FAST-1 antisense RNA epigenetically alters FXN expression

Friedreich ataxia (FRDA) is a multisystem genetic disorder caused by GAA repeat expansion mutations within the FXN gene, resulting in heterochromatin formation and deficiency of frataxin protein. Elevated levels of the FXN antisense transcript (FAST-1) have previously been detected in FRDA. To investigate the effects of FAST-1 on the FXN gene expression, we first stably overexpressed FAST-1 in non-FRDA cell lines and then we knocked down FAST-1 in FRDA fibroblast cells. We observed decreased FXN expression in each FAST-1 overexpressing cell type compared to control cells. We also found that FAST-1 overexpression is associated with both CCCTC-Binding Factor (CTCF) depletion and heterochromatin formation at the 5′UTR of the FXN gene. We further showed that knocking down FAST-1 in FRDA fibroblast cells significantly increased FXN expression. Our results indicate that FAST-1 can act in trans in a similar manner to the cis-acting FAST-1 overexpression that has previously been identified in FRDA fibroblasts. The effects of stably transfected FAST-1 expression on CTCF occupancy and heterochromatin formation at the FXN locus suggest a direct role for FAST-1 in the FRDA molecular disease mechanism. Our findings also support the hypothesis that inhibition of FAST-1 may be a potential approach for FRDA therapy.

(FXN-TSS) and silences the promotor activity 6 . In addition, severe depletion of the chromatin insulator protein CTCF has been identified at the 5′ untranslated region (UTR) of the FXN gene in FRDA patients. An antisense transcript named FAST-1 (FXN Antisense Transcript -1), whose sequence overlaps with the CTCF binding site, has also been discovered. FAST-1 expression is significantly increased in FRDA and is associated with the severe CTCF depletion and heterochromatin formation in the 5′UTR of the FXN gene 14,19 . Natural antisense transcripts (NATs) have long been described as 'junk DNA' or transcriptional noise due to their low expression and unknown function. However, in recent years, antisense transcripts have emerged as key regulators of gene expression in an epigenetic manner [20][21][22][23] . Literature supporting the notion that antisense transcripts are involved in heterochromatin formation and the regulation of their partner mRNA expression inspired us to further investigate the characteristics of FAST-1. We first identified a full-length FAST-1 transcript with a total length of 523 bp in size containing a poly (A) tail. Mapping the 3′ and 5′ ends of the FAST-1 transcript onto the genome showed that FAST-1 transcription overlaps with the FXN-TSS and CTCF binding site in the 5′UTR of the FXN gene. Therefore, we decided to investigate potential effects of altered FAST-1 expression on FXN expression in three different types of cell lines. We report that FAST-1 overexpression is consistently associated with reduced CTCF occupancy, heterochromatin formation and decreased FXN expression. We also show that knocking down FAST-1 expression results in increased FXN expression in FRDA fibroblast cells, thereby revealing FAST-1 to be a potential FRDA therapeutic target.

Results
Identification of FAST-1 by rapid amplification of cDNA ends. To determine the exact size and location of FAST-1, 5′ and 3′ rapid amplification of cDNA ends (RACE) experiments were performed with total RNA and poly(A) + RNA. Mapping the 5′ and 3′ ends of FAST-1 transcript onto the genome precisely localised them to nucleotides + 164 and −359 of the FXN gene, respectively, and the total length of FAST-1 was found to be 523 bp in size. A poly (A) signal was also identified in the FAST-1 sequence at FXN nucleotide positions −283 to −288 ( Fig. 1).

Stable overexpression of FAST-1 in non-FRDA cells.
To assess the effect of FAST-1 on FXN expression, we generated three non-FRDA cell lines -HeLa, HEK293 and fibroblast -that stably overexpress FAST-1 cDNA under the control of a CMV promoter. Twelve stable FAST-1 overexpressing HeLa cell lines, twelve stable FAST-1 overexpressing HEK293 cell lines and three stable FAST-1 overexpressing fibroblast cell lines were developed. To confirm FAST-1 overexpression in FAST-1 transfected cells, qRT-PCR measurements were performed, using primers designed to detect human-specific frataxin antisense transcript, FAST-1. Analysis of qRT-PCR measurement confirmed a very significant increase of FAST-1 expression in transfected HeLa cells (396%, P < 0.001), transfected HEK293 cells (660%, P < 0.05) and transfected fibroblast cells (332%, P < 0.01) compared with non-transfected control cells (Fig. 2). No such differences were detected with the empty vector-transfected cells.
Overexpression of FAST-1 reduces the FXN mRNA and protein expression. Analysis of FAST-1 overexpressing cells revealed a very significant reduction of FXN mRNA in the FAST-1 overexpressing HeLa cells (45%, P < 0.05), HEK293 cells (47%, P < 0.001) and fibroblast cells (42%, P < 0.001) compared with non-transfected control cells. There was no significant difference detected between the empty vector-transfected and non-transfected control cells (Fig. 3a).
To determine the levels of human frataxin expression in the FAST-1 overexpressing cells, frataxin protein expression levels were measured by lateral flow immunoassay with the Frataxin Protein Quantity Dipstick assay kit. Analysis of FAST-1 overexpressing cells revealed that frataxin expression levels were significantly decreased to (40%, P < 0.01) in the FAST-1 overexpressing HeLa cells, HEK293 cells (55%, P < 0.05) and fibroblast cells (58%, P < 0.001) compared with non-transfected control cells (Fig. 3b). No such differences were detected with the empty vector-transfected cells.
FAST-1 copy number associated gene expression changes. The FAST-1 copy number was investigated in FAST-1 overexpressing HeLa cell genomic DNA samples using TaqMan real-time PCR, compared to non-transfected HeLa cells that contain two copies of the FXN locus. To assess the reliability of each copy number call, the results were analysed by Applied Biosystems CopyCaller ™ Software v.2.0. Seven out of twelve FAST-1 overexpressing HeLa cell lines, together with the endogenous non-transfected HeLa cell control, had acceptable confidence values and absolute zero score to pass the test. The results indicated that FAST-1 copy number is positively correlated with increased FAST-1 expression (R = 0.7, P < 0.05) (Fig. 4a), which in turn is negatively correlated with FXN expression by the analysis of eleven transfected HeLa cell lines and the endogenous non-transfected HeLa cell control (R = −0.58, P < 0.05) (Fig. 4b). This suggests a concentration-dependent molecular mechanism of action for FAST-1, rather than a potential threshold effect.  (Fig. 5a). We also investigated acetylated and trimethylated histone H3K9 modifications by ChIP analysis at the FXN 5′UTR region. The ChIP results were normalized to the 'input' , and negative control antibody values were taken into account. ChIP assay results showed significantly decreased acetylation of H3K9 (Fig. 5b) and enrichment of H3K9me3 (Fig. 5c) at the FXN 5′UTR region in FAST-1 overexpressing HeLa cells compared to non-transfected control cells. There were no statistically significant differences in H3K9 acetylation and trimethylation levels between the empty vector-transfected and non-transfected control cells.

FAST-1 overexpressing HeLa cells show reduced occupancy of CTCF and marks of heterochro-
FXN gene DNA methylation levels are altered in FAST-1 overexpressing fibroblast cells. It has been reported that CpG sites, designated U6 and U11, at the upstream GAA repeat region show elevated levels of DNA methylation in FRDA patients compared to controls 11,19 . Therefore, we chose to investigate the DNA    methylation status at CpG U6 and U11 in two FAST-1 overexpressing fibroblast cell lines (FAST-1B, FAST-1C), together with FRDA fibroblasts and unaffected control fibroblasts. Results from MethylScreen assays revealed a non-significant small increase in DNA methylation at CpG site U6, with DM values increasing from 14% in unaffected control fibroblasts to 18% and 23% in FAST-1 B and FAST-1 C, respectively ( Fig. 6a). At CpG site U11, the proportion of densely methylated templates in FAST-1 overexpressing fibroblasts significantly increased from 8% in unaffected controls to 79% (P < 0.01) and 76% (P < 0.001) in FAST-1 B and FAST-1 C, respectively, similar to the levels found in FRDA fibroblasts (Fig. 6b). Restoring aconitase activity following knockdown of FAST-1. As a biomarker of frataxin function within cells, aconitase activity was measured in FAST-1 knockdown fibroblast cells 1 month after treatment with LV FAST-1 using the Aconitase Assay Kit (Cayman). Normal FRDA fibroblast cells (GMO7492) and FRDA cells  treated with scrambled shRNA were used as the controls. All data were calibrated to the mean of the aconitase activity in normal fibroblast cells, which was set to 100%. The aconitase activity in FRDA cells was determined to be 49% (P < 0.05) compared to normal control fibroblast cell levels. However, after knocking down FAST-1 in FRDA fibroblast cells, aconitase activity significantly increased to a value of 195% (P < 0.01). No significant difference was detected in FRDA fibroblast cells treated with scrambled shRNA (Fig. 9).

Discussion
Although it is more than 20 years since the identification of the genetic mutation underlying FRDA, the exact mechanisms by which GAA repeat expansions lead to FXN gene silencing are still not fully understood. However, several potential mechanisms have been proposed based on experimental evidence. For example, it has been suggested that abnormal triplex conformation, such as sticky DNA adopted by GAA repeat expansion and/or RNA•DNA hybrids, might impede the transcription of the FXN gene 24,25 . In addition, several previous studies of FRDA mouse models, human tissues and cells have highlighted the possibility of epigenetic changes being also either directly or indirectly involved in FXN gene silencing 12 . Significant enrichment of the repressive histone modifications at the 5′UTR of the FXN gene, together with elevated levels of the non-coding RNA (FAST-1) and hypermethylation of CpG regions upstream of the GAA repeat are the epigenetic changes that have been identified in FRDA 10,11,26 . Enrichment of these repressive chromatin marks in the FRDA alleles provides evidence of the heterochromatin formation in the promoter region, involving the critical +1 nucleosome and/or the first intron of the FXN gene. Bidichandani and colleagues showed that in FRDA patient-derived fibroblasts, the FXN Antisense Transcript (FAST-1) was expressed at higher levels. Previous in vitro studies in our lab also detected high levels of FAST-1 in the YG8R and YG22R-derived fibroblast cells 27 . In addition, it has been reported that higher levels of FAST-1 were associated with the severe CTCF depletion and coincidentally heterochromatin formation in the 5′UTR of the FXN gene 14 . It is important to note that FRDA cells do not have a generalized defect in CTCF binding 14 . Our results from overexpressing FAST-1 in three different non-FRDA cell lines now demonstrate that FAST-1 can act in trans in a similar manner to the cis-acting FAST-1 overexpression that has previously been identified in FRDA fibroblasts. This mechanism of action, where an antisense RNA acts both in cis and trans and induces epigenetic silencing of its partner sense gene, has been reported for the p15 and huntingtin (HTT) naturally occurring antisense transcripts 21,28 . We also observed that FAST-1 overexpressing HeLa cells have a reduced occupancy of CTCF, a decreased level of H3K9ac and increased levels of H3K9me3 at the 5′UTR of the  Thus, it is plausible that FAST-1 overexpression may displace CTCF from the 5′UTR of the FXN gene, which would result in secondary changes in chromatin structure that are incompatible with CTCF binding to DNA (Fig. 10) 29,30 . It is well established that it is the intrinsic property of the expanded GAA repeat that causes heterochromatin formation and exert its gene silencing effect 9 . It is not clear yet how GAA repeat expansion results in CTCF dislodgement. The expanded GAA repeat drives chromatin changes in intron 1 and this repeat-proximal heterochromatin spreads from the expanded GAA in intron 1 to the upstream regions of the FXN gene encompassing the FXN promoter and CTCF binding site 6,10 . One possibility is that the heterochromatin emanating from the expanded GAA repeat displaces CTCF from its binding site. Considering the reduced occupancy of CTCF, H3K9me3 enrichment and decreased H3K9ac levels, all hallmarks of heterochromatin formation, at the 5′UTR of the FXN gene in FAST-1 overexpressing HeLa cells, it can be speculated that FAST-1 RNA, via subsequent heterochromatin, may indirectly displace CTCF. Indeed, several studies have reported that association of H3K9me3 with loss of CTCF binding is a common event in epigenetic silencing of cancer-related genes such as p16 and p53 31,32 . When the chromosomal boundaries are destabilised through dissociation of CTCF and long-range epigenetic organisations are lost, repressive chromatin can spread passively into the promoter region of the FXN gene and inhibit its transcription 14,31 . In addition, given the enrichment of CTCF at the border of lamina-associated domains (LADs), loss of CTCF can affect the position of genomic loci relative to the nuclear lamina (NL), a location with a generally repressive environment 33,34 . With this in mind, it has been reported that the majority of expanded FXN alleles are positioned at the NL. Thus, CTCF loss and subsequent heterochromatin formation may contribute to FXN relocation to the NL and its repression 35 . Recent studies suggest that CTCF has a contradictory function to its classical enhancer-blocker function. It has been proposed that CTCF can tether distant chromatin sites together, create a loop and physically bridge enhancer-promoter interaction 36,37 . In addition, it has been demonstrated that gene looping plays an important role in restricting divergent transcription of non-coding RNA. It can be speculated that CTCF may facilitate enhancer-promoter interactions at the 5′UTR of the FXN gene to drive the FXN expression. FAST-1 overexpression and resultant local depletion of CTCF could abolish this CTCF-mediated enhancer-promoter interaction and consequently reduce FXN gene expression. Non-coding RNAs have also been found to interact with transcription factors and regulate local gene expression. Two transcription factors, SRF and TFAP2, have been found to regulate expression of the FXN gene. They bind directly to a region that is located close to the start of FAST-1 antisense transcript 38 . Therefore, it is possible that FAST-1 acts as a decoy, reducing the availability of these transcription factors required for the FXN gene transcription.
Antisense transcription is a very common phenomenon in mammalian transcriptomes. Although NATs can show mRNA-like characteristics such as having a poly A tail and a 5′ cap structure, the function of these transcripts is not well understood 39 . Experimental evidence suggests that non-coding RNA can influence gene expression by recruiting chromatin remodelling complexes to specific alleles and mediating heterochromatin formation and gene silencing. It has been shown that the polycomb repressive complex 2 (PRC2) associates with almost 3000 NATs in mouse embryonic stem cells 40,41 . Indeed, antisense transcripts can act as scaffolds or directly recruit PRC2 to chromatin in cis and trans manners to induce trimethylation of H3K27 and consequently establish a silent chromatin state. According to this mechanism, even a relatively low abundance of antisense transcript can mediate transcriptional repression 42 . PRC2 has also been reported to bind to ncRNA molecules in a size-dependent manner, predominantly around TSSs to maintain the repressed chromatin state 43,44 . In addition, recent studies suggest that there is a cooperative mechanism between H3K27 and H3K9 methylation marks, where H3K9me3 can crosstalk with the Polycomb H3K27me3 modification to cooperate in gene silencing. Indeed, H3K27me3-bound PRC2 stabilizes H3K9me3-anchored HP1 and reinforces heterochromatin formation and ensuing gene repression. Significant enrichment of H3K9 and H3K27 methylation and HP1 has been reported at the FXN locus of FRDA-derived fibroblast cells. Here in FAST-1 overexpressing HeLa cells, we also identified significant enrichment of H3K9me3 at the 5′UTR of the FXN gene. Moreover, knowing that genomic repeats are targets of Polycomb complexes and that repeat sequences might provide a binding platform for polycomb group proteins, it is conceivable to speculate that the expanded GAA repeat and high levels of FAST-1 either directly or indirectly recruit PRC2, and through a cooperative mechanism dictate heterochromatin formation and FXN gene silencing in FRDA 19,45 . Alternatively, overexpression of FAST-1 could be leading to the generation of endogenous short RNAs (endo-siRNAs) thus leading to promotion of the RNA interference pathway 46 .
Our results show that knocking down FAST-1 in FRDA fibroblast cells increases FXN gene expression (Fig. 10). Therefore, it can be concluded that, since FAST-1 is associated with epigenetic repression of the FXN gene, inhibition of FAST-1 may be an approach to increase the FXN transcripts and stimulate subsequent protein expression. Indeed, our results demonstrate that knocking down FAST-1 in FRDA results in a significant increase in aconitase enzyme activity, a good indicator of frataxin function within cells. Our data suggest that since FAST-1 is associated with FXN gene silencing, inhibition of FAST-1 may be an approach for FRDA therapy. Considering the nature of NATs and the fact that many currently available drugs would not affect the activity of non-coding RNA molecules, developing new methods to disrupt the function of NATs seems necessary.

Characterization of antisense FAST-1 transcripts by rapid amplification of cDNA ends (RACE)
. RACE experiments were done with DNase I-treated RNA from human FRDA fibroblast cell line (GM04078); 5′-and 3′-RACE were carried out by using the SMARTerTM RACE cDNA amplification kit (Clontech). The gene-specific primers used in each case are: Following the primary FAST-1 RACE PCR, nested PCRs were performed using the NUP primer (nested universal primer A, supplied with the SMARTer RACE cDNA amplification kit) and a gene specific primer: RACE-N-FAST-R2 5′-GACAGCAGCTCCCAAGTTCCTCCTG-3′ for the 5′nested PCR. RACE-N-FAST-F1 5′-GACCCAAGGGAGACTGCAGCCTGGTG-3′ for the 3′nested PCR.
The PCR products were gel purified and cloned for the sequencing analysis.
FAST-1 overexpression. Initially, a full-length FAST-1 sequence (523 bp) was cloned into the expression plasmid pcDNA3. One day before transfection, HeLa, HEK293 and fibroblast cells were seeded in 6 well plates and grown in DMEM containing 10% FBS and antibiotics (penicillin/streptomycin). Two micrograms each of pcDNA3 empty vector and pcDNA3-FAST-1 (Addgene) were transfected into cells by using Lipofectamine 3000 (Life Technologies) and Neon ™ transfection system. After 48 h, 200, 400 and 100 μg/ml of G418 (Gibco) were added to the medium of HeLa, HEK293 and fibroblast cells, respectively, to select for cells carrying the plasmid. After a further 16 days G418-resistant colonies were picked for individual culture, and were maintained in complete medium (DMEM medium containing 10% fetal calf serum) with appropriate G418 concentration.
MethylScreen assay. DNA methylation analysis was performed using the 'MethylScreen' method 19 , which uses combined restriction digestion of DNA with methylation sensitive and methylation dependent restriction enzymes, MSRE and MDRE respectively. MethylScreen was used to analyse the two CpG sites, CpG6 and CpG11 11 , at FXN locus upstream of the GAA repeat. 1 µg of genomic DNA was digested with: (1) a MSRE, (2) MDRE, (3) both MSRE and MDRE (double digest, DD), and (4) neither MSRE or MDRE (mock control). The MSREs used for CpGs 6 and 11 were AjiI (Fermentas) and Hpy188III (New England Biolabs), respectively 38 . The MDRE used for all two CpGs was McrBc (Fermentas). A 50 ng aliquot of digested DNA was then amplified by quantitative PCR using SYBR ® Green (Applied Biosystems) and an ABI 7900 Fast Real-Time PCR System (Applied Biosystems) with the following primers: CpG6 F 5′-GAAGATGCCAAGGAAGTGGTAG-3′ and R 5′-GAGCAACACAAATATGGCTTGG-3′; CpG11 F 5′-GATCCGTCTGGGCAAAGGCCAG-3′ and R 5′-ATCCCAAAGTTTCTTCAAACACAATG-3′. PCR quantification was carried out using the ΔCt method (values were calculated as 2 ΔCt (mock -digest) with the mock value set at 100%) and RQ Manager Software Knocking down FAST-1 expression. To knock down FAST-1 expression, Sigma custom cloning MISSION shRNA team designed a short hairpin RNA, cloned it into a pLKO.1-puro-CMV-tGFP vector and provided us with ready to use lentiviral particles for transduction. To achieve a stable FAST-1 knockdown, FRDA and normal fibroblast cell lines were transduced with lentiviral (LV) vectors expressing shRNAs targeting FAST-1 (pLKO.1-puro-CMV-tGFP). Several precautions were taken in order to control off-target effects and specificity, as follows: (i) pLKO.1-puro Non-Target shRNA was used as the negative control; (ii) The normal fibroblast cell line was also transduced with LV FAST-1 to assess the specificity of any FAST-1 knockdown effect on FRDA cells. Briefly, cells were seeded into 6-well plates with culture medium containing 10% FBS until 80% confluence, and the cells were then incubated with FAST-1 and scrambled LV particles (25ul-100ul of 5.7 × 10 7 to achieve MOI values of [5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20] in serum-free culture medium supplemented with polybrene (8 μg/ml) for 6 h. Then the supernatant was replaced with 1 ml fresh medium and incubated at 37 °C and 5% CO 2 for 72 hours to allow time for the expression of the reporter and puromycin resistance genes. After 72 hours, the selective antibiotic, puromycin, was added to the media at a final concentration of 0.2 μg/μl. The selection media was replaced every 3 days and the final cell harvest for RNA analysis was performed at 2 weeks post-transduction.
Aconitase assay. Aconitase activities were determined using the Aconitase Assay Kit (Cayman Chemical Company, 705502). To perform the assays, cultured cells were washed with cold PBS, scraped off the flask and pelleted by centrifugation at 800 g for 10 min. Cell pellets were suspended in cell lytic buffer (Cayman) for 15 min at 4 °C, then centrifuged at 10,000 g for 10 min, followed by BCA protein concentration determination and dilution of cell protein lysates to 0.5ug/µl in 1x assay buffer (Cayman). 50 µl of 0.5 ug/µl cell protein lysates were added to 200 µl of substrate mix (50 mM Tris/HCl pH 7.4, 0.4 mM NADP, 5 mM Na citrate, 0.6 mM MgCl 2 , 0.1% (v/v) Triton X-100 and 1 U isocitrate dehydrogenase) and the reactions were incubated at 37 °C for 15 min, followed by spectrophotometric 340 nm absorbance measurements every minute for 15 min at 37 °C to determine the reaction slope. Aconitase activities of samples were then normalized to citrate synthase activities, which were determined using a citrate synthase assay kit (Sigma, CS0720).
Statistical analysis. Statistical analyses that compared two groups of data were performed using the Student's t test, with a P value of <0.05 taken to indicate a statistically significant difference. Correlation and regression analyses of FAST-1 copy number, FAST-1 expression and FXN expression were performed using Microsoft Excel data analysis tools.