Introduction

Secretion systems, which are bacterial cell envelope-located protein complexes, are utilized by bacteria to produce virulence-related substrates that facilitate survival and pathogenicity1. Among secretion systems, the type II secretion system (T2SS) is broadly present and functional in Proteobacteria species, including non-pathogenic Escherichia coli (E. coli), Enterotoxigenic E. coli (ETEC), Enteropathogenic E. coli (EPEC), Vibrio cholerae, Klebsiella pneumoniae, and Aeromonas hydrophila2. T2SS substrates in non-pathogenic bacteria can facilitate nutrient absorption from the environment or symbiosis with plants or animals, whereas in pathogenic bacteria, T2SS substrates can aid in adhesion to hosts, intoxicate host cells, and suppress immunity in the host, causing various diseases3. With T2SS’s diverse substrate functions and close relevance to virulence and diseases, knowing its structure and working mechanism is necessary for understanding bacteria functions and developing antimicrobial strategies.

The outer membrane component of T2SS is the secretin, constituting a large channel structure, connecting to the protein scaffold on the inner membrane, and controlling the last step of substrate transportation2. Phylogenetic analysis of T2SS secretins from Proteobacteria species demonstrated two types: the Klebsiella-type secretins found in Klebsiella and Dickeya; and Vibrio-type secretins found in Vibrio, ETEC, and EPEC4. Structures of the two secretin types both appear as cylindrical channels containing the N0-N3 domains, the secretin domain with a central gate region, and the S domain5,6,7,8,9,10. But they are different in the transmembrane region: the Vibrio-type secretins have about 20 additional amino acids between the α7 and α8 helices, forming a loop extending upward and inward to the central axis of the channel like a roof, constituting a cap gate, while Klebsiella-type secretins do not have the cap gate5,6,7,8,9,10. This directly causes differences in the height of the proposed transmembrane regions of the two secretin types, which may give them different ways of interacting with the membrane. The in situ structure of T2SS has been visualized in Legionella pneumophila, which has indicated the architecture of T2SS and the membrane interaction of the secretin11. However, the resolution is limited to a few nanometers, hindering the observation of more details from the structure. Studying high-resolution structures of secretins in situ could reveal possible biological processes that occur in the cellular environment and offer a more complete understanding of their assembly process and mechanisms. This environment cannot be accurately simulated in vitro and may have a significant impact on secretin behavior.

The translocation of T2SS secretins from the inner membrane to the outer membrane is an essential step required for T2SS assembly, however, it is still unclear how this translocation is regulated in the bacterial cell envelope. Based on biochemistry experiments, there exist two possible pathways: (1) the scaffolding proteins GspA and GspB bind and increase the pore size of peptidoglycan, and translocate the secretin to the outer membrane, as found in Klebsiella-type secretins ExeD and OutD12,13,14; (2) a small pilotin GspS could bind to the S domain of secretin, and the pilotin itself can translocate to the outer membrane through the Lol sorting pathway, as found in the Vibrio-type secretins4,15,16. Besides these two pathways, since some T2SS exist without GspA and GspB or the pilotin11,17,18, other pathways also possibly exist. Despite the current biochemical evidence, this translocation process has not been visualized in living cells. By doing in situ electron cryotomography (cryo-ET) studies on T2SS secretins, we capture different intermediate states during the outer membrane translocation process, which helps to explain the translocation mechanism and provide a clearer blueprint of the biogenesis process of T2SS secretins.

There are two T2SS secretins encoded in two T2SS operons in the E. coli genome: the Klebsiella-type GspDα encoded in the T2SSα operon that can secrete chitinase19,20,21,22, representing non-pathogenic functions; and the Vibrio-type GspDβ encoded in the T2SSβ operon4,15, which can secrete toxins23,24, representing pathogenic functions. In this study, we investigate the in situ structures of GspDα from the E. coli K12 strain and GspDβ from the ETEC H10407 strain as representatives of the Klebsiella-type secretins and Vibrio-type secretins, respectively. We report four in situ structural states of specific secretin/secretin-pilotin complexes, determined within E. coli cells, using the cryo-ET subtomogram averaging. They are, respectively, GspDα on the inner membrane, GspDα on the outer membrane, GspDβ–GspS complex on the outer membrane, and GspDβ on the inner membrane (Supplementary Table 1). Together, these structures show interactions of secretins with inner and outer membranes and provide insights into the biogenesis process of the GspD secretin.

Results

Visualization of GspDα on the inner membrane through cryo-ET

To obtain thin cells for improved contrast under cryo-ET, we employed a bacteria minicell system to provide thin cells which retain physiological activities25. We induced overexpression of GspDα within E. coli BL21 (DE3) cells (Fig. 1j). Cryo-ET was performed to image the minicells and 250 tilt series were collected. From the raw tilt images and the reconstructed tomograms, cell features including the intact bacterial cell envelope and membrane-bound GspDα particles could be clearly recognized (Fig. 1a–d). GspDα multimer particles in different orientations with respect to the grid plane could be identified in the tomograms, and surprisingly, under these experimental conditions, the GspDα particles are naturally located on the inner membrane of the bacterial envelope, with the non-transmembrane domains facing the periplasmic space (Fig. 1d). As GspDα has an approximately cylindrical shape, we identified particles in the top view and side view as in circles and a pair of curved lines attaching to the inner membrane, respectively. The images of negative control cells further validated that these particle features found in the tomograms belong to our protein of interest (Supplementary Fig. 1).

Fig. 1: Visualization of GspDα multimers within E. coli cells and determination of the overexpression and membrane location of GspDα.
figure 1

a The tomogram z slice view of E. coli overexpressing GspDα. b Segmentation of the tomogram shown in (a), with OM colored purple, PG colored yellow, IM colored dark red, and GspDα on IM colored blue. c Zoom in tomogram z slice view of top view GspDα particles. d Zoom in tomogram z slice view of side view GspDα particles, indicated by white arrowheads with blue outline. eg Tomogram z slices showing tilted GspDα on the inner membrane or GspDα with only one side connecting the inner membrane (white arrowheads with blue outline). The outer and inner membranes are indicated by the purple line and the dark red line, respectively. h A tomogram z slice view of E. coli overexpressing GspDα and with D-methionine added (the GspDα particle on the OM is indicated by a white arrowhead with a green outline). i Segmentation of the tomogram shown in h, with OM colored purple, PG colored yellow, IM colored dark red, GspDα on IM colored blue, and GspDα on OM colored green. j Immunoblotting results of GspDα overexpression by using anti-His tag antibodies. BL21 (DE3) cells (WT) without plasmid transformation were used as the control. k, Inner and outer membrane fractions of the GspDα overexpressing cells treated with/without D-methionine were separated by sucrose density gradient centrifugation. The membrane location of GspDα was immunoblotted with anti-His tag antibodies. l Outer membrane proteins of the GspDα overexpressing cells treated with/without D-methionine were extracted using a bacterial membrane protein extraction kit. The total GspDα and outer membrane GspDα were examined by western blot analysis using anti-His tag antibodies. The expression and membrane location experiments of GspDα were repeated three times independently with similar results. IM inner membrane, OM outer membrane, PG peptidoglycan, D-Met D-methionine, WT wild type. Source data are provided as a Source Data file.

In situ structure of GspDα at subnanometer resolution

For subtomogram averaging, ~32,000 particles were manually picked, and data processing was carried out using EMAN226. In a series of previous in vitro structures, it has been verified that secretin has C15 symmetry5,7,9,27,28. However, the symmetry of secretin in vivo has not been verified. To confirm the symmetry of our particles, we developed an algorithm to measure the radius of individual particles and generated a histogram to look for potential multiple radii (Supplementary Fig. 2). The histogram shows only one peak with a diameter consistent with the C15 structure of GspDα (PDB: 5WQ7), strongly implying that there is only one diameter of GspDα particles in situ on the inner membrane, corresponding to its C15 structure. One challenge in subtomogram alignment of the secretin is that, due to the lack of low-resolution features along the symmetry axis, at the initial stage of orientation determination, it is difficult to distinguish top-view particles from bottom-view ones without information about the membrane position. Therefore, at the beginning of the subtomogram refinement, many top-view particles were misaligned and positioned upside down, limiting the final resolution, and causing the initial reconstructed map to exhibit some “D” symmetric features (symmetry along the central xy plane) in addition to the 15-fold symmetry. To resolve this ambiguity, we developed an algorithm in the EMAN2 tomography package, which considers the geometric location of the cell membrane as an additional factor in particle orientation determination (see “Methods”). The use of this methodology improved both visible features in the map as well as the measured resolution (Supplementary Fig. 3b, c).

The final C15 symmetrized subtomogram average achieved a measured resolution of 9 Å (Supplementary Fig. 4a), demonstrated by some visibility of alpha helices in the map (Supplementary Fig. 5b, Supplementary Movie 1). The inner membrane bilayer is clearly resolved and visible at the bottom of this density map. Above the inner membrane bilayer are the spool-shaped protein (cylinder with a larger diameter at both ends) with a sealed membrane adjacent surface and an open membrane distal surface. The N0–N3 domains can be clearly recognized in the density map (Supplementary Fig. 5b). In N1 and N2 domains, the two layers of alpha-helices are just resolved, consistent with the measured resolution of the map. With the position of the N domains identified, the transmembrane region of GspDα can be approximately modeled using the length of the secretin domain. The tip of GspDα (α7–β11, β14, and the membrane-buried region of β10 and β155) transits through one leaflet of the lipid bilayer but does not penetrate through or generate an opening on the membrane (Supplementary Fig. 5b). The sealed end of the cylinder corresponds to the gate region of the secretin. The connecting density between the protein and the inner membrane appears to have very low occupancy based on the low isosurface threshold required to visualize it. Interpreting this connection thus required some additional effort.

GspDα is flexible on the bacterial inner membrane

As both the membrane and GspDα particles are clearly resolved in the raw tomogram, we could directly observe that some particles were slightly tilted with respect to the membrane, instead of perpendicular to the membrane (Fig. 1e–g). To further confirm the membrane connection of GspDα, using the C15 density map and the corresponding orientation information of each particle, we generated an asymmetric structure by relaxing the symmetry (Supplementary Fig. 3e, f) (see “Methods”), producing a structure with a reduced resolution of 15 Å (Supplementary Fig. 5d). In this density map, the inner membrane leaflets are still resolved, as expected, and GspDα constitutes the middle-contracted cylindrical density. The membrane connecting density is now clearly visible with apparent full occupancy, but only on one side of the cylinder. The membrane connection covers a roughly 100° arc, depending on the threshold level, so when bound to the inner membrane, GspDα is only partially connected. When C15 symmetry is applied along the cylindrical axis, this partial connection is averaged out, yielding the lower apparent occupancy in the original symmetrized structure (Supplementary Fig. 5b).

In addition, we performed a focused refinement on the protein region, excluding the membrane, and compared the particle orientation with that of the integrated refinement result (Supplementary Fig. 3g) (see “Methods”). The orientation difference between the two refinements can then be determined for each particle. By classifying these differences, we recovered a trajectory for GspDα on the inner membrane, where the multimer swings around the membrane contact site (Supplementary Movie 2). We show the two endpoints’ conformations for this trajectory (Fig. 2b,c): in conformation A, the symmetry axis of GspDα is perpendicular to the membrane; and in conformation B, the symmetry axis of GspDα is tilted 2.86° compared to conformation A.

Fig. 2: In situ structures of the GspDα secretin on the outer and inner membranes.
figure 2

a The in situ structure of GspDα on the outer membrane, showing the central slice view and side view. The secretin, N3, N2, N1, and N0 domains are marked with dashed lines. b, c The in situ structures of GspDα conformation A and conformation B on the inner membrane respectively, showing the central slice view and side view. Density maps are fitted by the GspDα in vitro structure (PDB: 5WQ7). OM outer membrane, IM inner membrane.

In summary, these results indicate that the GspDα multimer is not integrated into the bacterial inner membrane. Instead, it only forms a loose connection with significant mobility, facilitating its further transportation to the outer membrane (see “Discussion”).

GspDα structure on the bacterial outer membrane

We hypothesize that the GspDα multimer is on the inner membrane because of the existence of the peptidoglycan, which is a meshwork whose pore size is too small for the GspDα multimer to pass through and translocate to the outer membrane. We attempted to reduce peptidoglycan crosslinking to see if this would facilitate GspDα outer membrane targeting. In previous studies, it was shown in Aeromonas hydrophila that decreasing the cross-linking of the peptidoglycan by glycine could localize the secretin ExeD to the outer membrane12,13. Biochemical evidence also showed that the growth of E. coli in media containing a high concentration of D-methionine decreases the cross-linking of the E. coli peptidoglycan29, the mechanism of which includes both D-methionine incorporating into the muropeptides and D-methionine directly influencing peptidoglycan remodeling. Therefore, we raised E. coli minicells in LB medium containing 40 mM D-methionine, induced expression of GspDα, and found that this increased the proportion of GspDα on the outer membrane compared to the control (Fig. 1k, l). Within the tomograms, we observed GspDα particles located on the outer membrane as well as the inner membrane (Fig. 1h, i). For subtomogram averaging, we only selected particles on the outer membrane. There were 309 particles, leading to a final refinement with C15 symmetry at 16 Å resolution (Supplementary Fig. 4b). Within the density map, the outer membrane bilayer is again resolved with an adjacent spool-shaped density. The membrane adjacent surface of the cylinder is sealed, indicating the gate region. The N0, N1, N2, and N3 domains could be recognized from the membrane distal to the membrane adjacent side of the cylinder (Fig. 2a).

To further investigate the protein-membrane interactions, we relaxed the C15 symmetry as previously for GspDα on the inner membrane. The resulting density map, seen from the cross-section at the membrane connection, shows GspDα connecting to the membrane with an even distribution of occupancy on the contour of the cylinder periphery (Supplementary Fig. 5c), instead of the one-side connection observed on the inner membrane (Supplementary Fig. 5d). To examine the symmetry of GspDα on the outer membrane, we did a refinement imposing only C5 symmetry if the structure has C15 symmetry, it should show 3 structural repeats within one C5 symmetry unit. The resulting density map retains clear C15 symmetrical features (Supplementary Fig. 5e), confirming the symmetry of GspDα in situ. For comparison, we did C5 refinement on the GspDα on the inner membrane dataset, and the resulting density map did not manifest clear C15 features (Supplementary Fig. 5f). Together, these results show that enlarging the pore size of peptidoglycan by adding D-methionine when raising E. coli permits GspDα to translocate to the outer membrane, where GspDα adopts a more consistent conformation and forms an evenly-distributed connection with the membrane.

GspDβ on the bacterial outer membrane co-exists with GspS

GspDβ, as a homolog of GspDα, has a similar structure except for an additional cap gate on the membrane adjacent side5,8, which may generate different transmembrane regions and membrane interactions. Also, GspDβ has a different outer membrane targeting mechanism. Instead of using scaffolding proteins or enlarging the peptidoglycan pore size, a small protein GspS, named pilotin, could bind to the S domain of GspDβ and translocate GspDβ to the outer membrane4,15. Therefore, to visualize GspDβ particles on the outer membrane and verify whether GspS forms a complex with GspDβ there, we performed two experiments. One is that the GspDβ and GspS expression are both induced in wild-type E. coli minicells, the other one is that only GspDβ expression is induced in wild-type E. coli minicells, and the protein expression was verified (Fig. 3h).

Fig. 3: In situ structures of the GspDβ secretin and determination of the overexpression and membrane location of GspDβ.
figure 3

a The tomogram z slice view of E. coli overexpressing GspDβ and GspS. b The in situ structure of the GspDβ–GspS complex on the outer membrane when both GspDβ and GspS are overexpressed, showing the side view and central slice view. c, d Tomogram z slice views of E. coli overexpressing only GspDβ. e The in situ structure of the GspDβ–GspS complex on the outer membrane when only GspDβ is overexpressed, showing the side view and central slice view. f The tomogram z slice view of ∆gspS E. coli overexpressing GspDβ. g The in situ structure of the GspDβ multimer on the inner membrane, showing the side view and central slice view. Particles on the outer and inner membranes are indicated by white arrowheads with a red outline and white arrowheads with a yellow outline, respectively. The density maps are fitted with the in vitro structure of the GspDβ–GspS complex (PDB: 5ZDH). The secretin, N3, N2, N1, and N0 domains are marked with dashed lines. h Immunoblotting results of GspDβ and GspS overexpression by using anti-His antibodies. BL21 (DE3) cells (WT) and BW25113-ΔgspS cells (ΔgspS) without plasmid transformation were used as the control. i Detection of the membrane location of GspDβ through sucrose density gradient centrifugation followed by western blot analysis using anti-His tag antibodies (the bacteria strains are marked on the left, and all the bands correspond to GspDβ). The expression and membrane location experiments of GspDβ were repeated three times independently with similar results. IM inner membrane, OM outer membrane, WT wild type. Source data are provided as a Source Data file.

In the reconstructed tomograms of GspDβ–GspS expressed cells, particles are all found on the outer membrane (Fig. 3a), verified by a membrane separation experiment (Fig. 3i). We boxed 514 subtomogram particles, and the refinement with C15 symmetry achieved a 19 Å resolution (Supplementary Fig. 4c). To verify the symmetry, we did a refinement with C5 symmetry applied, and the resulting density map still shows C15 features (Supplementary Fig. 6a), confirming that GspDβ in situ has C15 symmetry. In the density map (Fig. 3b), the outer membrane density is visible as the top layer. Below the outer membrane is the spool-shaped density, similar to that of GspDα. The membrane adjacent surface of the cylinder density is sealed, indicating the gate region. Notably, extra density appearing as 15 lumps can be seen connecting to the membrane adjacent periphery of the cylinder. These extra densities likely belong to GspS, indicating that GspDβ on the outer membrane exists in a complex with GspS. The in vitro structure of the GspDβ–GspS complex (PDB: 5ZDH) could be well fitted into the density map, and according to the position of the membrane density, the transmembrane region could be located at α7–α10, β11–β14, and the membrane buried region of β10 and β155,8. In contrast to GspDα which only transits through one leaflet of the membrane, GspDβ transit through two leaflets of the lipid bilayer, with the additional cap gate contacting the outer leaflet of the outer membrane.

We expect that in cells where only GspDβ is expressed, particles will be on the outer membrane since there should be the endogenous expression of GspS in the E. coli BL21(DE3) strain. Observed from the tomograms, most particles are on the outer membrane (Fig. 3c), but surprisingly, a few particle side views are seen on the inner membrane (Fig. 3d), which was verified with a membrane separation experiment (Fig. 3i). Within this dataset, 910 particles were picked, including a mixture of outer and inner membrane-located particles. After one round of refinement, we performed a multi-reference classification with two references rotated 180° with respect to each other, which produced two populations of particles with a ratio of 723 outer membrane particles to 187 inner membrane particles. We used the 723 particles to do the refinement with C15 symmetry and achieved a 16 Å resolution structure (Fig. 3e and Supplementary Fig. 4d). Refinement with C5 symmetry was also performed, and the density map showed C15 features (Supplementary Fig. 6b). This structure basically resembles that of the GspDβ-GspS expressed dataset (Fig. 3b), but notably, the GspS density is still seen outside the GspDβ channel, appearing as lumps connecting to the cylinder surface, although we did not induce overexpression of GspS.

Together, these results indicate that, when GspS is overexpressed together with GspDβ, they form a complex on the outer membrane. Without exogenous GspS expression, when GspDβ is expressed, the E. coli endogenous GspS is sufficient to locate GspDβ to the outer membrane, where GspS stays attached to GspDβ.

GspDβ multimer is visualized on the bacterial inner membrane

As biochemistry results have shown that GspDβ could form multimers on the inner membrane when gspS is knocked out4,15, to visualize GspDβ on the inner membrane, we induced expression of GspDβ within ΔgspS E. coli cells (Fig. 3h). Within the reconstructed tomograms, particles are all visualized on the inner membrane, with the non-transmembrane domains located in the periplasm (Fig. 3f, i). There are 370 particles, and the refinement with C15 symmetry achieved a 14 Å resolution structure (Supplementary Fig. 4e), rendering GspDβ on the inner membrane (Fig. 3g). The membrane density could be located at the bottom, and the cylinder density appears above, inserted into the membrane. The sealing at the lower end of the cylinder corresponds to the gate region. The lower end of GspDβ transits through two leaflets of the inner membrane, and there is no GspS density seen outside the cylinder density. These results indicate that, when gspS is knocked out, GspDβ could self-assemble into a multimer that has stable conformation on the inner membrane.

Discussion

In this study, we show the in situ structures of the GspDα secretin from bacterial T2SS and their interactions with the inner and outer membranes. We show that: (1) GspDα could self-assemble into its multimeric form on the inner membrane without overexpression of any other protein (Fig. 1d); (2) GspDα multimer forms flexible connections with the inner membrane, and does not generate an opening on the inner membrane (Fig. 2b, c); (3) the undisturbed normal peptidoglycan pore size does not allow the passage of a GspDα multimer30; (4) dissociative GspDα multimers exist in the periplasm of E. coli expressing GspDα, but only between the peptidoglycan and the inner membrane (Supplementary Fig. 7a–c); (5) D-methionine could reduce peptidoglycan crosslinking29, and adding D-methionine when raising E. coli could localize GspDα to the outer membrane (Fig. 1h); (6) dissociative GspDα multimers exist in the periplasm of E. coli raised with D-methionine and expressing GspDα (Supplementary Fig. 7d); (7) GspDα on the outer membrane exists in a more stable conformation, showing distinguishable symmetry, and forming evenly-distributed connections with the outer membrane (Supplementary Fig. 5c, e). Altogether, this evidence supports a translocation model in which GspDα first forms multimers on the inner membrane, where it stays in an intermediate state, unstable and swinging, which could favor its transportation to the outer membrane. When the peptidoglycan pore is enlarged by D-methionine, the multimer translocates to the outer membrane without disassembly. On the outer membrane, it exists in a consistent conformation, firmly attached to the membrane, which potentially facilitates substrate transportation (Fig. 4a). Notably, in our data, after we treat E. coli cells with D-methionine, not all GspDα particles are located on the outer membrane. This result indicates that other pathways or factors possibly exist to help GspDα translocate to the outer membrane, such as the GspA and GspB proteins encoded in the T2SSα operon. We acknowledge that the inner membrane-associated state results are observed from overexpression bacteria and the states we observed may not represent the exact scenario happening in an un-manipulated bacterium cell.

Fig. 4: Proposed models of the outer membrane translocation process of the GspDα secretin (a) and the GspDβ secretin (b).
figure 4

a Firstly, GspDα forms multimers on the inner membrane, and adopts an unstable and swinging conformation, changing between conformation A (light blue) and conformation B (dark blue). The multimer could dissociate from the inner membrane and enter the periplasm (light blue cartoon model), but it could not pass through the peptidoglycan and insert into the outer membrane. When the peptidoglycan pore size is enlarged by D-methionine, the multimer translocates to the outer membrane, where it exists in a more stable conformation (green). b GspDβ first forms multimers on the inner membrane (yellow); then the pilotin GspS (pink) approaches and binds the GspDβ monomer, transporting it to the outer membrane. On the outer membrane, GspDβ assembles into a multimer again and GspS stays associated, forming a GspDβ–GspS multimer complex (light red). IM inner membrane, PG peptidoglycan, OM outer membrane, D-Met D-methionine.

We observed GspDβ multimer structures on both the bacterial inner and outer membranes. We show that: (1) when gspS is knocked out in E. coli, GspDβ forms multimers on the inner membrane (Fig. 3f, g)4; (2) when gspS is not knocked out or overexpressed in E. coli, a small number of GspDβ multimers are still found on the inner membrane (Fig. 3d); (3) the undisturbed normal peptidoglycan pore size does not allow the passage of GspDβ multimers30; (4) when GspS exists, either from vector overexpression or endogenous genome expression, GspDβ multimers are found on the outer membrane (Fig. 3a, c), even if the peptidoglycan is undisturbed; (5) GspS binds to GspDβ monomer in a 1:1 ratio8; (6) GspS itself is a lipoprotein that could translocate from the inner to the outer membrane through Lol pathway15; (7) we do not observe disassociated GspDβ multimers in the periplasm; (8) GspDβ multimers on the outer membrane exist in a complex with GspS, that is, GspS does not dissociate after transportation of GspDβ to the outer membrane (Fig. 3b, e). Based on what we observed, we would propose a translocation model that, first GspDβ self-assembles into multimers on the inner membrane, then GspS on the inner membrane binds to GspDβ, which dissociates into monomers. GspS then transports GspDβ to the outer membrane, where GspDβ assembles into multimers again with GspS associated, forming a GspDβ–GspS multimer complex (Fig. 4b). Because of the limitations of cryoET observations, we did not directly visualize the process of GspDβ multimers on the inner membrane dissociating into monomers and entering the periplasm with GspS bound. Further investigations are needed to provide evidence for this process.

In our structures, the transmembrane region of GspDα is only buried inside one leaflet of the membrane, instead of penetrating the membrane. This transmembrane pattern is unexpected and would be biophysically unreasonable but can be clearly observed from our cryo-ET structures. Notably, the in situ structure of the whole T2SS has been visualized11, and it was indicated that the secretin of Legionella pneumophila, a Klebsiella-type secretin based on the secretin phylogenetic analysis4, also punctures through only one leaflet of the outer membrane. It has been suggested that the transmembrane region of the secretin may have a more extended conformation in vivo based on these findings. Further investigations are necessary in order to explain why this membrane interaction could exist, and what is the in vivo conformation that makes this kind of transmembrane pattern possible.

Comparing the inner and outer membrane-located structures of GspDα and GspDβ, we observe that GspDα only transits one leaflet of the lipid bilayer, but GspDβ transits through both leaflets (Supplementary Fig. 8). This difference could give GspDβ higher efficiency when transporting substrates. On the E. coli genome, there are two T2SS gene operons, T2SSα and T2SSβ. During evolution, T2SSβ may be generated by gene duplication of the T2SSα operon. However, T2SSα is endogenously silenced, while T2SSβ is actively expressed and functional15. Our result could provide a possible explanation, that is, the possibly higher substrate transport efficiency gives GspDβ an advantage during evolution, making it the preferred and expressed T2SS in many bacteria.

In summary, we determined four in situ structures of the T2SS secretin: both the inner and outer membrane structures of GspDα and GspDβ. GspDα exists in an unstable form on the inner membrane, but forms firm and stable connections with the outer membrane. The GspDβ multimer exists on the inner membrane in wild-type E. coli, and GspDβ on the outer membrane exists together with GspS. Taken together, these results add to the knowledge of the membrane interactions of the T2SS secretins and their outer membrane targeting process, providing a comprehensive model for the secretin biogenesis process of Proteobacteria.

Methods

Bacterial strains, plasmids, chemicals, and protein expression

E. coli BW25113-ΔgspS strain was obtained from the Nara Institute of Science and Technology in Japan, from the Keio collection31,32. E. coli BL21 (DE3)-ΔgspS strain was constructed by Ubigene Biosciences Co., Ltd. (Guangzhou, China). The DNA sequences encoding GspDα (E. coli K12 strain), GspDβ (ETEC H10407 strain), and GspS (ETEC H10407 strain) were synthesized by General Biosystems Co., Ltd. (Anhui, China). D-methionine was purchased from Sigma (catalog number: M9375-5G).

The gspDα from the E. coli K12 strain with no tag or with a C-terminal hexahistidine tag was cloned into pETDuet-1, yielding pETDuet-gspDα and pETDuet-gspDα-his, which were expressed in E. coli BL21 (DE3), respectively. The vector with no tag was used for cryo-ET imaging, and the vector with the hexahistidine tag was used to perform biochemistry tests. The gspDβ of ETEC H10407 strain with hexahistidine tag on the C terminal was inserted into pETDuet-1, yielding pETDuet-gspDβ-his, which was expressed in E. coli Rosetta (DE3) and used for cryo-ET imaging and expressed in E. coli BL21 (DE3) for biochemistry experiments. The gene of gspS was cloned into pETDuet-gspDβ-his with a hexahistidine tag at the C terminus, yielding pETDuet-gspDβ-his-gspS-his, which was expressed in E. coli BL21 (DE3) and used for the biochemistry experiments. The gspDβ with hexahistidine tag and gspS with no tag from ETEC H10407 strain were both inserted into pETDuet-1’s two multi-cloning sites, yielding pETDuet-gspDβ-his-gspS, which was expressed in E. coli BL21 (DE3) and used for cryo-ET imaging. The gspDβ of ETEC H10407 strain with hexahistidine tag was inserted into pBAD, yielding pBAD-gspDβ-his, which was expressed in BW25113-ΔgspS strain and used for cryo-ET imaging, and was expressed in BL21 (DE3)-ΔgspS strain and used for biochemistry experiments. For cryo-ET imaging experiments, we obtained minicells by co-transforming the E. coli strains with pBS58, which increases the frequency of cell division by constitutively expressing cell division-related genes33. The obtained minicell thickness ranges from 130 to 250 nm. E. coli cells only transformed with pBS58 and not with any protein expression vectors were used as negative control cells.

Cells were grown at 37 °C in LB medium supplemented with appropriate antibiotics (final concentrations of 100 µg/ml ampicillin and/or 50 µg/ml kanamycin), and 40 mM D-methionine when needed. Protein expression was induced at an OD600 of 0.8, by adding 0.5 mM IPTG or 0.02% arabinose at 20 °C overnight. Protein expression was examined by western blotting analysis using anti-His tag antibodies. For experiments with GspDα, the anti-His tag antibodies were diluted 10,000-fold. For experiments with GspDβ, the anti-His tag antibodies were diluted 1000-fold. The antibodies for the control EF-Tu were from previous publications34,35 and were diluted 20,000-fold.

Sample preparation

To perform minicell separation, E. coli cultures were first centrifuged at 1000 × g for 40 min to precipitate large cells. The supernatant was taken and centrifuged at 20,000 × g, for 10 min. The precipitation was washed with PBS buffer for one time, then resuspended to an OD600 of 10. Cells were mixed with 6 nm BSA fiducial gold (Aurion) and then, 3 µl mixture was deposited onto freshly glow-discharged, continuous carbon film-covered grids (Quantifoil Cu R3.5/1 with 2 nm continuous carbon film, 200 mesh). Using Vitrobot Mark IV (FEI), the grids were blotted with filter papers for 4 s and plunged into liquid ethane. Samples were stored in liquid nitrogen.

Cryo-ET data collection

All tilt series were collected using 5° angular step, from −50° to +50°, defocus from −1.5 to −5 µm, and a total dose of about 100 e2. For the GspDα on the inner and outer membrane dataset, the sample was imaged using the bidirectional data collection scheme (starting angle −30°) on a 300 kV FEI Titan Krios microscope with a Gatan K2 Summit direct electron detector camera, using the SerialEM software. The magnification was 81,000×, with a pixel size of 1.76 Å. For the other datasets and the control dataset, the sample was imaged using a dose symmetric data collection scheme on a 200 kV FEI Glacios microscope with a Falcon 4 direct electron detector camera, using the Tomography software. The magnification was 92,000×, with a calibrated pixel size of 1.52 Å (Supplementary Table 1).

Cryo-ET data processing of GspDα on the inner membrane dataset

EMAN2 refinement pipeline

The raw frames were aligned by MotionCorr236. The tilt series alignment, tomogram reconstruction, and CTF correction were performed in EMAN226. For the refinement, 32,915 particles were manually picked from 250 tilt series and extracted using a box size of 256. After generating an initial model with a small set of particles that have a good signal-to-noise ratio, 4 iterations of subtomogram refinement were performed. The worst 20% of particles were excluded based on their similarity to the averaged structure.

Using the cell membrane feature to assist particle alignment

Due to the shape of GspDα, its top view and bottom view are hard to distinguish, and after the refinement, many top-view particles were wrongly aligned by about 180°. To correct this, an algorithm was developed that decides a center for each tomogram based on all the particles picked in this tomogram and draws a vector from the particle position to the center (or the opposite direction if the parameter “invert” is toggled). If a particle’s orientation from the refinement result is not facing the same side as its corresponding vector, its orientation will be rotated 180°. After the correction, a new particle orientation file was written and used as input for the next iteration of refinement. In the next iterations, local refinement was used to make sure that the particle orientation will not rotate back to the wrong side. And three more iterations of refinement were performed to achieve the final structure.

Symmetry determination and particle radius measurement

To determine the particle symmetry, we first erected each particle based on its orientation from the refinement and projected them to 2D images, and then performed 2D classification in EMAN2, but the result did not show distinctly different classes or explicit symmetry. We also did refinement with C12, C14, and C16 symmetry, but the resulting density map did not show any obvious feature differences or feature improvements compared to the C15 structure. To measure the particle radius, we applied a low pass filter to the particle 2D projection image, calculated the mean radial intensity along the radial axis, and then recorded the index corresponding to the maximum intensity value (Supplementary Fig. 2). The particle radius is pixel number times angstrom per pixel.

Symmetry release and particle movement trajectory calculation

To visualize the membrane connecting region of GspDα more clearly, we did a focused classification, using a mask that is focusing on the membrane connecting region of the density map (Supplementary Fig. 3, dashed line box). Within the results, there is one class that shows C1 features (Supplementary Fig. 3, red line box). To achieve this C1 structure, we did a symmetry release using the whole dataset: with the orientation of the symmetry axis unaltered, each particle is allowed to rotate around the symmetry axis and search for a best-fitted symmetry unit for one iteration, followed by averaging of subtomograms. Then, the particle orientations are subject to two iterations of gold standard refinement, doing only a local search.

We have observed tilted inserted particles from the tomograms (Fig. 1e–g), which indicates that, for these particles in the alignment process, if we give a reference map where the membrane and the particle symmetry axis are perpendicular, the membrane density and the protein density could not be aligned correctly at the same time. If the membrane density plane is correctly aligned, the protein density orientation will be inaccurate. To resolve this problem, we did a focused refinement. A mask was generated enclosing the non-transmembrane part of the protein, excluding the membrane density, and one iteration of alignment was performed using the masked structure as a reference (Supplementary Fig. 3g). In this iteration, the protein density part should be aligned correctly. We then compared the orientation from the focused refinement with that from the overall refinement, and the orientation difference represents the protein tilt angle. The orientation difference for each particle could be plotted, and a trajectory was calculated. To show the movement, the full trajectory was sectioned into several intervals, and class averages for each interval were calculated representing the corresponding conformation (Supplementary Fig. 3h).

Cryo-ET data processing of the other datasets

The data processing workflow for the GspDα on the outer membrane dataset, GspDβ–GspS complex dataset, and GspDβ on the inner membrane dataset resembles the GspDα on the inner membrane dataset, but simpler. After motion correction and tomogram reconstruction, particles were picked from tomograms. After generating the initial model, the refinement was firstly done with C15 symmetry for 4 iterations, and then particle orientations were corrected using the cell membrane geometry. Then another 4 iterations of local refinement were done to achieve the final structure (for C5 refinements, we only changed the symmetry input to C5, and other parameters were the same). The symmetry release protocol of the GspDα on the outer membrane dataset followed that of the GspDα on the inner membrane dataset.

Extraction of outer membrane protein

Outer membrane proteins were extracted using a bacterial membrane protein extraction kit (BestBio). Briefly, E. coli cells expressing the protein of interest were harvested by centrifugation, washed twice with PBS, and suspended in Extracting Solution A (containing protease inhibitor mixture). After 1 h of shaking at 4 °C, the suspension was centrifuged at 12,000 × g for 5 min at 4 °C to remove the pellet. The supernatant was taken and incubated at 37 °C for 1 h, after which it could be observed that the liquid was divided into two layers. The liquid at the bottom was collected as the outer membrane proteins.

Membrane separation of the E. coli cell envelope

Separation of inner and outer membranes by isopycnic sucrose density gradient centrifugation was performed as described previously37. Briefly, E. coli cells expressing the protein of interest were harvested by centrifugation and suspended in buffer A (10 mM Tris-HCl, pH 7.5; 0.5 M sucrose; 10 mg/ml lysozyme; 1.5 mM EDTA). After incubation on ice for 7 min, the suspension was centrifugated at 10,000 × g for 10 min at 4 °C, and then the pellet was resuspended in buffer B (10 mM Tris-HCl, pH 7.5; 0.2 M sucrose; 1 M MgCl2) containing RNase/DNase nuclease reagent and protease inhibitor cocktail. The mixture was lysed by sonication and centrifuged at 6169 × g for 10 min at 4 °C to remove the cell debris. Subsequently, the supernatant was pelleted by ultracentrifugation at 184,500 × g for 1 h at 4 °C and then suspended in buffer C (1 mM Tris-HCl, pH 7.5; 1 mM EDTA,) containing 20% sucrose to get the total membrane fraction. After that, buffer C containing 73% sucrose and buffer C containing 53% sucrose were layered in a 13 ml tube (Beckman Coulter), starting from the bottom, and finally, 0.5 ml of the total membrane fraction was layered on the top, followed by buffer C containing 20% sucrose to fill up the 13 ml tube. Then, the density gradients were centrifuged at 288,000 × g for 16 h at 4 °C, and the tawny band at the interface between 20% and 53% sucrose layers and the white band at the interface between 53% and 73% sucrose layers were collected as inner membrane fraction and outer membrane fraction, respectively.

Reporting summary

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