Repeat-associated non-AUG (RAN) translation and other molecular mechanisms in Fragile X Tremor Ataxia Syndrome.

Fragile X-associated tremor/ataxia syndrome (FXTAS) is a late-onset inherited neurodegenerative disorder characterized by progressive intention tremor, gait ataxia and dementia associated with mild brain atrophy. The cause of FXTAS is a premutation expansion, of 55 to 200 CGG repeats localized within the 5'UTR of FMR1. These repeats are transcribed in the sense and antisense directions into mutants RNAs, which have increased expression in FXTAS. Furthermore, CGG sense and CCG antisense expanded repeats are translated into novel proteins despite their localization in putatively non-coding regions of the transcript. Here we focus on two proposed disease mechanisms for FXTAS: 1) RNA gain-of-function, whereby the mutant RNAs bind specific proteins and preclude their normal functions, and 2) repeat-associated non-AUG (RAN) translation, whereby translation through the CGG or CCG repeats leads to the production of toxic homopolypeptides, which in turn interfere with a variety of cellular functions. Here, we analyze the data generated to date on both of these potential molecular mechanisms and lay out a path forward for determining which factors drive FXTAS pathogenicity. Published by Elsevier B.V.

Introduction

Fragile X-associated tremor/ataxia syndrome (FXTAS) is an inherited neurodegenerative disorder caused by a “premutation” expansion of 55 to ∼200 CGG repeats, located on the X chromosome within the 5’untranslated region (5′UTR) of Fragile X Mental Retardation 1 (FMR1) (Hagerman et al., 2001). The prevalence of the CGG premutation varies among studies and populations, but is estimated to range between 1 in 110 to 250 in females and 1 in 260 to 800 in males (Tassone et al., 2012, Hunter et al., 2014 Jul). However, due to incomplete penetrance, it is estimated that 1 in ∼5000 to ∼10,000 men older than 50 years will develop FXTAS (Jacquemont et al., 2004, Tassone et al., 2012). Female carriers of the FMR1 CGG premutation have a lower risk of developing FXTAS, likely due to the random X-inactivation of the premutation allele.

Principle clinical characteristics of FXTAS include progressive intention tremor, gait ataxia, dementia (with prominent executive dysfunction), parkinsonism, autonomic dysfunction and peripheral neuropathy (Jacquemont et al., 2003). Neuropathological features of FXTAS include mild brain atrophy, Purkinje cell dropout and white matter lesions, especially in the splenium of the corpus callosum and the middle cerebellar peduncles. A key pathological characteristic is the presence of large eosinophilic nuclear inclusions in both neurons and astrocytes of individuals with FXTAS (Greco et al., 2002, Greco et al., 2006). These inclusions are present throughout the brain and are most numerous in the hippocampus (Greco et al., 2002, Greco et al., 2006, Ariza et al., 2016 Oct). Inclusions are also observed in tissues outside of the CNS (Greco et al., 2007, Hunsaker et al., 2011, Buijsen et al., 2014). These inclusions contain ubiquitin and various chaperone proteins such as HSP27, HSP70 and αB-crystallin, but do not contain tau proteins, α-synuclein and polyglutamine (Greco et al., 2002, Iwahashi et al., 2006).

At the molecular level, FMR1 CGG premutation carriers have a 2 to 8-fold increase in FMR1 mRNA compared to control individuals (Tassone et al., 2000, Kenneson et al., 2001, Tassone et al., 2007). This is in strict contrast to Fragile X syndrome (FXS), a neurodevelopmental disease characterized by intellectual disability and autism, where expansions of over 200 CGG repeats lead to hypermethylation and silencing of the FMR1 promoter (Bell et al., 1991). Importantly, expression of mutant RNAs containing expanded CGG repeats is toxic in cell and animal models (Willemsen et al., 2003, Jin et al., 2003, Arocena et al., 2005, Hashem et al., 2009, Entezam et al., 2007, Hukema et al., 2014, Hukema et al., 2015). These results suggest that the pathogenicity of the FMR1 premutation arises from expression of a mutant RNA containing expanded CGG repeats (Hagerman and Hagerman, 2004). This model is supported by the observation that FXTAS is not found in FXS individuals with fully silenced FMR1 alleles. However, how expression of an RNA containing 55 to 200 CGG repeats causes neuronal dysfunctions and cell death is not yet fully understood. Here we discuss two non-exclusive pathogenic mechanisms. The RNA gain-of-function model (Fig. 1), proposes that CGG repeats are toxic through binding and sequestration of specific RNA binding proteins (Iwahashi et al., 2006, Sofola et al., 2007, Jin et al., 2007, Sellier et al., 2013). In contrast, the repeat-associated non-AUG (RAN) translation model (Fig. 2) proposes that expanded CGG repeats cause pathogenicity through their translation into toxic proteins (Todd et al., 2013, Oh et al., 2015, Kearse et al., 2016, Krans et al., 2016, Sellier et al., 2017). This review will focus on the current literature that supports each model while defining the work needed to determine which process, if either, is the proximal driver of FXTAS disease pathogenesis.

Fig. 1

FMR1 RNA repeat-mediated toxicity in FXTAS. (A) CGG repeat RNA creates a repetitive motif that is bound directly by hnRNP A2/B1, DGCR8, and Purα. This could potentially lower the amount of these proteins available to perform normal functions. (B) CGG repeat RNA may also indirectly titrate the abundance of other proteins through interactions with these three repeat binding partners. (C) Both CGG repeat RNAs and RNA binding proteins may be capable of phase separation and RNA gelation. (D) Potential functional consequences of CGG repeat RNA interactions include sequestration of proteins involved in splicing (hnrNP A2), mRNA transport (hnrNP A2, Purα), miRNA processing (DGCR8 and Drosha), and chromatin maintenance (HP1). Each of these pathways could contribute to generation of unprocessed or mislocalized mRNAs and miRNAs, increased transposon expression, stress granule formation, and global translational blockade.

Fig. 2

Repeat associated non-AUG translation and toxicity in FXTAS. (A) RAN translation of CGG repeats within the 5′UTR of FMR1 occurs in a m7G cap and ribosomal scanning dependent manner. Initiation occurs 5′ to the repeat at near cognate codons in the +0 (blue) and +1 (green) reading frames, and within the repeat in the +2 reading frame (yellow), to produce three homopolypeptides: FMRpolyR (blue), FMRpolyG (green), and FMRpolyA (yellow). Abundance of each product is depicted based on data from differential translation rates (Kearse et al., 2016, Krans et al., 2016). (B) RAN translation from ASFMR1 CCG repeat RNA produces three different homopolypeptides: ASFMRpolyP (purple), ASFMRpolyR (blue), and ASFMRpolyA (yellow). Translation of ASFMRpolyP also occurs potentially through initiation at an AUG codon (purple). The initiation sites for the +1 (blue) and +2 (yellow) reading frames have not yet been determined. (C) RAN translation products elicit toxicity. FMRpolyG (green). FMRpolyG may sequester LAP2β within these inclusions, disrupting nuclear architecture. FMRpolyG and FMRPolyA expression elicit ubiquitin proteasome system impairment. Either CGG repeat mRNA or CGG RAN derived polypeptides promote stress granule formation impairing global protein translation.

The RNA gain-of-function model

The repeat RNA gain-of-function model was first proposed in myotonic dystrophy type 1 (DM1), a neuromuscular disease caused by large expansions of a CTG repeat located in the 3′UTR of DMPK. Mutant DMPK mRNAs accumulate in nuclear RNA foci that co-localize specifically with Muscleblind-like (MBNL) proteins (Mankodi et al., 2001, Fardaei et al., 2002). MBNL1, MBNL2 and MBNL3 are RNA binding proteins that recognize YGC RNA motifs and regulate pre-mRNA alternative splicing, mRNA localization and stability (Ho et al., 2004, Wang et al., 2012 Aug 17, Batra et al., 2014 Oct 23). MBNL proteins bind to CUG repeat RNA expansions, which sequesters them away from their normal mRNA targets (Goodwin et al., 2015). Consistent with this model, reduction of free MBNL1 and MBNL2 in cells, and knockout of Mbnl1 and/or Mbnl2 in mice, reproduces splicing alterations and key features of myotonic dystrophies (Savkur et al., 2001, Mankodi et al., 2002, Charlet-Berguerand et al., 2002, Fugier et al., 2011, Tang et al., 2012, Freyermuth et al., 2016 Apr, Kanadia et al., 2003, Charizanis et al., 2012, Lee et al., 2013, Nakamori et al., 2013 Dec, Thomas et al., 2017). Conversely, AAV-mediated overexpression of MBNL1 corrects myotonia in mice expressing expanded CUG repeats (Kanadia et al., 2006). Finally, increases in CUG repeat length lead to a greater titration of MBNL proteins, resulting in more RNA metabolism alterations that correlate with increased disease severity in individuals with myotonic dystrophy (Wagner et al., 2016 Sep 28, Thomas et al., 2017). These results highlight some key features that define an RNA gain-of-function disease, namely (1) the accumulation of expanded microsatellite repeats within RNA foci that co-localize with specific RNA binding proteins, (2) the sequestration of these RNA proteins away from their normal mRNA targets as evidenced by specific changes in mRNA metabolism regulated by these proteins, (3) the functional depletion of these RNA binding proteins by expanded microsatellite repeats leads to molecular and phenotypic features similar to the knockout or knockdown of these RNA binding proteins in model cell or organisms, and (4) overexpression of these RNA binding proteins corrects molecular and phenotypic features caused by the expression of expanded microsatellite repeats in model cells and organisms.

FMR1 mRNA is found within nuclear inclusions in brain sections of individuals with FXTAS and multiple RNA binding proteins associate with expanded CGG repeats (Tassone et al., 2004b, Iwahashi et al., 2006, Sofola et al., 2007, Jin et al., 2007, Sellier et al., 2010, Qurashi et al., 2011, Sellier et al., 2013). Thus, sequestration of specific RNA binding proteins by premutation FMR1 RNA has been proposed as a mechanism by which CGG repeats promote neurotoxicity in FXTAS (Jin et al., 2003, Hagerman and Hagerman, 2004, Tassone et al., 2004a).

A key protein involved in RNA metabolism and found to associate with CGG repeat RNA is heterogeneous nuclear ribonucleoprotein (hnRNP) A2/B1 (Iwahashi et al., 2006, Jin et al., 2007, Sofola et al., 2007, Bekenstein and Soreq, 2013 Sep, Jean-Philippe et al., 2013). Of interest, mutations that enhance hnRNP A2/B1 aggregation cause the cellular degenerative syndrome, Multisysem Proteinopathy (MSP), which is clinically and pathologically distinct from FXTAS (Kim et al., 2013). The strongest evidence to date implicating hnRNP A2/B1 in FXTAS pathogenesis comes from work in Drosophila. Overexpression of expanded CGG repeats outside of their native sequence context is sufficient to elicit neurodegeneration and a rough eye phenotype that is dosage and repeat length dependent (Jin et al., 2003). Overexpression of either human hnRNP A2/B1 or either of its two Drosophila orthologs suppresses CGG repeat toxicity (Sofola et al., 2007, He et al., 2014). In support of a deleterious role in cytoplasmic mRNA metabolism, sequestration of hnRNP A2/B1 by CGG RNA repeats in neurons impairs dendritic delivery of known hnRNP A2/B1 targeted mRNAs (Muslimov et al., 2011). hnRNP A2/B1 also recruits in trans other proteins to CGG repeat RNA, thus potentially sequestering these proteins away of their normal functions. Among these hnRNP A2/B1 binding proteins is the CUG binding protein (CUGBP1, also known as CELF1) (Sofola et al., 2007), involved in pre-mRNA alternative splicing, mRNA stability, and translation, whose expression and phosphorylation is elevated in cardiac samples of individuals with DM1 (reviewed in Barreau et al., 2006, Timchenko et al., 1999, Savkur et al., 2001, Kuyumcu-Martinez et al., 2007). Interestingly, overexpression of CUGBP1 rescues CGG repeat elicited neurodegeneration in Drosophila (Sofola et al., 2007). hnRNP A2/B1 also binds retrotransposon DNA and recruits heterochromatin protein 1 (HP1) to silence retrotransposon expression in Drosophila. Expanded CGG repeats sequester hnRNP A2/B1, preventing its binding to retrotransposons and subsequent HP1 recruitment, causing increased retrotransposon expression in a fly model of FXTAS (Tan et al., 2012a). Reducing this retrotransposon expression via activation of other pathways suppresses CGG repeat toxicity, suggesting that retrotransposon activation contributes in part to CGG repeat-induced pathology in Drosophila (Tan et al., 2012a). Thus, the ability of CGG repeats to bind hnRNP A2/B1 could promote neurotoxicity by both directly interfering with hnRNP A2/B1 function, and indirectly by sequestering hnRNP A2/B1 binding partners.

Other hnRNP A2/B1 binding partners can alleviate neurotoxic effects of CGG repeats by preventing hnRNP A2/B1 sequestration. The RNA and DNA binding protein TDP-43 is one such protein that is of special interest for neurodegenerative diseases as protein inclusions of TDP-43 in neurons are a histopathological hallmark of amyotrophic lateral sclerosis and some forms of frontotemporal dementia (Neumann et al., 2006). Two independent studies found that overexpression of TDP-43 partially rescues the neurodegeneration induced by expanded CGG repeats in a Drosophila model of FXTAS, despite the fact that expression of this protein in isolation is toxic itself (Galloway et al., 2014, He et al., 2014). TDP-43 is not found within CGG repeat RNA aggregates and does not bind the RNA directly; rather, TDP-43 titrates the hnRNP A2/B1 homologue proteins Hrb87F and Hrb98DE away from the repeats, altering their distribution and helping to restore their function (He et al., 2014). While TDP-43 dependent rescue requires these hnRNP A2/B1 orthologs, overexpression of hnRNP A2/B1 itself is capable of suppressing CGG repeat toxicity even when expression of Tbph, the fly homologue of TDP-43, is suppressed (He et al., 2014). It is also intriguing to note that TDP-43 expression is down-regulated in Purkinje cells of a mouse model expressing expanded CGG repeats, highlighting the potential importance of this protein in FXTAS (Galloway et al., 2014).

CGG repeat RNA also binds the purine-rich binding protein α (Purα), a cytoplasmic RNA-binding protein that regulates a number of RNA functions (e.g. RNA transport and translation) (Jin et al., 2007). Similar to the example above, overexpression of Purα rescues neurodegeneration in a Drosophila model of FXTAS (Jin et al., 2007). Purα localizes within cytoplasmic inclusions in Drosophila expressing 90 CGG repeats, but has not been consistently identified within the ubiquitin-positive intranuclear inclusions observed in brain sections of FXTAS patients or mouse models expressing expanded CGG repeats (Jin et al., 2007, Iwahashi et al., 2006, Sellier et al., 2010). Purα recruits Rm62 (also known as Dmp68), the Drosophila ortholog of the RNA helicase P68/DDX5 (Qurashi et al., 2011). Rm62 expression is decreased in FXTAS fly models while its mRNA levels are unchanged, suggesting that expanded CGG repeats influence Rm62 post-transcriptional regulation (Qurashi et al., 2011). Furthermore, flies expressing CGG RNA have increased nuclear retention of mRNAs normally exported by Rm62, including the chaperone Hsp70. Overexpression of Rm62 rescues CGG repeat-elicited neurodegeneration in flies, suggesting that dysfunctional nuclear export could also contribute to FXTAS pathogenesis (Qurashi et al., 2011).

In addition to the examples above, mass spectrometry has identified other proteins, including DiGeorge Syndrome Critical Region Gene 8 (DGCR8), that bind directly to CGG repeat RNA (Sellier et al., 2013). DGCR8 forms a protein complex with the ribonuclease DROSHA, which processes pri-miRNAs into pre-miRNAs, an essential step in miRNA biogenesis. CGG RNA aggregates reduce processing and expression of miRNAs in neuronal cells by partially sequestering DROSHA-DGCR8 complexes (Sellier et al., 2013). Indeed, miRNA expression changes have been observed in both FXTAS patients and in Drosophila models of FXTAS (Alvarez-Mora et al., 2013, Tan et al., 2012b). Furthermore, DGCR8 recruits SAM68 in trans to the expanded CGG RNA. SAM68 is an RNA binding protein that regulates pre-mRNA alternative splicing. Consistent with some degree of functional sequestration, CGG repeat overexpression leads to alterations in the splicing of SAM68 target transcripts. However, while overexpression of DGCR8 rescues neuronal cell death induced by expression of CGG repeats, overexpression of SAM68 failed to rescue this phenotype, suggesting that miRNA processing but not dysregulated SAM68-mediated splicing could be contributing to FXTAS pathogenesis (Sellier et al., 2010, Sellier et al., 2013).

Recently, new evidence has emerged that sequestration of RNA binding proteins by GC-rich repeats might be further enhanced by “RNA gelation.” Neurodegenerative causing repeats, GGGGCC, CAG and CUG have all been shown to form gels in vitro, and nuclear and cytoplasmic foci in cells, in a manner dependent on repeat size and RNA abundance (Jain and Vale, 2017, Fay et al., 2017). Similar to CGG-repeats, some of the nuclear foci co-localize with hnRNP A2/B1, suggesting that RNA gelation might be a precursor for sequestration of RNA binding proteins (Jain et al., 2017). Duplex stability and repeat secondary also influence RNA gelation dynamics. GGGGCC repeats, which have strong structure formation and can fold into G-quadruplexes display more stable foci formation while CAG repeats, with weaker structure formation are dynamic (reviewed in Ciesiolka et al., 2017, Jain and Vale, 2017, Fay et al., 2017). This secondary structure might further influence function. G-quadruplex forming sequences expressed in cells, were shown to form RNA granules that co-sedimented with classic stress granule markers G3BP1 and eIF3B, supporting an additional potential role for RNA gelation in stress granule assembly (Fay et al., 2017). While CGG-repeats have not yet been demonstrated to form gels in vitro, it would be consistent with the observation that CGG-repeat containing FMR1 mRNA has been found in some intranuclear inclusions in FXTAS brains, and in large nuclear foci in both cell and Drosophila model systems (Tassone et al., 2004b, Sellier et al., 2010, He et al., 2014). Furthermore, recent observations of increased stress granule formation in cells expressing reporters downstream of expanded CGG repeats could be indicative of a role for CGG repeats in stress granule assembly (reviewed below; Green et al., 2017). For now, it remains to be tested whether gelation occurs with expanded CGG RNA repeats, and whether it is an important mechanism contributing to FXTAS pathogenesis.

Overall, these observations suggest that premutation FMR1 mRNA could be pathogenic by binding and sequestering specific RNA binding proteins, ultimately reducing their normal functions (Fig. 1). Furthermore, it is possible that the partial sequestration of several diverse RNA binding proteins could have a cumulative effect and cause synergistic toxicity. However, the data published to date are not yet sufficient to definitively establish this mechanism as causative in FXTAS. First, it remains to be determined whether overexpression of hnRNP A2/B1, CUGBP1, TDP-43, DROSHA-DGCR8 or Purα specifically corrects any phenotype in mouse models expressing expanded CGG repeats. As groups embark on these studies, interpretation of the experiments require caution. For example, given that RNA binding proteins such as hnRNP A2/B1 may protectively destabilize the RNA structures formed by expanded CGG repeats, their overexpression might alter its metabolism, localization, or processing in a fashion that impedes its involvement in other pathogenic cascades rather than specifically replacing the sequestered factor (Khateb et al., 2004, Ofer et al., 2009). Indeed, overexpression of any repeat binding protein could potentially rescue all phenotypes nonspecifically by sequestering the repeat RNA away from other key cellular functions (Todd et al., 2014). Similarly, antisense oligonucleotides or small molecules targeting CGG repeat RNA may suppress their toxicity by precluding their translation or any other toxic gain-of-function mechanism induced by these repeats, as seen with other transcribed microsatellite expansions (Wheeler et al., 2012 Aug 2, Laurent et al., 2012, Disney et al., 2012, Tran et al., 2014, Pettersson et al., 2014, Todd et al., 2014, Yang et al., 2016, Ishiguro et al., 2017, Zu et al., 2017). Thus, functional rescue experiments teasing out these possibilities will be instrumental in determining the importance of these RNA binding proteins in FXTAS pathology.

Second, phenotypic similarities must be demonstrated between loss of function of sequestered proteins, overexpression of CGG repeats, and changes in FXTAS patient cells or brains. Studies to date provide little evidence that partial or complete loss of any of these proteins reproduces the specific molecular and clinical features of FXTAS. Furthermore, the CGG repeat expansion in FXTAS is relatively small (55 to 200 repeats) compared to the large CUG expansions (hundreds to thousands of repeats) in DM1. Whether a limited number of CGG repeats can sequester sufficient amounts of RNA binding proteins, particularly abundant ones, to inhibit their activity is not clear.

Third, whether the CGG repeat RNA is mis-localized in FXTAS remains unclear. FMR1 mRNAs are found primarily in the cytoplasm in both control and FXTAS lymphoblasts, yet co-localize in intranuclear inclusions in FXTAS brain tissue (Tassone et al., 2004b, Tassone et al., 2007). Of interest, the sequence surrounding the CGG repeat significantly influences its localization and thus its potential mode of pathogenesis. Transfection of plasmids containing expanded CGG repeats deprived of any natural FMR1 sequence leads to CGG repeat RNAs that are retained within the nucleus where they form large toxic RNA aggregates that sequester RNA binding proteins (Sellier et al., 2010). The ability of these CGG repeats, deprived of FMR1 sequence, to form nuclear RNA foci is cell type dependent, the reason for which is still unknown (Sellier et al., 2010). In contrast, expanded CGG repeats embedded in the natural 5′UTR sequence of FMR1 are largely exported into the cytoplasm (Sellier et al., 2017, Tassone et al., 2007). These results highlight the potential experimental bias of using artificial microsatellite repeat constructs separated from their natural sequence context. It also suggests that abortive RNAs generated through transcriptional stalling within the repeat may behave differently in terms of toxic mechanism (Haeusler et al., 2014). Thus, it is important to determine the behavior of the native FMR1 mRNA in patient derived neurons and model organisms to clarify this technical bias. In that aspect, FXTAS iPSC-derived cortical neurons and brain sections of mouse models expressing premutation FMR1 present no or very rare CGG RNA foci, despite having intranuclear inclusions (Hukema et al., 2015, Sellier et al., 2017). These negative results do not exclude the possibility that tissue type might dictate cellular localization of the premutation FMR1 mRNA, hence influencing its interacting partners and subsequent modes of toxicity. However, it remains to be determined whether expanded CGG repeats embedded within the natural FMR1 sequence can form nuclear RNA foci in specific cell types or tissues, and what factors influence this localization.

The RAN translation model

Recent reports indicate that the characteristic intranuclear inclusions found in FXTAS brain sections and in model organisms expressing premutation FMR1 form predominantly as a consequence of repeat associated non-AUG (RAN) translation of the CGG repeats (Todd et al., 2013, Sellier et al., 2017). This process was first identified at a CAG repeat expansion in the coding region of ATXN8, where the expanded repeats were translated in the absence of a canonical AUG start codon (Zu et al., 2011). Initiation occurred in all potential reading frames, producing three distinct homopolypeptides. RAN initiation in ATXN8 occurred within the repeat itself in at least one reading frame, although its exact initiation mechanism remains unclear (Zu et al., 2011).

A similar process was more recently observed at the CGG repeats within the 5′UTR of FMR1 (CGG RAN) (Todd, et al., 2013). RAN translation at this locus can generate three different homopolypeptides: polyarginine (FMRpolyR) from the +0 (CGG) reading frame, which would create an N-terminal extension on FMRP, polyglycine (FMRpolyG) from the +1 (GGC) reading frame, and polyalanine (FMRpolyA) from the +2 (GCG) reading frame. Two of these homopolypeptides, FMRpolyG and FMRpolyA are observed as aggregate prone proteins in cells transfected with constructs containing the human premutation FMR1 5′UTR fused to GFP in the three different frames. FMRpolyR was not observed in this system (Todd et al., 2013).

Using a series of nanoluciferase reporter constructs, Kearse and colleagues were able to investigate the mechanism underlying CGG RAN translation (Kearse et al., 2016). In a rabbit reticulocyte lysate system and in transfected cells and neurons, they found that CGG RAN translation, similar to AUG initiated translation, occurs in an eIF4E and m7G cap dependent mechanism. Moreover, CGG RAN translation required the RNA helicase eIF4A, which is critical for both cap binding and scanning of the small ribosomal subunit along the 5′UTR. CGG RAN translation is inefficient compared to translation initiating at a canonical AUG start codon, with the degree of efficiency dependent on the reading frame. The +1 reading frame (FMRpolyG) is the most efficient, with initiation primarily at two near-cognate codons (i.e. codons that differ from AUG by one nucleotide) upstream of the repeat: an ACG and a GUG, 36 nt and 12 nt 5′ of the repeat, respectively (Todd et al., 2013, Kearse et al., 2016, Sellier et al., 2017). Compared to AUG initiated translation of the same protein, RAN translation of FMRpolyG is approximately 40% as efficient. The +2 reading frame (FMRpolyA) is translated at ∼30% of the +1 reading frame. Initiation in this reading frame likely occurs within the repeat itself, as placement of a stop codon immediately above the repeat has no impact on its translation (Todd et al., 2013, Kearse et al., 2016). When no repeats are present, initiation in the +0 frame (FMRpolyR) can occur at an ACG codon located 60 nt upstream of the repeat insertion site (Kearse, et al., 2016). However, as CGG repeat codons are added, there is a strong reduction in FMRpolyR expression: FMRpolyR production is less than 5% of FMRpolyG expression at 25 repeats and less than 1% of FMRpolyG expression at 100 repeats (Todd et al., 2013, Kearse et al., 2016). This suppressive effect is mediated at least in part at the level of translation elongation, as incorporation of an AUG start codon in a good Kozak sequence context has only a modest impact on product generation (Kearse et al., 2016). In that aspect, CGG repeats form a strong hairpin structure in vitro that impedes ribosome scanning (Feng et al., 1995, Primerano et al., 2002, Chen et al., 2003, Sobczak et al., 2003, Napierala et al., 2005, Broda et al., 2005).

For all three reading frames, repeat length correlates with peptide size (Kearse, et al., 2016). Moreover, FMRpolyG is prone to length-dependent aggregation as FMRpolyG with expansions above ∼50 glycine repeats forms nuclear aggregates in both cell and animal models, and has been observed in the intranuclear inclusions in FXTAS brain tissue (Todd et al., 2013, Buijsen et al., 2014, Sellier et al., 2017). FMRpolyA is likewise prone to length-dependent aggregation (Todd et al., 2013, Oh et al., 2015, Kearse et al., 2016). Interestingly, while increased repeat length correlates with increased translation efficiency of the +2 (FMRpolyA) reading frame, there is only a modest correlation between repeat length and translation of the +1 (FMRpolyG) reading frame. On a technical note, fusion of the CGG repeats to large tags, such as the GFP, in the +1 frame allows detection of FMRpolyG even with short repeat lengths (Todd et al., 2013, Sellier et al., 2017). This is important because FMRpolyG is not readily detected at normal repeat sizes in control human brain samples despite evidence suggesting it is produced at these repeat sizes in reporter systems. Intriguingly, the threshold for FMRpolyG aggregation and detection corresponds to the CGG repeat threshold above which individuals are at greater risk of developing FXTAS (Tassone et al., 2012, Todd et al., 2013, Sellier et al., 2017).

The FMR1 locus is also transcribed in the antisense direction from multiple promoters producing a series of transcripts (Ladd et al., 2007, Khalil et al., 2008, Pastori et al., 2013). One of these transcripts, FMR1 antisense RNA 1 (ASFMR1), spans the CCG repeat region (Ladd et al., 2007). Similar to FMR1 mRNA, there is a 2 to 3-fold upregulation of ASFMR1 expression in lymphoblasts of premutation carriers, while ASFMR1 is silenced in FXS patients with full-mutation (Ladd et al., 2007). The ASFMR1 transcript contains an open reading frame initiating at an AUG start codon and encoding a putative polyproline protein (ASFMRpolyP). Expanded CCG repeats continue to exhibit RAN translation in all three potential reading frames when this AUG is removed from reporter constructs, producing polyarginine (ASFMRpolyR) and polyalanine (ASFMRpolyA) peptides in addition to an N-terminally truncated ASFMRpolyP (Krans et al., 2016). Similar to CGG RAN translation, increases in CCG repeat length correlate with increased peptide size for all reading frames. In contrast to CGG translation of the +1 (FMRpolyG) reading frame, increased CCG repeat length correlates with increased translation in all three reading frames. Increased CCG repeat length causes ASFMRpolyR to relocalize from the cytoplasm to the nucleolus, and ASFMRpolyA to relocalize from the cytoplasm to the nucleus (Krans et al., 2016). Both ASFMRpolyP and ASFMRpolyA are present in intranuclear inclusions and perinuclear aggregates in FXTAS patient brain tissue (Krans et al., 2016).

Evidence for toxicity of FMRpolyG

Currently, the best characterized product of CGG RAN translation is FMRpolyG. This small protein comprises a 12 amino acid N-terminus, a central glycine stretch with length equivalent to the number of CGG repeats, and a 42 amino acid C-terminus. FMRpolyG is present in intranuclear inclusions in patient tissue (i.e. brain tissue of individuals with FXTAS, iPSC-derived neurons from FXTAS patients, and ovarian tissue of an individual with Fragile X-associated primary ovarian insufficiency (FXPOI)) as well as in model organisms (i.e. Drosophila, mouse, and human cell culture) expressing expanded CGG repeats embedded within the human FMR1 sequence (Todd et al., 2013, Buijsen et al., 2014, Buijsen et al., 2016, Hukema et al., 2015, Sellier et al., 2017). A series of studies first conducted in Drosophila, and later in transgenic mice, suggest that FMRpolyG production is, at a minimum, necessary for both inclusion generation and CGG repeat-mediated toxicity in these model organisms (Todd et al., 2013, Oh et al., 2015, Sellier et al., 2017). In both mammalian cells and in Drosophila, placing an AUG codon in the FMR1 5′UTR upstream of the CGG repeat in the +1 (FMRpolyG) frame enhances inclusion formation, cell death in cell culture, and retinal degeneration in Drosophila (Todd et al., 2013). In contrast, placing a stop codon just 5′ to the repeat but 3′ to the FMRpolyG near-cognate initiation codons precludes toxicity, with loss of inclusion formation and correction to normal Drosophila viability (Todd et al., 2013). Similarly, moving the repeat to the 3′UTR, where RAN translation is markedly reduced, also precludes its toxicity in Drosophila. In mice, knockin of the expanded CGG repeat with a natural mouse stop codon upstream of the repeat and in the +1 FMRpolyG reading frame results in a significant reduction in inclusions that correlates with lower expression of FMRpolyG (Todd et al., 2013).

More recently, evaluation of transgenic mice expressing expanded CGG repeats with or without FMRpolyG has provided additional support for a direct role of RAN translation in FXTAS pathogenesis (Sellier et al., 2017). Two transgenic mouse models were generated. In both, the repeat expansion was placed upstream of the GFP in the +1 FMRpolyG reading frame. In one model however, the 5′ region, where FMRpolyG (and FMRpolyR) initiation occurs, was deleted. Transgenic mice expressing the full 5′UTR of human FMR1 with 100 CGG repeats exhibited intranuclear brain inclusions that co-stained for FMRpolyG, early death, Purkinje cell loss and locomotor abnormalities (Sellier et al., 2017). However, mice lacking the 5′ region above the repeat exhibited neither formation of intranuclear inclusions nor signs of toxicity despite comparable CGG mRNA expression levels (Sellier et al., 2017). These effects were seen with both ubiquitous expression as well as with expression confined to the nervous system.

A number of factors appear to influence the toxicity of FMRpolyG across model systems. First, the 42 amino acid C-terminal region of FMRpolyG appears to modulate the toxicity of this protein. Expressing the full length FMRpolyG or just the region C-terminal to the repeat induces cell death in mouse cortical neurons (Sellier et al., 2017). Furthermore, expressing the polyglycine stretch alone, without its surrounding native sequence, has a milder phenotype in neuronal cell cultures and in Drosophila, despite aggregation of the polyglycine in aggregates (Sellier et al., 2017). These results suggest that the C-terminal region of FMRpolyG might be critical to certain pathogenic cascades. Consistent with this model, the C-terminal region of FMRpolyG binds to a series of proteins, including the lamin associated protein Lap2β. Overexpression of either FMRpolyG or its C-terminal region is sufficient to disrupt the nuclear lamina architecture and elicit toxicity. Conversely, overexpression of Lap2β is capable of suppressing FMRpolyG toxicity in neuronal cell cultures (Sellier et al., 2017). Along with evidence for accumulation of other lamin-associated proteins, such as lamin A/C, in FXTAS inclusions and abnormal nuclear lamina architecture in the disease state (Arocena et al., 2005, Iwahashi et al., 2006), these findings suggest a role for RAN translation-induced nuclear architecture disruption in FXTAS pathogenesis. These findings are reminiscent of other microsatellite expansion diseases in which expanded repeats are RAN translated into pathogenic proteins that disrupt nuclear architecture and nucleocytoplasmic transport (Zhang et al., 2015, Zhang et al., 2016, Jovičić et al., 2015, Freibaum et al., 2015, Grima et al., 2017, Gasset-Rosa et al., 2017).

A second mechanism by which FMRpolyG could elicit toxicity is through alterations of the protein quality control pathways. Proteasomal components such as ubiquitin, heat shock proteins, and αβ-crystallin accumulate in intranuclear inclusions in FXTAS. (Iwahashi et al., 2006). In Drosophila, CGG repeat-associated toxicity is enhanced by genetic impairment of the ubiquitin proteasome system (UPS), but not of the autophagy pathway (Todd et al., 2010, Oh et al., 2015). Consistent with this, overexpression of chaperone proteins such as HSP70, reduces CGG-mediated toxicity (Jin et al., 2003, Oh et al., 2015). This modulation by UPS impairment only occurs when FMRpolyG is generated and there is no genetic interaction associated with expression of the CGG-repeat RNA alone (Oh et al., 2015). Moreover, expression of FMRpolyG impairs the UPS in transfected cells, with enhanced impairment elicited with greater FMRpolyG production. Overall, these data suggest that, as with other proteins prone to aggregation, modulation of the protein degradation pathways may be of therapeutic interest in FXTAS.

Potential toxic effects of other RAN-derived proteins in FXTAS

The role of other RAN translation products in CGG repeat associated toxicity is less well studied. FMRpolyA is expressed at lower levels in reporter assays compared to FMRpolyG (Todd et al., 2013, Kearse et al., 2016). To date, there is no published evidence that FMRpolyA is present in inclusions in model systems or in pathology samples from FXTAS patients (Sellier et al., 2017). However, ASFMRpolyA generated from the antisense CCG repeat is observed within nuclear aggregates in patient brains by immunohistochemistry (Krans et al., 2016). Importantly, expansions of polyalanine are aggregation-prone and are involved in various human genetic diseases, suggesting that FMRpolyA and/or ASFMRpolyA could have a toxic contribution in FXTAS. Expansions of polyalanine in transcription factors are responsible for at least 8 congenital disorders and an expansion of 7 to 17 alanines in PABPN1 causes the neurological disorder Oculopharyngeal Muscular dydrophy (OPMD) (Blondelle et al., 1997, Abu-Baker et al., 2003, Albrecht et al., 2004). As few as 7 alanines in a row leads to conformational changes of a peptide from a monomeric α-helix to an aggregated species formed primarily by β-sheets (Blondelle et al., 1997). Intriguingly, polyalanine aggregates were shown to recruit and alter proteasomal components including ubiquitin and chaperone proteins (Abu-Baker et al., 2003, Albrecht et al., 2004). Consistent with this, driving expression of FMRpolyA via a canonical AUG start codon in a good Kozak sequence induces cytoplasmic aggregation and enhances CGG repeat elicited impairment of the UPS pathway (Oh et al., 2015). Thus, FMRpolyA and ASFMRpolyA could contribute to FXTAS disease pathogenesis, but their relative stoichiometry and toxicity compared to FMRpolyG and other RAN proteins remain to be determined.

Expanded sense CGG and antisense CCG repeats are also potentially translated into two polyarginine-containing peptides, FMRpolyR and ASFMRpolyR (Todd et al., 2013, Krans et al., 2016). Interestingly, polyPR and polyGR, two arginine containing RAN proteins generated in another repeat disorder, C9orf72-associated Amyotrophic Lateral Sclerosis and Frontotemporal dementia (C9ALS/FTD), are toxic and interfere with a variety of cellular processes including pre-mRNA splicing and ribosomal biogenesis (Mizielinska et al., 2014, Kwon et al., 2014, Tao et al., 2015, Yin et al., 2017). Furthermore, polyGR has also recently been shown to induce oxidative stress by binding mitochondrial ribosomal proteins, and polyPR has been shown to inhibit the proteasome and also bind to the nuclear pore, causing disruptions in nuclear transport (Lopez-Gonzalez et al., 2016, Shi et al., 2017 Feb 14, Gupta et al., 2017). The positive charge of a polyarginine peptide gives it an unusual property that allows it to translocate through cell membranes and localize to the nucleolus (Futaki et al., 2001). For reasons not fully understood, polyarginine alone has also been shown to have some neuroprotective properties (Meloni et al., 2015). Whether polyarginines are toxic or require interruption by glycine or proline to drive toxicity is still unknown. Thus, it is important to consider whether FMRpolyR and/or ASFMRpolyR contribute to disease pathogenesis in FXTAS. Studies in cell-based systems using reporter assays suggest that FMRpolyR is not generated at detectable levels (Todd et al., 2013, Kearse et al., 2016, Sellier et al., 2017), although initiation can occur in the reading frame at an ACG near-cognate codon located upstream of the repeats when the expansion is absent or reduced (Kearse et al., 2016). Addition of normal or expanded repeats appears to suppress translation in this frame, which is different from what has been reported in all other CGG RAN reading frames. The reason for this effect remains unclear. RAN translation in this reading frame would generate an N-terminal extension on FMRP itself, as there is no stop codon between the repeat and the AUG start codon starting FMRP. Importantly, FMRP is not found in inclusions in FXTAS tissue and its expression does not modulate CGG repeat toxicity in Drosophila model systems (Jin et al., 2003, Iwahashi et al., 2006). In contrast, ASFMRpolyR is readily detected in reporter assays (Krans et al., 2016), so its role in FXTAS pathology warrants investigation.

Finally, the ASFMR1 transcript with expanded CCG repeats contains an open reading frame initiating at an AUG start codon and encoding a polyproline protein, ASFMRpolyP, observed in intranuclear inclusions in brain sections of individuals with FXTAS (Krans et al., 2016). Similar to polyarginine, polyproline has previously been shown to possess some protective properties. Polyprolines can protect against misfolded proteins, notably preventing aggregation of polyglutamine-containing huntingtin, and even aiding in oligomerization of neuroprotective proteins (Dehay and Bertolotti, 2006, Larson et al., 2014, Siwach et al., 2011 Aug). Polyproline induced toxicity so far has only been described in the context of the C9ALS/FTD proline-arginine dipeptide protein, polyPR. Thus, it remains to be determined whether ASFMRpolyP is playing a protective or toxic role in FXTAS. Similarly, it could be instructive to investigate whether RAN translated proteins from CGG and CCG repeats, notably FMRpolyG and ASFMRpolyP, interact with each other, and whether this interaction promotes neuronal toxicity.

Expression and contribution of the various RAN translation products in FXTAS

The extent to which individual CGG and CCG RAN translation products contribute to FXTAS pathogenesis is a key question. Parameters to consider are the absolute abundance of each of the mRNA species (e.g. FMR1 CGG repeat mRNA versus ASFMR1 CCG repeat mRNA), the relative translation efficiency in each reading frame from each mRNA, (which includes the near cognate initiation translation of FMRpolyG versus the AUG initiated translation of ASFMRpolyP), and the relative toxicity of each RAN translation protein at equimolar concentrations. While both FMR1 and ASFMR1 mRNAs are significantly increased in FXTAS patient tissue compared to controls, their relative abundance to each other has not been determined (Ladd et al., 2007). Similarly, although, relative translation efficiencies of each RAN product can be roughly estimated (Table 1), at this stage, the abundance of each RAN protein in patients are not yet accurately known.

Table 1

RAN translation at CGG and CCG repeats in FXTAS. Evaluation of both the mechanism and relative contributions of each RAN translation product to the pathogenicity of FXTAS is ongoing. For each potential RAN product, rough estimation of the relative abundance of each product in nLuciferase or GFP assays in transfected cells and pathological detection in patient derived tissues or model systems is shown. NT indicates not tested. ND indicates tested but not detected. X’s are included where mechanistic data are available to support a specific mode of initiation. (−) signs are included where data preclude use of such a mechanism. Question marks are included where data are either not available or inconclusive.

	Proposed initiation Mechanism			Observed in
	AUG codon	Near cognate codon	Within repeats	Transfected cells (nanoluciferase)	Transfected cells (GFP)	Patient brain	Neurons from patient iPS cells	Drosophila models	Mouse models (human FMR1)
FMRpolyG	–	X	X	+++	+++	++	+	+++	+++
FMRpolyA	–	–	X	++	±	NT	NT	NT	NT/ND
FMRpolyR	–	X	?	±	–	NT	NT	NT	NT
ASFMRpolyP	X	?	?	++	NT	+	NT	NT	NT
ASFMRpolyA	–	?	?	+	NT	+	NT	NT	NT
ASFMRpolyR	–	?	?	+	NT	NT	NT	NT	NT

RAN translation and the integrated stress response

A current key question in the microsatellite expansion field is whether RAN translation is a regulated mechanism, and if so, can it be modulated. Two recent studies suggest that cellular stress may indeed control RAN translation (Cheng et al., 2017, Green et al., 2017). The integrated stress response (ISR) is an adaptive pathway by which various types of cellular stress (e.g. viral infection, misfolded proteins, oxidative stress) activate one of four stress-specific kinases (i.e. PKR, PERK, HRI, and GCN2) which then phosphorylates the canonical translation initiation factor eIF2α. This phosphorylation inhibits eIF2B from recycling the GDP bound to eIF2α for GTP, preventing formation of new ternary complexes, effectively inhibiting global translation. Recently, it was shown that CGG RAN translation in the +1 (FMRpolyG) and +2 (FMRpolyA) reading frames is selectively upregulated in response to a variety of ISR activators (Green et al., 2017). Intriguingly, phosphorylated eIF2α is necessary for ISR-enhanced RAN translation, and phosphorylated eIF2α alone, in the absence of any stress, is sufficient to upregulate RAN translation while simultaneously blocking canonical mRNA translation. ISR-induced RAN translation of FMRpolyG is dependent on the presence of near cognate codons, as mutating these codons to AUG abolishes this effect. Importantly, expression of expanded CGG repeat containing constructs is sufficient to induce formation of stress granules in cells. This stress granule induction requires eIF2α phosphorylation and is associated with a reduction in global protein translation in repeat expressing cells and neurons (Green et al., 2017). These findings are consistent with previous work demonstrating that expression of either C9ORF72 GGGGCC repeats or DM1 CUG repeats elicits stress granule formation and global protein translational blockade (Huichalaf et al., 2010, Lee et al., 2016, Boeynaems et al., 2017). Together, these findings suggest a model whereby cellular stress selectively upregulates RAN translation, which in turn further activates cell stress pathways to create a feed-forward loop that potentially drives excess RAN translation and promotes neurodegeneration.

Other potential contributors to neurodegeneration in FXTAS

While not the focus of this review, it is important to note that other mechanisms have been proposed as contributors to disease pathogenesis in FXTAS (Hagerman and Hagerman, 2016). First, mitochondrial dysfunction (e.g. decreases in ATP production, oxygen uptake, and mitochondrial protein expression) and elevated levels of reactive oxygen species (ROS) have been reported in both cell and animal models of FXTAS as well as in fibroblasts and brain tissue of premutation carriers (Ross-Inta et al., 2010, Napoli et al., 2011, Napoli et al., 2016 Oct, Kaplan et al., 2012, Hukema et al., 2014, Song et al., 2016, Alvarez-Mora et al., 2016, Giulivi et al., 2016, Robin et al., 2017 Apr 21). Second, other studies support a model whereby transcribed CGG repeat expansions create R-loops, (i.e. RNA:DNA hybrids formed between the transcribed CGG RNA and the DNA) causing DNA breaks and subsequent recruitment and activation of the DNA damage checkpoint kinase, ATM, and the histone variant γH2AX, both of which are critical for DNA repair (Reddy et al., 2011, Loomis et al., 2014, Groh et al., 2014, Aguilera and García-Muse, 2012). In support of both of these models, increased ROS and hyperphosphorylation of ATM have been observed in FXTAS cell and animal models, and γH2AX was identified in nuclear inclusions in FXTAS patient tissue (Iwahashi et al., 2006, Hoem et al., 2011, Robin et al., 2017 Apr 21). These data suggest that mitochondrial dysfunction and hyper-activation of the DNA damage response may contribute to aspects of repeat toxicity. An open question is whether these alterations represent proximal events in disease pathogenesis or late-stage events triggered in sick cells (e.g. Is DNA damage is a direct consequence of R-loops or an indirect consequence of oxidative stress triggered by other pathogenic processes?) Thus, determining which aspect of CGG repeats—formation of R-loops, expression of CGG RNA, FMRpolyG or other RAN translated proteins, or some combination of these factors—induces mitochondrial damage and oxidative stress will be important for understanding the mechanism(s) driving pathogenicity in FXTAS.

RAN translation in Fragile X-associated primary ovarian insufficiency (FXPOI)

FXPOI is characterized by amenorrhea, estrogen deficiency in association with menopausal FSH levels and arrest of menstruation before the age of 40. FXPOI is observed in ∼ 20% of female carriers of a CGG premutation. Interestingly, increased expression of sense and antisense FMR1 transcripts are detected in granulosa cells of female carriers of the CGG premutation (Elizur et al., 2014, Elizur et al., 2016). Thus, it is proposed that FXPOI, like FXTAS, is caused by a toxic effect of the mutant FMR1 RNA. This hypothesis is supported by observations of ovulation defects in mouse models expressing expanded CGG repeats embedded in the human or mouse FMR1 gene (Lu et al., 2012, Hoffman et al., 2012, Conca Dioguardi et al., 2016 Jun, Buijsen et al., 2016). However, whether this mutant RNA is pathogenic through an RNA gain-of-function mechanism or through RAN translation is unknown. In support of RAN translation, FMRpolyG is detected in ovarian stromal cells of a woman with FXPOI, as well as in the ovarian stromal cells of a knockin mouse model expressing expanded CGG repeats in a human FMR1 context (Buijsen et al., 2016). However, defects in ovulation are also observed in a different CGG premutation knockin mouse that cannot produce FMRpolyG due to a natural murine stop codon located before the CGG repeats (Hoffman et al., 2012, Conca Dioguardi et al., 2016 Jun). In these mice the ovarian defects are associated with mitochondrial alterations (Conca Dioguardi et al., 2016), while in a YAC mouse model expressing expanded CGG repeats embedded within the human FMR1 gene, defects in ovulation are associated with alterations of the AKT and mTOR pathways (Lu et al., 2012). Thus, multiple mechanisms might contribute to pathogenicity in FXPOI.

Conclusions

Individuals expressing CGG repeat expansions beyond 55 repeats within the 5′UTR of the FMR1 gene are at risk of developing the late-onset neurodegenerative disease FXTAS. Whether expansions of microsatellite repeats are pathogenic through RNA gain-of-function mechanisms, RAN translational mechanisms, both, or neither is currently a key question in the field. In model systems that overexpress CGG repeats in mammalian neurons, Drosophila, and mice, the current data supports a direct role for RAN translation of FMRpolyG as an important pathogenic event (Todd et al., 2013, Sellier et al., 2017). However, these models are based largely on non-physiological overexpression of transgenes and do not eliminate potential contributions from the repeat RNA. Moreover, the sufficiency of FMRpolyG in disease pathogenesis is not yet established. Thus, novel crispr/CAS9 edited iPSC and animal models such as knockin mice or zebrafish are required to confirm and investigate further FMRpolyG pathogenicity. Evaluation of each potential RAN translated protein independent of its repeat context is also be needed to establish if all pathogenicity observed in patients can be explained by the FMRpolyG product alone or if other factors contribute. Similarly, whether endogenous levels of FMRpolyG are sufficient to trigger disease phenotypes and whether selective blockade of FMRpolyG synthesis can ameliorate FXTAS pathology in human warrants investigation. Finally, the contributions of other potential disease mechanisms such as toxicity from ASFMR1 RNA, mitochondrial dysfunctions, formation of R-loops and altered DNA damage response pathways need to be incorporated into a holistic disease model that integrates the potential impacts of different contributors to disease pathogenesis.

Despite these caveats, recent progress supports a tentative model where selective blockade of RAN translation at CGG repeats may act as a potential therapeutic target in FXTAS. This would require development of a potent and selective RAN translation inhibitor. Given the potential applications across a range of disease states, such a tool would have significant impact on both a research and clinical level, opening a route to a novel and effective treatment in patients with FXTAS or with other microsatellite expansion diseases.