Expanded GAA repeats impede transcription elongation through the FXN gene and induce transcriptional silencing that is restricted to the FXN locus

Abstract

Friedreich's ataxia (FRDA) is a severe neurodegenerative disease caused by homozygous expansion of the guanine-adenine-adenine (GAA) repeats in intron 1 of the FXN gene leading to transcriptional repression of frataxin expression. Post-translational histone modifications that typify heterochromatin are enriched in the vicinity of the repeats, whereas active chromatin marks in this region are underrepresented in FRDA samples. Yet, the immediate effect of the expanded repeats on transcription progression through FXN and their long-range effect on the surrounding genomic context are two critical questions that remain unanswered in the molecular pathogenesis of FRDA. To address these questions, we conducted next-generation RNA sequencing of a large cohort of FRDA and control primary fibroblasts. This comprehensive analysis revealed that the GAA-induced silencing effect does not influence expression of neighboring genes upstream or downstream of FXN. Furthermore, no long-range silencing effects were detected across a large portion of chromosome 9. Additionally, results of chromatin immunoprecipitation studies confirmed that histone modifications associated with repressed transcription are confined to the FXN locus. Finally, deep sequencing of FXN pre-mRNA molecules revealed a pronounced defect in the transcription elongation rate in FRDA cells when compared with controls. These results indicate that approaches aimed to reactivate frataxin expression should simultaneously address deficits in transcription initiation and elongation at the FXN locus.

Introduction

Friedreich's ataxia (FRDA, FA, OMIM229300) is a severe neurodegenerative disease caused by transcriptional repression, which is induced by expanded guanine-adenine-adenine (GAA) repeats located in intron 1 of the FXN gene (1). The GAA repeat tract is polymorphic and in unaffected individuals it is typically shorter than 30 triplets whereas disease-causing expanded alleles can harbor up to 1500 GAA repeats (2,3). FRDA patients homozygous for GAA expansions have 5–30% of frataxin mRNA and protein when compared with healthy individuals. This deficiency of frataxin leads to a serious imbalance in mitochondrial metabolism, especially in the synthesis of iron-sulfur clusters—inorganic cofactors involved in many cellular pathways ranging from cellular respiration to DNA synthesis and repair (4–6). The clinical severity and the age of onset of FRDA correlate with the size of the expanded GAA tracts, and in particular, with the length of the shorter allele (7,8).

It has been demonstrated, using various model systems along with patients' autopsy samples, that expanded GAA repeats induce epigenetic silencing of the FXN gene (9–14). Post-translational histone modifications that typify heterochromatin (histone H3K9me2, H3K9me3) are enriched in the vicinity of the repeats, whereas active chromatin marks (histone H3K9ac, H3K14ac and H4K5ac) in this region are underrepresented in FRDA samples when compared to controls. Aberrant DNA methylation of the CpG island located in intron 1 of the FXN gene has also been identified in FRDA samples (12,15). Interestingly, the heterochromatinization of the FXN locus induced by GAA repeat expansion can be recapitulated in model systems by inserting long GAA tracts into reporter genes (10,16,17). Knock-in of a GAA repeat tract in the mouse FXN gene led to a decrease of FXN expression (18). It has been postulated that formation of non-B DNA or DNA/RNA hybrid structures may represent a signal that initiates chromatin modifications leading to FXN inactivation. Thus far, silencing triggers as well as the exact molecular mechanism of FXN silencing remain unknown (9).

Spreading of heterochromatin and transcriptional silencing has been observed in the proximity of the highly repetitive regions of the genome such as telomeres and centromeres (19–21). However, the extent of silencing induced by expanded GAAs beyond the direct vicinity of intron 1 has not been determined. Results indicating both silencing of the FXN promoter associated with a transcription initiation defect as well as studies demonstrating transcription elongation dysfunction have been reported (12,22–24). Moreover, recent phenotypic observations conducted mostly in FRDA patient-derived fibroblast and lymphoblast cell lines indicate the possibility of a long-range cis silencing mechanism, spanning a larger region of the FXN locus (25,26). The extent of GAA repeat-induced silencing is of particular importance considering two major therapeutic strategies for FRDA: reversal of FXN silencing and gene replacement therapy (27). In the case of extensive transcriptional silencing of a larger region of chromosome 9, restoring frataxin expression using gene therapy strategies would alleviate only the part of the disease phenotype arising solely from frataxin deficiency. On the other hand, strategies based on epigenetic reactivation of the FXN locus should result in complete reversal of FRDA phenotype. However, if the effect of GAA expansion is localized to the FXN gene, both therapeutic avenues should be equally efficacious.

Herein, we present the results of a comprehensive analysis of the transcription status and epigenetic environment of the FXN locus in a large set of 18 primary fibroblast cell lines derived from FRDA patients and 17 lines from unaffected individuals. Using next-generation RNA sequencing, we demonstrated a remarkable variability in FXN expression within FRDA as well as control sample groups. Additionally, we showed using RNA-seq and chromatin immunoprecipitation (ChIP) analyses that the epigenetic silencing effect induced by the expanded GAA repeats is confined to the FXN locus and does not affect expression of upstream or downstream neighboring genes. Finally, analysis of FXN pre-mRNA expression between FRDA and control samples combined with ChIP analyses revealed a pronounced transcription elongation defect at the expanded GAA region.

Results

Frataxin expression and GAA repeat length in FRDA and control fibroblasts

We first characterized 35 primary FRDA and control fibroblast lines. Sixteen FRDA lines were obtained from skin biopsies, as described in Materials and Methods. The remaining 2 FRDA lines along with 17 control lines were acquired from Coriell Cell Repositories. The FRDA cohort included 7 females and 11 males with a mean age of onset 17.7 years (4–41) and mean age of sampling 36.3 years (13–70) (Table 1). Clinical data were available for 16/18 FRDA patients and 50% of them (8/16) developed cardiomyopathy and hearing loss, and diabetes was confirmed in one patient (Table 1). The control group included nine females and eight males with average age at sampling of 30.1 years (11–64). To minimize variability in our molecular analyses, all fibroblast lines were cultured simultaneously, in the same batch of media to the same cell density. Molecular characterization of FRDA and control fibroblasts included the following: determination of the GAA repeat size by PCR with two different primer sets, analysis of the GAA interruption status using MboII digestion which recognizes and cleaves GAAGA sequence (28) and quantitative analyses of FXN expression using quantitative real-time (qRT)-PCR and western blot (Fig. 1 and Supplementary Material, Figs S1–S3). Results of the GAA repeat PCR showed bands corresponding to two GAA alleles in most cases. However, somatic instability of the expanded GAAs was observed in a few fibroblast lines (e.g. sample 4259, Fig. 1). In such instances, the two longest PCR products were considered parental GAA alleles. The number of GAAs found in FRDA fibroblast samples varied between 110 and 1470 repeats with the average size of the shorter allele (GAA1) being 454 GAAs and longer (GAA2) 898 GAAs (Table 1). No GAA expansions were found in the control lines, thus excluding the possibility of asymptomatic FRDA carriers being included in the control cohort (Supplementary Material, Fig. S1). All PCR reactions were conducted under conditions allowing for simultaneous amplification of short and expanded GAAs (Fig. 1A, lane CR-carrier). The results of MboII digestion consistently showed two bands corresponding to 206- and 242-bp flanking sequences, which remain after complete digestion of the repeat region (Fig. 1B), thus indicating lack of complex interruption of the GAA tracts at the FXN gene in the fibroblast cell lines used in this study.

Table 1.

FRDA fibroblast cell lines used in this study

Cell line	Gender	Age of sampling	Age of onset	No. of GAA repeats allele 1, allele 2	FXN level versus average controlsa				Phenotype
					SYBR	TaqMan	RNA-seq	Frataxin by western blot	C	D	H
GM03665b	F	13	nd	816, 1410	0.11	0.15	0.33	0.17	nd	nd	nd
68	F	21	7	570, 1200	0.13	0.15	0.39	0.13	+	−	+
4230	F	28	6	870, 1470	0.15	0.16	0.26	0.17	+	+	+
4497	F	44	30	526, 826	0.22	0.23	0.47	0.20	+	−	+
88	F	50	16	422, 520	0.35	0.35	0.50	0.31	−	−	+
4627	F	50	22	468, 807	0.42	0.55	0.45	0.65	−	−	−
203	F	31	14	916, 1382	0.24	0.31	0.41	0.26	−	−	−
GM04078b	M	30	nd	341, 480	0.53	0.59	0.88	0.46	nd	nd	nd
4259	M	37	15	404, 920	0.24	0.26	0.53	0.16	+	−	+
4192	M	33	16	400, 967	0.15	0.20	0.49	0.40	−	−	−
4654	M	19	16	190, 500	0.35	0.40	0.62	0.62	+	−	+
188	M	47	11	490, 680	0.25	0.29	0.55	0.29	−	−	+
130	M	56	41	136, 540	1.52	1.16	0.78	0.55	+	−	−
156	M	41	15	495, 505	0.32	0.24	0.33	0.16	−	−	−
281	M	19	11	630, 806	0.28	0.28	0.33	0.28	+	−	+
4675	M	28	4	185, 1130	0.23	0.26	0.47	0.38	−	−	−
66	M	70	41	110, 590	0.93	0.94	1.01	0.87	−	−	−
4509	M	36	18	211, 1428	0.41	0.56	0.58	0.55	+	−	−

Eighteen FRDA fibroblast lines were analyzed.

C, clinically diagnosed cardiomyopathy; D, diabetes; H, hearing loss/deficit; nd, no data available.

^aControls are listed in Supplementary Material, Table S1.

^bFibroblast lines obtained from Coriell Cell Repositories.

Figure 1.

Characterization of the FRDA fibroblast lines. (A) Determination of the number of GAA repeats using PCR; CR-GAA repeat expansion carrier harboring one expanded and one short GAA allele. (B) Analysis of the GAA interruption status using MboII digestion. (C and D) Correlation between the length of the shorter of the two expanded alleles (GAA1) and frataxin deficiency. Relative FXN mRNA and protein levels determined in 18 fibroblast lines using qRT-PCR (TaqMan) (C) and western blot (D) are plotted against the length of GAA1. Expression of the FXN transcript and frataxin was calculated relative to the average expression of FXN in 17 control fibroblast lines listed in the Supplementary Material, Table S1. The correlation coefficient value (r) is indicated in each graph.

Analyses of FXN mRNA expression in FRDA samples using TaqMan qRT-PCR assays (Fig. 1C and D and Table 1) revealed a strong correlation (correlation coefficient r = −0.75) between the length of the shorter GAA1 allele and FXN mRNA levels (Fig. 1C). FXN mRNA expression as determined by qRT-PCR was normalized to the average FXN expression in all 17 control cell lines. As expected, correlation of FXN expression with the number of repeats in GAA2 was much weaker (r = −0.54 to −0.64; Supplementary Material, Fig. S2A and B). Similarly, steady-state levels of mature frataxin protein correlated very well with GAA1 size (r = −0.75) but not with GAA2 (r = −0.4) (Fig. 1D and Supplementary Material, Fig. 2C). Thus, this cohort of FRDA fibroblast cell lines demonstrated typical correlation of FXN expression as a function of GAA1 size (8,29).

RNA-seq analysis of FRDA and control fibroblasts

To determine the expression of FXN mRNA using an unbiased approach, we isolated RNA from all 35 fibroblast cell lines and performed high-throughput RNA sequencing. Importantly, total RNAs were purified using a ribosomal RNA depletion protocol to allow for analysis of both mature transcripts as well as unspliced pre-mRNAs, as described in Materials and Methods. Approximately 27–39 M pairs of reads were generated for each fibroblast RNA preparation and 82–93% of them were mapped on both ends to the human genome. The landscape profile of RNA-seq signal was uploaded into UCSC Genome Browser (GRCh37/hg19), and a representative snapshot was taken of ∼0.5 Mbp of chromosome 9 in the vicinity of the FXN locus (Fig. 2A). Differential expression analysis using R/Bioconductor package DESeq showed a highly significant (false discovery rate, FDR = 1.5⁻²²) decrease of FXN mRNA expression in FRDA fibroblasts when compared with the control cells (Fig. 2B and E). Interestingly, the expression level of FXN mRNA in FRDA samples showed high variability between lines, with some FRDA fibroblasts expressing frataxin transcript at levels similar to control cells (e.g. cell lines GM04078, 66 and 130, Table 1 and Fig. 2B). Notably, all of these lines harbor shorter GAA repeats (especially GAA1) and, in the case of FRDA 66 and 130, are associated with late onset of the disease. Although higher expression of the FXN gene was observed in proliferating fibroblasts, it is very likely that FXN silencing is significantly augmented in terminally differentiated disease-relevant cells (e.g. neurons or cardiomyocytes) in these patients. Additionally, somatic expansions of the GAAs (e.g. in dorsal root ganglia neurons) can increase the number of repeats in already expanded alleles leading to a progressive decrease of FXN levels and evoking disease symptoms (30). The RNA-seq data showed the highest correlation between FXN mRNA expression and the length of GAA1 (r = −0.8) when compared with qRT-PCR and western blot data (compare Figs 1C and D and 2C). Although overall much less significant, a modest correlation (r = −0.57) between FXN expression and the length of the GAA2 allele could also be detected (Supplementary Material, Fig. S2D).

Figure 2.

Transcriptional silencing induced by expanded GAA repeats is restricted to the FXN gene. (A) RNA-seq data from the analyses of two control and two FRDA fibroblast lines were aligned to GRCh37/hg19 and visualized in UCSC Genome Browser (http://genome.ucsc.edu). A representative snapshot was taken of ∼0.5 Mbp of chromosome 9 encompassing the PIP5K1B, FAM122A, FXN and TJP2 genes. No expression of the PIP5K1B and PRKACG mRNAs was detected in FRDA or control lines. (B) The expression level of FXN mRNA in 17 control (black bars) and 18 FRDA (white bars) fibroblast lines was determined using RNA-seq. Quantitative analysis was conducted using DESeq method. (C) Correlation between length of the GAA1 and FXN mRNA expression as determined using RNA-seq. (D, E, F) A cumulative analysis of FAM122 (D), FXN (E) and TPJ2 (F) gene expression in 17 control and 18 FRDA fibroblast lines. The FDR is shown.

Expanded GAA-induced FXN silencing is confined to the frataxin locus and does not affect expression of neighboring genes

To determine whether epigenetic silencing associated with the GAA repeat expansion in FRDA samples can spread beyond the FXN locus, we used RNA-seq data to assess mRNA expression of neighboring genes PIP5K1B, FAM122A, PRKACG and TJP2 (Fig. 2). Expression of TJP2, FXN and the FAM122A pseudogene was easily detectable in all control and FRDA samples; however, expression of PIP5K1B and PRKACG mRNA was not detected among a total of 16 531 genes expressed in human fibroblasts. To ensure that the RNA-seq reads were not misaligned to the homologous PIP5K1A mRNA, we conducted uniqueness and alignability analyses in UCSC Genome Browser (GRCh37/hg19) using DUKE Uniq 35 and CRG Align 100 tracks (31). The results clearly demonstrated that the entire PIP5K1B transcript can be unambiguously mapped (data not shown), thus confirming that expression of PIP5K1B mRNA is below the detection level of deep sequencing in human fibroblasts. On the other hand, in fibroblasts, the highly expressed FAM122A pseudogene located upstream of the FXN locus in an intronic region of PIP5K1B was transcribed at similar levels in both control and FRDA groups (Fig. 2D). Also, no statistically significant differences in the expression of TJP2 between these two groups were detected (Fig. 2F). Thus, the GAA expansion affects the expression of FXN mRNA without apparent consequences to the expression of the neighboring genes. Interestingly, increased expression of TJP2 resulting from duplication of a 270-kb fragment encompassing the TJP2 locus has been implicated in dominant, adult-onset, progressive nonsyndromic hearing loss (32). Our FRDA cohort included eight samples derived from patients diagnosed with hearing impairment and eight without hearing deficit. On average, expression of TJP2 mRNA in the FRDA fibroblast obtained from hearing impaired patients was 2.2-fold higher (P = 0.01) than that in the cells derived from patients not diagnosed with hearing deficit (Supplementary Material, Fig. S4).

To confirm that the expanded GAAs do not affect the chromatin landscape of neighboring genes, we conducted ChIP analyses with antibodies specific to two histone marks associated with active transcription (histone H3K4me2 and H3K9ac), along with a histone H3K9me3 antibody to detect a histone mark enriched in transcriptionally silenced loci. ChIP analyses were performed in three FRDA and three control cells lines chosen based on FXN mRNA expression levels (average FXN mRNA expression typical for FRDA and control fibroblasts). As previously reported, the chromatin landscape immediately upstream of the GAA repeats significantly differed between control and FRDA samples (Fig. 3A–C), with characteristic overrepresentation of histone H3K9me3 and lower levels of histone H3K9ac and H3K4me2 in FRDA cells. In spite of the variability in immunoprecipitation efficiency for histone modification-specific antibodies, no differences in chromatin status between representative FRDA and control fibroblasts were observed at PIP5K1B, FAM122A and TJP2 loci, indicating that GAA expansion in FRDA does not have an effect on the chromatin landscape at the neighboring loci.

Figure 3.

Epigenetic changes induced by expanded GAAs in FRDA cells are restricted to the FXN locus. Experiments were conducted in three control fibroblast lines (GM08399, GM01650 and GM02169; black bars) and three FRDA fibroblasts (4497, 281 and 203; white bars). (A–C) ChIP data for the indicated histone marks at the PIP5K1B, FAM122A and TJP2 genes in the vicinity of their transcription start sites and upstream of the GAA repeat region for the FXN locus. Statistically significant differences are denoted by asterisks (P < 0.05).

To test for potential long-range effects of the expanded GAA repeats, we analyzed the expression profile of genes located within a large region 25 Mbp upstream and 25 Mbp downstream of the FXN locus (Supplementary Material, Fig. S5). Expression of 38 annotated transcripts was detected in this region by RNA-seq among all samples; however, RNA levels of only two genes (CTSL1 and CENPP) were higher in the FRDA cohort than those in the control (at FDR ≤ 0.05; Supplementary Material, Fig. S5). No silenced loci could be detected within the 50-Mbp region in FRDA samples. Moreover, the expression of only three genes located on chromosome 9, approximately 52, 59 and 65 Mbp from the FXN locus is downregulated in FRDA cells as compared with controls (RPL12, RPL7A and RPS6).

Expanded GAA repeats impede transcription elongation rate

Epigenetic changes in the vicinity of the expanded GAA repeats have been documented in different eukaryotic model systems including FRDA patient samples (9,10,12,13,33); however, the mechanism of transcriptional silencing and contribution of specific defects in initiation, elongation or termination of transcription remain unclear. It has been demonstrated in vitro that long GAA repeats can impede transcription progression (34,35). In cell culture settings, arguments for both initiation and elongation impairment have been presented (12,22–24).

Herein, we took advantage of our ultra-deep RNA-seq data set consisting of ∼5 × 10⁸ mapped pairs of reads for each group, FRDA and control, to analyze transcription progression through the expressed FXN gene. Critically, the RNA-seq reactions were conducted using total RNA preparations depleted of ribosomal RNAs, as opposed to RNA samples prepared by polyA selection, thus allowing for comprehensive analyses of intronic reads and transcription progression. In order to determine the elongation rate of RNAPII through intron 1 of the FXN gene, RNA-seq reads from all sequencing reactions were combined into one FRDA group and one control group to ensure sufficient sequence coverage and to obtain an average FRDA intronic transcription gradient (Fig. 4A) (36,37). To test whether the expanded GAA repeat region can inhibit transcription progression, we quantified the ratio of RNA-seq signal in intron 1 upstream of the GAA repeats, within the first ∼1.3-kbp, to the RNA-seq signal in ∼9-kbp region of intron 1 downstream of the GAAs. The ratio is ∼2.2-fold greater (P = 0.0052) for FRDA compared with control, indicating accumulation of FXN pre-mRNA upstream of the expanded GAAs (Fig. 4B).

Figure 4.

Expanded GAA repeats impede transcription elongation rate at intron 1 of the FXN gene. (A) A combined landscape representation of RNA-seq tags for all 18 FRDA and 17 control samples mapped to the exon 1/intron 1 region of the FXN locus. Exon 1 and intron 1 are labeled below the plots, and the GAA repeat region is indicated by a dashed line. The nucleotide position on chromosome 9 is indicated above the plots, along with a 2-kb scale bar. RNA-seq tags were mapped to the human reference sequence GRCh37/hg19 lacking the expanded GAAs tract. Both landscapes are presented in the same scale (left of plots). (B) The ratio of RNA-seq signal measured upstream of the GAA repeats versus downstream of the GAAs was calculated for the control (black bar) and FRDA (white bar) cohorts. (C) RNA-seq intron gradient relative to 3′ splice site region (3′ss) for FXN, ZNF169 and TMEM38A. The RNA-seq read count was averaged for samples in the control group (left panel) and FRDA group (right panel). The RNA-seq reads are plotted in 200-bp windows and normalized to the number of reads at the 3′ end of the first intron to compensate for expression differences between mRNAs. The linear regression equation is shown for each plot.

RNA-seq reads mapped to intronic regions of co-transcriptionally spliced transcripts form a characteristic 5′ to 3′ gradient that corresponds to the elongation rate of RNAP II over the intron (Supplementary Material, Fig. S6). A steeper gradient indicates slower RNAPII elongation as more reads accumulate prior to splicing and degradation of intronic sequences (36). Conversely, rapid transcription elongation results in a smaller number of reads over the intron, corresponding to a less steep gradient (Supplementary Material, Fig. S6).

Considering the previously established correlations mentioned earlier, we plotted the number of reads generated by RNA-seq in intron 1 of the FXN gene in the FRDA and control groups. Only fragments having both ends mapped within intron 1 and that were aligned to the direction of FXN gene transcription were considered. The reads were combined into 200-bp windows. To account for differences in gene expression, the signal in each window was normalized to the average signal measured in the 10 windows proximal to the 3′ end of the intron (details in Materials and Methods). The transcript gradient in the FRDA cohort is much steeper than that for the control group, indicating a slower rate of RNAPII elongation in intron 1 of FRDA cells harboring expanded GAAs (Fig. 4C). As a comparison, we plotted the intron 1 gradients of RNA-seq reads for transcripts that met the following criteria: similar expression level to FXN mRNA as measured by normalized RNA-seq reads, no statistically significant difference in expression between the FRDA and control groups, an intron 1 of 10 kbp or longer, lack of overlapping intron 1 from the opposite strand, and a minimum of 10 RNA-seq reads for each individual sample in the control and FRDA cohorts. Only three transcripts TMEM38A, ZNF169 and GMCL1 fulfilled these requirements of similarity to the FXN locus. The intron 1 RNA-seq gradient for each of these genes was similar between the FRDA and control groups, indicating very similar rates of RNAPII elongation (Fig. 4C).

Histone marks associated with transcription elongation are reduced in FRDA fibroblasts

To further validate the defect in transcription elongation observed through FXN intron 1 in FRDA fibroblasts, we conducted ChIP analyses using an antibody specific for histone H4K20me1, a modification tightly linked to the rate of transcription elongation (36,38) and chromatin isolated from FRDA and control fibroblast cells. Interestingly, a progressive underrepresentation of histone H4K20me1 was observed between chromatin upstream and downstream of the expanded GAAs in FRDA cells when compared with control cells (Fig. 5). This effect supports the conclusion of the RNA-seq results that transcription elongation rate, along with initiation, is affected by pathologically expanded GAA repeats. As a reference, we conducted ChIP analyses of histone H4K20me3, a mark of constitutively silenced chromatin, and histone H4K5ac, a mark of transcriptionally active chromatin. In agreement with previous reports, we observed decreased histone H4K5ac upstream and downstream of the GAA tracts in FRDA fibroblasts when compared with the control cells (9,24). The opposite pattern was detected for histone H4K20me3, which was significantly enriched in patient cells in the vicinity of expanded GAAs (Fig. 5).

Figure 5.

Histone H4K20me1 is decreased downstream of the expanded GAA repeats in FRDA cells. ChIP analyses of histone H4K20me1, H4K20me3 and H4K5ac enrichment in control and FRDA fibroblasts (black and white bars, respectively) upstream (UP) and downstream (DN) of the GAA repeats. The data are expressed as the mean ± SEM.

Discussion

Major potential avenues of therapeutic intervention in FRDA include: (i) stimulation of FXN expression, e.g. by histone deacetylase inhibitors; (ii) contraction or excision of the expanded GAA repeats; (iii) supplementation of frataxin (e.g. protein, gene therapy); (iv) stabilizing FXN mRNA or protein; (v) ‘metabolic’ therapy (e.g. iron chelation, mitochondrial function enhancement or ROS scavenging) (27). Selecting the most appropriate and complete therapeutic approach depends on understanding the molecular mechanism of FRDA. For instance, spreading of the GAA-induced silencing beyond the FXN locus could limit the efficacy of frataxin supplementation strategies, such as gene therapy or protein delivery approaches, which would not be able to alleviate the effects of expanded GAAs on genes other than FXN. Also, specifically targeting transcription initiation at the FXN gene is unlikely to fully reactivate its expression without also addressing impaired elongation.

Spreading of chromatin silencing in cis from the initial nucleation region over short- or long-range distances is frequently observed (19). Our deep RNA-sequencing analyses of a large cohort of FRDA and control primary cells together with ChIP results demonstrate a lack of a significant effect of the expanded GAAs on the expression of neighboring genes. We were unable to detect expression of the PIP5K1B gene in FRDA and control cohorts irrespective of the great sequencing depth achieved by 35 RNA sequencing reactions (∼16 500 genes detected). Similarly, expression of PRKACG, a small intronless gene located upstream of the FXN gene, was not detected in FRDA nor control fibroblasts. Conversely, the FAM122A pseudogene positioned in the intron of the PIP5K1B gene upstream of the FXN locus was highly expressed in human primary fibroblasts regardless of FXN expression level. A potential for spreading of the GAA-induced silencing resulting in phenotypic consequences was recently suggested (25). However, our unbiased, comprehensive gene expression analyses across chromosome 9 demonstrate a lack of long-range silencing effects of the expanded GAA repeats throughout the FRDA sample group (Supplementary Material, Fig. S5). Perhaps chromatin insulators, such as CTCF, are involved in restricting heterochromatin to the FXN locus (39).

We did not observe a significant difference between FRDA and control cohorts in the expression of the TJP2 gene, indicating no spreading of the GAA-induced silencing effect downstream of FXN. However, when analyzed within the FRDA cohort, TJP2 expression was notably higher in patients diagnosed with hearing loss. Perhaps a cumulative effect of stronger TJP2 expression (although still within the range observed in unaffected individuals) combined with low FXN expression facilitates development of a hearing deficit, or increased TJP2 expression could be a response to other pathways mediating hearing deficiency. Additional studies on larger patient and control cohorts will be required to corroborate this finding and perhaps include evaluation of TJP2 expression as a predictive biomarker of hearing loss in FRDA.

Transcription is a multistep, highly regulated process that can be divided into three carefully orchestrated stages: initiation, elongation and termination. Simultaneously, newly synthesized RNAs undergo additional processing, such us splicing and cap synthesis, further complicating production of mature transcript. The transcription rate of the first few thousands of base pairs from initiation is typically 4–10 times lower than transcription rate after the initial ∼15 kbp (36–38). It has been proposed that this difference is caused by ‘maturation’ of the transcription machinery and involves an increase of RNAPII serine 2 phosphorylation and/or gradual removal of pausing factors (37). The expanded GAA repeats are located ∼1.7 kbp downstream of the transcription start site, within the zone of productive elongation of the FXN transcript. The deleterious effect of long GAAs on elongation can be potentiated when these repeats are located in the region of ‘early’ elongation. Moreover, progression of transcription machinery can be impeded by the low complexity of the DNA repeats and formation of stable non-canonical DNA conformations (34,35,40–44). Perhaps also local and transient depletion of specific nucleotide substrates, as proposed for replication of the expanded simple repeats (45), can lead to transcriptional pausing. Taken together, an encounter between not yet fully potent transcriptional machinery and extremely difficult sequences/structures can result in effective inhibition of transcription elongation.

It is likely that both the elongation defect and the previously reported transcription initiation deficiency caused by the expanded GAAs are tightly linked. Pausing during transcription elongation is frequently associated with RNAP backtracking, and the probability of backtracking depends on the efficiency of transcription initiation (46). It has been demonstrated in Escherichia coli coli and yeast that the elongation rate of the leading RNAP complex can be facilitated by elongation complexes trailing behind and effectively stimulating efficiency of transcript synthesis by reducing pausing-backtracking (46,47). Thus, robust initiation results in faster transcription elongation. Considering that similar mechanisms maybe present in human cells, decreased initiation at the FXN promoter is likely to negatively affect the elongation rate. Paradoxically, in such a scenario, FXN expression may be most severely affected in cells/tissues producing a lower basal level of frataxin (i.e. lower level of FXN transcription). Recently, genome-wide transcription profiling experiments have identified histone H4K20me1 and histone H3K79me2 as modifications that positively correlate with transcription elongation rate (36,38). We previously found significantly decreased histone H3K79me2 upstream and downstream of the GAA repeats in FRDA lymphoblast cell lines (24). We now discovered that the histone H4K20me1 mark of transcription elongation rate is decreased in FRDA samples downstream of the expanded GAA repeats, further substantiating an impediment to transcription elongation. Additionally, defects in elongation rate associated with formation of RNA–DNA hybrids at the expanded GAAs and antisense transcription can potentially facilitate a premature termination of the transcription in the vicinity of the repeats (48).

In summary, we demonstrated using next-generation RNA sequencing of a large cohort of FRDA and control fibroblast lines an impediment to transcription elongation through the FXN gene in the presence of expanded GAA repeats. Additionally, we showed that epigenetic silencing induced by the expanded GAAs is limited to the FXN locus and does not spread to nor affect expression of upstream and downstream genes. The limited success of approaches aimed to reactivate the mutated FXN gene (49,50) suggests that simultaneous targeting of different enzymatic activities involved in FXN silencing combined with direct targeting of conformational obstacles impeding transcription may result in reactivation of its expression.

Materials and Methods

FRDA patient samples

A total of 18 FRDA patient fibroblast cell lines were used in this study. Sixteen fibroblast lines were derived from skin biopsies performed at The Children's Hospital of Philadelphia. All patients' signed informed consent forms and the study were approved by The Committees for the Protection of Human Subjects of The Children's Hospital of Philadelphia (IRB 10-007864) and the Institutional Review Board for Human Use (IRB) at the University of Alabama at Birmingham (IRB N131204003). All fibroblast lines designated GM (FRDA and controls) were purchased from NIGMS Human Genetic Cell Repository at The Coriell Institute for Medical Research, Camden, NJ, USA.

Establishing FRDA fibroblast lines

Isolation of human fibroblast lines from skin biopsies was conducted, as described in Li et al. (51). Briefly, biopsy material was washed two to three times using ∼5 ml HBSS with 1× penicillin and streptomycin (Hyclone, Logan, UT) for ∼1–2 min, suspended in an enzyme mixture containing Collagenase IV 1 mg/ml, dispase 1 mg/ml (Stem Cell Technologies, Vancouver, British Columbia, Canada) and trypsin+EDTA (0.05%), cut to small pieces and incubated for ∼30 min at 37°C with gentle shaking. Digested tissue fragments were transferred to 15-ml conical tubes, and DMEM/F12 media (Life Technologies, Carlsbad, CA) containing glutamine, 20% FBS (Hyclone), 1× penicillin/streptomycin and 1× of non-essential amino acids (Life Technologies) was added prior to centrifugation and subsequent plating onto gelatin-treated six-well plates. Typically outgrowths of fibroblasts were observed 7–14 days after initial plating. Established cultures were maintained in DMEM/F12 media containing 10% FBS at 37°C, 5% CO₂.

Determination of GAA repeat number and interruption status

The number of GAA repeats was determined by PCR using two sets of primers FXN_short (forward and reverse; 498 bp of the GAA repeats flanking sequences) and FXN_long (forward and reverse; 1370 bp of the GAA repeats flanking sequences; Supplementary Material, Table S2) using previously described conditions (51). Products were separated on 1% agarose gels. The band sizes were determined using GelAnalyser. The presence of non-GAA interruptions in the repeat tract was determined using MboII digestion exactly as described in (28).

Quantitative real-time-PCR

Total RNA was extracted using RNeasy Mini Kit (Qiagen, Valencia, CA) and rigorously treated with DNase I (TURBO DNA-free; Ambion, Carlsbad, CA). The qRT-PCR reactions were conducted using Power SYBR Green RNA-to-C_T 1-Step Kit (7500 Fast Real Time-PCR System; Applied Biosystems, Carlsbad, CA). All primers are listed in Supplementary Material, Table S2. Reactions were also performed without reverse transcriptase to confirm removal of genomic DNA. Reverse transcription was conducted at 48°C for 30 min, followed by 40 cycles of denaturation at 95°C for 15 s, annealing at 55°C for 20 s and elongation for 1 min at 60°C. To verify FXN mRNA expression, qRT-PCR reactions were also conducted using a TaqMan Gene Expression Assay (FXN: IDHs00175940_m1 and GAPDH: Hs02758991; Applied Biosystems) according to manufacturer's protocol. All reactions were conducted in triplicate, and reported data are the results of at least two independent analyses.

Western blot

Lysates were prepared using Passive Lysis Buffer (Promega, Madison, WI) and protease inhibitor cocktail (Sigma–Aldrich, St. Louis, MO). Protein concentration was determined with Bradford Protein Assay kit (Bio-Rad, Hercules, CA). Twenty micrograms of whole cell extract was electrophoresed on 4–12% SDS–polyacrylamide gels followed by electrophoretic transfer onto nitrocellulose membranes (Hybond-C; GE Healthcare, Pittsburgh, PA). Human frataxin was detected with the anti-frataxin H-155 polyclonal antibody (Santa Cruz, Dallas, TX) at 1:100 dilution for 12 h at 4°C. GAPDH was detected with mouse anti-GAPDH monoclonal antibody (Millipore, Billerica, MA) at 1:10 000 dilution for 12 h at 4°C. Horseradish peroxidase-conjugated rabbit anti-mouse immunoglobulin and donkey anti-rabbit immunoglobulin (GE Healthcare) were used as secondary antibodies for 1 h at room temperature at 1:20 000. Signal was quantified using Image Lab (BioRad). Special attention was devoted to the membrane exposure to avoid saturation of the signal for appropriate quantitation.

Chromatin immunoprecipitation

ChIP was performed according to the EZ ChIP instructions (Upstate Biotechnology, Billerica, MA), as described in Li et al. (51). The antibodies used for ChIP analyses were as follows: anti-rabbit IgG as a negative control (Cell Signaling, Danvers, MA), anti-total histone H3 (Cell Signaling), anti-total histone H4 (Abcam, Cambridge, MA), anti-histone H3K9/14ac (Upstate Biotechnology), anti-histone H3K9me3 (Upstate Biotechnology or Active Motif, Carlsbad, CA), anti-histone H3K4me2 (Abcam), anti-histone H4K20me3 (Abcam), histone H4K20me1 (Abcam) and histone H4K5ac (Abcam). Immunoprecipitated chromatin was purified using phenol/chloroform extraction and ethanol precipitation before qRT-PCR. The qRT-PCR was conducted using the Power SYBR Green-C_T Kit (7500 Fast Real Time-PCR System or Step-One Plus System, Applied Biosystems). The qRT-PCR was carried out as follows: 10 min at 94°C, 40 cycles of 30 s at 94°C followed by 60 s at 60°C. The relative abundance of histone modifications was determined by normalizing the quantity of the immunoprecipitated sample to the quantity of total histone H3 or H4 after normalization to the input reactions (%Input).

Next-generation sequencing of RNA (RNA-seq)

A total of 1 µg of each RNA sample was used for RNA-seq analyses. A second DNase I treatment was performed for all samples to eliminate genomic DNA contamination. RNA-seq analyses were conducted using HiSeq2000 (Illumina, Inc., San Diego, CA) at the University of Texas MD Anderson Cancer Center Molecular Biology core facility, Science Park Department of Molecular Carcinogenesis. Library preparation was conducted using TruseqStranded Total RNA Sample Prep kit (Illumina) according to the manufacturers’ instructions. Importantly, this protocol included an rRNA depletion step instead of polyA selection allowing for expression analysis of pre-mRNAs. The cDNA synthesis was conducted using SuperScript II reverse transcriptase followed by second strand synthesis, adenylation of 3′ ends and ligation of adapters. The libraries were sequenced using a 2 × 76 bases paired end protocol on Illumina HiSeq 2000 instrument. Totally, 35 libraries (18 from FRDA patients and 17 from unaffected controls) were sequenced in 6 lanes, generating 27–39 million pairs of reads per sample. Each pair of reads represents a cDNA fragment from the library. The reads were mapped to human genome (hg19) by TopHat (version 2.0.7) (52) and bowtie2 (Version 2.1.0) (53). By reads, the overall mapping rate is 89–97%. A total of 81–94% fragments have both ends mapped to human genome. The number of fragments in each known gene from RefSeq database (54) was enumerated using htseq-count from HTSeq package (version 0.5.3p9, HTSeq: http://www-huber.embl.de/users/anders/HTSeq/). Genes with <10 fragments in all the samples were removed before differential expression analysis. The differential expression between conditions was statistically assessed by R/Bioconductor package DESeq (version 1.10.1) (55). Genes with FDR ≤0.05 were called as differentially expressed.

Landscape profile of RNA-seq signal

For the fragments that have both ends mapped, the first reads were kept. Together with the reads from the fragments that have only one end mapped, every read was extended to its 3′ end by 200 bp in exon regions. For each read, a weight of 1/n was assigned, where n is the number of positions the read was mapped to. The sum of weights for all the reads that cover each genomic position was rescaled to normalize the total number of fragments to 1 m and averaged over 10 bp resolution. The averaged values were displayed using UCSC genome browser (http://genome.ucsc.edu/). For the combined landscape, the individual landscape profiles were averaged over all the FRDA samples and control samples.

RNA-seq intron gradient relative to 3′ splice site regions

To ensure the fragments were truly from normal gene transcription, only the fragments with both ends mapped within the gene and in the same direction of the gene transcription were used. For each sample, the signal at each position was calculated as the coverage of fragments, normalized to 100 total fragments in the gene. Then the signal was averaged over all FRDA samples and control samples separately. To smooth the signal and compensate for expression difference between different genes, the maximal signal over each 200-bp window was normalized by the last 10 windows at the 3′ end of the first intron.

Statistical analyses

Statistical analyses were conducted using GraphPad Prism 6. Statistical significance was determined by performing paired, two-tailed Student's t-test or Fisher exact test and, unless otherwise indicated, P < 0.05 was considered significant.

Supplementary Material

Supplementary Material is available at HMG online.

Funding

These studies were supported by National Institute of Health (7R01NS081366) from NINDS to M.N., a grant from the Muscular Dystrophy Association (MDA0789 to M.N.), Friedreich's Ataxia Research Alliance (FARA to M.N. and separately to D.L.) and National Ataxia Foundation postdoctoral fellowship to A.D.B.