Molecular Mechanisms Underlying Nucleotide Repeat Expansion Disorders

Abstract

Short tandem repeats, a class of DNA elements composed of 2–12 bp motifs repeated in tracts of variable lengths, are abundant throughout the human genome. Expansions of these repeats underlie over four dozen human diseases, the first of which was described 30 years ago. This paradigm-shifting discovery radically changed how we conceive DNA, RNA, and protein dynamics and their interactions across human disease states. Here, we dive into the four major modes by which short tandem repeat expansions cause disease: loss of function through transcriptional silencing, RNA-mediated gain of function through gelation and RNA-binding protein sequestration, gain of function of AUG-initiated repeat-harboring native proteins, and repeat-associated non-AUG (RAN) translation of toxic repetitive peptides. Somatic repeat instability, driven by replication- and repair-dependent processes, titrates toxicity with age and across tissues. We focus on the crosstalk between these mechanisms, with particular examples which emphasize that targeting a single pathway alone may be insufficient to fully address disease pathogenesis. We also discuss the emerging native functions of repeat elements and how their dynamics might contribute to disease at a larger scale than currently appreciated. Lastly, we use these molecular insights to suggest holistic approaches to known and novel repeat expansion disorders as a method for better understanding disease mechanisms and expediting therapeutic development.

INTRODUCTION:

Short tandem repeats [G] (STRs, or microsatellites) are stretches of repeated 2–12 bp units of DNA found within both coding and non-coding regions of the genome. STRs make up about 3% of the human genome and are intrinsically unstable, showing a high degree of variation across species and within human populations^1,2. Small fluctuations in length enable STRs to function as important adaptive regulators of gene expression by modulating DNA methylation, alternative splicing efficiency, and transcription factor binding^3,4—yet, this instability comes at a cost.

In 1991, four groups independently identified instability and expansion of an STR composed of CGG repeats in the 5’ untranslated region (UTR) of the FMR1 gene as the underlying cause of Fragile X syndrome^5–8. In parallel, a fifth group described a CAG repeat expansion encoding a polyglutamine stretch within the androgen receptor AR as the underlying cause of spinal and bulbar muscular atrophy (SBMA)⁹. These seminal findings introduced nucleotide repeat expansions as a cause of human disease and redefined how gene mutations can occur and contribute to neurological dysfunction. Moreover, these two disease-associated repeats highlight distinct modes of pathogenesis, opening the door to a wide spectrum of mechanisms by which expanded repeats lead to disease.

Since those initial discoveries, there has been significant growth in our understanding of both how repeat expansions arise and how they contribute to more than fifty human disorders. Disease-causing expansions can reside within gene promoters, protein-coding exons, 5’ and 3’ UTRs, or introns. Expansion motifs range from trinucleotides to dodecameric units [Table 1], and pathogenic expansion lengths vary from a few repeat units to thousands. These repeats have had an outsized influence on our understanding of molecular and cellular biology and neuroscience, providing novel insights into chromatin biology, DNA repair, RNA-protein condensates, non-canonical translation [G], and protein aggregation. Furthermore, although more than 1.5 million STRs have been identified in the human genome to date⁴, our knowledge of their native functions at non-pathogenic lengths remains in its infancy^10–12.

Table 1:

Repeat expansion diseases by location and pathomechanism.

Disease	Abbrev.	Inheritance	Host gene	Repeat motif	Location	Somatic instability	LOF	RBP sequestration	AUG-initiated protein GOF	RAN translation	Other
Unverricht-Lundborg disease	EPM1	AR	CSTB	CCCCGCCCCGCG	promoter		+
Baratela-Scott syndrome	BSS	AR	XYLT1	CGG	promoter/5’ UTR	+	+
Glutaminase deficiency	GAD	AR	GLS	CAG	5’ UTR	+	+
Spinocerebellar ataxia type 12	SCA12	AD	PPP2R2B	CAG	5’ UTR				+
Fragile XE syndrome	FRAXE	XL	AFF2	CCG	5’ UTR	+	+
Jacobsen syndrome	FRA11B	AD	CBL2	CCG	5’ UTR						+ a
Intellectual disability, associated with fragile site FRA2A	FRA2A	AD	AFF3	CGG	5’ UTR	+	+
Intellectual disability, associated with fragile site FRA12A	FRA12A	AD	DIP2B	CGG	5’ UTR		+
Fragile X syndrome	FXS	XL	FMR1	CGG	5’ UTR	+	+
Fragile X-associated primary ovarian insufficiency	FXPOI	XL	FMR1	CGG	5’ UTR	+				+
Fragile X-associated tremor/ataxia syndrome	FXTAS	XL	FMR1	CGG	5’ UTR	+		+		+
Neuronal intranuclear inclusion disease	NIID	AD	NOTCH2NLC	CGG	5’ UTR
Oculopharyngodistal myopathy 1	OPDM1	AD	LRP12	CGG	5’ UTR	+
Oculopharyngodistal myopathy 2	OPDM2	AD	GIPC1	CGG	5’ UTR
Oculopharyngeal myopathy with leukoencephalopathy	OPML	AD	LOC642361/NUTM2B-AS1	CGG/CCG	lncRNA	+
Cerebellar ataxia, neuropathy, vestibular areflexia syndrome	CANVAS	AR	RFC1	AAGGG	intron
Spinocerebellar ataxia type 10	SCA10	AD	ATXN10	ATTCT	intron	+		+
X-linked dystonia-parkinsonism	XDP	XL	TAF1	CCCTCT	intron	+	+
Myotonic dystrophy type 2	DM2	AD	CNBP	CCTG	intron	+		+		+
Autism spectrum disorder, associated with fragile site FRA7A	FRA7A	AD	ZNF713	CGG	intron	+	+
Fuchs endothelial corneal dystrophy	FECD	AD	TCF4	CTG	intron	+		+		+
Friedreich’s ataxia	FA	AR	FXN	GAA	intron	+	+
Spinocerebellar ataxia type 36	SCA36	AD	NOP56	GGCCTG	intron			+		+
C9ORF72 amyotrophic lateral sclerosis/frontotemporal dementia	C9 ALS/FTD	AD	C9ORF72	GGGGCC	intron	+	+	+		+
Spinocerebellar ataxia type 31	SCA31	AD	BEAN1/TK2	TGGAA/TTCCA	intron			+		+
Familial adult myoclonic epilepsy 1	FAME1	AD	SAMD12	TTTCA	intron	+
Familial adult myoclonic epilepsy 2	FAME2	AD	STARD7	TTTCA	intron
Familial adult myoclonic epilepsy 3	FAME3	AD	MARCH6	TTTCA	intron	+
Familial adult myoclonic epilepsy 4	FAME4	AD	YEATS2	TTTCA	intron
Familial adult myoclonic epilepsy 6	FAME6	AD	TNRC6A	TTTCA	intron
Familial adult myoclonic epilepsy 7	FAME7	AD	RAPGEF2	TTTCA	intron
Spinocerebellar ataxia type 37	SCA37	AD	DAB1	TTTCA	intron
Dentatorubral-pallidoluysian atrophy	DRPLA	AD	ATN1	CAG	CDS	+		+	+
Huntington’s disease	HD	AD	HTT	CAG	CDS	+		+	+	+
Spinal and bulbar muscular atrophy	SBMA	XL	AR	CAG	CDS	+	+		+
Spinocerebellar ataxia type 1	SCA1	AD	ATXN1	CAG	CDS	+	+		+
Spinocerebellar ataxia type 2	SCA2	AD	ATXN2	CAG	CDS		+	+	+
Spinocerebellar ataxia type 3	SCA3	AD	ATXN3	CAG	CDS	+		+	+
Spinocerebellar ataxia type 6	SCA6	AD	CACNA1A	CAG	CDS				+
Spinocerebellar ataxia type 7	SCA7	AD	ATXN7	CAG	CDS	+			+
Spinocerebellar ataxia type 17	SCA17	AD	TBP	CAG	CDS	+	+		+
Pseudoachondroplasia and multiple epiphyseal dysplasia	PSACH/MED	AD	COMP	GAC	CDS		+		+
Blepharophimosis, ptosis, and epicanthus inversus syndrome	BPES	AD	FOXL2	GCN	CDS		+
Cleidocranial dysplasia	CCD	AD	RUNX2	GCN	CDS		+
Congenital central hypoventilation syndrome	CCHS	AD	PHOX2B	GCN	CDS	+ c	+
Hand-foot-genital syndrome	HFGS	AD	HOXA13	GCN	CDS	+ c	+
Holoprosencephaly 5	HPE	AD	ZIC2	GCN	CDS	+ c	+
Oculopharyngeal muscular dystrophy	OPMD	AD	PABPN1	GCN	CDS		+		+
Synpolydactyly 1	SPD	AD	HOXD13	GCN	CDS		+
X-linked hypopituitarism	XH	XL	SOX3	GCN	CDS		+
X-linked intellectual disability b	XLID	XL	ARX	GCN	CDS	+ c	+
Spinocerebellar ataxia type 8	SCA8	AD	ATXN8OS/ATXN8	CTG/CAG	3’ UTR/CDS			+	+	+
Myotonic dystrophy type 1	DM1	AD	DMPK	CTG	3’ UTR	+		+		+
Huntington’s disease-like 2	HDL2	AD	JPH3	CTG	3’ UTR		+	+

^aFRA11B expansion causes hypermethylation and is associated with chromosomal breakage and deletion of the telomeric end of 11q.²⁷¹

^bLoss-of-function mutations in ARX are associated with a spectrum of clinical phenotypes, including X-linked infantile spasm syndrome, X-linked lissencephaly with ambiguous genitalia, X-linked myoclonic epilepsy and intellectual disability, Partington syndrome, Ohtahara syndrome, and Proud syndrome.²⁷²

^cSomatic mosaicism documented in carriers only.^273–276

STR expansions induce architectural changes in DNA and elicit a series of concurrent molecular processes through either loss-of-function (LOF) or gain-of-function (GOF) mechanisms at the DNA, RNA, and/or protein levels [Figure 1]. The toxicity of each of these mechanisms is modulated by somatic instability [G] of the repeat, a stochastic process that titrates repeat length differentially across a patient’s tissues and cells over their lifetime^13–15.

Figure 1: Molecular mechanisms driving nucleotide repeat expansion pathogenesis.

Hyper-methylation of promoter regions can lead to transcriptional silencing, resulting in partial or complete loss of the native protein harboring the repeats. In contrast, active transcription through the repeats can trigger formation of R-loops (RNA:DNA hybrids that lead to DNA damage response pathway activation) and potentially exacerbate somatic repeat instability. Transcribed repeat RNAs fold into complex structures, which aberrantly interact with and sequester cellular RNA-binding proteins (RBPs). Trinucleotide repeat expansions in protein-coding sequences generate mutant proteins that elicit gain-of-function toxicity. Finally, the coding and non-coding repeat RNAs are translated in the absence of canonical AUG-mediated initiation through repeat-associated non-AUG (RAN) translation, producing toxic polymeric peptides.

In this review, we detail each of the core molecular mechanisms of pathogenesis in repeat expansion diseases, and we discuss how synergy between mechanisms produces the unique and complex pathology for each disease. Given the breadth of research in the field over the past three decades, comprehensive coverage of all important discoveries within a single article is not feasible, as attested to by multi-chapter books on the topic^16,17. As such, we for the most part do not discuss the downstream sequelae of these events, which are myriad, but instead focus on understanding the mechanisms most proximal to the underlying cause of these disorders. We largely focus on a subset of disorders (fragile X-associated disorders, myotonic dystrophies, and C9ORF72 amyotrophic lateral sclerosis and/or frontotemporal dementia (C9 ALS/FTD)) which our own groups study and which we feel offer a representative through-line in the field. We place a particular emphasis on emerging roles for how repetitive RNAs generate aberrant ribonucleoprotein (RNP) complex [G] formations and trigger RAN translation, given recent advances in these spaces. We also highlight critical points of synergy across disease mechanisms, potentially revealing novel therapeutic targets. Throughout, we include references to more detailed recent reviews of specific topics and disorders, and we apologize in advance to the many investigators whose excellent work we were not able to highlight.

Pathomechanisms of repeat expansions

The proximal pathomechanisms of repeat expansion diseases can be broadly classified into four interrelated categories [Figure 1]: (1) DNA-based mechanisms, including LOF via repeat-induced transcriptional silencing and GOF via R-loop [G] formation and DNA damage response activation; (2) toxic RNA-mediated GOF through gelation and RNA-binding protein (RBP) sequestration; (3) GOF of AUG-initiated repeat-harboring native proteins; and (4) repeat-associated non-AUG (RAN) translation of toxic repetitive peptides. Here, we unravel each of these categories, providing historical foundations and highlighting the current state of research in the contexts of representative diseases.

Transcriptional silencing, R-loop formation and somatic instability

Fragile X syndrome (FXS) was originally described in the 1940s as a form of intellectual disability and autism in young males¹⁸, and it was linked to the rare (familial) eponymous folate-sensitive fragile site [G] on the X chromosome in the late 1960s¹⁹. The causative mutation was mapped to a CGG STR in the 5’ UTR of the FMRP translational regulator 1 (FMR1) gene^5–8. Almost all mammals have a repeat element at this locus, with an average size of around 30 repeats in humans²⁰. In FXS, this repeat expands to >200 CGG units, inducing hyper-methylation of the repeat tract and the neighboring FMR1 promoter and epigenetic silencing of the FMR1 locus (Figure 2B). The end result of these transformations is the loss of FMR1 transcription and the absence of the FMRP protein^21,22.

Figure 2: Repeat-induced transcriptional silencing, R-loops and somatic instability.

(A) Allelic classes of the FMR1 gene containing normal to pathogenic CGG repeats. FMR1 normally has ~30 CGG repeats in its 5’ UTR that are not included as a part of the mature protein product FMRP. Pre-mutation (55–200 repeats) expansions result in production of large CGG repeat-containing RNAs that underlie the age-related neurodegenerative disorder fragile X-associated tremor/ataxia syndrome (FXTAS). Full mutation (>200 repeats) expansions in subsequent generations lead to silencing of the FMR1 locus and fragile X syndrome (FXS). (B) CGG repeat methylation (mC) may direct transcriptional silencing by favoring histone methylation and heterochromatin formation through mechanisms similar to those typically active at CpG islands (left panel). Alternatively or cooperatively, nascent RNA may trigger epigenetic silencing by hybridizing to the complementary CGG-repeat DNA to form RNA:DNA duplexes that recruit polycomb repressive complex 2 (PRC2) (right panel). (C) Transcription-induced R-loops also support formation of DNA slip-out structures that contribute to repeat instability. For CAG/CTG trinucleotide repeat expansions, extended stable hairpins form in both strands. Normally, mismatch repair (MMR) pathways keep the repeat tract length stable by melting the slip-outs, followed by gap-filling by DNA polymerase across the region. Inefficient repair or formation of multiple slip-outs leads to retention of the slip-out structures and expansion of the repeat region by incorporation of the looped DNA. Small molecules that target slip-out structures in CAG repeat DNA inhibit repeat expansion and bias instability toward contraction.

Loss of FMRP drives pathology in FXS. Disease severity correlates with repeat methylation and transcriptional silencing [G], and FMR1 mutations that disrupt FMRP function replicate the clinical features of FXS in patients and animal models^23–25. However, the exact mechanism by which the repeat expansion triggers transcriptional silencing remains unresolved²⁶. One proposal is that expanded FMR1 RNA instigates formation of R-loops, stalling RNA polymerase II (Pol II) transcription complexes and triggering recruitment of the histone methyltransferase polycomb repressive complex 2 (PRC2)^27,28 (Figure 2B). PRC2 then drives epigenetic changes at the locus through H3K27 histone methylation that favor stable silencing of FMR1. Consistent with this hypothesis, embryo-derived full mutation hESCs often exhibit active FMR1 transcription and a lack of DNA methylation prior to neuronal differentiation, but FMR1 is silenced later in development²⁹. Interestingly, depletion of nascent FMR1 mRNA favors an open chromatin state in FXS hESCs during neuronal differentiation and maintenance of FMR1 expression, suggesting a role for either R-loops or RNA-induced transcriptional silencing in the heterochromatinization process^27,28,30. However, this model largely ignores a direct role for DNA methylation in the silencing of the locus. Delivery of demethylating agents or targeting of DNA methyltransferases to FMR1 in patient iPSCs is sufficient to reactivate FMR1 expression^28,31,32. Moreover, methylation of CGG repeats in vitro favors heterochromatin formation and nucleosome assembly^22,33, suggesting that DNA methylation itself plays an active role in transcriptional silencing. As such, CGG repeat-mediated transcriptional silencing is likely triggered by a combination of DNA hyper-methylation and histone methylation elicited by R-loops and Pol II stalling, leading to heterochromatinization of the locus and loss of FMRP expression²⁶. Similar methylation-mediated silencing mechanisms may contribute to reduced C9ORF72 expression at GGGGCC repeats in C9 ALS/FTD^34–38, as well as other disease-causing repeats containing CpG elements^39,40. Key next steps are to define the exact interplay of these epigenetic processes and determine the proximal trigger for repeat DNA methylation.

Other repeats trigger transcriptional silencing and LOF phenotypes largely in the absence of DNA methylation. Most prominent among these is Friedreich’s ataxia, the most common autosomal recessive ataxia, which is typically caused by a homozygous GAA repeat expansion in the first intron of FXN⁴¹. This expansion triggers heterochromatin protein (HP1)-sensitive silencing of the gene^42–44. At least three proposed mechanisms appear to contribute to FXN silencing. First, the expanded GAA/TTC repeats form “sticky” triplex H-DNA structures that directly impede transcription⁴⁵. Second, an FXN antisense transcript (FAST-1) is upregulated in the disease and triggers a loss of CTCF binding in the FXN 5’ UTR, which in turn favors heterochromatization⁴⁶. Interestingly, FAST-1 expression in isolation is sufficient to suppress FXN transcription⁴⁷. Third, R-loops similar to those discussed above for FMR1 are also thought to form at expanded GAA repeats, where they can trigger RNA-induced transcriptional silencing^48,49. Once heterochromatinized, transcription elongation through the repeat is markedly diminished^50–52. Importantly—and regardless of the etiology of silencing—direct targeting of heterochromatinization by histone deacetylase and methyltransferase inhibitors appears sufficient to reactivate FXN transcription in many contexts and is the basis for multiple ongoing clinical trials⁴².

While epigenetic silencing causes loss of expression of the repeat-harboring gene, active transcription across repeats can encourage R-loop formation that does not ultimately lead to transcriptional shutdown. These R-loops trigger DNA damage response (DDR) cascades in cells that may contribute to pathogenesis⁵³. In Fragile X-associated tremor/ataxia syndrome (FXTAS), an age-related neurodegenerative disorder caused by moderately sized (50–200 CGGs) unmethylated repeat expansions in FMR1, R-loop formation rate increases with repeat length and is associated with DDR activation in both patient cells and mouse models^54–56 (Figure 2A). R-loops are also observed at other disease-causing repeat expansions, including the GGGGCC repeat implicated in C9 ALS/FTD and the CAG repeat in HTT that causes Huntington disease [G] ^57–59. Once activated, these DNA damage response cascades can trigger mitochondrial dysfunction and apoptosis^60,61. While Huntington disease cells and other cells expressing CAG repeat expansions are more sensitive to DNA damage^62,63, it is not clear whether resolution of these events is sufficient to preclude repeat toxicity.

DNA mismatch repair (MMR) [G] pathways that typically resolve abnormal DNA-DNA and DNA-RNA structures formed during transcription can exacerbate instability and cause further expansion of repeat elements in somatic cells^15,64,65 (Figure 2C). This somatic instability is observed in the majority of repeat expansion diseases [Table 1] and produces variation in repeat size and toxicity across tissues within the same patient. This process notably does not require DNA replication, as somatic expansions are observed in terminally differentiated cells such as neurons and myofibers^66–69. The precise mechanisms of somatic instability in repeat expansion diseases remain under intensive study and have been recently reviewed¹⁵. A role for somatic instability in disease pathogenesis is supported by genetic studies implicating MMR proteins as modifiers of age of onset in Huntington disease^70,71 and spinocerebellar ataxias (SCAs)⁷². In rodent models of Huntington disease, modulation of MMR is sufficient to suppress somatic instability and can reduce toxicity^73,74. Indeed, small-molecule targeting of slip-out structures in CAG repeat DNA, which recruit MMR, induces contractions in disease models (Figure 2C)⁷⁵. These results suggest that these processes may serve as cross-platform therapeutic targets.

RNA multivalency and RNA-binding protein interactions

Most repeat expansion diseases do not involve significant transcriptional silencing; in these cases, they are often inherited in a dominant fashion [Table 1]. In some contexts, RNA expressed from expanded repeats accumulates within cells into complexes commonly referred to as RNA foci [G] [Figure 3a]. These foci were first observed in myotonic dystrophy type 1 (DM1) patient fibroblasts and myofibers, where nuclear clumps of DMPK mRNA with expanded CUG repeats (CUG^exp) were detected by fluorescence in situ hybridization⁷⁶. Since this discovery, RNA foci have been identified as a hallmark of many repeat expansion diseases, including DM2, C9 ALS/FTD, Fuch’s endothelial corneal dystrophy (FECD), FXTAS, and many SCAs⁷⁷. These RNA foci are often retained in the nucleus, but cytoplasmic foci are also detected in some cases, such as in SCA10 fibroblasts⁷⁸ and C9 ALS/FTD neurons^79–81. Bidirectional transcription of expanded repeats produces both sense and antisense transcripts, leading to formation of both sense and antisense foci⁸². In C9 ALS/FTD, GGGGCC and CCCCGG foci can coexist within the same nuclei and have even been observed to colocalize⁷⁹. In some diseases, including C9 ALS/FTD⁷⁹ and DM1⁸³, the presence or number of RNA foci is directly correlated with onset or pathology⁸⁴.

Figure 3: Mechanisms of RNA toxicity in repeat expansion diseases.

(A) Long repetitive RNAs and RNA-binding proteins (RBPs) interact to form complex nuclear-retained RNA foci. (B) RNA foci are formed and maintained through a stochastic combination of intramolecular and intermolecular interactions. (C) A conceptual phase diagram [G] describes the thermodynamics of RNA foci in repeat expansion diseases. The transition from soluble RNA to RNA-protein phase separation is defined by the sum of RNA-RNA, protein-protein, and RNA-protein interactions (phase boundary isolines drawn as solid lines). (D) RNA processing is impaired by sequestration of RBPs on repetitive RNA, the extent of which is a cell-specific function of repeat length, host gene expression, and RBP expression. (E) Effects of nuclear retention on RBP localization can be exacerbated by autoregulatory dynamics, which may additionally disrupt cytoplasmic processes mediated by RBPs. (F) Competition between RBPs at RNA foci may modulate disease-associated sequestration. In DM2, both MBNL and RBFOX proteins bind the expanded CCUG repeat RNA, and overexpression of RBFOX partially displaces MBNL from RNA foci in muscle cells.

RNA foci in repeat expansion diseases are likely formed and maintained by a combination of intramolecular and intermolecular interactions [Figure 3b]. Through Watson-Crick and non-canonical base pairing, repetitive RNAs can form stable intramolecular secondary structures [G] in vitro and in vivo, including via imperfect stem-loops [G]⁸⁵ and G-quadruplexes [G]^86–88. For example, short (CUG)₆ RNA folds into a single stable A-form double helix [G] with U-U mismatches⁸⁹, while longer CUG RNAs likely explore an ensemble of stem-loop orientations^90,91. Intermolecular interactions may also drive repeat RNA aggregation [Figure 3b]. Through multivalent base pairing alone, (CUG)₄₇, (CAG)₄₇, and (GGGGCC)₅ RNAs undergo sol-gel transition to form phase-separated [G] RNA droplets in vitro⁹². Similar liquid-like behaviors are also observed in cell models: in C2C12 mouse myoblasts expressing (CUG)₁₄₅ RNA, spontaneous division and coalescence of foci have been detected⁹³. In addition, (CAG)₄₇ RNA foci in transfected U2OS cells are solubilized when treated with inhibitors of RNA base pairing, including ammonium acetate and doxorubicin⁹².

Repetitive RNAs are not the only inhabitants of these structures. Expanded RNAs attract RNA-binding proteins [G] (RBPs) with expected motif and/or structure preferences, and these RBPs can coat the mutant transcripts, resulting in high local concentrations. In DM1, CUG foci co-localize with the muscleblind-like (MBNL) family of RBPs⁹⁴. In C2C12 cells, knockdown of Mbnl1 by RNAi reduces aggregation of (CUG)₁₄₅ RNA⁹³, suggesting that MBNL proteins are necessary to stabilize intermolecular CUG^exp RNA interactions in a cellular context. RBPs with multiple RNA-binding domains may contribute to RNP granule stability through multivalent protein-RNA interactions as well as higher order protein-protein interactions⁹⁵. This concept, when combined with intermolecular base pairing of long repetitive RNAs, evokes a model in which RNA-RNA, RNA-protein, and protein-protein interactions each play an important part in RNA foci formation [Figure 3c]. The effects of biopolymer multivalency on phase separation dynamics are well appreciated^96–99, and RNA acts as a scaffold for RNP phase separation in many contexts¹⁰⁰. Thus, gelation of repeat RNA and RBPs in repeat expansion diseases appears to be an exacerbation of the general thermodynamic processes that normally regulate RNP granule self-assembly.

Repeat expansion RNAs can be toxic if recruitment of an RBP sufficiently depletes the protein from the nucleoplasm [Figure 3d]. The most notable example of this phenomenon is myotonic dystrophy, in which MBNL proteins are sequestered by CUG^exp or CCUG^exp RNA in the nucleus⁹⁴. MBNL proteins are global alternative splicing factors, and their sequestration and inactivation in myotonic dystrophy produces a transcriptome-wide spliceopathy¹⁰¹. In mice, knockout of Mbnl1 produces a phenotype of myotonia, myopathy, and cataracts¹⁰². Mice expressing ~250 CUG repeats in a human skeletal actin transgene (HSA^LR) develop myotonia¹⁰³ that is ameliorated by overexpression of Mbnl1 administered by AAV to skeletal muscle¹⁰⁴.

Impaired splicing has also been observed in other disease contexts, such as in SCA10 fibroblasts via sequestration of hnRNP K⁷⁸. In C9 ALS/FTD, GGGGCC RNAs associate with many proteins, including Pur-α, ALYREF, SRSF2, RanGAP1, and hnRNPs^105–108, and sequestration of hnRNP H causes mis-splicing⁸⁸. Antisense CCCCGG RNAs also interact with SRSF2, ALYREF, and hnRNPs⁸⁰. In FXTAS neurons, CGG repeat RNA accumulates within large ubiquitinated inclusions and co-localizes with DGCR8, hnRNP A2/B1, and Pur-α, and overexpression of any of these factors reduces neurodegeneration in a Drosophila model expressing (CGG)₉₀ RNA^109–111. In addition to splicing, miRNA biogenesis¹⁰⁹ and alternative polyadenylation pathways^112,113 controlled by RBPs can be undermined by sequestration.

Besides regulating RNA maturation in the nucleus, RBPs play important roles in the cytoplasm as modulators of mRNA stability¹¹⁴ and mediators of a broad RNA transport program that enables the cell to shuttle RNAs to their proper locations^115–117. Indeed, abundance and subcellular localization of RBPs are often autoregulated to confer robustness in a dynamic and stochastic cellular environment¹¹⁸. However, upon sequestration, these autoregulatory loops can shift RBP localization away from the cytoplasm to compensate for reduced nuclear activity, exacerbating the effects of RNA toxicity¹¹⁹. Mis-localization of RBPs into the nucleus could therefore disrupt both nuclear and cytoplasmic RBP functions [Figure 3e].

By inactivating a small number of nodes in the RNA processing network, non-coding repeat expansions can have potent and entangled consequences on a large number of cellular processes. However, it is important to note that the presence of RNA foci alone does not confirm a central role in disease pathology; for example, while sense RNA foci have been observed in Huntington disease¹²⁰ and SCA3¹²¹, polyglutamine proteotoxicity (see below) seems to dominate as a pathomechanism¹⁶. In addition, RNA foci may play protective roles by reducing RAN translation of toxic peptide repeats (see below) through inhibition of nuclear export^122–124.

What makes RBP sequestration such a defining mechanism of toxicity in DM1, and is it unique among repeat expansion diseases? Clearly, stoichiometry of RBP binding is a function of the number of available target sites on the RNA, and somatic instability can produce hundreds or thousands of tandem repeats in DM1 muscle and neurons^125,126. Furthermore, the toxic CUG repeats are present in mature mRNA, which likely enhances their stability relative to intronic expansions, such as the GGGGCC repeats in C9 ALS/FTD or the UUUCA repeats in familial adult myoclonic epilepsy (FAME) [Table 1]. In diseases where repeat lengths are commonly shorter or where expression of the repeat element is low across most cell types, the impact of RBP sequestration may be limited. Competition between RBPs for binding sites on expanded RNAs may also limit exhaustion of any particular protein. In DM2, RBFOX proteins also bind to the CCUG^exp RNA and actively compete with MBNL, and overexpression of RBFOX1 in C2C12 cells expressing (CCUG)₁₀₀₀ RNA partially restores Mbnl-directed splicing¹²⁷ [Figure 3f]. Perhaps this phenomenon plays a role in other diseases as well, including C9 ALS/FTD and FXTAS, in which the repeated RNA motif attracts a large number of proteins. Finally, the lack of severe pathology in mice upon knockout of Dmpk suggests that haploinsufficiency is not a substantial contributor to DM1¹²⁸; in contrast, diseases caused by expansions in essential genes, such as the SCA17-linked CAG repeat in TBP, naturally exhibit another layer of complexity, as inhibition of host gene expression is likely to produce a phenotype¹²⁹.

Protein-mediated gain of function and the role of native gene context

The formation of insoluble neuronal protein aggregates is a common pathological feature across many neurodegenerative disorders. In repeat expansion diseases, aggregated proteins play a direct role in pathology^13,16. At least nine disorders, including HD, multiple SCAs, SBMA, and dentatorubro-pallidoluysian atrophy (DRPLA), are caused by CAG repeat expansions in protein-coding sequences, resulting in expression of polyglutamine proteins (polyQ)¹³⁰ [Table 1]. Although short glutamine stretches are present throughout the proteome, larger polyQ-containing proteins undergo conformational changes to form insoluble aggregates^131–134, and these aggregates can induce cellular proteotoxicity independent of the functions of their host proteins¹³⁵. For example, in mice, expression of either HTT exon 1 containing an expanded CAG repeat or insertion of a polyQ repeat in the unrelated gene HPRT is sufficient to produce intranuclear inclusions, neuritic aggregates and neurological phenotypes, which are not observed at normal repeat sizes^135–138.

Most polyQ disease genes encode multifunctional proteins involved in various stages of gene expression, RNA metabolism, and proteostatic pathways¹³⁹. Therefore, although polyQ toxicity is shared among these conditions, disease-specific phenomena are modulated by the functions of the protein containing the expansion. This concept is best exemplified by SBMA, which is caused by a CAG expansion in the androgen receptor gene AR. AR is a transcription factor whose nuclear entry is ligand-activated. In SBMA, androgens promote translocation of the polyQ-containing AR to the nucleus, where it aggregates and triggers transcriptional dysregulation and cytotoxicity^140,141. Indeed, genetic or pharmacological blockade of androgen binding causes cytoplasmic retention of mutant AR, enhancing clearance by autophagy and ameliorating disease phenotypes¹⁴². Similarly, phosphorylation of mutant AR that inhibits ligand-activated nuclear translocation also suppresses disease-relevant phenotypes¹⁴³.

Other diseases also demonstrate how native protein context can modulate repeat toxicity. For example, SCA1 is caused by expanded CAG repeats in ATXN1, resulting in polyQ expression. ATXN1 normally shuttles between the nucleus and cytoplasm and plays active roles in gene regulation¹⁴⁴. When expanded, polyQ ATXN1 is enriched in the nucleus, where it can interact with nascent RNAs and protein regulators of transcription, including the repressor Capicua (CIC), a critical mediator of toxicity¹⁴⁵. Similarly to SBMA, mice with mutations in the NLS domain of polyQ ATXN1 do not develop disease phenotypes, suggesting that nuclear localization is critical for pathogenesis¹⁴⁶. Furthermore, mutation of phosphorylation site S776 to a phosphomimetic aspartic acid impedes its interactions with the 14-3-3 chaperone in the cytosol, resulting in nuclear translocation and toxicity¹⁴⁷. Conversely, replacement of S766 to a phospho-dead alanine prevents neuronal toxicity^148,149.

Ultimately, while native protein context importantly influences pathology unique to each disease, symptoms and clinical outcomes of polyQ diseases begin to converge at longer expansion lengths, with more prominent pathology and earlier onset of motor dysfunction, dystonia, parkinsonism and dementia^16,150,151. Accordingly, transgenic polyQ disease models with large repeats outside of their normal protein context tend to show diffuse patterns of neurodegeneration¹³⁵. It thus appears that at larger repeat sizes, disease-specific contributions of repeat-harboring genes are overwhelmed by proteotoxic impacts of polyQ expression.

Repeat-associated non-AUG (RAN) translation

Dominantly inherited diseases caused by repeat expansions located outside protein-coding regions were initially thought to manifest solely via haploinsufficiency or RNA gain of function. However, the seminal discovery of non-AUG-initiated translation of repetitive elements raised the possibility of yet another mechanism of pathology¹⁵² [Figure 4]. During study of CAG repeats in SCA8, a serendipitous observation was made that an AUG start codon was not required to generate polyQ protein from an ATXN8 minigene, even in the presence of multiple stop codons upstream of the repeat. Repeat-associated non-AUG (RAN) translation [G] was observed in all three reading frames to produce three homopolymeric proteins: polyQ, polyS, and polyA¹⁵². CUG repeat RNA also supported RAN translation from reporter constructs, but a CAA repeat did not, suggesting that secondary structure of the repeat RNA may be required¹⁵². Subsequently, RAN translation was described at repeat loci associated with FXTAS, C9 ALS/FTD, FECD, DM1, DM2, Huntington disease and multiple SCAs^82,153–159. For many of these diseases, RAN translation occurs on both sense and antisense transcripts, and RAN peptides accumulate in patient tissues^{82,122,157,158,160,161}.

Figure 4: Mechanisms of repeat-associated non-AUG (RAN) translation.

(A) Canonical AUG-mediated initiation and some forms of RAN translation require binding of eIF4F complex (eIF4E, eIF4G and eIF4A) to the 5’ m⁷G cap with eIF4B and/or eIF4H. After assembly, the 43S pre-initiation complex (PIC) scans 5’ to 3’ along the mRNA until selecting an AUG or near-AUG codon (for example: CUG) for initiation. eIF2α phosphorylation (eIF2α-P) under stress blocks ternary complex recycling and inhibits canonical translation, but allows for continued RAN translation. RBPs regulate RAN initiation by binding and altering repeat RNA structures. Known RAN-associated factors are depicted with solid lines, while canonical and IRES initiation factors involved in RAN translation are depicted with dashed lines. (B) RAN translation may also initiate through IRES-like mechanisms in a cap-independent manner, supported by RPS25 and other IRES-trans acting factors (ITAFs). (C) RAN translation from the C9ORF72 GGGGCC sense and CCCCGG antisense transcripts generates multiple dipeptide repeats (DPRs). While all DPRs are detected in patient tissues or generated by cellular reporters, arginine-containing DPRs show the highest intrinsic toxicity in model systems. (D) Stable RNA secondary structures formed by GGGGCC repeats induce ribosomal frameshifting during RAN translation, leading to production of chimeric DPRs. 40S and 60S = ribosomal subunits, eIF = eukaryotic initiation factor, IRES = internal ribosome entry site, m7G = 7-methylguanosine, PKR = Protein kinase R, RBPs = RNA-binding proteins, uORF = upstream open reading frame.

Initial studies of RAN translation suggested functional overlap with canonical mechanisms of translation initiation [Figure 4a]. Translation is typically a highly regulated step-wise process that starts with binding of the eukaryotic initiation factor complex 4F (consisting eIF4E, eIF4G, and eIF4A) to the 5′ m⁷G cap of the mRNA, recruitment of the small 40S ribosomal subunit, eIF2 binding to methionine tRNA (Met-tRNA^Meti) to form the preinitiation complex (PIC), and finally scanning of the assembled PIC along the mRNA¹⁶². Recognition of the AUG start codon is promoted by eIF5, which triggers eIF2-GTP hydrolysis and initiation factor release, coupled with 60S subunit recruitment and formation of the first peptide bond. RAN translation at FMR1 CGG repeats proceeds efficiently in two of the three reading frames and is largely cap-dependent, requiring eIF4A in both cell-free systems and transfected cells¹⁶³. In the GGC (polyG) reading frame, initiation occurs predominantly at near-cognate start codons [G] (ACG or GUG) just 5’ of the repeat^163,164. In the GCG (polyA) reading frame, initiation occurs within the repeat itself, akin to observations at SCA8 CAG repeats¹⁵². Initiation in both reading frames is suppressed by overexpression of eIF1, which favors AUG codon usage, and enhanced by overexpression of eIF5, which relaxes the stringency of start codon selection¹⁶⁵. Thus, RAN translation at CGG repeats mimics upstream open reading frames (uORFs) [G] that are ubiquitous in human genomes¹⁶⁶ and results predominantly from repeat-induced decrements in start codon fidelity [Figure 4A].

However, at other repeats, findings suggest contributions from alternative initiation mechanisms. Sense and antisense transcripts from expanded C9ORF72 are translated into five different dipeptide repeat (DPR) peptides: polyGA, polyGR, polyGP, polyPR, and polyPA^{122,154,155,160} [Figure 4C]. As in the FMR1 CGG repeat, RAN translation of GGGGCC repeats within monocistronic mRNA reporters shows strong dependence on the 5’ m⁷G cap and on eIF4A-dependent scanning^167,168. Similarly, a near-cognate CUG codon just 5’ of the repeat is critical for initiation in the polyGA reading frame^167–170. However, studies using bicistronic reporters support RAN translation in all reading frames in a repeat length-dependent fashion^169,171, suggesting that cap-independent internal ribosomal entry site [G] (IRES)-like initiation can occur within GGGGCC RNAs. Classically described in viral mRNAs, IRES-mediated translation bypasses the need for a 5’ cap by directly recruiting initiation factors and ribosomal components onto a structured mRNA sequence¹⁷². In some cases, the mRNA itself can mimic the initiator tRNA to enable initiation in the absence of any AUG or near-AUG codons. Such a mechanism may explain how intronic repeats could be translated in C9 ALS/FTD patient neurons. In support of this model, knockdown of RPS25, an IRES-associated 40S ribosomal subunit, strongly and selectively modifies RAN translation of GGGGCC and CAG repeats¹⁷³ [Figure 4B].

RAN peptides generated from CGG, CAG, GGGGCC and CCCCGG repeats cause toxicity in various model systems^{153,156,160,174–176}. Overexpression of pathogenic CGG repeats that support RAN translation of FMRpolyG leads to toxic phenotypes in cultured neurons, fly, and mouse models of FXTAS^153,164,177. This toxicity depends on the ability of these repeats to be translated, as mutation of the near-AUG codons that support translation strongly suppresses phenotypes in flies and transgenic mice, while mutating these near-cognate start codons to AUG boosts FMRpolyG production and toxicity^{153,164,178,179}. Moreover, antisense oligonucleotides (ASOs) [G] that selectively impede RAN translation of FMRpolyG enhance survival in human neurons with expanded CGG repeats¹⁸⁰.

In C9 ALS/FTD, AUG-initiated expression of DPRs alone in the absence of the native repeat RNA sequence (accomplished via use of alternative codons for the DPR protein sequence) elicits toxicity that is equivalent to or often greater than that of the pure tandem repeats themselves^{175,176,181,182}. Arginine-rich DPRs (polyGR and polyPR) in particular are highly toxic, inducing cell death in cultured primary and iPSC-derived neurons as well as eye degeneration and early mortality in Drosophila^{175,176,183–191}. In mice, expression of polyGR or polyPR in isolation leads to early-onset and severe neurodegeneration with motor dysfunction and memory impairment^189–191. PolyGA DPRs may also be important in disease, as they form filamentous amyloid-like structures and are moderately toxic in mammalian neurons^{176,183,192–194}, while polyGP and polyPA exhibit limited toxicity in model systems. The mechanisms by which DPRs elicit toxicity is extensively discussed in recent reviews^195–197.

Despite observations that DPRs and other RAN-generated proteins are sufficient to elicit toxicity, most of the evidence demonstrating their roles in disease pathogenesis to date have relied on overexpression systems, which may not faithfully recapitulate actual disease states. The relative toxicity of DPRs do not correlate with their relative rates of production in reporter systems^167,168,171 or, for the most part, their relative abundance and sites of accumulation in pathological analyses^198–201. Thus, it is not yet clear whether repeat RNA or DPRs alone are sufficient to explain human disease phenotypes. Studies selectively expressing DPRs at levels equivalent to the endogenous state or selective blockade of RAN translation in model systems are still needed to delineate these roles.

Modifiers of RAN translation

Multiple translation initiation factors, including orthologs of eIF4E, eIF4B, eIF4H, eIF5, eIF3D1 and eIF3I, modulate GGGGCC RAN translation in Drosophila²⁰², and eIF3F modifies RAN translation of CAG and GGGGCC repeat reporters in mammalian cells²⁰³ [Figure 4A–B]. The eIF4A helicase co-stimulatory factors eIF4B and eIF4H also suppress CGG RAN translation in flies and mammalian reporter systems¹⁶⁵, suggesting convergence around the activity of this helicase. Consistent with this, DDX3X, a DEAD-box helicase that binds CGG and GGGGCC repeat RNA, emerged as a key modulator of RAN translation from two independent screens^165,204. However, the effect of DDX3X on RAN translation is complex: knockdown of DDX3X or its homolog in Drosophila inhibits CGG RAN translation and repeat-associated toxicity¹⁶⁵, but it enhances GGGGCC RAN translation and toxicity in multiple systems, including patient iPSC-derived neurons²⁰⁴. These differences may reflect dependence on sequence context surrounding the repeats. The FMR1 5’ UTR is highly GC-rich outside the CGG repeat region; thus, DDX3X helicase activity is likely required to facilitate initiation at the canonical FMRP start codon even in the absence of expanded repeats. In contrast, if an IRES-like mechanism is critical for RAN initiation on GGGGCC repeats, then the loss of DDX3X may induce formation of RNA structures that support IRES mediated-ribosomal recruitment.

External stimuli, such as ER stress, viral infection and amino acid starvation, activate the integrated stress response (ISR), which leads to stress granule [G] formation and inhibition of global translation through phosphorylation of eIF2α^205–207 [Figure 4A]. A subset of transcripts that initiate using non-AUG codons or via IRES-dependent mechanisms escape this translational suppression. Accordingly, ISR activation significantly enhances RAN translation of both CGG and GGGGCC reporters in a process dependent on eIF2α phosphorylation^{167,169–171,208}. In C9ORF72, stress-induced escalation of GGGGCC RAN translation occurs with both cap-dependent and cap-independent bicistronic constructs^{167,169–171,208}. Repeat RNAs and DPR proteins can independently activate the ISR, creating a potential positive feedback loop where cellular stress enhances RAN translation, which in turn elicits further stress¹⁶⁷. Consistent with this concept, pharmacological or genetic suppression of PKR, which phosphorylates eIF2α to activate the ISR, reduces RAN translation in cells and improves disease phenotypes in a RAN mouse model of C9 ALS/FTD²⁰⁸. Similarly, knockout of the alternative initiation factor eIF2A, which allows for initiation when eIF2α is phosphorylated, partially suppresses RAN-initiated polyGA expression in C9 ALS/FTD models¹⁶⁹ [Figure 4a].

Mechanistic synergy may drive disease

As alluded to in several cases above, multifactorial pathomechanisms may perhaps be the rule rather than the exception. However, each of the four mechanisms previously described has most commonly been studied separately, either due to limited availability of appropriate disease models and reagents, or to simplify interpretation by separating confounding variables. Indeed, the pathomechanisms described above are roughly chronological in their discovery, and newer mechanisms often have not been thoroughly investigated in diseases for which earlier mechanisms have provided reasonable explanations for pathology. More recently, new animal models and combinations of models have facilitated deliberate investigation of combinatorial effects. Several examples of multifactorial interactions among disease pathways are described below [Figure 5].

Figure 5: Synergy across pathogenic mechanisms in repeat expansion diseases.

(A) In multiple diseases, the four major mechanisms detailed in this review can co-exist and/or synergize to drive complex pathology. For example, in C9 ALS/FTD, expanded GGGGCC repeats can induce intron retention, which leads to haploinsufficiency of C9ORF72, as well as exacerbates RBP sequestration by increasing the half-life of the repeat RNA. In addition, intron retention may increase the production of dipeptide repeats that activate numerous downstream pathogenic pathways. In DM2, expanded CCTG repeats lead to intron retention, which also results in reduction of mRNA available to generate full-length CNBP protein. RAN translation products can be generated from the intron-retained mRNA. In Huntington disease (HD), expanded CAG repeats alter RNA processing to impair recognition of the exon 1 donor splice site; this results in the formation of a truncated polyQ-containing HTT protein that is more toxic than full-length polyQ-containing HTT. RAN translation can also occur across the CAG repeat. In FXTAS/FXS, the CGG repeat can not only sequester RBPs, but can also enhance RAN translation of the uORF such that translation initiation for FMRP is reduced. (B) A more detailed view of pathways activated in C9 ALS/FTD shows that some pathogenic mechanisms can exacerbate or feed into other mechanisms. A complex network of cause and effect, including feed-forward loops, may synergize to drive disease pathology. (C) A more detailed view of pathways activated in HD also similarly reveals feedback loops in both the nucleus and cytoplasm.

Loss or gain of protein function due to alterations in RNA processing

Repeat expansions can alter RNA processing, and this phenomenon occurs in multiple repeat expansion diseases, resulting in intron retention^209,210, changes in alternative transcription start site usage^211,212, or premature polyadenylation²¹³. Alterations to RNA processing can change the repertoire of transcribed isoforms, with some leading to LOF and others leading to production of proteins with GOF activity. For example, repeats can trigger intron retention, allowing for efficient export of repeat RNA into the cytoplasm¹⁷¹, where it may elicit toxicity directly and/or undergo RAN translation. Simultaneously, intron retention can trigger depletion of the full-length protein product. As such, haploinsufficiency can potentially compound the effects of expanded repeat expression, even when haploinsufficiency alone may be benign. In C9 ALS/FTD, multiple studies indicate that loss of C9ORF72 can synergize with GGGGCC expression to exacerbate symptoms in both C9BAC mice^214,215 as well as an AAV-based model²¹⁶. In contrast, in Huntington disease, expanded CAG repeats impair recognition of the donor splice site of HTT exon 1, leading to premature polyadenylation and translation of a truncated polyQ-containing HTT peptide²¹³. This peptide may be much more toxic than full-length HTT containing expanded polyQ. Whether alterations to RNA processing and protein gain- or loss-of-function effects modulate symptoms in other repeat expansion diseases remains to be fully explored.

RNA gain of function and RAN translation

Although DM1 and DM2 have served as excellent examples in which repeat RNA acquires new functions to sequester RBPs, a clear pathogenic role for RBP sequestration has not been fully established in many other diseases. Many RBPs associate with GGGGCC, CAG, CGG, and other repeat RNAs, but in these contexts, the repetitive RNAs also generate pathogenic peptides. Efforts have been made to separate these effects, for example by using alternative codons to preserve protein sequence yet disrupt the simple tandem RNA repeats. In BACHD mice, a model of Huntington disease, a clear role for polyQ toxicity is suggested by the observation of progressive neurodegeneration upon CAGCAA repeat expression²¹⁷. However, other studies highlight a role for RNA toxicity independent from protein²¹⁸. For example, targeting CAG RNA with locked nucleic acid ASOs can ameliorate phenotypes in the R6/2 mouse model of Huntington disease, even when Htt protein is not perturbed²¹⁹. In DM2, in which a role for RNA toxicity is well established, RAN peptides polyLPAC and polyQAGR also occur in various brain regions¹⁵⁷ and may be modulated by the extent to which MBNL associates with CCUG RNA and prevents it from exiting the nucleus. However, clear pathogenic roles for these peptides remain to be defined. In C9 ALS/FTD, GGGGCC RNAs bind to hnRNP H and SRSF proteins and trigger splicing changes^88,106,220, but it remains unclear whether RBP sequestration synergizes or competes with the effects of DPRs. A potential interaction between RNA sequestration and RAN translation has been proposed, in which GGGGCC RNAs associate with nucleolin⁵⁸, which can also associate with DPRs¹⁷⁵. In contrast, SRSF protein binding to GGGGCC repeats appears to influence the nucleocytoplasmic transport of repeat RNA and thus titrates its ability to undergo RAN translation²²¹.

Production of multiple proteins from a single repeat-containing message

Because RAN translation can initiate in a variety of reading frames²²², the contribution of each potential peptide repeat in disease has been challenging to separate. In C9 ALS/FTD, studies of each potential DPR have suggested that some DPRs are more toxic than others¹⁹⁵. However, additional mechanisms such as frameshifting [G]¹⁶⁸ may generate chimeric species and further complicate pathomechanisms [Figure 4D]. In vitro, mutation of the near-cognate CUG start codon that initiates polyGA translation modulates translation efficiency of not only polyGA, but of other reading frames as well¹⁶⁸. Indeed, polyGA:polyGP chimeric peptides, presumably produced by frameshifting events, accumulate in C9 ALS/FTD and may account for differential toxicity compared to polyGP produced in SCA36²²³. Even in HD, the presence of non-canonical protein species from both sense and antisense transcripts^{156, 224–227} raises questions about whether polyQ is the sole protein-based driver. Sorely needed are basic studies to probe fundamental mechanisms of RAN translation initiation and frameshifting, as well as careful studies using disease samples to characterize the distribution and abundance of each potential species, to gain clarity on pathogenesis.

Further work is needed to determine how, and if, RAN proteins contribute to pathogenesis in humans. Most studies to date have relied on overexpression systems or peptides generated from AUG-initiated constructs. This ignores key elements related to the inefficiency of RAN translation from different repeats and the endogenous stoichiometry of their protein products in patient tissues. For example, while polyGR and polyPR are more toxic in isolation, polyGA is more efficiently translated in the absence of an AUG start codon and is more abundant in human patient brains^{155,167–171,198}. PolyGA may act as a seeding factor for aggregation of other C9ORF72 RAN peptides such as polyGP and polyPA, which are otherwise soluble, or as a suppressor of polyGR toxicity¹⁸⁶. In FXTAS, the absolute abundance of FMRpolyG in patients may be low, and its presence does not always correlate with relevant phenotypes in mice^228,229. Moreover, both repeat RNA and RAN proteins are found in FXTAS inclusions, and these molecules may directly interact²³⁰. Thus, for each disease and RAN product, their relative contributions to toxicity will need to be correlated with their abundance and interaction with relevant pathways.

Mechanisms of tissue-specific pathogenesis

Despite often shared repeat sequences, repeat expansion diseases are largely syndromic, with incomplete overlap for a given repeat unit and its clinical presentation or its cell-type specific dysfunction¹⁶. In DM1, for example, while myotonia is a widely recognized characteristic symptom, many patients report that neurological symptoms, such as fatigue, hypersomnolence, and cognitive difficulties, are more debilitating²³¹. Differences in pathogenicity of repeat expansions across tissues emerge from multiple interacting variables, including differential rates of somatic instability, host gene expression, and expression of trans factors. Somatic instability of the CTG repeat in DM1 produces alleles in skeletal muscle and brain as much as 13 times larger than in leukocytes^125,232, likely exacerbating RNA toxicity in those tissues. Further, although neuromuscular symptoms are often more severe in DM1 than in DM2, alternative splicing biomarkers in blood are more pronounced in DM2, likely as a result of higher expression of CNBP in this tissue than DMPK²³³.

Recently, six of the seven known subtypes of familial adult myoclonic epilepsy (FAME, or BAFME), a slowly progressing disease primarily exhibiting myoclonic seizures and cortical tremor, were mapped to long intronic TTTCA expansions in various genes^234–237. While much remains unknown about the pathogenesis of FAME, RNA toxicity has been proposed to play a role, as host gene expression appears generally unaffected and nuclear accumulations of UUUCA RNA have been observed in FAME1 patient brain²³⁴. Interestingly, an intronic TTTCA expansion is also associated with SCA37, an entirely different disease characterized by gait instability, limb ataxia, and nystagmus²³⁸. If GOF mechanisms of a TTTCA expansion drive pathogenesis in both diseases, how do they differ so extensively in disease symptoms? A leading hypothesis is that host gene expression patterns may partially explain this phenomenon, as while the FAME-linked genes are generally expressed throughout the brain, expression of the SCA37-linked DAB1 gene is more specific to the cerebellum²³⁵.

Still, much about the precise mechanisms of tissue-specific pathology remains unclear. For example, in Huntington disease, HTT is expressed in most tissues, and while its expression is indeed highest in the nervous system, HTT protein is not confined to areas susceptible to neurodegeneration²³⁹. In fact, while many other polyQ diseases cause pronounced cerebellar ataxia, this symptom is considered rare in HD, even though mutant HTT inclusions are observed in the cerebellum²⁴⁰.

A broader role for short tandem repeats

Most of the >50 disease-causing repeat expansions identified to date were found in association with highly penetrant clinical syndromes, which allowed for careful genetic analysis and exclusion of other mutations as potential causes. However, these relatively rare syndromic conditions may represent the tip of a much larger iceberg. STRs account for ~3% of genomic DNA, and a significant fraction reside within genes and their regulatory regions^1,241 [Figure 6a]. As such, STRs can impact the structure and function of DNA, RNA, and proteins, with a range of molecular and cellular consequences. While such sequences have traditionally been viewed as “junk” DNA, evolutionary genomics suggests that STRs and other classes of DNA repeats have evolved under tight selective constraints²⁴².

Figure 6: Roles of repeats in human disease and neuronal function.

(A) Short tandem repeats represent ~3% of the human genome, with enrichment of specific elements within 5’ UTRs, ORFs, and introns. STR mutation rates are orders of magnitude higher than single nucleotide polymorphisms and their size can influence gene expression. (B) CAG repeats in ATXN2, which when fully expanded cause spinocerebellar ataxia type 2 (SCA2), act as risk alleles for development of ALS and other neurodegenerative disorders when the repeats are of intermediate size. Loss of ATXN2 suppresses ALS phenotypes in model systems. (C) The normal-length CGG repeat in FMR1, which when expanded causes FXS and FXTAS, serves to regulate translation of the FMR1 gene product, FMRP, in response to synaptic stimuli.

All of these factors suggest that variations in STRs could serve as risk alleles for non-Mendelian human disorders^10,11. Consistent with this concept, some disease-associated repeats have been identified as risk alleles for other neurological conditions at smaller expansion sizes (Figure 6B). For example, intermediate-length CAG expansions in the ATXN2 gene, which at larger sizes are associated with SCA2, serve as a common risk allele for development of ALS and other neurodegenerative disorders²⁴³. A similar relationship with ALS and FTD phenotypes was recently identified for intermediate CAG expansions in ATXN1^244–246 and for both intermediate and full mutations in HTT²⁴⁷. Recently, polyalanine expansions in NIPA1 and a more complex intronic tandem repeat in WDR7 were also associated with ALS with incomplete penetrance^248,249, as was a CGG repeat expansion in NOTCH2NLC²⁵⁰.

While it is possible that these disease loci are unique examples for how variation in STR length might contribute to multiple diseases, recent studies suggest that other STRs broadly contribute to multiple diseases and biological phenotypes. First, improved measurement of STR length and instability genome-wide with standard next generation sequencing platforms^4,251 suggests that a significant fraction of the signal for SNP markers used in genome-wide association (GWAS) studies may derive from tight linkage with STRs that modulate neighboring gene expression and RNA processing⁴. Second, gene-associated tandem repeat expansions at 2,588 loci are more prevalent among individuals with autism than their siblings or controls, particularly in exons and near splice junctions of genes related to nervous system development^252,253. Future studies aimed at linking STR variation to disease risk across a broad spectrum of human disease are likely to be fruitful.

Emerging data suggests that STRs likely perform important native functions in the genes in which they reside, implying that their loss may also contribute to human disease^10–12. The majority of repeat expansion disorders identified to date impact the nervous system, suggesting that STRs may have particular roles in these specialized cell types. STRs mutate more rapidly than single nucleotides, enabling them to hasten evolutionary adaptation by acting as tunable regulators of gene expression and function²⁵⁴. For example, yeast take advantage of the relative instability of repeats to allow proteins to have pleiotropic behaviors across a population of cells in response to environmental stressors^255,256.

As such, the native roles of repeats in humans remain understudied, even in the context of known disease-causing loci. At normal sizes, CAG-encoded polyQ repeats in proteins such as HTT serve as flexible hinges that link functional domains²⁵⁷. This flexibility enables the multifunctional proteins to play key roles in events mediated by large biomolecular complexes, such as transcriptional regulation. As an example, elimination of CAG repeats in ATXN3, a deubiquitinase with roles in autophagy, impedes its normal function in cells²⁵⁸. The CGG repeat in FMR1 plays an active role in regulating FMRP translation at normal repeat sizes¹⁸⁰. This CGG repeat serves as a RAN-translated uORF in the FMR1 5’ UTR that suppresses basal FMRP translation, while allowing for upregulation of FMRP expression in response to specific stimuli (Figure 6C). This finding is particularly intriguing, as nearly 100 other human genes, many involved in neuronal function, have CGG or CCG repeat elements in their 5’ UTRs.

Approaching newly discovered diseases

As a result of continuous technological improvements in recognition and detection, 28% of known disease-associated repeat expansions were mapped in the last four years alone, and more will undoubtedly be discovered. As new associations emerge, the advancements made over the past three decades should enable rapid evaluation of potential disease mechanisms and streamlining of therapeutic development. Initial assessments should draw comparisons to similar repeat loci and to diseases with similar clinical syndromes and pathologies. As an example, the discovery of a CCTG repeat expansion as the cause of DM2²⁵⁹ was rapidly followed by assessment of the association of MBNL proteins with CCUG RNA and identification of similar MBNL splicing abnormalities between DM1 and DM2²⁶⁰. However, the two repeats exhibit differential affinities for RBPs and reside in different genomic contexts. These aspects may help to explain clinical differences between the conditions and the lack of a congenital phenotype in DM2¹²⁷.

Mechanistically, the first key question to answer for each new disease-associated repeat is whether dysfunction occurs primarily through GOF or LOF mechanisms, while recognizing that both modes may act in concert or in competition. This step is critical, as the emergence of modular technologies aimed at knockdown (eg. ASOs and siRNAs) or upregulation (eg. gene therapy or CRISPR-mediated activation) of specific genes can enable rapid movement toward therapies once this separator is defined. However, even when a primary loss of function is clear, it is important to consider the potential negative impacts of gene reactivation, as upregulation of repeat mRNA or RAN-translated proteins might elicit additional toxicity. In certain contexts, combinatorial approaches may be needed to target simultaneous pathomechanisms²⁶¹.

Initial assessments in simple model systems, such as human and rodent cell lines, Drosophila, and C. elegans, have reliably generated valuable insights into these modes of toxicity and are useful for rapid screens of genetic modifiers and suppressors of relevant phenotypes. These preliminary observations are often robust and conserved across phylogenetic lines, but a rigorous approach must be taken to confirm such findings in more complex in vivo models, including rodents, larger mammals and human iPSC-derived neurons or organoids, which express the repeats from endogenous loci. Appropriate validation of the relative contributions from each branch of the repeat toxicity tree (DNA, RNA, AUG-initiated translation, and RAN translation) early in the therapeutic pipeline will significantly streamline development.

As an example of how this approach might be applied, we can examine a recently discovered GGC repeat expansion in the 5’ UTR of NOTCH2NLC^262–264 [Table 1]. Expansion of this repeat from normal lengths (~20 GGCs) to >50 GGCs causes neuronal intranuclear inclusion disease (NIID) in East Asian populations, and this expansion has been linked to a number of other neurological disorders. Even prior to the discovery of a shared repeat motif, its pathological, radiological and clinical phenotypes were noted to overlap with FXTAS²⁶⁵. This suggests shared pathomechanisms between these two conditions, with specific differences likely dependent on the properties of the repeat-harboring genes. In silico analysis of the sequence just 5’ of the repeat in NOTCH2NLC reveals an AUG codon that could generate a polyG protein, resembling the FMRpolyG that is produced by RAN translation in FXTAS. Given that FMRpolyG readily forms inclusions in FXTAS patient tissue and that AUG-initiated translation is more efficient than RAN translation, this observation suggests that polyG proteotoxicity may be a major pathomechanism in NIID. If this is the case, then therapeutic strategies in development for FXTAS could be explored early on for NIID to determine whether mechanistic convergence might lead to a clinical breakthrough.

Future perspective

After three decades of research on nucleotide repeat expansion disorders, we now have a roadmap for many of the central mechanisms that drive disease pathogenesis. Yet, it remains important to stay humble in the face of what we do not know. A salient example is cerebellar ataxia, neuropathy, and vestibular areflexia syndrome (CANVAS), caused by a recently discovered biallelic expansion of intronic AAGGG repeats in RFC1^266,267. This condition is unusual in at least two ways. First, the expansion often accompanies a shift in repeat sequence from AAAAG or AAAGG to AAGGG, but the exact repeat motif and structure varies geographically among patient populations^268–270. Interestingly, the repeat tract falls within the poly(A) tail of an AluSx3 element, raising the possibility that genomic instability engendered by a retrotransposon may drive pathogenic expansion in RFC1. Second, despite the autosomal recessive inheritance pattern of CANVAS, the repeat does not appear to impact expression of RFC1 protein²⁶⁶, making loss of function less likely. Yet, to date, studies of pathology have not revealed evidence of RNA foci or aggregated proteins in affected tissues, drawing classical gain-of-function mechanisms into question as well. The pathomechanisms that drive CANVAS, a disease caused by a unique and complex repeat expansion, remain undefined and yearn to be understood. Based on the past three decades, we expect that the solution to this newest conundrum will again change the way we think about expanded repeats and human disease.