Main

Prohibitin 1 (PHB1 or BAP32) and prohibitin 2 (PHB2, prohibitone, BAP37, REA) belong to the ancient and universally conserved stomatin/prohibitin/flotillin/HflK/C (SPFH) family of proteins1,2. A diverse set of critical functions has been attributed to the prohibitins. Although the localization of the prohibitins within the highly convoluted mitochondrial inner membrane (MIM) is undisputed, additional cellular localizations are controversially discussed1. On the basis of biochemical data and on negative-stain electron microscopy (EM) of prohibitin assemblies purified from yeast, it has been suggested that 12–20 PHB1/PHB2 heterodimers assemble into large ring complexes with a diameter of ~20 nm (refs. 3,4). Such ring-like structures might serve as scaffolds for proteins and lipids to form functional membrane domains. Still, direct experimental evidence for the existence, exact localization and number of ring-like prohibitin arrangements in mitochondria is missing.

Structural insights into the assembly of the SPFH family have been provided by studies expressing and purifying oligomers of the Escherichia coli HflK/C proteins in complex with the client bacterial AAA+ protease FtsH. Single particle cryogenic electron microscopy (cryo-EM) analysis of these reconstituted assemblies revealed that alternate HflK/C units assemble in a unique bell-shaped closed cage, containing 12 copies of each subunit5,6. A recent study using cryo-EM of isolated cellular vesicles demonstrated a similar cage-like assembly of flotillin-1/2, another member of the SPFH protein family, indicating that this intriguing structure resembles a common and unifying feature of the entire SPFH family7. However, it remained unclear whether the prohibitin subfamily adopts a different configuration.

Results

Prohibitins localize exclusively to mitochondria

In this study, we performed in situ cryo-EM tomography to visualize the structure of the prohibitin assembly within the MIM. Owing to the conflicting reports on the subcellular localizations of the prohibitins, we first aimed to characterize their localization in human U2OS cells8,9. We found that transient expression of PHB1 or PHB2 fused to the fluorescent protein Dreiklang (DK) resulted in aberrant mitochondrial structures or cytoplasmic localization of the fusion protein, respectively (Extended Data Fig. 1a,b)10. To explore whether these effects are caused by the overexpression of the prohibitins or by their tagging with a fluorescent protein, we employed clustered regularly interspaced palindromic repeats (CRISPR)–Cas9-mediated genome editing to generate human U2OS cell lines expressing PHB1 or PHB2 fused with DK from their respective native genomic loci, ensuring close to endogenous expression levels (Extended Data Figs. 2–4). The mitochondrial network of both heterozygous cell lines (PHB1–DK and PHB2–DK) displayed wild-type (WT)-like morphologies and PHB1–DK and PHB2–DK fusion proteins were localized to the mitochondria (Fig. 1a,b). This suggests that the overexpression, but not the tagging of the prohibitins with fluorescent proteins, induced artefacts.

Fig. 1: Human PHB1 and PHB2 are primarily localized at the CMs.
figure 1

a,b, Confocal microscopy of living cells expressing PHB1–DK (a; n = 3 independent experiments) and PHB2-DK (b; n = 3 independent experiments) at endogenous levels. Mitochondria were labelled using MitoTracker deep red FM (magenta). DK fluorescence is shown in green. c, Immunogold EM of PHB1–DK and PHB2–DK (top) and WT (bottom) cells. Representative micrographs for each decoration with the indicated antibody are shown and the localization of gold particles is indicated by boxes. d, Gold particle density in the CM and IBM (normalized to the same membrane length unit). e, FRAP analysis of PHB2–DK compared with mito-DK. The circle represents the area before (0 s) and after photobleaching in a representative experiment using mito-DK (1.9 s, n = 16 cells) or PHB2–DK (43 s, n = 20 cells), respectively. f, Recovery curves of PHB1–DK and PHB2-DK compared with mito-DK (inset). Scale bars, 10 µm (a and b), 200 nm (c) and 2 µm (e). Data are presented as mean values ± s.d.

Source data.

Full size image

To investigate whether endogenously expressed tagged PHB1 or PHB2 interacted with untagged prohibitins, co-immunoprecipitation experiments using nanobodies against DK were performed on mitochondria purified from knock-in cells (Extended Data Fig. 5). When PHB1–DK and PHB2–DK were used as bait proteins in co-immunoprecipitation, untagged PHB1 and PHB2, but not inner membrane proteins, such as ATP5A or COX2, were pulled down. This indicates that the tagged prohibitins formed complexes with endogenous untagged PHB1 and PHB2.

Prohibitins are enriched at CMs

To determine the distribution of PHB1–DK and PHB2–DK within the MIM, we performed quantitative immunogold EM on both knock-in cell lines11,12. A highly specific antibody against DK was employed to label the target proteins. The localization of each gold particle associated with the mitochondria was assigned either to the inner boundary membrane (IBM) or the crista membrane (CM). For PHB1–DK and PHB2–DK we found 84.9% and 90.6% of the gold particles at the CMs, respectively. Additionally, with antibodies directed against endogenous PHB1 and PHB2 we found 88.6% and 88.9% of the gold particles at the CMs, respectively (Fig. 1c). When considering that the ratio of IBM to CM is 0.64:1 in the U2OS cells (Extended Data Fig. 6a), it can be calculated that the concentration of prohibitins is three to five times higher in the CMs than in the IBM (Fig. 1d). This finding was somewhat surprising as in budding yeast the prohibitins are almost evenly distributed between CM and IBM13.

Crista-resident proteins, such as the respiratory chain complexes, have been shown to exhibit low mobility within the mitochondrial network14,15,16. To test whether PHB1 and PHB2 also exhibit low mobility, fluorescence recovery after photobleaching (FRAP) experiments were performed using genome-edited cell lines expressing either PHB1–DK or PHB2–DK (Fig. 1e,f). The fluorescence recovery was very slow for both proteins (in the minute range), with an immobile fraction ranging from 85% to 90%. In comparison, matrix-targeted DK exhibited subsecond recovery rates with an immobile fraction of <10%. These findings are consistent with the localization of the majority of prohibitins to the CMs.

Abundance of prohibitins

Next, we aimed to determine the average number of PHB1 and PHB2 molecules per cell. To this end, recombinant His-tagged PHB1 and PHB2 proteins were purified from E. coli and used as a reference in quantitative western blotting (Extended Data Fig. 6b). It was determined that, on average, each U2OS cell contains approximately (3.38 ± 0.23) × 106 molecules of PHB1 and (3.46 ± 0.15) × 106 molecules of PHB2. These numbers are in good agreement with previous mass spectrometry studies using various human cell lines17,18. In budding yeast, prohibitins have been postulated to form ring-like complexes made of 12–20 PHB1/PHB2 dimers3,4. For human cells, no direct evidence for prohibitin rings was available, but if we were to assume that human cells also contain rings of 16 prohibitin dimers, a single U2OS cell would contain approximately 2.14 x 105 prohibitin rings. As 75–84% of all PHB1/PHB2 molecules are localized to the CMs, it could be estimated that, on average, about 1.7 × 105 prohibitin rings formed by 16 dimers would be located on the CMs. To estimate how many prohibitin rings might be located on a single crista, we next determined the average number of cristae in U2OS cells. These cells contain mostly stacked lamellar cristae. To determine their average number, the average total length of the mitochondrial network and the average number of cristae per micrometre mitochondrial tubule was determined using light microscopy and EM, respectively (Extended Data Fig. 6c,d). The total mitochondrial network length of U2OS cells was 1,082 ± 360 μm and the stacked CMs had a distance of 74 ± 15 nm. As groups of stacked lamellar cristae are separated by membrane voids19, which cover 21 ± 8% of the mitochondrial tubules, the corrected average distance between two CMs was 94 ± 19 nm (Extended Data Fig. 6e,f). Based on these measurements, an average of 11,510 cristae is estimated per U2OS cell.

On the basis of these data and the assumption that U2OS cells exhibit prohibitin rings consisting of 16 dimers, it can be calculated that each lamellar crista is expected to carry, on average, ~15 prohibitin rings, so that each crista side would exhibit 7 or 8 prohibitin rings. Assuming a ring diameter of 20 nm, these rings would cover approximately 0.7–1.4% of the available crista surface4. We reasoned that if in U2OS cells indeed about 1% of the cristae surface is covered by a sizeable ring-like structure, these structures should be visible on cryo-electron tomography (cryo-ET) and potentially amenable to subtomogram averaging.

Consequently, we aimed to identify about 20-nm-sized ring-shaped structures in the CMs of human cells. To this end, vitrified human U2OS cells were thinned into ~150-nm-thick lamellae using cryo-focused ion beam milling and then imaged with cryo-ET20.

Convex structures in the inner membrane are formed by prohibitins

We observed numerous convex structures within the MIM. The majority of these prominent structures were localized in the CMs and occasionally in the IBM (Fig. 2a(i)). From a top view, these structures appear as a ring with a diameter of about 20 nm, whereas the side view reveals a convex shape with a height of about 9 nm, with the body of the structure pointing towards the intermembrane space, that is, the crista lumen or the space between the IBM and outer membrane (Fig. 2a(ii,iii) and Extended Data Fig. 7a). We concluded that the size, localization and orientation of these structures within the inner membrane were consistent with the proposed prohibitin structure.

Fig. 2: Prohibitins form bell-shaped complexes at the MIM.
figure 2

a, Cryo-tomograms of U2OS cells reveal bell-shaped assemblies at the inner membrane showing (i) a central tomographic slice of a tomogram showing multiple convex structures at the MIM (white arrowheads); (ii) a central tomographic slice at higher magnification (the convex structures feature a hollow core (side view); the lumen of the cristae is indicated with white asterisks); and (iii) a central tomographic slice at higher magnification. The structure appears as a ring (top view). b, Bell-shaped structures are a frequent feature of mammalian mitochondria, with (i) a central slice of an axonal mitochondrion of a rat hippocampal neuron and (ii) a central slice of a mitochondrion of a COS-7 cell. c, Quantification of convex complexes upon downregulation of PHB1 and PHB2. The boxes represent median, 25th and 75th percentiles, whiskers represent 10th and 90th percentiles. Three independent cultures for each condition were analysed (n = 3 independent experiments). Statistical comparisons using a two-sided Student’s t-test. P = 5 × 10−7 and P = 3.1 × 10−8. d, Rendering of the prohibitin subtomogram average (U2OS cells, 2.5 Å pixel size) results in a round bell-shaped assembly, with (i) the side view, (ii) the top view indicating 11 units making up the bell-shaped prohibitin complex and (iii) a clipping of the side view average. The dashed grey lines indicate the location of the MIM. Scale bars, 200 nm (in a (i) and b), 25 nm (in a (ii and iii)) and 10 nm (in d).

Source data.

Full size image

As prohibitins are well conserved, we suspected that similar structures should also be visible in mitochondria of other organisms. Hence, we used cryo-ET to inspect mitochondria from cultured rat hippocampal neurons and COS-7 cells derived from monkey kidney tissue. These features were also present in both cell types, indicating a common structural element within the inner membranes of mammalian mitochondria (Fig. 2b).

To investigate whether the identified structures are prohibitin complexes, we measured their abundance in tomograms taken from WT and PHB1 or PHB2 knockdown (KD) cells. Small interfering RNA-mediated silencing of either PHB1 or PHB2 downregulated the levels of both prohibitins (Extended Data Fig. 7b,c)21,22 and strongly reduced the number of convex structures identified in PHB1 KD cells (257 ± 123 particles µm−3) and PHB2 KD cells (202 ± 92 particles µm−3) compared with WT cells (743 ± 328 particles µm−3; Fig. 2c and Extended Data Fig. 7d). We also observed one to three additional densities at most (>90%) convex structures identified in the tomograms of U2OS PHB1–DK and PHB2–DK cells (Extended Data Fig. 7e–h). These extra densities sat at the top of the convex structure but were absent in WT cells, thereby indicating that they originate from the DK fluorescent protein fused to the C-termini of the PHB1 and PHB2 proteins. Thus, on the basis of these observations, we concluded that the convex structures identified by cryo-ET are indeed prohibitin complexes.

Subtomogram averaging of the prohibitin complex

We next performed subtomogram averaging to obtain an average three-dimensional (3D) map from subvolumes23 and gain high-resolution in situ structural information of the prohibitin complexes. To this end, we manually picked 817 particles and aligned them along the z axis followed by subsequent particle pose optimization in Dynamo (Extended Data Fig. 8a)24. An alignment of two independent half maps resulted in a resolution of 28.1 Å for the initial average (0.143 criterion, 3.97 Å pixel size; Extended Data Fig. 8b). This cryo-EM map revealed a bell-shaped structure that resembles a ring in the top view. We proceeded to extract the contrast transfer function (CTF)-corrected subtomograms in Warp for subsequent automated 3D refinement in Relion 4.0 with no initial symmetry implied25,26. Initially, we aligned all particles at 5 Å pixel size. The resulting cryo-EM map revealed 11 densities in the top view, probably representing subunits of the prohibitin complex (Extended Data Fig. 8c).

On the basis of this refined cryo-EM map, we proceeded with automated 3D refinement of the dataset with C11 symmetry implied. The resulting map showed a sharper representation of the lipid bilayer and provided a clearer view on the protein bell. A Fourier shell correlation revealed a resolution of 18.3 Å for the cryo-EM map (0.143 criterion, pixel size 5 Å; Extended Data Fig. 8d,e). Finally, we re-extracted the subtomograms at 2.5 Å pixel size for automated 3D refinement with implied C11 symmetry, resulting in a final cryo-EM map of the prohibitin assembly at 16.3 Å resolution (FSC, 0.143 criterion; Fig. 2d and Extended Data Fig. 8f–h). Subsequent 3D classification in Relion 4.0 did not lead to further separation of the particles into distinct structural classes.

The final cryo-EM map reveals a circular bell-shaped structure that is embedded into the lipid bilayer with a diameter of 190.2 Å directly above the lipid bilayer. The whole bell has a height of 84.0 Å measured from the lipid bilayer to the top of the particle and a narrowed top with a diameter of 89.0 Å. As the particles were embedded into either the IBM or the CM, the membrane opposing the top of the prohibitin bell was thus the mitochondrial outer membrane or the opposite CM of the crista. Notably, the map reveals a large empty cavity on top of the bilayer enclosed by the bell. The absence of distinct structural classes hints to no permanent occupancy of the gap by large proteins.

The molecular architecture of the prohibitin complex

Next, we sought to model the molecular architecture of the prohibitin complex on the basis of the final cryo-EM map. As the structures of PHB1 and PHB2 have yet to be solved experimentally, we relied on AlphaFold2 to predict the structures of full length PHB1 and PHB2 (Extended Data Fig. 9a). We placed these predicted structures manually into the experimentally obtained cryo-EM map. Numerous possible arrangements were tested but we found that each of the 11 densities in the cryo-EM map can accommodate only one prohibitin molecule. Therefore, the entire complex would consist of 11 prohibitin molecules in total. Ultimately, because biochemical evidence suggests a complex assembly with alternating PHB1 and PHB2 molecules3, successive PHB1 and PHB2 molecules were fitted into the cryo-EM map with the N-terminal transmembrane domains embedded in the lipid bilayer3 and the PHB domains perpendicular to the lipid bilayer. The final molecular model comprises 11 monomeric PHB molecules forming a bell-shaped assembly (Fig. 3a and Extended Data Fig. 9b,c). Obviously, the uneven symmetry implies that at least one pair of adjacent subunits comprises a repeat of the same molecule (PHB1–PHB1 or PHB2–PHB2).

Fig. 3: Molecular modelling of the human prohibitin complex.
figure 3

a, A cartoon representation of the predicted structures of PHB1 (green) and PHB2 (white) molecules positioned into the final cryo-EM map (light grey, transparent) in side view (left) and top view (right), with (i) a complex consisting of six PHB1 and five PHB2 molecules and (ii) a complex consisting of five PHB1 and six PHB2 molecules. b, The surface representation (left) and interfaces of neighbouring prohibitin molecules (right) with the hydrophobicity surface. Orange colours represent hydrophobic side residues (i and ii); the hydrophobic transmembrane domains anchor prohibitins to the membrane (i), the top of the prohibitin complex is stabilized by the hydrophobic C-termini (ii) and the electrostatic potential surface (iii); blue indicates negatively charged side chains and red indicates positively charged side chains.

Full size image

According to our model, PHB1 and PHB2 are both embedded into the lipid bilayer through their N-terminal transmembrane domains in an arrangement that only supports weak interactions between neighbouring PHB N-terminal domains (Fig. 3b(i)). On the contrary, adjacent PHB C-termini interact more strongly and stabilize the centre of the bell top. The coiled-coil domains that form the bell top are stabilized by alternating negatively (PHB1: 232–252, net −3) and positively (PHB2: 244–264, net +5) charged amino acid sidechains (Fig. 3b(ii,iii) and Extended Data Fig. 9c), consistent with previous biochemical data showing that the strongest interaction between subunits is between the helical coiled-coil domains of adjacent prohibitins3,27.

To further test the molecular model, we queried publicly available crosslinking mass spectrometry (XL-MS) data. XL-MS provides in situ structural information by measuring the proximity of amino acids and distance restraints between protein residues28. By analysing crosslinked peptides from previous XL-MS studies, we identified a total of 22 and 60 unique crosslinked peptide pairs within the prohibitin heterometric complex of mice29,30 and humans31, respectively. Among all crosslinks identified, 49 were unique to a single study, 12 were identified in two studies and 5 appeared in all three studies (Fig. 4a and Extended Data Fig. 10a). Owing to the remarkable degree of conservation between mouse and human PHB proteins (a single amino acid difference for PHB1 and 100% sequence identity for PHB2), all identified crosslinks could be successfully mapped to our prohibitin model. In total, 11 unique crosslinks were identified in both human and mouse species, which highlights the high structural conservation of the prohibitin complex. In good agreement with the molecular model, the majority of crosslinks mapping to the prohibitin bell top were under a 30 Å Cα–Cα distance, an empirically determined upper bound distance for most crosslinking studies32 (Extended Data Fig. 10b). This reinforces the idea that the C-terminal alpha-helices of the alternating PHB proteins interact with each other to stabilize the centre of the bell top. We also observed a high number of long-distance links between the globular domains of alternating PHB1 and PHB2 proteins (Extended Data Fig. 10b). However, these crosslinks exceeded the 30 Å distance constraint, suggesting higher flexibility and large-scale protein dynamics within this region.

Fig. 4: XL-MS analysis of the prohibitin complex.
figure 4

a, A schematic representation of the distribution of the intra-subunit crosslinks (magenta), inter-subunit crosslinks (green) and self-crosslinks (orange) identified in Ryl et al.31. The predicted coiled-coil domains are shown in green. b, The self-links identified were only satisfied at the PHB1–PHB1 or PHB2–PHB2 interfaces of the 6PHB1/5PHB2 (i) and 5PHB1/6PHB2 (ii) heteromeric arrangements, respectively. PHB1 is shown in green, PHB2 is shown in white, self-links <30 Å are shown in blue and unsatisfied self-links >30 Å are shown in red.

Source data.

Full size image

We also identified self-linked peptides where the same residues were crosslinked in overlapping peptides (Fig. 4b and Extended Data Fig. 10a), thereby indicating the occurrence of homomeric interactions33. Interestingly, these self-links only satisfied the distance constraint at the PHB1–PHB1 or PHB2–PHB2 (Fig. 4b(i,ii)) interfaces and were violated when mapped to the rest of the PHB heteromeric structure with alternating PHB1 and PHB2 subunits. These results strongly support the 11-fold symmetry model characterized by an uneven number of alternating PHB1 and PHB2 subunits. Importantly, self-links were identified in both mouse and human cells, indicating that the oligomeric organization of the prohibitin complex is conserved across species30,31. Overall, the crosslinking data strongly support our molecular model of the human prohibitin complex.

To test the plausibility of the final model further, we employed molecular dynamics (MD) simulations of the entire predicted complex in GROMACS (Extended Data Fig. 9d–f)34. The PHB domains showed fluctuations relative to the position of the globular domains. These fluctuations included small tilts of the PHB domain to the sides and are more prominent to the N-terminal part of the PHB domain, which is in line with weaker densities in the cryo-EM map at these positions (Extended Data Fig. 9g). This suggests that the N-terminal part including the PHB domain is intrinsically dynamic, while the coiled-coil domains within the prohibitin bell are more stable, consistent with our XL-MS observations. The molecular model remained stable over 100 ns of MD simulation without major alterations, suggesting that it represents a plausible model for the molecular structure of the prohibitin complex in human cells.

Discussion

The overall bell-shaped structure of the prohibitin complex resembles the structure formed by the bacterial prohibitin homologues HflK/C5,6. The HflK/C structure associates with three hexameric complexes of proteases. Intriguingly, the final cryo-EM map of the prohibitin complex did not identify additional proteins, despite the fact that the complex has been demonstrated to be involved in the regulation of mitochondrial AAA proteases35,36,37. Although an association with the mitochondrial AAA protease cannot be ruled out on the basis of the subtomogram averages, the interaction with the prohibitin complex in U2OS cells may adopt a more dynamic configuration, potentially involving fewer protease molecules than the large bacterial HflK/C–FtsH complex5,6.

As the overall bell-shaped structure formed by the prohibitins resembles the structure formed by the HflK/C and the flotillin-1/2 complexes, this arrangement seems to be a core structural principle underlying the function of SPFH proteins5,6,7,38. The molecular model of the prohibitin complex provides insight into how prohibitins establish membrane microdomains within the MIM. Quantification of the number of prohibitin complexes suggests that, on average, about 43 complexes are localized in a single crista (Fig. 5a,b). Their abundance and the fact that the top of the bell-shaped complexes is in close spatial proximity to the opposing CM or IBM might suggest an unexpected role of the prohibitin complexes in the large-scale spatial organization of CMs.

Fig. 5: Prohibitins in the average mitochondrion.
figure 5

a, A cartoon representing the abundance and distribution of prohibitin complexes in a mitochondrion. Right: the view into the crista lumen. b, The mitochondrial key numbers and prohibitin abundance determined in this study. The mitochondrial network length was determined in Extended Data Fig. 6c, intercristae distance determined in Extended Data Fig. 6d–f and PHB molecules per cell determined in Extended Data Fig. 6b.

Full size image

Methods

All procedures with live animals were conducted in the animal facility at the Max Planck Institute for Multidisciplinary Sciences. According to the German Animal Welfare Law, sacrificing an animal is not regarded as an experiment on animals. All requirements of section 4 TierSchG together with sections 2 Satz 2, Anlage 1, Abschnitt 2 and Anlage 2 TierSchVersV were implemented.

The facility is conducted under all aspects of animal welfare. The facility is headed by a veterinarian with special education in laboratory animal science as well as gene technology and molecular genetics.

Only professionally educated animal technicians are in charge of animal husbandry and care. The facility is registered according to section 11 Abs. 1 TierSchG (Tierschutzgesetz der Bundesrepublik Deutschland, Animal Welfare Law of the Federal Republic of Germany) as documented by 33.23-42508-066-§11, dated 16 November 2023 (Erlaubnis, zum Halten von Wirbeltieren zur Versuchszwecken, permission to keep vertebrates for experimental purposes) by the Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit (Lower Saxony State Office for Consumer Protection and Food Safety). According to the Animal Welfare Law of the Federal Republic of Germany (TierSchG) and the Regulation about animals used in experiments, dated 20 December 2022 (TierSchVersV) an animal welfare officer (specialized veterinarian in laboratory animal science) and an animal welfare committee for the institute was established.

Plasmids

Overexpression plasmids

Cloning of constructs for fusion protein overexpression was carried out using the primers listed in Supplementary Table 1. The DK DNA sequence was amplified via PCR from the plasmid pQE31–DK10. PHB1 and PHB2 coding sequences were amplified from plasmids provided by the Human ORFeome clone collection with the internal IDs 6030, 394 and 56919, respectively. The PCR products were purified and used for one-step isothermal assembly with EcoRV-digested pFLAG-CMV-5.1 (Sigma)39.

Nuclease plasmids

Design of guide RNAs (gRNAs) was done using the CRISPR Design Tool (http://crispr.mit.edu). For each gRNA, a forward and reverse oligonucleotide (Supplementary Table 2) was annealed in 1x T4 ligation buffer at a final concentration of 10 µM per oligo in a thermocycler using the following parameters: 95 °C for 5 min and ramp down to 25 °C at 5 °C min−1. The annealed oligoduplex was used in a ligation reaction for insertion into BbsI-digested pX330, a vector described elsewhere40,41. Plasmids containing the gRNA of interest were verified using Sanger sequencing. The final bicistronic vector encoded a gRNA and Cas9 nuclease.

Donor plasmids

DNA sequences for the left and right homology arms were amplified from genomic DNA using the primer pairs listed in Supplementary Table 3. The DK DNA sequence was PCR amplified from the plasmid pQE31–DK10. The PCR products were purified and cloned into EcoRV-digested pUC57 (Fisher Scientific) using one-step isothermal assembly followed by site-directed mutagenesis to remove Cas9 recognition sites39.

Cell culture

U2OS (HTB-96, American Type Culture Collection) and COS-7 cells (87021302, Sigma-Aldrich) were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Invitrogen) supplemented with 10% foetal bovine serum (PAA), 100 units ml−1 penicillin, 100 μg ml−1 streptomycin (Biochrom) and 1 mM sodium pyruvate (Sigma) under constant conditions at 37 °C and 5% CO2. Plasmid transfections were done using FuGENE HD (Promega) according to the manufacturer’s instructions. Mitochondria of living cells were stained using 100 nM MitoTracker Deep Red FM (Fisher Scientific) at 37 °C for 30 min.

Cell culture for cryo-EM

Confluent cultures of either U2OS WT or Cos7 WT cells were first detached by trypsination and subsequently suspended in DMEM complete medium. Next, 3 ml of the cell suspension was placed in ibidi glass-bottom dishes and four coated grids were added onto the bottom of the dish. Cells were allowed to adhere to the grid surface for at least 18 h before plunge freezing.

Hippocampus isolation and dissociation

Rat hippocampi were obtained through resection from Wistar rats. Newborn rats (P0 or P1) were decapitated and the brain was carefully extracted to preserve its integrity. The isolated brain was placed in a 3.5 cm culture dish filled with ice-cold Hank’s balanced salt solution (HBSS) without calcium and magnesium (Thermo Fisher Scientific). After separating the hemispheres and removing meninges, the hippocampus was isolated, placed in a 15-ml Falcon tube and stored on ice.

For dissociation, the hippocampi were placed in 4.5 ml ice-cold HBSS without calcium and magnesium, and 0.5 ml of freshly thawed 2.5% trypsin (Thermo Fisher Scientific) was added. Dissociation occurred for 18 min in a water bath set to 37 °C with occasional gentle shaking. Digestion was halted by adding 10 ml of pre-warmed DMEM supplemented with 5% foetal bovine serum to each Falcon tube, followed by centrifugation at 100g for 3 min. After discarding the supernatant, the tissues were washed twice with 10 ml of prewarmed HBSS without calcium and magnesium.

Subsequently, tissues were dissociated using a 5 ml plastic Pasteur pipette in approximately 5 ml Neurobasal plating medium. The cell suspension was filtered through a 40 µm cell strainer into a 50 ml Falcon tube. The final cell suspension was achieved by diluting the strained cell suspension with pre-warmed Neurobasal plating medium to reach a final cell count of 1 × 106 cells ml−1. Cells were then plated by adding 1 ml of the final cell suspension onto coated Quantifoil grids. Then, 1 h after plating, the supernatant was replaced with standard cultivation medium. Hippocampal neurons were cultivated until 18 days in vitro before freezing.

Generation of knock-in cells

U2OS or HeLa cells were cotransfected with the respective combination of nuclease and donor plasmids for targeting either PHB1 or PHB2. At 7–10 days after transfection, the cells were subjected to single-cell sorting into 96-well plates using a FACSAria II (BD Biosciences). About 2 weeks after sorting, the cells of each well were split and equally distributed into two wells of a 24-well plate with one of the two wells containing a glass cover slip for microscopic inspection. Cells that expressed the fluorescent fusion protein were identified using an epifluorescence microscope (DM6000B, Leica Microsystems) equipped with an oil immersion objective (1.4 NA, 100×, Planapo, Leica) and a BGR filter cube (excitation: 495/15, emission: 530/30). Successfully targeted clones were expanded and genotyped via PCR using the primers listed in Supplementary Table 4.

For DNA sequencing, genomic DNA was isolated from selected clones, the respective on- or off-target site was PCR amplified (Supplementary Tables 4 and 5) followed by purification and ligation into a pCR Blunt II-TOPO vector using a Zero Blunt TOPO Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. Plasmids containing an insert were identified via colony PCR and 15–20 plasmids were sequenced per locus.

siPool-mediated KDs of prohibitins

We transfected U2OS cells with 3 nM of PHB1 or PHB2 siPools (siTOOLs Biotech), respectively following the manufacturer instructions. The cells were first cultivated for 2 days in standard six-well plates. After 2 days, the cells were detached and the cell suspension transferred to poly-l-lysine/fibronectin-coated SiO2 R1/4 grids and allowed to adhere for 24 h before plunge freezing in liquid ethane at melting point. The duration of siPool KDs for both prohibitins (PHB1 and PHB2) was 3 days in total.

Immunoblotting

Cells were grown to 80–85% confluence and detached from the growth surface and counted using a Scepter 2.0 Cell counter (EMD Millipore). Cells were collected by centrifugation followed by lysis of 106 cells in 100 μl of radioimmunoprecipitation assay (RIPA) buffer supplemented with 1 mM EDTA, 1 mM PMSF, 10 U ml−1 universal nuclease (Thermo Fisher Scientific) and 1× complete protease inhibitor cocktail (Roche). After the addition of RIPA buffer, the cell suspension was placed on ice for 30 min with vortexing steps every 10 min. The suspension was centrifuged at 16,000g at 4 °C for 30 min. The supernatant was collected, and the protein concentration measured using the Pierce BCA protein assay kit (Thermo Fisher Scientific). Samples were diluted to 1.2 µg µl−1 with RIPA buffer and mixed with the respective amount of 6× Laemmli buffer (375 mM Tris pH 6.8, 12% sodium dodecyl sulfate (SDS), 60% glycerol, 0.6 M dithiothreitol and 0.06% bromophenol blue) to a final concentration of 1 µg µl−1. The suspension was boiled at 95 °C for 5 min, flash frozen in liquid nitrogen and stored at −20 °C for further use.

Extracts corresponding to a known number of cells as well as recombinant 6xHis-PHB1 or 6xHis-PHB2 (Abcam) were separated on 4–15% Mini-Protean TGX Precast Gels (Bio-Rad) according to the manufacturer’s instructions. Separated proteins were transferred to a nitrocellulose membrane (GE Healthcare) in transfer buffer (25 mM Tris, 190 mM glycine and 20% methanol) at 4 °C for 16 h. The membrane was rinsed in Tris-buffered saline (TBS) with 0.1% Tween-20 (TBST) and incubated in 5% blocking buffer (5 g skim milk per 100 ml TBST) at room temperature (RT) for 30 min. Primary antibodies were diluted in blocking buffer and incubated with the membrane at RT for 1 h. The following primary antibodies were used: anti-PHB1 (EP2803Y, 1:2,000, Abcam), anti-PHB2 (EPR14523, 1:5,000, Abcam), anti-GFP (JL-8, 1:3,000, Clontech), anti-Actin (AC74, 1:3,000, Sigma-Aldrich) anti-COX2 (ab203912, 1:500, Abcam), anti-ATP5A (ab14748, 1:1,000, Abcam), anti-tubulin (ab15246, 1:1,000, Abcam) and anti-ESR1 (sc-8005, 1:1000, Santa Cruz Biotechnology). After washing with TBST, the membranes were incubated at RT for 1 h with horseradish peroxidase-conjugated anti-rabbit or anti-mouse secondary antibodies (Dianova) diluted 1:5,000 in blocking buffer. After washing with TBST, the membrane was incubated with Pierce ECL western blotting substrate (Thermo Fisher Scientific) and exposed to a charge-coupled device (CCD) camera. Membranes were stripped by incubation with Restore (Thermo Fisher Scientific, Waltham) at 37 °C for 30 min followed by the described protocol for reprobing with a different antibody.

Isolation of mitochondria and immunoprecipitation

Mitochondria were isolated from WT, PHB1–DK or PHB2–DK U2OS cells grown to 80–85% confluence, detached from the growth surface and collected by centrifugation. The cell pellet was resuspended in trehalose/hepes/EDTA (THE) buffer (300 mM trehalose, 10 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid–KOH, pH 7.7, 10 mM KCl and 1 mM EDTA) supplemented with 0.1% (w/v) bovine serum albumin (BSA) and lysed using a Dounce homogenizer. Homogenized cells were subjected to centrifugation at 11,000g at 4 °C for 10 min. The mitochondria-containing pellet was resuspended in THE buffer without BSA and the protein concentration was measured using Bradford Protein Assay (Bio-Rad). The protein concentration was adjusted to 10 µg µl−1 with THE buffer.

GFP-Trap-Agarose beads (Chromo-Tek) were equilibrated in ice-cold dilution buffer (20 mM Tris–HCl pH 7.5, 50 mM NaCl, 0.5 mM EDTA, 10% (v/v) glycerol, 0.3% (w/v) digitonin and 1 mM PMSF) and mixed with 800 µg of mitochondrial extract. After rotation at 4 °C for 1 h, beads were centrifuged and washed ten times using dilution buffer. The beads were resuspended in 2× Laemmli buffer (125 mM Tris pH 6.8, 4% (w/v) SDS, 20% (v/v) glycerol, 0.2 M dithiothreitol and 0.02% (w/v) bromophenol blue), boiled at 95 °C and centrifuged. The supernatant was used for subsequent SDS–polyacrylamide gel electrophoresis and western blot analysis.

Fluorescence microscopy

Sample preparation

Cells were cultured on glass cover slips until they reached a confluence of about 70–85% and fixed in 37 °C prewarmed 4% (w/v) formaldehyde in PBS at RT for 5 min. The cells were permeabilized using 0.5% (v/v) Triton-X-100 in PBS for 5 min followed by subsequent incubation in blocking buffer (5% (w/v) BSA in PBS containing 100 mM glycin) for 15 min. Primary antibodies were diluted in blocking buffer and cover slips were incubated with that solution at RT for 1 h. The following primary antibodies were used: rabbit anti-PHB1 (EP2803Y, 1:250, Abcam), rabbit anti-PHB2 (EPR14523, 1:500, Abcam), mouse anti-ESR1 (D12, 1:500, Santa Cruz Biotechnology) and rabbit anti-GFP (ab290, 1:1,000, Abcam). After three washing steps in PBS, fluorophore-coupled secondary antibodies were diluted 1:1,000 and added for incubation at RT for 1 h. The following secondary antibodies were used: sheep anti-mouse and goat anti-rabbit (Dianova) coupled to KK114 (ref. 42) or Alexa 594 (Atto-Tec). After three PBS washing steps, cells were embedded in Mowiol 4-88 mounting medium containing 1 µg ml−1 4,6-diamidino-2-phenylindole and 2.5% (w/v) 1,4-diazabicyclo-[2,2,2]-octane.

Confocal microscopy

Confocal imaging was done using the Leica TCS SP8 Confocal Microscope (Leica). All recordings were performed using a pinhole diameter of one Airy unit (1.22λ/NA), a scan speed of 400 Hz and a 63× oil immersion objective (HCX PL APO CS 63×/1.40-0.60 oil). The following laser lines were used for fluorescence excitation: a 405 Diode (405 nm), an argon laser (458 nm/476 nm/488 nm/496 nm/514 nm) and a helium–neon laser (633 nm). Fluorescence detection was done using photomultipliers and a hybrid detector operated within the dynamic range. Separation of excitation and emission light was accomplished using an acousto-optic tunable filter. Multicolour imaging was done using sequential acquisition between frames. For image digitization a sampling rate according to the Nyquist criterion was chosen. Each image was recorded at least twice for averaging.

FRAP analysis

FRAP measurements were done using a Leica TCS SP5 Confocal Microscope and the FRAP Wizard application (Leica). All recordings were performed using an open pinhole and a 63× oil immersion objective (HCX PL APO CS 63×/1.40-0.60 oil). Living U2OS cells were mounted in a custom-built live-cell chamber and maintained at 37 °C in CO2-independent Leibovitz L-15 medium (Thermo Fisher Scientific) for imaging.

Mitochondria of U2OS cells expressing mitochondrial matrix targeted fluorescent protein Dreiklang (mito-DK) were photobleached in a circular region of interest with a diameter of about 0.5 µm using an argon laser at 20% laser power. Imaging after bleaching was performed at a time interval of 19 ms for about 2 s with the 514-nm line of the argon laser. Mitochondria of U2OS knock-in cells expressing PHB1–DK or PHB2–DK were bleached in a region of interest with a diameter of about 2.5 µm using an argon laser at 20% laser power. Imaging after bleaching was performed at a time interval of 8 s for about 1 min with the 514 nm line of the argon laser.

EM

Plastic embedding

U2OS cells were grown on Aclar polymer cover slips until 80–85% confluence. Cells were prefixed in 2.5% (w/v) glutaraldehyde in 0.1 M sodium cacodylate (pH 7.4) at RT for 15 min postfixed in the same buffer at 4 °C for 15 h. Cells were washed three times in 0.1 M sodium cacodylate (pH 7.4) and incubated in 1% (w/v) OsO4 in 0.1 M sodium cacodylate (pH 7.4) for 3 h. Cells were washed once in 0.1 M sodium cacodylate (pH 7.4) and then twice in water. The cells were place in 0.1% (w/v) uranyl acetate (in H2O) for 30 min. Uranyl acetate was washed out by subjecting the cells to 30% ethanol three times for 5 min followed by dehydration through a 50%, 70% and 100% ethanol series. Afterwards, the cells were placed in 100% propylene oxide for 5 min and then transferred to 50%/50% propylene oxide/Epon for 1 h followed by placement to 100% Epon overnight. Samples were sectioned to 50 nm thickness with a Leica EM UC6 ultramicrotome (Leica EM UC6, Leica Microsystems). Each section was transferred to 0.7% (w/v) Pioloform-coated 200 mesh carbon grids. Samples were subjected to post contrasting using 1% (w/v) uranyl and lead acetate. EM recordings were acquired using a Philips CM 120 transmission electron microscope equipped with a TVIPS 2K × 2K slow-scan CCD camera (Philips).

Immunogold labelling

U2OS cells were grown to 80–85% confluence and fixed in 37 °C prewarmed 4% (w/v) formaldehyde in PBS at RT for 30 min. Further sample processing was done according to previous work12. Samples were sectioned into 80-nm-thin slices and incubated with diluted primary antibodies for 30 min. The following antibodies were used: anti-GFP (JL-8; 1:20, Clontech), anti-PHB1 (EP2803Y, 1:20, Abcam) and anti-PHB2 (EPR14523, 1:40, Abcam). Subsequently each sample was incubated with protein A coupled to 10 nm gold particles for 20 min followed by multiple washing steps and additional contrasting using uranyl acetate/methylcellulose on ice for 10 min. EM recordings were done using a Philips CM 120 transmission electron microscope equipped with a TVIPS 2K × 2K slow-scan CCD camera (Philips).

Determination of crista area occupied by PHB1/2 complexes

First, we determined that the inner diameter of an average mitochondrial tubule in human U2OS cells is 564 nm (Extended Data Fig. 6g). Thus, the area of a circular CM (A = πr²) is around 249,832 nm2. Each lamellar crista consists of two opposing membranes hence the total membrane area per crista is around 499,664 nm2. Second, we assumed a ring-like prohibitin complex with a diameter of about 20 nm which corresponds to an area of about 314 nm2 per prohibitin ring2,4. Thus, a single prohibitin ring occupies 0.063% of the total available crista area while 11–22 prohibitin rings per crista occupy between 0.7% and 1.4% of the available crista surface, respectively.

Cryo-EM

Grid preparation for cell culture

Quantifoil R2/1 or SiO2 R1/4 gold grids (Quantifoil Mirco Tools GmbH) were briefly washed in chloroform and placed on drops of poly-d-lysine (100 µg ml−1). After incubation for 1 h at RT the grids were washed three times on drops of HBSS for 10 min each. Subsequently, the grids were placed on drops of fibronectin (20 µg ml−1) and incubated for 1 h at RT. Finally, the grids were washed three times on drops of HBSS for 10 min each and used directly for cell plating.

Preparation of frozen-hydrated cells

Grids containing cells were picked up with a pair of tweezers and any residual medium was removed by manual blotting at the tweezer tips. The sample was then rapidly loaded into a Vitrobot (Thermo Fisher) set at 37 °C and 95% humidity. We set a wait time of 120 s and applied 3 µl of a solution containing 10% glycerol in HBSS (Thermo Fisher). The grids were then backside blotted for 30 s with blot force 20, and subsequently plunged into liquid ethane at melting point. The grids were kept at liquid nitrogen temperatures until further processing.

Automated preparation of cryo-lamellae

Cryo-lamellae preparation was automated using an Aquilos 2 cryo FIB-SEM (Thermo Fisher). Grids were initially coated with GIS platinum for 20 s. Lamella positions were identified around the centre marker of the grids after automated tile acquisition with Maps (200× magnification, 5 kV, 13 pA). These positions were then transferred to the Cryo AutoTEM with a target lamella size of 14 µm length and approximately 170 nm thickness, with a 4 nm offset after final polishing, resulting in lamellae of approximately 180 nm thickness. Following the automated procedure, the lamellae were additionally sputter-coated with metallic platinum (30 mA, 1 kV, for 1–2 s). The final lamellae were stored at liquid nitrogen temperatures until further imaging.

Cryo-ET and tomogram reconstruction

Cryo-electron tomograms were recorded on a Titan Krios G2 microscope (Thermo Fisher) operated at 300 kV, equipped with a FEG, BioQuantum imaging filter (Gatan) with a 20 eV energy width, and a K3 direct electron detector (Gatan). For automated acquisition of the tomograms we used SerialEM43. Recorded frames were motion corrected using Warp and dose filtered before tomogram reconstruction. Aligned frames from Warp were utilized for patch tracking or fiducial-based alignment of small platinum particles in IMOD for tomogram reconstruction. CTF correction was performed through phase flipping in IMOD. Tomograms for visualization and IsoNet training were reconstructed with binning 6 (11.90 Å pixel size) and an additional SIRT-like filter in IMOD, set to 50 iterations. A comprehensive overview of all acquired tomograms used in this study is provided in Supplementary Table 6.

Missing wedge correction in IsoNet

For missing wedge correction in IsoNet, we employed the tool at binning 6 (11.90 Å pixel size) for visualization and prohibitin quantifications44. IsoNet was trained separately using four tomograms from U2OS WT cells, four from Co7 WT cells, and four from rat hippocampal neurons. CTF deconvolution was not applied before masking. A mask was generated with density and s.d. percentages set at 60, and a z-crop of 0.05. Subtomograms (120 per input tomogram) were extracted, and the network underwent training for 35 iterations with 15 epochs each. Noise was introduced at iterations 16, 21, 26 and 31 with levels of 0.05, 0.1, 0.15 and 0.2, respectively. The resulting model after 35 iterations was used for predicting tomograms depicted in this study.

Subtomogram averaging in Dynamo

For subtomogram averaging, 37 tomograms from U2OS WT cells were utilized, and 817 particles were manually picked in Dynamo at a pixel size of 3.87 Å (ref. 24). An initial average was created by manually aligning 50 particles in dgallery. Subsequently, particles were globally aligned to the initial average using Dynamo’s ‘global search’ pre-set. Randomization of particle orientations was performed using the ‘dynamo_table_randomize_azimuth’ function, and the resulting table was used for particle averaging. Pose-optimized particles were extracted in Warp at a 5 Å pixel size and aligned in Relion 4.0 through 3D auto-refinement, revealing an 11-fold symmetry of the prohibitin complex. All particles were then aligned to a C11-fold symmetrized template. The resulting star file from auto-refinement was used to re-extract particles in Warp at a 2.5 Å pixel size, followed by another round of 3D auto-refinement in Relion. Resolution estimation for all alignments was performed using the EMDB FSC server with two independent half maps generated from the Relion 4.0 3D auto-refinement45. Visualization of 3D models of the cryo-EM maps was conducted using ChimeraX46.

Modelling

For the modelling of the prohibitin complex, we used the Alphafold predicted structures of PHB1 (Uniprot P35232) and PHB2 (Uniprot P50093), respectively. As starting point, the structure of Prohibitin 1 was choosen47. After placement of the globular domain in the density map, backbone rotations between residues 174–178 and 229–233 were made, analogous to the bacterial homologous complex, while a C11 symmetry was maintained using PyMOL6,48. Afterwards, the resultant model was fitted into the cryo-EM derived density map using the MD method from Flex-EM, with a cap-shift of 0.5 Å, 20 runs and 100 iterations per run49. Additionally, all secondary structure elements from the Alphafold model with high confidence in the predicted local distance difference test (pLDDT >70), were treated as rigid bodies. This provided the C-terminal structured part of the interacting helices (residues 176–253), while a new placement of more globular and transmembrane part (residues 1–174) was made, since owing to their membrane interaction, a fitting in a density map including membranes is not anticipated. Using this arrangement, a homology model of an alternating complex of five PHB1 and six PHB2 was built using MODELLER 10.4 (ref. 50). Stereochemical restraints were added as well as helical restraints on the transmembrane domain for PHB2 (residues 19–39). To avoid interference of the longer disordered C- and N-terminal part of PHB2, these areas were redirected during modelling by setting a z-coordinate upper limit for residues 1–19 of 107 Å with a s.d. of 0.3 Å and a lower limit for residues 286–291 of 158 Å with the same s.d. The coordinates frame was identical to the cryo-EM derived map. From 300 models, after ranking by the discrete optimized protein energy score, two were selected and visually inspected51. All-atom systems including solvent and a lipid membrane were built using CHARMM-GUI as described previously52,53. The membrane consisted of POPC lipids in a hexagonal setup. After equilibration, 100 ns of a MD simulation were computed for each setup using GROMACS 2021.3 and the CHARMM36m force filed as described before52,54,55. Density maps were computed using GROmaps using a pixel size of 2.5 Å (ref. 56).

Blender modelling of an average mitochondrion

The mitochondrion was drawn to scale in Blender 4.0.2 using the following data: length, 1.5 µm; outer membrane diameter, 565 nm; inner membrane diameter, 540 nm; distance outer and inner membrane, 10 nm; crista diameter (without junctions), 480 nm; crista thickness, 10 nm; intercristae distance, 94 nm; thickness of all membranes, 5 nm.

Identification and validation of crosslinked peptides

Crosslink peptides were obtained from previously published studies28,29,30 and were reformatted using custom R scripts for visualization using the web-based visualization tool xiVIEW29,30,31,57. We evaluated the crosslinks by mapping the residue pairs onto the prohibitin Protein Data Base structures using the ChimeraX bundle XMAS33. The crosslinks were mapped to the shortest distance while allowing a difference of 2 Å unless otherwise stated. For each residue pair, we extracted the Euclidean distance between the Cα atoms of the structure using XMAS and considered a crosslink in agreement with the structure if the Cα–Cα distance was smaller than or equal to 30 Å.

Statistics and reproducibility

All experiments were conducted at least in triplicate using independent sample preparations. Sample size was chosen on the basis of previous experience and standards in the field. For cryo-ET data sets the maximum available samples were used for data acquisition. For the data shown in Fig. 1a,b, cells were tested on a regular basis throughout the study in three independent experiments. For the data shown in Fig. 1e,f, 16 cells were used for mito-DK and 20 cells for PHB2–DK analysis. For the data shown in Fig. 3a, four independent experiments were used. For the data shown in Fig. 3b two independent experiments each were used. For the data shown in Fig. 3c, we used three independent experiments of PHB1 and PHB2 KDs. The Jarque–Bera test was used to assess the normal distribution of the data followed by the two‐tailed unpaired t‐test (T) to determine P values in Excel 2016. These results are summarized in Supplementary Table 7. Data collection was not randomized and data collections and analysis were not performed blind to the conditions of the experiments.

Reporting summary

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