Structure of a Complete ATP Synthase Dimer Reveals the Molecular Basis of Inner Mitochondrial Membrane Morphology

1The mitochondrial F1Fo-ATP synthase produces most of the ATP in the cell by rotary catalysis and plays a crucial role in severe human neurodegenerative disorders. The proton motive force (pmf) across the inner membrane drives the c-ring rotor in the membrane-embedded Fo subcomplex, generating the torque that powers a sequence of conformational changes in the membrane-extrinsic F1 subcomplex, resulting in ATP generation. The Fo subcomplex is connected to F1 by the central stalk, which transmits torque to the catalytic head, and the peripheral stalk, which acts as a stator to prevent idle rotation of the F1 head with the c-ring.

2Dimers of the ATP synthase shape the inner mitochondrial membrane and mediate cristae formation. The ATP synthase forms rows of V-shaped dimers along the highly curved edges of inner membrane cristae. The dimer angle is 86° in yeasts and metazoans, but different in mitochondria of plants and algae. Recently, the complete structure of the dimeric mitochondrial ATP synthase of the chlorophyll-less green alga Polytomella sp. was reported at 6.2 Å resolution, revealing the unexpected feature of a horizontal four-helix bundle in the a subunit of the Fo subcomplex. The long horizontal helices are conserved not only in mammalian mitochondria and bacteria but also in the more distantly related V-type and A-type ATPases, and are thus a fundamental feature common to all rotary ATPases. Together with the c-ring rotor, the horizontal helices of subunit a create two aqueous half-channels on either side of the membrane. The c subunits in the rotor ring bind and release protons as the ring rotates through the alternating hydrophobic environment of the lipid bilayer and the aqueous environment of the half-channels, thereby generating the torque for ATP synthesis.

3The recently reported structures include the dimeric form of an ATP synthase that has unusual peripheral stalks, and the monomer of the bovine complex as well as a bacterial ATP synthase, which both appear to be incomplete. There is currently no structure of an ATP synthase dimer that closely resembles the mammalian complex. Mitochondrial ATP synthases from yeasts have a subunit composition very similar to the mammalian (human) ATP synthase and form the same V-shaped dimers. By a combination of cryoelectron microscopy (cryo-EM) and X-ray crystallography, we have obtained the structure of the complete ATP synthase dimer from the aerobic, genetically accessible yeast Yarrowia lipolytica, in which ATP synthase dimers were previously reported. The combined maps resolve 58 of the 60 known protein subunits and the inhibitor protein IF1. The structure reveals the previously unknown subunit architecture of the dimer interface in the membrane, thereby providing major new insights into mitochondrial membrane architecture.

4ATP synthase dimers from Y. lipolytica were purified from dodecylmaltoside (DDM)-solubilized mitochondrial membranes by centrifugation in a digitonin-containing glycerol gradient, followed by anion exchange chromatography. Two-dimensional gel electrophoresis and liquid chromatography-mass spectrometry (LC-MS) indicated that the 2YLF1Fo (Y. lipolytica ATP synthase dimer) fraction contained all ATP synthase subunits, including e, g, and k, which are known as dimer specific (Figure S1; Table S1, available online). The DDM-purified monomeric Y. lipolytica ATP synthase (1YLF1Fo) lacks subunits e, g, and k. The ATP hydrolysis activity of both 1YLF1Fo and 2YLF1Fo is ∼2.25 U/mg. Fo is coupled to 95% and 75%, respectively, as determined by oligomycin inhibition. The lower percentage of coupled complexes in 2YLF1Fo is most likely due to free F1 subcomplexes and detergent in the dimer preparation (Figure S1A). The similarly high activities of 2YLF1Fo and 1YLF1Fo indicate that the two ATP synthase monomers within the dimer operate independently in ATP hydrolysis mode.

5Crystals of the YLF1c10 subcomplex were obtained from the 1YLF1Fo complex. Whereas previous crystallographic studies of similar complexes used an excess of nucleotide substrates or inhibitors to trap functional states, we crystallized YLF1c10 without any such additives to ensure similar conditions for crystallography and cryo-EM. The 3.5 Å X-ray structure of YLF1c10 (Table 1; Figures S2A–S2D) reveals that all three non-catalytic α subunits bind Mg·ATP in their nucleotide sites. Of the three catalytic β subunits, one is empty (βE), while both the βDP and βTP sites contain Mg·ADP (Figures 1 and S2E).

6We determined the structure of the 2YLF1Fo by single-particle cryo-EM (Figure 2A). After 2D and 3D classification, 38,679 particles were selected for reconstruction of a 3D map with C2 symmetry imposed. The central stalks of the two monomers include an angle of ∼100°. Masking one monomer in the dimer during 3D refinement improved the resolution to 6.9 Å for the F1 subcomplex and masking the Fo dimer improved it to 6.2 Å, as determined by gold-standard Fourier shell correlation (Figure S4). The long helices in the peripheral stalks and the Fo part of the stator are the best-resolved features (Movie S1). The resolution of the F1c10 subcomplex in the cryo-EM map is slightly lower, due to minor variations in the dimer angle (Figures S3D and S3E) and to differences in rotational position of the rotor assembly. Further classification revealed that the position of the central stalk varies independently in both monomers (Figures S3F and S3G), confirming that the two ATP synthase assemblies in the dimer function independently, as already suggested by the similar ATPase hydrolysis activities of 1YLF1Fo and 2YLF1Fo.

7Classification of the same dataset with one monomer in the dimer masked enabled us to distinguish three different rotational states of the F1 head assembly, with two out of three positions favored, in which the positions of the central stalk differ by ∼120° or 240° (Figures S3H and S3I). Interaction with the central stalk affects the nucleotide binding domains and C-terminal domains of the β subunits (Figure S3I). The three conformations show the three “Boyer states,” open, loose, and tight, of the Y. lipolytica complex as seen in the YLF1c10 crystal structure (Movies S2 and S3), similar to the crystal structure of the bovine F1 complex. The three states are trapped in energy wells, which stall the rotor in defined positions upon dissipation of the pmf by the membrane-solubilizing detergent.

8In the most populated class (45% of the particles; subclass 2 in Figures S3H and S3I), a rod-like density protrudes from the αDPβDP pair close to the peripheral stalk (Figure 2D). This density superposes precisely on the inhibitor IF1 in an X-ray structure of the bovine mitochondrial F1Fo-ATP synthase with IF1 bound. The presence of IF1 in ATP synthases prepared from large-scale yeast fermentations is not unexpected, as oxygen concentration of these cultures can decrease, which reduces the matrix pH and triggers IF1 binding, as observed with yeast grown on non-fermentable substrates. The fact that IF1 is found in only one of the three classes is, however, surprising.

9The peripheral stalk consists of several long, well-resolved α helices, which were traced without ambiguity (Figures 2A–2C). Homology models based on crystal structures of the bovine subunits b, d, and OSCP were fitted to the soluble sector of the Y. lipolytica peripheral stalk, which has the same subunit composition (Table S1). Subunit h has only 20% sequence identity to the equivalent bovine F6, accounting for the observed structural differences. The overall curvature of the peripheral stalk differs from that in the bovine crystal structure, but resembles that in the cryo-EM map of the monomeric bovine complex, suggesting that crystal contacts affect stalk curvature. As in the bovine complex, helices 1 and 5 of OSCP on the F1 head are in contact with the N terminus of αE. Two further close contacts are found at the N terminus of αTP, which interacts with helices 4 and 5 of OSCP, and at the N terminus of αDP, which intercalates between the peripheral stalk helices. The N terminus of this α subunit forms a previously unrecognized four-helix bundle with b, h, and the C terminus of OSCP, which positions the F1 head and bonds it to the peripheral stalk (Figure 2B). The contacts in this interaction are mainly hydrophobic, except for those mediated by the conserved residues αGlu33 and αArg41. The d subunit interacts with the C terminus of the αDP subunit, displacing it toward the peripheral stalk by 5 Å relative to the YLF1c10 X-ray structure (Figure 2C). Below the F1 head, peripheral stalk subunits d and b bend toward the central stalk. The density of subunit b, which is thought to have two trans-membrane helices at its N terminus (Figure S4), continues without interruption into the membrane.

10The Y. lipolytica Fo stator subcomplex comprises the eight membrane protein subunits a, b, e, f, g, i, k, and 8. The Fo part of each monomer contains ten well-defined α-helical densities enveloped by a detergent micelle that features the characteristic ∼90° dimer membrane curvature (Figure 2A). Six of these densities indicate trans-membrane α helices, numbered 1–6 in Figures 3A and 3B and assigned in Figure 3C. The loops connecting the helices are, for the most part, not visible at this resolution, but the helix segments can be identified on the basis of sequence comparison, secondary structure predictions, proximity, and known helix topology.

11Helix 1 is the continuation of the peripheral stalk subunit b and is thus the second trans-membrane helix of b. Helix 2 is close to it and is the most likely candidate for the first trans-membrane helix of this subunit. The second-nearest helix 3 is too far for the short, six-residue loop connecting the two trans-membrane helices of subunit b (Figure S4A).

12Sequence alignment of subunit a indicates a consistent pattern of seven characteristic consecutive protein regions (Figure S5A): (i) the hydrophilic N terminus; (ii) a ∼20 residue hydrophobic stretch indicative of a trans-membrane helix; (iii) a region rich in hydrophilic and polar residues, prone to form an amphipathic helix (Figure S5B); (iv) two hydrophobic sequences separated by charged or polar side chains; (v) a region with several positively charged residues followed by (vi) a proline-rich region; and finally (vii) an extensive hydrophobic stretch with interspersed, highly conserved charged and polar residues.

13We can assign region (iii), the amphipathic helix aH2, to the straight helix density on the matrix side just above the horizontal four-helix bundle (Figure 4). Region (ii), the trans-membrane helix of subunit a, which we refer to as aH1, would thus be helix density 3 in the map (Figures 3 and 4). The N-terminal region (i) of subunit a is small, is without clear predicted secondary structure, and has no discernible map density. Regions (iv) to (vii) are assigned to the two membrane-intrinsic helix hairpins of subunit a (which we refer to as aH3 to aH6) on the basis of their striking similarity to the same feature in the Polytomella dimer map. The assignment of the two shorter helices as aH3 and aH4 follows from their proximity to the amphipathic helix aH2 (Figure 4). The non-helical regions (v) and (vi) link the two helix hairpins, but only limited density is visible for them in the map. We assign the longest helix in the four-helix bundle, which follows the curve of the c-ring closely, to aH5 in the first half of region (vii), and the second helix in this hairpin to helix aH6 in the C-terminal half of this region (Figure 4). Helix aH5 contains the essential Arg182, which interacts with the protonatable c-ring glutamate. Our assignment places this residue and a series of conserved charged or polar residues in the long horizontal hairpin at the subunit a/c interface (see Discussion). Our a subunit assignment is fully consistent with that of the bovine and Paracoccus ATP synthase, but the order of helices aH5 and aH6 with respect to the Polytomella assignment is reversed.

14Helix 4 (green in Figure 3) has a short matrix extension with a slight kink toward the c-ring. We assign this density to the small, 48-residue subunit 8 (Figure S4B). Subunit 8 has a conserved N-terminal MPQL motif located in the intermembrane space (IMS), followed by a trans-membrane helix, terminated in yeasts by the conserved Pro33, and a short hydrophilic C-terminal stretch. This sequence fits the density well, with Pro33 at the kink. The trans-membrane helix of subunit 8 has a short connecting density in the IMS toward the c-ring and below the first helix hairpin of the a subunit, which accommodates the conserved MPQL motif. Thus, the N terminus of subunit 8 appears to anchor the horizontal four-helix bundle of subunit a in its position within the Fo assembly.

15The longer mammalian subunit 8 has been shown to interact at its C terminus with the peripheral stalk subunits b and d; in plant mitochondria, subunit 8 is as long as a typical b subunit. This subunit was thought to have no prokaryotic equivalent, but comparison with the b subunit of α-proteobacteria, which share a common ancestor with mitochondria, strikingly reveals the same N-terminal MPQL motif. Therefore, the mitochondrial subunit 8 derives from one of the two b subunits of its bacterial ancestor and is truncated in the mammalian and fungal lines. Subunit 8 is one of the few mitochondrially encoded ATP synthase components in Y. lipolytica, together with the a and c subunits, consistent with its bacterial origin.

16Helix 5 (lavender in Figure 3) is the most likely candidate for the nuclear-encoded trans-membrane subunit f. In yeast, this 100-residue subunit has a hydrophilic N-terminal domain on the matrix side and a predicted C-terminal trans-membrane helix (Figure S4C). Three curved densities at the base of the peripheral stalk (Figure 3B) that surround the matrix extension of subunit 8 are assigned to the N terminus of subunit f. The sharp changes in direction between the densities assigned to this subunit are consistent with the positions of conserved prolines in the f subunit sequence alignment.

17Finally, the density of helix 6 (orange in Figures 3A and 3B) is weaker than the others. Based on its position next to the a subunit, we assign it to the yeast-specific, non-essential subunit i, which is present in both the monomer and the dimer in Y. lipolytica (Table S1) and has been shown to interact with subunits a, f, d, and g.

18Our assignments are fully consistent with all previously reported chemical crosslinking results of ATP synthases from yeasts, metazoans, and bacteria (Figure S6).

19We collected a cryo-EM dataset of 1YLF1Fo and generated a 3D map of the monomeric Y. lipolytica ATP synthase at 8.4 Å resolution (Figure 5). Unlike the dimer, 1YLF1Fo does not contain the dimer-specific subunits e, g, and k (Table S1). The bovine monomer has subunits e and g, but not k. A comparison of the 3D map volumes therefore reveals the location of e and g in the dimer map (Figures 5D and 5F). They occupy a roughly triangular density on the dimer interface next to the N-terminal trans-membrane helices of subunit b, with a narrow extension that protrudes ∼40 Å into the IMS. This density is similar to the e/g density assigned in the bovine monomer, but the orientation of the IMS extension is different (see below).

20Subunit e is predicted to have an N-terminal trans-membrane helix with a conserved, essential GxxxG motif, a signature of helix-helix interaction, and a hydrophilic C terminus that would account for the IMS extension. The g subunit can be crosslinked to the N terminus of b in the matrix. Deleting the first trans-membrane helix of b results in the loss of g and dissociation of the dimer, indicating that g contributes to dimer stability.

21Subunit g consists of an N-terminal matrix domain and a predicted C-terminal trans-membrane helix that likewise contains a conserved GxxxG motif. Subunits e and g may thus form a tight heterodimer in the membrane via their GxxxG motifs. The helices in such a tight heterodimer would not be resolved at 6.2 Å, like the inner helices of the c-ring, which are known to interact through such motifs. There is no contact between the e/g density of one monomer to any subunit of the other, so e and g do not participate directly in dimer formation. Side views of the bovine and Y. lipolytica maps (Figure S7A) indicate that each e/g heterodimer bends the membrane by ∼50°, resulting in the ∼100° kink observed in the dimer. The most prominent direct dimer contact is formed by the C-terminal domain of subunit f (Figure 6). The C terminus of subunit f contains conserved charged and polar residues that would mediate this interaction (Figure S3C). The membrane curvature induced by subunits e and g appears to be necessary to position the C-terminal domains of the f subunits in both monomers for interaction across the interface, resulting in dimer formation.

22A detailed comparison of the YLF1c10 crystal structure to the bovine and Saccharomyces cerevisiae F1 and F1c10 complexes (Figure S8) reveals that the three conformational states of the corresponding α/β heterodimers are very similar in the two yeast species, with root-mean-square deviation (RMSD) values below 1.7 Å, while nucleotide binding and C-terminal regions of the bovine β subunits differ (Table S2; Movies S4 and S5). In YLF1c10 both the βE and the βDP site are more open than in the bovine complex, while their βTP sites are similar. Overall, the three β subunits resemble one another more closely in Y. lipolytica than in the bovine and S. cerevisiae complexes (Figure S8B).

23Aligning the γ subunits in all F1 X-ray structures and comparing the relative positions of the conserved P loop in the βDP subunit (Figure S8A), we find that the YLF1c10 P loop is shifted to a position that, in bovine F1, would indicate a post-hydrolysis or pre-product release state. The post-hydrolysis position of the γ subunit in YLF1c10 agrees with the presence of bound Mg·ADP in the catalytic βDP and βTP sites. Since YLF1c10 was crystallized without added nucleotides, the ADP originates from ATP hydrolysis during isolation or crystallization. The fact that ADP is present in the Y. lipolytica βDP at this late stage of hydrolysis without addition of nucleotide-stabilizing azide is surprising, as ADP was not found in the binding sites of other F1 complexes crystallized under similar conditions. This might indicate a higher nucleotide affinity of the Y. lipolytica βDP site. In contrast, phosphate (Pi) was not detected in the βDP and βTP sites, in line with a possible alternative leaving route for Pi, as described for S. cerevisiae F1.

24We found the inhibitor protein IF1 bound to one of the three different rotational F1 states in the cryo-EM maps, but not in the YLF1c10 X-ray structure, indicating that it was lost during crystallization. By contrast, IF1 was present in all seven rotary states in the cryo-EM maps of monomeric bovine F1Fo-ATP synthase, as would be expected since the complex was purified by IF1 affinity chromatography. Apart from the absence of IF1 in the βDP region (Figure 2D), the 3.5 Å YLF1c10 crystal structure matches the dimer map closely (Figures 2A, S3H, and S3I). Like the X-ray structure, the cryo-EM map therefore shows a post-hydrolysis state.

25The subunit a structure is remarkably conserved in F1Fo-ATP synthases. Densities for all six helices of our Y. lipolytica structure are also present in the same orientations in the cryo-EM map of the bovine heart monomer, while four helices are present in the recent 4 Å X-ray structure of the bacterial complex (Figure S7B). The Polytomella cryo-EM map has elements that correspond to each of the six a subunit helices in Y. lipolytica, even though the polypeptide sequences diverge. The other Polytomella stator subunits do not resemble those of yeasts, metazoans, or bacteria.

26The cryo-EM map of the bovine monomer shows four trans-membrane helices, two of which were identified as belonging to the peripheral stalk subunit b, and one each to subunits a and A6L (subunit 8 in fungi). All helices superpose well on our map and the assignment agrees with ours, except that there is no density for the f subunit in the bovine map. Although the bovine complex prepared according to Runswick et al. should contain the f subunit, this subunit was not identified by Zhou et al. Subunit f may have dissociated during isolation of the bovine monomer, suggesting that it is not firmly attached. As the f subunit is responsible for direct dimer contacts in our assignment (Figure 6), its dissociation from the bovine complex may explain why dimers from mammalian mitochondria are, in our experience, less stable.

27The X-ray structure of the bacterial F1Fo complex from Paracoccus denitrificans shows only two of the trans-membrane helices in the Fo stator, which superpose well on the trans-membrane helix of subunit 8 and aH1 in the Y. lipolytica dimer (panel (iv) in Figure S7B). Therefore, these helix densities, which were unassigned in the Paracoccus map, belong to one of the two bacterial b subunits and the trans-membrane helix of subunit a (aH1), lending strong support to our conclusion that mitochondrial subunit 8 derives from a bacterial b subunit. Surprisingly, the trans-membrane helix of the second b subunit seems to be completely absent in the Paracoccus structure, indicating that it is flexible or disordered in the 4 Å X-ray map.

28Ion translocation through Fo is mediated by the a subunit and the c-ring (Figure 7). A number of high-resolution X-ray structures of ATPase c- or K-rings have shown that the conserved glutamate residues in the c subunits lock the protons (or Na+) in the hydrophobic environment of the lipid bilayer but open to release the ions in a hydrophilic environment. The recent cryo-EM structures of the Polytomella ATP synthase and S. cerevisiae V-type ATPase indicate two aqueous half-channels at the subunit a/c interface that are thought to conduct protons to and from the c-ring protonation sites. We find similar aqueous half-channels in equivalent positions of the Y. lipolytica dimer map (Figures 7A and 7B). Conserved hydrophilic residues of aH5 and aH6 line the aqueous cavity on the matrix side (Figures 7C and 7D). The conserved charged and polar residues of aH5, starting with Glu168 four helix turns upstream of Arg182, create the hydrophilic environment to release the proton from the opposing c subunit glutamate into the matrix. The ∼20° tilt of the aH5/aH6 hairpin places the hairpin loop close to the IMS surface, and the C terminus of aH6 on the matrix side (Figure 7A). Consequently, the lumenal half-channel near the hairpin loop and the matrix half-channel at the C terminus of subunit a are laterally offset, as anticipated. The proton entrance channel on the IMS side is likely to include the conserved Asn186 in aH5, Asn106 in aH3, and the exchangeable pair His191/Glu229 in aH5 and aH6, one of which is Glu or Asp in all ATP synthases (Figure S5). Rather than mediating proton release, the essential Arg182 would allow only deprotonated c subunit glutamate side chains to pass (Figure 7D). Removal of this arginine by mutagenesis uncouples ion translocation from ATP synthesis, as it results in futile proton translocation without c-ring rotation.

29Clinical studies show that mutations in aH5 and aH6 impair the functionality or assembly of ATP synthase in human mitochondria, giving rise to severe neuropathological disorders, such as the maternally inherited Leigh syndrome or retinitis pigmentosa. A molecular understanding of the exact ion translocation mechanism is essential for exploring future therapy. The structure of the genetically accessible Y. lipolytica ATP synthase now provides a basis for structural and functional studies to combat these diseases at the molecular level.

30In the Y. lipolytica F1Fo dimer, which contains subunits e and g, the detergent belt is bent by roughly 100°. In the bovine F1Fo monomer, which contains subunits e and g, it is bent by ∼50°, but in the Y. lipolytica monomer, which lacks e and g, the belt is more or less straight (Figure S7A; Table S1). We conclude that e and g are chiefly responsible for inducing the membrane curvature that results in mitochondrial cristae morphology.

31There is no evident function for the C-terminal IMS helix of subunit e in the dimeric complex (Figure 6A). In contrast to the bovine map, this elongated density protrudes straight out of the Fo stator region into the IMS, while in the bovine map it is curved and contacts the central plug of the c-ring. A role in the formation of higher-order ATP synthase oligomers and dimer rows seems unlikely, as the extension is not easily accessible and appears to point in the wrong direction for interaction between adjacent dimers (Figure 5D). Moreover, the dimer spacing along the rows is irregular, which suggests that the inter-dimer protein contacts are dynamic. Instead, the subunit e extension with its predicted coiled-coil propensity may play a role in complex assembly.

32The two Fo subcomplexes in the membrane are separated by a wedge-shaped gap that is ∼40 Å wide on the matrix side and narrows to ∼15 Å on the IMS side (Figures 5G and 6A). On the matrix side, this wedge appears to be filled by lipid or detergent, as there is no distinct protein density. The tip of the wedge on the IMS side contains the dimer contact domain assigned to the C terminus of subunit f (Figure 6). The center of the wedge-shaped gap is bridged by a conspicuous sheet of density connecting the trans-membrane helices of subunits a and 8 (Figure 5G), which may contribute to dimer formation. It is tempting to speculate that this density belongs to the so far unassigned yeast-specific subunit k, a small, partly hydrophilic protein without predicted trans-membrane helix (Figure S4D). The sheet may form a barrier between the membrane leaflet on the matrix side and hydrophilic protein domains on the IMS side of the dimer interface. The matrix half of the wedge-shaped space has the thickness of one membrane leaflet, implying a new and unusual membrane architecture in this region of the dimer.

33The comparatively simple bacterial and chloroplast ATP synthases consist of eight or nine different subunits, which are sufficient for ATP production. The chloroplast ATP synthase has been shown to be monomeric, and no ATP synthase dimers have been reported in bacteria. By contrast, all known mitochondrial ATP synthases form dimers in the membrane that self-assemble into rows. Mitochondrial ATP synthases of yeasts and metazoans have eight additional subunits of so far unexplained structure and function. Our map shows how the mitochondria-specific subunits in the mitochondrial Fo subcomplex are arranged, and that most of them have a role in dimer formation: the N-terminal trans-membrane extension of b anchors e and g to the complex, and the e/g heterodimer induces local membrane curvature, which in turn appears to enable the IMS domain of f (and possibly subunit k) to establish protein-protein contacts across the dimer interface.

34ATP synthase dimer rows are a prerequisite for the formation of inner membrane cristae, a hallmark signature of mitochondrial morphology. Cristae formation extends the membrane surface to accommodate a large number of respiratory chain complexes, making it possible to meet the high energy demands of eukaryotic cells. They also form a mitochondrial sub-compartment that supports a locally increased proton concentration in the confined cristae space. The structure of the mitochondrial ATP synthase dimer thus offers new insights into how mitochondria became the efficient power plants of eukaryotic cells.

35ATP synthase dimers fully competent for oligomycin-sensitive ATP hydrolysis were isolated from mitochondria prepared from large-scale Yarrowia lipolytica cultures and purified by glycerol gradient centrifugation, anion exchange, and gel filtration chromatography. Cryo-EM grids of 2YLF1Fo dimers and 1YLF1Fo monomers were prepared and images were recorded on an in-column energy-filtered JEOL 3200 FSC electron microscope with a Gatan K2 direct electron detector in movie mode. Global beam-induced motion was corrected by movie frame processing. Two- and three-dimensional classification and 3D map refinement were carried out with RELION 1.3. Crystals of the F1c10 subcomplex were grown from concentrated samples of the 1YLF1Fo. X-ray data were collected at beamline PX-II X10SA (Swiss Light Source), and the structure was determined by molecular replacement with a model based on the S. cerevisiae F1c10 complex (PDB: 2XOK).

36T.M. initiated the study; T.M. and W.K. directed the project; V.Z. provided mitochondrial membranes; A.H. purified the protein; D.J.M. devised the cryo-EM data collection procedure; A.H. and D.J.M. collected cryo-EM data; A.H. and J.V. analyzed cryo-EM data; K.P. grew crystals and collected X-ray data; K.P., M.B., and T.M. analyzed X-ray data; T.M., J.V., W.K., K.P., M.B., and A.H. wrote the paper.