Structure of the torque ring of the flagellar motor and the molecular basis for rotational switching

The flagellar motor drives the rotation of flagellar filaments at hundreds of revolutions per second^1,2, efficiently propelling bacteria through viscous media³. The motor uses the potential energy from an electrochemical gradient of cations^4,5 across the cytoplasmic membrane to generate torque. A rapid switch from anticlockwise to clockwise rotation determines whether a bacterium runs smoothly forward or tumbles to change its trajectory^6,7. A protein called FliG forms a ring in the rotor of the flagellar motor that is involved in the generation of torque^8–13 through an interaction with the cation channel forming stator subunit MotA¹². FliG has been suggested to adopt distinct conformations that induce switching but these structural changes and the molecular mechanism of switching are unknown. Here we report the molecular structure of the full-length FliG protein, identify conformational changes that are involved in rotational switching and uncover the structural basis for the formation of the FliG torque ring. This allows us to propose a model of the complete ring and switching mechanism in which conformational changes in FliG reverse the electrostatic charges involved in torque generation.

The structure of the full-length (FL) FliG protein (FliG_FL) from Aquifex aeolicus is entirely α-helical and consists of distinct amino-terminal (N), middle (M) and carboxy-terminal (C) globular domains, which are connected by two long helices (helix_NM and helix_MC) (Fig. 1a). The structure of the middle to C-terminal domain of FliG from Thermotoga maritima (FliG_MC) has been previously reported¹⁴ (Supplementary Fig. 1). Like FliG_MC, the middle domain of FliG_FL contains a single armadillo repeat motif (ARM_M) whereas the C-terminal domain can be further divided into a second ARM (ARM_C) and a six-helix bundle (helices_C1–6). An overview of the full-length FliG structure, assembly of the FliG ring and switching mechanism is illustrated in Supplementary Movie 1. In the C-terminal domain, charged residues that are clustered around helix_C5 result in a functionally important electrostatic charge distribution (Fig. 1b) that is involved in torque generation^9–12,14. Surprisingly, the fold around these charges, comprising helices_C3–6, is repeated in the N-terminal domain (helices_N1–4) (Fig. 1c). These helices are moderately conserved and contain four groups of conserved amino acid triplets (Fig. 1d). Charged residues on helix_N3 seem to invert the charges on the structurally equivalent torque helix_C5 (Fig. 1c, d). However, in the N-terminal domain, this does not result in an obvious polar electrostatic charge distribution, indicating that the repeated fold is not echoed with a repeated torque-generating function (Supplementary Fig. 2).

Figure 1. Structural overview of the full-length FliG monomer.

a–c, Residues are coloured from N to C terminus as a spectrum of colours from blue to red. Torque helix_C5 and helix_N3 are labelled with a red and blue asterisk respectively. Helices_C3–6 are shown with charged residues and the electrostatic potential on helix_C5 in b adjacent to helices_N1–4 in c to highlight the conserved fold. d, Sequence alignment of the residues shown in b and c. Conserved or similar residues are highlighted, and conserved amino acid triads are underlined. Helix_C5 and helix_N3 are encircled and charged residues on these helices are in red (negative) and blue (positive).

FliG is required for assembly of an intact flagellar motor⁸ where it binds to a structural subunit called FliF, thereby coupling torque to the rest of the flagellar filament^15,16 (Supplementary Fig. 3). Escherichia coli cells containing a FliF–FliG fusion mutant can form fully assembled flagella, and the deletion of groups of 10 residues within the N terminal 36 and 46 residues of FliG from Salmonella typhimurium disrupt flagellar assembly and binding to FliF, respectively¹⁷. In A. aeolicus, the equivalent residues map to the N-terminal three helices (helices_N1–3) of the protein (Fig. 1c), indicating that these promote binding to FliF.

FliG_FL and FliG_MC have been captured in different conformations. The most obvious conformational difference is in the middle domain. Whereas ARM_M is well conserved (Cα root mean squared deviation (r.m.s.d.), 1.0 Å), helix_MC is packed tightly against ARM_M and helix_NM in FliG_FL (‘closed’ conformation, Fig. 2a), but dissociated from ARM_M in FliG_MC (‘open’ conformation, Fig. 2b). The closed conformation is stabilized by 15 mostly conserved hydrophobic residues, which form a highly complementary hydrophobic interface between helix_MC and ARM_M (Supplementary Fig. 4a, b) that is disrupted in the open conformation. A second conformational difference is between the C-terminal domains of FliG_MC (Cter_MC) and FliG_FL (Cter_FL). Here, ARM_C and helices_C1–6 are very similar (Cα r.m.s.d., 1.0 Å and 0.77 Å, respectively) and are connected by a highly conserved loop, but their relative orientations differ between Cter_FL and Cter_MC (Fig. 2c, d). They are related by rotations of 77.8° around the F237 phi angle and 25.6° around the M236 phi angle (Supplementary Movie 2). This rotation alters the relative orientation of the torque helix_C5, which has been proposed as a mechanism for rotational switching¹³.

Figure 2. Conformational differences between A. aeolicus FliG_FL and T. maritima FliG_MC.

a, b, The middle domains of FliG_FL (a; A. aeolicus, closed) and FLiG_MC (b; T. maritima, open). The equivalent position of the extreme clockwise-biased deletion mutant is coloured in magenta. c, d, The C-terminal domains of A. aeolicus FliG_FL (Cter_FL) (c) and T. maritima FliG_MC (Cter_MC) (d). Torque helix_C5 is labelled with a red asterisk. Hinge residues M236 and F237 are shown as sticks.

We mapped all known FliG mutations that bias the direction of rotation of the flagellar motor onto FliG_FL (Supplementary Movie 3). These are clustered in three regions of the protein. Two of these clusters indicate that the conformational differences between FliG_FL and FliG_MC may reflect changes associated with motor switching for the following reasons. The first cluster is around helix_MC, where a majority of clockwise-biased mutations occur at the interface between helix_MC and the middle domain (Supplementary Fig. 4), and introduce charged or bulky residues that are likely to interfere with the closed conformation. Furthermore, deletion of a moderately conserved 169PAA171 motif (Fig. 2a, b, in magenta) in S. typhimurium results in an extreme clockwise bias¹⁸. This mutation is also likely to destabilize the closed conformation as it shortens the loop between helix_MC and ARM_M and alters the register of helix_MC, thereby changing the orientation of the hydrophobic ridge along helix_MC that binds to ARM_M (Supplementary Fig. 4). In contrast to clockwise-biased mutants, anticlockwise-biased mutations do not occur at the helix_MC–middle-domain interface, with the exception of the conservative I122L mutation in E. coli. Combined, these rotationally biased mutants indicate that the closed conformation represents FliG during anticlockwise rotation and that switching to clockwise rotation may involve the dissociation of helix_MC from ARM_M to an open conformation via a putative hinge loop between ARM_M and helix_MC (loop_M). This may explain why mutations at almost any residue in loop_M can bias the rotation direction (Supplementary Movie 3). The middle domains of FliG_FL and FliG_MC seem to represent anticlockwise and clockwise states respectively, indicating that Cter_FL and Cter_MC may follow the same trend. Indeed, a second cluster of mutants occurs between the two subdomains (ARM_C and helices_C1–6) of the C-terminal domain. However it is less clear whether these favour either the Cter_FL or the Cter_MC conformation (Supplementary Fig. 5). A third cluster of mutations occurs at the loop between helix_MC and the C-terminal domain (loop_C), which is another putative hinge loop that has been implicated in switching^14,19.

In the FliG_FL crystal lattice (Fig. 3a), the base of ARM_C is packed against the base of ARM_M of an adjacent monomer (ARM_M+1) (Supplementary Fig. 6a). The arrangement forms a stack of tandem ARM motifs resulting in a right-handed superhelix consisting of seven α-helices (Fig. 3b). The otherwise exposed hydrophobic patch on the base of ARM_C¹⁴ is completely buried in the large interacting surface and forms part of a continuous hydrophobic core that extends over the entire superhelix. The ARM_C–ARM_M+1 stacking has marked parallels with eukaryotic ARM motifs²⁰ (Supplementary Movie 4). In all structures containing ARM motifs, these form tandem repeats that interact extensively, resulting in a right-handed superhelix that creates a surface for protein–protein interactions^21,22. The stacking is mediated typically by nine hydrophobic residues, usually consisting of leucine, valine or isoleucine²⁰. Similarly, the stacking of ARM_C–ARM_M+1 is mediated by eleven leucine, valine or isoleucine residues, eight of which are absolutely conserved as hydrophobic residues across all known FliG sequences. This indicates that the interaction is not specific to A. aeolicus but extends to all flagellated bacteria. Indeed, in the crystal packing of the T. maritima FliG_MC structure, the ARM_C–ARM_M+1 interaction forms an identical (Cα r.m.s.d., 1.0 Å) right-handed superhelix (Fig. 3b). Importantly, this is despite the FliG_MC crystal originating from different species, constructs and crystal forms with no other conserved crystal contacts.

Figure 3. Structural basis for the formation of FliG multimers.

a, The FliG multimer in the FliG_FL crystal lattice. The ARM_C–ARM_M+1 superhelix is encircled and expanded in b, which shows the ARM superhelix from FliG_FL (left) adjacent to the ARM superhelix from FliG_MC (right). The positions of mutations that inhibit FliM binding and flagellar assembly are shown in magenta. c, Same as a, but highlighting the orientation of a FliG monomer and a FliG_UNIT in the crystal lattice. Repeated FliG_UNIT protomers in the same orientation are shown linearly in d, and with the curvature of a 45-nm diameter ring in e.

Combined, these data indicate that the ARM_C–ARM_M+1 interaction is a real biological interaction and this has several profound implications. First, FliG forms part of the flagellar motor known as the switch complex, which contains two other proteins, FliM and FliN. All three proteins are required for flagellar assembly (Supplementary Fig. 3). Mutations at the base of ARM_C and ARM_M and on the face of ARM_M, which contains a highly conserved EHPQR motif¹⁴, can disrupt flagellar assembly^8,10,23 and FliM binding²⁴. On the basis of the structure of a single FliG monomer, it seems evident that ARM_C and ARM_M are separate FliM binding sites¹⁴. However the ARM_C–ARM_M+1 superhelix indicates that ARM_C and ARM_M from adjacent monomers interact to create a surface for a single FliM binding site, which is consistent with all other known ARM superhelices that stack to form a surface for protein–protein interactions. Second, in addition to its interactions with FliF and FliM, proper assembly of the bacterial flagellum also requires a FliG–FliG interaction that is independent of other components of the flagellar motor. This is well supported by mutagenesis studies that demonstrate the requirement of at least five hydrophobic residues at the ARM_C–ARM_M+1 interface for flagellar assembly^10,23,24 (Fig. 3b). Last, because ARM_C forms a right-handed superhelix with ARM_M+1, it follows that ARM_C+1 has the same interaction with ARM_M+2. Thus, the ARM_C–ARM_M+1 superhelix is the structural basis for the formation of FliG multimers (Supplementary Fig. 6a, b).

As mentioned earlier, a cluster of rotationally biased mutations highlight the importance of the putative hinge loop_C in switching^14,19. Similarly, such mutations also occur at almost every residue on hinge loop_M, which mediates the transition from the closed to open conformation. Remarkably, in the FliG multimer, loop_C and loop_M+1 form an intermolecular anti-parallel β-sheet (Supplementary Fig. 6d, f) explaining how mutations in loop_C may influence switching.

The FliG multimer resembles beads on a string, linked by residue A193, one of the few amino acids unrestrained by secondary structure (Supplementary Fig. 6c, e). The repeating ‘beads’ are formed by an L-shaped protomer (FliG_UNIT) that contains two halves of adjacent FliG monomers (Supplementary Fig. 6a–c). Although loop_C probably allows interdomain flexibility in the FliG monomer¹⁴, the FliG_UNIT is a more robust arrangement of the three domains in FliG. In a FliG_UNIT the intersubunit β-sheet between loop_C–loop_M+1, in combination with the superhelix (ARM_C–ARM_M+1), restrains the relative movement of the C-terminal domain (Supplementary Movie 5). The main body of the FliG_UNIT contains a complete ARM_C/M+1 superhelix that is tethered to the base of the C-terminal helices_C1–6 through the M236–F237 hinge loop. The torque helix_C5 is perpendicular to the axis of the superhelix at the top of the FliG_UNIT (Fig. 3c) and the two long helices are at the base. Helix_MC connects adjacent FliG_UNIT protomers and helix_NM places the N-terminal domain next to the FliG_UNIT, thereby forming the L-shaped protomer. In the crystal lattice, adjacent FliG_UNIT protomers are in opposing orientations (Fig. 3a and Supplementary Fig. 6a–c).

To model the FliG ring, we applied a curvature to the multimer in the crystal lattice, creating a ring with a 45-nm diameter that corresponds to a region of the flagellar motor known as the C ring, where the switch complex is thought to reside (Supplementary Fig. 3). The model contains 40 FliG_UNIT protomers, where adjacent protomers are in opposing orientations resulting in a 20-fold symmetry (Supplementary Fig. 7a). However, this is inconsistent with the apparent 34-fold symmetry of the C ring²⁵. Furthermore, the opposing arrangement of alternating N- and C-terminal domains is inconsistent with the location of their respective interacting proteins, FliF and MotA. These are both on the periplasmic side of the C ring (Supplementary Fig. 3). Therefore, we generated rings by repeating the FliG_UNITs in the same orientation (Fig. 3d). These rings satisfy the spatial restraints imposed by the location of the MotA stators and the FliF ring (Fig. 3e). Furthermore, the 45-nm ring contains 34 monomers, which is in agreement with the 34-fold symmetry of the C ring²⁵ (Fig. 3e). Attempts to generate rings that match the size and symmetry of the MS ring, a 30 nm ring above the C ring that is integrated into the cytoplasmic membrane, were unsuccessful (Supplementary Fig. 7c, d).

When protomers are arranged in the same orientation, the continuous FliG polypeptide chain that spans adjacent FliG_UNIT protomers is broken with a distance of around 34 Å between A193 and E192 (Fig. 4b). Remarkably, dissociation of helix_MC from ARM_M into an open conformation brings these residues to within a peptide-bond distance (Fig. 4a, b). Similarly, this can be achieved with FliG in the closed conformation by rotating the main body of the FliG_UNIT (Fig. 4b, c). These conformational changes may explain why FliG alone does not spontaneously form an oligomer without an interaction with the FliF scaffold.

Figure 4. Molecular basis of rotational switching.

a–c, An expanded view of the encircled region in Fig. 3e is shown with one coloured FliG polypeptide chain. E192 and A193 are shown as yellow and red spheres respectively. a, b, Transition from the closed to the open conformation of helix_MC. b, c, An alternative conformational change when helix_MC remains in the closed conformation. d, e, The FliG ring with the monomers in the FliG_MC and the FliG_FL crystal structures, respectively. Charged residues on torque helix_C5 are shown in blue (positive) and red (negative).

We generated two FliG rings, one with each monomer in the FliG_FL (closed, anticlockwise) conformation and the other with each in the FliG_MC (open, clockwise) conformation (Fig. 4d and e, respectively). In the context of the complete FliG ring, the conformational differences between FliG_FL and FliG_MC result in a reversal of the electrostatic charges on torque helix_C5 of FliG, providing a model for rotational switching of the flagellar motor. Furthermore, to visualize the transition from FliG_FL to FliG_MC, we generated 100 intermediate structures between each state by interpolating the different phi and psi torsion angles. This illustrates that the entire conformational change occurs without any significant clashes in the ring (Supplementary Movie 6).

As the FliG monomer spans two adjacent FliG_UNIT protomers, neighbouring protomers are restrained to the same open or closed conformation. This indicates that switching entails a bistable global conformational switch where adjacent FliG monomers are restrained to the same rotational state. However, there is sufficient flexibility in the model to allow for opposing conformations in the ring that are bridged by protomers in intermediate states (Supplementary Fig. 8). This is compatible with the conformational spread model of cooperativity²⁶, the hallmarks of which were recently observed in the flagellar motor²⁷.

The FliG ring is consistent with three-dimensional electron microscopy reconstructions. The highest resolution three-dimensional electron microscopy reconstruction of the flagellar motor so far has been generated from single-particle averaging of a clockwise-locked 69PAA171-deletion mutant from S. typhimurium²⁵. We docked the FliG ring into the C ring, which has 34 distinct lobes on the periplasmic side at the outer periphery. The shape and size of these lobes complement that of the ARM superhelix and the C-terminal domain in the clockwise FliG ring (Supplementary Fig. 9a). Notably, the anticlockwise FliG ring is not consistent with the clockwise-locked electron microscopy reconstruction (Supplementary Fig. 10).

The vertical cross-section of the C ring from various electron micrographs has two lobes on the periplasmic side of the S. typhimurium C ring. The inner lobe is missing in the same micrograph of a flagellar motor containing a FliG–FliF fusion deletion mutant that does not contain the first 94 residues of FliG (N-terminal domain and part of helix_NM) and the position of the C ring is shifted up towards the MS ring compared to the native or full-length FliG–FliF mutants²⁸. We overlayed the cross-section of the FliG ring onto these micrographs and the C-terminal domain fits well on the outer lobe, the N-terminal domain on the inner lobe and the ARM superhelix in between (Supplementary Fig. 9b). Furthermore, we deleted the equivalent of the first 94 residues from FliG and overlayed this truncated structure onto electron micrographs of the vertical cross-section of the FliG–FliF deletion fusion mutant. The micrographs precisely shadow the structure of the truncated FliG ring, and the shift in the position of the C ring relative to the MS ring reflects the shortened link between these rings as a result of deleting the N-terminal domain (Supplementary Fig. 9c).

The orientation and location of the FliG rings on the periplasmic side of the C ring satisfies spatial restraints, which have been determined from structural, mutagenesis and electron microscopy data. First, in situ cryo-electron tomograms indicate that the MotA/B stator interacts with the outer circumference of the periplasmic side of the C ring^29,30 (Supplementary Fig. 3). Supplementary figure 9b, 4d and e illustrate how helix_C5 forms the outer circumference of the periplasmic side of the C ring and how the electrostatic charges that are involved in torque generation point towards the periplasm. Second, all FliF-binding helices_N1–3 are located in the inner circumference of the ring in an optimal position to interact with the MS ring (Fig. 3e and Supplementary Figs 3, 9b). Third, in the FliG ring, mutations that affect the interaction with FliM segregate to a single location on the cytoplasmic side of the ring on the surface of the ARM superhelix (Supplementary Fig. 9b), pointing towards the expected location of FliM (Supplementary Fig. 3). Finally, by docking structures of FliM and FliN subunits into the remaining density of the C ring we demonstrate that this density is sufficiently large to accommodate rings of both subunits in an arrangement that is consistent with biochemical studies (Supplementary Fig. 11).

METHODS SUMMARY

The full-length FliG protein from A. aeolicus was expressed in E. coli and crystallized as detailed in Supplementary Information. Crystals belong to space group P2₁ with one FliG monomer in the asymmetric unit. Native X-ray diffraction data were collected at the European Synchrotron Radiation Facility (ESRF) in Grenoble, France, and single or multiple wavelength anomalous dispersion (SAD or MAD) data were collected at the Advanced Photon Source (APS) in Chicago, USA. The methods used for structure determination are detailed in Supplementary Information. X-ray data and refinement statistics are given in Supplementary Tables 1 and 2, respectively. Still images and movie frames of protein structures were generated with PyMOL (http://www.pymol.org). Morphing and modelling software were written in C++.