Transcriptional profiling of isogenic Friedreich ataxia neurons and effect of an HDAC inhibitor on disease signatures

Abstract

Friedreich ataxia (FRDA) is a neurodegenerative disorder caused by transcriptional silencing of the frataxin (FXN) gene, resulting in loss of the essential mitochondrial protein frataxin. Based on the knowledge that a GAA·TTC repeat expansion in the first intron of FXN induces heterochromatin, we previously showed that 2-aminobenzamide–type histone deacetylase inhibitors (HDACi) increase FXN mRNA levels in induced pluripotent stem cell (iPSC)-derived FRDA neurons and in circulating lymphocytes from patients after HDACi oral administration. How the reduced expression of frataxin leads to neurological and other systemic symptoms in FRDA patients remains unclear. Similar to other triplet-repeat disorders, it is unknown why FRDA affects only specific cell types, primarily the large sensory neurons of the dorsal root ganglia and cardiomyocytes. The combination of iPSC technology and genome-editing techniques offers the unique possibility to address these questions in a relevant cell model of FRDA, obviating confounding effects of variable genetic backgrounds. Here, using “scarless” gene-editing methods, we created isogenic iPSC lines that differ only in the length of the GAA·TTC repeats. To uncover the gene expression signatures due to the GAA·TTC repeat expansion in FRDA neuronal cells and the effect of HDACi on these changes, we performed RNA-seq–based transcriptomic analysis of iPSC-derived central nervous system (CNS) and isogenic sensory neurons. We found that cellular pathways related to neuronal function, regulation of transcription, extracellular matrix organization, and apoptosis are affected by frataxin loss in neurons of the CNS and peripheral nervous system and that these changes are partially restored by HDACi treatment.

Introduction

Friedreich ataxia (FRDA⁵; OMIM 22930) is an autosomal recessive neurodegenerative disorder, characterized by progressive spinocerebellar ataxia, dysarthria, sensory neuropathy, and pyramidal weakness (1). Most patients show signs of hypertrophic cardiomyopathy (2); about half present with glucose intolerance, and 12% are clinically diagnosed with diabetes (3). The initial sites of neurodegeneration are the dorsal root ganglia (DRG), followed by spinocerebellar tracts and the dentate nucleus of the cerebellum (2). The DRG show a reduction in size and number of sensory neurons accompanied by proliferation of satellite cells (4). Larger neurons of the proprioceptive type are mainly affected, resulting in loss of position sense, but pain, light touch, and temperature perception can also decrease in advanced stages of the disease (1). Analyses of the cell sizes in the DRG are consistent with a pathogenic mechanism of hypoplasia rather than atrophy and support the idea that normal development of the DRG and the spinal cord is hindered in FRDA (5).

An expanded GAA·TTC triplet repeat sequence, located in the first intron of the nuclear FXN gene, causes inhibition of transcription and the loss of the essential mitochondrial protein frataxin in affected individuals. Available evidence supports a role for frataxin in the biogenesis of iron–sulfur (Fe-S) clusters in mitochondria, resulting in impaired activities of Fe-S enzymes, altered cellular iron metabolism, decreased mitochondrial energy production, and increased oxidative stress (6, 7). To counteract these abnormalities, antioxidants, iron chelators, and stimulants of mitochondrial biogenesis have been proposed as therapeutics (8). However, no clear results supporting the benefit of any of these drugs have so far been obtained in randomized human trials (9). Other avenues for therapeutic development, however, are being pursued, including strategies aimed at increasing frataxin expression by preventing frataxin degradation (10), repeat-targeted oligonucleotides (11), and synthetic transcription elongation factors (12), together with protein replacement therapy (13), stem cell therapy (14) and gene therapy (15). Based on the knowledge that GAA·TTC expansion leads to heterochromatin formation and gene silencing, we have shown that members of the 2-aminobenzamide family of histone deacetylase inhibitors (HDACi) reproducibly increase FXN mRNA levels in FRDA lymphoblast cell lines (16), primary lymphocytes from FRDA patients (17), FRDA mouse models (18, 19), and human FRDA neuronal cells derived from patient-induced pluripotent stem cells (iPSCs) (20). A phase I clinical trial with HDACi 109 (RG2833) demonstrated increased FXN mRNA in peripheral blood mononuclear cells from patients treated with the drug (20), providing a proof-of-concept for this therapeutic approach.

Although loss of frataxin is believed to be the main driver of the disease, the complex pathophysiology of FRDA is still not fully elucidated. For example, the roles of oxidative stress and iron metabolism in FRDA pathology are unclear (21, 22). Additionally, and similar to other neurodegenerative diseases, only certain cell types and tissues are affected, despite frataxin being ubiquitously expressed. Previous gene expression profiling studies aimed at addressing the molecular basis of FRDA pathophysiology have been conducted in mouse models that do not to fully recapitulate the human disease (18, 23–27) or human cells that do not represent an affected tissue (28–34). The advent of induced pluripotent stem cell (iPSC) technology (35) has allowed in vitro modeling of diseases that involve inaccessible human tissue (36). Moreover, advances in genome editing techniques allow the establishment of isogenic lines that overcome inter-individual variabilities in genome-wide studies. Here, we present the first transcriptomic study in FRDA of human iPSC–derived CNS and isogenic sensory neurons (SNs) and identify distinct but linked dysregulated pathways that are partially restored by HDACi treatment.

Results

Transcriptional profiling of FRDA iPSC-derived neuronal cells

We previously derived iPSC lines from FRDA fibroblasts (37) and showed that they can be differentiated into functional β-III tubulin-positive neurons (20). Using a modified version of our published protocol (adapted from Ref. 38), we differentiated four iPSC lines, two from unaffected individuals (KiPS, (39) and GM08333, Coriell Institute) and two from FRDA patients (from Coriell fibroblast lines GM03816 and GM04078) into CNS neurons. To investigate the effect of loss of frataxin on global gene expression and the effect of HDACi 109 (20) on such transcriptional changes, 14-day-old neurons were treated for 48 h with 5 μm 109 or DMSO. Similar to previous studies (20), FXN expression is lower in FRDA neurons compared with controls and is increased upon 109 treatment in affected cells (Fig. 1A). Total RNA was extracted, and rRNA-depleted RNA samples were sequenced using the Illumina HISeq Analyzer 2000 and mapped to the human genome (see “Experimental procedures”). Over 10 million single-end 75-bp reads were generated from each sample, with ∼70–75% of the reads mapping to exons. Principal component analysis (PCA) shows clustering dependent on FXN expression and HDACi treatment (Fig. 1B). 5545 genes were found to be differentially expressed (DE) between FRDA lines and controls (with a false discovery rate (FDR) <0.01), of which 2939 were up-regulated and 2606 were down-regulated (Fig. 1, C and D). Remarkably, and as reported previously (18, 30), over 50% of dysregulated genes are reverted toward normalization when patient cells are treated for 48 h with HDACi 109 and about 10% of these changes are statistically significant (Fig. 1E). Top enriched pathways for these reverted genes, identified within the GO biological process database, are reported in File S1. To investigate the nature of DE genes and the effect of HDACi 109 on their expression, we performed weighted gene coexpression network analysis (WGCNA) to identify modules of coexpressed genes that share similar functions or cellular localization. Dendrograms of clustered genes, identified modules, and their correlation are shown in Fig. 2, A and B, and module membership for each gene is indicated in File S2. Based on an FDR cutoff of <0.1 and on the module eigengenes (Fig. 2C), which summarize each module expression data across all samples, six modules were selected for pathway analysis. The brown, magenta, and black modules include genes that are down-regulated as a consequence of repeat expansion, and the pink, turquoise and yellow modules include genes that are up-regulated in the disease state. In the magenta and pink modules, the expression of dysregulated transcripts changed toward normalization with 109 treatment, which is possibly a consequence of restoration of FXN transcription. Top enriched pathways identified within the GO biological process and cellular component databases are shown in Fig. 2D. It is notable that over-represented genes in the pink module are related to transcriptional regulation and that these genes are reverted toward normalization by treatment with the HDAC inhibitor. Other identified pathways are extracellular matrix (ECM) organization (brown), chemical synapsis transmission (turquoise), regulated exocytosis (magenta), mitochondrion (black), and vesicle-mediated transport (yellow). Although the black module points to the well-established mitochondrial dysfunction in FRDA, we failed to see a beneficial effect of 109 on the expression of these genes.

Figure 1.

Transcriptomic analysis of FRDA and unaffected iPSC-derived neurons. A, frataxin expression in two FRDA and two control neuronal lines, plotted as voom-transformed values from RNA-seq counts. B, PCA plot of samples. C, number of differentially expressed genes with FDR <0.01, for the comparisons shown. Green (left panels) are down-regulated genes, and red (right panels) are up-regulated genes. D, volcano plots of DE genes in FRDA versus control cells (left) and 109-treated versus DMSO-treated FRDA neurons. The x axis shows the log fold change (logFC) and the y axis show the −log₁₀ p values for each gene. Points in green represent genes that are significant at FDR <0.01, and points in red represent genes that were not significant. E, changes in DE genes between FRDA and control neurons upon HDACi 109 treatment.

Figure 2.

Network analysis. A, hierarchical clustering dendrogram of all genes in the co-expression network. Module assignments are shown in the bar at bottom. B, top, hierarchical clustering dendrogram of module eigengenes. Bottom, heatmap of correlation coefficients between each pair of module eigengenes. C, eigengenes for the six most significant pathways (FDR <0.1). D, top overrepresented pathways in the six selected WGCNA modules.

Creation of isogenic iPSC lines by adenovirus-mediated homologous recombination

The comparison of transcriptomes of unaffected and diseased iPSC-derived neuronal cells can be confounded by their different genetic backgrounds (40). We therefore sought to derive isogenic iPSC lines that differ exclusively in the number of GAA·TTC repeats in the first intron of the FXN gene. Compared with nuclease-based methods such as zinc finger nucleases, TALENs and CRISPR–Cas9 systems, homologous recombination offers the advantage of virtually scar-less gene editing. We selected helper-dependent adenovirus (HdAV) due to its capacity of accommodating large amounts of genetic material, absence of genomic integrations, and its minimal mutation rate (41–44). We constructed a correction vector plasmid, FXN-HdAV, that contained ∼19 kb of the human FXN gene with six GAA·TTC repeats in intron 1. The vector also included an excisable neomycin cassette flanked by Frt sequences that was used for selection and subsequently removed seamlessly by Flp recombinase excision (Fig. S1A).

Adenoviruses packaged with the linearized FXN-HdAV construct were used to infect FRDA patient-derived iPSCs, which we had previously generated and characterized (37) (from Coriell fibroblast line GM03816; hereby denoted as FXN ^exp/exp). We then screened for clones that were corrected at one allele, using standard PCR methods. The removal of the neomycin cassette produced the isogenic heterozygous iPSC line FXN^exp/6. By repeating this process, we then generated iPSC clones where both alleles were corrected (generating FXN ^6/6; see Fig. S1, B and C for correction scheme). These isogenic iPSCs were characterized by karyotyping, embryoid body formation, and pluripotency marker expression, confirming that they retained pluripotency (Fig. S2).

Correction of the GAA·TTC repeats restores FXN expression and reverses heterochromatin formation in iPSCs and iPSC-derived neurons

The presence of short GAA·TTC repeats in the isogenic iPSCs was confirmed by standard PCR (Fig. 3A). FXN gene and frataxin protein expressions were measured by quantitative real-time PCR (qRT-PCR) (Fig. 3B) and Western blotting (Fig. 3C), respectively, demonstrating that correction of the GAA·TTC repeats restores FXN mRNA and protein levels to those of unaffected cells. To demonstrate reversal of heterochromatin silencing of the FXN gene, chromatin immunoprecipitation (ChIP) studies were performed and revealed that H3K9 acetylation, a critical histone mark down-regulated in FRDA cells (20), was restored to normal levels in fully corrected FXN ^6/6 iPSCs (Fig. 3D). The expression of the FXN transcript and of frataxin upon repeat shortening clearly demonstrates that the repeat expansion is solely responsible for the transcriptional defect. Moreover, the acetylation of histones around the GAA·TTC repeats is directly correlated to repeat length.

Figure 3.

Characterization of isogenic iPSC lines. A, genotyping of lines FXN^exp/exp (lane 1), FXN^exp/6 before neomycin cassette removal (lane 2), FXN^exp/6 after neomycin cassette removal (lane 3), FXN^6/6 (lane 4), and one nonisogenic unaffected control (lane 5). Lane 6 is a no DNA negative control. Ethidium bromide stained agarose gel is shown. B, qRT-PCR quantification of FXN mRNA levels in the isogenic lines and in one unaffected nonisogenic line. C, Western blotting of cell lysates derived from isogenic iPSCs and one nonisogenic control, probed with anti-frataxin and anti-RPL13 antibodies. The position of molecular mass markers is indicated at left. D, levels of H3K9 acetylation as measured by ChIP upstream (UPGAA) and downstream (DOWNGAA) of the GAA·TTC repeats in the same iPSC lines as in B.

The three isogenic iPSC lines, FXN ^exp/exp, FXN ^exp/6, and FXN ^6/6, were differentiated by dual SMAD inhibition (38, 45) into neuronal cells that expressed β-III tubulin (Fig. S3A) and other neuronal markers such as HUC and MAP2 (as measured by qRT-PCR, Fig. S3B). Similarly to the isogenic iPSCs, the isogenic corrected neurons showed a restoration of FXN gene expression when compared with FRDA neurons (Fig. S3C) and absence of repressive chromatin marks near the repeats (Fig. S3D). The analysis of eight histone postsynthetic modifications (20) by ChIP at the FXN promoter and in the regions upstream and downstream of the GAA·TTC repeats shows increased acetylation and decreased methylation in the corrected FXN ^6/6 neurons compared with the isogenic FXN ^exp/exp FRDA neurons (Fig. S3D). Transcriptomic analysis of these isogenic β-III tubulin–positive cells was performed (data not shown) and has been deposited in the Sequence Read Archive (accession PRJNA495860).

Transcriptional profiling of FRDA iPSC-derived sensory neurons

Signature neuropathological elements of FRDA are the thinning of dorsal root fibers of the spinal cord due to progressive loss of large DRG sensory neurons and low myelination of peripheral sensory nerves (5). We therefore sought to derive SNs from FRDA and isogenic control iPSCs to perform transcriptional profiling by RNA-Seq. We adapted published protocols (46, 47) that involve the forced expression of two transcription factors BRN3A and NGN1 and the use of small molecules that have been shown to induce the differentiation of sensory neurons of the nociceptor type. The combination of the two protocols (see “Experimental procedures”) produced a more efficient differentiation than the two individual methods, as measured by the expression of the sensory neuron marker ISL1 (data not shown). Immunostaining analysis at day 11 after differentiation showed that these cells express β-III tubulin and peripherin (Fig. 4A). We also detected similar levels of expression of the SN marker ISL1 and NTRK3 (TRKC), together with POU4F1 (BRN3A), in both the FXN ^exp/exp and FXN ^6/6 lines, as measured by qRT-PCR (Fig. 4B). Transcriptomic analysis was performed as described above to compare 11-day-old sensory neurons derived from the FXN ^exp/exp and FXN ^6/6 lines and their corresponding iPSCs. PCA plot of these samples shows clustering dependent on cell identity and FXN expression (Fig. 5, A and B). When comparing iPSCs to sensory neurons from both patient and its isogenic control, over 12,000 DE genes for each comparison were identified (Fig. 5C), denoting the expected major changes in the transcriptome upon differentiation. The expression of a panel of sensory neuron markers in iPSCs versus SNs, as measured by RNA-seq voom-transformed values, confirms the identity of these cells (Fig. 4D). Na_v1.7 (SCN9A), TRPV1, and P2RX3, some of the most selectively expressed ion channels in human DRG (48), were up-regulated in these neurons, together with SLC17A7 and SLC17A6 (glutamate transporters VGLUT1 and VGLUT2), whereas markers of fetal brain (SATB2 and POU3F2), astrocytes (GFAP), cerebellum (EOMES), melanocytes (MITF), and Schwann cells (MPZ) were not highly expressed (Fig. 4D).

Figure 4.

iPSC-derived sensory neuron characterization. A, immunofluorescence staining of 11-day-old sensory neurons with the pan neuronal marker β-III tubulin (red) and the peripheral nervous system marker peripherin (green). Nuclei were counterstained with DAPI. Scale bar, 50 μm. B, expression of FXN, the pluripotency marker OCT4, and sensory neuron markers POU4F1, ISL1, and NTRK3 as measured by qRT-PCR in iPSCs and SNs. C, heatmap showing expression of a panel of sensory neuron markers in isogenic iPSC lines (both FRDA and control, right) and isogenic SNs (both FRDA and control, left). Shown is also the expression of two housekeeping genes (ACTB and GAPDH) together with markers of other tissues and cell types: SATB2 and POU3F2 (fetal brain), EOMES (cerebellum), GFAP (astrocytes), MITF (melanocytes), MPZ (Schwann cells).

Figure 5.

Transcriptomic analysis of FRDA and unaffected iPSC-derived sensory neurons. A, frataxin expression in isogenic iPSCs and SNs, plotted as voom-transformed values from gene counts. B, PCA plot of samples. C, number of differentially expressed genes with FDR <0.01, for the comparisons shown. Green (left panels) are down-regulated genes, and red (right panels) are up-regulated genes. D, volcano plots of DE genes in FRDA versus control SNs (left) and FRDA versus control iPSCs. The x axis shows the log fold change, and the y axis shows the −log₁₀ p values for each gene. Points in green represent genes that are significant at FDR <0.01, and points in red represent genes that were not significant. E, Venn diagrams of up-regulated (left) and down-regulated (right) DE genes common to CNS and PNS neurons. F, pathway analysis of shared DE genes between CNS and PNS neurons.

Notably, when comparing FRDA and unaffected cells, only 855 genes differed in iPSCs, whereas 4886 were differentially expressed in SNs (Fig. 5, C and D). This could represent an expansion of the FRDA signature in a relevant cell type but also be the result of differentiation-induced variability in the two lines, as reported in a recent study (40). We identified 2431 genes that were down-regulated and 2455 genes that were up-regulated in FXN^exp/exp SNs compared with their isogenic controls, and among these, 678 and 737 genes, respectively, were common to CNS neurons (Fig. 5E). Top enriched pathways, identified within the GO biological process database, are reported in File S1. Interestingly, the common down-regulated genes are uniquely enriched for regulation of apoptosis pathways, whereas ECM organization-related terms were also identified within the brown module (Figs. 2D and 5F). Up-regulated genes that are common to CNS and the peripheral nervous system (PNS) neurons are over-represented in genes for axonogenesis (Fig. 5F). Moreover, these same pathways were identified in genes reverted toward normalization by HDACi 109 treatment (File S1).

Gene expression signature of FRDA

Our RNA-Seq studies of iPSC-derived neurons provide evidence of metabolic changes in FRDA cells versus controls. These changes are small but involve multiple aspects of cell metabolism. FRDA neurons show a decrease in mitochondrial protein transcript levels, such as components of the ATP synthase complex and of complex I (black module), all of which have been described in FRDA (23, 27, 49–51). We observe consistent changes in ECM organization, focal adhesion, and related signaling (brown and magenta modules and SNs), possibly linked to cytoskeletal abnormalities that have been reported in FRDA cells (52–54). We also identify changes in chemical synapsis transmission (turquoise module, Fig. 1D) and axonogenesis (SNs). Finally, we detect widespread changes in the regulation of transcription (pink module). This is very intriguing because metabolism and signaling events are intimately entwined, and recent evidence indicates that the two regulate each other (55–57). Remarkably, the comparison of DE genes in CNS and sensory neurons provides evidence of involvement of apoptotic processes. Taken together, these transcriptional changes represent a gene expression signature of frataxin loss and, possibly, of the disease.

Discussion

It is widely accepted that the loss of frataxin is the causative event of FRDA pathophysiology, albeit other genetic modifiers of disease severity could exist. Despite this apparent simple etiology, the mechanisms of disease pathogenesis remain unclear. This is in part because although frataxin has been shown to be an essential component of Fe-S cluster assembly (6), its function is not yet fully understood. Defects in Fe-S proteins like aconitase and respiratory chain complex components, together with oxidative stress induced by iron overload, defects in lipid metabolism, cytoskeleton assembly, and heme biosynthesis, have been implicated in FRDA pathophysiology (21, 22, 58, 59). The aim of this study was to further disease understanding, by providing transcriptome analysis of a relevant affected tissue in FRDA. The compelling hypothesis that the tissue specificity of human disease is linked not simply to the expression of a particular gene, but to a “module” or subnetwork of genes that all have to be expressed in the same tissue in order for the disease to manifest (60), renders transcriptional profiling studies essential to understand the tissue specificity of FRDA.

Examples of studies employing iPSC-derived disease-relevant cells are ample in the literature (36, 61–63), but the use of these models also uncovers numerous pitfalls such as the lack of maturity or sign of aging in these cells (64–67) and the variability introduced by genetic background and differentiation (40, 68). We used relatively “young” or immature cells, focusing on a recent report that implicates hypoplasia rather than atrophy in the demise of the DRG (5). According to this hypothesis, young and immature iPSC-derived neurons could provide information on early disease mechanisms. Our FRDA CNS neurons do not show any phenotypical deficit like the ones described for other FRDA cell types or animal models (27, 50, 51, 53, 69–78). Complex I, complex III, and aconitase activity were all similar to unaffected neurons and so were oxygen consumption, spare respiratory capacity, ATP production, reactive oxygen species formation, and mitochondrial membrane potential (data not shown). Others have shown reduction in mitochondrial membrane potential (79), increased oxidative stress, and decreased levels of Fe-S cluster-containing and lipoic acid-containing proteins (80) in iPSC-derived FRDA neurons compared with controls. We therefore expected the transcriptional changes to be subtle. To distinguish between disease-related changes and noise introduced by line-to-line variability during differentiation, we treated both unaffected and FRDA neurons with the HDACi 109, thus restoring FXN transcription in the FRDA neurons (Fig. 1A) (20). We reasoned that changes that are a true consequence of the disease state would also be reverted by treatment with 109. In fact, over 50% of DE genes were reverted toward normalization by 109 treatment, similarly to what we have shown previously (18, 30). WGCNA also allowed us to focus pathway analysis on modules of genes that shared similar expression profiles. This led to cleaner and more interpretable gene set annotation, as genes in the same pathway tend to show highly correlated gene expression patterns.

As noted above, our RNA-Seq data revealed subtle changes in expression of genes involved in cell metabolism. Changes in metabolism, like glucose utilization, can translate to changes in protein post-translational modifications, such as glycosylation and acetylation, and these modifications are sensors in the cell of nutrient availability (55). For example, acetyl-CoA is highly compartmentalized, and although its levels are not limiting in the cell, fluctuations in local availability can result in changes in protein acetylation and epigenetic alterations (81) and can account for changes in the regulation of transcription identified in the pink module (Fig. 2D). It is notable that hyperacetylation of mitochondrial proteins occurs in the heart of a conditional Fxn-knockout mouse model of FRDA (82, 83). Changes in transcription regulatory activity were also detected in the heart and cerebellum of an inducible mouse model of frataxin deficiency (FRDAkd mouse (27)). As noted above, we also identified pathways related to chemical synapsis transmission (turquoise module, Fig. 1D). Up-regulation of synapsis-related genes was also reported in the FRDAkd mouse (27). Axonogenesis was the most enriched term for up-regulated genes shared between the CNS and sensory neurons. This term was also enriched in the turquoise module, but not as significantly (data not shown). Whether this up-regulation is a compensatory event (for example because of defects in ECM–cytoskeleton interactions that could cause axonal retraction) or a result of variability induced by differentiation (40) remains to be determined.

As shown above, the expression of more than half of the DE genes between FRDA and unaffected neurons changes toward normalization upon 109 treatment, although only about 10% of these changes are statistically significant. Although we detected 3419 DE genes when comparing DMSO and HDACi 109-treated FRDA cells (with FDR >0.01), only four genes were significantly different in unaffected cells (Fig. 1C). This is unlike previous results, where we showed that hundreds of genes were changed upon HDACi 106 treatment (a reverse amide of HDACi 109 (84)) in unaffected lymphocytes (30) and is probably due to line-to-line variability in the response to HDACi (see PCA plot in Fig. 1B), as well as the biological differences between iPSC-derived neurons and blood cells. Among the most significant changes upon 109 addition were genes involved in regulated exocytosis (Fig. 2D, magenta module) and regulation of transcription (pink module) but not mitochondrial proteins (black module). It is conceivable that a more prolonged 109 treatment and sustained increase in frataxin protein are necessary to see more extensive correction of the transcriptional changes.

To address the concern of line-to-line variability due the different genetic backgrounds, we created isogenic iPSC lines. We elected to utilize homologous recombination as opposed to nuclease-based methods (such as ZFNs, TALENs, and CRISPR–Cas9) because of the nature of the FXN gene near the GAA·TTC repeats. Because the repeats are embedded in an Alu element (85), selection of nuclease sites is problematic. Moreover, it is not known whether the sequence flanking the repeat might be important for FXN gene expression or RNA processing. For example, epigenetic indicators of regulatory regions like H3K27ac peaks are detectable upstream of the repeat.⁶ We selected HdAV because of its large capacity to incorporate genetic correction material (up to 30 kb) and no possibility of nuclease-mediated off-targeted cleavage (41–44). This methodology utilizes a “helper” virus that aids in packaging the desired vector into replication-deficient adenoviruses. The packaged adenovirus possesses no ability of replication but retains the ability of infection and highly efficient delivery of its genetic content. We used two rounds of homologous recombination to correct both alleles in FXN ^exp/exp. The resulting FXN^6/6 iPSC line showed a restoration of FXN transcription and histone acetylation around the GAA·TCC repeats, comparable with that of unaffected cells. This unequivocally demonstrates that the repeat expansion is solely responsible for the transcriptional defect and that acetylation of histones in this region is determined by repeat length.

Given that the peripheral nervous system, and in particular the DRG, is the initial site of neurodegeneration, it was important to perform additional transcriptional profiling by RNA-seq using the most relevant cellular model, noting that β-III tubulin–positive neurons differentiated by dual SMAD inhibition represent a model of the central nervous system (45). We therefore derived SNs from FRDA and isogenic control iPSCs by adapting protocols from Baldwin and co-workers (46) and Studer and co-workers (47) that involve the forced expression of two transcription factors, BRN3A and NGN1, and the use of small molecules that have been shown to induce the differentiation of sensory neurons of the nociceptor type. The sensory neuron identity of these cells is demonstrated by overexpression of SN markers like ISL1 and POU4F1 (BRN3A) mRNAs, detected both by qRT-PCR and RNA-seq (Fig. 4). Receptors specific to SN subtypes (TRKA-C) are all overexpressed compared with iPSCs, although we could not detect the expression of RUNX3, which controls the development of proprioceptive, TRKC-positive neurons (86). Transcriptomic analysis showed that other markers of sensory neurons are expressed in these neurons like Na_v1.7 (SCN9A), TRPV1, and P2RX3 (Fig. 4D). These genes encode some of the most selectively expressed ion channels in the human DRG (48). Markers of fetal brain (SATB2 and POU3F2), astrocytes (GFAP), cerebellum (EOMES), melanocytes (MITF), and Schwann cells (MPZ) are either very poorly expressed or unchanged during SN differentiation.

Despite the creation of isogenic lines being expected to reduce differences associated with diverse genetic backgrounds, we anticipated that we would still detect some level of differentiation variability in the two isogenic lines.

Almost 30% of the DE genes in sensory neurons were common to CNS neurons, and they represent some of the same pathways that were discussed above. These common genes, however, were uniquely enriched for genes involved in the regulation of apoptosis. Involvement of apoptotic mechanisms has been previously reported in FRDA cells (53, 70, 72, 87, 88). This suggests that to overcome the pitfalls of studies centered on iPSC-derived cell models, both multiple cell lines and isogenic lines are necessary. Genes that are specifically dysregulated in SNs but not CNS neurons could represent a true disease signature rather than the result of changes that compensate for frataxin deficit in cell types that do not appear to be affected in the disease. Pathway analysis of these genes showed enrichments very similar to the genes in the WGCNA modules. This could indicate that, because of the immature nature of these cells, the same transcriptional changes are occurring in the two neuronal types. The question remains whether and how any of the identified changes in gene expression contribute to pathogenesis. All or some could be compensatory mechanisms of frataxin deficiency and an adaptive response to metabolic changes (because no overt phenotype was detectable). It is conceivable that, at some point during development or aging, this adaptive response could become ablated and the cell overwhelmed by frataxin loss, and neurodegeneration occurs.

We hypothesize that failure to observe an FRDA phenotype in iPSC-derived neuronal cells could be due to the immature nature of such cells, and the fact that most FRDA patients present with symptoms between the ages of 8 and 15 years, with no symptoms at birth. Because generation of iPSCs from patient fibroblasts resets the developmental clock back to the embryonic stem cell stage, it is not that surprising that iPSC-derived cells fail to recapitulate the degenerative hallmarks of FRDA. Several approaches are currently available to generate “aged” neurons from either donor fibroblasts (89) or forced aging of iPSC-derived cells by expression of progerin (64). Future studies in FRDA disease modeling could use these methods to attempt to recapitulate an FRDA phenotype in vitro.

Experimental procedures

Neuronal differentiation

iPSC line generation and characterization were described previously (37). Neuronal differentiation was performed as described (38) with minor modifications. Briefly, iPSCs grown on Matrigel (Corning) and mTeSR (Stem Cell Technologies, Vancouver, Canada) were treated for 10 days in E6 medium (Stem Cell Technologies, Vancouver, Canada) with 0.5 μm LDN-193189, 10 μm SB431542, and 20 μg/ml FGF2. iPSC colonies were dissociated with Accutase (Innovative Cell Technologies, San Diego) and plated in AggreWell plates (Stem Cell Technologies, Vancouver, Canada) at a density of 1000 cells per microwell. Neurospheres were grown in suspension for 1 week in Neurobasal A medium (ThermoFisher Scientific, Waltham, MA), supplemented with N2 and B27 supplements (ThermoFisher Scientific) and FGF2 and EGF, both at 20 μg/ml. Neurospheres were then plated on Matrigel and grown in the same medium as above until rosettes appeared. Rosettes were manually isolated, grown for 4–7 days in suspension, then dissociated with Accutase and plated on Matrigel at 200,000 cells/cm². To induce neuronal differentiation, cells were grown in the same medium as above without FGF2 and EGF for 14 days.

For sensory neuron differentiation, iPSCs were plated at 50,000 cells/cm² in mTeSR and infected with lentiviruses expressing a bicistronic construct of POU4F1 (Brn3a) and Neurog1 (Ngn1) under the control of a doxycycline-inducible promoter and the transactivator rtTA (90) (see under “Plasmids”). The next day (day 1), medium was changed to mTeSR supplemented with 5 μg/ml doxycycline. On day 2, medium was changed to N2 medium (46) with 0.5 μm LDN-193189, 10 μm SB431542, and 5 μg/ml doxycycline. On day 4, medium was changed to N2 medium with 0.5 μm LDN-193189, 10 μm SB431542, 5 μm CHIR99021, 10 μm SU5402, 10 μm DAPT, and 5 μg/ml doxycycline. On day 5, medium was changed to N3 medium (46) with 5 μm CHIR99021, 10 μm SU5402, 10 μm DAPT, and 5 μg/ml doxycycline. On day 7, medium was changed to N3 medium with 10 μm SU5402 and 10 μm DAPT. On day 10, medium was changed to Neurobasal A supplemented with N2 and B27. Cells were collected at day 12 for total RNA isolation.

RNA-Seq, differential expression, and WGCNA

RNA was isolated using the RNeasy mini kit (QIAgen, Hilden, Germany). Single-end 75-bp reads were generated by the NextSeq (Illumina, San Diego) located at the Scripps Next Generation Sequencing Facility. Image analysis, base calling, and demultiplexing was done using bcl2fastq. Cutadapt is used to trim the adapter and low bp called scores.

FASTQ files were aligned to BAMs and counts were generated using STAR (91). Raw counts were transformed to counts per million using limma voom (92) and were normalized for library size using TMM normalization implemented by edgeR (93). Differential expression was analyzed using linear models with limma (94), and no additional covariates were added to the models. Network analysis was run with the WGCNA package (95), using a soft power threshold of 7 for the adjacency matrix and a minimum module size of 20. Modules with a correlation higher than 0.8 were merged. Gene sets analyzed for enrichment analysis were submitted to Enrichr (96, 97), with pathways being identified in the GO Biological Process database and cell compartments from GO Cellular Component.

Plasmids

HdAV–FXN plasmid

A neomycin cassette containing the neomycin resistance gene flanked by Frt sequences under the control of bacterial and mammalian promoters was amplified from plasmid pgk-gb-frt-neo-frt together with 50-bp homology arms from the FXN gene and inserted by recombineering (98) 108 bp downstream of the GAA·TTC repeats in the bacterial artificial chromosome plasmid RP11-265B. 18.7 kb of this construct containing the human FXN gene locus with six GAA·TTC repeats and the neomycin cassette was cloned in pGEX-4T1 by recombineering (98). To construct the HdAV–FXN plasmid, the 18.7-kb fragment was isolated by restriction digestion and subcloned in the adenoviral plasmid pCIHDAdGT8-3.

Bicistronic plasmid expressing Brn3a and Ngn1

Mouse Neurog1 cDNA (83% homologous to human cDNA peptide) was cloned without the stop codon and replaced with an F2A self-cleaving peptide sequence. Human BRN3A cDNA (POU4F1) was added directly after the F2A peptide to generate an Ngn1–F2A–Brn3a bicistronic sequence. The bicistronic sequence was inserted into a lentiviral construct under the control of the tetracycline operator, as described previously (99). Replication-incompetent VSVg-coated lentiviral particles were packed in HEK293T cells (ATCC), collected 48 h after transfection, and filtered through a 45-μm membrane before use.

HdAV virus particle production

Helper virus particles were produced by infecting HEK293T cells (multiplicity of infection of 3). After 48–72 h of incubation when the cytopathic effect was greater than 90%, the cells were lysed by three freeze–thaw cycles, and the virus was purified by cesium chloride gradient centrifugation (100). The HdAV particles were assembled according to previously published methods (101). Briefly, the HdAV–FXN plasmid was transfected into 116 cells (a modified HEK293 cell line that expresses high levels of Cre recombinase (102)), and these cells were subsequently infected with the purified helper virus at a multiplicity of infection of 0.2–0.3. When 90% cytopathic effect was detected, the crude cell lysate was collected and use to re-infected 116 cells. Multiple rounds of infections were performed until the desired titer was achieved as measured by β-gal assay. The cell lysate was purified and concentrated by CsCl centrifugation (100).

PCR, qRT-PCR, quantitative Western blotting, and ChIP

GAA·TTC repeat length determination was performed as described previously (103). qRT-PCR and quantitative Western blotting were described previously (37). Primer sequences for mRNA quantification were as follows: for FXN transcript, 5′-CAGAGGAAACGCTGGACTCT-3′ and 5′-AGCCAGATTTGCTTGTTTGG-3′; for POU4F1 transcript, 5′-GGGCAAGAGCCATCCTTTCAA-3′ and 5′-CTGTTCATCGTGTGGTACGTG-3′; for ISL1 transcript 5′-GCGGAGTGTAATCAGTATTTGGA-3′ and 5′-GCATTTGATCCCGTACAACCT-3′ and for NTRK3 transcript 5′-ACGAGAGGGTGACAATGCTG-3′ and 5′-CCAGTGACTATCCAGTCCACA-3′. Primer sequences for MAP2, ELAVL3 (HUC), and OCT4 were described previously (37). The following antibodies were used in quantitative Western blot analysis: anti-frataxin (Santa Cruz Biotechnology, Dallas, TX); anti-RPL13 (Cell Signaling Technologies, Danvers, MA); and goat anti-mouse IR800 and goat anti-rabbit IR680, both from LI-COR Biosciences (Lincoln, NE). ChIP procedure and antibodies used are described previously (20).

Immunocytochemistry and FACS analysis

Immunocytochemistry and FACS analysis were described previously (37). Primary antibodies included anti-β-III tubulin (Tuj1; Covance, Princeton, NJ), anti-peripherin, anti-Oct4, anti-SSEA4, and anti-Tra-1–60 (all from Millipore, Temecula, CA). Secondary antibodies were anti-mouse Alexa 594 and anti-rabbit Alexa 488 (Invitrogen).

Author contributions

J.-I. L., J. M. G., and E. S. conceptualization; J.-I. L., D. N., and E. S. data curation; J.-I. L., D. N., F. G., G. C., J. M. G., and E. S. formal analysis; J.-I. L., D. N., L. P., E. C., and E. S. investigation; J.-I. L., D. N., and E. S. visualization; J.-I. L., L. P., E. C., and E. S. methodology; J.-I. L., J. M. G., and E. S. writing-original draft; D. N. and G. C. writing-review and editing; B. T. and K. K. B. resources; K. K. B., G. C., and J. M. G. funding acquisition.