Architecture of the flagellar rotor

Abstract

Rotation and switching of the bacterial flagellum depends on a large rotor-mounted protein assembly composed of the proteins FliG, FliM and FliN, with FliG most directly involved in rotation. The crystal structure of a complex between the central domains of FliG and FliM, in conjunction with several biochemical and molecular-genetic experiments, reveals the arrangement of the FliG and FliM proteins in the rotor. A stoichiometric mismatch between FliG (26 subunits) and FliM (34 subunits) is explained in terms of two distinct positions for FliM: one where it binds the FliG central domain and another where it binds the FliG C-terminal domain. This architecture provides a structural framework for addressing the mechanisms of motor rotation and direction switching and for unifying the large body of data on motor performance. Recently proposed alternative models of rotor assembly, based on a subunit contact observed in crystals, are not supported by experiment.

Introduction

The flagellar motor of bacteria is a rotary device (Berg and Anderson, 1973) energized by the membrane ion gradient (Larsen et al, 1974; Glagolev and Skulachev, 1978). Its mechanism has been studied for >30 years and a great deal is known about the performance of the motor under various circumstances (for a review, see Sowa and Berry, 2008). The identities and stoichiometries of components that form the various substructures in the flagellum are also well established (Macnab, 2003). The stators consist of the membrane proteins MotA and MotB, which form complexes with subunit composition MotA₄MotB₂ (Sato and Homma, 2000; Kojima and Blair, 2004). Each motor has several independent stator complexes that function to conduct the energizing ions (Blair and Berg, 1990) and harness ion movement to rotation. In a current model, protons move on and off a conserved aspartate residue in MotB (Zhou et al, 1998b), to drive conformational changes that apply torque to the rotor (Kojima and Blair, 2001). The rotor and stator engage in electrostatic interactions that involve conserved charged residues in a cytoplasmic domain of MotA and in the C-terminal domain of FliG (Lloyd and Blair, 1997; Zhou and Blair, 1997; Zhou et al, 1998a; Yakushi et al, 2006).

Flagellar rotation and direction is controlled by a large protein assembly on the rotor called the switch complex (Yamaguchi et al, 1986a). It is formed from the proteins FliG, FliM and FliN, each present in many copies, and corresponds structurally to the C-ring of the flagellar basal body (Francis et al, 1992, 1994; Zhao et al, 1996a, 1996b; Thomas et al, 2006; Figure 1). The lower part of the switch complex is formed from FliN and the FliM C-terminal domain (FliM_C); FliN is organized in doughnut-shaped tetramers that alternate with the FliM_C domains in an array at the membrane distal region of the C-ring (hereafter referred to as the ‘bottom’) (Brown et al, 2005; Paul and Blair, 2006; Thomas et al, 2006; Sarkar et al, 2010b). The switch complex functions in flagellar assembly as well as in rotation (Yamaguchi et al, 1986b). FliN interacts with components of the type III secretion apparatus housed in the basal body (Gonzalez-Pedrajo et al, 2006; McMurry et al, 2006; Paul et al, 2006) and may facilitate assembly by assisting in the delivery of protein subunits that form exterior parts of the structure (the filament and hook). FliN is also critical for direction switching and contains a binding site for the signalling molecule phospho-CheY that promotes clockwise (CW) rotation (Sarkar et al, 2010a). The thinner side-wall of the C-ring, above the FliN₄FliM_C array, is formed from the FliM middle domain (FliM_M; Park et al, 2006; Brown et al, 2007). Mutational analyses indicate that FliM has a large role in direction switching (Sockett et al, 1992) and its N-terminal domain, which is predicted to have an extended conformation, binds phospho-CheY (Welch et al, 1993; Lee et al, 2001). FliG is proximal to the membrane (hereafter referred to as the ‘top’ position) and comprises three domains (Irikura et al, 1993; Lloyd et al, 1996; Lee et al, 2010), each with distinct functions: The N-terminal domain (FliG_N) interacts with the FliF protein which forms the MS-ring, the middle domain (FliG_M) interacts with FliM, and the C-terminal domain (FliG_C) contains a set of conserved charged residues that interact with charged residues in the cytoplasmic domain of MotA (Lloyd and Blair, 1997; Zhou et al, 1998a; Yakushi et al, 2006).

Figure 1.

Hypotheses for FliG organization in the flagellar rotor. (A) The flagellar basal body of wild-type Salmonella (Thomas et al, 2001). The dashed box indicates the region shown magnified in the other panels. (B) FliG-domain arrangement discussed by Thomas et al (2006). (C) Hypothesis of Brown et al (2007). (D) Arrangement based on a FliG_M–FliG_C contact observed in crystals. The contact is postulated to involve either two different FliG subunits (Lee et al, 2010) or a single FliG subunit (Minamino et al, 2011).

Current electron microscopic reconstructions of the flagellar basal body are highly detailed and provide strong constraints on the overall shape of the switch complex (Figure 1; Thomas et al, 2006). Whereas it is clear that FliG must lie at the top of the C-ring to enable interaction with the stator, presently there is no consensus regarding the assignment of specific FliG domains to the features observed in electron micrographs. Thomas and co-workers suggested that FliG_C might correspond to the inner lobe of density at the top of the C-ring, with the other parts of FliG falling in the bottom part of the MS-ring (Figure 1B). Brown and co-workers favour an assignment of FliG_C to the outer lobe of density at the top of the C-ring, and the N-terminal and middle domains of FliG to the inner lobe (Figure 1C). Lee et al (2010) and Minamino et al (2011) take this a step further and propose that both the middle- and C-terminal domains of FliG lie in the outer part of the C-ring, with the N-terminal domain alone accounting for the inner lobe (Figure 1D). Their models are based on an interaction between FliG_C and FliG_M observed in FliG crystals, which was judged to be biologically relevant on the grounds that it involves surfaces with conserved hydrophobic character and occurred in more than one crystal form. Experimental tests of the crystal contact-based model have not been reported.

The known features of switch-complex organization have been deduced from crosslinking and mutational experiments guided by structures of the individual components (Park et al, 2006; Brown et al, 2007; Sarkar et al, 2010b). The lower part of the C-ring has been studied most fully; systematic disulphide crosslinking studies of this region produced a structural model that fits well with the EM reconstructions (Thomas et al, 2006), and that additionally revealed a subunit movement that occurs upon CW/CCW direction switching (Sarkar et al, 2010b). Similar information on the upper part of the switch complex is needed to provide a structural framework for understanding motor rotation and switching. Uncertainties regarding the arrangement of FliG in particular must be addressed, as it is the component that functions directly in rotation. Herein, we report a range of experiments that culminate in a specific, firmly grounded model of FliG and FliM organizations. The crystal structure of a complex between major domains of FliG and FliM is described, together with several biochemical and mutational experiments to probe their arrangement within the flagellar motor. The results support an architecture like that proposed by Brown et al (2007). The crystal contact-based models (Lee et al, 2010; Minamino et al, 2011) are not supported; several results indicate that the FliG_M–FliG_C contact observed in crystals does not occur within the flagellum and that FliG_C interacts directly with FliM_M instead. The new structural model provides a long-sought framework for addressing molecular details of the motor mechanism.

Results and discussion

Structure of a FliG_M:FliM_M complex

In the model of Thomas et al (2006) only the C-terminal domain of FliG resides within the C-ring, whereas the FliG_N and FliG_M domains lie within the lower part of the MS-ring (Figure 1B). This model would thus preclude a direct interaction between FliG_M and FliM, because FliM is believed to lie within the C-ring. However, our previous binding studies revealed an interaction between FliG_M and FliM, occurring through a conserved ‘EHPQR’ surface motif on FliG_M and a conserved ‘GGXG’ motif in the middle domain of FliM (Mathews et al, 1998; Brown et al, 2002, 2007). To examine this interaction further, we determined the crystal structure of a complex of the FliG_M and FliM_M domains, using thermostable proteins from Thermotoga maritima. The T. maritima proteins show high sequence similarity to those of Escherichia coli, including conservation of hydrophobic character at core positions, and in some cases can partially complement E. coli mutants (Lloyd et al, 1999; Brown et al, 2002, 2005). Thus, the T. maritima proteins provide relevant structural models for the proteins from the otherwise better-characterized enteric species.

The co-crystal structure (Figure 2) shows FliM_M and FliG_M domains essentially like the previously described individual structures (Brown et al, 2002; Park et al, 2006), except that a helix at the C-terminus of FliG_M is shortened and packed more closely against the body of the domain than in the previous FliG_MC structure (Brown et al, 2002) or in the just-reported structure of a 3-residue deletion variant of FliG_MC (Minamino et al, 2011). The close-packed conformation of this helix is stabilized in part by contacts with FliM_M in the complex (Figure 2E), and is similar to what is observed in the Aquifex aeolicus FliG structure (Lee et al, 2010). Most of the FliG_M:FliM_M interface is formed from the EHPQR motif of FliG and the GGXG motif and adjacent regions of FliM, as was suggested by the binding and mutational studies (Mathews et al, 1998; Brown et al, 2007). These motifs generate an interface from the inter-digitation of two surface loops that link helices on both proteins. Within the contact, FliM Met131, which immediately precedes the FliM GGXG motif, inserts into a hydrophobic patch composed of FliG Val172, Val176 and Val133 (T. maritima numbering is used here in the discussion of the T. maritima protein structures). Gln129 of the FliG EHPQR motif hydrogen bonds to the peptide backbone of the FliM GGXG motif, and FliG-His127 contacts FliM-Tyr124 and hydrogen bonds to FliM-Asp128. FliM-Asp128, whose conformation is also stabilized by interaction with the GGXG motif, further sets the interface orientation by forming a highly conserved salt bridge with FliG-Arg161. Interactions between conserved residues at the top of FliM α1′ (Thr144 and Ile146) and the end of FliG_M αA (Phe 122 and Glu 126) also contribute to the association. In total, the FliM_M–FliG_M contact buries ∼750 Ų surface area per subunit, with a calculated free energy of association ΔG=−6.3 kcal mol⁻¹, hydrophobic surface specificity of 0.32 and surface complementarity of 0.51 (Lawrence and Colman, 1993; Krissinel and Henrick, 2007). These parameters reflect an interface of medium affinity (ca. >μM) often characteristic of binding partners that associate and dissociate as part of their function (Park et al, 2004).

Figure 2.

Structure of the T. maritima FliM_M:FliG_M complex. (A) Overall shape of the complex. The N- and C-termini of FliM_M are oriented towards the bottom in this view; these parts of FliM are directed towards the bottom of the C-ring in the flagellar basal body (Park et al, 2006; Sarkar et al, 2010b). (A stereo version of this figure is provided in Supplementary data.) (B) The FliG_M:FliM_M interface. The EHPQR residues of FliG and the GGXG motif of FliM are indicated. Orange circles mark positions where Cys residues were introduced to confirm the interaction by crosslinking. Numbers are for the E. coli protein. Unbiased electron density for the EHPQR and GGXG motifs is shown in Supplementary data. (C) Crosslinking through the introduced Cys residues. Crosslinking was induced using Cu-phenanthroline. (D) Packing of the helix near the C-terminus of FliG_M against the body of the domain. The helix is shown in lighter colour. In the previous crystal structure of FliG_MC (Brown et al, 2002), this helix is detached from the domain and makes extensive inter-subunit crystal contacts instead. (E) Hydrophobic contacts between the helix and FliM_M that stabilize the close-packed conformation of the helix.

As a further check that the interaction seen in the co-crystal structure is relevant to the protein arrangement in the motor, Cys residues were introduced in the E. coli FliG and FliM proteins, at positions that are close in the crystal structure (FliG-129 and FliM-149, in E. coli numbering), and oxidative crosslinking was induced in cells. The Cys-substituted proteins crosslinked in high yield (Figure 2C). Taken together with the binding and mutational studies, this result indicates that the FliG_M–FliM_M arrangement observed in the crystal resembles that actually occurring in the motor. Switch-complex models with FliG_M in the MS-ring, and thus removed from FliM (Figure 1B), are therefore unlikely.

Interaction between FliM and FliG_C

The binding study that identified the FliM–FliG_M interaction also gave evidence of a binding interaction between FliM and FliG_C. The interaction involves a conserved hydrophobic patch on the surface of FliG_C opposite the stator-interaction site (Brown et al, 2007). Mutations in the FliG_C hydrophobic patch, like mutations in the FliG_M EHPQR motif, were found to weaken the FliG–FliM binding. Stock and co-workers (Lee et al, 2010) re-interpreted these findings to mean that FliG binds FliM through a surface composed jointly from the FliG_M and FliG_C domains, which were assumed to associate together in the motor in the same way as occurs in the FliG crystals. The FliG_M:FliM_M structure shows, however, that FliG_M has a FliM-binding surface distinct from the surface that associates with FliG_C in the crystal contact (Figure 2; Supplementary Figure S1). The disruption of FliM binding by the hydrophobic-patch mutations is, thus, more readily explained in terms of a direct interaction between FliM and FliG_C (Brown et al, 2007), and appears incompatible with models based on the FliG_M–FliG_C crystal contact (Lee et al, 2010; Minamino et al, 2011), where FliG_M intervenes between FliG_C and FliM. The hydrophobic-patch mutations were studied in the context of the full-length FliG protein, however, which might complicate the interpretation in terms of the individual domain interactions. To characterize the FliM–FliG_C interaction more directly, we expressed FliG_C (consisting of residues 185–331) as a separate domain and tested its binding to FliM in a pull-down assay. The separately expressed FliG_C domain showed clear binding to FliM (Figure 3). Using collections of surface-residue mutations, the interaction was mapped to the hydrophobic patch of FliG_C and the GGXG motif and adjacent regions on FliM_M (Figure 3; Supplementary Figure S2). The FliM mutations that weakened the FliG_C–FliM_M interaction also disrupted function, as assayed either by motility in soft agar or by export of flagellin (Supplementary Figure S2). The co-crystal structure shows that the GGXG motif also binds FliG_M; thus, essentially the same part of FliM is involved in interactions with both FliG_M and FliG_C (Supplementary Figure S2) and loss of function in the GGXG mutants might reflect disruption of either or both contacts.

Figure 3.

Interaction between FliM and FliG_C detected in GST pull-down assays. (Representative results are shown; see Supplementary Figure S2 for additional data.) Blots were probed with anti-FliM antibody. (A) Effects of FliG_C mutations on the binding to FliM. Positions where mutations eliminated binding are coloured red; black indicates positions where mutations had no effect. (B) Effects of FliM mutations on binding to FliG_C. Colouring as in panel (A), plus orange to indicate positions where binding was weakened.

We used disulphide crosslinking to verify the occurrence of the FliM_M–FliG_C interaction in cells and to obtain constraints on its geometry. Single-Cys residues were introduced at seven positions in the E. coli FliG_C protein in the vicinity of the hydrophobic patch, and six positions in the E. coli FliM_M protein near the GGXG motif, and all the pairwise combinations were studied. Oxidative crosslinking was induced in cells and products were examined on immunoblots. Several Cys pairs gave reproducible, moderately strong FliG–FliM crosslinking (Figure 4A; Supplementary Tables 3 and 4). A model for the FliM_M:FliG_C complex, constructed by bringing into proximity the highest yielding pairs (Figure 4B), places the hydrophobic patch of FliG_C in contact with residues on FliM_M with conserved strongly hydrophobic (Val127, Phe131 and Val153) or partially hydrophobic (Thr149) character (amino-acid residues and numbering are for the FliM protein of E. coli). One of the high-yielding Cys pairs also exhibited intergenic suppression: the Cys replacement at FliG residue 225 caused a complete loss of motility that was substantially rescued by the Cys replacement at residue 149 of FliM (Supplementary Figure S3). The size and shape of the FliM_M:FliG_C assembly provide an acceptable match to features observed in the upper part of the C-ring in EM reconstructions (Supplementary Figure S3B).

Figure 4.

Crosslinking experiments to probe the FliM_M–FliG_C relationship. (A) Positions of Cys replacements and summary of the crosslinking results. Dotted blue lines connect Cys pairs of residues that formed disulphide crosslinks, with the thickness of the line indicating relative yield. Representative gels are shown below; blots were probed with anti-HA antibody. The red dashed line connects a Cys pair that, in addition to crosslinking, showed mutational suppression (see the text and Supplementary Figure S3A). (B) Model for the FliM_M–FliG_C assembly based on the crosslinking results. The highest yield Cys pairs are indicated. (C) Tests of the crystal contact-based model for FliG organization. The 117/166 Cys pair in FliG_M was shown previously to crosslink efficiently (Lowder et al, 2005) and is included as a positive control. The 159/218 and 162/196 Cys pairs are in close proximity in the crystal contact model (Lee et al, 2010; see Supplementary Figure S4 for an illustration). These failed to crosslink, using either Cu-phenanthroline (shown) or iodine (data not shown).

FliG_N and FliG_M are in proximity

The binding and crosslinking results establish that FliG_C interacts directly with FliM_M, and thus argue against the FliG_C–FliG_M interaction that has been postulated on the basis of crystal contacts (Lee et al, 2010; Minamino et al, 2011). Complicated architectures involving both types of interaction might still be imagined, and so we introduced Cys pairs at positions that are in close proximity in the crystal contact models and tested for disulphide crosslinking in cells. Two Cys pairs were made, both of which are predicted to allow close approach of the sulphur atoms (van der Waals distance or nearer; see Supplementary Figure S4). One pair (residues 162/196; E. coli numbering) retained about half of wild-type function in a soft agar motility assay, while the other (159/218) functioned at about 10% of wild type. Neither Cys pair showed detectable crosslinking in cells, using either Cu-phenanthroline (Figure 4C) or iodine as oxidizing agents (data not shown). The bifunctional reagent bis-maleimidohexane (BMH) that can bridge more distant thiols was also tried, and also showed no crosslinking through these positions (data not shown).

These results indicate that FliM_M, and not FliG_M, is located under FliG_C in the outer part of the C-ring, in accordance with the proposal of Brown et al (2007). FliG_M must then occupy a more-inward location, nearer FliG_N (as in Figure 1C). The FliG_N and FliG_M domains are widely separated in the crystal structure of A. aeolicus FliG, but might adopt a more-compact conformation in the motor where the protein can engage in its normal interactions with FliM and FliF. In the A. aeolicus structure, both FliG_N and FliG_M, as well as the helix joining them, display sizable hydrophobic surfaces that appear to be stabilized by crystal contacts (Supplementary Figure S5). To test whether FliG_N and FliG_M actually lie near each other in the motor, we constructed three double-Cys mutants (31/146, 43/147 and 50/147), each with a replacement near an edge of FliG_M and an edge of FliG_N, and examined crosslinking in cells using the bifunctional reagent BMH. These Cys pairs are distant (Cβ-to-Cβ distances in the >30 Å range) in the rotor model of Lee et al (Supplementary Figure S4). The 43/147 Cys pair was crosslinked by BMH to form both dimer and trimer products. The corresponding single-Cys mutants either failed to crosslink (position 147) or formed dimer but none of the larger multimers (position 43) (Figure 5A). Previously identified crosslinks between FliG_M and FliG_M provide constraints on the relative orientation of the FliG_M domains (Lowder et al, 2005). If FliG_N is positioned in an appropriate orientation near FliG_M, the 43–157 crosslink can be accounted for while simultaneously satisfying the previous FliG_M–FliG_M constraints (Figure 5B). We conclude that the FliG_N and FliG_M subdomains are not widely separated as observed in the A. aeolicus crystal structure, but are in relatively close proximity in the flagellar motor of E. coli.

Figure 5.

Proximity of FliG_N to FliG_M. (A) Crosslinking of position 43 in FliG_N to position 147 in FliG_M by bis-maleimidohexane. Crosslinking was carried out at 23°C for 10 min. (B) A hypothetical arrangement of the FliG_N and FliG_M domains that could account for the observed FliG_N–FliG_M and FliG_M–FliG_M crosslinking. The FliG_N domain (residues 5–89) is pale-cyan and FliG_M (residues 104–184) is cyan. The segment linking the domains (residues 90–103) is yellow and the positions to which it would connect (carboxy-terminus of FliG_N and amino-terminus of FliG_M) are red and blue. The relative orientation of the FliG_M domains is based on a previous study (Lowder et al, 2005), which identified positions giving efficient FliG_M–FliG_M crosslinking; one such pair (117–166) is shown. The orientation of FliG_N, which is intended to be approximate only, is based on the observed FliG_N–FliG_M crosslink (A) and constraints imposed by the inter-domain connection (the length of the connecting helix and the positions it must connect). Spheres indicate positions of Cβ positions, (grey in FliG_N and black in FliG_M). Residue numbers are for the E. coli protein. For previously identified instances of crosslinking, including the indicated 117/166 Cys pair, see Lowder et al (2005).

Organization of torque-generating elements of the rotor

FliG binds to the MS-ring protein FliF (Oosawa et al, 1994; Kihara et al, 2000; Grunenfelder et al, 2003) and the available evidence indicates that both FliG and FliF are present in about 26 copies per motor (Jones et al, 1990; Francis et al, 1992; Sosinsky et al, 1992; Thomas et al, 2001, 2006; Suzuki et al, 2004). FliM is believed to be present in more, about 34, copies per motor (Thomas et al, 1999, 2006; Young et al, 2003). A subunit arrangement that can accommodate the different FliG and FliM copy numbers has been proposed (Brown et al, 2007). The dual FliM–FliG interactions that have been characterized here are key elements in the model. The FliM_M domains are proposed to occur in two kinds of structural setting. Most are in an approximately vertical orientation, forming the outer wall of the C-ring and interacting with the hydrophobic patch of FliG_C. A subset of FliM_M domains, typically 8 or 9 (equal to the number of FliM subunits present in excess over FliG subunits), are tilted slightly inward where they interact with FliG_M instead (Figure 6). FliG subunits are, therefore, also of two kinds; most are supported by a single FliM that is positioned under FliG_C and binds through the hydrophobic patch, while a subset (again about 8 or 9) is bound to two FliM subunits and is thus supported through both the middle- and C-terminal domains (Figure 6A).

Figure 6.

Structural model for the upper part of the C-ring. (A) Overall plan of FliG and FliM organizations. The arrangement is similar to that proposed by Brown et al (2007), with adjustments to reflect more-current information on FliG structure. FliM is light brown and FliG is cyan. (B) More detailed view of a section of the rotor. Colouring is as in (A), but with the three parts of FliG (FliG_NM, linking helix and FliG_C) coloured with increasing intensity, and the active-site ridge shown in atom colours to highlight the conserved charged residues that interact with the stator (Zhou et al, 1998a). The dashed line indicates the hypothesized path of the stator (relative to the rotor) as the motor turns (see the text). (C) Stereo-view (crossed-eye) of a section of the rotor. The view is in a roughly radial direction (out-to-in). The active-site ridge on FliG_C is coloured white.

The results here, combined with results in the previous structural study of FliM_M (Park et al, 2006), allow us to develop this structural model in detail. To construct the model explicitly, FliM_M domains were first positioned as they would be in a 34-member ring, in the relative orientation determined by crosslinking experiments of Park et al (2006). FliG domains (either FliG_C or FliG_NM, as dictated by the model) were positioned on top of the FliM_M domains, in the orientations determined in the FliM_M:FliG_M co-crystal structure (Figure 2) or by FliM_M:FliG_C crosslinking (Figure 4). The structure of the complex when placed in the rotor simultaneously satisfies crosslinking constraints between adjacent subunits of FliM_M and FliG_M. The helix that joins FliG_M and FliG_C was extended straight to residue 193 (just before the Gly-Gly linker that joins it to the C-terminal domain), as observed in the FliG_MC crystal structure (Brown et al, 2002). The subsets of FliM_M domains that are associated with FliG_NM were then tilted inward. This tilt was sufficient to bring the end of the linking helix close to the Gly-Gly linker of the adjacent FliG_C domain, to which it connects in the model. This matching of termini required no other assumptions but did depend on the linking helix assuming the roughly tangential orientation observed in the present co-crystal structure, rather than the roughly radial orientation observed in the previous T. maritima FliG_MC structure (Brown et al, 2002). The other FliG_NM domains (those not bound to FliM_M) were oriented similarly and were positioned to maintain, as nearly as possible, the same relationship with the attached FliG_C. Like the other elements in the assembly, the FliG_NM domains occur in slightly varied situations, in this case consisting of close groups of 3 or 4 separated by slightly larger gaps at the position of the inward-tilted FliM_M (Figure 6). All of the FliG_NM domains, including those not bound to FliM_M, would also be held in place by attachment to FliF. The protein subsets in different environments might have different stabilities within the structure, and consistent with this, Delalez et al (2010) recently reported that about two thirds of the FliM subunits in the motor are in relatively rapid exchange whereas the rest are more stably bound.

EM reconstructions indicate that features in the outer part of the C-ring have ∼34-fold symmetry, whereas the inner lobe has roughly 26-fold symmetry (Supplementary Figure S6). In the structural model, the symmetry transition occurs between FliG_NM (assigned to the inner lobe) and FliG_C (assigned to the outer lobe) (Figure 6). The linking helix connects to FliG_NM through an extended segment (residues 162–168 in E. coli numbering) that is relatively non-conserved, and to FliG_C through the aforementioned Gly-Gly motif. Either of these linkages might provide flexibility to accommodate the symmetry mismatch between FliG and FliM (see Supplementary Figure S9 for an illustration).

The structural model developed here agrees well with the electron microscopic reconstructions (Thomas et al, 2001, 2006). The unified FliG_NM domain has a size matching the inner lobe of density at the top of the C-ring (Supplementary Figure S7). The model also accounts for the effects of a deletion/fusion mutation removing large parts of FliG and FliF (Supplementary Figure S8). The ∼26-fold symmetry observed for the inner lobe in the reconstructions (Supplementary Figure S6) is well explained by an individual FliG_NM domain in each lobe. FliG_C can satisfactorily account for the outer lobe at the top of the C-ring (see Supplementary Figure S3 for shape comparison). While FliG is present in only about 26 copies, the 34-fold symmetry observed for the outer part of the C-ring arises because the FliG_C domains are held in position by the underlying FliM_M domains, which are present in 34 copies. The gaps in the upper edge of the C-ring (where FliM_M is tilted inward and FliG_C is absent) would have been obscured by the symmetry averaging of the EM reconstructions. The electron density in the outer lobes should reflect roughly three-fourths occupancy, and consistent with this, the outer lobe is substantially less intense than the inner lobe in most reconstructions (Figure 1; Supplementary Figure S3; Thomas et al, 2001, 2006).

To evaluate the consistency of our molecular model with the EM reconstructions, we built full representations of the upper C-rings composed from FliG and FliM_M and then averaged the outer lobes over 34-fold symmetry and the inner lobes over 26-fold symmetry. Electron density for the protomer subunits was then calculated to 20 Å resolution and compared with the EM maps (Supplementary Figure S6). Our model reasonably recapitulates the general shape of the electron density in the outer and inner C-ring lobes and the greater weighting of electron density in the inner lobe. In contrast, the FliG organization proposed by Lee et al (2010) does not fit the density or shape of the inner lobe as well, nor does it account for the varied symmetries in the C-ring (Supplementary Figure S6). Their model would also result in overlap between adjacent FliM subunits when the FliM_M domain is docked onto FliG_M according to the co-crystal structure (Supplementary Figure S6). The model of Minamino and co-workers (2011) was not developed in sufficient detail to allow for detailed comparison with the EM reconstructions. We emphasize, however, that neither of the crystal contact-based models appears consistent with the present binding, crosslinking and structural results.

FliG_C interacts with the stator protein MotA (Zhou et al, 1998a), and models for the rotation mechanism are typically focussed on the FliG_C–MotA interface. In the structural model developed here, FliG_C is absent in several positions around the rotor. Extensive physiological measurements have so far not given evidence of halting motor performance, provided the membrane is normally energized; even motors operating with a single stator unit (the full motor has about 10; Block and Berg, 1984; Blair and Berg, 1988; Reid et al, 2006) can rotate smoothly under high load (Reid et al, 2006) or rapidly under low load (Yuan and Berg, 2008). One of the charged residues of FliG that is important for rotation and that interacts with the stator (Arg 297 in the protein of E. coli) is at the inner edge of FliG_C, close to FliG_NM. We propose that the stator complexes are centred roughly above this position, where they could interact with both FliG_C and FliG_NM (Figure 5B). Interactions with more-inward parts of the rotor (the FliG_NM domains) might be important for propelling the rotor through the gap positions. An interaction with the FliG_NM domains might also be more consistent with the measured size of rotational steps, which average about 1/26th of a revolution (Sowa et al, 2005). While the FliG_NM domains have not previously been implicated in motor rotation in the same way as FliG_C, presently there is no evidence against their involvement, and the occurrence of some Mot⁻ (immotile but flagellate) mutations near the C-terminus of FliF, in segments that are known to bind to FliG_N (Grunenfelder et al, 2003), would be in accordance with this proposal.

NMR experiments gave evidence of a FliG–FliM interaction different from any found here (Dyer et al, 2009). That interaction involves surfaces of FliG and FliM that, if brought together, would orient the charged ridge of FliG downward and away from the stator. Such an interaction appears unlikely to occur in the fully assembled motor but may nevertheless occur, and have a useful role, in the cell. In a FliG molecule with its middle domain bound to FliM, the C-terminal domain could be re-oriented (by rotations in the Gly-Gly linker) to interact with FliM_M in the way observed in the NMR experiments. This more-compact conformation might provide a means of stabilizing the protein before its assembly into the C-ring. The FliG_M–FliG_C interaction observed in FliG crystals could have a similar role, helping to shield hydrophobic surfaces from inappropriate interactions until the normal interaction partners become available in the later stages of C-ring assembly.

In a recent proposal for the motor mechanism, rotation occurs as one part of the stator presses inward against angled surfaces of the rotor, and as another part, engaged through electrostatic interactions, moves tangentially (Blair, 2009). The structural information here is consistent with the essential elements of that hypothesis, provided the stator is positioned to allow interactions with both FliG_C and FliG_NM, as proposed above (Supplementary Figure S10). Most importantly, the present structural model for the torque-generating elements should provide a useful framework for addressing the molecular details of rotation and switching.

Materials and methods

Protein preparation

Coding sequences for T. maritima FliM residues 1–249 (FliM_NM, which contains the CheY-binding peptide and CheC-like domain) and FliG residues 117–195 (FliG_M195, which includes the middle domain and the segment linking it to FliG_C) were PCR cloned into the vector peT28a (Novagen) and expressed with a 6-histidine (His) tag in E. coli strain BL21-DE3 (Novagen) in LB broth with kanamycin selection (25 μg/ml). The proteins were purified on Nickel-NTA columns and the His tags were removed by thrombin digestion. The proteins were combined and run on a Superdex-200 sizing column (Pharmacia), followed by pooling of fractions and concentration (Centriprep; Amicon) in GF buffer (50 mM Tris pH 7.5, 150 mM NaCl and 4.5 mM DTT). The complex of FliM_NM and FliG_M was co-eluted a second time on Superdex-200 column and further concentrated for crystallization trials.

Crystallization and data collection

Multiple initial conditions for growing FliG_M195/FliM_NM complex crystals were found in commercial screening solutions (Hampton). The crystals with the best morphology appeared in a 2-ml drop (1:1 mixture of protein in GF buffer and reservoir) from a sealed well under vapour diffusion against a reservoir of 0.1 M MES pH 6.5, 10% dioxane and 1.6 M ammonium sulphate (Hampton Research). Diffraction data were collected under 100 K nitrogen stream at Cornell High-Energy Synchrotron Source (A1) on a CCD detector (Quantum-210, Area Detector System). The data sets were reduced and scaled using HKL200 (Otwinowski and Minor, 1997).

Structure determination and refinement

The FliG_M195/FliM_M complex structure was determined by molecular replacement (MR) with PHASER (McCoy et al, 2007) using as a model the RCSB deposited coordinates PDB codes 2HP7 (T. maritima FliM) and 1LKV (T. maritima FliG). Two FliM_M domains and one FliG_M domain were found by MR; the second FliG_M was placed manually in the residual electron density. Several residues of FliG_M195 (helix E) were removed from the initial model and rebuilt manually in XFIT (McRee, 1992). The final model was refined with the program CNS amidst cycles of manual model building (Brunger et al, 1998). Given the 3.5 Å resolution, only grouped B-factor refinement was applied. The model consists of two FliM_M and two FliG_M units that form an antiparallel dimer in the asymmetric unit through association of the FliM_M α1 and α1’ helices and the truncated FliM_M C-termini with the opposing FliG_M. Data collection and refinement statistics are summarized in Supplementary Table 1.

Protein interface analysis

Protein interfaces were analysed by the Protein Interfaces, Surfaces and Assemblies service PISA at European Bioinformatics Institute (http://www.ebi.ac.uk/msd-srv/prot_int/pistart.html), authored by Krissinel and Henrick (2007). Conservation of surface-exposed residues on FliM_M and FliG_MC was mapped with the Consurf server (Ashkenazy et al, 2010) and interface complementarity was evaluated with SC (Lawrence and Colman, 1993).

Strains

E. coli strains and plasmids used are listed in Supplementary Table 2.

Site-directed mutagenesis and assays of motility

Mutagenesis was performed using the QuikChange method (Stratagene) with oligonucleotides synthesized in core facilities of the University of Utah. Mutations were confirmed by sequencing. For assays of function, strains with deletions in the relevant genes were transformed with wild-type or mutant plasmids, and motility in soft agar, swimming in liquid, and flagellation were measured as described previously (Tang and Blair, 1995). Motility plates contained tryptone broth and 0.27% bacto agar, appropriate antibiotic(s) and IPTG at concentrations of 0, 40 and 100 μM to allow function to be tested over a range of expression levels. Plates were incubated at 32°C and swarm diameters were measured at regular intervals. Rates were determined from plots of diameter versus time.

Binding assays

Binding of FliM to FliG_C was measured using a pull-down assay with GST fused to residues 185–331 of FliG. Proteins were expressed separately in two strains, using plasmid pHT100 (Tang et al, 1996) derivatives to express the GST fusions to FliG_C, pDB72 (Tang et al, 1996) to express FliM (or its variants) and pKP41 to express FliN. For most experiments, FliN was coexpressed with FliM because FliM alone is prone to aggregation (Mathews et al, 1998). Control experiments used GST only, expressed from plasmid pHT100. Most binding experiments used strain RP3098, a ΔflhDC mutant that expresses no flagellar genes from the chromosome (Tang et al, 1996).

Cells were cultured overnight at 32°C in 40 ml TB or LB containing appropriate antibiotics and 400 μM IPTG for expressing GST (pHT100) or GST-fused FliG_C constructs. FliM and its mutant variants (pDB72) and FliN (pKP41) were cultured at the same condition containing appropriate antibiotics. IPTG (40 μM) was used to induce expression of FliM and 10 μM Na-salicylate to induce expression of FliN. Cells were harvested and resuspended in lysozyme-containing phosphate-buffered saline (140 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄, 1.8 mM KH₂PO₄, 5 mM EDTA and 0.2 mM APMSF (4-amidinophenylmethanesulphonyl fluoride)) and 0.1% CHAPS. Following a 1-h incubation on ice, the cells were further disrupted by sonication, debris was pelleted (16 000 g, 40 min, 4°C), and 50 ml of the supernatant was stored for use in estimating the amount of FliM present before addition of affinity beads. The rest (∼1 ml) was transferred to a clean tube, mixed with 150 μl of a 50% slurry of glutathione Sepharose 4B (Pharmacia) prepared according to the manufacturer's directions, and incubated for 1 h at room temperature with gentle rotation to allow binding. The Sepharose beads were then pelleted by a 1-min microcentrifuge spin, and washed twice with 1 ml of phosphate-buffered saline, each time pelleting again with a brief spin. The beads were then incubated with 50 μl of elution buffer (50 mM reduced glutathione in 50 mM Tris–HCl (pH 8.0)) for 10 min at room temperature with gentle rotation to release the GST–FliG_C and associated proteins. Beads were then pelleted and the supernatant was collected for analysis by SDS–PAGE and immunoblotting using anti-FliM antibody.

Crosslinking

Initial crosslinking experiments were carried out using the catalyst Cu [1,10-phenanthroline]₃. Plasmids expressing the Cys-substituted FliG and FliM proteins were co-transformed into the fliGM deletion strain DFB247. FliG-only, single-Cys mutants were transformed into the fliG deletion strain DFB225. Cells were cultured at 37°C for 4–5 h in LB medium containing required antibiotics and then diluted 100-fold into LB broth (containing antibiotics) and grown overnight with 50 μm IPTG at 37°C. Using A₆₀₀ readings to estimate culture density, equal numbers of cells from each culture were transferred to a centrifuge tube, pelleted (3000 g, 10 min) and resuspended in 200 μl of motility buffer (0.067 M sodium chloride, 0.01 M potassium phosphate pH 7.0 and 10⁻⁴ M EDTA), then divided into two 100 μl fractions. Crosslinking reagent (11 μl in 50% ethanol) was added to one sample, and non-crosslinked controls received just the 50% ethanol. The crosslinking reagent contained 4 mM CuSO₄ and 16 mM 1,10-phenanthroline, and was freshly prepared from a 1-M stock of phenanthroline in 95% ethanol and a 400-mM stock of CuSO₄ in water. Samples were rotated gently for 5 min at room temperature. Reactions were quenched after 5 min by addition of N-ethylmaleimide (2.2 μl from a 1-M stock in 95% ethanol) and EDTA (12.6 μl from a 0.5-M stock). Cells were then mixed with non-reducing gel-loading buffer, boiled and used for electrophoresis.

Some crosslinking experiments used the bifunctional reagent BMH. In all, 100 μl of cells was mixed with 2 μl of 50 mM BMH (dissolved in dimethyl sulphoxide and stored at −20°C), and incubated at room temperature for 10 min. Reactions were quenched with N-ethylmaleimide (2 μl from a 1-M stock in 95% ethanol). Control samples received just DMSO. Cells were mixed with reducing gel-loading buffer, boiled, and used for electrophoresis.

SDS–PAGE and immunoblotting

Protein samples were separated on 10% SDS–PAGE gels and transferred onto nitrocellulose using a semidry transfer apparatus (Bio-Rad). Rabbit polyclonal antibody against FliM was prepared as described previously (Tang and Blair, 1995; Tang et al, 1995) and used at 1500-fold dilution. Haemagglutinin-tagged FliG was detected using mouse anti-HA antibody at 1000-fold dilution (Covance, USA). Bands were visualized using the Super Signal West Picoluminol system (Pierce) and X-ray film.