The importance of CELF control: molecular and biological roles of the CUG-BP, Elav-like family of RNA binding proteins

CELF protein subfamilies

CELF homologs are found in deuterostomes, protostomes, and plants, but not in bacteria, Archaea, or yeast.^{6, 10} Phylogenetic analysis of human CELF proteins divides them into two distinct subfamilies: CELF1-2, and CELF3-6.^6-8 Each subfamily is represented by at least one member in plants and animals, suggesting the ancestral CELF gene underwent a duplication soon after it arose.¹⁰ Most invertebrates, including primitive chordates, have only two CELF proteins. Drosophila, however, has three paralogs, Bruno (Bru), Bru-2, and Bru-3.⁶ Bru and Bru-2 likely result from a lineage-specific tandem duplication and belong to the CELF1-2 subfamily. While phylogenetic analysis suggests that Bru-3 belongs to the CELF3-6 subfamily⁶, it has been suggested that Bru-3 is functionally more similar to CELF1 than Bru or Bru-2.¹¹ The number of CELF proteins expanded in vertebrates. Mice have six CELF proteins as in humans. Chicken has five due to partial deletion of the CELF3 gene.¹⁰ Paralogs of celf1-5, but not celf6, have been identified in Xenopus tropicalis and laevis.¹² As a pseudotetraploid X. laevis has two of each gene, and cDNAs have been cloned for second paralogs of celf1, celf2, and celf3.¹² Only celf1-5 are listed in the NCBI database for zebrafish, although ESTs and predicted protein sequences suggest there could eight Celf proteins due to expansion of the CELF3-6 subfamily.¹⁰

The CELF1-2 subfamily is broadly expressed, while the CELF3-6 subfamily exhibits more restricted expression. CELF1 and CELF2 have overlapping yet distinct expression profiles, both being highest in heart, skeletal muscle, and brain. ^{3, 6-8} CELF3 and CELF5 are detected primarily in the nervous system, while CELF6 is found in nervous system, kidney, and testis.^{6-8, 13} There are conflicting reports of CELF4 expression. In one study, CELF4 transcripts were broadly detected in most human tissues⁸, but in several others CELF4 mRNA was detected only in the nervous system.^13-15 This discrepancy may result from using probes against different regions of the CELF4 transcript. CELF1 and CELF2 are also broadly expressed in the developing embryo with high expression in myogenic tissues, while CELF3-6 are primarily found in embryonic neural tissues.^{5, 10, 12, 16-18} Differential expression of the subfamilies is conserved. In C. elegans ETR-1, the CELF1-2 homolog, is expressed in developing muscle¹⁹ while UNC-75, the CELF3-6 homolog, is expressed exclusively in the nervous system.¹³

CELF protein structure and function

CELF proteins share a common domain structure with the Elav family (which includes the Hu antigen proteins), but are evolutionarily distinct.⁶ Each CELF protein contains three RNA binding domains of the RNA recognition motif (RRM) variety, two at the N terminus and one at the C terminus (Figure 1). A unique linker region called the “divergent domain” separates RRM2 and RRM3. The divergent domain was so named because sequence similarities in this domain divide CELF proteins into the two distinct subfamilies.⁷

Alternative splicing

CELF proteins have been shown to activate inclusion of some alternatively spliced regions, but silence others. They typically bind to intronic sequences in their pre-mRNA targets. While early studies relied on the use of in vitro assays and minigene reporters in cultured cells, later studies have expanded to include regulation of endogenous transcripts in vivo.

RNA editing

In mammals, the apolipoprotein B (APOB) mRNA undergoes C to U editing in the nucleus.⁵⁴ APOBEC-1, an RNA cytidine deaminase, and APOBEC-1 complementation factor (ACF) form the core of the C to U editing holoenzyme. Function of this holoenzyme is regulated in part by CELF2.⁴³ CELF2 was identified as an apobec-1 interacting protein in a yeast two-hybrid screen.⁴² Co-immunoprecipitation and co-localization studies in liver cells indicated that CELF2 forms a complex with APOBEC-1, ACF, and the APOB RNA.⁴² Addition of CELF2 to a cell-free system inhibits C to U editing, while antisense inhibition of CELF2 in hepatoma cells promoted editing.⁴² Likewise, knockdown of CELF2 in an intestinal cell line stimulated APOB editing.⁴³ CELF2 levels inversely correlate with levels of APOB editing in mouse tissues and cultured human cells.⁴³ Other transcripts undergo nuclear C to U editing in mammalian cells⁵⁵, but a role for CELF proteins in these events has not yet been reported.

Deadenylation

In Xenopus, celf1 was first identified by its ability to bind a U(A/G)-containing embryo deadenylation element (EDEN) and promote rapid EDEN-mediated deadenylation and silencing of maternal transcripts.²⁵ When RNAs were co-immunoprecipitated with celf1 complexes from egg extracts and hybridized on microarrays over 150 putative targets were identified.⁵⁶ More than 90% of these contained EDEN motifs in the 3′ UTR, and included transcripts encoding proteins involved in cell cycle control and oocyte maturation. Human CELF1 also binds EDEN motifs, and can functionally replace the deadenylation activity of Xenopus celf1 in immunodepleted egg extracts.^{57, 58}

Deadenylation is not only important for silencing of maternal transcripts, but is also a rate-limiting step in the turnover of most eukaryotic mRNAs. CELF1 was shown to bind to C-FOS and TNFα transcripts and promote deadenylation in HeLa cell extracts.⁵⁹ CELF1 also interacts with poly(A)-specific ribonuclease (PARN), suggesting CELF1 may facilitate deadenylation by recruiting or stabilizing the deadenylase on its substrates.⁵⁹

While it is clear that CELF1 can promote mRNA deadenylation, it might also induce polyadenylation. Some putative targets of celf1 (e.g. frl1) are polyadenylated in the zygote.⁶⁰ PARN has been shown to be part of the cytoplasmic polyadenylation element binding protein (cpeb) complex in Xenopus oocytes, where it deadenylates transcripts as the poly(A) tail is added.⁶⁰ Dissociation of PARN occurs following phosphorylation of cpeb, allowing the poly(A) tail to lengthen. Recruitment of transcripts into this complex by celf1 could theoretically promote deadenylation or polyadenylation, depending on whether PARN is present. It remains to be determined whether celf1 stimulates the polyadenylation of any targets.

mRNA stability

CELF1 binds to a conserved GU-rich element (GRE) that is enriched in the 3′ UTRs of short-lived transcripts in human primary T cells, and promotes GRE-dependent mRNA decay.⁶¹ Interestingly, a CELF2 antibody failed to supershift GRE-containing complexes, suggesting this activity is specific to CELF1. Transcripts that co-purify with CELF1 in HeLa and C2C12 cytoplasmic extracts were significantly enriched for GRE and GU-repeats in the 3′ UTRs.^{62, 63} Many of these encode proteins involved in cell cycle, growth, and apoptosis. CELF1-associated transcripts in myoblasts have shorter half lives, and are stabilized by siRNA-mediated knockdown of CELF1.^{62, 64} As mRNA decay often follows poly(A) tail shortening, it remains to be determined whether CELF1-mediated degradation is a consequence of its deadenylation activity.

While CELF1 promotes mRNA decay, CELF2 binds to A/U-rich sequences in the 3′ UTR of cycloxygenase-2 (COX2) transcripts and stabilizes them, while simultaneously inhibiting their translation.⁶⁵ COX-2 is an enzyme that synthesizes prostaglandins, which provide a protective effect against ionizing radiation. CELF2 is induced following irradiation, whereas knockdown of CELF2 up-regulates COX-2 expression and decreases radiation-induced apoptosis.^{65, 66} Prostaglandins inhibit CELF2 expression in intestinal cells, indicating COX-2 may regulate CELF2 in a negative feedback loop.⁶⁷ CELF2 similarly stabilizes and inhibits translation of MCL1, an anti-apoptotic factor in the BCL2 family, via binding to its 3′ UTR.⁶⁸

Translation

CELF proteins have been shown to regulate translation through multiple mechanisms (Figure 2). In Drosophila, Bru was first described as a repressor of oskar (osk) translation via binding to its 3′ UTR.⁶⁹ Bru interacts with Cup, which associates with the 5′-cap binding initiation factor eIF4E, suggesting translational repression occurs via a 5′/3′ interaction mediated by an eIF4E-Cup-Bru complex.⁷⁰ Bru also binds to gurken and Cyclin A 3′ UTRs and inhibits translation.^{71, 72} As described above, CELF2 has likewise been shown to inhibit the translation of COX2 and MCL1 mRNAs.^{65, 68} The mechanism is not well understood, but CELF2 has been shown to compete with HuR, an activator of translation, for binding to A/U-rich elements in the COX2 3′ UTR.⁷³ Under conditions of environmental stress, CELF1 is routed to stress granules in HeLa cells, suggesting it may also play a role in translational silencing.³⁶

Like Bru, Xenopus celf3 was shown to bind to the 3′ UTR of cyclin A transcripts, but unlike Bru, celf3 binding stimulated translation.³⁴ Mammalian CELF1 induces translation of p21, a cell cycle inhibitor, by binding to a G/C-rich sequence in its 5′ UTR.^{74, 75} CELF1 also binds to sequences in the 5′ region of C/EBPβ, where it promotes the use of a downstream AUG translation initiation codon. This gives rise to a truncated form of the C/EBPβ protein, liver inhibitory protein (LIP), in a variety of contexts.^{23, 39, 76-78} Phosphorylation of CELF1 activates an interaction with subunits of the initiation factor eIF2, leading to recruitment of ribosomes to the downstream AUG start site.^{39, 79} A CELF1-eIF2 complex also binds to the 5′ UTR of HDAC1 and increases HDAC1 protein levels in aging liver.⁸⁰

Binding of CELF1 to the 3′ UTR of SHMT, a folate-dependent enzyme, has been implicated in internal ribosome entry site (IRES)-mediated translation.^{81, 82} A mechanism whereby CELF1-hnRNP H complexes promote circularization of the transcript by mediating 5′/3′ interactions has been proposed.⁸¹ Whether CELF1 recruits the translation machinery to the SHMT 5′ UTR via its interaction with eIF2 or another initiation factor remains to be determined.

Integration of cytoplasmic functions

Polyadenylation status, transcript stability, and translation are intimately related. For example, deadenylated transcripts are often degraded or translationally silent. How cytoplasmic CELF functions are mechanistically coupled is unclear. In addition, opposing effects may be observed in different contexts. Tethering CELF1 to the 3′ UTR of a reporter reduced RNA levels while maintaining protein production.⁸³ This suggests a decrease in mRNA stability and increase in translation, the opposite of the effects of CELF2 on COX2 RNA. As with alternative splicing, the construction of RNA binding maps combining 5′ and 3′ UTR binding sites with regulatory information may help elucidate the mechanisms by which the functions of CELF proteins are integrated in the cytoplasm.

Mutations in invertebrate CELF proteins

Three mutant alleles in the arrest (aret) locus have been shown to disrupt the coding region of the Drosophila Bru protein.⁸⁴ The aret mutants have gametogenesis defects in both males and females.^{85, 86} Bru translationally represses osk, which is important for germ cell formation and posterior body patterning.⁶⁹ Bru also represses gurken, which is essential for axis formation, and Cyclin A, which drives the release of meiotic arrest and mitotic reentry.^{71, 72} Celf1 is maternally expressed in zygotes and egg extracts in zebrafish and frogs^{12, 87}, and a cluster of mRNAs involved in the meiotic process of oocyte maturation were identified as potential targets of celf1 deadenylation.⁵⁶ These data suggest that CELF1 may play a conserved role in the switch from meiosis to mitosis and oocyte to zygote.

In C. elegans, inactivation of ETR-1 by RNAi caused elongation defects and embryonic lethality.¹⁹ Worms that survived to hatching had paralysis and defects in muscle attachment. Mutations in C. elegans unc-75 caused defects in motor neuron axon sprouting and synaptic function, but not neuron differentiation.¹³ Mosaic analysis indicated that UNC-75 affects cholinergic and GABAergic synaptic transmission between motor neurons and target muscles. Unc-75 mutants also had behavioral defects in egg laying and feeding that are characteristic of cholinergic transmission defects.

CELF proteins in the frog embryo

While initially uniform in early Xenopus embryos, celf1 becomes enriched in dorsal mesoderm at neurula and tailbud stages, particularly presomitic mesoderm.⁸⁸ Somites are transient structures that give rise to multiple tissues including body muscles. Inhibiting celf1 function by injecting antisense oligonucleotides or neutralizing antibodies leads to severe defects in somite segmentation, but not differentiation.⁸⁸ This phenotype has been attributed to de-repression of Su(H), a component of the notch signaling pathway.⁸⁹ CELF1 is also expressed in the somites in chick and mouse embryos, perhaps indicating a conserved role in segmentation.^{10, 18}

Although initially isolated as a neural-enriched gene⁵, knockdown of Xenopus celf3 in embryos inhibited proliferation and subsequent differentiation in the endoderm.³⁴ Conversely, over-expression of celf3 stimulated proliferation.

Knockout mice

Similar to Bru mutants in flies, Celf1 and Celf3 knockout mice both indicate a role in gametogenesis. Mice lacking CELF3 are fertile with normal testicular morphology.⁹⁰ A significant reduction in the total sperm count was observed, however, suggesting that CELF3 deficiency affects the progression of spermatogenesis. Since fertility was not adversely affected, CELF3 is not essential for spermatogenesis, but compensation by another CELF protein is possible. In contrast to Celf3-null mice, disruption of Celf1 resulted in severely impaired fertility in both males and females.¹⁸ Homozygous males are hypofertile, with a blockage in spermiogenesis prior to spermatid elongation, increased apoptosis, and down-regulation of germ cell markers. Histological analysis of the ovaries revealed no obvious changes in the oocytes or follicles, so the basis for reduced fertility in females remains unknown. Celf1 knockout mice also exhibit a high rate of perinatal mortality and growth retardation, perhaps due to impaired placental function. Their failure to thrive and fertility deficits may make future characterization of the effects of loss of CELF1 on other organ systems such as heart, muscle, and brain more difficult in this model.

Consistent with neurotransmission defects in unc-75 mutants, CELF4 deficiency is associated with a seizure disorder in the Frequent flier (Ff) mouse model of epilepsy, so named for its phenotype of convulsions followed by a running-bouncing phase.¹⁵ The Ff mutation arises from a transgenic insertion in the Celf4 gene, which results in loss of function. A significantly reduced seizure threshold was observed in heterozygotes with progression of recurrent limbic and tonic clonic seizures. Homozygotes did not usually survive in the original C57BL/6J background, but those that survived in F₂ hybrid backgrounds displayed spike-wave discharges, the hallmark of absence epilepsy. This suggests that although all six CELF proteins are expressed in brain, CELF4 plays a non-redundant role in neural function.

Transgenic dominant negative CELF protein models

Two mouse models have been generated that use tissue-specific expression of a nuclear dominant negative CELF protein, NLSCELFΔ, to investigate CELF-mediated alternative splicing programs in vivo. The dominant negative approach offers several advantages. First, all CELF proteins present in a tissue (e.g. CELF1 and CELF2 in striated muscle) are simultaneously inhibited. NLSCELFΔ has been shown to inhibit the splicing activity of all six family members.³² Second, because NLSCELFΔ is restricted to the nucleus, it presumably blocks only the nuclear function of CELF proteins, leaving cytoplasmic functions intact. Third, tissue-specific repression is easily achieved through use of tissue-specific promoters. Disadvantages are that individual functions of family members cannot be separated, and repression may not be complete.

MHC-CELFΔ mice express NLSCELFΔ under a cardiac muscle-specific promoter.⁴⁸ MHC-CELFΔ mice develop defects in CELF-mediated alternative splicing, dilated cardiomyopathy, and severe cardiac dysfunction. These changes were attributed to loss of CELF activity because over-expression of human CELF1 in bitransgenic mice partially rescued both splicing defects and overt heart disease. Myo-CELFΔ mice express NLSCELFΔ under a skeletal muscle-specific promoter.⁹¹ Unlike the MHC-CELFΔ mice, these mice display small changes in alternative splicing and have a mild phenotype. Myo-CELFΔ mice exhibit reduced muscle interstitia, increased variability in fiber size, and a slight increase in slow twitch fibers. These modest changes may indicate that CELF proteins are more determinative for alternative splicing in heart than in skeletal muscle. Alternatively, NLSCELFΔ may more effectively inhibit CELF activity in MHC-CELFΔ hearts than in Myo-CELFΔ muscles.

CELF proteins in cardiac pathogenesis

The human CELF2 gene is located at chromosome 10p13-14. CELF2 has been proposed as a candidate gene for two genetic disorders affecting the heart that arise from mutations in this vicinity. Arrythmogenic right ventricular dysplasia (ARVD) is characterized by progressive myocardial cell loss and fibrosis.⁹² A familial form of ARVD, designated ARVD6, maps to the locus 10p12-14, which contains CELF2. ARVD6 has high penetrance, early onset, and a high incidence of sudden death. Although no mutation was found in ARVD6 family members that would be predicted to change the CELF2 protein sequence, variations were detected that cosegregate with the disease.⁹² Partial monosomy 10p is a rare chromosomal aberration associated with symptoms of the DiGeorge syndrome spectrum. Haploinsufficiency within the proximal region of 10p, DGCR2, has been linked to heart and thymus defects. CELF2 was the only known gene found within a critical 300 kb region of DGCR2 identified by deletion mapping of partial monosomy 10p patients.⁹³ Although loss of CELF2 was not proven to cause heart defects in either ARVD6 or partial monosomy 10p, it is a promising candidate for future clinical and functional studies.

CELF proteins in muscle diseases

The best-studied example of involvement of a CELF protein in disease is that of CELF1 in myotonic dystrophy (DM). DM is a form of muscular dystrophy that affects several systems, including skeletal muscle and heart. DM results from expression of RNAs containing expanded CUG or CCUG repeats that form double-stranded hairpins, accumulate in nuclear foci, and disrupt the processing of other RNAs.⁹⁴ CELF1 was first identified as a CUG-repeat binding protein¹, leading to the proposal that the mutant RNAs sequester CELF1. Later studies indicated that another splicing factor, muscleblind-like 1 (MBNL1), is sequestered.^{95, 96} CELF1, however, does not interact with long CUG repeats or co-localize with mutant RNAs in cells.^{97, 98} Rather, CELF1 is up-regulated in DM heart and skeletal muscle.^{99, 100} PKC-mediated phosphorylation of CELF1 stabilizes the protein without a change in CELF1 transcript levels.^{38, 101} Increased CELF1 activity disrupts alternative splicing and translation events that are directly linked to disease symptoms.^{75, 99, 102} DM pathogenesis is now believed to result from a combination of effects including CELF1 over-expression and MBNL1 sequestration.

Transgenic mouse models that over-express CELF1 in striated muscle tissues support its contribution to DM pathogenesis. CELF1 is constitutively over-expressed in skeletal muscle in two models.^{103, 104} Both exhibited defects in muscle histology and physiology that recapitulate features of DM. Timchenko et al. showed that CELF1 over-expression enhanced translation of p21 and MEF2A¹⁰⁴, while Ho et al. demonstrated changes in alternative splicing of muscle transcripts.¹⁰³ Because embryonic muscle development is impaired in these mice, the effects of CELF1 over-expression in adult muscle are difficult to assess. Bitransgenic mice that inducibly over-express CELF1 demonstrated that elevated CELF1 in adult skeletal muscle is sufficient to induce impaired muscle performance, muscle wasting, and alternative splicing changes seen in DM.¹⁰⁵ Inducible over-expression of CELF1 in heart likewise exhibits echocardiographic, electrocardiographic, and histological abnormalities seen in DM patients.¹⁰⁶

While the dysregulation of CELF1 in DM is thought to be a primary pathogenic event, a recent study found that changes in alternative splicing and increases in CELF1 and CELF2 are also seen in other types of muscular dystrophy and muscle injury.¹⁰⁷ Elevated levels of CELF2 have also been reported in Duchenne and Becker muscular dystrophies.^{108, 109} These changes may reflect the reiteration of early developmental programs during muscle regeneration.

CELF proteins in nervous system disorders

Although DM is primarily considered a striated muscle disease, patients also have symptoms affecting the nervous system, including cognitive impairment.⁹⁴ Several exons of tau, which is involved in forming protein aggregates in Alzheimer’s disease and other dementias, are regulated by CELF proteins and are dysregulated in DM.^110-114 Interestingly, skipping of tau exon 10, which encodes a microtubule binding domain, is induced by over-expression of CELF2, but not over-expression of CELF1 or MBNL1 knockdown.¹¹⁵ CELF1 and CELF2 are both up-regulated in DM brains, and increased exon 10 skipping correlates with elevated CELF protein levels.¹¹⁵ Polymorphisms in CELF2 were also identified in a genome-wide association study of late-onset Alzheimer’s disease that are significantly associated with high risk alleles of APOE.¹¹⁶

Expression of an expanded CUG repeat-containing RNA in the neurodegenerative disorder spinocerebellar ataxia type 8 (SCA8) may follow a similar mechanism to DM, as MBNL1 is sequestered and CELF/MBNL alternative splicing is disrupted in neurons.⁵⁰ Fragile X syndrome, a form of hereditary mental retardation, is caused by a CGG repeat expansion in the 5′ UTR of the FMR1 gene that leads to silencing and loss of the FMRP protein.¹¹⁷ In a transgenic fly that expresses 90 CGG repeats over-expression of CELF1 suppresses the neurodegenerative phenotype through interaction with the CGG-binding protein hnRNP A2/B1.¹¹⁸

The mechanistic similarities between CELF-mediated pathogenesis in the nervous system and striated muscle suggest that although CELF family members are differentially expressed throughout the body, they fulfill similar functional niches in different cell types. As additional CELF targets are identified, it will be interesting to determine whether common pathways are involved in pathogenesis in different cell types.

Conclusion

CELF proteins have been shown to regulate RNA processing at several important steps in the nucleus and cytoplasm, but little is known about whether or how these disparate activities are coordinated. Dysregulation of CELF1-mediated alternative splicing and translation have been shown to contribute to pathogenesis in DM.^{103, 104} It is easy to imagine that shuttling of CELF1 between subcellular compartments serves to coordinate these processes in healthy tissues.

Many tissues express multiple CELF proteins, but the extent of functional redundancy and individual contributions of specific family members have not been determined in most cases. Many studies have focused on only a single CELF protein, and CELF1 and CELF2 have received greater attention than CELF3-6. The distinct phenotypes of CELF knockout mice despite overlapping expression patterns suggest that different family members do indeed play different roles in vivo. The elucidation and comparison of targets regulated by each family member will help to delineate specific CELF regulons.