The consequence of ATP synthase dimer angle on mitochondrial morphology studied by cryo-electron tomography

The F₁F_o ATP synthase is a molecular motor ubiquitous to all living organisms, required for the essential conversion of an electrochemical gradient into the universal energy currency ATP [1]. The ATP synthase is composed of a catalytic F₁ head connected to a membrane-embedded F_o motor by a central stalk; the entire assembly is visualised as a lollipop shape when examined by electron cryo-microscopy (cryoEM) [2,3]. The central stalk transmits the torque generated by the rotation of F_o to the F₁ head, and a peripheral stalk acts as an elastic spring, ensuring malleable coupling between F₁ and F_o [4]. Mitochondrial ATP synthases across species share the same complement of core subunits with varying nomenclature (Supplementary Table S1) [5,6]. In metazoans studied to date, the F₁ head consists of α and β subunits, the central stalk of γ, δ and ε subunits, the peripheral stalk of b, d, F₆ and oligomycin sensitivity conferral protein (OSCP) subunits, and the F_o motor contains the c-ring and subunit a.

Mitochondrial ATP synthases can assemble into dimers [7], of which there are four types [8]: Type I is present in both multicellular [9–11] and unicellular organisms [12] and type II–IV are present in various unicellular organisms [13–18], reviewed in [8]. When compared with type II–IV dimers, previously studied type I dimers contain an additional set of subunits at the dimer interface: e, f, g, i/j, k and 8 (Supplementary Table S1) [8]. Based on biochemical and imaging experiments, subunits e and g were shown to be essential for dimer formation [7,11,19–21]. Dimers of ATP synthases assemble into oligomeric rows (or ribbons) along the curved ridges of crista membranes, observed by cryo-electron tomography (cryoET) [9,11,14,22]. The formation of dimer rows is mediated by an ancestral motif in subunits e and g [20,21] with assistance from subunit k [5,23]. Dimer rows are required for crista membrane curvature, and thus maintenance of lamellar or tubular-shaped cristae [11,12,14,22,24]. Cristae undergo balloon-like deformation following the knockdown of subunits e or g [11] and on the disassociation of dimers into monomers [12], supporting the role of subunits e and g in influencing cristae morphology. Moreover, molecular simulations indicate that ATP synthase dimers have an innate propensity to induce membrane curvature [25]. This was confirmed experimentally when dimers reconstituted into liposomes spontaneously self-assembled into oligomeric rows to engender this curvature, maintaining identical dimer angles to those observed in whole mitochondria [22].

In situ structures of type I ATP synthase dimers have been determined from native membranes [10–12,22,26]. Mammals and fungi both display an average angle between the dimer heads of ∼86° [10]. Interestingly, higher-resolution single particle analysis of the purified bovine ATP synthase dimer reveals that dimer angles likely vary around this average (between 76° and 95°), depending on the catalytic state [27]. Further atomic-detail structures of purified mitochondrial type I ATP synthase dimers have also been determined from mammals (Bos taurus [23]) and fungi (Saccharomyces cerevisiae [28] and Yarrowia lipolytica [24]). Thus, whilst the structure and organisation of ATP synthase dimers have been studied across a range of different species, our knowledge of ATP synthases in invertebrates is lacking. The free-living nematode worm Caenorhabditis elegans is a well-established invertebrate model system for the study of cell and developmental biology [29], including the role of mitochondria in metabolism, health, disease and aging [30]. To complement in vivo physiological studies, intact mitochondria can be stably prepared [31,32] for biochemical and structural analyses [33]. Interestingly, studies have shown that nematodes lack the dimer-specific subunits i/j, k and 8 [34,35] found in mammals and fungi (Supplementary Table S1). Subunit 8 is encoded by one of two overlapping ATP synthase genes on the mitochondrial genome [36]. Proteins encoded on the mitochondrial genome are translated from essential genes [37,38]; thus it follows that subunit 8 is likely to be essential for respiration in mammals and fungi. The lack of dimer-specific subunits in C. elegans provides a unique opportunity to investigate how certain subunits influence ATP synthase dimer angles and mitochondrial morphology.

In this study, we employ cryoET and sub-tomogram averaging to determine the molecular architecture and organisation of the C. elegans ATP synthase in the membrane, revealing a novel average dimer angle of 105°. We also discover extra mass at the dimer interface compared with an equivalent S. cerevisiae structure [11]. We subsequently compare whole mitochondria from both C. elegans and S. cerevisiae to investigate the relationship between ATP synthase dimer angle and crista morphology. Finally, we use AlphaFold [39] and AlphaFold multimer [40] to predict how protein chains in the C. elegans ATP synthase dimer may be arranged. This allows us to analyse subunit differences at the dimer interface and postulate the cause of variations in angle. We speculate that an evolutionary divergence at the dimer interface and corresponding widening of the dimer angle may be an adaptation to more variable-oxygen environments.

To determine the arrangement and molecular architecture of ATP synthase dimers in C. elegans, tomograms of whole mitochondria (Figure 1A) and of isolated crista membranes (Figure 1B) were collected and analysed. ATP synthases were unambiguously identified by the characteristic lollipop shape of the 10 nm diameter F₁ heads positioned ∼10 nm away from the membrane. We confirmed the presence of oligomeric ATP synthase dimer ribbons, localised at the sharp curved ridges of crista membranes, in both samples (Figure 1A,B). Due to the obscuring presence of a dense matrix in whole mitochondria, many more dimers could be visualised in isolated crista membrane samples. Therefore, 3234 dimer pairs were extracted from the crista membrane data for sub-tomogram averaging. After classification, a map of the C. elegans ATP synthase dimer was determined from 1755 dimer pairs (Figure 1C, Supplementary Figures S1 and S2). Both the central and peripheral stalks were resolved clearly (Figure 1C).

Previous studies revealed a type I dimer angle of ∼86° across a range of mammalian and fungal species [10–12,22,26]. The architecture of the membrane-bound C. elegans ATP synthase dimer is unlike any other species studied so far, with an average angle of 105° between the dimer heads (Figure 2A). A comparison to the structure of the membrane-bound S. cerevisiae dimer [11] revealed that the wider dimer angle in C. elegans corresponds with a sharper angle of membrane curvature (50° compared with 74°) (Figure 2B). Accordingly, a shorter distance is measured between the two ATP synthase central stalks in C. elegans dimers compared with S. cerevisiae (16.5 nm compared with 20 nm), which has the effect of bringing the crista membranes closer together. Intriguingly, the dimer interface in the C. elegans map is also visually different from its S. cerevisiae counterpart (Figure 2B), and indeed all other type I dimers studied to date [10–12,24,26]. This difference is likely attributable to the different complement of dimer interface subunits present in C. elegans compared with S. cerevisiae (Supplementary Table S1, Figure 2C). We also analysed the inter-dimer distance and angle between dimer heads in consecutive dimers in the oligomeric rows. This revealed an inter-dimer distance of 12.5 nm and angle between dimer heads of 20° (Supplementary Figure S3). Despite differences in dimer angle, these values are consistent with those reported previously for the type II dimer from green algae (Polytomella sp.) [22], suggesting that dimer angle does not influence oligomerisation of ATP synthases into rows.

We hypothesised that the wider dimer angle associated with sharper membrane curvature in the C. elegans ATP synthase dimer (Figure 2B) would produce flatter cristae with a larger surface area to volume ratio. To test this, tomographic data of whole mitochondria from C. elegans and S. cerevisiae were collected and quantified. Qualitatively, C. elegans mitochondria appear to have more lamellar-shaped (or flatter) cristae, with sharp curved ridges, compared with mitochondria from S. cerevisiae (Figure 3A, Supplementary Movies S1 and S2). The surface area and volume of the crista membranes were quantified, to reveal that the surface area to volume ratio of the average crista membrane was significantly higher (∼1.5 fold, ****P ≤ 0.0001) in C. elegans than in S. cerevisiae (Figure 3B). In accordance with this, the average crista width in C. elegans was less than that observed in S. cerevisiae (Figure 3C–E), suggesting that dimer angle exerts influence on mitochondrial morphology at the level of membrane curvature.

Mitochondria are dynamic organelles, and crista morphology can be influenced by a wide range of factors such as metabolic state [41–44]. However, the average ATP synthase dimer angle remains consistent when imaged in membranes or on purification in detergent [10,27]. Our findings on crista width in whole mitochondria (Figure 3E) are corroborated by results obtained in isolated cristae containing either C. elegans or S. cerevisiae ATP synthase dimers (Figure 2B) [11]. This indicates that the dimer angle and corresponding angle of membrane curvature are consistent, irrespective of the method employed for sample preparation or analysis.

We observed extra mass at the C. elegans dimer interface (Figure 2B) not previously observed in other type I structures determined to date [10,26]. Nematodes are missing subunit 8 [34] (Supplementary Table S1, Figure 2C), which plays a key structural role in other species [22,23,28,45] and is considered essential for respiration owing to its position on the mitochondrial genome [37,38]. Therefore, it is likely that other subunits undergo rearrangements at the dimer interface to compensate for the lack of subunit 8 in nematodes, which could contribute to the observed change of dimer angle. To explore this possibility, we performed multisequence alignments of dimer interface and peripheral stalk subunits with C. elegans, S. cerevisiae and B. taurus [46–48]. This revealed significant extensions in three C. elegans subunits located at the dimer interface (e, f and g), and in subunit b in the peripheral stalk (Supplementary Figure S4). A range of more subtle changes were also identified in subunits d and F₆. To investigate if the changes in the dimer interface and peripheral stalk subunits could account for the extra mass observed at the dimer interface (Figure 2B), we built a homology model of the C. elegans ATP synthase. Prior to model building, mitochondrial targeting sequences were predicted and removed to allow structure prediction of mature protein subunits [49,50]. Mass spectrometry confirmed that the extensions identified by sequence alignment in the dimer interface subunits are present in the mature proteins and corroborate mitochondrial targeting sequence predictions (Supplementary Figure S5).

The ATP synthase dimer is too large to predict the structure as a single multimer; therefore, we used AlphaFold [39] and AlphaFold multimer [40] to predict the structures of individual or small groups of subunits (Supplementary Figure S6 and Table S2). Considering that protein–protein interactions are likely important at the dimer interface, we predicted the dimer interface and peripheral stalk subunits both as individual subunits and as multimers. The peripheral stalk subunits were predicted successfully as a multimer, whereas the multimeric prediction for the dimer interface was poor. This could be explained by a limitation of the AlphaFold multimer, which does not take stepwise assembly of complexes into account, instead assembling all proteins into a multimeric complex simultaneously [40]. The result may also be attributable to the unique dimer interface in C. elegans compared with previously determined structures. The predicted C. elegans structures were then fitted sequentially into a scaffold provided by the B. taurus ATP synthase dimer (PDB 7AJB [27]) (Supplementary Figure S7 and Table S2). The atomic model of B. taurus was chosen as a scaffold, as like C. elegans, it is a metazoan, and contains an equivalent number of subunits (eight) in the c-ring [8]. The C. elegans ATP synthase dimer model was then split into monomers and each was fitted sequentially into our sub-tomogram average dimer map (Supplementary Figure S7, Figure 4A), improving the fit considerably (Supplementary Figure S8). The C. elegans homology model correlated well to the sub-tomogram averaging map (Supplementary Figures S8, S9 and Table S3), providing us with a useful working model to investigate potential protein–protein interactions and allow a comparison of S. cerevisiae and C. elegans ATP synthase dimers (Figure 4B).

To investigate the relative position of subunits required for the oligomerisation of dimers into rows, we fitted the C. elegans ATP synthase dimer model into a row of oligomeric dimer pairs along the curved edge of a crista membrane (Figure 4C–F). This reveals potential inter-dimer interactions mediated by subunit e (Figure 4D,F and Supplementary Figure S10). This is supported by recent work demonstrating the key role that subunit e plays in oligomerisation and row formation [21].

To analyse the dimer interface and peripheral stalk subunits in more detail, we overlayed the individual C. elegans predictions with their equivalents in the S. cerevisiae atomic model (PDB 6B8H) [28]. This revealed extensions in the C. elegans dimer interface subunits e, f and g, and peripheral stalk subunits b, d and F₆ (Figure 5A), which is corroborated by our sequence alignments and mass spectrometry data (Supplementary Figures S4 and S5). The extra mass identified at the C. elegans dimer interface (Figure 2B) appears to be filled by a rearrangement of these extended subunits (Figure 5B–E). Specifically, extensions and rearrangements to subunits e and g in C. elegans relative to S. cerevisiae appear to induce a sharper membrane curvature (arrowhead in Figure 5B,C). The rearrangement of subunit f places its C-terminal alpha helix in the approximate position of the missing subunit 8 (boxed in Figure 5B,C). However, we cannot exclude the possibility that there are additional subunits yet unidentified in C. elegans that may also contribute to the dimer interface.

Owing to the essential and universal role of the ATP synthase across eukaryotic species, it is remarkable that the dimeric interface can be so variable [10,26]. Until now, the arrangement of ATP synthases in invertebrates was unknown, as was the correlation between dimer angle and whole mitochondrial morphology. In this work, a novel dimer angle for the ATP synthase from the nematode worm C. elegans was discovered. By comparing worm and yeast mitochondria, we correlated a wider ATP synthase dimer angle with flatter crista membranes. Since dimer row formation is known to be instrumental in the formation of curved ridges in crista membranes [11,12,14,22,24], it is consistent that dimer angle influences the extent of membrane curvature.

The C. elegans ATP synthase dimer shows clear extra mass at the dimer interface when compared with other determined structures, which can be attributed to changes in subunit composition. Using sequence analysis, we detected extensions in three C. elegans dimer interface subunits (e, f and g), an extension in the peripheral stalk component subunit b, and a range of more subtle changes in subunits d and F₆. To investigate whether these could bulk out the width of the dimer interface, we built a homology model using AlphaFold [39] and AlphaFold multimer [40]. A recently proposed alternative method employs the prediction of subcomponent structures using AlphaFold multimer based on known assembly intermediates [51]. While conceptually advantageous for constructing a homology model of the ATP synthase dimer, accuracy decreases with increasing chain number and the approach has yet to be evaluated on protein complexes exceeding 30 chains [51].

The extensions in C. elegans ATP synthase subunits e, f and g appear to result in the rearrangement of proteins at the dimer interface relative to S. cerevisiae. In addition, the extension in peripheral stalk component subunit b, and the changes to subunits d and F₆, appear to bulk out the width of the dimer interface. These alterations are associated with an increased level of membrane curvature in C. elegans compared with S. cerevisiae. In particular, the C-terminal extension in C. elegans subunit g is located in an area of sharp membrane curvature. Previous site-specific mutagenesis work of yeast subunit g revealed that the conserved C-terminal GXXXG motif is a key player in ATP synthase dimerisation and maintenance of lamellar crista membrane morphology [52]. Therefore, the C. elegans C-terminal extension to subunit g could be key in determining the extent of membrane curvature.

Some dimer interface subunits present in S. cerevisiae (j, k and 8) are absent from C. elegans. Whilst it cannot be completely excluded that yet unidentified subunits may substitute for one or more of these, we speculate that the absence of subunit 8 in worms [34] highlights an interesting evolutionary divergence. Subunit 8 is usually encoded by the mitochondrial genome, indicating that it is essential [37,38]. Additionally, subunit 8 appears to have a key structural role in joining the dimer interface to the peripheral stalk in all species studied to date [22,23,28,45]. We observed that an alpha helix at the C-terminus of C. elegans subunit f, projecting from the dimer interface to the peripheral stalk, appears to replace the missing density for subunit 8. It is, therefore, possible that subunit f has assumed the structural role of the missing subunit 8, supported by extensions in one or more of the dimer interface or peripheral stalk subunits close by (b, d, e, f or g).

The lack of subunit k in C. elegans (DAPIT in mammals), highlights another intriguing evolutionary deviation. In other species with type I dimers, subunit k is considered to play a key role in cross-linking dimers to generate oligomeric rows [5,23]. We speculate that the lack of subunit k in C. elegans may be compensated for by the extension in subunit e, which we observe to mediate tight inter-dimer interactions in dimer rows.

Mitochondria have evolved their highly convoluted crista membranes to increase their surface area [53], hence accommodating the maximum amount of respiratory chain complexes. This has made it possible for eukaryotic organisms to deal with higher energy demands than prokaryotes [53]. Whilst protein packing is dependent on the protein and lipid composition of the membrane [54], flatter cristae (in C. elegans) compared with wider cristae (in S. cerevisiae) could allow greater packing of respiratory chain complexes along the more lamellar membrane surfaces [10].

It has been suggested that cristae serve as proton concentrators that facilitate a directed flow from the source (proton pumping complexes) to sink (ATP synthase) [9,10]; protons have been proposed to preferably migrate from source to sink along membrane surfaces. In mitochondria with narrower cristae, the higher membrane surface area to a given volume could lead to reduced dissipation of protons into the crista lumen [41], increasing the efficiency of ATP synthesis (Figure 6). This is supported by data establishing a correlation between narrower cristae and heightened metabolic activity in human cells [42–44].

We show that the ATP synthase dimer architecture defines the degree of cristae membrane curvature. We thus propose that the maximal rate of metabolism in an organism is dictated by its ATP synthase dimer angle. Narrow cristae in C. elegans could allow for both efficient packing of respiratory chain complexes, and reduced proton dissipation, allowing for maximum energy production in its soil-based habitat [55], where conditions range from hypoxic to near atmospheric [56,57]. In such a variable environment, narrower cristae may allow C. elegans to capitalise on oxygen when it is available. Interestingly, an even wider dimer angle (120°) has been identified in potato tubers [10], which like C. elegans, tend to reside in low-oxygen environments. Mammals, on the other hand, tend to have a more reliable source of oxygen and a wider dimer angle may not be a necessity. Energy must be invested in forming membrane curvature, and the more highly curved the membrane, the higher the cost [58]. In the case of mammals, the energetic cost of consistently maintaining a high degree of curvature may outweigh the benefit of increased respiratory efficiency.

In summary, we suggest that a range of ATP synthase dimer angles have evolved to meet the energetic needs of different organisms. Future studies geared towards investigating dimer subunit composition, angle and corresponding crista morphology across a range of species inhabiting different environments will be key in providing further support for this hypothesis. We demonstrate that the divergence in ATP synthase dimer architecture relative to yeast and mammalian systems makes C. elegans an ideal model system for further investigation of the role of dimer angle in mitochondrial physiology, health and disease.

All standard reagents were purchased from Sigma–Aldrich (Burlington, U.S.A.).

The C. elegans N2 Bristol strain was maintained at 20°C on 60 mm nematode growth medium (NGM) plates seeded with Escherichia coli OP50. For large-scale preparations, a semi-synchronised population of C. elegans (achieved by starving so that they entered the dauer stage) [59,60] were grown in a liquid suspension of E. coli NA22 in S-basal complete medium [61] at 20°C, shaking at 200 rpm for 3 days to achieve adults. For further details see [32]. S. cerevisiae ‘Bakers's yeast’ S288C derivative strains YPH499 were cultured at 19–24°C in YPGal or YPG medium (1% w/v yeast extract, 2% w/v bactopeptone, 2% w/v galactose or 3% w/v glycerol) until OD 2–2.5 was reached. For further details see [62].

C. elegans and S. cerevisiae were both harvested from liquid cultures by low-speed centrifugation. C. elegans preparation required an additional sucrose flotation step to remove debris. To soften the C. elegans cuticle, the pellets underwent collagenase treatment (1 U/ml collagenase, 100 mM Tris–HCl pH 7.4 and 1 mM CaCl₂), whilst S. cerevisiae pellets underwent dithiothreitol (10 mM DTT, 100 mM Tris-SO₄ pH 9.4) and zymolyase treatment (4.5 mg/g zymolyase, 1.2 M sorbitol, 20 mM potassium phosphate, pH 7.4) to disrupt the cell wall. Pellets from both species were re-suspended in homogenisation buffers. For C. elegans, this was STEG/M (220 mM mannitol, 70 mM sucrose, 5 mM Tris–HCl pH 7.4 and 1 mM EGTA) supplemented with 1 mM PMSF in methanol and 1% (w/v) fatty acid-free BSA. For S. cerevisiae the homogenisation buffer contained 0.6 M sorbitol, 10 mM Tris–HCl pH 7.4, 1 mM PMSF, 0.2% (w/v) BSA, 2 mM magnesium acetate. The re-suspended C. elegans or S. cerevisiae samples were homogenised in a glass-Teflon Potter homogeniser to break open cells. Both samples were subsequently spun at low speed (750–3000×g for 5–15 min) to remove cell debris and nuclei, and then at higher speed (12 000×g for 15 min) to pellet mitochondria. Purified mitochondria were re-suspended in buffers that were optimised to maintain intact mitochondria: STEG/M for C. elegans or 250 mM sucrose, 2 mM magnesium acetate, 10 mM MOPS-KOH pH 7.2 for S. cerevisiae.

Crista membranes used for the sub-tomogram averaging experiments were generated by successive freeze–thaw cycles of mitochondria at −80°C. To purify mitochondrial membranes from other cellular material, membrane extracts were incubated for 1 h at 4°C with an anti-NDUFS3 primary antibody (ab14711; Abcam) against the matrix arm of complex I from C. elegans, followed by a 3 h incubation with an anti-mouse secondary conjugated to a quantum dot emitting at 625 nm (Q22085; Invitrogen). Crista membranes were separated from unbound antibodies and other cellular material on an Optiprep gradient with 10 layers (200 µl volume each) ranging from 0 to 27% v/v of iodixanol in STEG/M buffer, by centrifugation at 80 000×g for 30 min at 4°C using a TLS-55 rotor (Beckman Coulter Inc., Miami, FL, U.S.A.). Crista membranes were identified and removed based on fluorescence under a UV lamp. Samples were then diluted in STEG/M buffer to wash out the iodixanol, and spun at 20 000×g for 15 min at 4°C to pellet the membranes. The enriched cristae were again re-suspended in the STEG/M buffer.

Whole mitochondria or crista membranes were mixed 1:1 with 10 nm gold fiducials (Aurion, Wageningen, The Netherlands), applied to glow-discharged holey carbon EM grids (Quantifoil, Jena, Germany), and blotted for 5–6 s, followed by plunge-freezing in liquid ethane using a Vitrobot Mark IV (ThermoFisher, MA, U.S.A.) for C. elegans, or a home-made device for whole S. cerevisiae mitochondria. Pre-screening of C. elegans grids was carried out using an FEI Tecnai Spirit 120 kV microscope (ThermoFisher), with a OneView CCD Camera (Gatan, Pleasanton, U.S.A.). CryoET was performed using the same microscope for whole mitochondria, or using a 200 kV Talos Arctica (ThermoFisher) for crista membranes, equipped with a K2 direct electron detector camera and a GIF Quantum LS energy filter (Gatan). CryoET of whole S. cerevisiae mitochondria was performed using a 300 kV Titan Krios (ThermoFisher), K2 direct electron detector camera and a GIF Quantum LS energy filter (Gatan). Single tilt image series’ (±60, step size 1.5°–2°) were collected at −5 to −8 µm underfocus at nominal magnification of 21 000× for whole mitochondria and 39 000× for crista membranes, corresponding to 5.4 and 3.58 Å pixel sizes, respectively, for C. elegans, or 26 000× for whole mitochondria from S. cerevisiae, corresponding to a 4.51 Å pixel size. The total dose per tomogram was ∼120 e⁻/Ų for whole mitochondria, and ∼80 e⁻/Ų for isolated cristae. Tomograms were aligned using the gold fiducials in IMOD (University of Colorado, U.S.A.) [63] and volumes reconstructed via weighted back-projection. Contrast was enhanced by nonlinear anisotropic diffusion (NAD) filtering [64], followed by manual segmentation, also in IMOD. ImageJ [65] was used to generate movies of segmentations generated in IMOD.

3234 C. elegans ATP synthase dimers were picked manually in IMOD, using NAD-filtered tomograms. Sub-volumes containing the ATP synthase dimer were then extracted from tomograms that had not been NAD filtered. These sub-volumes were CTF corrected and imported into RELION 3.1 [66] using the approach and script described in [67]. A reference-free initial model was generated using 3× binned sub-volumes and 2481 dimers were selected by 2D classification for an unbinned refinement. Finally, 1755 dimers were selected from a 3D classification of this refined model to enter a final round of refinement and post-processing, resulting in a 36 Å resolution map. Supplementary Figure S1 details the full workflow.

AlphaFold was used to predict five structural models of each ATP synthase subunit in C. elegans based on their mature protein sequence [39]. Mature sequences were determined using MitoFates [49] and TargetP-2.0 [50] to predict mitochondrial targeting sequences. Predicted cleavage sites were interrogated manually to confirm whether consensus site rules were adhered to. If the mean probability of cleavage in both softwares was >0.5, and a valid consensus sequence identified, the corresponding mitochondrial targeting sequence was removed before submitting the FASTA file to AlphaFold for structure prediction. All ATP synthase subunits known to be present in C. elegans were included, excepting a putative homologue of subunit j, on account of its poor alignment with other homologues, and absence of any corresponding peptides in mass spectrometry analysis of the C. elegans dimer. The structures of peripheral stalk subunits b, d and F₆ were predicted using AlphaFold multimer [40]. The models for each subunit with the highest average pLDDT score were fitted sequentially to a scaffold provided by the atomic model of the B. taurus ATP synthase dimer (PDB 7AJB) [27] in ChimeraX [68] using the Matchmaker tool. B. taurus was chosen as an organism to provide the scaffold because, like C. elegans, it is also a metazoan and contains the same number of subunits (eight) in the C-ring [8]. Both monomeric and dimeric B. taurus ATP synthase structures exist in different catalytic states [23,27]; there were no significant differences apparent in the homology model depending on the model chosen. Therefore, we selected the B. taurus dimer as a scaffold [27], enabling clearer visualisation of differences in dimer angles between species. Where a subunit had more than one isoform, the version with the highest pLDDT score was used. In the case of subunit b, the isoform with the highest pLDDT score is also the only isoform expressed in somatic tissues [69]. Due to the differences in dimer angle between the C. elegans and B. taurus ATP synthases, the resulting dimeric structure was divided into monomers to fit the C. elegans homology model into the C. elegans map accurately. This was performed sequentially using the ‘fit in volume’ tool in ChimeraX [68]; the workflow is shown in Supplementary Figure S7 and the result in Supplementary Figure S8. To check the reliability of the fit, the resulting homology model was converted into an MRC map using the molmap command in ChimeraX [68]. This map could then be fitted into the sub-tomogram average map of the C. elegans dimer for comparison in Supplementary Figure S9. The yeast monomeric atomic model (PDB 6CP6) [70] was used for analysis in Figure 5A since it contains a more complete chain for F₆ than the dimeric atomic model from yeast (PDB 6B8H) [28].

The ATP synthase was purified from C. elegans mitochondria using a method described previously [71,72], and analysed by Nano-LC mass spectrometry. Briefly, isolated mitochondria were solubilised and mixed with a His-tagged inhibitor protein IF₁. This suspension was applied to a Nickel column to capture inhibited ATP synthase. The fraction most enriched in ATP synthase subunits was taken for mass spectrometry analysis. Further details are given in Supplementary Material.

The sub-tomogram averaging maps generated in this study have been deposited in the Electron Microscopy Data Bank (EMDB) under accession code EMD-18991. The raw data have been deposited to the Electron Microscopy Public Image Archive (EMPIAR) under accession number EMPIAR-11542. The Source Data accompanying Figure 3B,E can be found in the accompanying Source Data file.

The authors declare that there are no competing interests associated with the manuscript.

E.B. was supported by the Biotechnology and Biological Sciences Research Council-funded South West Biosciences Doctoral Training Partnership [DTP2: BB/M009122/1], supervised by V.A.M.G. M.M. was supported by a BBSRC responsive mode grant [BB/R008639/1] grant awarded to V.A.M.G. P.B. was supported by the Foundation for Polish Science First TEAM Programme co-financed by the European Union under the European Regional Development Fund [POIR.04.04.00-00-3F36/17]. B.D. received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme [grant agreement no. 803894]. This work was also funded by the Wellcome Trust (a Wellcome Investigator award [104632] to I.C.), which supported H.F.

Vicki A.M. Gold: Conceptualisation, Resources, Data curation, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualisation, Methodology, Project administration, Writing — review and editing. Emma Buzzard: Data curation, Formal analysis, Validation, Investigation, Visualisation, Writing — original draft. Mathew McLaren: Data curation, Supervision, Validation, Investigation, Methodology. Piotr Bragoszewski: Resources, Methodology. Andrea Brancaccio: Data curation, Formal analysis, Investigation, Methodology. Holly C. Ford: Data curation, Formal analysis, Investigation. Bertram Daum: Resources, Formal analysis, Methodology. Patricia Kuwabara: Conceptualisation, Resources, Supervision, Methodology, Writing — review and editing. Ian Collinson: Conceptualisation, Resources, Supervision, Funding acquisition, Investigation, Methodology.

We thank Rebekah White in the laboratory of Cameron Weadick for sharing equipment, resources and knowledge for ongoing nematode maintenance. We acknowledge Werner Kühlbrandt at the Max-Planck Institute of Biophysics, Frankfurt, Germany, where the S. cerevisiae data were collected. We thank Agnieszka Chacinska at IMol Polish Academy of Sciences, Warsaw, Poland, for supporting the S. cerevisiae-based experiments. We acknowledge access and support of the GW4 Facility for High-Resolution Electron Cryo-Microscopy, funded by the Wellcome Trust (202904/Z/16/Z and 206181/Z/17/Z) and BBSRC (BB/R000484/1), and are grateful to Ufuk Borucu of the GW4 Regional Facility for High-Resolution Electron Cryo-Microscopy for help with screening and data collection. We thank Kate Heesom from the Bristol Proteomics Facility for collecting and analysing mass spectrometry data. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. For the purpose of Open Access, the authors have applied a CC BY public copyright license to any Author Accepted Manuscript version arising from this submission.

cryoET

cryo-electron tomography

NAD

nonlinear anisotropic diffusion

OSCP

oligomycin sensitivity conferral protein