In vivo structure of the Legionella type II secretion system by electron cryotomography.

The type II secretion system (T2SS) is a multiprotein envelope-spanning assembly that translocates a wide range of virulence factors, enzymes and effectors through the outer membrane of many Gram-negative bacteria1-3. Here, using electron cryotomography and subtomogram averaging methods, we reveal the in vivo structure of an intact T2SS imaged within the human pathogen Legionella pneumophila. Although the T2SS has only limited sequence and component homology with the evolutionarily related type IV pilus (T4P) system4,5, we show that their overall architectures are remarkably similar. Despite similarities, there are also differences, including, for example, that the T2SS-ATPase complex is usually present but disengaged from the inner membrane, the T2SS has a much longer periplasmic vestibule and it has a short-lived flexible pseudopilus. Placing atomic models of the components into our electron cryotomography map produced a complete architectural model of the intact T2SS that provides insights into the structure and function of its components, its position within the cell envelope and the interactions between its different subcomplexes.

Evolutionarily-related to bacterial type IV pili[4,5], canonical T2SS are mainly distributed among genera of Proteobacteria, including representatives within α-, β-, γ-, and δ-proteobacteria[3,6,7] The T2SS apparatus generally employs 12 “core” components, which, for simplicity, will be referred to here as T2S C, D, E, F, G, H, I, J, K, L, M, and O[7,8]. These components have been described as being part of four subcomplexes[7,8]. The first subcomplex is a multimer of the T2S D protein that provides the ultimate portal or gate for substrate transit across the OM and out of the cell; i.e., the so-called secretin[9]. Interacting with the secretin to create a periplasm-spanning channel is an IM-anchored subcomplex or platform comprised of T2S F, L, and M[10]. The OM-associated subcomplex and the IM-associated subcomplexes are coupled by the IM-associated “clamp protein” T2S C[10]. The third subcomplex is a pseudopilus that consists of the major pseudopilin T2S G and minor pseudopilins T2S H, I, J, and K[10]. The pseudopilus is thought to span the periplasm within the channel created by the interaction of the IM platform with the secretin and is believed to act in a “piston-” or “screw-like” fashion to drive substrate through the OM portal[7,11,12]. The fourth subcomplex is a hexamer of T2S E, a cytoplasmic ATPase that is recruited to the IM in order to “power” the secretion process[10]. The T2S O protein is an IM- prepilin peptidase[7,8].

In vitro structural studies of purified components and subcomplexes have significantly improved our understanding of the T2SS[13-26]. Despite these advances, it is not clear how the different components of the T2SS are positioned and interact with each other within an intact cell envelope (in situ).

ECT has recently emerged as a powerful tool to investigate macromolecular complexes in situ at nanometer resolution[27-31]. To reveal the intact structure of the bacterial T2SS in situ, we used ECT and image nearly 2000 frozen-hydrated L. pneumophila cells (Supplementary Table 1). In our tomograms, we observed multiple electron dense “hour-glass” shaped particles (Fig. 1A–D), reminiscent of the secretin structure[9], in the periplasm. These structures were primarily localized in the vicinity of cell poles and were not associated with any exocellular filaments (Fig. 1A, B). We also observed several top views of these particles with a diameter ~10 nm (Fig. 1E, F). While L. pneumophila cells encode for a type IVa pilus (T4aP) system, which also has a secretin[32], several lines of evidence suggest that all the structures we saw were T2SSs. First, in our ~2000 tomograms, we did not see any T4aP filaments coming out of the L. pneumophila cells, suggesting that under our growth conditions, T4aP is not expressed. This is consistent with the earlier observation in which 30 °C was found to be better than 37°C for the assembly of the T4aP systems[32]. Second, no particles were visible in a strain lacking L. pneumophila T2SS components T2S D and E (ΔlspDE) (Fig. 1G), but they were still present in a strain lacking the T4aP secretin (ΔpilQ) (Fig. 1H). In a double deletion strain lacking the T2SS secretin as well as the T4aP secretin (ΔlspDE, ΔpilQ), again no particles were visible (Fig. 1I, Supplementary Table 1). Finally, our analysis of the T2SS secretin revealed distinct features as compared to the T4aP secretin (see Extended Data Fig. 1 and discussion below), confirming that the “hour-glass” shaped particles are L. pneumophila T2SSs.

Fig. 1.

Visualization of the type II secretion system (T2SS) in frozen-hydrated L. pneumophila cells.

(A) Cryo-EM image of a frozen-hydrated L. pneumophila cell on a Quantifoil grid. (B) Central slice through a tomographic reconstruction of a L. pneumophila cell pole. White arrows point to “hour-glass” shaped putative T2SS structures. (C and D) Tomographic slices showing individual T2SS particles. (E and F) Top views of individual T2SS particles. (G-I) Tomographic slices of L. pneumophila mutants lacking the T2SS secretin (G), the type IVa pili secretin (H), or both (I). (J) Composite subtomogram average of the wild-type L. pneumophila T2SS. For each strain, number of tomograms recorded and number of particles found are listed in the SI Table-1. (K and L) Subtomogram averages of the Vibrio cholerae T4bP and Myxococcus xanthus T4aP[33,34] for comparison. (M) Domain architectures of the secretin protein in the L. pneumophila T2SS (T2S D or LspD), M. xanthus T4aP (PilQ) and V. cholerae T4bP (TcpC). Protein regions corresponding to different densities are indicated by arrowheads. PG: Peptidoglycan binding domain, SPOR: Sporulation related repeat, STN: Secretin and TonB N terminus short domain. Scale bars, 500 nm (A), 50 nm (B), 20 nm (C, D), 10 nm (E, F), 50 nm (G-I), 10 nm (J-L).

Extended Data Fig. 1:

Comparison between the T2SS and related molecular machines T4aP and T4bP systems.

To reveal the in situ molecular architecture of the intact L. pneumophila T2SS, we generated a subtomogram average using 440 particles. Our analysis showed substantial flexibility between the OM-associated secretin complex and the rest (Extended Data Fig. 2A). To overcome this, we performed focused alignment on the OM- and IM-associated subcomplexes separately and produced a composite structure (Fig. 1J, Extended Data Fig. 2A–C) which had a variable local resolution between ~2.5–5.5 nm as estimated by ResMap (Extended Data Fig. 2D). The final average revealed that the T2SS is composed of a ~23 nm long vestibule that reaches down most of the way through the periplasm and opens up in the OM with an ~8 nm wide pore (Fig. 1J). There are two distinct densities in the lumen of the secretin channel near the two ends: (i) a “gate” just underneath the OM pore and (ii) a previously unreported structure, “plug”, near the base (Extended Data Fig. 2E). Although the gate structure resolved well in our subtomogram average, the plug structure was poorly visible. Further analysis revealed that the plug is either not present or so dynamic in a subset of the particles that it is almost invisible in the final average (Extended Data Fig. 2E). Below the secretin channel, 7.5 nm away from the IM, there is a lower- periplasmic ring and a short stem at the same level (Fig. 1J). The lower-periplasmic ring is 17 nm wide in diameter (peak to peak) and the stem structure is in the middle of this ring (Fig. 1J). On the cytoplasmic side, there are three distinct densities: a dome-like density immediately below the IM (cytoplasmic dome), a ring-like density of diameter 20 nm (peak to peak) (cytoplasmic ring) surrounding the dome and finally, 13 nm away from the IM a 12-nm wide (peak to peak) cytoplasmic disk (Fig. 1J). A comparison between the subtomogram averages of the T2SS and the non-piliated in situ structures of the Myxococcus xanthus T4aP (MxT4aP) and the Vibrio cholerae toxin-coregulated (TCP) type IVb pilus (VcT4bP)[33,34] machines revealed that despite varied gene organization and limited component homology, the overall architectures of these three classes of molecular machines are strikingly similar (Fig. 1J–L, note cytoplasmic disks for the MxT4aP and VcT4bP are seen in other states. See Extended Data Fig. 2F).

Extended Data Fig. 2.

T2SS Flexibility.

Subtomogram averages of all particles aligned on (A) the OM-associated complex and (B) the IM-associated complex. The distribution of the green dots in (B) indicates the translations imposed on the OM complexes to align the IM complexes. (C) A composite average using the upper and lower halves of (A) and (B), respectively. D) Local resolution of (C) calculated by Resmap. (E) Focused alignment near the base of the secretin channel revealed the presence of a plug-like structure. 20% of the particles with highest cross-correlation showed this distinct density. In the rest of the particles, the plug density is either not present or so dynamic that including them makes the plug almost invisible. (F) Previously reported in situ averages of the T4aP (WT, piliated) and T4bP (WT, piliated and ΔtcpR mutant) machines in states with cytoplasmic dome, ring and disks for comparison[33,34]. White arrows indicate cytoplasmic disks. Scale bars, 10 nm (A-C), (E), (F).

Intriguingly, the L. pneumophila secretin (T2S D, aa = 791) sequence is shorter than the MxT4aP secretin (PilQ, aa = 901), but in the subtomogram averages, the L. pneumophila T2SS secretin channel is ~1.5 times longer than the MxT4aP. To rationalize this, we predicted domain architectures of secretins of the L. pneumophila T2SS (T2S D or LspD), VcT4bP (TcpC) and MxT4aP (PilQ) systems using Motif-search, Phyre and the CD-vist programs[35,36] (Fig. 1M). Our analysis revealed that all three secretins (LpT2S D, MxPilQ and VcTcpC) have conserved C- terminal secretin domains. However, preceding the secretin domain, TcpC and PilQ have only one N-domain, but T2S D has four (N3, N2, N1 and N0) (Fig. 1M, Extended Data Fig. 3A). N- domains are known to fold into rigid structures and thus resolved well in the subtomogram average[33,37]. In contrast, the N-terminal AMIN domains of PilQ are not ordered and thus remained indiscernible in the subtomogram average.

Extended Data Fig. 3.

Position of the T2SS secretin with respect to the OM.

(A and B) Atomic models of the V. cholerae (PDB ID: 5WQ8) and E. coli (PDB ID: 5WQ7) T2SS secretins superimposed on our subtomogram average based on the position of the gate. (C) Positions of the OM on these structures as suggested in earlier publications[13,15]. The widths of the suggested OM spanning regions were only ~1.8 nm, but real membranes are known to be 5–7 nm wide. In all reported atomic models, the secretin channel is suggested to extend beyond the OM[13,9]. However, when we overlaid the secretin atomic models on our subtomogram average, it only reached through the inner leaflet of the OM. Scale bars, 10 nm. D) Tomographic slices of mutant L. pneumophila cells lacking all major and minor pilins (ΔlspGHIJK). Showing representative individual T2SS particles. No pseudopilus or lower-periplasmic ring is visible. A similar result was obtained when we examined a L. pneumophila ΔlspHIJK mutant. Scale bar, 10 nm (D). For each strain, number of tomograms recorded and number of particles found are listed in the SI Table-1.

In our subtomogram average, we saw densities for the cytoplasmic dome, ring and disk of the T2SS (Fig. 1J). The cytoplasmic dome is also visible in the piliated as well as in the non-piliated state of the MxT4aP and VcT4bP machines (Fig. 1K,L, Extended Data Fig. 2F), but the cytoplasmic ring and the disk are only visible in certain states of the MxT4aP and VcTcpC systems interpreted previously as assembly states (Fig. 1J–L, Extended Data Fig. 2F).

Our in situ structure revealed how the T2SS is positioned in the cell envelope with respect to the OM, peptidoglycan (PG) and IM. In our average, the secretin complex forms an OM-spanning pore. However, when we aligned available cryoEM structures of assembled secretin channels onto our density map based on the position of the gate (Extended Data Fig. 3A–C), the tip of the secretin penetrated only the inner leaflet of the OM, suggesting secretin assemblies might have a different, more extended conformation in vivo, and collapse or change conformation during detergent extraction and purification.

To investigate if the T2SS is activated under the growth conditions used, we looked for T2SS- effector release in the L. pneumophila culture supernatant. Western blot analysis of the L. pneumophila culture supernatant showed consistent presence of the effectors CelA, ProA and LegP during mid-log phase and late-log phase, suggesting that at least some of the T2SSs are active (Fig. 2P). This effector release is T2SS-specific because a T2SS-nonfunctional mutant (ΔlspDE) did not release any of these effectors (Fig. 2P). Interestingly, in a strain lacking the functional Dot/Icm type IV secretion system (Δdot/icm), we found 25% more T2SS particles. However, this does not correlate with an increase in T2SS effector release (Fig. 2P).

Fig. 2.

Individual T2SS subcomplexes.

Tomographic slices showing individual T2SS particles with: (A-E) pseudopili (yellow arrowheads), (F-J) lower-periplasmic rings (white arrowheads) and (K-O) only secretin channels. Schematic representations (right). Scale bars, 10 nm (A-O). (P) Western blot analyses of T2SS effector (CelA, ProA and LegP) release in L. pneumophila culture supernatants at mid- and late-log phases. Different strains were analyzed: lanes- 1 and 5) WT L. pneumophila (Lp02), and mutant strains lacking 2 and 6) a functional T2SS (ΔlspDE), 3 and 7) a functional Dot/Icm T4BSS (ΔdotH, ΔdotG, and ΔdotF: JV7058), 4 and 8) all components of the Dot/Icm T4BSS (JV4044). The top set of immunoblots assesses the proteins in the culture supernatants, whereas the bottom set of immunoblots assesses the proteins within cell lysates. Blots assessing the presence of the cytoplasmic protein isocitrate dehydrogenase (ICDH) are presented as controls. In several rounds of experiments, we did not notice any consistent difference (increase or decrease) in T2SS effector release in Dot/Icm T4SS nonfunctional mutants. Shown here are representative western blots. Experiments were done in duplicates and representative images are shown.

In the T2SS, four minor pseudopilins (T2S I, J, K and H) and a major pseudopilin (T2S G) constitute a pseudopilus that is thought to act as a piston to extrude exoproteins through the T2SS[7,11,12]. In mutant strains lacking the pseudopilins, the activity of the T2SS is severely impaired[38]. Since we did not see a pseudopilus in our subtomogram average (Fig. 1J), we carefully looked through all the individual T2SS particles in our tomograms and found that a fraction (~20%) of the particles had density likely from pseudopili (Fig. 2A–E). In these particles, the putative pseudopili extend (~14 nm) from the IM to the base of the T2S D channel. The positions of these densities are variable, however, so they remain invisible in the average. To confirm the identity of these pseudopili-like densities, we imaged a L. pneumophila mutant strain lacking all the minor pseudopilins (ΔlspHIJK) and another one lacking all the pseudopilins (ΔlspGHIJK). No pseudopilus-like density was visible associated with the T2SS particles in any of these mutant strains confirming that the rod-like densities just below the secretin channel are indeed pseudopili (Extended Data Fig. 3D).

In our tomograms, the lower-periplasmic ring densities in individual particles were just as visible as the secretin density, but their positions varied between particles, explaining why this density resolved poorly in the subtomogram average (Fig. 2F–J). Furthermore, there were also many particles with just the periplasmic vestibule and without a detectable pseudopilus or a lower- periplasmic ring (Fig. 2K–O). These particles are likely either assembly or disassembly intermediates.

Our in situ structure of the T2SS showed that the basic architectures of the MxT4aP, VcT4bP and LpT2SS are clearly very similar (Fig. 1J–L, Extended Data Fig. 1). There are also many interesting differences, however, which lead to specific hypotheses about the roles and adaptations of many of the components as discussed below (Extended Data Fig. 1).

The T2SS has a much longer periplasmic vestibule (23 nm) than the T4aP and T4bP systems (15 nm). It is unclear if this difference is because the T2SS translocates dozens of exoproteins through the secretin channel, whereas the T4aP and T4bP systems primarily secrete their pili. In our structure, we observed a poorly-resolved plug density near the base of the secretin vestibule (Extended Data Fig. 2E). T2S C is known to interact with the NO domain of T2S D through its HR domain and recruit T2SS effectors into the complex[22]. Therefore, the plug density could be the HR domain of T2S C[22] or, perhaps cargo that is waiting to be extruded by the pseudopilus.

Unlike theMxT4aP and the VcT4bP systems, the L. pneumophila T2SS does not encode a pilotin[3,39]. Although the T2S D sequence has a pilotin binding S-domain, no putative pilotin sequence was found in the L. pneumophila genome. This is consistent with the idea that pilotins are not essential for all secretin pores to assemble. It remains unclear how the L. pneumophila T2S D is targeted to the cell pole and associates with the PG. In the MxT4aP system, the N- terminal AMIN domains of PilQ tether the T4aP machine to the PG layer[34]. In our L. pneumophila tomograms, the PG layer is located immediately beneath the OM (Extended Data Fig. 4) but the putative PG binding domain of T2S D is N-terminal to the NO domain, which is ~20 nm below the PG (Fig. 1M). One possibility is that the linker between the NO domain and the putative PG binding domain bridges this gap. Curiously, the L. pneumophila T2SS also lacks other PG binding complexes like those present in Vibrio spp., Aeromonas spp. and in E. coli T2SS (e.g. gspAB homologs)[3,40].

Extended Data Fig. 4.

Position of the T2SS with respect to the PG layer in L. pneumophila.

(A and B) Tomographic slices through L. pneumophila cells showing T2SS particles (red arrowheads) and the peptidoglycan layer (PG, yellow arrowheads). (C and D) Tomographic slices through L. pneumophila cells showing T4BSS particles (red arrowheads) and the peptidoglycan layer (PG, yellow arrowheads). (E) Tomographic slice through a L. pneumophila cell showing both T4SS and T2SS particles (red arrowheads) and peptidoglycan (PG, yellow arrowheads) in the same cell. (F and G) Subtomogram averages of the T4SS and T2SS, respectively. DotK (shown as green arrow) in the T4BSS is known to interact with the PG layer confirming its location just a few nm below the OM (F). We therefore conclude that the PG layer surrounds the T2SS at approximately the level of the gate (G). For each strain, number of tomograms recorded and number of particles found are listed in the SI Table-1. Scale bars, 100 nm (A-E), 10 nm (F-G).

In the L. pneumophila T2SS, the periplasmic domains of T2S L and M form a lower-periplasmic ring similar to those present in the MxT4aP and VcT4bP systems. PilO and PilN (the two proteins that form the lower-periplasmic ring in the MxT4aP), and T2S L and T2S M all begin with an N-terminal transmembrane helix, followed by a ferredoxin-like fold in the periplasm. Previously, it was hypothesized that in the MxT4aP, the lower-periplasmic ring likely transmits information about the status of the pilus to the cytoplasmic-subcomplex[34]. Given that T2S L and T2S M have recently been shown to interact with secreted effectors[41], T2S L and M may likewise sense cargo in the periplasm and signal the cytoplasmic complex to extend the pseudopilus. Signal sensing and transmission are likely simpler in the T2SS, however, since while the lower-periplasmic ring in the MxT4aP is composed of 12 or more copies of both PilO and PilN, in L. pneumophila, the lower-periplasmic ring is apparently comprised of only six copies each of T2S L and M. Based on these stoichiometries, the conservation of fold, and the appearances of the rings in the subtomogram averages, we predict that the basic structures of these lower-periplasmic rings are the same, but there are two closely-packed rings in the MxT4aP, as opposed to only one in the T2SS. Signaling likely involve transitions between homodimer and heterodimer associations, as the presence of cargo biases the interactions between T2S L and M towards heterodimers[41], and the VcT4bP system, which may not signal (the pilus is never retracted by an ATPase), has only a single lower-periplasmic-ring protein (TcpD)[33]. It remains elusive how the symmetry mismatch between the secretin channel, T2S C and the lower-periplasmic ring is coordinated and what the role of symmetry mismatch is in effector loading and export.

The conservation of transmembrane helices in all the lower-periplasmic-ring proteins suggests an important role in signal transmission or structure, but the number of helices present and their connections to the cytoplasmic ring vary. Assuming there are six copies of both T2S L and M, a ring of 12 transmembrane helices will be present in the IM. This contrasts with the 24 or more helices from PilN and PilO in the MxT4aP. Perhaps in partial compensation, the T2SS clamp protein T2S C begins with a transmembrane helix, adding 6–12 more helices to the ring (according to current estimates of stoichiometry), while its counterparts in the MxT4aP and FcT4bP systems (PilP and TcpS, respectively) are simply lipoproteins. Curiously, in the FcT4bP system, there are again two proteins contributing to alpha helices in the IM, but in this case one is from the periplasmic ring (TcpD) and the other is from the cytoplasmic ring (TcpR). The L. pneumophila T2S exhibits yet another variation, in that one lower-periplasmic-ring protein (T2S L) forms part of the cytoplasmic ring, with the transmembrane helix in between. We speculate that the fusion allows another adaptation, which is that in L. pneumophila the cytoplasmic domain of T2S L appears to not form a continuous ring. Being fused to the periplasmic ring may be important to hold it in place. Overall, seemingly a network of 18 or more interacting transmembrane helices in the IM mediate signal transmission between the lower-periplasmic ring and the cytoplasmic ring.

Just as in the MxT4aP and FcT4bP systems, the T2SS ATPase sits directly below the IM on the machine’s axis, however, is positioned further away from the IM and is less well resolved. We speculate that as the T2SS rapidly switches from extending to retracting the pseudopilus (the piston model), it does so by moving the ATPase up against the adaptor T2S F to extend the pseudopilus, and then dropping it away from T2S F to let the pseudopilus retract. Our interpretation of the data is that most of our particles were in the retraction state, with the ATPase disengaged from T2S F. This mechanism is of course different than the MxT4aP system that has two (extension and retraction) ATPases[34] In the MxT4aP case, the two ATPases are sequentially exchanged and were seen in contact with the adaptor in the cryotomograms, suggesting they are held there continuously for long periods of time. In contrast, the T2SS likely evolved to rapidly switch between short bursts of extension and retraction, and so it does not release its ATPase completely - it appears to merely “clutch” it on and off. In the retraction state, we propose that the N1E domain is close to the IM and interacts with the cytoplasmic domains of T2S L and T2S F[17,20,42-44], while the N2E and CTE domains dangle on flexible linkers, causing them to resolve poorly.

In our tomograms, several individual particles exhibit putative pseudopili. No pseudopilus was seen to extend up to the gate, suggesting either the pseudopilus only pushes exoproteins into the periplasmic vestibule or an extended form of a pseudopilus (reaching up to the gate or beyond) is short-lived. Our model showed that only ~5 copies of the major pseudopilin T2S G and one copy of each of the minor pseudopilins (T2S I, J, K, H) is sufficient to span the distance between the IM and the base of the secretin. We also found that ~12–15 copies of the major pseudopilin and one copy of each of the minor pseudopilins T2S I, J, K, H would be required for the pseudopilus to extend to the secretin gate. It is not clear how the length of the pseudopilus is controlled. Intriguingly, the C-terminal domain of T2S L has sequence similarity (Phyre: 52 amino acids, 68% confidence and 15% identity) to the C-terminal domain of FliK that controls flagellar hook length[45]. Thus, T2S L may have multiple functions.

Overall, this study highlights commonalities and key differences between the evolutionarily- related T2SS, T4aP and T4bP machines and provides insights into its structure and function.

Materials and Methods

Strains, growth conditions, and mutant generation

All experiments were performed using the L. pneumophila Lp02 strain (thyA hsdR rpsL), a derivative of the clinical isolate L. pneumophila Philadelphia-1. Cells were grown as described previously[27,46]. Briefly, cells were grown in ACES buffered yeast extract (AYE) broth or on buffered charcoal yeast extract (CYE) plates. The culture media were always supplemented with thymidine (100 pg/ml), ferric nitrate and cysteine hydrochloride.

A T4SS mutant of Lp02 that lacks all 26 of the dot/icm genes (i.e., JV4044) was previously described[47,48], as was a mutant of Lp02 that lacks dotH, dotG, and dotF (i.e., JV7058)[48]. To generate a lspDE mutant lacking the T2SS (i.e., strain NU438) and apilQ mutant lacking T4P (i.e., strain NU439) of L. pneumophila, previously reported plasmids were introduced into Lp02 by natural transformation and then the desired mutations were introduced into the bacterial chromosome via allelic exchange[49]. Plasmid pOE4Kan was used to make the lspDE mutant, and pGQ::Gm was employed for making the pilQ mutant[50,51]. In a similar way, an lspDE pilQ double mutant lacking both T4SS and T4P (i.e., strain NU440) was constructed by introducing pGQ::Gm into strain NU438.

To generate an lspHIJK mutant lacking the minor pseudopilins (i.e., strain NU442) and an lspGHIJK mutant lacking both the major and minor pseudopilins (i.e., strain NU444) a form of allelic exchange was used. First, mutagenized alleles were generated using overlap extension PCR. The 5’ and 3’ flanking regions of lspHIJK and lspGHIJK were PCR amplified from L. pneumophila 130b DNA. Then the kanamycin-resistance cassette flanked by Flp recombination target sites was PCR amplified from pKD4. We performed two-step overlap extension PCR to combine the 5’ and 3’ regions of IspHIJK and IspGHIJK with the respective resistance cassettes. PCR products corresponding to the correct sizes were gel purified and ligated into pGEM-T Easy (Promega) to yield either pGlspHIJK::Kn or pGlspGHIJK::Kn. These plasmids were then introduced into strain Lp02 by natural transformation and mutants were obtained by plating on CYE agar containing kanamycin and verified by PCR.

Assay for secreted proteins

L. pneumophila strains were grown in AYE broth at 37°C to either mid-log or late-log phase, at which times the OD660 was measured and the cultures were centrifuged. The resulting supernatants were filtered through a 0.2 μm syringe filter. The supernatant and pellets were diluted in PBS and 2X Laemmli buffer according to the OD660 to normalize for bacterial numbers. The samples were subjected to SDS-PAGE and then analyzed by immunoblot as previously described[52]. Briefly, after blocking in 5% milk for 1 h, the blots were incubated overnight with 1:5,000 dilutions of rabbit anti-CelA, anti-ProA, anti-ICDH or 1:1,000 dilutions of anti-LegP antibodies, washed, and then incubated for 1 h with 1:1,000 dilution of secondary HRP-conjugated goat anti-rabbit IgG antibody (Cell Signaling Technology). Images of the immunoblots were developed with ECL™ Western Blotting Detection Reagent (GE Healthcare).

Sample preparation for electron cryotomography

L. pneumophila Lp02 cells were grown till early stationary stage (OD600 ~2.8) and harvested. Cells were mixed with 10-nm colloidal gold beads (Sigma-Aldrich, St. Louis, MO) precoated with bovine serum albumin. Four μl of this mixture was applied onto freshly glow-discharged copper R2/2 Quantifoil holey carbon grids (Quantifoil Micro Tools GmbH, Jena, Germany). Using an FEI Vitrobot Mark IV, grids were then blotted (under 100% humidity conditions) and plunge-frozen in a liquid ethane/propane mixture.

Electron tomography and subtomogram averaging

The frozen grids were subsequently imaged in an FEI Polara 300 keV FEG transmission electron microscope (Thermo Fisher Scientific) coupled with a Gatan energy filter and a Gatan K2 Summit direct electron detector. Energy-filtered tilt series of cells were collected automatically from −60° to +60° at 1.5° intervals using the UCSF Tomography data collection software[53] with a cumulative total dosage of 100 e− Å−2, a defocus of −6 μm and a pixel size of 3.9 Å. Using the IMOD software package[54], the images were then binned by 2, aligned and contrast transfer function corrected. Subsequently, SIRT reconstructions were produced using the TOMO3D program[55]. T2SS structures on cell envelopes were visually identified by their characteristic “hour-glass” like shape. Subtomogram averages of the L. pneumophila T2SS were generated by the PEET program[56]. The T2SS subtomogram averages exhibited a gross twofold symmetry around the central midline in the periplasm. Based on this observation, we applied a two-fold symmetry on the periplasmic complex. No symmetry was applied on the cytoplasmic complex. The T2SS exhibited a substantial flexibility between the OM- and IM-associated parts; binary masks were applied to perform focused alignments on the OM and on the IM complexes separately and finally combined to generate a composite structure. The numbers of tomograms collected and number of particles used are summarized in Supplementary Table 1.

Identifying homologues between the L. pneumophila T2SS and the T4P system

L. pneumophila Lp02 strain T2SS protein sequences were selected from the NCBI and UniProt databases. To identify corresponding components between the L. pneumophila T2SS and the MxT4aP and FcT4bP systems, we utilized two different programs: Phyre and Blast search[35,57]. This confirmed L. pneumophila T2SS components T2S C, D, E, F, G and M are homologues of the T4aP components PilP, PilQ, PilB, PilC, PilA and PilO respectively. Using a combination of Phyre, MOTIF-search, and Blast programs, we also predicted distinct domains and motifs within different T2SS components.

Building an architectural model of the intact L. pneumophila T2SS

Recently, Chang et al used a combination of ECT, subtomogram averaging and genetic manipulations to determine the locations of all the major components in the MxT4aP and FcT4bP systems and produced an architectural model of the MxT4aP machinery[33,34]. However, in this case, because there is only about one T2SS particle in every -five L. pneumophila tomograms, a thorough mutant analysis (where individual components are either deleted or fused to a GFP) was not possible. Therefore, in order to assign the locations of T2SS components in our density map and build an architectural model of the T2SS, we performed a thorough sequence analysis and confirmed that the L. pneumophila T2SS components T2S C, D, E, F, G, and M are homologues of the T4aP components PilP, PilQ, PilB, PilC, PilA and PilO respectively[58]. And used this information to building an architectural model of the intact L. pneumophila T2SS.

Interestingly, the cytoplasmic domain of T2S L is homologous to the T4aP component PilM (Phyre: 30% coverage, 97% confidence and 15% identity) and the periplasmic domain of T2S L is homologous to the T4aP component PilN (Phyre: 23% coverage, 78.6% confidence and 13% identity) implying T2S L is a fusion of the T4aP proteins PilM and PilN as previously suggested in other systems[59].

Since atomic models are available for all the soluble domains of the T2SS components[9,10], we sought to build an architectural model of the intact T2SS guided by the T4aP model (Fig. 3). The L. pneumophila T2SS has a total of 12 components. We began by placing an atomic model of the OM protein T2S D/secretin (PDB ID: 5WQ8) in our density map. The position of T2S D along its axis (perpendicular to the cell envelope) was set by aligning the gate densities. The N1, N2 and N3 domains of 5WQ8 matched well with the subtomogram average density map (Extended Data Fig. 3A–C). While the NO domain of secretin was not resolved in most of the previously reported single particle reconstructions, presumably due to flexibility, in our subtomogram average we saw an additional density for the NO domain just below N1 (Extended Data Fig. 3A–C) as recently seen by Chernyatina et al[17]. Our interpretation is that the NO domain is stabilized by the presence of its connecting subcomplexes and the rest of the cellular envelope. A co-crystal structure of the T2S D NO domain and the C-terminal homology region (HR) domain of the clamp protein T2S C (PDB ID: 3OSS) was therefore placed in this density below the N1 domain. Since no oligomeric structure is available for T2S C, we used known information about symmetry, connectivities and orientation (e.g. towards membrane etc.) to first generate numerous 15-mer (the known symmetry of T2S D) ring models for 3OSS using SymmDock[60] and placed the ring model in the electron density map that fit the density best and satisfied all other criteria. Recently, Chernyatina et al showed that Klebsiella pneumoniae T2S CELM constitute a ~2:1:1:1 complex and 6–12 (or maybe up to 15) copies of T2S C are present in this assembly (personal communication)[17]. Therefore, we removed 9 copies of T2S C randomly around the ring in the model to reflect this lowest stoichiometry and emphasize that not all T2S D’s are bound by a T2S C[17]. Other than the C-terminal HR domain, L. pneumophila T2S C contains a periplasmic disordered region and an N-terminal transmembrane helix. We modeled the disordered region of T2S C and placed the N-terminal helix in the IM.

Fig. 3.

Architectural model of the T2SS.

Atomic models of T2SS components are superimposed on the central slice of the T2SS subtomogram average based on known connectivities and interfaces (see text). Transmembrane domains of the IM proteins are shown as cylinders. Linkers between N1E domain and N2E/CTE domains of T2S E are represented by dotted lines. OM = outer membrane, IM = inner membrane, PG = peptidoglycan. Lipids are shown in dark cyan and peptidoglycan dark yellow.

The periplasmic domains of T2S L (PDB ID: 2W7V) and M (PDB ID: 1UV7) both fold into ferredoxin-like domains and are known to bind each other[41]. We used a T2S M dimer structure (PDB ID: 1UV7) to first generate a T2S ML heterodimer and then used SymmDock to produce several hexameric ring models of the T2S ML heterodimer. We selected the model that best matched the lower-periplasmic ring-density. To build a working model for the pseudopilus, we used a cryoEM structure of the major pseudopilin filament (PDB ID: 5WDA), a co-crystal structure of the minor pseudopilin complex T2S IJK (PDB ID: 3CI0) and a crystal structure of the minor pseudopilin T2S H (PDB ID: 2KNQ) and their known connectivities and interfaces[12,22,61]. Since in our tomograms we only found pseudopili extending from the IM to the base of the secretin channel, we found that ~5 copies of major pseudopilin (T2S G) and one copy of each of the minor pseudopilins (T2S I, J, K, H) were sufficient to extend this distance.

To begin to generate a working model for the cytoplasmic complex, we modelled a T2S F dimer structure based on the proposed structure of the T4aP homologue PilC[34] and placed this model in the cytoplasmic dome density. The T2SS ATPase T2S E has three distinct domains (N-terminal domains N1E, N2E and a C-terminal ATPase domain CTE) connected by two extended flexible linkers. We used a hexameric structure of the N2E+CTE domains of T2S E (PDB ID: 4KSS) and a co-crystal structure of the cytoplasmic domain of T2S L and the N1E domain of T2S E (PDB ID: 4PHT) to build working models for the cytoplasmic disk and ring densities, respectively. The cytoplasmic ring is 20 nm wide in diameter in our average. Given the experimentally determined stoichiometry of six T2S L molecules per secretin channel, we first used SymmDock to generate candidate ring assemblies of six 4PHT complexes. Interestingly, out of 5000 candidate ring models, none had a diameter more than 17 nm. Our interpretation of this result is that the T2S L:N1E complex does not form a continuous ring in situ. Rather we simply placed six separate copies of 4PHT around the cytoplasmic ring with gaps in between. The 4KSS structure fits nicely into the cytoplasmic disk as expected. Finally, T2S O is an IM peptidase that processes the pseudopilins prior to their assembly. This protein has no significant domain outside the membrane and is not visible in our subtomogram average. Therefore, we did not place this protein in our architectural model.

Cryo-EM data collection, refinement and validation statist