RNA-protein interactions in unstable microsatellite diseases

Abstract

A novel RNA-mediated disease mechanism has emerged from studies on dominantly inherited neurological disorders caused by unstable microsatellite expansions in non-coding regions of the genome. These non-coding tandem repeat expansions trigger the production of unusual RNAs that gain a toxic function, which involves the formation of RNA repeat structures that interact with, and alter the activities of, various factors required for normal RNA processing as well as additional cellular functions. In this review, we explore the deleterious effects of toxic RNA expression and discuss the various model systems currently available for studying RNA gain-of-function in neurologic diseases. Common themes, including bidirectional transcription and repeat-associated non-ATG (RAN) translation, have recently emerged from expansion disease studies. These and other discoveries have highlighted the need for further investigations designed to provide the additional mechanistic insights essential for future therapeutic development.

1. Introduction: Unstable microsatellites in neurological disease

Approximately 50% of the human genome consists of repetitive elements of which 3% consists of simple sequence repeats (SSR) (Treangen and Salzberg, 2012). These SSRs, more commonly referred to as microsatellites, are composed of short tandem repeats of 2–10 base pairs (bp) that are dispersed throughout the genome. Microsatellites have been proposed to function at multiple steps in gene expression when present within or near genes in both prokaryotes and eukaryotes (Usdin, 2008). These steps include transcription where tandem repeats act as transcriptional regulatory elements (Meloni et al., 1998; Punga and Buhler, 2010), pre-mRNA splicing with modulation of splicing activity by microsatellite polymorphisms (Pagani et al., 2000) and translation by influencing 5′ UTR ribosomal scanning (Ludwig et al., 2011). Errors in DNA replication, recombination and mismatch repair cause microsatellite instability leading either to repeat tract expansion or contraction. Importantly, the majority of microsatellite expansion diseases are neurological/neuromuscular disorders although many of the affected genes are expressed ubiquitously. For example, polyglutamine (polyQ) disorders, including Huntington disease (HD), spinocerebellar ataxias (SCA1, 2, 3, 6, 7, 17) and spinobulbar muscular atrophy (SBMA or Kennedy’s disease), are caused by protein coding region CAG repeat expansions. These polyQ expansions induce protein aggregate formation leading to altered homeostasis of multiple cellular pathways (La Spada and Taylor, 2010; Nalavade et al., 2013). Alternatively, repeat expansion mutations also occur in the non-coding regions of genes, which include 5′ and 3′ untranslated regions (UTRs) and introns, and cause dominantly inherited disorders such as myotonic dystrophy types 1 and 2 (DM1, DM2), Fragile X-associated tremor/ataxia syndrome (FXTAS) and SCA types 8, 10 and 12 (Ranum and Cooper, 2006; Poulos et al, 2011). The recent discovery of a repeat expansion mutation in chromosome 9-linked amyotrophic lateral sclerosis and frontotemporal dementia (C9ORF72 ALS/FTD) has highlighted the importance of elucidating the molecular mechanisms underlying these non-coding expansion disorders (DeJesus-Hernandez et al, 2011; Renton et al, 2011). One disease model proposes that non-coding mutations are deleterious at the RNA level (RNAopathy) because they fold into stable RNA structures that either inhibit or enhance the normal activities of important cellular factors. In contrast, RNA repeat expansions are also prone to a non-canonical type of protein translation, or repeat-associated non-ATG (RAN) translation, and the resulting unusual peptides, which for SCA8 are composed of ATXN8 polyQ, polyserine and polyalanine tracts, may induce disease-associated pathology (proteinopathy) (Zu et al, 2011). These and other studies suggest that the full range of molecular pathways that are compromised in these diseases remains to be determined. In this review, we discuss the potential interplay between these pathogenic mechanisms with a focus on RNA gain-of-function, RNA binding proteins and the available model systems to study RNA toxicity.

2. Origins of microsatellite expansions

Microsatellite intergenerational instability is a major feature underlying expansion diseases with subsequent generations experiencing anticipation, which is characterized by increased disease severity and earlier onset age triggered by an increase in repeat length (Pearson et al., 2005). Even within an individual, repeat tract length exhibits somatic mosaicism, or heterogeneity within and between tissues, with a positive correlation existing between repeat number and tissue-specific pathology in diseases like HD and DM1 (Kennedy et al., 2003). Although our understanding of the mechanisms underlying repeat instability is incomplete, considerable evidence points to pivotal roles for errors induced during DNA replication and repair depending on tissue developmental and proliferative state (Budworth and McMurray, 2013; Lopez Castel et al., 2010; Mirkin, 2007; Usdin, 2008). During DNA replication, instability is influenced by both cis- and trans-acting factors and cell studies have shown that cis-elements (repeat sequence, tract length, distance from the replication origin and replication direction) can affect repeat instability by modulating replication fork dynamics during lagging strand synthesis (Cleary et al., 2002). Additionally, epigenetic factors, including the CpG methylation status of CTCF binding sites flanking the repeat locus and histone modifications that affect chromatin structure and arrangement, impact repeat instability (Dion and Wilson, 2009; Libby et al., 2008). Repeats also have the potential to form non-canonical structures, such as non-B-form DNA-like triple helices, G-quadruplexes, intra-strand hairpins and slipped strand structures, which result in replication fork stalling and template switching (Lopez Castel et al., 2010).

Repeat instability in terminally differentiated cells is also affected by transcription. The majority (>80%) of the genome undergoes transcription (Hangauer et al., 2013) and widespread antisense transcription has been reported for many loci, including a majority of microsatellite disease-associated genes (Batra et al., 2010; Budworth and McMurray, 2013). Both CTG and CAG repeat expansions have been shown to enhance repeat instability by several fold upon unidirectional and bidirectional transcription induction in mammalian cells (Lin et al., 2006; Nakamori et al., 2011). Although the mechanisms underlying transcription-induced repeat instability are poorly defined, it is likely that the separation of DNA strands during transcription results in secondary structure formation by the repeats followed by protein-induced stabilization of these structures (McIvor et al., 2010). During bidirectional transcription, head-on collision between RNA polymerases may also occur and cause stalling and activation of downstream DNA damage response pathways (Lin and Wilson, 2011).

Trans-acting factors, including the mismatch match repair (MMR) proteins MSH2, MSH3, MSH6 and PMS2, are also critical drivers of repeat instability. Repeat contractions and stabilization occur in mice harboring CTG•CAG repeats when they are crossed with either Msh2 or Msh3 null mice or mice deficient in Msh2 ATPase activity indicating a role for these proteins in promoting instability in DM1 (Pearson et al., 2005). Similarly, Msh2 is important for promoting both intergenerational and somatic repeat expansions in a FXTAS model expressing CGG•CCG repeats (Lokanga et al., 2014). Naturally occurring Msh3 polymorphisms may act as a modifier of CAG repeat instability by interfering with the stability of the Msh3 protein (Tome et al., 2013) and the resulting variations in protein levels may account for some aspects of region-specific instability seen in the striatum of HD patient brains (Gonitel et al., 2008; Pinto et al., 2013). How do these proteins influence repeat expansions? Studies in S. cerevisiae demonstrate that Msh2 and Msh3 alter the activities of proteins mediating Okazaki fragment processing resulting in small yet incremental expansions (Kantartzis et al., 2012). Since repeat instability is a complex phenomenon regulated by multiple DNA metabolic pathways that influence the various tissue and developmental stage specific expansion/contraction patterns, it is critical to understand the underlying mechanism. Although repeat instability plays the primary role in determining the course of disease, functional impairment for some non-coding expansion disorders lies at the next level upon transcription of the DNA to produce RNAs with expanded repeats.

3. Overview of RNA gain-of-function disease mechanisms

RNA sequence and higher order structures are two critical features that determine how the processing of a particular RNA occurs (Bugaut and Balasubramanian, 2012). Alterations in normal RNA sequence/structure by mutations can interfere with multiple steps in RNA biogenesis as well as the functions of the mature RNA and result in widespread dysregulation. This phenomenon is exemplified in non-coding repeat expansion diseases in which repeat expansions lead to a deleterious RNA gain-of-function (Fig. 1).

Fig. 1.

Neuronal cell model illustrating pathogenic effects of non-coding microsatellite expansions in neurologic disease. Both sense (blue line) and antisense (red line) DNA repeat expansions result in three major downstream deleterious effects (labeled 1–3): (1) protein loss-of-function (LOF) results from hypermethylation of a CG-rich expanded microsatellite repeat in the 5′ UTR and/or promoter region of an affected gene and loss of transcriptional activity (e.g., FXS); (2) RNA gain-of-function (GOF) occurs following bidirectional transcription of expanded repeats and the synthesis of sense (sRNA) and antisense (asRNA), which fold into RNA hairpins (sRNA, blue; asRNA, red) or other stable structures and gain toxic functions either by sequestering an RNA binding protein(s) (RBP) and inhibiting pre-mRNA splicing (2a), pre-mRNA editing (2b), mRNA localization (2c), pri-miR processing (2d) or by triggering aberrant cellular activities such as protein kinase C (PKC) mediated CELF1 hyperphosphorylation (2e); (3) protein GOF due to altered post-translation modifications of other RNA binding proteins (3a) (CELF1 hyperphosphorylation, purple oval with white P in orange star) or RAN translation (3b) (three homopolymeric repeat proteins are shown with one undergoing translation by the ribosome (orange). In addition to these mechanisms, other pathways, including chromatin remodeling, proteome dysregulation and vesicular trafficking, may also contribute to disease pathogenesis, particularly in the CNS (Hernandez-Hernandez et al., 2013).

3.1. RNA foci

A hallmark of most non-coding repeat expansion diseases is the formation of distinct cellular aggregates of mutant RNA, such as the nuclear foci in DM1, DM2 and C9ORF72 ALS/FTD, or the inclusions containing mutant RNA in FXTAS (DeJesus-Hernandez et al., 2011; Greco et al., 2002; Taneja et al., 1995). RNA foci are dynamic structures characterized by random formation and dissociation with the formation phase mediated by RNA binding proteins, as shown for CUG RNA foci in stably transfected C2C12 myoblasts (Querido et al., 2011). The dynamics of RNA foci for mutant repeat RNAs associated with other repeat expansion diseases have not yet been determined. These foci contain multiple copies of mutant repeat RNA and its interacting protein partners in a complex (Taneja et al., 1995; Wojciechowska and Krzyzosiak, 2011). Morphologically, RNA foci and inclusions are distinct. The C(C)UG^exp foci in DM and the rGGGGCC^exp in C9ORF72 ALS/FTD appear more compact, likely due to tighter interactions between the RNA and bound proteins, while rCGG^exp-containing inclusions in FXTAS are larger and more diffuse structures that contain many proteins, including lamins and MBNL1 (Hagerman, 2013). FXTAS brains also contain ubiquitin-positive neuronal protein inclusions that contain polyglycine (FMRpolyG) generated by RAN translation (Todd et al., 2013). An important question to address in future studies will be whether the structural distinctions between RNA foci and inclusions reflect fundamental differences in the pathogenic pathway.

Although there is a lack of information on the structures of specific RNA-protein complexes in RNA foci, several studies have analyzed synthetic repeat oligomers in vitro. Biochemical and enzymatic structure probing, as well as biophysical studies using CD and UV spectroscopy, reveal that (CNG)₂₀ transcripts form stable, but slippery, hairpins with the exception of the most stable CGG triplet (Krzyzosiak et al., 2012; Sobczak et al., 2010). Crystal structures of short CUG, CAG, CGG oligomers indicate that they adopt an A-form helical structure with non-canonical G-G, A-A and U-U wobble base pairing flanked by stabilizing GC base pairs (Kiliszek et al., 2010; Kiliszek et al., 2011; Mooers et al., 2005). DM2-associated CCUG repeats also form RNA stem-loop structures like CNG repeats but with CU mismatches and lower stability (Sobczak et al., 2003). Notably, some sequestered proteins have been suggested to preferentially recognize CNG RNA sequences and not secondary structures. For example, the MBNL proteins bind to GC base pairs and then fold the CUG^exp RNA into a pseudo A-form helix (Teplova and Patel, 2008). Recent reports have also proposed a more complex structure for ALS/FTD-associated GGGGCC^exp RNA. Studies using 1D ¹H NMR and CD spectroscopy suggest that this expansion RNA forms thermostable inter- and intra-molecular G-quadruplexes with a parallel orientation (Fratta et al., 2012; Reddy et al., 2013). The structural details of GGGGCC^exp RNA and its antisense transcript CCCCGG^exp, both of which form RNA foci in patient cells and tissues, requires additional characterization although limitations exist with these in vitro analyses. For example, prior studies have used repeats in the normal size range and the multiple RNA and protein factors that likely influence expansion RNA structures in vivo are absent. Nevertheless, in vitro evidence for higher-order structure formation by RNA repeat expansions provides intriguing insights into the nature of RNA toxicity.

Another outstanding question is whether RNA foci have a causal role in disease progression or are innocent bystanders. For DM1, the expression of mutant RNAs above a repeat length threshold (Hamshere et al., 1997) leads to foci formation and MBNL protein sequestration followed by disruption of normal MBNL functions, including alternative splicing regulation (Ranum and Cooper, 2006). For C9ORF72 ALS/FTD, antisense oligonucleotides (ASOs) targeting rGGGGCC^exp cause a reduction of sense RNA foci in patient-derived iPS cells with a concomitant rescue of the cells from glutamate-induced toxicity (Donnelly et al., 2013). These results indicate that RNA foci are not simply biomarkers but important mediators of toxicity. However, an opposing viewpoint comes from studies on a DM mouse model expressing a GFP transgene containing the DMPK 3′ UTR with a normal length (CTG)₅ repeat (Mahadevan, 2012). These mice fail to develop RNA foci but still exhibit pathological features associated with DM1 disease including muscle pathology, splicing deficits and cardiac conduction defects (Mahadevan et al., 2006). Interestingly, another DMPK transgenic model expressing a normal length (CTG)₂₀ repeat does not exhibit a pathological phenotype (Seznec et al., 2001). Since it is not currently clear if the formation of RNA foci is an essential step in pathogenesis, the possibility that overexpression of the GFP-DMPK 3′ UTR transgene also results in Mbnl functional deficiency should be tested by either rAAV-induced Mbnl1 overexpression or crossing this transgenic line to MBNL1 overexpression mice (Chamberlain and Ranum, 2012; Kanadia et al., 2006).

3.2. RNA gain-of-function

Mounting evidence indicates that microsatellite expansions in DM1/DM2, FXTAS and C9ORF72 ALS/FTD are pathogenic at the RNA level (Echeverria and Cooper, 2012; Poulos et al., 2011; Sicot and Gomes-Pereira, 2013) (Fig. 1). These diseases are characterized by a dominant inheritance pattern and are caused by mutations in the non-coding regions of their respective genes, making a conventional protein loss-of-function mechanism unlikely. Additionally, C(C)UG^exp, rCGG^exp, rGGGGCC^exp RNAs accumulate in either distinct nuclear foci or inclusions in patient cells. Although full-length mutant DMPK mRNA accumulates in nuclear RNA foci, the majority of DM-associated disease symptoms are caused by mutant RNA expansions and not by DMPK haploinsufficiency (Jansen et al., 1996; Mankodi et al., 2000; Orengo et al., 2008; Reddy et al., 1996). For the CCTG^exp and GGGGCC^exp intronic expansions in DM2 and C9ORF72 ALS/FTD, respectively, the repeats undergo splicing and accumulate in nuclear RNA foci in the absence of flanking intronic sequences (Donnelly et al., 2013; Margolis et al., 2006). Similarly, FXTAS is characterized by ubiquitin-positive, neuronal and astrocytic intranuclear inclusions (2–5 μm) containing FMR1 mRNA and various proteins coupled with a 2–10 fold elevation in FMR1 mRNA levels (Hagerman and Hagerman, 2013). In addition, Drosophila and mouse models expressing expansion RNAs exhibit disease-relevant symptoms irrespective of the gene context or the flanking sequence. Neurodegeneration and ubiquitin-positive intranuclear inclusions are observed in Drosophila expressing rCGG₉₀ repeats flanking an EGFP transgene (Jin et al., 2003) and mice expressing CTG₂₅₀ repeats driven by the human skeletal actin promoter (HSA^LR) develop myotonia and numerous ribonuclear foci in a transgene expression-dependent manner providing strong evidence for a gain-of-function at the RNA level (Mankodi et al., 2000). In the following sections, we discuss specific mechanisms of RNA-mediated toxicity and its potential adverse effects using three neurological diseases as examples, DM, FXTAS and C9ORF72 ALS/FTD.

4. Myotonic dystrophy

Myotonic dystrophy has served as a paradigm for repeat expansion diseases caused by RNA gain-of-function mechanisms, also referred to as RNA dominance (Caillet-Boudin et al., 2014; Poulos et al., 2011; Ranum and Cooper, 2006). DM1 is the most common form of adult onset muscular dystrophy (1/8000 worldwide) (O’Rourke and Swanson, 2009). DM1 and DM2 are multisystemic disorders with symptoms that include progressive skeletal muscle wasting, delayed muscle relaxation (myotonia), cardiac arrhythmias, insulin resistance, gastrointestinal problems and CNS symptoms (cerebral atrophy, hypersomnolence, memory deficits, cognitive/behavioral abnormalities and intellectual disability in the congenital disease). DM is caused by unstable repeat expansions in the non-coding region of two genes: 1) a CTG^exp in the 3′ UTR of the dystrophia myotonica protein kinase (DMPK) gene in DM1; 2) a CCTG^exp in the first intron of CCHC-type zinc finger nucleic acid binding protein (CNBP/ZNF9) gene in DM2 (Liquori et al., 2001; Ranum and Cooper, 2006). For DM1, the normal CTG repeat length ranges from 5–37, while individuals with >50 repeats exhibit the classical disease symptoms of DM1 and >1000 repeats is associated with the severe congenital form (CDM). For DM2, the pathogenic range of CCTG repeats varies from 75 – ~11,000 in patients compared to 11–26 repeats in normal individuals. Anticipation is a prominent feature of DM1 but not DM2, although somatic mosaicism and intergenerational instability have been widely reported in both DM1 and DM2 (Udd and Krahe, 2012).

4.1. DM pathogenic mechanisms

The expression of C(C)UG^exp RNA, which folds into a stable stem-loop structure, alters the activities of two developmentally regulated RNA binding protein families, MBNL and CELF, causing misregulation of multiple cellular pathways (Echeverria and Cooper, 2012; Fernandez-Costa et al., 2013; Kalsotra et al., 2014; Rau et al., 2011; Wang et al., 2012) (Fig. 1). Although CELF1, an alternative splicing factor that promotes fetal splicing patterns, is not sequestered by CUG^exp RNAs, CELF1 protein levels increase in DM1 heart and muscle tissues through a mechanism mediated by protein kinase C (PKC) (Kuyumcu-Martinez et al., 2007). The details of how CUG^exp RNA activates PKC, and if this activation is specific to DM1, still needs to be clarified. In contrast, the MBNL proteins, which activate adult splicing patterns, are sequestered by C(C)UG^exp RNAs (Miller et al., 2000) and thus the combination of an increase in CELF1 and a decrease in MBNL activity causes a shift to fetal splicing patterns of specific target transcripts in adult tissues (Osborne and Thornton, 2006). These two events make alternative splicing dysregulation a major pathogenic event in DM. In addition, abnormal DNA methylation (Castel et al., 2011), altered miRNA and mRNA expression (through decreased expression of transcription factors) (Fernandez-Costa et al., 2013; Kalsotra et al., 2014; Rau et al., 2011), bidirectional transcription of the repeats (Batra et al., 2010), and repeat associated non-ATG translation (RAN translation) (Zu et al., 2011) have also been implicated in DM pathogenesis (Fig. 1).

4.2. RNA gain-of-function models for DM

The first convincing proof that DM is an RNA-mediated disease was obtained from the HSA^LR mouse model, which expresses a human skeletal muscle (HSA) transgene with CTG₂₅₀ repeats specifically in the skeletal muscle (Mankodi et al., 2000). This model shows features reminiscent of DM including myopathy, myotonia, intranuclear CUG RNA foci and centralized myonuclei with a positive correlation between pathology and transgene expression. Additional mouse models have also been generated to overcome the limitations of this model, which include a relatively short CTG^exp, the absence of the endogenous DMPK 3′ UTR sequence and restricted muscle expression (Gomes-Pereira et al., 2011). These models include Cre-inducible transgenic mice expressing interrupted CTG₉₆₀ repeats (EpA960) or CTG₀ repeats (EpA0) and transgenic mice carrying insertions of a ~45 kb human genomic region with 20, 55, and 300 CTG repeats (Seznec et al., 2000; Wang et al., 2007). DMSXL, derived from the DM300 line, is another transgenic model with >1000 CTG repeats due to intergenerational instability that led to ‘big jumps’ in CTG repeat number (Gomes-Pereira et al., 2007). These models exhibit different aspects of DM disease, such as muscle pathology, cardiac conduction problems, behavioral abnormalities accompanied by intranuclear RNA foci pathology and some tissue-specific splicing deficits (Gomes-Pereira et al., 2011; Hernandez-Hernandez et al., 2013; Huguet et al., 2012). Taken together, these mouse models serve as valuable tools to gain insight into DM disease mechanisms and for preclinical assessment of therapeutics targeting repeat expansion RNAs.

An important consequence of C(C)UG^exp RNA toxicity is the sequestration of MBNL proteins resulting in their functional insufficiency. The MBNL proteins were first discovered as factors that bind to CUG repeat RNA in a length-dependent manner in vitro and accumulate in nuclear RNA foci in patient myoblasts (Miller et al., 2000). Several Mbnl knockout (KO) mouse models have validated this sequestration hypothesis including Mbnl1^ΔE3/ΔE3 isoform KO mice, which develop DM-relevant and multisystemic symptoms including myotonia, ocular dust-like cataracts and abnormal splicing of developmentally regulated splicing events (Kanadia et al., 2003). Moreover, recombinant adeno-associated virus (rAAV)-mediated Mbnl1 overexpression in the HSA^LR model results in phenotypic rescue of myotonia and correction of dysregulated splicing (Kanadia et al., 2006). In contrast to Mbnl1, Mbnl2 is highly expressed in the adult CNS but not in skeletal muscle and Mbnl2 is the major Mbnl family member that regulates alternative splicing in the brain. Mbnl2^ΔE2/ΔE2 isoform KO mice develop DM-associated neurological symptoms including hypersomnia, learning/memory deficits, altered GABA sensitivity and disease specific mis-splicing (Charizanis et al., 2012). The differences observed in these Mbnl KO models indicate tissue-specific splicing regulation by the Mbnl family. Interestingly, Mbnl2 compensates for loss of Mbnl1 activity in heart and skeletal muscle while the absence of both proteins leads to embryonic lethality (Lee et al., 2013a). Mbnl1; Mbnl2 conditional double KO (DKO) mice are viable but exhibit phenotypes not observed in single Mbnl KOs, such as muscle wasting and cardiac conduction defects, suggesting that DM is caused by compound loss of MBNL function triggered by the production of toxic C(C)UG^exp RNAs (Lee et al., 2013a).

A major manifestation of DM disease in the brain is neurofibrillary degeneration associated with the aggregation of abnormally phosphorylated Tau and reduced splicing of MAPT exons 2, 3 and 10 (Jiang et al., 2004; Sergeant et al., 2001). In agreement with the possibility that DM results from compound loss of MBNL function, both MBNL1 and MBNL2 are required for enhancement of MAPT exon 2 splicing (Carpentier et al., 2014). This result suggests that heterotypic interactions between these MBNL proteins, possibly mediated by MBNL C-terminal regions (Tran et al., 2011; Yuan et al., 2007), are required for MAPT splicing regulation (Carpentier et al., 2014). Still, questions remain about the role of MBNL proteins in CDM pathogenesis and if MBNL titration affects other biochemical pathways, including RNA localization (Adereth et al., 2005; Wang et al., 2012). In addition to the MBNL family, other RNA-binding proteins have been implicated in either abnormal splicing regulation in DM1 (hnRNP H, STAU1), modulation of MBNL binding to CUG repeats (p68/DDX5) and inhibition of CUG^exp mRNA export from the nucleus (Kim et al., 2005; Laurent et al., 2012; O’Rourke and Swanson, 2009; Paul et al., 2006; Ravel-Chapuis et al., 2012). Further development of both loss-, and gain-of-function animal models are required to test disease-specific roles for these proteins.

5. Fragile X and Fragile X-associated tremor/ataxia syndrome

Prominent examples where microsatellite expansion length determines disease presentation and onset are the neurological disorders Fragile X syndrome (FXS) and Fragile X-associated tremor/ataxia syndrome (FXTAS) (Hagerman and Hagerman, 2013). FXS, a neurodevelopmental disorder that is the most common form of inherited intellectual disability, is triggered by (CGG)_>200 expansions in the 5′ UTR of the FMR1 gene leading to transcriptional silencing and subsequent loss of the encoded protein, Fragile X mental retardation protein (FMRP). Interestingly, intermediate expansions of 55–200 CGG repeats, previously termed the premutation range, result in the late adult-onset neurodegenerative disorder FXTAS. The clinical phenotypes of FXTAS are distinct from FXS and >30% of these ‘premutation’ carriers experience gait ataxia, progressive intention tremor, parkinsonism and cerebral atrophy (Hagerman, 2013).

5.1. FXTAS pathogenic mechanisms

After the initial discovery of the FXTAS premutation, the observation that mutant FMR1 transcripts exhibit increased levels in FXTAS (Tassone et al., 2000) and form relatively large intranuclear inclusions in multiple organs (Greco et al., 2002) led to the suggestion that an RNA toxicity mechanism underlies FXTAS pathogenesis (Hagerman and Hagerman, 2013). Further studies have shown the presence of additional proteins, including splicing regulators and miRNA processing factors in these inclusions, but the significance of the potential interactions of the factors with CGG RNAs remains elusive. Moreover, rCGG^exp expression has been proposed to dysregulate lamin A/C, the major component of the nuclear lamina (Arocena et al., 2005). In neurons and skin fibroblasts cultured from FXTAS patients, both the percentage of soluble lamin and the nuclear lamin architecture is disrupted suggesting that certain features of FXTAS, such as peripheral neuropathy could result from a functional laminopathy (Garcia-Arocena et al., 2010). In large premutation carriers (>120 CGG repeats), moderate reduction in FMRP protein levels is observed, in spite of elevated FMR1 mRNA levels (Hagerman, 2013). It has been reported that some aspects of the disease, like cognitive/behavioral dysfunction, are due to a decrease in FMRP levels in the amygdala (Hessl et al., 2011). Similar to DM and SCA8, CGG^exp RNA undergoes RAN translation producing polyglycine in patient tissues and rCGG^exp-induced toxicity is seen in cell and animal repeat expressing models (Todd et al., 2013). The transcription of long non-coding (lnc) RNAs and the occurrence of antisense transcription at the FMR1 locus, as well as RAN translation of those transcripts, further widens the repertoire of processes associated with neurotoxicity (Ladd et al., 2007; Pastori et al., 2014). Therefore, a combination of factors, some of which are direct effects of the rCGG^exp expression, may act synergistically to promote disease onset in FXTAS.

5.2. FXTAS RNA gain-of-function models

One proposed disease mechanism for FXTAS is an rCGG^exp gain-of-function similar to that previously described for DM1, in which mutant FXTAS-associated RNAs sequester protein factors required for normal cellular pathways. To delineate the effects of RNA toxicity several transgenic and knockin model systems have been generated (Hunsaker et al., 2012). A Drosophila transgenic model expressing CGG₉₀ repeats flanking EGFP has been developed which shows progressive eye neurodegeneration in a repeat-length and transgene dose-dependent manner (Jin et al., 2003). The CGG₉₈ knockin mouse, in which the endogenous (CGG)₈ repeats of the Fmr1 gene are replaced with (CGG)₉₈, exhibits FXTAS-relevant phenotypes including increased Fmr1 mRNA levels, reduced Fmrp protein expression, ubiquitin-positive neuronal and glial intranuclear inclusions, age-dependent cognitive decline and behavioral abnormalities (Willemsen et al., 2003). Recent studies with another knockin line, (CGG)₁₅₀ mice, show that this expansion causes impaired migration and differentiation of various embryonic neocortical cells, thus affecting early brain development (Cunningham et al., 2011). This embryonic defect could account for the developmental problems observed in premutation carrier children. The contribution of CGG^exp RNA toxicity to the neurodegenerative phenotype in FXTAS is not clearly understood from these mouse models due to the decreased Fmrp protein levels. Thus, a transgenic mouse model ectopically expressing CGG₉₀ repeats outside of context of Fmr1 gene in Purkinje neurons, while maintaining normal Fmrp levels, has been generated (Hashem et al., 2009). Interestingly, in addition to intranuclear inclusions, the mice exhibit Purkinje cell loss accompanied by axonal swelling and age-dependent neuromotor learning deficits demonstrating the importance of rCGG^exp toxicity in inclusion formation and in mediating neurotoxicity.

An early attempt to define the composition of FXTAS inclusions isolated from autopsied FXTAS brains used mass spectrometry and identified ~20 proteins, including MBNL1, hnRNP A2/B1 and lamin A/C (Iwahashi et al., 2006). Alternative experimental approaches such as RNA affinity pulldowns, have uncovered additional rCGG repeat binding factors (Sam68, Purα, DGCR8, DROSHA) (Hagerman, 2013; Jin et al., 2007; Sellier et al., 2010; Sellier et al., 2013). Overexpression of Purα, a transcriptional activator with probable roles in DNA replication and recombination, and hnRNPA2/B1, a ubiquitously expressed hnRNP involved in pre-mRNA processing and RNA trafficking, mitigates the neurodegenerative phenotypes seen in Drosophila transgenic rCGG^exp lines (Jin et al., 2007; Sofola et al., 2007). Although Purα knockout mice undergo premature death and show severe neurological features, such as spontaneous seizures and tremors (Khalili et al., 2003), the relevance of Purα sequestration to FXTAS phenotypes remains unclear. For Sam68, a KH domain-containing alternative splicing factor, there is evidence for splicing alterations of some of its targets (SMN2, ATP11B) in FXTAS patients (Sellier et al., 2010), though the downstream effect of this dysregulation and the effect of sequestration on other Sam68 targets remains to be determined. In addition, DGCR8 and DROSHA, the proteins central to pri-miR processing are partially sequestrated in intranuclear rCGG^exp-containing inclusions in FXTAS patients and mammalian cells expressing premutation repeats, accompanied by global reduction of mature miRNA levels (Sellier et al., 2013). Overexpression of DGCR8, but not DROSHA, Sam68 or MBNL1, rescues CGG^exp-induced toxic phenotypes including reduced cell viability and decreased dendritic complexity in E18 cortical neurons suggesting that DGCR8 loss-of-function is an important feature of FXTAS pathogenesis. However, important questions remain. Are rCGG^exp-containing inclusions functionally equivalent to RNA foci? How do these various proteins interact with rCGG^exp in these inclusions and what is the extent of their sequestration? What are the pathogenic roles of the antisense transcript ASFMR1 and does this mutant RNA also sequester factors, form inclusions and undergo RAN translation? Additional mechanistic approaches using different cellular and animal models are required to better understand these fundamental questions.

6. C9ORF72-linked amyotrophic lateral sclerosis and frontotemporal dementia

Recently, a non-coding GGGGCC^exp mutation in the first intron of the C9ORF72 gene was discovered as a major cause of ALS/FTD (DeJesus-Hernandez et al., 2011; Renton et al., 2011). ALS is a debilitating neurodegenerative disorder in which upper and lower motor neuron death results in muscle weakness, wasting, difficulty in swallowing/breathing leading to paralysis and death usually within 3–5 years of onset. FTD causes the second most common form of presenile dementia (age of onset <65 years of age) and is a complex disorder affecting language, cognitive and behavioral skills. Approximately 40% of FTD patients suffer from ALS-like motor symptoms and 50% of ALS patients exhibit FTD-associated behavioral and personality changes (Ling et al., 2013). Although there is an overlap in patients affected with both of these conditions, it is interesting to note the degree of phenotypic variability existing in the remainder of the C9ORF72 expansion carrier population. It is likely that genetic modifiers play a major role in determining disease presentation. For example, an allelic variant of TMEM106B protects carriers from developing FTD but not ALS (van Blitterswijk et al., 2014). Importantly, the differential effects of genetic modifiers is highlighted by the discovery that TMEM106B variants also act as risk factors for frontotemporal lobar degeneration with neuronal inclusions of hyperphosphorylated and ubiquinated TDP-43 (FTLD-TDP) (Van Deerlin et al., 2010).

6.1. C9ORF72 ALS/FTD disease mechanisms

Evidence that both C9ORF72 sense rGGGGCC^exp and antisense rCCCCGG^exp RNAs form nuclear RNA foci in neuronal and non-neuronal cells suggests a toxic RNA gain-of-function and protein sequestration mechanisms for ALS/FTD (Lagier-Tourenne et al., 2013). In situ hybridization experiments suggest that the incidence of sense foci is greater than antisense foci, but the number of antisense foci per cell is greater (Mizielinska et al., 2013). This result should be considered with caution because of differences in FISH probe affinity for sense and antisense RNAs. Nonetheless, it will be interesting to differentiate the pathogenic effect exerted by these individual entities. Transcriptome analysis of patient fibroblasts treated with ASOs targeting C9ORF72 sense RNA reveals that the disease-specific RNA signature is not reversed suggesting a possible role for antisense transcripts in pathogenesis (Lagier-Tourenne et al., 2013).

The observation that expression of certain C9ORF72 transcripts are downregulated upon repeat expansion due to epigenetic modifications suggests that C9ORF72 haploinsufficiency is a plausible disease mechanism (Belzil et al., 2013). Homology modeling predicts that C9ORF72 is related to DENN proteins belonging to the Rab GEF protein family, which are important for vesicular trafficking (Levine et al., 2013). The physiological function of the protein, and its distribution across tissues, is the focus of current studies and model systems are also being developed to address this question. LacZ reporter mice reveal C9ORF72 expression in neuronal regions sensitive to neurodegeneration, (ventral horn of the spinal cord, cortical layers, hippocampus) but absent in non-neuronal cells like microglia and astrocytes (Suzuki et al., 2013). For C9ORF72 haploinsufficiency models, knockdown of a zebrafish C9ORF72 orthologue by antisense morpholinos and C. elegans null mutations of Alfa-1, a C9ORF72 orthologue, have been developed (Ciura et al., 2013; Therrien et al., 2013). These models exhibit neurodegeneration phenotypes such as age-dependent motility defects, paralysis and motor neuron axonal degeneration. Contrary to the prediction of the haploinsufficiency model, reduction of C9ORF72 RNA levels in mice by ASOs specifically in the CNS is well tolerated with no significant pathological/behavioral changes (Lagier-Tourenne et al., 2013). However, the effect of sense RNA reduction on the corresponding C9ORF72 protein levels is not clear in this model because of the lack of specific anti-C9ORF72 antibodies. Cumulatively, these results support the need for additional C9ORF72 loss-of-function models to clarify the contribution of haploinsufficiency to C9ORF72 ALS/FTD.

Another pathogenic mechanism that has been linked to C9ORF72 ALS/FTD mutations is RAN translation of GGGGCC^exp/ CCCCGG^exp RNAs and the production of six different dipeptide proteins that form intranuclear and cytoplasmic aggregates in various brain regions (Ash et al., 2013; Mori et al., 2013b; Zu et al., 2013). The majority of cells expressing these RAN proteins do not contain C9ORF72 sense/antisense RNA foci indicating a mutually exclusive mechanism in which the transcribed repeats are either sequestered into foci or are exported to the cytoplasm for RAN translation (Gendron et al., 2013). This observation suggests that some cell types may preferentially express factors that promote the formation and nuclear retention of toxic RNAs while other cells, which are deficient in these factors, are permissive for nuclear export and RAN translation. However, the detailed relationship between RNA foci and RAN translation requires additional studies in cell and animal models of C9ORF72 ALS/FTD disease.

6.2. RNA gain-of-function models for C9ORF72 ALS/FTD

A Drosophila model expressing (GGGGCC)₃₀ repeats flanking EGFP shows severely disrupted eye morphology and locomotor defects compared to control flies expressing (GGGGCC)₃ – EGFP (Xu et al., 2013). One limitation of this model is that this repeat size may be in the normal range so the effects of expanded repeats are still unknown. Efforts to generate vertebrate zebrafish and mouse models expressing expanded GGGGCC and CCCCGG repeats are ongoing.

A number of RNA binding proteins have been identified that interact with rGGGGCC^exp RNA. Examples include: 1) hnRNP A3, a protein involved in cytoplasmic RNA trafficking; 2) Purα, a transcriptional activator; 3) ADARB2, a protein with homology to adenosine deaminases involved in adenosine to inosine RNA editing; 4) hnRNP H, a protein involved in pre-mRNA processing (Gendron et al., 2014). Both hnRNP A3 and Purα bind to GGGGCC RNA in vitro, but colocalization with nuclear RNA foci and loss-of-function is not observed in patients. Interestingly, hnRNP A3 is a component of the p62 positive cytoplasmic inclusions observed in patient brain and Purα forms intranuclear inclusions and rescues the neurodegenerative eye phenotype in (rGGGGCC)₃₀-EGFP expressing flies implicating them in pathogenesis (Mori et al., 2013a; Xu et al., 2013). A proteome array hybridized with (rGGGGCC)_6.5 RNA identified ADARB2, a protein with a possible regulatory role in RNA editing. ADARB2 colocalizes with C9ORF72 sense RNA foci in both patient and iPS cell lines and is important for RNA foci formation in iPS cells (Donnelly et al., 2013). However, the extent of ADARB2 sequestration and the downstream pathways possibly affected by sequestration of this protein requires further study. In addition, hnRNP H, a protein originally classified as a poly(G) binding protein (Swanson and Dreyfuss, 1988), has been shown to interact with rGGGGCC^exp RNA in vitro and in patient brain sections (Lee et al., 2013b). In contrast, the colocalization of hnRNP H and rGGGGCC^exp RNA is not observed in patient-derived iPS cells (Almeida et al., 2013). Further studies examining hnRNP H specific splicing alterations in patient tissues, either by microarray or RNA-seq, will be important to validate the potential effects of hnRNP H sequestration. Additionally, the presence of antisense CCCCGG^exp RNA foci in ALS/FTD cells spotlights the need to identify proteins sequestered by C9ORF72 antisense RNA.

7. Concluding remarks and therapeutic perspective

While the mechanisms underlying non-coding repeat expansion diseases are complex, the commonalities that exist between these disorders hint that similar molecular events are involved in pathogenesis. Furthermore, the discoveries of RNA-mediated toxicity, bidirectional transcription across repeat expansions, RAN translation and miR dysregulation serve as a foundation to delineate the relative contribution of each of these mechanisms to disease pathogenesis. For some expansion diseases, it is not clear how titration of potentially sequestered proteins leads to the disease and loss-of-function animal models for these factors must be generated to validate their pathogenic roles. In diseases such as DM, where animal models have validated the sequestration hypothesis, there is widespread splicing misregulation but the potential pathogenic effects of the majority of these splicing alternations remain unknown.

An appreciation of the pathogenic complexities of microsatellite expansion disorders has informed therapeutic development but also posed additional challenges. For example, ASO approaches have been used to correct specific mis-splicing events, such as reversal of Clcn1 mis-splicing and myotonia in a DM mouse model (Wheeler et al., 2007), but this type of targeted therapy cannot address the hundreds, or thousands, of disrupted RNA processing events that may occur following sequestration of RNA processing factors by repeat expansion RNAs. Modified ASO gapmers, composed of nuclease-resistant RNA flanking DNA designed to trigger RNase H-mediated degradation, have proven to be effective in knocking down wild-type and mutant transcripts in both C9ORF72 ALS/FTD fibroblasts and DM1 mouse models (Lagier-Tourenne et al., 2013; Wheeler et al., 2012). However, a caveat with this approach is that targeting both sense and antisense transcripts may be required for some diseases, such as ALS/FTD where current evidence suggests that bidirectional transcription plays an important pathogenic role. Haploinsufficiency is another concern for the ASO gapmer approach, since this strategy results in reduced expression of both normal and mutant alleles although nuclear-retained mutant RNAs are more susceptible to RNase H-mediated degradation (Wheeler et al., 2012).

Gene therapy approaches have been designed to replace the RNA-binding factors sequestered by toxic expansion RNAs, including Mbnl1 overexpression in a CTG^exp DM mouse model to rescue DM-associated mis-splicing and Purα overexpression to reverse disease phenotypes in FXTAS and C9ORF72 ALS/FTD in Drosophila and mouse neuronal cell models (Jin et al., 2007; Kanadia et al., 2006; Xu et al., 2013). However, effective gene delivery and expression in multiple tissues poses a major technical challenge. Another therapeutic strategy is to deploy small molecule inhibitors of specific protein-RNA interactions to release endogenous RNA binding proteins from toxic repeat RNAs (Arambula et al., 2009; Jahromi et al., 2013; Warf et al., 2009). However, a potential problem with this approach is the possibility that altering these protein-RNA interactions may result in more deleterious effects, possibly due to enhanced RAN translation (Childs-Disney et al., 2013).

Many fundamental questions remain. An intriguing feature of these microsatellite expansion diseases is that many of them are neurological disorders despite the fact that some mutations occur in ubiquitously expressed genes. What makes certain tissues more susceptible to overt pathology? Why do repeat expansion disorders often manifest later in life and what key events trigger symptom onset? What are the roles of penetrance and genetic modifiers in these diseases? Answering these, and additional questions, will provide essential insights that should lead to the development of new therapeutic modalities for these diseases.