In situ structure of the Caulobacter crescentus flagellar motor and visualization of binding of a CheY‐homolog

Abstract Bacterial flagellar motility is controlled by the binding of CheY proteins to the cytoplasmic switch complex of the flagellar motor, resulting in changes in swimming speed or direction. Despite its importance for motor function, structural information about the interaction between effector proteins and the motor are scarce. To address this gap in knowledge, we used electron cryotomography and subtomogram averaging to visualize such interactions inside Caulobacter crescentus cells. In C. crescentus, several CheY homologs regulate motor function for different aspects of the bacterial lifestyle. We used subtomogram averaging to image binding of the CheY family protein CleD to the cytoplasmic Cring switch complex, the control center of the flagellar motor. This unambiguously confirmed the orientation of the motor switch protein FliM and the binding of a member of the CheY protein family to the outside rim of the C ring. We also uncovered previously unknown structural elaborations of the alphaproteobacterial flagellar motor, including two novel periplasmic ring structures, and the stator ring harboring eleven stator units, adding to our growing catalog of bacterial flagellar diversity.

the bacterial flagellar motor. In enteric bacteria, additional ring-like structures, called the MS-, P-and L rings span the inner membrane, the peptidoglycan layer and the outer membrane. The MS ring serves as assembly platform for the cytoplasmic part of the motor and transmits the torque from FliG to the rod. P-and L-rings act as bushings for the axial driveshaft which transmits the torque to the extracellular flagellum (Berg, 2003;Morimoto and Minamino, 2014;Minamino and Imada, 2015;Beeby et al., 2020).
The chemotaxis response regulator CheY is the best studied effector protein directly interacting with the C ring. Extracellular changes of attractant and repellent concentrations are sensed by chemoreceptors, which alter the phosphorylation state of CheY. This allows binding to a conserved motif present at the N-terminus of FliM (Lee et al., 2001;Dyer et al., 2004;Dyer and Dahlquist, 2006), enabling "navigation," or chemotaxis, toward attractants and away from repellents. CheY-P-binding is thought to induce large conformational changes in the C ring between FliG and the stator complexes, leading to reversal of flagellar rotation. In the model bacteria Escherichia coli and Salmonella enterica Serovar Typhimurium with flagella situated randomly over the cell body, rotation switches from a counter-clockwise (CCW) to a clockwise (CW) direction to trigger a random cell reorientation by tumbling (Bren and Eisenbach, 1998;Stock et al., 2012). In polar flagellates like Campylobacter, Vibrio, Shewanella, or Caulobacter species, motor reversal leads to cell reversal and reorientation (Hugdahl et al., 1988;Xie et al., 2011;Bubendorfer et al., 2014;Lele et al., 2016).

A recent study revealed a stable interaction between several
CheY-like proteins and the flagellar motor in the relatively thin, polar flagellated freshwater bacterium Caulobacter crescentus (Nesper et al., 2017). Its characteristic asymmetric cell division leads to the formation of two different daughter cells: a surface attached "stalked" cell and a flagellated "swarmer" cell. Swarmer cells can disperse from biofilms, allowing the attachment on a new surface, and thus, the colonization of new environments. The small signaling molecule cyclic-di-GMP is known to orchestrate the transition from motile to sessile lifestyles in many species . The C. crescentus genome encodes 12 CheY homologs. In pusher mode, when the cell swims straight forward, its flagellar motor rotates by default in a CW direction. Binding of the major CheY protein involved in C. crescentus chemotaxis, CheYII (Nesper et al., 2017), reverses rotation to CCW, causing the cell to swim backward (Koyasu and Shirakihara, 1984) motor speed and promote smooth swimming upon surface encounter (Nesper et al., 2017). The protein CleD inhibits motility at high c-di-GMP conditions, enabling rapid surface attachment and biofilm formation. The interaction with FliM was strong enough to observe c-di-GMP-dependent colocalization of GFP-tagged versions CleD. It even allowed co-purification of FliM-CleD complexes (Nesper et al., 2017). Such a stable interaction between an effector protein and the flagellar motor presents the ideal model system for electron cryotomography (cryoET) and subtomogram averaging (STA).
While the intracellular signaling molecules responsible for switching of rotational direction are well-studied in diverse bacteria (Toker and Macnab, 1997;Bren and Eisenbach, 1998;Szurmant et al., 2003;Ahn et al., 2013;Henderson et al., 2020), the molecular mechanisms of action of effectors binding to the C ring are unclear. Transient interactions such as CheY binding to the flagellar motor are difficult to study with conventional structural biology approaches. Although the binding interface of CheY and FliM is known, the specific stoichiometry and three-dimensional arrangement of FliM in the C ring are unresolved (Kim et al., 2017). While it is possible to isolate basal bodies and even C rings in S. enterica (Thomas et al., 2006;Sakai et al., 2019), it is difficult to study their transient interactions with relatively small effector proteins. It is even more challenging to study the in vivo interactions between effector proteins and the bacterial flagellar motor (Sourjik and Berg, 2002;Di Paolo et al., 2016). CryoET offers a remedy, visualizing protein complexes in situ within flash-frozen cells. In a process called STA, details can be subsequently resolved by averaging many identical particles such as flagellar motors (Zhao et al., 2014;Rossmann and Beeby, 2018). Because cryoET resolutions are inversely proportional to sample thickness, the transient interaction of few CheY molecules is difficult to capture by cryoET. This becomes possible when focusing on targets at the polar regions of thin cells such as C. crescentus.
In this study, we examine the molecular architecture of the flagellar motor and specifically the C ring of C. crescentus when all cheY genes are deleted. This provides additional information about the flagellar motor structure of C. crescentus, which is an important model organism for motility and biofilm formation. Subsequently, we overexpressed one CheY-like protein, CleD, in the CheY-gutted strain and imaged the putative binding location of this protein in detail. This improves our understanding of the mechanism of action of these c-di-GMP-binding proteins, and we also present the first structure of a CheY-homolog bound to the flagellar motor at nanometer resolution. This demonstrates the potential of cryoET and STA to not only acquire in situ structures, but also capture interactions of multi protein complexes with effector molecules.

| Deletion of all cheY genes leads to straight swimming without reversals
To image binding of a single effector to the C ring, we first sought to remove all FliM-binding proteins so that we could subsequently reintroduce a single effector to study in isolation. We deleted all 12 cheY genes encoded in the genome (Figure 1a)

| Structure of the bare CW-locked C. crescentus C ring in situ
We next used the ΔcheYs strain to understand the molecular archi- Although the C. crescentus motor had the same conserved overall core structure as previously described (Chen et al., 2011), the higher resolution allowed us to discern finer details. The C ring, with a diameter of 35 nm, exhibited the known peanut-shape in a cross-section ( Figure 2b). The thicker lower part is probably composed of FliN, a thinner middle part might be formed by FliM and a slightly thicker upper section is commonly assigned to FliG.
This improved structure of the C. crescentus ΔcheYs C ring, therefore, allowed us to lay the groundwork for an in-depth molecular understanding of the C ring without the interference of CheY-like effectors.

| The C. crescentus motor features 11 putative stator complexes and a symmetry-mismatched 13-fold symmetric periplasmic ring
We also observed additional periplasmic ring structures in the C. crescentus motor. A ring above the MS ring with a diameter of 17 nm corresponds to the previously reported E ring (Stallmeyer et al., 1989). A 30-nm wide ring encircles the P and L rings beneath the outer membrane, which we name the outer membrane (OM) ring, and a second ring with a diameter of 40 nm located just below the first ring. Thirteen 7.5-nm long spoke-like structures extend from the wider ring, giving it a cog-like appearance with a total outer diameter of 55 nm in the horizontal central slice (Figures 2c, S3a,b).
This structure is described as "cogwheel" in Figure 2e.
Another density just above the inner membrane, corresponding to the location of the stator complexes, was clearly visible in a vertical section of the STA ( Figure S3a,b). However, similar to the previous structure (Chen et al., 2011), this density appeared fuzzy and the stator symmetry could not be resolved. Since the 13-fold symmetric cogwheel seemed to have driven the rotational alignment during STA, we speculated that the stator ring might exhibit a different rotational symmetry. The custom mask used for the alignment of the subtomograms in this STA included the whole periplasmic and cytoplasmic section of the motor ( Figure S3c). Thus, we attempted to redo the alignment, masking out the stator region ( Figure S2g) and exclusively allowing rotational searches around the y-axis.

| CleD overexpression leads to a bulkier C ring, indicating an outside-facing binding site for CheY-homologs
We were now poised to ask where CheY-homologs bind the C ring.
Since we suspected that basal expression of these individual proteins and physiological c-di-GMP concentrations were insufficient to saturate effector binding to the flagellar motor, we decided to overproduce CleD at high intracellular c-di-GMP concentrations (ΔpdeA).
CleD was selected because of its higher molecular mass and its ability to form stable complexes with FliM (Nesper et al., 2017). Most of the CheY residues known to be involved in interaction with FliM in E.
coli are conserved in CleD ( Figure S1b). To this end, we constructed a ΔcheYs ΔpdeA strain ectopically overexpressing CleD under the control of a xylose-inducible promotor (pMT375 CleD-GFP). GFP was fused to the C-terminus of CleD to confirm its subcellular localization to the flagellated cell pole and to increase its overall mass to generate more density in the STA.
The observation that this strain showed lower motility on semisolid agar plates ( Figure S4a) and that CleD-GFP localized to the pole opposite the stalk ( Figure S4b) confirmed that this CleD variant is able to interfer with flagellar motility at high c-di-GMP concentrations (Nesper et al., 2017). Thus, this overexpression strain ensured maximal occupation of CleD binding sites, increasing the chances to successfully detect additional densities derived from bound effector proteins.
We then used this strain to determine the structure of the C ring with bound CleD. 150 motors were averaged as indicated above, generating an independent motor structure without using the ΔcheYs structure as reference for alignment ( Figure S5a). This strain and all other strains used for generating subsequent STAs frequently formed two flagellar motors at the pole opposite the stalk.
Since both motors formed filaments, we assume that they are fully assembled and functional ( Figure S2c).
Comparison of the motor structure obtained from the ΔcheYs ΔpdeA pMT375 CleD-GFP strain and the subtomogram average of the motor from the ΔcheYs strain revealed a subtle but distinct thickening at the outer lobe of the middle part of the C ring ( Figure 3a).
While the width of the bottom part of the C ring density remains around 4 nm in all structures, the characteristic constriction of the C ring was not visible in the CleD overexpression strain. This strain exhibits a significant thickening and an additional density at the outer rim of the C ring. Exact measurements of the thickening of the C ring were difficult to obtain due to an additional diffuse density at the outer lobe of the C ring.
We hypothesized that this fading signal from the major density results from the conformational flexibility of the C-terminal GFPtag, which was fused to CleD by a flexible linker sequence. To increase the rigidity of the C-terminal part of this fusion protein, we redesigned the linker region of the construct by deleting the last six amino acid residues which are disordered according to PSIPRED (Jones, 1999) ( Figure S1b)and fused it to GFP via a rigid α-helix linker sequence (Chen et al., 2013). The expression of this protein on the plasmid pMT375 CleD-Δ6-α-GFP was supposed to boost the signal in this area. Subsequent, reference-free STA with 197 subtomograms revealed a structure with a similarly diffuse density outside the C ring. However, crucially, and reinforcing our proposed visualization of CleD binding ΔcheYs ΔpdeA pMT375 CleD-GFP, the bulky outside facing density remained present (Figures 3a, S5c).
These results strongly supported the hypothesis that the additional density at the outer lobe of the C ring is derived from bound CleD. However, the relatively modest density increase required a negative control experiment to exclude the possibility that this density is caused by experimental artifacts. STA on a ΔcheYs ΔpdeA strain with a plasmid expressing only GFP was chosen to control for GFP alone contributing to the additional density. This strain carrying the plasmid pMT375-GFP was treated and induced in the same way as the two CleD-GFP overexpression strains. 114 motors from this strain were used to generate a reference-free STA. The central slice of the C ring structure from the empty vector control did not show the extra density resulting in the same peanut-shape as the ΔcheYs C ring with a width of around 3 nm at the constriction (Figures 3a,   S5g). This confirms that the extra density is not caused by experimental artifacts from the xylose-inducible plasmid system, free GFP or other bound effector proteins responding to high c-di-GMP concentrations.
We further sought to test whether our additional density corresponded to recent findings that CleD interacts with FliM in the C ring via a conserved, N-terminal binding sequence which resembles the CheY binding motif in other flagellar systems (Nesper et al., 2017). We tested the prediction that our putative CleD additional

| D ISCUSS I ON
In this study, we aimed to image the structure of the C. crescentus C ring when a CheY-like effector protein is bound. This was achieved by acquiring the structure of the C. crescentus flagellar motor when all CheY homologs were deleted using cryoET and STA. We then The ΔcheYs strain was used to acquire a structure of a simplified flagellar motor without any bound accessory proteins. The reduced swimming speed of the CheY-gutted strain compared to the wildtype might resemble the speed reduction of ΔcleA, ΔcleA-E or ΔcheYI mutants in previous experiments (Nesper et al., 2017). However, the specific cause of the reduced swimming speed of the ΔcheYs strain is unknown. Directional switching, however, was absent beyond noise due to the lack of FliM-binding chemotaxis response regulators.
These experiments confirm that probably all potential c-di-GMP-dependent FliM-binding effector proteins modulating motor function were deleted in the ΔcheYs mutant. This drastically reduces the complexity of motor-associated effector network in the ΔcheYs mutant, allowing the isolated functional and structural examination of specific effector proteins. Together, this suggests that our ΔcheYs motor structure probably presents the molecular arrangement of a C ring without any c-di-GMP-dependent or chemotaxis-related, bound effector proteins.
Although the resolution of our C. crescentus motor structures remain modest, they are a considerable improvement upon previously published structure (Chen et al., 2011), revealing three previously unseen periplasmic ring structures as well as bound stator complexes, adding to our picture of the structural diversity in flagellar motors, including those from gammaproteobacteria (Zhu et al., 2017;Kaplan et al., 2019;Ferreira et al., 2019), epsilon-and deltaproteobacteria (Beeby et al., 2016;Chaban et al., 2018) and spirochetes (Qin et al., 2018). The composition and the function of these three periplasmic ring structures remain elusive. The presence of a "larger outer disk" had previously been hinted at (Stallmeyer et al., 1989), reporting enlarged 31-or 34-nm wide L rings, depending on the purification method, possibly corresponding to the 30 nm OM ring and/or the 40/55 nm cogwheel from our structure. The presence of two distinct rings and the variable stability of these structures under different purification conditions suggest that these rings are composed of at least two proteins. Outer membrane-associated rings found in other organisms (Beeby et al., 2016;Zhu et al., 2017;Chaban et al., 2018;Zhu et al., 2018;Kaplan et al., 2019;Ferreira et al., 2019) but so far have not been conclusively allocated a function. Previous studies in gammaproteobacteria associated the presence of outer rings with higher motor speed (Kaplan et al., 2019), penetration of the outer layer of the bacterial envelope (Zhu et al., 2018) or ejection of flagella (Ferreira et al., 2019). We could not, however, identify homologs of the putative major components of the H ring from polarly flagellated gammaproteobacteria or the basal disk in epsilonproteobacteria in C. crescentus. Because the 13-spooked cogwheel is located 5 nm away from the outer membrane, we speculate that it is unlikely to directly interact with the outer membrane. What other functions these ring structures might have remains to be understood.
The additional smaller ring above the C. crescentus MS ring, not found in gammaproteobacteria, may be the enigmatic E ring suggested to be involved in flagellar ejection (Stallmeyer et al., 1989;Kanbe et al., 2005). Interestingly, all currently known flagellar motor structures from epsilon-and deltaproteobacteria possess a similar, additional E ring structure above the MS ring. Phylogenetic analysis of core rod proteins revealed that these species are all in the same clade, indicating that the E ring might be an inherent feature in alpha-, epsilon-and deltaproteobacteria ( Figure S5).
However, we failed to identify suitable common candidate proteins, which might be responsible for the formation of this ring.
Nevertheless, the E ring represents yet another evolutionary elaboration in flagellar motors.
The 11-fold symmetry of the stator ring reveals new insights into the torque-generating part of the motor. This small stator ring with the fewest stator units encountered so far reflects the low torque condition this fresh water bacterium experiences in its environmental niche (Li and Tang, 2006;Beeby et al., 2016). This is also the first ever described stator symmetry in a motor where no scaffolding structures like MotXY in polar-flagellated gammaproteobacteria or PflAB in epsilonproteobacteria are present. This suggests discrete binding positions for the stator complexes independent of any scaffolding proteins. These locations might therefore be determined exclusively by interaction with FliG and steric effects. A mean occupancy of eleven stators was also predicted for E. coli (Khan et al., 1988;Reid et al., 2006;Leake et al., 2006) but so far stator symmetry has not been observed in situ. This could, in part, be due to decreased numbers of incorporated stator units under low motor loads (Lele et al., 2013). C. crescentus might exhibit maximal stator occupancy also at lower load. Together, this suggests that the C. crescentus stators might be less dynamic than their E. coli counterparts and are optimized for low torque conditions.

| Bacterial strains and growth conditions
C. crescentus CB15 strains were cultivated in rich medium (peptone yeast extract; PYE) under vigorous shaking at 30°C (Ely, 1991

| Strain constructions
Marker-less deletions in C. crescentus were generated using the standard two-step recombination sucrose counter-selection procedure based on pNPTS138-derivatives. E. coli S17.1 was used to transfer suicide plasmids by conjugation into C. crescentus strains (Ely, 1991). Transformation of cells with pMT375 was performed by electroporation. The bacterial strains, plasmids and oligos used in this work are summarized in Table S1, S2 and S3.

| CryoET sample preparation
All C. crescentus strains were cultivated to mid-exponential phase (OD 0.3-0.5) and induced as necessary. Cells were subsequently pelleted and resuspended to an OD600 of ~20. Ultraufoil R2/2 grids (200mesh) (Quantifoil Micro Tools GmbH) were glow-discharged for 60 s at 10 mA and a solution of 10 nm colloidal gold in 1% of BSA was pelleted and mixed with 25 μl cells immediately before plunge freezing. A 2.5 μl droplet of this sample solution was applied to the glow-discharged electron microscopy grid; the grid was blotted and plunge-frozen into a liquid ethane-propane mixture using a Vitrobot plunge-freezing robot (FEI Company) with a wait time of 60 s, a blot time of 5 s, and blot offsets −3 mm. Grids were stored under liquid nitrogen until data collection.

| CryoET of C. crescentus
Tilt series were collected on a 200-kV FEI Tecnai TF20 FEG transmission electron microscope (FEI Company) equipped with a Falcon II direct electron detector camera (FEI Company) using a Gatan 914 cryoholder.
Tilt series were recorded from an angle of −54° to +54° with an increment of 3° and a collected defocus between −3 μm and −4 μm using Leginon automated data-collection software (Suloway et al., 2009)

| STA
Tomograms were reconstructed automatically using RAPTOR (Amat et al., 2008) and the IMOD package (Kremer et al., 1996). Low-defocus images were low-pass filtered to remove data beyond 3.5 nm. Positions of flagellar motors in tomograms were initially aligned manually along their rotational axes using different custom masks ( Figure S3c,g).

| STA data visualization
For comparison of different structures, STAs were normalized and low-pass filtered as required using the EMAN (Ludtke et al., 1999) and EMAN2 software package (Tang et al., 2007). STAs were subsequently aligned in PEET (Nicastro, 2006). Difference maps were generated using the EMAN package (Ludtke et al., 1999). Isosurfaces were created using UCSF chimera (Pettersen et al., 2004). Intensity plots were made in ImageJ. Radial averaging was performed using C11, C13 or C100 symmetrization.

| Phylogenetics
Phylogenetic trees were generated essentially as described previously (Chaban et al., 2018). The protein sequences of four flagellar rod proteins FlgB, FlgC, FliE, FliF were aligned, trimmed and concatenated. Bootstrap sampling was done 1,000 times with a termination criteria of 10,000 generations without a topology improvement of 0.01 in the lnL score. Bootstrap support was converted to percentage and added to the base trees.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

AUTH O R CO NTR I B UTI O N S
FMR, IH, UJ and MB conceived and designed the study; FMR, IH and MS collected the data; FMR, IH, MS, UJ and MB analyzed the data; FMR, IH, UJ, MS and MB wrote the manuscript; and all authors approved the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
The subtomogram averaging data is available on EMDB (ΔcheYs