Activation of the Keap1/Nrf2 pathway suppresses mitochondrial dysfunction, oxidative stress, and motor phenotypes in C9orf72 ALS/FTD models

C9orf72 Drosophila models present a wide range of mitochondrial defects

Mitochondrial dysfunction has been suggested to be directly involved in disease pathogenesis in ALS/FTD, but evidence from animal models is limited (25). We sought to investigate this in vivo using Drosophila as a model system. Drosophila do not encode an orthologue of C9orf72; nevertheless, exogenous genes can be readily expressed in Drosophila based on the inducible GAL4-UAS expression system (26). Several groups have previously established different Drosophila models of C9orf72-related pathology, which express pure hexanucleotide repeats or alternative codon DPRs of various lengths, sometimes with additional genomic elements of C9orf72 (27). We focussed our initial efforts on readily available lines that express pure G4C2 repeats or alternative codon repeats that encode the DPRs. The expression of 36x G4C2 repeats (G4C2x36) and 36-repeat glycine–arginine DPR (GR36) is substantially neurotoxic compared with the 3x G4C2 repeat (G4C2x3) control (6). Indeed, we found that the strong pan-neuronal expression of G4C2x36 (via nSyb-GAL4) perturbed development, leading to larvae morphologically thinner compared with controls (Fig 1A), and with significantly reduced motor ability assessed by measuring larval crawling behaviour (Fig 1B). Consistent with previous reports, the pan-neuronal expression of GR36 caused even stronger phenotypes (Fig 1A and B), and was developmentally lethal at the third-instar (L3) larval stage.

To analyse C9orf72 pathology in the adult stage, where the impact of ageing can also be investigated, we found that the expression of the C9orf72 transgenes via a predominantly motor neuron driver, DIPγ-GAL4, was adult-viable and caused an age-related decline in motor ability assessed using the negative geotaxis “climbing” assay (Fig 1C). While the motor neuron expression of G4C2x36 did not impact motor behaviour at 2–d-old, by 10 d, G4C2x36 flies exhibited significantly impaired climbing ability (Fig 1C). As before, the motor neuron expression of GR36 was more toxic, causing a significant motor deficit at 2-d-old, which worsened by 10 d (Fig 1C). A similar pattern was observed in the lifespan of DIPγ-GAL4–driven G4C2x36 or GR36 flies with a mild phenotype from G4C2x36 but a markedly shortened lifespan with GR36 (Fig S1A).

Previous studies analysing mitochondrial defects in Drosophila C9orf72 models have mostly focussed on the muscle-directed expression of poly-GR (22, 28). Here, we aimed to explore the potential involvement of mitochondrial dysfunction in a neuronal context. Initially, we observed that the pan-neuronal expression of G4C2x36 caused a significant reduction in complex I– and complex II–linked respiration in lysates from young (5-d-old) fly heads (Fig 1D), indicative of a generalised mitochondrial disruption. To probe this in more detail, we next analysed mitochondrial morphology in larval neurons upon the pan-neuronal expression of the C9orf72 transgenes. As mitochondria are dynamic organelles, microscopy analysis of mito.GFP-labelled mitochondria typically reveals a mix of short round (fragmented) and long tubular (fused) morphologies (Fig 1E), which can be quantified using a scoring system that characterises the overall mitochondrial morphology on a cell-by-cell basis (Fig 1E and F). Using this approach, we found that mitochondria were more elongated and hyperfused in G4C2x36- and GR36-expressing neurons (Fig 1E and F).

Mitochondrial morphology is known to respond to changes in reactive oxygen species (ROS) levels, as well as other physiological stimuli. First, we tested the susceptibility of G4C2x36 flies to oxidative stress induced by paraquat (which generates superoxide anions) and hydrogen peroxide (H₂O₂, which generates hydroxyl radicals). We observed that G4C2x36 flies were hypersensitive to both types of oxidative stressors when compared to control animals (Fig S1B and C).

ROS can occur in different forms (e.g., superoxide anions, H₂O₂), which may vary in their prevalence and functional significance across cellular compartments (mitochondrial versus cytosolic). Therefore, it is important to study the effects of ROS species in each compartment to understand the tight regulation required to achieve redox homeostasis. We used genetically encoded redox-sensitive fluorescent protein (roGFP2) probes fused to glutaredoxin-1 (Grx1) or oxidant receptor peroxidase 1 (Orp1). Oxidation by oxidised glutathione (GSSG) or H₂O₂, respectively, leads to a shift in the excitation maxima of the roGFP2 fusion constructs from 488 to 405 (29). Importantly, these reporters can be targeted to specific subcellular compartments, that is, cytosol or mitochondria, giving insights into the compartment-specific redox status. We observed a significant increase in the oxidised status of mitochondrial roGFP2-Grx1 (Fig 1G and H) and mitochondrial roGFP2-Orp1 (Fig S1D and E) in larval neurons when combined with G4C2x36 or GR36, indicating an increase in mitochondrial H₂O₂ and GSSG in these animals. In contrast, no significant differences were observed in cytosolic glutathione redox potential (E_GSH) via cyto-Grx1-roGFP2 (Fig 1I and J), and little or no changes were detected with dihydroethidium (DHE) intensity, an indicator of cytosolic superoxide, in G4C2x36 and GR36 larvae (Fig S1F and G). To extend our analysis of the larval model to adult flies, and to investigate an independent measure of ROS, we used MitoSOX to investigate levels of mitochondrial superoxide in the adult brain. In agreement with the fluorescent reporter results, we observed an increase in MitoSOX intensity in 5-d-old G4C2x36 flies compared with the G4C2x3 control (Fig 1K and L). Taken together, these data suggest that the mitochondrial but not cytosolic redox state is disrupted in these Drosophila models of C9orf72-related pathology. These results are summarised in Table 1.

Changes in mitochondrial morphology and increases in ROS can be indicative of ongoing mitochondrial damage, which can be alleviated via mitophagy. Thus, we next used the mito-QC mitophagy reporter (30) to investigate the mitophagy status in the C9orf72 models. Briefly, the mito-QC reporter uses a tandem GFP-mCherry fusion protein targeted to the outer mitochondrial membrane. Mitolysosomes are marked when GFP is quenched by the acidic environment but mCherry fluorescence is retained, resulting in “red-only” puncta (30). Analysis showed that there were significantly fewer mitolysosomes in G4C2x36 and GR36 larval neurons compared with the G4C2x3 control (Fig 1M and N), suggesting that mitophagy is perturbed. However, this was not a mitophagy-specific defect as the analogous reporter for general autophagy, GFP-mCherry-Atg8a (fly Atg8a is homologous to LC3), also showed a reduction in the number of mCherry-positive autolysosomes in G4C2x36 and GR36 larval neurons (Fig S1H and I). Consistent with this, immunoblot analysis of protein lysates from 5-d-old G4C2x36 fly heads revealed an increase in ref(2)P levels (fly homologue of p62), as well as a reduction in lipidated Atg8a-II levels (Fig S1J and K). These results are consistent with a reduction in general autophagic flux, which could impact mitophagic flux, in agreement with previous literature using different C9orf72 models (31).

Recently, West et al (32) developed additional C9orf72 Drosophila models, which express more physiologically relevant repeat length (27), that is, ∼1,000 repeats. For comparison with the Mizielinska et al lines, we also analysed the GR1000-eGFP line (for simplicity, hereafter called GR1000). We observed an age-related decline in motor performance compared to control with the pan-neuronal expression of GR1000, where GR1000 flies were not able to climb at all by 20 d (Fig 2A), similar to that previously reported (32). To complement our results shown in the G4C2x36 and GR36, we measured mitochondrial superoxide with MitoSOX staining in GR1000 adult brains as it was not possible to use the roGFP2 reporters in conjunction with eGFP-tagged GR1000. Here, we also observed an increase in MitoSOX staining in 10-d-old GR1000 brains compared with the control (Fig 2B and C).

Taken together, the preceding data show that various aspects of mitochondrial form and function are disturbed in multiple Drosophila models of C9orf72 pathology, which correlate with the organismal decline and loss of motor function.

Genetic manipulation of mitochondrial dynamics or mitophagy does not rescue C9orf72 phenotypes

We next sought to determine whether disruptions to these observed changes in mitochondrial morphology, defective mitophagy, and oxidative stress are contributing factors to the neurodegenerative process. First, we evaluated whether genetic manipulations to counteract the elongated, hyperfused mitochondrial morphology caused by G4C2x36 or GR36 expression may rescue the larval locomotor defect. Thus, we combined the pan-neuronal expression of G4C2x36 or GR36 with genetic manipulations to promote fission (overexpression of pro-fission factors Drp1 or Mff [Tango11 in Drosophila]) or reduce fusion (loss of pro-fusion factors Opa1 and Mitofusin [Marf in Drosophila]). Although the elongated mitochondria observed in the G4C2x36 and GR36 neurons were partially reversed by these manipulations as expected (Fig S2A and B), this did not result in any improvement in larval locomotion (Fig S2C). These data suggest that excess mitochondrial fusion observed in the C9orf72 models does not play a key role in C9orf72 pathogenesis and may be a downstream consequence.

Similarly, since we observed reduced mitophagy in the C9orf72 models, we aimed to reverse this by boosting mitophagy by targeting the mitophagy inhibitor USP30. Knockdown of USP30 has been shown to rescue defective mitophagy caused by pathogenic mutations in PRKN and improve mitochondrial integrity in parkin- or Pink1-deficient flies (33). We have also seen that USP30 knockdown increases mitolysosome number in Drosophila neurons and muscle (34), indicative of increased mitophagy. As expected, we observed that USP30 knockdown significantly increased the number of mitolysosomes in control larval neurons (Fig S2D and E). However, when co-expressed with G4C2x36 or GR36, USP30 knockdown was not sufficient to rescue the reduced mitophagy (Fig S2D and E). Moreover, the pan-neuronal co-expression of G4C2x36 or GR36 with USP30 RNAi did not improve the larval locomotor deficit either (Fig S2F). These data suggest that mitophagy is also not a primary contributor to C9orf72 pathogenesis.

Overexpression of mitochondrial Sod2 and catalase ameliorates C9orf72 motor phenotypes

Previous studies have shown that targeting oxidative stress may be beneficial in C9orf72-related pathology (20). Cellular defence mechanisms such as antioxidants are targeted to different cellular and subcellular locations, due to many sources of ROS. This compartmentalisation also highlights the need for fine-tuning of ROS signalling for redox homeostasis, as well as the possibility for ROS to signal between compartments (35). Since an increase in mitochondrial ROS was observed in G4C2x36, GR36, and GR1000 flies, we hypothesised that the overexpression of antioxidants would suppress behavioural locomotor phenotypes. First, we used a pan-neuronal driver to co-express the major antioxidant genes—cytosolic Sod1, mitochondrial Sod2, catalase (Cat), and a mitochondrially targeted catalase (mitoCat)—with G4C2x36 and GR36. Contrary to our prediction, the overexpression of cytosolic Sod1 significantly worsened the morphology of G4C2x36 and GR36 larvae, becoming even thinner and more developmentally delayed (Fig 3A). This was reflected by a worsening of the locomotor deficit for G4C2x36 and GR36 larvae, where GR36 larvae co-expressing Sod1 did not crawl during the assay conditions (Fig 3B). In contrast, the overexpression of mitochondrial Sod2, Cat, or mitoCat significantly rescued G4C2x36 and GR36 larval crawling (Fig 3B). These manipulations also improved larval morphology, particularly for GR36 (Fig 3A). Interestingly, we found that the expression of Sod1 was reduced at both mRNA (Fig S2G) and protein levels (Fig S2H and I) in G4C2x36 flies, whereas Cat mRNA levels were increased and Sod2 mRNA levels were unaltered (Fig S2G).

We next asked whether these manipulations similarly affected the other mitochondrial phenotypes. We found that overexpressing Sod2, Cat, or mitoCat significantly reduced the elongated mitochondrial phenotype in G4C2x36 and GR36 larval neurons (Fig 3C and D). Interestingly, Sod1 overexpression also partially rescued this phenotype. Notably, Sod2 overexpression was also able to partially rescue the defective mitophagy in G4C2x36 larvae, though this did not reach significance for GR36 (Fig 3E and F). To validate these genetic interactions, we used the GR1000 lines where we also found that Sod2 or Cat overexpression significantly rescued the age-related locomotion defect (Fig 3G and H).

Thus, among all the different genetic manipulations used to reverse mitochondrial phenotypes observed, only the overexpression of certain antioxidant enzymes such as mitochondrial Sod2 and catalase was consistently beneficial in rescuing the locomotor deficits. This suggests that oxidative stress is an important mediator of pathogenesis.

The Keap1/Nrf2 pathway is partially activated in C9orf72 flies

Keap1 and Nrf2 define an important redox-sensing pathway which activates antioxidant gene expression upon oxidative stress. Keap1 and Nrf2 are conserved in Drosophila (Nrf2 is called cap-n-collar C isoform, CncC) (36), so we sought to explore whether this pathway is involved in C9orf72 pathology. Assessing the relative nuclear abundance of CncC in neurons of the larval ventral ganglion, we observed an increase in nuclear CncC in both G4C2x36 and GR36 compared with the G4C2x3 control (Fig 4A and C). We also observed a similar increase in CncC nuclear staining in 5-d-old adult brains of G4C2x36 animals (Fig 4B and C). This indicates that there is activation of the Keap1/CncC signalling pathway at early stages, which remains activated during disease progression.

To further analyse the activity of this signalling pathway, we used an antioxidant reporter transgene, which expresses GFP via the promoter of GstD1, an Nrf2 target and prototypical oxidative stress response gene (36), to monitor antioxidant responses. The pan-neuronal expression of both G4C2x36 (Fig 4D) and GR1000 (Fig 4E) caused an increase in GstD1-GFP levels compared with their respective controls, consistent with an increase in CncC activity. At the same time, we assessed whether G4C2x36 flies could also respond appropriately to ROS-inducing paraquat (PQ) treatment. We again saw that the GstD1-GFP reporter expression was significantly increased in G4C2x36 flies under basal conditions (−PQ) compared with controls (Fig 4F). However, although GstD1-GFP levels substantially increased upon PQ treatment in both control and G4C2x36 flies (Fig 4F), the response in G4C2x36 flies was similar to the control. These observations, together with the preceding genetic studies, suggest that although the C9orf72 flies can detect the elevated oxidative stress conditions and mount a response, it is not maximally induced and likely insufficient to confer full protection. The increased response of the GstD1-GFP reporter to PQ beyond the level seen in G4C2x36 alone indicates that an increased antioxidant response can be induced in these animals.

Genetic and pharmacological inhibition of Keap1 partially rescues C9orf72 phenotypes

Although the Keap1/Nrf2 pathway was activated in C9orf72 models, the response seemed to be insufficient. Therefore, we hypothesised that genetically reducing Keap1 levels to constitutively boost CncC activity could benefit C9orf72 phenotypes as has been seen in a Drosophila model of Alzheimer’s disease (37). Indeed, combining a heterozygous mutant of Keap1 with pan-neuronally driven G4C2x36 or GR36 significantly improved the larval motor ability of both G4C2x36 and GR36 compared with the control (Fig 5A), and visibly improved the larval morphology (Fig 5B). Furthermore, Keap1 heterozygosity was also able to fully rescue the GR1000 adult climbing deficit (Fig 5C).

Analysing the distribution of CncC, we observed again that G4C2x36 provoked an increase in CncC nuclear localisation (Fig 5D and E) with the trend towards a further increase with Keap1 heterozygosity (Fig 5D and E). Similarly, when performing high-resolution respirometry, there was a trend for improving both complex I– and complex II–linked respiration deficit in G4C2x36 with Keap1 heterozygosity (Fig 5F).

The positive effects of partial genetic loss of Keap1 on the behaviour of C9orf72 models support the potential for pharmaceutical agents that modulate Keap1 to also be beneficial in this context. To test this hypothesis, we used dimethyl fumarate (DMF), a Keap1-modifying Nrf2-activating drug with antioxidative and anti-inflammatory properties. DMF acts by inactivating Keap1 via succination of its cysteine residues (38), which has previously been shown to exhibit neuroprotective effects in animal models of neurodegeneration (39). Analysing the impact of DMF on the Keap1/Nrf2 pathway, we performed quantitative RT–PCR on transcriptional targets of CncC, including GstD1, GstD2, Gclc, and Cyp6a2 (40, 41), in the C9orf72 models with and without a modest dose of DMF treatment. The degree of transcriptional activation was variable between different targets, but DMF treatment caused a general up-regulation of CncC targets in G4C2x36 flies (Fig 6A).

To test whether DMF treatment improves mitochondrial function in the G4C2x36 model, we again used the mito-roGFP2-Orp1 reporter to assess mitochondrial H₂O₂ levels. Here, animals were raised on food containing DMF or vehicle (ethanol). We again observed an increase in reported H₂O₂ levels in G4C2x36 animals, which was partially rescued by DMF treatment (Fig 6B).

Finally, to extend these observations and assess the effects of DMF treatments on motor phenotypes, we expressed G4C2x3, G4C2x36, or GR36 via DIPγ-GAL4 and treated the adult flies with food supplemented with DMF or vehicle (ethanol) for 10 d. For both G4C2x36 and GR36, the progressive decline in locomotor performance in vehicle-treated flies was greatly alleviated by 10 d of DMF treatment (Fig 6C). Similarly, the progressive loss of motor function in pan-neuronal GR1000 flies was also significantly reduced with 20 d of DMF treatment, compared with the vehicle control (Fig 6D). Taken together, these data indicate that inhibiting Keap1 by DMF treatment can induce an antioxidant response transcriptional profile, alleviate mitochondrial dysfunction, and attenuate C9orf72 behavioural phenotypes.

DMF treatment is beneficial in C9orf72-ALS/FTD patient-derived neurons

Since we had observed elevated ROS in C9orf72 Drosophila models and found that DMF treatment provided phenotypic benefit, we wanted to test the translatable potential of DMF treatment on patient-relevant material. Fibroblasts from ALS patients carrying C9orf72 mutations and healthy age–matched controls (Table 2), previously reported and successfully differentiated (42), were reprogrammed into induced neural progenitor cells (iNPCs), then differentiated into induced neurons (iNeurons), as assessed by the expression of neuronal markers βIII-tubulin, MAP2, and NeuN (Fig S3A and B). Although these cultures did not show overt signs of excess toxicity or cell death (Fig S3C), assessing the level of oxidative stress using MitoSOX, we found a modest increase in mitochondrial ROS in C9orf72 iNeurons (Fig 7A and B), in line with that observed in Drosophila. We also found that there was a higher level of nuclear Nrf2 in patient iNeurons (Fig 7C and D) mirroring the C9orf72 Drosophila phenotypes.

Testing two different concentrations, 24-h treatment with 3 or 10 μM DMF was able to reduce the mitochondrial ROS levels in C9orf72 iNeurons (Fig 7A and B) without detrimentally affecting viability (Fig S3D). By comparison, we analysed treatment with edaravone (EDV), a known ROS scavenger and an FDA-approved ALS drug, as a positive control, and observed that 24-h treatment with 30 and 100 μM EDV reduced ROS levels in C9orf72 iNeurons to a similar extent as DMF (Fig S3E–G). Although significant, the effect size of DMF or EDV treatment is small and the impact on neuronal function or survival is currently unknown. Hence, more investigation is required to validate these initial results. Nevertheless, taken together, these data suggest that activating the Keap1/Nrf2 pathway using therapeutic interventions, such as DMF, can be beneficial across multiple model systems including in vivo C9orf72 Drosophila models and human C9orf72 patient-derived neurons.

Drosophila stocks and husbandry

Flies were raised under standard conditions in a temperature-controlled 12:12-h light/dark cycle incubator at 25°C and 65% relative humidity, on standard Drosophila food containing cornmeal, agar, molasses, yeast, and propionic acid (Genetics Department, University of Cambridge). Transgene expression was induced using the pan-neuronal driver nSyb-GAL4 or the predominantly motor neuron driver DIPγ-GAL4. The following strains were obtained from Bloomington Drosophila Stock Center: w¹¹¹⁸ (BDSC_6326), nSyb-GAL4 (BDSC_51635), UAS-mito-HA-GFP (BDSC_8443), UAS-luciferase RNAi (BDSC_31603), UAS-Sod1 (BDSC_33605), UAS-Sod2 (BDSC_24492), UAS-catalase (BDSC_24621), Opa1^s3475 (BDSC_12188), UAS-GFP.mCherry.Atg8a (BDSC_37749), UAS-mito-roGFP2-Orp1 (BDSC_67667), UAS-cyto-Grx1-roGFP2 (BDSC_67662), UAS-mito-roGFP2-Grx1 (BDSC_67664); the Zurich ORFeome Project: UAS-LacZ (F005035) and UAS-Tango11.HA (F002828); and the Fly Stocks of National Institute of Genetics: UAS-USP30 RNAi (3016R-2). Other Drosophila lines were kindly provided as follows: DIPγ (MI03222)-GAL4 (BDSC_90315) from Dr Robert Carrillo (56, 57), UAS-G4C2x3, UAS-G4C2x36, and UAS-GR36 from Prof. Adrian Isaacs (6), UAS-GR1000-eGFP from Dr Ryan West (32), UAS-mito.Catalase from Prof. Alberto Sanz (58), UAS-Drp1.WT from Prof. Jongkyeong Chung (59), Marf^B from Prof. Hugo Bellen (60), and GstD1-GFP (36) and Keap1^del (61) from Prof. Linda Partridge. The UAS-mito-QC reporter (BDSC_91640) has been described before (30).

Drug treatments

For paraquat treatment, standard Drosophila food was supplemented with paraquat (856177; Sigma-Aldrich) to a final concentration of 10 mM. For dimethyl fumarate (DMF, 242926; Sigma-Aldrich) adult treatment, DMF (or an equivalent volume of ethanol for the vehicle control) was added to a sugar–yeast (SY) medium (61) consisting of 15 g/litre agar, 50 g/litre sugar, and 100 g/litre yeast to a final concentration of 7 μM. All flies were transferred into freshly prepared supplemented food every 2–3 d. For larval DMF treatment, crosses of the parental genotypes were set up in standard food containing 1 μM DMF (or an equivalent volume of ethanol for the vehicle control), providing the larval offspring exposure to the treatment throughout development.

Locomotor and survival assays

Immunohistochemistry and sample preparation

For mitophagy analysis with the mito-QC reporter, autophagy analysis with the GFP.mCherry.Atg8a reporter, and mitochondrial morphology analysis, third-instar larval brains were dissected in PBS and fixed in 4% formaldehyde (FA) (28908; Thermo Fisher Scientific) in PBS for 20 min at RT. For the mitophagy and autophagy experiments, the 4% FA/PBS was adjusted to pH 7. Samples were then washed in PBS followed by water to remove salts. ProLong Diamond Antifade mounting medium (P36961; Invitrogen) was used to mount the samples and imaged the next day.

For immunostaining of third-instar larval brains, larvae were dissected and fixed as described, permeabilised in 0.3% Triton X-100 in PBS (PBS-T) for 30 min, and blocked with 1% BSA in PBS-T for 1  h at RT. Tissues were then incubated with mouse anti-CncC (63) (1:500; a kind gift from Dr Fengwei Yu), diluted in 1% BSA in PBS-T overnight at 4°C, then washed three times for 10 min with PBS-T, followed by incubation with secondary antibody—goat anti-mouse IgG H&L Alexa Fluor 488 (1:500, A11001; Invitrogen). The tissues were washed in PBS-T with 1:10,000 Hoechst (H3570; Invitrogen) for 20 min followed by two times of PBS washes and mounted on slides using ProLong Diamond Antifade mounting medium.

For immunostaining of adult brains, flies were dissected in PBS and fixed on ice in 4% FA in PBS for 30 min. Brains were washed three times for 20 min in PBS-T, before blocking for 4 h in 4% normal goat serum in PBS-T. Tissues were then incubated with mouse anti-CncC (63) (1:500; a kind gift from Dr Fengwei Yu), diluted in 4% normal goat serum in PBS-T overnight at 4°C, then washed three times for 10 min with PBS-T, followed by incubation with secondary antibody for 2 h. The tissues were washed in PBS-T with 1:10,000 Hoechst for 20 min followed by two times of PBS washes and mounted on slides using Prolong Diamond Antifade mounting medium.

ROS analysis

Microscopy

Fluorescence imaging was conducted using a Zeiss LSM 880 confocal microscope (Carl Zeiss MicroImaging) equipped with Nikon Plan-Apochromat 63x/1.4 NA oil immersion objectives. Images were prepared using FIJI (ImageJ). For mito-QC imaging, the Andor Dragonfly spinning disc microscope was used, equipped with a Nikon Plan-Apochromat 100x/1.45 NA oil immersion objective and an iXon camera. Z-stacks were acquired with 0.2-μm steps. For larval morphology, images were acquired using a Leica DFC490 camera mounted on a Leica MZ6 stereomicroscope.

Quantification and analysis methods

Mitochondrial respiration

Mitochondrial respiration was monitored at 25°C using an Oxygraph-2k high-resolution respirometer (OROBOROS Instruments). Standard oxygen calibration was performed before the start of every experiment. 25-d-old adult fly heads per replicate for each genotype were extracted using forceps and placed in 100 μl of respiration buffer (RB) (120 mM sucrose, 50 mM KCl, 20 mM Tris–HCl, 4 mM KH₂PO4, 2 mM MgCl₂, 1 mM EGTA, and 1 g/litre fatty acid–free BSA, pH 7.2). This was homogenised on ice using a pestle with 20 strokes. 1 ml of RB was added to the homogenate and passed through a 1-ml syringe with a piece of cotton wool inside to remove the debris. This was repeated with another 1 ml of RB. In total, 2.1 ml of homogenate was added to the respiratory chambers. For coupled “state 3” assays, saturating concentrations of substrates including 10 mM glutamate, 2 mM malate, 10 mM proline, and 2.5 mM ADP were added to measure complex I–linked respiration. 0.15 μM rotenone was added to inhibit complex I, and 10 mM succinate was added to measure complex II–linked respiration. Data acquisition and analysis were carried out using DatLab software (OROBOROS Instruments).

Immunoblotting

Proteins were isolated from either larval brains or adult heads using RIPA lysis buffer (50 mM [pH 7.4] Tris, 1 M NaCl, 0.1% SDS, 0.5% sodium deoxycholate, and 1% NP-40) supplemented with cOmplete mini EDTA-free protease inhibitors (Roche). After protein quantification using the bicinchoninic acid assay (23225; Thermo Fisher Scientific), Laemmli buffer (1610747; Bio-Rad) containing 1:10 β-mercaptoethanol (M6250; Sigma-Aldrich) was added. Samples were boiled at 95°C for 10 min, resolved by SDS–PAGE using 4–20% or 10% precast gels (4561093; Bio-Rad), depending on the molecular weight of the desired protein, and transferred onto a nitrocellulose membrane (1704158; Bio-Rad) using a Bio-Rad Trans-Blot semi-dry system. Membranes were blocked with 5% (wt/vol) dried skimmed milk powder (Marvel Instant Milk) in TBS with 0.1% Tween-20 (TBS-T) for 1 h at RT and probed with the appropriate primary antibodies diluted in TBS-T overnight at 4°C. After three 10-min washes in TBS-T, the membranes were incubated with the appropriate HRP-conjugated secondary antibodies diluted in 5% milk in TBS-T for 1 h at RT. Membranes were washed three times for 10 min in TBS-T, and detection was achieved with the Amersham ECL Prime detection kit (RPN2232; Cytiva). Blots were imaged using the Amersham Imager 680 with further analysis and processing using FIJI (ImageJ).

The following primary antibodies for immunoblotting were used in this study: GFP (1:100, ab294; Abcam), tubulin (1:5,000, T9026; Sigma-Aldrich), ref(2)P (1:1,000, ab178440; Abcam), and GABARAP (1:1,000, ab109364; Abcam). The following secondary antibodies for immunoblotting were used in this study: goat anti-mouse IgG H&L (HRP-conjugated, 1:10,000, ab6789; Abcam) and goat anti-rabbit IgG H&L HRP (1:10,000, G21234; Thermo Fisher Scientific).

qRT–PCR

Quantitative real-time PCR (qRT–PCR) was carried out as follows: 30 heads were collected and placed in a 2-ml tube containing 1.4-mm ceramic beads (15555799; Fisherbrand) for tissue preparation. 400 μl of TRI Reagent (T9424; Sigma-Aldrich) was added and placed into Minilys homogeniser (Bertin Technologies) where the programme was set to a maximum speed of 10 s. The samples were placed back on ice for 5 min before two further rounds of lysis. The Direct-zol RNA MiniPrep kit (R2050; Zymo Research) was used to extract RNA following the manufacturer’s instructions. TURBO DNA-free Kit (AM1907; Invitrogen) was used to remove contaminating DNA following the manufacturer’s instructions. cDNA synthesis was achieved using Maxima H Minus cDNA Synthesis Kit (M1681; Thermo Fisher Scientific) following the manufacturer’s instructions. Equivalent (500 ng) total RNA underwent reverse transcription for each sample. Finally, qRT–PCR was run using Maxima SYBR Green/ROX Kit (K0221; Thermo Fisher Scientific) following the manufacturer’s instructions using the QuantStudio 3 RT–PCR machine. The relative transcript levels of each target gene were normalised to a geometric mean of RpL32 and Tub84b reference genes; relative quantification was performed using the comparative C_T method entering into account PCR primer efficiency (65).

The following primers were used in this study:

Sod1: CCAAGGGCACGGTTTTCTTC, CCTCACCGGAGACCTTCAC.

Sod2: GTGGCCCGTAAAATTTCGCAAA, GCTTCGGTAGGGTGTGCTT.

Catalase: CCAAGGGAGCTGGTGCTT, ACGCCATCCTCAGTGTAGAA.

GstD1: CCGTGGGCGTCGAGCTGAACA, GCGCGAATCCGTTGTCCACCA.

GstD2: AACCAGCGTCTGTACTTCGA, TCAAGTCCTCATCCGATCCG.

Cyp6a2: AAGCCATGACCTACTTGAACC, TGCCCTTCTCAATCACAAGC.

Gclc: TCCTCTCAGTTCAGCCCG, TTCGTCTTTGTGTCCTTGAAAAC.

RpL32: AAACGCGGTTCTGCATGAG, GCCGCTTCAAGGGACAGTATCTG.

Tub84b: TGGGCCCGTCTGGACCACAA, TCGCCGTCACCGGAGTCCAT.

iNPC tissue culture, iNeuron differentiation, and compound treatment

Induced neuronal progenitor cells (iNPCs) were generated as previously described (66, 67). Details for the patient lines used in this study are provided in Table 2. iNPCs were maintained in DMEM/F-12 with GlutaMAX (31331028; Gibco) supplemented with 1% N-2 (17502001; Invitrogen), 1% B-27 (17504001; Invitrogen), and 20 ng/ml FGF-basic (100-18B; PeproTech) on fibronectin (FC010; Millipore)-coated cell culture dishes. Cells were routinely subcultured every 2–3 d using Accutase (15323609; Corning) to detach them. To achieve neuronal differentiation, iNPCs were plated into six-well plates and cultured for 2 d in DMEM/F-12 media with GlutaMAX supplemented with 1% N-2, 2% B-27, and 0.5 μM DAPT (D5942; Sigma-Aldrich). On day 3, DAPT was removed, and media were supplemented with 0.5 μM smoothened agonist (SAG, 566660; Millipore), 2.5 μM forskolin (11018; Cayman Chemical), and 1 μM retinoic acid (R2625; Sigma-Aldrich) for 16 d Cells were treated with edaravone (EDV; M70800; Sigma-Aldrich) and DMF (242926; Sigma-Aldrich) for 24 h before imaging. Cells were treated with 30 or 100 μM EDV, or DMF at concentrations of 3 or 10 μM.

Live-cell imaging assays of iNeurons

Cells were stained for 30 min with Hoechst (B2883; Sigma-Aldrich) and MitoSOX Red (M36008; Invitrogen) at concentrations of 20 μM and 500 nM, respectively. Cells were imaged using an Opera Phenix high-content imaging system, with analysis performed using a custom protocol on Harmony software (PerkinElmer). Nucleus counts after treatments with DMF and EDV were normalised to baseline counts per cell line. Cell viability was measured using a viability indicator from Neurite Outgrowth Staining Kit (A15001; Thermo Fisher Scientific) according to the manufacturer’s instructions.

Fixed cell imaging of iNeurons

On day 18 of differentiation, cells were fixed in 4% PFA for 30 min. After PBS washes, cells were permeabilised with 0.1% Triton X-100 for 10 min and blocked with 5% horse serum (H0146; Sigma-Aldrich) for 1 h. Cells were incubated with primary NRF2 antibody overnight at 4°C. Cells were washed with 0.1% Tween in PBS and incubated with secondary antibodies and 1 μM Hoechst before imaging. Cells were imaged using an Opera Phenix high-content imaging system, with analysis performed using a custom protocol on Harmony software (PerkinElmer).

The following primary antibodies for immunostaining of iNeurons were used in this study: NRF2 (1:1,000, ab31163; Abcam), βIII-tubulin (1:1,000, AB9354; Merck), MAP2 (1:1,000, ab32454; Abcam), NeuN (1:1,000, ab177487; Abcam). The following secondary antibodies for immunostaining of iNeurons were used in this study: donkey anti-rabbit IgG H&L Alexa Fluor 568 secondary antibody (1:1,000, A10042; Invitrogen) and goat anti-chicken IgG H&L Alexa Fluor 488 secondary antibody (1:1,000, A11039; Invitrogen).

Statistical analysis

GraphPad Prism 9 (SCR_002798; RRID) was used to perform all statistical analyses. Climbing data were analysed using the Kruskal–Wallis non-parametric test with Dunn’s correction for multiple comparisons. Data are presented as the mean ± 95% confidence interval (CI). Quantifications of larval crawling, the number of mitolysosomes and autolysosomes, WB, and qRT–PCR were analysed using one-way ANOVA with the Bonferroni post hoc test for multiple comparisons. Data are presented as the mean ± SD. Mitochondrial morphology was quantified using a chi-squared test. Unpaired t tests with Welch’s correction for unequal SD were used for respiratory analysis and immunoblots. Two-way ANOVA with Dunnett’s multiple comparison test was used for analysis of the iNeurons work. Lifespan experiments were analysed using the Mantel–Cox log-rank test. All statistical tests and n numbers are stated in the figure legends.