Activation of the Keap1/Nrf2 pathway suppresses mitochondrial 1 dysfunction in C9orf72 ALS/FTD in vivo models and patient iNeurons

20 Mitochondrial dysfunction such as excess production of reactive oxygen species (ROS) and 21 defective mitochondrial dynamics are common features of C9orf72 Amyotrophic Lateral 22 Sclerosis/Frontotemporal Dementia (ALS/FTD), but it remains unclear whether these are 23 causative or a consequence of the pathogenic process. To address this, we have performed 24 a comprehensive characterisation of mitochondrial dysfunction in vivo model, analysing 25 multiple transgenic Drosophila models of C9orf72 -related pathology, which can be correlated 26 to disease-relevant locomotor deficits. Genetic manipulations to reverse different aspects of 27 mitochondrial disruption revealed that only genetic upregulation of antioxidants such as 28 mitochondrial Sod2 and catalase were able to rescue C9orf72 locomotor deficits, suggesting 29 a causative link between mitochondrial dysfunction, ROS and behavioural phenotypes. By 30 analysing the Keap1/Nuclear factor erythroid 2–related factor 2 (Nrf2) pathway, a central 31 antioxidant response pathway, we observed a blunted response in the C9orf72 models. 32 However, both genetic reduction of Keap1 and its pharmacological targeting by dimethyl 33 fumarate (DMF), was able to rescue C9orf72 -related motor deficits. In addition, analysis of 34 C9orf72 patient-derived iNeurons showed increased ROS that was suppressed by DMF 35 treatment. These results indicate that mitochondrial oxidative stress is an upstream 36 pathogenic mechanism leading to downstream mitochondrial dysfunction such as alterations 37 in mitochondrial function and turnover. Consequently, our data support targeting the 38 Keap1/Nrf2 signalling pathway as a viable therapeutic strategy for C9orf72 -related ALS/FTD. 39


Introduction
Amyotrophic lateral sclerosis (ALS) is characterised by the loss of upper and lower motor neurons leading to symptoms such as muscle weakness and paralysis.A plethora of evidence supports a clinical, pathologic and genetic overlap between ALS and frontotemporal dementia (FTD) (1), which is characterised by the degeneration of frontal and temporal lobes leading to clinical symptoms such as cognitive impairment and changes in behaviour and personality.
Up to 50% of ALS patients report cognitive and behavioural changes; similarly, motor neuron dysfunction is a common feature in approximately 15% of FTD cases (2).A hexanucleotide repeat expansion consisting of GGGGCC (G4C2) in the first intron of C9orf72 is the most common pathogenic mutation in ALS/FTD (3,4).Several pathogenic mechanisms have been proposed including haploinsufficiency of the gene product and the sequestration of RNA binding proteins at accumulations (foci) of the transcribed RNA (5).However, although intronic, the expanded RNA can also be translated through a mechanism known as repeat associated non-AUG (RAN) translation which produces 5 different dipeptide repeat proteins (DPRs), with arginine DPRs exhibiting the most toxicity (6)(7)(8)(9)(10)(11)(12)(13)(14).
Abundant evidence supports that mitochondrial dysfunction is an early alteration in ALS (15,16), and mitochondrial bioenergetics and morphological changes have been observed in C9orf72 patient fibroblasts and iPSC-derived motor neurons (17)(18)(19).Cellular and mouse models expressing poly-GR have consistently shown mitochondrial perturbations such as redox imbalance and DNA damage (20).Evidence indicates that poly-GR binds to ATP5A1 thereby compromising mitochondrial function (21), and that poly-GR toxicity and aggregation can also occur due to frequent stalling of poly-GR translation on the mitochondrial surface, triggering ribosome-associated quality control and C-terminal extension (22).
Various cellular mechanisms exist to counteract disruptions in mitochondrial function and redox imbalance, from antioxidant defence mechanisms to the wholesale degradation of mitochondria via macroautophagy (mitophagy), which has been relatively understudied in the C9orf72 context.One of the most important upstream mechanisms that counteract redox imbalance is the Keap1/Nuclear factor erythroid 2-related factor 2 (Nrf2) pathway.Under basal conditions, Nrf2, a master regulatory transcription factor for antioxidant and cellprotective factors, is negatively regulated via targeted proteasomal degradation by an E3 ubiquitin ligase adaptor, Keap1.This interaction is alleviated upon oxidative and electrophilic stresses, and can also be provoked by Keap1 small-molecule inhibitors, causing Nrf2 to accumulate in the nucleus.This induces the expression of a repertoire of protective factors such as proteins with detoxification, antioxidant and anti-inflammatory properties, with the purpose to maintain mitochondrial function and redox balance (23,24).
In this study, we have conducted an extensive in vivo characterisation of mitochondrial dysfunction in multiple Drosophila models of C9orf72 ALS/FTD by studying mitochondrial dynamics, respiration, mitophagy and redox homeostasis, in conditions that cause the disease-relevant locomotor deficits.We found that only reversal of oxidative stress by the overexpression of antioxidant genes was able to rescue the progressive loss of motor function.
Focussing on a role for the key antioxidant Keap1/Nrf2 signalling pathway, we sought to investigate genetic and pharmacological inactivation of the negative regulator Keap1 in the C9orf72 Drosophila models as well as C9orf72 patient-derived iNeurons.Our results suggest that mitochondrial oxidative stress is an upstream pathogenic mechanism and activation of the Keap1/Nrf2 pathway could be a viable therapeutic strategy for ALS/FTD.

Drug treatments
For paraquat treatment, standard Drosophila food was supplemented with paraquat (Sigma, 856177) to a final concentration of 10 mM.For dimethyl fumarate (DMF, Sigma, 242926) adult treatment, DMF (or an equivalent volume of ethanol for the vehicle control) was added into a sugar-yeast (SY) medium (32) consisting of 15 g/L agar, 50 g/L sugar and 100 g/L yeast to a final concentration of 7 μM.All flies were transferred into freshly prepared supplemented food every 2-3 days.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
Larval crawling: Larval crawling was conducted by using wandering third instar (L3) larvae.
Each larva was placed in the middle of a 1 % agar plate, where they were left to acclimatise for 30 seconds.After, the number of forward and backward peristaltic waves were counted for 60 seconds and recorded.

Climbing:
The repetitive iteration startle-induced negative geotaxis (RISING, or 'climbing') assay was performed using a counter-current apparatus as previously described (Greene et al., 2003).Briefly, groups of 15-22 flies were placed in a temperature-controlled room for 30 minutes for temperature acclimatisation and transferred to test tubes for another 30 minutes.
Flies were placed into the first chamber, tapped to the bottom, and given 10 seconds to climb a 10 cm distance.20-day-old flies were given 20 seconds to climb to account for overall reduced mobility.Flies that reached the upper portion, i.e., climbed 10 cm or more, were shifted into the adjacent chamber.After five successful trials, the number of flies in each chamber was counted and the average score was calculated and expressed as a climbing index.
Survival: For survival lifespan experiments, groups of 20 males each were collected with minimal time (<30 s) under light anaesthesia and placed in separate food vials in standard food and transferred every 2-3 days to fresh food and the number of dead flies recorded.
Percent survival was calculated at the end of the experiment using https://flies.shinyapps.io/Rflies/(Luis Gracia).
Deaths were recorded two times per day, and flies were transferred every 2-3 days to fresh tubes.

Immunohistochemistry and sample preparation
For mitophagy analysis with mito-QC reporter, autophagy analysis with GFP.mCherry.Atg8a reporter and mitochondrial morphology analysis, third instar larval brains were dissected in PBS and fixed in 4% formaldehyde (FA) (Thermo Scientific, 28908) in phosphate buffered saline (PBS) for 20 minutes at room temperature (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 (Invitrogen, P36961) 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 and permeabilised in 0.3 % Triton X-100 in PBS (PBS-T) for 30 minutes, blocked with 1% bovine serum albumin (BSA) in PBS-T for 1 hour at RT. Tissues were then incubated with rabbit anti-CncC (kind gift from Dr Fengwei Yu), diluted in 1% BSA in PBS-T overnight at 4 °C, then washed 3 times 10 min with PBS-T, followed by incubation with secondary antibodygoat anti-rabbit IgG H&L Alexa Fluor™ 488 (1:500, Invitrogen, A11008).The tissues were washed in PBS-T with 1:10,000 Hoechst (Invitrogen, H3570) for 20 mins followed by 2 times 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 minutes.Brains were washed three times for 20 minutes in PBS-T, prior to blocking for 4 hrs in 4% normal goat serum (NGS) in PBS-T.Tissues were then incubated with rabbit anti-CncC (kind gift from Dr Fengwei Yu), diluted in 4 % NGS in PBS-T overnight at 4 °C, then washed 3 times 10 min with PBS-T, followed by incubation with secondary antibody for 2 hours.The tissues were washed in PBS-T with 1:10,000 Hoechst for 20 mins followed by 2 times PBS washes and mounted on slides using Prolong Diamond Antifade mounting medium.

ROS analysis
MitoSOX: 2-3 10-day-old adult brains were dissected at a time in PBS, incubated in 20 μM of MitoSOX Red (Invitrogen, M36008) for 30 minutes in the dark, washed with PBS for three times, mounted on poly-L-lysine coated wells on a 1.5 mm coverslip and imaged live immediately.The maximum intensity of projected z stacks from imaged brains was quantified using ImageJ.Genetic roGFP2 reporters: Mitochondrial and cytosolic ROS imaging was performed using the mito-roGFP2-Orp1, mito-roGFP2-Grx1 and cyto-Grx1-roGFP2 reporter lines.Third larval instar brains were dissected in HL3 and placed in a drop of HL3 on poly-L-lysine coated wells on 1.5mm coverslip and imaged by excitation at 488 nm (reduced) or 405 nm (oxidized), with emission detected at 500-530 nm.The maximum intensity of projected z-stacks from imaged brains was quantified using FIJI (Image J) and the ratio of 405/488 nm was calculated.

Microscopy
Fluorescence imaging was conducted using a Zeiss LSM 880 confocal microscope (Carl Zeiss MicroImaging) equipped with Nikon Plan-Apochromat 63x/1.4NA oil immersion objectives.
Images were prepared using FIJI (Image J).For mito-QC imaging, the Andor Dragonfly spinning disk microscope was used, equipped with a Nikon Plan-Apochromat 100x/1.45NA oil immersion objective and 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
Mitophagy: Confocal images were processed using FIJI (Image J).The quantification of mitolysosomes was performed as described in (33) using Imaris (version 9.0.2) analysis software.Briefly, a rendered 3D surface was generated corresponding to the mitochondrial network (GFP only).This surface was subtracted from the mCherry signal which overlapped with the GFP-labelled mitochondrial network, defining the red-only mitolysosomes puncta with an estimated size of 0.5 μm and a minimum size cut-off of 0.2 μm diameter determined by Imaris.
Autophagy: The quantification of autolysosomes was performed using FIJI (Image J) with the 3D Objects Counter Plugin.An area of interest was selected by choosing 6-10 cells per image.
The threshold was based on matching the mask with the fluorescence.A minimum size threshold of 0.05 μm 3 was set to select autolysosomes.
Mitochondrial morphology: After acquisition of images, each cell was classified using a scoring system where morphology was scored as fragmented, WT/tubular or fused/hyperfused.All images were blinded and quantified by three independent investigators.Data presented in Figure 3 and Supplementary Figure 2 were conducted concurrently, therefore the control groups are the same but replicated in the two figures for ease of reference.
Nuclear CncC quantification: All acquired images were taken with the same laser and gain during acquisition, which allowed a threshold to be set in FIJI (Image J) that was consistent for all images.For each brain, using the Hoechst signal, 10 nuclei from the central part of the larval CNS and 10 from the periphery were quantified to minimise bias.This was overlaid onto the CncC channel and the mean intensity within the nuclei was measured using the ROI manager.All images were blinded before quantification.

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.Twenty 5-day-old adult fly heads per replicate for each genotype was 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/L fatty acidfree 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 into 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 was 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).
Membranes were blocked with 5 % (w/v) dried skimmed milk powder (Marvel Instant Milk) in Tris-buffered saline (TBS) with 0.1 % Tween-20 (TBS-T) for 1 hour at RT and probed with the appropriate primary antibodies diluted in TBS-T overnight at 4 ℃.After three 10-minute washes in TBS-T, the membranes were incubated with the appropriate horse radish peroxidase (HRP-conjugated) secondary antibodies diluted in 5 % milk in TBS-T for 1 hour at RT. Membranes were washed three times for 10 minutes in TBS-T and detection was achieved with Amersham ECL-Prime detection kit (Cytiva RPN2232).Blots were imaged using the Amersham Imager 680 with further analysis and processing using FIJI (Image J).

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 (Fisherbrand, 15555799) for tissue preparation.400 μL of TRI Reagent (Sigma, T9424) was added and placed into Minilys homogeniser (Bertin Technologies) where the programme was set to maximum speed for 10 seconds.The samples were placed back on ice for 5 minutes before two further rounds of lysis.Direct-zol™ RNA MiniPrep kit (Zymo Research, R2050) was used to extract RNA following manufacturer's instructions.TURBO DNA-free™ Kit (Invitrogen, AM1907) was used to remove contaminating DNA by following manufacturer's instructions.cDNA synthesis was achieved by using Maxima™ H Minus cDNA Synthesis Kit (Thermo Scientific, M1681) following manufacturer's instructions.Equivalent (500 μg) total RNA underwent reverse transcription for each sample.Finally, qRT-PCR was ran using Maxima SYBR Green/ROX Kit (Thermo Scientific, K0221) following manufacturer's instructions using the Quant Studio 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 CT method entering into account PCR primer efficiency (35).
The following primers were used in this study:
Cells were treated with 30 μM or 100 μM EDV, or DMF at concentrations of 3 μM or 10μM.

Live cell imaging assays of iNeurons
Cells were stained for 30 minutes with Hoechst (Sigma, B2883) and MitoSOX Red (Invitrogen, M36008) at concentrations of 20 μM and 500 nM, respectively.Cells were imaged using an Opera Phoenix high content imaging system, with analysis performed using a custom protocol on Harmony software (PerkinElmer).

Fixed cell imaging of iNeurons
On day 18 of differentiation, cells were fixed in 4% paraformaldehyde for 30 minutes.After PBS washes, cells were permeabilised with 0.1% Triton X-100 for 10 minutes and blocked with 5% horse serum (Sigma, H0146) for 1 hour.Cells were incubated with primary NRF2 antibody (1:1000, Abcam, ab31163) overnight at 4 ℃.Cells were washed with 0.1% Tween in PBS and incubated with donkey anti-rabbit IgG H&L Alexa Fluor™ 568 secondary antibody (1:1000, Invitrogen, A10042) and 1 μM Hoechst prior to imaging.Cells were imaged using an Opera Phoenix high content imaging system, with analysis performed using a custom protocol on Harmony software (PerkinElmer).

Statistical analysis
GraphPad Prism 9 (RRID:SCR_002798) was used to perform all statistical analyses.Climbing

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 (38).To comprehensively assess different aspects of mitochondrial dysfunction in C9orf72-related pathology in vivo, we utilised several previously established Drosophila models based on the inducible GAL4-UAS expression system (39).Expression of 36x G4C2 repeats (G4C2x36) and 36-repeat glycinearginine DPR (GR36) are substantially neurotoxic compared to the 3x G4C2 repeats (G4C2x3) control (6).Indeed, we found that strong pan-neuronal expression of G4C2x36 (via nSyb-GAL4) perturbed development, where larvae were morphologically thinner compared to control (Fig. 1A), and with significantly reduced motor ability assessed by measuring larval crawling behaviour (Fig. 1B).Consistent with previous reports, pan-neuronal expression of GR36 caused even stronger phenotypes (Fig. 1A, B), and was developmental lethal at the third-instar (L3) larval stage.
In order to analyse C9orf72 pathology in the adult stage, where the impact of ageing can also be investigated, we found that 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 motor neuron expression of G4C2x36 did not impact motor behaviour at 2-days-old, by 10 days, G4C2x36 flies exhibited significantly impaired climbing ability (Fig. 1C).As before, motor neuron expression of GR36 was more toxic, causing a significant motor deficit at 2-days-old which worsened by 10 days (Fig. 1C).A similar pattern was observed in the lifespan of DIPγ-GAL4 driven G4C2x36 and 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 muscle-directed expression of poly-GR (22,40).Here we aimed to explore the potential involvement of mitochondrial dysfunction in a neuronal context.Initially, we observed that pan-neuronal expression of G4C2x36 caused a significant reduction in complex I-and complex II-linked respiration in lysates from young (5-day-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 pan-neuronal expression of the C9orf72 transgenes.As mitochondria are dynamic organelles, microscopy analysis of mito.GFPlabelled 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, F).Using this approach, we found that mitochondria were more elongated and hyperfused in G4C2x36 and GR36 expressing neurons (Fig. 1E, 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 (generates superoxide anions) and hydrogen peroxide (H2O2 -generates hydroxyl radicals).We observed that G4C2x36 flies were hypersensitive to both types of oxidative stressors when compared to control animals (Fig. S1B, 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 vs 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 utilised genetically encoded redox-sensitive fluorescent protein (roGFP2) probes where fusion of roGFP2 to oxidant receptor peroxidase 1 (Orp1) or glutaredoxin 1 (Grx1) results in roGFP2 oxidation by H2O2 or oxidised glutathione (GSSG), respectively, which leads to a shift in the excitation maxima of the fusion constructs from 488 to 405 (41).Importantly, these reporters can be targeted to specific subcellular compartments, i.e., cytosol or mitochondria, giving insights into compartment-specific redox status.We observed a significant increase in the oxidised status of mitochondrial roGFP2-Orp1 (Fig. 1G, H) and mitochondrial roGFP2-Grx1 (Fig. 1I, J) in larval neurons when combined with G4C2x36 and GR36, indicating an increase in mitochondrial H2O2 and GSSG in these animals.In contrast, no significant differences were observed in cytosolic glutathione redox potential (EGSH) (Fig. S1D), and little or no changes were detected with dihydroethidium (DHE) intensity, an indicator of cytosolic superoxide, in G4C2x36 and GR36 larvae (Fig. S1E).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-day-old G4C2x36 flies compared to G4C2x3 control (Fig. 1K, L).Taken together, these data suggest that mitochondrial but not cytosolic redox state is disrupted in these Drosophila models of C9orf72-related pathology.
These results are summarised in Table 2. 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 (33) 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 (OMM).Mitolysosomes are marked when GFP is quenched by the acidic environment but mCherry fluorescence is retained, resulting in 'red-only' puncta (33).
Analysis showed that there were significantly fewer mitolysosomes in G4C2x36 and GR36 larval neurons compared to G4C2x3 control (Fig. 1M, N), suggesting that mitophagy is perturbed.However, this was not a mitophagy-specific defect as the analogous reporter for general autophagy, GFP-mCherry-Atg8a autophagy reporter (fly Atg8a is homologous to LC3), also showed a reduction in the number of mCherry-positive autolysosomes in G4C2x36 and GR36 larval neurons (Fig. S1F).Consistent with this, immunoblot analysis of protein lysates from 5-day-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. S1G, H).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 (42).
Recently, West et al. (27) developed additional C9orf72 models which express physiologically relevant C9orf72-related DPRs, i.e., ~1000 repeats.For comparison to 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 pan-neuronal expression of GR1000, where GR1000 flies were not able to climb at all by 20 days (Fig. 2A), similar to that previously reported (27).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-day-old GR1000 brains compared to control (Fig. 2B, 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.

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 and GR36 expression may rescue the larval locomotor defect.Thus, we combined pan-neuronal expression of G4C2x36 and 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 was partially reversed by these manipulations as expected (Fig. S2A, 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 (43).We have also seen that USP30 knockdown increases mitolysosome number in Drosophila neurons and muscle (44), indicative of increased mitophagy.As expected, we observed that USP30 knockdown significantly increased the number of mitolysosomes in control larval neurons (Fig. S2D, E).However, when co-expressed with G4C2x36 and GR36, USP30 knockdown was not sufficient to rescue the reduced mitophagy (Fig. S2D, E).Moreover, pan-neuronal co-expression of G4C2x36 and GR36 with USP30 RNAi did not improve the larval locomotor deficit either (Fig. S2F).These data suggest that mitophagy is also not a primary cause but a downstream consequence in C9orf72 pathogenesis.

Overexpression of mitochondrial Sod2 and catalase ameliorates C9orf72 motor phenotypes
Previous studies have shown that targeting oxidative stress may be beneficial in C9orf72related 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 (45).Since an increase in mitochondrial ROS was observed in G4C2x36, GR36 and GR1000 flies, we hypothesised that 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) as well as a mitochondrially targeted catalase (mitoCat) -with G4C2x36 and GR36.Contrary to our prediction, 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 coexpressing Sod1 did not crawl during the assay conditions (Fig. 3B).In contrast, 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).Exploring the basis for these differential effects we found that expression of Sod1 was reduced at both mRNA (Fig. S2G) and protein levels (Fig. S2H) in G4C2x36 flies, while Cat expression was increased but Sod2 was 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, D).
Interestingly, here 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, F).To validate this genetic rescue, we used the GR1000 lines where we also found that Sod2 or Cat overexpression significantly rescued the age-related locomotion defect (Fig. 3G, H), while mitoCat expression did not reach significance.
Taken together, amongst all the different genetic manipulations used to reverse mitochondrial phenotypes observed, only overexpression of certain antioxidant enzymes such as mitochondrial Sod2 and catalase were beneficial in rescuing the locomotor deficits.This suggests that oxidative stress is an important upstream pathway in pathogenesis.

The Keap1/Nrf2 pathway is partially activated in C9orf72 flies
ROS act as important signalling molecules to coordinate cellular homeostasis and maintain redox balance.Potentially harmful levels of ROS activate redox sensor pathways such as Keap1/Nrf2 which can then upregulate a range of antioxidant genes upon stimulation.Under basal conditions, levels of Nrf2 are kept low by constitutive proteasomal degradation guided by Keap1 functioning as a substate adapter promoting Nrf2 ubiquitination.Upon oxidative stress, the modification of critical reactive cysteine residues on Keap1 leads to released binding of Nrf2, allowing Nrf2 to translocate to the nucleus and activate the expression of a series of antioxidative and cytoprotective genes (23,24).
Drosophila encode homologues of Keap1 and Nrf2 (called cap 'n' collar C isoform, CncC) (31), so we sought to explore whether this pathway is involved in C9orf72 pathology.First, by 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 to the G4C2x3 control (Fig. 4A, C).We also observed a similar increase in CncC nuclear staining in 5-day-old adult brains of G4C2x36 animals (Fig. 4B, C).This indicates that there is an activation of the Keap1/CncC signalling pathway at early stages which remains activated during disease pathogenesis.
To further analyse the activity of this signalling pathway, we made use of a reporter transgene which expresses GFP under the control of GstD1, an Nrf2 target and prototypical oxidative stress response gene (31), to monitor antioxidant responses.Pan-neuronal expression of both G4C2x36 (Fig. 4D) and GR1000 (Fig. 4E) caused an increase in GstD1-GFP levels compared to 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.Again, we saw that the GstD1-GFP reporter expression was significantly increased in G4C2x36 flies under basal conditions (− PQ) compared to controls (Fig. 4F, G).However, while GstD1-GFP levels further increased upon PQ treatment in both control and G4C2x36 conditions (Fig. 4F, G), this response was proportionally less than the response in a control background (Fig. 4H).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 appears to be insufficient to confer full protection.

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 (46).Indeed, combining a heterozygous mutant of Keap1 with pan-neuronally driven G4C2x36 and GR36 significantly improved the larval motor ability of both G4C2x36 and GR36 compared to 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).
The beneficial effects of partial genetic loss of Keap1 in the C9orf72 models supports the potential for pharmaceutical agents that modulate Keap1 to also be beneficial in this context.
We tested this hypothesis using dimethyl fumarate (DMF), a Keap1-modifying, Nrf2-activating drug with antioxidative and anti-inflammatory properties.DMF acts by the inactivation of Keap1 via succination of its cysteine residues (47), previously shown to exhibit neuroprotective effects in animal models of neurodegeneration (48).In order to assess the effects of extended DMF treatments on motor phenotypes, we expressed G4C2x36 and GR36 via DIPγ-GAL4 and treated the adult flies with food supplemented with DMF or vehicle (ethanol).After 3 days, G4C2x36 flies did not show a phenotype while the stronger phenotype of GR36 was not significantly improved (Fig. 5D).However, after 10 days of treatment, the decline in locomotor performance in vehicle treated G4C2x36 and GR36 flies was prevented by DMF (Fig. 5D).
Similarly, while no benefit was observed at day 3 or day 10 in GR1000 flies with DMF treatment, by day 20, GR1000 flies showed a partial rescue in climbing compared to vehicle control (Fig. 5E).
To test whether DMF treatment improves mitochondrial function in the G4C2x36 model, we employed the mito-roGFP2-Orp1 reporter to measure 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. 5F).Taken together, these data indicate that inhibiting Keap1 by DMF treatment can 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 healthy age-matched controls and ALS patients carrying C9orf72 mutations (Table 1) were reprogrammed into induced neural progenitor cells (iNPCs), then differentiated into induced neurons (iNeurons) (49).Assessing the level of oxidative stress in these cells using MitoSOX, we found a modest increase in mitochondrial ROS in C9orf72 iNeurons (Fig. 6A, B), in line with that observed in Drosophila.
We also found that there was a higher level of nuclear Nrf2 in patient iNeurons (Fig. 6C, D) mirroring the C9orf72 Drosophila phenotypes.
Testing two different concentrations, 24-hour treatment with 3 μM or 10 μM DMF was able to reduce the mitochondrial ROS levels in C9orf72 iNeurons (Fig. 6A, B).In comparison, we analysed as a positive control treatment with edaravone (EDV), a known ROS scavenger and FDA-approved ALS drug, and observed that 24-hour treatment with 30 μM and 100 μM EDV reduced ROS levels in C9orf72 iNeurons to a similar extent as DMF (Fig. S3A, B).Although significant, the effect size of DMF or EDV treatment is small and the impact on neuronal function or survival is currently unknown.Therefore, more investigation is required to validate these initial results.Nevertheless, taken together, these data suggest that activating the Keap1/Nrf2 pathway by 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.

Discussion
We have conducted a comprehensive characterisation of mitochondrial dysfunction using three different models of C9orf72 ALS/FTD and found disrupted morphology with hyperfused mitochondria, reduced mitophagy, impaired respiration and increased mitochondrial ROS production, all in a neuronal context in vivo.Genetic interaction studies showed that only overexpression of mitochondrial Sod2 and catalase were able to significantly rescue C9orf72 behavioural phenotypes, and further rescued other mitochondrial deficits such as mitophagy.
Together, these data suggest a causative link between mitochondrial dysfunction, ROS and behavioural phenotypes.
ROS has been well-characterised in the context of ALS (38), however, it is important to understand the relative contributions of different ROS species in different cellular compartments in order to identify more targeted treatment options.Addressing this, we have observed a robust increase in mitochondrial ROS whereas the effect on cytosolic ROS was more variable or unaltered.Consistent with the compartmentalised effects, we found that overexpressing mitochondrial Sod2 or a mitochondrially-targeted Cat partially rescued C9orf72 phenotypes, as did overexpression of cytosolic Cat, which suggests that mitochondrial superoxide produced is rapidly dismutated causing a cytosolic effect hence both cytosolic and mitochondrially-targeted Cat is beneficial.
We were surprised to find that Sod1 overexpression exacerbated C9orf72 phenotypes, contrary to previous findings from Lopez-Gonzalez et al. (20).SOD1 is a cytosolic ubiquitous enzyme with several functions, primarily involved in scavenging superoxide as well as modulating cellular respiration, energy metabolism and posttranslational modifications (50).
Mutations in SOD1 also cause ALS, most likely via misfolding of the protein, causing a toxic gain of function, although the exact mechanism leading to motor neuron death is still elusive (51).Although we observe reduced levels of Sod1 mRNA transcripts and protein, silencing SOD1 in mice does not itself cause neurodegeneration (52).However, it would be interesting to investigate whether Sod1 misfolding or aggregation may contribute to C9orf72 pathogenesis.Indeed, Forsberg et al. (53) found abundant inclusions containing misfolded WT SOD1 were found in spinal and cortical motor neurons from patients carrying mutations in C9orf72.Taken together, these findings highlight the importance of studying compartmentalised ROS effects in the context of disease pathology.
While defective autophagy has previously been noted in C9orf72 Drosophila and other models (54), we extend these observations to provide novel in vivo evidence that mitophagy is also perturbed.However, a mechanistic link between disrupted mitophagy and the upstream oxidative stress and modulation of the Nrf2 pathway is lacking.Nrf2 activity is known to promote mitochondrial homeostasis at multiple levels including biogenesis and turnover (55).
Notably, the autophagy adapter p62 both competes with Nrf2 for Keap1 binding (56) and is an Nrf2 target gene, thereby creating a positive feedback loop (56,57).Moreover, Nrf2 activation was shown to induce mitophagy independently of the Pink1/parkin pathway, and rescue Pink1/parkin phenotypes in vivo (58,59).Therefore, it will be interesting to investigate a potential link between C9orf72-Nrf2/CncC-p62/ref(2)P-mitophagy/autophagy as it may provide mechanistic insight, in all our model systems.
The protective role of the well-known antioxidant and cytoprotective Keap1/Nrf2 pathway has been discussed as a therapeutic target for treatment against many neurodegenerative diseases including ALS (24).Investigating this, we found increased nuclear localisation of Nrf2 and concomitant upregulation of the GstD1-GFP reporter, a proxy for Nrf2 activation, indicating that the pathway is upregulated upon C9orf72 toxicity.However, despite the initiated activation, this was clearly insufficient to prevent excessive ROS and the resulting neurotoxicity which could be prevented by transgenic overexpression of antioxidant enzymes.
While the reasons behind the blunted pathway activation are unclear, it is encouraging that both genetic reduction of Keap1, as well as its pharmacological targeting with DMF, could suppress C9orf72 toxicity in vivo, and in patient-derived iNeurons.
While some studies have reported the Nrf2 pathway to be disrupted in ALS patient samples (60)(61)(62), very limited data exists on C9orf72-related ALS samples.Recently, Jimenez-Villegas et al. (63) have shown that Nrf2 activation is impaired in cell culture models of arginine-DPR toxicity, where they also saw an improvement in cell viability upon treatment with DMF.We have extended these findings showing the benefit of targeting Keap1/Nrf2 in vivo as well as patient-derived iNeurons which further emphasises the importance of Nrf2 activation as a potential therapeutic target.Encouragingly, a recent phase 2 clinical trial to test the efficacy of DMF in sporadic ALS patients was conducted (64).However, while the study concluded there was efficacy on the primary endpoint (Amyotrophic Lateral Sclerosis Functional Rating Scale-Revised (ALSFRS-R) score), there was a reduced decline in neurophysiological index, suggesting preservation of lower motor neuron function.The authors also noted that participants showed an 'unusually slow disease progression' and that a larger trial is needed for verification (64).
In conclusion, our results provide compelling evidence that mitochondrial oxidative stress is an important upstream pathogenic mechanism leading to downstream mitochondrial dysfunction such as alterations in mitochondrial function and turnover.Consequently, targeting one of the main intracellular defence mechanisms to counteract oxidative stressthe Keap1/Nrf2 signalling pathway -could be a viable therapeutic strategy for ALS/FTD.While DMF treatment shows promise, more research is needed to understand the underlying mechanisms behind disease pathogenesis and progression.

Table 1 . Cell lines used in this study
was analysed using Kruskal-Wallis non-parametric test with Dunn's correction for multiple comparisons.Data are presented as mean ± 95% confidence interval (CI).Quantifications of larval crawling, number of mitolysosomes and autolysosomes, WB and qRT-PCR were analysed using one-way ANOVA with Bonferroni post hoc test for multiple comparisons.Data are presented as mean ± SD.Mitochondria morphology was quantified using Chi squared test.
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.