Fly stocks were maintained at 25°C on standard cornmeal-agar medium (water: 36 l; yeast: 720 g; cornmeal: 3200 g; soy flour: 400 g; light malt extract: 3200 g; sugar beet syrup: 880 g; agar: 320 g, Nipagin 120 g). The crosses between the GAL4 drivers and the UAS responder lines were carried out at 25°C unless stated. UAS constructs were used in heterozygous configurations for all experiments. The stocks used in this work as well as their origin are listed in Table S1 of Datasheet 1 in the Supplementary Material. Genetic interactions were carried out by first generating the stocks fhRNAi-1/CyO; Repo-GAL4/TM6B tub-GAL80 and fhRNAi-1/CyO; Mef2-GAL4/TM6B tub-GAL80.

cDNA for the mtRosella construct was kindly provided by Prof. Dr. Benedikt Westermann (Böckler and Westermann, 2014). A 1700 bp fragment was amplified using the primers mtRosellaFw (CTCGAGATGTGGACTCTCGGGCGCCG) and mtRosellaRv (GGTACCTCAAGCATCTTTTCCGGAATAGGCCAAG). The full length coding sequence of mtRosella was cloned into the Gateway^® entry vector pENTR/D-TOPO^® by BP-cloning (Invitrogen, Carlsbad, CA, USA) and was subcloned by LR-cloning (Invitrogen, Carlsbad, CA, USA) into PWH-attB plasmid vector (modified from PWH obtained from the Drosophila genomics Resource Center and kindly provided by Prof. Michael Krahn) generating the transgene UAS-mtRosella-PWH. The transgene was then integrated into the third chromosome of the Drosophila genome using the PhiC31 integrating cassette. Three independent lines were obtained for the construct.

Total RNA was extracted from 15 thoraces using peqGold TriFast reagent (PEQLAB Biotechnologie GMBH, Erlangen, Germany) following manufacturer’s instructions. Five-hundred nanogram mRNA were converted into cDNA using QuantiTect Rev. Transcription Kit (Qiagen GmbH, Hilden, Germany) and then used for qPCR with ORA qPCR Green ROX L Mix (HighQu, Kralchtal, Germany) on a CFX connect™ Real-Time PCR Detection System (Bio-rad, Hercules, CA, USA). The ribosomal protein 49 (rp49) was used as internal control. The results from at least four independent biological replicates were analyzed with the Bio-Rad CFX manager 3.1 software. Gene expression levels were referred to the internal control, the relative quantification was carried out by means of the ΔΔCt method and the results were plotted as relative mRNA expression. Each experiment consisted of 3–5 independent biological replicates. The genes and the pairs of primers used for the analysis are summarized in Table S2 of Datasheet 1 in the Supplementary Material.

For DNA extraction, thoraces from 15 to 20 male flies of desired age were homogenized in 400 μl Buffer A (100 mM Tris, pH7.5, 100 mM EDTA, 100 mM NaCl, 0.5% SDS). After incubation at 65°C for 60 min with continuous shaking, 800 μl of LiCl/KAc (5 M KAc and 6 M LiCl in a relation 1:2.5) solution were added, followed by 10 min incubation on ice. After centrifugation (15 min, 13,000 rpm, RT), 1 ml supernatant was transferred in a new tube. The DNA was precipitated by adding 600 μl of isopropanole and vortexing. After centrifugation (15 min, 13,000 rpm, RT) the resulting pellet was washed with 70% ethanol and resuspended in 200 μl H₂O. Isolated DNA was quantified by measuring the absorbance at 260 nm with a NanoDrop spectrophotometer (PEQLAB Biotechnologie GMBH, Erlangen, Germany) and frozen at −20°C. In order to estimate the amount of mtDNA, the levels of mitochondrial cytochrome oxidase I (Fw 5′-AACTGTTTACCCACCTTTATCTGCTB-3′ and Rv 5′-CCCGCTAAGTGTAAAGAAAAAATAGC-3′) were referred to the levels of the housekeeping gene GAPDH (Fw 5′-GACGAAATCAAGGCTAAGGTCG-3′ and Rv 5′-AATGGGTGTCGCTGAAGAAGTC-3′), using 10 ng of Template DNA per reaction. Reactions were performed with qPCR Green ROX L Mix (HighQu, Kralchtal, Germany) on a CFX connect™ Real-Time PCR Detection System (Bio-rad, Hercules, CA, USA). The relative quantification was carried out by means of the ΔΔCt method and the results were plotted as Relative mtDNA content. Each experiment consisted of 3–5 independent biological replicates.

For Western blots, 15 larvae (L2 and L3), adult male flies or thoraces were homogenized in 80 μl cold RIPA buffer (Sigma-Aldrich, St. Louis, MO, USA) with complete protease inhibitor mixture (Roche, Mannheim, Germany). The protein levels were determined using the BCA protein quantification kit (Thermo Scientific, Schwerte, Germany). Samples were diluted 1:4 with standard 4× Laemmli buffer, boiled for 5 min and approximately 25–30 μg of proteins were separated on 5% stacking-10% separating SDS polyacrylamide gels. The resolved proteins were electroblotted to a Protran BA 85 nitrocellulose membrane (Hartenstein, Würzburg, Germany) and were probed using rabbit anti-p62 (1:3000, kindly provided by Gábor Juhász, Pircs et al., 2012), rabbit anti-phospho eIF2αS1 (1:1500, ab32157, abcam, Cambridge, UK) and rabbit anti-GFP (1:2500, A6455, Invitrogen, Carlsbad, CA, USA). Mouse anti-α-tubulin (1:5000, T9026, Sigma-Aldrich, St. Louis, MO, USA) and rabbit anti-actin (1:2500, Sigma-Aldrich, St. Louis, MO, USA) were used as loading controls. Secondary fluorescent goat anti-mouse 680 LT and goat anti-Rabbit 800 CW (1:10,000 and 1:5000 respectively, Li-Cor Inc., Bad Homburg, Germany) were used in all cases. Detection and quantification was conducted using the Odyssey system (Li-Cor Inc., Lincoln, NE, USA). Arbitrary fluorescent units were normalized to the internal loading control. Data were expressed as percentage of average control values and plotted as relative protein levels. At least three independent biological replicates were used for each genotype and condition. The original images of representative blots shown or quantified in this manuscript highlighting the area displayed in the corresponding figures are available in the Datasheet 2 in the Supplementary Material.

Flies of appropriate age and genotype were fixed in 4% PFA for 2 h. Brains or thoraces were then dissected, washed with PBST and blocked with 10% Normal Goat Serum or 10% Normal Donkey Serum, depending on secondary antibody, and incubated 24 h at 4°C with the primary antibody: mouse anti-Repo (1:200, 8D12, Developmental Studies Hybridoma Bank, Iowa city, IA, USA), goat anti-GFP (1:100, 600-101-215, Rockland, PA, USA), rabbit anti-p62 (1:500, kindly provided by Sébastien Gaumer, Nezis et al., 2008) and rat anti-BiP (1:200, BT-GB-143P, Babraham Bioscience Technologies, Cambridge, UK). Then, samples were incubated 24 h at 4°C with secondary fluorescent antibodies: donkey anti-goat AF488, donkey anti-rabbit AF594, donkey anti-mouse AF555, goat anti-rabbit AF488, goat anti-rat AF555 (in all cases 1:200, Invitrogen, Carlsbad, CA, USA), goat anti-mouse Cy3 (1:200, Jackson Dianova, Hamburg, Germany), DAPI (10 μg/ml, Sigma-Aldrich, St. Louis, MO, USA) and TRITC-conjugated Phalloidin to visualize F-actin from muscle fibers (1:500, Sigma-Aldrich, St. Louis, MO, USA). Samples were embedded in Vectashield mounting medium (Vector Laboratories, Burlingame, CA, USA). At least 5–10 flies per genotype were scanned. GFP and antibody patterns were analyzed by scanning the samples using the multi-tracking setups of either a LSM 510 Confocal Microscope (Zeiss, Germany) or Confocal Laser Scanning Platform Leica TSC SP8 (Leica, Germany).

In order to analyze mitochondrial membrane potential in indirect flight muscles (IFM), flies were anesthetized with CO₂ and the head, abdomen, wings and legs were removed. The thorax was cut sagitally with a sharp injector blade (Stainless Steel Injector Blades, Cat. #71990, Electron Microscopy Sciences, Hatfield, PA, USA). The hemithoraces were then incubated with 1 μm tetramethylrhodamine, ethyl ester (TMRE; Cat. #89917, Sigma-Aldrich, Germany) in Relaxing solution (20 mM KH₂PO₄, 5 mM MgCl₂, 5 mM EGTA, 5 mM ATP) for 15 min, washed with Relaxing Solution three times for 5 min, mounted in VectaShield and immediately imaged.

All images were acquired using the same exposure, light intensity and filter settings for all conditions. Confocal images were further processed with the image processing software Fiji 2.0.0 (Schindelin et al., 2012). In detail, background was subtracted in ImageJ via the “Rolling Ball” method (Radius = 50 pixels). Maximum projections of three (muscle images) to 30 (brain images) slices were made and the resulting image was again subjected to background subtraction. Finally, contrast of each image was adjusted to improve quality of signal. Further details about confocal microscope settings and image handling are described in the Supplementary methods.

Paraffin sections were performed from 35-day-old adult flies. Flies were fixed with carnoy (ethanol:chloroform:acetic acid at a proportion 6:3:1), dehydrated in ethanol, and embedded in paraffin. Paraffin sections (7 μm) from 10 flies of each genotype were analyzed under a fluorescence microscope. Brain vacuolization was quantified using the ImageJ 1.48v software. The affected area was referred as % of total brain area that includes the lamina (La), the outer chiasm (OC) and the medulla (Me). For examination of lipid accumulation with light microscopy and analysis of degeneration of photoreceptor neurons, semithin epon plastic sections from 35-day-old and 5-day-old adult fly brains, respectively, were prepared. Lipids were identified by characteristically dark droplet-like structures.

Total aconitase activity was determined from whole individuals using Bioxytech Aconitase-340™ Spectrophotometric Assay kit (OXIS International Inc., Portland, OR, USA). Twenty-five adults or 15 thoraces were placed in a tube containing 500 μl of ice-cold homogenization buffer (0.2 mM sodium citrate and 50 mM Tris-HCl, pH 7.4) and gently pounded. The resulting mashes were centrifuged twice at 800 g for 10 min at 4°C discarding the pellet every time. Supernatants were disrupted by sonication for 30 s with 1 s interval between each pulse. One-hundred microliter of each extraction was used to measure the aconitase activity. The aconitase activity of each genotype was measured from four independent samples and each extraction was measured twice. The aconitase activity was finally referred to protein amount using the Bradford assay (Coomassie Plus™ Protein Assay Reagent, Thermo Scientific, Schwerte, Germany). ATP was determined using ATP Bioluminescence Assay Kit CLS II (Roche, Mannheim, Germany) according to manufacturer’s instructions with some modifications. Briefly, five thoraces were homogenized in 200 μl pre-heated ATP assay buffer (100 mM Tris, 4 mM EDTA, pH 7.75), boiled for 2 min and centrifuged for 1 min at 1000 g. The supernatant was diluted 1:3 in ATP assay buffer and 50 μl of the dilution were used to measure ATP levels. Luminescence was detected with a Tecan Infinite^® F500 plate reader (Tecan Trading AG, Switzerland). Each experiment consisted of 3–5 independent biological replicates.

In all experiments (normal conditions and food supplementation), flies were raised at 25°C under a 12 h:12 h light/dark cycle and male individuals were collected within 24–48 h posteclosion, placed at a density of 25 per vial and transferred to vials with the corresponding fresh food every 2–3 days. Lifespan experiments were conducted in standard cornmeal agar medium and 100–150 males per genotype were used in each experiment. For chemical treatments, Instant Drosophila Medium (Formula 4–24, Carolina Biological, Whitsett, NC, USA) was rehydrated either with 15 mM Tauroursodeoxycholic acid (TUDCA, Merck Miilipore, MA, USA), 7.5 mM 4-Phenylbutyric acid (4-PBA, Sigma-Aldrich, St. Louis, MO, USA) or 50 μm Rapamycin (LC Laboratories, Woburn, MA, USA). As control treatment, Instant Drosophila Medium was reconstituted with distilled water or DMSO diluted in distilled water. Hyperoxia treatment started 1 day post-eclosion and was performed by exposing flies in a glass container with a constant flux of 99.5% oxygen under a low positive pressure at 25°C (Botella et al., 2004). Flies were transferred every day to new vials. Locomotor assays were performed as previously described (Botella et al., 2004). Ten to twelve flies per genotype were assessed and each fly was recorded three times and the mean value of each fly was used for the subsequent analysis.

Survival data were analyzed using the Log-rank (Mantel-Cox) and Gehan-Breslow-Wilcoxon tests. In all further experiments, data is represented as mean ± SEM of 4–5 replicates. When comparing two samples, equal variances were confirmed by an F test. When comparing multiple samples, equal variances were confirmed by Bartlett and Brown-Forsythe tests. Normality of data was assessed in all cases with Shapiro and Wilk test and parametric or non-parametric tests were used accordingly. For data that passed normality test, significance was determined by two-tailed T-test or by One-way ANOVA with post hoc Dunnett or Tukey Multiple Comparison Test (***P < 0.001; **P < 0.01 and *P < 0.05). Statistical analysis was carried out using Prism version 7.02 for Windows, GraphPad Software, La Jolla, CA, USA¹.

All figures were assembled with Adobe Photoshop CC 2017.0.0 (Adobe Systems) by importing microscopy images from Fiji and graphs from Prism.

Downstream effects of frataxin downregulation such as altered transport, number, size and morphology of mitochondria (Puccio et al., 2001; Calmels et al., 2009; Shidara and Hollenbeck, 2010; Chen et al., 2016b) indicate defects in the stability, integrity and homeostasis of this organelle that might be contributing to the pathobiology of FRDA. However, the impact of mitochondrial dynamics on frataxin downregulation phenotypes has never been studied. This encouraged us to further investigate whether genetic manipulation of mitochondrial homeostasis might also provide beneficial effects in frataxin-deficient flies. We have reported previously that frataxin knockdown in glia affects fly locomotion, increases brain vacuolization due to cellular degeneration and triggers accumulation of lipids (Navarro et al., 2010). We now first assessed whether altered expression of genes involved in mitochondrial homeostasis had any influence on the locomotor performance of frataxin-deficient flies. Thus, fhRNAi-1/CyO; Repo-GAL4/TM6B Tub-GAL80 flies (Navarro et al., 2015) were mated with a collection of fly lines to increase or downregulate the expression of genes involved in mitochondrial fusion and fission (Marf, Opa1 and Drp1), mitochondrial quality control (Pink1 and Parkin) and mitochondrial biogenesis (Spargel, Srl). We decided to add a second group of genes involved in mitochondrial motility (Miro, Milton and Khc) since they have been shown to be important for glial function (Schmidt et al., 2012). Due to the known positive effects of antioxidant genes in FRDA flies (Anderson et al., 2008; Navarro et al., 2010), we also included in our analysis a constitutively expressed mitochondrial catalase (mitocatalase) construct as a secondary control of the reliability of our strategy. As expected, frataxin-deficient flies in glia showed reduced climbing ability (0.8 cm/s) compared to controls (1.5 cm/s). Importantly, coexpression of two neutral transgenes such as UAS-GFP and UAS-mcherryRNAi did not significantly alter the locomotion of FRDA flies. In addition, frataxin overexpression (UAS-fh) significantly rescued locomotor performance. As expected, expression of mitocatalase also ameliorated locomotor dysfunctions of frataxin-deficiency in glial cells (1.3 cm/s). All these results support the suitability of our experimental setup to test for genetic interactions.

Our results, summarized in Figure 1A, showed that among all lines tested only Mitofusin (Marf) downregulation, using an RNAi line extensively described in the literature (Deng et al., 2008; Dorn et al., 2011; Celardo et al., 2016), had a strong positive impact. Marf knockdown in glial cells rescued FRDA locomotion up to control levels (1.6 cm/s). Remarkably, compared to frataxin-deficient flies coexpressing the mcherryRNAi transgene (Figures 1B,H and quantification in Figure 1F), Marf silencing also suppressed brain vacuolization (Figure 1C and asterisks in Figure 1I and quantification in Figure 1F) and restored the lipid dysfunction as seen by the reduced number and size of lipid droplets in the fly brains (arrows in Figure 1I) with a similar aspect to controls (Figure 1G). In agreement, Marf overexpression worsened brain degeneration (Figure 1D and quantification in Figure 1F). A second RNAi line (TRiP) confirmed the ability of Marf silencing to rescue locomotor performance (Figure 1A) and the increased brain vacuolization (Figure 1E and quantification in Figure 1F) of frataxin deficiency.

Our results highlight Marf as a key molecule in the neurological dysfunction in FRDA.

Marf has been extensively described as a mediator of mitochondrial fusion in mammals as well as in Drosophila (Westermann, 2010; Corti et al., 2011) and it has been reported to be a substrate of the ubiquitin ligase Parkin in the mitophagy process (Pallanck, 2013; Wang et al., 2016). Therefore, the genetic interaction with Marf downregulation might indicate alterations in mitochondrial morphology and degradation upon frataxin silencing in glial cells.

In order to address this question, we decided to focus on one type of glial cell, the giant glial cells (GGCs) of the inner and outer optic chiasms on both sides of the medulla (Tix et al., 1997). These glial cells are a specialized form of ensheathing glia that wraps axonal tracts which project between neuropils. They are suitable for our analysis because of their large size and their location in the region that shows strong brain vacuolization upon frataxin silencing (Navarro et al., 2010). Mitochondria were visualized in glia by means of a mitochondria-targeted GFP (mitoGFP). The analysis of mitoGFP pattern between controls (Figures 2A,B; UAS-mitoGFP/+; Repo-GAL4/UAS-mcherryRNAi) and frataxin-deficient flies (Figures 2C,D; UAS-mitoGFP/+; Repo-GAL4/fhRNAi-1) did not reveal any remarkable difference in the distribution or morphology of mitoGFP signals throughout the glia. It is well known that reduction of frataxin expression affects mitochondrial function. However, the small cell size of glia and the structure of the Drosophila brain impedes the analysis of mitochondrial activity in our experimental scenario. Importantly, damaged mitochondria are normally degraded by a special form of autophagy called mitophagy (Westermann, 2010). Similarly to mammalian p62, the Drosophila ortholog Ref(2)P (hereafter referred to as p62) is also a classical marker of autophagic flux in the fly (Nezis et al., 2008; Pircs et al., 2012; Mauvezin et al., 2014). Interestingly, it has been recently reported that fly p62 might also be useful to monitor mitochondrial clearance by mitophagy upon organelle dysfunction (Wang et al., 2016). Reduction of frataxin expression in glial cells induced accumulation of p62 in our region of interest (denoted area, Figure 2F) compared to controls (Figure 2E). However, the difference could not be observed by western blot (Figure 2G), likely because of a general overall accumulation of p62 in aged brains (Nezis et al., 2008). Despite the accumulation of p62, targeted silencing of frataxin did not induce loss of GGCs (Figure 2H). Our results might suggest that frataxin deficiency in glial cells either blocks autophagy or increases the number of mitochondria labeled for degradation.

In order to address this question, we expanded our analysis to indirect flight musculature (hereafter referred to as muscles). Drosophila muscles comprise a large number of mitochondria and therefore, this tissue is extremely sensitive to mitochondrial perturbations which can be easily visualized and studied by histological approaches (Greene et al., 2003; Clark et al., 2006; Deng et al., 2008). By means of two different RNAi lines, our results demonstrate that frataxin knockdown in muscles reproduces the main features of the disease including reduced longevity and locomotion and impaired aconitase activity, ATP production and mitochondrial membrane potential (Supplementary Figure S1) in agreement with Lodi et al. (1999) and Vorgerd et al. (2000). Compared to controls (Supplementary Figures S2A, 3A), we also found that mitochondria in frataxin-deficient muscles display an age-dependent mitochondrial fragmentation from day 1 on (normal morphology, Supplementary Figure S2B) to day 10 (fragmented, Figure 3B). In line with the fragmentation, analysis of the expression profile of genes involved in mitochondrial dynamics and quality control showed important alterations (Supplementary Figure S2E). Interestingly, an increased mitochondrial fragmentation, as we have observed in frataxin-deficient muscles, is consistent with mitochondria being targeted for clearance by mitophagy (Palikaras and Tavernarakis, 2014). Indeed, frataxin-deficient muscles displayed a clear increase of p62 positive signals (Mef2-GAL4/fhRNAi-1; Figure 3D) compared to controls (Mef2-GAL4/UAS-mcherryRNAi; Figure 3C). Western blot analysis also showed that p62-levels were 4–8 times higher in frataxin-deficient flies than in controls and that they increased by 2-fold between 1-day-old and 10-day-old flies (Figure 3E). This was further confirmed by the analysis of number and size of p62-positive signals at three different ages (Supplementary Figures S2H–J). Interestingly, our results also show that mitochondrial dysfunction preceeds the organelle fragmentation since p62 starts to accumulate and mitochondrial TMRE signal is already decreased in 1-day-old flies (Figure 3E and Supplementary Figure S1L, respectively). Remarkably, p62 formed vesicle-like structures that are engulfing damaged mitochondria for degradation (Figure 3D and Supplementary Figure S2L). In most of the cases, similar results were observed with a second independent RNAi line for frataxin (fhRNAi-2, Supplementary Figure S2). Furthermore, mitochondrial DNA (mtDNA) content positively paralleled p62 levels (Supplementary Figure S2M) suggesting a clear mitochondrial accumulation. Our results show that frataxin deficiency in muscles alters mitochondrial dynamics and confirms that frataxin downregulation in the fly triggers accumulation of p62. Interestingly, the formation of p62-positive vesicles suggests a specific effect on mitophagy rather than a general autophagy dysfunction.

On the one hand, the detection of p62-positive vesicles indicates that the mitochondrial quality control mechanism driven by pink1 and parkin is working properly in frataxin-deficient flies. On the other hand, accumulation of p62 inversely correlates with autophagic degradation (Bartlett et al., 2011; Pircs et al., 2012) and might indicate a failure of the autophagosome-lysosomal pathway. Therefore, we decided to study the autophagic flux in frataxin-deficient cells analyzing the amount and cellular distribution of two autophagy markers, LAMP1GFP and GFPATG8a, in both tissues (Figure 4). Our Western Blot analysis in glial cells (Figures 4A,B) showed that the levels of both proteins were decreased by around 80% in frataxin-deficient brains. Accordingly, levels of free GFP were increased (Figure 4C). Appearance of cleaved GFP has been previously linked to an efficient flux (Mauvezin et al., 2014; Bargiela et al., 2015). These results suggested an increased mitophagic flux. The small size of lysosomes, autophagosomes and mitochondria in glia (Supplementary Figures S3A–D and Figure 5A) prevented a reliable and robust histological analysis to confirm our findings. We then decided to address this question in the fly muscles. Compared to controls (Supplementary Figures S3F,G), we found a strong accumulation of punctate signals of both autophagy markers in frataxin-deficient muscles (arrows in Figures 4D,E) along with a partial colocalization of p62 with LAMP1GFP or GFPATG8a (Figure 4F, 20% and 70%, respectively). This result indicated a correct recruitment of the autophagic machinery to form autophagosomes (GFPATG8a), although formation of the autophagolysosomes might be a limiting step (LAMP1GFP). Similarly to glial cells, molecular analysis in muscles also suggested enhanced cleavage and turnover of LAMP1GFP, albeit to a lesser extent (Supplementary Figure S3E). In order to study whether the problem may lie in the inability to complete the degradation of the mitochondria, we have generated transgenic flies to express the mtRosella transgene and perform analysis of mitophagy in vivo. This construct consists of a mitochondria-targeted dsRed fluorescent protein fused to a pH-sensitive GFP named pHluorin. Under normal conditions, the green and red signals overlap in mitochondria, whereas, upon mitophagy, degenerating mitochondria are transferred to lysosomal compartments and the green signal is lost due to the low pH (Rosado et al., 2008). We have observed that, in controls, all mitochondria clearly showed both signals (Figure 4G). In frataxin-deficient flies, dysfunctional mitochondria were first marked with p62, which still showed both mtRosella signals (arrow heads in Figure 4H). Then, damaged organelles were incorporated into the autophagosome and, after fusion with lysosomes, the acidic pH eliminated the green sensitive signal from the pHluorin (asterisks in Figure 4H). Finally, the process of mitochondrial clearance was completed and the red signal from mtRosella marker was also depleted (arrows in Figure 4H). As expected, all mitochondria that were not enveloped by p62-vesicles displayed both signals. Quantification of the population of mitochondria that have entered the mitophagy process (p62-positive mitochondria) showed three groups of organelles: a reduced fraction (7%) displayed both signals (Figure 4I, yellow bar), a substantial 23% only displayed dsRed indicating incorporation into lysosomal compartment (Figure 4I, red bar) and importantly, the majority (70%) of the labeled mitochondria seemed to have lost both mtRosella signals (Figure 4I, white bar).

Altogether, our results clearly exclude mitophagy dysfunction in frataxin-deficient flies and thus, accumulation of p62 may suggest that the normal cellular capacity is insufficient to deal with the increasing number of damaged mitochondria. Therefore, we tested whether promotion of autophagy might improve FRDA conditions. Expression of the autophagy marker GFPATG8a in glial cells was accompanied by a clear reduction of p62 levels (Supplementary Figure S4A) and of brain vacuolization (Supplementary Figures S4B,D) compared to the expression of LAMP1GFP (Supplementary Figures S3B, S4C,D, respectively). Since it might be controversial whether GFPATG8a is a functional protein, we repeated the experiment using an independent UAS-ATG8a line and reduction of p62 levels and brain vacuolization were also observed (Supplementary Figures S4D–F). The effect of genetic autophagy induction was also detected in muscles. Our experiments revealed that overexpression of GFPATG8a triggered a strong reduction of p62 levels (40%) in FRDA muscles compared to the expression of two other GFP markers such as LAMP1GFP and ERGFP (Supplementary Figure S4G). However, chemical induction of early autophagy by means of rapamycin failed to improve locomotion and brain degeneration upon depletion of frataxin in glia (Supplementary Figures S4H,I).

Since promotion of autophagy had a positive impact on FRDA phenotypes, we decided to analyze whether the rescue of frataxin-deficiency, which is triggered by Marf downregulation, was due to the enhancement of mitochondrial clearance or to any additional modification of the mitochondrial network. Marf knockdown in frataxin-deficient glial cells (Figure 5B) did not produce any visible effect on mitochondrial morphology and distribution compared to frataxin-deficient glia (Figure 5A). In agreement, neither reduction of Opa1 (Marf partner in mitochondrial fusion) nor overexpression of Drp1 (fission mediator) were able to ameliorate the locomotor dysfunction (Figure 1A) and the brain vacuolization triggered by glial frataxin deficiency (Figures 5E–H and quantification in Figure 5M). These results suggest that fusion and fission of mitochondria are not the driving force underlying mitofusin-mediated rescue. Next, we tested whether Marf silencing was able to improve autophagic flux and to reduce p62 accumulation. Marf knockdown in frataxin-deficient glial cells failed to reduce p62 levels (Figures 5C,D). In this line of findings, suppression or overexpression of the mitochondrial quality control genes Pink1 and Parkin (master regulators of mitophagy) failed to ameliorate any frataxin-deficient phenotypes such as negative geotaxis (Figure 1A), brain degeneration (Figures 5I–L and quantification in Figure 5M) and p62 accumulation (Figure 5N).

Remarkably, these results suggest that Marf silencing rescued FRDA phenotypes without improving mitochondrial function and that the protection mediated by Marf inhibition is mainly independent of Marf roles in mitochondrial fusion and clearance by mitophagy.

In mammalian cells, mitofusins are also involved in the interface between mitochondria and endoplasmic reticulum (ER). These connections facilitate the exchange of lipids and calcium and promote mitochondrial fission (van Vliet et al., 2014). Interestingly, Marf downregulation has been shown to induce ER stress in Drosophila muscles (Debattisti et al., 2014). ER stress activates in turn the unfolded protein response (UPR). Such a response is mediated by pancreatic ER kinase (PERK), inositol-requiring enzyme 1 (IRE1) and activating transcription factor 6 (ATF6), which trigger three cascades characterized by the increased expression of downstream transcription factors named ATF4, Xbp1 and the chaperone BiP/GRP78 (reviewed in Hetz and Mollereau, 2014). Importantly, the UPR is conserved in Drosophila and has already been associated with some fly models of neurological disorders (reviewed in Ryoo, 2015). This motivated us to determine whether frataxin silencing in flies was also affecting the ER morphology and stress.

First, we tested in vivo in 35-day-old brains whether frataxin-deficiency in glia cells induced some markers of ER stress such as increased levels of the protein BiP and of a Xbp1GFP reporter, which is only expressed in-frame with GFP when IRE1-mediated splicing of Xbp1 occurs (Sone et al., 2013). Surprisingly, we detected that spliced Xbp1 was constitutively expressed in nuclei of control cells (Figure 6A), whereas its expression was completely abolished upon frataxin deficiency (Figure 6B). The lack of Xbp1 induction in FRDA glial cells was already evident in young flies (Figure 6F) compared to age-matched controls (Figure 6E). Interestingly, we found that FRDA flies exhibited increased BiP immunoreactivity (Figure 6D) compared to controls (Figure 6C) in the region of interest (arrows in Figure 6C). Activation of ER stress pathways in the glia was not accompanied with evident morphological defects in the ER (Supplementary Figures S5A,B). In order to exclude the influence of off-targets in the ER stress response observed upon targeted silencing of frataxin in glia, altered levels of Xbp1 reporter as well as chaperone BiP were analyzed after coexpression of human frataxin (UAS-FXN). Human frataxin is insensitive to RNAi contructs against fly frataxin and has already improved frataxin-deficient conditions in other tissues (Navarro et al., 2011; Tricoire et al., 2014). As expected, expression of FXN restored Xbp1 expression (Figure 6G) and reduced BiP levels (Figure 6H) in frataxin-deficient glia. The altered ER stress response was not restricted to glial cells. Frataxin knockdown in muscles also triggered the activation of all three UPR branches (Figures 7A–D). In this case, frataxin-deficient muscles displayed a strong nuclear signal of the Xbp1 reporter, whereas controls did not show any activation (Figures 7A,B). Frataxin-deficient muscles also showed elevated levels of BiP (Figure 7C) and of phosphorylated α subunit of eukaryotic translation initiation factor 2 (eIF2α; Figure 7D), a known marker of ER stress via PERK signaling (Celardo et al., 2016). An increased clustering of ER cisternae (Supplementary Figures S5C,D) was also detected in FRDA muscles. Moreover, we also took advantage of a severe loss-of-function allele of fly frataxin (fh1, Chen et al., 2016b) and of a second RNAi line that decreases frataxin expression to levels similar to those observed in patients (fhRNAi-2, Llorens et al., 2007). In both cases (Figures 7E,F), severe and moderate ubiquitous loss of frataxin also increased the levels of phosphorylated eIF2α. However, since phospho-eIF2α levels can be also increased by other pathways (Baker et al., 2012), we cannot completely rule out that other mitochondrial stress responses are also influencing the increase detected in fh1 and fhRNAi-2 flies. These results clearly suggest that frataxin-depletion dramatically alters the ER stress response ubiquitously and specifically in glia and muscles.

Next, we addressed whether Marf was modulating the ER stress response in frataxin-deficient glia. Remarkably, Marf downregulation using two independent RNAi lines was sufficient to strongly reduce BiP expression in 80% of the frataxin-deficient flies analyzed (Figures 8B,D,E) compared to frataxin-deficient flies coexpressing mcherryRNAi (Figure 8A) and to restore expression of spliced Xbp1 in frataxin-deficient GGCs (Figure 8D). It is important to note that Marf downregulation in glia did not seem to trigger any ER stress response by itself (Figure 8C), whereas it does in Drosophila muscles (Debattisti et al., 2014). Our results suggest that mitofusin mediates the ER stress response in frataxin-deficient glia and therefore Marf downregulation is able to compensate the altered ER stress response and concomitantly reverses the locomotor dysfunction, the brain degeneration and the accumulation of lipid droplets (Figures 1A,C,I), the main hallmarks of frataxin downregulation in the Drosophila glia (Navarro et al., 2010).

Finally, to further analyze the impact of ER stress in our FRDA model, we decided to test whether two chemical chaperones, TUDCA and 4-PBA, which have been shown to attenuate ER stress (Debattisti et al., 2014; Celardo et al., 2016), were sufficient to improve frataxin-deficient phenotypes. Uptake of 15 mM TUDCA or 7.5 mM PBA was able to clearly reduce the BiP immunoreactivity of FRDA flies (Figures 9A–C) in the region of interest (arrows in Figure 9A) and concomitantly, both were able to promote a strong reduction of around 45% of the brain vacuolization triggered by frataxin depletion in glia (Figures 9D–G) which is similar to Marf downregulation (65%, Figure 1F). However, TUDCA and PBA did not improve the locomotion defects induced by glial-targeted frataxin silencing (Figure 9H) suggesting that other factors besides ER stress might be involved in this defect. Following the strategy described by Chen et al. (2016b), we generated mosaic mutant clones of Drosophila photoreceptor neurons by means of the Flippase/Flippase Recognition Target (FLP/FRT)-system and we analyzed the morphology of the photoreceptors in 5-day-old flies upon treatment with TUDCA. Compared to untreated flies (Figure 9I), application of the drug during development and adulthood helped to maintain the structural integrity of the photoreceptors and reduced the degeneration (Figure 9J). We and others have already shown that frataxin silencing with fhRNAi-2 decreases aconitase activity under oxidative stress (Llorens et al., 2007; Navarro et al., 2010, 2015). Moreover, this model has been reported as an interesting tool to test drugs that might improve FRDA conditions (Soriano et al., 2013; Calap-Quintana et al., 2015). As shown in Figure 9K, TUDCA and PBA were able to moderately increase aconitase activity (25%) in frataxin deficient flies (fhRNAi-2/+; actin-GAL4/+) without affecting the aconitase activity of controls (actin-GAL4/UAS-mcherryRNAi).

These experiments clearly indicate a positive effect by reducing the ER stress in frataxin-deficient flies and links ER stress with the neurodegeneration observed in the FRDA nervous system. Our results highlight ER stress as a crucial mediator in FRDA as well as a new cellular process for therapeutic interventions.