Repeat-Associated Non-AUG Translation and Its Impact in Neurodegenerative Disease

Abstract

Nucleotide repeat expansions underlie numerous human neurological disorders. Repeats can trigger toxicity through multiple pathogenic mechanisms, including RNA gain-of-function, protein gain-of-function, and protein loss-of-function pathways. Traditionally, inference of the underlying pathogenic mechanism derives from the repeat location, with dominantly inherited repeats within transcribed noncoding sequences eliciting toxicity predominantly as RNA via sequestration of specific RNA binding proteins. However, recent findings question this assumption and suggest that repeats outside of annotated open reading frames may also trigger toxicity through a novel form of protein translational initiation known as repeat-associated non-AUG (RAN) translation. To date, RAN translation has been implicated in 4 nucleotide repeat expansion disorders: spinocerebellar ataxia type 8; myotonic dystrophy type 1 with CTG•CAG repeats; C9orf72 amyotrophic lateral sclerosis/frontotemporal dementia with GGGGCC•GGCCCC repeats; and fragile X-associated tremor/ataxia syndrome with CGG repeats. RAN translation contributes to hallmark pathological characteristics in these disorders by producing homopolymeric or dipeptide repeat proteins. Here, we review what is known about RAN translation, with an emphasis on how differences in both repeat sequence and context may confer different requirements for unconventional initiation. We then discuss how this new mechanism of translational initiation might function in normal physiology and lay out a roadmap for addressing the numerous questions that remain.

Electronic supplementary material

The online version of this article (doi:10.1007/s13311-014-0292-z) contains supplementary material, which is available to authorized users.

Introduction

Nucleotide repeat expansions underlie over a dozen human neurological diseases, ranging broadly in severity, symptoms, sites of pathology, and prevalence [1, 2]. Repeat expansions are thought to elicit toxicity via 3 nonexclusive mechanisms. Repeats can alter transcription in cis, leading to suppressed RNA and protein expression from the gene in which they reside [3, 4]. Alternatively, transcribed repeats as RNA can bind to and sequester RNA binding proteins and prevent them from performing their normal functions [5, 6]. Lastly, translated repeats can alter the normal functions of the proteins in which they reside while also directly eliciting toxicity via alterations in proteostasis [5, 7].

The dominant mechanisms by which a given repeat acts to elicit toxicity is dependent on numerous variables, including the length of the repeat, its sequence context, and the native functions of the protein-coding gene with which it is associated. For example, large CGG repeat expansions (>200 repeats in length, i.e., “full mutations”) in the 5’ untranslated region (UTR) of FMR1 elicit transcriptional and translational silencing of the FMR1 locus (loss-of-function) resulting in absent expression of the fragile X protein (FMRP) [8]. The absence of FMRP leads to fragile X syndrome, a common cause of intellectual disability and autism. In contrast, moderate CGG expansions in FMR1 (50–200 repeats, i.e., “premutations”) are actively transcribed, allowing the CGG repeat as RNA to bind to a number of key RNA-binding proteins and alter their functions [9–12]. For example, the microRNA biogenesis-associated proteins DGCR8 and DROSHA are sequestered by the expanded CGG repeat, subsequently eliciting alterations in microRNA abundance [12]. Similarly, expanded CUG repeats in the 3’UTR of DMPK mRNA bind to muscleblind-like protein splicing factors, leading to alterations in alternative splicing of muscleblind-like protein target transcripts [5, 13–21]. For trinucleotide repeat expansions located within canonical open reading frames (ORFs), as occurs with CAG repeats in HTT, the repeat is translated into the native protein. In Huntington’s disease, this leads to an expanded polyglutamine repeat within the Huntingtin protein that aggregates in affected tissues and that elicits toxicity via alterations in both proteostasis and in the native functions of the protein [22–26].

However, recent data from multiple laboratories suggest that nucleotide repeat expansions are capable of eliciting translational initiation in the absence of a normal ORF or AUG start codon [27–30]. This process, referred to as repeat-associated non-AUG (RAN) translation, has the potential to significantly alter our understanding of nucleotide repeat disorder pathogenesis. In this review, we describe the discovery of RAN translation and its association with neurodegenerative diseases. We then discuss how RAN translation might occur, with specific insights into the roles of sequence and repeat context, and the potential for similar processes to expand the functional proteome. Lastly, we highlight specific questions that the field must address going forward.

Eukaryotic Translation Initiation

Upon transcription and RNA processing (e.g., capping, splicing, and polyadenylation), eukaryotic mRNAs are exported from the nucleus into the cytoplasm to serve as templates for translation. The majority of eukaryotic nuclear-encoded mRNAs rely on cap-dependent translational machinery for initiation. In this process, the 5’ m⁷GpppG cap acts as a critical cis-element that recruits the cap-binding protein eukaryotic initiation factor (eIF) eIF4E into a complex known as eIF4F to allow unwinding of the mRNA and loading of the 43S preinitiation complex (PIC; comprised of the small 40S ribosomal subunit and the eIF2 ternary complex) on the 5’ end of the mRNA (reviewed elsewhere [31–33]). This PIC then scans along the mRNA in a 5’ to 3’ direction with the help of RNA helicases until it reaches an AUG start codon in an optimal context (e.g., Kozak sequence) [34–40], at which point the 60S ribosomal subunit is recruited and translation begins.

Alternatively, some mRNAs utilize a cap-independent mechanism for translation initiation that allow for internal ribosome entry of the 43S PIC onto the mRNA at highly structured elements known as internal ribosome entry sites (IRES) [41–44]. IRES elements can also sometimes directly recruit a subset of eIFs to allow for unconventional translational initiation [43–46]. In one extreme example, the cricket paralysis virus-IRES initiates translation independent of any eIFs or initiator tRNA [47–50]. Although initially described only in viral RNAs, there are now numerous examples of eukaryotic cellular IRES in transcripts such as c-myc, p53 [51], FGF2 [52], eIF4G/eIF4G1 [53, 54], and Apaf-1 [55].

Additionally, while initiation typically begins at the first AUG codon in an appropriate sequence context, there are numerous documented examples where near-cognate start codons (differ from AUG by one nucleotide, e.g., CUG or GUG) are utilized [56]. Recent ribosome profiling studies, which map the position of ribosomes on mRNAs across the transcriptome, suggest that such near-cognate initiator codon usage may be quite prevalent [57–59]. These alternative initiation events play important regulatory roles in protein production from their associated transcripts, implying that a lack of stringency may be built into the system [60].

RAN Translation in Neurodegenerative Disease

Thus far, RAN translation has been linked to CTG•CAG, GGGGCC•GGCCCC, and CGG microsatellite expansions (Figs. 1 and 2). RAN translation was initially reported by Zu et al. [27] in association with spinocerebellar ataxia type 8 (SCA8). SCA8 is a dominant, slowly progressive neurodegenerative disorder caused by a CAG•CTG repeat expansion within the coding sequence of the ATXN8/ATXN8OS gene [61–63]. Zu et al. [27] found that removal of the only ATG start codon from a SCA8 minigene construct failed to prevent production of a poly-Gln protein. Using a series of elegant epitope-tagged constructs, mass spectrometry, and tritium-labeling experiments, Zu et al. [27] demonstrated that translation of CAG repeats could occur without an AUG start codon in all 3 reading frames to produce poly-Gln, poly-Ala, and poly-Ser homopolymeric proteins. Immunofluorescence-based experiments demonstrated accumulation of all 3 RAN products within a single transfected cell, suggesting that these processes can occur in parallel. For SCA8 RAN-translated poly-Ala, mass spectroscopy measurements identified a series of peptides lacking an N-terminal methionine but with differing lengths of alanine peptides, suggesting initiation occurring throughout the GCA repeat itself. Evidence for similar translation in all 3 reading frames across CUG repeats from the ATXN8OS transcript was also observed, as was initiation in alternative ORFs when the repeats were placed within an AUG-initiated ORF. Zu et al. [27] found that antibodies targeting the predicted poly-Ala product from the ATXN8 transcript selectively recognized a protein in cerebellum in human SCA8 cases and SCA8 model mice. A similar approach provided in vivo evidence of a poly-Gln RAN translation product from an antisense CAG repeat containing transcript associated with the DMPK locus in myotonic dystrophy type 1 (DM1) (Fig. 1) [27, 64].

Fig. 1.

Repeat-associated non-AUG (RAN) translation at CAG, CUG, and GGGGCC repeats. (a) Spinocerebellar ataxia type 8 (SCA8) results from a CAG•CTG expansion in exon A of the ATXN8 gene, with an expanded CAG repeat in the sense transcript and CUG repeat in the antisense transcript, ATXN8OS. RAN translation produces poly-Gln-, poly-Ala-, and poly-Ser-containing proteins from ATXN8 and potentially poly-Ala-, poly-Cys-, and poly-Leu-containing proteins from ATXN8OS. The box in the ATXN8 sense transcript represents the open reading frame (ORF). The CAG repeat comprises the entire ORF. (b) Myotonic dystrophy type 1 (DM1) results from a CTG•CAG expansion in the 3’ untranslated region (UTR) of DMPK, where the CUG repeat resides in the 3’UTR and the CAG repeat is part of an antisense transcript of unknown function. In vivo antibody-based evidence suggests that a poly-Gln RAN product is produced in a DM1 mouse model and human tissues. (c) RAN translation of the expanded GGGGCC repeat located in intron 1 of the C9orf72 transcript results in 3 dipeptide repeat proteins: poly-(Gly-Pro), poly-(Gly-Ala), and poly-(Gly-Arg). In addition, an antisense transcript (containing GGCCCC repeats) also acts as a substrate for RAN translation, producing poly-(Gly-Pro), poly-(Pro-Ala), and poly-(Pro-Arg) dipeptide repeat proteins. *Evidence of production in vitro. ^‡Evidence of production in vivo in patient tissues. ? = possible RAN products

Fig. 2.

Repeat-associated non-AUG (RAN) translation at CGG repeats in fragile X-associated tremor/ataxia syndrome. (a) Cap-dependent translational initiation at a canonical AUG start codon in FMR1 mRNA is likely required for production of fragile X protein (FMRP) in the setting of both normal and premutation repeat expansions. (b) Working model of RAN translation at CGG repeats. Premutation length repeats (50–200 repeats) form large thermostable hairpins or G-quadruplexes in FMR1 mRNA that stall scanning preinitiation complexes (PICs) [74]. Mutational analysis in the sequence upstream of the repeat suggests that initiation occurs at a non-AUG start codon just upstream of the repeat to produce FMRpolyG [30]. However, initiation in a different reading frame is not affected by placement of a stop codon just upstream of the repeat, suggesting that FMRpolyA initiation occurs within the repeat itself in a process that exhibits different repeat size requirements [30]. Of note, neither of these RAN products are fused to FMRP, as they exist in different open reading frames. (c) The mechanism of CGG RAN translation remains to be determined. Many open questions remain: 1) Is RAN translation cap-dependent or does it involve a previously defined internal ribosome entry site (IRES)?; 2) How does the initiation of FMRpolyG and FMRpolyA translational differ?; 3) What initiation codons are used to generate each RAN product?; 4) What role do trans-acting factors, such as G-quadruplex destabilizing proteins hnRNP A2 and CArG-box binding factor A (CBF), and other RNA helicases, play in RAN translation?; 5) Is the antisense transcript ASFMR1 through the repeat subject to RAN translation?; 6) Are FMRpolyA and FMRpolyG functional or pathogenic?; 7) What impact does RAN translation have on the production of FMRP? Transcript is not drawn to scale. *Evidence of production in vitro. ^‡Evidence of production in vivo in patient tissues. ? = possible RAN product

More recently, 2 independent laboratories have reported pathologic evidence for RAN translation of a GGGGCC hexanucleotide repeat expansion in intron 1 of C9orf72 [28, 29]. This repeat, which is normally <23 repeats in controls and is often expanded into the hundreds of repeats in patients, is the most common known genetic cause of amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) [65, 66]. Using antibodies specific for poly-(Gly-Ala), poly-(Gly-Pro), and poly-(Gly-Arg)—putative dipeptide repeat RAN products from GGGGCC—Mori et al. [29] selectively detected proteins in postmortem brain samples with the C9orf72 expansion mutations by slot-blot analysis and immunohistochemistry. Similarly, Ash et al. [28] detected GGGGCC RAN-positive proteins in C9orf72 mutated samples, but not control tissues via immunohistochemistry [28]. Both groups found that poly-(Gly-Ala), poly-(Gly-Pro), and poly-(Gly-Arg) co-localize with p62-positive, TDP-43-negative cytoplasmic inclusions—a hallmark in C9orf72-related FTD. In both cases, the proteins appeared to be either too large or too insoluble to enter into polyacrylamide gels, making sizing analysis and quantitation challenging. One of the groups provided at least a hint at how an intronic repeat might get translated by demonstrating an increase in intronic retention in some cases [29]. Moreover, as occurred with CAG repeats [27], in transfected cells GGGGCC RAN translation in each reading frame exhibited different repeat size requirements before they became detectable by Western blot [29].

Since these initial reports, multiple laboratories have investigated how GGGGCC RAN translation affects toxicity and pathogenesis. Using similar strategies that defined GGGGCC RAN proteins from the C9orf72 sense transcript, multiple independent reports have recently confirmed dipeptide repeat RAN proteins in vivo—including poly-(Gly-Pro), poly-(Pro-Ala), and poly-(Pro-Arg)—originating from the C9orf72 antisense transcript [67–69]. Mori et al. [68] and Zu et al. [69] demonstrated that RAN products from both the sense and antisense transcript can be found in individual hippocampal neurons in C9orf72 patients. Interestingly, while GGGGCC RAN proteins are produced from expanded repeats, Mackenzie et al. [70] showed that poly-(Gly-Ala) aggregate distribution within the central nervous system of C9orf72 FTD/ALS patients is similar across clinical phenotypes and is anticorrelated with both cytoplasmic TDP-43 aggregate formation and neuronal loss. One possible explanation for this finding is that GGGGCC RAN proteins, or at least poly-(Gly-Ala), do not play an active role in pathogenesis. Alternatively, the formation of detectable aggregates may play a protective role, with “surviving” neurons being marked by their ability to efficiently form inclusions of these potentially toxic proteins, as has been reported for polyglutamine proteins [71]. Consistent with this concept, cell culture-based experiments in HEK293T cells show a strong correlation between cellular toxicity and increased production of GGGGCC RAN products from both the sense and antisense transcript, specifically poly-(Gly-Pro) and poly-(Pro-Arg) [69]. Conversely, studies using RNA based knockdown of C9orf72 in patient-derived induced pluripotent stem cells primarily support a RNA gain-of-function toxicity mechanism associated with the expanded GGGGCC allele, although these studies did not control for potential differences between soluble and insoluble RAN protein toxicity [72, 73]. The more difficult experiments aimed at testing the relative toxicity of GGGGCC•GGCCCC RAN proteins in animal models at physiologic concentrations in the absence of repeat RNA await.

RAN Translation at CGG Repeats: Insights into Mechanism

Recent work form our laboratory described RAN translation at expanded CGG repeats in the 5’UTR of FMR1, which underlie the neurodegenerative disorder fragile X-associated tremor/ataxia syndrome (FXTAS) [30] (Fig. 2). Experiments in both flies and mammalian cells show RAN translation in 2 reading frames: GGC (+1, poly-Gly) and GCG (+2, poly-Ala). Using antibody-based analyses akin to those described above, we demonstrated that the predicted poly-Gly containing protein, which we termed FMRpolyG, was present in neuronal intranuclear inclusions in human FXTAS brains. Importantly, FMRpolyG production elicited intranuclear inclusions in transfected cells, and modulating its production altered CGG repeat-associated toxicity in both cell culture and fly models of disease, providing direct in vivo evidence that RAN translation products can elicit toxicity [30].

Studies in cell culture provide some mechanistic insights into what cis factors influence when RAN translation occurs at CGG repeats. In the +1 (GGC, Gly) frame, initiation appears to occur predominantly outside of the repeat itself [30]. Placing a stop codon 12 nucleotides, but not 21 nucleotides, upstream of the repeat blocked detectable FMRpolyG production. As a stop codon would not halt or interfere with the scanning PIC, but world terminate translation if initiation was already established, we reasoned that RAN translation in this reading frame must initiate upstream of the repeat. Yet serial mutation of near-cognate start codons or removal of this entire region (the 48 nucleotides just upstream of the repeat) hindered, but did not abolish, RAN translation. However, importantly, at least 1 in-frame near-cognate initiation codon was always required [30]. Thus, FMRpolyG translation is not strictly dependent on the sequence at the site of initiation, but does require something that looks like an AUG start codon and a secondary structure in the CGG repeat that is predicted to stall a scanning ribosomal PIC [74, 75]. A similar observation was seen with stalling of the PIC by a stable hairpin and initiation at non-AUG codon or an AUG codon in a nonoptimal context [35, 74, 75]. This model is further supported by ribosome profiling studies [57], which demonstrate ribosome-protected fragments over numerous near cognate start codons just upstream of the CGG repeat in both human cell lines and mouse embryonic stem cells [30].

However, these rules for one form of RAN translation effectively do not apply to the other CGG repeat reading frames. For example, while RAN translation in the +1 (Gly) frame occurs even at shorter CGG repeat sizes (25–30 repeats) within the “normal” range in humans, RAN translation in the +2 (GCG, Ala) frame was observed only at larger (>70 CGGs) repeat sizes. Moreover, placement of stop codons just prior to the repeat in this +2 (Ala) frame had no impact on translation, suggesting initiation within the repeat [27]. Further, even at larger repeat sizes (100 CGGs), no products were detectable in the +0 (CGG, Arg) reading frame. Consistent with this, there is no evidence for an N-terminal extension on FMRP in patients with large, unmethylated repeat expansions, and FMRP is not found in the ubiquitinated inclusions in patients [76], both of which would be predicted if such a product were made. Thus, the mechanisms by which translation initiation occurs at each repeat and, indeed, each reading frame of each repeat, may be different and interdependent on numerous factors, including the surrounding sequence context, the amino acid produced, and the length of the repeat expansion.

What other cis factors are required for RAN translation at different repeats remains unexplored. The FMR1 locus may provide some early insights. While in vitro studies suggest that the primary mode of FMR1 mRNA translational initiation is 5’-cap-dependent [77], the FMR1 5’UTR does contain a functional IRES upstream of the CGG repeat [78, 79]. Interestingly, Ludwig et al. [77] also showed that replacement of the CGG repeat with a stable hairpin blocked translation in a synthetic 5’UTR, but not in FMR1 5’UTR containing mRNA. This lack of blockade resulted from translation initiation at a near-AUG start codon (GUG) within the synthetic hairpin itself [77], suggesting the secondary structure within the FMR1 5’UTR is somehow permissive of nonconical translation initiation. Other components of the FMR1 5’UTR may also modulate FMRP translation. Premutation length repeats confer a propensity towards a different transcription start site than that typically used at normal length repeats [80, 81]. Use of this alternative transcription start site is associated with less efficient FMRP translation independent of repeat length [77]. These findings suggest that the FMR1 5’UTR may contain independent cis-elements that favor translational initiation at non-AUG start codons and RAN translational events, even at normal repeat sizes. However, the functional roles that these specific alternative transcriptional initiation sites and IRES sequences play in CGG RAN translation remain untested.

Trans-acting factors, such as specific initiation factors, RNA helicases, and other RNA-binding proteins, may be critical for RAN translation and may also differ between repeats. Interestingly, a few trans-acting factors, including hnRNPA2/B1 and CArG-box binding factor A, bind to CGG repeats (as either G-quadruplex [82–86] or hairpin structures [87, 88]) and augment translation of reporters placed downstream of the FMR1 5’UTR both in vitro and in cell culture systems [85, 89]. Although production of the +1 CGG RAN product and initiation at the downstream canonical AUG start codon appear to track together in cell-based systems [30], it is unknown whether the interaction of these RNA binding proteins might enhance or impair CGG RAN translation. Given that hnRNPA2/B1 overexpression suppresses CGG repeat-associated toxicity in Drosophila [10], these relationships deserve further exploration.

Impact on the Proteome and Translation Regulation

RAN translation has the potential to significantly contribute to proteome diversity by increasing the number of potential proteins generated from each transcript. The human genome harbors more repetitive elements than previously predicted [90], including microsatellites, and is pervasively transcribed [91–94]. Furthermore, there is evidence from many emerging techniques that generation of alternative protein products from noncanonical ORFs is common, even in some annotated “noncoding” RNAs [57, 59, 95, 96]. Ribosomal profiling studies suggest a significant underestimation of non-AUG ORFs and upstream ORFs (uORFs) in previous genome annotation [57, 97], and the use of these alternative uORFs appears to be a regulated event, given that alterations in internal states or environmental conditions can lead to shifts in ribosomal positioning on mRNAs in a transcription-independent manner [58]. If a portion of these uORFs initiate via a RAN translation-like mechanism, then understanding how they work may have a broader impact on our comprehension of the genome’s coding potential.

Additionally, RAN translation may play specific regulatory roles in certain sequence contexts [98]. In FMR1, translation through the repeat (the uORF) may assist in destabilizing the inhibitory CGG repeat RNA structure, allowing for normal scanning of subsequent ribosomes and proper initiation of the canonical ORF (Fig. 2) [30]. This is a particularly intriguing mechanism to consider for 3 reasons. First, uORFs within FMR1 appear to be conserved, given that the FMR1 5’UTR (outside of the repeat) is surprisingly invariant in humans [99], that ribosomal profiling peaks reveal multiple uORFs in mice just proximal to the CGG repeat, and that noncanonical upstream initiation occurs in association with dfxr, the Drosophila homolog of FMR1 [30, 100]. Second, the inhibitory effects of the FMR1 5’UTR on translation are not as significant as would be predicted based on RNA secondary structure modeling. Specifically, the GC rich repeat and surrounding sequence, even at “normal” repeat sizes, has a predicted minimal free energy in excess of hairpins that completely impair ribosomal scanning [35, 74, 101, 102]. This suggests that something (perhaps RAN translation?) must assist in unwinding this hairpin to allow FMRP production. Third, FMRP translation is itself a highly regulated event that is critical for appropriate spatiotemporal regulation of translation within neuronal dendrites [103], where it acts as a suppressor of translation [104]. In response to activation of metabotropic glutamate receptors, it is rapidly phosphorylated and degraded, allowing translation of the transcripts with which it is associated [105, 106]. Additionally, FMRP itself is rapidly synthesized by mGluR activation, and this appears critical for maintenance of normal synaptic responses [107–109]. However, how FMR1 translation is initiated and regulated remains poorly defined. CGG RAN translation may offer a mechanism by which the repeat and 5’UTR could allow for such regulation, and similar mechanisms may exist for other neuronal or cellular transcripts to provide a previously unappreciated level of translational regulation.

A Roadmap Forward in RAN Translation

Our understanding of how RAN translation occurs and what roles it plays in neurodegeneration and normal physiology is still in its infancy. Already, it is clear that there may well be multiple types of RAN translation, with different initiation requirements and pathologic consequences for each repeat—or, more specifically, for each reading frame associated with each transcript (sense and antisense) produced from each repeat [110–112]. For all of these, it will be important to address the following 3 questions.

First, what are the critical cis and trans factors that allow for RAN translational initiation to occur at repetitive RNA sequences? For each repeat in a setting as close as possible to its native context, we must evaluate what the key steps are in translational initiation, from whether the transcript must be capped and spliced to what impact neighboring sequence differences play, to what initiation factors are utilized to accomplish initiation of the translational event. Additionally, for certain repeat contexts, even more complicated questions must be answered. For example, the GGGGCC repeat-associated with C9orf72 is in an intron, which predicts that it should normally be rapidly targeted for degradation and excluded from the cytoplasm. Are introns retained in select mRNAs and exported from the nucleus, or is RAN translation a unique form of nuclear translation [113, 114]? Recent work from Haeusler et al. [115] demonstrates that both GGGGCC repeat DNA and RNA form G-quadruplex secondary structures, as well as RNA–DNA hybrids, which stimulate the accumulation of repeat-containing abortive transcripts. One intriguing possibility is that such transcripts might bypass typical intron degradation machinery, allowing for subsequent nuclear export and RAN translation [115].

One important question that has not been addressed is whether the canonical initiator tRNA (Met-tRNA^Met_i) is required, and, if so, how it is delivered to the PIC and non-AUG start codon. As part of the initiating ternary complex, Met-tRNA^Met_i is typically delivered to the 40S subunit by eIF2–guanosine triphosphate to form the canonical PIC. Importantly, the eIF2–guanosine triphosphate-bound PIC has very low selectively for non-AUG start codons [116–121], which may be inconsistent with the relatively high levels of RAN products seen in cell culture and fly models [27–30, 68, 69]. However, multiple translation factors—such as eIF2A [122, 123], eIF5B [124–127], and eIF2D [128, 129]—have been noted to bind and deliver Met-tRNA^Met_i to the 40S subunit, although still preferentially initiating at AUG start codons. Nevertheless, this raises the interesting possibility that atypical eIFs may have a unique role in RAN translation initiation. Alternative translation initiation approaches bypassing the need for Met-tRNA^Met_i have been reported, particularly for noncanonical initiation at the cricket paralysis virus–IRES. Not requiring Met-tRNA^Met_i or any known eIF [49], the cricket paralysis virus–IRES directly recruits the 40S subunit and allows for the first codon to be decoded by eEF1A-aminoacyl-tRNA at the A-site [130]. Whether similar mechanisms or translation factors are responsible for RAN translation remains to be determined.

Once a clearer picture of the minimal requirements for RAN translation emerges, we can begin to address questions related to the apparent differences between repeats [110–112]. Identifying the critical cis-elements and trans-factors that play roles in RAN translation may elucidate a core RAN translation machinery. In addition, determining what drives the nonuniformity in repeat length dependence between reading frames of a single transcript will provide crucial mechanistic insight.

Second, what are the normal physiologic functions, if any, for RAN translation? Although most RAN products appear at first glance to be purposeless, it is premature to assume that they lack normal functions. Characterization of their interacting partners and distribution within cells may provide insights into both their direct roles in toxicity and their native roles, if any [110–112]. Furthermore, it will also be important to define what roles RAN translation plays in regulating translation from other ORFs found in the same transcripts in which it occurs. A broader question is whether these same RAN translational mechanisms might mediate the use of non-AUG codons throughout the transcriptome on highly structured RNAs. While identifying such genes will not be trivial, characterizing the minimal cis elements that are required for RAN translation may allow for in silico identification of additional genes that harbor similar elements. Likewise, RAN translation-specific trans-acting factors could allow for identification of target transcripts from which RAN products might be produced. Together, such approaches may shed light on the prevalence of RAN translation and its potential roles in cellular homeostasis [110–112].

The repetitive nature of the RAN homopolymeric and dipeptide repeat proteins pose some specific challenges in terms of questions related to relative abundance, toxicity, stability, and cellular clearance. For example, degradation pathways such as the N-end rule pathway (reviewed in [131, 132]) can strongly influence the half-life of a given protein. Certain N-terminal amino acids, such as Arg and Gln, target proteins for rapid turnover [133, 134]. Thus the N-terminal amino acid utilized during RAN translational initiation may influence both the relative stability of the protein and its potential for cellular function. Such pathways could explain why, for example, no CGG RAN product is observed in the Arg (+0) frame [30], despite evidence for accumulation of RAN translation products in all other reading frames tested to date. Alternatively, some of the RAN proteins may be particularly resistant to endoproteolytic cleavage owing to a lack of enzymes that recognize the specific repetitive motifs they present. Detailed biochemical analysis of each RAN protein N-terminus and degradation pathways will likely be revealing.

Third, what is the pathological consequence of RAN translation and is its correction a meaningful therapeutic target? The accumulation of RAN translation products in aggregates in human disorders suggests that they may be pathologic, but the evidence for their direct roles in pathogenesis are largely lacking, especially in vivo in mammalian systems. To achieve this, it will be necessary to dissociate the toxicity associated with repeats as RNA from the potential toxicity of RAN products as protein. Given that many of these repeats exist in RNAs that are of low abundance (introns in pre-mRNAs and antisense transcripts typically are present at a fraction of the sense mRNAs with which they are associated), it will be important to study these processes at physiologic concentrations. Howe-ver, if they are highly toxic, even low-level production and accumulation may contribute meaningfully to pathogenesis.

As is true for many inherited neurodegenerative disorders, there is a decades-long delay between the initial expression of the toxic mutated genes and the subsequent development of clinical symptoms in FXTAS and ALS/FTD. If RAN proteins contribute meaningfully to neurodegeneration, then it will be important to determine if their delayed toxicity reflects a loss of a compensatory mechanism for dealing with toxic proteins or instead whether there are tissue-specific changes in the relative production of RAN proteins. For example, ribosome profiling of nonlog phase yeast demonstrates a dramatic increase in nonconical translation, suggesting that different cellular conditions might be more or less permissive of alternative translational initiation mechanisms [58]. Could the aging brain provide a similarly permissive environment for RAN translation?

If RAN products are important, then RAN translation may serve as a novel therapeutic target. The feasibility of such an approach will depend heavily on what native roles RAN translational initiation processes underlie and whether they are mechanistically separable from those involved in the translation of most mammalian proteins. However, it is tempting to speculate that if these translational events are truly aberrant, then identifying agents that can selectively block them is a potentially fertile area for drug development.

Conclusion

Discovery of RAN translation has shed light on a new facet of eukaryotic translation initiation and human disease. With further mining into the mechanism, a novel therapeutic target for multiple neurodegenerative disorders is plausible. It will be critical to determine how prevalent RAN translation is across the transcriptome and whether it contributes to normal cellular functions and human diseases.