase 1

EGFR

Epidermal growth factor receptor

EMT

Epithelial–mesenchymal transition

FBP1

Fructose-1,6-bisphosphatase 1

GBM

Glioblastoma

GC

Gastric cancer

GTEx

Genotype-Tissue Expression

HCC

Hepatocellular carcinoma

HHIP-AS1

Hedgehog-interacting protein antisense RNA 1

HHIP

Hedgehog-interacting protein

H3K9me3

Histone H3 lysine-9 trimethylation

HPV

Papillomavirus

INSR

Insulin Receptor

IR

Insulin resistance

IDR

Lysine/arginine-rich intrinsically disordered region

LUAD

Lung adenocarcinoma

LUSC

Lung squamous cell carcinoma

MTOR

Mammalian target of rapamycin

NLS

Nuclear Localization Signal

NES

Nuclear export signal

NSCLC

Non–small-cell lung cancer

OSCC

Oral squamous cell carcinoma

PDAC

Pancreatic ductal adenocarcinoma

PKM

Pyruvate Kinase M

RRMs

RNA recognition motifs

siRNA

Small interfering RNA

SMN1

Survival Motor Neuron 1

TCGA

The Cancer Genome Atlas

TNM

Tumor–Node–Metastasis

TREM2

Triggering Receptor Expressed on Myeloid Cells 2

Authors’ contributions

Y.M. and C.M. contributed equally to this work. Y.M. and C.M. conceptualized the review framework, designed the literature search strategy, performed literature selection and data extraction, and drafted the manuscript. X.X., W.L., and M.G. provided overall supervision and conceptual guidance, and critically revised the manuscript for important intellectual content. A.Y. and Y.C. assisted with the comprehensive literature search and conducted the initial data synthesis. J.G., Q.W., and Z.W. contributed to reference verification and participated in refining the final draft. All authors read and approved the final version of the manuscript.

Funding

This project is financially supported by grants from the National Natural Science Foundation of China (No. 82300219 and 82503702), Natural Science Foundation of the First Affiliated Hospital of Soochow University (BXQN2024038) and Suzhou Youth Fund for Basic Research Projects (SSD2025072). The project ZR2023QH249, ZR2023QH365, ZR2024MH230 supported by Shandong Provincial Natural Science Foundation.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.