Ras-ERK-ETS inhibition alleviates neuronal mitochondrial dysfunction by reprogramming mitochondrial retrograde signaling

Mitochondrial dysfunction activates the mitochondrial retrograde signaling pathway, resulting in large scale changes in gene expression. Mitochondrial retrograde signaling in neurons is poorly understood and whether retrograde signaling contributes to cellular dysfunction or is protective is unknown. We show that inhibition of Ras-ERK-ETS signaling partially reverses the retrograde transcriptional response to alleviate neuronal mitochondrial dysfunction. We have developed a novel genetic screen to identify genes that modify mitochondrial dysfunction in Drosophila. Knock-down of one of the genes identified in this screen, the Ras-ERK-ETS pathway transcription factor Aop, alleviates the damaging effects of mitochondrial dysfunction in the nervous system. Inhibition of Ras-ERK-ETS signaling also restores function in Drosophila models of human diseases associated with mitochondrial dysfunction. Importantly, Ras-ERK-ETS pathway inhibition partially reverses the mitochondrial retrograde transcriptional response. Therefore, mitochondrial retrograde signaling likely contributes to neuronal dysfunction through mis-regulation of gene expression.


Introduction
The use of ATP as a universal currency of energy transfer makes this molecule essential for life. ATP is generated either by glycolysis in the cytosol, or through the action of the tricarboxylic acid (TCA) cycle and β-oxidation of fatty acids coupled to oxidative phosphorylation (OXPHOS) in mitochondria. The mitochondrial electron transport chain (ETC) couples the transfer of electrons to the pumping of protons into the inter-membrane space [1]. This creates a membrane potential (ΔC), which is used by the mitochondrial ATP synthase to convert to mitochondria, but upon mitochondrial dysfunction a fraction of ATFS-1 localises to the nucleus where it regulates the expression of UPR mt genes [27,28].
ATFS-1 is not conserved in mammals and the RTG genes are not present in metazoans, but analogous retrograde signaling pathways have been identified that enable mitochondria to reprogram nuclear gene expression [29,30]. Transcriptional studies have shown that mitochondrial dysfunction elicits large scale changes in nuclear gene expression in diverse cell types in a range of model systems [22,23,31]. However, the molecular basis of mitochondrial signaling is still poorly understood, particularly in the nervous system [29,30]. Importantly, it is not known whether the retrograde transcriptional response contributes to mitochondrial dysfunction phenotypes or is protective. To address these questions, we performed a genetic screen and identified 30 genes that modify mitochondrial dysfunction in the Drosophila wing, implicating several new pathways in the mitochondrial retrograde response. Manipulation of one of the identified pathways, Ras-ERK-ETS signaling, also alleviates the effects of mitochondrial dysfunction in the Drosophila nervous system. Transcriptomic and functional analyses suggest that mitochondrial retrograde signaling is reversed and transcriptionally reprogrammed by Ras-ERK-ETS inhibition to restore neuronal function.

Knock-down and overexpression of TFAM cause mitochondrial dysfunction and activate mitochondrial retrograde signaling in the Drosophila wing
Mitochondrial retrograde signaling is activated in response to mitochondrial dysfunction. To induce mitochondrial dysfunction, we manipulated the expression levels of the mitochondrial DNA binding protein/transcription factor TFAM. TFAM expression is essential for mitochondrial DNA maintenance (S1A Fig) and gene expression, but overexpression of TFAM in mice and human cells also causes reduced mitochondrial gene transcription and mitochondrial dysfunction [32][33][34][35]. In Drosophila, both ubiquitous TFAM knock-down (S1B-S1E Fig) or TFAM overexpression [36] cause reduced mitochondrial gene expression and lethality at the larval stage. To develop a rapid, genetically modifiable assay for mitochondrial dysfunction in vivo we tested TFAM knock-down or TFAM overexpression in the wing using MS1096-Gal4. Strong TFAM overexpression in the wing causes late pupal lethality, while weak TFAM overexpression or TFAM knock-down (using several independent RNAi transgenes) results in a curved adult wing phenotype (Fig 1B, S1F, S1G, S1J-S1L Fig). Importantly, the TFAM knockdown curved wing phenotype is enhanced by heterozygosity for a loss-of-function mutation in TFAM (TFAM c01716 ) (Fig 1C, S1H Fig)  We previously showed that the gene Thor, encoding the eukaryotic initiation factor 4E binding protein (4E-BP), is a mitochondrial retrograde signaling response gene in neurons [36]. Both knock-down and overexpression of TFAM in the wing imaginal disc cause increased Thor expression (Fig 1H-1K). TFAM overexpression causes a strong increase in Thor expression throughout and beyond the dorsal compartment, while TFAM knock-down causes increased Thor expression in discreet patches within the dorsal compartment (Fig 1H-1K). As a result, using qRT-PCR from whole wing discs we could only detect increased Thor expression, or increased expression of the mitochondrial unfolded response pathway target gene Hsp22 [37] with TFAM overexpression (S2Y and S2Z Fig). These data show that mitochondrial dysfunction caused by TFAM knock-down and TFAM overexpression activate mitochondrial retrograde signaling in the wing.
In order to perform a genetic screen for modifiers of mitochondrial retrograde signaling, flies were generated that stably express the TFAM RNAi transgene, together with the TFAM c01716 mutation, in the wing. These flies, referred to as 'MitoMod' for 'Mitochondrial Modifier', have a distinctive~45˚curve at the tip of the wing (Fig 1C). To test the sensitivity of MitoMod flies to mitochondrial perturbation they were crossed to lines carrying RNAi or overexpression transgenes for genes associated with familial Parkinson's disease that have mitochondrial associated functions. Transgenes that cause a wing phenotype when expressed on their own were excluded to avoid additive effects (S1 Table). Overexpression or knockdown of Pink1, overexpression of parkin, knock-down of Lrrk and overexpression and knockdown of DJ-1α and DJ-1β all enhance the MitoMod wing phenotype (Fig 1L and 1M, S3 Fig and S1 Table). Therefore, the MitoMod wing phenotype provides a sensitised background for identifying mitochondrial retrograde signaling genes in vivo.

A genetic screen for genes that modify mitochondrial retrograde signaling identifies Aop
MitoMod flies were used to perform a genetic modifier screen of 646 RNAi lines, targeting 579 genes (Fig 2A). This RNAi collection was enriched for lines that target genes expressed in the nervous system and genes encoding chromatin remodelling factors (S2 Table). RNAi lines that cause a phenotype when expressed alone in the wing were excluded (S2 Table), to avoid additive effects with the MitoMod wing phenotype. Stringent criteria were used to identify interacting genes: only RNAi lines that caused a strong reproducible enhancement or suppression, which was also replicated by an independent RNAi line targeting the same gene were classed as hits. 25 genes were identified that enhance the MitoMod wing phenotype (Fig 2F-2I, S3 Table). Gene ontology (GO) analysis shows that these genes are involved in a range of biological processes and functions (Fig 2J, S3 Table). Five genes were identified that suppress the MitoMod phenotype (Fig 2C-2E, S4 Table). The suppressor genes function in chromatin remodelling or transcriptional regulation (Fig 2K, S4 Table). Overall, the variety in function of genes identified in the screen is consistent with the multifunctional cellular roles of mitochondria.
To test whether suppression of the mitochondrial dysfunction phenotype in the adult wing reflects reduced apoptosis, cleaved Drosophila effector caspase (Dcp-1) expression was analysed in the wing during development. Knock-down of TFAM in the wing causes increased Dcp-1 expression, but this is reduced to control levels by knock-down of the suppressors identified in the screen Aop, Ino80, Chrac-14, Ing3 and MTA1-like (S4A- S4I Fig). Moreover, knock-down of Aop or Ino80 suppress the increase in apoptosis caused by TFAM overexpression (S4J Fig). These data show that suppression of the adult wing phenotype in MitoMod flies reflects a reduction in apoptosis.

Inhibition of Ras-ERK-ETS signaling improves function in neurons with a mitochondrial deficit
The effects of mitochondrial dysfunction are particularly acute in the nervous system and manipulation of retrograde signaling may be a potential strategy to alleviate these effects. We aimed to use the wing screen to identify genes involved in mitochondrial retrograde signaling in the nervous system. Aop (anterior open, also known as Yan), one of the suppressor genes identified (Fig 2K), is an E-twenty six (ETS) transcription factor and a target of the highly conserved Ras-ERK (mitogen-activated protein kinase (MAPK)) pathway. Treatment of cultured neuronal cells with the mitochondrial uncoupler carbonyl cyanide p-(trifluoromethoxy) phenylhydrazone (FCCP) causes aberrant ERK activation [38]. We therefore hypothesised that Ras-ERK signaling is mis-regulated by mitochondrial dysfunction in the nervous system and that manipulation of Ras-ERK signaling would modify the effects of neuronal mitochondrial dysfunction.
TFAM knock-down in the wing causes similar but weaker effects to TFAM overexpression (S1J-S1L Fig). We have previously shown that TFAM overexpression in motor neurons causes a dramatic reduction in pre-synaptic tetramethylrhodamine methyl ester (TMRM) positive and mitochondrial-GFP labelled mitochondrial number and volume and a robust adult climbing phenotype [36]. Overexpression of TFAM in neurons with D42-Gal4 also causes a failure of wing inflation in approximately 50% of flies, due to dysfunction of the CCAP neurons that release the neuropeptide bursicon, which activates wing inflation (S5L- S5N Fig) [36]. TFAM knock-down in motor neurons also causes a reduction in mitochondrial volume and reduced adult climbing, but these phenotypes are much weaker than with TFAM overexpression (S5A- To test whether increasing Aop affects neuronal function we overexpressed Aop alone in motor neurons, or together with TFAM. Overexpression of Aop in neurons causes a strong climbing deficit and complete failure of wing inflation (Fig 3C and 3D). Moreover, overexpression of Aop and TFAM together causes pupal lethality (Fig 3C and 3D), which is not affected by heterozygosity for the Ras85D loss-of-function allele Ras85D ΔC40B , consistent with Aop acting downstream of Ras.
The ETS domain transcription factor pointed (Pnt) is a second target of Ras-ERK signaling in Drosophila. To test whether reducing Pnt expression also modifies neuronal mitochondrial dysfunction we used both a validated RNAi targeting Pnt (S6A- We next looked for direct evidence of Ras-ERK-ETS pathway mis-regulation in neurons caused by mitochondrial dysfunction. Motor neurons overexpressing TFAM were stained with an antibody that recognises the activated (di-phosphorylated) form of ERK (dpERK). dpERK expression is significantly increased in larval motor neurons overexpressing TFAM (Fig 4). Clonal analysis of TFAM overexpression shows that activation of dpERK is both cell

Knock-down of Pnt rescues the active zone phenotype caused by mitochondrial dysfunction
Similar to mitochondrial dysfunction in late larval sensory neurons, which does not affect ATP levels due to a compensatory increase in glycolysis [39], overexpression of TFAM in motor neurons does not alter ATP levels (S7A- S7C Fig). Active zones are the sites of pre-synaptic neurotransmitter release at chemical synapses and are enriched for the protein complexes that regulate synaptic vesicle release and recycling. TFAM overexpression in motor neurons causes altered mitochondrial morphology in the cell body, a dramatic loss of pre-synaptic mitochondria and a reduction in the number of active zones at the larval neuromuscular junction (NMJ) [36]. Therefore, to further investigate how modulation of Ras-ERK-ETS signaling affects neuronal mitochondrial dysfunction we focused on the synaptic compartment. Neither knock-down of Aop or Pnt affects the severe loss of pre-synaptic mitochondria caused by TFAM overexpression ( (Fig 5G and 5K). These data show that Ras-ERK-ETS pathway inhibition does not affect the primary mitochondrial defect, but modifies the active zone phenotype caused by neuronal mitochondrial dysfunction.

Inhibition of Ras-ERK-ETS signaling improves function in Drosophila models of mitochondrial disease and Parkinson's
Mitochondrial dysfunction in humans causes rare primary mitochondrial diseases and is also associated with more common neurodegenerative diseases, including Parkinson's. We  hypothesised that targeting the retrograde response, through inhibition of Ras-ERK-ETS signaling, would be beneficial in Drosophila models of human disease associated with mitochondrial dysfunction. To test this we used pan-neuronal (nSyb-Gal4) knock-down of the OXPHOS complex IV subunit Surf1, a model for the primary mitochondrial childhood encephalomyelopathy Leigh syndrome, and park 25 homozygous mutant flies, a model for familial Parkinson's disease [36,40,41]. dpERK expression is increased in the ventral nerve cord (VNC) by pan-neuronal knock-down of Surf1, but not in park 25 (Fig 6G). To test whether the beneficial effects in these two models are specific to reduced Aop and Pnt expression, or a general property of Ras-ERK-ETS pathway inhibition, we used a loss-of-function allele of Ras85D (Ras85D ΔC40B ). Heterozygosity for Ras85D suppresses the climbing and wing inflation defects caused by knock-down of Surf1 (Fig 6E and  6F) and the climbing deficit in park 25 homozygous flies ( Fig 6H). Thus, targeting the retrograde response, through inhibition of Ras-ERK-ETS signaling, improves function in two independent models of human disease caused by mitochondrial dysfunction.

Inhibition of Ras-ERK-ETS signaling partially reverses the transcriptional response to neuronal mitochondrial dysfunction
Mitochondrial dysfunction transcriptionally reprograms cells by altering nuclear gene expression [22,29]. It is not known whether this transcriptional reprogramming is protective or damaging. Ras-ERK-ETS pathway inhibition could benefit neurons either by enhancing the retrograde transcriptional response, or by reversing the expression of genes that are mis-regulated in response to mitochondrial dysfunction. To understand the mechanism by which Ras-ERK-ETS signaling alleviates the effects of neuronal mitochondrial dysfunction, we performed transcriptomic analysis using central nervous system (CNS) tissue from larvae with pan-neuronal TFAM overexpression, Pnt knock-down, Aop knock-down, or TFAM overexpression combined with Pnt or Aop knock-down. The expression of 606 and 519 genes are significantly altered by knock-down of Pnt and Aop respectively (Fig 7A, S7 Table). 189 genes were regulated by both Pnt and Aop knock-down and the expression of these genes is strongly positively correlated ( Fig 7B, S8 Table). These transcriptomic data support our epistasis analysis (S6N and S6O Fig) and are consistent with Pnt and Yan both acting as positive regulators of Ras-ERK-ETS signaling in neurons.
The expression of 494 genes were significantly altered in control versus TFAM overexpression conditions, 671 genes in control versus TFAM overexpression combined with Pnt knockdown and 560 genes in control versus TFAM overexpression combined with Aop knockdown conditions (Fig 7C, S7 Table). Around a third of these genes are commonly mis-regulated between these conditions and the expression of these common genes is very strongly positively correlated (Fig 7D and 7E, S9 and S10 Tables). Therefore, the expression of around a third of genes mis-regulated by TFAM overexpression are unchanged by Pnt or Aop knockdown. These data show that Pnt or Aop knock-down does not enhance the transcriptional changes caused by mitochondrial retrograde signaling.
To understand how Pnt and Aop knock-down modify the mitochondrial retrograde response we directly compared the TFAM overexpression transcriptome to TFAM overexpression combined with Pnt or Aop knock-down (S7 Table). This comparison shows that in  Table). Of these, 82 genes (TFAM o/e vs TFAM o/e, Pnt RNAi) and 78 genes (TFAM o/e vs TFAM o/e, Aop RNAi) are also mis-regulated in control versus TFAM overexpression conditions (S11 and S12 Tables). Interestingly, the expression levels of these genes are strongly negatively correlated (Fig 7F and 7G). Around half of these genes are misregulated by both Pnt and Aop knock-down (S13 Table). Therefore, Pnt or Aop knock-down reverses the expression of a minor subset of genes that are mis-regulated in response to mitochondrial dysfunction. The enriched GO terms for these genes shows that Pnt or Aop knockdown reverses the expression of a number of functional classes, including transcriptional regulation and transmembrane helix, which may contribute to improved neuronal function (S14 Table). Alternatively, Pnt and Aop knock-down may alleviate the effects of mitochondrial dysfunction through the action of the genes that are regulated independent of retrograde signalling. To test this directly we used RNAi to knock-down Hsc70-2. Hsc70-2 encodes a chaperone of the heat shock 70 family and is the most strongly upregulated gene in TFAM overexpression conditions (S7 Table). Knock-down of either Pnt or Aop dramatically reduces the retrograde induced upregulation of Hsc70-2 expression (S11 and S12 Tables). Knock-down of Hsc70-2 in motor neurons suppresses the climbing and wing inflation phenotypes caused by TFAM overexpression (Fig 7H and 7I). These results are consistent with a mechanism whereby inhibition of Ras-ERK-ETS signaling alleviates the damaging effects of neuronal mitochondrial dysfunction by reversing the expression of a specific subset of genes within the mitochondrial retrograde transcriptome.

Discussion
Fluctuations in mitochondrial activity and function occur during the cell cycle, throughout development and in disease states. Changes in mitochondrial function affect the cell at multiple levels, but the cellular response to these homeostatic changes is very poorly understood. We have devised a method to identify genes potentially involved in mitochondrial retrograde signaling in vivo. The number and function of the genes identified suggests an extensive and orchestrated cellular response to mitochondrial dysfunction. Inhibition of one of the pathways identified in the screen, Ras-ERK-ETS signaling, also alleviates the effects of mitochondrial dysfunction in the Drosophila nervous system. Targeting Ras-ERK-ETS signaling also improves function in Drosophila models of Leigh syndrome and Parkinson's. Inhibition of Ras-ERK-ETS signaling partially reverses the mitochondrial retrograde transcriptional response, evidence that retrograde signaling contributes to neuronal dysfunction.
Proteomic and genetic methods have been highly successful in characterising the complement of proteins that make up the mitochondrion [42][43][44]. Human mitochondria consist of around 1158 proteins, only 13 of which are encoded by the mitochondrial genome [42]. However, mitochondria do not function in isolation and participate in a variety of cellular functions, acting within a homeostatic network that responds to changes in the cellular environment [22]. We have developed a sensitised phenotypic assay in the wing to identify genes involved in the cellular response to changes in mitochondrial activity. Using this for a genetic screen we identified 30 modifier genes, the majority of which enhance the wing phenotype. It is possible that screening in this way could identify components of the OXPHOS complexes, or other core mitochondrial proteins. However, knock-down of such genes on their own in the wing is likely to (and in our experience does) cause a wing phenotype, and so these genes would be excluded from the screen. We did identify several genes encoding cytosolic metabolic proteins (Pgk, Adk1, Pgi) as enhancers of mitochondrial dysfunction, which did not cause a phenotype when knocked-down by themselves but enhanced the MitoMod phenotype (S3 Table). However, a key point of this study is that many of the genes identified have roles in signal transduction and regulation of gene expression, strongly suggesting that mitochondrial dysfunction modulates the activity of a variety of cell signaling pathways.
The ETS domain transcription factor Aop was identified as a suppressor in the modifier screen and also suppressed neuronal mitochondrial dysfunction phenotypes. Knock-down or heterozygosity for pnt also rescued neuronal mitochondrial dysfunction phenotypes. These results are surprising, as Aop and Pnt generally act antagonistically to each other, with Aop acting as an inhibitor of Ras-ERK signaling and Pnt as a positive effector. In the canonical model, activated ERK phosphorylates both Pnt and Aop, promoting cytosolic translocation and degradation of Aop, as well as enhancing the transcriptional activity of Pnt [45,46]. In the wing Pnt and Aop have opposite effects: knock-down of Pnt (using JF02227) enhances, while knock-down of Aop suppresses the MitoMod phenotype. However, recent systems biology approaches have shown that the dynamic interplay between Aop and Pnt is more complex than previously thought and is context dependent [47][48][49]. Aop stability is in fact regulated differentially by Ras-ERK signaling depending on the neuronal differentiation state [49]. We find that PntP2 and Aop overexpression and expression of a constitutively active form of Ras all strongly inhibit neuronal function and act synergistically with TFAM overexpression. Knockdown of either Pnt or Aop suppress Ras-ERK pathway activation and the gene expression changes caused by Pnt or Aop knock-down in neurons strongly correlate. Moreover, reduced expression of Ras, Pnt or Aop are protective against the effects of neuronal mitochondrial dysfunction. Aop and Pnt therefore both act as positive effectors of Ras-ERK signaling in motor neurons and in the context of neuronal mitochondrial dysfunction.
Mitochondrial activity plays a key role in healthy ageing. It was recently shown that inhibition of Ras, through ubiquitous expression of dominant negative Ras, or Ras knock-down in adult flies extends lifespan [50]. In contrast to our study, expression of an activated form of Aop in the gut and fat body of adult flies extended lifespan, while knockdown of Aop in these tissues had no effect on longevity [50]. The authors did not test whether inhibition of Pnt affected lifespan, but Pnt overexpression significantly reduced the lifespan of wild-type flies. Furthermore, administration of a pharmacological agent, Trametinib, which inhibits Ras activation of ERK kinase, increased Drosophila lifespan. Although there may be differences in the role of Aop in neuronal mitochondrial dysfunction versus healthy ageing, this previous study and our work together point to the exciting possibility that inhibition of Ras-ERK-ETS signaling may be beneficial to both healthy ageing and human diseases associated with mitochondrial dysfunction. Our data also suggest that Ras-ERK-ETS signaling acts both cell autonomously and non-cell autonomously in response to mitochondrial dysfunction. In the developing Drosophila eye, Ras-ERK signaling determines cell autonomous photoreceptor cell fate, acting downstream of the EGF receptor [51]. The secretion of Spitz, the ligand for the EGFR, from adjacent cells is also regulated by the Ras-ERK pathway [52]. Ras-ERK signaling thus acts both cell autonomously and non-cell autonomously to control photoreceptor differentiation. Mitochondrial signaling in Drosophila has previously been shown to act through non-autonomous systemic negative regulation of insulin signaling [53]. The mechanism of regulation of Ras-ERK-ETS in ageing is not known, nor do we know how mitochondrial dysfunction activates this pathway in neurons. A variety of mitochondrial retrograde signals have been identified, including ROS, Ca 2+ , AMP, nicotinamide adenine dinucleotide (NAD+) and acetyl coenzyme A [30]. Future studies will identify the factor(s) that mediate mitochondrial retrograde signaling in the nervous system.
Large scale alterations in transcription in response to mitochondrial dysfunction have been observed in a wide range of cell types [23]. However, whether these altered transcriptomes have a functional consequence is not clear. Inhibition of Ras-ERK-ETS signaling restores neuronal function and pre-synaptic active zones in our models but does not appear to affect the primary mitochondrial defect. This is consistent with our transcriptomic data, which do not show large scale alterations in the expression of mitochondrial genes with Aop or Pnt Knockdown. We suggest that the beneficial effects of Ras-ERK-ETS pathway inhibition on mitochondrial dysfunction result from transcriptional reprogramming that leads to improvement in pre-synaptic structure and function.
Reduced expression of Aop or Pnt, transcriptional targets of Ras-ERK signaling, alleviate the effects of neuronal mitochondrial dysfunction. We exploited this finding to test how the transcriptome is affected by Aop and Pnt knock-down in the background of mitochondrial dysfunction. Surprisingly, a significant number of the mitochondrial retrograde transcriptional changes are reversed by Pnt or Aop knock-down. This finding suggests that the transcriptional mis-regulation activated by mitochondrial retrograde signaling at least partly contributes to neuronal dysfunction. In support of this idea, we find that knock-down of Hsc70-2, a retrograde response gene whose expression is reduced by Aop and Pnt knockdown, alleviates neuronal mitochondrial dysfunction phenotypes. Future analyses of these retrograde response genes will help elucidate the cellular mechanisms that contribute to neuronal dysfunction.
Individual mitochondrial diseases are rare, but in total affect up to 1 in 4300 in the population [7]. The nervous system is frequently affected by mitochondrial mutations, resulting in a wide range of clinical outcomes including ataxia, epilepsy, neuropathy and deafness [54]. Treatments for mitochondrial diseases are limited and mostly symptomatic. Manipulation of the response to mitochondrial dysfunction in neurons may provide a new potentially curative strategy for mitochondrial diseases. Ras-ERK signaling has key roles in synaptic plasticity, learning and memory [55]. However, Ras-ERK-ETS signaling has not previously been identified as potential therapeutic target for mitochondrial disease. Determining how the transcriptional targets of Ras-ERK-ETS signaling contribute to neuronal dysfunction will provide important new insight into mitochondrial diseases such as Leigh syndrome. Fly stocks were UAS-TFAM3M [36], which was used for all TFAM overexpression experiments except where the weaker UAS-TFAM10M [36] is stated, park 25 [40], UAS-Surf1 23.4 RNAi [41], Ras85D ΔC40B [56] and UAS-mito-roGFP2-Grx1 [57]. UAS-ATeam1.03NL (AT[NL]) and UAS-ATeam1.03RK (AT[RK]) flies [39] were from the Kyoto Stock Center (DGRC). The following fly stocks were from the Bloomington Stock Center: DJ-1α, DJ-1β, Lrrk, parkin and Pink1 RNAi and overexpression lines (details in S1 Table) 25 , Ras85D ΔC40B and park 25 , pnt Δ88 lines used in Fig 6 were generated by recombination. Mosaic analysis with a repressible cell marker (MARCM) analysis of FRT82B, TFAM c01716 was performed using y, w,hs-flp;tub-Gal4,UAS-mCD8GFP;FRT82B,tub-Gal80 flies as in Avet-Rochex et. al. [59]. Gene names are according to Flybase [60].

Genetic screen
Virgin female flies carrying the MitoMod genotype balanced with TM6B,tub-Gal80 (MS1096-Gal4; TFAM RNAi,TFAM c01716 /TM6B,tub-Gal80) were crossed to males carrying RNAi transgenes. 1-2 days after eclosion of the progeny wings were observed and scored in males. RNAi lines were only classed as enhancers if most flies had a !90ᵒ wing curve. RNAi lines were classed as suppressors if most flies had a <45ᵒ wing curve (S2 Table). Crosses from RNAi lines that enhanced or suppressed the MitoMod wing phenotype were repeated to confirm the result. To exclude RNAi lines that cause a phenotype by themselves, MS1096-Gal4 virgin females were crossed to all RNAi lines. If the progeny of this cross had a wing phenotype then the RNAi line was omitted from the MitoMod screen (S2 Table). Genes for all interacting RNAi lines were tested with independent RNAi lines, where available (S2 Table), and only classed as positive hits if the phenotype was replicated by the independent RNAi (S5 and S6 Tables).
RNAi lines were selected using gene expression data available on FlyAtlas, to select genes that are expressed more strongly in the brain than in the whole body [61]. RNAi lines for Drosophila chromatin remodelling genes were also used [62]. GO analysis of genes identified in the screen was performed using the Panther Classification System [63].

Behavioral analysis
Climbing assays were performed as previously described [36]. Males were used for all climbing assays, apart from experiments involving knock-down of Surf1, where females were used for all genotypes.
To quantify wing inflation, flies were transferred into a new vial after eclosion and left for at least 24 hours to allow time for normal wing inflation to occur. Numbers of flies with fully inflated, semi-inflated and uninflated wings were then recorded. All flies that eclosed from the vial were counted. Statistical analysis was performed on raw data and data displayed as a percentage.
For Förster resonance energy transfer (FRET)-based ATP biosensor imaging, AT[NL] ATP biosensor expressing and AT[RK] ATP insensitive expressing control wing imaginal discs or larval CNS tissues were dissected in Schneider's medium (Thermo Scientific) and imaged immediately at 21˚C using a 458nm excitation laser and detecting emitted light between 460-499nm (CFP) and 535-650nm (FRET) using a Zeiss LSM 710 confocal microscope. For control experiments wing imaginal discs were incubated in Schneider's medium with 100μM oligomycin (VWR)/50 mM 2-deoxyglucose (SLS) for 40 minutes at 21˚C, then imaged immediately in the same medium. CFP and FRET channel signal intensity at the same three randomly selected points in each wing disc, or three cell bodies in each VNC, was determined using the ImageJ Point and Measure tools and used to calculate the FRET/CFP ratio as a measure of ATP levels [39]. Imaging using Mito-roGFP2-Grx1 was performed as described previously [36]. For DHE staining, wing discs were incubated in 2μM DHE (Cambridge Bioscience) in Schneider's medium for 10 minutes, rinsed twice in Schneider's medium, then fixed for 5 minutes in 4% formaldehyde/PBS, washed briefly in PBS, dissected and imaged. 20 wing imaginal discs per genotype were dissected in PBS, homogenized in 100μl extraction buffer (6M guanidine chloride, 100mM TrisHCL, 4mM EDTA, pH 8.0) and incubated at 70˚C for 5 mins. ATP levels were measured using the ATP Determination kit (Molecular Probes) according to the manufacturer's instructions.

Western blot analysis
Western blot analysis was performed as previously described [36]. Primary antibodies were diluted in TBS/0.1% Tween 20 (TBS-T) and incubated overnight at 4˚C and were Drosophila anti-TFAM (Abcam, 1/500), mouse anti-ATP5A (Abcam, 1/5000), mouse anti-MTCO1 (Abcam, 1/1000) and rabbit anti-Actin (Cell Signaling, 1/4000). After three ten minute washes in TBS-T, the membranes were incubated for 90 minutes with fluorescently labelled secondary antibodies (anti-mouse IRdye 680 and anti-rabbit IRdye 800, LI-COR, both at 1/5000) diluted in TBS-T, then washed three times for ten minutes in TBS-T. The membranes were then scanned and analysed using an Odyssey infrared scanner (LI-COR). Odyssey infrared imaging systems application software version 3.0.25 was used to quantify the intensity of the bands on the blots. Normalised expression level was calculated by determining the band intensity relative to Actin.

Microarray experiments and analysis
CNS tissue from 20 late third instar larvae per genotype were dissected in cold PBS and transferred directly into 100μl lysis buffer containing β-mercaptoethanol (Absolutely RNA Microprep kit, Agilent Technologies). The lysis buffer was kept on ice while all the brains were dissected. RNA was prepared following manufacturer's instructions, including DNase treatment and stored at -80˚C. Samples were prepared in triplicate.
RNA was measured for quantity and integrity on an RNA Pico Chip (Agilent Technologies). 10ng of RNA per genotype was converted into labelled cDNA with the Nugen Ovation System V2 (NuGEN Technologies Inc.). 7mg of labelled cDNA was hybridised to Affymetrix Drosophila genome v2 GeneChips for 20 hours at 45˚C. They were then washed, stained (Gen-eChip Fluidics Station 450) and scanned (GeneChip Scanner 3000 7G) according to the manufacturer's instructions (Nugen Technologies Inc & Affymetrix).
Microarray data was processed using the MAS5.0 algorithm using the Transcriptome Analysis Console (ThermoFisher). Means were calculated using Tukey's Bi-weight average algorithm and differential expression between groups was calculated using un-paired one way analysis of variance (ANOVA). A statistical cutoff of p<0.05 and a fold change cutoff of ±1.5 fold were used. Correlations between datasets were analysed using GraphPad Prism (Graph-Pad Software Inc.). The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE114054. GO analysis was performed using DAVID (the database for annotation, visualization and integrated discovery) bioinformatics resources [65].

Statistical analyses
GraphPad Prism (GraphPad Software Inc.) was used to create graphs and for statistical analysis. Data with a p-value less than or equal to 0.05 was considered significant. Comparisons of two samples of continuous data were analysed with an unpaired, two-tailed student's t-test, where appropriate. Data were analysed for normality using the D'Agostino & Pearson omnibus normality test. Data that did not pass the normality test were analysed with the Mann Whitney test. Variance of the samples was assessed with an F test. If the variances of the two samples were significantly different then the Welch's correction was applied to the t-test. In order to compare more than two samples of continuous data, one-way ANOVA was used with Tukey's post hoc test. If data did not pass the D'Agostino & Pearson omnibus normality test, the Kruskal-Wallis, followed by Dunn's post hoc test were utilised. Categorical data were analysed using chi-squared.  Table. RNAi lines screened in the MitoMod modifier screen. The outcome of the screen is in the Result column. Any lines that had a phenotype with MS1096-Gal4 alone were excluded from the screen, so the result reads 'Excluded'. The confirmed column refers to confirmation of the result with an independent RNAi for the same gene, Y: confirmed, N: not confirmed, see S5 and S6 Tables for details. If the Confirmed column is blank, then an alternative RNAi was not available. (XLSX) S3 Table. Enhancers identified in the screen with GO molecular function. (XLSX) S4  Table. Genes whose expression is significantly altered in the transcriptomic analyses. (XLSX) S8 Table. Genes whose expression is significantly altered in control vs Pnt RNAi and control vs Aop RNAi conditions. (XLSX) S9 Table. Genes whose expression is significantly altered in control vs TFAM overexpression and control vs TFAM overexpression combined with Pnt RNAi. (XLSX) S10 Table. Genes whose expression is significantly altered in control vs TFAM overexpression and control vs TFAM overexpression combined with Aop RNAi. (XLSX) S11 Table. Genes whose expression is significantly altered in control vs TFAM overexpression and TFAM overexpression vs TFAM overexpression combined with Pnt RNAi. (XLSX) S12 Table. Genes whose expression is significantly altered in control vs TFAM overexpression and TFAM overexpression vs TFAM overexpression combined with Aop RNAi. (XLSX) S13 Table. Genes whose expression is significantly altered in control vs TFAM overexpression and TFAM overexpression vs TFAM overexpression combined with Pnt RNAi and TFAM overexpression vs TFAM overexpression combined with Aop RNAi. (XLSX) S14 Table. Functional annotation of GO classes of genes whose expression is significantly altered in TFAM overexpression versus TFAM overexpression combined with Pnt knockdown or TFAM overexpression combined with Aop knock-down conditions. (XLSX)