Repeat instability during DNA repair: Insights from model systems

Abstract

The expansion of repeated sequences is the cause of over 30 inherited genetic diseases, including Huntington disease, myotonic dystrophy (types 1 and 2), fragile X syndrome, many spinocerebellar ataxias, and some cases of amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). Repeat expansions are dynamic, and disease inheritance and progression are influenced by the size and the rate of expansion. Thus, an understanding of the various cellular mechanisms that cooperate to control or promote repeat expansions is of interest to human health. In addition, the study of repeat expansion and contraction mechanisms has provided insight into how repair pathways operate in the context of structure-forming DNA, as well as insights into non-canonical roles for repair proteins. Here we review the mechanisms of repeat instability, with a special emphasis on the knowledge gained from the various model systems that have been developed to study this topic. We cover the repair pathways and proteins that operate to maintain genome stability, or in some cases cause instability, and the cross-talk and interactions between them.

Repeat expansions cause human disease

The expansion of trinucleotide repeats (TNRs) was first recognized as a cause of human disease in 1991 when the expansion of a CGG/CCG repeat in the 5′ UTR of the Fragile X mental retardation 1 (FMR1) gene was shown to be the cause of Fragile X syndrome (FXS) (Verkerk et al., 1991; Yu et al., 1991) and the expansion of a CAG/CTG repeat in the coding region of the androgen receptor gene was shown to be the cause of spinal-bulbar muscular atrophy (SMBA) (La Spada et al., 1991). Both these diseases exhibited a phenomenon described as anticipation, a worsening of the disease in each subsequent generation. Identification of several other TNR expansion diseases followed shortly, including Huntington disease (HD) caused by the expansion of a CAG repeat within the coding region of the Huntingtin gene, several spinocerebellar ataxias (SCAs) caused by CAG repeats in coding regions, myotonic dystrophy type I (DM1), caused by a CTG expansion in the 3′ UTR of the dystrophia myotonica protein kinase (DMPK) gene, and Friedreich ataxia (FRDA), caused by a GAA expansion in the first intron of the frataxin (FXN) gene (Table 1; note: the repeat on the coding strand is used to refer to the disease-causing repeat for a particular locus) (reviewed in Gatchel & Zoghbi, 2005; Lopez et al., 2010; McMurray, 2010; Mirkin, 2006; Pearson et al., 2005). In each case, a more severe disease phenotype and/or an earlier age of onset is caused by inheritance of a longer repeat allele in the offspring compared with the parent, confirming the more deleterious nature of longer repeat tracts and explaining the molecular basis of anticipation. Since the initial discoveries, many more diseases have been identified that are caused by repeat expansions, most of which are trinucleotide repeats, although several diseases are caused by repeats of larger units. Examples include SCA10 caused by an ATTCT expansion (Matsuura et al., 2000) and the recently identified G₄C₂ repeat expansion in the c9orf72 gene, which is the most common cause of familial amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) (DeJesus-Hernandez et al., 2011; Renton et al., 2011) (Table 1). In addition, it has recently been recognized that different sized expansions can in some cases result in different disease phenotypes. In particular, CGG expansions in the FMR1 gene result in a group of disorders known collectively as the Fragile X-related disorders (FXDs). Expansions that result in alleles with 55–200 repeats (the so-called premutation or PM alleles) confer elevated risk of Fragile X-associated primary ovarian insufficiency (FXPOI) and Fragile X-associated tremor/ataxia syndrome (FXTAS), while expansions that generate alleles with >200 repeats are associated with FXS (Loesch & Hagerman, 2012; Table 1).

Table 1.

Characteristics of expandable repeats that cause disease.

Disease	Repeat location	Normal repeat length	Pathogenic repeat length
CAG expansion (hairpin, slip-stranded DNA, R-loop*)
Dentatorubral-pallidoluysian atrophy (DRPLA)	ATN1 exon	7–34	29–88
Huntington disease (HD)	HTT exon	11–34	40–121
Spinal bulbar muscular atrophy (SBMA)	AR exon	9–36	38–62
Spinocerebellar ataxia 1 (SCA1)	ATXN1 exon	6–39	40–82
Spinocerebellar ataxia 2 (SCA2)	ATXN2 exon	15–36	32–200
Spinocerebellar ataxia 3 (SCA3)	ATXN3 exon	13–36	61–84
Spinocerebellar ataxia 6 (SCA6)	CACNA1A exon	4–20	20–29
Spinocerebellar ataxia 7 (SCA7)	ATXN7 exon	4–35	37–306
Spinocerebellar ataxia 12 (SCA12)	PPP2R2B 5′ UTR	7–45	55–78
Spinocerebellar ataxia 17 (SCA17)	TBP exon	25–42	47–63
CTG expansion (hairpin, slip-stranded DNA, R-loop)
Myotonic dystrophy type 1 (DM1)	DMPK 3′ UTR	5–37	50–≥1000
Huntington disease like 2 (HDL2)	JPH3 3′ UTR	7–28	66–78
Spinocerebellar ataxia 8 (SCA8)	ATXN8 3′ UTR	16–34	>74
CGG expansion (hairpin, slip-stranded DNA, G-quadruplex, R-loop)
Fragile X syndrome (FXS)†	FMR1 5′ UTR	6–55‡	>200
Fragile X-associated primary ovarian insufficiency (FXPOI)†	FMR1 5′ UTR	6–55‡	55–200
Fragile X tremor/ataxia syndrome (FXTAS)†	FMR1 5′ UTR	6–55‡	55–200
FRAXE mental retardation syndrome (FRAXE MR)	FMR2 5′ UTR	4–36	200–900
GAA expansion (triplex DNA, R-loop)
Friedrich ataxia (FRDA)	FXN intron	6–32	200–1700
CCTG expansion (hairpin, slip-stranded DNA)
Myotonic dystrophy type 2 (DM2)	ZNF9 intron	10–26	75–11 000
ATTCT expansion (DNA unwinding element)
Spinocerebellar ataxia 10 (SCA1O)	ATXN10 intron	10–20	500–4500
G2C2TG expansion (hairpin)
Spinocerebellar ataxia 36 (SCA36)	NOP56 intron	<8	1500–2500
G4CC expansions (hairpin, G-quadruplex, R-loop)
Amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD)	C9orf72 intron	<23	250–1600
C4GC4GCG expansion (hairpin, G-quadruplex)
Progressive myoclonic epilepsy 1 (EPM1)	CSTB promoter	2–3 12–17¶	30–80

^*For all repeats, experimentally demonstrated structures are listed.

^†Members of the Fragile X-related disorders.

^‡≤30 repeats: normal allele; 45–54 repeats: intermediate allele, uncertain pathogenicity.

^¶Rare, highly unstable.

This review will focus primarily on mechanisms that cause repeat expansion (gain of units) and contraction (loss of units) in the CAG/CTG, CGG/CCG, and GAA/TTC trinucleotide repeats, which cause the majority of diseases and have been most intensively studied in model systems. However, the principles are likely to apply to other repeat expansion diseases.

Disease-associated repeats form non-B DNA structures

The expandable repeats (both trinucleotide and larger units) are able to form stable non-B form DNA structures that can interfere with normal cellular processes. For the CAG/CTG repeat, both the CAG and the CTG strands can form hairpin structures, where intrastrand rather than interstrand base-pairing occurs, with the CTG strand forming a more stable hairpin than the CAG strand (Figure 1A) (Gacy et al., 1995; Mitas et al., 1995b). Evidence for in vivo cleavage by synthetic zinc-finger nucleases specific for these hairpins demonstrates that these structures form in vivo in human cells (Liu et al., 2010). Similarly, both strands of the CGG/CCG TNR can form stable hairpin structures in vitro with the CGG strand forming the more stable structure (Mariappan et al., 1998; Mitas et al., 1995a; Nadel et al., 1995; Usdin & Woodford, 1995; Yu et al., 1997). Melting of both strands in a repeat tract followed by reannealing can lead to hairpins on both strands, termed a slipped strand structure (Figure 1B) (Pearson et al., 1998). The CGG strand also has the potential to form G-quadruplex structures (Figure 1D) and the CCG strand to form i-tetraplexes (Fojtik & Vorlickova, 2001; Fry & Loeb, 1994; Kettani et al., 1995; Usdin & Woodford, 1995). Recent results indicate that it is the CGG hairpin structure that interferes with DNA replication in yeast cells (Anand et al., 2012). While quadruplexes are formed by many sequences in vitro (Biffi et al., 2013; Lam et al., 2013), whether G-quadruplex or i-tetraplex formation is relevant for CGG expansion in vivo is currently unknown; if they do form, it is likely that many of their biological properties would overlap with hairpins. Of note, the repeats responsible for progressive myoclonus epilepsy and the ALS/FTD-associated G₄C₂ expansion also form quadruplexes (Grigg et al., 2014; Potaman et al., 2004; Saha & Usdin, 2001). While hairpins have also been reported to be formed by very long GAA repeats (Heidenfelder et al., 2003), the major structures formed by GAA/TTC repeats are triplexes in which the third strand is either derived from the pyrimidine strand to form a pyrimidine:purine:pyrimidine triplex or derived from the purine strand to form a purine:purine:pyrimidine triplex (Figure 1E) (Grabczyk & Usdin, 2000; Potaman et al., 2004; Sakamoto et al., 2001). The ATTCT repeat expansion is not predicted to form structures, but rather acts as a DNA-unwinding element (Figure 1F) (Potaman et al., 2003). Interruptions to the purity of the repeat tract reduces the risk of expansion for many of these diseases (Chung et al., 1993; Eichler et al., 1994; Gao et al., 2008; Imbert et al., 1996; Liquori et al., 2003; Nolin et al., 2013; Yrigollen et al., 2012), likely because the interruption reduces the stability or length of the secondary structures that can be formed.

Figure 1.

Stable non-B form DNA structures formed by expandable repeats. (A) Hairpins can be formed by G–C-rich expanding triplet repeats with stability: CGG>CCG>CTG>CAG. (B) A slipped-strand structure occurs when hairpins form simultaneously on both strands. (C) RNA/DNA hybrids (R-loops) occur during transcription when the newly synthesized RNA transcript stably pairs with single-stranded DNA in the transcription bubble. Purine-rich repeats are especially prone to forming persistent R-loops. (D) G-quadruplex structures can be formed by CGG repeats and G₄C₂ repeats. G-quadruplexes can be parallel (not shown) or anti-parallel (shown). (E) Triple helical DNA structures form at GAA/TCC repeats and can be either the purine:purine:pyrimidine triplexes or pyrimidine:purine pyrimidine triplexes. (F) AT-rich ATTCT/AGAAT repeats are DNA-unwinding elements, melting the double helix to form a region of unpaired DNA.

In addition to unusual DNA:DNA pairing, the G-rich repeats are especially prone to forming R-loops (Figure 1C). R-loops are RNA:DNA hybrids that form during transcription. While all sequences form transient R-loops behind RNA polymerase II (RNAPII), those that are formed in regions with a strand asymmetry with respect to purines and pyrimidines are particularly prone to form stable and persistent hybrids (Belotserkovskii et al., 2013). R-loops preferentially form when G-rich sequences are on the non-template strand (Belotserkovskii et al., 2010). Persistent RNA–DNA hybrids have been demonstrated at a CAG/CTG repeat in human cells, where their levels were correlated with instability (Lin et al., 2010a,b). CGG/CCG, GAA/TTC, and GGGGCC/GGCCCC repeats also form stable RNA:DNA hybrids (Grabczyk et al., 2007; Groh et al., 2014; Loomis et al., 2014; Reddy et al., 2014), and bidirectional transcription can further increase R-loop formation (Reddy et al., 2014, 2011). These hybrids could favor formation of intrastrand-folded structures by the non-template strand that remains unpaired. It is also possible that the ssDNA on the non-template strand is vulnerable to nicking. Both these events could facilitate repeat instability. The common ability to form these secondary structures of various flavors and forms are thought to be at the basis of repeat instability in the genome.

Developmental timing of TNR expansions

At expandable TNR loci in the human genome, there is a normal-size (non-disease-causing) repeat tract of variable length in the population, generally 5–30 repeat units. There is a fairly sharp size threshold where the repeat tract becomes much more unstable and likely to expand inter-generationally: around 35 uninterrupted repeat units for most loci (Table 1). The existence of this threshold indicates that an inherent property of DNA is likely involved in the switch to an unstable state. This could be a threshold for forming stable DNA structures in vivo. Coincidentally, this threshold corresponds to the size of an Okazaki fragment as well as the length of DNA that is wrapped around one nucleosome. Nonetheless, in systems where rare expansions can be measured, CAG/CTG repeats as small as 13 units are more likely to expand than a non-structure-forming control (Rolfsmeier et al., 2001). Additionally, in vitro studies suggest that even a small number of repeats produce expansions (see, for example, Henricksen et al., 2000; Liu et al., 2009; Pluciennik et al., 2013). In yeast, measurement of expansion frequencies shows that there is a non-linear increase with repeat size: for a CTG repeat on the lagging daughter strand (Okazaki fragment), expansions occur in about 1 in 10⁵ cells for (CTG)₂₅ (Miret et al., 1998) but in about 1 in 10² cells for (CTG)₇₀ (Callahan et al., 2003), a 1000× increase. In humans, CAG expansion frequencies also increase dramatically once above the threshold length of ~35 repeats (reviewed in McMurray, 2010). In the case of Fragile X syndrome, the CGG expansion frequency approaches 100% when alleles with 85 repeats are maternally transmitted (Nolin et al., 2011; Yrigollen et al., 2014). The increased risk of expansion with longer sizes suggests that therapies that slow expansion rate or increase contractions could have an impact on disease inheritance or progression.

Intergenerational expansions of TNRs, which are inherited in all tissues, could occur at several developmental stages: in the dividing germ cell precursors, during germ cell maturation, during meiosis, during post-meiotic sperm maturation, or even during very early embryonic development. For HD, which has a paternal bias for inheritance of expanded alleles, an illuminating study of repeat size in human male spermatocytes concluded that the majority of expansions occur as a pre-meiotic event, although some additional expansions were observed to occur post-meiotically (Yoon et al., 2003). Yet in a mouse model of HD, it was concluded that most expansions arise after meiosis in the haploid gamete (Kovtun & McMurray, 2001), and recent results indicate that these post-meiotic expansions arise during remodeling of spermatid chromatin (Simard et al., 2014). A factor that could explain the difference between the mouse and human data is that the human spermatogonial cells studied were estimated to have undergone many more cell divisions (~600–800 over 40 years) compared with those in the mice (~35 over <3 months). Thus the post-meiotic expansion phase may be more prominent in the mouse model (Yoon et al., 2003). A study of male germ line instability in a DM1 mouse model with >300 CTG repeats also showed expansions occurring in pre-meiotic spermatogonia (Savouret et al., 2004). Together, these studies show that paternally inherited expansions can occur in dividing spermatogonia, and thus potentially could be coupled with replication. Large intergenerational expansions such as those that cause FXS and a congenital form of DM1 are seen almost exclusively when the allele is maternally transmitted (De Temmerman et al., 2004; Oberle et al., 1991; Yu et al., 1991). Expansions can be seen in oocytes from human DM1 patients (De Temmerman et al., 2004; Dean et al., 2006). In the case of both FXS and DM1, an increase in expansions is seen with maternal age, consistent with the idea that expansions occur in the post-natal oocyte in humans (Morales et al., 2014; Yrigollen et al., 2014). Furthermore, in a mouse model of the FXDs, the likelihood of expansion in the progeny of mice heterozygous for mutations that eliminate expansions depends on the parental genotype rather than the offspring genotype, consistent with expansion occurring prezygotically (Lokanga et al., 2014b). Since oocytes are arrested in prophase I of meiosis until ovulation, these data would be compatible with an expansion mechanism that occurs independently of chromosomal replication.

In addition to intergenerational expansions, significant instability in somatic tissues can be seen in some TNR diseases, which can contribute to disease progression (for reviews see Dion, 2014; Lopez Castel et al., 2010; Pearson et al., 2005)). For example, in human HD patients, dramatic expansions (gains of up to 1000 repeats) are observed in striatal cells, the brain region most affected by the disease (Kennedy et al., 2003). Many mouse models also show extensive repeat expansion in the brain (for example, Libby et al., 2003; Lokanga et al., 2013; Mangiarini et al., 1997) and specifically in post-mitotic neurons (Gonitel et al., 2008). Since even expansion-prone somatic cells that are not post-mitotic proliferate very slowly, non-replicative expansion mechanisms are likely to explain many somatic expansions. Supporting this idea, inhibition of cell proliferation using a number of different approaches does not reduce expansions in cultured transgenic DM1 mouse cells (Gomes-Pereira et al., 2014). However, there may be replication-coupled somatic expansions in some tissues. A study of CTG repeat sizes in DM1 patient tissues, where the largest expansions are observed in muscle, pointed toward expansions occurring in a window during early development when muscle cells are proliferating prior to terminal differentiation (Zatz et al., 1995). Even in differentiated tissues, there are replicating cell populations that could provide a source of continued expansions, such as stem cells, where some triplet repeats are highly expandable (Du et al., 2013; Ku et al., 2010). In the brain, dividing glial cells show evidence of ongoing CAG/CTG instability in situ and in culture (Claassen & Lahue, 2007; Watanabe et al., 2000). Because there is a relationship between repeat length and disease severity, understanding the mechanisms of both intergenerational and somatic instability in replicating and non-replicating cells is relevant for understanding disease progression, and may provide an entry point for therapies.

Cis and trans-acting factors that modulate repeat instability

Even with the same repeat length, some cell types are more expansion prone than others. For example, pluripotent stem cells from individuals with FRDA and DM1 show expansion while the fibroblasts from which they are derived do not (Du et al., 2012, 2013). In mouse models of the TNR diseases, some tissues are more expansion prone than others and these differ between different disease models (Clark et al., 2007; Fortune et al., 2000; Goula et al., 2009; Kennedy et al., 2003; Lokanga et al., 2013), suggesting that a combination of cell-type-specific factors and locus-specific factors must play a role in the determination of expansion frequency.

A role for cis-acting factors in expansion is suggested by the fact that in many of the repeat expansion diseases some haplotypes are more likely to expand than others (Ennis et al., 2007; Martins et al., 2008; Murray et al., 2000; Richards et al., 1992; Takiyama et al., 1995; Warby et al., 2009). In addition, a comparison of different CAG/CTG triplet repeat loci indicated that the flanking CG content and proximity to CpG islands modifies the repeat expandability (Brock et al., 1999; Nestor & Monckton, 2011). Another cis-acting factor-influencing repeat instability is orientation with respect to replication direction; in bacterial cells, yeast cells, and HeLa cell extract contractions were more frequent with the CTG repeat was on the lagging template strand, while CTG on the nascent lagging strand (Okazaki fragment) favored expansions (Cleary et al., 2002; Freudenreich et al., 1997; Kang et al., 1995; Maurer et al., 1996; Miret et al., 1998; Panigrahi et al., 2002). These data fit with the known areas of single-stranded character at the replication fork and the greater stability of the CTG hairpin: on the exposed lagging strand template, bypass of a CTG hairpin would lead to contraction, whereas a CTG hairpin on the 5′ flap of the nascent lagging strand would lead to an expansion (see Figure 2 and section ‘‘General mechanisms of TNR expansion and contraction’’). It also predicts that a switch in origin location with respect to the repeat could change the balance between expansions and contractions, the ‘‘ori switch model’’ (Cleary & Pearson, 2005; Mirkin & Smirnova, 2002). Indeed, recent data on the replication patterns at the DM1 and FMR1 loci in human cells shows evidence for changes in origin usage at unstable alleles (Cleary et al., 2010; Gerhardt et al., 2014a; Mirkin & Mirkin, 2014). At the FMR1 locus, a common SNP identified in one of the flanking origins has been shown to be associated with an expansion-prone haplotype (Gerhardt et al., 2014b).

Figure 2.

General mechanisms of trinucleotide repeat instability. (A) Polymerase slippage can lead to formation of secondary structures on the extending nascent strand (expansion) or template strand (contraction). Slippage can also occur at replication barriers, and structures can form on the nascent stands or template strands at the stalled replication fork. Reversed forks can be processed into a hairpin on the leading strand, causing an expansion. (B) Hairpins or quadruplexes caused by unprocessed 5′ flaps can form during gap repair or on an Okazaki fragment, leading to expansions. (C) Misalignment of repeat units (gray blocks) during homologous recombination can lead to the addition of repeat units (expansion event) or loss of repeat units (contraction event). (D) Resection during double-strand break repair exposes repeat units that can misalign and anneal, leading to repeat unit deletion (right). Misalignment followed by slippage during gap filling could lead to an expansion (left).

Another cis-acting factor shown to influence repeat instability is transcription. Interestingly, all expandable repeats (with the possible exception of the EPM1-causing dodecamer) are located in the transcribed part of their associated genes. A role for transcription in repeat expansion of CAG/CTG repeats is suggested by the comparison of two different transgenic mouse lines that exhibited a higher level of instability in the line showing the higher level of transcription (Goula et al., 2012b). Using controllable systems, it has been observed that induction of transcription in E. coli, Drosophila, and human tissue culture, all increase CAG repeat instability (Bowater et al., 1997; Jung & Bonini, 2007; Lin et al., 2006). In humans, silenced FX alleles are stable (Wohrle et al., 2001), and in mice with ~135 repeats in the endogenous FMR1 gene on the X chromosome, expansion is only seen in females when this allele is on the active X chromosome (Lokanga et al., 2014a). In a further twist, some repeat loci, such as the those that cause SCA8, FXS, SCA7, DM1, and ALS-FTD, are transcribed bi-directionally in humans (Ladd et al., 2007; Moseley et al., 2006; Nakamori et al., 2011; Sopher et al., 2011; Zu et al., 2013), and induction of bidirectional transcription in a human cell model further increased instability over unidirectional transcription (Lin & Wilson, 2012; Lin et al., 2010a,b). Transcription may be important because it creates an opportunity for the DNA hairpins, which are thought to be the substrates for expansion, to form. During passage through the repeat, the transcription complex would temporarily occlude the template strand, thus leaving the non-template strand transiently unpaired and able to form a structure, providing the trigger for expansion.

On the other hand, some animal models do not show a good correlation between the levels of transcription and the extent of expansion. Specifically, there is no correlation between the level of transcription of the disease locus and the extent of expansion in different tissues in either the DM1 or the FXS mouse model (Entezam et al., 2010; Lia et al., 1998) or between the amount of expansion seen in the fly model of SCA3 and the level of transcription induction (Jung & Bonini, 2007). In HD patients and an HD mouse model, expansion is more extensive in the striatum than the cerebellum despite the similar levels of HTT transcription (Dure et al., 1994; Goula et al., 2012b). In the case of the HD mouse, it has been suggested that tissue-dependent regulation of promoter proximal pausing at the HTT gene might underlie tissue-selective instability (Goula et al., 2013). A recent study in yeast suggests another model for the effect of transcription. Shah et al. (2014) showed that a GAA repeat located in a non-transcribed region upstream of an inducible promoter exhibited an increased rate of expansions when transcription was induced nearby. Further investigation revealed that induction of transcription resulted in lower nucleosome occupancy in the upstream region containing the GAA tract. The authors conclude that transcriptional state, rather than transcription elongation through the repeat, was responsible for the 10-fold increase in expansion rate that occurred under induced conditions (Shah et al., 2014). They speculate that the lower nucleosome occupancy occurring due to transcriptional activation allows more template switching during replication.

A less well understood cis-acting factor shown to influence repeat instability is the local chromatin structure. Alterations in the level of packing of the chromatin fiber, histone modifications, DNA methylation, chromatin binding proteins, and chromatin looping could all affect repeat instability via the ability either to influence transcription levels or facilitate the formation of secondary structures, or promote accessibility to DNA repair/recombination processes. See Dion & Wilson (2009) for a comprehensive review of this topic. For example, SCA7 mouse models only show instability of a (CAG)₉₂ tract if the ataxin-7 gene is flanked by human genomic DNA containing functional-binding sites for CTCF, a protein implicated in large-scale chromatin structure that has binding sites next to several expansion prone repeat loci (Libby et al., 2008). Many expandable repeats are near CpG islands, and several studies correlate changes in DNA methylation with repeat stability. For example, inhibition of DNA methylation increased expansions at the CTG tract in the DMPK locus in DM1 patient cells (Gorbunova et al., 2004), and deficiency of DNA methyltransferase Dnmt1 increased germ-line but not somatic expansions in a SCA1 mouse model (Dion et al., 2008). The effect of the Dnmt1 deficiency was not mediated by an increase in transcription through the repeat or by many genes thought to be important for expansion; therefore, a different aspect of the chromatin structure influenced by methylation may contribute to expansion propensity. Histone modifications may also affect repeat instability through changes in chromatin structure. In flies, lower histone acetylation levels (due to haploinsufficency of the histone acetyltransferase CBP) increase CAG repeat instability, while a higher level of acetylation (due to treatment with trichostatin A (TSA), a histone deacetylase inhibitor) decreases instability (Jung & Bonini, 2007). Both CpG methylation and histone acetylation influence nucleosome assembly on CGG repeats (Mulvihill et al., 2005), perhaps linking these two modifications. Acetylation of histone H3 at lysine 56 is important for nucleosome deposition during DNA replication and repair, and defects in H3-K56 acetylation were found to increase CAG contractions and fragility in yeast (Yang & Freudenreich, 2010). H3-K56 acetylation was hypothesized to facilitate proper nucleosome assembly at the replication fork to prevent DNA structure formation and subsequent slippage events on the template strand or fork breakage. An example of an epigenetic mark affecting repair within a repeat is provided by our recent study of factors that prevent CAG expansions in yeast, that revealed that acetylation of the histone H4 N-terminal tail by the NuA4 and Hat1 enzymes, and chromatin remodeling by RSC2, were important for the fidelity of repair within the repetitive DNA tract (House et al., 2014b) (see section ‘‘Post-replication gap repair and fork restart’’ of this review). In this case, the modification was shown to occur directly at the expanded repeat coincident with damage, but not at a no-repeat control in the same location. However, the effects of some epigenetic modifiers may be indirect. For example, knockdown of some histone deacetylases that suppress CAG expansions in yeast and human cells may do so by altering the stability of the Sae2 nuclease (human CtIP) (Debacker et al., 2012).

There is also evidence for trans-acting factors influencing expansion rate, and thereby affecting the risk of disease or age of onset. For example, the risk of expansion of some FMR1 alleles is lower in the general population than it is in families with a family history of FXS (Nolin et al., 2013). In myotonic dystrophy patients, the level of somatic instability is highly heritable, implying a role for individual-specific trans-acting genetic modifiers (Morales et al., 2012; Neville et al., 1994). Potential trans-acting factors include the absolute and relative levels of different proteins involved in the expansions as well as the levels of proteins involved in protecting the genome against these events. Recently, transcript and protein level analysis has revealed that there are significant expression differences in repair protein levels in tissues with different expansion propensities, including Msh2, Msh3, Msh6, PCNA, Rpa1, Fen1, Lig1, HMGB1, and Polβ (Du et al., 2012; Goula et al., 2009; Mason et al., 2014; Seriola et al., 2011; Tome et al., 2013), all proteins shown to influence TNR instability (see sections ‘‘The role of mismatch repair proteins in repeat instability’’, ‘‘The role of base excision repair proteins in repeat instability’’, ‘‘The role of nucleotide excision repair proteins in repeat instability’’, ‘‘Homology-dependent recombinational repair of gaps and stalled forks’’, and ‘‘The role of DSB repair in preventing repeat fragility and instability’’). Also, different alleles or expression levels of mismatch repair proteins have been identified in mouse strains that exhibit different expansion frequencies (Pinto et al., 2013; Tome et al., 2013).

General mechanisms of TNR expansion and contraction

Addition of repeat units by definition involves some sort of DNA synthesis. Early models of repeat expansion envisioned that the incorporation of additional bases might arise as a result of strand slippage during replication (Richards & Sutherland, 1994) (Figure 2A). It is now understood that strand slippage can occur in many different contexts, including DNA replication in dividing cells, as initially envisioned, or during various types of repair synthesis. In addition, three other mechanisms can create expansions: failure to remove displaced 5′ DNA flaps and their subsequent incorporation into DNA (Figure 2B), misalignment during recombinational repair (Figure 2C), and synthesis during double-strand break (DSB) repair (Figure 2D). These events can occur both during the duplication of genomes and during repair processes. In addition to expansions, these unstable elements can also lose repeat units, a process termed contraction. Contractions can occur by three basic mechanisms: replication over a template hairpin either during replication or during repair synthesis (Figure 2A), misalignment during recombinational repair (Figure 2C), or resection during double-strand break repair (Figure 2D). Again, each of these events can occur during several types of replication or repair processes.

Although the basic idea of how repeat units can be gained and lost is fairly straightforward, a more detailed understanding of how instability occurs in different repair pathways, and the identification of which mechanisms are operating in different cell types and developmental windows, has been surprisingly difficult to determine. The use of model organisms has been invaluable in this regard, as proteins can be manipulated to perturb particular pathways and measure any resulting repeat stabilization or destabilization. The picture that has emerged to date is complex and interesting, providing many potential targets for manipulating the process in human patients, with more certainly to be learned in the future. In addition, study of these structure-forming sequences has yielded valuable insights into basic mechanisms of replication and repair that have increased our understanding of these crucial cellular processes. The study of repeat instability and fragility mechanisms has been particularly valuable in providing insights into how normally genome-protective pathways can fail, leading to genomic changes, and ultimately to disease pathology.

DNA damage at structure-forming repeats

As illustrated in Figure 2, DNA structures can interfere with a number of DNA transactions. For example, GAA/TCC triplexes and GGC/CCG repeats strongly interfere with replication progression, acting as site-specific barriers (Anand et al., 2012; Krasilnikova & Mirkin, 2004; Voineagu et al., 2009b). CAG/CTG repeats are much weaker barriers, but replication bypass can still leave a gap that must be filled post-replicatively (Kerrest et al., 2009). Hairpins on 5′ flaps interfere with the ligation step of nick repair and Okazaki fragment processing, leaving an unligated nick (Liu & Bambara, 2003; Liu et al., 2004; Panigrahi et al., 2005; Spiro et al., 1999). Indeed, deletion of Fen1 (yeast Rad27), a key protein needed for 5′ flap processing, results in a large increase in CAG repeat fragility (Callahan et al., 2003). In addition to the difficulties in processing the secondary structures formed by the repeat, some of the disease-associated hairpins have been shown to be prone to oxidative damage (Jarem et al., 2009) and they may also be vulnerable to cleavage by single-stranded nucleases. Since even short TNRs can form DNA structures that interfere with DNA processes (see sections ‘‘Disease associated repeats form non-B DNA structures’’ and ‘‘Developmental timing of TNR expansions’’), cellular mechanisms that can facilitate repair of the nick, gap, or stalled fork are predicted to be important for tolerance of even normal-length TNRs in the genome, but particularly for expanded tracts, as well as other sites of structured DNA. Indeed, cellular checkpoint mechanisms are induced in cells with expanded TNRs, and checkpoint signaling proteins are important for TNR stability (see section ‘‘Role of the DNA damage checkpoint in sensing damage at TNRs and regulating repair’’).

Expanded repeats are fragile sites

Expanded TNRs are sites of increased chromosomal fragility, causing DNA breakage that can result in chromosome loss or rearrangement. A DSB at an expanded repeat can occur by multiple mechanisms: (1) processing of a stalled fork (e.g. by nucleases acting on the exposed ssDNA or a structured DNA, such as the four-way junction of a reversed fork or recombination intermediate); (2) conversion of an unligated nick to a DSB; (3) direct cleavage of the repeat, e.g. by structure-specific nuclease cleavage; (4) physical breakage or cleavage of an unreplicated region during cell division.

In humans, cells with expanded CGG repeats in the FMR1 or FMR2 genes exhibit a cytologically visible fragile site coincident with the repeat. The fragile site is seen as a gap, constriction, or break on metaphase spreads of chromosomes from cells grown with either very high or very low folate levels. This folate stress is thought to lead to a nucleotide pool imbalance that affects the rate of DNA replication through the repeat. Deletions and translocations at the site of CGG expansions have been documented, which provides indirect evidence that a double-strand break (DSB) occurred on the chromosome. Expanded repeats at the FMR1 gene are associated with very late replication and problems with replication initiation and/or elongation (Gerhardt et al., 2014a; Hansen et al., 1993, 1997; Subramanian et al., 1996; Yudkin et al., 2014).

For other expanded TNRs, there is no cytological appearance of a fragile site in human cells (Jalal et al., 1993; Wenger et al., 1996); however, loss of the chromosome with an expanded CTG repeat at the DM1 locus is increased (Casella et al., 2003), suggesting that breakage is occurring in vivo. In addition, data from yeast provide strong evidence that these sequences cause increased frequencies of chromosome breakage. CAG/CTG tracts of 45 units or longer can induce fragility of a yeast chromosome as measured by sensitive genetic assays that detect large deletions or chromosome end loss (Callahan et al., 2003; Freudenreich et al., 1998). Additionally, breakage of longer CAG tracts (79–250 repeats) are visible as physical chromosome breaks in mitosis (Freudenreich et al., 1998) and meiosis (Jankowski et al., 2000). GAA/TCC tracts and other AT-rich elements predicted to form hairpins or cruciforms also cause chromosome fragility (Kim et al., 2008; Lobachev et al., 2002; Zhang et al., 2012b), as does the ATTCT repeat unwinding element (Cherng et al., 2011). Comparison of CAG and CGG fragility using the same yeast genetic assay reveals that CGG repeats are 3–10-fold more fragile than CAG repeats of equivalent size, offering one explanation of why CGG fragility is more evident in human cells (Balakumaran et al., 2000). Chromosome context or location may also play an important role in determining levels of fragility. For example, an impediment to replication generated by repeats that are located in genes that normally replicate late may be more likely to result in cells entering mitosis before replication of that region is complete. Expanded CGG repeats seem to escape the DNA damage checkpoint in yeast, whereas CAG repeats do not, providing another explanation for the apparent increased fragility of the CGG repeat (Voineagu et al., 2009a).

Protective and mutagenic roles of DNA repair at repetitive DNA

Chromosome breakage at an expanded TNR represents the failure of normal repair processes. However, most of the time that a nick, gap, or DSB occurs at a TNR tract the cell will succeed in repairing that damage without chromosome arm loss, deletion, or rearrangement. The repair may occur with fidelity, i.e. with no loss or gain of genetic material, or with a change in repeat number, but with no consequence to the rest of the genome. This outcome represents successful repair (no fragility), but with the possibility of a detrimental repeat expansion; e.g. the ‘‘protective’’ repair process has gone awry because of the challenges of replicating or aligning DNA across a repetitive region, resulting in repeat instability. Lastly, there is evidence that some ‘‘repair’’ pathways, such as mismatch repair (MMR) and transcription-coupled repair (TCR) may inappropriately recognize extruded TNR hairpins or triplex structures as DNA damage, and initiate an unnecessary or ‘‘gratuitous’’ repair event that is in no way protective and leads to repeat instability (Gomes-Pereira et al., 2004; Pluciennik et al., 2013; Salinas-Rios et al., 2011). In the remainder of the review, we summarize what is known about DNA repair pathways whose components have been implicated in repeat instability.

The role of mismatch repair proteins in repeat instability

Mismatch repair (MMR) is normally involved in the repair of base mismatches or insertions/deletions (IDLs) (see Modrich, 2006 for review). It is initiated by recognition of the mismatch by MSH2/MSH6 (MutSα), a heterodimer that binds to single base mismatches and small IDLs or MSH2/MSH3 (MutSβ), a complex that is involved in the repair of longer IDLs. A MutL complex is then recruited to form a ternary complex. Three different MutL complexes are found in mammals: MutLα a MLH1/PMS2 heterodimer and perhaps the most important of the MutL complexes for MMR outside of meiosis; MutLβ, a MLH1/PMS1 heterodimer; and MutLγ, a heterodimer of MLH1/MLH3. The MutS/MutL complex then interacts with the homotrimeric-proliferating cell nuclear antigen (PCNA). This process activates the latent endonuclease activity of the MutL complex, resulting in the generation of nicks nearby. MutL nuclease activity is strand directed if there is a pre-existing nick to signal the nascent strand, or it can cleave either strand if no such signal exists. The Exo1 exonuclease is then loaded at the nicks by the MutS/MutL complex and generates a single-stranded gap starting at the nick and extending ~150 nucleotides past the mismatch. Gap filling mediated by Polδ then occurs, with DNA ligase I (Lig1) sealing the nick.

MMR proteins are important for repeat expansion in a wide variety of model systems, including mammalian cells and mouse models. MMR proteins also play a role in CTG/CAG repeat expansions in yeast, although to a lesser extent than in some of the mouse models (Kantartzis et al., 2012; Schweitzer & Livingston, 1997). Importantly, hairpins or slip-stranded structures formed by CTG/CAG repeats bind MutSβ in vitro (Lang et al., 2011; Owen et al., 2005, 2009), and binding to these repeats (Gannon et al., 2012) and to GAA/TTC-repeats has been reported in vivo (Du et al., 2012; Ezzatizadeh et al., 2012).

As can be seen in Table 2, the preponderance of evidence supports a role for a MutS complex in repeat expansion. MutS complexes, and in particular MutSβ, are required for both intergenerational and somatic expansion in most mouse models of repeat expansion. Some differences are seen between specific models that may reflect the effect of sequence context, since cis-acting sequences affect expansion (Brock et al., 1999; Cleary et al., 2002; Libby et al., 2003), or genetic background, since mouse strains differ significantly in the levels of expression of key MMR proteins (Pinto et al., 2013; Tome et al., 2013). Even in the case of FRDA in which none of the MutS components seem to play a role in intergenerational expansion in the transgenic mouse model (Ezzatizadeh et al., 2012), loss of either of the MutSα components reduces somatic expansions (Bourn et al., 2012). Furthermore, knockdown of MutSα reduces expansions of the endogenous FRDA allele in patient derived induced pluripotent stem cells (iPSCs) (Du et al., 2012; Ku et al., 2010) and expression of MutSβ promotes GAA expansions in a human kidney cell line (Halabi et al., 2012). Thus far FRDA appears to be unique in the contribution of MutSα along with MutSβ to expansion. This may reflect the fact the FRDA repeat forms a triplex while the other repeats examined do not.

Table 2.

The requirement of MMR proteins for repeat expansion in different mouse models of the TNRs.

Genes (complex)	Disease model	Germ line	Somatic	References
MSH2	DM1 (TG)	Yes (95%)	Yes	Savouret et al. (2003) and Tome et al. (2009)
(MutSα; β)	HD (TG)	Yes	Yes	Manley et al. (1999) and Kovtun & McMurray (2001)
	HD (KI)	Yes (♂), no (♀)	Yes	Wheeler et al. (2003)
	FXS (KI)	Yes	Yes	Lokanga et al. (2014b)
	FRDA (TG)	No	Yes	Ezzatizadeh et al. (2012) and Bourn et al. (2012)
MSH3	DM1 (TG)	Yes (~80%)	ND	Foiry et al. (2006)
(MutSβ)	DM1 (KI)	ND	Yes	van den Broek et al. (2006)
	HD (TG)	ND	Yes	Tome et al. (2013)
	HD (KI)	No	Yes	Kovalenko (2012) and Dragileva et al. (2009)
	FRDA (TG)	No	No	Bourn et al. (2012)
MSH6	DM1 (TG)	Yes (♀)*, no (♂)	ND	Foiry et al. (2006)
(MutSα)	DM1 (KI)	ND	Protects	van den Broek et al. (2006)
	HD (TG)	No	No	Kovtun et al. (2004) and Tome et al. (2013)
	HD (KI)	No	No	Kovalenko et al. (2012) and Dragileva et al. (2009)
	FRDA (TG)	Protects	Yes	Ezzatizadeh et al. (2012) and Bourn et al. (2012)
MLH1	HD (KI)	ND	Yes	Pinto et al. (2013)
(MutLα; β; γ)	FRDA (TG)	Yes	Yes	Ezzatizadeh et al. (2014)
MLH3 (MutLγ)	HD (KI)	ND	Yes	Pinto et al. (2013)
PMS2	DM1 (TG)	ND	Yes (50%)	Gomes-Pereira et al. (2004)
(MutLα)	FRDA (TG)	Protects	Protects†	Ezzatizadeh et al. (2012) and Bourn et al. (2012)

TG, transgenic; KI, knock-in; ND, not determined; %, fraction of expansions that are dependent on the gene in question.

^*Effect may be indirect via loss of MSH3.

^†In neural tissue but not non-neuronal tissue.

The requirement for MMR proteins is not limited to the MutS complex. Mlh1, the protein common to all three MutL complexes found in mammals, and Mlh3, the Mlh1-binding partner in MutLγ, are required for all somatic expansions in an HD mouse model (Gomes-Pereira et al., 2004; Pinto et al., 2013). Since MutLγ is thought to interact with MutSβ not MutSα (Charbonneau et al., 2009), this supports a key role for MutSβ in expansion of CAG-repeats. However, in a DM1 mouse model, PMS2, the binding partner of Mlh1 in MutLα, is required for 50% of somatic expansions (Gomes-Pereira et al., 2004). Despite the lack of a role for MutSα or MutSβ in intergenerationally transmitted expansions in the FRDA mouse (Bourn et al., 2012; Ezzatizadeh et al., 2012), Mlh1 is required for both germ line and somatic expansions (Ezzatizadeh et al., 2014). Since PMS2 (MutLα), protects against expansions in this model (Bourn et al., 2012) this suggests that there is also a role for MutLγ in FRDA expansions.

A role of MutLγ in repeat expansion is intriguing since Mlh3 is present in mammalian cells at levels 60 times lower levels than PMS2 (Cannavo et al., 2005). While its role in meiotic recombination has been known for some time, the role of MutLγ in typical MMR is not well understood. MutLγ is able to carry out only low levels of MMR in vitro, and recent reports suggest that while MutLγ can localize to sites of DNA damage, it is unable to substitute for MutLα in MMR in vivo (Roesner et al., 2013). It is of interest to note that the MutLγ complex shows a marked preference for binding to Holliday junctions (Ranjha et al., 2014). It may be that the requirement of the MutLγ complex reflects a role in the recognition and processing of a similar DNA structure.

The requirement of a MutL complex suggests that the role of MMR proteins is not limited simply to lesion recognition. Since loading of the proliferating cell nuclear antigen (PCNA) clamp that is required for MutL activation during MMR requires a strand discontinuity, MMR was once thought to be confined to replication when leading and lagging strand termini are available for PCNA loading. However, it has recently been shown that both small CAG and CTG loopouts of 2–3 repeats support PCNA loading, and thus the MutLα activation that could allow MMR to occur even in post-mitotic cells (Pluciennik et al., 2013). Thus expansions might arise in such cells by a process like that outlined in Figure 3(A) in which PCNA loading would be facilitated by the presence of the hairpins. These hairpins could be preformed slipped strand structures as depicted in Figure 3(A), or formed by another process (e.g. gap filling, or transcription; see Figures 3B, 4, and 5). PCNA loading in turn results in activation of the latent endonuclease activity of MutLα and the generation of nicks that can be used by Exo1 to remove the region of DNA across from the template hairpin. Polδ would then fill in the gap. To avoid a contraction, the template hairpin would have to be unwound. We suggest the WRN DNA helicase (yeast Sgs1), which interacts with Polδ and Mlh1 and has been shown to prevent contractions (see section ‘‘The role of helicases in resolving hairpins during repair and fork restart’’), as a good candidate for acting at this step. In the absence of Exo1 digestion, repair could also occur via strand displacement synthesis by Polδ (Kadyrov et al., 2009) with or without strand slippage (dotted arrow to Figure 3B). After fill-in, Lig1 would then seal the nick generating a heteroduplex with a single loopout. A repair reaction originating on the complementary strand would result in the resolution of the heteroduplex by fill-in, resulting in an expansion. Since MutLα nicking on a template without preexisting nicks occurs without strand bias (Pluciennik et al., 2013), nicking could occur on either strand. The location of the nicks would determine the order of events and whether expansion, contraction or error-free repair results. Although this pathway can occur in non-replicating cells, if it did occur in the context of replication, recognition of a template hairpin by MutSβ followed by MutL cleavage could create a DSB; expansions and contractions could then occur during fork restart or DSB repair (Pluciennik et al., 2013).

Figure 3.

Model for generation of repeat expansions during MMR and BER. (A) Mismatch repair (MMR) is initiated when MutSβ (a heterodimer of MSH2/MSH3) binds to and stabilizes the hairpin formed by the repetitive DNA element. PCNA loading (facilitated by RFC, not shown) also occurs at the extrusion. The interaction between MutSβ and PCNA activates the cryptic endonuclease activity of the MutLα complex (a heterodimer of MLH1/PMS2). MutLα can nick either strand; here, a nick on the top strand is shown. Exo1 exonucleolytic activity digests the end to form a single-stranded gap that extends ~150 nt past the hairpin. The hairpin is unwound by helicase activity, potentially Sgs1/WRN (see section ‘‘The role of helicases in resolving hairpins during repair and fork restart’’), and Polδ fills the gap. The nick is sealed by Lig1. These steps are repeated on the other strand to resolve the second hairpin; DNA synthesis through the hairpin at this stage leads to an expansion on both strands. Note that after nicking, repair could also proceed by strand displacement (indicated by dotted gray arrow to part B), leading to an expansion by slippage during replication or inefficient flap cleavage by Fen1. (B) BER is initiated to repair 8-oxoG lesions caused by oxidative damage. 8-oxoG is first recognized by a DNA glycosylase (Ogg1 or NEIL1). ApeI creates a nick, and strand slippage during fill in by Polβ, Polδ, Polε, or Polη can lead to an expansion (left pathway). Alternatively, hairpins can impede Fen1 cleavage (indicated by dotted Fen1), leading to inefficient or ‘‘alternative’’ flap cleavage (indicated by solid Fen1) and expansions (right pathway). A nick on the complementary (bottom) strand followed by DNA synthesis through the hairpin leads to an expansion on both strands. Alternatively, direct nicking of the hairpin, or oxidative damage within the hairpin followed by nicking, could start the BER cycle again, leading to further expansions (toxic oxidation cycle, not shown). (see colour version of this figure at www.informahealthcare.com/bmg).

Figure 4.

Repeat instability through transcription-coupled repair. In this model, RNA polymerase stalls due to R loop formation and/or the formation of secondary structures on the non-template strand, which could be facilitated by an R-loop on the template strand. The folded structures could also facilitate the stall by sequestering the non-template DNA strand, thus decreasing its propensity to displace RNA (Belotserkovskii et al., 2013). An alternate model (not shown) is that the RNAPII stalls at pre-formed template hairpins (e.g. see Lin et al., 2009). Structures formed on the non-template strand during transcription could be stabilized by MutSβ binding, further increasing the strength of RNAPII stalling. Stalled transcription recruits the transcription arrest factors, including CSB and XPG, which initiate TCR. Alternatively, CSB may facilitate glycosylase activity to initiate BER, or the MutSβ complex could recruit MutLα to initiate MMR (left and right arrows). RNAPII is displaced and TFIIH is recruited; RPA and XPA stabilize the denatured bubble. The RNAPII-blocking lesion then undergoes dual incisions, the first carried out by XPF-ERCC1 that cleaves 5′ of the lesion. The second cleavage occurs downstream of the lesion and is carried out by XPG. The result would be a 25–30 nt gap that is filled by Polδ, Polε, and/or Polκ; repair replication may begin before 3′ cleavage (Staresincic et al., 2009). If the polymerase fills the gap faithfully no tract length change occurs (left). If strand slippage occurs during DNA synthesis, this leads to an expansion (center). If the polymerase replicates over a template hairpin formed due to exposed ssDNA on the bottom strand or the hairpin is excised, this will result in a contraction. (see colour version of this figure at www.informahealthcare.com/bmg).

Figure 5.

Post-replicative repair at repetitive DNA elements. Repeat units indicated reflect experimental constructs, but the mechanisms described here likely work along a continuum based on repeat length. (A) Repeats that cause fork stalling or fork reversal can initiate Rad5-dependent template switching, leading to expansions. In rad5Δ cells, these expansions are eliminated. In the absence of fork stabilization proteins (Mrc1, Tof1, and Ctf18) instability and fork breakage are both increased. (B) Hairpins at medium CAG repeat tracts are bypassed during replication and Rad5-dependent template switching is initiated at post-replication gaps. Rad5-, Rad51-, Rad52-, Rad57-, and Rad54-mediated expansions occur during sister chromatid recombination. Excessive recombination is inhibited by Srs2 anti-recombinase function. (C) At short repeat lengths, Rad5-dependent PRR and unwinding by the Srs2 helicase are sufficient to fill gaps and prevent expansions without sister chromatid recombination. (see colour version of this figure at www.informahealthcare.com/bmg).

In addition to canonical MMR, a variety of non-canonical roles for MMR proteins are known, including their involvement in the repair of oxidative damage, homologous recombination (HR), double-strand break (DSB) repair, immunoglobulin class switching, and somatic hypermutation. It may thus be that the MutS and MutL complexes affect expansions via their participation in other repair pathways. Of interest in this regard is the observation that while MutSβ binding to a CAG-hairpin in bulk solution has been reported to be indistinguishable from binding to a mismatch that is efficiently repaired by MMR (Tian et al., 2009), single molecule FRET has shown that MutSβ binding results in a subset of molecules having a different conformation (Lang et al., 2011). This complex is proposed to trap MutSβ and inhibit its dissociation from the hairpin. This has led to the suggestion that hairpin processing by MMR would be inefficient and could result in the hairpin being repaired by an alternative DNA repair pathway (McMurray, 2008). On preformed loop/hairpin substrates in vitro, expansion can be observed even in the absence of MMR proteins (Panigrahi et al., 2005, 2010). This in vitro repair is also independent of proteins involved in base excision repair and nucleotide excision repair. Also, in a HeLa nuclear extract, excess MutSβ does not inhibit or promote CTG or CAG repair of preformed hairpin substrates (Tian et al., 2009; Zhang et al., 2012a). These results have led to the suggestion that MutSβ may act to promote the formation of the secondary structures rather than determining repair outcome.

The role of base excision repair (BER) proteins in repeat instability

BER is the major pathway by which oxidized bases are repaired in mammalian cells. This process is initiated either by the removal of the damaged base by a DNA glycosylase or by the spontaneous hydrolysis of the glycosidic bond connecting the damaged base to the sugar phosphate back-bone (see Liu & Wilson, 2012 for a review). The resultant abasic site is processed by the apurinic/apyrimidinic (AP) endonuclease 1 (APE1) or the AP lyase function of some glycosylases to generate a single nucleotide gap (Figure 3B). If the AP site or the 5′-sugar phosphate group is not oxidized, the gap is processed by the single-nucleotide BER (SN-BER) pathway. In SN-BER, a single nucleotide is inserted by DNA polymerase β (Polβ) and the nick is sealed by DNA ligase III/XRCC1 or DNA ligase 1 (Lig1). In the absence of a ligatable nick, a second BER pathway, the long patch (LP)-BER pathway, is used instead. The most efficient form of the LP-BER pathway has been suggested to involve the coordinated actions of Polβ and Fen1 acting via a ‘‘Hit and Run’’ mechanism that consists of stepwise and distributive gap-filling by Polβ and single-nucleotide gap formation by Fen1 to generate a ligatable nick that is then sealed by Lig1 (Liu et al., 2005). This process yields repair patches of two nucleotides or more without the generation of a long 5′ flap resulting from strand displacement synthesis. LP-BER can also occur by strand-displacement synthesis that involves Polβ or the replicative polymerases Polδ and/or Polε, and perhaps Polη or other TLS polymerases. In this case, Fen1 removes the resultant flap (Figure 3B).

A role for BER in repeat expansion is suggested by the fact that treatment of HD fibroblasts with H₂O₂ increases the expansion frequency (Kovtun et al., 2007), as does exposure of FX premutation mice to potassium bromate (Entezam et al., 2010). The fact that the loss of either OGG1, the major DNA glycosylase responsible for removing 7,8-dihydro-8-oxoguanine (8-oxoG), the most common DNA oxidation product, or the endonuclease VIII-like 1 (NEIL1) glycosylase, decreases somatic expansions in a transgenic HD mouse model (Kovtun et al., 2007; Mollersen et al., 2012) would be consistent with this idea.

Biochemical experiments on CAG/CTG-repeat substrates have led to a number of different models being proposed for expansion via the LP-BER pathway. One model is based on the observation that during repair of an 8-oxoG lesion in the vicinity of a CAG repeat tract, multinucleotide gap filling occurs that exhibits periodic pausing sites consistent with hairpin formation (Liu et al., 2009). According to this model, expansion occurs during Polβ-mediated LP-BER if the optimal coordination between Fen1 and Polβ is disrupted. Expansions would arise because strand displacement and hairpin formation is able to occur before ligation, and Fen1 cannot cleave at the 3′ base of the resultant hairpin (Figure 3B, right pathway). Polβ would then carry out multinucleotide gap-filling synthesis and Fen1 would be forced to use its alternate cleavage activity to process only a few nucleotides off of the 5′ flap to generate a substrate suitable for ligation (Liu et al., 2009). The net result would be that many of the flap bases would be incorporated into the nascent strand (Figure 3B, right pathway; Fen1 ‘‘alternate’’ cleavage of flap 5′ to hairpin). A second model invokes strand-slippage on the nascent strand in the repeat region during LP-BER in the presence of both Polβ and Polδ. This model is based on the observation that while in the absence of Polβ, Polδ is able to use its 3′–5′ proof-reading ability to excise hairpins that lack a 3′ tail, in the presence of Polβ excision of this hairpin does not take place. Instead Polβ adds a few nucleotides to the 3′ end of the hairpin thus generating an effective primer for Polδ-catalyzed DNA synthesis that occurs without hairpin excision. The net result is that the hairpin becomes fixed in the nascent strand and can result in a repeat expansion (Chan et al., 2013) (Figure 3B, left pathway). In either case, since the hairpins formed by CAG and CTG repeats are hot spots for oxidation, and since OGG1 has a reduced affinity for 8-oxoG in these hairpins and excises them at a significantly lower rate compared with duplexes (Jarem et al., 2011), hairpin formation could result in a ‘‘toxic oxidation cycle’’ in which the repair of one lesion would increase the opportunity for the generation of additional oxidized bases.

A key step in the LP-BER pathway is flap cleavage by FEN1. The formation of secondary structures inhibits FEN1 cleavage of 5′ flaps in vitro (Spiro et al., 1999) and FEN1 has been shown to be important in creating expansions in an in vitro system that models BER (Liu et al., 2009). Fen1 heterozygosity resulted in fewer intergenerational contractions in an HD mouse model (Spiro & McMurray, 2003). However, Fen1 heterozygosity did not affect the frequency of expansions or contractions in either a FX or a DM1 mouse model (Entezam et al., 2010; van den Broek et al., 2006), or were expansions affected in DM1 Fen1^−/− blastocysts that stained negatively for Fen1 protein (van den Broek et al., 2006). Since Fen1^−/− blastocysts do not survive, and the level of Fen1 in heterozygotes may still be high enough to sustain the normal level of expansion, the question of a role for Fen1 in germ line expansions in mouse models remains open. In a yeast study, cells heterozygous for Fen1/Rad27 maintained (CAG)_70-130 repeat lengths; expansions increased only when the tract reached 155 repeats, and the frequency was still less than in fen1Δ cells (Yang & Freudenreich, 2007). In a mouse model of DM1, a hypomorphic mutation of DNA ligase 1 (46BR), reduced intergenerational repeat expansions and increased contractions but only when the expansion prone allele was maternally transmitted (Lopez Castel et al., 2011). However, the same mutation had no effect on intergenerational and somatic expansions in the FX mouse model (Entezam et al., 2010). It may be that a delicate balance exists between Fen1 and Lig1 (and perhaps other factors) so that an effect on expansions is not always seen in heterozygous animals. In the HD mouse model, the stoichiometry of proteins involved in BER has been suggested to explain why the striatum is more prone to expansion than the cerebellum. Specifically, a low ratio of proteins like FEN1, APE1, and LIG1 relative to Polβ would favor strand displacement that would lead to expansions (Goula et al., 2009, 2012a). Whether this correlation holds up when a larger range of tissues are tested remains to be seen.

Expansions in the HD mouse model can still occur in the absence of OGG1 or NEIL1. However, this does not necessarily invalidate a BER model. It could be that other glycosylases are able to compensate for the loss of these enzymes or that other sorts of DNA damage also contribute to expansion. Alternatively, Polη has recently been shown to be recruited by the MutS complex to participate in BER of tandem lesions that are not recognized by DNA glycosylases (Zlatanou et al., 2011), but whether MutSβ is also able to recruit Polη in response to hairpin lesions has not, to our knowledge, been tested. One advantage of a mechanism such as this is that it could account for the requirement of MMR proteins for expansion in vivo, using a combination of MutS-mediated lesion recognition and the BER pathway (Figure 3). A role for Polη in expansion is intriguing given the demonstration that this polymerase is involved in replicating past DNA structures present at common fragile sites (Bergoglio et al., 2013). However, it is not necessary to invoke this particular pathway in order to account for the role of MMR proteins. For example, it may be that the MMR proteins act to facilitate the formation or to stabilize the hairpin intermediates arising during LP-BER.

In addition to a role in generating expansions, there is evidence to suggest that BER can result in contractions. Oxidative damage to human cells containing plasmids carrying CTG-repeats, or treatment of FRDA lymphoblasts with temozolomide, an alkylating agent, result in the generation of contractions (Lai et al., 2013, 2014). BER of the lesion caused by oxidative damage would involve the generation of an abasic site, and the presence of such a site within a synthetic (CTG)₂₀ or (GAA)₂₀ tract results in both deletions and expansions in in vitro reactions with reconstituted with BER proteins; the location of the damaged base has been proposed to govern whether BER results in expansion or contraction (Lai et al., 2013). Contractions have been suggested to occur via hairpin bypass by Polβ (Xu et al., 2013). There is also evidence to suggest that BER can protect against expansions (Xu et al., 2014).

The role of nucleotide excision repair (NER) proteins in repeat instability

NER is involved in the repair of damaged bases including oxidized bases. NER is composed of two different pathways: one that occurs globally (global genome repair; GGR) and one that is confined to actively transcribed genes (transcription coupled repair; TCR) (see Hanawalt & Spivak, 2008 for review). In GGR, proteins like the XPC–Rad23B complex are responsible for DNA damage recognition, while in TCR, RNA polymerase stalls at a lesion and the blocked polymerase acts as the DNA damage signal, leading to the recruitment of proteins specific to TCR such as the Cockayne Syndrome B (CSB) protein (also known as ERCC6). The two pathways converge downstream of the DNA damage signal step and share a common lesion excision process in which an incision by XPF-ERCC1 5′ of the lesion and by XPG 3′ of the lesion lead to removal of 25–30 nucleotides. Repair is completed by Polδ, Polε, and/or Polκ and the nick is sealed by Lig1 or Lig3/XRCC1. A role for TCR in repeat expansion is appealing given the evidence for the importance of an open chromatin configuration and/or transcription for repeat instability (Goula et al., 2012b; Jung & Bonini, 2007; Lin & Wilson, 2012; Lin et al., 2006, 2010a,b; Lokanga et al., 2014a; Nakamori et al., 2011) and the fact that in HeLa cell extracts, slip-outs of (CAG)₂₀ or (CTG)₂₀ repeats on either strand cause stalling of RNA polymerase (Salinas-Rios et al., 2011).

Loss of XPC has little effect on repeat expansion in an HD mouse model (Dragileva et al., 2009) and siRNA knockdown of XPC had no effect in a human cell model for CAG/CTG contractions (Lin et al., 2006). This suggests that GGR is not responsible for generating either expansions or contractions, at least in these model systems. In contrast, there are a number of lines of evidence implicating TCR proteins in repeat instability. The loss of the fly XPG homolog, Mus201, reduced expansion in a Drosophila model of SCA3 (Jung & Bonini, 2007) while the loss of CSB in a mouse model for FXS led to a lower frequency of germ-line expansions in older females and a reduction in the extent of somatic expansions in some tissues (Zhao & Usdin, 2014). A role for CSB in expansions is appealing given the demonstration that a CSB variant is associated with increased expansion risk in humans with SCA3 (Martins et al., 2014). However, in a HD mouse model, the loss of CSB alone had no significant effect on somatic expansions in a wild-type background, although it did increase the somatic expansion frequency in Ogg1 null mice (Kovtun et al., 2011). An increase in the germ-line expansion frequency was also seen in Csb^−/− HD mice, although this increase did not reach statistical significance. Loss of XPA, a protein involved in stabilizing the denatured transcription bubble during TCR, also reduced somatic expansions in the neuronal tissue of a mouse model for SCA1, but had no effect on intergenerational expansion frequencies (Hubert et al., 2011). SiRNA knockdown of XPG, ERCC1, and CSB reduced repeat contractions in a human cell model of CAG repeat contractions (Lin & Wilson, 2007). A role for XPG in generating contractions would be consistent with the observation that XPG stimulates hairpin removal in vitro (Hou et al., 2011). Thus, evidence suggests that TCR proteins may contribute to both expansions and contractions in some model systems.

It is possible to envision a model for TCR in the generation of both expansions and contractions, as illustrated in Figure 4. TCR could be induced during transcription stalling either as a result of the R loops that are known to be formed by many disease associated repeats (see section ‘‘Disease-associated repeats form non-B DNA structures’’) or the formation of secondary structures on the non-template strand (Belotserkovskii et al., 2013; Duquette et al., 2004). It is also possible that RNAPII stalls at hairpins formed during passage of a previous transcription complex. CSB and other factors are then recruited to the stalled transcription complex to initiate TCR, RNAPII is displaced, and the denatured bubble is stabilized by XPA and RPA. The RNAPII-blocking lesion is then removed by dual incision, the first carried out by XPF-ERCC1 that cleaves 5′ of the lesion, the second carried out by XPG occurring downstream of the lesion. Since XPG does not require a free 5′ single-stranded end (Hohl et al., 2003), cleavage may not be impeded by the structure in the flap. The result would be a 25–30 nt gap that is filled by Polδ, Polε, and/or Polκ. If the polymerase fills the gap faithfully, no tract length change occurs and the original allele is regenerated. If strand slippage occurs during gap filling DNA synthesis, an expansion would result if slippage involved the nascent strand and a contraction would result if it was the template strand that slipped (Figure 4).

The knockdown of MutSβ as well as various components of the TCR pathway both reduced the mutation rate in an in vitro assay for repeat contractions (Lin et al., 2006). It may be that MutSβ acts to facilitate repeat instability by binding to transcription-induced hairpins, thereby blocking RNAPII progression and inducing TCR. It is interesting that in yeast, Msh2 has physical and genetic interactions with multiple members of the TCR pathway (Saccharomyces Genome Database). Since MutLα knockdown increases contractions (Lin et al., 2009), it was suggested that engagement of MutLα may divert the repair away from the contraction-prone TCR pathway (Lin & Wilson, 2012), perhaps by favoring repair of the hairpin by MMR as illustrated in Figure 3.

There is also the intriguing possibility that some TCR components are acting outside of the TCR pathway to promote expansions (Figure 4). Since CSB is essential for TCR, the fact that loss of Csb in the FXD mouse model does not eliminate expansions completely, would be consistent with the idea that Csb plays an auxiliary role in expansion (Zhao & Usdin, 2014). For example, it could act in a BER-based pathway via its ability to increase the incision activity of NEIL1 and OGG1 (Muftuoglu et al., 2009; Tuo et al., 2002) and to positively regulate hOGG1 gene expression (Javeri et al., 2011; Tuo et al., 2002). CSB also facilitates annealing of single strands (Muftuoglu et al., 2006), which could promote hairpin formation.

Homology-dependent recombinational repair of gaps and stalled forks

Somatic expansions in differentiated cells are most likely occurring by one or more of the non-replicative mechanisms described in the previous sections. However, as summarized in section ‘‘Developmental timing of TNR expansions’’, there is evidence that CAG/CTG repeat expansions occur in replicating pre-meiotic cells, resulting in intergenerational transmission of disease, and that some replicating somatic tissues incur expansions. Instability in these cell types could arise either by a non-replicative repair pathway or in a replication-coupled repair process. Studies in model systems, summarized below, confirm that replication-coupled repair processes can be a source of large expansions.

As discussed earlier, DNA structures formed by repetitive DNA sequences are impediments for DNA synthesis (Kerrest et al., 2009; Krasilnikova et al., 2007; Liu & Leffak, 2012; Pelletier et al., 2003; Shah et al., 2012; Voineagu et al., 2009b) and can cause fork stalling or, if the lesion is bypassed, a gap behind the replication fork. Such ssDNA gaps occur when the replication fork encounters a lesion and leading and lagging strand synthesis become uncoupled (reviewed by Heller & Marians, 2006)). Also, nicks or small gaps that arise independently of DNA replication, for instance during BER or NER, can be enlarged or become DSBs if replicated (Byun et al., 2005; Hashimoto et al., 2010).

Post-replication gap repair and fork restart

DNA damage tolerance (DDT) pathways, also referred to as post-replication repair (PRR), are dependent on ubiquitination of PCNA, and can be subdivided into two categories: translesion synthesis (TLS) and error-free template switching (reviewed by Boiteux & Jinks-Robertson, 2013; Saugar et al., 2014). The TLS branch is dependent on the PCNA ubiquitin ligases, Rad6 and Rad18, which together monoubiquitinate Lys164 of PCNA. The monoubiquitin signal is required for the recruitment of translesion polymerases (e.g. Polζ or Polη) that synthesize across the lesion. TLS does not require invasion into a homologous template and can be mutagenic. However, mutations in the TLS polymerases had no effect on CAG repeat instability (Dixon & Lahue, 2002; Freudenreich et al., 1998) or GAA repeat stability (Shah et al., 2012) in budding yeast, indicating that the TLS pathway is not a significant source of expansions. Recent data suggests that when replicative polymerases are compromised, some GAA repeat expansions do occur by a Polζ-dependent mechanism (Shah et al., 2012), as do short duplications initiated by small hairpins (Northam et al., 2014), so this conclusion could be revisited.

The error-free branch of PRR is dependent on the mono-ubiquitination of PCNA by Rad6–Rad18 E2–E3 ubiquitin ligases but then requires additional ubiquitylation action by Ubc13–Mms2–Rad5 E2–E3 ubiquitin ligases (mammalian HLTF/SHPRH), leading to a poly-ubiquitinated PCNA molecule (Boiteux & Jinks-Robertson, 2013; Saugar et al., 2014). Error-free PRR further requires the action of the homologous recombination proteins Rad51, Rad52, Rad57, and Rad54 (Gangavarapu et al., 2007; Minca & Kowalski, 2010; Vanoli et al., 2010). The requirement for HR proteins is consistent with the use of the undamaged sister chromatid as a template for synthesis through a poorly understood template-switch mechanism that is similar to sister chromatid recombination (Boiteux & Jinks-Robertson, 2013; Saugar et al., 2014). Indeed, Rad5 is required for sister chromatid recombination at non-repetitive sequences (House et al., 2014b; Zhang & Lawrence, 2005). In Saccharomyces cerevisiae, spontaneous sister chromatid recombination (SCR) is proposed to occur because of gaps formed behind the replication fork (Mozlin et al., 2008), and is induced by both CAG repeats and inverted repeats (Nag et al., 2004). It is the propensity of these sequences to form secondary structures that is thought to impede replication and induce sister chromatid recombination, which is dependent on the presence of Rad52 (Nag et al., 2004).

Rad5, Rad18, HTLF, and PCNA ubiquitination have all been shown to inhibit (CTG)₁₃ or (CAG)₂₅ expansions in yeast and human cells (Daee et al., 2007; Frizzell et al., 2014). At a longer (CAG)₈₅ repeat, deletion of Rad5 also mildly increased expansions, 3-fold over wild-type (House et al., 2014b). The role of Rad5 at short CAG repeats was epistatic to the Srs2 protein, a 3′–5′ helicase and anti-recombinase, implicating a role for Srs2 in DNA damage tolerance as well, perhaps by unwinding hairpins (Daee et al., 2007; Dhar & Lahue, 2008). Yeast lacking Srs2 display an increase in CAG and CTG fragility and instability, which for longer repeats is dependent on the presence of Rad51 (Bhattacharyya & Lahue, 2004; Kerrest et al., 2009). In a haploid strain, two-dimensional (2D) gel electrophoresis analysis showed the formation of joint molecules at a (CAG)_45-55 tract (Kerrest et al., 2009). The joint molecules were dependent on the presence of Srs2 but not Rad51, indicating that they could be reversed forks or other template switch events between sister chromatids (Kerrest et al., 2009). Altogether these data indicate that CAG tracts induce both SCR and template switch events, and that, in yeast, the Srs2 helicase is important to prevent instability during this process. This template switch event must play an important role in repairing TNR-related gaps, because in its absence (e.g. in rad5Δ or rad18Δ strains) expansions occur, by an alternative unknown pathway.

The chromatin environment at gaps also contributes to efficient repair by error-free PRR (reviewed in House et al., 2014a). The absence of Anc1, a subunit of the chromatin modifying complexes INO80, SWI/SNF, and NuA3, leads to an increase in (CAG)₂₅ expansion frequency that is equivalent to the increase in rad5Δ and mms2Δ mutants (Daee et al., 2007; Erlich et al., 2008). However, Anc1 may play a role in an as yet undefined branch of PRR since it was not epistatic to all members of a single branch of error-free damage tolerance (Erlich et al., 2008). Indirect effects of chromatin modifiers also need to be considered. For example, knockdown of histone deacetylases (HDACs) Rpd3L and Hda1 (hHDAC3) suppresses CAG expansions in yeast and human cells, potentially by altering the stability of the Sae2 nuclease (Debacker et al., 2012). Our group recently demonstrated that histone H4-K16 acetylation by the chromatin-modifying complex NuA4 is specifically enriched at expanded CAG repeats in yeast, and is required for high-fidelity error-free PRR and (CAG)₈₅ repeat maintenance (House et al., 2014b). In the absence of Esa1, the catalytic subunit of the NuA4 complex, repeat expansions were significantly increased compared with wild-type cells and this increase was dependent on Rad52, Rad57, and Rad5, indicative of a template switching event that requires invasion into the sister chromatid (House et al., 2014b). Further, Rsc2, an acetyl-lysine-binding subunit of the RSC chromatin remodeler, was recruited to the CAG repeat coincident with the peak in H4-K16 acetylation, suggesting a possible recruitment mechanism for chromatin modifiers to promote template switching without expansions (House et al., 2014b). Thus, although template switch is a protective pathway, it can be a source of repeat length changes if it occurs without accessory factors such as helicases (Srs2) and chromatin remodelers or modifiers (Rsc2, H4-K16ac, and Anc1) that allow it to occur with fidelity.

In contrast to short (CAG)_13-25 repeats, ATTCT and GAA repeat expansions are promoted by the presence of Rad5 in yeast (Cherng et al., 2011; Shishkin et al., 2009), analogous to the situation described above for expansions of (CAG)₈₅ that occur in the absence of H4-K16ac. In addition, a rad5Δ mutant displays decreased ATTCT fragility (Cherng et al., 2011), suggesting that template switching events can lead to chromosome breaks at this DNA unwinding element. Rad5-dependent expansions of the GAA repeat were proposed to occur by a template switching mechanism in which the GAA repeat expansions arise from dissociation of the leading strand from its normal template and aberrant copying from the newly synthesized Okazaki fragment (Shishkin et al., 2009). The model predicts that copying would not be dependent on DNA structure per se, but would be facilitated by pausing of the replication fork (Cherng et al., 2011; Shishkin et al., 2009).

How can one explain the different dependencies on Rad5 and Rad51/Rad52 observed for different types and sizes of repeats? We propose the following model to account for the somewhat contradictory roles of proteins in the PRR pathway on repeat instability (Figure 5). For longer or more ‘‘slippery’’ repeats (GAA, ATTCT, and longer CAGs), the fork stall could be strong enough to mediate a template switching event directly at the stalled fork, which could be equivalent to a fork reversal, hypothesized to be facilitated by Rad5 (Figure 5A). There is experimental evidence for fork reversal at both CAG and GAA repeats by direct visualization of replication intermediates by 2D gel electrophoresis and electron microscopy (EM) (Follonier et al., 2013; Fouche et al., 2006; Kerrest et al., 2009). For CAG repeats, the size needed to produce a fork stall stable enough to be visualized on a 2D gel is approximately 90–100 CAGs (G. -F. Richard, personal communication). Two models are drawn for the template switch: for hairpin-forming sequences, fold-back of the leading strand would allow DNA synthesis from the leading strand, resulting in a repeat expansion upon fork restart (Figure 5A, left) (first proposed by Mirkin, 2006). For non-hairpin-forming sequences, copying off of the lagging nascent strand provides the extra DNA synthesis, as proposed in Shishkin et al. (2009) for large-scale GAA expansions (Figure 5A, right; see Kim & Mirkin, 2013 for review). On the other hand, a single hairpin is more likely to be bypassed, leading to a post-replicative template switch that initiates from a gap, and looks more like SCR (Figure 5B; House et al., 2014b; Kerrest et al., 2009). This latter event may be more common for mid-length CAGs, at or above the expansion threshold of 35 repeats but still less than the size needed to produce a stable fork stall. For very short (CAG)_13–25 tracts, simple post-replicative hairpin unwinding by the Srs2 protein could be sufficient most of the time, with less engagement of the full recombination pathway (Figure 5C; Bhattacharyya & Lahue, 2004). This idea of length-dependent differences is supported by the fact that Tof1, a subunit of the replication-pausing complex, protects against instability for both ATTCT repeats (Cherng et al., 2011) and (CAG)_130–155 repeats, but has a lesser effect at (CAG)₈₅ (L. Gellon, M. Lahiri, C.H. Freudenreich, unpublished data). Thus, the Tof1 complex may act to limit the template-switching events at stalled forks that can allow for repeat expansions.

The role of helicases in resolving hairpins during repair and fork restart

As described above, the Srs2 protein plays a key role in preventing fragility and instability of expanded CAG repeats in yeast. Srs2 was shown to have a robust ability to unwind CTG and CCG hairpins in vitro (Anand et al., 2012; Bhattacharyya & Lahue, 2005). Additionally, Srs2 is needed to facilitate replication past a (CGG)₄₅ repeat that causes a barrier to replication via hairpin formation, determined using direct visualization of fork stalling in vivo by 2D gel electrophoresis (Anand et al., 2012). Srs2 had no activity on replication barriers due to G-quadruplex structures or protein binding, thus it is specific to DNA hairpins. This Srs2 function was unique among the helicases tested (Sgs1, Pif1, and Rrm3), and was dependent on its helicase activity and its ability to interact with PCNA, but not on its Rad51 displacement activity. Altogether, at least three roles for the Srs2 protein at hairpin-forming sequences have been defined: (1) to facilitate fork progression through hairpin DNA structures (Anand et al., 2012) (Figure 6B), (2) to effect hairpin unwinding during PRR, independent of Rad51 displacement (Daee et al., 2007; Dhar & Lahue, 2008) (Figure 6A), and (3) to prevent CAG expansions and contractions by restraining Rad51-dependent recombination (Kerrest et al., 2009) (Figure 5B). In addition, deletion of Srs2 significantly increases chromosome fragility of both CAG and CGG repeats in yeast (Anand et al., 2012; Kerrest et al., 2009), likely because of increased fork stalling due to defective hairpin unwinding (Anand et al., 2012; J. Nguyen, R. Anand, C.H. Freudenreich, unpublished data).

Figure 6.

Helicases resolve hairpins to promote repeat stability. (A) The helicases Srs2, RTEL1, Sgs1, and WRN unwind hairpins at gaps to prevent expansions of repeat DNA. Based on instability profiles in mutants, Sgs1/WRN are hypothesized to act on the template strand to prevent contractions, and Srs2/RTEL1 on the nascent strand to prevent expansions. (B) Helicases promote replication fork progression through hairpin unwinding. Sgs1/WRN and RTEL1 may also unwind G-quadruplex structures. The placement of Srs2 at the advancing fork is based on its ability to bind the sumoylated form of PCNA, found at the replication fork, and to lessen fork stalling at CGG repeats (Anand et al., 2012). (see colour version of this figure at www.informahealthcare.com/bmg).

Recent evidence indicates that the RTEL1 helicase performs at least some of these functions in human cells (Frizzell et al., 2014). RTEL1 knockdown resulted in an increase in CAG expansions to a similar level as knockdown of Rad18 and HLTF, homologs of yeast Rad18 and Rad5, and RTEL1 could also partially substitute for Srs2 in yeast cells to prevent CAG expansions. Fbh1, another potential Srs2 ortholog, did not complement any of these phenotypes in yeast or protect against expansions in human cells. Although RTEL1 and Srs2 lack protein sequence homology and have opposite DNA unwinding polarities, these results indicate a functional conservation between the two enzymes with respect to CAG repeat expansions. Both helicases are able to unwind CAG and CTG hairpin structures in vitro, although RTEL1 additionally acts at G4 DNA and is important in telomere maintenance (Frizzell et al., 2014; Vannier et al., 2012, 2013). There may be other proteins in human cells that share in the ability to unwind hairpins and prevent repeat expansions. One candidate is PARI, which, like Srs2, interacts with sumoylated PCNA and prevents recombination, but since PARI is not an active helicase, it would have to work with another as yet unknown factor to actively unwind DNA structures (Moldovan et al., 2012). Another helicase that has been shown to prevent CAG instability and fragility in yeast is Sgs1 (Kerrest et al., 2009). Sgs1 can unwind hairpins and G4 DNA in vitro, as can its human homologs WRN and BLM (Anand et al., 2012; Bhattacharyya & Lahue, 2005; Huber et al., 2002; Shen & Loeb, 2001; Sun et al., 1998). Sgs1 and WRN interact with Polδ, and thus are well positioned to unwind structures on the lagging strand template. Indeed, deletion of Sgs1 led to a large increase in repeat contractions, but no increase in expansions, consistent with such a role (Bhattacharyya & Lahue, 2004; Kerrest et al., 2009). Interestingly, WRN was identified in a screen for activities in HeLa extracts that were able to stimulate repair of a CTG hairpin on the template strand (Chan et al., 2012). The WRN helicase efficiently unwound CTG hairpins in this system to promote Polδ-catalyzed DNA synthesis across the gap and prevent deletions (Chan et al., 2012). Thus Srs2/RTEL and Sgs1/WRN may work together to resolve hairpins on different strands (Figure 6). This mechanism could also be relevant during gap repair in non-dividing cells (Figure 3).

The role of DSB repair in preventing repeat fragility and instability

Homologous recombination pathways of DSB repair

Recombination is a shared mechanism between gap repair and DSB repair that uses a homologous template, which can be a sister chromatid, homologous chromosome, or ectopic site, to copy and replace missing information (Symington & Gautier, 2011). Two-ended double-strand breaks (DSBs) are first bound by the MRX (yeast) or MRN (human) complex, consisting of Mre11, Rad50, and yeast Xrs2 or human Nbs1. MRN controls processing of the DNA ends by endonucleases Mre11 and Sae2 and exonucleases Exo1 and Dna2, to expose a single-stranded 3′ end and initiate HR (Mimitou & Symington, 2011). MRX/MRN also has a structural role in bridging DNA ends, as well as a role as a sensor and a mediator of the checkpoint response (Williams et al., 2007). In addition, Mre11 binds to stalled forks (Tittel-Elmer et al., 2009), where it could act to tether one-ended breaks to facilitate repair of the broken fork and fork restart.

Early studies in human patients showed no exchange of flanking markers at expanded repeat loci, and cases of conversion to the shorter allele were not documented, leading to the conclusion that meiotic crossovers or gene conversion between homologs were not involved in repeat expansions (reviewed in Monckton & Caskey, 1995). However mitotic recombination between sister chromatids does not lead to a detectable change in flanking sequence (Symington & Gautier, 2011). Furthermore, some human minisatellite loci show germ line repeat expansions that seem to arise from complex gene conversion events involving the homologous chromosome but that rarely are associated with the exchange of flanking markers (Jeffreys et al., 1994). Experiments in E. coli showed that long CAG/CTG repeats (67–175 repeat units) induce recombination, and the authors proposed that this occurs because the polymerase stalls within the repeat tract, which can cause nicks or DSBs that stimulate HR (Hebert & Wells, 2005; Napierala et al., 2002; Pluciennik et al., 2002). Expanded GAA/TTC repeats also stimulate mitotic crossovers in yeast (Tang et al., 2011). In yeast, the effect of HR on CAG repeat stability was assessed directly by induction of a DSB and selection for repair events that used an ectopically provided (CAG)₉₈ tract for repair (Richard et al., 2000). This experiment showed that repeat instability occurred during HR repair, resulting in a much higher percentage of expansions (13%) and contractions (30%), than a control that did not undergo repair (0% expansions and 10% contractions). Mre11 was required for efficient repair, and the proportion of expansions increased in its absence (Richard et al., 2000). Thus, synthesis-dependent HR from a DSB is a mechanism that can produce both repeat expansions and contractions.

Studies from the Freudenreich lab further investigated the role of the various DSB repair proteins and pathways in CAG repeat maintenance. CAG tracts of either 70 or 155 repeats contained on a yeast artificial chromosome (YAC) (Callahan et al., 2003) were evaluated for repeat fragility and instability (Sundararajan et al., 2010). Proteins needed for HR repair of DSBs (Rad51, Rad52, Mre11, Sae2, and Exo1) as well as those required for non-homologous end joining (NHEJ) repair (Mre11 and Dnl4) were important for preventing CAG fragility. Deletion of both Rad52 and Dnl4 (human Lig4) was additive, indicating that both pathways play important but separate roles in healing breaks within repetitive DNA. Indeed, in the absence of these repair pathways, yeast cells containing long CAG tracts showed increased cell death compared with control cells with no repeat, underscoring the importance of these pathways in repair of DSBs that occur at expanded repeat tracts (Sundararajan & Freudenreich, 2011; Sundararajan et al., 2010).

DSB repair proteins were also required to protect against CAG/CTG instability in this yeast model. Consistent with the earlier study, expansions were increased 11-fold in the absence of Mre11, pointing to a key role for this protein in preventing repeat instability (Sundararajan et al., 2010). These expansions were suppressed in the absence of Rad52, and, therefore, occurred through aberrant recombination (Sundararajan et al., 2010). It was proposed that MRX plays an important role in stabilizing one-ended DSBs that occur at stalled forks, promoting correct strand alignment during HR-dependent fork restart (similar to Figure 2C). One-ended DSBs could form by fork breakage, or alternatively fork reversal is predicted to create a double-stranded DNA end to which MRX could bind. Deletion of the Mre11-associated Sae2 and Exo1 nucleases also increased both CAG fragility and expansions, although to a lesser extent than mre11Δ, reinforcing a role for appropriate MRX-mediated processing of ends to promote efficient and correct repair of repeat-associated breaks (Sundararajan et al., 2010). Deletion of the gene encoding the Ctf18 protein, another factor speculated to have a role in fork stabilization, also increases CAG fragility and instability, with the expansions and contractions occurring during a Rad52-dependent process (Gellon et al., 2011). In summary, although HR is needed to repair DSBs within expanded repeats, it can also be a source of repeat instability. This could occur through misalignment (Figure 2C) and/or because of low fidelity of the repair process.

Despite the strong evidence for HR-induced repeat instability in yeast, in a DM1 mouse model, loss of the Rad54 gene product did not significantly suppress CTG instability, and the absence of the Rad52 gene decreased the size of expansions, but did not eliminate them (Savouret et al., 2003). These results have been interpreted to mean that HR has no effect on CAG/CTG instability in mammalian cells. So how can they be reconciled with the data from E. coli and yeast model systems that support a role for HR? First, consistent with the mouse data, deletion of HR proteins in wild-type yeast does not eliminate expansions. The effect of HR seems to be most important in repairing breaks, and it comes into play primarily when breaks are increased, such as in the mre11 or ctf18 backgrounds, at long repeats that break more frequently, or when gap repair by HR is compromised (such as during template switching in the absence of H4-K16ac). Furthermore, the knockouts used in the DM1 mouse studies, Rad52 and Rad54, do not eliminate HR in mammalian cells. Rad52 knockouts exhibit only a slightly reduced HR frequency and are not hyper-sensitive to DSB-inducing agents; Rad54 is not an essential HR protein, but serves to facilitate chromatin remodeling during HR (Suwaki et al., 2011; van Veelen et al., 2005). Despite this, the mean size of expansions was significantly decreased in Rad52^−/− mice, supporting the idea that rare large expansions could occur during HR in these mice. In summary, data in multiple systems are in agreement that HR is not the major expansion mechanism operating in wild-type cells. Nonetheless, it is important in repairing breaks at long TNRs, and can produce expansions in some situations. Due to the technical difficulties of eliminating HR in mammalian cells, the role of HR in repeat instability in mammals remains to be fully assessed.

Recent results show that replication in the context of HR repair or HR-dependent fork restart proceeds with less fidelity and more mutations than normal replication, even without the complication of copying DNA repeats (Deem et al., 2011; Hicks et al., 2010; Iraqui et al., 2012). Recently, GAA repeats have been shown to induce mutagenesis up to 8 kb from the repeat site in yeast, which is enhanced in strains with polymerase defects and largely dependent on the Polζ TLS polymerase (Saini et al., 2013; Shah et al., 2012; Tang et al., 2013). The evidence indicates that DSBs are formed at the repeat, followed by strand invasion of the sister chromatid, and mutagenic synthesis by Polζ during HR repair (Saini et al., 2013). This repeat-induced mutagenesis (RIM) has also been observed for H-DNA- and Z-DNA-forming sequences introduced into mammalian cells (Wang & Vasquez, 2004, 2009; Wang et al., 2006). Thus, the cis-effects of structure-forming DNA extend even beyond the repeat itself, influencing genome mutation rates over a large region.

Of note, the absence of HR may also create repeat instability (Sundararajan et al., 2010). In this case, since HR is defective, the tract length changes must occur through another pathway, such as end-joining or single-strand annealing (SSA).

The role of single-strand annealing (SSA) and end-joining mechanisms (NHEJ and MMEJ) in repair at repeats

NHEJ does not use a homologous sequence as a template for repair, and instead directly ligates the broken ends together by the Dnl4/Lif1 ligase complex in yeast, Lig4/XRCC4 in mammals. End joining is an error-prone process, which can result in insertions or deletions of nucleotides or translocations (Chapman et al., 2012; Lieber, 2010). Although an error-prone pathway, NHEJ is favored over HR in G₁ or non-replicating cells, as copying from the homologous chromosome could result in loss of heterozygosity (Shrivastav et al., 2008). NHEJ does not require 5′ end resection, and in fact, occurs when end resection is blocked by binding of the Ku70–Ku80 (Ku) heterodimer to the broken ends (Lieber, 2010). If resection does occur to expose homology between the 3′ ends, those ends can anneal in a pathway called SSA, resulting in deletion of the resected sequence (Figure 2D). SSA is Rad51 independent and can be Rad52 independent in some circumstances (Ira & Haber, 2002; Mott & Symington, 2011; Prado & Aguilera, 1995).

Just as HR is required for error-free repair of CAG repeats in yeast, so too is intact NHEJ. In the absence of Dnl4 (Lig4), CAG repeat contraction frequency is significantly increased and remains so in dnl4Δrad52Δ cells, indicating that these contractions were not occurring through HR (Sundararajan et al., 2010). The absence of HR proteins Rad52, Rad51, and Rad54 also increased repeat contractions ~2.5-fold (Sundararajan et al., 2010). Because broken DNA ends are less protected by Ku in both rad52 and dnl4 backgrounds, end processing followed by SSA between repeats is an attractive mechanism to explain these contractions (Figure 2D, right); another possibility is MMEJ (see below). Indeed, in a yeast system where breaks are induced within a CAG repeat, contraction via SSA is a prominent outcome (Richard et al., 1999), and induction of a DSB by a CAG/CTG-specific TALEN induced 100% contractions in a highly specific manner (Richard et al., 2014). The same mechanism appears to be operating in human cells, as a CAG-targeted DSB by a zinc finger nuclease also induced frequent contractions (Mittelman et al., 2009). This makes nuclease-directed cleavage of TNRs an attractive method for inducing contractions, which could potentially be used therapeutically.

Repeat expansions are also increased in the absence of HR proteins in some systems (Sundararajan et al., 2010) although not others (Miret et al., 1998; Savouret et al., 2003; Schweitzer & Livingston, 1999). In the yeast study by Sundararajan et al. (2010), more than half of the expansions in a rad52Δ strain were eliminated in the rad52Δdnl4Δ background, indicating that an end-joining pathway may contribute to expansions in this system (Figure 2D, left). The observation of large expansions created during repair of DSBs by HR or end-joining repair is intriguing, as large expansions that occur during maternal transmission of the DM1 CTG and FXS CGG repeat appear to happen during oogenesis. A prominent stage of oogenesis is meiosis, where many breaks occur and are repaired. In yeast, breaks occur frequently at CAG repeats during meiosis and are repaired to give both expansions and contractions (Jankowski et al., 2000; Schweitzer et al., 2001).

A second pathway of end joining is microhomology-mediated end-joining (MMEJ), which uses short stretches of homology (5–25 bp) to align the broken DNA ends and promote ligation to resolve the lesion (McVey & Lee, 2008). This process is error-prone and favors contraction events, but can also result in templated insertions during Polβ or Poly-dependent fill-in (Chan et al., 2010; McVey & Lee, 2008). Using human proteins in a reconstituted in vitro system, it was revealed that Polβ-mediated MMEJ promotes CAG expansions, and the frequency of these events is limited by the 9-1-1 (Rad9/Hus1/Rad1) DNA damage checkpoint complex (Crespan et al., 2012; Figure 2D, left).

Role of the DNA damage checkpoint in sensing damage at TNRs and regulating repair

Since expanded repeats can form unusual structures with regions of single stranded DNA, gaps, stalled forks, and DSBs, it is reasonable to think that they may activate the DNA damage checkpoint. The DNA damage checkpoint is the cellular response to damaged DNA. It consists of damage sensors, mediators, and transducers whose signal is mediated primarily through phosphorylation signaling cascades (Errico & Costanzo, 2012; Gobbini et al., 2013). One of the key inducers of the checkpoint is extensive single-stranded DNA. ssDNA can occur at large gaps, stalled forks (especially if the replisome is uncoupled from the replicative helicase), or DNA breaks that have been processed to expose ssDNA ends. Briefly, the RPA ssDNA-binding complex recruits the mammalian ATRIP/ATR complex (S. cerevisiae (sc) Ddc2/Mec1). ATR/Mec1 is a kinase that phosphorylates downstream effector kinases (Chk1 and scRad53 (scChk1 playing a lesser role)), and is thus a key transducer of the DNA damage checkpoint response. ssDNA gaps as small as 35 bp can weakly induce an ATR-dependent DNA damage checkpoint response; ssDNA greater than 200 nt induces a more robust response (MacDougall et al., 2007). In addition, ssDNA–dsDNA junctions bind the alternative PCNA clamp 9-1-1 (scRad17, Mec3, and Ddc1), loaded by clamp loader complex Rad17–RFC (scRad24–RFC), which contributes to ATR/Mec1 activation. At stalled forks, the mediator Claspin/scMrc1 facilitates ATR activation of the effector kinases. A second branch of the DNA damage checkpoint response flows through the ATM kinase and is responsive to DSBs. Unprocessed DSB ends activate the checkpoint primarily through MRN/MRX binding, activating the ATM/scTel1 kinase, which phosphorylates Chk2/scRad53, the main effector kinases of the DSB response cascade. MDC1 and 53BP1 are primary mediators of the DSB response in mammalian cells; Rad9 assumes this role in S. cerevisiae. The identity of checkpoint proteins important for maintaining repeats can provide insight into the prominent type of DNA damage they cause in a particular cell type or situation.

The role of the DNA damage checkpoint in cells with expanded CAG repeats was first investigated in yeast (Lahiri et al., 2004). In this study, it was found that CAG repeats of 85 and 135–155 units were highly sensitive to deletion of the genes encoding either the Mec1 and Ddc2 proteins (ATR/ATRIP) or the Rad9 protein; fragility and instability were increased in their absence as well as in cells with defective Rad53 signaling (Freudenreich & Lahiri, 2004; Lahiri et al., 2004). These data suggest that both stalled forks and DSBs exist in cells with expanded CAG tracts, and the checkpoint response is important in facilitating repair of these lesions. Of interest, (CAG)₈₅ tracts, but not the longer repeats, were also sensitive to the absence of a functional 9-1-1 complex (rad17Δrad24Δ mutant), supporting the model depicted in Figure 5 that there are less stalled forks and more gapped structures for this CAG repeat length. The most prominent instability phenotype in all these checkpoint mutants was an increase in contractions, ranging from 2 to 6-fold over wild-type cells, consistent with the increase in DSBs occurring. Related, shorter (CAG)₂₀ repeats that do not increase fragility did not show increased contractions in checkpoint mutants (Razidlo & Lahue, 2008). For (CAG)₈₅, expansions were modestly increased in some backgrounds, ~3-fold for rad53, rad17, and rad24 mutants. Strengthening this result, in a more sensitive system where expansions could be selected for, a 3- to 6-fold increase was observed at a (CAG)₁₃ tract in checkpoint mutants (Razidlo & Lahue, 2008). It is interesting that a repeat this short would induce enough damage (presumably gaps) to require checkpoint proteins for appropriate repair. A recent study showed that the 9-1-1 complex might reduce CAG expansions more directly by inhibiting Polβ expansion-prone synthesis during MMEJ (Crespan et al., 2012).

In a FXS mouse model, the loss of one copy of the Atr gene was associated with an increased risk of expansion on maternal transmission (Entezam & Usdin, 2008). The paternal expansion frequency was not significantly affected in Atr^+/− mice, perhaps because the expansion frequency was much higher to start with in males or because of sex differences in the levels of Atr in heterozygous animals. However, loss of one copy of the Atr gene resulted in the transmitted alleles being significantly larger in the offspring of both Atr^+/− males and females relative to the offspring of wild-type animals. This would be consistent with the idea that ATR protects the genome against repeat expansion in animals of both sexes. Loss of one Atm allele also increased the expansion frequency in both males and females (Entezam & Usdin, 2009). Thus, both yeast and mouse models show that the ATR and ATM checkpoint proteins are important for preventing repeat instability, although the ratio of expansions to contractions was somewhat different in the various systems. Consistently, inhibition of these kinases leads to an increase in the incidence of chromosome fragility in both yeast and human cells (Freudenreich & Lahiri, 2004; Kumari & Usdin, 2009; Lahiri et al., 2004).

The Mrc1/Tof1/Csm3 (human Claspin/Timeless/Tipin) complex travels with the replisome and has an important role in coupling Polε to the MCM replicative helicase in both yeast and mammalian cells (Katou et al., 2003; Lou et al., 2008; Tourriere et al., 2005; Zegerman & Diffley, 2009). In addition, the Mrc1 protein has multiple amino acid motifs (SQ or TQ residues) that are phosphorylated in an ATM/Mec1-dependent manner, and facilitate CHK2/Rad53 activation in response to stalled replication forks (Errico & Costanzo, 2012; Naylor et al., 2009; Voineagu et al., 2009a). By studying an Mrc1 mutant specifically deficient in checkpoint function (Mrc1AQ), it was determined that that Mrc1 checkpoint function plays a role in preventing (CAG)₁₃ repeat expansions in yeast, consistent with an earlier study using a mrc1-1 allele and a (CAG)₈₅ tract (Freudenreich & Lahiri, 2004; Razidlo & Lahue, 2008). However, (CAG)₂₀ contractions are increased in mrc1 cells independent of Mrc1 checkpoint function (Razidlo & Lahue, 2008). In human DM1 and Hela cells, knockdown of Claspin, Timeless, or Tipin all significantly increased expansions and contractions of an expanded CAG tract, either (CAG)₁₀₀ at an ectopic locus or (CTG)_400–1000 at the DM1 locus, confirming the importance of this complex in patient-derived cells (Liu & Leffak, 2012). The Mrc1/Claspin complex also has an important role independent of its checkpoint function in stabilization of forks stalled at CGG repeats. In yeast cells lacking Mrc1 and Tof1, stalling at an expanded (CGG)₄₀ repeat was increased, however, cells with the checkpoint deficient but replication proficient Mrc1AQ allele replicated the CGG tract normally (Voineagu et al., 2009a). The stall caused by CGG repeats may escape the replication checkpoint due to stable hairpin structures preventing exposure of ssDNA, resulting in fragility even in wild-type cells (Voineagu et al., 2009b). In summary, all results to date consistently support that the DNA damage checkpoint response is important for preventing triplet repeat expansions for all lengths, and additionally some checkpoint proteins have an additional role in preventing stalling, fragility, and contractions of longer repeats that interfere with fork progression.

Recently, more direct evidence has accumulated that the checkpoint response to DNA damage induced by expanded repeats may be a significant burden on cells, reducing their viability and lifespan. In yeast cells, (CAG)_70–155 repeat tracts can induce a length-dependent DNA checkpoint response that is increased in the absence of DNA repair pathways (Sundararajan & Freudenreich, 2011). In the absence of HR (rad52Δ), CAG-dependent Rad53 phosphorylation is persistent, indicating that intact DSB repair is required for recovery from the checkpoint response. The fact that the checkpoint response was increased when DNA repair was deficient implies that it occurred as a response to DNA damage (rather than RNA or protein toxicity). Additionally, CAG tract-induced Rad53 phosphorylation is abolished in an mre11 mutant, implicating this complex in sensing damage at the repeat to initiate a checkpoint response (Sundararajan & Freudenreich, 2011). Human neuronal cells are sensitive to DNA damage and induce apoptosis in response, so it would be important to test whether expanded repeat tracts impact cell health in humans. This is complicated by the fact that the RNA transcribed from the repeat can also be toxic, as can the translated protein, especially for the polyglutamine diseases. It has been shown that bidirectional transcription through an expanded CAG repeat in human cells induces a cell stress response, cell cycle arrest, and eventually massive cell death via apoptosis (Lin et al., 2010a,b). This response is signaled through RPA, ATR, CHK1, and p53, indicating that it is due to single-stranded DNA of some nature that occurs with convergent transcription. The aberrant structure that induces the response is speculated to be a ‘‘double bubble’’, with stalled RNAPII complexes on both strands; indeed the toxicity is worse in TCR mutants (Lin & Wilson, 2012). Together, these results point to a potential mechanism of cell death in the terminally differentiated cells that are affected in repeat diseases, and suggest this is an area that deserves future study.

Conclusions

As evident from this review, work in various model organisms and in vitro model systems has demonstrated that there are many ways in which repeats can expand and contract. Which pathway is used seems to depend on a variety of cis- and trans-acting factors, including the transcriptional activity of the locus containing the repeat, the chromatin structure of the region, the level of oxidative stress, and the level of expression of repair proteins in that particular cell type. However, many questions remain as to which of these pathways occur at the disease locus in humans and which of them are disease relevant. Because of the importance of locus-specific factors, tissue type, and developmental window in the choice between pathways, none of the model systems, although valuable, can be assumed to recapitulate completely the situation in patients. Nonetheless, knowledge of the pathways and proteins that can contribute to repeat instability in these systems can provide the basis for therapeutic ideas. In addition, through study of repeat expansion and contraction mechanisms, scientists have learned a great deal about DNA replication and DNA repair pathways and how they operate in the context of structure-forming sequences and in different cells and tissues. Insights into protein functions outside of their canonical pathways have also been made, and contributed to our understanding of DNA repair. The field has contributed many novel concepts, such as the possibility of structure-induced gratuitous repair, the importance of 5′ flap processing for preventing insertions in genomes, methods of fork progression through structural barriers and helicase unwinding preferences, the toxicity of bidirectional transcription, and insights into how chromatin structure and modifications prevent genome instability, to name a few. More surprises surely remain to be discovered, providing new insights into repeat expansion diseases and the sophisticated yet imperfect mechanisms of genome maintenance.