OMA1-mediated integrated stress response protects against ferroptosis in mitochondrial cardiomyopathy

SUMMARY Cardiomyopathy and heart failure are common manifestations in mitochondrial disease caused by deﬁciencies in the oxidative phosphorylation (OXPHOS) system of mitochondria. Here, we demonstrate that the cardiac-speciﬁc loss of the assembly factor Cox10 of the cytochrome c oxidase causes mitochondrial cardiomyopathy in mice, which is associated with OXPHOS deﬁciency, lysosomal defects, and an aberrant mitochondrial morphology. Activation of the mitochondrial peptidase Oma1 in Cox10 (cid:1) / (cid:1) mice results in mitochondrial fragmentation and induction of the integrated stress response (ISR) along the Oma1-Dele1-Atf4 signaling axis. Ablation of Oma1 or Dele1 in Cox10 (cid:1) / (cid:1) mice aggravates cardiomyopathy. ISR inhibition impairs the cardiac glutathione metabolism, limits the selenium-dependent accumulation of the glutathione peroxidase Gpx4, and increases lipid peroxidation in the heart, ultimately culminating in ferroptosis. Our results demonstrate a protective role of the Oma1-Dele1-mediated ISR in mitochondrial cardiomyopathy and link ferroptosis to OXPHOS deﬁciency and mitochondrial disease. For blue-native electrophoresis analysis we isolated mitochondria from heart tissue in homogenization buffer (220 mM mannitol, 20 mM sucrose, 2 mM EGTA, 0.1% BSA, 20 mM HEPES-KOH, pH 7.4, and protease inhibitor cocktail (Roche)) homogenized with ten strokes using a glass Teﬂon homogenizer at 1000 rpm on ice. The homogenates were centrifuged at 1,000 x g for 10 min at 4 º C and supernatant was collected. Homogenisation was repeated twice. Mitochondrial fraction was isolated with centrifugation at 8000 x g for 10 min at 4 º C. Protein concentration was determined using a Bradford assay. Mitochondria were solubilized with 6 g/g digitonin


In brief
Ahola et al. demonstrate that OXPHOS deficiency in the heart of Cox10 À/À mice causes cardiomyopathy and elicits an integrated stress response along the Oma1-Dele1-Atf4 axis. Oma1-dependent stress signaling preserves the glutathione metabolism and Gpx4 accumulation to limit lipid peroxidation, suppress ferroptosis, and delay cardiomyopathy.

INTRODUCTION
Mitochondrial diseases are rare disorders that together represent the most common group of inherited metabolic diseases with exceptional clinical variety (Fernandez-Vizarra and Zeviani, 2021). Typically, mitochondrial diseases arise from a primary defect in oxidative phosphorylation (OXPHOS), but a general energy deficit is not sufficient to explain the extraordinary cell and tissue specificity in mitochondrial disease. Differences in the various cellular responses to mitochondrial dysfunction appear to, at least partially, determine how cells and tissues cope with mitochondrial stress (Suomalainen and Battersby, 2018). Studies in patients, mouse models, and cultured cells have revealed a conserved transcriptional program that is induced upon mitochondrial dysfunction, termed the integrated stress response (ISR) (Bao et al., 2016;Mick et al., 2020;Quiró s et al., 2017;Tyynismaa et al., 2010). A comparative analysis of mouse models with cardiac-specific defects in mitochondrial DNA expression pointed to the activation of the ISR as a common signature in mitochondrial cardiomyopathies (K€ uhl et al., 2017). Besides metabolic adaptations, impaired OXPHOS activity results in the fragmentation of the mitochondrial network, which allows for the selective removal of damaged mitochondria by mitophagy and preserves cardiac homeostasis (Song et al., 2015). Despite compelling evidence for broad cellular responses to mitochondrial stress, it remains largely enigmatic to what extent these responses represent protective mechanisms or contribute to the progression of mitochondrial disorders.
Mitochondrial proteases govern mitochondrial fitness by degrading misfolded proteins and by balancing mitochondrial dynamics and mitophagy (Ahola et al., 2019;Deshwal et al., 2020). The inner membrane (IM) protease Oma1 is activated upon mitochondrial stress, such as mitochondrial depolarization, oxidative stress, or heat stress, or when misfolded IM proteins accumulate (Baker et al., 2014;Murata et al., 2020;Richter et al., 2015). Oma1 cleaves the dynamin-like GTPase Opa1, which mediates mitochondrial fusion, converting a long form of Opa1 into a short form (Gilkerson et al., 2021;MacVicar and Langer, 2016). Opa1 cleavage upon Oma1 activation limits mitochondrial fusion and causes fragmentation of the mitochondrial network. Genome-wide CRISPR screens identified Dele1 as another substrate that is cleaved by Oma1 upon mitochondrial dysfunction (Fessler et al., 2020;Guo et al., 2020). The stressinduced cleavage of Dele1 by Oma1 leads to the accumulation of Dele1 in the cytosol, where it activates the heme-regulated eIF2a kinase (HRI, EIF2AK1) to induce the ISR (Fessler et al., 2020;Guo et al., 2020). HRI is one of several kinases that phosphorylates eIF2a (peIF2a)   . p-eIF2a attenuates global translation and promotes the accumulation and nuclear translocation of the transcription factor Atf4, which is rapidly degraded in non-stressed conditions. Atf4 promotes amino acid and one-carbon metabolism, protects against oxidative stress, and induces autophagy (Bao et al., 2016;Pakos-Zebrucka et al., 2016). The ISR is therefore thought to support cellular recovery from stress, but prolonged ISR can activate apoptosis.
The concomitant regulation of Opa1-dependent mitochondrial fusion and Dele1-mediated ISR by Oma1 raises the question of the relative importance of mitochondrial fragmentation and stress signaling in disease. Ablation of Oma1 in mice does not cause gross phenotypes (Quiró s et al., 2012) but was found to protect against heart failure (Acin-Perez et al., 2018) and against dilated cardiomyopathy in mice, which lack the mitochondrial protease Yme1l (Wai et al., 2015). Similarly, the loss of Oma1 ameliorated ischemic kidney injury (Xiao et al., 2014) and delayed neurodegeneration in mice lacking prohibitin membrane scaffolds (Korwitz et al., 2016). By contrast, deletion of Oma1 aggravated neurodegeneration in neuron-specific Yme1l À/À mice despite suppressing mitochondrial fragmentation (Sprenger et al., 2019). These results point to cell-and context-specific roles of Oma1 in vivo.
Here, we used cardiac-specific Cox10 À/À mice to examine the role of Oma1 in mitochondrial disease caused by OXPHOS deficiency. Cox10 encodes a heme A-farnesyltransferase, which is required for the assembly of the cytochrome c oxidase (COX) complex. Mutations in COX10 are associated with Charcot-Marie-Tooth disease type 1A (Reiter et al., 1997). We demonstrate that Cox10 À/À mice develop early-onset lethal cardiomyopathy associated with Oma1-Dele1-Atf4-dependent ISR signaling in the heart, which suppresses ferroptosis and delays mitochondrial cardiomyopathy.

RESULTS
Loss of Oma1 aggravates dilated cardiomyopathy in Cox10 -/mice To examine the role of Oma1 in mitochondrial OXPHOS disease, we deleted Cox10 and Oma1 genes from skeletal and cardiac muscle by breeding the Cox10 fl/fl and Oma1 fl/fl animals with mice expressing Cre recombinase under the control of the creatine kinase promoter (Ckmm-Cre). Loss of Cox10 in heart and skeletal muscle led to early-onset dilated cardiomyopathy and death of the mice at a median age of 31 days ( Figure 1A). Cox10 À/À mice showed growth retardation, progressive loss of skeletal muscle mass, and enlarged hearts, which were accompanied by mild accumulation of connective tissue ( Figures 1B-1E). As expected, we observed drastically reduced levels of COX subunits, assembled COX complexes (Figures 1G, S1A, and S1B), and reduced COX activity in the heart lacking Cox10 ( Figure 1F). The analysis of cardiac tissue by transmission electron microscopy (TEM) revealed mitochondrial fragmentation and an altered mitochondrial ultrastructure with enlarged and swollen mitochondria and disrupted cristae (Figures 1H and S1C). The loss of Cox10 in the heart leads to the activation of Oma1, as indicated by the increased processing of Opa1 and the accumulation of short Opa1 forms c and e ( Figure 1I).
Mice lacking the COX assembly factor Cox15 in skeletal muscle accumulate aberrant lysosomes and benefit from an enhanced autophagic flux upon mTORC1 inhibition (Civiletto et al., 2018). Since we observed the accumulation of phosphorylated ribosomal protein S6 (pS6) and phosphorylated eukaryotic initiation factor 4E Figure 2. Loss of Oma1 in mitochondrial cardiomyopathy leads to lysosomal defects and accumulation of p62 (A and B) EM images of Cox10 À/À and Cox10 À/À Oma1 À/À hearts. (C) Quantification of the number of cells containing enlarged lysosomal structures in light microscope images for EM analysis (n = 3 mice).
We therefore generated Dele1 À/À mice by CRISPR-Cas9-mediated genome editing of embryonic stem cells and crossed them with heart and muscle-specific Cox10 À/À mice, examining a role of Dele1 for ISR signaling in vivo. Deletion of Dele1 did not broadly affect the survival of mice, which did not show any gross phenotype in the absence of Dele1 up to an age of 1 year ( Figure 4A). However, when combined with a deletion of Cox10, Dele1 ablation aggravated the phenotype of Cox10 À/À mice, which died at an age of about 4 weeks ( Figure 4A). Cox10 À/À Dele1 À/À mice exhibited a further decreased body weight and increased heart/body weight ratio when compared with Cox10 À/À mice ( Figure 4B) and further enlarged hearts ( Figure 4D). Dele1 ablation did not affect COX activity ( Figure 4E). Deletion of Dele1 completely abrogated the accumulation of Atf4 and peIF2a in the absence of Cox10 ( Figures 4F and 4G) and impaired the expression of ISR target genes ( Figure 4H). These results demonstrate that the loss of Dele1 or Oma1 affected phenotypes of heart-specific Cox10 À/À mice similarly. The ISR is activated in the heart of Cox10 À/À mice along the Oma1-Dele1 axis ( Figure 4I), which is associated with a prolonged lifespan of the mice. OMA1-mediated ISR response supports the glutathione metabolism in the heart The ISR has been shown to inhibit cytosolic translation and promote the expression of genes involved in amino acid synthesis (D and E) (D) Immunohistochemical staining of heart tissue for p62 and DAPI and (E) quantification of total area of p62 staining in three different images in n = 3 animals (scale bars, 20 mm). (F and G) (F) Immunoblot analysis and (G) quantification from mouse heart showing increased levels of p62 and Lamp1 in Cox10 À/À and Cox10 À/À Oma1 À/À mice hearts relative to the loading (vinculin) and relative to WT. (H and I) (H) Representative immunoblot for p62 and pS6 and (I) quantification of heart lysates from rapamycin-treated and control animals. Quantification relative to the loading (vinculin) and to WT (n = 5). (J) Survival curve of rapamycin-treated and control animals (Cox10 À/À , n = 9 [6 f, 3 m]; Cox10 À/À Oma1 À/À , n = 7 [5 f, 2 m]). (K) Representative EM images of hearts from rapamycin-treated and control animals. EM images; abnormal mitochondria (asterisk), lipid swirls (arrow head), and enlarged lysosomal structures (hashtag) (scale bars, 2 mm). Data represented as mean ± SD. *p < 0.05, **p < 0.01, and ***p < 0.001, unpaired two-tailed Student's t test and log-rank (Mantel-Cox) test for survival between treatment and ctrl.

RNA Protein
n.s ** Figure 3. Cox10 -/mice show Nrf2 and p53 activation and Oma1-and Dele1-mediated Atf4 signaling in cardiac muscle (A and B) (A) Ingenuity pathway analysis of RNA sequencing results from Cox10 À/À and wild-type (WT) hearts, showing the 10 most significantly changed pathways and (B) the 5 most significantly changed upstream transcription factors when comparing Cox10 À/À and WT mice (n = 5).
(D) Heatmap of the ingenuity pathway analysis annotated Atf4 target genes in different mouse lines.
(E) Heatmap of the abundance of 40 proteins, whose steady-state levels were determined by mass spectroscopy and differed most significantly between Cox10 À/À and Cox10 À/À Oma1 À/À hearts. (F and G) (F) Immunoblot analysis of Atf4 and p-eIF2a in heart lysates from mouse hearts and (G) quantification relative to the loading control (vinculin) and WT. Heatmaps, Euclidian clustering for proteins and samples; Z score of the log 2 values; n.a., non-applicable; Ig kappa = chain V-II region 26-10; C1-C4, clusters 1-4. Data represented as mean ± SD. *p < 0.05, **p < 0.01, and ***p < 0.001, unpaired two-tailed Student's t test and two-sided t test followed by permutation-based FDR correction for multiple testing (q values) for proteome.  Nikkanen et al., 2016). The deletion of Oma1 did not affect the steady-state level of these metabolites, nor did it strongly impair their accumulation in Cox10 À/À hearts ( Figure 5A). Similarly, intermediates of the glycolytic and pentose phosphate pathways were present at reduced levels in Cox10 À/À hearts regardless of the presence of Oma1 ( Figure S3D), while nucleotide levels were not significantly altered in any of the mouse models ( Figure S3E). In striking contrast, the accumulation of the reduced form of glutathione (GSH) in Cox10 À/À hearts was significantly augmented in Cox10 À/À Oma1 À/À hearts ( Figure 5B), suggesting that Oma1 regulates GSH levels in OXPHOS deficiency.
The GSH metabolism is closely linked to the one-carbon metabolism and is regulated cooperatively by Atf4 and Nrf2 (Figure 5C) (Kasai et al., 2020). Consistent with an impaired ISR in the absence of Oma1, the expression of enzymes involved in GSH metabolism were increased in Cox10 À/À hearts, while the expression of several of them was significantly reduced in Cox10 À/À Oma1 À/À hearts compared with Cox10 À/À hearts (Figure 5D). Expression of enzymes that are controlled by Nrf2 and Atf4 together was increased in Cox10 À/À Oma1 À/À hearts. Moreover, our proteomic analysis revealed the accumulation of many of these enzymes in Cox10-deficient hearts and lower steadystate levels in the absence of Oma1 ( Figure 5E). These proteins include the GSH-dependent lipid peroxidase Gpx4, the GSH-degrading enzyme Chac1 and Chop1, whose accumulation was also monitored by immunoblot analysis of heart tissues ( Figures 5F and 5G). Chop1 modulates Atf4 activity (Kaspar et al., 2021) and, together with Atf4, has been shown to regulate Chac1 expression to induce apoptosis in prolonged ISR (Mungrue et al., 2009). Together, we conclude from these experiments that the Oma1-mediated ISR supports the GSH metabolism in Cox10 À/À hearts. A ferroptotic signature of Cox10 -/-Oma1 -/hearts While acute ISR promotes metabolic reprogramming supporting cell survival, persistent ISR can trigger apoptotic cell death (Bar-Ziv et al., 2020;Galehdar et al., 2010;Lange et al., 2008). Unsupervised clustering of our heart RNA-seq data for genes associated with cell death revealed differential regulation of these genes in Cox10 À/À and Cox10 À/À Oma1 À/À hearts, including the pro-apoptotic proteins Chop and Chac1 ( Figures 6A and  S4A). TUNEL staining of heart sections, which detects DNA fragmentation in apoptotic cells, showed more TUNEL-positive cardiomyocytes in Cox10 À/À than in Cox10 À/À Oma1 À/À hearts ( Figures 6B and 6C). However, the low number of the detected TUNEL-positive cells unlikely explains the severity of cardiomyopathy in Cox10 À/À mice, nor the aggravated phenotype in Cox10 À/À Oma1 À/À mice. Moreover, our gene expression analysis pointed to an inflammatory response specifically in Cox10 À/À Oma1 À/À hearts ( Figures 6D and S4A), while apoptosis is generally considered as a non-inflammatory form of cell death. We therefore reasoned that a different form of cell death may explain the increased vulnerability of cardiomyocytes in Cox10 À/À Oma1 À/À mice.
The accumulation of GSH and the reduced steady-state level of Gpx4 point to an impaired redox homeostasis in Cox10 À/À Oma1 À/À mice. Gpx4 limits lipid peroxidation and protects cells against ferroptosis, a non-apoptotic form of cell death driven by iron-dependent phospholipid peroxidation (Dixon and Stockwell, 2019;Gan, 2021;Jiang et al., 2021). We therefore monitored the accumulation of malondialdehyde (MDA), the degradation product of lipid peroxidation, in Cox10 À/À and Cox10 À/À Oma1 À/À hearts ( Figures 6E and 6H). MDA-positive areas were detected in Cox10 À/À hearts but significantly accumulated in Cox10 À/À Oma1 À/À hearts ( Figures 6E and 6G). Oma1 À/À hearts showed wild-type levels of MDA ( Figure S4B). Immunostaining with antibodies directed against 4-hydroxy-2nonenal (4-HNE), a product of endogenous lipid peroxidation, confirmed these findings ( Figure S4C). Increased lipid peroxidation correlated with the appearance of lipid swirls detected in TEM images ( Figure 5F), raising the possibility that these structures represent oxidized mitochondrial membranes. We conclude from these experiments that the loss of Oma1 is associated with an inflammatory response, the accumulation of GSH, reduced Gpx4 levels, and increased lipid peroxidation in the hearts of Cox10 À/À mice, which thus show hallmarks of ferroptosis.
To demonstrate the protective role of ISR against ferroptosis, we treated Cox10 À/À cells with ISRIB, an inhibitor of eIF2a phosphorylation and ISR signaling (Sidrauski et al., 2013) before we examined the vulnerability of the cells for erastininduced ferroptosis. Short treatment with ISRIB increased the susceptibility of wild-type cells to erastin, whereas the survival of Cox10 À/À was not affected under these conditions (Figure 7J), likely due to the increased levels of cysteine and GSH in these cells. However, we observed increased erastininduced ferroptosis in Cox10 À/À cells upon prolonged ISR inhibition with ISRIB ( Figure 7J). Similarly, depletion of Atf4 or Dele1 increased the vulnerability of Cox10 À/À cells for erastininduced ferroptosis (Figures 7K and S4K). These experiments demonstrate a protective effect of the Oma1-Dele1-mediated ISR against ferroptosis. Oma1 supports selenium utilization to promote Gpx4 accumulation ISR signaling regulates the one-carbon metabolism as well as the metabolism of amino acids and of GSH, broadly promoting the oxidative defense. Gpx4 has been identified as central lipid peroxidase protecting against ferroptosis. However, although impaired ISR signaling caused decreased Gpx4 protein levels in Cox10 À/À hearts ( Figures 5F and 5G), transcription of Gpx4 was not altered in Cox10 À/À Oma1 À/À hearts when compared with Cox10 À/À hearts ( Figure 5D), indicating that Gpx4 is not a transcriptional target of Atf4 in cardiomyocytes. Gpx4 is a selenoprotein (Ingold et al., 2018), whose translation depends on the availability of selenium (Li et al., 2022). The ISR is known to affect the trans-sulfuration pathway ( Figure 5C) (Suomalainen and Battersby, 2018), and the involved enzymes have been demonstrated to allow trans-senylation (Lazard et al., 2015). These enzymes include the cystathione g lyase Cth, which provides selenophosphate for the synthesis of selenocysteine ( Figure 5C) and whose increased expression in Cox10 À/À hearts depended on Oma1 ( Figure 5D). To examine whether the selenium availability limits the accumulation of Gpx4, we treated wild-type, Cox10 À/À , Cox10 À/À Oma1 À/À , and Oma1 À/À cells with selenium and monitored Gpx4 protein levels ( Figure 7L). Strikingly, we observed significantly increased Gpx4 levels in wild-type cells supplemented with sodium selenite, whereas Gpx4 was only moderately increased in Oma1 À/À cells (Figure 7L). Loss of Oma1 in Cox10 À/À cells also reduced the selenium-dependent accumulation of Gpx4, although to a lower extent ( Figure 7L). Depletion of the Atf4 target Cth prevented Gpx4 accumulation in Cox10 À/À cells ( Figure 7M). In agreement with recent findings in breast cancer cells (Li et al., 2022) and the observed Gpx4 protein levels, selenium supplementation reduced ferroptosis of wild-type MEFs, while the protective effect was ameliorated in Oma1 À/À cells ( Figure S4L).
Thus, the Oma1-mediated ISR increases the transcription of enzymes of the trans-sulfuration pathway, which supports the utilization of selenium and thereby promotes the accumulation of Gpx4 and resistance against lipid peroxidation and ferroptosis ( Figure 7N).

DISCUSSION
Our results demonstrate that Oma1-Dele1-dependent ISR signaling serves a protective function against ferroptosis and (D) Accumulation of Atf4 or Atf4 and Nrf2 targeted mRNAs of selected GSH metabolism enzymes in RNA-seq data shown as log 2 FC in comparison to WT. Data are marked with a hashtag when differences were too small to be visualized. (E) Heatmap of the abundance of GSH-related enzymes that were detected by mass spectrometry in the cardiac proteome (Euclidian clustering for proteins, data shown as Z score of the log 2 values). Atf4 target proteins marked with blue and ANOVA significant with gray. (F and G) (F) Immunoblot analysis of selected enzymes of the GSH pathway and (G) quantification relative to vinculin. GSH, reduced glutathione; GSSG, oxidized glutathione; C1-C3, clusters 1-3. (A and B) Data represented as box and whiskers where whiskers are min and max. (G) Data represented as mean ± SD. *p < 0.05, **p < 0.01, and ***p < 0.001, unpaired two-tailed Student's t test or ANOVA, permutation-based FDR < 0.05 (E). delays cardiomyopathy. The ISR promotes the one-carbon metabolism, which supports GSH synthesis via the trans-sulfuration pathway and increases the cellular resistance against oxidative stress (Harding et al., 2003;Torrence et al., 2021). We demonstrate that trans-sulfuration and the GSH metabolism are important metabolic targets of the ISR protecting OXPHOSdeficient cardiomyocytes against ferroptosis. Impaired ISR signaling in Cox10 À/À Oma1 À/À hearts disturbs the GSH metabolism and results in decreased Gpx4 levels, which together explain the increased lipid peroxidation and ferroptosis in these mice. Our findings therefore reveal the physiological relevance of the observed strong synthetic interaction of Gpx4 with mitochondrial dysfunction in genome-wide CRISPR screens . The Oma1-dependent increased transcription of enzymes of the trans-sulfuration pathway, such as Cth, in the heart supports the utilization of selenium and the synthesis of the selenoprotein Gpx4. Our experiments in cultured cells indicate that an impaired ISR limits available selenium pools and Gpx4 translation in cardiomyocytes, increasing their ferroptotic vulnerability.
Oxidative stress in mitochondria activates the Nrf2-mediated stress response that together with Atf4 induces GSH synthesis and antioxidant defense (Kasai et al., 2020). We observed increased expression of Nrf2 target genes in both Cox10 À/À and Cox10 À/À Oma1 À/À mice but only partial induction of the GSH defense system without Oma1-mediated Atf4 activation in Cox10 À/À Oma1 À/À hearts. While we detected Atf4 and Nrf2 activation in OXPHOS-deficient heart, mild OXPHOS deficiencies in mtDNA replication-deficient skeletal muscle signal via Atf5 and manifest with a disbalance of one-carbon metabolites and nucleotides (Forsströ m et al., 2019;Nikkanen et al., 2016). Mitochondrial translation defects in heart and skeletal muscle lead to Atf4-and Atf5-mediated stress responses (Dogan et al., 2014). It thus appears that the combined and timely controlled expression of different transcription factors in defined metabolic settings determines the stress response in a tissuespecific manner.
Although persistent ISR signaling can induce apoptosis and although Oma1 protects against apoptosis under defined conditions in vitro (Anand et al., 2014;Jiang et al., 2014), we only observed apoptotic cell death of a low number of cardiomyocytes in vivo, unlikely driving cardiomyopathy. Indeed, the mitochondrial apoptotic machinery is age regulated both in mice and human and downregulated in the first weeks of life in mice (Sarosiek et al., 2017). OXPHOS deficiency can increase ROS production and decrease the levels of coenzyme Q (K€ uhl et al., 2017), which serves protective functions against lipid peroxidation and ferroptosis (Bersuker et al., 2019;Doll et al., 2019). It is conceivable that this increases the dependency of OXPHOS-deficient cardiomyocytes on ISR signaling and renders them susceptible for ferroptosis. Our results provide rich in vivo support for ferroptosis in mitochondrial cardiomyopathies, adding to the emerging evidence for important roles of ferroptosis in cardiovascular diseases in general (Li et al., 2021;Tadokoro et al., 2020).
We observed increased Opa1 processing by Oma1 and mitochondrial fragmentation in Cox10-deficient hearts. However, although ablation of Oma1 stabilizes long Opa1 forms, it did not ameliorate mitochondrial defects or cardiomyopathy in these animals. However, our results do not exclude that the fragmentation of the mitochondrial network contribute to cardiomyopathy. Indeed, disturbances in mitochondrial morphology have been linked to heart disease (Dorn, 2015). Thus, the role of both Oma1-dependent pathways-ISR signaling and the regulation of mitochondrial dynamics-for cell survival and their relative pathophysiological importance appear to vary between different tissues and to depend on the metabolic state of the cell. This variability may contribute to the cell-and tissue-specificity in mitochondrial disease.

Limitations of study
Deletions of either of the two key components of mitochondrial ISR signaling phenocopy each other and aggravate cardiomyopathy in Cox10 À/À mice with hallmarks of ferroptosis. These findings correlate with our experiments in cultured cells, which reveal the protective function of the ISR against ferroptosis. To unambiguously demonstrate that the activation of ferroptosis aggravates cardiomyopathy in vivo requires us to pharmacologically suppress ferroptosis using peritoneal injections. However, these experiments are hampered by the early-onset phenotype of cardiac-specific Cox10 À/À mice, offering only a short therapeutic window for treatments of newborn mice. Another limitation of our study concerns other possible substrates of Oma1, whose impaired proteolysis may contribute to mitochondrial cardiomyopathy. Substrate-specific mouse models will be required to delineate the role of individual substrate proteins for cardiac health.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

ACKNOWLEDGMENTS
We wish to thank Dominique Diehl for her excellent technical help in proteomic analysis; Katharina Ried for help in optimizing the ferroptosis assays; Fiona Mayer for help in designing the Dele1 knockout animals; and Jun Kim, Kat Folz-Donahue, and Lena Schumacher for their technical help in FACS experiments. Microscopy and FACS analyses were performed in the FACS & Imaging Core Facility at the Max Planck Institute for Biology of Ageing (MPI Ageing). RNA sequencing was performed in Cologne Centre for Genomics. Bioinformatic analysis for RNA-seq data were performed in the Bioinformatics Core Facility at the MPI Ageing. Metabolomics analyses were performed in the Metabolomics Core Facility at the MPI Ageing. CRISPR-Cas9 gene editing was performed in the Transgenesis Core Facility at the MPI Ageing. We thank Christina Lienkamp, Corinna Schwierzy-Kr€ amer, and Patrick Wollek for their valuable help with animal work and permits. Animal housing and treatment experiments were performed in the Comparative Biology Core Facility at the MPI Ageing and in CECAD's in vivo Research Facility. We wish to specially thank Tobias Nespital for his help with the rapamycin experiment. EM was performed in CECAD Imaging Facility, and we specially want to thank Janine Heise for her excellent technical help. The work was supported by postdoctoral grants from the Alexander Von Humboldt Foundation (1163782-HFST-P) to S.A. and grants from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) -SFB 1218 -Projektnummer 269925409 and SFB1403 -Projektnummer 414786233) and the German-Israel-Project (DIP, RA1028/10-2) to T.L.

DECLARATION OF INTERESTS
The authors declare no competing interests.  Institute (SwRI). The floxed Cox10 fl/fl mice originated from previously published line (Diaz et al., 2005) and floxed Oma1 fl/fl mice from (Wai et al., 2015) crossed with mice expressing Ckmm-Cre-recombinase (Wang et al., 1999) to create cardiac and skeletal muscle specific knock-out of Cox10 and Oma1 in the C57BL/6N background. Dele1 -/mice were created with the CRISPR-Cas9 system by electroporation of mouse zygotes with a NEPA21 Electroporator (CUY501P1-1.5 electrode) and the IDT Alt-R CRISPR-Cas9 System (Integrated DNA Technologies) with guide RNAs listed in the key resources table. Groups included male and female animals. Samples for protein and RNA extraction were taken after cervical dislocation and snap frozen in liquid nitrogen. For protein, metabolome and RNA analysis as well as histological analysis we used mice of similar age (approximately 4 weeks).
Electron microscopy 1-2 mm piece from heart tissue was fixed in 2% formaldehyde/ 2% glutaraldehyde in 0,1 M cacodylic acid at least 48 h at 4 C. Samples were then washed four times 15 min in 0,1 M cacodylic acid and fixed with 2% osmiumtetroxid (Science Services) in 0,1 M cacodylic acid and washed again four times 15 min in 0,1 M cacodylic acid. After changes in ethanol 50%-100%, a mixture ethanol/propyleneoxid and 100% propyleneoxid the tissue was embedded in Epon fixative. Fixed tissue was cut in 70 nm sections on the ultramicrotome (UC6, Leica) on a grid and contrasted with 1,5% uranylacetate aqueous solution 15 min at 37 C. Cuts were washed five times in water, incubated 4 min in lead citrate and washed again five times in water and dried on a filter paper. Images were acquired with a transmission electron microscope (JEM 2100 Plus, JEOL), a OneView 4K camera (Gatan) with DigitalMicrograph software at 80 KV at room temperature.

Protein Digestion for proteomics
For lysis of heart samples, 4% SDS in 100 mM HEPES pH = 8.5 was used as a buffer system and the Precellys tissue homogenizer was utilized for mechanical disruption of the tissue following the manufacturer's instructions. The protein concentration was determined. 10 mg of protein was subjected for tryptic digestion. Proteins were reduced (10 mM TCEP) and alkylated (20 mM CAA) in the dark for 45 min at 45 C. Samples were subjected to SP3 based digestion including a pooled set consisting of all heart samples utilized which served as an internal standard for each TMT batch (n=2 using 126 Channel). Washed SP3 beads (SP3 beads (Sera-Mag (TM) Magnetic Carboxylate Modified Particles (Hydrophobic), Sera-Mag (TM) Magnetic Carboxylate Modified Particles (Hydrophylic) from Thermo Fisher Scientific) were mixed equally, and 3 mL of bead slurry were added to each sample. Acetonitrile was added to a final concentration of 50% and washed twice using 70 % ethanol (V=200 mL) on an in-house made magnet. After an additional acetonitrile wash (V=200mL), 5 mL digestion solution (10 mM HEPES pH = 8.5 containing 0.5mg Trypsin (Sigma) and 0.5mg LysC (Wako)) was added to each sample and incubated overnight at 37 C. Peptides were desalted on a magnet using 2 x 200 mL acetonitrile and labelled with TMT (11 plex) on SP3 beads in 10 mL 100 mM HEPES pH=8.5 for 1h at 37 C. 0.4 mg TMT reaction reagent in acetonitrile was used per sample. The reaction was stopped using 0.5 % final concentration of hydroxylamine for 30 min at room temperature. Beads were pooled and peptides were eluted together using 5% DMSO. Next, we performed high pH offline peptide fractionation for mouse heart samples. TMT-labelled peptides were pooled to a total peptide amount of 50 mg. The sample was desalted using the StageTip technique, dried in a SpeedVac completely and resuspended in 10 mL of 10 mM ammonium hydroxide in 5% acetonitrile. The instrumentation consisted out of a ZirconiumTM Ultra HPLC and a PAL RTC autosampler system using the binary buffer system. AB) 10 mM ammonium hydroxide and ) 80% acetonitrile and 10 mM ammonium hydroxide. Peptides were separated according to their hydrophobicity using an in-house packed column (length = 40 cm, inner diameter = 175 mm, 2.7-mm beads, PoroShell, Agilent Technologies) column. The instrument was controlled using the software Chronos (Axel Semrau GmbH). The total gradient length was 40 min and in total 36 fractions were collected (1/30 s) and subsequently concentrated using a SpeedVac to complete dryness.
Liquid Chromatography and Mass Spectrometry for heart proteomics Eluted peptides in each fraction were dissolved in 10 mL of 2% formic acid and 2% acetonitrile in LC-MS H 2 O. 3 mL were injected for each fraction. The instrumentation consisted of a nanoLC 1200 (Thermo Fisher) coupled via a nano-electrospray ionization source to a Qexactive HF-x mass spectrometer. Peptides were separated on a 20 cm in-house packed column (75 mm inner diameter, PoroShell 2.7 mm beads) using a binary buffer system: AB) 0.1% formic acid and ) 0.1% formic acid in 80% acetonitrile. The gradient time was 25 min. MS1 spectra were acquired using a mass range from 250 to 1650 Th, a resolution (at 200 m/z) of 60,000 and an AGC target of 3e6 allowing a maximum injection time of 20 ms using profile mode. MS2 spectra were acquired at a resolution of 45,000 using a maximum injection time of 96 ms. The AGC target was to 1e5. The isolation window was set to 0.8 m/z. The fixed first mass was set to 110 m/z. The spectra were acquired in centroid mode.
Analysis of proteomic data Acquired mass spectra were subjected to MaxQuant (2.0.3.0) based analysis using the implemented Andromeda search engine. The input Fasta file contained reviewed Uniprot protein entries of the Mus musculus reference proteome (downloaded 12.2021, 17.029 protein entries). TMT MS2 Quantification was selected using the first channel (126) as a reference (all intensities values are normalized to this value by division) since it contained a pool protein digest from all samples. Oxidation at methionine residues and protein N-term acetylation were defined as variable modifications. The false discovery rate (FDR) was controlled to 1% at the protein and peptide-spectrum-match level using the implemented 'revert' algorithm. The mass tolerances were used as defined by default for FTMS instruments: 20 ppm at MS/MS level. The match between runs algorithm was enabled using default settings. A set of common contaminants were included and filtered out from the proteinGroup.txt file output. Gene ontology annotations were downloaded from Uniport website. Two-sided t test as well as one-way analysis of variance (ANOVA) were performed followed by permutation-based FDR correction (s0 = 0.1, number of permutations=500) were performed to identify significant differently expressed proteins in the Perseus software. Visualization was performed in the Instant Clue software suite (v0.11.0).
Metabolite extraction of polar and lipophilic metabolites from mouse heart tissues For the extraction of total lipids, 15-20 mg of snap-frozen murine heart tissue was homogenized to a fine powder using a ball mill-type grinder (Tissue Lyser2: http://www.quiagen.com/) equipped with a 10-sample holder (Retsch). For the homogenization of the cell pellets one liquid nitrogen cooled 5 mm stainless steel metal balls was added to each tube and the frozen material was disintegrated for 1 min at 25 Hz. Polar metabolites and lipids were extracted by adding 1 mL of pre-cooled (-20 C) extraction buffer (methyl tert-butyl ether (MTBE): methanol: UPLC-grade water 5:3:2 [v:v:v]), containing 20 mL of EquiSplash Lipidomix (https://avantilipids.com/) as internal standard. The tubes were immediately vortexed until the sample was well re-suspended in the extraction buffer. The homogenized samples were then incubated on a cooled (4 C) orbital mixer at 1500 rpm for 30 min. After this step, the metal ball was removed and the samples were centrifuged for 10 min at 21.100 x g in 4 C. The supernatant was transferred to a fresh tube and 150 mL of MTBE and 100 mL of UPLC-grade water were added to each sample. The tubes were immediately vortexed before incubating them for an additional 10 min on a cooled (15 C) orbital mixer at 1500 rpm. After this step the samples were centrifuged for 10 min at 15 C and 16.000 x g, which provided a 2-phase separation of lipid and polar metabolites. The upper, MTBE phase, contains the lipids, while the lower, methanol-water, phase contains the polar and semi-polar metabolites. For the analysis of rapamycin, 600 mL of the upper lipid phase were collected into a fresh tube, which was stored at -80 C for the mass spectrometric analysis. The remaining lower (polar) phase ($800 mL) was immediately dried in a SpeedVac concentrator and stored dry at -80 C until mass spectrometric analysis.

Metabolite extraction of polar and lipophilic metabolites from MEFs
MEFs were plated 24 h before collecting. Growth media was removed from each well on the day of sample collection and the cells were washed twice, using 1 mL of 75 mM ammonium carbonate pH 7.4 (Sigma) wash buffer warmed to 37 C. MTP plates were then placed on ice and metabolites were extracted by adding 400 mL of pre-cooled (-20 C) metabolite extraction buffer 40:40:20 [v:v:v] acetonitrile:methanol:water (Optima LC/MS grade, Thermo Fisher Scientific) to each well. Cells were immediately scraped and the whole volume, including the precipitated cellular material, was transferred to a labelled 1.5 ml Eppendorf tube stored on ice. This extraction procedure from the plates was repeated another two times, leading to the collection of a total of 1.2 mL of metabolite extract from each sample.
Once the extracts were collected, they were centrifuged for 10 min at 4 C and 21.000 x g. The cleared supernatant was transferred to a fresh 1.5 mL Eppendorf tube and the polar metabolite extract was dried down immediately in a SpeedVac concentrator (ScanVac) set to 20 C and 1000 rpm until tubes were completely dried. These samples can then be used directly for analysis of they can be stored at -80 C until measured on the diverse LC-MS systems. The cellular pellet, which contains the precipitated protein, was used for protein quantification (BCA Protein Assay Kit) and later on these values were utilized for metabolite normalization.
Targeted liquid chromatography-high-resolution mass spectrometry-based (LC-HRS-MS) analysis of aminecontaining metabolites from mouse heart tissue and MEF cell cultures The LC-HRMS analysis of amine-containing compounds was performed using an adapted benzoylchlorid-based derivatization (Wong et al., 2016). In brief: The polar fraction of the metabolite extract was re-suspended in 300 mL of LC-MS-grade water (Optima-Grade, Thermo Fisher Scientific) and incubated at 4 C for 15 min on a thermomixer. The re-suspended extract was centrifuged for 5 min at 16.000 x g at 4 C and 50 mL of the cleared supernatant were mixed with 25 ml of 100 mM sodium carbonate (Sigma), followed by the addition of 25 ml 2% [v/v] benzoylchloride (Sigma) in acetonitrile (Optima-Grade, Thermo Fisher Scientific). Samples were vortexed and kept at 20 C until analysis. For the LC-HRMS analysis, 1 ml of the derivatized sample was injected onto a 100 x 2.1 mm HSS T3 UPLC column (Waters). The flow rate was set to 400 ml/min using a binary buffer system consisting of buffer A (10 mM ammonium formate (Sigma), 0.15% [v/v] formic acid (Sigma) in LC-MS-grade water (Optima-Grade, Thermo Fisher Scientific). Buffer B consisted solely of acetonitrile (Optima-grade, Thermo Fisher-Scientific). The column temperature was set to 40 C, while the LC gradient was: 0% B at 0 min, 0-15% B 0-4.1min; 15-17% B 4.1 -4.5 min; 17-55% B 4.5-11 min; 55-70% B 11 -11.5 min, 70-100% B 11.5 -13 min; B 100% 13 -14 min; 100-0% B 14 -14.1 min; 0% B 14.1-19 min; 0% B. The mass spectrometer (Q-Exactive Plus, Thermo Fisher Scientific) was operating in positive ionization mode recording the mass range m/z 100-1000. The heated ESI source settings of the mass spectrometer were: Spray voltage 3.5 kV, capillary temperature 300 C, sheath gas flow 60 AU, aux gas flow 20 AU at a temperature of 330 C and the sweep gas to 2 AU. The RF-lens was set to a value of 60.
Liquid Chromatography-High Resolution Mass Spectrometry-based (LC-HRMS) analysis of rapamycin The stored (-80 C) lipid extracts were dried in a SpeedVac concentrator before analysis and lipid pellets were re-suspended in 200 mL of UPLC-grade acetonitrile: isopropanol (70:30 [v:v]). Samples were vortexed for 10 seconds and incubated for 10 min on a thermomixer at 4 C. Re-suspended samples were centrifuged for 5 min at 16.000 x g and 4 C, before transferring the cleared supernatant to 2 ml glass vials with 200 ml glass inserts (Chromatography Zubehö r Trott, Germany). All samples were placed in an Acquity iClass UPLC (Waters) sample manager at 6 C. The UPLC was connected to a Tribrid Orbitrap HRMS, equipped with a heated ESI (HESI) source (ID-X, Thermo Fischer Scientific). and proteins were separated in 3-9% native gradient gels, transferred to PVDF membrane and immunoblotted with the antibodies listed in key resources table.
Immunofluorescence and histological staining Heart samples were freshly frozen in liquid nitrogen cooled isopentane in O.C.T Compound Embedding Medium (TissueTek). Frozen sections (8 mm) were blocked with 5% horse serum for one prior to an incubation overnight at 4ºC in 2% BSA with selected antibodies that are listed in the key resources table. Sections were washed three times with PBS before incubation with secondary antibodies (AlexaFluor (Invitrogen)) for 1 h in the dark, washed again three times with PBS, stained with 4,6-diamidino-2-phenylindole (DAPI) and mounted with ProlongH Gold. For TUNEL staining, frozen sections (8 mm) were stained according to the guidelines of the manufacturer (BrdU, abcam) with counterstaining for DAPI. Images were acquired using a Leica SP8-X laser-scanning confocal microscope equipped with an 63x oil HC PL APO CS2 objective (NA 1.4) Four images/animal were analyzed by MDA staining and the total area of MDA positive stain was determined using the FiJi software. Cells from four images/animal were calculated for TUNEL analysis (200-300 cells in total).
For COX staining, frozen sections (8 mm) were incubated in 0,05 M phosphate buffer containing 220 mM sucrose, 2,5 mM 3'3 diaminobenzidine, 1 mg/ml catalase, 0,5mg/ml cytochrome c for 12 min. Masson's Trichrome (Abcam) was performed to frozen sections (8 mm) according to the guideline of the manufacturer. Sections for COX and Masson's were dehydrated in ascending alcohols and in xylene and mounted with Cytoseal Xyl (Thermo Scientific). Images were acquired using Nicon Eclipse Ci with a CFI P-Achromat 4x/0.10 and CFI P-Fluor 20x/0.50 objectives.

QUANTIFICATION AND STATISTICAL ANALYSIS
Sample size was chosen according to our previous experience and common standards. Power analysis was used to predetermine sample size for mouse experiments. The sample size included at least three independent cell culture wells or mice where statistical evaluation was performed. For metabolomic analysis of anionic metabolites, we excluded one WT mouse sample and two Cox10 samples as clear outliers based on PC clustering blots. Experiments were repeated as detailed in the figure legends. Mice were assigned to experimental groups based on genotypes available. Analyses were not blinded because experiments were performed and analysed by the same researcher except for microscopy imaging where samples were numbered. The N number for all MEF cell experiments represent independent experimental cell cultures. Data analysis was performed with Prism GraphPad 9 and Instant Clue. Images were processed with ImageJ and schematics were created with Adobe Illustrator 26.0.2 and BioRender.com.