Peripheral blood samples were collected in purple-top/EDTA tubes from 50 study subjects with a confirmed DNA diagnosis of FRDA (homozygosity for expanded GAA triplet-repeat alleles; Table 1), 14 unaffected individuals, and five heterozygous carriers at The Children’s Hospital of Philadelphia (CHOP). Blood samples were shipped overnight on ice via courier service for processing and analysis at The University of Oklahoma Health Sciences Center (OUHSC) in Oklahoma City. Research Protocols were approved by the Institutional Review Boards at both institutions (CHOP IRB# 01-002609 and OUHSC IRB# 8071). Informed consent was obtained from all participants and/or their legal guardian(s) in accordance with the Declaration of Helsinki. PBMCs were isolated using Ficoll-Paque PLUS (GE Healthcare) and incubated in lymphocyte growth medium (RPMI with Glutamax, 15% FBS, pen/strep) at 37°C in 5% CO₂ for 24 h. Cells were treated with 10 μM HDACi-109 (RG2833, Selleckchem, #S7292) dissolved in dimethyl sulfoxide (DMSO) or with DMSO only for 48 h. The final concentration of DMSO in the cell culture medium for HDACi-109 or DMSO-only treatment was 0.1%.

Total RNA (400 ng) was reverse transcribed following manufacturer instructions for the QuantiTect^® reverse transcription kit (Qiagen). Transcript levels were quantified by real-time PCR with Power SYBR green PCR master mix (Applied Biosystems) on a Roche LightCycler^® 96 System. Primers spanned the splice junction of FXN exons 3 and 4 values were normalized to expression of RPL27 using the ΔΔCt method. Primer sequences and reaction conditions for FXN Ex3-Ex4 and RPL27 were as previously described (Herman et al., 2006; de Jonge et al., 2007).

Response to HDACi-109 was calculated in two ways: (1) individual fold change was calculated as the FXN transcript level for HDACi-109 treated PBMCs calculated relative to that individual’s DMSO-only level, which was artificially set to 1 (see Figure 1A); (2) individual reactivation response relative to heterozygous carrier was calculated using the FXN transcript level in DMSO-only and HDACi-109 treated PBMCs for all patients, relative to the FXN transcript level in heterozygous carriers, artificially set to 0.5 (see Figure 1B). The individual reactivation response relative to heterozygous carrier was used in all correlation analyses in this study.

Bisulfite deep sequencing to assay DNA methylation at the FXN locus was previously described in detail (Rodden et al., 2021). Briefly, genomic DNA (0.5 μg), isolated from freshly isolated PBMCs, was bisulfite converted and prepared for targeted deep sequencing. Four amplicons were analyzed to cover all the CpG dinucleotides from the 3′ end of the CpG island to the expanded GAA triplet-repeat (n = 39 CpG sites; numbered 57–95 in Figure 2A), i.e., the region of the FXN gene known to be hypermethylated in FRDA. Amplicons were dual-indexed and pooled to create a library which was sequenced on an Illumina MiniSeq platform. n = 1000 sequence reads per sample were used to calculate the percentage of methylated cytosines at individual CpG dinucleotides and plotted with LOWESS regression to generate trend lines (Cleveland, 1981). Methylation in the FRDA-DMR, a region previously identified as having the maximal difference in methylation between FRDA (typically >90%) and non-FRDA controls (typically <10%), was calculated using n = 1000 sequencing reads covering the 11 CpG sites that map within the FRDA-DMR (numbered 72–82 in Figure 2A). Deep sequencing of a single amplicon that spanned all 11 CpG dinucleotides in the FRDA-DMR and analyzing the methylation status in cis within 300 individual reads (representing 300 FXN molecules per patient) was used to identify individual FXN epialleles. Epialleles were defined as follows: fully methylated (all 11 CpG sites methylated), unmethylated (≤2 CpG sites methylated), and partially methylated (>2 and <11 CpG sites methylated). The definition of partially methylated alleles was further refined in the analysis in Figure 4. Unmethylated epialleles were permitted to have up to 2 methylated CpGs because of the occasional and sporadic methylation seen in FXN epialleles from non-FRDA controls. Epiallele proportions were calculated as the percentage of n = 300 individually sequenced FXN DNA strands.

Statistical tests were performed using GraphPad, Prism v9. Student’s t-test and Mann–Whitney test were used to compare means and medians where appropriate. Linear correlations were evaluated to determine the correlation coefficient “R” where appropriate, and the method of Benjamini and Hochberg (1995) for false discovery rate (FDR = 10%) was used to account for multiple linear regression comparisons for the data in Figure 4A, and depicted as the adjusted p values in Table 3 (which had a modified discovery threshold of p < 0.04).

A prospective cohort of 50 FRDA patients was enrolled with the goal of testing the hypothesis that FXN DNA hypermethylation predicts response of FXN gene reactivation via HDACi-109 treatment. This cohort was generally representative of the broader FRDA patient population in terms of age of onset and sizes of expanded GAA triplet-repeats (Table 1; Regner et al., 2012). Consistent with previous studies, HDACi-109 treatment of PBMCs increased the median steady-state FXN transcript level to 3.9-fold over the corresponding DMSO control (Figure 1A), indicating that it reactivates the silenced FXN gene in FRDA. However, individual responses varied considerably in the increase in FXN transcript following HDACi-109 treatment, ranging from 1.7- to 12-fold over DMSO control (Figure 1A). A desirable goal of FXN reactivation therapy is to increase FXN transcript levels to at least the level seen in asymptomatic carriers, i.e., to approximately 50% of non-FRDA levels. In FRDA there is considerable variability in the residual level of FXN transcript, mostly driven by the large variability in length of the expanded GAA repeats, especially the shorter of the two expanded alleles (GAA1) (Pianese et al., 2004; Chutake et al., 2014; Li et al., 2015). Thus “fold-change over DMSO control” as a measure of drug response (as seen in Figure 1A) lacks the ability to demonstrate the potential clinical usefulness of the magnitude of increase in FXN transcript level. Therefore, the increase in FXN levels with HDACi-109 was analyzed relative to the levels seen in heterozygous carriers, which was artificially set at 0.5 to represent half the level seen in non-FRDA controls (Figure 1B). Using this measure, the cohort showed an increase in FXN transcript level from a median value of 0.15 (range: 0.05–0.57) to 0.59 (range: 0.21–1.62) following HDACi-109 treatment (p < 0.0001; Figure 1B), and resulted in samples from 42 of 50 patients (84%) achieving or surpassing the steady-state levels of FXN transcript seen in heterozygous carriers. Thus, HDACi-109 effectively reactivates the silenced FXN gene in FRDA, but with considerable inter-individual variability, whereby a substantial minority of individuals fail to achieve a level that is considered clinically meaningful.

In order to test the hypothesis that inter-individual variability in FXN DNA methylation predicts variability in HDACi-109 response, FXN DNA methylation was analyzed in PBMCs from the cohort of 50 patients. The expanded GAA triplet-repeat in FRDA induces DNA hypermethylation in intron 1 (Castaldo et al., 2008; Evans-Galea et al., 2012; Rodden et al., 2021). The maximal difference of DNA methylation between FRDA and non-FRDA controls is seen in a region of intron 1 with 11 contiguous CpG sites, which form a defined FRDA-specific differentially methylated region (FRDA-DMR) located just downstream from the FXN CpG island (Figure 2A; Rodden et al., 2021). This cohort showed the expected variation in magnitude of FRDA-DMR methylation, ranging from 56 to 97% in FRDA (n = 50; Figures 2B,C; Supplementary Table 1), compared with 4% in non-FRDA controls (n = 14; p < 0.0001; Figures 2B,C) and 68% in heterozygous carriers (n = 5; p < 0.001; Figure 2C). Deep sequencing of a single amplicon spanning the FRDA-DMR to assess methylation of all 11 CpG sites in cis was performed to analyze methylation within individual FXN DNA strands, i.e., somatic FXN epialleles. DNA methylation of 300 such FXN epialleles, representing the somatic variability of methylation in the FRDA-DMR, was determined for all 50 patients in our cohort (Figure 2D shows a representative set of 300 stacked FXN strands per individual; black dash = methylation; sorted with highest methylation at the bottom). Individually sequenced FXN molecules revealed three types of epialleles: fully methylated (with all 11 CpGs methylated), unmethylated (≤2 CpGs methylated), and partially methylated (with >2 and <11 CpGs methylated) (Figures 2D,E). The majority of epialleles in FRDA patients are fully methylated or partially methylated, both displaying considerable inter-individual variability in our cohort (Figures 2D,E). Patients also showed a smaller and variable proportion of unmethylated epialleles (Figure 2E). We hypothesized that the highly variable prevalence of FXN epialleles may help explain the inter-individual variability in response to HDACi-109 treatment. Specifically, that the relative prevalence of partially methylated and unmethylated epialleles in FRDA may represent the individual potential for FXN reactivation.

FXN DNA hypermethylation in FRDA is known to be primarily dependent on the length of the expanded GAA repeat, and does not correlate with patient age and disease duration, or differ between males and females (Rodden et al., 2021). Our hypothesis that variability in HDACi-109 response is predicted by variability in FXN DNA methylation requires that FRDA-DMR methylation and the distribution of somatic FXN epialleles remain stable. This requires that drug treatment per se does not alter the level of DNA methylation or relative proportion of FXN epialleles. Five FRDA patients in our cohort were systematically tested to investigate if HDACi-109 alters the level of FRDA-DMR hypermethylation, or the proportional distribution of somatic FXN epialleles. These individuals were selected for their wide range of GAA repeat lengths, DNA methylation levels, and broad distribution of the three types of FXN epialleles (Table 2). These five patients also showed robust reactivation responses following HDACi-109 treatment, achieving at least the level seen in heterozygous carriers (set at 0.5; Table 2), indicating adequate modification of the FXN locus. DNA methylation was assayed with and without drug treatment, which did not appreciably change in these patients upon treatment with HDACi-109 (Figure 3A). We also did not observe any appreciable change in the prevalence of unmethylated, partially methylated, and fully methylated epialleles in these patients following treatment (Figures 3B–E). Therefore, at least in non-dividing PBMCs, HDACi-109 treatment does not alter the magnitude or pattern of FRDA-DMR hypermethylation in FRDA.

Bisulfite deep sequencing provided a rich dataset of DNA hypermethylation in FRDA, including 300 FXN epialleles representing the somatic heterogeneity of FRDA-DMR hypermethylation in each of 50 patients. The level of DNA methylation at each of the 39 CpG sites (numbered as 57–95 in Figure 2A) spanning the region of intron 1 upstream of the expanded GAA repeat was tested for correlation with response to HDACi-109. This revealed that most CpG sites within the FRDA-DMR (and a few outside of the DMR) showed statistically significant negative correlations with HDACi-109 response (Figure 4A and Table 3), indicating that lower methylation in the FRDA-DMR correlates with a higher magnitude of response. We then tested for correlation between the prevalence of variably methylated FXN epialleles and response to treatment with HDACi-109. HDACi-109 response showed a statistically significant positive correlation with the prevalence of unmethylated (≤2 CpGs methylated) epialleles and partially methylated epialleles containing 3–7 (of 11) methylated CpG sites (Figure 4B and Table 4). Interestingly, methylated epialleles with ≥8 CpGs methylated and fully methylated FXN epialleles did not show this correlation, or were even inversely correlated (Figure 4B and Table 4), suggesting that highly methylated epialleles (>70%) represent FXN genes that are relatively resistant to reactivation. These data support the hypothesis that the variable prevalence of unmethylated and partially methylated FXN epialleles (0–64% methylation) underlies the potential for FXN reactivation via HDACi-109 treatment in FRDA. Indeed, such alleles constitute a substantial, albeit variable, subset of somatic FXN genes in FRDA patients (Figure 4C).

Previous studies did not find any correlation of GAA1 with response to HDACi-109 treatment (Plasterer et al., 2013). In our cohort we noted weak correlation of response to HDACi-109 with GAA1 (p = 0.04). No correlation was seen with GAA2 (p = 0.62) or GAA1 + 2 (p = 0.12). Furthermore, no correlation of HDACi-109 response was seen with either disease duration (p = 0.12) or patient age at sampling (p = 0.74).