Mitochondrial Fusion: the Machineries in and Out

Trends in Cell Biology OPEN ACCESS

Review Mitochondrial Fusion: The Machineries In and Out

Song Gao 1,2,* and Junjie Hu 3,*

Mitochondria are highly dynamic organelles that constantly undergo ﬁssion and Highlights fusion. Disruption of mitochondrial dynamics undermines their function and Crystal structures of truncated mitofusin causes several human diseases. The fusion of the outer (OMM) and inner mito- (MFN)1 and MFN2, the dynamin-like chondrial membranes (IMM) is mediated by two classes of dynamin-like protein fusogens of the outer mitochondrial membrane, reveal their structural kinship (DLP): mitofusin (MFN)/fuzzy onions 1 (Fzo1) and optic atrophy 1/mitochondria to bacterial dynamin-like protein (BDLP). genome maintenance 1 (OPA1/Mgm1). Given the lack of structural information Human MFN1 and MFN2 bear subtle on these fusogens, the molecular mechanisms underlying mitochondrial fusion differences that govern the distinct bio- remain unclear, even after 20 years. Here, we review recent advances in struc- chemical properties. tural studies of the mitochondrial fusion machinery, discuss their implication GTP-dependent dimerization and con- for DLPs, and summarize the pathogenic mechanisms of disease-causing muta- formational changes are key features of tions in mitochondrial fusion DLPs. MFNs in driving outer membrane fusion.

Short optic atrophy 1/short mitochon- The Mitochondrial Fusion Machinery dria genome maintenance 1 (s-OPA1/ Mitochondria are double-membrane organelles that confer various essential cellular functions, s-Mgm1) resembles fission dynamins including energy production, metabolism, apoptosis, and innate immunity [1]. They form a highly in 3D architecture, highlighting a mechanism of inner mitochondrial membrane fi dynamic network and constantly undergo cycles of fusion and ssion. The balance between (IMM) merging distinct from that of other fission and fusion has critical roles in maintaining mitochondrial homeostasis in response to known types of homotypic fusion. metabolic or environmental stresses, and is linked to cell division, apoptosis, and autophagy [2–5]. In particular, fusion promotes the capacity of oxidative phosphorylation and allows redistri- OPA1/Mgm1 uses multiple intermolecular assemblies to achieve either IMM bution of mitochondrial (mt)DNA between damaged and healthy mitochondria [6,7]. Disruptions fusion or cristae shaping. of mitochondrial dynamics are implicated in aging as well as in several human diseases, including neurodegenerative and metabolic disorders, and cancer [8,9]. Mitochondrial fusion is a two-step Structures of the mitochondrial fusion machinery provide important rules for Fzo fl MFN process; fusion of the OMM is mediated by Fzo1 in yeast, (see Glossary)in ies, and in comparing fusion DLPs with fission mammals, whereas the IMM is fused by Mgm1 in yeast and OPA1 in mammals [10,11]. DLPs.

MFN and Fzo1 were identiﬁed some 20 years ago [12–15]. Mammals have two MFNs, namely MFN1 and MFN2. Functionally, MFN1 and MFN2 share not only a certain amount of redundancy, 1State Key Laboratory of Oncology in but also substantial differences [15,16]. Mice with deletion of either MFN1 or MFN2 die in utero in South China, Collaborative Innovation Center for Cancer Medicine, Sun mid-gestation [14]. MFN2 mutation accounts for most cases of Charcot-Marie-Tooth disease Yat-sen University Cancer Center, type 2A (CMT2A), a neuromuscular disorder [17]. The yeast IMM fusogen Mgm1 was initially 510060 Guangzhou, China 2 identiﬁed as a key regulator of mtDNA maintenance [18,19]. The human OPA1 gene was mapped Guangzhou Regenerative Medicine and Health Guangdong Laboratory, 510530 in genetic studies of patients with autosomal dominant optic atrophy (ADOA) [20], a hereditary Guangzhou, China neurodegenerative disease. Around the same time as the initial characterization of Fzo1 and 3National Laboratory of MFNs, Mgm1 and OPA1 were annotated as membrane-bound GTPases and subsequently Biomacromolecules, CAS Center for Excellence in Biomacromolecules, found to serve as IMM fusogens and cristae-shaping proteins [18,21,22]. Linkage of mitochon- Institute of Biophysics, Chinese Academy drial fusogen mutations to neurodegenerative diseases emphasizes the physiological importance of Sciences, Beijing 100101, China of mitochondrial membrane dynamics in neuronal cells, likely explained by the high demand for energy there [4,8].

*Correspondence: The mechanism of mitochondrial fusion is unclear, partly due to the lack of structural information [email protected] (S. Gao) and on these fusogenic proteins. Recently, several crystal and cryo-electron microscopy (EM) [email protected] (J. Hu).

62 Trends in Cell Biology, January 2021, Vol. 31, No. 1 https://doi.org/10.1016/j.tcb.2020.09.008 © 2020 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Trends in Cell Biology

structures of MFN and OPA1 have been reported [23–30]. In this review, we summarize these Glossary advancements and discuss the new insights in mitochondrial fusion therefrom. Amphipathic helix: an α-helix with hydrophobic residues aligning on one Structural Relationship of Mitochondrial Fusogens with the Dynamin Superfamily side and hydrophilic residues on the other side. Typically, these helices can MFNsandOPA1belongtothedynaminsuperfamily of multi-domain GTPases engaged in induce curvature by inserting their various membrane-remodeling events in eukaryotic cells [31]. Of these so-called DLPs hydrophobic face shallowly into (see Table 1 for gene names of key DLPs discussed herein), the best known function of dynamin membranes. is cleavage of clathrin-coated vesicles from the plasma membrane during endocytosis [32]. In Atlastin (ATL): an ER-resident DLP that mediates fusion of ER membranes. ATL addition, dynamin-1-like protein (DNM1L), also known as dynamin-related protein 1 (Drp1), tethers membranes by GTP-dependent mediates mitochondrial fission [33]; myxovirus-resistant (Mx) proteins and guanylate-binding dimerization and fuses them by proteins (GBPs) restrict several types of RNA virus and retrovirus [34,35]; and atlastin (ATL) conformational changes. Mutations of catalyzes endoplasmic reticulum (ER) fusion [36]; EH domain-containing proteins (EHDs) are human ATL1 causes hereditary spastic paraplegia. involved in endosomal trafficking [37], and a recently identified member, neurolastin, is a neuronal Autosomal dominant optic atrophy differentiation mediator during embryonic development [38]. In addition, bacteria also have (ADOA): a hereditary neuronal bacterial DLPs (BDLP), although their functions are not yet fully understood [39]. degenerative disease characterized by reduced visual acuity, vision loss, or vision impairment. It is caused by The DLPs all contain a GTPase (G) domain and a helical region (Figure 1A,B). Previous structural mitochondrial dysfunction that leads to and biochemical studies, especially over the past decade, revealed important common features the degeneration of optic nerve fibers. of DLPs [40–49]. First, compared with the canonical small GTPases, such as Ras and Rab, DLPs Bacterial dynamin-like protein fi (BDLP): bacterial ancestors of DLP. have an enlarged G domain that binds guanine nucleotides with weaker af nity (micromolar range The structures of BDLP resemble that of versus nanomolar range for Ras-like GTPases in terms of the dissociation constant Kd)[31]. MFN. Some BDLP appears in a tandem Second, DLPs generally do not need specific GTPase-activating proteins (GAPs) or guanine nu- manner. The functions of BDLP remain cleotide exchange factors (GEFs) during the GTP hydrolysis cycle [31,50]. Instead, their GTPase elusive. Cardiolipin: a type of phospholipid that activity is stimulated by homodimerization of the G domains [42,51], and the reloading of GTP for is enriched in, and almost exclusive to, DLPs appears to be spontaneous, in accordance with their relatively low affinity for guanine nu- IMM. It contains two phosphatidic acid cleotides [45,46,48,51]. Third, the function of DLPs relies on the relative movement between groups linked by a glycerol. the domains regulated by GTP hydrolysis [50,52,53]. MFN and OPA1 were the last members Charcot-Marie-Tooth disease type 2A (CMT2A): a hereditary neuronal of the dynamin superfamily to have information about their structure revealed [52]. For structural degenerative disease characterized by features of fission DLPs, refer to Box 1. distal weakness, atrophy, sensory loss, decreased deep-tendon reflexes, and variable foot deformity. It is caused by MFN Structures and Mitochondrial Outer Membrane Fusion mutations in MFN2. With a negatively charged OMM, mitochondrial fusion does not occur spontaneously. Genetic Fuzzy onions (Fzo): aMFNhomologin data suggest that MFN is needed on both opposing OMMs to allow fusion; thus, the trans Drosophila. Fzo was the first reported interaction between MFNs is essential [14]. In earlier studies, MFNs were depicted as a MFN [97], and its mutations disrupt mitochondrial fusion during spermatid V-shaped molecule that anchors to the OMM via two transmembrane (TM) helices [60]. The G differentiation, making the normal onion- domain and two conventionally termed heptad repeats (HR1, from residues 317-400 and like fused mitochondria (the Nebenkern) ‘fuzzy onions’. – – GDP•BeF3/GDP•AlF4: analogs used for mimicking the transition state of GTP Table 1. Gene Names of Key DLPs in Various Speciesa hydrolysis when the γ-phosphate is Function DLP name hydrolyzed off from GTP but remains in the active site of the GTPase. Saccharomyces Arabidopsis thaliana Drosophila Homo sapiens G-HB1: a truncated construct used in cerevisiae melanogaster structural and functional studies of MFN;

OMM fusion Fzo1 FZL Fzo/Marf MFN1/2 also known as MFNIM or MGD. It contains the G domain and an IMM fusion Mgm1 N/A Opa1 OPA1 associating HB. Given interdomain Mitochondrial fission Dnm1 N/A Drp1 DNM1L stabilization, G domain often aggregates (DLP1, DRP1) when expressed and purified alone. Vesicular fission N/A DRP1/2 shi (shibire) DNM1/2/3 Similar constructs have been reported for dynamin (termed GG), OPA1, and ER fusion SEY1 RHD3 Atl ATL1/2/3 several other DLPs. aN/A, not applicable.

Trends in Cell Biology, January 2021, Vol. 31, No. 1 63 Trends in Cell Biology

Helix bundle (HB): a set of more than (A) two α-helices bundling into a stalk-like BSEG domain BSE Stalk PH Stalk BSE PRD Dynamin domain. These helices are glued through 6 33 293 314 321 499 518 631 653 708 746 864 (1–864) hydrophobic interactions and often act BSE G domain BSE Stalk BI GEDStalk BSE DMN1L/Drp1 as molecular lever to rely motions. 220 299321329 491503610642703729 (1–736) Heptad repeat (HR): a repeated sequence with a set of seven amino BSE G domain BSEStalkL4 Stalk BSE Mx acids. With the seven residues, 44 69 340 367 529 574 632 662 (1–662) hydrophobic ones appear at every three TM BSEG domain BSE StalkPaddle/LIS Stalk BSE OPA1/Mgm1 or four positions. HR routinely forms 74 223 254 524 549 703 710 809 877 911 (1–939) coiled coil structures using its aligned HB1 G domain HB1 HB2TM A HB2 HB1 MFN/Fzo1 hydrophobic face. 6 75 336 365596 628 637 695 741 (1–741) Mitofusin (MFN): a membrane-bound DLP named after its mitochondrial fusion G domain 3HB TM A Atlasn activity. MFN localizes to the OMM and 33 338 347 432 447 503 (1–558) mediates merging of apposing OMMs in G domain Helical region GBP1 a GTP-dependent manner. 6313583 (1–592) Nostoc punctiforme BDLP He G domain Helical EH EHD2 (NpBDLP): a model DLP molecule the 20 59 288 402 443 534 (1–543) full-length structure of which greatly NeckG domain Neck TrunkPa Trunk Neck BDLP advanced early understanding of DLPs 568 300311 358 571 606 660 693 (1–693) in terms of domain organization and structural dynamics. The domain movement around hinge 1 and hinge 2 upon nucleotide binding was ﬁrst (B) observed for NpBDLP [39,65]. Being the Mgm1 (OPA1) MFN1 closest structural homolog to MFN TM – G domain [23 26], NpBDLP serves as a prototype Paddle/LIS Predicted for the working mechanism of MFN. G domain Hinge 2 HB2/HD2 Optic atrophy type 1 (OPA1): agene N Stalk found on chromosome 3, region q28-qter that is linked to optic atrophy. Hinge 2 OPA1 localizes to the IMM and mediates Hinge 1 C HB1/HD1 mergingofapposingIMMsina C BSE N Hinge 1 GTP-dependent manner. PH domain G domain G domain ATL1 3HB Dynamin Stalk PRD TM BSE

G domain Helical Stalk EH DNM1L domain domain

BSE EHD2 BI G domain G domain MxA Stalk Paddle

Trunk BSE L4 G domain BDLP GBP1 Helical Head region (G domain) Neck

Trends in Cell Biology

(See ﬁgure legend at the bottom of the next page.)

64 Trends in Cell Biology, January 2021, Vol. 31, No. 1 Trends in Cell Biology

Box 1. Structure of Fission DLPs The structures of fission DLPs have been extensively studied, As an example, DNM1L comprises the G domain, a BSE, and a stalk (see Figure 1B in the main text). The BSE, connected to the G domain via the so-called hinge 2, is formed by three helices derived from widely dispersed sequence regions. The stalk is a four-helix-bundle connected to the other side of the BSE via hinge 1 [48]. This domain organization is shared by dynamin and Mx proteins [45–47], representing the typical architecture of DLPs with membrane fission activity (fission DLPs). At the far end of the DNM1L molecule, a flexible region called the B-insert, which mediates the anchoring of DNM1L to target the OMM, bulges from the stalk [48]. The positional and functional equivalent of the B-insert is the pleckstrin homology (PH) domain in dynamin for binding phos- phatidylinositol lipids, and loop L4 in Mx proteins for recognizing viral ribonucleoparticles [45,46,51](seeFigure 1Bin the main text). These fission DLPs are able to deform and tubulate liposomes in vitro in the absence of nucleotide [48,54,55]. In cells, cytosolic DNM1L can form helical oligomers via the stalk around the OMM. In the GTP hydrolysis cycle, the DNM1L molecules undergo vigorous domain movement via hinge 1 and hinge 2, collectively leading to constriction of the helical oligomer and nonleaky scission of the tubulated membrane [49,56]. This functional model shared by dynamin and Mx proteins is endorsed by data from crystal structures, cryo-EM helical reconstruction, and other biophysical approaches [45–48,57–59].

HR2, from residues 662-737 of human MFN1) flanking the TM helices facing the cytosolic side. According to the first structural report on MFN, the 75-residue HR2 of mouse MFN1 folds into alongα-helix and dimerizes to form an antiparallel coiled coil in the crystal [60]. Based on this structure, a SNARE-like tethering model was deduced for MFN1 HR2 [60]. However, biochemical characterization of mammalian MFNs revealed that they form homo- and hetero-oligomers when tethering isolated mitochondria in vitro. Importantly, tethering activity is dependent on GTP hydrolysis [61]. These results achieved before the burst of structural information for DLPs over the past decade raised a question about the role of GTP hydrolysis in the HR2-mediated tethering model [62]. Structures of another fusion DLP, atlastin (ATL), which fuses ER membranes, highlighted a G domain-mediated tethering model [43,44]. The N-terminal cytosolic domain of ATL comprises a G domain, followed by a three-helix bundle (3HB) [43,44]. During GTP hydrolysis, the G domains form trans dimers, and the swing of neighboring 3HB domains drags the opposing ER membrane close to allow fusion [43,44]. MFN is similar to ATL in that they are both integral membrane fusion DLPs with no evidence of forming helical oligomers observed for fission DLPs. Therefore, the G domain-mediated tethering model has also been presciently proposed for MFN before the report of its structure [62].

Structures of MFN Recent structural studies on MFNs, including both MFN1 and MFN2, provided new insights into the mechanism of OMM tethering [23–26]. The constructs used for crystallization were designed based on existing DLP structures [23,24,26]. They all contain the G domain and the ﬁrst helical domain termed HD1 [also known as helix bundle 1 (HB1)], and the predicted HD2 (also known

Figure 1. Structural Comparison of the Dynamin-Like Protein (DLPs). (A) Schematic drawing showing the domain organization of DLPs. Domains with structural homology and/or functional relation of these proteins are shown in the same color. The domains shown in blue are helix-rich. The positions of the domains in the primary structures of the DLPs are indicated by amino acid residue numbers. These numbers are based on following protein sequences: Dynamin-1 (Uniprot accession code Q05193), DNM1L (O00429), MxA (P20591), Mgm1 (G0SGC7), MFN1 (Q8IWA4), Atlastin-1 (Q8WXF7), GBP1 (P32455), EHD2 (Q8BH64), and BDLP (B2IZD3). Numbers of the total amino acids of these proteins are given. Abbreviations: A, amphipathic helix; BI, B insert; BSE, bundle-signaling element; EH, Eps15 homology domain; G domain, GTPase domain; HB, helical bundle; He, helical; L4, the L4 loop; LIS, lipid interacting stalk; Pa, paddle. PH, pleckstrin homology domain; PRD, proline-rich domain; TM, transmembrane domain. (B) Structures of the DLPs, with domains labeled and colored as in (A) Domains without structural information are shown as broken lines. Note the structural similarity between mitofusin (MFN) and bacterial DLP (BDLP), and between optic atrophy 1 (OPA1) and the ﬁssion DLGs [dynamin, dynamin-1-like protein (DNM1L), and myxovirus-resistant A (MxA)]. Structures of mitochondrial genome maintenance 1 (Mgm1) (Protein Data Bank 6QL4), MFN1 (5GO4), dynamin (5A3F),DNM1L(4BEJ);MxA(3SZR), guanylate-binding protein 1 (GBP1) (1DG3), atlastin 1 (ATL1) (3Q5D), EH domain-containing protein 2 (EHD2) (2QPT), and Nostoc punctiforme BDLP (NpBDLP) (2J68) are presented.

Trends in Cell Biology, January 2021, Vol. 31, No. 1 65 Trends in Cell Biology

as HB2) and TM helices have been removed [23,24,26]. HB1 and HB2 of MFN are equivalent to the bundle-signaling element (BSE) and stalk of ﬁssion DLPs [50]. HB1, which comprises two α- helices from the N terminus, one after the G domain and one from the C terminus, is also reminis- cent of the neck domain of the cyanobacterial DLP from Nostoc punctiforme (NpBDLP)[39] (Figure 2A). Notably, the majority par of conventional HR2 accounts for the C-terminal helix of HB1 [23,24,26]. The G-HB1 constructs of MFN1 and MFN2, also termed internally modiﬁed

MFN (MFNIM)orminimalGdomain(MGD),aremonomericinsolutionintheapostate [23,24,26]. GTP loading triggers the rearrangement of residues around the nucleotide-binding pocket to constitute an interface (G interface) through which the G domains dimerize in the tran- – – sition state of GTP hydrolysis (in the presence of GDP•BeF3 or GDP•AlF4 )[25,26]. Simulta- neously, G-HB1 dimers undergo a major conformational change, as observed for many other DLPs; HB1s move from protruding in opposite directions (the open conformation) to the same di- rection (the closed conformation) via hinge 2 [24–26](Figure 2A). The dimerization and conformational changes were subsequently conﬁrmed by ﬂuorescence resonance energy transfer (FRET)- based analysis [25,26], which was also used for studying the domain movement of ATL [63].

Hints for OMM Fusion The structural features described earlier support an ATL-like tethering model [43,63]. Compared with ATL, MFN has an ‘extra’ predicted HB2 that is essential for the fusion activity but has no structural information available [64]. The possible behavior of HB2 during OMM fusion can be modeled based on the equivalent NpBDLP trunk domain, which moves around hinge 1 during the GTP hydrolysis cycle [23,24,65]. Mutation of hinge 1 residues in human MFN1 leads to aberrant mitochondrial fusion, suggesting that a similar conformational change is essential for MFNs [24]. Thus, GTP binding and hydrolysis regulate conformational changes in MFN to allow trans association of G domains and domain movements via the two hinges, eventually leading to the tethering of opposing OMMs from a reasonable distance (Figure 2B) [24]. An alternative HR2-mediated tethering model has been hypothesized in which the conventional HR2 helix detaches from HB1 [66], which would need to overcome a high energy barrier to break up the tight hydrophobic network [23,24,26].

MFN achieves and maintains membrane tethering by continuous GTP hydrolysis. G-HB1- anchored vesicles form clusters only when GTP is supplied [23,24]. The next question is how tethered OMMs merge at the molecular level. A cryo-electron tomography (cryo-ET) study using isolated yeast mitochondria illustrated a possible scenario for how OMMs evolve from tethering to local merger with Fzo1 [67]. A cluster of proteins, likely enriched with Fzo1, tether the opposing OMMs at a distance of ~6 nm via trans interactions. To allow the lipid mixing needed for fusion, tethering complexes were only seen at the orbit of the tightly apposed membrane interface. Fusion pores were observed near the edge of the docking ring, where the membranes are supposed to be highly curved [67].

Membrane deformation, including curvature generation, affects fusion [68]. The conventional HR1 of MFN1, which mainly comprises the predicted HB2, has been found to destabilize membranesifisolatedasapeptide[69]. More plausibly, an amphipathic helix neighboring the TM domain of human MFN1 may facilitate the ﬁnal step of the OMM merger and mitochondrial targeting [64]. A similar role for membrane destabilization has also been reported for a C-terminal amphipathic helix of ATL [70,71].

Homologs of MFN1 have common and unique features in mediating fusion. For Fzo1, a switch in the intramolecular salt bridge has been suggested to promote membrane curvature and allow the ubiquitination of Fzo1 needed to trigger eventual membrane merger [72,73]. For more details about structural differences and cooperation between MFN1 and MFN2, see Box 2.

66 Trends in Cell Biology, January 2021, Vol. 31, No. 1 Trends in Cell Biology

(i) (ii) G domain (A) NpBDLP HB1 Neck MFN G-HB1 Apo state Apo state (’Folded’) G domain Paddle GTP binding Dimerizaon via Trunk and hydrolysis the G interface GTP analog binding Hinge 2

Hinge 2 Transion state (’Closed’)

GMPPNP-bound Hinge 1 state Release of the MFN1: G domains dissociate (’Stand-up’) phosphate MFN2: dimerizaon sustains GDP-bound state (’Open’)

MFN (B) HB1 (i) (ii) G GTP HB2 APH Tethering OMM Cytosol TM Mito OMM G domain dimerizaon

Domain Docking movement GDP (iv) (iii)

Merger

Membrane destablizaon

Trends in Cell Biology Figure 2. Structure and Functional Model of Mitofusin (MFN). (A) Domain movement and dimerization of MFN G-helix bundle 1 (HB1) during GTP hydrolysis cycle [(i) Protein Data Bank codes from top to bottom: 5GO4, 5YEW, and 5GOM] and the domain movement of bacterial dynamin-like protein (BDLP) [(ii) from top to bottom: 2J69 and 2W6D]. The behavior of BDLP has provided important clues for the conformational ﬂexibility of full-length MFN. (B) A model for MFN-mediated outer mitochondrial membrane (OMM) fusion. (i) The loading of GTP triggers the domain movement of MFN from a folded state to a stand-up state. (ii) The GTP-bound MFN molecule from opposing mitochondrial OMM dimerizes via the G domains to accomplish the OMM tethering step. (iii) Upon GTP hydrolysis, the G domain and HB1 of the trans MFN dimer change from an open conformation to a closed conformation, thereby dragging the opposing OMMs in close proximity (docking). (iv) Docked OMMs merge, and the MFN molecules become GDP-bound for next GTP hydrolysis cycle.

Trends in Cell Biology, January 2021, Vol. 31, No. 1 67 Trends in Cell Biology

Box 2. Comparison and Cooperation between MFN1 and MFN2 MFN1 and MFN2 have different roles in mediating OMM fusion. Relevant evidence has been gleaned with regard to tissue- specific expression profiles [15], knockout phenotypes in animal models [14], causation of human diseases [17], post- translational modifications [74], and biochemical properties [61]. Recent studies revealed subtle yet important structural differences between the two human MFNs that may define their functional differences.

MFN1 G-HB1 has self-stimulating GTPase activity coupled with domain dimerization and dissociation [24]. MFN2 has a much tighter G interface than MFN1. MFN2 G-HB1 stays dimerized even after GTP hydrolysis, which prevents the GTP/GDP exchange, resulting in a negligible apparent GTPase activity [26]. Human MFN1 has >20-fold higher GTPase activity than MFN2, which is largely dependent on variance in a single amino acid (MFN1-I108 versus MFN2-T129) at the G interface [26]. More interestingly, this variance is only found in primates. Mouse MFN1 and MFN2 have GTPase activity comparable to that of human MFN2, and they both have a threonine at corresponding positions. This observation highlights the special feature of primate MFN1 and suggests the functional complexity of MFNs during evolution [26].

In addition, an ~190-residue region across HB1 and HB2, comprising the conventional HR1, has been reported to deter- mine the speciﬁc functions of MFN1 and MFN2, because swapping this region between the two MFNs exchanged the phenotypes of mitochondrial elongation [75]. MFN1 and MFN2 form complexes in vitro [61] and co-immunoprecipitate in MEFs [14]. The MFN1-MFN2 trans heterocomplexes have greater fusion efﬁcacy on isolated mitochondria compared with homocomplexes [76]. This heterotypic interaction appears to be dependent on GTP loading and hydrolysis, and can be mediated by the G interface [26,61].

– When MFN1 and MFN2 G-HB1s are mixed in the presence of GDP•BeF3, the heterocomplex forms more efﬁciently than do the MFN1 or MFN2 homodimers. MFN1-MFN2 heterodimerization via the G interface at the transition state of GTP hydrolysis is also coupled with the domain movement of HB1 from the stretching state to the folded state. Moreover, some CMT2A-related mutations on MFN2 do not affect formation of the heterocomplex, suggesting different mechanisms of MFN2-related CMT2A [26].

Similar to the stalk of ﬁssion DLPs mediating dimerization/oligomerization, HB2 may constitute another interface for the homo- and heterocomplexation of MFN1 and MFN2. Evidence can be found on the G domain-HB1-HB2 dimer of MFN1 and the heterotetramer of two bacterial DLPs, CjBDLP1 and CjBDLP2 [77]. Formation of the MFN1-MFN2 heterocomplex is competitive with that of homocomplexes.

OPA1 Structures and Mitochondrial Inner Membrane Fusion The mechanism of IMM fusion is likely to be more complicated than that of the OMM. The success of OMM fusion, similar to ER fusion, is not necessarily 100%, because these membranes can be temporarily tethered and subsequently untethered without causing problems [12]. By contrast, once the OMMs of two mitochondria are fused, their IMMs should merge quickly to complete the process [10,78,79]. In addition, the IMM fusogens Mgm1 and OPA1 undergo partial proteolytic processing before being fusion competent, yielding a TM-containing long form and a soluble short form [80–85]. Such arrangements have not been reported for any other DLPs that mediate homotypic membrane fusion.

IMM fusion is even more complicated by the IMM fusogen OPA1 shaping the cristae of the IMM [78]. Deletion of OPA1 not only causes mitochondrial fragmentation, which is evidently linked to its fusion activity, but also diminishes cristae biogenesis, in that IMM becomes vesicular, as observed by EM [78]. Puriﬁed short Mgm1 (s-Mgm1) and short OPA1 (s-OPA1) are able to interact with IMM-like liposomes, induce local membrane bending, and deform them into tubule-like structures [86,87]. Tubulation of membranes by these DLPs is consistent with the formation of IMM protrusions between cristae. As with other membrane-tubulating DLPs, including dynamin-1, membrane-based s-Mgm1 or s-OPA1 assembly stimulates GTPase activity [88,89]. Intriguingly, these characteristics of OPA1/Mgm1 are often observed only with ﬁssion DLPs [90].

Structures of Mgm1 and OPA1 Low-resolution structures of membrane-bound s-Mgm1 derived from Saccharomyces cerevisiae (Sc) sequences and puriﬁed from either Escherichia coli or Sf9 insect cells, were initially obtained

68 Trends in Cell Biology, January 2021, Vol. 31, No. 1 Trends in Cell Biology

using EM analysis, both of which showed a threefold symmetry and possibly a hexameric ring assembly [88,89].

Recently, three crystal structures and four EM structures were concurrently reported for Mgm1 and OPA1 by three independent groups [27–30]. Mgm1 from Sc and thermophilic fungus (Chaetomium thermophilum, Ct), and human OPA1 were used in these studies (Figure 3A) [27– 30]. In general, s-Mgm1 and s-OPA1 exhibit dynamin-1-like folding [27,28,30]. The crystal structure of ScMgm1 [GDP bound, Protein Data Bank (PDB) code: 6JSJ] [28] and CtMgm1 (nucleotide-free, PDB code: 6QL4) [27] are almost superimposable and similar to the structures of nucleotide-free dynamin-1 (PDB codes: 3SNH and 3ZVR) [45,46]. These proteins all comprise two central helical regions (BSE/HB1 and stalk/HB2), with the N-terminal region closely attaching to the G domain and the other end of the molecule linking to a lipid-interacting module (Figure 1B). As with MFNs, both termini of s-Mgm1 and s-OPA1 are in close proximity. Compared with dynamin-1, the stalks of Mgm1 and OPA1 have a different orientation relative to the BSE. In addition, the PH domain in dynamin-1 is replaced by a lipid-interacting stalk (LIS, or paddle) that re- sides in the distal end of the stalk and is nearly perpendicular to it [27,28]. The OPA1-MGD – (equivalent to G-HB1/BSE) structure (GDP•BeF3 bound, PDB code: 6JTG) can be nicely overlaid with the Mgm1 MGD region and dynamin-1 GG structure (e.g., PDB code: 2X2F) [29]. These structural findings not only confirm conservation between these DLPs, but also raise the question of how a fission DLP-like fold executes IMM fusion.

Fusion versus Cristae Formation As discussed earlier, the swing of HB1 relative to the G domain and bending between HB1 and HB2 likely have a key role in MFN-mediated fusion [25]. However, structural details of theMgm1hinge1and2suggestthatMgm1/OPA1donotfollowthispath[28]. These domains maintain the same relative orientations in all structures. By contrast, IMM fusion DLPs utilize a variety of intermolecular interactions. At least three interfaces are present in the structures: a BSE stalk-mediated ‘back-to-back’ interface; a G-BSE-stalk-mediated ‘head-to-tail’ interface; and a G-G-mediated ‘head-to-head’ interface. In the crystal structures of ScMgm1 and CtMgm1, the ‘back-to-back’ interfaces are slightly shifted, but conserved residues are used [27,28](Figure 3B). These interactions were conﬁrmed in the CtMgm1 and OPA1 EM structures [27,30]. A similar 'back-to-back' assembly is seen with ﬁssion DLPs, including dynamin-1. Thus far, the ‘head-to-tail’ interface has only been observed with ScMgm1 [28]. Whether it is applicable to other IMM fusogens is yet to be determined. The interface chains s-Mgm1 into trimeric rings and might extend to low-order spirals [28]. The ‘head-to-head’ interface is seen in most DLPs [25]. In the case of MFN and ATL, it is the foundation of membrane tethering/fusion. In the case of dynamin-1, it is thought to form transiently during membrane-based helical assembly and related conformational changes [42]. All three interfaces are critical in structure–function analysis using yeast cells or mammalian cells. However, their precise roles remain elusive.

Purified s-Mgm1 and s-OPA1 are known to tubulate membranes and form helical assemblies in a fission DLP-like manner [27,30]. It is plausible that Mgm1 or OPA1 wraps around the protruding IMM that eventually forms cristae. Strikingly, CtMgm1 is able to self-associate on the inner side of liposomal membranes and form similar helical assemblies, which fits the shaping process of the indented side of the cristae [27](Figure 3C). Regardless of the orientation, it is clear that the ‘back-to-back’ interface is essential in cristae biogenesis [27,28]. The ‘head-to-head’ interface mutants, which are currently only available for OPA1, are yet to be tested in membrane tubulation by s-OPA1 [29]. Based on the results of equivalent mutants in dynamin-1, the ‘head-to-head’ interface is probably also important for cristae formation [29].

Trends in Cell Biology, January 2021, Vol. 31, No. 1 69 Trends in Cell Biology

(i) (ii) (A) Mgm1/OPA1 G domain Paddle/LIS

Stalk

BSE G domains Dimerizaon via BSE the G interface

(iii) Assembly for Polymerizaon Assembly for (iv) inner surface decoraon outer surface decoraon

(B) (C) OMM IMM Cristae ScMgm1 CtMgm1 ‘Back-to-back’ LIS

Stalk G BSE Mgm1/OPA1 ScMgm1 OPA1 ‘Head-to-tail’ ‘Head-to-head’

OMMIMM Assembly on the outer OMM IMM IMM fusion and (D) surface of IMM disassembly

GTP hydrolysis

l-Mgm1/OPA1 s-Mgm1/s-OPA1

Trends in Cell Biology Figure 3. Structure and Functional Model of Mitochondria Genome Maintenance 1/Optic Atrophy 1 (Mgm1/OPA1). (A) Assembly of the short (s)-Mgm1/s-OPA1 polymers. (i) The full-length s-Mgm1 structure in the apo state [Protein Data Bank (PDB) code 6QL4]. (ii) Dimerized s-OPA1 G-bundle-signaling element (BSE) in the transition state (6JTG). s-Mgm1/s-OPA1 can assemble into helical polymers via the stalk in two forms, one decorating the inner surface of the lipid tubule [(iii) 6RZW], the other decorating the outer surface of the lipid tubule [(iv) 6RZU]. The G domain association takes place in an inter-ring manner in the inner surface- decorating polymers. (B) Schematic showing the various assemblies of Mgm1/OPA1 oligomers, as represented by the structures of Saccharomyces cerevisiae (Sc)Mgm1 (6JSJ), Chaetomium thermophilum (Ct)Mgm1 (6QL4), and OPA1 G-BSE (6JTG). (C) A model for Mgm1/OPA1-mediated cristae formation. Helical assembly of Mgm1/OPA1 induces membrane curvature needed for cristae formation. The ‘back-to-back’ and ‘head-to-head’ interfaces are involved. (D) A model for Mgm1/ OPA1-mediated inner mitochondrial membrane (IMM) fusion. Transient ‘head-to-tail’ assembly of Mgm1/OPA1 induces membrane curvature that producing unstable tips on two opposing IMMs. Disassembly reverses membrane bending. When opposite unstable tips encounter each other, lipid mixing results in a fusion pore that can be expand to complete IMM fusion. The L-form and s-form are likely mixed proportionally in the ‘head-to-tail’ chain.

70 Trends in Cell Biology, January 2021, Vol. 31, No. 1 Trends in Cell Biology

For OPA1/Mgm1-mediated membrane fusion, the consensus is that, when they tubulate membranes, the resulting unstable tips are spontaneously fusogenic [27,28]. Based on assembly-stimulated GTPase assays, the ‘head-to-tail’ interface is less likely to be involved in extensive assembly and could not contribute to cristae formation [28]. However, it is capable of organizing Mgm1 into a short spiral that, when it sits on the membrane, generates unstable tips for subsequent fusion (Figure 3D) [28]. It is not clear which assembly is specifically needed for IMM fusion or whether they work in cooperation. When these assemblies fit into the geometry of mitochondria, the dynamin-like assembly would be closely associated with cristae, with the resulting tubular tips as cristae junctions covered by protein complexes, and no longer fusogenic [27]. By contrast, the ScMgm1 trimeric assembly would be sufficient for the tight space between newly merged OMMs and, thus, more likely to mediate fusion by generating tubular tips there [28]. Notably, trimers of Mgm1 can be formed without nucleotide, but, as with other DLPs, the fusion process is inevitably GTP dependent. A possible scenario is that the GTP cycle resets the ‘head-to-tail’ assembly and allows the generation of sufficient unstable tips to ensure the pairing of opposing IMMs [28].

Nucleotide-Independent Dimerization of OPA1 In these structural studies, both s-Mgm1 and s-OPA1 need to be truncated at the N terminus due to the ﬂexible nature of the region [27–30]. In both ScMgm1 and CtMgm1, one helix extension is found before the G domain, which results in a 3HB for the BSE. In OPA1, an extra helix was identiﬁed in the N terminus; it does not join BSE to make a four-helix-bundle. Instead, it forms a weak dimerization motif [89] and complicates the possible assembly of OPA1 by adding one more interface [29].

Both s-Mgm1 and s-OPA1 bind to membranes in the absence the l-form. When CtMgm1 decorates membranes, either outside or inside, only the LIS domain makes direct contact with the lipid bilayer, just as with assembled dynamin-1 [59,91]. Similarly, assembled s- OPA1 attaches to membranes via its LIS domain, and the remaining regions of the molecule are away from the lipid surface [30]. However, lipid-interacting assays, either flotation or sedimentation, have confirmed that the G domains of Mgm1 and OPA1 are also capable of membrane engagement. The dual binding sites of Mgm1 fit well on the membrane when the ‘head-to-tail’ interface is used for assembly [28]. Notably, the nucleotide-independent dimerization described earlier has a critical role in the efficient membrane association of s-OPA1 [29].

Lessons from Mitochondrial Fusogens Disease-Associated Mutations in Mitochondrial Fusogens Over 100 different single-point mutations in MFN2 have been linked to the occurrence of CMT2A, and most are distributed in the G domain and HB1 [26]. The crystal structure of MFN G-HB1 offers clues for understanding the pathogenic mechanisms associated with such mutations. Mutations at hinge 2 (including the hotspot R94Q/W and K357N) and the G-HB1 interface in the transition state (L248V, P251A/R, and R364W/Q/P) promote the GTP turnover of MFN2 G-HB1, possibly by regulating the relative movement between the G domain and HB1 [23,26]. Thus, excessive activation of MFN2 is harmful to the physiological homeostasis of mitochondria, causing a pathological outcome [26]. Another important observation is that many CMT2A-related MFN2 mutants are able to efﬁciently associate with wild-type MFN1 or MFN2 via the G interface [26]. This type of heteroassociation highlights the complexity of disease-related MFN2 mutants in a cellular environment. On the one hand, the disturbed functions of these mutants may be partly compensated for by normal MFN1 or MFN2. On the other hand, these mutants can hijack normal MFN molecules and cause a dominant negative effect.

Trends in Cell Biology, January 2021, Vol. 31, No. 1 71 Trends in Cell Biology

Many optic atrophy-related mutations have been reported for human OPA1 [22,92]. Similar to Outstanding Questions MFN2, most of these mutations are mapped to the G-BSE region, tied to nucleotide binding, G There are still a few critical structures to domain dimerization, and folding [29]. As mentioned earlier, the GTP cycle likely has an indirect be determined for the mitochondrial role in OPA1 functions, both cristae formation and IMM fusion [29]. Thus, it differs from ATL1 fusion machinery. How are the HB2 of MFN and the stalk of OPA1 organized, (HSP-causing) and MFN2 (CMT2A-causing) disease mutations in which fusion-related mecha- and what do the full-length molecules nisms are directly compromised [26,43]. look like? How does OPA1 assemble homotypically? What are the nucleotide- Comparison between Fusion DLPs and Fission DLPs dependent conformational changes for Membrane fusion or fission are two completely opposite reactions. It is difficult to tell the desig- full-length MFN and Mgm1/OPA1? nated activity of a DLP based solely on its structural configuration. Detailed biochemical compari- Why would the HB2 of MFN be func- sons have revealed two clues. First, fusion and fission DLPs may prefer different oligomerization tionally essential, if the dimerization states: fusion DLPs preferentially form G domain dimers upon nucleotide addition [23,24]. It may and swing motion of the G-HB1 appears to be sufficient? Does bending cluster within the same membrane via the TM region [23], but HB-based nucleotide-independent at hinge 1 seen in BDLP have a role in assembly has not been reported. By contrast, fission DLPs have prominent nucleotide- MFN-mediated fusion? independent oligomerization [93], even though dynamin requires G domain dimerization for fission [34]. Second, fusion and fission DLPs may be differentially associated with the membrane How do Mgm1 and OPA1 utilize different assembly interfaces to mediate either (Figure 1). Almost all fusogens, including SNAREs, viral proteins, and fusion DLPs, are integral IMM fusion or cristae formation? How membrane proteins. However, fission DLPs are usually peripheral membrane proteins [93]. They do s-form and l-form of these fusogens attach to membranes using lipid-interacting modules, but dissociate once fission is complete. work cooperatively?

How would fusion of OMM and IMM Interestingly, as fusion DLPs, both Mgm1 and OPA1 do not follow these rules because they can coordinate? Is it orchestrated by the undergo extensive nucleotide-independent oligomerization. In addition, they exist in both integral and previously identified Ugo1 or by other peripheral membrane forms and the soluble s-form is capable of membrane binding, as seen with mechanisms? Would the fusion ma- fi fi chinery discussed here be suf cient ssion DLPs. We suspect that this exception is partly due to the needs of cristae biogenesis, because for overall shaping of mitochondria? tubular shaping of the cristae is fission DLP-like behavior. Mgm1 and OPA1 adapt to these changes by modifying the fusion mechanism. Finally, IMM fusion may not require the dedicated tethering step that How is mitochondrial fusion regulated? is essential for other fusion events, because two IMMs are physically close enough after OMM fusion. How would lipid composition, including the presence of cardiolipin, stress conditions, or apoptotic factors influence Concluding Remarks and Future Perspectives fusion activity? Structural analysis has a key role in understanding mechanisms of mitochondrial fusion mediated by DLPs. Several key structures, including the HB2 of MFN, full-length MFN/Fzo1, and s-OPA1, are still required to complete the puzzle (see Outstanding Questions). Rigorous domain movement, the oligomerization tendency, and the membrane-associating feature are main obstacles for the structural study of these proteins, which entails a combined application of multiple techniques, such as cryo-EM and single-molecular FRET. These structures will help decipher how full-length MFN rearranges during fusion, why HB2 is essential even when G-HB1 is sufficient for dimerization and the ATL-like HB swing, and how OPA1 utilizes multifaceted interfaces to achieve fusion. Recent in vitro fusion assays of OPA1 suggest that a 1:1 ratio of the s- and l-form would be optimal for membrane fusion [94], but how they cooperate is still not clear. Although it is clear that negatively charged lipids are required for membrane association of s-Mgm1 or s-OPA1 [88,95],theabsoluterequire- ment for cardiolipin, a unique negatively charged lipid enriched in the IMM, is still debatable [96]. Using assembly-stimulated GTPase as an example, cardiolipin is essential for ScMgm1 and OPA1, but acts adversely for CtMgm1 [27,28]. How specific lipids, stress conditions, or apoptotic factors regulate fusion remains elusive. Finally, it is critical to understand how OMM and IMM fusion are coordinated. A co-reconstitution experiment with two-layered proteoliposomes containing both MFN and OPA1, although extremely challenging, would address these key questions.

Acknowledgments This work was supported by grants from the National Key R&D Program of China to S.G. (2018YFA0508300) and J.H. (2016YFA0500201), National Natural Science Foundation of China to S.G. (31722016 and 81772977) and J.H.

72 Trends in Cell Biology, January 2021, Vol. 31, No. 1 Trends in Cell Biology

(91854202, 31630020, and 31421002), the Strategic Priority Research Program of the Chinese Academy of Sciences to J.H. (XDB39000000), Natural Science Foundation of Guangdong Province to S.G. (2019TX05Y598), and Innovative Team Program of Guangzhou Regenerative Medicine and Health Guangdong Laboratory to S.G. (2018GZR110103002).

References 1. Friedman, J.R. and Nunnari, J. (2014) Mitochondrial form and 28. Yan, L. et al. (2020) Structural analysis of a trimeric assembly of function. Nature 505, 335–343 the mitochondrial dynamin-like GTPase Mgm1. Proc. Natl. 2. Cerveny, K.L. et al. (2007) Regulation of mitochondrial fusion Acad. Sci. U. S. A. 117, 4061–4070 and division. Trends Cell Biol. 17, 563–569 29. Yu, C. et al. (2020) Structural insights into G domain dimeriza- 3. Chan, D.C. (2012) Fusion and fission: interlinked processes tion and pathogenic mutation of OPA1. J. Cell Biol. 219, critical for mitochondrial health. Annu. Rev. Genet. 46, 265–2687 e201907098 4. Mishra,P.andChan,D.C.(2016)Metabolicregulationof 30. Zhang, D. et al. (2020) Cryo-EM structures of S-OPA1 reveal mitochondrial dynamics. J. Cell Biol. 212, 379–387 its interactions with membrane and changes upon nucleotide 5. Wai, T. and Langer, T. (2016) Mitochondrial dynamics and binding. eLife 9, e50294 metabolic regulation. Trends Endocrinol. Metab. 27, 105–117 31. Praefcke, G.J. and McMahon, H.T. (2004) The dynamin 6. Youle, R.J. and van der Bliek, A.M. (2012) Mitochondrial fission, superfamily: universal membrane tubulation and fission fusion, and stress. Science 337, 1062–1065 molecules? Nat. Rev. Mol. Cell Biol. 5, 133–147 7. Labbe, K. et al. (2014) Determinants and functions of mitochon- 32. Antonny, B. et al. (2016) Membrane fission by dynamin: what we drial behavior. Annu. Rev. Cell Dev. Biol. 30, 357–391 know and what we need to know. EMBO J. 35, 2270–2284 8. Chan, D.C. (2020) Mitochondrial dynamics and its involvement 33. Sprenger, H.G. and Langer, T. (2019) The good and the bad of in disease. Annu. Rev. Pathol. 15, 235–259 mitochondrial breakups. Trends Cell Biol. 29, 888–900 9. Itoh, K. et al. (2013) Mitochondrial dynamics in neurodegeneration. 34. Kim, B.H. et al. (2012) IFN-inducible GTPases in host cell Trends Cell Biol. 23, 64–71 defense. Cell Host Microbe 12, 432–444 10. Cipolat, S. et al. (2004) OPA1 requires mitofusin 1 to promote mito- 35. Haller, O. et al. (2015) Mx GTPases: dynamin-like antiviral chondrial fusion. Proc.Natl.Acad.Sci.U.S.A.101, 15927–15932 machines of innate immunity. Trends Microbiol. 23, 154–163 11. Song, Z. et al. (2009) Mitofusins and OPA1 mediate sequential 36. Moss, T.J. et al. (2011) Fusing a lasting relationship between ER steps in mitochondrial membrane fusion. Mol. Biol. Cell 20, tubules. Trends Cell Biol. 21, 416–423 3525–3532 37. Naslavsky, N. and Caplan, S. (2011) EHD proteins: key conduc- 12. Hermann, G.J. et al. (1998) Mitochondrial fusion in yeast requires tors of endocytic transport. Trends Cell Biol. 21, 122–131 the transmembrane GTPase Fzo1p. J. Cell Biol. 143, 359–373 38. Lomash, R.M. et al. (2015) Neurolastin, a dynamin family 13. Santel, A. and Fuller, M.T. (2001) Control of mitochondrial GTPase, regulates excitatory synapses and spine density. Cell morphology by a human mitofusin. J. Cell Sci. 114, 867–874 Rep. 12, 743–751 14. Chen, H. et al. (2003) Mitofusins Mfn1 and Mfn2 coordinately 39. Low, H.H. and Löwe, J. (2006) A bacterial dynamin-like protein. regulate mitochondrial fusion and are essential for embryonic Nature 444, 766–769 development. J. Cell Biol. 160, 189–200 40. Prakash, B. et al. (2000) Structure of human guanylate-binding 15. Eura, Y. et al. (2003) Two mitofusin proteins, mammalian homo- protein 1 representing a unique class of GTP-binding proteins. logues of FZO, with distinct functions are both required for mito- Nature 403, 567–571 chondrial fusion. J. Biochem. 134, 333–344 41. Daumke, O. et al. (2007) Architectural and mechanistic insights 16. Detmer, S.A. and Chan, D.C. (2007) Complementation between into an EHD ATPase involved in membrane remodelling. Nature mouse Mfn1 and Mfn2 protects mitochondrial fusion defects 449, 923–927 caused by CMT2A disease mutations. J. Cell Biol. 176, 405–414 42. Chappie, J.S. et al. (2010) G domain dimerization controls 17. Züchner, S. et al. (2004) Mutations in the mitochondrial GTPase dynamin's assembly-stimulated GTPase activity. Nature 465, mitofusin 2 cause Charcot-Marie-Tooth neuropathy type 2A. 435–440 Nat. Genet. 36, 449–451 43. Bian, X. et al. (2011) Structures of the atlastin GTPase provide 18. Pelloquin, L. et al. (1998) Identification of a fission yeast dynamin- insight into homotypic fusion of endoplasmic reticulum related protein involved in mitochondrial DNA maintenance. membranes. Proc. Natl. Acad. Sci. U. S. A. 108, 3976–3981 Biochem. Biophys. Res. Commun. 251, 720–726 44. Byrnes, L.J. and Sondermann, H. (2011) Structural basis for 19. Shepard, K.A. and Yaffe, M.P. (1999) The yeast dynamin-like the nucleotide-dependent dimerization of the large G protein protein, Mgm1p, functions on the mitochondrial outer mem- atlastin-1/SPG3A. Proc. Natl. Acad. Sci. U. S. A. 108, brane to mediate mitochondrial inheritance. J. Cell Biol. 144, 2216–2221 711–720 45. Faelber, K. et al. (2011) Crystal structure of nucleotide-free 20. Eiberg, H. et al. (1994) Dominant optic atrophy (OPA1) mapped dynamin. Nature 477, 556–560 to chromosome 3q region. I. Linkage analysis. Hum. Mol. Genet. 46. Ford, M.G. et al. (2011) The crystal structure of dynamin. Nature 3, 977–980 477, 561–566 21. Wong, E.D. et al. (2000) The dynamin-related GTPase, Mgm1p, 47. Gao, S. et al. (2011) Structure of myxovirus resistance protein a is an intermembrane space protein required for maintenance of reveals intra- and intermolecular domain interactions required for fusion competent mitochondria. J. Cell Biol. 151, 341–352 the antiviral function. Immunity 35, 514–525 22. Alexander, C. et al. (2000) OPA1, encoding a dynamin-related 48. Frohlich, C. et al. (2013) Structural insights into oligomerization GTPase, is mutated in autosomal dominant optic atrophy linked and mitochondrial remodelling of dynamin 1-like protein. to chromosome 3q28. Nat. Genet. 26, 211–215 EMBO J. 32, 1280–1292 23. Qi, Y. et al. (2016) Structures of human mitofusin 1 provide 49. Kalia, R. et al. (2018) Structural basis of mitochondrial receptor insight into mitochondrial tethering. J. Cell Biol. 215, 621–629 binding and constriction by DRP1. Nature 558, 401–405 24. Cao, Y.L. et al. (2017) MFN1 structures reveal nucleotide- 50. Jimah, J.R. and Hinshaw, J.E. (2019) Structural insights into the triggered dimerization critical for mitochondrial fusion. Nature mechanism of dynamin superfamily proteins. Trends Cell Biol. 542, 372–376 29, 257–273 25. Yan, L. et al. (2018) Structural basis for GTP hydrolysis and con- 51. Gao, S. et al. (2010) Structural basis of oligomerization in the formational change of MFN1 in mediating membrane fusion. Nat. stalk region of dynamin-like MxA. Nature 465, 502–506 Struct. Mol. Biol. 25, 233–243 52. Daumke, O. and Praefcke, G.J. (2016) Invited review: Mechanisms 26. Li, Y.J. et al. (2019) Structural insights of human mitofusin-2 into of GTP hydrolysis and conformational transitions in the dynamin mitochondrial fusion and CMT2A onset. Nat. Commun. 10, 4914 superfamily. Biopolymers 105, 580–593 27. Faelber, K. et al. (2019) Structure and assembly of the mito- 53. Faelber, K. et al. (2013) Oligomerization of dynamin superfamily chondrial membrane remodelling GTPase Mgm1. Nature 571, proteins in health and disease. Prog. Mol. Biol. Transl. Sci. 429–433 117, 411–443

Trends in Cell Biology, January 2021, Vol. 31, No. 1 73 Trends in Cell Biology

54. Roux, A. et al. (2006) GTP-dependent twisting of dynamin 76. Hoppins, S. et al. (2011) The soluble form of Bax regulates mito- implicates constriction and tension in membrane fission. Nature chondrial fusion via MFN2 homotypic complexes. Mol. Cell 41, 441, 528–531 150–160 55. von der Malsburg, A. et al. (2011) Stalk domain of the dynamin- 77. Liu, J. et al. (2018) Structural basis for membrane tethering by a like MxA GTPase protein mediates membrane binding and bacterial dynamin-like pair. Nat. Commun. 9, 3345 liposome tubulation via the unstructured L4 loop. J. Biol. 78. Meeusen, S. et al. (2006) Mitochondrial inner-membrane fusion Chem. 286, 37858–37865 and crista maintenance requires the dynamin-related GTPase 56. Mears, J.A. et al. (2011) Conformational changes in Dnm1 sup- Mgm1. Cell 127, 383–395 port a contractile mechanism for mitochondrial fission. Nat. 79. Westermann, B. (2010) Mitochondrial fusion and fission in cell life Struct. Mol. Biol. 18, 20–26 and death. Nat. Rev. Mol. Cell Biol. 11, 872–884 57. Mattila, J.P. et al. (2015) A hemi-fission intermediate links two 80. McQuibban, G.A. et al. (2003) Mitochondrial membrane remod- mechanistically distinct stages of membrane fission. Nature elling regulated by a conserved rhomboid protease. Nature 423, 524, 109–113 537–541 58. Chen, Y. et al. (2017) Conformational dynamics of dynamin-like 81. Ishihara, N. et al. (2006) Regulation of mitochondrial morphology MxA revealed by single-molecule FRET. Nat. Commun. 8, 15744 through proteolytic cleavage of OPA1. EMBO J. 25, 2966–2977 59. Kong, L. et al. (2018) Cryo-EM of the dynamin polymer assem- 82. Griparic, L. et al. (2007) Regulation of the mitochondrial bled on lipid membrane. Nature 560, 258–262 dynamin-like protein Opa1 by proteolytic cleavage. J. Cell Biol. 60. Koshiba, T. et al. (2004) Structural basis of mitochondrial 178, 757–764 tethering by mitofusin complexes. Science 305, 858–862 83. Song, Z. et al. (2007) OPA1 processing controls mitochondrial 61. Ishihara, N. et al. (2004) Mitofusin 1 and 2 play distinct roles in fusion and is regulated by mRNA splicing, membrane potential, mitochondrial fusion reactions via GTPase activity. J. Cell Sci. and Yme1L. J. Cell Biol. 178, 749–755 117, 6535–6546 84. Head, B. et al. (2009) Inducible proteolytic inactivation of OPA1 62. Knott, A.B. et al. (2008) Mitochondrial fragmentation in mediated by the OMA1 protease in mammalian cells. J. Cell neurodegeneration. Nat. Rev. Neurosci. 9, 505–518 Biol. 187, 959–966 63. Byrnes, L.J. et al. (2013) Structural basis for conformational 85. Ehses, S. et al. (2009) Regulation of OPA1 processing and mito- switching and GTP loading of the large G protein atlastin. chondrial fusion by m-AAA protease isoenzymes and OMA1. EMBO J. 32, 369–384 J. Cell Biol. 187, 1023–1036 64. Huang, X. et al. (2017) Sequences flanking the transmembrane 86. Ban, T. et al. (2010) OPA1 disease alleles causing dominant optic at- segments facilitate mitochondrial localization and membrane fusion rophy have defects in cardiolipin-stimulated GTP hydrolysis and by mitofusin. Proc. Natl. Acad. Sci. U. S. A. 114, E9863–E9872 membrane tubulation. Hum. Mol. Genet. 19, 2113–2122 65. Low, H.H. et al. (2009) Structure of a bacterial dynamin-like 87. Rujiviphat, J. et al. (2015) Mitochondrial genome maintenance 1 protein lipid tube provides a mechanism for assembly and (Mgm1) protein alters membrane topology and promotes local membrane curving. Cell 139, 1342–1352 membrane bending. J. Mol. Biol. 427, 2599–2609 66. Franco, A. et al. (2016) Correcting mitochondrial fusion by 88. DeVay, R.M. et al. (2009) Coassembly of Mgm1 isoforms re- manipulating mitofusin conformations. Nature 540, 74–79 quires cardiolipin and mediates mitochondrial inner membrane 67. Brandt, T. et al. (2016) A mitofusin-dependent docking ring fusion. J. Cell Biol. 186, 793–803 complex triggers mitochondrial fusion in vitro. eLife 5, e14618 89. Meglei, G. and McQuibban, G.A. (2009) The dynamin-related 68. McMahon, H.T. and Boucrot, E. (2015) Membrane curvature at a protein Mgm1p assembles into oligomers and hydrolyzes GTP glance. J. Cell Sci. 128, 1065–1070 to function in mitochondrial membrane fusion. Biochemistry 48, 69. Daste, F. et al. (2018) The heptad repeat domain 1 of Mitofusin 1774–1784 has membrane destabilization function in mitochondrial fusion. 90. Warnock, D.E. et al. (1995) Dynamin GTPase is stimulated by EMBO Rep. 19, e43637 crosslinking through the C-terminal proline-rich domain. EMBO 70. Liu, T.Y. et al. (2012) Lipid interaction of the C terminus and J. 14, 1322–1328 association of the transmembrane segments facilitate atlastin- 91. Hinshaw, J.E. and Schmid, S.L. (1995) Dynamin self-assembles mediated homotypic endoplasmic reticulum fusion. Proc. Natl. into rings suggesting a mechanism for coated vesicle budding. Acad. Sci. U. S. A. 109, E2146–E2154 Nature 374, 190–192 71. Faust, J.E. et al. (2015) The Atlastin C-terminal tail is an 92. Delettre, C. et al. (2000) Nuclear gene OPA1, encoding a mito- amphipathic helix that perturbs the bilayer structure during en- chondrial dynamin-related protein, is mutated in dominant optic doplasmic reticulum homotypic fusion. J. Biol. Chem. 290, atrophy. Nat. Genet. 26, 207–210 4772–4783 93. Wang, M. et al. (2019) Mycobacterial dynamin-like protein IniA 72. Anton, V. et al. (2019) Plasticity in salt bridge allows fusion- mediates membrane fission. Nat. Commun. 10, 3906 competent ubiquitylation of mitofusins and Cdc48 recognition. 94. Ge, Y. et al. (2020) Two forms of Opa1 cooperate to complete Life Sci. Alliance 2, e201900491 fusion of the mitochondrial inner-membrane. eLife 9, e50973 73. Schuster, R. et al. (2020) Dual role of a GTPase conformational 95. Ban, T. et al. (2017) Molecular basis of selective mitochondrial switch for membrane fusion by mitofusin ubiquitylation. Life fusion by heterotypic action between OPA1 and cardiolipin. Sci. Alliance 3, e201900476 Nat. Cell Biol. 19, 856–863 74. Chen,Y.andDornII,G.W.(2013)PINK1-phosphorylated 96. Joshi, A.S. et al. (2012) Cardiolipin and mitochondrial phosphatidyl- mitofusin 2 is a Parkin receptor for culling damaged mitochondria. ethanolamine have overlapping functions in mitochondrial fusion in Science 340, 471–475 Saccharomyces cerevisiae. J. Biol. Chem. 287, 17589–17597 75. Sloat, S.R. et al. (2019) Identification of a mitofusin specificity 97. Hales, K.G. and Fuller, M.T. (1997) Developmentally regulated region that confers unique activities to Mfn1 and Mfn2. Mol. mitochondrial fusion mediated by a conserved, novel, predicted Biol. Cell 30, 2309–2319 GTPase. Cell 90, 121–129

74 Trends in Cell Biology, January 2021, Vol. 31, No. 1