Structural modeling of the flagellum MS ring protein FliF reveals similarities to the type III secretion system and sporulation complex

Sequence mining, analysis and alignment

All protein sequences were identified in the National Centre for Biotechnology Information protein database RefSeq (Pruitt et al., 2012). Multiple sequence alignments were generated with ClustalW (McWilliam et al., 2013) using default parameters. For secondary structure-based multiple alignments, a composite alignment was generated manually based on the individual pairwise alignments. Alignment figures were produced with ESPript (Gouet, Robert & Courcelle, 2003).

Secondary structure elements, signal sequences, transmembrane helices and structural homologues were predicted with the PSIPRED server (Buchan et al., 2013). Protein sequence identity was calculated with the Needleman-Wunsch algorithm on the EBI server (Li et al., 2015), using default parameters. Signal sequences were predicted with SignalP 4.1 (Petersen et al., 2011).

Modeling and refinement

Structure-based alignment and initial models were obtained with Phyre (Kelley et al., 2015), and the models were further refined by performing 1,000 cycles of the Relax procedure (Rohl et al., 2004) in Rosetta 3.4 (Leaver-Fay et al., 2011), using the following flags:

• database ∼rosetta/rosetta_database

• in:file:s input.pdb

• in:file:fullatom

• relax:thorough

• nstruct 1000

• out:file:silent relax.silent

The geometry of the obtained models was analyzed with the PSVS suite (Bhattacharya, Tejero & Montelione, 2007).

EM map docking and symmetry modeling

The flagellum, injectisome, and FliF EM maps (EMDB ID 1887, 1875 and personal communication from K Namba), were docked with the MatchMaker tool in Chimera (Goddard, Huang & Ferrin, 2007). Models of the FliF RBMs were placed in their putative location of the FliF EM map manually using Chimera.

For the EM-guided symmetry modeling, 24-fold, 25-fold and 26-fold symmetry definition files were generated, and used for the rigid-body step of the EM-guided symmetry modeling procedure described previously (Bergeron et al., 2013). Briefly, the individual domains were manually placed in the corresponding region of the EM map, and 1,000 rigid-body decoy models were generated with imposed symmetry and with a restraint for fit into the FliF EM map density, at 22 Å resolution. The obtained models were isolated with the Cluster procedure in Rosetta, and scores calculated with the Score procedure using the lowest-energy model as a template for RMS calculations.

The flags used for the modeling procedure are listed in the Supplementary Methods section.

Characterization of the FliF domain organization

In order to identify conserved features, I gathered FliF sequences from a number of human pathogens, spanning gram-negative, gram-positive and spirochete bacteria. Most of the FliF sequences are similar in length (∼560 amino acids), but show limited sequence conservation (Table 1). I next used sequence analysis servers to predict secondary structure elements and other structural and functional features. Two transmembrane (TM) helices are predicted, between residues 20–45 and residues 445–470, and a secretion signal peptide targeting it for secretion and inner-membrane localization (Kudva et al., 2013) is predicted at the N-terminus. All FliF sequences show a very similar secondary sequence prediction pattern (Fig. S1), and similarity to SctJ was identified for residues 50–220 in all sequences (hereafter referred to as the SctJ homology domain), but no structural homologues were found for residues 220–440 (FliF-specific domain). Within this region, residues 305–360 of the FliF-specific domain are predicted unstructured in all sequences.

I then generated a multiple alignment of all FliF sequences, and used it to map the predicted secondary structure and identified domains (Fig. 2). This further illustrates the conserved domain organization in all FliF orthologues. The two RBMs of the SctJ homology domain (labeled RBM1 and RBM2) are well conserved, as is the linker L1 between these. It has been shown that in SctJ this linker plays a role in ring assembly (Bergeron et al., 2015), suggesting that this may also the case in FliF. In contrast, the linker region L2, separating the SctJ homology domain to the FliF-specific domain, is highly variable.

The Chlamydiacae FliF-FliG fusion protein

While the FliF domain organization described above was found in most FliF sequences, one notable exception was identified, for the Chlamydiaceae family, where the FliF sequence is notably shorter (∼330 amino acids). Sequence analysis and multiple sequence alignment from all available Chlamydiaceae homologues (Fig. 3A) revealed that the N-terminal signal sequence and the two TM (residues 16–33 and 250–275) are present, but the periplasmic region is significantly shorter. Residues 60–145 show some sequence similarity to SctJ, but only encompassing RBM2. No structural homologues could be identified for residues 165–235, however the predicted secondary structure matches that of the canonical RBM.

In most orthologues, the C-terminus cytosolic region of FliF (residues 520–560) binds to the protein FliG (Levenson, Zhou & Dahlquist, 2012; Ogawa et al., 2015), a ∼37 KDa protein possessing three domains, labeled N, M and C. FliF interacts with domain N, while both M and C bind to the C-ring component FliM (Brown et al., 2007; Minamino et al., 2011), as illustrated on Fig. 3B. However, a sequence similarity search revealed that in the Chlamydia FliF, the cytosolic region is actually homologous to the M region of FliG (not shown), revealing that in this species the protein is in fact a FliF-FliG fusion. Chlamydia is not thought be a flagellated bacterium, and possesses only a few flagellar genes, namely FliF, FliL and FlhA. It does however possess a functional injectisome that is essential for virulence (Peters et al., 2007), and it has been shown that Chlamdia flagellar proteins interact with components of its injectisome (Stone et al., 2010). FliM (and FliN, another C-ring component) is homologous to the Chlamydia injectisome component SctQ (Notti et al., 2015). It is therefore possible that the FliF-FliG fusion interacts with SctQ (Fig. 3B). While the Chlamydia injectisome possesses both SctJ and SctD homologues (Nans et al., 2015), expression of fliF was shown to be concomitant with that of the injectisome (Hefty & Stephens, 2007). It remains to be tested experimentally if FliF is included in the Chlamydia injectisome.

Interestingly, a FliF-FliG fusion is not entirely unprecedented, as two artificial FliF-FliG fusions have been reported in S. typhimurium (Francis et al., 1992; Thomas, Morgan & DeRosier, 2001). In both cases, the fusion does not impair flagellum assembly and rotation, but induces a bias in the rotation direction. Since it is not known if the Chlamydia FliF-FliG fusion protein is part of a proto-flagellum, or contributes to the injectisome, the implications for this observation is not clear, but suggests that such a fusion protein may be functional.

Structural modeling of the SctJ homology domain

Structures of the periplasmic domains of both SctJ from EPEC (Crepin et al., 2005; Yip et al., 2005) and S. typhimurium (Bergeron et al., 2015) named EscJ and PrhK respectively, have been reported. Exploiting this information, a structural model for FliF was generated using the prototypical S. typhimurium FliF sequence. Despite the predicted structural homology, the sequence conservation between FliF and SctJ is low (Fig. 4A). I therefore employed the secondary structure alignment-based server Phyre (Kelley et al., 2015) for modeling the SctJ homology domain, spanning residues 50–221. However, the relative orientation of the two RBMs in FliF is not known, and may differ from that of SctJ. I therefore modeled the two RBMs independently, and refined the obtained models with Rosetta (see ‘Materials and Methods’ for details). As shown on Figs. 4B and 4C, both RBMs converged to a local energy minimum within an RMSD of 1 Å to the lowest-energy model in the refinement procedure. The models possess good geometry (Table 2), and their overall architecture (Figs. 4D and 4E) is expectedly similar to that of SctJ.

The FliF-specific domain is a RBM

I next sought to generate a structural model for the FliF-specific domain (residues 228–443). As mentioned above, secondary structure prediction indicated that residues 228–309 and residues 356–443 possess defined secondary structure, while residues 310–355 are predicted as intrinsically disordered (Fig. 2). I therefore hypothesized that the FliF-specific region consists of two globular domains (D1 and D2) separated by a flexible linker. I then attempted to identify structural homologues to these two regions, using the Threading server PsiPred (Buchan et al., 2013). Surprisingly, D1 is predicted to have structural homology to SctJ, although only for the first two secondary structure elements (residues 229–268, Fig. S2A). Similarly, D2 was identified to have structural homology to the sporulation complex protein SpoIIIAH, which also possesses a RBM fold and has been proposed to oligomerize into ring structures (Levdikov et al., 2012; Meisner et al., 2012) (Fig. S2B). However, the structural homology was limited to the last three secondary structure elements (residues 386–436). Based on these observations, I postulated that the FliF-specific domain is a “split” RBM that possesses a large insert in the loop between the first and second strands (Fig. 5A). A secondary structure similarity search for the FliF-specific domain with this insert removed (FliF_{228 –443}Δ274–378) confirmed overall fold similarity to both SctJ and SpoIIIAH (Fig. S2C).

A “split” RBM is not unprecedented, as it is also predicted in the sporulation complex component SpoIIIAG (Fig. 5A). In both proteins an insert is predicted between the first and second strand of the RBM, with four putative β-strands in the inserted domain. This insert is large enough to accommodate a small globular domain; alternatively, it is possible that the insert adopts an extended loop conformation stabilized by β-strands.

In order to build an atomic model for the FliF RBM3, the SctJ-derived model for FliF_{230 –275} was combined with the SpoIIIAH-derived model for FliF_{380 –440}, and the resulting structural model was refined with Rosetta. As shown on Fig. 5B, the RMSD to the lowest energy model is significantly higher than for the RBM1 and RBM2 models (around 3.5 Å for most decoys), which is perhaps expected, as the starting model was a presumably poorer, composite model. Nevertheless there is a clear energy funnel, indicative that the modeling procedure is converging. The obtained FliF RBM3 model, shown on Fig. 5C, possesses the canonical RBM fold, with good overall geometry (Table 2) and an elongated architecture similar to the RBM2 of SctJ and SpoIIIAH, with the insert between strands 1 and 2 located on one side of the structure.

Localization of the FliF domains

Previous EM studies have shown that in isolation, FliF forms a doughnut-shaped oligomer with two side rings corresponding to the MS rings seen in the intact basal body (Suzuki et al., 1998; Thomas et al., 2006). The FliG-binding domain forms the cytoplasmic M ring, but little is known about the organization of the periplasmic domains. However, the organization of the injectisome basal body, and that of SctJ in particular, is well characterized (Schraidt & Marlovits, 2011; Bergeron et al., 2015). By comparing the EM reconstructions of the flagellar (Thomas et al., 2006) and injectisome (Schraidt & Marlovits, 2011) basal body complexes, localization of the FliF RBMs can be inferred. As shown on Fig. 6A, density attributed to the two globular domains of SctJ has clear equivalence in the flagellum, suggesting this location corresponds to the SctJ homology domain of FliF. EM map density for the flagellum S-ring corresponds to the injectisome protein SctD localization, and is also present in purified FliF (Suzuki et al., 1998), confirming that it is not attributed to another flagellar component. The periplasmic region of SctD is composed of three domains with an RBM fold (Spreter et al., 2009; Bergeron et al., 2013), including the region corresponding to FliF density, suggesting that in the flagellum, this density can be attributed to the FliF-specific domain, which also possesses an RBM fold.

Based on these observations, I propose that FliF is akin to a SctJ-SctD fusion (Fig. 6B), including the two RBMs of SctJ and its C-terminal TM helix, and the N-terminal TM and first RBM from SctD. This likely explains a major difference between flagellar basal body, where FliF assembles on its own in the inner membrane, and injectisome basal body, where co-expression of SctJ and SctD is required (Kimbrough & Miller, 2000).

Further sequence analysis will be required to decipher the exact evolutionary pathway leading to the emergence of two genes in the injectisome. It is however interesting to note that in the ancestral Myxococcus injectisome, the SctD homologue lacks a periplasmic domain, and only consists in the cytoplasmic FHA-fold domain found in all SctD homologues, Bergeron et al. (2013) followed by a predicted transmembrane domain. It is therefore likely that this gene was fused to a FliF duplication gene in subsequent injectisomes, leading to the SctD homologue including the FHA cytosolic domain and the three BRMs in the periplasm.

One can wonder how having two separate proteins is beneficial for the injectisome. It may allow for a more dynamic assembly/disassembly process. Indeed, most injectisomes are only transiently functional, and while the regulation of injectisome genes transcription is well characterized, the fate of the injectisome complex after its required function (such as, for example, the SPI-1 injectisome of S. typhimurium after insertion in the Salmonella-Containing Vacuole) is not known. It is possible that having two proteins allow for disassembly of the basal body and recycling of the corresponding proteins (Diepold & Wagner, 2014).

It is also interesting to note that unlike SctJ and FliF, SctD lacks a secretion signal at its N-terminus. Indeed, deletion of the complete N-terminus of the S. typhimurium SctD homologue allows the formation of intact basal body complexes (Bergeron et al., 2013). It remains to be shown how the periplasmic domain of this protein is translocated across the inner membrane, but co-export with SctJ is an intriguing possibility, and to my knowledge such co-secretion via the Sec pathway (through which both FliF and SctJ are thought to be translocated) is unprecedented.

Modeling of the FliF oligomer

I next exploited the domain localization proposed above, the structural models of the three FliF RBMs, and the previously determined EM map (Suzuki, Yonekura & Namba, 2004), to generate a structural model of the FliF monomer. To that end, I positioned all three domains so that their termini point towards the correct region of the map, and fitted into the EM map density with Chimera (Goddard, Huang & Ferrin, 2007). Figure 7A shows that RBM1 is located near the inner-membrane region, while RBM2 forms the neck of the structure. Finally, the RBM3 is located in the region of density forming the S-ring. The insert in RBM3 points towards the lumen of the ring, and could correspond to the gate density observed in the FliF EM structure (Suzuki et al., 1998; Suzuki, Yonekura & Namba, 2004).

I next attempted to generate a model for the FliF oligomer. I applied a previously described EM-guided modeling procedure (Bergeron et al., 2013) to all three domains individually, using the FliF map as a restraint, and applying 24-, 25- or 26-fold symmetry. In most cases the procedure led to ring models that were too large for the EM map (not shown). This is unsurprising, as it had been reported previously that the procedure requires experimental structures (as opposed to homology models) to converge (Bergeron et al., 2013). The one exception was the modeling of the RBM3, using 25-fold symmetry. This procedure generated three distinct clusters of models with clear energy funnels and good fit to EM density (Fig. 7B). Close inspection of the clusters reveals that they correspond to similar oligomerization modes and use the same interface, but with a slightly different angle between the subunits (Fig. 7C). I therefore exploited this 25-mer mode for the RBM3 to generate a complete 25-mer model for FliF. The corresponding model, shown in Fig. 7D and included in the Supplemental Information, matches well with the cryo-EM structure, with only a few loop regions of RBM2 and RBM3 located outside of the density.

I emphasize that while only 25-fold symmetry led to a convergent model for RBM3, this should not be considered as evidence that FliF forms a 25-mer. Indeed, the limited amount of data used in the modeling procedure is not sufficient to generate an accurate structural model, and at the resolution of the EM map (∼24 Å) the difference between 25 and 26 subunits is indistinguishable. It is also possible that the overall diameter of the complex appears smaller due to a small error in the pixel size of the detector used, thus biasing the modeling process towards a lower oligomeric state. However, the converging energy landscape for RBM3, and fit to EM density for all three domains, suggest that the reported model may be exploiting the native general orientation and oligomerization interface for all three domains. Additional experimental data, including experimentally determined RBM1, RBM2 and RBM3 atomic structures, as well as a higher resolution EM map of FliF and of the intact flagellum basal body, will be required to further refine the FliF model.

Despite these significant limitations, several aspects of the FliF model are worth commenting on. Firstly, the oblong shape of RBM2 forms an angle close to perpendicular to the membrane plane (Fig. 7A). In contrast, this domain is almost to parallel to the membrane plane in SctJ (Yip et al., 2005; Schraidt & Marlovits, 2011; Bergeron et al., 2013). While the proposed orientation of FliF RBM2 is driven by the EM map density of the isolated FliF oligomer, it is possible that it may differ in the intact flagellum basal body, and correspond to a structural rearrangement upon recruitment of other flagellar components (rod and/or LP ring). Higher resolution EM reconstruction of the full flagellum basal body will be required to identify structural rearrangements associated with flagellar assembly.

It is similarly noteworthy that in the FliF model, RBM2 is located approximately 20 Å away from RBM1, unlike SctJ where the two domains form a direct interaction (Bergeron et al., 2015). The later domain arrangement was shown to be mediated by the linker region, with in particular a conserved phenylalanine residue playing essential role in oligomerization (Phe 72 in the S. typhimurium SctJ homologue PrgK). Interestingly, the linker between RBM1 and RBM2 is also well conserved in FliF (linker L1 in Fig. 2), with in particular a Phe found in most orthologues at position 121 (and replaced with a different large hydrophobic residue in some species), which could perform a similar role. It is therefore likely that this linker plays a role on FliF assembly.

Finally, it is interesting to note that in the FliF model, RBM1 and RBM3 are in close proximity, and would likely form direct interactions. Considering the limitations of the modeling procedure described above, it would be premature to use this model to identify residues that are part of the interaction interface. Nonetheless, continuous density in this region of the injectisome EM map (Schraidt & Marlovits, 2011) suggests that a direct interaction does exist between these two domains, thus supporting a similar interaction in FliF.