Structure of the flotillin complex in a native membrane environment

Flotillin Cages on Cell-Derived Membranes.

Fig. 1A outlines three separate methods we used to prepare membrane vesicles in this study. In one method, we produced giant PM vesicles by applying N-ethylmaleimide (NEM) and calcium (15) and sonicated them to reduce their size for cryo-EM analysis (16). In another, we isolated membrane fragments without NEM and calcium application as described in Materials and Methods. In a third, we isolated spontaneous exosomes and sonicated them. All three methods yielded vesicles with Flotillin cages attached to membranes. We note that the first and third methods presumably produce vesicles originating from the PM, whereas the second method produces vesicles from both plasma and organellar membranes. As depicted in Fig. 1A and shown in Fig. 1B and SI Appendix, Fig. S1 B–D, Flotillin cages sometimes attach to the inside of a vesicle and sometimes to the outside. We learned while studying other proteins in these kinds of vesicles that PM vesicles probably start out with an outside-out orientation, but the orientation in some vesicles can reverse with sonication.

The Flotillin cages occur naturally in HEK293 GnTl⁻ cells as no exogenous genes were expressed in this study. We observed Flotillin cages in other cells as well, for example, A431 cells and human erythrocytes. Fig. 1C shows images of erythrocyte vesicles after removal of hemoglobin and cytoskeletal elements followed by sonication. Flotillin cages inside the vesicles look like those from the HEK293 GnTl⁻ cells. Whole erythrocytes were also treated with a hypotonic buffer to induce swelling and then unroofed on a cryo-EM grid using fluid flow–mediated shear force. Cryo-EM tomographic images of the erythrocyte membrane patches remaining on the grid showed membrane-attached Flotillin cages (Fig. 1D).

It is likely that vesicles with exterior Flotillin cages, e.g., Fig. 1 A and B, are flipped inside-out vesicles and that the cages are naturally present on the cytoplasmic face of PMs and possibly organellar membranes as well.

Structure of a Flotillin Complex.

We processed images from vesicles prepared using the NEM, Ca²⁺ method of GPMV formation followed by sonication (Fig. 1A and Materials and Methods). From a dataset consisting of 19,024 micrographs, we manually selected Flotillin cages in numerous orientations. Top views show a circular structure with a diameter of about 30 nm. Side views resemble a truncated cone (Fig. 2 A–C and SI Appendix, Fig. S1 B–D). The wide end of the truncated cone is always attached to a membrane surface and typically alters the curvature of the membrane at the attachment region relative to that of the surrounding membrane. We will show later that the curvature of the membrane also alters the shape of the Flotillin cage. Thus, Flotillin and the membrane mechanically interact with each other. 40% of Flotillin cages reside inside vesicles (outside-out vesicles) and 60% reside outside vesicles (inside-out vesicles). Many Flotillin cages on the outer surface of inside-out vesicles contain a second, smaller vesicle attached to the opposite end of the truncated cone (Fig. 2 A and B).

Manually selected particles were used to train multiple Topaz models (17). To minimize false positive selections, multiple rounds of 2D classification were carried out. If selected particles were separated by less than 16 nm (half the cage diameter) they were treated as duplicates and removed. An ab initio reference map was generated with the remaining 3,730 particles, without symmetry constraints applied, and refined to ~16 Å (SI Appendix, Fig. S2).

Two-dimensional classification pointed to rotational symmetry (Fig. 2C), and in one class there appeared to be 44 subunits (Fig. 2D). A rotational symmetry search from C1 to C51 yielded the highest resolution with C44 symmetry (SI Appendix, Fig. S2). 3D classification without alignment followed by further refinement and local motion correction improved the resolution to 3.5 Å. Still, some regions of the map, particularly at the ends of the truncated cone, were not well resolved (SI Appendix, Fig. S3A).

Two proteins, Flotillin1 and Flotillin2, which are 69% identical, were identified in western blot analysis of vesicles isolated from HEK293 GnTl⁻ cells (SI Appendix, Table S1) along with several other members of the Stomatin, Prohibitin, Flotillin, HflK/C (SPFH) family (3). The structure of one member of this family, HflK/C from bacteria, forms a cage that attaches to membranes (14). We hypothesized that the Flotillins might form the larger cages in our samples. Past studies showed that Flotillin1 and Flotillin2 colocalize in puncta (6), that knockdown of either one decreases expression of the other (18), and that the two can be cross-linked (4). We therefore considered the possibility that both Flotillins form a pseudosymmetric structure. We applied C22 symmetry with a mask over the poorly defined narrow end of the truncated cone. This procedure distinguished between two layers of helical density (SI Appendix, Fig. S3C). One layer contains 44 “cap” helices and the other 22 “CTD” helices, explicable if one Flotillin encodes a CTD helix and the other does not. We conclude that the cage consists predominantly of 22 each of Flotillin1 and Flotillin2 in an alternating pattern.

C22 symmetry applied to the entire structure improved density in many regions. For example, residues R291 of Flotillin 1 and A294 of Flotillin2 are poorly resolved with C44 symmetry but improved with C22 symmetry (SI Appendix, Fig. S3D). Similar improvements in side-chain density were observed for corresponding residues in Flotillin1/Flotillin2, including K360/A363, D355/A358, A339/K342, and F325/A328 (SI Appendix, Fig. S3E). With a final density map for the overall Flotillin complex at 3.5 Å resolution, atomic models could be built for much of the molecule. Regions with poor or weak density were removed. The final model consists of Flotillin-1:1-422 and Flotillin-2:1-402. (SI Appendix, Fig. S4A), except for the SPFH1 domains (SI Appendix, Fig. S4B). The density extends from the 44-helical layer to the 22-helical layer in every other subunit, permitting us to identify the origin of the 22-helical layer of Flotillin1 (SI Appendix, Fig. S4C). The segment following the cap-helix domain in both Flotillin1 and Flotillin2 shows some differences. For example, the length of the linker connecting the cap helix to the CTD helix aligns with the sequence of Flotillin1. The model is consistent with constraints derived through lysine–lysine cross-link data (4). Thus, it is likely that 22 pairs of Flotillin 1 and Flotillin 2 form the cage, however, we cannot exclude the possibility that other stoichiometries can occur.

The Flotillin protein complex resembles a basket, with its wide end always connected to a membrane and its narrow end sometimes connected to a separate membrane (Fig. 3A). The wide end has a diameter of 32 nm, the narrow end 19 nm, and height 22 nm (Fig. 3B). The wide end contains SPFH1 and SPFH2 domains at the N terminus (Fig. 3C). The wall is a single-layered α-helical barrel, and the narrow end consists of the two α-helical layers described above, one with 22 and the other with 44 α-helices (Fig. 3D). A β-strand from each Flotillin makes a 44-stranded β-barrel inside the helical layer. C-terminal amino acids form a poorly resolved structure that further constricts the opening of the narrow end to about 4 nm diameter.

Structural Basis of Flotillin Oligomerization.

The C terminus of Flotillin was suggested to be involved in oligomerization (19). The narrow end is made by the C terminus of Flotillin arranged in two layers of helices that run in opposite directions and are linked by a β-strand (Fig. 4 A and B). From the structure, we observed strong interactions between Flotillin1 and Flotillin2 near the narrow end. The first layer has 44 helices from both Flotillin1 and 2 (22 each). The interaction between Flotillin1 and Flotillin2 is enhanced by ionized hydrogen bonds (Fig. 4 C and D). The center of the narrow end is a parallel β-barrel containing 44 β-strands from both Flotillin1 and 2. After the β-barrel, Flotillin1 and Flotillin2 diverge. Flotillin1 extends outward, while Flotillin2 goes further in toward the center. Flotillin1 adds a new layer of 22 helices on top of the first layer, placing hydrophobic residues between the two helical layers (Fig. 4E).

The wall of the Flotillin cage is made of a right-handed α-helical barrel with 44 helices arranged side by side (Fig. 4F). This formation spans 186 amino acids, constituting the core of the Flotillin structure. These helices are characterized by repeating patterns of amino acids (A/G, E, A/G, E), which are believed to form coiled-coil bundles that facilitate the clustering of Flotillin molecules (1). Flotillin does not contain heptad-repeat sequences found in classical coiled-coil helices. Instead, Flotillin is characterized by the presence of charged amino acids both inside and outside its α-helical barrel. These charged residues form specific clusters of ionized hydrogen bonds, likely serving to glue adjacent subunits together (Fig. 4 F and G and SI Appendix, Fig. S5).

The structure of Flotillin’s right-handed helical barrel differs from those found in other proteins like SNARE (20) and motor proteins such as myosin, kinesin, and dynein (21). These proteins typically have helices that intertwine in pairs or groups, unlike Flotillin’s parallel arrangement. This difference notwithstanding, the buried area per residue in Flotillin’s wall structure is like that in these other helical structures (detailed in Materials and Methods).

Interactions between Flotillin and Membranes.

In our micrographs, the wide end of the Flotillin cage is mostly bound to a membrane surface. Adherence to the membrane is mediated by the SPFH1 domains, which penetrate one leaflet of the lipid bilayer (Fig. 5A). We show in Fig. 5B that the SPFH1 domains expose to the membrane a surface containing many hydrophobic amino acids (Fig. 5B). We note the prevalence of Trp residues, which are well known to be most stable at the interface between the membrane’s headgroup layer and hydrophobic core (Fig. 5B, solid line). We also note the many cysteine residues on the hydrophobic surface. Some of these are palmitoylated (19), and Flotillin-2 is N-myristoylated on Gly2. This myristoylation is required for membrane association (19). The bacterial HflK/HflC complex attaches to membranes through transmembrane helices (14). The Flotillin cage, by contrast, uses a hydrophobic surface that contains covalent lipid molecules.

When Flotillin cages were located outside a vesicle, they often connected to a second membrane at their narrow end (Fig. 2B). The unstructured C terminus regions of Flotillin1 and Flotillin2 are likely involved in this membrane interaction, although the low resolution of this region precludes a detailed description (Fig. 5C).

Flotillin cages in outside-out and inside-out orientations showed conformational differences at the membrane interface (Fig. 5D), where inward bending of the SPFH domains may induce membrane invagination and initiate a clathrin-independent endocytic pathway (6). This bending may be initiated by the phosphorylation of tyrosine residues Tyr160 in Flotillin1 and Tyr163 in Flotillin2 by kinase Fyn. Located at the hinge region, phosphorylation of Tyr160 and Tyr163 is essential for altering membrane curvature and initiating endocytosis, as demonstrated by the reduced endocytosis observed when these residues are mutated to phenylalanine (22). Further research is needed to explore how these phosphorylation-induced changes impact Flotillin’s functionality in endocytosis and membrane dynamics.

Structural Insights for SPFH Protein Family.

The SPFH protein family members likely share a similar architecture. Using negative stain cryo-EM Prohibitin, Slr1128/SYNY3, and Erlin-1/2 were all shown to form ring-shaped structures (9, 23–25). More recently, the aforementioned HflKC and FtsH complex was determined at high resolution (14, 26).

Using Alphafold2 (27), we predicted SPFH proteins’ hetero/homo-tetramer structures based on their sequences from SPFH2 to the C-terminal cap helix which they share in common (Fig. 6A). Different SPFH family proteins have various lengths of wall helices ranging from 189 to 44 amino acids. Following the long helix, they share a similar cap helix forming the first helical layer, resembling the narrow end of the Flotillin cage. The differing lengths of the wall helix and cap helix likely influence the overall size and the number of subunits in their protein complexes (Fig. 6B). Additionally, we observed that two other truncated-cone structures in the NMVs of HEK293 GnTl⁻ cells, presumably stomatin and stomatin-like protein 2, exhibit distinct sizes and shapes compared to the Flotillin complex (Fig. 6C). This suggests that SPFH family proteins can form a variety of cage structures.

Cell Culture.

HEK293S GnTl⁻ cells (ATCC CRL-3022) were cultured in Freestyle 293 medium supplemented with 2% fetal bovine serum at 37 °C. A431 cells (ATCC CRL-1555) were cultured in Dulbecco’s Modified Eagle’s Medium, Catalog No. 30-2002 supplemented with 10% fetal bovine serum. Human erythrocytes were purchased from Innovative Research (IWB3ALS40ML).

PM Vesicle Preparation.

Four liters of live HEK293 GnTl⁻ cells (3 × 10⁶/mL) were harvested by centrifugation at 500×g for 10 min. The cell pellet was resuspended in 500 mL GPMV buffer containing 10 mM K-HEPES, pH 7.4, 140 mM NaCl, 10 mM KCl, and 2 mM CaCl2. The cells were centrifuged again at 500×g for 10 min and then resuspended in 500 mL GPMV buffer supplemented with NEM (Thermo Scientific™ Pierce™, 23030) at 7.5 mM. The cell resuspension was transferred to four 500 mL baffled flasks to contain 125 mL in each flask. The flasks were then incubated in a 37 °C incubator shaking at 130 rpm for 2 h. The fast shaking helps to dislodge the NMVs from the cells, resulting in smaller unilamellar vesicles. At the end of the incubation, the flasks were shaken by hand for ~30 s. The suspension was then spun down at 500×g at 4 °C for 10 min to remove cells and large vesicles. Next, 10% glycerol was added to the supernatant to prevent aggregation, and it was sonicated using a Branson probe sonicator (1/2” tip with Branson102-C converter) at 40% power in three pulses of 30 s each, with approximately 1 min of cooling on ice between each pulse. The vesicles were then pelleted by ultracentrifugation at 100,000×g at 4 °C for 1 h in a Ti70 rotor.

The membrane vesicle pellet from every ~25 mL sample was triturated in ~0.2 mL GPMV buffer supplemented with 10% glycerol by gently squirting the buffer toward the pellet (with a 100 μL tip). The resuspended vesicles in the ultracentrifugation tubes were then sonicated in a water bath sonicator (Branson M1800) with ~10 s pulses at room temperature until the solution was opalescent. The vesicles were subsequently spun down at 3,500×g at 4 °C for 10 min to remove any remaining aggregates.

The supernatant was mixed with two volumes of 90% sucrose GPMV buffer and loaded at the bottom of a step gradient that contained 2 mL of 45%(w/v) and 35% sucrose, 4 mL of 25%, and 12% sucrose GPMV solutions in a 14-mL ultraclear tube. Next the tubes were centrifuged at 150,000×g for 18 h at 4 °C in a SW40Ti rotor. After centrifugation, 25 to 45% fraction was collected and then concentrated to OD280 ~2 using an Amicon 2 mL concentrator (molecular weight cutoff of 100 kDa).

For western blot analysis, vesicles were loaded onto sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) using Bio-Rad Mini-PROTEAN® TGX™ Precast Gels (4 to 15%) and subjected to electrophoresis. Following electrophoresis, the proteins were transferred to a nitrocellulose membrane. The blots were then blocked for 30 min at room temperature in TBS-T (25 mM Tris-HCl pH 7.4, 0.9% NaCl, and 0.02% Tween-20) containing 4% dry milk. Subsequently, the membranes were probed with anti-Flotillin-1 antibody (Protein tech, 15571-1-AP), anti-Flotillin-2 antibodies (Protein tech, 28208-1-AP), Stomatin (Santa Cruz, sc-376869), Stomatin-like protein 2 (Abcam, ab191883), and detected using fluorescence.

Exosomes Preparation.

Four liters of conditioned medium was harvested from live HEK293 cultured cells (3 × 10⁶/mL). Cells were removed by centrifugation at 3,000×g for 20 min followed by 10,000×g for 30 min. The supernatant was centrifuged at ~100,000×g for 1 h in a Ti70 rotor.

The membrane vesicle pellet from every ~25 mL sample was triturated in ~0.2 mL GPMV buffer supplemented with 10% glycerol by gently squirting the buffer toward the pellet (with a 100 μL tip). The resuspended vesicles in the ultracentrifugation tubes were then sonicated in a water bath sonicator (Branson M1800) with ~10 s pulses at room temperature until the solution was opalescent. The vesicles were then centrifuged at 3,500×g at 4 °C for 10 min to remove the remaining aggregates.

The supernatant was mixed with two volumes 90% sucrose GPMV buffer and loaded at the bottom of a step gradient that contained 2 mL of 45%(w/v) and 35% sucrose and 4 mL of 25% and 12% sucrose GPMV solutions in a 14-mL ultraclear tube. Then, the tubes were centrifuged at 150,000×g for 18 h at 4 °C in a SW40Ti rotor. After centrifugation, 25 to 45% fraction was collected and then concentrated to an absorbance at 280 nm of ~2.0 using an Amicon 2 mL concentrator (molecular weight cutoff of 100 kDa).

Preparation of Total Membrane Vesicles.

All procedures were conducted at 4 °C within a cold room. Pellet from 4 L of cells was resuspended in 160 mL of lysis buffer [20 mM K-HEPES pH 7.4, 300 mM KCl, 0.5 mM MgCl2, and 5 mM dithiothreitol (DTT)] supplemented with 10 µg/mL leupeptin, 10 µg/mL pepstatin A, 1 mM benzamidine, 2 μg/mL aprotinin, 0.3 mg/mL AEBSF, ~100 µg/mL DNase, and ~100 µg/mL RNase A. The cell suspension was homogenized using a Dounce homogenizer for approximately 30 to 60 strokes. The resulting homogenate was transferred to a metal beaker and sonicated in an ice bath with a probe sonicator (1/2” tip using the Branson102-C converter) at 60% power. The homogenate was disrupted by applying 30-s pulses four times, with a 30-s interval between each pulse. Then the supernatant was centrifuged at ~100,000×g for 1 h in a Ti70 rotor. The membrane vesicle pellet from every ~25 mL sample was triturated in ~0.2 mL GPMV buffer supplemented with 10% glycerol by gently squirting the buffer toward the pellet (with a 100 μL tip). The resuspended vesicles in the ultracentrifugation tubes were then sonicated in a water bath sonicator (Branson M1800) with ~10 s pulses at room temperature until the solution was opalescent. The vesicles were then centrifuged at 3,500×g at 4 °C for 10 min to remove the remaining aggregates. The supernatant was mixed with two volumes 90% sucrose GPMV buffer and loaded at the bottom of a step gradient that contained 2 mL of 45%(w/v) and 35% sucrose, 4 mL of 25% and 12% sucrose GPMV solutions in a 14-mL ultraclear tube. Then, the tubes were centrifuged at 150,000×g for 18 h at 4 °C in a SW40Ti rotor. After centrifugation, 25 to 45% fraction was collected and then concentrated to OD280 ~2 using an Amicon 2 mL concentrator (molecular weight cutoff of 100 kDa).

Erythrocytes Vesicle Preparation

Washed human erythrocytes (Innovative Research, IWB3ALS40ML) were hemolyzed in 5 volumes of 1 mM EDTA and 10 mM Tris-HCl, pH 7.4. Then they were centrifuged at 18,000×g for 10 min. The resulting “ghost” membranes were then washed five times in the hemolysis buffer with 1 mM EDTA and 10 mM Tris-HCl, pH 7.4. Ghost membranes were pelleted (18,000×g, 10 min) and then sonicated using a probe sonicator (Branson, 1/2” tip with Branson102-C converter) at 60% power for three 30 s pulses with ~1 min chilling on ice in between pulses. The vesicles were then centrifuged at 18,000×g at 4 °C for 10 min twice to remove the remaining aggregates. The vesicles were then pelleted by ultracentrifugation at 100,000×g at 4 °C for 2 h in a Ti70 rotor.

The membrane vesicle pellet from every ~25 mL sample was triturated in ~0.2 mL of GPMV buffer supplemented with 10% glycerol by gently squirting the buffer toward the pellet (with a 100 μL tip). The resuspended vesicles in the ultracentrifugation tubes were then sonicated in a water bath sonicator (Branson M1800) with ~10 s pulses at room temperature until the solution was opalescent. The vesicles were then centrifuged at 3,500×g at 4 °C for 10 min to remove the remaining aggregates.

The supernatant was mixed with two volumes 90% sucrose GPMV buffer and loaded at the bottom of a step gradient that contained 2 mL of 45%(w/v) and 35% sucrose, 4 mL of 25% and 12% sucrose GPMV solutions in a 14-mL ultraclear tube. The tubes were then centrifuged at 150,000×g for 18 h at 4 °C in a SW40Ti rotor. After centrifugation, the 25 to 45% fraction was collected and concentrated to an absorbance at 280 nm ~2.0 using an Amicon 2 mL concentrator (molecular weight cutoff of 100 kDa).

Cryo-EM Grid Preparation for Unroofed Erythrocytes

Washed human erythrocytes (Innovative Research, IWB3ALS40ML) were diluted at a ratio of 1:100 and deposited onto carbon-coated 200-mesh copper grids. These grids had previously undergone glow discharge and were precoated with a poly-D-lysine solution. Following a 30-min incubation, the human erythrocytes were rinsed in a low salt buffer [a 1:3 dilution of Dulbecco’s phosphate-buffered saline (DPBS)]. The unroofing process involved the application of a stream of DPBS through a 20-gauge needle. Subsequently, the unroofed erythrocytes were washed three times at room temperature with DPBS.

3 μL DPBS was then applied to the grids containing the unroofed erythrocytes. The grids were manually blotted from the edge using a piece of filter paper and were plunge-frozen using a Vitrobot Mark IV. The blotting was performed with a blot force of 0 and a blot time of 3 s.

Model Building, Refinement, and Analysis

To build the Flotillin model in the high-resolution density map, the structure of the dimer of Flotillin-1 and Flotillin-2 predicted by AlphaFold was fitted into the map using molecular dynamic flexible fitting in ISOLDE (62). We modeled four Flotillin dimers in the map to capture the entire interaction interface because a subunit interacts with multiple neighboring protomers in the region close to the narrow end. The four Flotillin dimer positions in the map were built manually in Coot (63) and the other Flotillin subunits were adjusted in Coot by applying noncrystallographic symmetry, using the four Flotillin dimers as master chains. The structure underwent iterative refinement using Coot and Phenix real-space refine (64) with noncrystallographic symmetry restraints while omitting geometry restraints. Regions with poor or weak density were removed. The final model consists of Flotillin-1:1-422 and Flotillin-2:1-402. Refinement quality summary is in SI Appendix, Table S2. Cryo-EM density maps and protein structures were visualized using UCSF ChimeraX (65).