Optic Atrophy 1 Controls Human Neuronal Development by Preventing Aberrant Nuclear DNA Methylation.

Optic atrophy 1 (OPA1), a GTPase at the inner mitochondrial membrane involved in regulating mitochondrial fusion, stability, and energy output, is known to be crucial for neural development: Opa1 heterozygous mice show abnormal brain development, and inactivating mutations in OPA1 are linked to human neurological disorders. Here, we used genetically modified human embryonic and patient-derived induced pluripotent stem cells and reveal that OPA1 haploinsufficiency leads to aberrant nuclear DNA methylation and significantly alters the transcriptional circuitry in neural progenitor cells (NPCs). For instance, expression of the forkhead box G1 transcription factor, which is needed for GABAergic neuronal development, is repressed in OPA1+/- NPCs. Supporting this finding, OPA1+/- NPCs cannot give rise to GABAergic interneurons, whereas formation of glutamatergic neurons is not affected. Taken together, our data reveal that OPA1 controls nuclear DNA methylation and expression of key transcription factors needed for proper neural cell specification.

Introduction

Mitochondria are double-membrane-bound subcellular organelles required for numerous cellular functions. In response to metabolic and physiological signals, mitochondria undergo fusion and fission, processes that are catalyzed by conserved GTP-hydrolyzing enzymes and their associated proteins (Chan, 2012). Fission of mitochondria is needed for their fragmentation. This is triggered by dynamin 1-like protein (DNM1L), a cytoplasmic protein that is recruited to the mitochondrial surface where it constricts mitochondrial tubules. Fusion of the outer membranes is catalyzed by mitofusin 1 and 2 (MFN1, MFN2), whereas fusion of the mitochondrial inner membranes is catalyzed by the mitochondrial dynamin-like GTPase optic atrophy 1 (OPA1). Due to RNA splicing and protein processing, different isoforms of OPA1 exist. The long isoforms localize to the mitochondrial inner membrane through their transmembrane domains, whereas the short isoforms are expressed as soluble proteins in the mitochondrial intermembrane space. A balance of long and short OPA1 isoforms is required to maintain a healthy mitochondrial network (Del Dotto et al., 2017, MacVicar and Langer, 2016).

OPA1 is crucial for tissue development. Opa1 heterozygous animals are viable, but experience a progressive loss of retinal ganglion cells (RGCs), optic nerve degeneration, and abnormal brain development (Davies et al., 2007, Alavi et al., 2007), whereas Opa1-null mice die early during embryonic development due to growth retardation and morphological abnormalities (Davies et al., 2007, Alavi et al., 2007). In humans, heterozygous mutations in OPA1 cause the most common form of autosomal dominant optic atrophy (ADOA), a neuropathy wherein the majority of patients experience impaired vision (Delettre et al., 2000, Alexander et al., 2000). Homozygous mutations in OPA1 result in severe and fatal infantile disorders with neurodevelopmental deficits, multi-organ complications, encephalopathy, cardiomyopathy, and optic atrophy (Spiegel et al., 2016, Nasca et al., 2017).

To elucidate the molecular mechanisms by which OPA1 contributes to human neural development, we used OPA1 haploinsufficient pluripotent stem cell lines and differentiated them into neural progenitor cells (NPCs) and forebrain neurons. Although we were able to generate NPCs and glutamatergic neurons, OPA1 haploinsufficiency interfered with GABAergic interneuron formation. We then explored the molecular changes associated with the observed altered neural cell specification and identified a novel function for OPA1.

Results

OPA1 Haploinsufficiency Induces Oxidative Stress in hESCs

To study the role of OPA1 during neural development, we used human embryonic stem cells (hESCs) and the CRISPR-Cas9 gene editing technology and deleted a stretch of nucleotides, which induces a frameshift and a premature stop codon in the second exon of the OPA1 transcript (Figure S1A). Because the majority of human disorders linked to OPA1 are caused by heterozygous mutations, we targeted one allele only. We found a 50% reduction in OPA1 mRNA transcript levels in heterozygous OPA1+/− compared with OPA1+/+ hESCs, confirming that only one OPA1 allele was transcribed (Figure 1A). The reduction in mRNA expression levels correlated with a 50% reduction in OPA1 protein levels in hESCs, indicating a non-sense-mediated RNA decay in OPA1+/− cells (Figures 1B and 1C). Expression levels of OCT4 and NANOG were comparable between OPA1+/− and parental hESCs, as shown by immunostaining and immunoblotting (Figures S1B and S1C). Moreover, hESC proliferation rates, assessed by automated live-cell analysis system, were similar between the two genotypes (Figure S1D). To further analyze cell cycle dynamics in OPA1+/− hESCs, we performed flow cytometry (fluorescence-activated cell sorting [FACS]) analysis of hESCs labeled with the fluorescent DNA-intercalating dye propidium iodide (PI) and Ki-67 antibody. The ratio of cells in G0/G1, S, and G2/M phases (Figures S1E and S1F), as well as Ki-67 labeling (Figure S1G), were similar between OPA1+/+ and OPA1+/− hESC, indicating that OPA1 haploinsufficiency does not affect pluripotency or proliferation rates in hESCs. Using transmission electron microscopy (TEM), we then assessed mitochondrial morphology and distribution in OPA1+/− hESCs. The TEM images showed no obvious differences in mitochondrial morphology between OPA1+/+ and OPA1+/− hESCs (Figure S2A). Next, we measured mitochondrial length in OPA1+/+ and OPA1+/− hESCs and binned the data into categories of intervals. Although this analysis revealed no significant change, we found a trend of more fragmented mitochondria in OPA1+/− hESCs (Figure S2B). We therefore determined the circumference of mitochondria, and found a small, but significant decrease in circumference in OPA1+/− compared with OPA1+/+ hESCs (Figure S2C). In addition, we found a significant increase in the intercristae distance in OPA1+/− compared with OPA1+/+ hESCs (Figures S2D and S2E). Mitochondria-targeted GFP (mito-GFP)-transduced hESCs also demonstrated that there was no obvious difference in mitochondrial shape or size among the two genotypes (Figure S2F). The mitochondrial DNA content, as analyzed by copy number of the mitochondrial genes NADH-ubiquinone oxidoreductase chain 1 (MT-ND1) and MT-ND4, was similar between OPA1+/+ and OPA1+/− hESCs (Figure S2G). Using mitochondrial respiration assays, we found similar ATP production and basal respiration rates (Figures 1D and 1E) and an increase in maximal respiration in OPA1+/− compared with OPA1+/+ hESCs (Figures 1D and 1E). Next, we analyzed whether reactive oxygen species (ROS) were altered in OPA1+/− hESCs. Flow cytometry analysis using the hydrogen peroxide-sensitive ROS indicator CM-H2DCFDA revealed significantly increased ROS levels in OPA1+/− compared with OPA1+/+ hESCs (Figures 1F and 1G). Taken together, our results reveal subtle changes in the mitochondrial morphology and increased levels of ROS in OPA1 haploinsufficient hESCs.

Figure 1

OPA1 Haploinsufficiency Induces Oxidative Stress in hESCs

(A) OPA1 mRNA expression levels in hESCs determined by qRT-PCR analysis.

(B) Representative immunoblotting images showing OPA1 protein levels in hESCs, week 3 NPCs, and neurons. Beta-actin was used as loading control.

(C) Densitometric analysis of immunoblots in OPA1+/+ and OPA1+/− cells. Mean ± SEM, N = 4 independent experiments.

(D) Oxygen consumption rate changes under mitochondrial stress in OPA1+/+ and OPA1+/− hESCs. Oligomycin, FCCP, antimycin, and rotenone were sequentially applied. Mean ± SEM, N = 2 independent experiments.

(E) Basal respiration rates, maximum respiration rates, and ATP production calculated in OPA1+/+ and OPA1+/− hESCs. Mean ± SEM, N = 2 independent experiments were performed and N ≥ 9 technical replicates were analyzed for each genotype. Student's t test was used to analyze the difference between two groups. n.s. not significant. ∗p < 0.05.

(F) Representative flow cytometry analysis of OPA1+/+ and OPA1+/− hESCs labeled with the hydrogen peroxide-sensitive oxidative stress indicator CM-H2DCFDA. Unlabeled hESCs (−) were used as negative control. hESCs that were incubated with 0.5 mM H2O2 were used as positive control.

(G) Quantification of flow cytometry measurements in OPA1+/+ and OPA1+/− hESCs. Arithmetic mean values were normalized to the values of labeled OPA1+/+ hESCs.

Mean ± SEM, N ≥ 3 independent experiments and N = 2 technical replicates. Student's t test was used to analyze difference between two groups. n.s. not significant. ∗∗p < 0.01, ∗∗∗p < 0.001. See also Figures S1, S2, and S4.

OPA1 Expression Is Upregulated during Neural Differentiation

Next, we investigated how OPA1 heterozygosity affects neural differentiation. Using the dual SMAD inhibition protocol (Chambers et al., 2009, Shi et al., 2012), hESCs were differentiated into neuroepithelial progenitors and NPCs (Figure S3A). On day 11 of the differentiation protocol, neural commitment was evident by morphological transformation and expression of the neural stem cell marker PAX6 (day 11 NPCs, Figure S3B). After passaging and further differentiation, PAX6-positive NPCs reorganized themselves into neural rosette structures (week 3 NPCs, Figure S3C) before being differentiated into beta-tubulin III-positive neurons (Figure S3D). To assess the neurogenic ability of hESCs, we first compared rosette formation of NPCs. Immunostainings showed that OPA1+/− NPCs formed tight junction protein Zonula-occludens-1 (ZO-1)-positive rosette structures with comparable efficiency as OPA1+/+ cells (Figure S3E). As the self-organization of neural rosettes is reminiscent of neural tube formation from neuroepithelium (Deglincerti et al., 2016), these results reveal that early neural development is not affected by OPA1 haploinsufficiency.

Quantification of OPA1 transcript levels by qRT-PCR showed a 25% increase of total OPA1 mRNA levels during neuronal differentiation (Figure S4A). OPA1 is alternatively spliced in different tissues. Of these splice variants, OPA1 transcript variant 1 (NM_015560), which lacks exons 4b and 5b (Figure S1A), is the most abundantly expressed isoform in the mouse brain and retina (Delettre et al., 2001, Akepati et al., 2008). In line with this, we found that OPA1 isoform 1 levels markedly increased in our neuronal cultures (Figure S4B). Moreover, we found that isoform 1 became the predominant splice form during neural differentiation. In hESCs, the ratio of transcript variant 1 to total OPA1 mRNA levels was approximately 1:1, whereas in neurons the majority of OPA1 transcripts were isoform 1 (Figure S4C). Immunoblotting experiments showed a 25% and 2-fold increase of OPA1 protein in OPA1+/+ NPCs and neurons when compared with hESCs, respectively (Figures 1B, 1C, and S4D).

OPA1 Heterozygous NPCs Fail to Generate DLX1/2+ Neurons

To investigate the role of OPA1 during differentiation and neuronal function further, we analyzed neurons derived from OPA1+/+ and OPA1+/− NPCs. Immunostaining experiments showed that pan-neuronal marker beta-tubulin III-positive neurons could be efficiently generated from both genotypes, indicating that OPA1 haploinsufficiency is not affecting overall neuronal development (Figure 2A). Patch-clamp experiments demonstrated that OPA1 haploinsufficiency did not interfere with the ability of neurons to fire action potentials in response to current stimuli (Figure 2B). To determine neuronal lineage specification, we selected T-box brain 1 (TBR1) as a marker for glutamatergic neurons (Hevner et al., 2001), and DLX1 and DLX2 as they are important for specification of RGCs and GABAergic neurons (Anderson et al., 1997, Letinic et al., 2002, de Melo et al., 2005). As expected, TBR1, DLX1, and DLX2 mRNA expression levels increased during the neuronal differentiation of OPA1+/+ hESCs (Figures 2C–2E), demonstrating that our neuronal cultures included cell types representing forebrain development. During neuronal differentiation of OPA1+/− hESCs, we found that TBR1 was expressed at similar levels as in OPA1+/+ cells (Figure 2C). However, OPA1+/− cells notably failed to induce mRNA expression of DLX1 and DLX2 during neuronal maturation (Figures 2D and 2E). Using immunostaining, we showed that there were significantly less DLX2 immunoreactive cells in OPA1+/− compared with OPA1+/+ neuronal cultures (Figures 2F and 2G). The absence of DLX2 protein in lysates of OPA1+/− but not OPA1+/+ neurons was further confirmed using immunoblotting (Figure 2H). Neurons derived from both genotypes expressed comparable levels of beta-tubulin III protein (Figure 2H).

Figure 2

OPA1 Heterozygous NPCs Fail to Generate DLX1/2+ Neurons

(A) Beta-tubulin III (TUJ1) staining of neurons. Scale bar, 100 μm.

(B) Representative traces of patch-clamped neurons. Action potentials were elicited in OPA1+/+ and OPA1+/− neurons upon depolarizing current injection. Black bar indicates time of depolarizing current injection.

(C–E) (C) TBR1, (D) DLX1, and (E) DLX2 mRNA levels determined by qRT-PCR analyses in OPA1+/+ and OPA1+/− hESCs, NPCs, and neurons. Mean ± SEM, N ≥ 3 independent experiments for each genotype and cell type.

(F) Immunostaining images of OPA1+/+ and OPA1+/− neurons. DAPI was used to stain nuclei. Scale bar, 100 μm.

(G) Quantification of DLX2+ neurons. Mean ± SEM, N = 3 independent experiments.

(H) Immunoblotting showing OPA1, DNM1L, GAD1/GAD67, DLX2, and beta-tubulin III protein expression in OPA1+/+ and OPA1+/− hESCs and neurons. Beta-actin was used as loading control.

Student's t test was used to analyze difference between two groups. n.s. not significant. ∗p < 0.05, ∗∗∗p < 0.001. See also Figures S3 and S4.

OPA1 Haploinsufficient NPCs Do Not Give Rise to GABAergic Interneurons

DLX1 and DLX2 are transcription factors (TFs) important for GABAergic interneuron and RGC development (Letinic et al., 2002, Anderson et al., 1997, de Melo et al., 2005). As their downregulation indicated an altered neuronal function and differentiation potential when OPA1 was reduced, we performed RNA sequencing (RNA-seq) in 7-week-old neurons. We found several genes encoding for TFs with known function in neurogenesis differentially regulated in OPA1+/− neurons (Table S1). For instance, FOXG1, DLX1, DLX2, DLX5, and DLX6 were genes highly downregulated in OPA1+/− compared with OPA1+/+ neurons (Figure 3A; Table S1). Moreover, expression of SFRP1 and SEMA5A, which are known to control RGC development and axonal growth, as well as CTIP2/BCL11B and NEUROD6, which are TFs that are highly expressed in the developing brain and retinal cells (Cherry et al., 2011), were also significantly downregulated. Other genes downregulated included SLITRK2, SLITRK4, and SLITRK6, which are plasma membrane proteins expressed in the retina, and several GABA receptor subunits (GABRA2, GABRA4, GABRB1, and GABRG1). Among the upregulated genes in OPA1+/− neurons we found NR2F1, NR2F2, and NEUROD4, which are TFs known to regulate cell fate specification in the retina (Figure 3A).

Figure 3

OPA1 Haploinsufficient NPCs Do Not Give Rise to GABAergic Interneurons

(A) Log2 fold-change and expression heatmaps of selected down- (green) and up- (red) regulated genes in OPA1+/− compared with OPA1+/+ neurons. FC = fold-change. Expression intensities are displayed from blue (low expression) to yellow (high expression).

(B) Immunoblotting showing OPA1, DNM1L, TBR1, and GAD1/GAD67 protein expression in OPA1+/+ and OPA1+/− hESCs and neurons. OCT4 was used as pluripotency marker; beta-actin was used as loading control.

(C) Densitometric analysis of GAD1 protein levels in OPA1+/+ and OPA1+/− neurons. Mean ± SEM, N = 2 independent experiments.

(D) Immunoblotting showing GAD2/GAD65 and beta-tubulin III protein expression in OPA1+/+ and OPA1+/− hESCs and neurons.

(E) Densitometric analysis of GAD2 protein levels in OPA1+/+ and OPA1+/− neurons. Mean ± SEM, N = 2 independent experiments.

Student's t test was used to analyze difference between two groups. ∗∗∗p < 0.001. See also Figures S3 and S4, and Table S1.

The absence of DLX1/2 expression in OPA1+/− neurons prompted us to further investigate GABAergic neuronal markers. Glutamate decarboxylase 1 (GAD1/GAD67) and 2 (GAD2/GAD65), which catalyze the production of GABA from L-glutamic acid, were expressed in OPA1+/+ neurons, whereas they were absent or markedly reduced in OPA1+/− neurons (Figures 3B–3E and S4E). TBR1 and beta-tubulin III were expressed at comparable levels (Figures 3B and 3D). Overall, our transcriptional analysis demonstrates that reduced OPA1 protein levels alter the expression of genes that are important for GABAergic interneuron formation and retinal development.

We imaged mito-GFP-transduced neurons to assess mitochondrial morphology. Confocal microscopy analysis demonstrated that mitochondrial morphology and distribution were similar in OPA1+/− and OPA1+/+ neurons (Figure 4A). Quantification of mitochondrial length and binning of measurements into categories of intervals showed no evident differences in mitochondrial distribution (Figures 4B and 4C). Next, we analyzed energy metabolism in OPA1+/− and OPA1+/+ neuronal cultures using mitochondrial respiration assays. Our data revealed that neurons exhibited an increased mitochondrial respiration compared with hESCs (Figure 4D). However, basal respiration, maximal respiration, and ATP production rates were similar in OPA1+/− and OPA1+/+ neurons (Figure 4E), suggesting that OPA1 haploinsufficiency does not affect mitochondrial respiration in neurons. We then investigated oxidative stress levels in neuronal cultures. Flow cytometry analysis showed significantly increased ROS levels in OPA1+/− compared with OPA1+/+ neurons (Figures 4F and 4G). FACS analysis of neurons labeled with the early apoptosis indicator Annexin V and late apoptosis indicator PI (Figure 4H) demonstrated no differences in apoptosis in neuronal culture conditions (Figures 4I and 4J). Importantly, when neurons were exposed to stress by pre-treatment with hydrogen peroxide we observed significantly more Annexin V+ cells in OPA1+/− compared with OPA1+/+ neurons (Figures 4I and 4J).

Figure 4

OPA1 Haploinsufficiency Induces Oxidative Stress and Apoptosis in Neurons

(A) Confocal microscopic images of mito-GFP-transduced OPA1+/+ and OPA1+/− neurons. ∗ denotes nucleus. Scale bar, 10 μm.

(B) Quantification of mitochondria length in OPA1+/+ and OPA1+/− neurons. N = 680 mitochondria for OPA1+/+ and N = 645 mitochondria for OPA1+/− were measured.

(C) Mitochondrial length was measured and binned into categories of length intervals.

(D) Oxygen consumption rate changes under mitochondrial stress in OPA1+/+ and OPA1+/− neurons. Oligomycin, FCCP, antimycin, and rotenone were sequentially applied. Mean ± SEM, N = 2 independent experiments.

(E) Basal respiration rates, maximum respiration rates, and ATP production calculated in OPA1+/+ and OPA1+/− neurons. Mean ± SEM, N = 2 independent experiments were performed and N ≥ 40 technical replicates were analyzed for each genotype.

(F) Representative flow cytometry analysis of OPA1+/+ and OPA1+/− neurons labeled with the hydrogen peroxide-sensitive oxidative stress indicator CM-H2DCFDA. Unlabeled neurons (−) were used as negative control. Neurons that were incubated with 0.5 mM H2O2 were used as positive control.

(G) Quantification of flow cytometry measurements in OPA1+/+ and OPA1+/− neurons. Arithmetic mean values were normalized to the values of labeled OPA1+/+ neurons. Mean ± SEM, N ≥ 3 independent experiments and N = 2 technical replicates.

(H) Representative flow cytometry analysis of live OPA1+/+ and OPA1+/− neurons labeled with the early apoptosis indicator annexin V and late-apoptosis indicator propidium iodide (PI). Neurons were incubated with 50 mM H2O2 to assess sensitivity to apoptotic stimuli.

(I) Quantification of early apoptotic annexin V+/PI− neurons. Percentage of labeled cells in total gated cells is shown.

(J) Quantification of late-apoptotic Annexin V+/PI+ neurons. Percentage of labeled cells in total gated cells is shown.

Mean ± SEM, N = 4 independent experiments and N = 2 technical replicates. Student's t test was used to analyze difference between two groups. n.s. not significant. ∗p < 0.05, ∗∗∗ p < 0.001. See also Figures S3 and S4.

OPA1 Heterozygous NPCs Exhibit a Significantly Altered Transcriptional Signature

Next, we addressed whether the impaired formation of DLX1/2+ neurons in OPA1+/− cultures could be explained by an altered transcriptional circuitry at the NPC level. Successful formation of day 11 NPCs was verified by the loss of the pluripotency markers OCT4 and NANOG, and the gain of the neural stem cell marker PAX6 (Figure S5A). We then performed RNA-seq analysis in OPA1+/+ and OPA1+/− NPCs after 11 days and 3 weeks of differentiation to elucidate potential transcriptional changes associated with OPA1 haploinsufficiency. Principal-component analysis was used to assess and visualize the neural differentiation trajectory (Figure 5A). When we analyzed differentially expressed genes (DEGs) at day 11 OPA1+/− versus OPA1+/+ NPCs, we found FOXG1 already highly downregulated at this early NPC stage (Figure 5B and Table S2). Another gene we found highly downregulated encodes for TAF9B (Figure 5B and Table S2), a protein that is part of the neuronal core promoter complex and known to regulate the induction of specific neuronal genes (Herrera et al., 2014). Moreover, we found SLITRK2, a transmembrane protein expressed in proliferative neuroblastic layer of the retina (Beaubien and Cloutier, 2009), to be significantly downregulated in OPA1+/− NPCs (Figure 5B and Table S2). In addition, GABRA2, the gene coding for GABRA receptor alpha 2; ATP10D, a putative phospholipid transporting ATPase; and several neural cell adhesion genes, including NCAM2, PCDHGA3, and PCDHGA6, were also downregulated (Figure 5B and Table S2). Significantly upregulated genes in day 11 OPA1+/− versus OPA1+/+ NPCs included genes coding for neural guidance proteins, such as SEMA3E, SLIT3, Ephrin type-A receptor 1 (EPHA1), and its ligand ephrin A1 (EFNA1) (Figure 5B and Table S2).

Figure 5

OPA1 Haploinsufficiency Significantly Alters the Transcriptional Circuitry in NPCs

(A) Principal-component analysis of hESCs, day 11 and week 3 NPCs, and neurons.

(B) Log2 fold-change and expression heatmaps of selected down- (green) and up- (red) regulated genes in OPA1+/− compared with OPA1+/+ day 11 NPCs. FC = fold-change. Expression intensities are displayed from blue (low expression) to yellow (high expression).

(C and D) (C) PAX6 and (D) FOXG1 mRNA levels in day 11 and week 3 NPCs, normalized to transcript levels in OPA1+/+ hESCs.

(E) Immunoblotting showing OPA1, FOXG1, and PAX6 protein expression in NPCs. Beta-actin was used as loading control. N = 3 independent experiments.

(F) ATP10D, GABRA2, SLITRK2, TAF9B, and ZNF280D mRNA expression levels in day 11 NPCs. mRNA expression levels were normalized to transcript levels in OPA1+/+ NPCs.

(G) GRHL2, GRHL3, SIX6, VAX1, and VSX2 mRNA expression levels in day 11 NPCs. mRNA expression levels were normalized to transcript levels in OPA1+/+ NPCs.

Mean ± SEM, N ≥ 4 independent samples for each genotype and cell type. Student's t test was used to analyze the differences between two groups. n.s. not significant. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. See also Figures S3, S5, and S6, and Tables S2 and S3.

DEG analysis of OPA1+/− compared with OPA1+/+ NPCs in week 3 revealed that FOXG1, ATP10D, GABRA2, SLITRK2, ZNF280D, and TAF9B also remained highly downregulated at this later stage of differentiation (Figure S5B and Table S3). In addition, several genes encoding for TFs involved in primary neurulation, forebrain development, and retinal development were upregulated in OPA1+/− NPCs. These included SIX6, GRHL2, GRHL3, VAX1, and VSX2 (Figure S5B and Table S3). Overall, our RNA-seq analyses identified significant alterations in the transcriptional circuitry of OPA1+/− NPCs (Figures 5B and S5B, Tables S2 and S3), which can explain the observed altered neuronal cell fate specification.

We then assessed the expression levels of selected genes in more detail. qRT-PCR analysis at three different time points showed continuous increase of PAX6 mRNA expression levels during NPC derivation (Figure 5C). OPA1+/+ NPCs also showed significant upregulation of FOXG1, starting 11 days after differentiation (Figure 5D), a finding in line with its previously described critical role in early brain development (Xuan et al., 1995). OPA1+/− NPCs expressed similar PAX6 mRNA levels as OPA1+/+ NPCs (Figure 5C), whereas FOXG1 transcription was absent in OPA1+/− cells during all stages of NPC differentiation (Figure 5D). This was also true at the protein level (Figure 5E). In addition, we used qRT-PCR analyses to confirm the downregulation of ATP10, GABRA2, SLITRK2, TAF9B, and ZNF280D (Figure 5F) and upregulation of GRHL2, GRHL3, SIX6, VAX1, and VSX2 mRNA expression in OPA1+/− compared with OPA1+/+ NPCs (Figure 5G). Using a second, independently derived OPA1+/− hESC clone #63, we confirmed reduced OPA1 protein (Figure S6A) and OPA1 mRNA expression levels in OPA1+/− compared with OPA1+/+ hESC and NPCs, respectively. Moreover, PAX6 mRNA expression was comparable between genotypes (Figure S6B), whereas FOXG1 and TAF9B mRNA expression levels were also markedly reduced in OPA1+/− #63 NPCs (Figure S6B).

To investigate whether altered transcriptional regulation affects proliferation or cell survival of OPA1+/− NPCs, we performed FACS analysis of NPCs labeled with PI. Figures S5C and S5D showed no differences in cell cycle dynamics in OPA1+/+ and OPA1+/− NPCs. Similarly, Ki-67 expression was also comparable between the two genotypes (Figure S5E). Next, we analyzed the levels of apoptosis in NPCs under normal conditions and upon stimulation with hydrogen peroxide (Figure S5F). We found no significant difference in apoptosis between OPA1+/+ and OPA1+/− NPCs (Figure S5G), whereas pre-treatment of NPCs with hydrogen peroxide increased Annexin V+/PI+ late apoptotic cells in OPA1+/− NPCs (Figure S5H), supporting our results that OPA1 haploinsufficiency increases the sensitivity of cells to apoptotic stimuli. Collectively, our data demonstrate that OPA1 is essential for proper transcriptional regulation in NPCs, and that OPA1+/− NPCs exhibit a developmental defect that causes impaired neural subtype specification.

OPA1 Protects against Aberrant DNA Methylation

Recently, it has been shown that metabolic stress and accumulation of mitochondria affect hematopoietic stem cell fate decisions through altered nuclear DNA methylation (Ho et al., 2017). Our transcriptome analyses showed that several genes important for NPC differentiation, GABAergic interneuron formation, and retinal development such as the pioneer transcription factor FOXG1 were strongly downregulated in OPA1+/− NPCs. To address whether this downregulation was caused by epigenetic alterations, we analyzed the CpG content in annotated sequences and found CpG-enriched promoters in genes encoding FOXG1, ATP10D, TAF9B, SLITRK2, GABRA2, and ZNF280D. Figure 6A depicts the location and CpG-enriched regions in the promoter and 5′ UTR of FOXG1. Genomic DNA was isolated from NPCs, bisulfite treated, and the FOXG1 5′UTR region was sequenced. We found that all CpG bases within this region were methylated in OPA1+/− cells, whereas this was not the case in OPA1+/+ cells (Figure 6B). This result demonstrates that DNA hypermethylation was the reason for the observed repression of FOXG1 transcription. The amplification of the FOXG1 promoter CpG island was not possible due to the long CpG stretches, which prevented optimal primer design (Figure 6A). Bisulfite sequencing of the ATP10D locus also revealed a high content of methylated promoter CpG in OPA1+/− but not in OPA1+/+ NPCs (Figures 6C and 6D), further confirming that DNA methylation patterns were altered in OPA1+/− NPCs. Taken together, our results suggest that OPA1 is necessary for NPC transcriptional regulation by preventing aberrant methylation.

Figure 6

OPA1 Protects against Aberrant DNA Methylation

(A) Location and sequence of CpG-enriched regions in the FOXG1 gene promoter and 5′ UTR. ATG is indicated as translation start codon. Numbers denote sequences upstream (−) or downstream (+) of the start codon. The yellow box indicates the 5′ UTR, and blue lines represent CpG enriched regions. CpGs are marked in red.

(B) Bisulfite sequencing of the FOXG1 5′UTR in OPA1+/+ and OPA1+/− NPCs. Empty circles represent unmodified cytosines, and filled circles represent methylated cytosines.

(C) Location and sequence of CpG-enriched regions in the ATP10D gene promoter and 5′ UTR. ATG is indicated as translation start codon, and numbers denote sequences upstream (−) or downstream (+) of the start codon. Yellow box indicates the 5′ UTR flanked by a 20-kb intron, and blue lines represent CpG-enriched regions. CpGs are marked in red.

(D) Bisulfite sequencing of the ATP10D gene promoter in OPA1+/+ and OPA1+/− NPCs. Empty circles represent unmodified CpG bases, and filled circles represent methylated cytosines.

(E) FOXG1 mRNA levels in day 11 OPA1+/− NPCs treated with different concentrations of DNMT inhibitor RG108. Mean ± SEM, N = 4 independent experiments with N = 2 technical replicates for each condition. mRNA expression levels were normalized to transcript levels in untreated OPA1+/+ NPCs.

(F) Succinate and alpha-ketoglutarate levels measured in OPA1+/+ and OPA1+/− hESCs and day 11 NPCs measured by mass spectrometry analysis. Mean ± SEM, N ≥ 3 independent samples for each genotype and cell type.

(G) Oxygen consumption rate changes under mitochondrial stress in OPA1+/+ and OPA1+/− NPCs. Oligomycin, FCCP, antimycin, and rotenone were sequentially applied. Mean ± SEM, N = 3 independent experiments.

(H) Basal respiration rates, maximum respiration rates, and ATP production in OPA1+/+ and OPA1+/− NPCs. Mean ± SEM, N = 3 independent experiments were performed and N ≥ 40 technical replicates were analyzed for each genotype.

(I) Representative flow cytometry analysis of OPA1+/+ and OPA1+/− NPCs labeled with the hydrogen peroxide-sensitive oxidative stress indicator CM-H2DCFDA. Unlabeled NPCs (−) were used as negative control. NPCs that were incubated with 0.5 mM H2O2 were used as positive control.

(J) Quantification of flow cytometry measurements in OPA1+/+ and OPA1+/− NPCs. Arithmetic mean values were normalized to the values of labeled OPA1+/+ NPCs. Mean ± SEM, N ≥ 4 independent experiments and N = 2 technical replicates.

(K) CAT mRNA expression levels in day 11 OPA1+/+ and OPA1+/− NPCs. mRNA expression levels were normalized to transcript levels in OPA1+/+ NPCs.

Mean ± SEM, N ≥ 3 independent samples were assessed for each genotype. Student's t test was used to analyze the difference between two groups. n.s. not significant. ∗p < 0.05, ∗∗∗p < 0.001. See also Figures S6 and S7.

Next, we tested whether inhibition of DNA methylation can restore the transcriptional silencing in OPA1+/− NPCs. Strikingly, incubation of cells with the DNA methyltransferase (DNMT) inhibitor RG108 during neural differentiation resulted in a partial rescue of FOXG1 expression in OPA1+/− NPCs, demonstrating that the transcriptional silencing caused by OPA1 haploinsufficiency could be rescued by inhibiting DNMTs and preventing DNA methylation (Figure 6E). To determine whether DNA methylation and hydroxymethylation levels were affected on a global level in OPA1+/− NPCs, we used mass spectrometry-based quantification and found similar total methylation (5me(dC)) and hydroxymethylation (5hm(dC)) levels in OPA1+/+ and OPA1+/− NPCs (Figure S7A). Moreover, immunoblotting for DNMT3A and DNMT3B proteins showed similar expression levels between different genotypes, suggesting that reduced OPA1 levels lead to loci-specific rather than global DNA methylation changes (Figure S7B).

To further explore the potential cause for the observed loci-specific changes in DNA methylation, we analyzed the concentration of the small metabolites S-adenosylmethionine (SAM) and S-adenosylhomocysteine (SAH), which serve as substrates for DNMTs. Using mass spectrometry analysis we demonstrate that the overall SAM/SAH ratio in NPCs of both OPA1+/− and OPA1+/+ genotypes was significantly higher than in hESCs, indicating an overall higher methylation capacity in NPCs (Figure S7C). However, the overall SAM/SAH ratio was not altered when OPA1 was reduced (Figure S7C). Next, we quantified succinate and alpha-ketoglutarate levels in hESCs and NPCs. Again, there were no significant changes in succinate and alpha-ketoglutarate levels in OPA1+/− versus OPA1+/+ cells (Figure 6F), indicating that altered availability of these small metabolites is likely not the reason for the observed changes in DNA methylation.

When analyzing the mitochondrial energy metabolism in NPCs, we found a small but significant increase in basal respiration and ATP production levels in OPA1+/− compared with OPA1+/+ NPCs (Figures 6G and 6H). In addition, flow cytometry analysis showed significantly increased ROS levels in OPA1+/− NPCs (Figures 6I and 6J), further confirming our results in hESC and neurons. In line with the observed increased oxidative stress level in OPA1+/− NPCs, we found catalase (CAT) and FOXC1 mRNA expression levels significantly upregulated in day 11 OPA1+/− NPCs (Figure 5B and Table S2). Catalase is a peroxisomal enzyme protecting cells against oxidative stress, and FOXC1 is a stress-responsive TF that provides resistance to oxidative stress in the eye (Mirzayans et al., 2007). The upregulation of CAT expression was confirmed by qRT-PCR in NPCs derived from two independent OPA1+/− lines (Figures 6K and S6B).

Patient-Derived iPSC Harboring a Pathogenic OPA1 Mutation Exhibit Increased ROS Levels and Downregulation of FOXG1

Heterozygous mutations in OPA1 are a common cause of ADOA (Delettre et al., 2000, Alexander et al., 2000). To ensure that the changes detected in our genetically engineered hESCs were not caused by off-target effects by CRISPR-Cas9, we generated induced pluripotent stem cells (iPSCs) derived from two male donors who carry the inactivating OPA1 c.2873_2876delTTAG mutation. First, we assessed OPA1 mRNA expression and found that mRNA expression levels were reduced by 35%–40% compared with OPA1+/+ NPCs (Figure 7A). Next, we found that PAX6 mRNA expression levels were similar between patient iPSC- and hESC-derived NPCs (Figure 7B). Moreover, FOXG1 (Figure 7C) and CAT (Figure 7D) mRNA expression levels were down- and upregulated in patient iPSC-derived NPCs, respectively. Next, we used the ROS indicator CM-H2DCFDA and further demonstrated that ROS levels were markedly increased in patient-derived iPSCs (Figures 7E and 7F). Taken together, these results indicate that OPA1 haploinsufficiency is accompanied by aberrant nuclear DNA methylation and changed transcriptional circuitry.

Figure 7

Patient-Derived iPSCs Harboring a Pathogenic OPA1 Mutation Exhibit Increased ROS Levels and Downregulation of FOXG1

(A–D) (A) OPA1 mRNA, (B) PAX6, (C) FOXG1, and (D) CAT mRNA expression levels in day 11 NPCs differentiated from iPSC lines derived from two donors diagnosed with ADOA.

(E) Representative flow cytometry analysis of OPA1+/+ hESCs, iPSC #1, and iPSC #2 labeled with the oxidative stress indicator CM-H2DCFDA. Unlabeled hESCs (−) were used as negative control. hESCs and iPSCs that were incubated with 0.5 mM H2O2 were used as positive control.

(F) Quantification of flow cytometry measurements in iPSC #1 and iPSC #2. Arithmetic mean values were normalized to the values of labeled OPA1+/+ hESCs.

Mean ± SEM, N = 3 independent experiments and N = 2 technical replicates. Student's t test was used to analyze difference between two groups. n.s. not significant. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

Discussion

Our study shows for the first time that OPA1 haploinsufficiency correlates with aberrant nuclear DNA methylation patterns as well as a significantly altered transcriptional circuitry and differentiation potential of NPCs (Figure 8).

Figure 8

Schematics Illustrating the Metabolic and Epigenetic Changes in OPA1 Haploinsufficient Cells, Which May Contribute to Altered Neuronal Differentiation

The role of OPA1 during physiologic and pathologic neural development remains incompletely understood. We observed a severe loss of DLX1/DLX2-positive neurons when OPA1 protein levels were reduced by 50%. Studies using mouse models have previously shown that Dlx1 and Dlx2 are expressed in the developing retinal neuroepithelium (Eisenstat et al., 1999) and are needed for the terminal differentiation of late-born RGCs (de Melo et al., 2005). Moreover, Dlx1/Dlx2 double-null mice exhibit up to 30% reduced numbers of RGCs, whereas formation of other retinal neuronal cell types is not affected. In addition to Dlx1/2, loss of Sfrp1 expression (Esteve et al., 2003) and forced expression of NeuroD factors (Cherry et al., 2011) are known to dysregulate retinal cell diversity in mice and chicken. Strikingly, we find impaired expression of DLX1/2 and SFRP1 as well as an upregulation of NEUROD4 when OPA1 levels are reduced, which may explain the loss of RGCs observed in patients with ADOA.

Our transcriptome analysis in NPCs substantiates the relevance of balanced OPA1 levels for proper neural cell homeostasis. One TF that we find consistently repressed in our study is the pioneer factor FOXG1, which is predominantly expressed in the developing neuroepithelium, retina, and optic chiasm. A previous study found that FOXG1 is crucial for GABAergic neuronal formation and retinal development because Foxg1−/− mice exhibit eye abnormalities (Xuan et al., 1995). Importantly, Foxg1−/− telencephalic progenitors fail to generate Dlx1-and Dlx2-positive neurons (Xuan et al., 1995). Other evidence that links FOXG1 to the generation of DLX1/2+ neurons stems from a recent report demonstrating that FOXG1 upregulation is responsible for the overproduction of GABAergic neurons in brain organoids derived from iPSCs obtained from patients with autism spectrum disorder (Mariani et al., 2015). In addition, studies in mice and chicken showed that loss of Foxg1 expression results in RGC axonal misprojection and pathfinding defects (Yuasa et al., 1996, Pratt et al., 2004). Our gene expression analyses in hESC-derived NPCs and neurons also suggest that FOXG1 is a direct regulator of human GABAergic neuronal formation and retinal development.

Recently, iPSCs derived from patients carrying heterozygous mutations in OPA1 and having an inherited form of Parkinson disease (PD) were generated and used to investigate dopaminergic neuronal development. The authors found fragmented mitochondria, a decrease in respiration, and a decrease in neuronal viability, as well as an increase in ROS (Iannielli et al., 2018). Although we also observe an increase in ROS and sensitivity to apoptotic stimuli, we only find subtle changes in mitochondrial morphology including increased intercristae distance, reduced circumference, and a trend to more fragmented mitochondria in OPA1+/− hESCs. Overall, the phenotype in PD patient-derived iPSCs seems more severe than the one observed in OPA1 haploinsufficient hESCs and ADOA patient-derived iPSCs. It is possible that this is because PD is a more complex disease with potential additional epigenetic and genetic changes in contrast to familial optic atrophy with a single-gene mutation. Other studies also show significant mitochondrial fragmentation in mouse cerebellar granule (Jahani-Asl et al., 2011) and HeLa cells (Arnoult et al., 2005). In these studies OPA1 expression was almost completely abolished, whereas in our cells 50% OPA1 proteins remain, which might explain the mild changes in mitochondrial morphology in our cells. In line with this, a recent study using OPA1 haploinsufficient iPSCs derived from patients with syndromic parkinsonism also shows only subtle changes in the mitochondrial ultrastructure in iPSC-derived dopaminergic neurons (Jonikas et al., 2018).

A fascinating question is how OPA1 haploinsufficiency impacts on nuclear DNA methylation at specific sites, such as the FOXG1 and ATP10D promoters. Epigenetic pathways and transcriptional regulation are influenced by non-protein-coding RNA molecules. For instance, the murine long non-coding RNA Kcnq1ot1 recruits the DNA methyltransferase Dnmt1 to CpG islands of imprinted genes (Mohammad et al., 2010), and the natural antisense transcript AS1DHRS4 controls localization of DNMTs to the promoter regions of transcripts DHRS4L1 and DHRS4L2 (Li et al., 2012). Given that the DNA methylation changes we observe in OPA1+/− NPCs are loci specific, it is possible that expression and availability of non-coding RNAs that regulate DNA methylation is affected in OPA1+/− NPCs.

Enzymes catalyzing epigenetic changes require specific metabolites as sources of acetyl or methyl groups (Katada et al., 2012, Lu and Thompson, 2012). Small metabolites such as SAM, SAH, and alpha-ketoglutarate establish a link between mitochondrial metabolism and epigenetics (Matilainen et al., 2017). We find similar SAM/SAH and succinate/alpha-ketoglutarate levels in OPA1+/− and OPA1+/+ cells, suggesting that the observed changes in DNA methylation are not caused by alterations of these small metabolites. Previously, it has been shown that oxidative stress causes aberrant DNA methylation and transcriptional silencing (O'Hagan et al., 2011). The authors reveal that DNMT1 is targeted to damaged chromatin, forming a complex with DNMT3B and polycomb repressive complex 4 (O'Hagan et al., 2011), and show that hydrogen peroxide treatment causes a relocalization of these proteins to promoter CpG islands. Importantly, we find that OPA1 haploinsufficiency results in increased oxidative stress and ROS accumulation in OPA1+/− hESCs and ADOA patient-derived iPSCs. It is well possible that altered ROS levels play a role in the DNA methylation changes we described. Supporting this notion, mitochondrial dysfunction and increased ROS levels have recently been described to cause specific promoter hypermethylation and decreased expression of microRNA-663 in cancer cell lines (Carden et al., 2017). Exogenous and endogenous ROS are also known to affect neural stem cell renewal and proliferation. For instance, Le Belle et al. show that brain-derived neural progenitors maintain a high ROS status and are responsive to ROS stimulation leading to posttranslational oxidative inactivation of PTEN, and activation of PI3K/Akt signaling pathway, which is required for NPC self-renewal (Le Belle et al., 2011). Another study by Chui et al. demonstrated that selective removal of histone-lysine N-methyltransferase PRDM16 in NPCs leads to abnormal neuronal composition and organization in the mouse brain by inducing excessive ROS (Chui et al., 2020). Our results together with these studies suggest that activation of specific pathways by ROS signaling is context dependent and may influence various pathways during cell fate specification.

In conclusion, we provide evidence that OPA1 haploinsufficient cells exhibit increased oxidative stress, aberrant nuclear DNA methylation, and a significantly altered transcriptional circuitry.

Limitations of the Study

Although we addressed DNA methylation changes in the promoter regions of selected down-regulated genes, we did not explore this for up-regulated genes or in a global context. Moreover, we did not assess cell death under stress conditions in iPSC-derived neurons, and performed the mitochondrial respiration assay in two not three independent experiments, although each with 10–40 replicates. In addition, although we observed an increase in ROS that might contribute to the aberrant DNA methylation, we were not able to identify the exact mechanism by which OPA1 haploinsufficiency contributes to these changes in nuclear DNA methylation. Defining primary versus secondary changes related to OPA1 will be exciting to explore in the future.

Resource Availability

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Judith Staerk (judith.staerk@ncmm.uio.no).

Materials Availability

There may be restrictions or delays to the availability of hESC and iPSC lines due to our need to maintain the stock and patient consent. Availability of iPSC lines will require a Materials Transfer Agreement.

Data and Code Availability

Access to data that support the findings of this study are available from the authors on reasonable request. RNA-seq information including all raw data has been deposited at Gene Expression Omnibus under GSE117976: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE117976.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.