Architecture of the ESX-3/Type VII secretion system

Abstract

Host infection by pathogenic Mycobacteria such as Mycobacterium tuberculosis is facilitated by virulence factors secreted by Type VII secretion systems. Here we report the cryo-electron microscopy structure of a membrane-embedded core complex of the ESX-3/Type VII secretion system from Mycobacterium smegmatis at 3.7 Å resolution, resolving the molecular architecture of a Type VII secretion machine and providing insights into the underlying secretion mechanism. The core of the ESX-3 secretion machine consists of four protein components, EccB3:EccC3:EccD3:EccE3 in a 1:1:2:1 stoichiometry, building two identical protomers. The EccC3 coupling protein, which interacts with the secreted substrates, links a flexible array of four ATPase domains to the membrane through a stalk domain. The “domain of unknown function” (DUF) adjacent to the stalk is identified as an ATPase domain essential for secretion. EccB3 is predominantly periplasmatic but a small segment crosses the membrane and contacts the stalk domain, suggesting that conformational changes triggered by substrate binding at the distal end of EccC3 and subsequent ATP hydrolysis in the DUF could be coupled to substrate secretion to the periplasm. Our results reveal that the architecture of Type VII secretion systems differs markedly from other known secretion machines.

Introduction

Pathogenic mycobacteria harbor up to five paralogous ESX/Type VII secretion systems (ESX, Early Secretory Antigen Target 6 System). Three of these systems, ESX-1, ESX-3 and ESX-5, mediate the secretion of specific sets of effector proteins that play defined roles in tuberculosis [1]. The ESX-3 secretion system is expressed in response to iron limiting conditions [2, 3] and it has been implicated in metal homeostasis [4–7], inhibition of T-helper cell (CD4+) activation, phagosome maturation and pathogen-induced phagosomal damage repair [8–10].

ESX secretion systems feature a set of five conserved core membrane components (EccB, EccC, EccD, EccE, MycP; Ecc, esx conserved component), which mediate the secretion of the EsxA:EsxB virulence factor family, the classical type VII substrate [11–16] or DNA [17]. Biochemical and structural studies on ESX-5 secretion systems demonstrated that four of these components (EccB, EccC, EccD, EccE) assemble into a stable hexameric secretion pore in the cell envelope while MycP, a membrane-anchored protease, is not tightly associated with the stable core [14, 18]. The coupling protein EccC recognizes effector proteins in the cytoplasm and energizes their transport [19].

Recent progress provided insights into the structural features of mycobacterial Type VII secretion systems. Crystal structures of soluble domains of three core membrane components (EccB, EccC, EccD) have been determined [19–21]. A low resolution negative-stain electron microscopy (EM) structure of ESX-5 revealed that the secretion system organizes into hexamers [18], but the use of staining agents limited the information to the external contour shape of the complex.

Here, we used single particle cryo-electron microscopy (cryo-EM) to reveal the atomic structure of the membrane embedded ESX-3/ Type VII core complex from M. smegmatis, a close ESX-3 homolog of M. tuberculosis (40.4-74.7% sequence identity), and show how the membrane components interact and derive mechanistic implications. We describe the organization of the Type VII coupling component EccC and discover a fourth ATPase domain as well as a stalk domain that brings EccC3 and EccB3 together close to the membrane.

Results

In order to gain insights into the molecular architecture of Type VII secretion systems, we co-expressed all 11 genes of the ESX-3 gene cluster of Mycobacterium smegmatis, encoding the secreted substrates and cytosolic components alongside the components for the membrane embedded secretion machinery in Mycobacterium smegmatis (plasmid pMyNT:ESX-3; Extended Data Fig. 1a). A ~900 kDa core complex was the most abundant and highest molecular weight species in detergent solubilized membrane extracts (Extended Data Fig. 1b). Subsequent purification revealed the ESX-3 core complex consisting of the membrane components EccB3, EccC3, EccD3 and EccE3 (Extended Data Fig. 1c).

A minimal expression construct (pMyNT:Mini) encoding the predicted membrane proteins enabled the purification of the ESX-3 core complex in high yields (Fig. 1a, b, right lanes; Extended Fig. 1a, d). For cryo-EM, the sample was exchanged into Amphipol A8-35 and the 900 kDa ESX-3 core complex was further purified by size exclusion chromatography (Fig. 1a, b, right lanes; Extended Fig. 1d, peak 2 (P2)). Higher oligomers of the ESX-3 core complex were observed after the exchange (Fig. 1a, b, left lanes; Extended Data Fig. 1d, peak 1 (P1)) and were heterogeneous in shape and size as shown by cryo-EM (Extended Data Fig. 1e), but a few 2D averages were compatible with a complex of 25 nm in diameter comparable to the ESX-5 hexamer [18] (Extended Data Fig. 1f).

Fig. 1. Purification and cryo-EM of the ESX-3 core complex.

a) Size-exclusion chromatogram after TeDM to Amphipol A8-35 exchange of the ESX-3 complex reveals two peaks, P1 and P2 (Extended Data Fig. 1d), which were analyzed by Blue Native (BN) PAGE. b) Peak 1 (P1) and peak 2 (P2) of the ESX-3 core complex expressed in M. smegmatis from the minimal expression construct pMyNT:Mini encoding only the membrane components of the system were analyzed by SDS-PAGE. c) Representative 2D class average of the ~900 kDa ESX-3 complex (peak 1). White arrows indicate flexible regions. The scale bar represents 10 nm. Top insert: 2D average of the fork obtained after subtraction from the images. d) Cryo-EM density map of the ESX-3 core complex built after merging the information of three processing strategies (Extended Data Fig. 2, Extended Data Fig. 3). In the model, EccB3 was placed in one orientation with respect to the membrane while maintaining the continuity with the EccB3 TMHs inserted into the membrane. The four core components are indicated in different colors: EccB3 (green), EccC3 (blue and light blue for flexible regions), EccD3 dimer (orange and red), EccE3 (purple). Scale bar, 2 nm. Left panel, side view. Middle panel, front view. Right panel, top view. EccB3 was removed for clarity. e) Schematic detailing the cross-linking results. Black bars indicate predicted trans-membrane helices. Abbreviations for EccC3: domain of unknown function (DUF), ATPase domains DI-DIII (DI-DIII); EccD3: Ubl domain (Ubl), membrane domain (MD); EccE3: domain structure unknown prior to this work. EccB3: domains labeled R1, R2, C, R3, R4.

The 2D class averages of the 900 kDa ESX-3 core complex reveal the distinctive shape of membrane-spanning proteins and two layers of density at each side of the membrane (Fig. 1c, Extended Data Fig. 1g). The smaller density with the shape of a fork appears blurred, and shows conformational flexibility with respect to the membrane (Extended Data Fig. 1h). Nonetheless, when the density of the fork is subtracted from the images, 2D averages indicate that this region is intrinsically rigid, but is flexibly attached to the membrane. Some heterogeneity seems to affect also the tip of the two arms on the other side of the membrane (Fig. 1c). This could not be resolved after image subtraction, thus indicating flexibility of this region.

The structure of the 900 kDa ESX-3 core complex was determined to an average resolution of 3.8 Å, following a complex classification/refinement strategy coupled to re-centering and particle subtraction of the flexible parts (Fig. 1d; Extended Data Fig. 1g, h, 2, 3, 4, 5) [22]. A merged reconstruction of the complex was built using one possible orientation for the fork (Extended Data Fig. 1h) and one of the conformations for the arms (Extended Data Fig. 3d). The composite map shows two similar protomers intimately interacting in a dimeric structure. Several transmembrane α-helices (TMHs) form a platform with a thickness that matches the inner mycobacterial membrane. A fork structure composed of two identical monomers is connected to the membrane. At the opposite end of the membrane, two arms, one from each protomer, interact with each other at their distal ends (Fig. 1d). Some flexibility between the two protomers was detected during image processing, and thus, we also resolved the structure of a single protomer (the one labeled as protomer 1) after particle subtraction, classification and refinement (Extended Data Fig. 2), which reached 3.7 Å resolution and improved the definition of several regions (Extended Data Fig. 3b, 4b).

Based on the cryo-EM density, which shows side-chain information in most regions (Extended Data Fig. 6), and the guide from cross-linking mass spectrometry (Fig. 1e), we modeled the structure for most of ESX-3 (Fig. 2). The cryo-EM map of the ESX-3 monomer was essential to build regions in EccC3 and EccE3, which were not well defined in the dimer (Extended Data Fig. 3, 4). Cross-linking mass spectrometry indicated the proximity of a small hydrophilic domain of EccD3 to EccC3 and EccE3 but no cross-links to EccB3 were found (Fig. 1e). Thus, we assigned the soluble domains of EccC3, EccD3 and EccE3 to the large density segment composing two cytoplasmic “arms” (Fig. 2a), while the soluble domains of EccB3 reside in the smaller segment forming the “fork” on the opposite side of the membrane layer (Fig. 2b).

Fig. 2. Overview of the atomic structure of the ESX-3 core complex.

a) The rigid core of the ESX-3 dimer excluding the flexible regions (fork and tips). The Ubiquitin-like domains (Ubl), membrane domain of EccD3 (MD1 and MD2), trans-membrane helix (TMH) for EccB3, EccC3, and EccE3, and the EccE3 C-terminal domain (CTD) are shown. b) Fit of a homology model of EccB3 into the 4.6 Å resolution cryo-EM map. Side view (top panel), top view (bottom panel). Right side, schematic of the EccB3 domain structure. c) Anchoring of EccB3 to the membrane of the ESX-3 dimer, showing the trans-membrane helix (TMH) and the N-terminal helix (NTH). d) Top view of protomer 1 membrane region with the assignment of all transmembrane helices. EccD3 dimer (orange and red); MD, membrane domain. The structure of protomer 1 was obtained after particle subtraction (Extended Data Fig. 2, 3), and the density for EccC3 TMH, shown in blue, is slightly better defined than in the dimer.

The crystal structures from either M. smegmatis (pdb ID 5CYU) or M. tuberculosis EccB1 (pdb ID 3X3M) are comprised of four domains (R1-R4) with a putative peptidoglycan binding fold (pdb ID 4F87) (labeled as domain C) [20, 23, 24]. Strikingly, two copies of an EccB3 homology model fit into the density of the fork (cross-correlation > 0.95) (Fig. 2b), supporting the previous proposal that the EccB core component appends Type VII secretion systems to the periplasmic peptidoglycan layer [20]. The EccB3 dimer exhibits two-fold symmetry, in which the dimer interface is formed by domains R1, R2 and C. Density is observed connecting the periplasmic domains to their respective transmembrane helices that anchor each of the two EccB3 monomers in the membrane layer (Fig. 2c). Interestingly, we find that each TMH is connected to a short N-terminal helix (NTH) that runs almost parallel to the membrane and makes strong electrostatic interactions with the Type VII coupling protein EccC3 (Fig. 2c).

The membrane-spanning region was resolved at 3.7 Å resolution (Extended Data Fig. 4a, b) and was modeled de novo including EccD3 and EccE3. Each protomer contains 25 trans-membrane helices corresponding to two copies of the EccD3 transmembrane domain (2 x 11 TMH) for each copy of EccB3 (1 TMH) and EccE3 (2 TMH) (Fig. 2d). The membrane domains of the two EccD3 monomers interact at one end in an antiparallel orientation, forming a hydrophobic crevice on the periplasmic side, which appears filled with lipids or amphipols. Homology models of the EccD1 Ubl domain from M. tuberculosis (pdb ID 4KV2) were fitted into the EM density and used as template for modeling. The EccD3 dimer provides the scaffold for the ESX-3 core complex. In the membrane, it is TMH 11 of the two respective EccD3 monomers that interacts with either EccB3 or EccE3 at the opposite ends (Fig. 2d). The EccD3 membrane domains are connected to their respective N-terminal Ubl domains (Fig. 3a) through TMH 1. In the cytoplasm, the Ubl domains and their linkers with the membrane domains of both EccD3 monomers form a scaffold for several cytoplasmic domains of EccB3, EccC3 and EccE3 (Fig. 3a).

Fig. 3. Structure and function of the ESX-3/Type VII secretion system.

a) Side view of the protomer 1 membrane region. The Ubl domains, membrane domains 1 and 2 of EccD3 (MD1 and MD2), trans-membrane helices (TMH) for EccB3, EccC3, and EccE3, and the EccE3 C-terminal domain (CTD) are shown. Same color code as in Fig. 2. b) Domain structure of EccC3 with the stalk domain and interaction with two EccD3 Ubl domains. c) The plasmid-encoded ESX-3 gene cluster (pMyNT:ESX-3) mediates secretion of the substrate heterodimer EsxG:EsxH into the culture filtrate of the transformed knockout strain M. smegmatis ΔESX-3. Whole cell lysate (WCL) and culture filtrate (CF) of MsmegΔESX-3 transformants were analyzed containing the following plasmids. Lane 1: pMyNT:no insert; lane 2: pMyNT:ESX-3; lane 3: pMyNT:ESX-3ΔeccE3; lane 4: pMyNT:ESX-3Δperiplasmic domain(eccB3), lane 5: pMyNT:ESX-3ΔDIII(eccC3); lane 6: pMyNT:ESX-3 (eccC3 D319), lane 7: pMyNT:ESX-3 (eccC3 D320), lane 8: pMyNT:ESX-3 (eccC3 R342T); lane 9: purified EsxG:EsxH. Loading and lysis control: αRNAP, RNA polymerase ß subunit (~150 kDa). The right panel shows a close-up of the structure of the DUF domain, highlighting residues mutated in the secretion assay.

A model of EccC3 was built into the density showing the domain organization of a Type VII coupling protein and the discovery of two unknown domains. Two N-terminal TMHs insert EccC3 into the membrane, for which we detect partial density and whose position suggests that they flank the membrane pore (Fig. 2a, d, 3a, 3b, Extended Data Fig. 6b). We found that a previously unassigned stalk domain at the N-terminus connects the EccC3 coupling component to the membrane. The stalk comprises two α-helices and a short antiparallel ß-sheet and contacts the EccB3 N-terminal helix (Fig. 3a, b). This interaction suggests that conformational changes of the stalk could be coupled to changes in EccB3. The stalk domain is followed by a conserved domain, which is referred to as “domain of unknown function” (DUF) in the literature [19]. The DUF structure exhibits an ATPase fold, which is in an ATP free state (Fig. 3a, b, Extended Data Fig. 6a).

The flexible tips in ESX-3 could only be visualized in reconstructions obtained after re-centering the particles (Extended Data Fig. 2b, 3, 5). This strategy was sufficient to fit a homology model of the EccC3-DI domain within the density immediately in contact with the DUF domain with great confidence (cross-correlation coefficient > 0.95) (Fig. 3b). Classification of the particles revealed that DII and DIII are extremely flexible and only partial density for the DII domain was detected (Extended Data Fig. 2b, 3d, 5). ATP binding to EccC3-DI but not DII and DIII is essential for secretion [19], whereas peptides representing the Type VII secretion signal bind to EccC3-DIII [19].

Crosslinking and mass spectrometry indicated that the soluble domains of EccC3, EccD3 and EccE3 are closely located to each other, and thus, the upper cytoplasmic arm was assigned to component EccE3 given the localization proximal to the two ubiquitin like domains (Ubl) of EccD3 and the cross-links to these domains (Fig. 2a, d, 3a). EccE3 is appended to the platform by two TMHs and forms an extensive interface with the EccD3 Ubl domain dimer in the cytoplasm (Fig. 2d, 3a). The core of the EccE3 cytoplasmic domain shows a mixed α/β-fold with a central antiparallel β-sheet surrounded by several α-helices (Fig. 3a).

Overall, the structure of the ESX-3 core complex forms a dimer that comprises two identical protomers each consisting of one copy of EccB3, EccC3 and EccE3 and two copies of EccD3. The calculated molecular weight of the complex is 651 kDa without the amphipol micelle. The two protomers converge at a hinge region, in which EccB3, EccC3 and EccD3 engage in intricate contacts to form the dimer. The EccB3 dimer is crucial to hold both protomers together as each periplasmic domain is linked to a TMH that is firmly anchored within the two respective protomers (Fig. 2c). In agreement, EccB3 is found in dimeric ESX-3 complexes but not in the monomeric species detected after image classification (Extended Data Fig. 2a).

The architecture of the ESX-3 core complex differs from other known secretion machines and we therefore performed several experiments to validate our structure. We first confirmed that the purified membrane embedded secretion machine is competent for secretion (Fig. 3c) and established a secretion assay to monitor ESX-3 dependent secretion into the culture medium using polyclonal anti serum against the two effector proteins, EsxG and EsxH. For our analysis, we used a M. smegmatis ESX-3 knock out strain [4] that is defective in secretion [4] (Fig 3c, lane 1). Complementation of this strain with plasmid pMyNT:ESX-3, that encodes the ESX-3 gene cluster (Extended Data Fig. 7a) restored secretion of EsxG and EsxH (Fig. 3c, lane 2). Analysis of detergent solubilized cell envelopes of the transformed knock out strain by BN-PAGE and Western Blot and subsequent purification confirmed that the 900 kDa ESX-3 core complex was assembled (Extended Data Fig. 7b), which permitted us to carry out further experiments to validate that the ESX-3 complex found in vivo is identical to the one from which the structure was derived (Extended Data Fig. 7c-h).

The periplasmic domain of EccB3 does not interact with the other components in the ESX-3 core complex and is well separated by the membrane. To study the function of the periplasmic domain, we removed most of it from EccB3 (pMyNT:ESX-3) and expressed it in secretion deficient M. smegmatis ΔESX-3. Purification of the EccB3Δfork:EccC3:EccD3:EccE3 complex showed a shift in molecular weight towards a single protomer (without fork) in size exclusion chromatography (Extended Data Fig. 7h). This was expected since a large part of the ESX-3 dimer interface is composed of the periplasmic EccB3. Despite the presence of some complexes with a molecular weight corresponding to dimers (without fork), the same plasmid did not mediate secretion of EsxG and EsxH, thus indicating that the fork structure is essential for secretion (Fig. 3c, lane 5).

To investigate the function of EccE3, we removed EccE3 from the plasmid encoded ESX-3 gene cluster (pMyNT:ESX-3) and expressed the remaining genes in M. smegmatisΔESX-3. Purification of the EccB3:EccC3:EccD3 subcomplex showed the assembly of a dimer, which was, however, prone to dissociation into monomers, confirming that EccE3 is a scaffold component of the ESX-3 protomer (Extended Data Fig 7h). This plasmid did not mediate secretion of EsxG and EsxH into the culture medium showing that EccE3 is an essential component of the ESX-3 secretion system (Fig. 3c, lane 3). We conclude, that the correct assembly and stability of the ESX-3 protomers is critical for the functioning of the ESX-3 secretion system.

Utilizing the established secretion assay, we investigated the function of EccC3. We first generated a mutant of EccC3 lacking the DIII domain. Deletion of the DIII domain completely abolished secretion (Fig. 3c, lane 4) in agreement with observations in the staphylococcal Type VII secretion system [25]. The DUF domain contains a canonical and conserved Walker B motif whereas the Walker A motif (336-G XXXX H R TT-344) includes non-canonical residues in positions 341 (H for G), 342 (R for K) and 344 (T for S). We performed an alanine scan to examine the effect of mutations in the Walker A and B motifs in the DUF domain (Fig. 3a). D319A and D320A mutations abolished secretion, confirming that the DUF is an ATPase domain with an essential function in secretion (Fig. 3c, lane 6, 7), whereas the R342T mutation in the non-canonical Walker A motif had no effect (Fig. 3c, lane 8). Thus, the domain architecture of Type VII coupling proteins comprises a stalk domain that connects a linear array of four ATPase domains (DUF, DI, DII, DIII) (Fig. 3b) to the membrane. Three out of the four ATPase domains are identified as the main constituents of the arms in the EM density (Fig 2a, 3a, b) while density for DIII is missing in the reconstruction due to high flexibility.

Unlike the hexameric ESX-5 core complex, we extracted the ESX-3 core complex in a dimeric state. The ESX-3 core complex is a stable dimer in which two identical protomers are held tightly together due to an intricate network of interactions that could be extended to complete the ESX-3 secretion pore. Three dimers are required to complete a full circle of approximately 250 Å in diameter whereas two dimers would leave an open ring, if the approximate angle between monomers were maintained. Hence, a hexameric arrangement agrees with the orientation between monomers in our structure, with some of the larger oligomers in the void of our purification (Extended Data Fig. 1e, f), and with the hexamers of the homologous ESX-5 [18].

Since we did not detect higher ESX-3 oligomers in membranes of secretion competent mycobacteria expressing the ESX-3 gene cluster (Extended data Fig. 7a, b), we tested, whether a stronger overexpression of the ESX-3 core complex using our “high yield” minimal expression construct (pMyNT:Mini) would enable detection. Blue-Native PAGE of detergent solubilized membrane extracts revealed species of higher molecular weight than the dimers (Extended Data Fig. 7i). When a crosslinking agent is used prior to detergent, it was possible to extract increased amounts of the larger oligomers, suggesting that ESX-3 forms larger complexes in the membranes that are disrupted during our extraction protocol. As a control, large oligomers are not found in the membranes of mycobacteria expressing only EccD3. Based on these results and the close relation between ESX-3 and ESX-5 systems, we conclude that ESX-3 dimers associate in the membrane to form higher oligomers.

Using the low-resolution map of the homologous ESX-5 as a template, three ESX-3 dimers can be fitted by following the contour shape provided by ESX-5 to model the ESX-3 hexamer (Extended Data Fig. 8). In our model, six protomers form a central, putative protein-conducting channel (Fig. 4a), which is restricted by the TMH2 of EccC3 to a diameter of around 25 Å in the membrane and closed by the EccB3 domain C at the periplasmic exit (Fig. 4a, b).

Fig. 4. Model of the oligomeric ESX-3 secretion machine.

a) Side and top views of the model for the ESX-3 oligomer. A vestibule-like structure is formed by the five EccC3 domains: Stalk, DUF, DI, DII and DIII. b) Inner membrane pore and vestibule structure of the ESX-3 oligomer. The inner membrane pore is composed of EccB3 and EccC3 TMHs and the stalk domains of EccC3. c) Model for the mechanism of secretion in ESX-3 T7SS. The membrane pore is formed by EccC3 and contacts between EccC3 and EccB3 facilitate the coupling of ATP hydrolysis and substrate binding to changes in conformation in both proteins and the opening of the pore.

The cryo-EM structure of the ESX-3 core complex revealed a distinct 1:1:2:1 stoichiometry compared to the 1:1:1:1 ratio proposed for the ESX-5 core complex, which was based on Intensity Based Absolute Quantification Mass Spectrometry (iBAQ MS) measurements [18]. However, this method is not sufficiently sensitive to define the stoichiometry of ESX-3 (Extended Data Fig. 7f). Furthermore, EccD3 elutes as a dimer of ~100 kDa (2 x 48 kDa) in size-exclusion chromatography combined with multi-angle light scattering (SEC-MALS) (Extended Data Fig. 7g) and, when superimposed, the volume of the EccD3 dimer fits perfectly into the corresponding volume of the ESX-5 core complex protomer while the EccD3 monomer would leave a gap (Extended Data Fig. 8). Noteworthy, the ratio of subunits in ESX-3 extracted from secretion competent cells estimated by Oriole staining (Extended Data Fig. 7e) is similar to the ESX-3 sample whose structure has been solved (Fig. 1a). Quantification of the bands from triplicate experiments agrees with the 1:1:2:1 stoichiometry in ESX-3 from secreting cells and the one produced in higher yields for the cryo-EM experiments (Extended Data Fig. 7e).

A mechanistic model for secretion can be proposed with the available information (Fig. 4c). The interior of the membrane pore is formed by two transmembrane helices of EccC3, for which we observe partial density in our reconstruction (Extended Data Fig. 6b). The secreted substrate heterodimer, EsxG:EsxH of the ESX-3 system as well as the known substrate structures of other ESX systems assumes a diameter of 22 Å. Hence, the secretion pore in our model is sufficient to permit the translocation of substrates in their folded state.

On the cytoplasmic side of the proposed ESX-3 secretion pore, six copies of the coupling component EccC3 pack into a unique vestibule-like structure consisting of a stalk domain and four ATPase domains of EccC3 (DUF, DI, DII, DIII) (Fig. 4c). The architecture of the DUF and the stalk domain linked to a membrane traversing helix connected to a periplasmic domain is reminiscent of ABC F transporters (Extended Data Fig. 9). DI, DII and DIII domains are very flexible, and this flexibility increases towards the outermost layer of the vestibule, composed of six ATPase domains (DIII domain). DIII domains were shown to have an additional function in substrate recognition [19] and thus the DIII layer provides six potential binding sites for the secreted substrates. The EccC3 stalk domains interact with the TMHs of EccB3 in our structure and the extensive contacts between the stalk and the two N-terminal helices in EccB3 suggests that conformational changes triggered by substrate binding to the DIII domain and subsequent ATP hydrolysis of the DUF can directly be transmitted to open or close the membrane pore as well as the periplasmic exit formed by EccB3 (Fig. 4c).

In summary, the ESX-3 core complex structure provides the first detailed architecture and topology for each of the Ecc components of a T7SS (Extended Data Fig. 10). Our data indicate that the Type VII secretion machine is built upon oligomerization of a building block made up of two tightly interconnected protomers forming a secretion pore sufficient in size to allow translocation of substrate heterodimers in their folded state. While EccD3 provides the central scaffold, EccB3 and EccE3 are essential assembly factors. Remarkably, the ATPase EccC3, which couples effector protein recognition and translocation [19], is found in a flexible conformation that becomes ordered closer to the membrane, and we identify the DUF domain as a new ATPase domain essential for secretion. Our model suggests that EccC3 forms the inner membrane pore and a cytoplasmic vestibule-like structure containing several ATPase domains, which establishes electrostatic contacts with EccB3 at the membrane that could provide the mechanistic basis for coupling substrate recognition and ATPase activity of EccC3 with the conformation of EccB3 and translocation. ESX-3/Type VII secretion systems are essential for pathogen growth and have been implicated in impeding key functions of the cellular immune system. This work provides a structural understanding of ESX-3 that can be used as a basis to design new anti-microbial strategies.

Methods

Molecular Biology

For generation of the ESX-3 expression plasmids used in this study (Extended Data Table 1) fragments of the ESX-3 gene cluster of M. smegmatis mc² 155 (eccA3 through eccE3) were amplified by PCR (CloneAmp HiFi PCR premix, Takara), combined and cloned into the vector pMyNT using restriction-free ligation (In-Fusion® HD cloning kit, Takara). The following In-Fusion® primers were used for amplification: (i) X390/X391 (eccA3- eccC3), (ii) - X392/ X391 (eccB3- eccC3), (iii) - X396/X394 (pe- eccE), (iv) - X393/ X394 (espG-eccE), (v) - X395/X394 (eccD3- eccE3), (vi) - X390/NX391 (eccA3-eccC3ΔD3). A C-terminal His-Tag (-GSMGGSHHHHHH*) was introduced on eccC3 using oligos X388/X389 (vii). The backbone of the pMyNT vector was amplified using the primer pairs X386/X387 (viii) or M37/M38 (ix). In order to clone the ESX-3 gene cluster carrying a His-Tag on eccC3 fragments (i) and (vii) were inserted into the pMyNT backbone (viii) first. The resulting construct was then linearized using the restriction enzyme SnaBI followed by insertion of fragment (iii) to produce plasmid “pMyNT:ESX-3”. For the generation of plasmid pMyNT:ESX3i, we excluded the acetamidase promoter region of the pMyNT:ESX3 using primers X430/X426 and replaced it with the IdeR promoter amplified with primers X427/X428.

To generate construct “pMyNT:ESX-3ΔS” lacking all substrate genes of the ESX-3 gene cluster, fragments (i) and (vii) were cloned into the pMyNT backbone (viii). The resulting construct was digested with SnaBI followed by ligation of fragment (iv). Plasmid “pMyNT:ESX-3ΔSC” was produced by insertion of fragments (ii) and (vii) into the pMyNT backbone (viii). After SnaBI digestion, fragment (v) was cloned into the construct.

To generate plasmid “pMyNT:Mini”, the MCS of the pMyNT vector, downstream of the acetamidase RBS, was replaced with the oligonucleotide pair X397/ X398.

The genes of eccB3, mycP3 and eccE3 were amplified from the genomic DNA of M. smegmatis mc² 155 using In-Fusion® primers X266/X267, X307/ X308 and X275/ X276. Gene eccC3 encoding a 3’-terminal His-tag (-GSMGGSHHHHHH*) was amplified using In-Fusion® primers X311/X312 and eccD3 encoding an N-terminal Strep-tag II (MASWSHPQFEKGS-) was amplified using In-Fusion® primers X274/X273. Genes were cloned via the following restriction sites: EcoRV for eccB3, HindIII for eccC3, SnaBI for eccD3 (N-terminal Strep-tagII), ScaI for mycP3 and HpaI for eccE3.

To generate plasmid pMyNT:StrepII-EccD3 the multi-cloning site of the pMyNT vector was removed by inverse PCR using primers X42/X43. Oligonucleotides (X44/X44C) encoding the Strep-tag II and the BamHI restriction site were annealed and then inserted into the linearized vector by ligation. Gene eccD3 was amplified with In-Fusion® primers X96/X97 and cloned into the pMyNT plasmid via the BamHI restriction site.

To generate construct “pMyNT:ESX-3-EccC3ΔDIII” lacking the DIII domain of EccC3, fragments (vi) and (vii) were cloned into the pMyNT backbone (viii). The resulting construct was digested with SnaBI followed by ligation of fragment (iii).

The plasmid EccB3Δfork was produced by amplifying the following fragments using “pMyNT:ESX-3" as template: NX1/NX2 (pMyNT-eccA3-eccB3Δfork), NX3/NX4 (eccB3Δfork-eccE3), NX5/NX6 (eccE3-pMyNT).

For generation of plasmids “pMyNT:ESX-3-EccC3-D319A”, “pMyNT:ESX-3-EccC3-D320A” and “pMyNT:ESX-3-EccC3-R342T” containing point mutations in the Walker A or B motifs of the domain of unknown function (DUF) of EccC3, the vector pMyNT was linearized by PCR first using primer pair M37/M38 (ix). Fragments (x) (eccA3-eccC3) and (xi) (eccC3-eccE3) were amplified from “pMyNT:ESX-3” using In-Fusion® primers M39/M40, or M47/M48, respectively. Oligonucleotides encoding parts of the eccC3 gene comprising point mutations D319A (M41/M42), D320A (M43/M44), or R342T (M45/M46) were annealed and ligated together with fragment (x) into the pMyNT backbone (ix), respectively. The resulting constructs were digested with SnaBI followed by ligation of fragment (xi).

Protein overexpression and purification of the EccB3:EccC3:EccD3:EccE3 dimer and the ESX-3 higher oligomer complex

M. smegmatis mc² 155 was transformed with the respective plasmid and cultured in LB (Luria/Miller) medium supplemented with 0.05 % (v/v) Tween 80 and 0.2 % (v/v) glycerol at 150 rpm. Induction was performed with 0.2 % (w/v) acetamide (or 200 μM 2,2’-dipyridyl in case of the pMyNT:ESX-3i) at OD_600nm 0.5 and cells were grown until a final optical density of 1.4-1.6. Cells were pelleted by centrifugation and washed in 1x PBS buffer. The cell pellet was resuspended in buffer A (30 mM Hepes pH 8.0, 300 mM NaCl and 10% glycerol) supplemented with EDTA-free protease inhibitors (Roche). Cells were lysed by three passages through an Emulsiflex-C3 homogenizer (Avestin). Unlysed cells were removed by centrifugation (10 min, 10 000 x g). The membrane fraction was separated by ultracentrifugation (1 h, 100 000 x g). Membranes were solubilized in buffer A supplemented with 0.5 % TeDM (n-Tetradecyl-β-D-maltopyranoside, Anatrace) at 4 °C for 1 h. Insoluble material was removed by ultracentrifugation (1 h, 100 000 x g).

The sample was supplemented with 30 mM imidazole and loaded on a HisTrap column (GE Healthcare). For wash and elution, buffer A supplemented with 0.0022 % TeDM was used with 30 mM and 250 mM imidazole, respectively. TeDM was exchanged for amphipol A8-35 at a 1:3 protein to amphipol ratio. TeDM was removed by incubation with BioBeads (Biorad) overnight. In order to separate the different oligomeric states of the complex size exclusion chromatography was performed using a Superose 6 Increase 10/300GL column (GE Healthcare). Fractions were analyzed by SDS-PAGE and BN-PAGE followed by Colloidal Blue staining. To confirm the identity of the purified proteins in the resolved complexes and to confirm the integrity of the sample, protein bands were cut out of the BN-PAGE and assessed by Intensity Based Quantification Mass Spectrometry (IBAQ) prior to grid preparations (not shown).

Protein expression and purification of the EccD3 dimer

M. smegmatis mc² 155 was used to produce higher yields of the complex, required for structural studies. These cells were transformed with the pMyNT:Strep-EccD3 and cultured in LB medium supplemented with 0.05 % (v/v) Tween 80 and 0.2 % (v/v) glycerol at 150 rpm. Induction was performed with 0.2 % (w/v) acetamide at an OD_600nm of 0.5 and cells were grown until a final optical density of 1.4-1.6 was reached. Cells were pelleted by centrifugation and washed in 1x PBS buffer. The cell pellet was resuspended in buffer A (50 mM Tris pH 8.0, 300 mM NaCl). Cells were lysed by three passages through an Emulsiflex-C3 homogenizer (Avestin). Unlysed cells were removed by centrifugation (10 min, 10 000 x g). The membrane fraction was separated by ultracentrifugation (1 h, 100 000 x g). Membranes were solubilized in buffer A supplemented with 1% DDM (n-Dodecyl-β-D-maltopyranoside, Anatrace) at 4 °C for 1 h. Insoluble material was removed by ultracentrifugation (1 h, 100 000 x g).

The sample was loaded on a StrepTrap column (GE Healthcare). For wash and elution, buffer A was supplemented with 0.05 % DDM and with 0.05 % DDM and 2.5 mM D-Desthiobiotin respectively. Eluted protein was subjected to size exclusion chromatography using a Superose 6 Increase 10/300GL column (GE Healthcare).

Generation of polyclonal anti-serum against the substrates EsxG and EsxH

The EsxG:EsxH protein complex of M. smegmatis was expressed from plasmid pMAPLe3, which was kindly provided by the Eisenberg lab (University of California, Los Angeles), and purified to homogeneity as described in Arbing et al [26]. Purified EsxG:EsxH was used to immunize rabbits (Immunoglobe, Himmelstadt, Germany). The polyclonal anti-serum was subjected to affinity purification using a column coupled to the purified antigen.

Protein secretion assay

M. smegmatis strains were grown in LB-Tween medium at 37 °C to an OD_600nm of 0.8. For whole cell lysate analyses, cells were pelleted by centrifugation (5 000 x g, 10 min), washed with PBS buffer and disrupted by bead-beating with 0.1 mm glass beads (Sigma). Samples were supplemented with Novex® Tricine SDS Sample Buffer (Invitrogen), boiled for 2 min and then separated on a 10-20% Novex® Tricine gel for Western Blot analysis. Culture supernatants were centrifuged twice (5 000 x g, 4 °C, 10 min) and passed through a 0.2 μm filter to remove remaining unlysed cells. Proteins were precipitated by addition of 10% trichloroacetic acid (overnight, 4 °C) followed by centrifugation (10 000 x g, 4 °C, 15 min). The precipitate was washed with ice-cold acetone, resuspended in Novex® Tricine SDS Sample Buffer, boiled for 2 min and then separated on a 10-20% Novex® Tricine gel. The amount of whole cell lysates loaded correspond to 0.26 OD_600nm units of cells while culture supernatants loaded correspond to 7.5 x 0.26 OD_600nm units of cells. Subsequent Western Blots were carried out according to the manufacturer’s protocol (Invitrogen). PVDF membranes were either incubated using polyclonal rabbit serum against the EsxG-EsxH protein complex (~4.5 μg/ml) or a monoclonal mouse antibody against the RNA polymerase β subunit (1:2 000; BioLegend). Secondary Horse Radish Peroxidase (HRP) conjugated antibodies were used in the following concentrations: anti-rabbit IgG-HRP (1:10 000; Carl Roth), and anti-mouse IgG-HRP (1:10 000; Carl Roth). Proteins were visualized using the Enhanced Chemiluminescence substrate kit (Pierce) and a FujiFilm LAS-3000 luminescent image analyzer. Band intensities were quantified using the ImageJ software [27].

Cryo-EM sample preparation

The sample was concentrated to 0.3 mg/ml and a volume of 3.5 µl was applied on glow discharged holey carbon grids (Quantifoil Cu/Rh R1.2/1.3 400 mesh). Excess liquid was removed by blotting for 3 s (blot force -10) using filter paper followed by plunge freezing in liquid ethane using a FEI Vitrobot Mark IV at 100% humidity and 4 °C.

Cryo-EM data acquisition and image processing of larger complexes

For the analysis of the large complexes eluting in the void volume of the size-exclusion chromatography (Extended Data Fig. 1), images were collected on a 200 kV FEI Talos Arctica electron microscope equipped with a Falcon III direct electron detector at the Spanish National Centre for Biotechnology (CNB-CSIC, Madrid). Images were recorded using the integrative mode and a nominal magnification of 75 000x, corresponding to 1.42 Å/px at the specimen level. A total dose of 50 electrons per Ų was used, fractionated in 50 frames over a 2.98 s exposure. Images were collected with a nominal defocus range of 1.4-3.2 µm.

3 387 movie stacks were corrected for drift (5 x 5 patches) and dose-weighted using MotionCor2 [28]. The contrast transfer function (CTF) parameters were determined for the drift-corrected micrographs using Gctf [29]. A first set of 2D-references were generated from manually picked particles in RELION [30] and these were then used for subsequent automatic particle picking using Gautomatch (provided by Kai Zhang, Faculty of Arts and Sciences, Yale University). 241 990 automatically selected particles were extracted and thoroughly analyzed using 2D-classification routines available in RELION and cryoSPARC [31]. A subset of 36 055 particles corresponding to side views of the multimeric complex was aligned using a 2D-reference and subsequently classified in 2D using a mask to focus classification on the density outside the membrane region. The dimensions of a representative average with density at both sides of the membrane region were measured.

Cryo-EM data acquisition of dimeric complexes

An Initial data set of the purified ESX-3 complex was collected at The Netherlands Center for Electron Nanoscopy (NeCEN). Processing of these data reached 7 Å resolution for the core of the dimer (Extended Data Fig. 2a). For the high-resolution structure, micrographs were collected at the Rudolf Virchow Center, Universität Würzburg (Germany) using a 300 kV FEI Titan Krios electron microscope (Extended Data Fig. 2b). Micrographs were recorded using a Falcon III direct electron detector in counting mode (Extended Data Table 2).

11 903 movies were with an underfocus of 1.6 to 2.6 μm, a calibrated magnification of 1.0635 Å/px, 1 e^-/Ų/frame and a total dose of 50 e^-/Ų fractionated in 55 frames.

High-resolution cryo-EM data processing

The dataset was processed making use of tools available in RELION versions 2.1 and 3.0 [30], cryoSPARC [31], cisTEM [32] and SCIPION [33]. Initially, movies coming from the different sessions were aligned and local motion was corrected using MotionCor2 with dose weighting. Micrographs exhibiting defects in the Thon rings due to excessive drift, ice contamination or astigmatism were discarded. Details can be found in Extended Data Table 3.

206 6007 particles were automatically selected and submitted to several rounds of 2D-classification in order to discard bad particles using RELION (Extended Data Fig. 2b). After cleaning, 325 205 particles were used in all subsequent analyses. Next, an initial 3D-model was generated from the selected particles using routines available in RELION and this volume was used to further classify each of the datasets in 3D. After classification 165 412 particles were refined using RELION to generate a cryo-EM map with an estimated average resolution of 3.8 Å, using gold-standard refinement methods and the Fourier shell correlation (FSC) cutoff of 0.143 (Extended Data Fig. 4). Local resolution ranges were analyzed within RELION. Refinement of one of the monomers after particle subtraction to remove the influence of the other monomer during refinement yielded 3.7 Å and further improved the resolution of some regions.

The cryo-EM map of the ESX-3 complex exhibited a lack of details at both ends of the complex (tip of the arms and the fork), in part due to flexibility but possibly also due to the increasing distance from the center of the box. It has been observed before that the accuracy of the alignments improves at the center of rotation compared to more distal regions, especially in flexible and large multi-component macromolecular complexes. We applied a re-centering strategy in which the group of 325 205 particles previously selected was re-extracted and centered in either of the two distal regions of the map (Extended Data Fig. 2b), as detailed below. To improve the density at the distal arm region, the selected particles were first re-extracted with their center at this region. Particles were next classified using a mask that only accounted for the variability of the density corresponding to the most distal region of the molecule, and this classification was performed without further alignment of the particles. These two classes included 80 173 and 69 919 particles each, and they were further processed using gold-standard refinement methods in RELION to generate cryo-EM maps at 5.4 Å and 5.3 Å resolution (gold-standard FSC= 0.143 criterion) (Extended Data Fig. 5).

Similarly, particles were subsequently re-extracted and centered in the fork region (Extended Data Fig. 2b). Particles were first classified using a circular mask that comprised the fork region and permitting a large angular search. The 126 308 particles assigned to one of the two resulting sub-groups were then subtracted in order to remove the density that did not correspond to the fork. These particles were analyzed with cryoSPARC v2 and three different models were generated using the ab-initio routine available in this package, and without assuming any symmetry. One of the maps, corresponding to 55 342 particles, exhibited features for the fork already observed in the 2D averages and preliminary processing of the data, and it was selected for further refinement using RELION and applying C2 symmetry. The map showed an estimated average resolution of 4.6 Å (Gold-standard FSC=0.143 criterion) (Extended Data Fig. 4). Differences in resolution across the reconstructed map were analyzed using the local resolution routine available in RELION.

Model building and refinement

The programs RaptorX [34], I-Tasser [35] or PHYRE2 [36] were used to generate homology models of the periplasmic domain of EccB3 (aa 92-465), the Domain of Unknown Function (aa 201-400), the two ATPase domains DI-DII (aa 401-1052), Ubl domain of EccD3 (aa 6-97) based on crystal structures of the periplasmic domain of EccB1 from M. tuberculosis (pdb ID 3X3M), the Domain of Unknown Function (pdb ID 2IUUA), the Ubl domain of EccD1 from M. tuberculosis (pdb ID 4KV2) and the three ATPase domains of EccC from T. curvata (pdb ID 4N0H) as templates, respectively. As a first step rigid body fitting of these models into the cryo EM density was performed using CHIMERA [37]. To improve the fit of the DUF and the Ubl domains, five cycles of morphing were carried out using PHENIX Refine v1.12 [38] before parts of the structures were rebuilt using COOT [39]. De novo models of the cytoplasmic segment and the transmembrane region of EccB3 (aa 28-91), the N-terminal stalk domain of EccC3 (aa 4-56 and 88-210), the transmembrane domain of EccD3 (90-468) and EccE3 (aa 2-285) were built guided by Quick2D [40] and Haruspex (https://www.biorxiv.org/content/10.1101/644476v1). EccC3-DI is located at the flexible arms and the resolution in one of the structures obtained after re-centering of the particles indicated clear secondary structure elements and was sufficient to fit a homology model with high cross-correlation (> 0.95).

For EccC3 domain DII density is present only in protomer 1 without secondary structure information. A homology model of DII was fitted into this density as a rigid body using CHIMERA. Density for DIII (aa 1052-1325) was missing in both protomers. EccC3 contains 2 TMHs for which partial density was detected in the map of the ESX-3 monomer. For TMH1, amino acids 36-45 of the helix were built, whereas no density was present for amino acids 46-62. A poly-alanine model of TMH2 (aa 72-90) was placed into the density and rigid body fitting and morphing were carried out using Phenix Refine v1.12 (Extended Data Fig. 6b).

Model building and structure refinement were performed iteratively and using the information from the map of the dimer and the monomer. Five macro cycles of real-space and grouped B factor refinement of main and side chains of the entire model of the ESX-3 core complex were performed in PHENIX Refine v1.12 at 3.7 Å resolution using restraints for secondary structure, rotamers as well as Ramachandran angles. Statistics for the model can be found in the Extended Data Table 4.

To model the ESX-3 oligomer, a composite EM map of the EccB3:EccC3:EccD3:EccE3 complex was chosen, in which the EccB3 dimer was rotated by 10° along the 2 fold symmetry axis with respect to the membrane layer permitting the fit of all three copies of this composite map into the ESX-5 EM map without clashes (EMDB-3596). To model the ATPase domains DII and DIII, the crystal structure of DI-DIII (pdb ID 4N0H) was superimposed onto DI in the ESX-3 structure placing DII into the density adjacent to DI.

Cross-linking mass spectrometry (XL-MS) of ESX-3

The purified EccB3:EccC3:EccD3:EccE3 complex was cross-linked with the N-hydroxysuccinimide (NHS) ester disuccinimidyl dibutyric urea (DSBU also known as BuUrBu). The cross-linking reactions were incubated for 45 min at room temperature at a final excess of 100-, 50- and 25-fold with respect to the protein concentration. The reactions were quenched by adding NH₄HCO₃ to a final concentration of 50 mM and incubation for 15 min.

The cross-linked samples were freeze-dried and resuspended in 50 mM NH₄HCO₃, reduced with 10 mM DTT and alkylated with 50 mM iodoacetamide. Following alkylation, proteins were digested with trypsin (Promega, UK) at an enzyme-to-substrate ratio of 1:20, overnight at 37 °C. The samples were acidified with formic acid to a final concentration of 2% (v/v) and the peptides fractionated by peptide size exclusion chromatography, using a Superdex Peptide 3.2/300 column (GE Healthcare) with 30% (v/v) acetonitrile/ 0.1% (v/v) TFA as mobile phase and at a flow rate of 50 μl/min. Fractions were collected every 2 min with elution volumes of 1.0 ml to 1.7 ml, lyophilized and resuspended in 2% (v/v) acetonitrile and % (v/v) formic acid.

Fractions were analyzed by nano-scale capillary LC–MS/MS using an Ultimate U3000 HPLC (ThermoScientific Dionex, USA) to deliver a flow of approximately 300 nl/min. A C18 Acclaim PepMap100 5 μm, 100 μm × 20 mm nanoViper (ThermoScientific Dionex, USA), trapped the peptides before separation on a C18 Acclaim PepMap100 3 μm, 75 μm × 250 mm nanoViper (ThermoScientific Dionex, USA). Peptides were eluted with a gradient of acetonitrile. The analytical column outlet was directly interfaced via a nano-flow electrospray ionization source, with a hybrid quadrupole orbitrap mass spectrometer (Q-Exactive HF-X, ThermoScientific, USA). MS data were acquired in data-dependent mode. High-resolution full scans (R=120 000, m/z 350-2 000) were recorded in the Orbitrap and after H activation (stepped collision energy 30 ± 3) of the five most intense MS peaks, MS/MS scans (R=15 000) were acquired.

For data analysis, Xcalibur raw files were converted into the MGF format through MSConvert (Proteowizard) [41] and used directly as input files for MeroX [42]. Searches were performed against an ad hoc protein database containing the sequences of the complexes and a set of randomized decoy sequences generated by the software. The following parameters were set for the searches: maximum number of missed cleavages 3; targeted residues K, S, Y and T; minimum peptide length 5 amino acids; variable modifications: carbamidomethyl-Cys (mass shift 57.02146 Da), Met-oxidation (mass shift 15.99491 Da); BuUrBu modification fragments: 85.05276 Da and 1 11.03203 Da (precision: 5 ppm MS and 10 ppm MS); False Discovery Rate cut-off: 5 %. Finally, each fragmentation spectrum was manually inspected and validated.

Mass determination of purified EccD3 using size exclusion chromatography (SEC) coupled with multi-angle light scattering (MALS)

SEC-MALS was performed at room temperature using a Superose 6 Increase 10/300 GL column coupled to an AKTA Purifier system (GE Healthcare) in-line with Dawn 8+ MALS and Optilab T-rEX refractive index detectors (Wyatt Technology, Santa Barbara, CA). Purified EccD3 (100 μL, 1 mg/ml) was applied to a Superose-6 Increase size exclusion column (10/300 GL) in a buffer containing 50 mM Tris pH 8.0, 300 mM NaCl and 0.05 % (w/v) DDM. The molecular weight was determined with the ASTRA 6 software (Wyatt Technology). For the protein conjugate analysis, we used a refractive index increment (dn/dc) of 0.185 ml/g for the protein fraction and 0.133 for DDM.

Stoichiometric Analyses of ESX-3 protein complexes by the Intensity-Based Absolute Quantification (iBAQ) method and Oriol staining

For in-gel digestion, the excised gel slices were de-stained with 30% acetonitrile, shrunk with 100% acetonitrile, and dried in a vacuum concentrator. Trypsin digest was performed overnight at 37 °C in 0.05 M NH₄HCO₃ (pH 8), using 0.1 µg of protease per slice. Peptides were extracted from the gel slices with 5% formic acid.

NanoLC-MS/MS analyses were performed on an Orbitrap Velos Pro (Thermo Scientific) equipped with a PicoView Ion Source (New Objective) and coupled to an EASY-nLC 1000 (Thermo Scientific). Peptides were loaded on capillary columns (PicoFrit, 30 cm x 150 µm ID, New Objective) packed with ReproSil-Pur 120 C18-AQ 1.9 µm (Dr. Maisch), and separated with a 30 min linear gradient from 3% to 30% acetonitrile and 0.1% formic acid at a flow rate of 500 nl/min.

MS scans were acquired in the Orbitrap analyzer with a resolution of 30 000 at m/z 400; MS/MS scans were acquired in the Orbitrap analyzer with a resolution of 7 500 at m/z 400 using HCD fragmentation with 30% normalized collision energy. A TOP5 data-dependent MS/MS method was used; dynamic exclusion was applied with a repeat count of 1 and an exclusion duration of 30 s; singly charged precursors were excluded from selection. The minimum signal threshold for precursor selection was set to 50 000. Predictive AGC was used with an AGC target a value of 1e6 for MS scans and 5e4 for MS/MS scans. The lock mass option was applied for internal calibration in all runs using background ions from protonated decamethylcyclopentasiloxane (m/z 371.10124).

For raw data file processing, database searches and quantification, MaxQuant version 1.5.7.4 was used [43]. The search was performed against the Mycobacterium smegmatis reference proteome databases (Uniprot) and, additionally, a database containing common contaminants. The search was performed with tryptic cleavage specificity with three allowed miscleavages. Protein identification was under control of the false-discovery rate (<1 % FDR on protein and peptide level). In addition to MaxQuant default settings, the search was performed allowing the following variable modifications: Protein N-terminal acetylation, Gln to pyro-Glu formation (N-terminal Gln), and oxidation (Met). Carbamidomethylation (C) was set as fixed modification. iBAQ intensities were used for protein quantitation. Proteins with less than two identified razor/unique peptides were dismissed.

The purified ESX-3 core complex was separated by SDS-PAGE. Gels were stained with Oriol (BioRad) according to the manufacturer’s instructions. Protein band intensity was measured with the Gel Doc XR+ (Biorad) using the UV transiluminator. ImageJ was used for quantification.

Cross-linking of ESX-3

Cross-linking experiments were carried out using the same protocol as for the ESX-5 core complex [14]. In brief, mycobacterial membranes were isolated as described in the purification section, resuspended in PBS-250 mM sucrose, and incubated with 1 mM DSS (disuccinimidyl suberate) or dimethylsulfoxide (DMSO; control) for 30 min on ice. The cross-linking reaction was quenched by addition of 100 mM glycine, 10 mM Na₂HPO₄ (pH 8.5) for 30 min. The cross-linked proteins were extracted from the membrane by detergent-solubilization with 0.5 % n-dodecyl β-D-maltoside (DDM) for 1 h on ice. Aggregates and insoluble material were removed by ultracentrifugation (100 000 x g, 1 h). The solubilized membrane proteins were separated by BN-PAGE using NativePAGE Novex 3–12% BisTris gels followed by Western Blot analysis using antibodies against the Strep-tag II (StrepMAB-Classic HRP conjugate, IBA life-sciences, Germany) according to the manufacturer's protocol (Invitrogen).

Data deposition

Cryo-EM maps corresponding to the central core region dimer and monomer, cytoplasmic distal region, and the fork have been deposited in the Electron Microscopy Database under accession codes EMD-X, EMD-X, EMD-X and EMD-X respectively and the model of ESX-3 with accession code PDB XXX (To be provided during review).

Extended Data

Extended Data Fig. 1. Purification and cryo-EM of ESX-3 protein complexes.

a) Overview of the constructs tested for co-expression studies. The illustrated gene cassettes were cloned into plasmid pMyNT and expressed under the control of the acetamidase promoter (Ace). Ace RBS, acetamidase ribosome binding site. From top to bottom: the ESX-3 gene cluster (pMyNT:ESX-3), the ESX-3 gene cluster without substrate genes (pMyNT:ESX-3ΔS), without substrate genes and cytosolic proteins (pMyNT:ESX-3ΔSC), only the ESX-3 membrane proteins (pMyNT:Mini).

b) Western Blot of TeDM solubilized membrane extracts containing ESX-3_EccC3-His6 complexes expressed from constructs listed in a) and separated by BN-PAGE. Anti-His₆-tag antibodies were used to detect the His-tagged ESX-3 complexes.

c) The ESX-3 core complex was expressed in wild type M. smegmatis mc2 155 from plasmid pMyNT:ESX-3 encoding the entire ESX-3 gene cluster. Left: BN-PAGE. Right: SDS-PAGE of the purified ESX-3 core complex.

d) Size exclusion chromatogram (Superose-6 Increase 10/300 GL) after the TeDM to Amphipol A8-35 exchange of the ESX-3 core complex. The positions of peak 1 (P1) and peak 2 (P2) are indicated with arrows.

e) Overview micrograph of the higher oligomeric ESX-3 core complex (> 1.2 MDa; P1). Some particles are highlighted by circles. The scale bar represents 100 nm.

f) Representative 2D class average of the higher oligomeric ESX-3 core complex.

g) Overview micrograph of the ~900 kDa ESX-3 core complex (P2). Some particles are highlighted within circles. The scale bar represents 100 nm.

h) Several 2D averages of the ESX-3 core complex after focused centering on the flexible fork. A flexible attachment of the fork to the membrane region can be observed.

Extended Data Fig. 2. Image processing and classification strategy of the cryo-EM data for the ESX-3 core complex.

a) Tools in RELION and cryoSPARC were used to clean a preliminary dataset and classify the images. 85% of particles corresponded to complexes made of 2 protomers and one fork, and these were grouped and refined, while the single protomer did not exhibit the fork (EccB3). This analysis suggests that EccB3 is involved in dimerization.

b) A larger data set was collected for high-resolution analysis. After removing bad particles, using 2D and 3D classification methods, a clean data set was further classified and refined to the core ESX-3 dimer, and the monomer was obtained after particle subtraction. The structure of the fork was resolved after density subtraction and refinement of the new generated particles.

Extended Data Fig. 3. Cryo-EM maps of the ESX-3 core complex.

a) Two views of the ESX-3 dimer

b) Two views of the ESX-3 monomer

c) Two views of the fork region (EccB3)

d) One view of each of the two conformations of the flexible tips of the arms, corresponding to EccC3, obtained after 3D classification. The density of each cryo-EM map is rendered at a low threshold and represented as a transparent density. Within each density, the same cryo-EM map is represented at higher threshold to highlight the structural details at higher resolution in the map.

Extended Data Fig. 4. Estimation of resolution and local resolution EM maps.

a) Left panel, Fourier shell correlation (FSC) plot and average resolution estimation for the core structure of the ESX-3 core complex (dimeric structure). Right panel, local resolution map and color scale.

b) As in (a) for the structure of protomer 1.

c) As in (a) for the structure obtained for the fork region.

d) model to map correlation plot for the ESX-3 dimer

e) model to map correlation plot for the ESX-3 protomer 1

Extended Data Fig. 5. Estimation of resolution and local resolution EM maps for the flexible arms.

a) Fourier shell correlation (FSC) plot and average resolution estimation for the two conformations (conformation 1 and 2) obtained for the complex after re-centering and processing around the flexible tips, corresponding to EccC3.

b) Local resolution maps and color scale for conformation 1, represented at two different thresholds. At lower threshold density for all the visible tip region is represented but the higher resolution details of EccC3 domain DI are not visible. Representations at higher threshold reveal these details at the expense of the disappearance of the regions of lower resolution.

c) As in (b) for conformation 2.

Extended Data Fig. 6. Details of the cryo-EM density and atomic models.

a) Representative regions in the cryo-EM density of protomer 1 showing details of high resolution and the atomic model for several domains in the structure.

b) In the cryo-EM map of the ESX-3 dimer there is partial density for one TMH of EccC3 for each protomer. The two TMHs (described in b) were fitted as poly-alanine models into the density using PHENIX Refine [38].

c) Plot revealing that density for the two EccC3 TMHs is well defined only at one end of the helix. Of note, the density for one of the EccC3 TMHs is better defined in the structure of the subtracted monomer, and this is represented in the panels showing the monomer structure in Fig. 2 and Fig. 3.

Extended Data Fig. 7. Validation experiments.

Experiments to validate that the ESX-3 core complex extracted from actively secreting mycobacteria is the same as the one whose structure was solved. We investigated the expression of the ESX-3 complex in the mycobacterial membrane, the stoichiometry of the extracted and purified complex and the orientation of the complex in the membrane.

a) Overview of the constructs used in secretion assays and for the validation of the ESX-3 core complex structure. Plasmid pMyNT:ESX-3 encodes the ESX-3 gene cluster under acetamidase promoter control and pMyNT:ESX-3i under control of the native IdeR promotor. The expression of the latter construct is induced in iron depleted culture medium.

b) BN-PAGE and Western Blot analysis (EccC3-His) indicate the presence of the 900 kDa ESX-3 core complex in the membrane of secretion-competent cells that either harbor the ESX-3 gene cluster under control of the acetamidase promoter or of the native IdeR promotor (see Extended Data Fig. 7a).

c) Comparison of size exclusion chromatography profiles of ESX-3 core complexes purified from secretion-competent cells transformed with plasmids pMyNT:ESX-3 and pMyNT:ESX-3i, both expressed in M. smegmatisΔESX-3; see Extended Data Fig. 7a), with the ESX-3 core complex purified from the minimal expression construct (pMyNT:Mini) expressed in WT M. smegmatis, which was used to determine the cryo EM structure (see Extended Data Fig. 1a).

d) ESX-3 core complexes purified from secretion-competent cells, in the presence and absence of ß-mercaptoethanol. A disulfide bridge is found in the EccB3 protein in complexes purified from secretion-competent cells. The two positions for EccB3 in the SDS-PAGE are indicated.

In the ESX-3 core complex structure, EccB3 is periplasmic and contributes to the assembly between protomers. We observe a different mobility for EccB3 present in the ESX-3 core complex purified from secretion-competent cells when the SDS-PAGE is performed in the presence or absence of ß-mercaptoethanol. This is compatible with an intra-molecular disulfide bridge in EccB3 expected for the oxidative environment of the periplasm.

e) Determination of the stoichiometry of native ESX-3 complexes in secretion-competent cells. Oriole staining of the SDS-PAGE bands in order to estimate the ratio of subunits obtained showed that EccD3 is over-represented in agreement with the observed 1:1:2:1 stoichiometry in the ESX-3 core complex structure. The SDS-PAGE shows that the density of the bands for EccD3 is higher than for EccE3, despite both proteins having a similar molecular weight (MW). The bands intensities were integrated and the relative ratio of subunits was estimated, taking into account the differences in MW of the proteins. The reported means and standard deviations are presented as bar diagram. The results obtained indicated a stoichiometry of 1:1:2:1 (EccB3:EccC3:EccD3:EccE3).

f) Analysis of the protein complex subunit composition using intensity-based absolute quantification (iBAQ). The size of the dots correlates with the number of identified unique peptides.

The Intensity Based Absolute Quantification Mass spectrometry (iBAQ MS) method was in our hands not sufficiently sensitive to define the stoichiometry of ESX-3. We used the ESX-3 complex from structural studies as internal control since the stoichiometry of the ESX-3 structure is fully determined as 1:1:2:1 at the high resolution of our cryo-EM maps. The iBAQ method estimates a 1:1:1:1 for the purified complex that we use to resolve the structure by cryo-EM, thus indicating that the method does not report the correct stoichiometry for ESX-3 in our hands. When the stoichiometry of the native ESX-3 core complex in secretion-competent cells was estimated by the iBAQ method used previously for the ESX-5 core complex [18], a 1:1:1:1 ratio was also obtained (data not shown).

g) Size-exclusion chromatography (Superose-6 Increase 10/300 GL column) coupled with multi-angle light scattering of purified EccD3. De-convolution of the EccD3-DDM protein-detergent complex: total complex mass (dot-dashed line), protein contribution (dotted line) and detergent micelles (dashed line) contribution. This experiment demonstrates that EccD3 dimerizes as observed in the ESX-3 core complex structure.

h) Size exclusion chromatography of the ESX-3 core complex derivates. Removing the periplasmic fork of EccB3 or EccE3, leads to dissociation towards single protomers showing that both components are essential for ESX-3 complex stability. The elution peak of the ESX-3 core complex dimer is indicated.

i) Isolated mycomembranes from M. smegmatis WT harboring the “high yield” minimal expression construct (pMyNT:pMini) and extraction of ESX-3 core complexes after cross-linking using DSS. ESX-3 complexes were analyzed by Blue-Native PAGE and Western Blot analysis (EccD3-StrepII). As a control complexes formed by EccD3 expressed alone were also analyzed simultaneously. Only ESX-3 assembled into large complexes, including those of higher molecular weight (MW) than the ESX-3 dimer.

Extended Data Fig. 8. Building the ESX-3 hexamer using the ESX-5 negative-stain structure as template.

a) Modeling of the ESX-3 secretion machine. The negative stain map of the hexameric ESX-5 core complex (EMDB-3596) was used to generate a model for the ESX-3 secretion machine. For this, the ESX-3 dimer (shown in red) was fitted into ESX-5 (shown as white transparent density), using the contour information of the negative-stain structure. Subsequently, another 2 dimers were fitted in the remaining subunits to form a hexamer. The atomic structure of each ESX-3 dimer fits approximately the dimensions of ESX-5, as shown in top and side views of ESX-5.

Since the EMDB-3596 reconstruction was generated assuming a perfect 6-fold symmetry, all subunits in the ESX-5 complex are identical, whereas the ESX-3 hexamer is based on a trimer of dimers.

Extended Data Fig. 9. The structure of the EccC3/EccB3 contact is reminiscent of ABC type F transporters.

a) Side by side comparison between the structure of the DUF and stalk domains of EccC3 interacting with EccB3 in the structure of the ESX-3 core complex, and the MacB ABC transporter (pdb ID 5NIL) [44]. The open (ATP unbound) state of MacB is shown. ATP binding and hydrolysis trigger long-range conformational changes in the stalk domains and the periplasmic domains of the MacB dimer, which close and open the transmembrane channel as well the periplasmic exit.

Extended Data Fig. 10. Comparison between the negative-stain structure of ESX-5 (EMDB-3596) and the cryo-EM structure of ESX-3.

a) Side view of the ESX-5 structure solved by Beckham et al. (EMDB-3596) [18]. The structure by Beckham et al. [18] was obtained after applying 6-fold symmetry. The use of staining agents limits the obtainable information to the external shape of the complex, while secondary structure elements and internal details are not resolved in negative-stain structures. EccC is not visualized in the structure, possible due to its flexibility and the use of 6-fold symmetry.

b) Top view of the ESX-5 structure.

Extended Data Table 1. Primers and oligonucleotides used for cloning.

X42	TAACTAGCGTACGATCGACTGCC
X43	GGTGGACTCCCTTTCTCTTATCGG
X44	GAAAGGGAGTCCACCATGGCTAGCTGGAGCCACCCGCAGTTCGAAAAAGGATCCTAACTAGCGTACGAT
X44C	ATCGTACGCTAGTTAGGATCCTTTTTCGAACTGCGGGTGGCTCCAGCTAGCCATGGTGGACTCCCTTTC
X96	GTTCGAAAAAGGATCCATGTCCGAGAACACTGTGATGC
X97	ACGCTAGTTAGGATCCTCACCTGTCGAGCACGAGG
X266	AGTCCACCGGATCGATATCATGACCGGCCCCGTCAAC
X267	GACTCCCTTTCTCGATATCTCATCGGGAGGCCTCCATACG
X273	GACTCCCTTTCTCTACGTATCACCTGTCGAGCACGAGG
X274	AAGGGAGTCCACCTACGTAATGGCTAGCTGGAG
X275	AAGGGAGTCCACCGTTAACATGACCGCCCGGATAGC
X276	TGGCTCCAGGATCGTTAACTCACGCCGGATGACC
X307	GGAGTCCACCGTTAGTACTGTGATCCACAAGAGTCTGGGC
X308	GACTCCCTTTCTCAGTACTTTATGTGGTCTTGTCCTTCCGACG
X311	GGGAGTCCACCAAGCTTATGAGCCGGCTC
X312	CTCCCTTTCTCAAGCTTTTAGTGGTGGTGGTGGTGGTGC
X386	TTATCGGGTGGTGGCCGC
X387	TAACTAGCGTACGATCGACTGCC
X388	GGATCTATGGGAGGATCGCACCACCACCACCACCACTAATACGTATAACTAGCGTACGATC
X389	GATCGTACGCTAGTTATACGTATTAGTGGTGGTGGTGGTGGTGCGATCCTCCCATAGATCC
X390	GGCCACCACCCGATAAGTTAGCTTAGGCATACATAAGGGAGAGAG
X391	ATCCTCCCATAGATCCTCGGTATTCCCCTCCTCGGTTG
X392	GGCCACCACCCGATAAGCTGCGTGGCCTCGGACT
X393	CCACCACCACTAATACAGGCCGAAGGCGCCAAG
X394	CGTACGCTAGTTATACCTACGCCGGATGACCCGC
X395	CCACCACCACTAATACGAGAACCAGTTCTCCAGAGAAGAACC
X396	CCACCACCACTAATACGAGCGGAAACATCGGGAACAAC
X397	GGGAGTCCACCGGATCGATATCGAGAAAGGGAGTCCACCAAGCTTGAGAAAGGGAGTCCACCTACGTAGAGAAAGGGAGTCCACCGTTAGTACTGAGAAAGGGAGTCCACCGTTAACGATCCTGGAGCCACCC
X398	GGGTGGCTCCAGGATCGTTAACGGTGGACTCCCTTTCTCAGTACTAACGGTGGACTCCCTTTCTCTACGTAGGTGGACTCCCTTTCTCAAGCTTGGTGGACTCCCTTTCTCGATATCGATCCGGTGGACTCCC
X426	ACGCTCGACGCTTGAGAC
X427	CTCAAGCGTCGAGCGTGACGCTGAACGAGTGTTTACATGAG
X428	TCCTCGTCGGTGAAATCTTCTTGGTCGG
X430	ATTTCACCGACGAGGAGATG
NX1	GCTAGAGGGGGCGTCAGGCG
NX2	CTCCATACGCGCAGAGTTGCCCGGCACGCCGC
NX3	GGCGTGCCGGGCAACTCTGCGCGTATGGAGGCCT
NX4	CGTCGAGAAGACTCATGACTGTTTCCT
NX5	AGGAAACAGTCATGAGTCTTCTCGACG
NX6	CGCCTGACGCCCCCTCTAGC
NX391	ATCCTCCCATAGATCCCGGGTACCGCGCGTTGATG
M37	CCGCGAGCAGTACGTATAACTAGCGTACGATCGACTGC
M38	GTATGCCTAAGCTAACTTATCGGGTGGTGGCCG
M39	GTTAGCTTAGGCATACATAAGGGAGAG
M40	GAGCACCACCAGCAGATGC
M41	CTGCTGGTGGTGCTCGCCGACCCCGATGCCGATCCCGACGACATCGCCCGCAAGCCCGGGCTGACCGGTGTCACCGTCATCCACCGCACCACGGAACTGCCCAACCGCGAGCAGTACGTA
M42	TACGTACTGCTCGCGGTTGGGCAGTTCCGTGGTGCGGTGGATGACGGTGACACCGGTCAGCCCGGGCTTGCGGGCGATGTCGTCGGGATCGGCATCGGGGTCGGCGAGCACCACCAGCAG
M43	CTGCTGGTGGTGCTCGACGCCCCCGATGCCGATCCCGACGACATCGCCCGCAAGCCCGGGCTGACCGGTGTCACCGTCATCCACCGCACCACGGAACTGCCCAACCGCGAGCAGTACGTA
M44	TACGTACTGCTCGCGGTTGGGCAGTTCCGTGGTGCGGTGGATGACGGTGACACCGGTCAGCCCGGGCTTGCGGGCGATGTCGTCGGGATCGGCATCGGGGGCGTCGAGCACCACCAGCAG
M45	CTGCTGGTGGTGCTCGACGACCCCGATGCCGATCCCGACGACATCGCCCGCAAGCCCGGGCTGACCGGTGTCACCGTCATCCACACCACCACGGAACTGCCCAACCGCGAGCAGTACGTA
M46	TACGTACTGCTCGCGGTTGGGCAGTTCCGTGGTGGTGTGGATGACGGTGACACCGGTCAGCCCGGGCTTGCGGGCGATGTCGTCGGGATCGGCATCGGGGTCGTCGAGCACCACCAGCAG
M47	CAACCGCGAGCAGTACCC
M48	GATCGTACGCTAGTTATACCTACGCC

Extended Data Table 2. Cryo-EM data collection for all three cryo-EM maps obtained in this work.

Cryo-EM	(EMDB-XXXX for protomer 1, EMD-XXX for the ESX-3 dimer; EMD-XXXX for the EccB3 fork)
Data collection
Microscope	FEI Titan Krios
Detector	Falcon III (counting mode)
Voltage (kV)	300
Electron exposure (e-/Å2)	50 (55 fractions)
Underfocus range (μm)	1.6 to 2.6
Pixel size (Å)	1.0635

Extended Data Table 3. Cryo-EM data refinement and validation statistics for all the cryo-EM maps obtained in this work.

Processing of ESX-3 core dimer (EMD-XXX)
Symmetry imposed	C1
Initial particle images (no.)	325,205
Final particle images (no.)	126,308
FSC threshold	0.143
Map resolution (Å)	3.8
Processing of ESX-3 core monomer (EMD-XXX)
Symmetry imposed	C1
Initial particle images (no.)	325,205
Final particle images (no.)	126,308
FSC threshold	0.143
Map resolution (Å)	3.7
Processing of fork (EccB3) (EMD-XXX)
Symmetry imposed	C2
Initial particle images (no.)	325,205
Final particle images (no.)	55,342
FSC threshold	0.143
Map resolution (Å)	4.6
Processing of tails (EccC3) conformation 1 (EMD-XXX)
Symmetry imposed	C1
Initial particle images (no.)	325,205
Final particle images (no.)	80,173
FSC threshold	0.143
Map resolution (Å)	5.4
Processing of tails (EccC3) conformation 2 (EMD-XXX)
Symmetry imposed	C1
Initial particle images (no.)	325,205
Final particle images (no.)	69,919
FSC threshold	0.143
Map resolution (Å)	5.3

Extended Data Table S4. Statistics for the atomic models of ESX3 monomer (protomer 1) and dimer.

Statistics for the ESX-3 model (protomer 1)
Refinement software	PHENIX Real-space Refinement
Initial model used (PDB code)	4KV2, 4N0H
Map sharpening B factor (Å2)	-114.21
Model composition
Non-hydrogen atoms	11 281
Protein residues	1516
Ligands	0
B factors (Å2)
Average	52.02
R.m.s. deviations
Bond lengths (Å)	0.005
Bond angles (°)	1.042
Validation
MolProbity score	1.84
Clashscore	4.91
Poor rotamers (%)	0.52
Cbeta outliers (%)	0.0
Ramachandran plot
Favored (%)	88.64
Allowed (%)	11.02
Disallowed (%)	0.34
Mask CC	0.78
Statistics for the ESX-3 model (dimer)
Refinement software	PHENIX Real-space Refinement
Initial model used (PDB code)	4KV2, 4N0H
Map sharpening B factor (Å2)	-101.54
Model composition
Non-hydrogen atoms	22 291
Protein residues	2 998
Ligands	0
B factors (Å2)
Average	83.67
R.m.s. deviations
Bond lengths (Å)	0.005
Bond angles (°)	1.068
Validation
MolProbity score	1.91
Clashscore	5.87
Poor rotamers (%)	0.48
Cbeta outliers (%)	0.0
Ramachandran plot
Favored (%)	88.54
Allowed (%)	11.05
Disallowed (%)	0.41
Mask CC	0.78