The Architectural Dynamics of the Bacterial Flagellar Motor Switch

The rotary bacterial flagellar motor is remarkable in biochemistry for its highly synchronized operation and amplification during switching of rotation sense. The motor is part of the flagellar basal body, a complex multi-protein assembly. Sensory and energy transduction depends on a core of six proteins that are adapted in different species to adjust torque and produce diverse switches. Motor response to chemotactic and environmental stimuli is driven by interactions of the core with small signal proteins. The initial protein interactions are propagated across a multi-subunit cytoplasmic ring to switch torque. Torque reversal triggers structural transitions in the flagellar filament to change motile behavior. Subtle variations in the core components invert or block switch operation. The mechanics of the flagellar switch have been studied with multiple approaches, from protein dynamics to single molecule and cell biophysics. The architecture, driven by recent advances in electron cryo-microscopy, is available for several species. Computational methods have correlated structure with genetic and biochemical databases. The design principles underlying the basis of switch ultra-sensitivity and its dependence on motor torque remain elusive, but tantalizing clues have emerged. This review aims to consolidate recent knowledge into a unified platform that can inspire new research strategies.

protein structures [33,34,36,41] provided important clues into possible interactions. (B) The period witnessed the timely application of new NMR methodologies for structure determination of large macromolecules [49] to CheY complexes [42], as well as cryo-EM allied single-particle image processing of multi-subunit assemblies [19,20] to isolated basal bodies [15,23,50]. The Thermatoga maritama FliG structures [33,34] set an important precedent for X-ray crystallography of thermophile switch proteins. The first applications of live cell imaging with GFP biotechnology [51] to determine CheY bias modulation [45] or localization [43]-and together with single-molecule, force microscopy to characterize the load-dependent switching [31]-were to prove equally influential.
Biomolecules 2020, 10, x FOR PEER REVIEW 4 of 26 macromolecules [49] to CheY complexes [42], as well as cryo-EM allied single-particle image processing of multi-subunit assemblies [19,20] to isolated basal bodies [15,23,50]. The Thermatoga. maritama FliG structures [33,34] set an important precedent for X-ray crystallography of thermophile switch proteins. The first applications of live cell imaging with GFP biotechnology [51] to determine CheY bias modulation [45] or localization [43]-and together with single-molecule, force microscopy to characterize the load-dependent switching [31]-were to prove equally influential. Four fundamental issues could now be addressed. First, what was the nature of the coupling between the FliGC motor domain and the FliMN CheY binding target? Secondly, how were the dynamics of the C ring, a large multi-subunit assembly, synchronized for smooth rotation and rapid reversal? Thirdly, how did CheY activation trigger an ultrasensitive switch response? Fourthly, how was the switch regulated by the motor operation as dictated by the torque-velocity relations? suppressor mutations in swarm plate assays. Its interactions with chemotaxis components and Mot proteins were also identified. {a} Schematic of a swarm plate-the native (WT) strain forms a swarm with chemotactic rings. Strains carrying mot mutations (Mot-) do not swarm while those with che mutations (Che-) have reduced swarms. Suppressor mutations yield pseudo-revertant strain (PR) with partially restored swarming. {b} Color codes are followed in subsequent Figures for the switch complex components (FliG (green), FliM (gold), FliN (cyan)), the CheY protein (salmon) and the MS-ring scaffold (orange) (adapted from [6]). (ii) Gene sequencing identified the mutations. The fliM gene (N-C terminal residue numbers) predominantly contained the che lesions, clustered into distinct CW (green) and CCW (magenta) regions. Arrows mark mot lesions (adapted from [8]). (B) Structural identification: (i) An extended cytoplasmic structure contiguous with the basal body MS-ring (yellow arrow) was isolated using gentler protocols and subsequently established as the switch complex by immuno-EM and biochemistry (from [14]). (ii) Assembly of switch complex by overproduction of plasmid-encoded components allowed biochemical characterization culminating in the determination of the C ring subunit stoichiometry (n = 33-34) (from [23]). (iii) Single-particle analysis resolved FliG domain substructure (yellow arrows) from differences in central sections from wild-type (WT) and ∆FliFFliG (∆) 3D basal-body reconstructions (from [22] with permission). (C) Motor function and mechanism: (i) Temporally resolved measurement of filament rotation, as a sinusoidal variation of laser dark-field spot intensity, characterized aberrant phenotypes in switch complex mutant strains. Panels (top to bottom) show slow rotation (S), pausing (P) and reversal (R) episodes (reproduced from [32] with permission). (ii) The first atomic structure of a switch component (FliGc [33]) followed by the FliG MC structure localized much of the mutant library then available ((mot lesions (black); CW lesions (red); CCW lesions (yellow); CW or CCW, depending on the residue substitution, orange; and motB suppressors (purple)) to generate chemically explicit ideas for motor reversal (PDB: 1lkv (modified from [34])). (D) Switch chemotactic signal transduction: (i) {1}-Determination of switch "ultra-sensitivity" (Hill coefficient, H = 10.3) by simultaneous measurement of the CW bias of beads on flagellar stubs (red) and concentration of a fluorescent GFP-CheY fusion (green) locked in the active state (*) in engineered strains (reproduced from [45] with permission). {2}-Plots show non-cooperative binding of acetate-activated CheY to overproduced C rings [46] compared to the in-vivo change in CW bias. (ii) The atomic structure of beryllium-fluoride (BeF3 (black))-activated CheY (salmon) bound to the FliM N-terminal peptide (yellow) initiated structure guided mutagenesis to explain the switch ultra-sensitivity. Aromatic residue (W58, Y106 (orange)) motions were early diagnostics for activation. Magnesium ion (red) (PDB: 1f4v (modified from [41])).
Four fundamental issues could now be addressed. First, what was the nature of the coupling between the FliG C motor domain and the FliM N CheY binding target? Secondly, how were the dynamics of the C ring, a large multi-subunit assembly, synchronized for smooth rotation and rapid reversal? Thirdly, how did CheY activation trigger an ultrasensitive switch response? Fourthly, how was the switch regulated by the motor operation as dictated by the torque-velocity relations?

Switch Physiology and Mathematical Models
Motor dynamics are the direct outcome of the architectural dynamics of the molecular machinery. Knowledge of motor dynamics, and the associated development of mathematical models, has advanced concurrently with knowledge of the molecular architecture and dynamics. The spatiotemporal resolution and mechanical range of the rotation assays have continued to increase for characterization of speed fluctuations, the load dependence and stochastic properties of the switch machinery in unprecedented detail. These advances provide additional constraints that must be addressed by the study of the molecular mechanism. Motor dynamics have been reviewed recently [52]. An overview is given in this short section, summarized in Figure 2, to provide additional context for the main body of the review.
Focal back-plane interferometry with high spatiotemporal resolution (1 • , 10 −3 s) recorded incomplete, intermediate switching events to extend the temporally resolved measurement of switch transitions [53]. More recently, gold nanospheres (d = 60 nm) conjugated to genetically engineered, rigid Salmonella flagellar hooks, imaged by dark-field and rapid (5000 Hz) CMOS cameras, have been exploited to measure motor rotation [54]. This study interpreted the large speed fluctuations recorded near the zero-torque speed (~400 Hz) that persisted over many revolutions as the association-dissociation of individual stator units. The highlight of such studies was that the resolution of the angular step periodicity per revolution was resolved in motors resurrected by single or a few stator units; first for a sodium powered chimeric motor [55] and then for the S. enterica proton motor [56]. In each case, 26 steps per revolution were reported for both CW and CCW rotation. However, while the step size was symmetrical for the Salmonella motor, the CW steps were smaller than the CCW steps in the chimeric motor. A study of the unidirectional R. sphaeroides motor reported a similar estimate of 27-28 discrete stopping angles [57]. The steps may be due to modulation of the elastic potential well by periodic contacts between the static and mobile motor modules rather than individual energy coupling events. A formal model of this idea successfully predicted the difference in CW versus CW step size [58].
The long-standing tethered cell measurements of the Poisson-distributed CCW and CW intervals had been the bedrock for theoretical models. Filament-associated bead rotation measurements in the phospho-mimetic E. coli CheY13DK mutant strain first revealed gamma distributions for both CW and CCW intervals [59] to challenge the view of switch transitions as a single Poisson process based on tethered cell data. Subsequent measurements of the rotation of beads attached to flagellar stubs [53] and nanoparticles attached to flagellar hooks [60] obtained exponential distributions in apparent contrast to [59], fueling speculation that filament polymorphic transitions [61] may have complicated the rotation of filament-associated latex beads (0.5 µm). The use of different-sized latex spheres extended the dependence of switching kinetics on load [62]. The study reported that at near zero-load speed both CCW<->CW and CW<->CCW transitions increased with the load. Furthermore, the torque-velocity curve for CW rotation was linear in contrast to CCW rotation [63]. The load dependence raised the possibility that it may underlie the apparent discrepancy between the gamma and Poisson interval distributions. The comprehensive analysis of the rotation of different-sized beads in the double mutant phospho-mimetic strain CheY13DK106YW showed this was indeed the case. Both exponential and peaked distributions were obtained, depending on the different load, proton potential and torque [64], to rule out speculation that the gamma distribution was an artefact.
The knowledge about the regulation of (CW/CCW) rotation bias by CheY has also advanced. Temporally resolved recordings of bead rotation over minutes re-evaluated motor individuality in terms of its rotation bias. Long-term recordings of variation in single bead rotation compared to population variation of multiple beads concluded that individuality was due to intracellular chemistry. The bias of individual motors had a bimodal distribution with cells with high switching frequencies binned into a central trough. The CW and CCW fractions were dominant, consistent with the ultrasensitive switch in rotation bias by CheY [65]. Single bead CW and CCW rotational intervals had earlier been reported to be asymmetrically distributed around the bias midpoint ((CW/(CW + CCW)) = 0.5), with the CCW but not CW intervals varying with rotation bias [66]. These different relations argued against the determination of transition rates by CheY or any other single parameter. The ultra-sensitivity has also been re-evaluated. The local GFP-CheY concentration around single flagellar motors was measured and correlated with (CW/CCW) rotation bias in a ∆cheY E. coli strain [67] for an estimate of the mean CheY rotor occupancy during CW rotation. GFP-tagged fusion proteins also provided evidence that the basal C ring components FliM and FliN undergo dynamic exchange [68,69]. An experiment was designed to correct for the FliM copy variation that would be obtained based on the exchange being adaptive. It estimated the Hill coefficient, H, for the ultra-sensitivity from bias changes due to attractant removal and addition in strains lacking receptor adaptation was as high as 21 [70].
The experiments motivated the development of the conformational spread model [48] and the formulation of a fundamentally distinct non-equilibrium model [71]. The initial interpretation for the peaked gamma distribution was based on multiple Poisson events [59] to refine, but not radically alter the conformational spread model. The asymmetry in the experimental torque-velocity curves and the load dependence of the switching kinetics could also be accounted for by an integrated model that combined torque generation models with the switch ultra-sensitivity as formalized by conformational spread [72]. Theoretical models, in general, seek to explain the interdependence between rotation bias, switching frequency and interval distributions. The experimental relations between these motor parameters were set as constraints to model the coupling between synchronous switching and signal amplification as a function of subunit number [73]. More recent simulations explained the correlation between local CheY concentration and single motor rotation bias as well as the utility of dynamic subunit exchange in the maintenance of switch ultra-sensitivity [74]. Detailed balance within the equilibrium conformational spread framework was followed in both studies [73,74]. Tu's non-equilibrium model, where the detailed balance is broken based on energy input, also explains the gamma distribution data [71,75]. The model is consistent with the expanded set of interval distributions obtained under different load and energization conditions as detailed in [64]. The energy input for switching need only be a small fraction of the total input that is dominantly utilized to power rotation.
Biomolecules 2020, 10, x FOR PEER REVIEW 7 of 26 switching and signal amplification as a function of subunit number [73]. More recent simulations explained the correlation between local CheY concentration and single motor rotation bias as well as the utility of dynamic subunit exchange in the maintenance of switch ultra-sensitivity [74]. Detailed balance within the equilibrium conformational spread framework was followed in both studies [73,74]. Tu's non-equilibrium model, where the detailed balance is broken based on energy input, also explains the gamma distribution data [71,75]. The model is consistent with the expanded set of interval distributions obtained under different load and energization conditions as detailed in [64]. The energy input for switching need only be a small fraction of the total input that is dominantly utilized to power rotation. body (reproduced from [54] with permission). (B) CW and CCW interval distributions: (i) Idealized time series of a motor alternating between CW and CCW rotation. (ii) Interval (τ CW , τ CCW ) distributions measured under low and high torque (reproduced from [75] with permission). (C) Coupled energization and switching of rotation: (i) Torque velocity curves for CW and CCW rotation [63]. (ii) Free energy diagram of CW <-> CCW transitions and the mechanical work (blue arrow) contribution (reproduced from [72] with permission). (D) Models for non-Poisson interval distributions: (i) Conformational spread seeded from multiple CheY binding events (reproduced from [59] with permission). (ii) Breakdown of detailed balance from motor energy dissipation (reproduced from [64] with permission).

Architecture and Molecular Mechanism
Structural work on the issues outlined in the previous two sections can be grouped into four major areas based on the protein-protein interactions: the central processing unit (FliM M .FliG N ), the trigger machinery (CheY/C ring), the torque reversal mechanism (FliG C /FliG M ) and the torque transmission platform (FliG N /FliF C ).

The Central Processing Unit-The FliM M .FliG M Complex
X-ray crystallography led the effort guided by mutagenesis, in situ crosslinking and EM reconstruction to characterize the linkage between the domains responsible for torque generation and CheY signal reception. The atomic structure of the T. maritima FliG middle (FliG M ) and C-terminal (FliG C ) domains had an established ARM architecture for both domains and localized hotspots for E. coli che mutations to the sequence encoding the linker region between FliG M and FliG C [34]. The atomic structure of the T. maritima FliM middle domain (FliM M ), the central element in switch function, showed that the CCW and CW substitutions localized to residue positions at the intradomain contact interface, as deduced by subunit crosslinking in situ [76]. The CW substitutions aligned differently to the CCW substitutions, suggesting distinct FliM M orientations for CW versus CCW rotation ( Figure 3A). In situ crosslinking had been introduced in earlier work on FliG M domain organization [77]. The in-situ crosslinking experiments probed the subunit organization of the switch proteins as appropriate controls established that cross-linking was negligible in non-flagellate strains [76,77]. The structures of FliM M .FliG M complexes from T. maritima [78,79] and H. pylori [80] determined the functional importance of the FliM GXXG and FliG EHPQR loop motifs at the FliM M .FliG M interface. Comparison of the T. maritima and H. pylori interfaces showed conservation of essential GXXG and EHPQR loop contacts and their stabilization by complex formation. Additional interfacial residues important for coupling were identified in H. pylori, supported by mutagenesis.
Protein dynamic simulations showed that the core architecture and dynamics of FliM M do not change upon complex formation and are conserved between T. maritima and H. pylori. FliM M conformational ensembles generated from the T. maritima structure had a bimodal distribution, while the FliG M domain movements also alternated between the bi-stable states strongly coupled to FliM M [81]. Targeted tryptophan substitutions had identified FliG M residue positions important for association with FliM M [82]. In addition to the EHPQR loop, these residues were in a FliGc ARM-C hydrophobic patch adjacent to the FliG MC GG linker as well as non-interfacial FliG M residue positions. Recent NMR measurements of the FliG M dynamics in the sodium V. alginolyticus motor have provided important validation, supported by mutagenesis, of the conformational plasticity of the FliG M domain between the bi-stable conformational states [83]. The EHPQR E144D residue substitution increased the switching frequency, a similar response to one obtained upon phenol repellent addition. It did not alter FliG M conformation. In contrast, residue substitutions in the GG hinge both locked the bias and reduced the FliG M dynamics, as assessed by NMR. The E144D and CCW-biased G214S motors were both CCW-locked in the ∆che strains, implying a close linkage between CheY occupancy and FliG M conformational fluctuations ( Figure 3B).
In H. pylori, co-crystallization of putative spermidine synthase (SpeE) with FliM M , combined with mutagenesis and motility assays, has shown how this protein affects motile speed and switching behavior. The SpeE.FliM N contact interface partially overlaps with the FliG M .FliM M interface [84]. While several proteins important for cell metabolism have been identified in diverse bacteria to influence motile behavior via direct interaction with FliG (cited in [84]), SpeE is the first known to act at the FliG M .FliM M interface. In conclusion, the crystal structure, allelic mutations, in situ crosslinking, and protein dynamic simulations all argue for the alternation of FliM M between bi-stable conformational states, inter-species conservation of its fold and interfacial FliG M coupling.
Biomolecules 2020, 10, x FOR PEER REVIEW 9 of 26 and protein dynamic simulations all argue for the alternation of FliMM between bi-stable conformational states, inter-species conservation of its fold and interfacial FliGM coupling. highlighted with homologous FliGM residues to those implicated by E. coli tryptophan mutagenesis (red. [82]) and V. alginolyticus switch mutants G214S, G215A and E144D (orange [83]). (ii) 2D-NMR 1 H-15 N correlation spectra reveal that the native (WT) V. alginolyticus FliGM architecture is substantially altered by substitutions in the conserved glycine pair that alter rotation bias, but is unaffected by the EPQR E144D residue substitution that increases the switching frequency. Axes indicate the shift in the magnetic field, in parts per million (ppm), relative to a reference compound for resonance (reproduced from [83] with permission).

The Trigger Machinery-CheY and the Basal C Ring
Motor response to chemotactic signals can be parsed into CheY activation and binding to the C ring to bias FliMM conformational fluctuations. There is a 20-fold increase upon phosphorylation in highlighted with homologous FliG M residues to those implicated by E. coli tryptophan mutagenesis (red [82]) and V. alginolyticus switch mutants G214S, G215A and E144D (orange [83]). (ii) 2D-NMR 1 H-15 N correlation spectra reveal that the native (WT) V. alginolyticus FliG M architecture is substantially altered by substitutions in the conserved glycine pair that alter rotation bias, but is unaffected by the EPQR E144D residue substitution that increases the switching frequency. Axes indicate the shift in the magnetic field, in parts per million (ppm), relative to a reference compound for resonance (reproduced from [83] with permission).

The Trigger Machinery-CheY and the Basal C Ring
Motor response to chemotactic signals can be parsed into CheY activation and binding to the C ring to bias FliM M conformational fluctuations. There is a 20-fold increase upon phosphorylation in CheY affinity for E. coli FliM N [41]. The molecular analysis of CheY activation by the phospho-mimetic CheY 13DK106YW double residue substitutions used to study motor response was initiated by co-crystallization of the E. coli protein alone and in complex with the FliM N-terminal peptide (FliM N ) [85]. Strikingly, while the CheY13DK106YW in the complex was in a conformation akin to BeF3 activated CheY, the free CheY13DK106YW adopted the inactive conformation. Subsequently, the atomic structure of the native CheY.FliM N complex revealed that the electron density of a critical hinge, the α4-β4 loop, in this complex partitioned between its active and inactive states [86]. Quantitative comparison of MD conformational ensembles from crystal structures of the CheY superfamily representatives has extracted common signatures, including the α4-β4 hinge, for allosteric communication between the phosphorylation and target binding sites [87]; the importance of the α4-β4 loop further emphasized by MD simulations of acetate-activated CheY [88]. MD simulations supported by oxygen-radical foot-printing solution measurements, a sensitive probe for sidechain solvent accessibility [89], have now shown that closure of this loop hinge buried the allosteric relay aromatic sidechains to both stabilize the global CheY fold and increase the local FliM N affinity.
The motor response is likely to depend on the intervening disordered linker between FliM N and FliM M that may, in principle, serve as a flexible tether to deliver FliM N -bound CheY to second binding sites. Occupancy of these sites could control FliM M conformational fluctuations and inter-subunit coupling. Candidates have been identified in two species. There is NMR evidence for an interaction between T. maritima FliM N -tethered CheY and FliM M [90]. Alternatively, in E. coli, pull-down assays have reported evidence for an association of the CheY-FliM N fusion with FliN [91]. This evidence is summarized in Figure 4A.
There is a close homology between FliN and FliM C . The atomic structure of the T. maritima FliN homodimer [92] motivated mutagenesis and in situ crosslinking experiments to infer FliN tetrameric quaternary organization in the E. coli flagellar motor [93]. These experiments reported cross-link changes upon addition of chemotactic stimuli, consistent with domain motions, but the connectivity between FliM N and the FliM C 1 .FliN 3 module is not well-understood as the FliM inter-domain linkers have not been structurally characterized. The atomic structure of the FliM C .FliN heterodimer is similar to the FliN homodimer [94], consistent with a FliM C 1 .FliN 3 tetramer identified by mass spectroscopy and structural homology with the Type III injectosome components [95]. The C-terminal tail of the FliH component of the flagellar export ATPase assembly has been co-crystallized and bound to the FliM C .FliN heterodimer ( Figure 4B). The co-crystal validated early predictions of the interactions of the C ring with the flagellar export apparatus [96], as well as the proposal based on cryo-tomographic (cryo-ET) reconstruction and FliH sequences from diverse species that FliH acts as a spacer to set C ring diameter [97]. The switch complex contains FliY instead of, or in addition to, FliM and FliN in numerous species [98]. The structure of T. maritima FliY reveals a middle domain with strong structural similarity to FliM M , while the FliY C-terminal domain is similar to FliN [98]. This study also reported solution assays to show that T. maritima FliY homodimerizes via its N-terminal domain and does not have an increased affinity for activated versus inactive CheY, and that its middle domain does not bind FliG. FliY was first reported in the Gram-positive B. subtilis [99] where CheY phosphorylation enhances CCW, not CW rotation [100], in contrast to the Gram-negative γ-proteobacteria E. coli and S. enterica. The characterization of the B. subtilis rotor module by in situ crosslinking found that the FliM M FliG M interface is conserved as in other bacteria with FliY proposed to form an external ring adjacent to the FliM M ring [101].
Thus, FliM C interacts with FliN to form the C ring base and can, in principle, transmit structural perturbations triggered by CheY-FliN association to FliM M . Other bacteria, such as T. maritima, may utilize a different signal strategy and a distinct basal ring architecture. The variable orientations of the FliGC domain relative to FliGM inform on both rotation reversal in addition to assembly. A striking example has been a pair of H. pylori FliGMC structures that show FliGC in orthogonal orientations relative to FliGM [102]. The underlying motions have been characterized by experimental probes and atomistic protein dynamics simulations since the snapshots of the kinetic mechanism provided by the crystal structures, though valuable, are too sparse to determine reaction trajectories. FliGC reorientation was first deduced from in situ crosslinking in E. coli [77]. Reorientation of the S. enterica α-helical linker (helixMC) adjacent to the GG pair was inferred from in vivo crosslinking experiments and simulations, to regulate the switching between the CW and CCW states. The crosslinking experiments found that the helixMC G174C residue substitution formed crosslinks in the CW-locked FliG PEV  but not the native, dominantly CCW S. enterica strain [103]. Analyses of conformational ensembles simulated from the FliGMC structures had identified helixMC as a central hinge and predicted its melting regulated the FliGC orientation relative to FliGMC [81]. These studies, taken together, supported the large reorientations deduced from the superimposition of different crystal structures.
The simulations also reported that the T. maritima MFXF linker between FliGC ARM-C and Cα1- 6 amplifies thermal fluctuations of the coupled FliMM-FliGM central processing unit [81]. This idea was supported and extended in important ways by NMR and MD simulations of V. alginolyticus FliGC. The conformational dynamics of FliGC-Cα1-6 helix α1 in FliG A282T were reduced relative to the native protein to give detectible peaks in the NMR spectra. The A282T strain has a CW bias in

Bidirectional Torque Generation-FliG M -FliG C Interactions
Vital clues driven by FliG multi-domain crystal structures have emerged since 2003 on the mechanism of rotation reversal, albeit from a limited set of organisms (S. enterica, T. maritima, H. pylori and A. aeolicus). Mutagenesis screens, experimental and computational analyses of protein dynamics and coevolutionary information have supplemented the crystal structures in important ways, as outlined below.
The variable orientations of the FliG C domain relative to FliG M inform on both rotation reversal in addition to assembly. A striking example has been a pair of H. pylori FliG MC structures that show FliG C in orthogonal orientations relative to FliG M [102]. The underlying motions have been characterized by experimental probes and atomistic protein dynamics simulations since the snapshots of the kinetic mechanism provided by the crystal structures, though valuable, are too sparse to determine reaction trajectories. FliG C reorientation was first deduced from in situ crosslinking in E. coli [77]. Reorientation of the S. enterica α-helical linker (helix MC ) adjacent to the GG pair was inferred from in vivo crosslinking experiments and simulations, to regulate the switching between the CW and CCW states. The crosslinking experiments found that the helix MC G174C residue substitution formed crosslinks in the CW-locked FliG PEV but not the native, dominantly CCW S. enterica strain [103]. Analyses of conformational ensembles simulated from the FliG MC structures had identified helix MC as a central hinge and predicted its melting regulated the FliG C orientation relative to FliG MC [81].
These studies, taken together, supported the large reorientations deduced from the superimposition of different crystal structures.
The simulations also reported that the T. maritima MFXF linker between FliG C ARM-C and Cα 1-6 amplifies thermal fluctuations of the coupled FliM M -FliG M central processing unit [81]. This idea was supported and extended in important ways by NMR and MD simulations of V. alginolyticus FliG C . The conformational dynamics of FliG C -Cα 1-6 helix α1 in FliG A282T were reduced relative to the native protein to give detectible peaks in the NMR spectra. The A282T strain has a CW bias in contrast to the CCW bias of the native strain. MD simulations further showed the reduction was due to additional hydrogen bonding contacts between the ARM-C and FliG C α1-6 that constrained the flexibility of the intervening linker. MFXF 254 hinge orientation, monitored by residue F254, partitioned into conformational clusters that overlapped the subsets of the crystal structures representing either the CW or CCW rotation states based on other criteria ( Figure 5A,B). A prescient early analysis of E. coli FliG residue substitutions in the helix MC linker, sensitized by a serendipitous E232G substitution a few positions upstream of the MFXF hinge, had reported these gave rise to diverse phenotypes with altered switching frequencies, pausing or altered bias [104].
The complete Aquifex aeolicus FliG structure revealed an ARM fold for the N-terminal domain (FliG N ) in addition to FliG M and FliG C, separated by α-helical linkers [105]). Sequence similarity suggests these domains arose from gene duplication [106]. The structure reported inter-molecular stacking between the FliG M and FliG ARM_C domains. In contrast, the S. enterica FliG PEV [107] and T. maritima [78] crystal structures of the FliG MC complexes showed intramolecular stacking of these domains. The consensus view now is that the assembly of the FliG middle and C-terminal domain is mediated by intermolecular stacking. Solution pulsed dipolar ESR spectroscopy combined with residue substitutions first indicated that, in T. maritima, these domains self-assemble via the FliG M -FliG ARM_C intermolecular stacking contact; then, direct assembly of FliM [108]. The central and membrane-proximal sections of the S. enterica C ring map was fit well by the ESR-derived model ( Figure 5C). A domain-swap mechanism for FliG ring assembly in the E. coli motor was subsequently determined with SAXS analysis supported by in situ cross-linking [109]. Binding energy from FliG association with the FliF MS ring was speculated to alter the conformational equilibrium between the compact and extended conformation of the malleable helix MC to favor polymerization of a chained FliG MC ring in the flagellar motor but not in solution.
Coevolutionary information has emerged as an important high throughput tool to assess the design principles of bacterial multi-protein complexes at the single residue level. Residue coevolution supported the design of the switch machinery framed by experiments and simulations. First, the role of FliM M as the central relay was consistent with its strongly coevolved inter-subunit and FliG M interfacial contacts [110]. Two distinct T. maritima FliM M dimer configurations were obtained when dimer formation was simulated based on coevolved subunit couplings [111], although one orientation did not match either the CW and CCW dimers deduced from S. enterica residue substitutions [76], either because of limited sampling and/or because the bi-directional switch is not universal across species. Secondly, coevolution provided strong support for conservation of the FliG C -FliG M stacking contact [109,112] relative to other contacts identified by cross-link data [77]. Thirdly, the coevolved coupling was mapped, in part, onto the identified dynamic couplings and modules. Notably, the coevolved FliM Mc -FliG M contacts was mapped onto the dynamic couplings across the T. maritima FliM M .FliG M interface [81] and FliG MC , as a coevolved network with distinct nodes as allosteric sectors [110]. The nodes included the well-characterized EHPQR and PEV (in S. enterica) at the FliM M .FliG M interface as well as α-helices (helix MC, Cα 1-6 helix α 1 ). The melting of these α-helices has been noted above. The coevolution signal from FliG C ARM-C is sparse. This sub-domain may be the converter element that encodes different species-specific outputs from a conserved input signal ( Figure 5D). (ii) Bimodal hinge planar angle distributions from the dominant principal collective motions 1-3 extracted from the generated conformational ensemble. Lines plot the 3D angular displacement distribution (solid) and unimodal fit (dotted). The distribution of the FlIM M FliG M interfacial hinge (1 (gold)) has a narrow spread relative to the MFXF hinge distribution (2 (green)), indicating most of the amplification occurs at the latter hinge (from [81]). (B) MFXF hinge dynamics: (i) The representative structure of the V. alginolyticus FliGc A282T homology model from cluster analysis of the MD ensemble. The predicted dynamic helix α 1 (pale green), the MFXF 254 motif, and residues G214, G215 (orange) and T282 (magenta with an asterisk) are marked. 2D-NMR reports the CW-biasing A282T residue substitution melts helix α 1 and reorients MFXF 254 . (ii) The 2D plot of F254 dihedral angle versus helix α 6 orientation relative to the FliG M. FliG C interface obtained from the MD trajectories. The wild-type (WT) and mutant (A282T) conformational clusters map onto different subsets of the FliGc crystal structures. Labeled circles (magenta) mark structures from T. maritima (CW-locked. PDB: 3ajc; wild type. PDB: 1lkv; complexed with FliM M (PDB: 4fhr)), Helicobacter pylori (PDB: 3usy and 3usw) and A. aeolicus (full-length. PDB: 3hjl) (from [113] with permission). (C) The FliG M -ARMc stacking interaction: The best-fit of the pulsed dipolar ESR spectroscopy model of the T. maritima FliG M -ARMc inter-domain stack to the 3D S. enterica electron density map (from [108] with permission). (D) Residue coevolution model for rotation reversal: Strong, coevolved FliM M (yellow) contacts mediate conformational spread (arrows) in the FliM M ring. The FliG αC3-6 "motor domain" (dark green) is organized around the αC5 "torque helix" with charged residues (red) important for rotor-stator interactions. The primary nodes of the coevolved network form a relay of allosteric sectors (numbered grey patches) across ARM-M and ARM-C (light green). ARM-C has sparse coevolved contacts implying a variable fold to generate different motor responses from a conserved FliM M switch transition Box. Coevolved interfacial pairs (residues (white spheres) and contacts (red)) link the FliM M GGXG motif with the FliG M EHPQ motif. The contacts and nodes are mapped onto PDB: 4fhr (from [110]).
In conclusion, the library of crystal structures now available, together with experimental and computational measurements of their dynamics, position FliM, FliG and FliN in the S. enterica C ring map and charts out their connectivity and conformational plasticity. FliM M has a central location in the C ring with subunit contacts designed for propagation of both the CW and CCW conformational states in the species studied thus far. FliM M forms a tightly coupled complex with FliG M , with a central location in the C ring, connected via FliG M to FliG C adjacent to the membrane stator complexes, and via FliM M to FliM C in the basal C ring.

The Association of the Switch with the Mot Stators and the FliF C Scaffold
The interactions of the FliGc domain with the MotA.MotB stator complexes determine torque. The biochemical studies of rotor-stator interactions in E. coli [11] were soon extended to other bacterial motors, notably the sodium-driven V. alginolyticus motor [114]. The FliG C torque helix and adjacent segments in the V. alginolyticus motor contain more charged residues compared to E. coli [115]. Furthermore, substitutions at two residue positions selectively impair only CCW rotation [116]. Interestingly, the V. alginolyticus FliG C also determines the correct polar localization and assembly of the stator complexes [117]. The mutagenesis of the residues involved in rotor-stator interactions in the E. coli motor was extended in the related S. enterica [118]. Notably, these investigators coupled fluorescent localization of the GFP-MotB fusion proteins with motility assays to parse out charged residues at the FliG c -MotA interface that influence stator assembly from those dedicated to torque generation. The first cryo-ET images documenting the alteration of C ring morphology by stator complexes have now been obtained in the spirochete Borrelia burgdorferi. Comparative study of motB mutant strains with defective versus restored proton conduction suggests that functional stators are required for full expression of this effect [119] ( Figure 6A). Thus, torque affects switch architecture, in addition to CW/CCW rotation bias (Section 2), while the FliG rotor ting, in turn, affects assembly of the force-generating stator complexes.
The CCW and CW torque generated by stator interactions with the FliG ring must be transmitted via the intervening FliF MS-ring to the external components of the filament whose rotation is the physiologically relevant parameter. The fliF and fliG genes are adjacent in one operon for middle gene expression in the S. enterica flagellar regulon [120]. Strains with these fusions are motile and form C rings but have aberrant CW/CCW bias [121]. Their analysis has provided valuable clues on the effects of the FliF-FliG association on switch operation and C ring morphology.
The bias of the full-frame fusion was restored by suppressor mutations in FliM M and FliG N , as well as by insertion of a flexible, nine-residue glycine-rich linker at the fusion site [121], emphasizing the influence of FliG N and FliG M conformational plasticity on the sign of the transmitted torque. The deletion the FliF-FliG fusion (∆FliF.FliG) strain lacked a major part of FliG N ,and formed smaller C rings with lower (n = 31) subunit stoichiometry [122]. In contrast, the FliG M ∆PAA deletion strain formed smaller C rings with altered packing as their subunit stoichiometry was unchanged. The bias of the ∆FliF.FliG strain could be also be compensated by residue substitutions localized to the FliM M .FliG M interface (FliG D124Y ) and a likely FliM M dimerization interface (F188Y, V186A). Cryo-electron microscopy of isolated FliG D124Y basal bodies showed that D124Y substitution partly restored the C ring diameter towards wild-type values. These results demonstrate that either subunit number or spacing variation can change C ring size. While FliF.FliG N association primarily transmits torque, it also has downstream effects on the switch machinery for torque reversal. There seems to be a synergistic relationship between the supramolecular organization of the FliG N and FliM M rings with defects in one restored by compensating alterations in the other ( Figure 6B).
FliGM fold [125], and possibly also conformational plasticity compatible with the compensatory interactions between the two domains in the regulation of C ring size. The crystal structure of the H. pylori co-folded domain corroborated the essential features reported for T. maritima [126]. The cytoplasmic FliF segment contiguous with the co-folded FliGN.FliFC-tail domain is predicted to form a predominantly α-helical connector to the periplasmic C-terminal FliF (FliFC per ) [112]. The recent 3D-cryoelectron microscopy reconstruction of overproduced S. enterica FliF rings has shown that FliFC per forms a periplasmic scaffold with a split ring-building motif (RBM) that staples together an anti-parallel β-barrel to form the external periplasmic modules of the MS ring while the N-terminal FliF forms inner RBM modules structurally homologous to RBMs characterized for Type-III How does FliG N transmit torque generated at the FliG ring periphery to the axial rod and filament via FliF? The architecture of the S. enterica hook basal body revealed by the work of DeRosier and colleagues (see [50] and references therein) notably shows different symmetries for the internal and external modules; the cytoplasmic C ring (n = 33-36) and the external hook (n = 11), for example. The architectural design by which torque transmission is achieved and symmetry mismatch accommodated is starting to be understood. The structural adaptation for torque transmission was determined in T. maritima [123] and, subsequently, the sodium-powered V. alginolyticus motor [124]. NMR reported extensive conformational changes in the T. maritima FliG N ARM fold upon interaction with 46 FliF C-terminal residues. These changes were due to formation of the co-folded FliG N and FiF C-tail domain. FliG N in the crystallized co-folded domain alters conformation to closely match the FliG M fold [125], and possibly also conformational plasticity compatible with the compensatory interactions between the two domains in the regulation of C ring size. The crystal structure of the H. pylori co-folded domain corroborated the essential features reported for T. maritima [126].
The cytoplasmic FliF segment contiguous with the co-folded FliG N .FliF C-tail domain is predicted to form a predominantly α-helical connector to the periplasmic C-terminal FliF (FliF C per ) [112]. The recent 3D-cryoelectron microscopy reconstruction of overproduced S. enterica FliF rings has shown that FliF C per forms a periplasmic scaffold with a split ring-building motif (RBM) that staples together an anti-parallel β-barrel to form the external periplasmic modules of the MS ring while the N-terminal FliF forms inner RBM modules structurally homologous to RBMs characterized for Type-III injectosomes [127]. Both the split RBM and the β-barrel were predicted by residue coevolution in the course of ongoing work on the full-frame FliF.FliG fusion ring [112]. Thus, the FliF flagellar motor scaffold has evolved to add a particularly stable, periplasmic C-terminal domain to the injectosome RBMs. The map further reveals that FliF C per symmetry (n = 33-34), within the design tolerance reported for other biomolecular assemblies, matches the C ring symmetry to dispel the MS and C ring symmetry mismatch conundrum raised by initial estimates of a lower FliF subunit stoichiometry. Figure 7 summarizes these advances.
Biomolecules 2020, 10, x FOR PEER REVIEW 16 of 26 RBMs. The map further reveals that FliFC per symmetry (n = 33-34), within the design tolerance reported for other biomolecular assemblies, matches the C ring symmetry to dispel the MS and C ring symmetry mismatch conundrum raised by initial estimates of a lower FliF subunit stoichiometry. Figure 7 summarizes these advances.  Figure 5D (modified from [112]). (B) The 3D model of the S. enterica hook-basal body complex. The basal body part of the structure was from a 3D reconstruction as published [50]. The hook with attached FliD cap structure was done by single-particle methods using the entire hook with the cap as the single-particle (Dennis Thomas, unpublished results (with permission)). (C) The 3D map of the S. enterica FliFC torque transmitter module (EMD-10143, 3.1-angstrom resolution (modified from [127])). The en-face view resolves 33-subunits.
In summary, rotor-stator interactions employ a core set of charged residues and control large scale changes in stator assembly and C ring morphology either side of the interaction interface. The FliFC per module presumably templates the assembly of the FliG ring via FliFC-tail that is part of the cofolded domain. FliGM then dictates FliM assembly via contacts with FliMM. The FliMM ring drives reorientation of FliGC for bidirectional torque generation and remodels in response to lesions in the co-folded FliGN.FliFC-tail domain, Thus, there is conformational coupling of FliMM to these distant  Figure 5D (modified from [112]). (B) The 3D model of the S. enterica hook-basal body complex. The basal body part of the structure was from a 3D reconstruction as published [50]. The hook with attached FliD cap structure was done by single-particle methods using the entire hook with the cap as the single-particle (Dennis Thomas, unpublished results (with permission)). (C) The 3D map of the S. enterica FliF C torque transmitter module (EMD-10143, 3.1-angstrom resolution (modified from [127])). The en-face view resolves 33-subunits.
In summary, rotor-stator interactions employ a core set of charged residues and control large scale changes in stator assembly and C ring morphology either side of the interaction interface. The FliF C per module presumably templates the assembly of the FliG ring via FliF C-tail that is part of the co-folded domain. FliG M then dictates FliM assembly via contacts with FliM M . The FliM M ring drives reorientation of FliG C for bidirectional torque generation and remodels in response to lesions in the co-folded FliG N .FliF C-tail domain, Thus, there is conformational coupling of FliM M to these distant domains via FliG M . The correlation between C ring size and rotation state, driven by defined lesions, is comparable to the rotation-fluorescence intensity correlation reported for GFP-tagged motors. It provides the first structural clue for the difference reported for the CW versus CCW torque-velocity relations.

Current Challenges
This review draws overwhelmingly from the E. coli, S. enterica and sodium V. alginolyticus motors for the elucidation of their molecular mechanisms, with a notable contribution from thermophile and H. pylori crystal structures. The fundamental issues regarding switch synchrony and ultra-sensitivity had been posed by 2003. The present knowledge of the structural basis for torque generation and transmission does not answer these issues but takes their study to a new level. Other species with diverse phylogeny have revealed the diversity of the switch operation. The bacillus B. subtilis has inverted motile responses to CheY. V. alginolyticus is one of several bacteria that alter switch frequency rather than rotation bias in response to CheY activation. The flagellum of the α-proteobacterium R. sphaeroides stops and starts. The thermophile A. aeolicus has been reported to lack FliM, consistent with its mostly smooth-swim motile behavior [128]. A major new challenge as illustrated by the phylogenetic tree of FliG C , the "motor" domain ( Figure 8A), is to explain how such diversity can arise from a small core of protein components.
Cryo-electron tomography (cryo-ET) has been an important role for the appreciation of the diverse morphology of flagellar motors (reviewed in [129,130]), even though it is presently limited to thin bacteria or mini-cells. Crucially, the technique provides 3D-reconstructions of complete motors that capture rotor-stator interactions [131], the effect of the cell membrane upon rotor flexure [132] and novel cell wall motor components [133]. Mutagenesis guided Cryo-ET has, interestingly, established diversification of C ring architecture for various cellular functions within a single species (Campylobacter jejuni) [133], and has led to the realization that high-torque motors have larger diameter stator and C rings [134].
The emergence of informatics due to developments in high-throughput sequencing, mass spectroscopy and high-performance computing is the highlight of this era, as appreciated by the increase in protein sequences (135,850 (2003) -> 177,754,527 (2020) (www.ebi.ac.uk/uniprot)) and atomic structures (<20,000 (2003) -> >140,000 (2018) [135]). Computational strategies for protein dynamic simulations [87,136] have benefited from the expanded databases. Most core switch components belong to a larger superfamily. The remarkable diversity of the CheY response regulator superfamily has been reviewed recently [137]. Several response regulators are expressed concurrently within one species. Many species have multiple CheYs of which typically one interacts with the flagellar motor: a study on V. cholerae being an early example [138]. FliF, FliM, FliN and FliY are members of larger families that include Type III injectosome components (reviewed in [139]). FliG has distant homology to the MgtE transporter family [140]. The diversity of the basal C ring and the CheY atomic structures suggests that bacteria utilize multiple strategies for signal reception.
The E. coli and S. enterica bacteria remain the primary source for motor rotation assays in conjunction with native and chimeric V. alginolyticus sodium motors. Structural knowledge, while anchored in these bacteria, now has important contributions from other species driven by the developments in cryo-ET and informatics. The current understanding of the flagellar motor switch is based on a generic, integrated assembly with contributions from multiple species, as schematized in Figure 8B.    Charged residues on a single FliG C α-helix (torque helix) are the primary determinant of rotor-stator interactions. FliG N co-folds with a FliF C-terminal fragment (FliG N .FliF C-tail ) to form a domain with a similar architecture to FliG M ; FliG N full-frame or deletion fusions can be compensated by engineered modifications in the FliG M domain. The co-folded domain connects to FliF C per , the assembly scaffold for the C ring. The matching 33-34 symmetries of FliF C per and the C ring support the 1:1 stoichiometry for FliG N ,FliF C-tail seen in the crystal structures. A chained FliG C -FliG M stack is the working model for rotor ring organization, but its assembly is probably modulated by other C ring components whose genes encode mot alleles [8,9]. There is not a straightforward match of the FliG subunit symmetry with the 26 steps per revolution resolved in rotation assays. Other stator-rotor contacts may determine step periodicity. The in-situ conformation and dynamics of the linkage between the chained FliG C -FliG M , the co-folded FliG N .FliF C-tail domain and FliF C per remain to be determined.
The connectivity of the molecular linkage between the FliM N CheY binding site and the FliG C torque helix is now understood in broad outlines. FliM N tethered the CheY associates with secondary binding sites on the basal C ring that may differ with species. FliM M monomers fluctuate between bi-stable conformations that likely reflect the CCW and CW operational states. These states propagate across the C ring via inter-domain contacts and across a conserved interface to FliG M . FliG M and its stacking contact with FliG C ARM-C are dynamic. Helix MC melting within the flexible inter-domain GG linker in conjunction with adjacent α-helices orchestrates these dynamics. The FliGc intra-domain MFXF hinge amplifies FliG M motions to cause a large re-orientation of the FliG C αC 3-6 torque helix subdomain. These reorientations may reflect CCW <-> CW transitions. CW and CCW-locked motors have different torque velocity relations. Substitutions that alter the fold of FliG C αC 3-6 affect MFXF hinge dynamics and selectively impair CCW rotor-stator interactions. Residue substitutions that enhance CW or CCW bias have different FliG M dynamics. Rotation state is influenced by changes in C ring size due to subunit number or packing variations. C ring morphology is also altered by the presence and activity of the stator complexes. The molecular basis of the coupling between FliG M dynamics and CCW <-> CW transitions is not defined, let alone the coupling between the C ring dynamics and the torque-velocity relation. Adaptive subunit exchange of GFP-tagged basal-ring components may influence switch ultra-sensitivity to CheY activity. The C-ring is disrupted by single residue perturbations [141] and might also be impaired by GFP, which has a similar size to the proteins tagged. However, the CCW adaptive increase in basal body fluorescence is difficult to explain by GFP perturbation of C ring assembly. Nevertheless, the correlation between adaptation and subunit turnover is qualitative and determination of whether it is determinative or incidental must await structural elucidation of the turnover mechanism. Finally, the integrated picture is based on studies of a few species. The complete diversity of the species under current study is substantially more, as glimpsed in motor output, CheY function, basal C ring composition and C ring architecture. The determination of the conformational transitions of the stator complexes during the work cycle is a likely prerequisite for the explanation of their long-range effects on C ring morphology. More generally, elucidation, as opposed to description, of switch diversity is a severe challenge that may be intractable is the absence of the fundamental signal and energy transduction mechanism.
Mathematical models have closely tracked the progress in motor physiology. The conformational spread model has been modified and extended, while a fundamentally different non-equilibrium motor model has been developed. These advances have led to the appreciation that switch operation cannot be understood in isolation from motor mechanics. However, the details of the chemical machinery that these models seek to explain has remained at the 2003 level. There is an urgent need for a top-down development of these models to discriminate between possible molecular mechanisms. MD simulations based on X-ray structures, NMR, ESR spectra and foot-printing techniques offer a bottoms-up approach to supply the kinetics required to correlate the molecular level descriptors known thus far with motor physiology. An atomic-level model of a rotor module from even one species will be a game-changer for a kinetic model with the necessary predictive power, although the scale-up of the simulations will be a logistic challenge with present-day resources. Nevertheless, research on this remarkable biomolecular machine has consistently advanced in unforeseen ways with fundamental implications for protein energetics and allostery. The innovation of its research community will ensure that it continues to do so.
Funding: This article received no external funding.