Regulation of OPA1 processing and mitochondrial fusion by m-AAA protease isoenzymes and OMA1

Mitochondria are dynamic organelles that undergo continuous remodeling through fusion and fission events. Conserved protein machineries mediate fusion and fission of mitochondrial membranes (Cerveny et al., 2007; Hoppins et al., 2007; Westermann, 2008). The dynamin-related GTPase OPA1, which resides in the mitochondrial intermembrane space, is involved in inner membrane fusion, in the regulation of mitochondrial cristae morphology, and in protecting cells from apoptosis (Olichon et al., 2003; Griparic et al., 2004; Frezza et al., 2006; Meeusen et al., 2006; Lenaers et al., 2009). Mutations in OPA1 cause dominant optic atrophy, a progressive neurological disease characterized by degeneration of the retinal ganglion cells and atrophy of the optic nerve (Amati-Bonneau et al., 2009). Mitochondria contain various OPA1 isoforms that arise from alternative splicing and proteolytic processing events at two sites, S1 and S2, generating shorter isoforms (Delettre et al., 2001; Ishihara et al., 2006).

A combination of long and short OPA1 isoforms is required for mitochondrial fusion (Song et al., 2007), pointing to a crucial regulatory role of OPA1 processing. Roughly equimolar concentrations of long and short isoforms of OPA1 are formed under normal conditions. However, low mitochondrial ATP levels, the dissipation of the membrane potential across the inner membrane (Δψ_m), or apoptotic stimuli induce OPA1 cleavage, resulting in the loss of long isoforms (Duvezin-Caubet et al., 2006; Ishihara et al., 2006; Baricault et al., 2007). Various proteases in the inner membrane have been linked to the processing of OPA1. The intermembrane space AAA (ATPase associated with diverse cellular activities) protease YME1L1 regulates OPA1 cleavage at S2, which is only present in a subset of OPA1 variants (Griparic et al., 2007; Song et al., 2007). Debate exists about the protease cleaving OPA1 at S1, as both the rhomboid protease PARL (presenilin-associated rhomboid-like protein) and the matrix AAA (m-AAA) protease, an oligomeric ATP-dependent metallopeptidase in the inner membrane, have been proposed to be involved in OPA1 processing (Cipolat et al., 2006; Ishihara et al., 2006). However, normal OPA1 processing in mouse embryonic fibroblasts (MEFs) lacking PARL or the m-AAA protease subunit paraplegin raised doubts about the role of either protease for OPA1 processing in vivo (Duvezin-Caubet et al., 2007).

Notably, human and murine mitochondria possess different m-AAA protease isoenzymes, which differ in their subunit composition but exert at least partially redundant activities (Koppen et al., 2007). Three homologous subunits are expressed in the mouse, AFG3L1 and -2 and paraplegin. Whereas paraplegin was exclusively detected in heterooligomeric complexes with AFG3L1 and -2, homooligomeric complexes composed of AFG3L1 or -2 do exist. Therefore, functionally redundant isoenzymes might substitute for the loss of paraplegin and maintain OPA1 processing in the absence of paraplegin. A complementation study in yeast indeed demonstrated functional conservation of different mammalian m-AAA protease isoenzymes and revealed their ability to cleave OPA1 expressed heterologously in yeast (Duvezin-Caubet et al., 2007). This notwithstanding, they have been linked to different neurodegenerative diseases in human: mutations in Spg7 encoding paraplegin cause a recessive form of hereditary spastic paraplegia, whereas heterozygous mutations in Afg3l2 are responsible for a dominant form of spinocerebellar ataxia, SCA28 (Cagnoli et al. 2008. Annual meeting of the American Society of Human Genetics. Abstr. 1501; DiBella et al. 2008. Annual meeting of the American Society of Human Genetics. Abstr. 216). It is presently unclear whether this reflects functional differences of m-AAA isoenzymes or results from a tissue-specific expression of m-AAA protease subunits (Koppen et al., 2007).

In this study, we have assessed the function of different m-AAA isoenzymes in vivo using RNAi and demonstrate that OPA1 processing and mitochondrial fusion depend on AFG3L1 and -2. Moreover, we identify a novel peptidase in the inner membrane, OMA1, which mediates OPA1 processing if m-AAA proteases are absent or mitochondrial activities are impaired.

We used RNAi in MEFs to down-regulate m-AAA protease subunits individually or in combination. Immunoblotting revealed effective depletion of AFG3L1 and -2 and paraplegin (Fig. 1 A). Mitochondrial morphology was monitored by fluorescence microscopy in cells cotransfected with a mitochondrially targeted DsRed variant. Neither paraplegin-depleted mitochondria nor mitochondria from Spg7^−/− MEFs showed morphological abnormalities (Fig. S1 and not depicted). Similarly, mitochondria of cells transfected with either Afg3l1- or Afg3l2-specific siRNA appeared largely normal in morphology (Fig. 1, B and C). However, concomitant down-regulation of both subunits caused fragmentation of the mitochondrial network in ∼80% of the cells (Fig. 1, B and C). Examination of these cells by electron microscopy revealed a severe disturbance of the mitochondrial ultrastructure (Fig. 1 D). Lamellar cristae were either completely absent or severely reduced in number in cells lacking AFG3L1 and -2, whereas the structure of cristae was not significantly affected upon down-regulation of AFG3L1 or -2 (Fig. 1 D). These findings reveal overlapping activities of AFG3L1 and -2 and demonstrate an essential role of the m-AAA protease for the ultrastructure of mitochondria.

An impaired fusion or increased fission may cause fragmentation of the mitochondrial network in the absence of AFG3L1 and -2. To distinguish between these possibilities, we transfected interfered cells with a dominant-negative variant of the fission component DRP1, DRP1^K38A. In agreement with previous observations (Smirnova et al., 1998; James et al., 2003), expression of DRP1^K38A impaired fission and resulted in the elongation of mitochondrial tubules as the result of ongoing fusion (Fig. 2, A and B). In contrast, cells lacking AFG3L1 and -2 maintained punctiform mitochondria upon expression of DRP1^K38A, strongly suggesting that the organelles are unable to fuse. As independent evidence, we found that cells interfered for Afg3l2 and -1 did not form a highly interconnected network when exposed to low concentrations of cycloheximide (CHX; Fig. 2, C and D), indicating that they have lost the ability to undergo stress-induced mitochondrial hyperfusion (SIMH), which is an adaptive response against stress (Tondera et al., 2009). Therefore, we conclude that the fusion of mitochondrial membranes depends on AFG3L1 and -2.

OPA1 regulates both mitochondrial fusion and cristae morphogenesis (Lenaers et al., 2009). MEFs express mainly two alternative transcripts of OPA1, variants 1 and 7, which are cleaved, resulting in the accumulation of five OPA1 forms: two long (a and b) and three short isoforms (c–e; Ishihara et al., 2006; Duvezin-Caubet et al., 2007). Down-regulation of individual subunits of the m-AAA protease isoenzymes did not reproducibly affect the processing of OPA1 (Fig. 3 A). However, the concomitant loss of AFG3L1 and -2 significantly affected the accumulation of OPA1 variants (Fig. 3 A): the long OPA1 isoforms were no longer detected, whereas short forms, mainly e, accumulated (Fig. 3 A). This shift was not significantly enhanced in Spg7^−/− cells (Fig. 3 A). These results substantiate the overlapping activities of AFG3L1 and -2 and point to an impaired OPA1 processing in m-AAA protease–depleted MEFs.

Both long and short OPA1 isoforms are necessary to support mitochondrial fusion (Song et al., 2007). Therefore, the loss of long OPA1 isoforms can account for the fragmentation of the mitochondrial network upon depletion of AFG3L1 and -2. To test this hypothesis, we assessed mitochondrial morphology after coexpression of Afg3l1 and -2 siRNAs with a Flag-tagged version of OPA1 splicing variant 1 (v1), in which the amino acid residues flanking the processing site had been deleted (v1ΔS1; Ishihara et al., 2006). Mitochondrial morphology was also analyzed in AFG3L1- and AFG3L2-depleted MEFs expressing S-OPA1, a hybrid protein composed of a short form of OPA1 fused to the mitochondrial target sequence of apoptosis-inducing factor (Ishihara et al., 2006). We observed a partial but statistically significant restoration of a tubular network in m-AAA protease–deficient cells upon expression of v1ΔS1 but not of S-OPA1 (Fig. 3 B), suggesting that the m-AAA protease regulates mitochondrial fusion via OPA1.

These findings link, for the first time, the function of AFG3L1 and -2 to the biogenesis of OPA1. Surprisingly, depletion of AFG3L1 and -2 caused a loss rather than a stabilization of long OPA1 isoforms. Therefore, we directly monitored the efficiency of OPA1 processing in the absence of m-AAA protease subunits by expressing Flag-tagged v1 or v1ΔS1 in MEFs (Ishihara et al., 2006). Down-regulation of AFG3L1 and -2 promoted processing of v1 and led to the loss of long OPA1 forms (Fig. 3 C). v1ΔS1 was present at severely decreased levels in cells lacking both m-AAA protease subunits, suggesting degradation of the OPA1 variant (Fig. 3 C). Indeed, if protein synthesis was inhibited by CHX, v1ΔS1 was stable upon further incubation of control cells (Fig. 3 D). In contrast, v1ΔS1 level decreased in the absence of AFG3L1 and -2 (Fig. 3 D), explaining the only partial restoration of the tubular mitochondrial network upon expression of this variant in AFG3L1/AFG3L2-depleted cells (Fig. 3 B). We conclude from these experiments that turnover, as well as cleavage, of OPA1 at S1 is enhanced upon down-regulation of m-AAA protease subunits.

Disruption of the mitochondrial membrane potential and decreased ATP levels within mitochondria have been shown to induce processing of OPA1 (Duvezin-Caubet et al., 2006; Baricault et al., 2007). Respiratory chain deficiencies have recently been reported in brain mitochondria of Afg3l2^−/− mice (Maltecca et al., 2008). Therefore, we assessed the formation of the mitochondrial membrane potential in cells depleted of m-AAA protease subunits using the membrane potential–dependent fluorescent dye JC-1 (Fig. 3 E). Whereas fluorescence dropped upon dissipation of the membrane potential by carbonyl cyanide m-chlorophenyl hydrazone (CCCP), the membrane potential was maintained upon depletion of m-AAA protease subunits (Fig. 3 E). Similarly, TMRE (tetramethylrhodamine, ethyl ester) measurements in Afg3l2^−/− MEFs grown on glucose or galactose did not reveal an impaired membrane potential (Fig. S2).

In subsequent experiments, we measured oxygen consumption in MEFs lacking m-AAA protease subunits to evaluate phosphorylating respiration with endogenous substrates. Down-regulation of AFG3L1 and -2 did not significantly affect oxygen consumption, irrespective of the presence of paraplegin (Fig. 3 F). However, respiration was significantly increased upon inhibition of the mitochondrial F₁F_O-ATP synthase with oligomycin, suggesting a higher proton leakage under conditions of membrane hyperpolarization (Fig. 3 F). In m-AAA protease–depleted cells, we also observed a small but significant decrease of the maximum respiratory capacity, which was measured under uncoupled conditions (Fig. 3 F). Accordingly, the ratio of uncoupled respiration and oligomycin-inhibited respiration, which is the respiratory control ratio (RCR), was reduced to ∼50% when AFG3L1 and -2 were down-regulated (Fig. 3 F). The loss of paraplegin did not further decrease the RCR (Fig. 3 F). These results indicate that the capacity of the oxidative phosphorylation system to supply ATP under conditions of high energy demand is limited in the absence of the m-AAA protease. However, under normal conditions, respiration was not impaired upon down-regulation of m-AAA protease subunits, suggesting that AFG3L1 and -2 directly affect mitochondrial fusion.

To substantiate the role of the m-AAA protease for OPA1 processing, we generated mutant variants of AFG3L2, in which glutamate 408 within the Walker B (WB) motif of the AAA domain or glutamate 575 within the proteolytic site (PS) were replaced by glutamine (WB, E408Q; PS, E575Q). Although glutamate 575 is essential for proteolytic activity, glutamate 408 is conserved within AAA+ ATPases and is required to activate a water molecule for nucleophilic attack on bound ATP. AAA+ ATPases harboring a corresponding mutation in the WB motif lack ATPase activity and trap ATP (Weibezahn et al., 2003; Dalal et al., 2004; Augustin et al., 2009). We established stable HEK293 cell lines allowing the tetracycline-inducible expression of C-terminally tagged AFG3L2 variants. In contrast to mouse fibroblasts, human cells express only two m-AAA protease subunits, AFG3L2 and paraplegin (Kremmidiotis et al., 2001). Induction of AFG3L2 expression in the presence of tetracycline resulted in an approximately twofold overexpression of AFG3L2 (Fig. 4 B). Although overexpression of AFG3L2 or AFG3L2^E575Q (PS) did not affect cell growth, cells expressing AFG3L2^E408Q (WB) showed a prominent reduction of their growth rate (Fig. 4 A), indicating that AFG3L2^E408Q (WB) but not AFG3L2^E575Q (PS) exerts a dominant-negative effect on other m-AAA protease subunits.

Long OPA1 isoforms were destabilized in the presence of AFG3L2^E408Q (WB), although they accumulated at normal levels upon expression of AFG3L2 or AFG3L2^E575Q (PS) (Fig. 4 B). Consistently, the analysis of mitochondrial morphology in these cell lines revealed fragmentation of the mitochondrial network in the majority of cells expressing AFG3L2^E408Q (WB) (Fig. 4, C and D). In contrast, expression of AFG3L2 of AFG3L2^E575Q (PS) did not affect the morphology of mitochondria (Fig. 4, C and D). Expression of the dominant-negative AFG3L2 variant did not dissipate the mitochondrial membrane potential in HEK293 cells (Fig. 4 E), nor was the oxygen consumption of HEK293 cells impaired (Fig. 4 F). However, similar to m-AAA protease–deficient MEFs, we observed a reduced RCR in cells expressing AFG3L2^E408Q (WB) (Fig. 4 F), suggesting an impaired energy metabolism under conditions of high energy demand. We conclude from these experiments that expression of a dominant-negative variant of AFG3L2 and depletion of m-AAA protease subunits by RNAi similarly affect mitochondrial morphology and OPA1 processing.

Our results suggest that an unidentified protease mediates processing and degradation of the long OPA1 isoforms in m-AAA protease–deficient cells. Notably, lack of paraplegin (Fig. 3 A), PARL, or HtrA2/Omi did not stabilize L-OPA1 in the absence of AFG3L1 and -2 (Fig. S3 A). Furthermore, concomitant down-regulation of AFG3L1 and -2 with YME1L1 or LON did not rescue long OPA1 isoforms (Fig. S3 B). To restrict the number of possible proteases involved in this process, we incubated AFG3L1/AFG3L2-deficient MEFs expressing v1 with different protease inhibitors and monitored the accumulation of Flag-tagged v1 and endogenous OPA1 by immunoblotting (Fig. 5 A). The stability of OPA1 was not affected upon incubating cells with pepstatin A, E-64d, or the broad-spectrum serine protease inhibitor Pefabloc (Fig. 5 A). In contrast, o-phenanthroline (o-phe), a membrane-permeable metalloprotease inhibitor, stabilized v1, resulting in the accumulation of the long as well as the precursor form of OPA1 (Fig. 5 A). This is consistent with previous experiments on CCCP-induced OPA1 processing (Ishihara et al., 2006). Similarly, the proteasome inhibitor MG132 stabilized v1 in AFG3L1/AFG3L2-depleted cells (Fig. 5 A). Thus, OPA1 levels in mitochondria depend critically on the activity of 26-S proteasomes and an unknown metallopeptidase. It is conceivable that inhibition of the general mitochondrial processing peptidase causes the accumulation of the precursor form of OPA1 within mitochondria in the presence of o-phe. Similarly, 26-S proteasomes may degrade newly synthesized OPA1 in the cytosol before import into mitochondria rather than mediating the turnover of preexisting mitochondrial OPA1. In agreement with this hypothesis, MG132 did not stabilize v1ΔS1 in the absence of AFG3L1 and -2 when cytosolic protein synthesis was inhibited by CHX (Fig. S3 C). Thus, the inhibitor experiments did not allow us to identify proteases controlling the stability of L-OPA1 within mitochondria.

We have previously identified Oma1 in the inner membrane of yeast mitochondria as a metallopeptidase, whose activity overlaps with the m-AAA protease (Oma1 for overlapping activity with m-AAA protease 1; Käser et al., 2003). Oma1 is conserved in evolution with homologous proteases present in vertebrates. However, human OMA1 has been localized to the ER membrane (Bao et al., 2003). To reexamine the subcellular localization of OMA1, we expressed an OMA1 variant harboring a C-terminal GFP domain in MEFs (Fig. 5 B). OMA1-GFP was exclusively localized to mitochondria and not detected in the ER membrane (Fig. 5 B). Similarly, a tubular mitochondrial network was visible in immunofluorescence experiments after transient expression of an OMA1 variant harboring a C-terminal Flag tag (unpublished data). We conclude that OMA1, as yeast Oma1, is localized in mitochondria.

To examine whether OMA1 affects OPA1 processing in the mitochondrial inner membrane, we used three different oligonucleotides directed against Oma1 for RNAi. Quantitative RT-PCR experiments revealed efficient down-regulation of Oma1 transcripts by >80% (Fig. S4 A). Depletion of OMA1 slightly reduced the level of OPA1 isoforms c and e, which are generated by S1 cleavage and accumulate at low levels under these conditions in MEFs (Fig. 5 C). However, we observed a significant stabilization of L-OPA1 in AFG3L1/AFG3L2-deficient MEFs upon down-regulation of OMA1, whereas transfection of the cells with scrambled oligonucleotides had no effect (Fig. 5 C).

We monitored OPA1 processing directly after transient expression of the Flag-tagged OPA1 variant v1 in MEFs (Fig. 5 D). In the presence of AFG3L1 and -2, Flag-tagged L-OPA1 accumulated in OMA1-depleted cells (Fig. 5 D), indicating that OMA1 is able to cleave overexpressed OPA1. In contrast, in the absence of AFG3L1 and -2, down-regulation of OMA1 stabilized L-OPA1, demonstrating that OMA1 mediates turnover and increased OPA1 processing in the absence of the m-AAA protease (Fig. 5 D).

OPA1 isoforms accumulated virtually as in wild-type (WT) cells in MEFs depleted of both OMA1 and the m-AAA protease. Therefore, we examined the morphology of the mitochondrial network after expression of a mitochondrially targeted DsRed variant in these cells (Fig. 5, E and F). Depletion of OMA1 did not significantly affect mitochondrial morphology in the presence of the m-AAA protease but restored the tubular network in AFG3L1/AFG3L2-deficient MEFs (Fig. 5, E and F). Electron micrographs revealed a normal, lamellar cristae morphology of mitochondria lacking both the m-AAA protease and OMA1 (Fig. 5 E). These findings substantiate our conclusion that the m-AAA protease regulates mitochondrial fusion via OPA1 and demonstrate that OMA1 is epistatic to the m-AAA protease.

To further define the functional interrelationship of OMA1 and the m-AAA protease, we assessed the assembly of m-AAA proteases in OMA1-depleted mitochondria by blue-native gel electrophoresis (Fig. S4 B). Down-regulation of OMA1 did not affect the formation of the proteolytic complexes containing AFG3L2 and/or -1. Moreover, depletion of OMA1 did not interfere with the maturation of AFG3L1 or -2 (Fig. 5 C and Fig. S5), which occurs autocatalytically upon import into mitochondria (Koppen et al., 2009). Similarly, processing of an OMA1 variant carrying a C-terminal myc tag was not impaired in mitochondria depleted of AFG3L1 and -2 (Fig. S4 C).

Mutations in the m-AAA protease subunits paraplegin and AFG3L2 cause neurodegeneration in human, but molecular details of the pathogenesis are currently not understood. Our experiments revealed an at least partially redundant function of AFG3L1 and -2 for mitochondrial fusion in MEFs (Fig. 1). However, as AFG3L1 is a pseudogene in humans (Kremmidiotis et al., 2001) and is expressed at low levels in the mouse brain (Koppen et al., 2007; Martinelli et al., 2009), it is conceivable that the regulation of OPA1 processing by the m-AAA protease is of potential pathogenic relevance. To examine whether the loss of AFG3L2 impairs OPA1 cleavage in vivo, we monitored OPA1 processing in mitochondria isolated from tissues of a recently established Afg3l2^−/− mouse line (Maltecca et al., 2008). The abundance of individual OPA1 isoforms varies in different tissues, as previously described (Fig. 6 A; Akepati et al., 2008). In the absence of AFG3L2, L-OPA1 isoforms accumulated at strongly reduced levels in the brain, heart, and kidney, demonstrating that the stability of L-OPA1 depends on AFG3L2 in vivo. Notably, deletion of Afg3l2 hardly affected OPA1 isoforms in the liver (Fig. 6 A), which is consistent with a compensatory effect of AFG3L1, which is expressed at higher levels relative to AFG3L2 in the liver (Koppen et al., 2007).

To analyze whether depletion of OMA1 can suppress deleterious effects of the loss of the Afg3l2 gene, we monitored OPA1 processing and mitochondrial morphology in Afg3l2^−/− MEFs (Fig. 6, B–D). Deletion of Afg3l2 only slightly affected cleavage of OPA1 in MEFs grown on glucose-containing medium, most likely because of the presence of AFG3L1 in these cells (Fig. 6 B). However, L-OPA1 was absent in Afg3l2^−/− MEFs grown in the presence of galactose, i.e., conditions with an increased demand for mitochondrial respiratory function (Fig. 6 B). This was accompanied by a fragmentation of the mitochondrial network in the majority of these cells (Fig. 6, C and D). Depletion of OMA1 in immortalized Afg3l2^−/− MEFs, cultured either in glucose- or galactose-containing medium, stabilized L-OPA1, providing further evidence that OMA1 mediates OPA1 processing if the function of the m-AAA protease is impaired (Fig. 6 E).

We next examined whether OMA1 also cleaves OPA1 under other cellular conditions known to induce OPA1 processing. Current models suggest that a dysfunction of mitochondria accompanied by decreased mitochondrial ATP levels stimulates OPA1 processing and results in mitochondrial fragmentation (Duvezin-Caubet et al., 2006; Baricault et al., 2007). Therefore, we compared the accumulation of OPA1 isoforms in a human osteosarcoma cell line containing (WT) or lacking mitochondrial DNA (mtDNA; ρ⁰; Fig. 7 A). Although long and short OPA1 isoforms accumulated in these cells in the presence of mtDNA, short isoforms accumulated in ρ⁰ cells (Fig. 7 A). Strikingly, long OPA1 isoforms were detected after RNAi-mediated down-regulation of OMA1, suggesting that the loss of mtDNA induces OPA1 cleavage by OMA1.

OPA1 is converted into short isoforms upon dissipation of Δψ_m with the uncoupler CCCP or upon inhibition of the F₁F_O-ATP synthase with oligomycin (Fig. 7, B and C; Duvezin-Caubet et al., 2006; Ishihara et al., 2006). Down-regulation of OMA1 before the addition of CCCP or oligomycin significantly stabilized long OPA1 isoforms, indicating that OPA1 processing is mediated by OMA1 under both conditions (Fig. 7, B and C). In conclusion, these experiments identify OMA1 as the protease responsible for the inducible cleavage of OPA1 at processing site S1.

Mutations in m-AAA protease subunits cause neurodegeneration, but pathogenic mechanisms remain enigmatic. Our results identify the m-AAA protease as a novel regulator of mitochondrial membrane fusion and cristae morphology, suggesting that m-AAA protease–associated diseases may represent a group of neurological disorders caused by deficiencies in mitochondrial dynamics. Murine AFG3L1- and AFG3L2-containing m-AAA protease isoenzymes exert at least partially overlapping activities during membrane fusion. Therefore, the relative expression of AFG3L1 and -2 in different tissues will determine whether or not a functional impairment of one m-AAA protease subunit affects the morphology of the mitochondrial network. Notably, the m-AAA protease subunit predominantly expressed in the mouse brain is AFG3L2, whereas AFG3L1 is only present at lower levels (Koppen et al., 2007; Martinelli et al., 2009). This explains severe neurological defects (Maltecca et al., 2008) and cerebellar degeneration in mice with reduced levels of AFG3L2-containing m-AAA isoenzymes (Martinelli et al., 2009), which is reminiscent of patients with heterozygous mutations in human AFG3L2 (Cagnoli et al. 2008. Annual meeting of the American Society of Human Genetics. Abstr. 1501; DiBella et al. 2008. Annual meeting of the American Society of Human Genetics. Abstr. 216).

Whereas mitochondrial fusion depends on the presence of either AFG3L1 or -2, the third m-AAA protease subunit paraplegin is not required to maintain a reticulated mitochondrial network. Deletion of Spg7-encoding paraplegin did not affect mitochondrial morphology in MEFs. Moreover, phenotypes observed upon down-regulation of AFG3L1 and -2 were not worsened in paraplegin-deficient cells. However, it should be noted that an efficient depletion of AFG3L1 and -2 inactivates paraplegin that only exists in heterooligomeric complexes with AFG3L1 and -2 (Koppen et al., 2007). Thus, it is conceivable that paraplegin-containing isoenzymes may control mitochondrial morphology in cells expressing limiting levels of AFG3L1 and -2. This may explain the cell type–specific axonal degeneration associated with mutations in paraplegin both in mouse and human (Casari et al., 1998; Ferreirinha et al., 2004).

The m-AAA protease controls mitochondrial fusion via OPA1 (Fig. 8). The loss of AFG3L2 in various tissues of the mouse, down-regulation of AFG3L1 and -2 in MEFs, or the expression of a dominant-negative AFG3L2 variant in human cells decreases OPA1 stability and results in an increased processing and turnover of long OPA1 isoforms. This causes fragmentation of the tubular mitochondrial network, which is at least partially restored upon expression of an OPA1 variant lacking the cleavage site S1. Thus, our experiments link, for the first time, the function of AFG3L1 and -2 to OPA1 in vivo.

Strikingly, inactivation or loss of the m-AAA protease does not result in the accumulation of long OPA1 isoforms, as predicted by experiments in yeast (Duvezin-Caubet et al., 2007), but decreases their stability. The accelerated cleavage of OPA1 in the absence of the m-AAA protease is mediated by OMA1, a new member of a conserved and widespread family of membrane-bound M48 metallopeptidases. We have previously identified yeast Oma1 in the mitochondrial inner membrane as a peptidase that can substitute for the function of the m-AAA protease during degradation of a misfolded membrane protein (Käser et al., 2003). Similarly, synthetic growth defects have been observed when mutations in the bacterial AAA protease FtsH were combined with mutations in HtpX, which is a distant homologue of OMA1, indicating overlapping proteolytic activities (Shimohata et al., 2002). In agreement with these findings, we demonstrate in this study that mammalian OMA1, similar to the m-AAA protease, can cleave OPA1. The substrate specificity of OMA1 is distinct from yeast Oma1, which cleaves neither OPA1 when expressed in yeast nor the yeast OPA1 homologue Mgm1 (Duvezin-Caubet et al., 2007). Therefore, differences in substrate recognition by OMA1-like peptidases may explain why the loss of the m-AAA protease is not accompanied by increased OPA1 processing in yeast. Interestingly, mitochondrial rhomboid proteases, although highly conserved, also differ in their ability to cleave OPA1. Both yeast Pcp1 and its mammalian homologue PARL mediate processing of Mgm1, but neither of them cleave OPA1 when expressed in yeast (Duvezin-Caubet et al., 2007).

The regulation of OPA1 stability and processing at site S1 by two peptidases raises the question of which peptidase mediates constitutive and stress-induced OPA1 cleavage. Our experiments reveal that OMA1 promotes OPA1 processing if mitochondrial activities are impaired (Fig. 8). The loss of mtDNA, dissipation of Δψ_m, or a decrease of mitochondrial ATP levels induces OPA1 processing by OMA1. The molecular mechanism of OMA1 activation under stress conditions remains to be determined. Mitochondrial dysfunction may alter the accessibility of the S1 cleavage site in OPA1, allowing stress-induced cleavage by OMA1. Alternatively, stress conditions may increase the proteolytic activity of OMA1 itself. It is also possible that stress conditions lower mitochondrial ATP levels and thereby m-AAA protease activity, resulting in the accumulation of an activator of OMA1.

In contrast to the stress-induced OPA1 cleavage, constitutive cleavage of OPA1 during mitochondrial biogenesis affects only a subset of the newly imported OPA1 molecules and results in an approximately equimolar accumulation of long and short isoforms. As both forms are required for mitochondrial fusion (Song et al., 2007), their balanced formation is critical for the maintenance of the mitochondrial network. Although our experiments identify OMA1 as the peptidase responsible for stress-induced OPA1 cleavage, two possibilities must be considered for constitutive OPA1 processing.

According to the first scenario, OMA1 is responsible for both constitutive and induced OPA1 cleavage. However, the m-AAA protease is required to stabilize long OPA1 isoforms and inhibit OMA1 activity under normal conditions, possibly exerting functions beyond a role as a processing enzyme. These may include the ATP-dependent membrane dislocation of OPA1 preceding its proteolytic cleavage. A similar nonproteolytic activity has been assigned to the m-AAA protease during the biogenesis of yeast cytochrome c peroxidase (Tatsuta et al., 2007). In yeast, the ATP-dependent mitochondrial import motor was proposed to regulate Mgm1 processing by the rhomboid protease in response to mitochondrial ATP levels in a process termed alternative topogenesis (Herlan et al., 2004). Constitutive OPA1 processing might be regulated in a similar manner. Notably, membrane dislocation could also be affected by an altered phospholipid composition of mitochondrial membranes, which has recently been observed in yeast mitochondria lacking the m-AAA protease (Osman et al., 2009). Several observations indeed suggest that OMA1 can cleave OPA1 in the presence of the m-AAA protease. Depletion of OMA1 in human osteosarcoma cells or in MEFs cultured in galactose-containing medium leads to the disappearance of OPA1 isoforms c and e, indicating that OMA1 can affect constitutive OPA1 processing at S1. Moreover, OMA1 is required for S1 cleavage of OPA1 overexpressed in MEFs. However, an OMA1-like peptidase appears to be absent in Caenorhabditis elegans and Drosophila melanogaster, although mitochondrial morphology depends on OPA1 in both organisms. In addition, neither down-regulation nor overexpression of OMA1 affects mitochondrial morphology in MEFs, suggesting that OMA1 is dispensable for the balanced formation of long and short OPA1 isoforms.

According to the second scenario, constitutive and stress-induced OPA1 processing are mediated by two distinct peptidases, which differ in their enzymatic properties: the ATP-dependent m-AAA protease ensures the balanced accumulation of long and short isoforms under constitutive conditions, whereas OMA1 acting in an ATP-independent manner completely converts long OPA1 isoforms into short forms under stress conditions. In agreement with a role of the m-AAA protease as a processing enzyme under normal conditions, reconstitution experiments in yeast demonstrate that AFG3L1 and -2 can cleave OPA1 in a heterologous environment (Duvezin-Caubet et al., 2007). Moreover, paraplegin binds OPA1 when overexpressed and stimulates OPA1 processing (Ishihara et al., 2006). OPA1 processing by ATP-dependent and -independent peptidases under different physiological conditions would resolve the apparent caveat of previous studies suggesting that processing of OPA1 may involve the m-AAA protease but is accelerated at decreased mitochondrial ATP levels (Duvezin-Caubet et al., 2006, 2007; Ishihara et al., 2006; Baricault et al., 2007). It also offers an attractive possibility for regulation: low ATP levels may impair membrane dislocation of OPA1 possibly mediated by the m-AAA protease and thereby make OPA1 accessible for OMA1 cleavage under stress conditions. Clearly, further experiments are required to define the relative contribution of the m-AAA protease and OMA1 to constitutive OPA1 cleavage, which may also depend on relative expression levels in different tissues and cell types. Regardless, the identification of proteases involved in OPA1 processing now allows us to assess the relevance of an impaired OPA1 cleavage and mitochondrial fragmentation for the maintenance of cellular functions and for various disease states.

All animal procedures were performed according to protocols approved by the Institutional Animal Care and Use Committee. The generation of Spg7^−/− mice (Ferreirinha et al., 2004) and Afg3l2^Emv66/Emv66 genetic characterization (Maltecca et al., 2008) were previously described. Afg3l2^Emv66/Emv66 mice were provided by G. Cox (The Jackson Laboratory, Bar Harbor, ME). Primary MEFs were established from embryonic day (E) 14.5 Spg7^−/− or E13.5 Afg3l2^Emv66/Emv66 and littermate WT embryos and immortalized by SV40 transformation. MEFs were cultured in DME containing 4.5 g/liter glucose and supplemented with 10% FCS. When specified, MEFs were cultured in DME with 6 mM galactose. The human osteosarcoma-derived cell line 143B and its ρ⁰ derivative were provided by R. Wiesner (University of Cologne, Cologne, Germany) and grown in medium supplemented with 50 µg/ml uridine.

RNAi-mediated knockdown of paraplegin, AFG3L1 and -2, and OMA1 was performed using specific Stealth RNAi and the nontargeting Stealth RNAi negative control (Invitrogen). The sequences are 5′-GCGCGUCAUUGCUGGUACUGCUAAA-3′ for paraplegin, 5′-CCUGCCUCCGUACGCUCUAUCAAUA-3′ for AFG3L2, 5′-GCGAAACCAUGGUGGAGAAGCCAUA-3′ for AFG3L1, and 5′-GGAUACAGUCAAAGUUGCAGAGUA-3′, 5′-AGUUCUCAGAGAGUCUUCAUGGCUA-3′, and 5′-GCUUCUUGGUCUGAGUGCAUUUGGA-3′ for murine OMA1. For human OMA1, Stealth siRNAs with the sequences 5′-TGGACTACTGCTTGCTGCAAAGGCT-3′ and 5′- TGGCAGCAAATGGAGTTCGTTGATA-3′ were used. 10–20 nM of each siRNA was transfected twice with Lipofectamine (Invitrogen). Down-regulation was monitored by immunoblot analysis of cell lysates generated after 2 d if not indicated otherwise.

Human Afg3l2 was cloned into pcDNA5/FRT/TO, and mutations were introduced into the WB motif (E408Q) or the proteolytic center (E575Q). Stable transformants in FlpIn T-REx 293 cells were obtained by cotransfection with the plasmid pOG44, allowing expression of Flp-recombinase (Invitrogen). Expression of fusion proteins harboring a C-terminal hexahistidine epitope tag was induced with 1 µg/ml tetracycline. After 24 h, we analyzed cell lysates by immunoblotting and assessed mitochondrial morphology. The coding region of murine Oma1 (GenBank/EMBL/DDBJ accession no. NM_025909) was cloned into the EcoRI–BamHI sites of the vector pEGFP-N2 in frame with GFP.

5×10³ FlpIn T-REx 293 cells expressing AFG3L2 variants were seeded in medium with or without 1 µg/ml tetracycline. To assess the number of living cells in culture, we used the One Solution Cell Proliferation Assay (Promega) and monitored the conversion of a tetrazolium compound to formazan, determining the absorbance at 490 nm.

Mitochondrial morphology was examined after expression of mito-DsRed using a microscope (DeltaVision; Applied Precision; Merkwirth et al., 2008). Fluorescence images were acquired using a camera (CoolSNAP HQ/ICX285; Sony) and a Plan-Apochromat N 60× NA 1.42 oil objective (Olympus). Image stacks were deconvoluted with the SoftWoRx Imaging Suite (Applied Precision). Image files were further processed with the CorelDRAW 11 Graphics Suite software (Corel). For colocalization of OMA1-GFP and mitochondria, we stained cells with MitoTracker red CMXRos (Invitrogen) and used a microscope (Eclipse E600; Nikon) and a Plan-Apochromat A 60× NA 1.40 oil objective (Nikon). Images were acquired with a confocal microscope (Radiance 2100; Bio-Rad Laboratories). For detection of the ER, cells were stained with an antiserum directed against calnexin and examined using an imaging station (Eclipse TE2000-S [Nikon] + CARV II [BD]) and a Plan-Apochromat VC 100× NA 1.40 oil objective (Nikon), and images acquired using a camera (QuantEM 512 SC; Photometrics) and deconvoluted with MetaMorph software (MDS Analytical Technologies).

MEFs were plated on glass coverslips, transfected with siRNA, and after 48 h, flat embedded (Merkwirth et al., 2008). Ultrathin sections (60–70 nm) were cut and mounted on pioloform-coated copper grids (Plano). Sections were stained with lead citrate and uranyl acetate and viewed with a transmission electron microscope (JEM-2100; JEOL Ltd.) operated at 80 kV. Micrographs were taken using a charge-coupled device camera (Erlangshen ES500W; Gatan Inc.).

Respiration of intact cells was measured at 37°C with a Clark-type electrode oxygraph (Hansatech Inc.). 2.5 × 10⁶ cells were assayed in 250 mM saccharose, 20 mM Hepes, pH 7.4, 10 mM KH₂PO₄, 4 mM MgCl₂, 1 mM EDTA, 5 mM glucose, 2 mM pyruvate, and 4 mM glutamate. After recording endogenous respiration, ATP synthase was inhibited with 2 µM oligomycin followed by uncoupling of oxidative phosphorylation using CCCP with concentrations of 250–750 nM. Cellular respiration was inhibited with 2 mM KCN and was corrected to KCN-insensitive respiration.

To induce SIMH, MEFs were exposed to 1 µM CHX for 3 h 2–3 d after transfection with siRNAs and plasmids and were incubated for 5 h with the following protease inhibitors: 5 µM MG132, 0.5 mM o-phe, 50 µM pepstatin A, 0.5 mM Pefabloc SC, and 50 µM E-64d. To induce OPA1 cleavage, MEFs were treated either with 2 µM oligomycin 60 h after transfection or with 20 µM CCCP 48 h after transfection.

Polyclonal antibodies directed against murine m-AAA protease subunits were described previously (Koppen et al., 2007). An antiserum recognizing murine LON was provided by C. Suzuki (University of Medicine and Dentistry of New Jersey, Newark, NJ). The following commercially available antibodies were used: α-OPA1 (Invitrogen), α-Flag M2 (Sigma-Aldrich), α–SLP-2 (GenWay Biotech Inc.), α-calnexin (Stressgen), and α–complex II (Invitrogen).

Fig. S1 shows the mitochondrial morphology in paraplegin-deficient MEFs. Fig. S2 demonstrates the maintenance of the mitochondrial membrane potential in Afg3l2^−/− mitochondria using TMRE fluorescence. Fig. S3 shows the stability of OPA1 in AFG3L1/AFG3L2-deficient cells lacking various mitochondrial proteases and the effect of an inhibition of cytosolic protein synthesis on OPA1 isoforms accumulating in m-AAA protease–deficient MEFs in the presence of MG132. Fig. S4 shows the quantification of OMA1 mRNA levels in MEFs after down-regulation of OMA1 by RNAi, demonstrates the assembly of AFG3L2 and -1 in m-AAA protease complexes in OMA1-deficient mitochondria by blue-native–PAGE analysis, and reveals normal processing of OMA1 in m-AAA protease–deficient mitochondria. Fig. S5 shows the accumulation of AFG3L1 and -2 in OMA1-deficient MEFs.

We thank Dr. G. Cox for providing the Afg3l2^Emv66 mouse line, Dr. K. Mihara for OPA1 expression plasmids, Drs. J. Downward, B. de Strooper, and R. Wiesner for cell lines, Dr. C. Suzuki for LON-specific antibodies, and Dr. T. Tatsuta for discussion.

This work was supported by grants from the Deutsche Forschungsgemeinschaft, the German-Israeli-Project (DIP grant F.5.1), and the European Research Council to T. Langer and the Muscular Dystrophy Association, United Mitochondrial Disease Foundation, and Fondazione Cariplo to E.I. Rugarli.