OPA1 helical structures give perspective to mitochondrial dysfunction

Dominant optic atrophy is one of the leading causes of childhood blindness. Around 60–80% of cases1 are caused by mutations of the gene that encodes optic atrophy protein 1 (OPA1), a protein that has a key role in inner mitochondrial membrane fusion and remodelling of cristae and is crucial for the dynamic organization and regulation of mitochondria2. Mutations in OPA1 result in the dysregulation of the GTPase-mediated fusion process of the mitochondrial inner and outer membranes3. Here we used cryo-electron microscopy methods to solve helical structures of OPA1 assembled on lipid membrane tubes, in the presence and absence of nucleotide. These helical assemblies organize into densely packed protein rungs with minimal inter-rung connectivity, and exhibit nucleotide-dependent dimerization of the GTPase domains—a hallmark of the dynamin superfamily of proteins4. OPA1 also contains several unique secondary structures in the paddle domain that strengthen its membrane association, including membrane-inserting helices. The structural features identified in this study shed light on the effects of pathogenic point mutations on protein folding, inter-protein assembly and membrane interactions. Furthermore, mutations that disrupt the assembly interfaces and membrane binding of OPA1 cause mitochondrial fragmentation in cell-based assays, providing evidence of the biological relevance of these interactions. Cryo-electron microscopy structures of OPA1, mutations of which are associated with the disease dominant optic atrophy, provide insight into how structural features of OPA1 enable this protein to mediate mitochondrial-membrane fusion and remodelling.

Dominant optic atrophy is one of the leading causes of childhood blindness. Around 60-80% of cases 1 are caused by mutations of the gene that encodes optic atrophy protein 1 (OPA1), a protein that has a key role in inner mitochondrial membrane fusion and remodelling of cristae and is crucial for the dynamic organization and regulation of mitochondria 2 . Mutations in OPA1 result in the dysregulation of the GTPasemediated fusion process of the mitochondrial inner and outer membranes 3 . Here we used cryo-electron microscopy methods to solve helical structures of OPA1 assembled on lipid membrane tubes, in the presence and absence of nucleotide. These helical assemblies organize into densely packed protein rungs with minimal inter-rung connectivity, and exhibit nucleotide-dependent dimerization of the GTPase domains-a hallmark of the dynamin superfamily of proteins 4 . OPA1 also contains several unique secondary structures in the paddle domain that strengthen its membrane association, including membrane-inserting helices. The structural features identified in this study shed light on the effects of pathogenic point mutations on protein folding, inter-protein assembly and membrane interactions. Furthermore, mutations that disrupt the assembly interfaces and membrane binding of OPA1 cause mitochondrial fragmentation in cell-based assays, providing evidence of the biological relevance of these interactions.
OPA1 is the most frequently mutated gene in the disease dominant optic atrophy (DOA), a major cause of childhood blindness 1,[17][18][19][20] . The LOVD database at present has more than 640 unique mutations of OPA1, and pathogenic mutations can cause progressive vision loss, hearing loss, ataxia and myopathy 1,[17][18][19][20] . Molecularly, this results from mitochondrial dysfunction, including fragmentation of the mitochondrial network, loss of mitochondrial DNA and loss of respiratory function 15,17,20 . The urgent need to develop a treatment for DOA has attracted interest in OPA1, but the mechanics of its roles in membrane remodelling remain elusive 2,14,15,21 . Previous structural studies of OPA1 have suggested similarities with other DSPs, but insight from these studies on pathological OPA1 mutations is limited by the use of shortened constructs, or by the absence of a membrane environment that facilitates the formation of higher-order assemblies 22,23 .
The eight isoforms of OPA1 all contain an N-terminal S1 cleavage site for the protease OMA1, which responds to membrane depolarization. Four isoforms contain a second S2 cleavage site for YME1L, coupled to mitochondrial respiration, and one isoform contains a third S3 cleavage site for YME1L (refs. 14-16,24) ( Fig. 1b and Supplementary Fig. 2). These cleavage events result in a dynamic ratio of l-OPA1 and s-OPA1, which regulates mitochondrial morphology and function 14,15,25 . During fusion, the two forms of OPA1 must bring together two IMM leaflets from distinct mitochondria 5,15,25 .
Two X-ray structures have been solved for the s-OPA1 homologue (s-MGM1) from Chaetomium thermophilum and Saccharomyces cerevisiae 26,27 . As in other DSPs, the C. thermophilum structure is tetrameric, with similar protein-protein interfaces 26,28 (Protein Data Bank (PDB) ID: 6QL4). The S. cerevisiae structure captured s-MGM1 as a trimer in the presence of guanosine diphosphate 27 (GDP) (PDB ID: 6JSJ). However, the low (around 30%) sequence identity of MGM1 with OPA1 limits the extent to which structures of MGM1 can inform on the human pathologies associated with DOA 26,27 . There are also low-resolution cryo-electron microscopy (cryo-EM) structures of s-MGM1 and an Article N-terminally cleaved OPA1 construct (ΔN-OPA1, residues 253-960), solved as helical assemblies 22,26 . Notably, s-MGM1 assembled on both the outside and inside of lipid tubes 26 , in which the inner helical lattice presents a possible mechanism for remodelling of cristae 26 (PDB ID: 6RZV and 6RZW). The outer s-MGM1 and ΔN-OPA1 helical assemblies were solved for two states (apo and GTPγS) at a low resolution (around 15-23 Å). The dimerization of the GTPase domain that is typical of DSPs was not observed in either of these structures 25,26 , although a nucleotide-dependent GTPase domain dimer was seen for a minimal GTPase-bundle signaling element (BSE) construct in the presence of GDP-BeF x (ref. 23) (PDB ID: 6JTG).

Cryo-EM reveals s-OPA1 interfaces
Here we present two cryo-EM structures of human s-OPA1 assembled around a cardiolipin (CL)-containing lipid bilayer (DOPS:CL) in the absence and presence of nucleotide (GDP-AlF x ) ( Fig. 1c-g and Extended Data Fig. 1). The full helical maps were resolved to 5.48 Å with nucleotide and 9.68 Å without (Fig. 1d,f and Extended Data Fig. 1a,b). Z-clipping to 50% of the full map improved the resolution to 3.86 Å and 5.8 Å, respectively (Extended Data Fig. 1c-f). We further improved the resolution using local non-uniform refinements 29,30 (Extended Data Fig. 2 and Supplementary Fig. 3) to 3.1-6.5 Å, from which we built models of s-OPA1 (Fig. 1c,e,g and Extended Data Figs. 3 and 4). The observed domain organization of s-OPA1 parallels other DSPs with GTPase, BSE and stalk domains 2,23,28 (Fig. 1b,c). However, s-OPA1 has a uniquely kinked stalk 26 , an N-terminal coiled-coil juxta-membrane linker and a cardiolipin-specific paddle domain 20,31 . Using these models, we could locate DOA-associated point mutations and hypothesize how clinically significant mutations disrupt the function of OPA1 ( Fig. 1c and Extended Data Table 1).
The s-OPA1 assemblies have outer diameters of 38.2 nm and 40.5 nm for the nucleotide-bound and nucleotide-free states, respectively, whereas the inner lumens are nearly identical in both states (around 9.4 versus 9.5 nm) (Fig. 1d,f and Extended Data Fig. 1a,b). From these structures, we were able to define oligomeric and membrane-binding interfaces, identify the canonical nucleotide-dependent dimerization of the GTPase domains and observe an inter-rung N-terminal coiled-coil interface (Figs. 1e,g and 2). Although the s-OPA1 maps have substantial flexibility, owing to minimal inter-rung connectivity, as evidenced by the inherent curvature of the s-OPA1 tubes, they also show dense intra-rung packing (Extended Data Fig. 2a and Supplementary  Fig. 3h,p).
Local refinements also improved the resolution of interfaces that define the oligomerization of s-OPA1. The building block of s-OPA1 helical assemblies is a dimer, which is defined by three interfaces (interfaces 1-3) ( Fig. 1e and Extended Data Figs. 3 and 4a). A pair of dimers then assembles into a tetramer through a paddle-paddle contact, interface 4, and a smaller interface near the BSE domain (interface 9) (Figs. 1g and 2a,f and Extended Data Fig. 4b,g), with successive tetramers forming the full biological rung (Supplementary Fig. 3h,p). In the nucleotide-bound state, the helical assembly is further stabilized by the GTPase domain dimer (interface 5) and GTPase-stalk contact (interface 6) within the biological rung (Fig. 2b,c, Extended Data Fig. 4c,d and Supplementary Fig. 3h). Finally, two inter-rung contacts involve the paddle (interface 7) and N-terminal (interface 8) domains (Fig. 2d,e and Extended Data Fig. 4e,f), further stabilizing the oligomerization of s-OPA1.
Although the s-OPA1 helix is overall similar to other DSPs, there are unique features across each stage of assembly. In the dimer, the stalk-dimer interface (interface 1) occurs along the opposite face of the stalk's four-helix bundle compared to s-MGM1 (ref. 26). This results in a compressed, W-shaped interface, owing to the tightly interlocked, kinked stalk (interface 1; Fig. 1e,g and Extended Data Figs. 3a,d and 5a). Furthermore, the position of interface 1 within the s-OPA1 GDP-AlF x dimer causes the buried surface area between the stalk and the paddle domains to extend into two additional interfaces: between the stalk and paddle (interface 2), and between two Pα5 paddle helices 32 (interface 3) (Fig. 1e Table 1). After nucleotide binding, the helical assembly compacts owing to dimerization of the GTPase domain (Supplementary Video 1). This GTPase-dimer interface (interface 5) is the largest interface in the s-OPA1 oligomer and resides within a rung, whereas in other DSPs the GTPase dimer is between rungs (Fig. 2b, Extended Data Fig. 4c and Supplementary Fig. 3h). Formation of interface 5 involves the GTPase domain and BSE swinging 25-27° along hinge 1, between the BSE and stalk domains (Fig. 3a,b). The GTPase interface and nucleotide-binding pocket resemble the crystal structure of the minimal OPA1 GTPase domain dimer bound to GDP-BeF x (PDB ID: 6JTG; root-mean-square deviation (RMSD) = 2.995 Å), but with increased asymmetry along the dimer interface 23 (Extended Data Fig. 5b,c). Incorporation into the helical assembly also results in a small swing of the BSE region, which is less resolved, probably owing to heterogeneity or pull from the N-terminal region. Consistent with this, mapping the per-particle heterogeneity of both assemblies in cryoDRGN showed continuous transitions within the N-terminal and GTPase-dimer regions 30 (Supplementary Fig. 3r-t and Supplementary Video 2). Relative to the crystal structure, the BSE helices of the GDP-AlF x -bound assembly are offset by 30.31° in the A-chain and 22.26° in the B-chain (Extended Data Fig. 5c). These shifts in the BSE suggest that potential energy is stored in the assembled state to assist in the eventual disassembly after GTP hydrolysis, as shown by our supernatant and pellet assay (Extended Data Fig. 2b).
Previous structural studies of dynamin observed the swing of the BSE domain along hinge 2, referred to as the powerstroke, during the GTPase hydrolysis cycle 33 . This is in contrast to the swing we observed in s-OPA1 around hinge 1 after nucleotide binding. In the proposed dynamin powerstroke, the BSE domain is in the up position in the GTP-bound state and swings down after hydrolysis 33 (Extended Data Fig. 5d). The s-OPA1 BSE domain is in the down position relative to the GTPase domain, similar to dynamin in the presence of GDP-AlF x (ref. 34). The positioning of the BSE in a pre-powerstroke or up position has not been observed for OPA1 or MGM1, even in the presence of GTPγS, a GTP-bound state 22,26 .
In the paddle domain, interface 7 provides a rare inter-rung contact through Pα4 ( Fig. 2d and Extended Data Figs. 3a,d and 4e), with a similar buried surface area in the two states 32 (Supplementary Table 1). Notably, in the GDP-AlF x -bound state, we also resolved a second inter-rung interface (interface 8) that resides above interface 7 and consists of a coiled-coil domain assembled from the N-terminal α-helix, Nα1 ( Fig. 2e and Extended Data Fig. 4f). This region is unique to s-OPA1 homologues and has not been previously resolved in other ΔN-OPA1 structures. In addition, despite its inherent flexibility, we found that the N-terminal domain also interacts with the BSE domain 32 (interface 9) (  Table 1). We also observe electron density for the N-terminal region in the apo state, and, given that it is known to dimerize in the absence of nucleotide, we expect this region still to have a major role in the apo assembly 22 . Overall, the decreased buried surface area and fewer interfaces of the apo state show that s-OPA1 is more loosely packed before nucleotide addition, and correlate with increased variance in apo-state diameters, explaining the reduced resolution of the apo structure 32 (Extended Data Fig. 2b and Supplementary Table 1).

The s-OPA1 paddle is primed to bind the IMM
Our s-OPA1 structure reveals a unique lipid-binding motif in the paddle domain that has evolved to include both membrane-proximal and monotopic, membrane-inserting helices (Fig. 3c,d). These features result in more extensive membrane contact and insertion than in s-MGM1 (ref. 26). In both s-OPA1 and s-MGM1, the paddle domain connects to the stalk through hinge 3, which includes a conserved disulfide bond that defines the paddle hinge (OPA1 residues C856-C874) (Extended Data Fig. 6b,d). However, the paddle hinge of s-OPA1 is extended by the membrane-inserting Pα6 helix, with three more positively charged or aromatic residues 26 (Extended Data Fig. 6b-d). In addition, in the paddle tip along the opposite surface of the paddle, the turn between the Pα1 and Pα3 helices is extended in s-OPA1 by a flexible  Article membrane-inserting loop that contains the Pα2 helix, which has nine extra charged or aromatic residues. In the GDP-AlF x state, Pα2 inserts up to 16 Å into the bilayer and appears to interact with heterogeneous cardiolipin density (Fig. 3b-d and Extended Data Fig. 6c-j). s-OPA1 also has two central helices, Pα1 and Pα5, which are interfacial and similar in length to s-MGM1, but OPA1 has five more positively charged residues (Extended Data Fig. 6c,d). These extended helices better align the s-OPA1 paddle with the membrane surface. Finally, s-OPA1 has a Pα4 helix that forms interface 7 with a neighbouring paddle domain and is anchored to the membrane through residue R824 ( Fig. 2d and Extended Data Fig. 6c). s-MGM1 might compensate for the lack of Pα4 through the presence of an extended loop projecting out of the stalk, with five positively charged or aromatic residues 26 (glycine-rich loop; Extended Data Fig. 6b). Overall, these differences in paddle composition and structure suggest that the OPA1 paddle has evolved to enhance interactions with the IMM, through an extended paddle domain that supports a higher degree of hydrophobic insertion.
Comparing the apo and GDP-AlF x states of s-OPA1, the paddle domains are more closely associated to the membrane in the nucleotide-bound case, with more insertion of the paddle-tip (Pα2-4) and paddle-hinge (Pα6) helices (Fig. 3c,d). With nucleotide, the Pα2 helix is embedded 16 Å into the bilayer, 5 Å more than the average insertion in the apo state (around 11 Å). Furthermore, the Pα1, Pα3 and Pα5 helices rest on the membrane surface in the GDP-AlF x -bound case, but are shallowly associated in the apo state (Fig. 3c,d). We propose that the increased membrane interactions after nucleotide binding are driven by the dimerization of the GTPase domain, leading to a more compact structure that forces the paddle deeper into the membrane.
This insertion is most marked for the membrane-inserting loop with Pα2, which appears to have a crucial role in membrane curvature remodelling and destabilization for fusion. As with other monotopic proteins, embedding the Pα2 helix displaces lipid molecules in the outer leaflet of the bilayer, driving the positive curvature of the bilayer and helical assembly 35,36 . To enhance the resolution of the paddlemembrane interaction, we performed local refinements focusing on a section of the outer leaflet of the bilayer. This improved the resolution around the inserted Pα2 helix (Extended Data Fig. 6e-j). Altering the threshold of the resulting maps also revealed conical, wedge-like densities that we propose are heterogeneous cardiolipin colocalization around the membrane-inserting loop and Pα2 helix 37,38 (Extended Data Fig. 6e-j). These features are located at the protein interface within the outer leaflet of the bilayer and allow for the charged and hydrophobic regions of the flexible cardiolipin molecule to align with the paddle (Extended Data Fig. 6g-j). A similar mode of cardiolipin binding has been observed or proposed in other cardiolipin-binding proteins [38][39][40][41][42] and is supported by the flexibility of Pα2 and its enrichment in positively charged, hydrophobic, and aromatic residues (Extended Data Fig. 6d,j and Supplementary Video 2). Residues near the modelled cardiolipin include T782, R774, K779, N780, W775 and W778.

Electrostatics in s-OPA1 assembly
Electrostatics seem to have a major role in the two s-OPA1 assemblies compared to other DSPs (Extended Data Fig. 7a-f). Putatively, the initial attraction between s-OPA1 and the cardiolipin-containing membrane is aided by both electrostatic contact and hydrophobic insertion (Extended Data Fig. 7a-l). The Pα1, Pα3 and Pα5 helices are largely positively charged (blue), which would be interfacial to the negatively charged lipid headgroups of the bilayer (Extended Data Fig. 7b). The hydrophobic (gold), aromatic-residue-rich helices (Pα2 and Pα6) would then be stabilized by insertion into the lipid tails of the membrane (Extended Data Fig. 7h). Overall, the OPA1 paddle has a greater propensity for charged and hydrophobic membrane interactions, and this is likely to optimize its affinity for the IMM (Extended Data Figs. 6d,h-j and 7).   The s-OPA1 dimer and tetramer interfaces also have large contributions from aligned electrostatic potential. The s-OPA1 dimer is striated in positive and negative charge and the alignment of interface 1 results in alternating charges throughout the helical bundle, electrostatically stitching together in-register interfaces while contributing to heterogeneity when they fall out of register (Extended Data Fig. 7c). This results in a large buried surface area throughout the s-OPA1 dimer in the GDP-AlF x -bound state (1,330 Å 2 ), which is considerably reduced in the apo state (1,014 Å 2 ) owing to a looser interface 1 between stalk domains. The looser interface in the apo state reduces the in-register charge within interface 1 (Extended Data Fig. 7d) and results in approximately half the buried surface area (433 Å 2 compared to 743 Å 2 ) and a wider stalk dimer (48 Å) than in the GDP-AlF x state (38 Å) (Extended Data Fig. 3a,d and Supplementary Table 1). There is very little change in buried surface area between the paddle and the stalk (interface 2), but the interface between Pα5 helices (interface 3) is substantially smaller in the apo state (40 Å 2 versus 90 Å 2 ), demonstrating the tighter packing when nucleotide is bound (intra-rung GDP-AlF x interfaces bury 4,278 Å 2 ; apo interfaces bury 1,493 Å 2 ) (Extended Data Fig. 3a,d and Supplementary Table 1). The inter-rung interfaces, in particular interface 7 between the paddles and interface 8 between the N-terminal domains, also seem to have electrostatic contributions that are in better register for the GDP-AlF x -bound state (inter-rung GDP-AlF x interfaces bury 1,085 Å 2 ; apo interfaces bury 756 Å 2 ) (Extended Data Fig. 7e,f). By contrast, hydrophobic burial does not appear to have such a strong effect on assembly (Extended Data Fig. 7i-l), with the exception of tetramer interface 4 (Extended Data Fig. 7h).

DOA-associated mutations in s-OPA1 assemblies
The interfaces and membrane-binding surfaces identified within our s-OPA1 structures allow us to predict the molecular basis of how 86 mutants identified in patients with DOA lead to disease (Extended Data Fig. 8 and Supplementary Tables 1 and 2). Of the 238 sites with point mutations (Fig. 1c) in OPA1, 37 have multiple substitutions, resulting in 284 mutations that are expected to disrupt OPA1 function 1,18-20 . As expected, many (121) are in the GTPase domain 19 , with the rest scattered throughout the structure. In the monomer, 84 mutations are within the hinge regions and could disturb protein folding, oligomerization and the putative powerstroke. There are 101 DOA-associated mutations that disrupt charged residues, 39 proline or glycine helix breakers and 15 that introduce cysteines. For s-OPA1 GDP-AlF x , most sites (187; 87%) are surface-exposed, with 86 in membrane-binding sites or assembly interfaces identified in this paper. Of these, 65 are in the protein-protein interfaces that are involved in helical assembly, 17 are membrane facing and could disrupt lipid binding and 21 are likely to be involved in GTP hydrolysis. Of note, 24 more DOA mutations are just 1 or 2 amino acids away from our tightly defined interfaces. These observations show the importance of membrane binding and helical assembly in OPA1 pathogenicity.

Assembly mutants fragment mitochondria
To determine the cellular effect of disrupting the s-OPA1 helical interfaces and membrane-binding regions, we transfected OPA1-GFP mutants into HeLa-M cells and characterized their mitochondrial morphology using a single-cell-based assay ( Fig. 4 and Extended Data Figs. 9 and 10). In total, we assayed 14 interfacial and membrane-binding mutants that span 8 interfaces and the Pα2 (in the membrane-inserting loop) and Pα6 helices. Of these, 13 included known atrophy mutants (bold in Extended Data Fig. 9). We also assayed four control mutations that lie outside the identified interfaces within the N-terminal, GTPase and stalk domains. We hypothesized that the expression of even small amounts of GFP-tagged OPA1 mutants in wild-type cells would alter mitochondrial morphology by poisoning the function of   . Notably, quantification of the expression of OPA1-GFP in every cell scored for mitochondrial phenotype showed that there was no significant variation in expression between wild-type OPA1 and mutant versions of OPA1, indicating that our results are not confounded by possible differences in expression level (Extended Data Fig. 10d-g). The OPA1 mutants causing the most and the least fragmented mitochondria were the stalk-stalk dimer interface mutant (interface 1, I1e) and the interface 4 mutant (I4b), respectively. This latter result could be due to the apparent heterogeneity of interface 4. Mutations in the novel s-OPA1 N-terminal interfaces (mutants I8 and I9) produced substantial fragmentation, implying that N-terminal contacts are essential for stabilizing a fusion-competent assembly. The presence of strongly fragmented mitochondria for all tested interfaces, pertinent to all stages of assembly from dimer to higher-order intra-and inter-rung interfaces, underlies the apparent importance of OPA1 helical assemblies in vivo. With regard to membrane binding, both membrane-inserting loop (Pα2) and Pα6 mutants fragmented the mitochondrial network, highlighting the interactions that occur between OPA1 and the inner mitochondrial membrane during fusion. Notably, one of the most fragmented phenotypes was seen for Pα2, possibly owing to its direct interaction with cardiolipin. Further more, the presence of atrophy mutations within the identified s-OPA1 interfaces and membrane-binding regions supports the assertion that many atrophy mutations arise from the disruption of OPA1 assemblies.
Although the fragmentation phenotypes exhibited by cells expressing OPA1 with interfacial or membrane-binding mutations could be caused by defects in the assembly of OPA1 on membranes, they could also be caused by defects in OPA1 processing or mitochondrial cristae organization. The ratio of long to short OPA1-GFP isoforms present in transfected cells (around 0.78:1) was comparable to the ratio for endogenous OPA1 in the same samples (that is, comparable to the OPA1 present in both transfected and non-transfected cells; around 0.8:1), and to the ratio in the non-transfected control (around 0.8:1) (Extended Data Fig. 10c). These results, together with the fact that the transfection efficiencies for all the mutants tested by western blotting were very similar (Extended Data Fig. 10c, bottom), indicate that the fragmentation phenotypes are not caused by alterations in OPA1 processing. Finally, we examined cristae morphology in control cells and cells co-transfected with mutant OPA1-GFP and mScarlet mitochondrial matrix marker by correlating light microscopy with both focused-ion-beam scanning electron microscopy (FIB-SEM) and transmission electron microscopy (TEM) (see Methods for details). Representative images revealed normal cristae morphology in all the mutants tested, arguing that mitochondrial fragmentation was not secondary to disrupted cristae architecture 43,44 (Extended Data Fig. 10h-m). Together, our results show that mutating the interfacial and membrane-binding regions of OPA1 results in fusion-deficient phenotypes, and suggest that these phenotypes arise from the disruption of OPA1 assembly on membranes rather than from altered OPA1 processing or defective cristae morphology 45 .

Discussion
Our results support a model in which IMM fusion is driven by fluid helical assemblies of OPA1 (refs. 22,26,46) (Fig. 5). As shown previously, c, Nucleotide binding orders and tightens the helical assemblies, bringing the IMMs close enough to promote fusion. During this stage, OPA1 scaffolds the positively curved membrane and the Pα2 paddle insertion deepens, which enhances the curvature of the outer-leaflet membrane. The inserted Pα2 helix also binds to and sequesters negative-curvature-inducing cardiolipin, destabilizing the bilayer. d, Fusion occurs and the IMM is merged, producing a single mitochondrion. As the merged IMM collapses back towards the OMM, OPA1 helices disassemble and diffuse from the site of fusion (shown in a magnified and partially magnified view; right). Images generated by Falconieri Visuals, LLC.
Nature | Vol 620 | 31 August 2023 | 1115 OPA1 interacts with the IMM through transmembrane tethering and its cardiolipin-specific paddle domain 5,25,[47][48][49] . The transmembrane domain of l-OPA1 might provide the initial localization of OPA1 to the fusion site and nucleate further s-OPA1 assembly 2,15,25 . Once at the membrane surface, the formation of s-OPA1 helices would induce bilayer curvature 36 . At this stage, the loosely connected assemblies of varying diameters are stabilized by electrostatic contacts and shallow membrane insertion. During the GTP hydrolysis cycle, the GTPase domains dimerize, compacting the oligomer by tightening the stalk and paddle interfaces, increasing the buried surface area, aligning interfacial charges and forcing deeper insertion of the hydrophobic paddle helices into the bilayer 48,50 . This monotopic insertion of the paddle and concurrent recruitment of cardiolipin to Pα2 would drive mechanical destabilization and further curve the bilayer 35,36,51-53 . The helical assembly would also provide a pre-fusion scaffold for the membranes from each mitochondrion until they become close enough for fusion. The destabilized outer leaflets would then have a lower energy barrier for lipid mixing, facilitating fusion, after which the negatively curved, fusion pore-stalk would be stabilized by the previously recruited cardiolipin 54-56 . In addition, insertion of the l-OPA1 transmembrane domain might lower the fusion energy barrier, amplifying the lipid remodelling caused by the s-OPA1 paddles 47 . After GTP hydrolysis, the protein falls off the membrane, although the structural mechanism that leads to disassembly remains elusive. As with other DSPs, it is possible that OPA1 undergoes a powerstroke to aid the disassembly of the organized helix and contribute energy toward fusion, but a similar s-OPA1 powerstroke has not yet been observed. After fusion, the membrane surface would rapidly relax from its highly curved state as the fusion pore collapsed. Considering that the GTPase dimer has only been observed in the presence of a transition-state mimic 23 (PDB ID: 6JTG), it is likely to dissociate after fusion. Together, the collapsing pore and GTP hydrolysis would induce conformational changes in OPA1 assembly, leading to rapid dissociation after the fusion event. Alternatively, the elastic nature of OPA1 and its propensity for forming helical structures of varied diameters and conformations could temporarily accommodate a wider, loosely packed post-fusion assembly. This would result in a slower disruption of the OPA1 helix, or reflect alternative pathways along the mitochondrial fusion landscape, such as transient pore kiss-and-run or organelle-hitchhiking events 45,57 .
Compared to the previously published helical structure 22 , our assemblies have narrower diameters. We propose that our structures represent a state just before membrane fusion, whereas the previous apo assembly of ΔN-OPA1 and tomogram of ΔN-OPA1 with GTPγS are an earlier state of OPA1 oligomerization on the membrane 22 . Consistent with this, the ΔN-OPA1 assembly was generated at a lower temperature and in low-salt conditions, in which two additional interfaces (back stalk-I1 and BSE-BSE) suggest a more electrostatically driven, looser OPA1 lattice 22 . Also, the ΔN-OPA1 assembly contained a higher percentage of negative-curvature-inducing lipids 42 (PE:CL, 22:25), which would stabilize a wider-diameter lipid tube 22 . Although the resolution is limited, fitting our s-OPA1 model into the previous apo densities highlights several similar interfaces, allowing us to render a morph towards our narrower-diameter assembly, and revealing a potential pathway towards fusion (Supplementary Video 3). This morph highlights a decrease in assembly diameter, accompanied by a helical stretch, that would bring the pre-fusion membranes into closer apposition. One contention with this model is that the prior ΔN-OPA1 helix becomes wider after the addition of GTPγS, possibly due to the back stalk-I1 and BSE-BSE interfaces that would need to break before GTPase-dimer formation; this is likely to be disfavoured by the low temperature and ionic strength used in the previous study 22 .
Overall, the present s-OPA1 assemblies provide a window into the pre-fusion IMM intermediate. They reveal a novel GTPase-dimer association-interface 5-consistent with other dynamin-family proteins, and establish a potential pathway between previous assemblies of OPA1 (ref. 22), which might represent the early stages of assembly, and our pre-fusion state captured using GDP-AlF x . Several new interfaces are present in our assemblies, in comparison to previous work with ΔN-OPA1 and other DSPs. In addition to the novel GTPase-dimer association (interface 5), the s-OPA1 assembly in the presence of GDP-AlF x exhibits a markedly different orientation of interface 1 compared with s-MGM1. A further five interfaces (interfaces 1, 2, 3, 4 and 7) were similar to the previous ΔN-OPA1 assembly but absent in other DSPs, with the remaining three (interfaces 6, 8 and 9) identified in this work. Moreover, these assemblies reveal a previously unknown role for the N terminus of s-OPA1 in stabilizing the helical assembly, and highlight the plasticity of membrane interactions with the s-OPA1 paddle domain.
The relevance of these assemblies and membrane interactions was further supported by OPA1 mutagenesis in cells, in which all tested interfacial mutants resulted in markedly more fragmented mitochondrial networks than did wild-type OPA1 or control mutants. Crucially, increased fragmentation was seen for transfection levels that were much lower than the levels of endogenous OPA1, without alteration of the ratio of s-OPA1 to l-OPA1 or changes in cristae morphology 43,44 . This indicates that the interfacial mutants exhibited dominant-negative phenotypes that did not arise from an altered equilibrium of OPA1 isoforms or from OPA1-GFP overexpression or variability 45 . Of note, the mutations within interfaces 1 and 4 also fall into similar apo dimer and tetramer interfaces within the previously published OPA1 assembly 22 , supporting the idea that these interfaces are common to OPA1 assemblies across diameters. Furthermore, mutation of the membrane-inserting Pα6 and Pα2 paddle helices also fragmented mitochondrial networks. The Pα2 helix is unique to OPA1 and might serve to recruit cardiolipin. These results are not only a high-resolution observation of the paddle domain of OPA1 binding and inserting into the membrane, but also a direct observation of a paddle domaincardiolipin contact. Finally, they also provide the direct evidence that mutating the paddle domain disrupts IMM fusion.
Given the apparent relevance of these assembly interfaces and membrane-binding regions to OPA1 function, we also examined their correspondence with known DOA-associated mutations. We found that 86 previously identified determinants of DOA were located within the membrane-binding sites or assembly interfaces identified in this paper. Most (65) were located within s-OPA1 helical interfaces, 17 potentially disrupt membrane contacts within the paddle and 21 might disrupt GTPase-dimer formation and thus GTP hydrolysis. Because our cellular data showed substantial phenotypes for small disruptions of all tested interfaces and membrane-binding helices in the paddle domain, we propose that these DOA determinants arise from impaired protein-protein or protein-membrane contacts that affect the ability of OPA1 to function as a higher-order helical assembly. A further 24 known DOA mutations are proximal to our interfaces, possibly reflecting the conformational plasticity of OPA1 assemblies. Still, the interfacial and membrane contacts of s-OPA1 appear to be tightly tuned, as often a single mutation is sufficient to produce DOA. Overall, the OPA1 structures provide insight into the mechanism of mitochondrial IMM fusion and a foundation for the classification of pathogenic DOA mutations in the structural context of OPA1 and its higherorder assemblies.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-023-06462-1. Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Expression and purification of the s-OPA1 construct
The s-OPA1 isoform 1 was expressed and purified with modifications from the plasmid and protocols provided by D. Chan. s-OPA1 (residues 195-960) was expressed in BL21-DE3(RIL) cells using a pET28a vector with an N-terminal 6×His tag. Cells were grown in terrific broth (TB) medium to an optical density at 600 nm (OD 600 nm ) of 1.2-1.3. Cells were then cold-shocked for 30 min at 4 °C and induced with 0.5 mM IPTG. After an overnight incubation at 16 °C with shaking at 300 rpm, cells were collected by centrifugation. s-OPA1 was purified using TALON metal affinity resin (Takara Bio) and size-exclusion chromatography (SEC). In brief, pelleted cells were resuspended in lysis buffer (20 mM Tris pH 8, 500 mM NaCl and 5 mM imidazole) with the addition of 1 mg ml −1 of lysozyme, 0.01% DNAse and a protease inhibitor tablet (EDTA-free, Roche), incubated for 1 h at 4 °C and lysed by probe sonication (1 min process time, 1 s on, 5 s off at 90% amplitude). Cell debris was pelleted by ultracentrifugation at 230,000g for 1 h at 4 °C. Clarified lysate was transferred to TALON resin and treated with wash buffer (20 mM Tris pH 8, 500 mM NaCl and 20 mM imidazole) and elution buffer (20 mM Tris pH 8, 500 mM NaCl and 250 mM imidazole). Fractions containing s-OPA1 were pooled, concentrated and washed into SEC buffer (20 mM Tris pH 8, 300 mM NaCl and 1 mM DTT) using a 30-kDa AmiconUltra concentrator. The protein precipitation was pelleted by centrifugation at 13,000g for 1 min, and the sample was run over an S650 size-exclusion column (Bio-Rad). After examination by SDS-PAGE, fractions containing s-OPA1 were pooled, exchanged into HCB150 buffer (20 mM HEPES, pH 7.2, 150 mM KCl, 1 mM MgCl 2 , 2 mM EGTA and 1 mM DTT) and concentrated to around 2 mg ml −1 . Protein was quantified by absorbance at 280 nm, flash-frozen in liquid nitrogen and stored at −80 °C.

GTPase assay
Basal GTP hydrolysis of s-OPA1 was measured using the colorimetric malachite-green assay described previously 58  Post-sonication tubes were incubated between 25 min and 60 min. For tubes in the presence of nucleotide, GDP-AlF x was then added to a final concentration of 1 mM and the tubes were allowed to incubate for another 10 min. A GDP-AlF x stock was generated by combining GDP, AlCl 3 and NaF to final concentrations of 10 mM, 60 mM and 300 mM, prepared within the 10 min before addition to s-OPA1:lipid tubes.
Negative stain s-OPA1:lipid mixtures were loaded onto carbon-coated, 15-second glow-discharged, nickel formvar (FCF400 Ni-NA) grids and incubated for 1 min, followed by staining with 1% uranyl acetate. Grids were visualized on a FEI Technai 12, as part of the NIDDK core facility.

Cryo-EM
Samples were screened with a TF20 microscope (FEI) at 200 kV at a magnification of 29,000×, with a nominal defocus range of 1.5 to 3.0 μm, using a K2 summit camera (Gatan) in counting mode. High-resolution images of OPA1 lipid tubes in the presence of GDP-AlF x were recorded on a Titan Krios G3 microscope (Thermo Fisher Scientific) operating at 300 kV on a Gatan K3 camera. A total of 6,319 images were collected at a magnification of 105,000×, with a calibrated pixel size of 0.415 Å, nominal defocus range of 0.3-2.4 μm, 24 frames and 60 e − /Å 2 electron exposure per movie. Images of OPA1 lipid tubes in the apo state were recorded on a Glacios cryo-TEM microscope (Thermo Fisher Scientific) operating at 200 kV on a Gatan K3 camera. A total of 2,830 images were collected at a magnification of 36,000×, with a calibrated pixel size 0.58 Å, nominal defocus range 0.6-2.4 μm, 22 frames and 24.86 e − /Å 2 electron exposure per movie. All data were collected using Serial EM.

Preliminary cryo-EM processing and determination of helical parameters
Preliminary micrographs from the FEI TF20 of s-OPA1:lipid tubes with GDP-AlF x were motion-corrected and dose-weighted with MotionCor2 in RELION 3.1 (refs. 59,60). CTF estimation was determined in RELION 3.1 using Ctffind4 (ref. 61). Particle selection was performed manually in RELION 3.1 (refs. 60,62). Extracted particles were used to generate initial two-dimensional (2D) reference classes for real-space analysis of helical parameters. Motion-corrected particles generated in RELION 3.1 were also used to generate stacks of averaged power spectrum for Fourier-Bessel analysis in Fiji v.2.9.0 (refs. 63,64). Suspected helical parameters were then input into HELIXPLORER (L. Exterozi, French National Centre for Scientific Research) to determine a range of potential solutions 63 .

Cryo-EM processing
Final images of s-OPA1:lipid tubes in the GDP-AlF x or apo state were processed using cryoSPARC v.3.2.0 and v.3.3.0-3.3.2 (refs. 29,65). Images were gain-corrected, motion-corrected, binned and dose-weighted using patch motion correction in cryoSPARC 29,65 . CTF estimation was determined using Path CTF estimation in cryoSPARC 29,65 . Particle selection was performed by generating a template, by manually selecting a subset of desired s-OPA1 tubes, 2D classifying, then template tracing and extracting particles with a 1.2018 Å/pixel size and a 500-pixel box size, for the GDP-AlF x -bound dataset, and a 1.2518 Å/pixel size and a 480-pixel box size for the apo dataset. Selected particles were then examined and filtered according to a 6-Å CTF resolution using the curate exposures job. Particles were then pruned using one round of 2D classification. Initial tube alignment was performed using the ab Article initio reconstruction job and helical maps were generated using the helical refinement (Beta) pipeline. Suspected helical symmetry, determined above, was cross-examined using the symmetry search tool in cryoSPARC 65 . Owing to heterogeneity in tube diameter for apo-OPA1 images, multiple rounds of iterative 2D classification, selection and helical refinement were performed to produce the final helical map. After generating an initial helical map, a local refinement was performed to focus in on 30% of the box size in the z direction to better compare to maps generated in RELION 3.1 (refs. 60,62,66). Local refinements on s-OPA1 and the outer leaflet of the bilayer were performed using the local refine (Beta) pipeline in cryoSPARC using masks generated in Chimera 29,65 . Masks resulting from MolMaps were the product of iterative s-OPA1 model building or hand-built lipid monolayers. Local-resolution maps were also generated in cryoSPARC 29,65 .
To sort the heterogeneity of the apo sample, a subset of particles was selected from outputs in cryoDRGN v.0.3.4 (ref. 30), which were re-imported and refined using cryoSPARC's helical and local refine Beta pipelines 29,65 . Maps generated in cryoSPARC were post-processed using deepEMhancer to sharpen protein densities and also sharpened in cryo-SPARC as a cross-comparison for model building 67 . DeepEMhancer was not used for ligand densities. Neither deepEMhancer nor cryoSPARC sharpening was applied to the lipid-monolayer local refinements.

Diameter measurements
The diameter distribution of s-OPA1 tubes in the apo and the GDP-AlF x -bound state was determined using 2D classifications in cry-oSPARC. Particles were divided into around 500 2D classes per state. The tube diameter of each class was measured using Fiji. The resulting stacked bar chart was created using Plotly.
Supernatant and pellet assay s-OPA1 lipid tubes, generated as described above, were incubated for 30 min with and without GTP. Samples were centrifuged at 100,000g for 15 min at 4 °C in a Beckman TLA 100 rotor. The resulting supernatant and pellet fractions were then analysed by SDS-polyacrylamide gel electrophoresis, stained with Coomassie Brilliant Blue and quantified using Fiji. Raw images for Extended Data Fig. 2b are available in Supplementary Fig. 1.

Model building and refinement
Initial fitting of s-OPA1-MGD (GTPase-dimer reference) into cryo-EM helical maps was done manually in UCSF Chimera 68 with the PDB ID: 6JTG (ref. 23). To perform local refinements in cryoSPARC, initial models were then extended from the s-OPA1-MGD, de novo, using Coot v.0.9 (ref. 69). The initial model was iteratively refined in Coot as the map resolution improved and features became more apparent 69 . After the release of the AlphaFold database, the initial model was hybridized with the AlphaFold OPA1 structure, AF-O60313-F1, in Coot v.0.9.0 and v.0.9.2 to produce a model for refinement 69

Summary of helical assemblies
Visually, the biological twist and rise of the densely packed rungs of both s-OPA1 assemblies are left-handed ( Supplementary Fig. 3h,p). However, we chose to define the helical parameters along the opposing lattice in right-handed coordinates, which allowed us to minimize the rise and pitch of a full turn. This maximized the number of units for a given box size, which had computationally limited our attempts at left-handed solutions 65,66 .
In the right-handed solution space, both assemblies are onestart, with a pitch of 12.59 nm (Supplementary Fig. 3a-d) in the GDP-AlF xbound map and 13.60 nm in the apo map ( Supplementary Fig. 3i-l). The GDP-AlF x -bound assembly had a twist of 37.439° and a rise of 12.933 Å ( Supplementary Fig. 3e), whereas the apo assembly had a twist of 37.25° and a rise of 14.07 Å (Supplementary Fig. 3m). The helical maps were z-clipped in cryoSPARC to 50% of the full map 29,65 (Extended Data Fig. 1c,d). The flexibility and continuous heterogeneity were examined and minimized for the apo assembly using cryoDRGN 30 ( Supplementary  Fig. 3q-t and Supplementary Video 2).
An alternative minimal left-handed lattice of the GDP-AlF x -bound assembly would be a four-start helix with a pitch of 33.29 nm (Supplementary Fig. 3f), compared to an eight-start helix with a pitch of 64.45 nm for the apo assembly ( Supplementary Fig. 3n). The left-handed GDP-AlF x -bound assembly has a twist of −49.0° and a rise of 45.33 Å (Supplementary Fig. 3g), compared to a twist of 62.3° and a rise of 111.6 Å for the apo assembly ( Supplementary Fig. 3o).
The apparent biological rung of the helical assembly is based on visual packing. The biological assemblies form left-handed, 9-start helical lattices of around 16 tetrameric dimers-of-dimers. A full turn for the GDP-AlF x -bound map has a pitch of 182.3 nm (Supplementary Fig. 3h) and the apo case has a pitch of 179.5 nm (Supplementary Fig. 3p). The GDP-AlF x -bound helix has a twist of −22.93° and a rise of 116.1 Å (Supplementary Fig. 3h), whereas the apo helix has a twist of −25.16° and a rise of 125.5 Å (Supplementary Fig. 3p).

Mutations of full-length OPA1 for imaging
Full-length human OPA1, isoform 1 cDNA was obtained, courtesy of D. Chan, through Addgene (70173). The OPA1 cDNA sequence was then ligated into a pEGFP-N1 vector (Clonetech), attached through an eight-amino-acid linker, GLALPVAT, to an adjoining GFP fluorescent protein tag using the In-Fusion HD Cloning kit (Clonetech, 638911) following the manufacturer's instructions. Alanine mutations were introduced into the full-length OPA1-GFP sequence as single mutations by site-directed mutagenesis using the Quikchange II XL Site-Directed Mutagenesis Kit, following the manufacturer's instructions, for the following mutants: E444A (interface 5), Q659A (interface 1(c)), E679A (interface 1(d)), H631A (interface 1(e)), I735A (interface 4(b)), D716A  4(a)). Primers were synthesized by Eurofins. Plasmid concentration and purity was checked using a NanoDrop (Thermo Fisher Scientific). The coding sequences were verified by Sanger sequencing (Psomagen) and the integrity of the full-length plasmids was further verified by nanopore sequencing (Plasmidsaurus). Primers can be found in Supplementary Table 3.

Evaluation of OPA1 expression levels and mitochondrial network morphologies
HeLa-M cells were cultured in Dulbecco's modified Eagle's medium (DMEM; Gibco Life Technologies) supplemented with 10% heatinactivated fetal bovine serum (FBS; Gibco Life Technologies) and 1% antibiotic-antimycotic solution (Gibco Life Technologies), and were maintained at 37 °C in a humidified, 5% CO 2 incubator. Cells were plated in glass-bottomed, four-or eight-well chamber slides (Cellvis, C4-1.5H-N) that had been coated with Fibronectin (10 ng ml −1 ; Gibco Life Technologies, 33016015), and were transfected with GFP-tagged wild-type and mutant OPA1 constructs using Lipofectamine 3000 according to the manufacturer's instructions (Invitrogen, Thermo Fisher Scientific, L3000-001). After 24-28 h, cells were fixed with 4% paraformaldehyde, 0.25% glutaraldehyde in 0.1 M sucrose and 0.1 M cacodylate (pH 7.0) for 15 min at 37 °C. The samples were then permeabilized for 10 min using 0.1% Triton X-100, and blocked with 5% bovine serum albumin in phosphate-buffered saline (PBS; Thermo Fisher Scientific, WL335677) for 1 h. To enhance the GFP signal present in transfected cells, and to visualize mitochondrial morphology, fixed samples were stained with a 1:200 dilution of rabbit anti-GFP (Rockland, 600-401-215) and a 1:200 dilution of mouse anti-TOM20 (F-10) (Santa Cruz., sc-17764) for 1 h at room temperature. After washing three times with 5% bovine serum albumin (BSA) in PBS, the cells were stained with a 1:500 dilution of both Alexa-488 anti-rabbit secondary antibody (Invitrogen, A11034) and Alexa-594 anti-mouse secondary antibody (Invitrogen, A32742) for 1 h at room temperature, and then washed with PBS. All images were captured using a Zeiss LSM 780 confocal microscope equipped with a 60× 1.40 NA oil objective. A zoom setting of 1 was used to give 8-15 cells in each field of view (around 135 μm 2 ). For each field of view, a z-stack spanning the full thickness of the cells was taken in 0.6-μm intervals. Maximum projections of the z-stacks were then used to quantify OPA1 expression and to score mitochondrial morphology, as described next. As regards the quantification of the expression of wild-type and mutant OPA1 constructs, the transfected cells within each field of view were first outlined on the basis of their footprints. The total fluorescence in the 488 channel was then determined for each cell by summing the fluorescent intensities in each plane of the z-stack using Fiji. Cells that showed a 488 signal that was above the background exhibited by non-transfected cells in the same field of view, and that showed colocalization between the anti-GFP signal and the TOM20 signal, were then scored for mitochondrial length. Representative examples of each category (fragmented, intermediate or filamentous mitochondria) are shown in Extended Data Fig. 10a,b. The data in Extended Data Fig. 10d-g are representative of at least three independent experiments. The fold increases in anti-GFP fluorescence over background for each category of mitochondrial morphology obtained were compared using the Student's t-test (two-tailed) (Extended Data Fig. 10g). HeLa-M cells (a gift from A. Peden) were early passaged stocks, which tested mycoplasma-free. Raw images for Extended Data Fig. 10c are available in Supplementary Fig. 1.

Examining mitochondrial cristae morphology with FIB-SEM and TEM
HeLa-M cells were cultured as stated above, with the modification that cells were plated on 35-mm dishes with a gridded coverslip bottom (MatTek, P35G-1.5-14-C-GRD). The cells were co-transfected with GFP-tagged wild-type and mutant OPA1 constructs together with an mScarlet-tagged Mito Matrix construct using the same transfection method as above. After 24 h transfection, live cells were imaged using a Zeiss LSM 880 microscope equipped with a 40× 1.40 NA oil objective and an environmental chamber to maintain the cells at 37 °C and 5% CO 2 . To localize transfected cells on the grid, a 5 × 5-tile scan was performed to cover an area of 1 mm 2 using both DIC and fluorescence channels. FIB-SEM and TEM were performed on mutant-OPA1-expressing cells possessing mitochondria that were both red and fragmented.
After light-microscopy imaging, cells were fixed with a mixture of 2.5% glutaraldehyde and 1% paraformaldehyde in 0.1 M sodium cacodylate buffer at 4 °C overnight. The next morning, the cells were washed with 0.1 M sodium cacodylate buffer five times for 3 min each. Cells were then incubated in 2% osmium tetroxide in 1.5% potassium ferricyanide made in 0.1 M sodium cacodylate buffer for 1 h on ice and then rinsed with water five times for 3 min each. All subsequent water washes mentioned here consist of five washes for 3 min each. The cells were then incubated in 1% thiocarbohydrazide in water for 20 min at room temperature, followed by water washes. This was followed by an incubation in 2% osmium tetroxide in water for 30 min on ice and subsequent water washes at room temperature. Cells were then incubated in 1% uranyl acetate in water overnight at 4 °C. The following day, cells were washed in water and then placed in a solution of 0.066% lead aspartate for 30 min at 60 °C. After water washes, cells were dehydrated with a series of graded ethanol in water incubations consisting of 30%, 50%, 70%, 90% and 3 times with 100% ethanol for 5 min each. This was followed by infiltration with Epon resin (Embed 812, Electron Microscopy Sciences) using increasing concentrations of resin dissolved in ethanol over the course of 16 h. The cells were then embedded in a thin layer of fresh Epon resin and cured in an oven at 60 °C for two days.
After cell embedding, the glass coverslip at the bottom of the dish was dissolved in hydrofluoric acid (Sigma-Aldrich). This left the resin with the alphanumeric grid exposed on the surface of the block and visible with a dissecting microscope. The region at the centre of the 35-mm dish containing the grid and cells was separated from the remaining plastic dish using a thin saw. This was mounted on an SEM stub with the grid facing up and then coated with a 20-nm layer of gold using a sputter coater (Electron Microscopy Sciences). FIB-SEM imaging was done using a Crossbeam 540 (Zeiss). Because the alphanumeric grid was visible under the SEM beam, the cells of interest for FIB-SEM imaging were easily identified on the basis of the coordinates seen in the light-microscopy images. FIB-SEM images were acquired at 3 nm and 6 nm in XY and with a slice thickness of 6 nm. The SEM beam was set to 1.5 kV and 1.2 nA and the FIB was set to 30 kV. Images were acquired with the back-scatter detector with a grid voltage set to 500 V. After the volume acquisition, the contrast of the images was inverted and the stack was registered using the Atlas 5 (Zeiss) registration module.

Article
For TEM, cells were prepared as above. Transfected cells were relocated on a large resin surface based on the grid imprint and coarsely trimmed with a jeweller's saw. Coarsely trimmed blocks were attached to a prepared Epon block using cyanoacrylate glue and finely trimmed close to the cell of interest using a Diatome TrimTool (Electron Microscopy Sciences). Ultrathin sections (65 nm) were cut from the block surface using a Leica EM UC7 ultramicrotome and digital micrographs were acquired on a JEOL JEM 1200 EXII TEM operating at 80 kV and equipped with an AMT XR-60 digital camera.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
Structural data supporting findings in this study have been deposited in the PDB and the Electron Microscopy Data Bank (EMDB). The accession codes of the cryo-EM. maps and accompanying atomic models are provided for the following:

Code availability
Custom code was not central to the conclusions of the paper.

Article
Extended Data Fig. 1 | Helical assemblies of s-OPA1 coating cardiolipincontaining membranes with and without GDP-AlF x . a, Left, top view of s-OPA1 with GDP-AlF x bound, with 38.2 nm outer diameter and a 9.4 nm inner lumen diameter of the membrane. Centre, side view of the helical map with 10 modelled tetramers which define a single helical turn, coloured by domain (see Fig. 1). Right, a central slice of the helix is shown in comparison to a 2D classification generated in RELION (right corner). A radial profile plot is shown above the sliced helix. b, Left, top view of apo s-OPA1, with 40.5 nm outer diameter and a 9.5 nm inner lumen diameter of the membrane. Centre, side view of the helical map with 10 tetramers which define a single turn, coloured as in a. Right, a central slice of the helix is shown in comparison to a corresponding apo 2D classification generated in RELION (right corner). A radial profile plot is shown above the sliced helix. c, A z-clipped refinement focusing on the central around 30% of the GDP-AlF x -bound helix is coloured by domain, shown as a top view (left), side view (centre) and centrally sliced (right) with a single s-OPA1 dimer highlighted in darker hues. d, Domains of s-OPA1, moving from the outer diameter of the GDP-AlF x helix toward the membrane bilayer. Left, GTPase domains and BSE regions coloured green and pink, respectively. Centre, stalks shown in blue. Right, paddles coloured orange. e,f, same as c,d for the apo helix. g, s-OPA1 GDP-AlF x map with density coluored by domain as in a and refined in cryoSPARC without dynamic masking. Insets highlight the N terminus, coloured grey. Fig. 2 | See next page for caption.

Article
Extended Data Fig. 2 | Workflow for determining s-OPA1 helical assembly and corresponding local refinements using cryoSPARC. a, Sample preparation workflow for s-OPA1 purification, GTPase activity assay, and tube formation. Images for the GDP-AlF x and apo states were collected on a Titan Krios and a Glacios, respectively. GTPase activity data are presented as mean values ± s.d. of n = 3 independent experiments. b, Left, s-OPA1 tube diameters for both the apo (orange) and GDP-AlF x (blue) states. Asterisks mark diameters used for cryo-EM image processing. Right, Sup/Pellet assay showing GTPmediated protein disassembly from membrane. SDS-PAGE gel (top) of Sup/ Pellet assay with protein alone (*), and s-OPA1 tubes before (Apo) and after GTP addition (GTP). Bar graph (bottom) of Sup/Pellet assay with supernatant (blue) and pellet (orange) for the same conditions (data are presented as mean values ± s.d. of n = 3 independent experiments). c, Outline of data processing. Each dataset followed a standard workflow including patch motion correction and patch CTF correction. Segments of tubes were then manually selected, 2D classified to generate templates to trace the particles, extracted and curated to a 6 Å CTF cut-off. Particles were 2D classified and particles of similar identity and diameter were selected and aligned through ab initio modelling. The selected ab initio models were then refined with non-uniform helical refinement. For GDP-AlF x tubes, helical symmetry was determined using a combination of Fourier and real-space analysis using HELIXPLORER and RELION 60,62,63 . For Apo tubes, symmetry parameters were determined in cryoSPARC. Each set of parameters was examined with the cryoSPARC helical search tool 29,65 . To assess the interfaces of each OPA1 in the assembly, local refinements were performed by first symmetry expanding the particles in the helical refinements. Areas of interest were masked in chimera. Masks were then expanded and padded within cryoSPARC for particle subtraction, followed by non-uniform local refinements. The resulting maps were used to build the model of s-OPA1, which was generated using Coot, Rosetta, and PHENIX 69,71,74 . Fig. 3 | Closer examination of the stalk and paddle interfaces within the s-OPA1 dimer. a, A side view of a GDP-AlF x s-OPA1 dimer fit into a locally refined map. Monomers coloured by domain (see Fig. 1). Chain A is coloured a dark hue and chain B a light hue. The interlocked stalk interface 1 is shown above and paddle-paddle interface 3, where the paddle is rotated 180° showing the membrane-facing region, is displayed below. Insets: interface 4 between the paddle hinges and interface 7 between the inter-rung paddles are highlighted. b, Top view of isolated interfaces for the s-OPA1 stalk and paddle regions. The full dimeric interface between the stalks and paddles is shown fit into the locally refined density (centre) and coloured as in a. Interface 2 between the stalk of an s-OPA1 monomer and the paddle of the second monomer is shown for the A-to B-chain (top) and the B-to A-chain (bottom). c, A representative fit of each domain of the s-OPA1 model into the map density. d,e, Same as a,b for the apo dimer.

Article
Extended Data Fig. 6 | Human s-OPA1 compared to s-MGM1 and enhanced s-OPA1 membrane binding. a, Sequence comparison between s-OPA1 and s-MGM1 (ref. 26) (grey). s-OPA1 is coloured by domain (see Fig. 1). b, Overlay of s-OPA1 with s-MGM1 (grey) (PDB ID: 6QL4) highlighting similar domains. s-OPA1 domains are coloured according to a. Unique regions in s-MGM1 and s-OPA1 are highlighted: kinked stalks, paddle domains, and the stalkpaddle hinge disulfide bond. Inset shows a 180° rotation of the zoomed paddle tip highlighting differences in secondary structure and domain angle. Top left, s-OPA1 monomer highlighted in helical map. Helices are labelled as in Supplementary Fig. 2. c, A 90° rotation of s-OPA1 and s-MGM1 (grey), with an s-OPA1 monomer coloured in helical map, top left. The membrane is drawn in grey highlighting increased s-OPA1 membrane insertion, the membraneinserting loop (Pα2) and Pα6 and closer OPA1 paddle-tip membrane association (inset). d, Table comparing structural features and sequences between OPA1 and MGM1. Positively charged residues are blue and aromatics are orange. e, Locally refined map of the bilayer outer leaflet with proposed cardiolipin density proximal to the Pα2 helix. s-OPA1 paddles are coloured as in a, cardiolipin is coloured magenta, unidentified lipids are white. To orient, a cylinder representing the membrane with a box representing the refined region appear, top left. f, A view of the map, rotated 180° from e. g, Three views of cardiolipin modelled into the map. h, s-OPA1 paddle with a possible cardiolipin orientation shown (sticks and surface views). Membrane-inserting, positively charged (blue) and aromatic (white) residues appear (sticks). A representative cylinder with the boxed region (top left). i, A 90° rotation of h. j, A magnified view of the s-OPA1 and two cardiolipin models (dark and light magenta) highlighting potential residues (sticks) involved in the Pα2-CL interaction and the cardiolipin headgroup swing (curved arrow). c, A s-OPA1 GDP-AlF x dimer coloured by electrostatic potential and oriented as in a in the side view of the helical assembly. Charge striation is particularly apparent along the dimer within the interlocked stalk interface 1 and at the site of N-terminal coiled-coil interface 8. Zoomed, the striated stalk-stalk interface 1 is highlighted, charges are labelled in green. d, The striated s-OPA1 apo stalk-stalk interface 1 is shown with charges (green) slightly out of register. e, A magnified view of the electrostatic potential for the N-terminal interface 8. f, The inter-rung interface 7 coloured by electrostatic potential, highlighting charge dependence. g-l, Same as in a-f for calculated lipophilicity: hydrophilic, teal; hydrophobic, gold; neutral, white. Hydrophobic regions (gold) within the paddle domain are membrane-inserted (h). Coulombic electrostatic potential and molecular lipophilicity were determined using ChimeraX (ref. 68). Fig. 10 | Cell-imaging statistics examining the functional effects of OPA1 mutants in more detail. a, Example images of filamentous, intermediate and fragmented mitochondrial networks used for categorizing single-cell data. Left, mitochondria immunostained against TOM20. Middle, OPA1-GFP mitochondrial localization. Right, colocalization of OPA1-GFP and mitochondria immunostained against TOM20. Scale bars, 10 μm. b, Magnified regions of the anti-TOM20 (yellow dash) and merged (pink dash) filamentous, intermediate and fragmented mitochondria. Scale bars, 10 μm. c, Western blot of HeLa-M cells showing the ratios of GFP-tagged and endogenous long-and short-OPA1. Top, long-and short-OPA1-GFP (117 kDa, 129 kDa) detected with an anti-GFP antibody (long-to short-OPA1 ratio: around 0.78:1). The relative expression between WT versus mutant OPA1-GFP is 1:0.8-1.1. Middle, long-and short-ratios of endogenous OPA1 isoforms (~89 kDa, ~101 kDa) detected with anti-OPA1 antibody (long-and short-OPA1 ratio: around 0.8:1). Bottom panel, actin control detected with anti-actin antibody. Below, transfection efficiencies, measured in parallel samples using fluorescence imaging. d, Plot of cells transfected with OPA1-GFP expression levels estimated from differences in intensity relative to background cells. OPA1-GFP mutations are found in Fig. 4 and Extended Data Fig. 9. e, Box and whisker plot of transfected OPA-GFP mutants intensity over background cells.   Last updated by author(s): Jun 16, 2023 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted

Software and code
Policy information about availability of computer code Data collection CryoEM data was collected on either a FEI Krios or TFS Glacios microscope, wtih Gatan K3 direct detectors, using the program Serial EM. For light microscopy, experiments were collected using a Zeiss LSM 880 microscope using the confocal setting with 60 x 1.40 NA oil objective and acquired using the software ZenBlak version 2.3
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information.

April 2023
Data Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A description of any restrictions on data availability -For clinical datasets or third party data, please ensure that the statement adheres to our policy There are no restrictions on availability of unique materials. Note that full information on the approval of the study protocol must also be provided in the manuscript.

Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design
All studies must disclose on these points even when the disclosure is negative.

Sample size
Our cryoEM reconstructions of GDP-AlFx and apo sOPA1 polymers contained 598,503 and 88,155 asymmetric units respectively. These sample sizes were determined through image processing software, and were sufficient to converge the structure calculation to detailed maps that are consistent with crystal structures and functional studies. Furthermore, all Euler angles were well sampled in the reconstructions. For the GTPase and supernatant/pellet assays 3 independent samples were analyzed. For the light microscopy experiments, the N value (the number of cells scored) for each sample is reported in the bar graph in Figure 4 and comes from at least three independent experiments, all of

April 2023
Alexa Fluor 488 dye is pH-insensitive over a wide molar range. Probes with high fluorescence quantum yield and high photostability allow detection of low-abundance biological structures with great sensitivity. Alexa Fluor 488 dye molecules can be attached to proteins at high molar ratios without significant self-quenching, enabling brighter conjugates and more sensitive detection. The degree of labeling for each conjugate is typically 2-8 fluorophore molecules per IgG molecule; the exact degree of labeling is indicated on the certificate of analysis for each product lot. Target Information Anti-Rabbit secondary antibodies are affinity-purified antibodies with well-characterized specificity for rabbit immunoglobulins and are useful in the detection, sorting or purification of its specified target. Secondary antibodies offer increased versatility enabling users to use many detection systems (e.g. HRP, AP, fluorescence). They can also provide greater sensitivity through signal amplification as multiple secondary antibodies can bind to a single primary antibody. Most commonly, secondary antibodies are generated by immunizing the host animal with a pooled population of immunoglobulins from the target species and can be further purified and modified (i.e. immunoaffinity chromatography, antibody fragmentation, label conjugation, etc.) to generate highly specific reagents.
Alexa-594 secondary antibody (anti-mouse polyclonal, Invitrogen, A32742): Product Specific Information To minimize cross-reactivity, the goat anti-mouse IgG whole antibodies have been cross-adsorbed against IgG from bovine, goat, rabbit, rat, and human. Cross-adsorption or pre-adsorption is a purification step to increase specificity of the antibody resulting in less background staining and cross-reactivity. The secondary antibody solution is passed through a column matrix containing immobilized serum proteins from potentially cross-reactive species. Only the nonspecific-binding secondary antibodies are captured in the column, and the highly specific secondaries flow through. Further passages through additional columns result in highly crossadsorbed preparations of secondary antibody. The benefits of these extra steps are apparent in multiplexing/multicolor-staining experiments where there is potential cross-reactivity with other primary antibodies or in tissue/cell fluorescent staining experiments where there may be the presence of endogenous immunoglobulins. Using conjugate solutions: Centrifuge the protein conjugate solution briefly in a microcentrifuge before use; add only the supernatant to the experiment. This step will help eliminate any protein aggregates that may have formed during storage, thereby reducing nonspecific background staining. Because staining protocols vary with application, the appropriate dilution of antibody should be determined empirically. Specificity: This antibody binds to heavy chains on mouse IgG and light chains on all mouse immunoglobulins. This antibody does not bind non-immunoglobulin mouse serum proteins or IgG from bovine, goat, human, rabbit, or rat. Target Information Anti-Mouse secondary antibodies are affinity-purified antibodies with well-characterized specificity for mouse immunoglobulins and are useful in the detection, sorting or purification of its specified target. Secondary antibodies offer increased versatility enabling users to use many detection systems (e.g. HRP, AP, fluorescence). They can also provide greater sensitivity through signal amplification as multiple secondary antibodies can bind to a single primary antibody. Most commonly, secondary antibodies are generated by immunizing the host animal with a pooled population of immunoglobulins from the target species and can be further purified and modified (i.e. immunoaffinity chromatography, antibody fragmentation, label conjugation, etc.) to generate highly specific reagents.
Peroxidase conjugated AffiniPure Goat Anti-Rabbit IgG (H+L), 111-035-003): Based on immunoelectrophoresis and/or ELISA, the antibody reacts with whole molecule rabbit IgG. It also reacts with the light chains of other rabbit immunoglobulins. No antibody was detected against non-immunoglobulin serum proteins. The antibody may cross-react with immunoglobulins from other species. Whole IgG antibodies are isolated as intact molecules from antisera by immunoaffinity chromatography. They have an Fc portion and two antigen binding Fab portions joined together by disulfide bonds and therefore they are divalent. The average molecular weight is reported to be about 160 kDa. The whole IgG form of antibodies is suitable for the majority of immunodetection procedures and is the most cost effective. The antibody was purified from antisera by immunoaffinity chromatography using antigens coupled to agarose beads. Horseradish peroxidase (HRP) conjugates are prepared by a modified Nakane and Kawaoi procedure (J. Histochem. Cytochem. 1974Cytochem. . 22, 1084. Peroxidase conjugates are commonly used for immunohistochemistry, Western blotting, and ELISA. Affinity-purified antihorseradish peroxidase and conjugates are available for detection of horseradish peroxidase antigen or for signal amplification of HRP-containing reagents. For immunostaining of mammalian cells, an advantage of using anti-horseradish peroxidase is reduced background, since the antibody does not recognize the endogenous peroxidase-like enzymes found in those cells.
Peroxidase AffiniPure (Goat Anti-Mouse IgG (H+L), 115-035-003): Based on immunoelectrophoresis and/or ELISA, the antibody reacts with whole molecule mouse IgG. It also reacts with the light chains of other mouse immunoglobulins. No antibody was detected against non-immunoglobulin serum proteins. The antibody may cross-react with immunoglobulins from other species. Whole IgG antibodies are isolated as intact molecules from antisera by immunoaffinity chromatography. They have an Fc portion and two antigen binding Fab portions joined together by disulfide bonds and therefore they are divalent. The average molecular weight is reported to be about 160 kDa. The whole IgG form of antibodies is suitable for the majority of immunodetection procedures and is the most cost effective. The antibody was purified from antisera by immunoaffinity chromatography using antigens coupled to agarose beads. Horseradish peroxidase (HRP) conjugates are prepared by a modified Nakane and Kawaoi procedure (J. Histochem. Cytochem. 1974Cytochem. . 22, 1084. Peroxidase conjugates are commonly used for immunohistochemistry, Western blotting, and ELISA. Affinity-purified antihorseradish peroxidase and conjugates are available for detection of horseradish peroxidase antigen or for signal amplification of HRP-containing reagents. For immunostaining of mammalian cells, an advantage of using anti-horseradish peroxidase is reduced background, since the antibody does not recognize the endogenous peroxidase-like enzymes found in those cells.

Eukaryotic cell lines Policy information about cell lines and Sex and Gender in Research
Cell line source(s) HeLa M cells (a gift from A. Peden, University of Sheffield) were used for fluorescence microscopy.