Expression and purification of Vipp1

Full-length Vipp1 (sll0617) of Synechocystis sp. PCC 6803 and associated mutants (Vipp1 (α1–α6) (truncation of helix α0 (aa 1–23)), Vipp1 dL10 (truncation of aa 157–166) and Vipp1 dL10Ala (substitution of aa 157–166 to alanine)) were heterologously expressed in E. coli C41 cells in terrific broth medium using a pET50(b) derived plasmid with a C-terminal His-tag and 3C protease cleavage site. For purification of Vipp1 and Vipp1 (α1–α6) under denaturing conditions, cells were resuspended after protein expression in lysis buffer containing 6 M urea (10 mM Tris-HCl pH 8.0 and 300 mM NaCl) supplemented with a protease inhibitor. Cells were lysed in a cell disruptor (TS Constant Cell disruption systems 1.1 kW; Constant Systems). The crude lysate was supplemented with 0.1% (v/v) Triton X-100 and incubated for 30 min at room temperature. Subsequently, the lysate was cleared by centrifugation for 15 min at 40,000g. The supernatant was applied to Ni-NTA agarose beads. The Ni-NTA matrix was washed with lysis buffer and lysis buffer with additional 10 mM imidazole. The protein was eluted from the Ni-NTA with elution buffer (10 mM Tris-HCl pH 8.0, 1,000 mM imidazole and 6 M urea). The fractions containing protein were pooled, concentrated (Amicon Ultra-15 centrifugal filter, 10-kDa molecular weight cutoff (MWCO)) and stored at −20 °C. For cleavage of the C-terminal His-tag, the protein was diluted 1:1 in cleavage buffer (10 mM Tris-HCl, 300 mM NaCl, 1 mM DTT and 0.1 mM EDTA) and dialyzed overnight against cleavage buffer together with 3C protease, including three buffer exchanges. After the cleavage reaction, the protein was diluted 1:3 with 10 mM Tris-HCl pH 8.0, 8 M urea, 200 mM NaCl and 7 mM imidazole and applied to the Ni-NTA matrix. The Ni-NTA matrix was washed with 10 mM Tris-HCl pH 8.0, 6 M urea, 100 mM NaCl and 4 mM imidazole. Flowthrough and wash fractions containing the cleaved protein were pooled, concentrated (Amicon Ultra-15 centrifugal filter, 10-kDa MWCO) and stored at −20 °C. Protein concentrations were determined by measuring the absorbance at 280 nm of Vipp1 diluted in 4 M guanidine hydrochloride using the respective molar extinction coefficient calculated by ProtParam35.

Liposome preparation and lipid reconstitution

Chloroform-dissolved EPL extract was purchased from Avanti Polar Lipids. Lipid films were produced by evaporating the solvent under a gentle stream of nitrogen and vacuum desiccation overnight. For Vipp1 reconstitution with lipids, urea-unfolded Vipp1 was added to the EPL film (Extended Data Table 1) and incubated with shaking for 20–30 min at 37 °C until the lipid film was resolved. Subsequently, the mixture was dialyzed overnight against 10 mM Tris-HCl pH 8.0 (8 °C, 20-kDa MWCO) including three buffer exchanges to refold the protein.

For analyzing membrane binding of Vipp1 helix α0 peptides, buffer was added to the EPL film to a final lipid concentration of 8 mM and incubated at 37 °C with shaking for 30 min. The liposomes were then prepared by sonication (Branson tip sonifier).

Characterization of the refolded protein

Vipp1 prepared as described above was refolded by dialysis against 10 mM Tris-HCl pH 8.0 overnight, with three buffer exchanges. Protein concentrations were determined on the basis of protein absorption at 280 nm as described above. For comparison, Vipp1 purified under native conditions (as described previously13) was studied. The protein concentration was adjusted to match the concentrations of the refolded protein on the basis of a comparison of the tryptophan fluorescence intensity at 340 nm determined in 6 M urea. The thermal stability was studied by monitoring the change in secondary structure through CD spectroscopy (J-1500, Jasco). The CD spectrum between 210 and 230 nm was measured at different temperatures (2 °C temperature steps, 1-nm wavelength steps, overall mean temperature ramp of 10 °C h−1). Samples of each protein were prepared in duplicate and measured in parallel in a six-cell holder. Interaction of the refolded protein with preformed liposomes was monitored by following changes in tryptophan fluorescence. To this end, an EPL film, prepared as described above, was rehydrated with Tris buffer. Unilamellar liposomes were either formed by sonication (Branson tip sonifier) to yield small liposomes or by extrusion through a 200-nm filter (Avanti Polar Lipids). The size of the liposomes was determined by dynamic light scattering (Zetasizer Pro, Malvern). According to the z average, the extruded liposomes had a diameter of 143 ± 0.5 nm and the sonified liposomes had a diameter of 56 ± 0.44 nm. A concentration of 3 µM protein was incubated with liposomes for 45 min at room temperature. Fluorescence emission was determined from 300 nm to 450 nm at 25 °C after excitation at 280 nm, with both excitation and emission slit widths set to 2.5 nm (FP-8500, Jasco).

Membrane binding of Vipp1 helix α0 peptides

The peptides Vipp1 helix α0 WT (MGLFDRLGRVVRANLNDLVSKAED), Vipp1 helix α0 R6A;R9A (MGLFDALGAVVRANLNDLVSKAED), Vipp1 helix α0 F4A;V11A (MGLADRLGRVARANLNDLVSKAED) and Vipp1 helix α0 amph(−) (MGAADRAGRAARANANDAASKAED) were purchased from PSL. The changes in secondary structure upon membrane binding were determined by CD spectroscopy (J-1500, Jasco). Then, 40 µM peptide (10 mM Tris pH 8.0) was incubated with increasing amounts of liposomes (up to 4 mM EPL). After incubation at room temperature for 15 min, spectra were recorded between 190 and 250 nm in 1-nm steps at 25 °C. Three individual samples were measured for each sample, with 12 accumulations per sample. The mean and s.d. of the value from the three samples at each lipid concentration were used to create a binding curve.

Negative-staining EM

For negative-staining EM, 3 µl of the sample was applied to glow-discharged (PELCO easiGlow glow discharger, Ted Pella) continuous carbon grids (CF-300 Cu, EM Sciences). The sample was incubated on the grid for 1 min. Then, the grid was side-blotted using filter paper, washed with 3 µl of water, stained with 3 µl of 2% uranyl acetate for 30 s and air-dried. The grids were imaged with a 120-kV Talos L120C EM instrument (Thermo Fisher Scientific, FEI) equipped with a CETA camera at a pixel size of 2.49 Å per pixel (×57,000 magnification) at a nominal defocus of 1.0–2.5 µm.

Cryo-EM

Grids were prepared by applying 4 μl of sample (with or without 5-nm gold fiducials) (Extended Data Table 1) to glow-discharged (PELCO easiGlow glow discharger, Ted Pella) Quantifoil grids (R1.2/1.3 Cu 200 mesh, EM Sciences). The grids were plunge-frozen in liquid ethane using a Leica EM GP2 plunge freezer set to 80% humidity at 10 °C (sensor-guided backside blotting, blotting time of 3–5 s). Videos were recorded in underfocus on a 200-kV Talos Arctica G2 (Thermo Fisher Scientific) EM instrument equipped with a Bioquantum K3 (Gatan) detector or a 300 kV Titan Krios G4 (Thermo Fisher Scientific) EM instrument equipped with a Biocontinuum K3 (Gatan) and a Falcon4i (Thermo Fisher Scientific) detector operated by EPU (Thermo Fisher Scientific) or Tomo (Thermo Fisher Scientific). Tilt series for cryo-electron tomography (cryo-ET) were collected at −60° to 60° with 3° in a dose-symmetric scheme. Tilt images were acquired at a magnification of ×53,000 (pixel size of 1.7 Å) with a nominal underfocus of 3.0 µm. The total dose for each tilt series was 152 e− per Å2. A total of 176 tilt series were collected (details of the single-particle and cryo-ET acquisitions in Tables 1 and 2 and Extended Data Table 1).

Single-particle image processing and helical reconstruction

Video frames were gain-corrected, dose-weighted and aligned using cryoSPARC Live36. Initial 2D classes were produced using the autopicker implemented in cryoSPARC Live. The subsquent image-processing steps were performed using cryoSPARC. The classes with most visible detail were used as templates for the filament tracer. For the Vipp1 EPL dataset, the resulting filament segments were extracted with a 600-pixel box size (Fourier-cropped to 200 pixels) and subjected to multiple rounds of 2D classification. The remaining segments were sorted by filament class: (1) stacked rings; (2) type I tubes; and (3) type II tubes. For each filament class, the remaining segments were re-extracted with a box size of 400 pixels (Fourier-cropped to 200 pixels) and subjected to an additional round of 2D classification. The resulting 2D class averages were used to determine filament diameters and initial symmetry guesses in PyHI37. Initial symmetry estimates were validated by helical refinement in cryoSPARC and selection of the helical symmetry parameters yielding reconstructions with typical Vipp1 features and the best resolution. Subsequently, all segments for each filament class were classified by heterogeneous refinement and followed by 3D classifications using the initial helical reconstructions as templates. The resulting helical reconstructions were subjected to multiple rounds of helical refinement including the symmetry search option. Reconstructions of the stacked rings were treated similarly; however, instead of helical symmetry, the respective rotational symmetry was applied in helical reconstruction jobs (C11–C14). Segments with poor visible details were discarded by heterogeneous refinement. For the final polishing, the segments were re-extracted at 600 pixels with Fourier cropping to 400 pixels (stacked rings and type I tubes) or at 500 pixels without Fourier cropping (type II tubes). The Vipp1 (α1–α6) and Vipp1 (α1–α6) EPL datasets were preprocessed as described above. The filament segments were extracted with a 700-pixel box size (Fourier-cropped to 200 pixels) and subjected to multiple rounds of 2D classification. The subsequent image-processing steps were identical to the above-described workflow of helical reconstruction of type I and type II filaments. Bad segments were discarded by heterogeneous refinement. For the final polishing, the segments were re-extracted at 700 pixels with Fourier cropping to 350 pixels.

The Vipp1 dL10Ala EPL dataset was preprocessed using WARP38 (motion correction, binning to physical pixel size and contrast transfer function (CTF) estimation) and further processed in cryoSPARC as described above. The filament segments were extracted with a 500-pixel box size (Fourier-cropped to 200 pixels) and subjected to multiple rounds of 2D classification. The subsequent image-processing steps were identical to the above-described workflow of helical reconstruction of type I and type II filaments. Poor segments were discarded by heterogeneous refinement or 3D classification. For the final polishing, the segments were re-extracted at 500 pixels without Fourier cropping. The ring structures were picked using a template picker and extracted at 600 pixels with Fourier cropping to 200 pixels. The extracted particle stack was cleaned after multiple rounds of 2D classification. Rings with different rotational symmetries were sorted by multiclass ab initio reconstruction and further classified by multiple rounds of heterogenous refinement and ab initio reconstruction with imposed rotational symmetry (C15–C20). Poor particles were discarded by heterogeneous refinement. For the final polishing, the particles were re-extracted at 600 pixels with Fourier cropping to 450 pixels and subjected to nonuniform refinement with imposed symmetry (C15–C20). The local resolution distribution and local filtering for all maps were performed using cryoSPARC (Extended Data Figs. 2a, 3a, 4e, 6d and 7a). The resolution of the final reconstructions was determined by Fourier shell correlation (FSC; automasked, 0.143).

Cryo-ET image processing

Tilt image frames were gain-corrected, dose-weighted and aligned using WARP38. The resulting tilt series were aligned, 8× binned and reconstructed by the weighted backprojection method using AreTomo39. For segmentation, the reconstructed tomograms were filtered with a recursive exponential filter followed by a nonlocal means filter using Amira (Thermo Fisher Scientific). The filtered tomograms were segmented in Dragonfly (Object Research Systems) by progressively training a U-Net with an increasing number of manually segmented tomogram frames (5–15). Then, the trained U-Net was used to predict those features of the tomogram. For visualization, the segmentation was cleaned up by removing isolated voxels of each label group (islands of <150 unconnected voxels). The resulting segmentation was imported to ChimeraX40 for 3D rendering. For quantitative analysis of membrane curvature, eight tomograms were used in the same manner to train the U-Net, which was then used to predict 123 tomograms included in the analysis. Once the resulting segmentations were coarsely corrected for errors, they were exported as .tif files and assembled into .mrc files for further analysis. For the membrane segmentations, the surface morphometrics toolkit41 was used to determine curvature and membrane normals. The resulting coordinates and their corresponding curvature estimates were used to filter for adjacent carpet segmentations, accepting a distance up to 50 Å. To minimize errors, coordinates close to tomogram edges were excluded. For further analysis, coordinates were sorted into fixed-size bins according to their curvedness and, for each bin, the ratio of membrane coordinates with a present proximal carpet segmentation was determined resulting in the corresponding occupancy value between 0 and 1 (0, no coverage; 1, full coverage).

For subtomogram averaging, RELION 4 (ref. 42) was used. CTF estimation was conducted with ctffind4 (ref. 43). As initial particles, the coordinates resulting from the previously described membrane analysis were used. Coordinates were selected for the presence of Vipp1 carpet structures (within a radius of 100 Å) and a minimal interparticle distance of 15 Å was enforced to avoid excessive overlaps. The orientation of particles was initially estimated using their corresponding membrane normal to restrain angular searches for all subsequent refinement procedures. To create an initial model from a subset of homogenous particle picks, all coordinates were clustered according to their corresponding membrane curvature, using the Scikit-Learn implementation of k-means. After particle averaging and several rounds of 3D classifications and refinements in 4× binning, a low-resolution model emerged. This model was then used as an initial reference and, from here on, all particles were included. Multiple rounds of 3D classifications and refinements for cleaning up the dataset followed, initially using 4× binning and later on 2× binning. As a further means to exclude poor particles, particles with diverging orientations compared to the majority of particles in close proximity were excluded. Finally, the resulting maps were used to improve predetermined parameters, using RELION’s frame alignment and CTF refinement procedures followed by a final round of refinements. For the highest-resolution map, the average curvedness of all points from which the final average was calculated (~0.054) indicated a substantially higher diameter than the currently existing highest diameter Vipp1 ring model (C18, Protein Data Bank (PDB) 7O3Z). Therefore, PDB 7O3Z was used to initially set the distance to the symmetry axis and rotation. To find the correct symmetry parameters (C symmetry, distance and rotation in relation to the symmetry axis), the map was fitted in ChimeraX40 along the x axis. The correlation of the original map with versions of itself with different symmetry parameters applied was then calculated and the correlation was maximized. First, translational searches along the x axis in 1-Å increments were conducted for C18–C22 symmetries. The best correlation for the 0.054 curvedness reconstruction was identified for a C20 symmetry. When optimal distance and symmetry parameters were found, rotational searches were conducted (Supplementary Fig. 6e). Finally, the symmetry, shift and rotational parameters were applied to the reconstruction and PDB 7O3Z was rigid-body fitted into the resulting map.

Cryo-EM map interpretation and model building

The handedness of the maps was determined by rigid-body fitting Vipp1 reference structures (PDB 7O3W) into the final maps using ChimeraX40,44 and flipped accordingly. For models of the Vipp1 type I and type II and Vipp1 dL10Ala type IIb tubes, one of the monomers of PDB 7O3W underwent molecular dynamics flexible fitting (MDFF) to the 3D reconstructions using ISOLDE45. The models were subjected to autorefinement with ‘phenix.real_space_refine’ (ref. 46). The autorefined models were checked and adjusted manually in Coot47 and ISOLDE45 before a final cycle of autorefinement with ‘phenix.real_space_refine’ (ref. 46). After the final inspection, the models were validated using ‘phenix.validation_cryoem’ (ref. 48) and MolProbity49. Then, the respective helical symmetry was applied to all models to create assemblies of 60 monomers each to generate ‘biomt’ matrices for deposition. For models of the stacked C11 and C12 rings PDB 6ZVR and PDB 6ZVS were used as ref. 3. Side chains were adjusted to the sequence of Synechocystis sp. PCC 6803 Vipp1 and the resulting models were rigid-body fitted to the maps. For models of the stacked C13 and C14 rings, PDB 7O3W was used as a ref. 1. The nucleotide was deleted and the monomers were first rigid-body fitted and then underwent MDFF to the 3D reconstructions using ISOLDE45. For models of the Vipp1 (α1–α6) rods, PDB 7O3W was used as a ref. 1. Only the central monomer was selected and aa 1–22 were removed. The truncated model was rigid-body fitted to the maps. For each map, the monomers then underwent MDFF using ISOLDE45. The models were subjected to autorefinement with ‘phenix.real_space_refine’ (ref. 46). The autorefined models were checked and adjusted manually in Coot47 and ISOLDE45 and subjected to a final cycle of autorefinement with ‘phenix.real_space_refine’ (ref. 46). After the final inspection, the models were validated using ‘phenix.validation_cryoem’ (ref. 48) and MolProbity49. Then, the respective helical symmetry was applied to all models to create assemblies of 60 monomers each to generate ‘biomt’ matrices for deposition. The Vipp1 dL10Ala C15 to C20 ring complex models were generated using Vipp1 WT C15 and C16 ring models as references (PDB 7O3X and 7O3Y)1. The nucleotide was removed and the monomers were first rigid-body fitted and then underwent MDFF to the 3D reconstructions using ISOLDE45. Image processing, helical reconstruction and model building were completed using SBGrid-supported applications50. In this manner, from a total of 37 cryo-EM maps determined in four samples (3× Vipp1 + EPL rods, 4× Vipp1 + EPL stacked rings, 10× Vipp1 (α1–α6) rods, 10× Vipp1 (α1–α6) + EPL rods, 4× Vipp1 dL10Ala EPL tubes and 6× Vipp1 dL10Ala rings), a total of 17 unique PDB coordinates were refined and deposited originating from four samples (Table 1). The four Vipp1 + EPL stacked-ring maps and the six Vipp1 dL10Ala ring complex maps were not of sufficient quality to refine models. The remaining complementary Vipp1 (α1–α6) + EPL maps were of identical symmetry consisting of either similar or poorer resolution densities; therefore, atomic model refinement was not pursued further. Diameters of the Vipp1 rod and tube reconstructions were measured as radial density profiles using ImageJ51. Outer and inner leaflet radii of engulfed membrane tubes were determined from the same radial density profiles at the peak maxima from each bilayer leaflet.

Quantification and statistical analysis

Data and statistical analysis were performed using OriginPro 2022b (OriginLab). Detailed descriptions of quantifications and statistical analyses (exact values of n, dispersion and precision measures used and statistical tests used) can be found in the respective figures and figure legends and the Methods.

Reporting summary

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