m-AAA and i-AAA complexes coordinate to regulate OMA1, the stress-activated supervisor of mitochondrial dynamics

The mitochondrial network optimizes different activities by continuously changing its morphology by the essential and antagonistic activities of fission and fusion (Pernas and Scorrano, 2016; Westermann, 2010). These dynamic processes are regulated by the pro-fission dynamin-related proteins (DRPs), DRP1, Fis1 and MFF1 (Gandre-Babbe and van der Bliek, 2008; James et al., 2003; Smirnova et al., 2001), and pro-fusion proteins, MFN1, MFN2 and OPA1, which are involved in the fusion of the outer and inner mitochondrial membrane (OMM and IMM), respectively (Cipolat et al., 2004; Pernas and Scorrano, 2016).

The GTPase protein OPA1 exists in different splicing isoforms that can be partially or totally proteolytically processed at one or two distinct sites (S1 and S2) (Delettre et al., 2001). The combination of splicing isoforms and proteolytic cleavage results in five electrophoretically distinguishable protein bands, named L1, L2 (the long forms) and S1, S2 and S3 (the short forms), which promote fusion and fission (L and S forms, respectively) (Anand et al., 2014). The balance between long and short OPA1 forms is finely regulated by two mitochondrial inner membrane proteases, OMA1 (Ehses et al., 2009) and YME1L1, which cleave OPA1 at different sites (Song et al., 2007).

The intermembrane (i)-AAA protease YME1L1 exposes its catalytic domain to the intermembrane space (Leonhard et al., 1996) and is responsible for generation of the S2-OPA1 form by proteolytic cleavage, whereas OMA1 gives rise to the S1 and S3 forms (Anand et al., 2014; MacVicar and Langer, 2016; Quirós et al., 2012). OMA1 harbours an M48 metallopeptidase domain and is the major player in OPA1 processing under conditions of stress (Quirós et al., 2012). Different stress stimuli, such as dissipation of Δφ, increased ROS production, decreased ATP level, heat shock and loss of mtDNA have been demonstrated to overactivate OMA1 (Anand et al., 2014; Baker et al., 2014; Ehses et al., 2009; Head et al., 2009). Recently, a threshold of Δφ was identified as a determinant of mitochondrial homeostasis mediated by OMA1 and DRP1, which cooperatively regulate OPA1 maintenance and processing, and therefore control fission and fusion pathways (Jones et al., 2017).

The absence of YME1L1, which generates organellar stress, also activates OMA1 (Stiburek et al., 2012); in fact, YME1L1 and OMA1 are reciprocally regulated in response to specific stress stimuli. OMA1 is degraded by YME1L after toxic insults that depolarize the mitochondrial membrane and cause ATP depletion. The balance of degradation activities of these two proteins tunes the proteolytic processing of OPA1, which, in turn, affects the mitochondrial morphology state, thus profoundly modulating mitochondrial functions (Rainbolt et al., 2016).

OMA1 is synthesized as a pre-pro-protein of 60 kDa that undergoes proteolytic processing upon import into mitochondria to generate a mature form of 40 kDa (Baker et al., 2014; Head et al., 2009). It has been proposed that 40 kDa OMA1 is a stress-sensitive pro-protein that undergoes autocatalytic cleavage of a C-terminal subunit peptide upon stress stimuli to generate the active form of OMA1 that is ultimately responsible for OPA1 processing (Baker et al., 2014). All these findings indicate that OMA1 is a key sensor for a plethora of mitochondrial stress events.

Nevertheless, clear information on the generation of pro-OMA1 and the proteases involved in this mechanism is still missing. Here, we add an additional tile to this complex system of interlaced and regulated proteases by showing that (1) AFG3L2, the essential component of the matrix (m)-AAA complex, processes OMA1 by mediating its conversion from pre-pro-OMA1 to pro-OMA1, whereas (2) YME1L1 is tightly involved in pro-OMA1 catabolism, as previously suggested (Rainbolt et al., 2016).

With the aim of identifying the peptidase involved in OMA1 processing, HeLa cells were transfected with OMA1-HA, and inhibitors of different classes of peptidases were added to monitor the accumulation of the pre-pro-OMA1 form: PMSF to inhibit serine proteases, E64 for cysteine proteases and O-phe to block metallopeptidases (see the Materials and Methods). Only O-phe treatment induced the accumulation of the pre-pro-OMA1 form, indicating that the class of metallopeptidase is the one involved in the initial processing of the protein (Fig. 1A).

A total of 380 putative mitochondrial proteases were selected from four different protein databases: UniProt (UniProt Consortium, 2013), MitoCarta (Pagliarini et al., 2008), MitoMiner (Smith and Robinson, 2009) and Gene Ontology (Ashburner et al., 2000). Combining O-phe sensitivity with the presence of the protease in at least two of the databases, we reduced candidates to 27 proteins. Among them, all proteins inhibited by metal ions or recognizing specific consensus sequence not present in OMA1 or showing deacetylase/hydrolase activity were excluded. Since OMA1 is localized in the IMM (Baker et al., 2014) and presumably its cleavage occurs in this site, four proteins localized in the IMM were prioritized: paraplegin (encoded by the SPG7 gene), AFG3L2, YME1L1 and OMA1 (Fig. 1B).

We excluded OMA1 itself from this initial step since the expression of the proteolytic inactive mutant (OMA1-Q-HA) does not affect the pre-pro OMA1 to pro-OMA1 cleavage (Fig. 1C). However, we confirmed that OMA1 has an active role on its own processing, in particular in the conversion from the pro-OMA1 to the fragment of ∼30 kDa, as previously reported (Baker et al., 2014). Indeed, we observed that the expression of OMA1-Q-HA prevents the processing to the 30 kDa OMA1 fragment, demonstrating that it is generated by auto-processing (Fig. 1C). By transfecting OMA1-HA in HSP patient fibroblasts lacking paraplegin (Atorino et al., 2003), we found no alteration of OMA1 processing (Fig. 1D), thus excluding paraplegin from candidates and confirming previous findings (Ehses et al., 2009).

We then evaluated OMA1 processing in the absence of AFG3L2 in MEFs (Fig. 1E). Interestingly, we found a significant accumulation of pre-pro-OMA1 and a corresponding reduction of the pro-OMA1 form. Pre-pro-OMA1 accumulation is not due to alterations in the mitochondrial import machinery since we previously demonstrated that the absence of AFG3L2 does not affect the import of nuclear encoded proteins into mitochondria (Maltecca et al., 2008). Importantly, similar results were also obtained with endogenous OMA1 (Fig. 1F,G) in both HeLa cells and MEFs. AFG3L2-knockdown HeLa cells showed accumulation of the pre-pro-OMA1 form and reduction of the pro-OMA1 form (Fig. 1F). Afg3l2^−/− MEFs (Fig. 1G) show a decreased pro-OMA1 form, indicating an active role of AFG3L2 in pre-pro-OMA1 processing.

The diminished conversion of pre-pro-OMA1 into the pro-OMA1 form in the absence of AFG3L2 can be explained by a direct cleavage of OMA1 by AFG3L2 or by an indirect cleavage mediated by another protease. To discriminate between these alternative possibilities, we co-transfected Afg3l2^−/− MEFs with OMA1-HA and AFG3L2-myc or its proteolytic inactive mutant (AFG3L2-Q₁-myc) and then monitored the processing of both OMA1 and OPA1. In the absence of AFG3L2, OPA1 is processed to the short forms (Ehses et al., 2009; Maltecca et al., 2012) and at the same time pre-pro-OMA1 accumulates (Fig. 2A).

The re-expression of AFG3L2 induces a slight increase of OPA1 L1, so rescuing the physiological processing of OPA1 and inducing an increase of pro-OMA1. In contrast, the expression of the proteolytically inactive form of AFG3L2 cannot restore OMA1 processing (Fig. 2A). We therefore explored the possible physical interaction. To stabilize the interaction, we decided to use AFG3L2-Q₁-Myc, which is still capable of substrate binding but lacks proteolytic activity (Atorino et al., 2003), as a trap for interacting proteins. As hypothesized, AFG3L2 co-immunoprecipitates with OMA1 (Fig. 2B).

To gain further evidence of the role of AFG3L2 in OMA1 processing in human cells, which unlike murine cells do not express the AFG3L2 homologue AFG3L1, we tested the accumulation of pre-pro-OMA1 in HeLa cells silenced for AFG3L2 and demonstrated that they behave similarly to Afg3l2^−/− MEFs. In fact, upon silencing of AFG3L2, we detected the accumulation of the pre-pro-OMA1 and the presence of intermediate bands between pre-pro-OMA1 and pro-OMA1 (Fig. 2C, first two lanes). Moreover, expression of proteolytic siRNA-insensitive AFG3L2 mutants (AFG3L2-ins-Q₁-Myc, AFG3L2-ins-Q₂-Myc) failed to reduce the amount of the pre-pro-OMA1 form (Fig. 2C, right panel).

In order to identify the putative target sequence, a series of segmentally deleted OMA1 mutants revealed that pre-pro-OMA1 to pro-OMA1 maturation is the consequence of an N-terminal trimming effect operated by the m-AAA complex, similar to maturation of MRPL32, another m-AAA-specific substrate (Bonn et al., 2011) (Fig. 2D). Moreover, since the N-terminus of 40 kDa OMA1 has been mapped at amino acid 140 (Baker et al., 2014), we generated a vector expressing a truncated form of OMA1 that mimics pro-OMA1 by removing 53 amino acids between the mitochondrial leader sequence and the first N-terminal amino acids of pro-OMA1. This mutant (OMA1-Δ92-144-HA) encodes a protein of 50 kDa and is processed like the wild-type protein to generate pro-OMA1 (Fig. 2E). We also found that OMA1-Δ92-144 is proteolytically active since it is able to cleave L-OPA1, restoring the bands S1 and S3, which both disappeared after OMA1 silencing (Fig. 2E).

Our findings demonstrate the major role of AFG3L2 in OMA1 processing, and in particular, in the conversion from the pre-pro-protein to the pro-protein. However, the presence of a residual amount of pro-OMA1 in the absence of the m-AAA complex indicates that an alternative unknown salvage pathway operated by metalloprotease is active to ensure the generation of the pro-protein.

Based on our results, we propose that the absence of AFG3L2 hampers the pre-pro-OMA1 to pro-OMA1 conversion, leading to the accumulation of the pre-pro-protein, but at the same time, the stress caused by the absence of AFG3L2 (decreased assembling of respiratory complexes, ATP depletion, increased ROS production, accumulation of peptides) activates OMA1 auto-catalytic cleavage, leading to a reduction of pro-OMA1. The activated pro-OMA1 is generated by an autocatalytic processing that involves the final C-terminal amino acids of pro-OMA1 (Baker et al., 2014), causing the removal of the HA moiety from the HA-tagged OMA1 protein. In support of this, reintroducing AFG3L2 in Afg3l2^−/− cells reduces the mitochondrial stress, hence decreasing pro-OMA1 C-terminal auto-cleavage and increasing the amount of pro-OMA1 (Fig. 2A).

We tested the involvement of YME1L1 in the generation of pro-OMA1. We failed to demonstrate pre-pro-OMA1 accumulation by silencing YME1L1, hence excluding the direct involvement of YME1L1 in pre-pro-OMA1 processing (Fig. 3A). On the contrary, depletion of YME1L1 strongly increased the amount of pro-OMA1 (Fig. 3A), confirming a role for YME1L1 in the regulation of pro-OMA1 turnover (Rainbolt et al., 2016).

We excluded the increased amount of m-AAA or enhanced OMA1 autocatalytic activity (Fig. 3A) or the increased protease activity of the m-AAA complex (Fig. 3B) as possible causes of pro-OMA1 accumulation following YME1L1 knockdown. YME1L1 is indeed the specific regulator of pro-OMA1 since in YME1L1-knockdown cells pro-OMA1 accumulates (Fig. 3C). Therefore, YME1L1 is directly involved in regulating the pro-OMA1 level, thus preventing OMA1 accumulation, OPA1 over-processing and the consequent mitochondrial network fragmentation.

Taken together, these data support for the first time that the m-AAA and i-AAA complexes actively cooperate in controlling mitochondrial dynamics through the fine regulation of OMA1 processing and stability. The vital necessity of accurately controlling OMA1 proteolytic activity is underlined by the findings of a ‘tipping-point’ threshold of Δφ due to both DRP1 and OMA1, which act coordinately to balance the fusion–fission equilibrium under Δφ stimulus (Jones et al., 2017).

We showed that AFG3L2 is crucial for OMA1 maturation, hence directly allowing its conversion from pre-pro-OMA1 to pro-OMA1 by a trimming mechanism involving the solvent-exposed N-terminal part of the protein. Possibly, another as yet unknown protease could be involved in the generation of pro-OMA1, highlighting the existence of a compensatory mechanism acting when AFG3L2 is absent or mutated. We indeed demonstrated that the i-AAA complex is involved in the regulation of pro-OMA1 turnover, preventing its accumulation and, consequently, mitochondrial network fragmentation.

A possible model of functional interactions between m-AAA, i-AAA and OMA1 is shown in Fig. 4. Under physiological conditions, AFG3L2 cleaves pre-pro-OMA1 to generate abundant pro-OMA1, which is further controlled by the activity of i-AAA, which regulates pro-OMA1 accumulation and ensures a balance between long and short OPA1 forms.

In contrast, the absence of AFG3L2 induces a wide range of organellar damage, including defective assembly of respiratory complexes, ROS production, ATP depletion and inner membrane proteostatic stresses (Maltecca et al., 2008; Richter et al., 2015) that possibly turn on an emergency pathway operated by other proteases that activate OMA1 and the downstream effect of mitochondrial fragmentation. The same effect is caused by the absence of YME1L1 and the resultant pro-OMA1 accumulation, which enhances its autocatalytic activity and, again, induces mitochondrial fragmentation. Considering the critical role of OMA1 in the mitochondrial fusion–fission homeostasis, it is conceivable that multiple alternative effectors, including AFG3L2, YME1L1 and OMA1 itself, together with Δφ, are needed to maintain the effective fine-tuning of this crucial stress sensor protein.

For the generation of pcDNA3.1-Oma1-HA (herein OMA1-HA), Oma1 mouse cDNA clone was obtained from Thermo Scientific (Waltham, MA, USA) and an HA tag sequence introduced. The proteolytically inactive OMA1E324Q-HA was obtained by site-specific mutagenesis of pcDNA3.1-Oma1-HA. To generate pcDNA3.1-deleted-Oma1-HA mutants (in the text: Δ1-Δ9) we introduced the indicated deletions performing site-directed mutagenesis: Δ1, pcDNA3.1-Oma1-Δ147-149-HA; Δ2, pcDNA3.1-Oma1-Δ150-152-HA; Δ3, pcDNA3.1-Oma1-Δ153-155-HA; Δ4, pcDNA3.1-Oma1-Δ169-172-HA; Δ5, pcDNA3.1-Oma1-Δ170-HA; Δ6, pcDNA3.1-Oma1-Δ171-HA; Δ7, pcDNA3.1-Oma1-Δ172-HA; Δ8, pcDNA3.1-Oma1-Δ175-178-HA; Δ9, pcDNA3.1-Oma1-Δ177-180-HA.

In pcDNA3.1-Oma1-Δ92-144-HA (herein OMA1-Δ92-144-HA), we introduced a deletion between amino acids 92 and 144.

pcDNA3.1-AFG3L2-Myc (AFG3L2-myc), was previously described (Maltecca et al., 2012). pcDNA3.1-AFG3L2-E408Q-Myc and pcDNA3.1-AFG3L2-E575Q-Myc appear in the text as AFG3L2-Q₁-Myc and AFG3L2-Q₂-Myc, respectively.

In order to make them insensitive to the stealth RNAi against AFG3L2, these constructs were mutagenized without changing the sense of encoded amino acid sequence, generating: pcDNA3.1-AFG3L2-Ins-Myc, pcDNA3.1-AFG3L2-Ins-E575Q-Myc (AFG3L2-Ins-Q₁-Myc), pcDNA3.1-AFG3L2-Ins-E408Q-Myc (AFG3L2-Ins-Q₂-Myc).

pMT21-YME1L1-Myc (YME1L1-Myc) was previously described (Coppola et al., 2000). pMT21-YME1L1-E543Q-Myc (YME1L1-Q-Myc) was generated by site-directed mutagenesis. To make them insensitive to the stealth RNAi against YME1L1, these constructs were mutagenized without changing the sense of encoded amino acid sequence. MrpL32 cDNA was isolated from HeLa cells, HA-tagged and cloned into pCDNA3.1 (MRPL32-HA in the text). All primers are available on request. We confirmed all constructs by Sanger sequencing.

Human immortalized fibroblasts, MEFs and HeLa cells were maintained in DMEM containing penicillin-streptomycin 10 μg/ml, 10% fetal bovine serum, 2 mM L-glutamine and 1 mM sodium pyruvate. Transient transfections for the overexpression of the indicated constructs or for the silencing of the indicated genes were performed using Lipofectamine 2000 (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's instructions.

Knockdown of AF3GL2, OMA1 and YME1L1 was performed using specific stealth RNAi siRNAs and scrambled stealth RNAi siRNAs as a negative control (Invitrogen, Carlsbad, CA, USA). The sequences of the stealth RNAi siRNAs are 5′-acgacuuccaaucugucuacuacuc-3′ for human AFG3L2, 5′-uggacuacugcuugcugcaaaggcu-3′ for human OMA1, 5′-uccagaaacccaaucugccaucgaa-3′ for human YME1L1; 100 pmoles of each siRNA were transfected. Down-regulation of the target genes was monitored by immunoblot analysis on cell lysates 72 h after transfection.

Western blot was performed using standard protocols. Briefly, total cell lysate was prepared by protein extraction using the following buffer: 50 mM Tris-HCl, pH 8, 150 mM NaCl, 2% Triton X-100, 0.5 mM EDTA, pH 8, 1× protease inhibitor cocktail (Roche Applied Science, Penzberg, Upper Bavaria, Germany). 25 µg of protein extracts were dissolved in sample buffer (60 mM Tris-HCl, pH 6.8, 5% glycerol, 1.7% SDS, 0.1 M DTT, 0.002% Bromophenol Blue), were resolved by SDS-PAGE and analysed by standard immunoblotting procedures. Anti-HA antibody was from Sigma-Aldrich (H6908, 1:2000), anti-OPA1 was from BD Transduction Laboratories (Franklin Lakes, NJ, USA, 612606, 1:2500), anti-Hsp70 antibody and anti-GAPDH were from Santa Cruz Biotechnology (sc-32239, 1:1000 and sc-32233, 1:10,000), anti-c-Myc and anti-OMA1 were from Novus Biologicals (Littleton, CO, USA, NB600-335, 1:1000 and NBP2-30971, 1:500), anti-YME1L1 antibody was from ProteinTech Group (Chicago, IL, USA, 11510-AP, 1:1000), anti-Hsp60 was from Enzo Life Science (Farmingdale, NY, USA, ADI-SPA-806, 1:7000) and anti-AFG3L2 was previously generated in the lab (Atorino et al., 2003) and used at 1:2000.

HeLa cells were transfected with OMA1-HA for 14 h and subsequently incubated for 10 h with the following protease inhibitors: 0.5 mM O-phe, 100 µM E-64d and 400 µM PMSF (Sigma-Aldrich, St Louis, MO, USA).

One mg protein was extracted from HeLa cells in lysis buffer (50 mM Tris-HCl, pH 8, 150 mM NaCl, 0.5 mM EDTA pH 8, 1× protease inhibitor cocktail, 0.15% Sarkosyl), after mild sonication. After a pre-clearing step (incubation with Protein-G–Sepharose beads) proteins were incubated with anti-Myc or anti-HA (Clone 16B12 monoclonal antibody, Affinity Matrix beads, Covance, Princeton, NJ, USA) antibodies for 1 h with constant gentle agitation. Subsequently, Protein-G–Sepharose was added. Precipitates were used for western blot analysis.

HeLa cells were treated with 10 µM carbonyl cyanide-4-(trifluoromethoxy)phenylhydrazone (FCCP) (Sigma-Aldrich) for 1 h.

We are grateful to Laura Cassina for scientific discussion; Maurizio De Fusco for technical assistance; Michela Riba and Davide Gaudesi for bioinformatic analyses. OMA1^−/− MEF cells were kindly provided by Carlos Lopez-Otin, Universidad de Oviedo, Oviedo, Spain.