The central role of DNA damage and repair in CAG repeat diseases

DNA damage

It has been estimated that each mammalian cell sustains as many as 10,000 single-strand and 10-50 double-strand DNA breaks per day (Madabhushi et al., 2014). Exogenous sources of DNA damage, such as UV light, ionising radiation and chemical mutagens, predominantly affect exposed and dividing cells, but in neurons, and especially in neurons of the CNS, endogenous metabolic processes are more relevant sources of DNA damage. The high oxygen demands of the brain (20% of total body oxygen consumption, but only 2% of body mass) expose its cells to numerous reactive oxygen species (ROS; see Glossary, Box 1) produced by normal mitochondrial respiration (Cooke et al., 2003), and further oxidative damage can arise as a result of inflammation. Over 100 different types of DNA base damage have been identified as being caused by ROS, the most abundant of which is 8-oxo-2′-deoxyguanosine (8-oxo-dG). One study of human lymphocytes found 8-oxo-dG at a steady state of ∼10,000 damaged bases per cell nucleus (Ohno et al., 2006). This altered base can be premutagenic in replicating cells and has inhibitory effects on transcription in postmitotic neurons (Iyama and Wilson, 2013). In addition to ROS, cellular metabolism generates endogenous alkylating compounds (e.g. S-adenosyl methionine), lipid peroxidation products, and reactive nitrogen and carbonyl species that can directly damage DNA. DNA is also susceptible to hydrolysis, which can directly cause base loss or deamination, particularly in single-stranded regions. CAG repeats might be especially susceptible to damage as hydrolytic depurination (i.e. loss of A or G) and cytosine deamination are frequent events (Lindahl, 1993).

In addition to pathological DNA damage, there is increasing recognition of the role of physiological, ‘programmed’ DNA strand breakage. For example, outside the CNS, the generation of antibody and T cell receptor diversity depend on programmed double-strand breaks, and their subsequent repair, as part of V(D)J recombination (see Glossary, Box 1) (Slean et al., 2008), and meiotic crossing over is initiated by SPO11-induced double-strand breaks (see Glossary, Box 1) (Keeney et al., 2014). Topoisomerases, which can introduce temporary single- or double-strand breaks in DNA to regulate supercoiling (see Glossary, Box 1), are essential for the expression of long genes and are particularly important in the brain, where expressed genes are longer than elsewhere in the body (Gabel et al., 2015; King et al., 2013; Zylka et al., 2015). Recently, it has also been shown that neuronal activity in cells and in animals can trigger topoisomerase-induced double-strand DNA breaks in the promoters of neuronal early-response genes (Ju et al., 2006; Madabhushi et al., 2015; Suberbielle et al., 2013). These DNA breaks might relax topological constraints and stimulate promoter activity. Topoisomerase mutations can lead to various neurodevelopmental and neurodegenerative conditions, highlighting the importance of torsional regulation of DNA in brain function (King et al., 2013; McKinnon, 2016).

Therefore, neuronal DNA strands are continually broken and repaired in vivo. The tight regulation of DNA repair is crucial for maintaining neuronal gene expression and function; indeed, many Mendelian neurological diseases result directly from defects in the DNA damage response (Madabhushi et al., 2014).

DNA repair

Unrepaired DNA damage can have profound consequences on cells. For example, point mutations or chromosomal rearrangements can lead to cancer in dividing cells, such as glia, and can induce cell death in nondividing cells, such as CNS neurons. Moreover, lesions or strand breaks can stall DNA and RNA polymerases, leading to impaired replication or transcription, respectively, and potentially triggering cell death or senescence (Hoeijmakers, 2009). Therefore, an elaborate damage response has evolved to identify and repair DNA damage in conjunction with cell cycle regulation. The DDR involves >450 genes in humans, with subsets of these genes deployed depending on the type of DNA damage, type of cell, stage of cell cycle and stage of organism development (Pearl et al., 2015).

Human CNS development involves the proliferation and subsequent migration and differentiation of neural progenitor cells, beginning a few weeks after conception and continuing to ∼6 months after birth (Rulten and Caldecott, 2013). Most neurons then enter a postmitotic phase (and are required to survive for a lifetime), although there is clearly a high turnover of molecules within these cells. A small subset of neurons in the human hippocampus and lateral ventricle of the mature CNS can divide and contribute to ongoing neurogenesis (McKinnon, 2013). The phase of cell cycle determines which DNA repair pathways are utilised in neurons. For example, dividing progenitors in S/G2 phase use accurate homologous recombination (HR) for double-strand break repair (see Glossary, Box 1) and replication fork maintenance (Fig. 1). Mutations in these repair systems are embryonic lethal or lead to profound neurodevelopmental defects (McKinnon, 2013; Rulten and Caldecott, 2013). By contrast, mature postmitotic neurons in G0/G1 rely upon less accurate nonhomologous end joining (NHEJ) for double-strand break repair in the absence of a sister chromatid (Fig. 1). In addition, DNA damage in these neurons is predominantly single stranded, which can lead to the inhibition of transcription, as well as compromised genomic integrity. Pathways such as single-strand break repair (SSBR; see Glossary, Box 1) and transcription-coupled nucleotide excision repair (TC-NER; see Glossary, Box 1) are essential to cellular function, and mutations in these pathways lead to late neurodevelopmental defects, such as microcephaly, and, more commonly, neurodegeneration (McKinnon, 2013; Rulten and Caldecott, 2013). The type and location of affected neurons determine the clinical phenotypes observed (Madabhushi et al., 2014).

Many parts of the DDR are highly conserved from prokaryotes to eukaryotes, with extra layers of regulation and redundancy found in higher organisms. Individual linear repair pathways, such as base excision repair (BER; see Glossary, Box 1), nucleotide excision repair (NER; see Glossary, Box 1) and mismatch repair (MMR; see Glossary, Box 1), involve a series of analogous steps: lesion recognition, repair factor recruitment, lesion excision leading to DNA strand breakage, processing of DNA ends and DNA synthesis, to complete repair (Fig. 2). In addition, it is increasingly recognised that there is significant redundancy between pathways, presumably arising through divergent evolution in order to repair a wide range of lesions whilst maintaining genomic integrity (O'Brien, 2006; Pearl et al., 2015). Pathways can also be involved in noncanonical repairs. For example, MMR enzymes, which usually act at replication forks in dividing cells to correct DNA polymerase errors, have been shown to recognise and repair mispaired bases in nondividing yeast cells (Rodriguez et al., 2012). Therefore, DNA repair systems have activities that depend not only on the type of DNA damage but also on the cell/tissue context, the background genetics of the cell, and the prevailing environmental conditions. CAG repeats in the DNA of adult postmitotic neurons constitute a particular substrate for repair systems, and the regulation of these repair activities is entwined with CAG repeat disease pathogenesis, as discussed below.

Intrinsic instability of CAG repeat number

Biophysical studies in vitro have shown that disease-causing CAG repeats can form unusual DNA structures, including stable hairpins. The stability of these structures correlates with the propensity for CAG repeat expansion (Gacy et al., 1995). CAG:GAC base pairing in the stem of a hairpin contains a middle A:A base pair mismatch, which in silico modelling predicts will adopt an unusual Z DNA structure, perhaps via the flipping out of bases (Khan et al., 2015). This could predispose hairpin structures to both increased DNA damage and MMR protein binding. Indeed, bases damaged by ROS (e.g. 8-oxo-dG) can affect the formation and stability of hairpins, as well as having consequences on DNA repair fidelity (Volle et al., 2012). MSH2, part of the MutS mismatch recognition complexes (see Glossary, Box 1), binds directly to slipped-strand DNA structures formed by (CAG)_n in vitro (Pearson et al., 1997). The processing of these artificial DNA substrates by human cell extracts requires various repair factors, principally MMR proteins (including MSH2, MSH3, PMS2), although the repair outcomes depend on the starting DNA structure and not just its sequence (Panigrahi et al., 2005, 2010). It is hard to draw physiological mechanistic conclusions from these cell-free systems, but putative pathways can be identified; for example, MutSβ (MSH2/MSH3 complex) is required for repeat expansion in some assays (Nakatani et al., 2015; Stevens et al., 2013), and long-patch BER deficiency has been implicated elsewhere (Goula et al., 2009).

Further insight into CAG repeat stability has come from bacteria and yeast. These microorganisms are attractive models as they are genetically tractable, most of their DNA repair factors have been identified, and high-throughput screening assays exist for them (Bichara et al., 2006; Dixon et al., 2004; Kim et al., 2016). Repeats are unstable in these microorganisms, and seem to have a similar length threshold to that observed in human diseases, although they exhibit a propensity for repeat contractions over expansions. There is also a requirement for MMR in (CAG)_n instability: when MMR factors are knocked out, repeats are stabilised (Jaworski et al., 1995; Williams and Surtees, 2015). Given the phylogenetic conservation of MMR and other DNA repair factors from bacteria to humans (Fishel, 2015), these results might be relevant to human disease. However, bacterial and yeast cells divide in culture, and have much simpler DNA-damage response systems than human cells. They also contain eukaryotic triplet repeats out of genomic context, either on plasmids or integrated into DNA that lacks human chromatin structure and organisation. More disease-relevant data have come from multicellular organisms, as detailed below.

Germline and somatic instability of CAG repeats

Heritable, stochastic CAG repeat expansions in germline cells (i.e. sperm or egg) are reported in many CAG and non-CAG repeat disorders (Budworth and McMurray, 2013; Durr, 2010). Given that repeat length correlates inversely with age at disease onset, this intergenerational propensity for CAG expansion explains the phenotypic observation of anticipation, whereby disease onset tends to get earlier over generations. This has been shown in all CAG repeat disorders, and is usually more marked through the paternal line (Durr, 2010), although other non-CAG trinucleotide repeat disorders, such as Huntington's disease-like 2 (HDL2) and myotonic dystrophy, show increased anticipation through the maternal line (Gövert and Schneider, 2013). Repeat length analysis has shown that significant CAG mosaicism is present in the sperm of male HD patients and that this correlates with repeat expansion on transmission (Telenius et al., 1995). Evidence from a transgenic mouse model of HD (R6/1) suggests that repeat expansion in sperm occurs after meiosis, as haploid spermatids are maturing into spermatozoa (Kovtun and McMurray, 2001). This implicates DNA repair rather than DNA replication in the DNA synthesis needed for repeat expansion.

Variations in CAG repeat length have also been found in somatic (i.e. nongermline) cells. Repeat length stability varies across cell types and can be associated with phenotype. For example, in transgenic and knock-in mouse models of HD, CAG repeat length tends to be increased in cells from the striatum, cortex and liver, but stable in cells from the cerebellum, blood and tail. Maximum expansion is observed in the striatum, which correlates well with the degeneration of striatal MSNs that underpins the disease (Gonitel et al., 2008; Lee et al., 2011; Møllersen et al., 2010). Large CAG repeat expansions have also been demonstrated in postmortem human brain neurons from HD patients, and increased expansion of repeats correlates with younger age at onset of symptoms (Kennedy, 2003; Shelbourne et al., 2007; Swami et al., 2009). DNA repair is implicated in affected neurons, as these are postmitotic. However, similar tissue-specific patterns of repeat expansion are seen in SCAs and, in these diseases, degeneration is observed in the cerebellar Purkinje neurons rather than the striatal MSNs (Chong et al., 1995; Hashida et al., 1997; Tanaka et al., 1996). The relatively high levels of expression of certain DNA repair factors in the cerebellum might prevent significant repeat expansion in this tissue. Genomic context is also important. For example, SCA7 CAG repeats in a transgenic mouse model were stable when present in complementary DNA (cDNA), but unstable if contained in a human genomic fragment. In addition, repeat stability did not correlate with neurodegeneration (Libby et al., 2003). Therefore, although somatic CAG repeat expansion can occur, and might correlate with neurodegeneration in HD, it remains unclear whether expansion drives pathology in individuals or results from downstream DNA repair defects in affected cells (Chong et al., 1995).

Modifiers of CAG repeat stability in the genome

Various cis and trans factors can affect the stability of CAG repeats in the genome. Cis factors include the length of the CAG repeat, the presence of repeat interruptions, haplotype and genomic context. Longer repeats, particularly those that are over the disease-causing threshold, are more unstable and tend to expand (Fig. 3), although repeat interruptions can temper these effects (Usdin et al., 2015). Interruptions have been identified and associated with increased repeat stability, and later age at disease onset, in various SCAs and in HD (Table 1), as well as in other trinucleotide repeat disorders such as fragile X syndrome, Friedreich's ataxia and myotonic dystrophy (Chong et al., 1997; Chung et al., 1993; Eichler et al., 1994; Gao et al., 2008; Yu et al., 2011). Within the CAG repeat diseases, interruptions of a pure CAG repeat almost always involve alteration of the third base in the codon (the most tolerant of change). Interruptions were first observed in SCA1 with variable numbers of CAT codons replacing CAG (Chung et al., 1993). These reduce the stability of hairpin structures formed by pure (CAG)_n repeats in DNA, inhibiting repeat expansion, as well as inserting a number of histidines into the polyglutamine stretch of protein (Menon et al., 2013; Pearson et al., 1998; Sobczak and Krzyzosiak, 2004). More commonly, CAG repeat tracts can be interrupted by CAA codons, which alter the DNA/RNA sequence but do not affect the translated polyglutamine. In SCA2, pure CAG repeats of intermediate length (∼27-35 repeats) can expand into the disease range but repeats interrupted by CAA expand to a lesser extent and are associated with different neurological phenotypes (amyotrophic lateral sclerosis or parkinsonism, depending on the repeat number) (Charles et al., 2007; Elden et al., 2010; Yu et al., 2011). In SCA17, complex CAG repeat alleles of the TATA-box binding protein gene (TBP) that contain various CAA interruptions are associated with later disease onset (Gao et al., 2008). In HD, the penultimate codon of the CAG tract is usually CAA, followed by one more CAG before the (CCG)_n repeat encoding polyproline begins. The relevance of the penultimate interrupting CAA is uncertain, but its absence correlates with earlier onset (and with repeat expansion) in a few cases (Goldberg et al., 1995). Conversely, in a bacterial artificial chromosome mouse model of HD, a huntingtin (HTT) allele with 97 codons of alternating CAA/CAG is stable over 12 months in both germline and somatic cells, but the mice still develop a neurodegenerative pathology (Gray et al., 2008).

It appears that if a CAG repeat is above a threshold length and of appropriate codon structure and context, it is licensed to become unstable. Trans factors, such as DNA repair proteins, can then act on this DNA substrate to increase or decrease the repeat number in specific tissues. Work on trans factors has mostly been carried out in mouse models of HD. Transgenic and knock-in mouse models of HD, all with long tandem CAG tracts of >100 repeats, develop progressive neurological impairment, leading to reduced abilities in tests of motor, coordination and cognitive function (Brooks et al., 2012; Ferrante, 2009). The development of somatic CAG repeat expansion in the striatum of mouse HD models (for example, R6/1 transgenic mice carrying exon 1 of the human HTT gene) (Mangiarini et al., 1997) correlates with symptom development. Crossing mouse models of HD with mice that carry different DNA repair gene mutations has shown that deficiencies in specific MMR genes (e.g. Msh2, Msh3, Mlh1 or Mlh3), or BER genes (e.g. Ogg1 or Neil1) can abrogate somatic and/or germline CAG repeat expansion and, in some cases, ameliorate HD-like phenotypes (Budworth et al., 2015; Kovalenko et al., 2012; Pinto et al., 2013; Usdin et al., 2015; Wheeler et al., 2003). These effects seem gene specific, as knockouts of other DNA repair factors in the same pathways [e.g. of Msh6 (MMR) or Mpg glycosylase (BER)] have no effect on repeat stability. In addition, results from different diseases and models have been inconsistent. For example, NER factors have been implicated in CAG repeat diseases. A study of 137 SCA3 parent-child repeat transmissions identified variants in NER factors Cockayne syndrome protein B (CSB; see Glossary, Box 1; Fig. 2C), RPA proteins and CDK7, as being associated with intergenerational repeat expansions (Martins et al., 2014). However, knockouts of different NER factors in different CAG repeat disease models have variable effects: Csb (Ercc6) knockout in HD mice promotes germline repeat instability (Kovtun et al., 2011); XPG (mus201) knockout in a Drosophila model of SCA3 abolishes repeat instability (Jung and Bonini, 2007); Xpa knockout in a SCA1 mouse reduces somatic repeat instability in many areas of the brain (although not in the cerebellum) (Hubert et al., 2011). The reasons for these differences are not fully understood, although the human homologues might not contain relevant variation, and some repair factors have additional functions outside NER; for example, CSB has chromatin remodelling and transcriptional regulation activities.

Downstream deficits in DNA repair in CAG repeat diseases

The proteins encoded by CAG repeat-containing genes have a wide range of functions in different cellular processes (Table 2). However, most are ubiquitously expressed and have links to transcriptional regulation, tying in with earlier work that identified polyglutamine stretches of 10-30 amino acids as being potent transactivators (Gerber et al., 1994). The recent realisation that repeat length polymorphism in DNA can modulate gene expression suggests that expanded repeats might affect transcription through both DNA- and protein-mediated mechanisms (Gymrek et al., 2015). The target genes that are differentially expressed as the result of repeat instability are not known. Fibroblasts, lymphoblasts or lymphocytes cultured from patients with HD or other neurodegenerative conditions accrue more DNA damage than wild-type controls; this could be caused by defective DNA repair (Moshell et al., 1980; Robison and Bradley, 1984; Scudiero et al., 1981). In support of this hypothesis, various polyglutamine-containing proteins, including HTT, androgen receptor (AR), ataxin (ATXN) 1, ATXN2 and ATXN3, have been shown to have roles in DNA repair (Table 2), and there is evidence that at least some of the pathogenesis of CAG repeat expansion might arise from the loss of wild-type protein function (Ashkenazi et al., 2017; Chatterjee et al., 2015; Gao et al., 2015; Saudou and Humbert, 2016; Zeitlin et al., 2000). As an example, regulated phosphorylation of HTT at various sites is involved in the DNA damage response. The N-terminus of HTT (particularly methionine 8) can act as a direct sensor of oxidative stress, leading to its phosphorylation at serines 13 and 16 and translocation from its usual cytoplasmic location to the nucleus (DiGiovanni et al., 2016). In the nucleus, HTT is recruited to sites of DNA damage, in a process dependent on ATM serine/threonine kinase (ATM), and might function as a scaffold for DNA repair complexes (Fig. 4A). Analysis of patient fibroblasts shows that an expanded polyglutamine in HTT does not prevent its recruitment to sites of DNA damage, but is associated with increased DNA damage, consistent with a dominant-negative effect on repair (Fig. 4B) (Maiuri et al., 2016). In addition, wild-type HTT is phosphorylated by cyclin-dependent kinase (CDK) 5 on serines 1181 and 1201 in response to DNA damage, and this seems to have a protective role in inhibition of p53 (TP53)-induced cell death. Loss of this protective phosphorylation in ageing and/or disease-associated neurons could lead to increased cell death, although the exact molecular mechanisms involved are not understood (Fig. 4B) (Anne et al., 2007).

Recent work on ATXN3 has reinforced the importance of subcellular localisation and protein-protein interactions in the regulation of DNA repair. ATXN3 is a deubiquitinase that has a neuroprotective role mediated by its regulation of autophagy (via wild-type polyglutamine; see Glossary, Box 1) (Ashkenazi et al., 2017), the ubiquitin-proteasome system, and DNA repair. Heat-shock or oxidative stress can stimulate the nuclear translocation of ATXN3, where it has roles in transcriptional regulation, possibly through its deubiquitinase activity, and DNA repair (Fig. 4A) (Orr, 2012). The latter was first suggested by the interaction of ATXN3 with human RAD23 homologues in a yeast two-hybrid screen, as RAD23 proteins are involved in NER (Fig. 2C) (Wang et al., 2000). More recently, ATXN3 has been shown to protect cells from DNA damage through its interaction with polynucleotide kinase 3′-phosphatase (PNKP), a key enzyme involved in the processing and repair of DNA strand breaks (Figs 1 and 2) (Chatterjee et al., 2015). Rare mutations in PNKP can cause ataxic syndromes (Bras et al., 2015). Interestingly, the expansion of a polyglutamine tract in ATXN3 leads to both the sequestration of PNKP outside the nucleus and the inhibition of nuclear PNKP activity, which together lead to impaired DNA repair. Persistent DNA damage (and DNA damage signalling through ATM) is observed in mouse and cellular models of SCA3, as well as in postmortem human SCA3 brains, and this can trigger cell death through p53-mediated and other pathways (Fig. 4C) (Chatterjee et al., 2015; Gao et al., 2015).

A similar theme is continued in cell models of SBMA and SCA1. Expanded polyglutamine within AR or ATXN1, respectively, lead to the sequestration of proteins involved in DNA repair and the subsequent accumulation of DNA damage (Fujita et al., 2013; Xiao et al., 2012). As well as being intrinsically deleterious to cells, unrepaired DNA damage also leads to persistent activation of the DDR, which in itself can be toxic, as seen in other neurodegenerative disorders (Hoch et al., 2016). The importance of accurate and timely DNA repair in the cerebellum has long been noted, given that mutations in enzymes such as aprataxin and tyrosyl DNA phosphodiesterase 1 (TDP1), responsible for processing damaged DNA ends to permit repair (Fig. 1), lead specifically to cerebellar neurodegeneration (Ward and La Spada, 2015).

Therefore, there is increasing evidence in a range of CAG repeat diseases that defective DNA repair might be involved in disease pathogenesis. An expanded polyglutamine tract could lead to the inactivation and/or inappropriate sequestration of repair proteins, such that DNA damage builds up in neurons. If the wild-type proteins themselves have functions in DNA repair, then pathogenesis could be the result of a combination of loss of function and dominant-negative gain of function (Fig. 4). Accrual of DNA damage would also predispose CAG repeats, which are already susceptible to damage as discussed above, to further strand breaks, the repair of which could result in further repeat expansion, thus setting up a toxic cycle (Fig. 3C,D).