Non-coding RNAs in the nervous system
Corresponding author J. S. Mattick: ARC Special Research Centre for Functional and Applied Genomics, Institute for Molecular Bioscience, University of Queensland, St Lucia, QLD 4072, Australia. Email: firstname.lastname@example.org
Increasing evidence suggests that the development and function of the nervous system is heavily dependent on RNA editing and the intricate spatiotemporal expression of a wide repertoire of non-coding RNAs, including micro RNAs, small nucleolar RNAs and longer non-coding RNAs. Non-coding RNAs may provide the key to understanding the multi-tiered links between neural development, nervous system function, and neurological diseases.
The nervous system is unique among organs in its precise and sophisticated patterns of regional cellular morphogenesis, cellular diversity, membrane electrical properties, responses to changing environmental inputs and perturbations, neural network connections, and dynamic activity-dependent alterations in synaptic strength underlying higher order cognitive functions including learning and memory (Abrous et al. 2005). These functional properties are, in turn, orchestrated by a corresponding set of multilayered developmental mechanisms (Mehler, 2002a,b), including neural induction, neural patterning and axis formation within the evolving neural plate and neural tube, elaboration of stem cell generative zones throughout the neuraxis and the evolution of connections between specialized regional neuronal and glial cell types.
Alterations of specific components of these developmental stages and maturational processes result in a broad spectrum of neurodevelopmental disorders and predispose to an equally complex array of adult neurological and neuropsychiatric disorders of unknown aetiology, underscoring the levels of complexity in developmental and mature brain–behaviour relationships. However, we have little understanding of the genetic programs and molecular mechanisms that orchestrate nervous system development, plasticity and function, or how these programs and mechanisms are perturbed in disease.
Although only about 1.2% of the mammalian genome encodes proteins, most of the genome is transcribed, in complex patterns of interlacing and overlapping transcripts from both strands (Carninci et al. 2005; Cheng et al. 2005a; Frith et al. 2005; Katayama et al. 2005; Engstrom et al. 2006; Mattick & Makunin, 2006), at least some of which are processed to form small regulatory RNAs such as microRNAs and small nucleolar RNAs (reviewed in Mattick & Makunin, 2005). A range of evidence suggests that these RNAs form complex networks that direct the trajectories of differentiation and development, via regulation of chromatin modification, transcription, RNA modification, splicing, mRNA translation, and RNA stability (Mattick & Gagen, 2001; Mattick, 2003, 2004a) as well as other mechanisms (Prasanth et al. 2005; Willingham et al. 2005). It is also clear that multiple classes of non-coding RNAs (ncRNAs) are overly represented in the central and peripheral nervous system (Hsieh & Gage, 2004; Kim et al. 2004; Rogelj & Giese, 2004; Cheng et al. 2005b; Davies et al. 2005; Klein et al. 2005; Rogaev, 2005; Cao et al. 2006; Ravasi et al. 2006), underscoring the likelihood that nervous system development and function is heavily dependent on RNA regulatory networks, and that perturbations of these networks underlie many neurological diseases.
MicroRNAs (miRNAs) are short 21–23 nucleotide regulatory sequences that inhibit the translation or stability of target RNAs (reviewed in Mattick & Makunin, 2005; Zamore & Haley, 2005). In mice, there are numerous brain-specific miRNAs (Krichevsky et al. 2003; Cheng et al. 2005b; Lim et al. 2005; Xie et al. 2005), a significant subset of which have been directly implicated in neural development and neural cell differentiation (Kawasaki & Taira, 2003; Smirnova et al. 2005). A wide variety of miRNAs are localized to neuronal subtypes with the highest concentration in the cerebral cortex and the cerebellum (Kosik & Krichevsky, 2005; Krichevsky et al. 2006). Additional miRNAs are present within glial cell subtypes with others exhibiting more ubiquitous or neural progenitor cell-specific patterns of expression (Krichevsky et al. 2003; Klein et al. 2005; Smirnova et al. 2005). In zebrafish, the miRNA miR-430 rescues defects of neurulation, neural tube formation, segmental morphogenesis, neural stem cell maintenance and axonal pathfinding observed in dicer mutants that are defective in miRNA processing – although not completely, indicating that that other miRNAs are involved in later stages of neural development (Giraldez et al. 2005).
miRNAs are also abundantly expressed in the adult brain and appear to regulate the maintenance of mature neural traits and synaptic plasticity (Krichevsky et al. 2003; Jin et al. 2004; Sempere et al. 2004; Cheng et al. 2005b; Kosik & Krichevsky, 2005; Smirnova et al. 2005; Conaco et al. 2006; Schratt et al. 2006). Numerous studies suggest that miRNAs are intimately involved in synaptic function and input specificity during memory formation (Martin & Kosik, 2002; Schaeffer et al. 2003; Kim et al. 2004; Lugli et al. 2005; Ashraf et al. 2006; Schratt et al. 2006). Moreover, transcripts encoding synapse-associated proteins also comprise the largest subgroup of predicted miRNA targets, including synapsin 1 and the fragile X mental retardation protein (FMRP) (John et al. 2004).
A novel RNA called dsNRSE (double-stranded neuron-restrictive silencing element) that resembles a miRNA in structure and length acts as a transcriptional activator of neuronal differentiation genes by converting the neuronal silencer factor (REST/NRSF) from a transcriptional repressor in undifferentiated and non-neuronal cells to a transcriptional activator during neuroblast differentiation (Kuwabara et al. 2004). Interestingly, recent studies have revealed that REST/NRSF modulates the expression of a family of miRNAs including the CNS-specific miR-124a (Conaco et al. 2006).
Perturbations in miRNAs are associated with a number of neural diseases. Deletion of DGCR8, which encodes a component of the complex that processes miRNAs (Gregory et al. 2004; Landthaler et al. 2004), results in DiGeorge syndrome, a multi-system disorder associated with significant learning disabilities (Shiohama et al. 2003). Dysregulation of miRNAs also occurs in a mouse knockout of presenilin 1, the gene mutated in a subset of early familial forms of Alzheimer's disease (AD) (Krichevsky et al. 2003). Further, miR-175 has been implicated in a form of X-linked mental retardation (MRX3) and in a type of early onset Parkinson's disease (Waisman syndrome) (Dostie et al. 2003). Other studies have implicated miRNAs in diverse neuropsychiatric conditions, particularly those associated with developmental pathogenesis (Rogaev, 2005). In addition, predicted miRNA targets include numerous proteins implicated in neurodevelopmental and neurodegenerative diseases (Rogaev, 2005). Sequence variations in the binding site for miR-189 in the SLIT and Trk-like family member1 (SLITRK1) mRNA have been associated with Tourette's syndrome (Abelson et al. 2005). SLITRK1 is essential for neuronal growth, guidance and neurite branching and is also differentially expressed in many different neural tumours (Aruga & Mikoshiba, 2003; Aruga et al. 2003). Profound over-expression of miR-21 is seen in glioblastoma multiforme, a highly malignant tumour of the brain, whereas less dramatic degrees of miR-21 over-expression are seen in other neural-specific tumour types (Chan et al. 2005).
It is likely that there are many more miRNAs that control neural differentiation and cell-type specificity. Most known miRNAs have been identified on the basis of sequence conservation, ostensibly because they regulate many targets (making co-variation difficult), which in turn suggests that there may be many more that may not be so constrained (Mattick & Makunin, 2005). Deep sequencing has indicated that many if not most miRNAs remain to be discovered (Cummins et al. 2006), some of which appear to be primate-specific (Bentwich et al. 2005). The left–right patterning of the chemosensory organs of the nematode worm Caenorhabditis elegans is mediated by the asymmetric expression of the miRNA lsy-6, which is only expressed in a single neuron (Johnston & Hobert, 2003; Chang et al. 2004), suggesting that miRNAs may play important roles in cell-type specificity, a possibility supported by the enormous increase in the length of 3′-untranslated regions (UTRs) in higher organisms (Frith et al. 2005). The existence of additional layers of complexity in the miRNA regulatory network is further indicated by the observation that some miRNA precursors undergo adenosine–inosine (A–I) RNA editing that affects their processing and stability (Luciano et al. 2004; Yang et al. 2006) as well as the diversity of miRNAs and their targets (Blow et al. 2006).
Small nucleolar RNAs and RNA modification
Small nucleolar RNAs (snoRNAs) guide the site-specific modification of nucleotides in target RNAs, by 2′-O-ribose methylation and pseudouridylation, directed by two large families termed box C/D and box H/ACA snoRNAs, respectively (Bachellerie et al. 2002; Meier, 2005; Lestrade & Weber, 2006). The common snoRNAs are involved in rRNA modifications during ribosomal biogenesis and are localized in the nucleolus, hence the name. Related RNAs, termed scaRNAs, guide modifications of spliceosomal RNAs, are localized in cytoplasmic Cajal bodies (Meier, 2005). SnoRNAs and sno-like RNAs have recently been implicated in a spectrum of biological processes including RNA splicing, transcription, cell cycle regulation, chromosome maintenance and segregation and genomic imprinting (Huttenhofer et al. 2002; Rogelj & Giese, 2004; Royo et al. 2006). Thus it seems likely that RNA modification is employed widely as another layer of gene regulation important for developmental and functional complexity, and that, like miRNAs, many more remain to be discovered and characterized.
This appears to be particularly true in the brain. A number of brain-specific snoRNAs have been identified in mice including MBI-36, MBII-13, MBII-48, MBII-49, MBII-52, MBII-78 and MBII-85 (Cavaille et al. 2000; Huttenhofer et al. 2001; Rogelj & Giese, 2004). At least some of these miRNAs show differential expression in different areas of the brain, such as the hippocampus and amygdala, areas associated with learning and memory, and are transiently modulated during contextual memory consolidation (fear conditioning) (Rogelj et al. 2003). Human homologs of these snoRNAs are also highly enriched in brain (Cavaille et al. 2000). Certain snoRNAs (RBI-36) exhibit genus-specific functions in rat brain, further attesting to the potential complexity of non-housekeeping snoRNA functions in the nervous system (Cavaille et al. 2001).
In addition it has recently been shown that the snoRNA HBII-52 modifies the A–I RNA editing and alternative splicing of the serotonin 5-HT (2C) receptor subunit (Kishore & Stamm, 2006). HBII-52 is not expressed in the Prader-Willi developmental syndrome and 5-HT (2C) receptor isoforms distinct from the normal expression pattern are present, suggesting that anomalous splicing may contribute to disease pathogenesis (Cavaille et al. 2000; Kishore & Stamm, 2006). In humans, HBII-13, HBII-52 and HBII-85 have been mapped to the Prader-Willi syndrome locus suggesting that snoRNAs may be involved in, or regulated by, genomic imprinting (Rogelj & Giese, 2004). Many larger non-coding RNAs are also imprinted and also implicated in the genetic transactions which underlie imprinting, which clearly affects brain development and function in a variety of ways (see below).
Adenosine to inosine (A–I) RNA editing catalysed by ADARs (adenosine deaminases acting on RNA) is particularly active in the brain, especially in transcripts encoding proteins involved in nerve cell function (Bass, 2002), such as voltage-gated ion channels, ligand-gated receptors, intracellular transduction molecules, apoptosis and cell cycle arrest proteins and modulators of presynaptic terminal integrity (Morse et al. 2002; Hoopengardner et al. 2003; Maas et al. 2003; Athanasiadis et al. 2004; Gelbard, 2004; Levanon et al. 2004; Wang et al. 2004a; Valente & Nishikura, 2005). A–I editing has the capacity to change the coding capacity of mRNA (Bass, 2002), to modulate splice site choice (Laurencikiene et al. 2006), miRNA and miRNA target diversity (Blow et al. 2006), miRNA processing (Yang et al. 2006), and perhaps other targets including chromatin architecture (Fernandez et al. 2005; Valente & Nishikura, 2005), A–I editing may also be inhibited by snoRNAs (Vitali et al. 2005), further evidence of the complexity of RNA regulatory networks.
In mammals, ADARs are differentially expressed during organogenesis with ADAR3 restricted to brain and ADAR2 preferentially expressed in the nervous system (Chen et al. 2000; Bass, 2002). RNA editing also exhibits precise CNS regional specificity and essential regulatory roles during neuronal maturation (Lai et al. 1997; Kohr et al. 1998; Bernard et al. 1999; Paupard et al. 2000). RNA editing can also affect multiple sites on the same RNA with diverse functional outcomes catalysed by different ADARs (Valente & Nishikura, 2005). ADAR mutants exhibit complex behavioural defects in C. elegans, Drosophila and mice (Reenan, 2001; Tonkin et al. 2002). Moreover, abnormalities in RNA editing have been implicated in a spectrum of nervous system disorders including Alzheimer's and Huntington's diseases, amyotrophic lateral sclerosis, epilepsy, schizophrenia, depression, suicidal ideation, autosomal dominant episodic ataxia type I and Prader-Willi and Angelman syndromes (reviewed in Valente & Nishikura, 2005).
Intriguingly, in humans, A–I editing occurs far more frequently in transcripts than had been previously appreciated, with the vast majority of the editing occurring in inverted Alu repeats predicted to form intramolecular duplexes in non-coding RNA sequences in introns, intergenic transcripts and UTRs (Athanasiadis et al. 2004; Blow et al. 2004; Kim et al. 2004; Levanon et al. 2004). These observations raise the intriguing possibility that the predominance of Alu elements in the human genome (10.5% of which is comprised of Alu elements) may not be simply an accident of history, but the result of positive selection for these sequences as a natural substrate for A–I editing, in turn driven by selection for increased cognitive capacity in the primate lineage (Mattick, 2004b).
It is also worth noting that other small brain-specific trans-acting RNAs such as the primate-specific dendritic BC200 RNA and the analogous rodent dendritic BC1 RNA are both descended from retrotransposed sequences (Martignetti & Brosius, 1993; Ohashi et al. 2000), and it appears likely that many, if not most, transposon-derived sequences in our genome have been exapted into function, primarily at the regulatory level (Brosius, 1999).
Longer non-coding RNAs
There are tens of thousands of larger ncRNAs, both polyadenylated and non-polyadenylated, that are transcribed from the mammalian genome (Carninci et al. 2005; Cheng et al. 2005a; Kapranov et al. 2005; Engstrom et al. 2006), many of which appear to be evolving rapidly (Pang et al. 2006). Many of these ncRNAs are developmentally regulated, alternatively spliced and physiologically responsive, and show particular abundance in the brain (Inagaki et al. 2005; Ravasi et al. 2006). A subset of these longer ncRNAs has recently been categorized as macroRNAs or long expressed non-coding regions of mouse (ENOR) (Furuno et al. 2006). Some ENOR loci (ENOR28, 31) produce several macroRNAs, all enriched in brain. Many identified loci display evidence of imprinting and antisense transcription, represent host genes for miRNAs and snoRNAs, exhibit greater nuclear than cytoplasmic localization and are overly represented within the nervous system (Furuno et al. 2006).
Multiple classes of ncRNAs are involved in dendrite development, mRNA transport and targeting, and local protein synthesis associated with synapse-specific forms of plasticity and accompanying long-term changes in synaptic strength (reviewed in Jin et al. 2004). Central to this process is the RNA binding protein, FMRP, whose absence results in mental retardation, epilepsy, autism and anxiety disorders. FMRP is part of an elaborate ribonucleoprotein complex that includes FXR1P/2P, nucleolin, YB1/p50, Purα, staufen, IMP1 (RNA transport factor) and kinesin 5 (Ceman et al. 1999, 2000; Ohashi et al. 2000; Kanai et al. 2004; Rackham & Brown, 2004). Purα links the cytoplasmic mouse and human ncRNAs BC1/200 to microtubules whereby they are transported to dendrites to participate in ‘synaptic tagging’, sites of coincident electrical and molecular activities essential to encode synapse specificity for the targeted propagation of informational signals within regional neural networks (Johnson et al. 2006; Xu et al. 2006). FMRP also binds to BC1/200 to promote dendritic localization (Kobayashi et al. 1998; Muslimov et al. 1998; Wu & Hecht, 2000; Brosius & Tiedge, 2001; Rogelj & Giese, 2004).
In Drosophila, large ncRNAs are involved in multiple stages of organogenesis and cellular differentiation. These ncRNAs exhibit rapid evolution, preferential expression during embryogenesis and dramatic central and peripheral nervous system tissue specificity (Inagaki et al. 2005). For example, the bereft RNA is essential in the Drosophila peripheral nervous system for extrasensory organ development and the maintenance of interommatidial bristles of the eye (Hardiman et al. 2002).
Most mammalian genes also have antisense transcripts (Katayama et al. 2005), many of which are abundant in the nervous system, although the extent and complexity of their regulatory actions are only beginning to be understood (reviewed in Korneev & O'Shea, 2005). These antisense RNAs exhibit dynamic developmentally regulated and spatially discrete expression profiles, and modulate the expression of genes involved in brain morphogenesis, stem cell renewal and proliferation, stress responses, cell polarity and cytoskeletal functions, and neuronal survival, maturation and synaptic plasticity (Korneev & O'Shea, 2005).
In both schizophrenia and bipolar illness, susceptibility loci are present within the disabled in schizophrenia 1 (DISC1) gene and in the large antisense DISC2 RNA that modulates its expression (Millar et al. 2000, 2004). DISC1 is involved in intracellular transport, cell polarity and neuronal migration and disruption of function during cortical developmental may, in part, underlie the developmental pathogenesis of these heterogeneous neuropsychiatric diseases (Kamiya et al. 2005).
Non-coding RNAs and brain imprinting
Imprinted genes have essential roles in both neural development and adult CNS functioning, and alterations in their expression profiles are associated with a spectrum of complex neurodevelopmental and neuropsychiatric disorders (Costa, 2005; Davies et al. 2005, 2006). These allele-selective genes exhibit preferential and exquisite cell-specific patterns of expression within the brain, and are frequently processed from larger transcriptional units encompassing multiple tandemly repeated snoRNAs and miRNAs (Sleutels et al. 2000; Seitz et al. 2004; Davies et al. 2005; Lewis & Reik, 2006). These imprinted loci also generate a complex spectrum of spliced and unspliced larger ncRNAs of presently unknown function (Sleutels et al. 2000; Davies et al. 2005; O'Neill, 2005; Furuno et al. 2006). Additional ncRNAs associated with imprinted loci include the production of antisense RNAs to reciprocally imprinted neighbouring protein-coding genes (Sleutels et al. 2000; Davies et al. 2005). The seminal role of imprinted genes in regulating distinct brain signalling systems and in mediating brain–behaviour relationships is illustrated by the spectrum of neurological diseases associated with parent of origin effects and caused by disruptions in imprinted loci: autism, schizophrenia, attention deficit hyperactivity disorder, bipolar disorder and Tourette's syndrome (see Davies et al. 2004, 2005, 2006; Wang et al. 2004b).
Transfer RNAs and ribosomal RNAs
Transfer RNAs (tRNAs) and ribosomal RNAs (rRNAs) have recently been implicated in a broad array of neural developmental and mature CNS functions as indicated by the effects of mutations in these two classes of ncRNAs which underlie a range of neurodevelopmental, neurodegenerative and neuropsychiatric diseases, including chronic progressive external ophthalmoplegia (CPEO), Kearn-Sayre syndrome (KSS: CPEO with retinal degeneration), MELAS syndrome (mitochondrial encephalopathy with stroke-like syndromes and migraine headaches), MERRF syndrome (myoclonus epilepsy, mitochondrial myopathy, cerebellar ataxia and less commonly dementia, hearing loss and peripheral neuropathy) (reviewed in Dimauro, 2004; Dimauro & Davidzon, 2005; Fattal et al. 2006) and motor neuron disease (Borthwick et al. 2006). MELAS syndrome and other tRNA-mediated diseases are also associated with prominent neuropsychiatric diseases including schizophrenia, psychosis, delirium, personality disorders, major depressive disorders, and anxiety disorders (Fattal et al. 2006).
RNA trinucleotide expansions
A range of neurodevelopmental and neurodegenerative diseases associated with trinucleotide repeat expansion appear to be caused by RNA-mediated mechanisms (reviewed in Gallo et al. 2005; Gatchel & Zoghbi, 2005). These include fragile X syndrome which results from dramatically expanded (> 200) CGG repeats in the 5′ UTR of the Fmr1 gene and the related disease associated with smaller (60–200) trinucleotide repeat expansions called FXTAS (fragile X tremor/ataxia syndrome) (FXTAS) associated with tremor, cerebellar ataxia, cognitive decline, peripheral neuropathy, Parkinson's disease, autonomic dysfunction, proximal muscle weakness, multisystem atrophy and dementia (Hagerman et al. 2005; Van Esch, 2006). Trinucleotide repeat expansions also underlie myotonic dystrophy, which is predominantly a muscle disorder but exists in two forms with associated CNS pathology: DM1 with mental retardation, memory and visuo-spatial and executive dysfunction and DM2 with preferential executive dysfunction (D'Angelo & Bresolin, 2006). DM1 is associated with CTG expansion within the 3′ UTR of the dystrophia myotonica protein kinase gene, DMPK, and DM2 is linked to CCTG expansion in intron 1 of the zinc finger protein gene, ZNF9 (Brook et al. 1992; Fu et al. 1992; Mahadevan et al. 1992; Ranum et al. 1998; Liquori et al. 2001). These mutant RNAs orchestrate different forms of pathogenesis through the degree and type of repeat length expansion and their molecular interactions with RNA-binding proteins of the muscleblind-like (MBNL) family (Jiang et al. 2004; Pascual et al. 2006).
Several forms of spinocerebellar ataxia (SCA) may also be caused by different RNA-mediated pathological mechanisms. SCA8 results from CTG expansion of the 3′ UTR of an untranslated antisense RNA with partial overlap with the Kelch-like 1 (KLHL1) gene (Koob et al. 1999; Nemes et al. 2000; Mutsuddi et al. 2004; Gatchel & Zoghbi, 2005). Moreover, using SCA8 as a sensitized background in a modifier screen resulted in the identification of four novel ncRNAs with preferential neuronal expression (Mutsuddi et al. 2004). SCA10 is mediated by an unstable ATTCT repeat expansion in the 3′ end of a large intron of a gene of presently unknown function that may result in transcriptional silencing or in a different RNA-associated toxic mechanism (Matsuura et al. 2000). SCA12 is caused by CAG expansion in the non-coding 5′ promoter/5′ UTR of the PPP2R2B gene, which encodes a brain-specific regulatory subunit of protein phosphatase 2A (Holmes et al. 1999). Depending on the precise location of the expanded trinucleotide repeat, disease pathogenesis may be mediated by distinct trans-dominant RNA or alternate toxic gain of function mechanisms (Holmes et al. 2003).
The known list of both small and large ncRNAs that are involved in the nervous system almost certainly represents only a tiny fraction of the total transcriptome devoted to RNA-mediated mechanisms underlying the development, functional complexity and plasticity of the mammalian brain. Indeed, it appears that the majority of human genomic programming is devoted to RNA-based regulatory circuitry (Mattick, 2003; Mattick & Makunin, 2006). It also appears that the traditional presumption that most genetic information is transacted by proteins has led to a fundamental misunderstanding of the genetic programming of human differentiation and development, both generally and specifically in the brain, where RNA transactions appear to be at their most complex.
This work was supported by grants from the National Institutes of Health (NS38902, MH66290, HD01799) and the F. M. Kirby, the Skirball, the Rosanne H. Silbermann and the Roslyn and Leslie Goldstein Foundations (M.F.M.), and by the Australian Research Council (Federation Fellowship Grant FF0561986), the Queensland State Government and the University of Queensland (J.S.M.). We thank Marcel Dinger and Ryan Taft for helpful comments on the manuscript.