Loss of the intermembrane space protein Mgm1/OPA1 induces swelling and localized constrictions along the lengths of mitochondria.

Mgm1 is a member of the dynamin family of GTP-binding proteins. Mgm1 was first identified in yeast, where it affects mitochondrial morphology. The human homologue of Mgm1 is called OPA1. Mutations in the OPA1 gene are the prevailing cause of dominant optic atrophy, a hereditary disease in which progressive degeneration of the optic nerve can lead to blindness. Here we investigate the properties of the Mgm1/OPA1 protein in mammalian cells. We find that Mgm1/OPA1 is localized to the mitochondrial intermembrane space, where it is tightly bound to the outer surface of the inner membrane. Overexpression of wild type or mutant forms of the Mgm1/OPA1 protein cause mitochondria to fragment and, in some cases, cluster near the nucleus, whereas the loss of protein caused by small interfering RNA (siRNA) leads to dispersal of mitochondrial fragments throughout the cytosol. The cristae of these fragmented mitochondria are disorganized. At early time points after transfection with Mgm1/OPA1 siRNA, the mitochondria are not yet fragmented. Instead, the mitochondria swell and stretch, after which they form localized constrictions similar to the mitochondrial abnormalities observed during the early stages of apoptosis. These abnormalities might be the earliest effects of losing Mgm1/OPA1 protein.

ciency (4). It was also discovered recently that the OPA1 gene encodes the human homologue of yeast Mgm1, thereby opening new avenues for research into the physiological disruptions that lead to atrophy of the optic nerve (5,6).
The Mgm1/OPA1 protein is a dynamin family member that was first discovered in yeast, where it affects mitochondrial genome maintenance (7). Members of the dynamin family are structurally similar but functionally diverse GTP-binding proteins that have a GTP binding domain, a middle domain, and a GTPase effector domain (GED or assembly domain) (8,9). These three domains bind to each other and are thereby capable of forming large multimeric assemblies (10). Different members of the dynamin family also have divergent sequences that appear to be specific for their particular functions. Dynamin, for example, has a pleckstrin homology domain and a prolinerich domain that bind to other molecules in clathrin-coated pits. A second member of the dynamin family, named Drp1, Dlp1, or Dymple in mammals and Dnm1p in yeast, is also cytosolic, but this protein is required for division of the mitochondrial outer membrane and is therefore likely to bind to proteins on the mitochondrial outer membrane (11)(12)(13)(14). In contrast, OPA1/Mgm1 has a mitochondrial leader sequence, suggesting that this protein is imported into mitochondria (15).
The submitochondrial localization of Mgm1/OPA1 has been a matter of debate. Although early reports suggest localization to the matrix (16) or the outer membrane (15), more recent reports indicate that Mgm1 is an intermembrane space protein (17)(18)(19)(20)(21)(22). There is also, however, ambiguity about the precise localization within the mitochondrial intermembrane space. It was reported that Mgm1 is peripherally associated with the mitochondrial inner membrane (17) or peripherally associated with both the outer and inner membranes (19), but it has also been reported that Mgm1 is an integral membrane protein (21). It is therefore not surprising that the exact function of Mgm1 has also been a matter of debate.
As indicated by the name of the yeast mutant, Mgm1 was originally proposed to be required for mitochondrial genome maintenance. More recent data indicates that yeast with mutations in their Mgm1 gene lose mitochondrial DNA as a secondary consequence of altered mitochondrial morphology (17). Mitochondria are converted from elaborate branched structures in wild type yeast cells into an aggregate of smaller fragments in cells with mutations in the Mgm1 gene. This phenotype is similar to that of yeast with mutations in the fuzzy onions protein Fzo1p, which is thought to affect fusion between mitochondria (23). This possible role of Mgm1 in mitochondrial fusion was tested with a cell fusion assay in yeast. Results obtained with these assays indicate that mutations in Mgm1 are capable of disrupting fusion between mitochondria (18). It has therefore been proposed that Mgm1 contributes to mitochondrial fusion (18,19) in a manner similar to the proposed role of the fuzzy onions protein in yeast and Drosophila and that of the mitofusin proteins in mammals (24 -26). There is, however, as yet no evidence for a direct role in mitochondrial fusion, leaving open the possibility that this protein affects other processes that indirectly affect the balance between mitochondrial fission and fusion.
Here we report our characterization of human Mgm1 at the cellular level. Our results confirm that this protein is targeted to the mitochondrial intermembrane space. Data obtained with subcellular fractionation and cryo-immuno-electron microscopy are consistent with localization of Mgm1/OPA1 to the mitochondrial intermembrane space, where it is tightly bound to the outer leaflet of the inner membrane. Silencing of Mgm1/ OPA1 expression results in fragmented mitochondria with aberrant inner membrane morphologies. At early time points, the only visible abnormalities are swelling and localized constrictions of the mitochondria.

MATERIALS AND METHODS
Cloning and Expression of the Human Mgm1 cDNA Expression Constructs-Human cDNA clone KIAA0567 was obtained from T. Nagase (Kazusa DNA Research Institute, Chiba, Japan). Full-length Mgm1/ OPA1 was recloned into the mammalian expression vector pCI (Promega, Madison, WI) by PCR amplification, adding KpnI and NotI sites to the ends of the PCR oligonucleotides. The K301A mutation was introduced by fusion PCR. The clones were sequenced to rule out errors introduced by PCR.
For bacterial expression, the GTPase domains of Mgm1 and Mgm1(K301A) were PCR amplified and cloned into NheI-and XhoI-cut pet21D (Novagen Inc., Madison, WI). These clones were transformed into BL21(DE3)pLysS bacteria (Novagen Inc.). Expression was induced with 40 M isopropyl-1-thio-␤-D-galactopyranoside added at an OD of 0.4 -0.6. The cells were then grown for 2 h at 27°C and harvested by centrifugation. The cell pellet was resuspended in 150 mM NaH 2 PO 4 and 300 mM NaCl, followed by the addition of 0.1% Triton X-100 and 1 mM phenylmethylsulfonyl fluoride and lysis using a French press. Bacterial debris was removed by centrifugation (15 min at 12,000 ϫ g), and the supernatant was passed through nickel-nitrilotriacetic acid agarose (Qiagen, Hilden, Germany). The GTPase fragments were subsequently eluted with 1 M imidazole, yielding Ͼ90% purity.
Antibodies-A 705-bp fragment encoding the Mgm1/OPA1 C terminus was cloned in the pET21d expression vector (Novagen Inc.) using NcoI and XhoI sites that were added to the ends of the PCR oligonucleotides. The resulting 26-kDa recombinant protein was expressed in BL21 bacteria and purified with nickel-nitrilotriacetic acid chromatography. Rabbit antisera were generated by Robert Sargeant (Ramona, CA). The antibodies were blot-purified using recombinant protein. ATP synthase subunit ␣ and cytochrome c oxidase subunit I antibodies were purchased from Molecular Probes (Eugene, OR), glutamate dehydrogenase from Rockland (Gilbertsville, PA), and HSP60 and TOM20 from Santa Cruz Biotechnology (Santa Cruz, CA). Porin antibodies were purchased from Calbiochem, tubulin from Sigma-Aldrich, prohibitin from Research Diagnostics, Inc. (Flanders, NJ), and TIM23 and cytochrome c antibodies from BD Biosciences.
Cell Culture and Transfection-HeLa and COS7 were grown in Dulbecco's modified Eagle's medium with 10% fetal calf serum. Vero cells were grown in Dulbecco's modified Eagle's medium with 5% serum. Plasmid DNA was transfected using FuGENE (Roche Molecular Biochemicals) and the protocol provided by the manufacturer. The siRNA 1 oligonucleotides were transfected using OligofectAMINE TM (Invitrogen) as described by the manufacturer. The effects of siRNA were observed by plating the cells in glass-bottomed dishes (Mat-Tek, Ashland, MA). HeLa cells used for electron microscopy were seeded on Thermanox® plastic coverslips (Nalge Nunc, Naperville, IL).
Immunofluorescence Microscopy-Cells grown on glass coverslips were fixed for 4 min at Ϫ20°C in methanol followed by permeabilization for 4 min at Ϫ20°C in acetone. Untransfected cells were incubated overnight with Mgm1/OPA1 antibody, followed by a 20-min incubation with a secondary antibody. Transfected cells were incubated for 20 min with an Mgm1/OPA1 antibody. Where indicated, cells were stained for 30 min with 0.1 M MitoTracker Red (Molecular Probes) prior to fixation. Images were acquired with a 100ϫ Neofluar/NA1.3 objective on a Zeiss Axiovert 200 M microscope, using a Hamamatsu ORCA ER camera controlled by Zeiss Axiovision software.
Subcellular Fractionation and Western Blotting-Fractionation procedures were as described previously (14). Briefly, 50 g of fresh bovine liver or brain was minced and homogenized in isolation buffer (70 mM sucrose, 220 mM mannitol, 2 mM HEPES, and 0.5 mg/ml bovine serum albumin) and protease inhibitor mix (Roche Diagnostics) by three passes with a loose fitting pestle in a Potter-Elvejhem homogenizer, followed by five passes with a tight fitting pestle. The homogenate was subjected to differential centrifugation as described previously (14). Mitochondria in the P2 fractions were further purified with a Percoll gradient, except when used for protease sensitivity experiments and carbonate washes. The P2 fractions used for protease sensitivity experiments were washed in isolation buffer without protease inhibitors.
Digitonin extraction was as described previously (27) with some minor modifications. Briefly, samples of the liver mitochondrial fraction P2 containing 1 mg of protein were mixed with equal volumes of digitonin solution to achieve the indicated concentrations of digitonin. The samples were then incubated for 30 min on ice with occasional mixing by inverting the tubes. The samples were then diluted with 3 volumes of isolation medium, and the mitochondria were pelleted by centrifugation for 10 min at 11000 ϫ g. The pellet was resuspended in 250 mM KCl and incubated for 5 min on ice, followed by centrifugation for 10 min at 11 000 ϫ g. This last pellet was resuspended in Laemmli sample buffer and size-fractionated by SDS-PAGE.
Protease protection experiments were done by resuspending 0.5 mg of mitochondrial proteins in 0.8 ml of isolation buffer. Trypsin and digitonin were added in that order at the indicated concentrations. After 30 min of incubation on ice, the reaction was stopped by adding trichloroacetic acid. Upon precipitation of the proteins, the pellets were resuspended in Laemmli sample buffer and size-fractionated by SDS-PAGE.
Alkaline extraction was performed by incubating the mitochondrial fraction for 30 min on ice with 0.2 M Na 2 CO 3 adjusted to pH 13 with 1 N NaOH. The membranes were then pelleted by centrifugation for 1 h at 100,000 ϫ g. Proteins in the supernatants were precipitated with trichloroacetic acid. The pellets were resuspended in Laemmli sample buffer and size-fractionated by SDS-PAGE.
The proteins that were size-fractionated by SDS-PAGE were then transferred to nitrocellulose for Western blotting. These blots were incubated with primary antibodies and developed with a horseradish peroxidase-conjugated secondary antibody and enhanced chemiluminescence reagents (Amersham Biosciences). The protein bands were scanned with a densitometer (Molecular Dynamics, Sunnyvale, CA), and the data were quantified using ImageQuant (Molecular Dynamics).
RNA Interference-The silencing of the Mgm1 gene in HeLa cells was done essentially as described (28). Synthetic oligonucleotides, matching the sequence from positions 511 to 533, were made by Genset Oligos (La Jolla, CA). A control duplex was designed by inverting this sequence. The siRNA duplexes were transfected at a concentration of 20 nM into HeLa cells grown to 40% confluency. The effect of this treatment was followed during the course of 3 days, at which point the cells were stained with MitoTracker, and their mitochondrial morphology was observed by fluorescence microscopy. The mitochondrial thickness was measured using the line tool in NIH Image 1.62. The measurements were done at multiple randomly selected sites along the mitochondrial length. The results are expressed as mean values Ϯ S.E.
GTP Hydrolysis-To assay GTP hydrolysis, 1-g samples of bacterially expressed wild type and mutant GTPase domains were incubated at 37°C in 30 l of 50 mM Tris, pH 7, 2 mg/ml bovine serum albumin, 0.1 mM dithiothreitol, and 5 mM MgCl 2 with 100 M non-radioactive GTP and trace amounts of [␣-32 P]GTP. At the indicated times, samples of 2.5 l were spotted onto polyethyleneimine cellulose F TLC plates (EM Science, Gibbstown, NJ). The reactions were stopped immediately with a hair drier. The plates were then developed with 1 M LiCl 2 and 1 M acetic acid and air-dried. Radiolabeled spots were quantified with a Typhoon™ PhosphorImager (Amersham Biosciences) and Image-Quant™ software (Molecular Dynamics).
Electron Microscopy-For negative staining, the cells were fixed for 30 min in 1% glutaraldehyde (Ted Pella Inc., Redding, CA), washed with PBS, and incubated for 1 h with 1% osmium tetroxide. The samples were then dehydrated and embedded in Epon resin. 70-nm thick sections were stained with uranyl acetate and lead citrate. The sections were viewed with a JEOL electron microscope (JEOL Ltd., Tokyo, Japan).
Cryo-immunogold Electron Microscopy-Subcellular localization of 1 The abbreviation used is: siRNA, small interfering RNA. antigens using high resolution cryo-immunogold electron microscopy was performed as described (29). Briefly, the mouse ileum was fixed by perfusion for 24 h with 2% paraformaldehyde or 2 h with 2% paraformaldehyde/0.2% glutaraldehyde. Fixed tissue was embedded in gelatin, and ultrathin sections were made with a microtome and mounted on grids. The sections were immuno-labeled with primary antibody and then with protein A-gold. The cryosections were examined with Philips CM10 and Technia 12 electron microscopes at 80 kV.

Localization of Human Mgm1/OPA1-A BLAST search of
GenBank TM for clones sharing homology with dynamin revealed a cDNA sequence, designated KIAA0567, which encodes human Mgm1/OPA1. To determine the subcellular and submitochondrial localization of the Mgm1/OPA1 protein, we generated polyclonal antibodies against a 26-kDa C-terminal fragment of Mgm1/OPA1 spanning amino acid residues 616 -850. This antibody recognizes ϳ95and ϳ85-kDa bands by Western blotting of C2C12 and HeLa cell extracts fractionated with 10% SDS-PAGE (Fig. 1A), consistent with previous observations in yeast, mouse, and humans. The 85-kDa band is often detectable as a closely spaced doublet. These bands may represent partial proteolytic products, mature processing products, or isoforms generated by alternative splicing. We then determined whether the proteins recognized by our antibodies were mitochondrial by differential centrifugation and further fractionation with a Percoll gradient. Our results were consistent with the mitochondrial localization of the Mgm1/OPA1 protein (Fig. 1B).
Immunofluorescence staining of fixed tissue culture cells shows that the Mgm1/OPA1 protein colocalizes with ATP synthase (Fig. 1C) and with MitoTracker (data not shown) on mitochondria. The pattern within mitochondria was punctate, but close inspection of enlarged images showed that these puncta were distinct from the staining patterns of other mitochondrial proteins (Drp1, COX, HSP60, prohibitin, and TIM23) viewed at high magnification. 2 We conclude that the Mgm1/ OPA1 protein is mitochondrial but that the submitochondrial distribution, as detected by immunofluorescence, is distinct from that of known mitochondrial proteins.
Submitochondrial Localization of Mgm1/OPA1-To determine the localization of the Mgm1/OPA1 protein within mitochondria, we analyzed mitochondrial fractions obtained from bovine liver by treating them with increasing concentrations of digitonin along with protease digestion and carbonate washes. The mitochondrial outer membrane is solubilized at a lower concentration of digitonin than the mitochondrial inner membrane, which makes it possible to selectively release intermembrane space proteins or make them accessible to protease digestion (27). Increasing concentrations of digitonin first released cytochrome c, but only when these mitochondria were also incubated with 250 mM KCl, as expected for a peripheral membrane protein ( Fig. 2A). Under these same conditions, Mgm1 was not released into the supernatant, suggesting that this protein is even more tightly bound to the membrane or enclosed in the mitochondrial matrix proteins. The Mgm1/ OPA1 protein was released at higher digitonin concentrations along with mitochondrial matrix proteins and inner membrane proteins. These data suggest that Mgm1/OPA1 is not an outer membrane protein or a soluble intermembrane space protein.
To determine whether Mgm1 is exposed to the intermembrane space or enclosed in the matrix, we combined the digitonin extractions with protease digestions (Fig. 2B). Relatively low concentrations of digitonin make holes in the outer membrane, making intermembrane space proteins accessible to cleavage by exogenously added proteases such as trypsin. Higher concentrations of digitonin also make holes in the inner membrane, thereby also exposing matrix proteins to protease digestion. Solubilization of the outer membrane is detectable by blotting for cytochrome c and TIM23, which shows protease digestion, even at the lowest digitonin concentration (Fig. 2B). The Mgm1/OPA1 protein is also digested at the lower digitonin concentrations, thereby indicating that Mgm1/OPA1 is exposed to the intermembrane space. The matrix proteins, glutamate dehydrogenase, HSP60, and the ␣-subunit of ATP synthase require relatively high digitonin concentrations before they become accessible to protease digestion (Fig. 2B). Because the antibody with which we detect the presence of the Mgm1/OPA1 protein was raised against the C-terminal half of this protein, we conclude that at least this half of the protein is exposed to the intermembrane space.
There is no obvious transmembrane segment in the sequence of Mgm1/OPA1, suggesting that this protein is a peripheral membrane protein even though it could not be removed with 250 mM KCl. To verify that Mgm1/OPA1 is indeed a peripheral membrane protein, we subjected mitochondria to carbonate washes. This procedure extracts peripheral proteins from the membrane, whereas integral proteins remain insoluble (30). Upon extraction of mitochondrial membranes with sodium carbonate pH 13, Mgm1/OPA1 was successfully solubilized along with the peripheral ␣-subunit of the ATP synthase complex, whereas the integral membrane protein TIM23 remained in the pellet (Fig. 2C). These results indicate that Mgm1 is peripherally associated with the mitochondrial inner membrane but that it binds more tightly to the membrane than cytochrome c, which can be released by 250 mM KCl. We conclude that Mgm1/OPA1 is a peripheral membrane protein, tightly bound to the mitochondrial inner membrane on the side facing the intermembrane space.
Cryo-immunogold electron microscopy was used to substantiate the submitochondrial localization of the Mgm1/OPA1 protein. Examples are shown in Fig. 3. Most gold particles label the interior of mitochondria, but some gold particles are on or near the rim (92 and 8%, respectively; n ϭ 768), similar to a previously published report on Mgm1/OPA1 distribution (21). To further localize Mgm1, we counted the number of gold particles on either side of the inner membrane. We found that 39% of the gold particles that could be assigned to one side or the other of the inner membrane were in the intermembrane space, while 61% were on the matrix side of the inner membrane (n ϭ 256). Those numbers were compared with numbers obtained with antibodies against proteins with known submitochondrial distributions. The intermembrane space protein cytochrome c gave 45 and 55%, respectively (n ϭ 200), the mitochondrial matrix protein GDH gave 22 and 78%, respectively (n ϭ 89), and a peripheral membrane subunit of ATP synthase gave 21 and 79%, respectively (n ϭ 46). The apparent displacement of gold particles labeling proteins that are unquestionably localized to one side or the other of the cristae membranes may be due in part to the added lengths of the primary and secondary antibodies, which can displace the gold particles by as much as 20 nm, and it could be due in part to the strong curvature of cristae membranes, which would make them more difficult to detect in thin sections. We can, nevertheless, conclude that the Mgm1/OPA1 protein is distributed throughout the interior of mitochondria in close association with the inner membrane. The distribution of gold particles obtained with Mgm1 is most similar to that obtained with cytochrome c antibodies, suggesting that Mgm1/OPA1 and cytochrome c have similar submitochondrial distributions.
Overexpression of Mgm1/OPA1 Protein in Mammalian Cell Lines-The effects of Mgm1 overexpression were investigated with a transient transfection construct encoding wild type Mgm1. When expressed in HeLa cells (Fig. 4) or Vero and COS7 cells (data not shown), the protein encoded by this construct was targeted to mitochondria. Staining with MitoTracker showed that Mgm1 overexpression caused the mitochondria to form relatively small fragments that were dispersed throughout the cytoplasm (76% of transfected cells; Fig. 4A) or clustered in the perinuclear region (24% of transfected cells; Fig. 4B). The formation of perinuclear clusters is likely caused by high levels of overexpression, as is also seen when certain other mitochondrial proteins are overexpressed (14). Disruption of mitochondrial morphology was already detected at 10 h after transfection, suggesting that precise expression levels of Mgm1 are critical for normal mitochondrial function.
To determine the effects of the mutant Mgm1/OPA1 protein on mitochondrial morphology, we introduced a point mutation in the G1 consensus motif of the GTPase domain (K301A) that is similar to the mutations used previously to analyze the functions of other dynamin family members. A construct encoding the Mgm1(K301A) mutant was transfected into HeLa cells, and its effect on mitochondrial morphology was detected with MitoTracker. Overexpression of the Mgm1(K301A) mutant caused mitochondria to fragment and aggregate (Fig. 4C), similar to the effects of overexpressing wild type protein. There were some subtle differences between mitochondria in cells that overexpress wild type or mutant protein. The fragmented mitochondria in cells with mutant protein had diameters larger than those obtained with the overexpression of the wild type protein, sometimes showing a crescent shape or hollow vesicle with MitoTracker staining (Fig. 4C).
Because the differences between cells that overexpress mutant or wild type Mgm1 were subtle, we needed to ensure that the K301A mutation indeed affects GTP hydrolysis, as was previously shown with the GTPase domain of dynamin. Therefore, we measured the ability of bacterially expressed GTPase domains to hydrolyze GTP in vitro. As shown in Fig. 4D, the mutant form of the Mgm1 GTPase domain is unable to hydrolyze GTP, unlike the wild type GTPase domain, which is highly active. The rate of hydrolysis as deduced from these experiments was 0.6 mol/mol/min, which is much higher than that of small GTPases such as Ras, but similar to rates that were previously reported for other members of the dynamin family (31,32). We conclude that Mgm1(K301A) has severely disrupted GTPase activity, even though the mutant phenotype in transfected cells is not all that different from the phenotype observed in cells overexpressing wild type Mgm1. It would thus appear that mitochondria are sensitive to the levels of Mgm1 expression regardless of whether the overexpressed protein hydrolyzes GTP.
To determine the effects of Mgm1/OPA1 loss of function on mitochondrial morphology, endogenous protein was depleted by transfecting HeLa cells with siRNA (28). The effects of Mgm1/OPA1 depletion on mitochondrial morphology were assessed by staining the transfected cells at selected time points with MitoTracker. At 60 h, almost all transfected cells (70 -90% cells in the dish) contained mitochondria that were fragmented and dispersed throughout the cytosol (Fig. 5B). Unlike the fragmented mitochondria in cells transfected with mutant Mgm1/OPA1, the mitochondria in cells transfected with Mgm1/ OPA1 siRNA were never clustered in the perinuclear region, suggesting that mitochondrial clustering is a side effect of protein overexpression, as was also observed previously with other mitochondrial proteins (14). When the transfected cells were stained at an earlier time point (generally at 42 h after transfection), they already showed marked changes in mitochondrial morphology, but very few of those cells contained fragmented mitochondria. Many transfected cells contained mitochondria with periodic constrictions along their lengths, giving them the appearance of beads on a string (Fig. 5, C-F).
Other cells contained mitochondria that were abnormally thick (Fig. 5, G and H). The mitochondria in these cells had an average diameter of 0.95 Ϯ 0.04 m (n ϭ 26), compared with 0.40 Ϯ 0.09 m (n ϭ 38) in wild type cells (Fig. 5A).
To determine the possible effects of Mgm1/OPA1 depletion on the morphology of the mitochondrial inner membrane, we processed siRNA-transfected cells for transmission electron microscopy. The most striking abnormalities were observed in cells fixed 3 days after transfection, at which time the level of Mgm1/OPA1 protein was reduced by 86% (Fig. 6D), and the mitochondria are fragmented. Electron microscopy showed that some of these mitochondria had reticular cristae (Fig. 6A), some had enlarged cristae (Fig. 6B), and some had cristae that were oriented parallel to the longitudinal axis of the mitochondria (Fig. 6C). Other mitochondria had cristae that formed curved or ring-shaped structures within the mitochondrial matrix or appeared to lack cristae altogether, although those organelles were still recognizable as mitochondria because of their double membrane (data not shown). A few of the transfected cells also had dilated ER, but this is likely a secondary consequence of the observed mitochondrial abnormalities, because Mgm1/OPA1 is clearly a mitochondrial protein.
Although we could not rule out subtle changes in inner membrane morphology, gross abnormalities of cristae structures were not observed in cells that were fixed at earlier time points (data not shown), suggesting that the thickening and stretching of the mitochondria observed with light microscopy is one of the earliest effects of Mgm1 depletion. The rearrangement of cristae and fragmentation of mitochondria occur at a later stage of Mgm1/OPA1 depletion, suggesting that these are secondary defects that might be induced by disruption of the mitochondrial inner membrane, since that is where the Mgm1/ OPA1 protein is localized.

DISCUSSION
Our results are consistent with the results of other groups that have worked with Mgm1/OPA1 in yeast (15,16,18,19) and mammalian cells (22,33,34). We confirmed that the Mgm1/OPA1 protein is a mitochondrial intermembrane space protein and that overexpression of wild type or mutant Mgm1/ OPA1 both cause mitochondria to become fragmented and form perinuclear clusters. It is surprising that there are only subtle differences between cells that overexpress mutant or wild type Mgm1/OPA1, even though we show here that GTP hydrolysis is severely impaired by the K301A mutation. These findings suggest that elevated levels of Mgm1, rather than a non-functional GTPase, cause mitochondria fragmentation. Depletion of Mgm1/OPA1 protein by siRNA also produces mitochondrial fragments, but those are dispersed, suggesting that the clus-tering observed in cells with overexpressed protein is an artifact of overexpression. Because mitochondrial fragments are observed with all three manipulations, the formation of fragments is likely to more closely reflect the primary function of Mgm1/OPA1.
As noted previously, mitochondrial fragmentation suggests that the balance between mitochondrial fission and fusion is shifted toward fission in cells that lack functional Mgm1 (17). The phenotype of the Mgm1 mutants thereby resembles the phenotype of another mitochondrial protein implicated in mitochondrial fusion, the fuzzy onions protein in Drosophila (24) and its homologues Fzo1p in yeast (23) and mitofusins 1 and 2 in mammals (25). Indeed, the mitochondria of yeast cells with a strong loss of Mgm1 function are incapable of fusion, even when the mitochondrial fragmentation phenotype is suppressed by mutations in the yeast mitochondrial division protein Dnm1p (18). It was therefore postulated that Mgm1/OPA1 is required for fusion between mitochondria (18,19).
Our results suggest that the Mgm1/OPA1 protein might also affect mitochondria in other ways. We observed gross morphological changes in the shape of mitochondria well before the onset of mitochondrial fragmentation. These morphological changes consist of swelling and stretching, similar to the changes in mitochondrial morphology that were observed previously in cells about to undergo apoptosis (35). At a later stage, the mitochondria of cells transfected with Mgm1/OPA1 siRNA form constrictions, arranged like beads on a string, which may then lead to their fragmentation. A similar sequence of events may also occur in cells undergoing apoptosis, but in those cells the different stages leading to mitochondrial fragmentation occur in rapid succession (35), whereas in our experiments they occur much more slowly as a result of the gradual depletion of Mgm1/OPA1 protein. Moreover, the majority of Mgm1/OPA1deficient cells do not undergo apoptosis, even though their mitochondria are at least as fragmented as those of apoptotic cells. 2 Although Mgm1/OPA1 deficient cells might not spontaneously undergo apoptosis, they are, nevertheless, more susceptible to apoptosis-inducing triggers (33). The external trigger for apoptosis in patients with dominant optic atrophy could be the reactive oxygen species produced by light. Such synergy might then help explain why the optic nerve is specifically affected in patients with dominant optic atrophy.
Coincident with mitochondrial fragmentation but well after the initial changes in external morphology were observed, we saw a variety of changes in cristae morphology. The cristae in some cells were reoriented such that they ran parallel to the longitudinal axis of the mitochondria, whereas in other cells the cristae were reticular. Similar longitudinal and reticular cristae were reported previously for various other pathological conditions, for example in patients treated with chloramphenicol (36), patients with mitochondrial encephalomyopathy (37), rats treated with a hypocholesterolemic agent (38), and in certain mutant mice (39). Besides longitudinal and reticular cristae, we also observed mitochondria with swollen cristae and a condensed matrix. Condensed matrices were previously observed in chloramphenicol-treated cells (40), anoxic cells (41), and lymphoblasts from a patient with Leber's disease. 2 Patients with Leber's disease show progressive degeneration of their optic nerves similar to the pathology of dominant atrophy. It is therefore tempting to speculate that Leber's disease and dominant optic atrophy reflect disruptions of the same cellular process. Leber's disease is caused by mutations in mitochondrial DNA affecting OxPhos complex I (42). It is possible that the mitochondria of patients with Leber's disease and dominant optic atrophy are both primed for apoptosis, for example, by producing elevated levels of reactive oxygen species. The Mgm1/OPA1 protein might somehow affect reactive oxygen species production by affecting cristae morphology, or it might act more directly on OxPhos, which then indirectly affects cristae morphology. Interestingly, cells undergoing apoptosis also show dramatic rearrangements of their cristae (43), but the precise relationship between Mgm1/OPA1and the early stages of apoptosis remains to be clarified.
To learn the cellular function of Mgm1/OPA1, we also need to know the precise localization of the Mgm1 protein. The initial reports on Mgm1 localization suggested that Mgm1 is localized to the mitochondrial outer membrane or the matrix. More recent data indicates that Mgm1 is localized to the mitochondrial intermembrane space. The results that we obtained with immuno-electron microscopy and protease sensitivity corroborate localization to the mitochondrial intermembrane space. Previous experiments with yeast showed that Mgm1 is peripherally bound to the inner membrane (17), although some of it appears to be peripherally bound to the outer membrane (19). However, mammalian Mgm1/OPA1 remained membrane bound, even with carbonate washes at pH 11 (21). In our hands, however, this tight association can be disrupted by raising the pH of the carbonate wash to 13, as was also used previously to analyze the signal recognition particle (SRP) receptor (30). We conclude that Mgm1/OPA1 is not an integral membrane protein but is more tightly bound to the membrane than cytochrome c, because the latter can be readily removed with 250 mM KCl, consistent with the previously reported internal distribution (18). We also conclude that Mgm1/OPA1 is bound to the inner membrane, because the gold particles used for immuno-electron microscopy were distributed throughout the mitochondria rather than strictly localized to the outer rims of mitochondria, as would be expected for a protein associated with the outer membrane. Taken together, our results indicate that the Mgm1/OPA1 protein is a peripheral membrane protein that is tightly bound to the outer leaflet of the mitochondrial inner membrane.
What might the cellular function of Mgm1/OPA1 be? Localization to the mitochondrial intermembrane space is consistent with both a role in fusion between mitochondria and a role in organizing cristae, as was suggested previously for yeast Mgm1 (17). The distribution of Mgm1/OPA1 throughout mitochondria, as shown by immuno-electron microscopy, is also suggestive of a role in organizing cristae. However, this distribution does not rule out a role in fusion, because the bulk of the Mgm1/OPA1 protein need not be active at any given time. Furthermore, the disruption of cristae morphology only occurs at later time points in our siRNA experiments, whereas in yeast the disruption of cristae morphology by mutant Mgm1p is suppressed by mutations in Dnm1, suggesting that Mgm1 is not required for normal cristae morphology (19). If, on the other hand, Mgm1 is directly involved in mitochondrial fusion, then only a very small amount of Mgm1/OPA1 may be active at any given time, because fusion is sporadic and occurs in localized spots along the rims of mitochondria. To investigate this issue, we will need to know whether Mgm1 exists in an active complex and an inactive pool, analogous to the membrane-bound and cytosolic fractions of dynamin and Drp1. In either case, Mgm1/OPA1 is likely to use a mechanism that is very different from that of other dynamin family members, because neither of these processes utilizes constrictions with the topology expected of a dynamin family member.