Introduction
Fragile X syndrome is one of the most common forms of inherited mental retardation with an estimated incidence of 1 in 4,000 men and 1 in 8,000 women1. The gene responsible for fragile X syndrome, FMR1 (for fragile X mental retardation 1), is situated on the X chromosome and was cloned in 1991 (refs 2–5). It was one of the first human genes in which expansions of triplet-nucleotide repeats were linked to disease. Although many other disease-associated trinucleotide repeats are found in the coding region of genes, FMR1 belongs to a small group where such repeats are located in a noncoding portion of the gene; in this instance, the 5' untranslated region6.
Full expansion of the FMR1 CGG repeat is the causative mutation in 99% of individuals with fragile X syndrome. In these cases, methylation of the CGG repeat and upstream CpG islands silences FMR1 expression, and results in loss of the encoded protein, FMRP. The length of the FMR1 CGG repeat is highly polymorphic in the general population, with a range of 6 to 60 and 30 repeats being most common. Close examination of fragile X pedigrees reveals two other allelic classes: premutations, with about 60–200 repeats; and full mutations, with >200 (and often >500) repeats that are completely penetrant in men (that is, every man who has the genetic mutation displays the symptoms of fragile X syndrome) and 50% penetrant in women. Unlike the FMR1 alleles in the general population, premutation alleles are unstable when transmitted to offspring1. This premutation instability gives rise to repeat expansions that usually yield slightly larger alleles in the premutation range, but in women it also can lead to massive expansion to a full mutation. In men, spermatogenesis seems to be unable to tolerate such large expansions. The likelihood of expansion to a full mutation is positively correlated with the length of the premutation in the transmitting female1, 7.
FMR1 attracts attention for a variety of reasons. At the DNA level, questions remain regarding how the CGG repeat expansion occurs and how cells distinguish full mutations from normal and premutation alleles. At the RNA level, it was recently recognized that carriers of the fragile X premutation can develop a tremor/ataxia syndrome that is distinct from the full mutation phenotype. Moreover, RNA encoded by CGG repeats (rCGG) was recently found in Drosophila to cause the first example of RNA-mediated neurodegeneration8 (see ref. 9 for a review). Finally, at the protein level, it is still not clear how the loss of a single protein, FMRP, leads to mental retardation and behavioural problems. FMRP is clearly implicated in many interesting biological phenomena, from the control of local protein synthesis in neuronal processes to microRNA-mediated translational suppression, and perhaps also chromatin remodelling.
Within the past few years, our understanding of the role of FMRP in learning and memory has been accelerated through use of multi-disciplinary approaches and the development of model systems. In this review, we will focus on the latest progress towards understanding the cellular functions of FMRP in neural development. In addition, we will discuss our current understanding of the molecular mechanism of methylation of expanded CGG repeats, for which we propose an RNAi-based model.
Neuronal functions of FMRP
FMRP is widely, but not necessarily ubiquitously, expressed in fetal and adult tissues, with the most abundant expression in brain and testes10. FMRP, along with its autosomal paralogues the fragile X-related proteins FXR1P and FXR2P, compose a small family of RNA-binding proteins (the fragile X-related gene family)11, 12. These proteins share more than 60% amino-acid identity and contain two types of RNA-binding motif: two ribonucleoprotein K homology domains (KH domains), and a cluster of arginine and glycine residues (RGG box). The fragile X-related gene family is well conserved throughout evolution; orthologues of FMR1, FXR1 and FXR2 exist in mouse, chicken and Xenopus. Drosophila, however, contains only a single fragile X-related gene, dfmr1 (ref. 13).
FMRP itself is involved in translational control. It associates with polyribosomes in an RNA-dependent manner through messenger ribonucleoprotein (mRNP) particles14, 15, and it can suppress translation both in vitro and in vivo16, 17. FMRP shuttles between the nucleus and cytoplasm using a nuclear localization signal (NLS) and a nuclear export signal (NES)18.
A current working model is that FMRP is involved in synaptic plasticity through regulation of mRNA transport and local protein synthesis at synapses (Fig. 1). This model is supported by the observation of abnormal dendritic spines in the brains of both humans with fragile X syndrome and Fmr1-knockout mice19, 20, 21. Moreover, FMRP is found associated with polyribosomes at the base of or within dendritic spines of wild-type neurons22. Recent studies using biochemical, genetic, genomic, electrophysiological and high-resolution imaging approaches in different model systems have advanced our understanding of FMRP cellular functions in neural development. Here we will focus on three major areas: the RNA cargoes associated with FMRP, FMRP-mediated translational control and neuronal dysfunction in the absence of FMRP.
Figure 1: Model of fragile X mental retardation protein (FMRP) function in the neuron.
FMRP (green hexagon) dimerizes in the cytoplasm and enters the nucleus of neurons by its nuclear localization signal. FMRP forms a messenger ribonucleoprotein (mRNP) complex, by interacting with specific RNA transcripts (hairpin structure) and proteins. The FMRP–mRNP complex is transported out of the nucleus by the FMRP nuclear export signal. Once in the cytoplasm, the FMRP–mRNP complex can either associate directly with ribosomes (purple ovals) in the cell body, or interact with members of the RNA-induced silencing complex (RISC; red star) before associating with ribosomes. Both FMRP–mRNP complexes (with or without members of RISC) regulate protein synthesis (string of blue circles) in the cell body of the neuron. Alternatively, both complexes can be transported into dendrites to regulate local protein synthesis of specific RNAs in response to synaptic stimulation signals such as activation of the metabotropic glutamate receptor (mGluR; orange oval).
Full size image (39 KB)RNA cargoes associated with FMRP
As FMRP is involved in mRNA transport and translational regulation, the identities of the mRNA and protein components of FMRP-containing mRNP complexes have been extensively investigated. Several groups have used different approaches to identify the mRNAs bound specifically by FMRP, and the structure required for FMRP–RNA interaction in mammals23, 24, 25, 26. To identify the in vivo mRNA targets for FMRP, our group used microarrays to perform a genome-wide search23. In parallel studies, we compared the mRNA profiles in polyribosomal fractions collected from normal cells and cells from individuals with fragile X syndrome. These studies identified 14 homologous mouse/human transcripts, that exhibited an abnormal polyribosome profile in the absence of FMRP23. Because the distribution of FXR1P and FXR2P in polysomes is not changed in fragile X cells, the altered profiles were specific to the absence of FMRP.
Another group developed a novel approach called antibody-positioned RNA amplification (APRA) to identify the direct mRNA substrates of FMRP in vivo26. These experiments used hippocampal neurons to identify a group of mRNAs participating in synaptic function and axon guidance as substrates of FMRP. Disruption of these processes through loss of FMRP is consistent with the abnormal dendritic spines found in both humans with fragile X and the Fmr1-knockout mouse23, 26.
Studies investigating the molecular requirements for FMRP–RNA interactions revealed the specific RNA sequences and protein domains responsible. Two groups found independently that an intramolecular G-quartet structure is involved in FMRP–RNA interactions24, 27. G quartets are hydrogen-bonded stem-loop structures formed from four guanosine residues in a square-planar array that are stabilized preferentially by K+ and disrupted by the presence of Li+. Recognition of RNA by FMRP requires not only the G-quartet structure, but also specific sequences in the loop surrounding the G quartet. Mutations of surrounding nucleotides that are not involved in either the G quartet or the stem are able to reduce FMRP binding markedly.
At the protein level, deletion analysis of FMRP demonstrated that the RGG box, but not the KH domain, is responsible for binding to mRNAs that contain the G quartet24. This specificity is rather surprising given that the RGG box has long been considered a non-specific RNA-binding domain28. Together, these data suggest that the RGG box of FMRP functions together with the target RNA structure and nucleotide sequence to facilitate efficient FMRP–RNA binding. However, the role of the G quartet in FMRP-mediated translation control remains elusive.
Drosophila is becoming an important model system to study the biological functions of FMRP, and several mRNA targets of the Dfmr1 protein have been identified, including futsch, rac1 and pickpocket1 (ppk1)29, 30, 31. These mRNAs are involved in either synaptic function or dendritic development. A genome-wide biochemical screen for RNA ligands to dfmr1 protein, similar to the one for the mammalian orthologue, would be useful.
The protein components of the FMRP-containing mRNP complex include FXR1P, FXR2P, nucleolin, YB1/p50, Pur 
and mStaufen32, 33, 34. Given the fragile X syndrome phenotype, it is interesting that Pur and mStaufen function in the transport of neuronal granules composed of RNA and associated proteins35, 36. A fusion of FMRP and green fluorescent protein (FMRP–GFP) allowed these granules to be followed, and they were found to move into the neurites of living cells. Motile FMRP–GFP granules are microtubule-dependent and exhibit two types of movement: oscillatory (bidirectional) and unidirectional anterograde37. Recently, a newly developed reporter of protein–RNA interactions allowed the FMRP–mRNA interaction to be followed. The system showed that FMRP and an RNA transport factor (IMP1) interact with common mRNAs in granular structures, and IMP1 is recruited to the mRNA by FMRP38. Moreover, these FMRP-associated granular structures are localized throughout dendrites and within spines of cultured hippocampal neurons in a manner that is dependent on neuronal activity and that can be modulated by KCl depolarization and metabotropic glutamate receptor (mGluR) activation39. Together, these observations suggest that FMRP-associated transportation of RNA cargoes is a dynamic and highly regulated process.
Despite the identification of several FMRP-associated proteins, the molecular motor responsible for the trafficking of FMRP and its associated mRNAs remained enigmatic until recent studies indicated that kinesin might be involved40. The kinesin superfamily of proteins (KIFs) are molecular motors that transport cargoes along microtubules. More than 40 KIFs have been identified in mammals, including kinesin. Recently, it was found that FMRP associates with a conventional kinesin (KIF5)40. Along with other proteins that interact with FMRP (including Pur and mStaufen), FMRP and its associated mRNAs co-localize to the kinesin-associated granules in dendrites. The kinesin-associated granules move bidirectionally, and the distally directed movement is enhanced by the overexpression of KIF5 and reduced by its functional blockage40. However, the question of whether and how the interaction between kinesin and the FMRP-containing mRNP complex is regulated remains unanswered.
MicroRNA-mediated translational regulation by FMRP
Early studies in non-neuronal cell lines demonstrated that FMRP associates with polyribosomes in an RNA-dependent manner15, 18, 41, 42. More recent studies show definitively that FMRP associates with polyribosomes at synapses in brain43, 44. Together with the fact that the FMRP–polyribosome association is sensitive to puromycin, a drug that targets actively translating ribosomes44, these and other studies suggest that FMRP is involved in protein translation.
FMRP functions as a translational repressor of reporter constructs both in vitro and in transfected cells. In Drosophila, the Dfmr1 protein regulates the translation of at least two of its mRNA targets, futsch and rac1, and functions as a translational suppressor29, 31. In mouse, the identification of FMRP mRNA targets provided a list of candidate genes to examine at the protein level in Fmr1-knockout mice. Recently, the translation of MAP1B, one of the mRNA targets of FMRP and the orthologue of futsch, was found to be increased in Fmr1-knockout compared with wild-type mice45. Interestingly, FMRP is also required for mGluR-dependent translation of presynaptic density protein 95 (PSD-95; ref. 46). Along with the finding of activity-dependent FMRP localization, these data suggest that FMRP might be involved in mRNA transport and local protein synthesis in a neuronal activity-dependent manner.
Thus far, all the data indicate that FMRP functions as a translational suppressor. Because microarray studies show that mRNAs associated with FMRP in vivo exhibit different polyribosome profiles in the absence of FMRP, it will be interesting to study these mRNAs to understand how the loss of FMRP might affect their translation23. One of the mechanisms by which this regulation could occur is through phosphorylation of FMRP, which might modulate the translation state of FMRP-associated polyribosomes47. Both mammalian FMRP and Drosophila Dfmr1 can be phosphorylated in vivo47, 48 at a phosphorylation site that is conserved throughout evolution (Ser 406 in Dfmr1 and Ser 500 in human FMRP). In non-neuronal cells, unphosphorylated FMRP is associated with actively translating polyribosomes, whereas a fraction of phosphorylated FMRP is associated with apparently stalled polyribosomes47. Thus, removal of this phosphate by an activated phosphatase might be the signal for FMRP to release the translational suppression and allow synthesis of a locally required protein.
The accumulation of work from several groups suggests that the RNA interference (RNAi) pathway is the major molecular mechanism by which FMRP regulates translation. The crucial observation came from biochemical studies in Drosophila showing that Dfmr1 associates with Argonaute 2 (AGO2) and the RNA-induced silencing complex (RISC), which mediate RNAi49, 50. RNAi is a conserved gene-silencing response to double-stranded RNA (dsRNA)51, initiated when the dsRNA triggers are processed into small interfering RNAs (siRNAs). This is catalysed by a group of RNase III enzymes known as the Dicer family. The siRNAs are incorporated into the effector complex, RISC, which uses siRNA as a guide to select complementary mRNA substrates51. Most components of RISC can also be used by endogenous microRNAs (miRNAs)52, a new class of noncoding RNAs that are believed to control translation of specific target mRNAs by base-pairing with complementary sequences in the mRNA 3' untranslated region52. Mature miRNAs are single strands of 20–25 nucleotides that are processed from roughly 70-nucleotide (or longer) stem-loop precursors by Dicer. The downstream processing and functions of miRNAs and siRNAs are facilitated by members of the PIWI/PAZ-domain protein (also known as Argonaute) family52.
Although recent data in Drosophila suggest Argonaute 1 (AGO1) is mainly involved in the endogenous miRNA pathway and AGO2 is required for siRNA-mediated gene silencing, the loss of dfmr1 or FMRP does not seem to affect the siRNA pathway49, 50, 53. Therefore, it is still unclear what role, if any, FMRP has in siRNA-mediated gene silencing. The association of Dfmr1 with RISC raises the possibility that FMRP might regulate the translation of its target genes through miRNAs. Indeed, FMRP does associate with microRNAs in both Drosophila and mammals49, 50, 54. FMRP also interacts with the mammalian Argonaute protein, EIF2C2, which is a component of microRNA-containing mRNP complexes54. In addition, FMRP is associated with Dicer activity, suggesting that FMRP might be involved in the processing of microRNA precursors as well49, 50, 54.
To test the functional importance of these interactions further, our group examined the genetic interaction between dfmr1 and AGO1, the Drosophila homologue of EIF2C2. We found that AGO1 is required for dfmr1-mediated regulation of synaptic plasticity in an AGO1 loss-of-function model. Moreover, partial loss of AGO1 could suppress neuronal apoptosis caused by the overexpression of dfmr1 (ref. 54). Together these data suggest that AGO1 is critical for the biological functions of FMRP in neural development and synaptogenesis54. It recently has been found that dfmr1 also interacts genetically with AGO2 (ref. 30). The level of ppk1 mRNA, an mRNA target of dfmr1 protein, seems to be regulated by dfmr1 and AGO2, although the mechanism of this regulation is unclear.
The current data strongly support the idea that FMRP could regulate the translation of its mRNA through miRNA interaction. A likely scenario is that once FMRP binds to its specific mRNA ligands, it recruits RISC along with miRNAs and facilitates the recognition between miRNAs and their mRNA ligands. Thus, FMRP might modulate the efficiency of translation of its mRNA targets using miRNAs. This mechanism would allow this activity to be rapid and reversible, a requirement for protein synthesis-dependent synaptic plasticity. The establishment of the link between fragile X mental retardation and the miRNA pathway provides a new avenue to study not only the molecular pathogenesis of fragile X mental retardation, but also miRNA-mediated translational regulation. Indeed, fragile X syndrome is the only disease that has been linked to the dysfunction of the miRNA pathway thus far.
Neuronal dysfunction in the absence of FMRP
Studies from human brain tissues and mouse and Drosophila models suggest that FMRP is involved in synaptic plasticity and dendritic development. This involvement was first suggested by the increased density of abnormally long dendritic spines in fragile X patients, which is consistent with the notion that dendritic spine dysgenesis is associated with human mental retardation19, 20. In Fmr1-knockout mice, the dendritic spines are abnormal early in postnatal life. These abnormalities are most pronounced in the somatosensory cortex, which undergoes a period of increased synaptogenesis during this early phase of development21. Interestingly, these anomalies seem to be transient and mostly subside by the end of the first postnatal month.
In Drosophila, dfmr1 is involved in synaptic function at the neuromuscular junction and in dendrite development in different types of neurons, such as lateral (LNv), dorsal cluster (DC) and multiple dendritic (MD) neurons, as well as mushroom bodies (MB)29, 30, 55, 56, 57. Loss of dfmr1 expression could alter certain behaviours related to these types of neurons.
Recent data suggest that the role of FMRP in neuronal dysgenesis is the regulation of microtubule stability through FMRP-mediated control of translation. In Drosophila, dfmr1 protein regulates microtubule stability during both synaptogenesis in the nervous system and spermatogenesis in the testes58. In mammals, one of the key components of microtubule stability, MAP1B, is an mRNA target of FMRP. Developmentally programmed FMRP expression represses MAP1B translation and is required for the accelerated decline of MAP1B during active synaptogenesis in murine neonatal brain development45. Indeed, the lack of FMRP results in misregulated MAP1B translation and delayed MAP1B decline in the brains of Fmr1-knockout mice. The aberrantly elevated MAP1B protein expression leads to an abnormal increase in microtubule stability in the neurons of the Fmr1-knockout mouse45. These results indicate that FMRP has critical roles in the control of cytoskeleton organization during neural development and that loss of FMRP leads to abnormal microtubule dynamics and altered plasticity.
There are two major types of synaptic plasticity that are dependent on protein synthesis: long-term potentiation (LTP) and long-term depression (LTD). LTP is a long-term increase in synaptic strength in response to high-frequency stimulation, whereas LTD is a decrease in the strength of the same synapses after prolonged, low-frequency stimulation. There are two well-studied forms of LTD, NMDA receptor (NMDAR)-mediated LTD (NMDAR-LTD) and mGluR-dependent LTD (mGluR-LTD)59. To investigate further the role of FMRP-mediated translation control in neuronal activity, local protein synthesis at the synapse was monitored in the Fmr1-knockout mouse. Normally, activation of mGluR5 by its agonist DHPG stimulates LTD in a process that requires new protein synthesis without needing new RNA transcription, suggesting that pre-existing messages might be held in a translationally repressed state until an activity-dependent signal releases the messages. Consistent with this idea, mGluR-LTD (but not NMDAR-LTD) is enhanced substantially in the hippocampus of Fmr1-knockout mice, whereas LTP was unaffected60. This led to the hypothesis that mGluR5 stimulates local protein synthesis while FMRP suppresses local protein synthesis, and the two perhaps work in concert to fine-tune activity-dependent local protein synthesis. In the absence of FMRP, mGluR5-stimulated messages could be translated at higher-than-normal levels. Indeed, excessive activation of mGluR5 by DHPG in normal neuron cultures leads to a dendritic morphology remarkably similar to that seen with endogenous mGluR5 activation in the Fmr1-knockout neurons61. The role of FMRP in mGluR5-dependent LTD suggests new treatment approaches for fragile X syndrome, based on antagonists of mGluR5 to temper the increased translation of mGluR5-stimulated messages that occurs in the absence of FMRP62.
RNAi-mediated methylation of expanded CGG repeats?
Investigations of the timing and molecular mechanisms underlying the specific methylation of the expanded CGG repeats (full mutation) provide us with a unique model to study epigenetic regulation in vivo. In individuals with fragile X syndrome, the expansion of the CGG repeat tract to more than 200 copies triggers FMR1 methylation of CpG cytosines and heterochromatin marks. This silences FMR1 transcription. Notably, treatment of fragile X cells with the DNA methylation inhibitor 5-aza-2'-deoxycytidine (azadC) reverses the methylation and leads to a modest but transient reactivation of FMR1 transcription63, 64.
Methylation of the expanded FMR1 CGG repeat occurs early in development and is a dynamic process. In intact ovaries from human fetuses that carry a full mutation, the FMR1 repeat is fully expanded and unmethylated. By contrast, in chorionic villus samples from fetuses carrying a full mutation, the expanded repeat is methylated to an increasing degree as development progresses7.
So far, the molecular mechanisms that trigger the methylation of cytosines in the repeat tract and in the flanking CpG islands during early development remain a mystery. However, recent studies suggest that RNAi provides RNA specificity for the precise targeting of gene-silencing chromatin complexes to particular genomic loci65. In Arabidopsis, both DNA methylation and histone modification require the RNAi machinery66, 67, and in Schizosaccharomyces pombe, Dicer, RdRp (RNA-dependent RNA polymerase) and ago1 are required for heterochromatin formation68, 69. In Drosophila, mutations in RNAi components (aubergine/sting, piwi and homoless/spindle-E) alleviate heterochromatin-mediated gene silencing70. The involvement of RNAi in these processes is further supported by the finding that endogenous small RNAs corresponding to repetitive elements and transposons are often associated with heterochromatic regions, and that these small RNAs might be derived from long dsRNAs generated by bidirectional transcription of repetitive sequences from adjacent promoters71. This led to the hypothesis that a RISC-like heterochromatin-targeting complex called RITS (for RNA-induced initiator of transcriptional gene silencing) containing Argonaute binds to siRNAs and promotes their pairing to either nascent transcripts or to homologous DNA sequences at the target locus72.
A premutation CGG repeat RNA was recently found to form a single stable hairpin structure73. The most prominent hairpin forms in the 3' part of the repeat and involves the long uninterrupted CGG repeats. Interestingly, the rCGG repeats can be cleaved by Dicer (the central component of RNAi complex) to produce small RNAs (
20 nucleotides). Based on this observation, and the role of the RNAi pathway in chromatin modification, we propose a model for the methylation of expanded CGG repeats (Fig. 2). During early development, the full mutation allele of the FMR1 gene is unmethylated and could be transcribed to produce mRNAs containing expanded CGG repeat RNAs. The expanded rCGG repeat forms a hairpin structure and becomes a Dicer substrate. Dicer cleaves the double-stranded rCGG repeats to produce small RNAs. These siRNA-like RNAs join the RITS complex and guide the complex to homologous sequences in a process that probably involves the pairing of siRNAs to nascent RNA transcripts or to DNA directly. The RITS-mediated recruitment of histone methyltransferase(s) and DNA de novo methyltransferase(s) leads to full methylation of the expanded repeat and the resulting transcriptional repression at the FMR1 locus.
Figure 2: Model for RNAi-mediated methylation of expanded CGG repeats in individuals with fragile X syndrome.
In the 5' untranslated region of the fragile X mental retardation-1 (FMR1) gene, CGG repeats can be expanded to up to 1,000 repeats. Having more than 200 repeats in FMR1 is associated with fragile X syndrome. Early in development, transcripts arising from the expanded allele form a hairpin structure that is cleaved by Dicer to produce small RNAs (20 nucleotides). These siRNA-like RNAs associate with the 'RNA-induced initiator of transcriptional gene silencing' (RITS) complex, and direct the complex to homologous sequences by base-pairing. RITS-mediated recruitment of DNA de novo methyltransferase(s) (DMTases) and/or histone methyltransferases (HMT) initiates local methylation (black circles) of the FMR1 5' untranslated region, which ultimately leads to heterochromatin formation and transcriptional repression of the FMR1 locus.
Summary
Although the FMR1 gene responsible for fragile X syndrome was cloned more than 13 years ago and has been extensively studied since, it continues to hold surprises. Fragile X syndrome has proven to be an excellent disease model not only to understand learning and memory, but also to provide an opportunity to study mRNA transport and local protein synthesis in neurons. By integrating biochemical, genetic, genomic, physiological and behavioural approaches, more exciting findings are expected.
