|Home | About | Journals | Submit | Contact Us | Français|
Fusion controls mitochondrial morphology and is important for normal mitochondrial function, including roles in respiration, development, and apoptosis. Key components of the mitochondrial fusion machinery have been identified, allowing an initial dissection of its molecular mechanism. Outer and inner membrane fusion events are coordinately coupled but are mechanistically distinct. Mitofusins are mitochondrial GTPases that likely mediate outer membrane fusion. The dynamin-related protein OPA1/Mgm1p is required for inner membrane fusion and maintenance of normal cristae structure. We highlight recent findings that have advanced our understanding of the mechanism, function, and regulation of mitochondrial fusion.
Mitochondria are essential organelles in most eukaryotic cells. Their most well-known biochemical functions are in intermediary metabolism and respiration, which result in the generation of adenosine triphosphate (ATP) through oxidative phosphorylation. In addition, they play important roles in apoptosis, cell signaling, iron metabolism, and steroidogenesis.
Recent studies indicate that the functions of mitochondria are coordinated with their dynamic behavior. Mitochondria frequently fuse and divide , and the balance of these processes determines overall mitochondrial morphology. When mitochondrial fusion is reduced, mitochondria fragment due to ongoing fission; conversely, mitochondria are long and overly interconnected when this balance shifts towards fusion. Mitochondrial fusion plays important roles in cell growth and development. Disturbance in mitochondrial fusion results in neurodegenerative diseases, such as Charcot-Marie-Tooth disease and autosomal dominant optic atrophy [2–4].
Over the last several years, key components of the mitochondrial fusion machinery have been identified. Mammalian cells require three mitochondrial GTPases -- the mitofusins (Mfn1 and Mfn2) and OPA1-- for mitochondrial fusion. In this review, we discuss recent advances in understanding how these molecules regulate mitochondrial fusion. The topic of mitochondrial fission has also seen rapid progress and is covered in other reviews [5,6].
The first player identified in mitochondrial fusion was the fuzzy onions gene (Fzo), which was isolated from a screen for genes involved in Drosophila spermatogenesis . Homologous proteins with similar function in mammals and yeast were later identified as mitofusins and Fzo1p, respectively [8–10]. There are two mammalian mitofusins, Mfn1 and Mfn2, which share about 60% sequence identity. The most obvious sequence feature among all of those proteins is an N-terminal GTPase domain, including canonical G1–G4 motifs (Figure 1). A bipartite transmembrane domain is located near the C-terminus, which spans the mitochondrial outer membrane twice. As a result, both N- and C-terminal portions of the proteins are oriented toward cytosol [11,12]. Two hydrophobic heptad repeat regions, HR1 and HR2, flank the transmembrane domain. Hydrophobic heptad repeats are sequence motifs in which apolar residues occur at the first and fourth positions of a seven-residue tandem repeat, and as discussed below, form helical coiled-coil structures. Yeast Fzo1p processes a third heptad repeat region (HRN) N-terminal of the GTPase domain.
Several lines of evidence demonstrate that mitofusins and Fzo1p are essential for mitochondrial fusion. In fzo1Δ yeast cells, mitochondria are highly fragmented due to ongoing mitochondrial fission [8,13]. Yeast carrying temperature-sensitive fzo1 alleles are deficient for mitochondrial fusion , a defect the secondarily leads to loss of mitochondrial DNA and therefore respiratory activity. Similarly, Mfn1-null or Mfn2-null mouse embryonic fibroblast (MEF) cells show predominantly fragmented mitochondria and have greatly reduced mitochondrial fusion in vivo [14,15].
In mouse, mitofusins are clearly important for early embryonic development. Loss of either Mfn1 or Mfn2 causes early embryonic lethality . Mfn2 mutant embryos die in midgestation due to impaired trophoblast giant cells, polyploid cells in the placenta that are critical for maintaining maternal supplies to the embryo. At least part of this requirement is likely due to the role of mitochondrial fusion in maintaining respiratory activity of mitochondria. MEFs lacking both Mfn1 and Mfn2 completely lose mitochondrial fusion activity, resulting in not only severely fragmented mitochondrial morphology, but also reduced respiratory capacity and slow cell growth .
Loss of Mfn1 or Mfn2 leads to distinct phenotypes, at both the cellular and whole animal levels. Mfn1-null MEFs display a pattern of mitochondrial fragmentation distinct from that observed in Mfn2-null MEFs [14,15]. More than 95% of MEFs lacking Mfn1 contain severely fragmented mitochondria that are very short tubules or small uniform spheres with diameters no larger than that of normal tubules. MEF cells lacking Mfn2 also display fragmented mitochondria, but such fragments vary greatly in size. The diameters of some fragments are several times larger than that of wild-type tubules. At the whole animal level, mice lacking Mfn1 or Mfn2 both die in midgestation. However, Mfn2-null placenta shows a disrupted trophoblast giant cell layer, a defect not observed in Mfn1-null embryos .
Why do mammals contain two mitofusin homologs? Mfn1 and Mfn2 share more than 70% sequence similarity and appear to have partially redundant activities. MEFs lacking either Mfn1 or Mfn2 alone retain some mitochondrial fusion activity, indicating that endogenous levels of a single mitofusin homolog can support low rates of fusion [14,15]. Tubular mitochondrial morphology can be restored in Mfn1-null cells by overexpression of Mfn2. Similarly, overexpression of Mfn1 in Mfn2-null cells can restore wild-type mitochondrial morphology. In addition, the morphology defects of double-null cells can be rescued by overexpression of either mitofusin . These results indicate that Mfn1 and Mfn2 have, at least partially, interchangeable activities for mitochondrial fusion. Based on this similarity of biochemical function, a simple explanation for the cell-type-specific phenotypes observed in Mfn1 mutant versus Mfn2 mutant mice is that the phenotypes reflect differences in their expression levels in different tissues. Alternatively, these observations could reflect biochemical differences between the two mitofusins, an idea supported by several observations. First, overexpression of OPA1, a dynamin-related protein on mitochondrial inner membrane, induces mitochondrial tubulation and fusion in MEF cells lacking Mfn2, whereas no such effect is found in Mfn1-null MEF cells . Second, in an in vitro mitochondrial tethering assay, Mfn1 shows more tethering activity than Mfn2 .
Emerging evidence suggests a connection between mitochondrial fusion and protection from apoptosis. During apoptosis, mitochondria undergo rapid fragmentation, a process dependent on the mitochondrial fission proteins Fis1 and Drp1 . During the remodeling of mitochondria and before the activation of caspases, Bax and Bak, two pro-apoptotic members of the Bcl-2 family, concentrate at foci on the mitochondrial outer membrane, which subsequently become mitochondrial fission sites . Interestingly, Drp1 and Mfn2 colocalize at these puncta upon induction of apoptosis. Down-regulation of mitochondrial fission proteins can inhibit apoptosis [20,21]. Conversely, overexpression of mitofusins or OPA1 increases the resistance of cells to death stimuli, whereas their loss makes cells more vulnerable [22–24].
Because mitofusins are critical for maintenance of mitochondrial fusion, it is likely that their levels are carefully regulated in response to the cellular environment. In yeast, stringent regulation of Fzo1p turnover is important for maintaining normal mitochondrial morphology. At least two mechanistically distinct mechanisms regulate Fzo1p degradation, one operating under steady-state growth and the other one involved during mating [25,26]. In the first pathway, Fzo1p levels are controlled by Mdm30p, an F-box containing protein. Mdm30p is a component of the Skp1-Cdc53-F-box (SCF) E3 ubiquitin ligase complex, which ubiquitinates a variety of proteins and thereby targets them for degradation by the 26S proteasome. Δmdm30 cells show accumulation of Fzo1p, accompanied by fragmentation and aggregation of mitochondria [25,27]. Mdm30p binds to Fzo1p and requires the F-box motif and ATP for Fzo1p degradation. Surprisingly however, Mdm30p-mediated Fzo1p turnover does not involve the SCF ubiquitin ligase complex, ubiquitination, or the 26S proteasome . In contrast, Fzo1p degradation upon α-mating pheromone treatment does not require Mdm30p but is both ubiquitin- and proteasome-dependent .
Recent results suggest that the cell death molecules Bax and Bak regulate Mfn2 function in mammalian cells. During apoptosis, Bax and Bak form puncta on mitochondria, and colocalize with Mfn2 . Beyond their roles in apoptosis, Bax and Bak also regulate mitochondrial fusion through interactions with Mfn2 in healthy cells . Bax/Bak double knockout cells have fragmented mitochondria and reduced mitochondrial fusion activity. These mutant cells also show a change in Mfn2 localization on mitochondria. In wildtype cells, Mfn2 is localized both uniformly on the mitochondrial outer membrane and also in punctate spots. In Bax/Bak mutant cells, there is reduced punctate localization of Mfn2. As a caveat, it should be noted that punctate localization is not observed with yeast Fzo1p or with some other studies of mammalian Mfn2 [8,14]. Expression of Bax induces focal localization of Mfn2 on mitochondrial outer membrane, reduces Mfn2 membrane mobility, and increases assembly of Mfn2 into a 440 kDa-complex. Thus, there is evidence that Bax and Bak have a non-apoptotic role in promoting mitochondrial fusion through Mfn2, as well as an apoptotic role in promoting fission upon receiving appropriate stimuli.
It is likely that mitofusins collaborate with other proteins to mediate mitochondrial fusion. In addition to Bax, recent studies have identified a number of potential mitofusin-associated proteins, including mitofusin-binding protein (MIB) , stomatin-like protein 2 (Stoml2)  and, membrane-associated RING-CH-V (MARCH-V) . Further biochemical characterization of these proteins may lead to new insights into mitochondrial fusion.
Intensive studies of intracellular membrane trafficking have yielded important insights into how biological membranes fuse [32,33]. Appropriate cognate membranes are brought together through specific protein-protein interactions that occur in trans between the membranes. At least two levels of specificity are provided through the action of Rab GTPases with Rab effectors and the formation of SNARE (soluble N-ethylmaleimide-sensitive factor attachment protein receptors) complexes. Both sets of proteins are localized to specific membrane compartments, and specific interactions in trans ensure that only the correct pairs of membranes can fuse. Rab GTPases are Ras-like small regulatory GTPases . They act as molecular switches by cycling between an active GTP-bound form and an inactive GDP-bound form. In the GTP-bound form, vesicle-associated Rabs are capable of interacting with Rab effectors on target membranes to form molecular bridges between the membranes. Such interactions lead to membrane tethering, in which the membranes are brought together but are not directly apposed [34,35]. Once cognate membranes are tethered, SNARE complexes form in trans to provide a second level of specificity and also to bring the membranes into direct contact [33,36]. SNAREs are a family of membrane-associated proteins containing a characteristic 60–70 residue SNARE motif composed of heptad repeats. During the process of fusion, an R-SNARE on the vesicle pairs with Q-SNAREs on the target membranes in trans. Such SNARE complexes form highly stable parallel four-helix bundles that pull the two bilayers into contact, a prerequisite for membrane fusion. Studies of vesicle trafficking, therefore, have shown that proteins interactions mediate membrane fusion by forming an early “tethered” state followed by a later “docked” state. Because mitochondrial fusion does not involve Rabs or SNAREs, it remains to be determined whether these features will be conserved.
Given that mitochondria are double-membraned organelles, fusion would first involve fusion of the outer membranes. Fzo1p and Ugo1p are the only outer membrane proteins identified in yeast as essential for mitochondrial fusion. No obvious ortholog of Ugo1p exists in mammals, leaving the mitofusins as the best candidates for mediating fusion of the mitochondrial outer membranes. As detailed above, mitofusins are large GTPases containing a GTPase domain, two or more hydrophobic heptad repeats regions, and a bipartite transmembrane domain (Figure 1). Mutations in any of these domains result in loss of activity [12,37–40].
Mitochondrial fusion requires mitofusins/Fzo1p on both mitochondria [38,41]. Mitochondrial fusion assay shows that mitochondria from Mfn double-null cells fail to fuse with those from wild type cells . In yeast, Fzo1p is required on both mitochondria as well in vitro . Therefore, mitofusins/Fzo1p complexes that support fusion are likely to function in trans.
Structural studies reveal that Mfn1 HR2 form a dimeric anti-parallel coiled coil that is 95 Å long . The anti-parallel orientation places the transmembrane segments at opposite ends of the helical bundle. Therefore, the outer membranes of adjacent mitochondria are tethered together, but a significant gap between the two membranes remains (Figure 2). This arrangement is quite different from the parallel SNARE helical bundles that draw two membranes into direct apposition for fusion. Therefore, the HR2 complex probably is functionally analogous to the Rab-based tethers that occur in SNARE-mediated vesicle fusion.
What role do the other regions of mitofusins play in mitochondrial fusion? The GTPase domain is required for fusion activity [7,8,14]. Expression of an Mfn1 construct lacking the GTPase domain results in tightly packed mitochondria that are separated by a distance comparable to the HR2 helical bundle . This structure may represent an intermediate that is trapped at tethered step and unable to proceed to membrane fusion. This observation suggests the GTPase domain may function downstream of mitochondrial tethering to mediate full fusion. In spite of intensive studies, the molecular mechanism of the GTPase domain during the fusion reaction remains unclear. Given its sequence similarity to dynamin GTPases, the GTPase domain may have a mechanochemical function similar to that of classical dynamins and depend on GTP hydrolysis to deform membranes. In vitro, fusion requires GTP hydrolysis, and non-hydrolysable GTP analogues inhibit fusion . Alternatively, the GTPase domain may function similarly to classical small G proteins. Depending on the nucleotide-bound state of the GTPase domain, it could regulate the activity of other domains or factors.
Inter-domain interactions exist within mitofusins and are crucial for fusion activity. In part, such interactions are important for the ability of mitofusin molecules to form oligomeric complexes. Fzo1p in yeast functions as a homo-oligomeric complex, whereas Mfn1 and Mfn2 form both homo-oligomeric and hetero-oligomeric complexes [14,15,37,40]. The intermolecular interactions mediating mitofusin oligomerization are multiple and complex; disruption of single domains typically do not disrupt complex formation . In addition, some of these domain interactions may be important for mediating steps in fusion beyond mitochondrial tethering. For example, the N-terminal region of Mfn2 including the GTPase domain and HR1 can interact with the C-terminal region, which includes HR2 [12,39].
MitoPLD, a mitochondrial member of phospholipase D superfamily, has recently been identified as an important regulator of mitochondrial fusion . Knockdown of MitoPLD significantly reduces mitochondrial fusion activity. When MitoPLD is over-expressed in NIH3T3 cells, mitochondria are aggregated but separated by a uniform gap. The distance of the gap is about 59 Å, about the half the size of the gap produced by HR2-mediated mitochondrial tethering. This unique mitochondrial phenotype requires mitofusins, whereas MitoPLD depletion does not affect tethering mediated by the HR2 region of Mfn1. These results suggests that MitoPLD functions downstream of the mitofusin-mediated tethering step . Although MitoPLD is homologous to bacterial cardiolipin synthase, it actually leads to hydrolysis of cardiolipin to generate phosphatidic acid. The role of phosphatidic acid in mitochondrial fusion is not well-understood, but this lipid promotes membrane fusion in certain types of exocytosis. These findings raise the intriguing possibility that divergent forms of membrane fusion have a conserved lipid requirement.
The yeast protein Mgm1p is a mitochondrial dynamin-related protein whose loss leads to mitochondrial fragmentation and mtDNA loss. These defects are associated with loss of mitochondrial fusion [43,44]. The mammalian ortholog, OPA1, is mutated in human autosomal dominant optic atrophy, a disease characterized by degeneration of the optic nerve [3,4]. OPA1 knock-down results in mitochondrial fragmentation and a loss of mitochondrial fusion [15,16]. These proteins have been biochemically localized to the mitochondrial intermembrane space and associated with the inner membrane [45,46]. Sequence analysis reveals that both OPA1 and Mgm1p are members of the dynamin-related protein family. They consist of an N-terminal mitochondrial targeting sequence (MTS), two consecutive hydrophobic segments, a coiled-coil domain, GTPase domain, middle domain, and a C-terminal coiled-coil domain that may correspond to GED domain (Figure 1). The pleckstrin homology and proline-rich domains, found in classical dynamins, are missing.
Both OPA1 and Mgm1p undergo extensive post-translational processing, but processing of the yeast isoform is less complex and better characterized. At steady state, Mgm1p exists as both long and short isoforms, l-Mgm1p and s-Mgm1p, respectively. Biochemical studies support a model of alternative topogenesis of Mgm1p . l-Mgm1p is produced by targeting of the precursor, which contains a mitochondrial targeting sequence (MTS), to the mitochondrial inner membrane and cleavage of the MTS by the mitochondrial processing pepidase (MPP). At the time of MTS cleavage, the precursor is presumably associated with the translocase of the inner membrane (TIM). Lateral exit from the TIM complex would generate l-Mgm1p, associated with the inner membrane through the first hydrophobic segment. Alternatively, if exit from the TIM complex were delayed, the Mgm1p precursor would be further pulled into the matrix, until the second hydrophobic segment is anchored in the inner membrane. At that point, the mitochondrial rhomboid protease, Rbd1p/Pcp1p, cleaves Mgm1p at a second site to produce s-Mgm1p [47,48]. Both l-Mgm1p and s-Mgm1p are necessary for mitochondrial fusion, and deletion of Rbd1/Pcp1 results in loss of fusion activity. Ups1p, a mitochondrial protein peripherally associated with the inner membrane, regulates the bifurcate sorting of Mgm1p .
In mammals, the diversity of OPA1 isoforms is much greater and their biogenesis is poorly understood. Alternative splicing generates at least eight mRNA isoforms of OPA1 . In addition, OPA1 is post-translationally processed through regulated proteolysis. The isoforms encoded by each of these RNA species can presumably be proteolytically processed, leading to many protein isoforms whose functions remain to be resolved. Identification of the protease responsible for OPA1 processing has been challenging (Figure 2). PARL (Presenilin associated rhomboid-like) has been an attractive candidate, because this mammalian ortholog can functionally replace Rbd1p/Pcp1p in yeast cells. Yeast two-hybrid and co-immunoprecipitation studies indicate that OPA1 physically interacts with PARL . PARL appears to be involved in cleavage of OPA1 into a soluble inter-membrane space form [23,51]. However, this form constitutes a minority of OPA1, and the bulk of OPA1 processing is unaffected in PARL-deficient cells. The m-AAA protease paraplegin also appears to play some role in OPA1 processing . Paraplegin is an ATP-dependent metallo-protease located in the mitochondrial inner membrane with its catalytic site exposed to the matrix. However, knockdown of paraplegin results in only modest effects on OPA1 processing. Therefore, the relative importance of specific proteases in OPA1 processing remains to be clarified. It is possible that multiple proteases can regulate OPA1 function depending on the cellular environment. Experiments with the uncoupling agent CCCP indicate, for example, that OPA1 processing is induced by loss of mitochondrial membrane potential .
Given its association with the mitochondrial inner membrane, Mgm1p is a good candidate for mediating mitochondrial inner membrane fusion. Recent results have provided strong support for this idea. In an in vitro mitochondrial fusion assay, mitochondria containing temperature-sensitive alleles of MGM1 are able to undergo outer membrane fusion but not full fusion . In vivo, yeast carrying a temperature-sensitive mgm1 allele show apparent fusion intermediates in which the outer membrane but not the inner membrane of mitochondria has fused. Additional observations suggest that Mgm1p might play a direct role in inner membrane fusion . Mgm1p molecules form complexes in trans between mitochondria, and such trans interactions are important for fusion activity. Because Mgm1p is a dynamin-related GTPase, it will be interesting to determine whether it uses GTP hydrolysis to modulate membrane curvature, as has been shown for classical dynamins.
In addition to its role in mitochondrial fusion, OPA1/Mgm1p is also important for maintaining normal cristae structure. Yeast cells containing Mgm1p temperature-sensitive alleles display decreased cristae number  and loss of Mgm1p results in swollen and poorly involuted cristae . Similarly, cultured mammalian cells lacking OPA1 have highly disorganized cristae .
OPA1 has been proposed to protect cells from apoptosis by restricting the diameter of cristae junctions (bottleneck structures between the cristae and the outer boundary membrane) and thereby preventing cytochrome c release . During some forms of apoptosis, opening of cristae junctions has been associated with cytochrome c release. Overexpression of OPA1 results in narrow cristae junctions and protection from cell death. This protective effect of OPA1 expression occurs even in Mfn1 and Mfn2 double-null cells, suggesting that it is independent of mitochondrial fusion . The anti-apoptotic effect of OPA1 has also been linked to the proteolytic activity of PARL . A small fraction of OPA1 is processed into a short soluble form in the intermembrane space, and this form is reduced in PARL-deficient cells. Oligomers of OPA1 include this soluble form, and such oligomers have been proposed to act with membrane-bound OPA1 to close cristae junctions .
Our understanding of Mfn/Fzo1p and OPA1/Mgm1p in mitochondrial fusion has advanced significantly in the last several years. Mitofusins/Fzo1p are the best candidates for mediating mitochondrial outer membrane fusion, whereas OPA1/Mgm1p is involved in inner membrane fusion and maintenance of normal cristae structure. With new cellular systems and an in vitro fusion assay in yeast, more progress can be expected in the future. Genetic screens in yeast have been very successful in identifying the core components of mitochondrial fusion. An important area of future research is the characterization of additional players and regulators of mitochondrial fusion (Figure 2). Some of these regulators will likely be critical in understanding how outer membrane fusion is coupled with inner membrane fusion. In addition, they will help to understand how mitochondrial dynamics is coupled to the cellular environment. An additional area of future research is the structural analysis of the core components, information that will be critical in elucidating molecular mechanisms. Considering the importance of mitochondrial fusion in mammalian development and neurodegenerative diseases, the answers to these questions not only are important in fundamental cell biology but also will ultimately benefit human health.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.