Mycobacterium tuberculosis encodes five homologous, but functionally distinct, T7SSs that are designated ESX-1 to ESX-5. These systems translocate a number of effector proteins across the unique and impermeable diderm cell envelope2,3. Because of their importance for mycobacterial physiology and virulence, T7SSs are considered to be promising targets for the development of drugs for the treatment or prevention of tuberculosis4. Although T7SSs have previously been shown to form hexameric complexes5, high-resolution structural information exists only for part of the T7SS—a dimeric ESX-3 subcomplex from the nonpathogenic species Mycobacterium smegmatis6,7. Here we reconstituted the ESX-5 T7SS of M. tuberculosis H37Rv in M. smegmatis to obtain a structural view of the entire T7SS membrane complex from this human pathogen.

Architecture and stoichiometry

The M. tuberculosis ESX-5 system showed robust expression in M. smegmatis and correct assembly of the membrane complex (Extended Data Fig. 1a, b). Purification of the M. tuberculosis ESX-5 membrane complex (using a C-terminal Strep tag on EccC5 and mild solubilization conditions) resulted in copurification of the conserved MycP5 protease (Extended Data Fig. 1c–e). To our knowledge, MycP (also known as mycosin) is absent in all previously reported T7SS structures5,6,7—although MycP is known to be essential for T7SS function and complex stability8. The addition of nucleotides and MgCl2 improved sample homogeneity, as judged by a more-distinct high molecular weight complex on native-PAGE, size exclusion chromatography and subsequent negative-stain electron microscopy analysis (Extended Data Fig. 1f–j).

Cryo-electron microscopy (cryo-EM) analysis of the M. tuberculosis ESX-5 complex purified in the presence of ADP–AlF3 showed clear hexameric particles (Extended Data Figs. 2b, 3). We performed an ab initio reconstruction without symmetry enforcement that yielded an average resolution of approximately 4 Å (Extended Data Fig. 4), which improved to an overall resolution of approximately 3.5 Å after further data processing; this allowed us to build around 78% of the stable complex de novo (Supplementary Tables 1, 2). The intact machinery comprises EccB5, EccC5, EccD5, EccE5 and MycP5 with a 6:6:12:6:3 stoichiometry (Fig. 1b, c, d, Supplementary Video 1) resulting in a 2.32-MDa complex that is anchored in the inner membrane through 165 transmembrane helices (TMHs) (Fig. 1e). The membrane assembly is best described as a trimer of dimers, in which each dimer comprises a single copy of MycP5 and two protomers each of one copy of EccB5, EccC5, EccE5 and two copies of EccD5 (Fig. 1f). The overall fold and stoichiometry of a dimeric building block of M. tuberculosis ESX-5 is similar to that of a dimer of ESX-3 from M. smegmatis6,7, albeit with notable differences on the periplasmic side and the angle between protomers. In the intact ESX-5 complex, the angle between protomers at the membrane level differs by about 0.5°, from 59.7° between protomers of one dimer to 60.2° between protomers of adjacent dimers. However, at the cytosolic level these angles differ by more than 10° (from 65.3° to 54.7°) (Extended Data Fig. 5). By contrast, the individual ESX-3 dimer displayed an overall angle of 72° between protomers6,7.

Fig. 1: Cryo-EM structure of the intact ESX-5 inner-membrane complex of M. tuberculosis.
figure 1

a, Genetic organization of the esx-5 locus of M. tuberculosis H37Rv, which was cloned and expressed in M. smegmatis MC2155. be, Cryo-EM density of the intact ESX-5 inner-membrane complex of M. tuberculosis, zoned and coloured for every individual component. Components are inner EccB5 (dark green), outer EccB5 (light green), EccC5 (blue), inner EccD5 (beige), outer EccD5 (orange), EccE5 (purple) and MycP5 (red). The full complex is 28.5 nm in width and 20 nm in height, and has an absolute stoichiometry of 6:6:12:6:3 for EccB5:EccC5:EccD5:EccE5:MycP5. be, Side (b), top (c) and bottom (d) views and a top cross-section (e) of the complex at the membrane level, highlighting the arrangement of the 165-TMH region. Inset, top cross-section of an extracted dimeric unit. f, Single dimer viewed from the centre of the intact complex, highlighting the central EccC5 TMH bundle and the position of MycP5 with its active site directed towards the inside of the periplasmic cavity. g, Ribbon model of the M. tuberculosis ESX-5 assembly.

Structural rearrangements of periplasmic domains

The periplasmic assembly of the ESX-5 membrane complex of M. tuberculosis is formed by three EccB5 dimers and three MycP5 proteases. The EccB5 dimers assemble in a triangle, which forms a central cavity (Fig. 2c). Within an EccB5 dimer, two slightly different conformations (that is, inner and outer) can be distinguished between monomers, depending on their position. EccB5 dimerization is mediated mainly through the R1 and R4 repeat domains and is further stabilized by the EccB5 C termini, which wrap around their interacting EccB5 partner to form intermolecular hydrophobic contacts with its R1 domain (Fig. 2d). The GIPGAP motif—which is a highly conserved region in EccB homologues—is central to these interactions (Fig. 2d). Compared to the EccB3 dimer from the ESX-3 subassembly6,7, the three EccB5 dimers are rotated by about 52° with respect to their corresponding EccC5–EccD5–EccE5 membrane dimers, which indicates that large conformational rearrangements are required during maturation into the fully assembled hexamer (Extended Data Fig. 6d). To form the triangle-shaped assembly, the inner EccB5 engages the outer EccB5 of the adjacent dimer by packing its R3 domain against the α-helices α5 and α8 of domains R2 and R3, respectively, which results in an asymmetric EccB5–tip arrangement (Fig. 2c). Consequently, domain R3 of the outer EccB5 does not form any interactions at its tip extremity and thus displays higher flexibility, consistent with previous observations9.

Fig. 2: MycP5 drives EccB5 hexamerization and stabilization of the membrane complex.
figure 2

a, Transparent assembly of intact M. tuberculosis ESX-5, with EccB5 and MycP5 coloured as in Fig. 1. b, Complete structure of monomeric M. tuberculosis EccB5, highlighting its overall fold and domains. c, Top and bottom view of the EccB5–MycP5 periplasmic assembly with one unit (EccB5 dimer and MycP5 monomer) as ribbon model, highlighting the active site of MycP5 in yellow. CD, central domain; TM, TMH. d, EccB5 dimerization site, highlighting the C-terminus of outer EccB5 that is wrapped around the R1 and R4 domains of the adjacent inner EccB5 monomer, the conserved GIPGAP motif of EccB5 (in yellow) and the interactions of the EccB5 dimer with loop 2 and the linker connection of MycP5. e, Transparent map of M. tuberculosis ESX-5 without copurified MycP5, with EccB5 highlighted in dark green. The high flexibility of EccB5 and the overall heterogeneity of the membrane complex in the absence of MycP5 is indicated by curved lines. f, EccB5–MycP5 interaction surface, highlighting the three buried tryptophans. g, Angle variation range between protomers of the MycP5-bound (+) and two unbound (−) states (I and II). Intra, between two protomers within a dimer: 62.7°, 62.4° and 62.4° (MycP5-bound); 61.8°, 61.3° and 61.5° (unbound, I); 61.4°, 62.3° and 60.2° (unbound, II). Inter, between two protomers of adjacent dimers: 57.5°, 57.3° and 57.5° (MycP5-bound); 58.5°, 58.3° and 58.3° (unbound, I); and 57.9°, 58.3° and 59.6° (unbound, II).

Periplasmic MycP5–EccB5 assembly

Our periplasmic M. tuberculosis ESX-5 map shows three MycP5 proteases that form a dome-like structure, which cap the periplasmic central cavity (Fig. 2a, c, Supplementary Video 2). Interactions between EccB5 and MycP5 are mediated mainly by the MycP5 protease domain and a composite interface that is generated by the R4 domain and loop 6 (residues Thr424 to Ser435) of the inner EccB5 (Fig. 2f). The MycP5–EccB5 interface covers a surface area of about 1,230 Å2, which leads to the burial of three conserved tryptophan residues (Trp437 and Trp469 of EccB5, and Trp523 of MycP5) (Fig. 2f). Additionally, loop 2 of MycP5 binds to the C terminus of the outer EccB5, which explains why a deletion of this loop previously showed reduced ESX-5 secretion in Mycobacterium marinum10 (Fig. 2d, Extended Data Fig. 6i). MycP5–MycP5 interactions are mediated mainly through loop 1 and the N-terminal extension (which run across the top of the MycP5 protomers), and loop 3, which contacts the neighbouring protease domain from the side (Extended Data Fig. 7c, d). Loop 5 (residues Ala151 to Val271), which is cleaved during ESX-5 maturation11, folds along the interface of two protease domains towards the central pore formed by the MycP5 trimer (Extended Data Fig. 7a). Although we could not build a complete model of loop 5 (owing to its high flexibility), this loop appears to cap the central periplasmic pore (Extended Data Fig. 7a). Notably, as loop 5 is not present in all mycosins and is dispensable for ESX-5 secretion10, a speculative role in gating remains to be identified. The active sites of the MycP5 proteases face towards the central lumen of the cavity (Fig. 2c), which implies that potential substrates of this protease are translocated through—and processed within—this periplasmic chamber.

The dimer interface between the inner and outer EccB5 is the largest in the periplasmic assembly, and covers a surface area of around 2,000 Å2 and provides a solvation-free energy gain of ΔG = −23 kcal mol−1 per dimer (Fig. 2d). By contrast, the interfaces formed between EccB5 dimers each bury a surface area of about 600 Å2, with a cumulative energy gain of only ΔG = −18 kcal mol−1 upon trimerization. This could provide an explanation as to why dimeric ESX subcomplexes are more stable than their fully assembled counterparts6,7. The intermolecular EccB5–MycP5 interactions (which have a surface area of around 395 Å2 and ΔG = −0.1 kcal mol−1) are even more modest, which provides a rationale for why interactions between MycP5 and the membrane complex have so far remained unknown.

MycP5 stabilizes the entire membrane complex

To further investigate the effect of MycP5 on the entire structure, we analysed MycP5-free M. tuberculosis ESX-5 complexes from the same preparation (Fig. 2e, g). These assemblies contained the same EccB5:EccC5:EccD5:EccE5 stoichiometry as the fully assembled complexes (Extended Data Fig. 8). Following 3D reconstruction, we obtained two MycP5-free M. tuberculosis ESX-5 maps that displayed resolution estimates of about 4.5 and about 6.7 Å (Extended Data Fig. 4). The differences were most notable on the periplasmic side, on which the six EccB5 copies showed high flexibility and did not form a stable triangular scaffold in the absence of MycP5 (Fig. 2e). This shows that MycP5 enables the trimerization of the EccB5 dimers in the periplasm. This result is highly interesting, because mycosins are subtilisin-like proteases without any additional domains apart from a TMH, some small loops and an N-terminal extension that wraps around the protein12,13. A more structural role for mycosin has previously been predicted8 but we now understand the essential role of mycosins in T7SSs.

In contrast to the periplasmic domain, the cytosolic and membrane regions of the MycP5-free maps were more similar to those of the MycP5-containing particles (Extended Data Fig. 8). However, the MycP5-free particles displayed an increased heterogeneity that affected the entire complex, which resulted in a slight waving of the membrane region and an increased angle variation between individual protomers (Fig. 2g, Extended Data Fig. 8, Supplementary Video 4). The protease domain and TMH of MycP5 synergistically reinforce the membrane complex. Their interactions with periplasmic inner EccB5 and membrane-embedded outer EccD5 (from separate protomers within a dimer) better anchor the periplasmic assembly to the membrane, while also stabilizing the dimeric unit (Extended Data Figs. 8f, 9). Additionally, by driving the formation of the periplasmic assembly, MycP5 stabilizes the entire complex by promoting cross-dimer MycP5–MycP5 and inner EccB5–outer EccB5 interactions (Extended Data Fig. 9). Our MycP5-free M. tuberculosis ESX-5 maps are reminiscent of the hexameric, low-resolution negative-stain structure of ESX-5 from Mycobacterium xenopi, in which the periplasm was similarly disorganized in the absence of MycP55.

EccC5 gates a potential secretion conduit

At the membrane level, six EccD5-dimer barrels (each of which contains 22 TMHs) together form a circular raft with an inner cavity (Extended Data Fig. 10). Within this raft, inner EccD5 monomers are situated closer to the centre, whereas outer EccD5 monomers face towards the periphery of the membrane complex. The EccD5 membrane barrels are structurally highly similar to the homologous EccD3 barrel in the ESX-3 subassembly6,7. The inner surface of each EccD5 barrel is decorated with densities that are attributable to stably bound lipids or detergent molecules, which suggests that, in their native membrane environment, these barrels are filled with membrane lipids (Extended Data Fig. 10e, f).

The TMH of each copy of EccB5 is anchored within the confinement of the EccD5 raft through hydrophobic interfaces that are provided by TMH6 and TMH11 of inner EccD5 and stably bound lipids (Fig. 3a, Extended Data Fig. 6f). Notably, no substantial intermolecular interactions can be found between adjacent EccD5 barrels. Instead, coupling between two neighbouring EccD5 barrels is achieved by the N-terminal loop and α-helix of EccB5 that run parallel to the cytoplasmic side of the inner membrane and engage in interactions with the TMHs of the neighbouring EccD5 barrel in a clockwise manner (Fig. 3a, Extended Data Fig. 6g, h). Because the TMH of EccB5 is slightly angled towards the centre of the complex, the architecture of the hexamer of EccB5 TMHs is reminiscent of a basket, the inner diameter of which shrinks from around 60 Å to around 45 Å towards the periplasmic side (Fig. 3b).

Fig. 3: A basket formed by the EccB5 TMHs holds three four-TMH bundles of EccC5.
figure 3

a, Angled view from the outside of the complex, showing the TMH and N terminus of an outer EccB5 interacting with a pocket formed by TMH8, TMH10 and TMH11 of inner EccD5 from the adjacent barrel. b, Side cross-section through the EccB5 basket that contains the EccC5 TMH bundles. Light blue densities depict the three copies of EccC5 TMH2 that form the central pyramid. Two TMHs of EccC5 were removed for clarity. Sizes indicate the inner diameters of the EccB5 basket. c, Side cross-section through an EccB5 basket, showing that the EccC5 TMH bundle does not interact with outer EccB5 from its own dimer, but instead forms lipid-mediated interactions with the outer EccB5 TMH of the adjacent dimer. Lipids are shown in gold. d, Top view of the central EccB5 basket and the EccC5 TMH bundles. Dashed line marks the TMHs that belong to one dimeric unit. e, As in d, highlighting the lipid-rich environment. In the central area that surrounds the EccC5 pyramid, lipids are not clearly distinguishable (which suggests fluidity in this area). f, g, Surface model displaying the hydrophobicity of an EccC5 TMH bundle (f) and the EccB5 basket (g). Hydrophilic amino acids are shown in turquoise, and hydrophobic residues are shown in sepia.

EccC is the only component that is present in all T7SSs (including in related systems in Firmicutes2), and is therefore thought to be the central component in this nanomachinery. Each EccC protein has—in addition to two TMHs—four FtsK/SpoIIIE-like ATPase domains (also known as nucleotide-binding domains (NBDs)) that are known to be important for secretion6,7,14. We fully resolved the twelve EccC5 TMHs in the intact M. tuberculosis ESX-5 complex; these form three four-TMH bundles, each of which belongs to the EccC5 molecules of one dimer (Fig. 3b–d, Extended Data Fig. 11). These bundles are held together by hydrophobic interactions and effectively seal the central space of the membrane complex, which is enclosed by the EccB5 basket (Fig. 3d, Supplementary Video 3). Two EccC5 TMHs from each bundle contact the TMH of the inner EccB5, which leaves the outer EccB5 TMH unbound by EccC5 (Fig. 3c, d). At the very centre of the complex, one TMH of each bundle contributes to the formation of a pyramidal assembly that aligns with the periplasmic chamber (Extended Data Fig. 11d).

The chamber within the EccB5 basket appears to be filled with lipids. However, the density for these lipids is more ambiguous than that of the lipids in and around the EccD5 barrels, which suggests that the lipids within this chamber are more fluid (Fig. 3e). Notably, the local resolution gradually increases when moving from the centre to the EccB5 basket, where the resolution is highest (Extended Data Fig. 11c). This indicates that the EccC5 TMH bundles display more flexibility, as compared to the rigid EccB5 basket. The entrance to the putative EccC5 pore widens on the cytoplasmic side, where the EccC5 stalk domains expand radially (Extended Data Figs. 5, 11a). Together, our data suggest that the six EccD5 barrels provide a stable scaffold for assembly of a secretion pore that is confined by the EccB5 TMHs and gated through three EccC5 TMH bundles. Secretion through the inner membrane complex would require rearrangement of the EccC5 TMHs. Such a proposed central pore would extend into the periplasmic chamber that is formed by EccB5 and MycP5.

Cytosolic EccC5 adopts two conformations

At the cytoplasmic side of the complex, EccC5 has a stalk helix that connects its second TMH to the first NBD (which is also known as the DUF domain). This NBD is bound to the cytosolic domains of inner and outer EccD5, which—in turn—are bound to EccE5 at the periphery, together forming a ‘cytosolic bridge’ (Fig. 1f).

The distal C-terminal part of EccC5, which comprises a string of three NBDs (NBD1, NBD2 and NBD3), adopts two main conformations: we refer to these as extended and contracted (Fig. 4a). In the extended state, the C-terminal three NBDs of EccC5 bend parallel to the membrane, and align with the cytosolic domains of inner EccD5 and EccE5 of the same protomer, and extend beyond the diameter of the membrane assembly. Although of considerably lower resolution, this density can confidently accommodate a homology model that consists of the three EccC5 NBD domains (Fig. 4a, Extended Data Fig. 12). Further classification of the extended state reveals EccC5 to be more heterogenous beyond NBD1, which suggests that NBD1 is more stably bound to components of its own protomer. Although we found only a relatively small number of particles in the contracted conformation, we solved the stable core of the membrane complex to sub-nanometre resolution (Extended Data Fig. 3). In the contracted state, the flexible arms of EccC5 extend from the interface between the DUF domain of EccC5 and the cytosolic domain of inner EccD5 (Fig. 4a, Extended Data Fig. 12). We observed three separate disc-like structures that gradually constrict from the top to the bottom. This density shows a gap at the interface between NBD1 and NBD2. This would allow the previously postulated14,15 binding of substrates to the linker 2 that connects NBD1 and NBD2, resulting in the displacement of this linker and the activation of NBD1. The highly dynamic cytoplasmic domains of the machinery may provide the basis for substrate selection, recognition or transport across the membrane.

Fig. 4: EccC5 adopts an extended and a contracted conformation.
figure 4

a, Side cross-section of density maps, showing the extended and contracted conformation of EccC5. The periplasmic and cytoplasmic chambers formed by EccB5–MycP5 and by EccC5 upon closing are highlighted. Homology models of the three C-terminal NBDs of EccC5 are fitted in the cytosolic densities. Cytosolic bridge components are coloured as in Fig. 1. b, Model of the intact T7SS inner-membrane complex, highlighting the two conformations of EccC5.

Our work provides a fully assembled structure of the ESX-5 inner-membrane complex of M. tuberculosis. As the membrane components of the five mycobacterial ESX systems show high sequence conservation, our results probably constitute a general structural blueprint for all of these T7SSs—including the virulence-related ESX-1 system (Supplementary Figs. 2, 3). Furthermore, our structure will serve as a platform for the identification of interactions that—if perturbed by small molecules—would aid in the treatment of tuberculosis.


No statistical methods were used to predetermine sample size. The experiments were not randomized, and investigators were not blinded to allocation during experiments and outcome assessment.

Molecular biology

Escherichia coli Dh5α was grown at 37 °C and 200 rpm in LB medium supplemented with 30 μg ml−1 streptomycin. Cloning was performed in E. coli Dh5α using IProof DNA polymerase from BioRad and restriction enzymes from New England Biolabs. A list of the primers used for amplification is available in Supplementary Table 3.

The plasmid expressing M. tuberculosis ESX-5 was built as follows: the backbone of the previously described pMV ESX-5mxen plasmid5 was modified to encode the unique restriction sites DraI and PacI upstream and SpeI and NdeI downstream of the TwinStrep tag sequence. The rv1782–rv1783 (eccB5–eccC5) region, including about 380 bp upstream of eccB5, of M. tuberculosis H37Rv was amplified while adding DraI and PacI restriction sites at the 5′ and 3′ ends, respectively (primers 1 and 2), and cloned into the modified plasmid upstream of the TwinStrep tag sequence, resulting in plasmid intermediate 1. The M. tuberculosis H37Rv region spanning rv1791–rv1798 (pe19–eccA5) was amplified while adding SpeI and NdeI unique restriction sites (primers 3 and 4) and cloned downstream of the TwinStrep tag sequence into the intermediate 1, resulting in plasmid intermediate 2. Plasmid intermediate 2 was digested with SpeI and SnaBI, removing the region rv1791–rv1794 (pe19–espG5), and the region encompassing rv1785–rv1794 (cyp143–espG5) was amplified as two individual PCR products (primers 5 and 6 and primers 7 and 8). The restricted backbone and PCR products were InFusion (Takara Bio)-ligated, resulting in the final pMV-ESX-5mtb containing the entire rv1782–rv1798 (eccB5–eccA5) locus.

Isolation of mycobacterial cell envelopes

Mycobacterium smegmatis MC2155 expressing M. tuberculosis ESX-5 was grown at 37 °C and 90 rpm in LB medium supplemented with 0.05% Tween 80 and 30 μg ml−1 streptomycin. Cultures were grown to an optical density (OD) at 600 nm of about 1.5, spun down for 15 min at 12,000g in a JLA-8.1000 rotor and subsequently washed in PBS. After culture collecting, all subsequent steps were performed at 4 °C. Washed cell pellets were resuspended in buffer A (50 mM Tris-HCl pH 8, 300 mM NaCl and 10% glycerol) at a concentration of about 50 OD ml−1 and lysed by passing two times through a high-pressure homogenizer (Stansted) using a pressure of 0.83 kbar. Unbroken cells were pelleted at 5,000g for 5 min and supernatants were transferred to ultracentrifugation tubes. Cell envelopes were separated from the soluble fraction by ultracentrifugation at 150,000g for 1.5 h. Following ultracentrifugation, supernatants were discarded, pellets were washed once with buffer A, resuspended in buffer A at a concentration of 750–1,000 OD ml−1, snap-frozen in liquid nitrogen and stored at −80 °C until further use. The protein concentration of the cell envelope fraction was measured by BCA assay (Pierce).

Purification of the M. tuberculosis ESX-5 membrane complex

All steps were performed at 4 °C. The M. tuberculosis ESX-5 was purified as follows: cell envelope fractions were diluted to 3 mg ml−1 in buffer B (50 mM Tris-HCl pH 8, 300 mM NaCl and 5% glycerol), supplemented with 0.25% DDM, 3 mM ADP–AlF3 and 6 mM MgCl2. Following solubilization, the cell envelope mixture was spun down at 100,000g for 20 min, supernatants were collected and incubated with StrepTactin resin (IBA). Beads were subsequently washed with buffer B supplemented with 0.03% DDM, 1 mM ADP–AlF3 and 2 mM MgCl2. Bound protein was eluted from the resin with buffer B supplemented with 0.03% DDM, 3 mM ADP–AlF3, 6 mM MgCl2 and 10 mM desthiobiotin. The protein concentration of the eluate was measured by Bradford assay and amphipol A8-35 was added in an amphipol:protein ratio of 5:1. After a 1-h incubation, the amphipol-containing eluate was incubated overnight (around 12–16 h) with BioBeads in a BioBeads:detergent ratio of 20:1. Subsequently, BioBeads were removed using gravity flow chromatography columns and the sample was concentrated using Amicon Ultra 0.5-ml 100-kDa spin concentrators. The concentrated sample was further purified through size exclusion chromatography (SEC), using a Superose 6 Increase column running in buffer C (20 mM Tris-HCl pH 8, 200 mM NaCl) supplemented with 1 mM ADP–AlF3 and 2 mM MgCl2. Size exclusion chromatography fractions were analysed by blue-native polyacrylamide gel electrophoresis (BN-PAGE) and negative-stain electron microscopy, after which the appropriate fractions were concentrated for cryo-EM using Amicon Ultra 0.5-ml 100-kDa spin concentrators. The initial Arctica sample was purified similarly, with the addition of 5% glycerol in the SEC purification step and the omission of ADP–AlF3 and MgCl2 from the purification protocol.


Samples consisting of either solubilized membranes or purified membrane complexes were mixed with 5% G-250 sample additive (Invitrogen), to a final concentration of about 0.2%, and run on 3–12% NativePage Bis-Tris Protein Gels (Invitrogen) according to manufacturer specifications. Gels were either stained with Coomasie R-250 or transferred to PVDF membranes and stained with appropriate antibodies (Supplementary Fig. 1). Antisera against EccB5, used in Extended Data Fig. 1b, was raised against the synthetic peptide CLPMDMSPAELVVPK and has previously been described16. Polyclonal rabbit antisera against the peptide was raised in rabbits by Innovagen using Stimune (Prionix) as adjuvants. The antibody was used as a 1:5,000 dilution. Blots were visualized on a ChemoStar TouchMotionCor2 using ChemoStarTS.

Negative-stain electron microscopy

Carbon-coated copper grids were glow-discharged for 30 s at 25 mA using a GloQube Plus Glow Discharge System (Electron Microscopy Sciences). Four microlitres of diluted sample was applied to the grids and incubated for 30 s. The sample was blotted off from the side and the grid was washed briefly with 4 μl of staining solution (2% uranyl acetate) and then stained with 4 μl of the staining solution for 30 s. The stain was blotted off from the side and grids were air-dried. Grids were imaged using a Thermo Fisher Scientific Talos L120C TEM equipped with a 4K Ceta CEMOS camera using TIA 4.1.5.

Cryo-EM sample preparation

For the main datasets, purified sample was applied to Quantifoil R2/2, 200 mesh, copper grids floated with an additional approximately 1.1-nm layer of amorphous carbon. Four microlitres of sample was applied onto glow-discharged grids (30 s at 25 mA) and allowed to disperse for 60 s at 4 °C and 100% humidity. Grids were blotted for 4–6 s with a blot force of −5 and plunge-frozen in a liquid propane–ethane mixture, using a Thermo Fisher Scientific Vitrobot Mark V. For the initial Arctica dataset, all steps were similar, with the exception of the additional layer of amorphous carbon.

Cryo-EM data acquisition

The initial cryo-EM dataset was collected on a 200-kV FEI Talos Arctica electron microscope equipped with a Falcon III direct electron detector running in counting mode and using Thermo Fisher Scientific EPU 1.11. A total of 853 movies were recorded with a nominal magnification of 150,000×, corresponding to a pixel size of 0.96 at the specimen level. Movies were recorded with a total dose of 40.28 electrons per A2, fractionated in 38 frames over a 40-s exposure time and with a nominal defocus range of 1–2.5 μm.

The two high-resolution datasets were recorded using Thermo Fisher Scientific EPU 2.4 software on a 300-kV Titan Krios TEM, equipped with a Gatan K3 direct electron detector running in counting mode and a Gatan Bioquantum energy filter (slit size 10 eV). We recorded 7,984 and 9,389 movies in counting mode in the two separate sessions with a nominal magnification of 81,000×, corresponding to a pixel size of 1.1 Å at the specimen level. Movies were recorded with a total dose of 59.5 electrons per A2, fractionated in 50 frames over a 3-s exposure time and with a nominal defocus range of 1–3 μm.

Cryo-EM data processing

Single-particle analysis was performed using Relion3.117, unless stated otherwise. For the initial Arctica dataset, movies were motion-corrected using MotionCor218 and dose-weighted, and the contrast-transfer function (CTF) was estimated using CTFFIND419. Automated particle-picking was performed using Cryolo20 and the pretrained Janni model. Following particle extraction and several rounds of 2D classification to remove obvious artefacts, an initial de novo model was generated. The dataset was further cleaned using 3D classification and the best class was subsequently used for reference-based particle-picking. Following 2D and 3D classification (and 3D refinement in C1), the map displayed an apparent threefold symmetry and was further refined in C3. This final map displayed an estimated 13.5 Å resolution.

In the first Krios dataset, movies were motion-corrected using MotionCor2, dose-weighted and the CTF was estimated using CTFFIND4. Automated particle-picking was performed using Cryolo with the pretrained Janni model and a low threshold. Particles were extracted and binned 4× and several rounds of 2D classification were performed followed by 3D classification with the 30 Å-filtered Arctica model as a template. The resulting particles were re-extracted without binning, CTF-corrected and polished and refined in C1, giving a map with an estimated overall resolution of 4.5 Å. For the cytosolic region, particles were recentred on the cytosolic region, re-extracted, CTF-corrected, polished and 3D-refined. Following refinement, the density accounting for individual cytosolic dimers was subtracted, resulting in a particle stack that was three times larger. Cytosolic dimers were first refined with a mask encompassing both cytosolic bridges. Subsequently, these were focus-refined using a soft mask around one of the cytosolic bridges. This map was refined using the default Relion value ‘--tau2fudge 2’ but also ‘--tau2fudge 4’, which increased the overall connectivity of the lower cytosolic area. The final map for the cytosolic bridge showed an estimated resolution of 3.3 Å and was sharpened using either Relion postprocessing or DeepEMhancer21. DeepEMhancer further helped to improve the observed anisotropy, overall map connectivity.

The second Krios dataset was processed similarly, with some exceptions. Following 3D classification of the binned data against the 4.5 Å Krios map filtered to 30 Å, the two maps with and without MycP5 were processed separately. The MycP5-unbound map displayed increased heterogeneity and—following unbinned re-extraction and refinement—the particles were 3D-classified without alignment, resulting in two obvious classes of 4.5 Å and 6.7 Å resolution. Model free density modification in Phenix.Resolve_Cryo_EM22 further improved the resolution to 4.3 Å and 5.8 Å, respectively. By contrast, a similar 3D classification for the MycP5-bound map did not further classify into structurally distinct classes. Following unbinned re-extraction and refinement, the MycP5-bound map showed an overall resolution of 4 Å, which was further improved to 3.8 Å after C3 refinement. Model free density modification in Phenix.Resolve_Cryo_EM further improved the resolution of the entire C1 map to 3.8 Å and of the C3 refined map to 3.56 Å. For the periplasmic map, the centre of mass for that region was determined using Chimera23 and the particles were recentred, extracted, 3D-refined and polished to obtain the periplasmic map at an estimated 3.8 Å resolution in C1. Following 3D classification without alignment and further refinement in C3, the estimated resolution of the periplasmic map improved to 3.5 Å. To separate the two states of EccC5, particles were recentred on the lower cytosolic region, at the level of the DUF domain, polished and 3D-refined. This was followed by a masked 3D classification in which the mask contained NBD1 and NBD2 of EccC5 in the extended conformation. The two main classes were further 3D-refined unmasked and subsequently masked, leading to a map of about 4.27 Å for the extended conformation and 7.6 Å for the contracted conformation.

Model building and refinement

Model building started by generating homology models for MycP5, EccB5, EccC5 and EccD5 with Phyre224. For MycP5, Protein Data Bank (PDB) entry 4J9412 served as a structural template, and PDB entries 4KK725, 4NH014 and 6SGW6, and 6SGZ6 served as reference models for EccB5, EccC5 and EccD5, respectively. To obtain atomic models of the periplasmic part (MycP5–EccB5) of the M. tuberculosis ESX-5 complex, homology models of MycP5 and EccB5 were rigid-body-fitted into a C1 symmetry, focus-refined periplasmic M. tuberculosis ESX-5 map (Electron Microscopy Data Bank (EMDB) code EMD-12518) using the fit-in-map tool in ChimeraX (v.1.0)26. Model building, extension and interactive refinement was performed with ISOLDE (v.1.0.1)27, a molecular-dynamics-guided structure refinement tool within ChimeraX (v.1.0). The resulting coordinate file (PDB 7NPS) was further refined with Phenix.real_space_refine (v.1.18.2-3874)28 using reference model restraints, strict rotamer matching and disabled grid search. Model validation was carried out using the MolProbity web server29 and EMRinger30 within the Phenix software package. Models for the membrane-embedded region (MycP5–EccB5–EccC5–EccD5) and cytoplasmic bridge (cytosolic domains of EccC5–EccD5) (PDB 7NPT) were built in the same way, using a reconstruction of the full M. tuberculosis ESX-5 complex (EMDB EMD-12517) and a focus-refined map of the cytoplasmic domains (EMDB EMD-12520) sharpened with DeepEMhancer, respectively. Finally, a composite model was assembled by fusing the periplasmic assembly and six copies of the cytosolic bridge to the membrane-embedded region model. This composite model was then refined against the full M. tuberculosis ESX-5 complex map with C1 symmetry (PDB 7NP7 and EMDB EMD-12514) and C3 symmetry (PDB 7NPR and EMDB EMD-12517).

Modelling into MycP5-free maps was performed with ISOLDE using the composite ESX-5 model, in which MycP5 and the periplasmic domain of EccB5 (residues 84–507) had been deleted. Adaptive distance restraints as well as torsion restraints were applied to all atoms to restrain short-range conformational changes but allow for long-range conformational movements. ISOLDE simulations for dynamic fitting of the coordinate file into EMD-12521 and EMD-12522 were performed (about 10 min, 10 K) after which the models showed satisfying fits to the new maps without further manual intervention. MycP5-free models were further refined against the maps using Phenix.real_space_refine (v.1.18.2-3874) as stated.

Visualization of atomic coordinates and map volumes was performed with ChimeraX (v.1.1) and PyMOL v.2.4031. Buried surface areas between subunits were calculated by PISA32.

Reporting Summary

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