Mitochondrial fusion may regulate mitochondrial morphogenesis and underlie complementation between mitochondrial genomes in mammalian cells. The nuclear encoded mitochondrial proteins Mfn1 and Mfn2 are human homologues of the only known protein mediators of mitochondrial fusion, the Drosophila Fzo GTPase and Saccharomyces cerevisiae yFzo1p. Although the Mfn1 and Mfn2 genes were broadly expressed, the two genes showed different levels of mRNA expression in different tissues. Two Mfn1 transcripts were detected at similar levels in a variety of human tissues and were dramatically elevated in heart, while Mfn2 mRNA was abundantly expressed in heart and muscle tissue but present only at low levels in many other tissues. Human Mfn1 protein localized to mitochondria and participated in a high molecular weight, detergent extractable protein complex. Forced expression of Mfn1 in cultured cells caused formation of characteristic networks of mitochondria. Introduction of a point mutation in the conserved G1 region of the predicted GTPase domain (Mfn1K88T) dramatically decreased formation of mitochondrial networks upon Mfn1 overexpression, suggesting that network formation required completion of the Mfn1 GTPase cycle. Conversely, a protein variant carrying a point mutation in the G2 motif of the Mfn1 GTPase domain acted as a dominant negative: overexpression of Mfn1T109A resulted in fragmentation of mitochondria. We propose that Mfn1T109A interferes with fusion activity of endogenous Mfn1 protein, possibly by binding necessary cofactors, so to allow unopposed mitochondrial fission. Thus, Mfn1 appears to be a key player in mediating mitochondrial fusion and morphology in mammalian cells.
Defects in mitochondria are thought to underlie a variety of degenerative disorders in humans. Myopathies, neuropathies, encephalomyopathies and many other diseases have been associated with mutations in the mitochondrial DNA (mtDNA). In many of these diseases onset of clinical symptoms is late, occurring as the ratio of mutant to wildtype mtDNA passes a critical threshold depending on energy demands of the particular tissue (Graff et al., 1999; Nakada et al., 2001a). Recent studies of the effect of mtDNA mutations on disease phenotypes in a transgenic mouse model (Inoue et al., 2000; Nakada et al., 2001a; Nakada et al., 2001b; Nakada et al., 2001c; Shoubridge, 2000) suggested an important protection mechanism against the onset of disease symptoms. In individual cells containing both wild-type and mutant mitochondria, immunoelectron microscopy revealed that most if not all of the mitochondria showed cytochrome c oxidase activity (Nakada et al., 2001c). This observation of complementation of mtDNA mutations in mice suggested that fusion between mitochondria with different genotypes may provide an important mechanism for protecting cells and tissues from expressing disease phenotypes caused by mitochondrial dysfunction based on mutations in the mtDNA. In addition, recent observations have suggested inter-mitochondrial complementation in cell culture (Enriquez et al., 2000; Ono et al., 2001). Although these findings are still controversial (Attardi et al., 2002), they have focused attention on molecular mechanisms that might mediate and regulate mitochondrial fusion in mammalian cells.
Mitochondria take on different characteristic morphologies in different cell types and under different conditions in the same cell type (Bereiter-Hahn and Voth, 1994; Griparic and van der Bliek, 2001). Although many examples of mitochondrial morphogenesis have been reported (reviewed in (Griparic and van der Bliek, 2001; Hermann et al., 1998; Shaw and Nunnari, 2002; van der Bliek, 2000; Westermann, 2002), the physiological roles of regulated changes in mitochondrial morphology are not yet understood. Formation of extensive mitochondrial networks has been proposed to be important for efficient intracellular energy transfer into different cell compartments (Skulachev, 2001). Mitochondrial morphology also changes during the process of apoptosis and remodeling of the mitochondrial cristae has been proposed to mobilize cytochrome c release during apoptosis (Scorrano et al., 2002). In addition, the mitochondrial fission protein Drp1 participates in regulating apoptosis by controlling mitochondrial fragmentation (Frank et al., 2001).
In Drosophila and yeast, mitochondrial fusion is mediated by the nuclear encoded mitochondrial GTPase fuzzy onions (fzo) (Hales and Fuller, 1997; Hermann et al., 1998). The Drosophila Fzo protein was identified through its role in a developmentally regulated massive mitochondrial fusion event leading to the formation of a specialized mitochondrial structure (the Nebenkern) during spermatogenesis. Its homologue in Sacharomyces cerevisiae, yFzo1, is required for mitochondrial fusion after mating and also for the continuously ongoing mitochondrial fusion events that maintain the dynamic network of mitochondrial filaments in vegetatively growing yeast (Bleazard et al., 1999; Shaw and Nunnari, 2002; van der Bliek, 2000). The Drosophila and yeast fzo proteins are the founding members of a family of conserved, large, transmembrane GTPases that constitute the only known protein mediators of mitochondrial fusion. We previously described the identification of two human Fzo homologues, Mitofusin1 (Mfn1) and Mitofusin2 (Mfn2) (Santel and Fuller, 2001) and reported that overexpression of Mfn2 influences mitochondrial morphology in cultured mammalian cells.
Here we provide evidence that human Mfn1 controls mitochondrial morphology by regulating fusion of mitochondria. The human Mfn1 gene encodes a ubiquitously expressed protein associated with mitochondria. Overexpression of Mfn1 by transient transfection of cultured cells resulted in formation of characteristic networks of interconnected mitochondria. A point mutation in the conserved G1 motif of the GTPase domain (Mfn1K88T) dramatically decreased formation of mitochondrial networks, while a point mutation in the G2 motif of the GTPase domain (Mfn1T109A) resulted in mitochondrial fission upon overexpression in cultured cells, suggesting that the predicted GTPase activity of Mfn1 plays an important mechanistic role in the effects of Mfn1 on mitochondrial morphology. Expression of different ratios of wild-type Mfn1 protein versus the apparently dominant negative, fission promoting, Mfn1T109A GTPase-mutant variant in the same cell appeared to titrate the effects of Mfn1 overexpression on mitochondrial morphology, resembling the dynamic balance of fusion and fission in vegetatively growing yeast. Such ongoing fusion and fission of mitochondria in mammalian cells may provide a mechanism for functional complementation of mutants in mtDNA by ensuring mixing of protein and membrane components among mitochondria throughout the cell.
Materials and Methods
Cell culture and transfection
COS-7 cells were maintained in DMEM with high glucose supplemented with 10% FBS and antibiotics (100 U/ml penicillin and 100 μg/ml streptomycin). HL-60 (human promyelocytic leukemia) cells were cultured at 37°C in 5% CO2 in Iscove's medium supplemented with 20% fetal calf serum, 2 mM glutamine, nonessential amino acids, 2.5 μM sodium-pyruvate, 100 U/ml penicillin and 100 μg/ml streptomycin. Media and antibiotics were obtained from Biofluid (Rockville, MD). Cells were seeded the day before transfection in two or four chamber Lab-Tek slides and transfected using SuperFect (Qiagen) or Fugene reagent (Roche) according to the manufacturer's instructions.
Immunofluorescence and microscopy
About 15-20 hours after transfection, cells were fixed for 15 minutes at room temperature in 4% formaldehyde/PBS. All subsequent steps for indirect immunofluorecence staining were carried out as described previously (Santel and Fuller, 2001). MitoTracker Red (CMXRos; Molecular Probes) was added to cell medium at a concentration of 0.1 μM for 15 minutes. Electron microscopy and live imaging of transfected cells by confocal microscopy was carried out as described (Frank et al., 2001).
To generate anti-Mfn-1 antibodies, an EcoRI-PstI-fragment from pBSfzoH1 (human EST HFBDS57) was subcloned in frame into pGEX-KG. The 35 kDa GST-fusion protein was expressed in E. coli XL-1 Blue cells with IPTG-induction under standard conditions. The soluble GST-fusion protein (2 mg) was purified under native conditions according to the manufacturer's protocol and used for a standard immunization of two rabbits (BabCo/Covance, CA). Polyclonal antisera from both rabbits were affinity-purified against the GST-Mfn fusion protein coupled to AminoLink column (Pierce) and subsequently tested in western blot experiments. Both purified antibodies gave identical results when used in western blot experiments.
Polyclonal antibodies specific to Mfn2 protein were raised against an internal peptide sequence [(C)-KNSRRALMGYNDQVQ-RPIPLTPAN] by Zymed Laboratories (South San Francisco, CA) and affinity-purified against the immunogenic peptide. The affinity purified Mfn1- and Mfn2-antibodies were used at a concentration at 1:500-1000 in western blot experiments and 1:50 for immunofluorescence staining
Polyclonal rabbit anti-rTOM40 antibodies (Suzuki et al., 2000) were kindly provided by Hiroyuki Suzuki and Katsuyoshi Mihara (Kyushu University Fukuoka, Japan) and used at 1:1000. Monoclonal anti-cytochrome c (clone 6H2.B4) purchased from Pharmingen (San Diego, CA) was used at 1:50. Monoclonal anti-actin antibody obtained from Chemicon International (Temecula, CA) and mouse anti-porin monoclonal antibody (Calbiochem) were used according to the manufacturer's recommendation. The monoclonal anti-myc antibodies (clone 9E10; kindly provided by Alan J. Zhu, Stanford University, CA) was used 1:1000 and monoclonal anti-HA (BabCo/Covance) at a concentration of 1:500.
For expression constructs carrying the Mfn1 coding region, the Mfn1 ORF was amplified by PCR using primers A (5′-ATGGCAG-AACCGTTTCTCCAC-3′) and B (5′-CATGGTCACCAAAGC-AATC-3′), A-tailed by Taq polymerase and subcloned in pcDNA3.1V5/HISTOPO resulting in pcDNA3.1::Mfn1. The GFP-Mfn1 expression construct (pEGFPC2::Mfn1) (Santel and Fuller, 2001) was altered by site-directed mutagenesis to generate mutant expression constructs pEGFPC2::Mfn1(K88T) and pEGFPC2::Mfn1(T109A) using the Quickchange Mutagenesis kit (Stratagene) with appropriate PCR primers. Generation of introduced point mutations was confirmed by sequencing. For N-terminal in frame fusion of an V5/HIS-tag to the Mfn1 ORF, B-primer was replaced by primer C lacking the stop codon (5′-GGATTCTTCATTGCTTGAAGG-3′) for PCR amplification resulting in pcDNA3.1::Mfn1V5-HIS. The expression construct pEGFPC2::Mfn1(1-418) was generated by excising a SalI fragment from the original pEGFPC2::Mfn1 construct. GFP-Mfn1ΔGTPase was generated by PCR and subsequent in frame subcloning into pcDNA-NT-GFPTOPO (Invitrogen). PCR reactions were performed using proof-reading polymerase pfu (Perkin Elmer) or the ThermalAce DNA amplification kit (Invitrogen). The HA-tagged Drp1(K38A) expression construct was kindly provided by A. van der Bliek (UCLA) (Smirnova et al., 1998).
Protein extracts and western blot analysis
Proteins were extracted from mammalian tissue-culture cells by treating cells with NP-40 buffer (150 mM NaCl, 1% NP-40, 50 mM Tris-HCl pH 8.0) for 15 minutes at 4°C and subsequent centrifugation at 3000 g for an additional 10 minutes. Human protein extracts were purchased from Clontech (Protein Medley Kidney and Heart). For mitochondrial protein extracts and subcellular fractionation, mitochondria were isolated and fractionated following a previously described protocol (Spector et al., 1998). Protein extracts were separated in an 8% SDS-PAGE and then blotted onto Hybond ECL membrane. Membranes were blocked in 5% nonfat dried milk (NFDM) in PBT (PBS with 0.1% Tween-20) for 2 hours at room temperature and probed with the primary antibody in PBT+NFDM overnight at 4°C. After washing with PBT for 30 minutes at room temperature, samples were incubated with HRP-peroxidase-coupled secondary antibodies (anti-mouse from Roche, anti-rabbit from Amersham) for 2 hours at room temperature, followed by a 30 minute wash step at room temperature. The reaction products were visualized by enhanced chemiluminescence using ECLplus (Amersham) and BioMax films (Kodak).
Cells were spun down, washed once in Iscove's medium and resuspended at a cell density of 1×108/ml in Lysis buffer (20 mM HEPES, pH 7.4, 10 mM KCl, 1.5 mM MgCl2, 1 mM EDTA, 1 mM EGTA, 1 mM DTT, 250 mM sucrose and 0.1 mM phenylmethylsulfonyl fluoride). After a 30 minute incubation on ice, the lysate was Dounce-homogenized and spun at 900 g to pellet the nuclei. The postnuclear supernatant was then spun at 10,000 g to pellet the mitochondria. Isolated mitochondria (2 mg protein) were lysed in 200 μl digitonin buffer [1% (w/v) digitonin, 150 mM potassium acetate, 30 mM HEPES-KOH pH 7.4, 1 mM phenylmethylsulfonyl fluoride] for 30 minutes on ice. Following a clearing spin (30 minutes, 226,000 g, TLA45 rotor, Beckman TL-100 ultracentrifuge), the supernatant was applied to a Superose 6 gel-filtration column (25 ml column volume, Pharmacia) equilibrated with the same digitonin buffer. Fractions (0.5 ml) were collected, precipitated by adding trichloroacetic acid to a final concentration of 12.5% (w/v) and analyzed by SDS-PAGE and western blotting.
Northern blot analysis and RNA-chip analysis
Human Multiple-Tissue-Northern-/MTN-Northern blots were purchased from Clontech and hybridized with radiolabelled probes (32PC-labeling kit from Amersham) specific for Mfn1 and Mfn2. The RNA Chip contained 128 different human poly-A+ RNAs derived from a wide range of human tissues and cell lines, normalized to six house-keeping genes, and printed in duplicate spots onto chemically treated microscope slides (BD Biosciences-Clontech). Several negative control spots were added to each RNA Chip: Yeast total RNA, Yeast tRNA, E. coli rRNA, E. coli DNA, Poly r(A), Human Cot-1 DNA, Human genomic DNA. 32P-labeled DNA-probes (specific activity of the probe was >108 cpm/μg DNA) were used for hybridization according to the manufacturer's protocol. The RNA Chip was analyzed by phosphorimaging and autoradiography. The following probes were applied as positive control to assess equal RNA quantity and specificity: actin, liver-specific TAT (data not shown).
Mitofusins form a conserved class of mitochondrial GTPases
Searches of the human genome revealed the existence of four human genes that could encode proteins homologous to Drosophila fzo. Two appeared to be functional genes: the human Mfn1 gene (UniGene # Hs. 197877) located on chromosome 3 at 3q27.1 and human Mfn2 (UniGene # Hs. 3363) located on chromosome 1 at 1p36.21. The predicted proteins encoded by cDNAs corresponding to the human Mfn1 and Mfn2 genes were quite homologous, showing 60% identity and 77% similarity to each other. Moreover, both predicted proteins shared features characteristic of the conserved family of mitofusin proteins, including an N-terminal predicted GTPase domain with the four signature GTPase motifs predicted to form the nucleotide binding pocket (Fig. 1). When expressed in S. cerevisiae, neither Mfn1 nor Mfn2 rescued the glycerol growth defect phenotype of the fzoΔ1 mutation in the homologous yeast gene fzo1 (A. Mozdy, A. Santel, M. T. Fuller and J. M. Shaw, unpublished).
The other two human potential Mfn genes appeared to be pseudogenes. PsiMfn3 and psiMfn4 were found on chromosome X (Xq23-24) and chromosome 9, respectively. Although psiMfn3 and psiMfn4 were highly homologous to the Mfn1 gene both showed characteristic features of pseudogenes, including deletions and multiple stop codons throughout the predicted coding region, suggesting that they no longer encode proteins.
Mfns are widely expressed but show tissue-specific differences in expression level
Human Mfn1 and Mfn2 mRNAs were expressed at low levels in all tissues tested (Fig. 2). However, the levels of Mfn1 and Mfn2 mRNA expression varied among different tissues, with Mfn1 mRNA levels being increased in more tissue and cell types than Mfn2 mRNA. An Mfn1-cDNA probe detected transcripts of 6.4kb and 3.6kb in a variety of adult human tissues, including heart, pancreas, skeletal muscle, brain, liver, placenta, lung, and kidney. An Mfn2-specific probe detected a single 5.5kb mRNA in the same range of tissues. (Fig. 2A). Both Mfn1 and Mfn2 mRNA levels were high in heart compared to other tissues (Fig. 2A). In addition, Mfn2 mRNA was also high in skeletal muscle (Fig. 2A). In contrast, Mfn1 mRNA levels appeared slightly elevated in pancreas and liver, but not in skeletal muscle (Fig. 2A). As expression of the actin mRNA loading control was also elevated in heart and skeletal muscle in Northern blots, we analyzed relative levels of Mfn mRNA expression using an RNA-chip with mRNA from 128 tissues (including fetal tissues) normalized to six different housekeeping genes (Fig. 2B). The RNA-chip results confirmed that Mfn1 and Mfn2 transcripts are widely expressed at low level in most tissues. Relatively elevated levels of Mfn2 mRNA expression were detected in heart and skeletal muscle, as well as in tongue (Fig. 2B). Expression of both Mfn1 and Mfn2 was detected in fetal tissues, with Mfn1 mRNA more abundant than Mfn2 (Fig. 2B, box 4). The chip results and Northern blot analysis (data not shown) revealed that both Mfn1 and Mfn2 mRNAs were present in several cultured cell lines, indicating that both genes can be co-expressed in the same cell type. Strikingly, the level of Mfn1 mRNA was elevated in certain carcinoma derived cancer cell lines while the level of Mfn2 mRNA was not (Fig. 2B; for example, compare Mfn1 expression to Mfn2 mRNA in human HL-60 carcinoma cells). In addition, Mfn1 and Mfn2 mRNAs appeared to be differentially expressed in two Burkitt's lymphoma cell lines (Fig. 2B).
Differential expression of mitofusin proteins
Mfn1 protein was widely expressed, based on immunoblotting experiments using antibodies that recognized Mfn1 but not Mfn2 (Fig. 3; see Materials and Methods). The affinity-purified anti-Mfn1 antibodies recognized a single endogenous protein with an apparent size of ∼86 kDa in mammalian cell extracts (Fig. 3A), matching the size of 84 kDa predicted from the Mfn1 primary sequence. The anti-Mfn1 antibodies also recognized recombinant GFP-tagged Mfn1 fusion protein transiently expressed in mammalian cells, but did not crossreact with Mfn2 protein similarly expressed from an epitope-tagged construct (Mfn2-myc) in COS-7 cells (Fig. 3A, lower panel) or expressed in yeast under control of an inducible GAL1 promoter (Fig. 3C). In addition, the anti-Mfn1 antibodies recognized overexpressed Mfn1 but not Mfn2 in immunofluorescence analysis of cells transiently transfected with various Mfn1 and Mfn2 expression constructs (data not shown). Immunoblotting experiments with anti-Mfn1 revealed expression of endogenous Mfn1 protein in HeLa cells, human kidney and heart (Fig. 3B), as well as in mouse heart, liver, kidney, NIH3T3 fibroblasts, mouse C2C12 myoblasts, and differentiated myotubes, and rat clone9 cells (data not shown). Strikingly, the level of Mfn1 protein appeared to be only slightly elevated in human heart protein extracts compared to kidney, using the level of the mitochondrial protein TOM40 as a loading control. Thus, the high level of Mfn1 mRNA detected in heart tissues may not be reflected in correspondingly elevated levels of Mfn1 protein.
Expression of Mfn2 protein appeared more tissue-specific than Mfn1 protein, as suggested from the mRNA expression studies. An Mfn2-specific antibody was raised against a peptide from an internal region of Mfn2 that had relatively low homology to Mfn1 (Fig. 1, bracket). The anti-Mfn2 antibody detected recombinant Mfn2 protein expressed in yeast and a doublet of proteins migrating at an apparent size of approximately 86 kDa in extracts from human tissues (Fig. 3C). Mfn2 protein appeared to be present in significantly higher levels in human heart than human kidney, in contrast to Mfn1, which was expressed at more similar levels in these two tissues. As Mfn1 mRNA and protein appeared to be more generally expressed than Mfn2, we investigated the subcellular localization and possible role of the Mfn1 protein in controlling mitochondrial morphology in mammalian cells.
Mitofusin 1 is a mitochondrial protein
Mfn1 protein from both HeLa cells and mouse heart co-fractionated with mitochondrial markers during differential centrifugation and was either not detected or was greatly reduced in the cytosolic postmitochondrial supernatant (Fig. 4A). Mfn1 protein released from mitochondria by detergent extraction migrated in an apparent high molecular mass complex through a size exclusion gel filtration column. A mitochondrial preparation derived from human HL-60 cells was treated with 1% digitonin to solubilize membrane proteins, fractionated by gel filtration, and fractions were probed for Mfn1 by western blotting. Mfn1 immunoreactivity peaked in the fraction corresponding to a molecular mass of ∼350 kDa (Fig. 4B). Mfn1 was not detected in the monomeric fraction corresponding to 86 kDa.
Both GFP-tagged and untagged Mfn1 protein localized to mitochondria when expressed transiently in cultured cells, as demonstrated by counterstaining with Mitotracker or antibodies against the mitochondrial intermembrane space protein cytochrome c (Fig. 5A,B). In contrast, a C-terminal truncated GFP-Mfn11-418 fusion protein lacking the conserved transmembrane domain and predicted coiled-coil tail did not localize to mitochondria (data not shown), suggesting that Mfn1, like Mfn2, carries the mitochondrial targeting signal in the C-terminal half.
Overexpression of Mfn1 in mammalian tissue culture cells induced formation of mitochondrial networks
Overexpression of either untagged or GFP-tagged Mfn1-fusion protein resulted in formation of perinuclear clusters of mitochondria in COS-7 cells (Fig. 5), as well as in HeLa cells, rat clone9 cells, C2C12 myoblasts, and NIH3T3 fibroblasts (data not shown). This dramatic change in mitochondrial morphology appeared to correlate with substantially elevated levels of Mfn1 protein: anti-Mfn1 antibodies strongly stained the protein in transfected cells with perinuclear networks of mitochondria, while under the same conditions endogenous Mfn1 protein was not detected by standard immunofluorescence microscopy in presumably untransfected cells, which had well spread mitochondria (Fig. 5B). Staining with the dye Mitotracker Red (CMXRos) indicated that the clustered mitochondria in Mfn1 transfected cells still had an active membrane potential and therefore most likely unaffected respiration activity (Fig. 5A).
Detailed analysis of transfected cells by high magnification light microscopy (Fig. 5C-E) and ultrastructural analysis by electron microscopy (Fig. 5F) revealed a peculiar and characteristic structural arrangement in the perinuclear networks of mitochondria. Simultaneous visualization of GFP-Mfn1 and cytochrome c, a marker of the mitochondrial intermembrane space, revealed that GFP-Mfn1 was adjacent to, but did not colocalize with, cytochrome c in the structures. Instead, anti-cytochrome c stained grape-like clusters of spherical structures, while the GFP-Mfn1 both surrounded the cytochrome c stained spheres in a thin layer and appeared concentrated in the interstices between the tightly packed spheres (Fig. 5C-E). Similarly, GFP-Mfn1 appeared to circumscribe the mitochondrial matrix as indicated by the Mitotracker marker in transfected cells (data not shown). Strikingly, the cytochrome c positive spheres were smaller in size toward the center of the array.
Ultrastructural analysis by electron microscopy confirmed the peculiar structure of the mitochondrial networks (Fig. 5F). Overexpression of GFP-Mfn1 after transient transfection induced formation of tight clusters of mitochondria, with smaller mitochondria near the center surrounded by larger mitochondria with typically scant cristae. Immunogold-labeling using antibodies against GFP indicated that the GFP-Mfn1 was largely associated with the mitochondrial outer membrane (Fig. 5F), consistent with the appearance of GFP-Mfn1 in a thin layer surrounding the cytochrome c containing spheres observed by light microscopy (Fig. 5C). GFP-Mfn1 protein appeared most abundant at the interfaces between the tightly packed small mitochondria inside the clusters. The enlarged mitochondria around the periphery of the cluster were less densely stained by immunogold particles. We propose that overexpression of Mfn1 by transient transfection triggers aggregation of mitochondria into tight networks through close interactions between outer membranes, followed by outer membrane fusion events that result in formation of the enlarged mitochondria characteristically observed around the periphery of the grape-like clusters.
Formation of mitochondrial networks required a wild type Mfn1 GTPase-domain
To test whether formation of the grape-like mitochondrial networks (Fig. 5, Fig. 6B) was dependent on the predicted GTPase activity of the overexpressed Mfn1 protein, we expressed a mutant form of GFP-Mfn1 carrying a K to T substitution at amino acid residue 88 in the G1 motif of the signature GTPase domain (GFP-Mfn1K88T; Fig. 6A). Substitutions in the analogous residue in the Drosophila Fzo and yeast Fzo1p proteins blocked mitochondrial fusion activity (Hales and Fuller, 1997; Hermann et al., 1998). The Mfn1K88T GTPase mutant greatly reduced the formation of mitochondrial networks. The majority of Mfn1K88T expressing cells exhibited mitochondria with apparently normal morphology and distribution, compared with mitochondria in adjacent non-transfected cells (Fig. 6C). Whereas 93% of the cells expressing detectable levels of wild type Mfn1 after transient transfection exhibited perinuclear clusters of mitochondria (Fig. 6B), only 30-40% of the cells expressing Mfn1K88T showed collapsed mitochondrial aggregates. In addition, most of these aggregates that were observed appeared structurally different from those mitochondrial clusters induced by wild-type Mfn1 expression. In the perinuclear mitochondrial aggregates observed in cells overexpressing GFP-Mfn1K88T, the GFP-Mfn1 and the cytochrome c intermembrane space marker appeared well aligned at the light microscope level (Fig. 6D), in contrast to the cluster-of-grapes appearance characteristic of mitochondrial networks observed in cells transfected with wild type Mfn1 (Fig. 5). We suggest, that in the case of Mfn1K88T induced mitochondrial aggregates, mitochondria might be tightly packed rather than `hyperfused' as observed for Mfn1 induced mitochondrial clusters. The difference in structure of the mitochondrial clusters/aggregates for wild type versus mutant Mfn1 could be accounted for if individual mitochondria in cells overexpressing GFP-Mfn1K88T did not fuse into the large mitochondria observed at the periphery of the grape-like clusters in cells transfected with wild type Mfn1. Expression of a GFP-Mfn1 fusion construct lacking the entire GTPase domain (Mfn1Δaa1-248) caused formation of mitochondrial aggregates resembling those observed in cells expressing the Mfn1K88T mutant variant (Fig. 6E) in almost all transfected cells (>80%).
The GTPase mutant Mfn1T109A blocked endogenous fusion activity
Mutation of the predicted G2 motif of the GTPase domain by a T to A substitution at amino acid residue 109 (Mfn1T109A) produced a form of the Mfn1 protein that appeared to act as a dominant negative and block endogenous mitochondrial fusion when expressed at high levels in COS-7 cells (Fig. 7). GFP-Mfn1T109A was still targeted specifically to mitochondria (Fig. 7A). Counterstaining with MitoTracker in live cells expressing GFP-Mfn1T109A showed an active membrane potential (Fig. 7A). However, in more than 80% of the transfected cells, mitochondria were distributed as small punctate structures throughout the cytoplasm, in contrast to the normal filamentous morphology observed in non-transfected cells (Fig. 7A, lower cell) or cells exhibiting only very low expression of the GFP-Mfn1T109A construct (Fig. 7B, upper cell).
To test whether mitochondria might be fragmented due to interference with the endogenous activity of Mfn1, we shifted the balance of mutant versus wild type Mfn1 in the cells by co-transfecting with GFP-Mfn1T109A and wild type Mfn1. Cells doubly transfected with GFP-Mfn1T109A and wild type Mfn1 showed a range of mitochondrial morphologies. However, mitochondria in these cells often exhibited a more filamentous morphology (Fig. 6C) than mitochondria in cells transfected with GFP-Mfn1T109A alone. To test whether the different mitochondrial morphologies observed correlated with relative levels of wild type and Mfn1T109A mutant protein expressed in individual cells, we co-transfected COS-7 cells with GFP-Mfn1T109A and a construct driving expression of a HIS-tagged wild type Mfn1 (Mfn1-V5/HIS) fusion protein and compared relative levels of GFP-Mfn1T109A and His-tagged Mfn1 in doubly transfected cells. Cells expressing GFP-Mfn1T109A and relatively low levels of Mfn1-V5/HIS tended to have fragmented and/or lightly fused mitochondria (Fig. 7D, upper panels, double arrow: cell with low Mfn1-V5/HIS; lower panels, cell with no detectable Mfn1-V5/HIS). Cells showing relatively similar levels of GFP-Mfn1T109A and Mfn1-V5/HIS tended to have more normal, filamentous mitochondria. High levels of expression of Mfn1-V5/HIS appeared to override and block the negative interfering activity of GFP-Mfn1T109A, resulting in formation of mitochondrial networks resembling those observed in cells overexpressing wild-type Mfn1 alone (Fig. 7D, arrow). We propose that Mfn1T109A encodes a dominant negative form of the protein able to block mitochondrial fusion activity of endogenous Mfn1. If so, then overexpression of Mfn1T109A may cause fragmentation of mitochondria due to unopposed mitochondrial fission, much as loss of function mutations in the Saccharomyces cerevisiae mitofusin gene yfzo1 causes fragmentation of mitochondria in vegetatively growing yeast. To test, whether the observed mitochondrial fragmentation in Mfn1T109A expressing cells can be prevented by blocking opposed mitochondrial fission, we co-expressed GFP-Mfn1T109A with the dominant-negative version of the fission protein Drp1 (HA-tagged Drp1K38A) (Smirnova et al., 1998) transiently in COS-7 cells. In doubly-transfected cells, mitochondrial fusion rather than fission occurred (Fig. 8). Fragmented mitochondria were no longer observed indicating that mitochondrial fission was blocked. The expression of Drp1K38A interfered with the expression of GFP-Mfn1T109A resulting in altered mitochondrial morphology (Fig. 8). Clustered and aggregated mitochondria as well as filamentous and interconnected mitochondria were observed. In a few cases, mitochondria displayed an interconnected network of tubular GFP-positive structures. (Fig. 8, right panel).
Members of the evolutionary conserved class of mitochondrial Fzo-GTPases have been identified in yeast, C. elegans, Drosophila and human. Here, we described the expression and function of the generally expressed human homologue Mfn1. The human genome contains two different functional genes encoding members of the Fzo-GTPase family, Mfn1 and Mfn2, which differ in their expression patterns. Interestingly, Drosophila also has two differently expressed genes encoding Fzo-GTPase family members: the founding member of the family fzo and dmfn (Hales and Fuller, 1997; Hwa et al., 2002). The Drosophila fzo protein is expressed only at the end of male meiosis and in early spermatids, where it is required for a developmentally regulated, massive mitochondrial fusion event in differentiating male germ cells (Hales and Fuller, 1997). In contrast, the Drosophila dmfn gene is broadly expressed in many tissues and stages of development (Hwa et al., 2002), where it presumably is involved in maintaining the balance between mitochondrial fission and fusion in many cell types.
Our results indicate that at both the mRNA and the protein levels human Mfn1 may be more widely and abundantly expressed than the previously described Mfn2 isotype (Santel and Fuller, 2001), which was expressed mainly in heart and skeletal muscle. The differential expression we observed by Northern and dot blot disagrees with the conclusions drawn by Rojo et al., who observed similar levels of Mfn1 and Mfn2 mRNA expression by an RT-PCR assay (Rojo et al., 2002). Northern blot analysis revealed two different transcripts detected by an Mfn1-specific probe. Database searches uncovered cDNAs derived from the Mfn1 locus that may represent alternative splice variants encoding truncated forms of the Mfn1 protein (A.S. and M.T.F., unpublished). Interestingly, it has been reported that expression of a corresponding Mfn1-splice variant was specifically up-regulated in tumor tissues (Chung et al., 2001). Our investigation did not independently either confirm or rule out the existence of such mRNA or protein variants, especially in diseased tissues or cancer cells.
The Fzo-GTPase family members Drosophila Fzo and yeast Fzo1p have been shown to be key mediators of mitochondrial fusion (Hales and Fuller, 1997; Hermann and Shaw, 1998). We propose that the human Mitofusin-1 protein (Mfn1) represents the main mediator of mitochondrial fusion in many human cell types. This idea is in agreement with two recent reports on the characterization of Mitofusin function (Chen et al., 2003) (Legros et al., 2002). Loss-of-function analysis of the mouse orthologs clearly demonstrated a contribution of Mfns in mitochondrial fusion (Chen et al., 2003). Consistent with our model, high level overexpression of Mfn1 in cultured cells led to formation of characteristic grape-like perinuclear clusters of mitochondria containing many large mitochondria around the outer edge. Strikingly, immunofluorescence staining for cytochrome c showed that the large peripheral mitochondria in the clusters had highly enlarged regions of intermembrane space (Fig. 5E,F). The large mitochondria with abnormal internal structure in Mfn1 overexpressing cells could arise if Mfn1 protein mediates fusion of outer membranes but not the inner membranes of adjacent and closely apposed mitochondria. Action at the outer mitochondrial membrane is consistent with a number of biochemical studies indicating that Mfn family proteins are associated with the mitochondrial outer membrane (Fritz et al., 2001; Rapaport et al., 1998; Rojo et al., 2002). In addition, several lines of evidence support the prediction that Mfn family proteins have a characteristic bipartite transmembrane domain that passes through the outer membrane twice, leaving both the N-terminal GTPase domain and the C-terminal coiled-coil motif facing the cytoplasm (Fritz et al., 2001), where they would be available to interact with Mitofusins or other proteins displayed on the surface of adjacent mitochondria. Fusion of inner mitochondrial membranes to allow mixing of matrix components could either be mediated by different proteins or require additional components not overexpressed in our studies.
The ability of overexpressed human Mfn1 to cause mitochondrial fusion in mammalian tissue culture cells was dependent on signature GTPase motifs, as demonstrated for Drosophila Fzo and yeast Fzo1p. Mutation of the conserved residue K88 to T in the G1 GTPase motif of Mfn1 abolished ability of the overexpressed Mfn1K88T protein to induce formation of the characteristic mitochondrial networks and grape-like clusters containing enlarged mitochondria. The G1 motif (GxxxxGKS/T) is the conserved core of the P-loop, which in Ras (Bourne et al., 1991) and several other members of the GTPase superfamily forms a critical part of the nucleotide binding pocket interacting with the α and β phosphates of GTP or GDP. The three dimensional structure of the G1 region of Ras was found to be similar in the GTP versus the GDP bound forms (Krengel et al., 1990; Pai et al., 1989). The K residue is conserved throughout the GTPase superfamily, as well as in ATPase motor proteins such as myosins and kinesins (Vale, 1996). In Ras, theϵ -amino group of the corresponding lysine residue (K16) with its positive charges stabilizes GTP phosphates by forming ionic interactions with theβ - and γ-phosphates of the nucleotide (Maegley et al., 1996; Prive et al., 1992). The loss of the corresponding K residue in the Mfn1K88T mutant form might affect affinity of guanine nucleotide binding, and therefore result in reduced hydrolysis rates for GTPase activity. Our finding that the K88 to T mutation in human Mfn1 blocked the ability of overexpressed Mfn1 to induce formation of the characteristic enlarged mitochondria suggests that mitochondrial fusion by Mfn1 protein depends on the GTPase cycle and raises the possibility that mitochondrial fusion in the cell may be regulated by additional proteins, much as GTPase activating proteins (GAPs) and guanine nucleotide exchange factors (GEFs) regulate the action cycle of many GTPases.
In contrast, overexpression of a version of the Mfn1 protein mutated in a different GTPase subdomain (Mfn1T109A) had a strikingly different effect on mitochondrial morphology. Amino acid residue T109 of Mfn1 is located in the predicted G2 motif, a region that in Ras is part of a large loop that forms a roof over the GTP binding pocket. A threonine is conserved in an analogous position throughout the GTPase superfamily (Bourne et al., 1991). In the crystal structure of Ras complexed with the GTP analogue Gpp(NH)p, this conserved threonine contacts the gamma phosphate of the nucleotide and also binds the critical Mg2+ ion that coordinates the γ and β phosphates of GTP (Bourne et al., 1991). Comparison of the structures of RAS in the GTP vs. GDP bound forms revealed that the G2 loop undergoes a conformational change upon GTP binding, most likely due to a dramatic reorientation of the conserved threonine (Bourne et al., 1991).
We found that a form of human Mfn1 in which the critical G2 motif threonine was mutated (Mfn1T109A) acted as a dominant negative when overexpressed by transient transfection in tissue culture cells. Cells expressing Mfn1T109A displayed many small mitochondria, resembling the punctate mitochondria that rapidly accumulate after loss of function of the yeast Mfn, yeast Fzo1p (Bleazard et al., 1999; Hermann et al., 1998). Consideration of the activity cycle of GTPase family members suggests a model for why Mfn1T109A may interfere with the function of endogenous Mfn protein in the transfected cells. Many GTPases rely on interaction with a guanine exchange factor (GEF) protein to induce release of bound GDP, which must occur to allow GTP to enter the nucleotide binding pocket after each activity cycle. Analysis of the GTPase family member EF-Tu suggested that the protein passes through a transient empty state, lacking nucleotide, while still bound to its GEF partner EF-Ts (Romero et al., 1985). Binding of GTP then allows release of EF-Ts, possibly due in part to the dramatic change in conformation of the GTPase upon GTP binding. If a similar GTPase cycle involving a partner GEF operates as a critical part of Mfn function in mitochondrial fusion, then high levels of the mutant Mfn1T109A protein may act as a dominant negative by binding up and titrating out the GEF. Without the critical G2 motif threonine residue, the Mfn1T109A protein may not undergo the dramatic conformational change upon GTP binding needed to release the GEF for use by endogenous Mfn proteins in the cell. As a consequence, loss of mitochondrial fusion activity could allow unbalanced mitochondrial fission, leading to accumulation of small, punctiform mitochondria, as observed in yeast lacking function of yeast Fzo1p.
Our results suggest that normal mitochondrial morphology in mammalian cells is maintained by a dynamic balance of Mfn-mediated mitochondrial fusion and Drp1-mediated mitochondrial fission. Consistent with this model, co-expression of Drp1K38A, a dominant interfering form of the dynamin GTPase Drp1 implicated in mitochondrial fission (Smirnova et al., 2001; Smirnova et al., 1998) with human Mfn1 resulted in formation of interconnected filamtentous mitochondrial tubules of mitochondrial outer membrane (S. Frank, R. J. Youle, A. Santel and M. T. Fuller, unpublished) much as observed for Mfn2 in similar experiments (Santel and Fuller, 2001). In addition, co-expression of dominant negative Drp1 (Drp1K38A) with the Mfn1 mutant form Mfn1T109A resulted in an inhibition of mitochondrial fission, but leading to mitochondria with altered morphology similar to those observed in cells expressing Drp1K38A alone or co-expressed with wild-type Mfn1/Mfn2, respectively (Santel and Fuller, 2001). This observation resembles the situation in yeast dnm1Δ/fzo1-1 double mutants, where mitochondrial morphology was similar to those observed in dnm1Δ-cells (Bleazard et al., 1999). Interestingly, Chan and colleagues also observed altered mitochondrial morphology in Mfn-mutant mouse embryonic fibroblasts expressing Drp1K38A compared to restored mitochondrial morphology in Mfn-deficient cells rescued by epitope-tagged mitofusins (Chen et al., 2003).
Mfn1 migrated in a gel filtration column as part of a high molecular protein complex, which may represent either a mitochondrial fusion machine or Mfn1 protein complexed with regulatory GEFs or GAPs. In extracts from S. cerevisiae, the Fzo1p behaved in biochemical experiments as part of an 800 kDa protein complex (Rapaport et al., 1998) rather than the 350 kDa complex observed in human HL-60 cells. The nature of the protein complex and the other components are not yet known for either yeast or human cells. Presumably due to the tissue-specific expression of Mfn2 we could not detect Mfn2 as a component of the Mfn1-complex by gel filtration fractionation in HL-60 cell extracts. In addition, preliminary experiments with extracts from cultured cells transiently transfected with Mfn1 and Mfn2 did not uncover biochemical interaction between these two isoforms. It is possible that Mfn1 and Mfn2 reside in different protein complexes, even in tissues were expression of both proteins overlaps.
The human Mfn1 and Mfn2 proteins had subtly different behavior with respect to their effects on mitochondrial morphology after transient overexpression in cell culture. Overexpression of Mfn1 alone was sufficient to cause GTPase dependent fusion of mitochondria, resulting in the formation of large mitochondria around the periphery of the perinuclear mitochondrial clusters. In contrast, GTPase dependent effects of overexpression of Mfn2 were best revealed in cells also undergoing overexpression of the dominant interfering Drp1K38A form of the mitochondrial fission mediator Drp1 (Santel and Fuller, 2001). Conversely, overexpression of Mfn2 caused mitochondrial clustering that was dependent on the C-terminal coiled-coil domain of Mfn2 (Rojo et al., 2002; Santel and Fuller, 2001). It may be that the more tissue specific Mfn2 regulates or mediates a specialized form of mitochondrial fusion characteristic of only certain cell types. Alternatively, the Mfn2 protein may play a completely different or more specialized function related to mitochondrial morphology. Recently, it has been proposed that Mfn2 may be in involved in regulating BAX-mediated apoptosis (Karbowski et al., 2002).
We thank J. Shaw, J. Nunnari and members of the Fuller and Youle labs for helpful discussions and comments on the manuscript. A. Mozdy is gratefully acknowledged for helping with the yeast experiments. We also thank K. Suyama, A. J. Zhu, H. Suzuki, K. Mihara and A. van der Bliek for sharing antibodies and plasmids. This work was supported in part by Deutsche Forschungsgemeinschaft postdoctoral fellowship #Sa 803/1-1 to A.S., NIH grant HD29194 to M.T.F and a generous gift from MitoKor (San Diego, CA).
↵* Present address: atugen AG, Robert-Rössle-Str.10, 13125 Berlin, Germany
- Accepted March 14, 2003.
- © The Company of Biologists Limited 2003