Replication dependent and independent mechanisms of GAA repeat instability

Abstract

Trinucleotide repeat instability is a driver of human disease. Large expansions of (GAA)_n repeats in the first intron of the FXN gene are the cause Friedreich’s ataxia (FRDA), a progressive degenerative disorder which cannot yet be prevented or treated. (GAA)_n repeat instability arises during both replication-dependent processes, such as cell division and intergenerational transmission, as well as in terminally differentiated somatic tissues. Here, we provide a brief historical overview on the discovery of (GAA)_n repeat expansions and their association to FRDA, followed by recent advances in the identification of triplex H-DNA formation and replication fork stalling. The main body of this review focuses on the last decade of progress in understanding the mechanism of (GAA)_n repeat instability during DNA replication and/or DNA repair. We propose that the discovery of additional mechanisms of (GAA)_n repeat instability can be achieved via both comparative approaches to other repeat expansion diseases and genome-wide association studies. Finally, we discuss the advances towards FRDA prevention or amelioration that specifically target (GAA)_n repeat expansions.

1. Introduction

DNA microsatellites, 1-to-9 base-pair long tandemly duplicated sequences, comprise up to 3% of the human genome [1–3]. Expansions of a subset of microsatellites are associated with over 50 repeat expansion diseases (REDs), and the number is ever-growing [4]. Disease-associated repeats are characterized by length variability (expansions and contractions) and are known to induce fragility and repeat-induced mutagenesis, resulting in genomic instability (for an extensive review, see [5]). In the first years after their discovery in 1991 [6], most repeat expansion diseases were identified as autosomal dominant disorders associated with expansions of (CNG)_n repeats, resulting in a toxic gain of function at the protein level. (CNG)_n expansion diseases include Huntington’s disease (HD), spinocerebellar ataxias (SCAs), myotonic dystrophy (MD) and Fragile X syndrome (FXS).

All expandable (CNG)_n repeats can form imperfect hairpins stabilized by CG base pairs or slipped strand DNA structures that result from the formation of two such hairpins in complementary DNA strands [7]. Thus, strand slippage and hairpin formation during DNA replication was initially proposed to be at the center of trinucleotide repeat (TNR) instability [8]. Therefore, the discovery of the genetic basis of Friedreich’s ataxia (FRDA) in 1996 came as a surprise [9]. FRDA is an autosomal recessive disease caused by the expansion of a (GAA)_n repeat, which, unlike (CGN)_n repeats, cannot form a hairpin structure [9]. Overall, (GAA)_n runs are amongst the most expansion-prone trinucleotide repeats in the human genome, the majority of which originated from 3′ poly(A) tracts of various Alu elements upon AAA to GAA transition [10], [11]. In the case of FRDA, however, the (GAA)_n repeat originated from the A_n(TAC)A_n sequence at the center of the Alu Sq element located in the first intron of the FXN gene [10].

FRDA is the most common form of hereditary ataxia in humans [12]. (GAA)_1-33 repeats are in the normal range, (GAA)_34-65 repeats are pre-mutational, and longer repeats are pathogenic. The length of the (GAA)_n repeat positively correlates with the age of disease onset, its severity and progression [13–17]. (GAA)_n expansions ultimately lead to chromatin changes and FXN gene silencing, resulting in a drastic reduction in the levels of the mitochondrial protein frataxin [18–22]. Reduced frataxin levels lead to increased oxidative stress, accumulation of iron species in the mitochondria and subsequent cell death, primarily affecting neuronal tissues [18], [23–25]. The loss of gene function upon repeat expansions accounts for the recessive mode of inheritance. FRDA is associated with cerebellar and sensory ataxia, diabetes mellitus, and cardiomyopathies, leading to early death (for a clinical review, see [26]). There is currently no effective cure or treatment for FRDA [27].

2. (GAA)_n repeats form triplex DNA structures during transcription and replication

Being a homopurine-homopyrimidine (hPu/hPy) mirror repeat, (GAA)_n runs can assume the non-B DNA structure termed DNA triplex or H-DNA. H-DNA is formed when a DNA strand corresponding to one half of the repeat folds back forming a triplex with the duplex half of the repeat via Hoogsteen base pairing, while its complementary strand remains single-stranded (Fig. 1A–D) [28], [29]. Like other alternative DNA structures, H-DNA is thermodynamically unfavorable in linear double-stranded DNA, but becomes favorable in negatively supercoiled DNA, which is topologically equivalent to unwound DNA, as it relieves torsional stress [30], [31]. Consequently, H-DNA is not a steady-state presence in genomic DNA but can rather form at different stages of the cell-cycle during specific genetic processes such as replication and transcription, hence it is a dynamic DNA structure.

Figure 1:

Various types of triplex DNA structures formed by long (GAA)_n repeats. Each H-DNA conformation, YRY or RRY, can exists in two isoforms, depending on whether the 3’ or the 5’ of a strand is donated to the triplex. Homopurine strands are in orange, homopyrimidine strands are in blue. Dashes indicate Watson-Crick base-pairing, circles indicate Hoogsteen or reverse Hoogsteen base-pairing, asterisks indicate protonated cytosines. See Section 2 for more detail.

H-DNA can exist in several conformations [32]. In an H-y triplex (YRY), the pyrimidine strand contributes to triplex formation by Hoogsteen base pairing (Fig. 1A–B). Normally, H-y formation is favored at low pH since it requires cytosine protonation, but because of their high AT-content, (GAA)_n repeats can readily form H-y triplexes under physiological conditions [31], [33]. In an H-r triplex (RRY), the purine strand contributes to triplex formation through reverse Hoogsteen base pairing (Fig. 1C–D). H-r triplexes, including those formed by the (GAA)_n repeat, were observed at neutral pH in the presence of divalent cations [34], [35]. Finally, two distant (GAA)_n repeats within the same supercoiled DNA molecule can form a structure termed sticky DNA (Fig. 1E) [34], [36]. In this case, a homopurine strand from one of the repeats forms an H-r triplex with another repeat, also in the presence of divalent cations [34], [37].

While all these structures can be formed under specific conditions in vitro, it remains unclear which one is the most common at the FRDA locus and how each contributes to repeat instability and disease. It is generally challenging to detect dynamic DNA structures at endogenous genomic loci in vivo, given that they may only be formed transiently. It is particularly difficult in mammalian cells, in which chemical probing has proven to be highly cumbersome, partially due to the extreme genome size (reviewed in [38]). Triplex-specific antibodies were shown to bind in situ to multiple sites in human chromosomes, some of which contained (GAA)_n repeats [39–41]. Note however, that the resolution power of this technique is not at the nucleotide level.

As discussed above, negative DNA supercoiling is the main driver for H-DNA formation in vitro. In mammalian nuclei, transcription is the main source of negative DNA supercoiling ([42] and references therein). Kouzine et al. used permanganate treatment, which oxidizes thymine residues in single-stranded DNA (ssDNA), in combination with S1 nuclease cleavage and high-throughput sequencing to detect non-B DNA structures in the genome of mouse B-cells [43]. This study identified approximately 17,000 sites of H-DNA out of ~728,000 predicted H-DNA motifs, the prevalence of which positively correlated with transcription levels (Fig. 2A) [43]. More recently, an S1-seq based method, which relies on mapping of S1-cleavage sites in permeabilized cells [44], identified about 144,000 H-DNA forming structures, preferably H-y5 triplexes (Fig. 2B) [45]. Notably, most H-DNA motifs were observed at relatively short homopurine-homopyrimidine repeats, and the authors believe that they could have formed ex vivo during sample preparation owing to the acidic pH of the S1-nuclease reaction buffer. In FRDA patient cells, a similar S1-END-seq approach revealed H-DNA at the expanded (GAA)_n repeats specifically when the FXN locus was transcribed (Fig. 2C) ([46] preprint). It is concluded, therefore, that in this case, the triplex is formed in vivo during transcription of the (GAA)_n repeat.

Figure 2:

Methods to detect in vivo formation of H-DNA in higher eukaryotes (see section 2). (A) Genome-wide H-DNA formation in cultured mouse B-cells (adapted from [43]). (B) Genome-wide H-DNA formation in primary mouse cells via S1-seq (adapted from [45]). (C) Genome-wide H-DNA formation in human cancer cell lines via S1-END Seq (adapted from [46] preprint).

Besides transcriptional supercoiling, formation of H-DNA can be promoted by DNA strand unwinding during DNA replication. During DNA polymerization in vitro, a DNA strand from a partially unwound repeat can fold into a triplex structure, effectively blocking further DNA polymerase progression. Consequently, triplex forming motifs were called “suicidal sequences” for DNA polymerization [47]. A similar mechanism accounts for the blockage of DNA polymerization by expanded (GAA)_n repeats [31], [48]. Recently, analysis of DNA synthesis through a reconstituted eukaryotic replication fork revealed a weak but reproducible stalling only when a long (GAA)_n repeat was located on the leading strand template [49].

Long (GAA)_n repeats were also shown to stall replication in vivo both in yeast and mammalian cells [50–54]. In yeast, (GAA)_n-mediated fork stalling is orientation-dependent, as it was only observed when the (GAA)_n run was on the lagging strand template [53], [55]. This was interpreted as the formation of an H-r triplex in front of the fork, followed by strand unwinding by the replicative CMG (Cdc45/Mcm2-7/GINS) helicase [53]. In human cells, (GAA)_n repeats cause fork stalling in SV40-based episomes in an orientation-independent manner [50], [51], [56] (preprint). This inconsistency between the two systems could be explained by the fact that in an SV40 replisome, the T-antigen helicase is used instead of the endogenous CMG helicase, and synthesis is performed by Pol δ on both the leading and the lagging strand [57]. Fork stalling at (GAA)_n repeats result in fork reversal [51], [56] (preprint),which can lead to subsequent fragility and instability [55], [56], [58].

It is important, therefore, to understand whether triplexes can cause the stalling of regular replication forks at endogenous chromosomal locations. The use of the S1-END-seq approached described above [44] identified two types of non-B DNA structures in human cells: DNA cruciforms formed by (AT)_n repeats and H-DNA formed by long hPu/hPy mirror repeats ([46] preprint). H-DNA was detected in multiple cancer cell types, and a considerable portion of sequences was shared between genomes, indicating the presence of highly conserved H-DNA regions (Fig. 2C). Most of the detected triplexes were formed by (GAAA)_n, (GAA)_n and (GGAA)_n repeats and corresponded to the H-r5 conformation. To address the concern that the acidic pH of S1-nuclease treatment could promote triplex formation ex vivo, the authors replaced S1-nuclease with P1-nuclease, which cleaves ssDNA at neutral pH. This new approach, called P1-END-seq, revealed a similar number and distribution of triplex peaks in cancer cell lines. Another important argument supporting H-DNA formation in vivo is that the triplex peaks spike in S-phase, which correlates with orientation-dependent replication fork stalling, similar to that observed in yeast.

Replication fork stalling at expanded (GAA)_n repeats at the FXN locus was directly shown using single-molecule analysis of replicated DNA (SMARD) isolated from FRDA patient cells [52]. Stalling was observed in both orientations, being particularly pronounced when the (TTC)_n run is on the lagging strand template. This polarity is opposite to that observed in yeast and in human cells via S1-END-seq. While the reason for this difference remains unclear, one possibility is that head-on collision of FXN transcription with replication going in the opposite direction adds to the strength of the stall. Importantly, GAA-specific polyamides that disrupt triplex DNA rescue replication fork stalling, implying that H-DNA causes fork stalling at the repeat (preprint) [46], [52].

Further characterization of non-B DNA structures at the single nucleotide level has the potential to provide insight into not only location and frequency of H-DNA, but also requirements for its formation, mutagenic potential and association with cell or tissue type. This information will be important to determine whether H-DNA is associated with genetic diseases such as cancer besides FRDA, as proposed by recent computational analyses [59], [60]. So far, individual sequencing methods show detection biases toward specific H-DNA isomers, and different types of sample processing might introduce artificial H-DNA formation to a certain extent. Newer methods based on the use of small molecules as modifiers are being developed to allow more sensitive and accurate detection of alternative DNA structures [61], [62]. Where, then, are the commonalities? Firstly, all studies agree that even though both replication and transcription can promote H-DNA formation, the major contributor varies based on the specific repeat and its genetic environment, such as chromatin context and specific location. In addition, interruptions in the repeat decrease its ability to form H-DNA and stall the replication fork [45], [46] (preprint), in accordance with previous in vitro studies for long interrupted (GAA)_n repeats [36]. Importantly, interrupted (GAA)_n repeats are rarely found in FRDA patients, and when present lead to delayed disease onset and milder phenotypes, strongly suggesting that the ability of the repeat to form a triplex is essential for disease pathogenesis [13], [63]. Finally, the prevalence of H-DNA forming sequence warrants extensive studies of their potential biological functions, which likely remain widely underestimated.

3. Genome instability mediated by (GAA)_n repeats

As is true for other repeat expansion diseases, the longer the (GAA)_n repeat is, the more prone to length instability it becomes [64], [65]. Instability is observed both during intergenerational transmission and in post-mitotic somatic cells [66–70]. Expansions predominantly happen during intergenerational transmission and cell division. In somatic cells, contractions are the most prevalent form of (GAA)_n instability, although expansions were observed in affected tissues, including the heart, pancreas and neuronal tissues [66], [71]. Somatic mosaicism – the presence of a variety of (GAA)_n lengths in the same patient – was observed in multiple patient tissues in an age-dependent manner, with the length of the largest allele determining the scale of the observed mosaicism [66], [71–74].

Since (GAA)_n repeat expansions cause a human hereditary disease, research has focused on the study of their expansion mechanisms. Somewhat less appreciated is the fact that (GAA)_n repeats can cause additional local and global genome rearrangements. The original examples came from studies in yeast. First, it was directly demonstrated that expanded (GAA)_n repeats are fragile, resulting in double-strand breaks [55]. Second, (GAA)_n repeats appear to cause mutagenesis at a distance, in a process that we called repeat-induced mutagenesis (RIM) [75]. Expanded (GAA)_n repeats at the FXN locus also increase mutagenesis in the area surrounding the repeat, likely through double-strand break (DSB) repair processes [75–79]. Note that other triplex-forming sequences were also associated with increased break-induced mutagenesis in mammalian cells [77], [80], [81].

Altogether, various types of (GAA)_n repeat-mediated instability contribute to the accumulation of mutations at and around the expanded locus, as well as rearrangements in other genomic regions. These events can modulate the age and onset of FRDA and/or lead to the emergence of other pathogenic mutations, which can in turn modulate (GAA)_n length stability. We will cover the mechanisms of (GAA)_n mediated genome instability during replication and in non-dividing cells in the next sections.

4. (GAA)_n repeat instability during DNA replication

4.1. Fragility

As is common for other disease-related repeats, (GAA)_n repeats were shown to be fragile in multiple model systems, causing DSBs and genome rearrangements. In yeast, (GAA)_n fragility was shown to be dependent on its orientation relative to the replication origin, being higher when the (GAA)_n is on the lagging-strand template – the same orientation that causes replication stalling [55]. In this case, fragility was dependent on the mismatch repair (MMR) machinery. We hypothesize that yeast MMR cleaves the single-stranded loops of H-DNA, erroneously perceiving them as mismatched loop-outs. It would be of great interest to substantiate this idea in biochemical studies. Indirect evidence also indicates that there is increased fragility at the endogenous (GAA)_n repeat in the FXN locus in FRDA patient cells [58].

4.2. Expansions

Expanded (GAA)_n repeats were shown to affect replication fork progression in every experimental system studied to date, likely owing to their triplex-forming potential. In yeast, (GAA)_n repeats start expanding at the carrier length of (GAA)₅₂, and the rate of expansion increases exponentially with the repeat’s length [65]. This begs the question, is there a link between replication through the repeat and its instability? An unambiguous affirmative answer came from a yeast experimental system. A genome-wide screen identified genes that modulate (GAA)_n instability, including its fragility and propensity for expansions [82]. The screen had hits in three main categories: replication-associated genes, transcription initiation genes, and two components of the CST (Cdc13-Stn1-Ten1) complex, which regulates telomere maintenance. Further studies expanded on the role of each category in (GAA)_n repeat instability.

First, an intact and processive core replisome was shown to counteract instability [82], [83], as mutations in Pol ε and δ, as well as in the fork stabilization complex (Tof1-Csm3-Mrc1), promote large-scale (GAA)_n repeat expansions [83]. Mutations in subunits of the CMG helicase also increase mid-scale expansions of short (GAA)₂₅ repeats, through a mechanism consistent with template switching (TS) and break-induced replication (BIR) [84], [85]. During lagging strand synthesis, each Okazaki fragment needs to be processed by 5’ flap endonucleases prior to ligation. The major flap endonucleases in yeast are Rad27 and Dna2, which cleave short and long flaps, respectively. Mutation of either flap endonuclease dramatically increases (GAA)_n repeat expansions, possibly due to an imbalance in the total amount of ssDNA in the cells which can result in increased formation of triplexes on the flaps, ultimately resulting in (GAA)_n repeat instability (Fig. 3A) [65], [83].

Figure 3:

Models of (GAA)_n repeat instability. (A) (GAA)_n repeat instability in dividing yeast mainly occurs via replication-dependent mechanisms, such as template switching (TS) or flap ligation, leading to expansions (see 4.2) [65], [83]. Pol δ dissociation during lagging strand synthesis causes contractions (see 4.3) [64]. During transcription, formation of a triplex-stabilizing R-loop (H-loop) can trigger break-induced recombination leading to repeat expansions (see 5) [113]. (B) (GAA)_n repeat instability in non-dividing yeast is characterized by two different types of events, depending on whether the MMR machinery is functional. MMR drives incisions in H-DNA and sticky DNA structures, which are then converted into DSBs. DSB repair by HR leads to gene conversions events, while repair by NHEJ results in deletions. In MMR-deficient strains, nick repair leads to expansions (see 6.1). (C) Replication dependent (GAA)_n repeat expansions in human cells is initiated by triplex formation ahead of the fork leading to its regression. Repeats can expand upon strand slippage during the restoration of the regressed fork [56] (preprint) (see 4.2). (D) (GAA)_n repeat instability in non-dividing and somatic human cells is promoted by the formation of R-loops during transcription and/or upon transcription-replication collisions (TRCs), which are then recognized by MMR and converted into DSBs. Other sources of DSBs can initiate this process as well. Error-prone repair results in (GAA)_n repeat expansions (see 6.1 and 6.2) [52], [114–116].

A strong replication stall could lead to dramatic consequences such as template switching or replication fork reversal [51]. The goal of both of these processes is to bypass a template “lesion”, but can result in (GAA)_n repeat instability [86]. Nevertheless, whereas fork stalling is orientation-dependent [52], [53], [65], [69], large-scale (GAA)_n repeat instability seems largely orientation-independent, even though there is a slight bias for increased instability when the (GAA)_n run is on the lagging strand template in all of the studied systems [64], [65]. This bias has been related to the asymmetrical nature of DNA replication and led to the proposal of the “ori-switch” model, in which origin activity and orientation relative to the position of the repeat influences predisposition for either expansions or contractions [87]. What other processes, in addition to fork stalling, are then causing instability during replication? Template-switching during DNA synthesis could also occur independently of stalling, especially at the site of long repetitive templates in which TS could happen at a higher rate (Fig. 3A).

In addition, factors other than the ones involved directly in the replication fork can contribute to (GAA)_n expansions during replication, as illustrated by the identification of the CST complex in the genetic screen described above [82]. Recently, mutations in the CST complex were shown to lead to large-scale (GAA)_n repeat expansions through a mechanism involving the Rad9-dependent G2/M DNA damage checkpoint activation and post-replicative repair (PPR) of gaps and nicks [88]. Even though (GAA)_n repeats, unlike expanded (CAG)_n repeats, do not activate the checkpoint by themselves, the checkpoint response likely becomes relevant in the context of other sources of replicative stress [89].

Very recently, an episomal experimental system was set up to study the link between replication and expansions of (GAA)₁₀₀ repeats in human cells [56] (preprint). Using siRNA to deplete specific proteins, this study shows that large-scale repeat expansions are promoted by proteins involved in replication fork reversal (SHPRH/SMARCAL1/HLTF) and counteracted by proteins that are involved in the restoration of reversed forks (RAD52/RECQL1/WRN). Further, the DDX11 helicase, which was shown to untangle triplex DNA in vitro [90], also counteracts repeat expansions. Altogether, this data indicates that expansions occur while the replication fork attempts to bypass a triplex formed by the (GAA)_n repeat (Fig. 3C).

4.3. Contractions

Repeat contractions are another type of (GAA)_n repeat instability which has been intimately tied with ongoing DNA replication. (GAA)_n repeats become more prone to contractions the longer they are, and contractions occur progressively in somatic cells of FRDA patients, as well as during intergenerational transmission [66], [67], [70]. The question of whether contractions and expansions occur through a shared mechanism remained a mystery for a long time. A recent study in yeast set up a genetic assay to study large-scale contractions (> 20 repeats) of long (GAA)₁₂₄ tracts [64]. First, contractions were associated with the ability of the repeat to form H-DNA, directly tying contraction events to triplex formation during DNA synthesis. Second, mutations of lagging strand synthesis polymerases (Pol α and Pol δ) and flap-processing nucleases (Rad27 and Dna2), which result in the accumulation of long ssDNA tracts, promoted large contractions in an orientation-independent manner. Third, the ssDNA-binding replication protein A (RPA) strongly counteracted triplex formation between the nascent lagging strand and its template, preventing contractions (Fig. 3A) [64]. At the same time, all major DSB-repair pathways did not influence (GAA)_n repeat contractions. This led us to conclude that large-scale contractions occur during triplex bypass during lagging strand synthesis.

Importantly, the average contraction size of the starting (GAA)₁₂₄ was of ca. 60 repeats in this system, and likely to be the result of a one-step process. This corresponds to a contraction of a (GAA)₁₂₄ repeat in the mutational range down to pre-mutational or normal repeat sizes, which is very encouraging from a potential therapeutic standpoint. The challenge remains to identify proteins which exclusively promote contractions or prevent expansions, without affecting the other side of instability or have genome-wide mutagenic effects.

Altogether, the expansion and contraction data point to a link between replication through (GAA)_n repeats and their instability. Both fork stalling at the repeat and instability become apparent at (GAA)_n repeat lengths corresponding to carrier sizes in patients and become more pronounced at disease length. That being said, in the experimental systems studied so far, there is no obvious direct correlation between the strength of fork stalling and repeat instability causing disease. This ambiguity warrants further studies of the mechanisms of replication-dependent repeat instability.

4.4. Complex genome rearrangements

Two classes of repeat-mediated genome rearrangements were revealed in a yeast experimental system. First, repair of DSBs within expanded (GAA)_n repeats occasionally leads to the formation of large deletions which encompass the repeat and its adjacent regions [65]. Second, genetic assays combined with Nanopore sequencing found that (GAA)_n repeats cause complex genome rearrangements (CGRs) of the yeast genome that are yet another byproduct of DSB repair [91]. These CGRs resulted from a mixture of reciprocal and non-reciprocal gene-conversion events, which involve both intra- and inter-chromosomal interactions [91].

We also want to emphasize that with the advent of long-read sequencing and dedicated computational tools, the sequencing of expanded repeats appears more practical, and provides the opportunity to widely survey both the general and affected populations to assemble a comprehensive picture of (GAA)_n repeat sizes and distributions [91–98]. Recently, Nanopore sequencing was used to detect replication fork stalling associated with structure formation during sequencing [99], [100]. The combination of long-read sequencing with GWAS, such as the ones conducted on (CAG)_n repeats and Huntington’s disease [101–103], could provide invaluable information on genetic modifiers of (GAA)_n repeat stability and FRDA onset and severity, revealing new mechanisms of instability and guiding the development of novel therapeutic avenues.

4.5. Repeat-induced mutagenesis

As mentioned above, expanded (GAA)_n repeats in the FXN locus increase mutagenesis in surrounding genomic regions in FRDA patients. The mechanisms of this mutagenesis were thus far studied only in a yeast experimental system. Repeat-induced mutagenesis (RIM) was detected up to 10 kb upstream and downstream of long (GAA)_n repeats [79]. Conditional mutations in Pol Ɛ dramatically elevated the rate of RIM, implicating DNA replication in the process [83]. It is not yet clear what causes the mutagenesis. In two studies, RIM depended on translesion synthesis (TLS) by DNA Polymerase ζ [79], [104]. In another study, translesion synthesis was only involved in RIM if Pol δ activity was compromised [83]. Two mechanisms are being considered: repair of post-replication gaps that involves Pol ζ [79], [104], and a BIR-like pathway [83].

5. Role of transcription and R-loops in (GAA)_n repeat instability

Expanded (GAA)_n repeats pose an obstacle to transcription in an orientation-dependent manner [54], [105], [106]. During transcription, a sense r(GAA)_n strand or an antisense r(UUC)_n strand can participate in the formation of stable DNA:RNA duplexes, or R-loops, consisting of two DNA and one RNA strands [107], [108]. Interestingly, recent analyses have shown that R-loop formation might be promoted at repetitive elements across eukaryotic genomes [109], [110]. Enrichment of unscheduled R-loops can result in genomic instability and has been shown to modulate stability of other TNRs (reviewed in [111], [112]).

R-loop formation during transcription of expanded (GAA)_n repeats has been proposed as a pathogenic mechanism contributing to transcriptional silencing at the FXN locus [117]. The formation of DNA:RNA hybrids at expanded (GAA)_n repeats is thermodynamically favorable since the sense RNA is homopurine [118], [119]. Interestingly, transcription has been shown to increase (GAA)_n repeat instability in both human and yeast experimental systems [82], [120–122]. In yeast, the transcription-dependent increase in instability was further exacerbated in the absence of the RNase H enzymes, which counteract the accumulation of R-loops [113]. Furthermore, this increased instability was caused by BIR. Altogether, these data point to the role of R-loops and transcription-replication collisions (TRC) in (GAA)_n repeat instability (Fig.3A).

It is generally believed that R-loops can cause genome instability when replication and transcription collide head-on. It was surprising, therefore, that R-loop-dependent instability of (GAA)_n repeats did not depend on relative orientation of replication and transcription [113]. To explain this difference, it was hypothesized that H-DNA transiently formed upstream of elongating RNA polymerase interacts with the repetitive RNA transcript forming the so-called H-loop [5]. This H-loop is much more stable than either H-DNA or R-loop alone. Therefore, DNA replication could be blocked notwithstanding its directionality. It is yet to be determined whether transcription can promote (GAA)_n repeat instability independently of DNA replication. The answer this question will allow us to determine whether transcription-mediated instability might be a universally shared mechanism in both dividing and non-dividing cells. It will be important to study whether other factors which contribute to R-loop balance and/or transcription-coupled repair (TCR) affect (GAA)_n repeat stability. One promising candidate is the Senataxin helicase (Sen1 in S. cerevisiae), which associates with the replication fork to resolve R-loops and prevents their accumulation during DNA replication and DNA damage repair [123–127].

6. Replication-independent pathways of (GAA)_n repeat instability

The tissues which are most severely affected in repeat expansion diseases are usually terminally differentiated and do not undergo cellular division. In such tissues, DNA replication-dependent processes cannot be a major source of repeat instability. Nevertheless, (GAA)_n repeats progressively expand in neuronal tissues, such as the cerebellum and dorsal root ganglia (DRG), as well as in cardiac muscles, in both human and mouse models [66], [128–132]. A longitudinal study in FRDA patients confirmed lifetime-long addition of (GAA)_n repeats in multiple non-dividing tissues and shows that longer starting repeat sizes lead to a greater magnitude in expansions over time [71]. DNA damage and subsequent repair, involving tracts of DNA synthesis through the repetitive tract, occur in post-mitotic tissues such as neuronal tissues at every stage of the cell cycle [133], [134]. This is relevant for (GAA)_n repeat instability since DNA damage repair can be affected by the presence of the triplex H-DNA structure, which may be perceived as a lesion by DNA repair machineries. It is also foreseeable that repeat instability could arise during cell-cycle reactivation in postmitotic neurons, which occasionally involves aberrant DNA synthesis [135–138]. In the next section, we focus on the DNA repair mechanism that has received most of the recent attention regarding (GAA)_n repeat instability – mismatch repair.

6.1. Mismatch repair in (GAA)_n repeat instability

Mismatch repair (MMR) is responsible for the correction of base mismatches and small insertions and deletions (indels), which can arise during DNA replication, recombination, and repair. In MMR, mismatches and small loops (1-3 bp) are recognized by MutSα (Msh2-Msh6), and larger loop-outs by MutSβ (Msh2-Msh3). Subsequently, the MutL complexes, MutLα (Mlh1-Pms2) and MutLγ (Mlh1-Mlh3), excise the mismatch, followed by DNA strand resection and fill-in synthesis. In this capacity, MMR is essential for the maintenance of genome stability. Among other things, defects in MMR lead to microsatellite instability – a characteristic feature of various cancers (reviewed in [139]). But not all microsatellites are equal in the eyes of the MMR machinery. Counterintuitively, functional MMR overall promotes instability of the microsatellites responsible for repeat expansion diseases, including (CAG)_n, (CGG)_n and (GAA)_n [140], [141]. The individual effects of the MMR components on repeat instability depend on both the model organism and whether somatic or intergenerational instability is under investigation (Table 1).

Table 1:

Role of MMR proteins during length instability of (GAA)_n and (CAG)_n repeats

Complex	MMR protein	GAA somatic and non-dividing	GAA intergenerational and replication-models	CAG somatic and non-dividing	CAG intergenerational and replication-models
MutLα MutLγ	MLH1	Promotes expansions Mouse model [143] No effect on expansions Promotes deletions Non-dividing yeast cells [149]	Promotes expansions Mouse model and human cells [116], [143]	Promotes expansions HD mouse model [156] Human HD population [157]	Prevents expansions S. cerevisiae [158]
MutLα	PMS2 (Pms1 in S. cerevisiae)	Prevents expansions Mouse model Non-dividing yeast cells [114], [149] Promotes deletions Non-dividing yeast cells [149]	Prevents expansions Promotes contractions Mouse model [142] Prevents expansions Human cells [116]	Promotes expansions Prevents deletions Mouse model [159]	Prevents expansions S. cerevisiae [158]
MutLγ	MLH3	Prevents expansions Human cells [116]	Prevents expansions Human cells [116]	Promotes expansions HD mouse models and patient derived cells [156], [160]	No effect S. cerevisiae [158]
MutSβ MutSα	MSH2	Promotes expansions Mouse model Human cells [114], [115]	Prevents contractions Mouse model [142] Promotes expansions iPSC GAA expansion model [144], [161]	Promotes expansions Mouse models Human cell model [126], [162], [163]	Promotes expansions S. cerevisiae Mouse model [164], [165] Prevents contractions S. cerevisiae [158]
MutSβ	MSH3	Prevents expansions Promotes deletions Non-dividing yeast cells [149] Promotes expansions Human cells and FRDA fibroblasts [115]	No effect on expansions G0 yeast cells [149] Prevents contractions Mouse model [142]	Promotes expansions Human cell model [126]	Promotes expansions HD mouse model DM1 mouse Human cell lines S. cerevisiae [164], [166–170]
MutSα	MSH6	Promotes expansions Mouse model [114] No effect on expansions and promotes deletions Non-dividing yeast cells [149]	Prevents expansions and contractions Mouse model iPSC model [142], [161] No effect on expansions Human cells [115]	Prevents expansions Human cell model [126]	Prevents contractions HD mouse model [166] Prevents expansions S. cerevisiae [164]

In humanized FRDA mice, MMR affects repeat instability, although fine molecular mechanisms are somewhat controversial. That is, MUTSβ prevents intergenerational contractions, MUTSα precludes both expansions and contractions, while MUTLα counteracts expansions but promotes contractions [142], [143]. In somatic cells, specifically in neurological tissues, MUTLα suppresses expansions, but MUTSα promotes expansions [114].

In contrast to mice, MUTSβ promotes small-scale expansions in cultured human cells [115]. This depends on the activity of MUTLγ, and MUTLα protected against expansions, as observed in mouse models (Fig. 3D) [116]. In FRDA patient-derived induced-pluripotent stem cells (iPSCs), MUTSβ promotes large-scale (GAA)_n expansions [144]. Whether allele variants of the individual proteins are actually modifiers of repeat stability in humans remains to be ascertained [145–148]. Notably, genetic analysis of (GAA)_n repeat instability in non-dividing somatic cells cannot be easily done in patient-derived tissues, as they present a static picture or are passaged derived cells, which underwent further replication cycles.

To address this problem, Neil et al. developed a novel experimental system to study instability of a long (GAA)₁₀₀ repeat in chronologically aging quiescent (G0) cells in S. cerevisiae [149]. Three categories of mutagenic events were observed: repeat expansions, NHEJ-dependent deletions, and HR-mediated gene conversions. Whereas the main mutational event in dividing yeast cells is repeat expansions, the balance rapidly shifted to large deletions frequently including the whole (GAA)_n repeat as the cells entered quiescence (Fig. 3B). Deletions were triggered by DSBs at the repeat generated during MMR as discussed in 4.1. These DSBs are then repaired by Exo1-mediated resection and NHEJ. Consequently, a functional MMR machinery suppresses expansions of (GAA)_n repeats specifically in non-dividing yeast cells (Table 1) [55], [65], [149]. Inactivation of MutSβ and MutLα and, above all, Exo1 increased both the frequency and the size of (GAA)_n repeat expansions in quiescent cells. Finally, expansions involved processive DNA synthesis by Pol δ, likely occurring during error-prone repair of nicks accumulated in the damage-susceptible single-stranded parts of H-DNA [150]. Whereas large-scale expansions occur in one step during replication [83], expansions in quiescent cells could result from multiple smaller-scale expansions [149].

In sum, MMR has been established as a major pathway modulating (GAA)_n repeat stability, albeit with dramatic differences when it comes to the role of individual MMR components between different model systems. Note, also, that the effects of the MMR machinery on (GAA)_n repeat instability are strikingly different from their role in (CAG)_n repeat instability, as is comprehensively displayed in Table 1. In short, the MMR machinery binds but cannot excise long stable hairpins, further stabilizing those hairpins. Consequently, MMR promotes (CAG)_n expansions in mice and humans [151–155]. The association between MMR and (GAA)_n repeat instability clearly does not fit the same scheme. Furthermore, the effects of MMR on (GAA)_n stability are clearly different in dividing and non-dividing cells, as evidenced by different effects of MutLα and MutSβ described above and in Table 1. The latter difference could result from different expression levels of MMR proteins as well as a different relationship between replication, transcription, and DNA repair.

6.2. Base excision repair in (GAA)_n repeat instability

Base excision repair (BER) is a DNA damage repair pathway which processes lesions initiated by oxidative damage, alkylation and base deamination. In short, the modified base is removed by a DNA glycosylase and the resulting abasic site is cleaved by the AP endonuclease creating a nick in a DNA strand. DNA polymerase β carries out repair DNA synthesis, which is followed by flap removal by flap-endonucleases and ligation. BER has been proposed to be a major pathway leading to small-scale, age-dependent somatic expansions of (CAG)_n repeats [171], [172]. Trinucleotide repeat stability is influenced by BER based on the site of DNA modifications within the repeat, the presence of additional proteins, and its balance with MMR processes [173].

Lai et al. showed that alkylation of (GAA)_n repeats by the chemotherapeutic temozolomide results in BER-mediated (GAA)_n repeat contractions both in vitro and in lymphoblasts of FRDA patients [174]. A structure-prone repetitive flap may compromise coordination between Pol β and flap cleavage by the flap endonuclease FEN1. BER was also proposed as a mechanism of repair of abasic sites within R-loops formed at (CAG)_n repeat [158]. Laverde et al. conducted a biochemical study of BER activity on R-loop substrates containing a (GAA)₂₀ repeat. The BER enzyme AP endonuclease 1 (APE1) was found to incise the abasic site in the (GAA)_n R-loop, creating a double flap intermediate that hinders synthesis by Pol β while stimulating 5’-flap cleavage by FEN1. Cleavage by FEN1 promotes R-loop resolution and contractions of about half the repeat length [175]. Therefore, the data so far indicate that processing of lesions in (GAA)_n repeats by BER primarily results in repeat contractions.

What remains to be determined is the direct contribution of individual oxidizing agents and lesions on (GAA)_n repeat stability. The position of the lesion relative to the repeat could also be involved in determining whether repair will result in expansions or contractions, as is the case for (CAG)_n repeats [176]. Oxidative damage is particularly relevant in the context of FRDA, as it is a prominent form of DNA damage in aging neurons and the frataxin protein itself is involved in the processing of oxidative damage [177], [178]. Mitochondrial dysfunction, accumulation of reactive oxygen species (ROS) and subsequent cell death has been proposed as a driving cause of FRDA [179]. ROS accumulation promotes elevated levels of oxidative damage in the cell. If oxidative damage repair modulates (GAA)_n repeat instability in somatic cells, it could be one of the main drivers of age-dependent somatic instability.

6.3. (GAA)_n repeats are hotspots of homologous recombination

(GAA)_n repeats were shown to promote homologous recombination in bacterial and yeast experimental systems [55], [180], [181], a feature they share with other trinucleotide repeats [182]. However, there are sensitive differences between the two systems. In bacteria, the recombinogenic potential of (GAA)_n repeats decreased with their length, which was attributed to the formation of sticky DNA by longer repeats (see Section 2). In yeast, in contrast, the repeat’s recombinogenic potential increased with its length. Furthermore, repeat-mediated recombination in bacteria, but not in yeast, occasionally led to length instability of the repeat itself. Finally, the data from the tetrad analysis in yeast indicated that repeat-mediated recombination occurred during the G1 phase of the cell cycle – it was replication-independent [180]. We want to emphasize that unlike (CAG)_n repeats, (GAA)_n repeat instability is not modulated by homologous recombination factors in yeast experimental systems, except for when the RNase H enzymes had been deleted, eliciting a BIR response [64], [82], [113], [182] (see Section 5).

7. Does FAN1 play a role in (GAA)_n repeat instability?

Cells contain a variety of structure-specific nucleases, which process flaps and other structures during DNA replication, repair, and recombination. Whereas each nuclease optimally processes a specific substrate, it is possible they can mis-recognize unusual DNA structures, influencing their stability. Two such nucleases are FEN1 (Rad27 in S. cerevisiae) and the Fanconi-associated nuclease FAN1. FEN1 processes 5’ flaps of Okazaki fragments during lagging strand synthesis and DNA damage repair [183]. FAN1 is an interstrand cross-link (ICL) repair protein which exhibits 5’-to-3’ exonuclease activity as well as endonuclease activity, participates in HR and processing of stalled forks [184]. FEN1 and FAN1 belong to the same class of enzymes and have been shown to have overlapping substrates, suggesting they might have similar or redundant roles in the regulation of repeat instability.

In yeast, Rad27 (FEN1) prevents instability of both (GAA)_n and (CAG)_n repeats [64], [185–188], and multiple models propose that it does so through its flap equilibration abilities, which likely counteracts the formation of non-B DNA structures. In contrast, mammalian FEN1 does not seem to fully share this important role, as its depletion does not affect (GAA)_n instability and has contrasting effects on (CAG)_n stability, and has not been identified as a disease regulator [189–193].

What could explain this striking difference between organisms? It is possible that flap processing during Okazaki fragment synthesis might not be a major contributor to somatic instability overall. Alternatively, flap processing performed by other nucleases might be more important in human cells. Recently, FAN1 has emerged as a prominent candidate. Genome-wide association studies (GWAS) of Huntington’s disease have identified FAN1 as a strong genetic modifier of disease onset, with FAN1 mutations being associated with earlier onset [103], [194]. In addition, FAN1 prevents expansions of (CGG)_n repeats in Fragile X syndrome mouse models and rare missense variants in FAN1 were associated with expanded (CGG)_n repeats present in individuals with autism spectrum disorders [195–197].

Goold et al. propose that FAN1 acts through a nuclease-independent pathway to stabilize (CAG)_n repeats and prevent expansions, possibly by recruiting DNA damage repair proteins to the repeat, promoting conservative repair [198]. Candidates for key FAN1 interactors are its physical interactors in the MutL family [199], [200]. FAN1 and MLH1 have been recently shown to have opposite but interdependent effects on (CAG)_n repeat instability. FAN1 sequesters MLH1 through a *SPYF* motif, leaving it unable to promote (CAG)_n expansions through its canonical MMR function [201–204] (Table 1). In these studies, the nuclease activity of FAN1 was needed to prevent expansions. Variants of FAN1 with reduced nuclease activity have been found in patients with particularly early HD onset and highlight the endo- and exonuclease activity of FAN1 in protecting against expansions [195], [205]. Therefore, it has become clear that FAN1 is a major regulator of (CAG)_n repeat stability [206].

Since MLH1 activity has been shown to regulate (GAA)_n repeat instability in yeast and mouse models as well as in cultured human cells, it is foreseeable that the FAN1-MLH1 interaction might influence the balance of (GAA)_n repeat stability possibly by the processing of 5’ flaps generated by strand displacement during DNA synthesis. The latter role would more closely mirror the effects of Rad27 on (GAA)_n repeat instability, as Rad27 does not interact with the MMR machinery in S. cerevisiae in otherwise unperturbed conditions [207]. Thus, we believe it is of great interest to study the role of FAN1 nuclease in (GAA)_n repeat instability.

8. Therapeutic avenues targeting (GAA)_n repeat stability

8.1. Gene editing and replacement therapies to modulate (GAA)_n repeat size and stability

Friedreich’s ataxia is a predominantly monogenic disease. Therefore, removal of expanded (GAA)_n repeats at the FXN locus via gene editing is a potentially promising approach to modulate and even prevent disease. Excision of the (GAA)_n repeat and some of the flanking sequence with zinc-finger nucleases in FRDA derived cells can partially rescue defects in FXN expression and ameliorates pathological phenotypes in iPSC-derived neuronal cells [208]. More recently, CRISPR technology has been applied to the study of repeat expansion diseases [209–211]. Removal of expanded (GAA)_n repeats in an FRDA mouse cell line containing one expanded (GAA)₁₉₀ allele promoted partial transcriptional rescue of FXN gene expression, with an associated increase in protein levels [211]. The same result, though, was not observed in a cell line with two expanded alleles [211]. We envision that both length of the expanded allele and the relative size of the other allele can influence the success of CRISPR targeting.

Genome editing techniques to restore frataxin levels can be combined with cell replacement therapies to overcome an additional hurdle in therapy. Before the advent of CRISPR, same-species (allogeneic) transplant of healthy cells had been explored as a method to treat FRDA symptoms in mice [212]. The major hurdles of allogeneic transplantation are immunosuppression and graft rejection. The possibility of using the patient’s own cells, modifying their genome, and reintroducing them into the patient (autologous graft) is therefore much more attractive, as it circumvents the mentioned issues. Rocca et al. removed the expanded repeat from human FRDA fibroblasts and hematopoietic stem and progenitor cells (HPSC), reaching a substantial increase in frataxin protein levels and rescue of mitochondrial defects. HPSCs then underwent hematopoiesis but displayed reduced cell proliferation rates [210].

Can CRISPR succeed also in the context of a functional brain? FRDA-derived iPSCs and embryonic stem cells differentiated into neuronal derivatives are able to withstand grafting into rodent brains, survive and even mature into dorsal root ganglia (DRG) – the primary tissue affected by neurodegeneration in FRDA [213], [214]. Consequently, they constitute a model in which the utility of gene editing can be more reliably tested. FRDA-derived iPSCs have also been used to develop an in vitro 3D DRG organoid (DRGO) model [215]. Removal of the expanded (GAA)_n repeat by CRISPR partially restored the frataxin protein levels and rescued FRDA-associated phenotypes, and the level of rescue was greater when shorter (GAA)_n repeats were deleted. On the other hand, an almost complete deletion of the first FXN intron restored frataxin expression levels to approximately wildtype levels. The repressing chromatin markers associated with FXN transcription silencing in FRDA were permanently removed when the whole intron was removed, indicating that long (GAA)_n repeats propagate chromatin silencing through its upstream and downstream regions [215]. Thus, removal of the expanded (GAA)_n repeat alone might not be sufficient and will only work in association with the concomitant loss of repressive chromatin marks in its surroundings.

What are some of the caveats? First, it seems that the extent of the FXN region to be removed will have to be tailored to the starting (GAA)_n repeat length of the individual patient, and different guides might be needed in each specific case. Secondly, off-target effects need to be carefully studied and minimized, and this will have to be tested for each target site. Third, whether the same approach can be applied to intergenerational instability remains to be determined. Finally, long-term studies are needed to test whether once the repeat has been shortened it remains stable over a long period of time, or whether it eventually becomes unstable again.

An alternative approach is gene addition, in which transgenic wild-type FXN is reintroduced into cells using viral vectors to rescue frataxin expression (reviewed in [216]). Since gene addition was successfully used for treating recessive genetic diseases in humans [217], [218], this approach has been broadly investigated for FRDA, using various mouse models, patient-derived fibroblasts and in non-human primates ([216] and references therein). The main caveats in the FRDA case, however, was toxicity upon delivery and failure to rescue frataxin expression in neurological tissues.

8.2. Oligonucleotide-based approaches

The availability of repeat-stabilizing agents, such as small molecules that stabilize H-DNA or promote contractions, is being investigated as a complementary approach for treating repeat expansion diseases [52], [219]. Since the protein coding sequence of frataxin remains unaltered in FRDA, upregulation of transcription and protein levels is a viable therapeutic avenue. Recently, the Napierala group has pioneered an oligonucleotide-based approach in which frataxin mRNA levels were stabilized resulting in increased frataxin protein levels in both FRDA fibroblasts and iPSC-derived neuronal progenitor cell lines [220]. Targeting the 5’ and 3’ untranslated regions of the FXN in combination led to a modest increase in mRNA half-life protein levels without altering the chromatin status of the FXN gene [220]. Since FRDA patients only have 5% to 35% of the control frataxin levels, it remains to be determined whether such an increase has the potential to alleviate frataxin-deficiency associated phenotypes [221], [222].

Oligonucleotides can also be used to directly try and prevent (GAA)_n repeat expansions. The first support for this idea comes from the use of locked nucleic acids (LNA), which have been shown to be able to interfere with triplex DNA formation [223]. LNA-DNA mixmers, which are not toxic for human cells, nearly completely prevented large-scale expansions of (GAA)_n repeats in human cells [224]. This approach directly inhibits repeat expansions and is therefore very promising, albeit their effectiveness in FRDA patients remains to be determined. One of the major challenges regarding patient treatment with these oligonucleotides, as well as other promising small molecules [225–227], is the inefficiency of drug delivery to the central nervous system. While delivery to other affected tissues such as heart or pancreas can be achieved with viral vectors, the current ways of delivering drugs to the central nervous system are very invasive [228]. We hope that a better understanding of the mechanism of (GAA)_n repeat instability and FXN expression during human development would lead to defining the most effective spatiotemporal windows for long-lasting treatment.

List of Abbreviations:

CGR

Complex Genome Rearrangement

FRDA

Friedreich’s ataxia

hPu/hPy

Homopurine/homopyrimidine DNA run

LNA

locked nucleic acid

RED

Repeat Expansion Disease

RIM

Repeat-induced Mutagenesis

TNR

Trinucleotide Repeat

TRC

transcription-replication collisions