Gene regulation and epigenetics in Friedreich's ataxia


  • Cihangir Yandim,

    1. Gene Control Mechanisms and Disease, Department of Medicine and MRC Clinical Sciences Centre, Imperial College London, London, UK
    Search for more papers by this author
  • Theona Natisvili,

    1. Gene Control Mechanisms and Disease, Department of Medicine and MRC Clinical Sciences Centre, Imperial College London, London, UK
    Search for more papers by this author
  • Richard Festenstein

    Corresponding author
    • Gene Control Mechanisms and Disease, Department of Medicine and MRC Clinical Sciences Centre, Imperial College London, London, UK
    Search for more papers by this author

Address correspondence and reprint requests to Richard Festenstein, Gene Control Mechanisms and Disease,Department of Medicine and MRC Clinical Sciences Centre, Imperial College London, Du Cane Road, W12 0NN, London, UK. E-mail:


This is an exciting time in the study of Friedreich's ataxia. Over the last 10 years much progress has been made in uncovering the mechanisms, whereby the Frataxin gene is silenced by (GAA)n repeat expansions and several of the findings are now ripe for testing in the clinic. The discovery that the Frataxin gene is heterochromatinised and that this can be antagonised in vivo has led to the tantalizing possibility that the disease might be amenable to a more radical therapeutic approach involving epigenetic modifiers. Here, we set out to review progress in the understanding of the fundamental mechanisms whereby genes are regulated at this level and how these findings have been applied to achieve a deeper understanding of the dysregulation that occurs as the primary genetic lesion in Friedreich's ataxia.

Abbreviations used

three prime end


five prime end




absent, small, homeotic discs related protein (methyltransferase)


base pair


CREB binding protein


complementary deoxyribonucleic acid


chromatin immunoprecipitation




CCCTC-binding factor


carboxy terminal domain


Duchenne Muscular Dystrophy 1


Dystrophia myotonica-protein kinase gene


DNA methyl transferase


downstream promoter element


double homeobox 4 gene


Epstein-Barr virus


histone-lysine N-methyltransferase


FXN antisense transcript 1


fluorescence in situ hybridisation


Fragile X mental retardation 1 gene


Friedreich's ataxia


Frataxin gene


Fragile X syndrome


green fluorescent protein


Gcn5-related N-acetyltransferase


histone H3 acetylation


histone H3 lysine 27 tri-methylation


histone H3 lysine 9 mono- methylation


histone h3 lysine 9 di-methylation


histone H3 lysine 9 tri-methylation


histone H4 acetylation


histone acetyl transferase


human cluster of differentiation 2 (surface marker)


histone deacetylase


histone demethylases




histone methyl transferase


heterochromatin protein 1




induced pluripotent stem cell






locus control region


long interspersed element


methylated CpG binding protein




mariner DNA transposon


mammalian interspersed repeats


mixed lineage leukaemia factor


messenger RNA




paired box (group of proteins)


peripheral blood mononuclear cells


polymerase chain reaction


position effect variegation


peroxisome proliferator-activated receptor-γ coactivator 1 α


polycomb repressive complex




RNA-induced transcriptional gene silencing complex


RNA interference


RNA polymerase


ribonucleic acid




reactive oxidative species




Schizosaccharomyces pombe


SET domain bifurcated 1 (methyltransferase)


short interspersed element


small interfering RNA


SRY-like box


serum response factor




Drosophila suppressor of variegation 3-9 (human homologue)


SWItch/Sucrose NonFermentable


TATA box (DNA)


transcription factor AP2


transcription start site


untranslated region

Being the most common autosomally recessive neurodegenerative disorder, Friedreich's ataxia (FRDA) presents an interesting example of so-called ‘triplet-repeat’ diseases. The vast majority of FRDA cases are linked to an expansion mutation of (GAA)n repeats within the Frataxin (FXN) gene which encodes the mitochondrial protein Frataxin (see the review written by Pastore and Puccio 2013). In the case of (GAA)n expansion, FXN is silenced and hence less Frataxin is produced causing the FRDA disease phenotype. Interestingly, this pathological expansion resides neither on the promoter nor in the coding region of FXN but in the first intron of the gene. At the time the original discovery was made (Campuzano et al. 1996), one could only speculate about how this non-coding region mutation causes Frataxin deficiency. However, the new millennium has witnessed much progress in uncovering important aspects of chromatin and gene regulation. Although there are many unanswered questions, harnessing current insights in molecular genetics as well as the efforts of numerous scientists in the FRDA field has led to a better understanding of this incurable and frequently devastating disease with implications for the design of more rational and radical therapeutic approaches. Here, we review the regulation of the FXN gene and its epigenetic silencing in FRDA. To understand the key molecular events in the silencing of FXN in disease, one needs to have an insight into how chromatin is regulated. Therefore, this review also aims to provide information about different aspects of chromatin and how it is dysregulated in FRDA.

Transcriptional and physiological regulation of the FXN gene

As illustrated in Fig. 1a, the human FXN gene resides on the positive strand of chromosome 9 (9q21.11; and contains five exons and four introns. There are three putative cytosine–phosphoguanine (CpG) islands ( and the first of them includes exon1 and extends towards intron1 (Takai and Jones 2002). Importantly, within the first intron are polymorphic (GAA)n repeats in the centre of an Alu element, from which they were thought to have arisen (Campuzano et al. 1996; Cossee et al. 1997; Clark et al. 2004). Consistent with the structure of Alu elements, a polymorphic poly(A) tract was detected in the preceding region of (GAA)n repeats (Cossee et al. 1997; Justice et al. 2001; Monticelli et al. 2004). Whereas the number of these repeats were reported to be between 10 and 66 in healthy individuals, the repeated copies could be up to 1700 in the case of FRDA associated alleles (Campuzano et al. 1996; Durr et al. 1996; Cossee et al. 1997). It is still noteworthy though, the cascade of molecular events at the level of (GAA)n instability which gives rise to expansion mutations are known to start to occur when the repeat number reaches a threshold of ~35. Therefore, repeat numbers between 35 and 66 are often referred as ‘pre-mutational’.

Figure 1.

The human Frataxin (FXN) gene and its regulatory components. (a) Full body of FXN. The FXN gene, which encodes the essential mitochondrial protein Frataxin, consists of five exons and four introns. Three putative cytosine–phosphoguanine (CpG) islands were identified on FXN by USC CpG island searcher ( Importantly, FXN's first intron contains a number of (GAA)n triplet repeats. There are approximately 10–66 repeats in healthy individuals but higher numbers (approximately 66–1700) are associated with Friedreich's ataxia, where the FXN gene is pathologically silenced. (b) Regulatory elements at the 5′ end of FXN. FXN is known to be transcribed from two main transcription start sites (TSS). TSS1 is 221 bp upstream of the translation start site (ATG), whereas TSS2 is only 62 bp upstream. The region between TSS1 and the first exon is thought to be a TATA-less downstream promoter, which contains Inr/downstream promoter element-like elements. Moreover, binding sequences were identified for transcription factors serum response factor, TFAP2 and EGR3 as well as for the insulator protein CCCTC-binding factor. An E-box is present in the vicinity of (GAA)n repeats. Although there is no direct evidence, this site could potentially be a target for the muscle specific factor MyoD or c-myc. Interestingly, apart from (GAA)n repeats, a number of repetitive DNA elements have been identified at the FXN locus. These include L2 (LINE) and Alu (SINE) elements as well as MIRb and mariner DNA transposon. Although the deletion of these elements significantly impaired a reporter FXN construct (Greene et al. 2005), their exact function on FXN regulation is not known.

So far, two major transcription start sites (TSS) were identified at the FXN gene (see Fig. 1b). 5′ rapid amplification of cDNA ends experiments identified a transcription start site (TSS1) 221bp upstream of the ATG translation start site (Campuzano et al. 1996) and primer extension approach recently revealed another transcription start site (TSS2) located 62bp upstream of the ATG (Kumari et al. 2011). Which TSS is dominant is unclear; however, Kumari et al. suggested that TSS2 could be the primary transcription start site in Epstein-Barr virus-transformed lymphoblastoid cell lines. Apart from these, several other TSSs were also reported by sequencing of cDNA derived from several human cell lines and tissues (Strausberg et al. 2002); but these were not confirmed with additional methods such as the ones described for TSS1 and TSS2.

The promoter of FXN is unusual in terms of core-promoter function. Initial experiments with reporter luciferase constructs in mouse skeletal muscle cells revealed that a region covering 1034bp upstream and 100bp downstream of TSS1 plays the major role in the regulation of the FXN (Greene et al. 2005). Notably, this region lacks a TATA box but contains Inr/downstream promoter-like elements starting 24bp downstream of TSS1 (For a review on regulatory promoter elements, see Maston et al. 2006). Greene et al. reported that the deletion of these elements significantly reduced the expression of FXN.

There are only limited results so far in terms of the identity of transcription factors regulating FXN. Chromatin immunoprecipitation (ChIP) and electromobility shift assay approaches revealed binding sites for a serum responsive factor (91–110 bp downstream TSS1) and a transcription factor AP2 (TFAP2; 139–154 bp downstream of TSS1), deletion of which significantly reduced the transcription of FXN reporter constructs (Li et al. 2010). Serum response factor is known to be important in neuronal development in maintaining synaptic activity and plasticity (Knoll and Nordheim 2009) whereas TFAP2 is implicated in neural-crest development (Eckert et al. 2005). Li et al. also report an early growth response protein factor-binding site (EGR3, 233-249 bp downstream of the translation start site ATG), whose deletion impairs FXN expression. However, there is, as yet, no direct evidence for the binding of this factor. Another independent study reported FRDA-like symptoms in EGR-3 deficient mice (Tourtellotte and Milbrandt 1998). Similarly, an E-box element was shown to be crucial for FXN expression via reporter assays (Greene et al. 2007). This element is likely to be bound by members of a large family of basic Helix-Loop-Helix E-box-binding proteins that includes the muscle specific factor MyoD and c-myc (Murre et al. 1989; Walhout et al. 1997). In addition to these, CCCTC-binding factor (CTCF) was reported to be bound between 154 and 173 bp downstream of TSS1 (De Biase et al. 2009). This will be discussed further in the following sections.

A number of repetitive DNA elements are found within the FXN locus. Greene et al. (2005) reported remnants of LINE (L2) elements as well as mammalian interspersed repeats and mariner DNA transposon elements in the downstream vicinity of TSS1 and the first intron of the gene (See Fig. 1b). Moreover, intronic (GAA)n repeats are thought to be rooted by an intronic Alu element, as mentioned earlier (Campuzano et al. 1996; Cossee et al. 1997; Clark et al. 2004). Although the exact functions of these repetitive elements were not elucidated, their deletions seem to impair FXN expression significantly according to Greene et al. (2005). Repetitive DNA elements are often thought to be ‘evolutionary junk’ material; however, numerous studies now suggest that they could have gained important roles in regulating nearby genes. Indeed, many human genes were reported to be transcribed from the antisense promoter of LINE elements (Nigumann et al. 2002), which may contain sequences for the binding of various transcription factors including SRY-like box proteins and YY1 (Nigumann et al. 2002; Yang et al. 2003; Athanikar et al. 2004). Similarly, Alu (SINE) elements were shown to act as enhancers or provide direct binding sites for hormone receptors (Norris et al. 1995; Piedrafita et al. 1996) and transcription factors such as PAX6 (Zhou et al. 2002). Promoters of Alu elements are recognised by RNA Polymerase III; for a review, see (Richard et al. 2008). Interestingly, an independent study reported that RNAPIII promoters may act as enhancers for RNAPII transcribed genes (Oliviero and Monaci 1988; Tomilin 1999); possibly by creating nucleosome positioning effects (Englander and Howard 1995).

FXN is considered to be expressed at a relatively low level even in healthy individuals with a variation less than 20% among the European population (Su et al. 2004; Boehm et al. 2011; Sacca et al. 2011). Its expression is slightly higher in dorsal root ganglia and the granular layer of the cerebellum as well as in tissues with high metabolic rate such as heart, brown fat and liver (Campuzano et al. 1997; Koutnikova et al. 1997; Rotig et al. 1997; Al-Mahdawi et al. 2006). It is interesting to note that over-expression of Frataxin was shown to be toxic for cells, perhaps implying that a fine balance is needed for the regulation of its expression (Fleming et al. 2005; Navarro et al. 2011). Nevertheless, there is still not much known regarding its physiological regulation. Studies so far suggest that there is a tight link between FXN regulation and iron concentration; consistent with the function of Frataxin protein on iron metabolism (see review written by Pastore and Puccio 2013). FRDA is associated with the accumulation of iron in mitochondria and its depletion in other cellular compartments (Rouault and Tong 2005; Li et al. 2008). In line with this, experimentally induced iron depletion resulted in reduced levels of FXN transcripts in healthy and FRDA cells (Li et al. 2008). This may suggest a negative feedback mechanism between the disease phenotype (i.e. less iron in the cytoplasm and nucleus) and FXN expression in FRDA; therefore accentuating FXN gene silencing and the disease phenotype. Still, the effect of iron on FXN expression dynamics and the molecular basis of this link remain to be investigated.

Low levels of FXN expression were also found in conjunction with the depletion of PGC1α, the coactivator of PPARγ (Coppola et al. 2009; Marmolino et al. 2010). The latter is a nuclear receptor activated by fatty acids and is a key regulator of mitochondrial biogenesis as well as reactive oxygen species metabolism (Wu et al. 1999; Kelly and Scarpulla 2004). The link between the PPARγ and FXN regulation has also not been resolved yet; however, a PPARγ agonist routinely used in the laboratory seems to up-regulate FXN in FRDA primary lymphoblasts (Marmolino et al. 2009). Importantly, PPARγ pathway is thought to be crucial in the pathogenesis of diabetes (Sugden et al. 2010). According to Coppola et al. (2009), the unresolved link between PPARγ and FXN gene regulation could potentially explain why diabetes is associated with FRDA (see review written by Cnop et al. in this issue).

Blockage of FXN transcription by expanded (GAA)n repeats

(GAA)n repeats are prone to expansion and contraction mutations potentially facilitated by mechanisms involved in DNA replication and repair as well as transcription. Critical tissues affected by Friedreich's ataxia (i.e. dorsal root ganglia and cerebellum) were shown to have a tendency to accumulate expansion mutations (Al-Mahdawi et al. 2004; Clark et al. 2007; De Biase et al. 2007). Expanded (GAA)n repeats (67-1700) were associated with decreased levels of FXN mRNA (4-29% of normal in FRDA, ~50% of normal in asymptomatic carriers) (Campuzano et al. 1997; Sacca et al. 2011). This correlates well with a drop in Frataxin protein levels (Sacca et al. 2011). The extent of (GAA)n expansion correlates positively with the disease severity and correlates negatively with the age of onset (Filla et al. 1996; Monros et al. 1997; Montermini et al. 1997a; Bhidayasiri et al. 2005; Evans-Galea et al. 2012).

It is generally accepted that the dysregulatory events resulting in reduced FXN levels occur during the process of transcription per se and are unlikely to take place at the post-transcriptional level. In line with this, the half-life of FXN mRNA (~11 h) was not found to be different between FRDA patients and healthy individuals (Punga and Buhler 2010). Furthermore, there is no substantial evidence that the expanded (GAA)n repeats affect the RNA splicing of the FXN gene. Minigene constructs transfected into mammalian cells (Baralle et al. 2008) as well as a yeast gene fused with expanded (GAA)n repeats (Shishkin et al. 2009) exhibited splicing differences in the case of expanded repeats; however, these findings were not reproduced in the native context of an intact FXN locus (Bidichandani et al. 1998; Punga and Buhler 2010).

One way of explaining reduced FXN transcription in FRDA could be the physical-blockage effects of (GAA)n repeats on the RNAPII transcription machinery. To understand this hypothesis, one should have a closer look at the unusual conformations formed by expanded (GAA)n repeats. These were reviewed and illustrated in detail by Kumari and Usdin (2012). One of the non-B DNA conformations inhabited by expanded (GAA)n repeats is the purine: purine: pyrimidine triplex, where a purine already paired with a pyrimidine based on Watson–Crick model, also forms reverse Hoogsten pairing with another purine (Mariappan et al. 1999; Grabczyk and Usdin 2000a,b; Jain et al. 2002; Potaman et al. 2004). Under the influence of negative supercoiling, these triplexes are likely to adopt secondary structures known as ‘sticky DNA’ (Gacy et al. 1998; Sakamoto et al. 1999). These were suggested to impair transcription by creating a physical blockage effect on transcription by making it more difficult for the elongating RNAPII complex to unwind the DNA template and move forward (Sakamoto et al. 2001). An illustrative model focusing on the transcriptional blockage effects is given in the discussion section (Fig. 5a).

Consistent with this phenomenon, polyamides which were suggested to block sticky DNA conformations by binding the (GAA)n tracts, could up-regulate FXN expression in FRDA lymphoblastoid cell lines but not in resting primary cells (Burnett et al. 2006). In addition to ‘sticky DNA’, hairpin structures are thought to be formed by unusual hydrogen bonds between G•G, G•A or A•A. (Suen et al. 1999; Chou et al. 2003; Heidenfelder et al. 2003; Kumari and Usdin 2012). (GAA)n repeats were also reported to form parallel duplex structures, where a reverse Watson–Crick orientation is inhabited by two hydrogen bonds between purines and pyrimidines as opposed to the normal conformation where three hydrogen bonds are formed (LeProust et al. 2000). Furthermore, triplex structures formed by (GAA)n repeats allow the formation of stable RNA•DNA hybrids (R-loops) during transcription (Grabczyk and Usdin 2000a,b; Grabczyk et al. 2007). Such hybrids are thought to stall RNAPII at the end of the repetitive DNA (Bentin et al. 2005; Tous and Aguilera 2007; McIvor et al. 2010; Kumari and Usdin 2012). Notably, unusual conformations described here are not only limited to (GAA)n expansions and could be seen on other triplet repeats such as (CNG)n. These were reviewed in detail elsewhere (Mirkin 2007).

Triplex structures inhabited by expanded (GAA)n repeats were shown to be recognised by the DNA repair machinery. Mismatch repair proteins (particularly MSH2/MSH3 dimers) were shown to be enriched flanking expanded (GAA)n repeats in various models including induced pluripotent stem cells derived from FRDA patients (Kim et al. 2008; Bourn et al. 2009; Ku et al. 2010; Du et al. 2012; Ezzatizadeh et al. 2012). These were implicated in the repeat expansion process. DNA damage mechanisms were associated with stalled RNAPII as well, especially in the case of transcription-coupled nucleotide excision repair (Mellon et al. 1987; Hanawalt and Spivak 2008).

Although the unusual conformations caused by (GAA)n repeats provide a logical explanation for reduced FXN mRNA levels, the existence of such structures were only confirmed in vitro or in Escherichia coli. Therefore, more direct evidence is still needed as to whether these actually take place in patient cells on the native FXN locus. In vitro transcription studies that involve RNAse protection assays and northern blots so far underlined a transcriptional elongation defect when (GAA)n repeats are expanded, consistent with the models suggested above (Bidichandani et al. 1998; Ohshima et al. 1998; Sakamoto et al. 1999, 2001; Grabczyk and Usdin 2000a,b; Krasilnikova et al. 2007). Studies on patient lymphoblastoid cells also reported a pronounced defect in the elongation of RNAPII complex rather than its initiation (Punga and Buhler 2010; Kim et al. 2011). Punga et al. found similar levels of the initiating form of RNAPII (phosphorylated at S5 on its CTD domain) on the promoter and immediately upstream of (GAA)n repeats in lymphoblastoid cells derived from healthy individuals and FRDA patients. They also observed chromatin related differences corresponding to a transcriptional elongation deficit in FRDA (discussed further within the next section). Consistently, Kim et al. reported reduced levels of primary transcripts only downstream of (GAA)n. They performed ChIP using an anti-RNAPII antibody that does not distinguish between its initiating and elongating forms and observed reduced levels of RNAPII immediately upstream of (GAA)n repeats but not at the FXN promoter. On the other hand, Kumari et al. (2011) suggest both initiation and elongation defects in the case of FRDA. It is noteworthy that the ChIP antibodies used to distinguish initiating and elongating forms of RNAPII in Kumari et al. (2011) study have recently been reported to cross-react (Brookes et al. 2012), making it more difficult to interpret the results.

Epigenetic silencing of FXN by expanded (GAA)n repeats

Overview of epigenetics and Friedreich's ataxia

In a eukaryotic nucleus, DNA is hierarchically packaged into chromosomes. This starts with the wrapping of DNA around histone proteins forming the so-called ‘nucleosome’ which is the basic subunit of chromatin. Chromatin could be less dense and easily accessible to RNAPII (i.e. euchromatin) or tightly packed and inaccessible (i.e. heterochromatin). Accessibility of chromatin is mainly mediated by the interaction of DNA with histone proteins which are subject to a plethora of biochemical modifications that determine the degree of such interactions and hence the compactness of chromatin (Kouzarides 2007; Bannister and Kouzarides 2011). In addition, direct methylation of DNA and RNA-interference based mechanisms have been shown to be involved in the dynamic regulation of chromatin. All of these key events are thought to be inherited independently of the genetic code; a phenomenon known as ‘epigenetics’ (Margueron and Reinberg 2010). Understanding of epigenetic mechanisms in the last decade elucidated novel layers in carcinogenesis as well as in the pathology of neurological diseases such as Alzheimer's or Parkinson's disease (Urdinguio et al. 2009).

Although epigenetics is defined to be independent of DNA sequence, the content of the genomic landscape seems to play an important role in determining the epigenetic chromatin status of a given gene. Importantly, 55% of the human genome is comprised of repetitive DNA sequences (Lander et al. 2001) and heterochromatin is actually thought to have evolved as a protective response against the disruptive effects of aberrant transcriptional noise which is likely to have been triggered by such elements (Yoder et al. 1997; Bashkirov 2002). Indeed, over-expression of repetitive DNA was linked to genomic instability that is associated with cancer and ageing (Schulz et al. 2006; Belancio et al. 2008; Ting et al. 2011; Zhu et al. 2011). Heterochromatinization of repetitive DNA is also important for the maintenance of nuclear architecture needed for the spatial organization of centromeres and sister chromatid segregation during cell division (Fisher and Merkenschlager 2002; Maison et al. 2002; Guenatri et al. 2004; Peters et al. 2008). Being first described by Heitz (1928), heterochromatin is visualised, after DNA staining, through the microscope as densely stained dark patches (Fig. 2a). Some of its important characteristics are that it is more prevalent in non-dividing differentiated cells, forms clusters close to the nuclear periphery, replicates late in S-phase, exhibits reduced recombination rates and is associated with transcriptional repression, particularly at centromeric and telomeric repeats (Dillon 2004).

Figure 2.

Heterochromatin and its link with (GAA)n repeats. (a) A mammalian cell nucleus during interphase. When looked at through the microscope after DNA staining, a mammalian cell nucleus (presented as grey) appears to have light and dark regions. Lightly stained parts represent accessible and transcriptionally active ‘euchromatin’ whereas densely stained regions are condensed and transcriptionally silent ‘heterochromatin’. (b) An archetypal example of position effect variegation (PEV) on Drosophila eye. The ‘white’ gene residing on a long chromosome arm is responsible for red eye-colour (left). A coding sequence mutation in this gene typically causes white eye-colour (middle). Radiation induced translocation of the ‘white’ gene into a region close to pericentromeric heterochromatin causes a patchy white/red eye-colour as a result of heterochromatin's invasive nature (right). (c) PEV-like heterochromatic effects induced by (GAA)n in mouse T-cells. The hCD2 transgene with a truncated locus control region allows one to study PEV in mammals by using flow cytometry. The expression of this transgene does not variegate when it is located in a long chromosomal arm (left). When the transgene is placed into close proximity of pericentromeric heterochromatin, the expression of hCD2 variegates in a similar fashion to that observed in the Drosophila eye (middle). Strikingly, insertion of a long (GAA)n tract next to the hCD2 transgene results in a position independent PEV-like variegation (right) implying that (GAA)n repeats could nucleate heterochromatin.

Heterochromatin is invasive and tends to affect the expression of nearby genes. First lines of evidence regarding this phenomenon were reported in Drosophila, where the euchromatic ‘white’ gene responsible for the red eye-colour was translocated close to pericentromeric heterochromatin as a result of exposure to radiation (Fig. 2b) (Muller 1930). The resulting phenotype was a mosaic eye colour because of the stochastic expression of white in some cells but not in others, a phenomenon known as ‘position effect variegation (PEV)’. PEV takes place when competing factors involved in the formation of euchromatin or heterochromatin are in equilibrium (Fig. 3). Such factors include epigenetic marks including DNA methylation, histone modifications and antisense transcripts as well as sequence elements such as enhancers, silencers, insulators or locus control regions and repetitive DNA (Zuckerkandl 1974; Tartof et al. 1984, 1989; Locke et al. 1988; Festenstein et al. 1999; Dillon and Festenstein 2002). The presence or over-expression of chromatin modifiers could therefore shift the chromatin status in the opposite direction. Notably, many key modifiers of chromatin were discovered using PEV as a model tool (Reuter et al. 1987; Reuter and Spierer 1992; Tschiersch et al. 1994; Elgin 1996; Rea et al. 2000; Dillon and Festenstein 2002).

Figure 3.

Competition between euchromatin and heterochromatin. Easily accessible euchromatin is associated with DNA demethylation, histone acetylation and H3K4 methylation. On the other hand, inaccessible and tightly packed heterochromatin is linked to DNA methylation, histone deacetylation, H3K4 demethylation, H3K9 and H3K27 methylation. ‘Glue proteins’ such as HP1 or PRC1 components are known to create a higher order chromatin structure in heterochromatin via their interactions with the nucleosome and each other. The concentration of modifiers responsible for these effects as well as the presence of the binding sites at a given locus determines the final status of transcription. It is thought that if heterochromatic and euchromatic factors are in balance, stochastic expression of genes (PEV) takes place.

Importantly, microsatellite (GAA)n repeats are shown to behave like pericentromeric heterochromatin. Ohno et al. (2002) reported that (GAA)n tracts show similar spatial nuclear organizations as centromeric repeats. Moreover, expanded (GAA)n repeats are able to silence a nearby gene as shown using the human hCD2 transgene, which encodes a surface marker in thymocytes and T-lymphocytes when introduced into mouse genome. The expression of an hCD2 transgene with a truncated locus control region, when placed close to pericentromeric heterochromatin, exhibits a variegating pattern evident on flow cytometry analyses; providing a tool to study PEV in a mammalian system (Fig. 2c) (Festenstein et al. 1996). Similar to pericentromeric repeats, long microsatellite (GAA)n as well as (CTG)n tracts also induced heterochromatic silencing when fused with the hCD2 gene even when the construct was integrated into a euchromatic location (Saveliev et al. 2003). This discovery pointed to the possibility of abnormal heterochromatinization and thereby silencing of the FXN gene in FRDA, which is associated with expanded intronic (GAA)n elements. Notably, such silencing was also found to be highly sensitive to dosage of the classical PEV modifier heterochromatin protein 1 (HP1), raising the possibility that PEV modifiers were potentially disease modifiers in FRDA. Since then, numerous studies searched for the proof of such heterochromatic effects on the endogenous FXN locus and investigated the chromatin related mechanisms involved in its pathological silencing. The following sections summarise the major findings within the context of epigenetic gene silencing in FRDA.

Direct methylation of DNA

A key layer in regulating gene expression is based on the fact that DNA could be covalently modified at 5′CpG3′ residues by the addition of methyl groups (*-CH3) at the cytosines (C) (Holliday and Pugh 1975; Riggs 1975; Bestor and Ingram 1983; Li et al. 1992). This is performed via the activity of DNA methyltransferases (DNMTs) DNMT-1, -3a and -3b. Whereas, DNMT1 is mainly responsible for the inheritance of this epigenetic mark during DNA replication (Pradhan et al. 1999), the others were shown to be crucial in regulating gene expression at various stages of embryonic development and X chromosome inactivation (Beard et al. 1995; Okano et al. 1999; Geiman and Muegge 2010). The methylation event seems to be reversible (Ramchandani et al. 1999); however, no DNA demethylating enzymes have been identified so far. It is generally accepted that CpG methylation decreases the probability of transcription factor binding or creates additional binding sites for specific proteins (e.g. MeCP2) that recognise methylated DNA and recruit chromatin modifying enzymes (e.g. HDACs) (Dhasarathy and Wade 2008). Whether histone modifications take place before or after the DNA methylation is still a chicken-egg problem. Nevertheless, DNMTs were reported to act in synergy with chromatin modifiers. One important example is the interaction of DNMTs with the heterochromatin protein, HP1 (Fuks et al. 2003; Lehnertz et al. 2003). Importantly, DNA methylation is thought to regulate the majority of the human genome (Shen et al. 2007) and dysregulation of this key modification was associated with cancer and neurodegenerative disorders (Robertson 2005).

The tight link between DNA methylation and gene silencing prompted FRDA scientists to investigate whether this epigenetic mark is affected in the case of expanded (GAA)n repeats. According to the University of Southern California's CpG island search algorithm (Takai and Jones 2002), the FXN gene contains a CpG island covering the promoter, first exon and the beginning of first intron (Fig. 1). Most research, so far, focused on this region and some other CpG residues flanking the (GAA)n repeats. These are summarised in Table 1. Greene et al. (2007) reported a low level of DNA methylation flanking the unexpanded (GAA)n repeats in healthy-derived lymphoblastoid cell lines using bisulphite sequencing. They thought that this methylation pattern in healthy might be because of the Alu element, in which the (GAA)n repeats reside, triggering DNA methylation and causing its bidirectional spread (Liu and Schmid 1993). Strikingly, the number of methylated residues was significantly higher in patient cells. Greene et al. (2007) also reported that some of the cytosines were fully protected from methylation in healthy samples whereas they were extensively methylated in FRDA. One such key CpG residue overlaps with the E-box mentioned earlier.

Table 1. A summary of major studies investigating chromatin changes on the FXN locusThumbnail image of

Al-Mahdawi et al. (2008) performed a similar type of analysis using FRDA autopsy samples including relevant tissues such as heart and brain, which exhibited elevated CpG methylation levels upstream of (GAA)n. The same study also reported similar methylation patterns in FRDA-affected tissues of two strains of FRDA disease model mice YG8 and YG22. In line with these findings, two large-scale studies on peripheral blood mononuclear cells also reported significantly higher levels of DNA methylation on the pathologically silenced FXN locus, particularly upstream of the expanded repeats (Castaldo et al. 2008; Evans-Galea et al. 2012). Both found a negative correlation with the age of disease onset and the level of DNA methylation on FXN. Castaldo et al. (2008) also reported that the degree of methylation is proportional to the extent of (GAA)n expansion and Evans-Galea et al. (2012) presented a positive correlation between FXN gene silencing, disease severity and the level of DNA methylation. Interestingly, 5′ untranslated region (UTR) region of the FXN gene did not show any difference in terms of DNA methylation between healthy and FRDA-patient fibroblasts (De Biase et al. 2009).

Histone modifications

The fundamental subunit of chromatin - the nucleosome - consists of DNA wrapped around the highly conserved histone H3/H4 tetramer and two pairs of H2A/H2B dimers (Luger et al. 1997). In addition to these core histones, the linker DNA between the nucleosomes is occupied by the linker histone H1 (Allan et al. 1980). Histones are highly basic and positively charged and this explains their strong affinity towards the negatively charged DNA (Luger et al. 1997). As mentioned earlier, biochemical modifications of histones (mainly protruding N-terminal tails of H3 and H4), modulate the degree of this affinity towards the DNA; either by changing the electrostatic charge of the histone or creating conformational alterations (Luger et al. 1997; Kouzarides 2007; Bannister and Kouzarides 2011). The strength of intermolecular interactions with DNA as well as interactions between other histone proteins from the neighbouring nucleosomes are thought to play an important role in regulating the extent of chromatin compaction. In addition, histones could also interact with other proteins such as HP1 to form a higher order compactness of DNA (Bannister et al. 2001; Lachner et al. 2001).

Among the many covalent histone modifications, the most prevalent and well-studied ones are acetylation and methylation of lysine (K) residues. All types of core histones can be acetylated and this correlates well with the transcriptionally active euchromatic state; most possibly because of the fact that the addition of acetyl groups (CH3CO-) reduces the positive charge and thereby the affinity towards the DNA (Allfrey et al. 1964; Sealy and Chalkley 1978; Halleck and Gurley 1981; Hebbes et al. 1988; Hong et al. 1993; Shahbazian and Grunstein 2007). In addition, acetylated histones are known to recruit chromatin-remodelling complexes such as SWI/SNF (Hassan et al. 2002). The occurrence of the acetylation mark on many different lysine residues is thought to create a cumulative effect in loosening up the chromatin structure. Among these many residues, H3K9 and H4K16 acetylation are known to be robust marks for transcriptionally active euchromatin (Mottus et al. 2000; Dion et al. 2005; Shogren-Knaak et al. 2006). While histone acetyltransferases catalyse the acetylation reaction, histone deacetylases (HDACs) remove this euchromatic mark. Phylogenetically classified, there are four major groups of histone acetyltransferases (GNAT, MYST, P300/CREB binding protein and nuclear receptor families) and also four groups of HDACs (HDAC I, II, III, IV – HDACIII also known as sirtuins). Various enzymes belong to each family and it is generally difficult to tell their specificity for a particular lysine residue. Still, some are known to dominate particular nuclear processes (e.g. heat shock or hormone response). These were reviewed in previous publications (Sterner and Berger 2000; Martin and Zhang 2005; Shahbazian and Grunstein 2007) and also presented in this issue by Pandolfo et al.

Deacetylated lysine (K) residues of histones could be subject to another biochemical modification; methylation, which can also take place on arginine (R) residues. Unlike acetylation, the addition of single or multiple methyl groups (*-CH3) does not change the electrostatic charge of histones but it introduces conformational changes because of hydrophobic alterations. Therefore, the outcome of methylation on transcription could be either activation or repression, depending on the place of the methylated residue (Lee et al. 2005; Kouzarides 2007). As reviewed by Kouzarides (2007) and Bannister and Kouzarides (2011), histone methylation is carried out by various histone methyltransferases (HMTs) and reversed by histone demethylases whose mechanisms of action were clearly defined for specific residues. Among the many different types of histone methylations, at least two of them are known to be hallmarks of heterochromatin. H3K9 trimethylation (H3K9m3) is associated with a highly compact chromatin structure and particularly enriched in constitutive heterochromatin such as satellite or telomeric repeats (Jenuwein and Allis 2001; Peters et al. 2001). On the other hand, mono- and di-methylated forms (H3K9me1 and H3K9me2) are mostly enriched on the silenced promoters of formerly euchromatic genes (Barski et al. 2007; Mikkelsen et al. 2007). In constitutive heterochromatin, the histone methyltransferase Drosophila suppressor of variegation 3-9 (human homologue) (SUV39H) catalyses H3K9 trimethylation which is recognised by heterochromatin protein 1 (HP1) (Rea et al. 2000; Bannister et al. 2001; Lachner et al. 2001). HP1 self-dimerises and is therefore thought to create a ‘glue effect’ on chromatin by holding adjacent nucleosomes together. Importantly, SUV39H methyltransferase was shown to interact with HP1 (Aagaard et al. 1999; Bannister et al. 2001; Schotta et al. 2002). This is thought to be the basis of the spreading behaviour of heterochromatin (Fig. 4). In addition to SUV39H, other HMTs such as G9a, SETDB1 and RIZ1 are known to act mostly on inactivated gene promoters (Kouzarides 2007).

Figure 4.

Spreading of heterochromatin via HP1 and SUV39H. Heterochromatin is invasive. In a classical example, the spreading of chromatin takes place via the dimerisation of HP1 proteins as well as intermolecular interactions between HP1 and the H3K9 methyltransferase SUV39H. Here, SUV39H catalyses the H3K9 methylation, which is subsequently recognised by HP1. While HP1 molecules self-dimerise and create a more compact chromatin, they also recruit more SUV39H to the locus to exacerbate heterochromatin's spreading.

Another vital methylated residue is H3K27. Di- and tri- methylated forms of this residue are associated with facultative heterochromatin formed at previously euchromatic genes and the inactivated X chromosome (Wutz and Jaenisch 2000; Barski et al. 2007; Mikkelsen et al. 2007). The activity of H3K27 methylation has been tightly linked to the polycomb system, where one polycomb complex (PRC2) catalyses the methylation and another one (PRC1) recognises methylated H3K27 residues. PRC1 monoubiquitinates the globular H2AK119, which prohibits chromatin remodelling and hence successive elongation of RNAPII (Fischle et al. 2003; Wang et al. 2004; Cao et al. 2005; Hublitz et al. 2009). Intramolecular interactions of PRC2 complexes are thought to be responsible for a low-level chromatin compaction and its spreading (Kim et al. 2002; Francis et al. 2004; Grau et al. 2011). Apart from H3K9 and H3K27 methylation, H4K20 methylation has also received particular attention recently for its association with heterochromatin, as reviewed by Balakrishnan and Milavetz (2010).

As opposed to the ones described above, some other methylation events are associated with the active state of chromatin. RNAPII is known to associate with specific SET methyltransfereases. Thus, the initiating form of RNAPII (phosphorylated on S5) recruits SET1 which catalyses the methylation of H3K4 residues (Ng et al. 2003). While the trimethylated residues of H3K4 are particularly enriched at active promoters, mono- and di- methylated forms are known to extend towards the coding regions (Santos-Rosa et al. 2002; Barski et al. 2007; Mikkelsen et al. 2007). Apart from SET1, there are other HMTs acting on H3K4 methylation (e.g. MLL and ASH1), which possibly act on the first nucleosome where the basal transcription machinery is initially recruited (Kouzarides 2007; Vermeulen et al. 2007). Furthermore, the elongating form of RNAPII (phosphorylated on S2) is known to interact with SET2 which methylates H3K36 residues (Xiao et al. 2003). This mark is therefore particularly enriched in the active coding regions towards the 3′ end (Saunders et al. 2006; Barski et al. 2007; Mikkelsen et al. 2007). Although not much is known about it, H3K79 methylation is another mark that is associated with active transcription (Nguyen and Zhang 2011).

FRDA scientists have collected elaborate ChIP data relating to the histone modification changes on the silenced FXN locus with expanded (GAA)n. In general, all of these studies reported high levels of heterochromatin marks in the first intron of the pathologically silenced FXN in a similar fashion with DNA methylation. An up-to-date summary of literature addressing histone modifications on the FXN locus is also given in Table 1. These studies establish that the expanded (GAA)n repeats are indeed linked to heterochromatinization, as had been suggested by Saveliev et al. (2003) and is consistent with the observations of Ohno et al. (2002). Because of the unstable nature of (GAA)n repeats and their non-specific distribution throughout the genome, ChIP DNA could not be analysed using primers specific to the (GAA)n. However, it was revealed that the classical heterochromatin marks H3K9me2, histone H3 lysine 9 tri-methylation (H3K9me3) and histone H3 lysine 27 tri-methylation (H3K27me3) are enriched particularly in the immediate flanking regions of expanded (GAA)n repeats whereas acetylation marks are reduced. This suggests that heterochromatin is bidirectionally spreading from the expanded (GAA)n tract within the FXN locus. The spreading of heterochromatin from (GAA)n repeats is particularly obvious at the ChIP results presented by Chan et al. (2013), a study that has been published while this review was in press.

The invasive nature of heterochromatin prompted scientists to pursue experiments to understand the boundaries of (GAA)n induced heterochromatin. Importantly, De Biase et al. (2009) discovered a CTCF binding site 154bp downstream of TSS1 within the 5′UTR of the FXN gene. CTCF is a chromatin insulator protein known to prevent the spreading of heterochromatin and generate long-range organization of genes via intramolecular interactions (Bushey et al. 2008). How CTCF exerts the chromatin insulation activity is not yet clearly defined; however, it is thought to recruit euchromatic enzymes and RNAPII or serve as a physical barrier against heterochromatin (Chernukhin et al. 2007; Bushey et al. 2008; Bartkuhn et al. 2009; Witcher and Emerson 2009). Importantly, De Biase et al. (2009) reported depletion in CTCF binding in FRDA patient fibroblasts and associated this with increased antisense transcription (see the next section).

Whether the CTCF binding prevents the spreading of heterochromatin from the expanded (GAA)n repeats towards the promoter, has not been fully answered. De Biase et al. (2009) showed increased H3K9me3 and H3K27me3 levels as well as HP1 on the CTCF binding site of the silenced FXN locus of FRDA derived fibroblasts. Nevertheless, there is still no substantial functional evidence yet to confirm that CTCF prohibits the spreading of (GAA)n centred heterochromatin towards the FXN promoter. Knocking-down CTCF with siRNAs further down-regulated FXN, but whether this is a specific or indirect effect has not yet been fully elucidated. To further check whether CTCF is acting as a barrier it would be interesting to examine the chromatin marks on the promoter region, upstream of the CTCF binding site in the presence and absence of CTCF.

Although there is consensus on the heterochromatinization of the intronic region close to (GAA)n, there is not complete agreement regarding to the promoter/5′ UTR region of the FXN. Al-Mahdawi et al. (2008) reported the enrichment of H3K9me2 and H3K9me3 on the promoter region based on ChIP using human brain tissue. The same study did not observe any reduction in the acetylation marks on the promoter of the silenced FXN. In contrast, Punga and Buhler (2010) and Kim et al. (2011) presented no significant enrichment in heterochromatic methylation marks on the FXN promoter region of FRDA patient derived lymphoblastoid cell lines compared to healthy. In line with this, FRDA lymphoblastoid cells also did not exhibit lower levels of acetylation on the FXN promoter region of FRDA patients compared to healthy (Herman et al. 2006; Kim et al. 2011). However, another study reported lower levels of the majority of acetylation marks in the case of expanded (GAA)n repeats in lymphoblastoid cell lines (Kumari et al. 2011). The discrepancy between these results could be because of technical differences between ChIP experiments and different control cell lines. Thus, whether the promoter of the pathologically silenced FXN is heterochromatinised, is still unclear.

As mentioned before, the effects of (GAA)n induced heterochromatin on the initiation and elongation of transcription is another matter of discussion. In line with other histone marks and RNAPII ChIP data, transcription related marks such as H3K4 methylation as well as H3K36 and H3K79 methylations also seem to be unaffected in the promoter region of silenced FXN alleles as suggested by Kim et al. (2011). The same study reports that all of these key methylated residues are reduced on the coding regions of FXN in FRDA lymphoblastoid cells. The results regarding to FXN promoter and the coding regions are in agreement with a previous study (Punga and Buhler 2010) and overall suggest a transcription elongation defect rather than a problem with initiation. It is noteworthy though, that Kumari et al. (2011) observed reduced H3K4me2 methylation levels at the promoter as well.

Antisense transcription

Transcription is not solely specific for protein-coding genes and the majority of the human transcriptome has been shown to consist of non-coding RNAs arising from intergenic regions, repetitive sequences and antisense strands of coding genes (Djebali et al. 2012; Dunham et al. 2012; Harrow et al. 2012). Various models have been suggested regarding how antisense transcription may regulate/dysregulate gene expression. As reviewed by Faghihi and Wahlestedt (2009), antisense transcripts could dominate transcription and stall the coding transcription by prohibiting the movement of the transcription bubble in the sense direction or by creating RNA•DNA hybrids. Antisense transcription is also likely to increase the chances of mutation or recombination, which may result in the recruitment of the affected gene to a different subnuclear environment (Bolland et al. 2004, 2007; Ronai et al. 2007). Furthermore, antisense transcripts are also important for genomic imprinting, alternative splicing and nuclear transport or retention (Faghihi and Wahlestedt 2009). Finally, one of the most intriguing discoveries regarding antisense transcription is the finding in the fission yeast, Schizosaccharomyces pombe. Here, double-stranded RNA targets heterochromatinization of the affected genomic region via the activity of the RNA-induced transcriptional gene silencing complex complex with an RNA-interference induced mechanism that involves the DICER enzyme (Fire et al. 1998; Caplen et al. 2001; Elbashir et al. 2001; Reinhart and Bartel 2002; Volpe et al. 2002; Djupedal and Ekwall 2009). Intense efforts are underway to discover whether a similar mechanism occurs in higher-eukaryotes; however, the results so far are inconclusive (Djupedal and Ekwall 2009).

Literature suggesting the presence of a link between heterochromatin and antisense transcription inspired De Biase et al. (2009) and they identified an FXN antisense transcript called ‘FXN antisense transcript 1′ (FAST1) which seems to be produced at higher levels in the fibroblasts obtained from FRDA patients compared to the levels obtained in healthy. Reverse transcription with strand specific primers revealed that FAST1 extends from the first exon and intron towards the promoter. However, it is as-yet unclear where this transcript initiates. Interestingly, siRNA knockdown of CTCF up-regulated FAST1 and reduced FXN sense transcription. As mentioned earlier, the question as to whether the depletion of CTCF actually provokes the spreading of heterochromatin arising from the (GAA)n repeats has not been addressed but was suggested as a potential mechanism. Accordingly, CTCF might prohibit the antisense transcription and therefore the spreading of heterochromatin. Importantly though, the tight link between antisense transcription and heterochromatin was predominantly shown in S. pombe, where heterochromatin is much simpler compared to mammals (Rando and Chang 2009; Beisel and Paro 2011). Although early reports suggest that the RNA interference pathway is indispensable for mammalian cellular lifespan and development (Bernstein et al. 2003; Fukagawa et al. 2004), DICER knock-out experiments did not result in significant changes in mammalian heterochromatin (Murchison et al. 2005; Wang et al. 2006; Ip et al. 2012). This implies that RNAi induced heterochromatinization mechanisms may not be fully conserved in mammals. Still, antisense transcription was linked to transcriptional silencing of oncogenes p15 (Yu et al. 2008) and p21 (Morris et al. 2008) as well as the progesterone receptor (Schwartz et al. 2008). Morris et al.(2008) and Schwartz et al. (2008) suggested a mechanism whereby Argonaute proteins recognise antisense transcripts in mammals.

Overall, it is tempting to think that (GAA)n induced heterochromatinization involves a mechanism where antisense transcription plays a key role; however, the elusive results so far obtained in mammals bring up the need for further characterisation of the link between FAST1 and the pathological heterochromatinization in FRDA. Moreover, the replication of data presented by De Biase et al. (2009) in different disease models (i.e. primary cells, induced pluripotent stem cells or mouse models) will strengthen this hypothesis further.

Discussion and future directions

Is FRDA truly an epigenetic disorder?

Although the hallmarks of heterochromatin are considered to be epigenetic in the case of FRDA, there are still no studies addressing directly how the (GAA)n induced heterochromatin is inherited by daughter cells. Inheritance of heterochromatic marks to newly formed cells could theoretically enhance the disease phenotype at least in early onset FRDA cases. Such epigenetic events may take place during DNA replication as reviewed by Probst et al. (2009). However, some patients may have disease modifier genes that protect against the (GAA)n induced heterochromatin. Intriguingly, French Acadian FRDA patients were reported to have a late-onset disease accompanied by mild symptoms even in the case of a long (GAA)n repeat (Montermini et al. 1997a,b). One of the theoretical reasons for this observation could be that such epigenetic differences act either in cis or trans. Therefore, it would be interesting to study DNA methylation levels and histone modifications in the Acadian population.

Is there PEV-like stochastic silencing in FRDA?

Another interesting phenomenon remaining to be investigated is whether PEV-like stochastic effects take place in the silencing of FXN. Unpublished RNA-FISH results from our group suggest that some FRDA cells are expressing FXN transcripts whereas others are not; this is reminiscent of PEV as exemplified by the hCD2 gene. Thus, patient cells have a higher proportion of non-expressing cells than healthy controls (N. Chapman-Rothe, Ph.D. Thesis, 2008). The dynamics of FXN silencing would therefore be valuable to study as the transcriptional silencing may take place temporarily at a single cell level – e.g. occur in bursts. However, all published studies on FXN silencing have been performed on a population of cells and therefore results were attributed to an average outcome. Interestingly, Facioscapulohumeral muscular dystrophy is one example of a disease caused by a so-called ‘position effect’. In this case, a repeating unit consisting of the silenced gene Double homeobox 4 (DUX4) is truncated by a chromosomal translocation leading to its de-repression and the de-repression of neighbouring genes both of which are implicated in the pathogenesis of the disease (van der Maarel et al. 2011; Cabianca et al. 2012). In the normal individual, the gene array appears to be silenced by the polycomb system (Cabianca et al. 2012). In patients, DUX4 mRNA is detectable but at very low levels and around 1 per 1000 Facioscapulohumeral muscular dystrophy myoblasts express DUX4 in culture (Snider et al. 2010); raising the intriguing possibility of aberrant stochastic expression as seen in PEV.

How exactly is heterochromatin triggered in FRDA?

FRDA is one of several diseases caused by triplet repeats located in non-coding regions. One important example of these tri-nucleotide repeat diseases is the neuropsychiatric disorder Fragile X syndrome, where a CGG•CCG tract resides within the 5′ UTR of the Fragile X mental retardation 1 (FMR1) gene (Pieretti et al. 1991; D'Hulst and Kooy 2009). DNA methylation, heterochromatic histone marks (e.g. H3K9 methylation) and antisense transcription were also reported to be raised on the FMR1 gene because of the presence of an expanded (CGG•CCG)n tract (Pietrobono et al. 2005; Ladd et al. 2007; Naumann et al. 2009; Kumari and Usdin 2010). Similarly, Myotonic Dystrophy 1 (DM1) is known to be triggered by expanded (CTG•CAG)n repeats located within the 3′UTR of the Dystrophia myotonica-protein kinase (DMPK) gene (Buxton et al. 1992; Mahadevan et al. 1992; Cho and Tapscott 2007). DM1 was also linked to the abnormal heterochromatinization of DMPK accompanied by DNA and H3K9 methylation as well as antisense transcription (Filippova et al. 2001; Cho et al. 2005; Lopez Castel et al. 2011).

Neither the exact mechanism behind triplet DNA induced heterochromatin nor the cascade of events during its formation have been fully elucidated yet within the context of triplet repeat diseases. Experiments with human embryonic stem cells concluded that heterochromatic histone modifications take place before the DNA methylation in the case of Fragile X syndrome (Eiges et al. 2007). However, there are no direct experiments addressing this issue in FRDA so far. Still, CpG methylation seems likely to happen secondarily to heterochromatin because the (GAA)n expansion mutation does not introduce more CpG residues to the FXN locus. One interesting observation on the silenced FXN locus is the dual presence of both high levels of H3K9 and H3K27 methylation. As previously explained, H3K9 methylation is associated with HP1 mediated silencing of highly heterochromatinised satellite repeats whereas H3K27 is linked to polycomb-mediated silencing of formerly euchromatic genes. Typically, these two marks do not overlap significantly in the mammalian genome (Barski et al. 2007; Wang et al. 2008). Nevertheless, H3K27me3 could partially substitute for H3K9me3 in mammals when the H3K9 specific SUV39H was knocked out (Peters et al. 2003). Whether there is redundancy or cooperation between these two heterochromatic marks on the silenced FXN locus is therefore an intriguing question.

It is also not known which heterochromatin-inducing enzymes are primarily responsible for the histone modifications observed in FRDA. Undoubtedly, revealing the key chromatin modifiers responsible for the silencing of FXN will increase the chances of developing rational therapies. As explained above, there are various enzymes known to catalyse heterochromatic marks in a cell nucleus. It is often difficult to distinguish which enzyme is responsible for a given gene locus by inhibiting or knocking-down a particular histone modifying enzyme, because of the high potential of non-specific effects. So far, chemical inhibition of H3K9 histone methyltransferase G9a was shown to decrease H3K9 methylation at the FXN locus but failed to up-regulate FXN to a significantly high level (Punga and Buhler 2010). This result suggests that H3K27me3 may be capable of efficient silencing at the FXN locus although the authors did not address this.

Because heterochromatinization of FXN takes place because of a microsatellite tract, it would be interesting to investigate the link between the silenced FXN and SUV39H; the crucial enzyme that catalyses H3K9 methylation on satellite repeats (Rea et al. 2000; Bannister et al. 2001; Lachner et al. 2001). Apart from this, other histone lysine methyltransferases (e.g. SETDB1, EZH2/PRC2) as well as histone deacetylases would be of interest in understanding the process of heterochromatinization. One interesting study revealed HDAC3 as a key enzyme for FXN silencing as its chemical inhibitors seem to activate the silenced FXN gene in FRDA lymphoblastoid cell lines (Xu et al. 2009). Although the availability of ChIP-grade antibodies is limited, performing ChIP against histone modifying enzymes (including G9a, SUV39H, HDAC3 and maybe others) will allow a more extensive analysis of the FXN locus and has the potential to identify specific factors.

As explained in this review, two models have been proposed for the silencing of FXN in FRDA. One is the ‘transcriptional blockage effect’ created by expanded (GAA)n repeats, which induce triplex DNA structures and R-loops. The other is the heterochromatin effect caused by expanded repeats. These models are not necessarily mutually exclusive. One may hypothesise that heterochromatin is triggered by as-yet-unrevealed factors which recognise the triplex structures responsible for blocking the elongation of transcription (Fig. 5a). So far, very little is known about how heterochromatin is formed de novo in a cell nucleus. Thus, the understanding of how expanded (GAA)n repeats induce heterochromatinization is mostly speculative. It is known that the mismatch repair machinery (MSH2/MSH3 dimers) recognises these triplets. Interestingly, some literature from lower eukaryotes as well as mammalian cells suggest that heterochromatin is formed right after DNA damage takes place as a protective response against the production of faulty transcripts (Ayoub et al. 2008; Goodarzi et al. 2008; Luijsterburg et al. 2009; Peng and Karpen 2009; Sun et al. 2009; Zarebski et al. 2009). These studies underlined increased H3K9 methylation and/or HP1 binding on the damaged site. In terms of de novo formation of heterochromatin, HP1α was reported to be recruited directly by its interaction with the pericentromeric repeat transcripts (Maison et al. 2011). Another explanation for the heterochromatin-inducing effect of repetitive DNA could be the strong preferential nucleosome affinity because of reduced DNA flexibility and curvature, as shown for (CTG)n repeats (Wang et al. 1994; Wang and Griffith 1995). Finally, antisense transcripts were implicated in the formation of heterochromatin although the link seems to be elusive in mammals as discussed earlier. Whether the antisense transcription of the FXN gene is a cause or a result of (GAA)n mediated heterochromatin, is also not known yet. Overall, one may expect that some of the mechanisms described here may be involved in the initial steps of heterochromatinization of the FXN gene in FRDA.

Figure 5.

A hypothetical model for the heterochromatinization of Frataxin (FXN) in Friedreich's ataxia. This non-scaled schematic model summarises up-to-date findings and ideas about how FXN might be silenced in disease. (a) Early events that may give rise to heterochromatinization. (GAA)n repeats (shown in red) are known to cause non-B conformations of DNA (e.g. hairpin, triplex) as well as RNA●DNA hybrids known as R-loops. Such unusual conformations may stall on-going transcription by a physical blockage effect. Moreover, such structures are recognised by the cell's mismatch repair mechanism that involves MSH2/MSH3 protein dimers. Also, it is hypothesised that there could be other unknown proteins which may potentially bind to the (GAA)n tract directly. Another interesting finding is the detection of antisense transcription, which seems to be elevated at the FXN locus in Friedreich's ataxia (FRDA). However, whether antisense transcripts recruit the RNAi machinery, is unknown. (b) Invasion of heterochromatin factors into the FXN locus. Although there is no direct evidence yet, unusual DNA conformations recognised by mismatch repair enzymes or yet-unknown proteins and/or antisense transcription are thought to trigger the recruitment of heterochromatic factors. Which chromatin modifiers are the primary drivers for the silencing of FXN, is yet to be determined. Modifiers related to histone deacetylation, H3K9 methylation (Drosophila suppressor of variegation 3-9 (SUV39H, human homologue), G9 and HP1). H3K27 methylation (Polycomb group) and DNA methylation (DNMTs) could be responsible for the silencing of FXN and therefore must be investigated further at the FXN locus. The co-operation between such chromatin modifiers are thought to be essential for the spreading of heterochromatin from the expanded (GAA)n repeats. (c) The binding of the chromatin insulator CCCTC-binding factor (CTCF) close to the promoter may prevent the spreading of heterochromatin towards the promoter as well as antisense transcription. Importantly, CTCF binding was reported to be decreased in FRDA cells (De Biase et al. 2009). The exact function of CTCF at the FXN locus is still under debate and needs to be studied further.

Is there an insulator element in the FXN gene?

Heterochromatic factors are known to interact with each other and this may facilitate the spreading of heterochromatin at the pathological FXN locus (Figs 4 and 5b). Whether an insulator somehow regulates this spreading effect still remains as an interesting question. The CTCF binding site identified by De Biase et al. (2009) seems to be an important regulatory element and it is tempting to hypothesise that this element functions as an insulator against (GAA)n induced heterochromatin (Fig. 5c). It is also important to note that CTCF is not only involved in the insulation of heterochromatin. Other well-described roles of CTCF include enhancer blocking (Chung et al. 1993, 1997), DNA loop formation (Splinter et al. 2006) and the activation of transcription by interacting with RNAPII at active gene promoters (Chernukhin et al. 2007; Bartkuhn et al. 2009). Therefore, these phenomena may be pertinent to CTCF-mediated regulation of FXN.

In addition to FRDA, other disease genes associated with trinucleotide repeats were also reported to be subject to CTCF mediated gene regulation. Both FMR1 (Fragile X syndrome) and DMPK (Myotonic Dystrophy) genes were found to have binding sites for CTCF flanking the trinucleotide tracts (Filippova et al. 2001; Ladd et al. 2007). The role of CTCF on the FMR1 gene has not yet been revealed although an enhancer blocking activity was described in the case of the DMPK albeit in vitro.

Towards a more radical therapy for FRDA

Understanding the molecular pathology of Friedreich's ataxia in a deeper context already opened new avenues for the treatment of this as-yet-incurable condition. While the previous therapeutic approaches for FRDA mostly address the consequences of Frataxin deficiency, a more radical strategy is emerging to antagonise the aberrant silencing of the FXN gene. One of the encouraging outcomes is the development of HDAC inhibitors for creating a more accessible chromatin by reducing histone deacetylation and thereby subsequent methylation (Festenstein 2006; Herman et al. 2006; Gottesfeld 2007; Rai et al. 2010; Sandi et al. 2011). Results so far seem promising in terms of up-regulating FXN in FRDA (Gottesfeld et al. 2013). In addition to commercially generated HDAC inhibitors, a more recent study reports that Nicotinamide (Vitamin B3; sirtuin HDAC inhibitor) has a potential therapeutic effect in up-regulating FXN in FRDA (Chan et al. 2013). Undoubtedly, identification of key enzymes/proteins involved in the pathological silencing of FXN will lead scientists to develop more efficient target-specific therapeutics in future.


The authors thank the members of the Gene Control Mechanisms and Disease group for helpful discussions. This study was funded by Imperial College, Ataxia UK, EFACTS (FP7) and MRC UK.

Conflicts of interest

The authors have no conflicts of interest to declare.