Modulation of flagellar rotation in surface-attached bacteria: A pathway for rapid surface-sensing after flagellar attachment

Attachment is a necessary first step in bacterial commitment to surface-associated behaviors that include colonization, biofilm formation, and host-directed virulence. The Gram-negative opportunistic pathogen Pseudomonas aeruginosa can initially attach to surfaces via its single polar flagellum. Although many bacteria quickly detach, some become irreversibly attached and express surface-associated structures, such as Type IV pili, and behaviors, including twitching motility and biofilm initiation. P. aeruginosa that lack the GTPase FlhF assemble a randomly placed flagellum that is motile; however, we observed that these mutant bacteria show defects in biofilm formation comparable to those seen for non-motile, aflagellate bacteria. This phenotype was associated with altered behavior of ΔflhF bacteria immediately following surface-attachment. Forward and reverse genetic screens led to the discovery that FlhF interacts with FimV to control flagellar rotation at a surface, and implicated cAMP signaling in this pathway. Although cAMP controls many transcriptional programs in P. aeruginosa, known targets of this second messenger were not required to modulate flagellar rotation in surface-attached bacteria. Instead, alterations in switching behavior of the motor appeared to result from direct or indirect effects of cAMP on switch complex proteins and/or the stators associated with them.


Introduction
Biofilms are clinically relevant, surface-associated multicellular bacterial communities. In the Gram-negative opportunistic pathogen Pseudomonas aeruginosa, a single polar flagellum and Type 4 pili (T4P) are important for biofilm initiation and maturation, respectively [1]. Reversible, flagellar-mediated attachment of bacterial cells to a surface is somehow sensed, resulting in cessation of flagellar rotation and a transition to irreversible attachment [2,3]. This step precedes surface-associated growth and T4P-mediated surface colonization, precursors to the formation of a matrix-encased biofilm community. The initial event of surface sensing is essential for this process, but how this bacterial sense of "touch" is perceived and transmitted remains unknown. A favored hypothesis is that increased flagellar load associated with surface tethering serves as a signal for surface attachment [4], but mediators that could transmit such a signal have not been definitively identified.
One of the earliest described examples of flagellar mechanosensing occurs in Vibrio parahaemolyticus. Lateral flagella (laf) genes, required for swarming motility, are transcribed when bacteria are transferred from liquid culture to high viscosity liquid or surface growth conditions [5]. Expression of a laf::lux transcriptional reporter can be induced by adding antibodies that specifically bind to and inhibit rotation of the organism's polar flagellum, suggesting that motion of the polar flagellum may be sensed by Vibrio [6]. Fla-mutants, which fail to assemble a polar flagellum, exhibit constitutive, surface-independent laf::lux expression, consistent with the hypothesis that flagellar rotation/function-and perturbations thereof-can be sensed by a bacterium [6].
Examples of flagellar mechanosensing have since been provided in other systems, where they control not only switches between swimming and swarming behavior (e.g. Proteus mirabilis [7]), but also between swimming and adhesion (e.g. Caulobacter crescentus [8]) or biofilm initiation (e.g. Bacillus subtilis [9]). The ability of the flagellar motor to respond to alterations in load is clearly demonstrated by the load-dependent, step-wise recruitment of force-generating units to the Escherichia coli flagellar motor [4,[10][11][12]. Thus most models of flagellar mechanosensing assume that alterations in flagellar load and function, as might occur in a high-viscosity environment or following flagellar-mediated attachment to a surface, initiate one or more potential signals such as altered proton flow across the motor, changes in stator conformation, or even cell envelope stress/deformation that are subsequently coupled to changes in cellular behavior [13].
The use of microfluidics to force bacterial proximity to a surface, coupled with real-time imaging of bacterial surface interactions, has allowed behaviors that bring bacteria to a surface to be separated from those required for surface sensing or surface adaptation [14,15]. Thus pilus retraction, implicated in the C. crescentus transition from a motile "swarmer" cell to a surface-attached "stalked" cell [16], may contribute by bringing bacteria close to a surface, as pili are dispensable for rapid synthesis of the adhesive holdfast when cells are physically

FlhF is required to stop rotation of bacteria tethered at a surface
FlhF is a highly conserved signal recognition particle (SRP)-like GTPase important for placement and function of the P. aeruginosa polar flagellum [23,24]. We observed that ΔflhF bacteria, which assemble a randomly placed but functional single flagellum, were as defective in biofilm formation as bacteria lacking a flagellum (ΔfliC) ( Fig 1A). As the flagellum influences initial steps in biofilm formation, we developed a single-cell tethering assay to observe bacteria after surface attachment (S1 Fig). The majority of tethered wild-type PAK bacteria (>80%) did not exhibit flagellar rotation after 5 minutes of incubation at an anti-flagellin antibody-coated surface ( Fig 1B). We continued to track the behavior of the cells initially observed to be spinning (Fig 1C, t = 0) and saw that most stopped rotation and became attached to the surface along their long axis in the next 5-15 minutes, which we scored as "attached". In contrast, 58% of ΔflhF bacteria exhibited flagellar rotation at the initial time-point (Fig 1B), and the decay in the number of bacteria still rotating was significantly slower than for wild-type organisms ( Fig  1C). Of note, flagellar rotation ceased with similar kinetics in wild-type bacteria and isogenic mutants lacking T4P (ΔpilA), suggesting that pilus-mediated adhesion did not make a significant contribution to this behavior. As expected, we saw fewer ΔpilA bacteria detach during this assay, consistent with prior observations of P. aeruginosa behavior after flagellar-mediated surface attachment [2]. Thus wild-type P. aeruginosa quickly stop flagellar rotation after tethering, and ΔflhF is required for this behavior.
interacted directly with components of the flagellar rotor or stators. Only FliG, the most membrane-proximal component of the C ring, interacted with FlhF in this assay (Fig 2). No significant variation in the strength of the B2H signal was observed when point mutant alleles of FlhF deficient in GTP binding or hydrolysis were tested for FliG binding (S2 Fig). Thus, FlhF localizes at the flagellar base [24] and appeared to interact with the C ring protein FliG. This placed FlhF in a position where it might be sensitive to interactions of FliG with the torquegenerating stator proteins and/or capable of altering or disrupting these interactions in some way.

Mutations in vfr and fimV suppress flagellar phenotypes of FlhF mutants
The deletion of flhF resulted in bacteria that showed abnormally persistent flagellar rotation at a surface (Fig 1B & 1C). To gain a more comprehensive understanding of how FlhF could control flagellar rotation, we made use of our prior observation that a mutation in the FlhF active site, of arginine 251 to glycine (R251G), results in loss of flagellar swimming motility [23]. FlhF(R251G) is "trapped" in a GDP-bound form [23] and has a dominant-negative (DN) effect on wild-type bacterial swimming motility (S3 Fig). We therefore hypothesized that FlhF interacted with other proteins to stop flagellar rotation, and that this interaction was absent in ΔflhF bacteria and inappropriately robust in bacteria expressing the FlhF(R251G) allele (S4 Fig). We tested this model by carrying out a screen for extragenic suppressors (described in Methods) that would restore swimming motility to paralyzed, FlhF(R251G)-expressing bacteria. 23 independent suppressors were mapped by whole-genome sequencing to vfr, a cAMPdependent transcriptional regulator that controls expression of many genes associated with P. aeruginosa virulence [31], while 4 additional suppressors mapped to fimV, a protein whose expression is positively regulated by Vfr [32](S1 Table). The Vfr mutations found in the suppressor strains either prematurely truncated this protein or altered residues implicated in Vfr dimerization, cAMP binding or DNA binding [33]. We tested these suppressors for their ability to twitch, a form of T4P-dependent motility that requires Vfr-dependent transcription of T4P structural and regulatory proteins, and found all to be twitching-negative, consistent with a loss of Vfr function ( Because of the strong association between Vfr and twitching, we tested whether the loss of T4P was sufficient to suppress the paralyzed swimming phenotype associated with FlhF (R251G). This was not the case, however, as a mutant lacking the major pilin, pilA, was unable to swim when FlhF(R251G) was overexpressed (S6 Fig). Loss of Vfr function leads to an approximately two-fold increase in transcription of the master flagellar activator, FleQ [32], raising the possibility that overexpression of flagellar genes could suppress the paralyzed phenotype of FlhF(R251G) bacteria. However, steady-state FleQ protein levels were indistinguishable in parental and extragenic suppressor strains (S7A Fig), and increased expression of FleQ was not sufficient to restore swimming motility to FlhF(R251G) over-expressing bacteria (S7B Fig). Thus, known effects of Vfr on expression of T4P and flagella did not account for the phenotype of the suppressor strains.
FimV, the second site of extragenic suppressor mutations, is also dependent on Vfr for its expression [32]. FimV is a large (919 aa) protein with an amino-terminal peptidoglycan binding domain, a single transmembrane helix, and a large cytoplasmic domain containing three tetratricopeptide repeats separated by an unstructured region between repeats 1/2 and 3 ( Fig  3A) [34]. FimV is homologous to Vibrio cholerae HubP, a polar organizer [35]; in P. aeruginosa FimV positively regulates adenylate cyclase activity via an unknown mechanism and promotes normal T4P assembly and function [36]. We mapped one extragenic suppressor to a missense mutation (L7P) within the predicted signal sequence of FimV. This single point mutation was introduced into the endogenous fimV gene of wild-type P. aeruginosa by homologous recombination and found to be sufficient to suppress the paralyzed swimming phenotype associated with FlhF(R251G) overexpression ( Fig 3B). We confirmed that fimV(L7P) bacteria had reduced twitching motility and decreased levels of intracellular cAMP, phenotypes described for fimV null mutants [36] and consistent with our observation that steady-state levels of FimV(L7P) were < 5% of the wild-type protein (S8 Fig). We also found that a loss-of-function transposon insertion mutant, PA14 fimV::Tn, was able to suppress the DN phenotype associated with FlhF(R251G) overexpression ( Fig 3B). The suppressor phenotype was specific to fimV, as transposon insertions that disrupted the T4P assembly secretin (pilQ::Tn), or that disrupted components of the Pil/Chp chemotaxis cluster that can also increase intracellular cAMP in bacteria (pilI::Tn, pilJ::Tn, pilG::Tn) [36] were unable to suppress FlhF(R251G)-associated paralysis ( Fig 3C). Thus, the loss of FimV specifically suppresses the DN swimming phenotype associated with FlhF(R251G) overexpression.

FlhF interacts with the carboxy terminus of FimV
We had predicted that our suppressor screen would identify interaction partners of FlhF important for the regulation of flagellar rotation. We used B2H to test for interactions between FlhF, FimV and Vfr. We found no evidence of direct protein-protein interactions between FlhF and Vfr in this assay but did demonstrate a robust interaction between FlhF and the cytoplasmic domain of FimV (aa 491-919) that was stronger when FlhF carried the R251G mutation ( Fig 4A). One extragenic suppressor mapped to a nonsense mutation predicted to truncate FimV after aa 826, suggesting that the extreme carboxy-terminus of FimV, which contains its third tetratricopeptide repeat motif, might be required for FlhF-FimV interactions. Deletion of the final 150 aa of the FimV B2H construct (FimV aa 491-769) abolished FlhF-FimV interaction in the B2H assay ( Fig 4A). We also used homologous recombination to replace P. aeruginosa fimV with the truncated fimV(ΔC) allele. This strain, PAK fimV(ΔC), was able to swim even when the DN FlhF(R251G) allele was overexpressed (S9 Fig). Together, these observations suggested that FlhF interactions with the carboxy-terminus of FimV lead to cessation of flagellar rotation.

FimV is required to stop flagellar rotation in surface-tethered bacteria
Surface-tethered ΔflhF bacteria are defective in stopping flagellar rotation. The hypothesis underlying our suppressor screen predicted that the protein that interacted with FlhF(R251G) to inhibit flagellar rotation inappropriately during swimming would also be required to stop flagellar rotation in surface-tethered cells (S1 Fig). We therefore tested whether a fimV mutant phenocopied ΔflhF bacteria tethered at a surface and found this to be the case for fimV(L7P) expressing cells (Fig 4B). Bacteria expressing the FimV allele lacking the carboxy-terminal domain required for FimV-FlhF interaction (PAK fimV(ΔC)) also showed persistent rotation after surface tethering ( Fig 4B). In aggregate, these observations strongly supported the hypothesis that FlhF interacts with FimV to modulate flagellar rotation. Both fimV(L7P) and fimV(ΔC) bacteria assembled a polar flagellum, unlike ΔflhF bacteria, demonstrating that defective slowing of surface-tethered bacteria is not an artifact associated with assembly of a non-polar flagellum ( Fig 4C).

Adenylate cyclase activity is required for FlhF(R251G) to inhibit swimming
FimV positively regulates the activity of the adenylate cyclase associated with T4P assembly and retraction in surface-associated P. aeruginosa, CyaB [34,37]. However, we did not isolate mutants mapping to cyaB in our extragenic suppressor screen. P. aeruginosa also encodes a second adenylate cyclase, CyaA, which must be deleted in tandem with CyaB to significantly reduce cAMP-dependent phenotypes such as twitching motility and Type 3 secretion system activity [32]. We therefore tested individual ΔcyaA and ΔcyaB mutants, as well as a double ΔcyaAB strain, for their ability to suppress the DN phenotype associated with FlhF(R251G). Only the deletion of both adenylate cyclases resulted in changes in swimming motility as compared to the wild-type parental strain ( Fig 5): overall swimming zone diameter was reduced for ΔcyaAB bacteria, but these cells retained swimming motility despite FlhF(R251G) overexpression. Thus the production of intracellular cAMP by either adenylate cyclase allows FlhF (R251G) to inhibit swimming. cAMP reverses the persistent rotation phenotype of flhF and fimV mutants P. aeruginosa, like other Gram-negative bacteria, can transport extracellular cAMP [38]. The addition of exogenous cAMP to P. aeruginosa adenylate cyclase mutants has been shown to complement phenotypes that depend on intracellular cAMP production [36]. If a cAMP signal is generated as a result of FlhF-FimV interaction in surface tethered bacteria, we reasoned that exogenous cAMP should rescue the phenotype of either ΔflhF or fimV(L7P) that were tethered to a surface. We therefore added extracellular cAMP to the media in our surface tethering assay and found that it was sufficient to stop the rotation of tethered ΔflhF or fimV(L7P) bacteria (Fig 6). This experiment established that the rotation of tethered cells could be modulated in the absence of FlhF or FimV, effectively ruling out the hypothesis that FlhF or FimV were themselves part of a "brake" or "clutch" that stopped flagellar rotation. Instead, FlhF and FimV were upstream of cAMP production.
The cAMP-dependent transcription factor Vfr was also identified as a suppressor in our screen (S1 Table). We confirmed that vfr mutants were defective in stopping flagellar rotation after surface tethering (Fig 6). The addition of exogenous cAMP nonetheless reversed this phenotype, indicating that cAMP does not mediate its effect on flagellar rotation via Vfr.
Polar pili have been implicated as just-in-time adhesins for Caulobacter crescentus, as their assembly by bacteria after surface binding stops flagellar rotation and plays a role in promoting permanent attachment of the organism [8,16]. We tested whether T4P played an analogous role in P. aeruginosa surface attachment by comparing the behavior of surface bound ΔflhF and ΔflhF pilA organisms. In the absence of T4P, flagellar rotation persisted even longer, suggesting that pili contribute to stopping rotation of tethered bacteria. However, exogenous cAMP could still reverse this defect, again arguing that cAMP is likely to have target(s) other than Vfr and T4P in this pathway (Fig 6).

Increased flagellar load is associated with elevated intracellular cAMP levels
Bacteria tethered to a surface by their flagellum experience increased flagellar load as compared to cells swimming in liquid. Increased load results in recruitment of additional stators to the motor [4,10,11], indicating that external load is "sensed". We sought to test whether increased flagellar load is associated with increased intracellular cAMP in the absence of an assay that would allow real-time measurement of intracellular cAMP in individual tethered bacterial cells. A recent cryo-electron tomography study of the polar flagellate Vibrio alginolyticus demonstrated that deletion of flhG, which causes the assembly of multiple polar flagella, was associated with high flagellar load and allowed multiple stator subunits to be visualized in motor structures [39]. We therefore deleted fleN, the homologous regulator of flagellar number in P. aeruginosa [40]. As expected, these bacteria assembled thick bundles of polar flagella ( Fig 7A) and were non-motile, in contrast to peritrichously flagellated bacteria in which multiple non-polar flagella form bundles that power swimming. Intracellular cAMP levels as measured by ELISA were increased in ΔfleN bacteria as compared to wild-type PAK when both were growing in liquid, a low-load condition ( Fig 7B). This increase in cAMP was not observed in the absence of FimV (ΔfleN fimV(L7P)), or when bacteria carried the ΔfleN mutation but could not assemble flagella (ΔfleN ΔflhA). Thus, endogenous cAMP levels increase in a FimV-dependent manner in bacteria experiencing high flagellar load as a result of fleN mutation. In aggregate, our results thus far support a model in which FlhF-FimV interact after bacterial tethering to a surface and bring about an increase in intracellular cAMP ( Fig  8). This signal is sufficient to stop flagellar rotation and promote attachment by acting on a target that is not FlhF, FimV, Vfr or T4P.

The MotCD stator is necessary for flagellar rotation in tethered bacteria, but the MotAB stator is required for rotation to stop
When swimming bacteria approach a surface, attach via their flagellum, and continue to spin, the flagellum experiences a significant increase in load. In E. coli, the flagellum responds to  increased load by recruiting additional stator pairs to the flagellum [4,41]. P. aeruginosa is unusual in encoding two stator pairs that interact with the polar flagellum to drive PMFdependent motility, MotCD (PA1460/PA1461) and MotAB (PA4954/PA4953) [18,19]. The two stators are not functionally equivalent, as only MotCD can support swarming motility Modulation of flagellar rotation in surface-attached bacteria [20]. We hypothesized that MotAB and MotCD might have distinct functions in swimming versus surface-tethered bacteria, and constructed strains lacking motAB or motCD to test this. We confirmed that our mutants displayed previously described phenotypes in plate-based swimming assays, where both are defective vis-a-vis wild-type bacteria, (S10 Fig) and in swarming assays (S11 Fig). Despite their similar phenotypes in swimming agar assays, the ΔmotAB and ΔmotCD bacteria exhibited markedly different swimming speeds and reversal behavior during swimming in liquid media. Speed and reversal frequency of the ΔmotCD mutant, which expresses only MotAB, were close to indistinguishable from those of wild-type PAK, suggesting that swimming may be largely powered by the MotAB stator in liquid (Fig 9). The ΔmotAB mutant had a lower reversal frequency than either wild-type or ΔmotCD bacteria, similar to that of a ΔcheR1 methyltransferase mutant [42] that we constructed as a control, suggesting that the stators influence chemotactic properties of the flagellum.
We hypothesized that bacteria expressing only one of the two P. aeruginosa stators might also differ in their behavior at a surface. ΔmotCD bacteria could be tethered by anti-flagellin antibodies, but very few were able to rotate (S12 Fig), an observation consistent with the proposed role of MotCD in flagellar rotation under conditions of high external load [20]. In , has stronger and/or more persistent interaction with FimV. This causes an inappropriate cAMP increase in liquid-grown bacteria that inhibits flagellar rotation. B. When bacteria tether at a surface, an undefined "signal" causes a conformational change in FlhF that allows it to interact with FimV. This leads to an increase in intracellular cAMP that stops flagellar rotation and promotes bacterial cell body attachment to a surface. If either FlhF or FimV are absent, the interaction does not occur, cAMP levels do not rise, and bacterial rotation persists in surface-tethered cells.
https://doi.org/10.1371/journal.ppat.1008149.g008 Cells of each genotype were imaged under dark-field microscopy at 30 fps. Image analysis was carried out as described in Methods using a custom MATLAB code. In (B), boxplots span 1 st and 3 rd quartiles with a line at the median. The number of cells tracked is shown above each sample in panel B; data from these tracked cells was also used to generate the histogram in panel A, in which the y-axis shows the probability of observing a given velocity.
contrast, tethered ΔmotAB bacteria exhibited prolonged flagellar rotation after tethering, despite the presence of FlhF and FimV (Fig 10A). We found that the addition of exogenous cAMP had a modest effect on the rotation of ΔmotAB bacteria and no effect at all on the rotation of ΔflhF motAB bacteria. These findings demonstrate that the MotCD stator is required for tethered bacteria to spin, but that the MotAB stator is necessary for rotation to stop (Fig 10A).

cAMP changes flagellar reversal behavior in tethered bacteria
We used high-speed video-microscopy to examine more closely the behavior of tethered ΔflhF and ΔflhF motAB bacteria before and after addition of exogenous cAMP. cAMP addition changed the behavior of the ΔflhF rotor in several ways. The median number of reversals observed in tethered bacteria increased significantly (Fig 10B), and CCW rotation speed increased significantly (Fig 10C). In contrast, when tethered ΔflhF motAB were exposed to cAMP, no change in median reversal frequency or rotation speed was observed. Thus observed changes in rotor behavior of tethered bacteria in response to cAMP depend on the presence of the MotAB stator. In aggregate, these findings suggest that the MotAB stator might itself be a direct or indirect target of cAMP.

Discussion
Bacterial binding to a surface via the flagellum is often a first step in the transition to surfaceassociated behaviors such as T4P mediated surface colonization and biofilm formation [2,17]. Ultimately, bacteria commit to these surface-associated behaviors by downregulating flagellar gene expression and upregulating the expression of adhesins and biofilm-matrix components [3]. However, modulation of flagellar rotation, rather than flagellar production, is a strategy that allows bacteria first to reversibly bind to and sample a surface [43][44][45][46].
We have demonstrated that FlhF, a GTPase required for polar flagellar placement, is necessary for flagellar rotation to stop after surface binding. FlhF localizes to the pole [24]and interacts with the C ring protein FliG by bacterial two-hybrid assay (Fig 2). The MotA stator component interacts with FliG to generate torque [47,48], and many regulators of flagellar rotation act by disengaging [45] or jamming [43,44,46,49] the stator-FliG interaction. Thus FlhF seems well-positioned to either alter this rotor-stator interaction or, alternatively, to sense it.
Previous structure-function analysis of the FlhF GTPase established that mutation of the catalytic site arginine results in a protein, FlhF(R251G), that inhibits flagellar rotation and bacterial swimming [23]. This allowed us to design an unbiased genetic screen to identify extragenic suppressors of FlhF(R251G) that restored swimming (S4 Fig). We hypothesized that our screen would identify proteins that interacted with FlhF to stop flagellar rotation, and indeed found that the cessation of flagellar rotation required the HubP-like protein FimV, which interacted with FlhF (Fig 4). An interaction between FlhF and HubP has previously been reported in V. cholerae, although neither protein affects the other's distribution and~95% of ΔhubP bacteria assemble a single polar flagellum [35]. In P. aeruginosa, FimV is a well-known positive regulator of adenylate cyclase activity [36]. Neither of the two adenylate cyclases encoded by P. aeruginosa was identified as a suppressor in our screen, but a mutant lacking both adenylate cyclases (ΔcyaAΔcyaB) did suppress FlhF(R251G) mediated inhibition of flagellar rotation, implicating cAMP in this signaling pathway. Exogenous cAMP could rescue the defect of either ΔflhF or fimV mutant bacteria in stopping flagellar rotation after surface-tethering, a key result indicating that FlhF and FimV participate in the upstream "sensing" part of this surface-adaptation response rather than in the response itself. cAMP is a pleiotropic regulator that interacts with the transcription activator Vfr to upregulate T4P gene expression and also positively regulates T4P-mediated motility, via proteins of the Chp/Pil chemotaxis-like cluster, as recently reviewed [50]. Nonetheless cAMP still stopped flagellar rotation in bacteria lacking Vfr or pili. These findings suggested that cAMP has targets other than T4P or Vfr in this pathway, while subsequent experiments demonstrated that cAMP alters flagellar reversal frequency and rotation speed in a MotAB dependent manner (Fig 10).
Intracellular cAMP production is also associated with a T4P-mediated mechanosensing pathway that allows P. aeruginosa to produce this second messenger after T4P attachment to a solid surface [51,52]. The activation of the Pil/Chp pathway by cycles of T4P attachment and retraction leads to increased CyaB activity and promotes cAMP-dependent virulence behaviors such as Type 3 secretion and biofilm formation [53]. FimV is implicated in T4P-mediated cAMP production and interacts with PilG via the scaffold protein FimL [37]. The FimVdependent pathway that we observed, however, seems to be distinct from T4P-associated cAMP production. In particular, only mutation of FimV-but not mutation of other Pil/Chp proteins associated with cAMP production, or of the CyaB adenylate cyclase linked to T4Pmechanosensing, or even of T4P themselves-was able to suppress FlhF(R251G) effects on the flagellum.
P. aeruginosa is unusual in encoding two stators that both interact with the polar flagellum to drive PMF-dependent motility [18,19]; in other bacteria, different stators may associate with polar vs. lateral flagella, or drive flagellar rotation in response to proton vs. Na + gradients [30]. Prior work has suggested that the two stators of P. aeruginosa are recruited to the flagellar rotor under distinct conditions [20], and our observations agree with this hypothesis. Wild-type and ΔmotCD bacteria swimming through a low viscosity medium have very similar swimming speeds and reversal frequencies, in contrast to ΔmotAB cells, suggesting that the MotAB stator primarily powers swimming under this condition (Fig 9). However, when bacteria were tethered to a surface via their flagellum, continued rotation in this condition of increased drag required the presence of MotCD (S12 Fig). Surface-tethered bacteria were less likely to stop rotating if they lacked MotAB, and exogenous cAMP could not complement this phenotype. High-speed tracking of tethered cells showed that cAMP increased reversal frequency of the flagellum through a mechanism that required MotAB (Fig 10). We have yet to identify the direct target of cAMP in this pathway.
Mechanisms to change flagellar behavior have been reported for many different bacteria, and often coincide with bacterial adaptation to surface-attached growth [43][44][45][46]54]. These changes in flagellar behavior are often mediated by cyclic-di-GMP (cdG) effectors that interact with the motor-switch complex [55]. Many mechanisms increase cdG levels in bacteria. These are often associated with activation of diguanylate cyclases in response to bacterial surface contact [56], and can occur quite rapidly, as has recently been reported following bacterial surface attachment in microfluidics devices [15]. The behavior that we have characterized precedes contact of the P. aeruginosa cell body with a surface and appears to increase the probability that such surface attachment will occur. Our findings are consistent with the recent observation that MotAB is required for rapid polar localization of FimW, an event that reflects a rapid rise in intracellular cdG in surface-attached bacteria [15], as we observe that MotAB is necessary for bacteria to stop flagellar rotation after tethering and become surface-attached.
We propose that swimming bacteria that bind to a surface via their flagellum experience increased load on the flagellar motor (Fig 11). P. aeruginosa cells tethered to a surface can continue rotating, but only if they possess the MotCD stator, suggesting that this stator is recruited to the flagellum under this condition. Wild-type bacteria quickly stop flagellar rotation and become surface-attached, a process that requires interaction between FlhF and FimV. Our data suggest that a cAMP signal is produced as a result of FlhF-FimV interaction, and that cAMP modulates reversal frequency and speed of the flagellar motor in a MotAB-dependent manner. Our data do not explain why the absence of MotAB changes the response of tethered bacteria to cAMP. The changes in flagellar reversal frequency, however, may help drive the bacteria toward the surface [57]. These events likely precede the cdG-dependent signaling cascade described by Jenal and colleagues, which requires MotAB for its initiation [15].
The roles of second-messenger signaling in P. aeruginosa surface-association are complex and likely highly redundant. cAMP signals in surface-associated bacteria are amplified by the Pil/Chp system, which mechano-senses via the retraction of T4P [51,52], and lead to surfaceassociated motility and virulence. Transient "surface-exposure" by swimming cells, however, also changes cAMP levels in individual bacteria that fail to commit to a surface, as recently demonstrated by Lee et al. [58]. These so-called "surface-sentient" cells exhibit higher levels of intracellular cAMP and are more likely to progress to surface association and irreversible attachment on subsequent encounters with a surface. Whether this cAMP-dependent "memory" requires the FlhF-FimV circuit that we have described remains to be tested.

Static biofilm formation
Static biofilm assays were performed as previously described [60]. Briefly, overnight cultures grown in LB were diluted into fresh medium (1:100); 100 μl of the dilution was used to inoculate 3 to 4 replicates per strain in a 96 well plate (Costar #2797). Plates were incubated for 24 h at 30˚C, then gently washed to remove non-adherent bacteria. Adherent biofilm was stained with crystal violet (0.1%), followed by biofilm dissolution with glacial acetic acid (1%). The amount of dissolved crystal violet was measured as absorbance at 550 nm. Each experiment was repeated independently at least 3 times.
To screen for suppressors of the paralyzed phenotype of the parental strain, aliquots of mutagenized cells were inoculated in LB medium overnight and then spotted on 0.3% LB swimming agar plus 2% arabinose. Swimming "plates" were cast in sterile reagent reservoirs (Costar #4870), allowing bacteria and serine chemoattractant to be spotted at opposite ends of the reservoir. After incubating swimming plates overnight at 30˚C, bacteria were sampled at 0.5 cm intervals from the origin by stabbing agar with a 200μl sterile pipet tip and resuspending the plug in 50 μl of LB. Each of these samples was plated onto LB agar, and the one furthest from the origin that still showed bacterial growth was used to inoculate swimming plates for a second round of selection. After this second swimming round, bacteria were again sampled at different distances from the origin and frozen as glycerol stocks at -80˚C. Pools were subsequently streaked for single colonies and approximately 200 single colonies were subjected to further analysis.
Intragenic flhF(R251G) suppressors were excluded by transforming each candidate with a plasmid overexpressing FlhF(R251G). Only candidates that were still able to swim were further evaluated. Twenty of the strongest suppressors were subjected to whole genome sequencing to map the sites of suppressor mutations, along with the parental strain PAK ΔflhF attB::flhF (R251G) and wild-type PAK.

Whole genome sequencing and sequence analysis
Whole genome sequencing was carried out as previously described [61]. Briefly, bacterial genomic DNA was prepared for sequencing on the Illumina MiSeq by the Yale Center for Genomic Analysis using the TruSeq DNA LT Sample Prep Kit (Illumina). Purified libraries were barcoded, pooled, and sequenced on the MiSeq using a 2 x 250 paired-end protocol. Initial basecalls were converted to fastq files using MiSeq CASAVA software suite, demultiplexed, and clipped for adaptors. Sequences were examined using FastQC (Galaxy) [62]. Reads were mapped to the P. aeruginosa PAO1 reference genome with the BWA (v. 0.6.2) software package using default parameters [63,64]. Bowtie (v. 2.1.0) was used to align reads [65], and SNPs were called using the Samtools mpileup [66].

Construction of chromosomal mutations
Mapped mutations were introduced into the chromosome of the parental strain, PAK ΔflhF attB::flhF(R251G), or of wild-type PAK (as indicated) by homologous recombination. Briefly, primers flanking the mutation of interest were synthesized to contain attB1 or attB2 sites compatible with the Gateway Clonase II (Invitrogen) system. The mutation plus a linked, "silent" loss or gain of a restriction site was engineered into overlapping internal primers, and these four primers were used to generate overlapping PCR products that were spliced together by overlap extension PCR (S2 Table). PCR products were cloned into the Gateway-adapted suicide vector pDONRX [36], screened by restriction mapping, and confirmed by DNA sequencing. Constructs containing only the desired mutation were transformed into E. coli S17-1 and mobilized into P. aeruginosa by mating.
Exconjugants were selected on VBM gentamicin and then streaked to VBM plus 10% sucrose to select for loss of vector backbone sequences through a second recombination event [67]. Sucrose-resistant, gentamicin-susceptible colonies were screened by amplifying the targeted gene and digesting the PCR product to look for the linked gain/loss of a restriction site.
Colonies that passed this screen were sequenced to confirm the presence of the desired mutation in the endogenous chromosomally encoded gene. This strategy was also used to introduce a BB2 epitope tag prior to the stop codon of the chromosomally encoded fimV and fimV(L7P) genes.
The PAK ΔflhF pilA strain was constructed by allelic recombination, using a previously described pEX18Gm pilA::aacC1 suicide vector in which the pilA gene is replaced by a gentamicin resistance cassette [68]. Briefly, this construct was mobilized into PAK ΔflhF by mating. Exconjugants were selected on VBM-gentamicin plus 10% (w/v) sucrose (to select for gene replacement and loss of the vector-carried sacB gene). The mutation was confirmed by PCR, Western blotting, and phenotype (loss of twitching motility).
Gene deletions of motA/motB, motC/motD, cheR1, cheZ, and fleN were introduced into the chromosome by homologous recombination. Briefly, primers flanking the deletion of the gene of interest were synthesized to contain attB1 or attB2 sites compatible with the Gateway Clonase II (Invitrogen) system. PCR products were cloned into the Gateway-adapted suicide vector pDONRX [36]. The procedures for generating ex-conjugants, screening candidates, and confirming the presence of the desired mutation(s) are described above.

Motility assays
Swimming assays were performed by spotting 5 μl of a fresh overnight culture of bacteria (grown at 37˚C with aeration in LB) diluted to OD 600 = 1 onto 0.3% LB agar plates supplemented with antibiotics or 20mM cAMP when appropriate. The diameter of the swimming zones was measured after overnight incubation (16 h) at 30˚C.
For swarming assays, M8 minimal medium plates (0.5% agar) supplemented with 0.4% glucose and 0.05% sodium glutamate were used. Swarming plates were inoculated with 5 μl of an overnight culture (diluted as for swimming assays), incubated at 30˚C overnight (16 h), and then incubated at room temperature for an additional 24 h. Pictures of the whole agar plates were taken for comparison with an Image Station 4000R (Kodak).
Twitching assays were performed by stabbing a colony through 1% LB agar to the agar/plastic interface with a sterile toothpick. The diameters of twitching zones were measured after 24 hours of incubation at 37˚C, at which time the twitching zone was visible on the plastic surface. All motility assays were performed at least three times using 4 or more replicates per experiment.

Bacterial two-hybrid assay
Genes of interest were subcloned into pBRGPω and pACTR-AP-Zif to generate carboxy terminal fusions, respectively, with the ω subunit of E. coli RNA polymerase or the Zn-finger DNA binding domain of murine Zif268 [69]. FlhF wild-type and mutant alleles were cloned into both bacterial two-hybrid vectors as previously described [23]. The cytoplasmic domain of FimV (FimV C or V C , amino acids 491-919) was PCR amplified from PAK genomic DNA and cloned the same way using the corresponding primers NdeI-FimV-forw and NotI-FimV-rev (S2 Table). A truncated version of this FimV domain lacking the C-terminal region (FimV ΔC or V CΔC , amino acids 491-769) was cloned using the reverse primer NotI-FimV2rev. Full-length Vfr was cloned using the primers NdeI-Vfr-forw and NotI-Vfr-rev. Flagellar genes of interest (fliF, fliG, fliM, fliN, motA, motB, motC, and motD) were PCR amplified from PAK genomic DNA and cloned the same way into both bacterial two-hybrid vectors using the corresponding primers NdeI-gene-forw and NotI-gene-rev. All constructs were confirmed by DNA sequencing.
Plasmids containing ω-and Zif fusions were electroporated simultaneously into E. coli KΔZif1ΔZ, which carries a lacZ reporter gene downstream of a promoter containing the Zif binding site [69]. Expression of the fusion proteins was induced by 50 μM IPTG. β-galactosidase activity was measured as described previously [70] and reported in Miller units. Samples were assayed in triplicate in three independent experiments.

Fluorescence microscopy
Flagella were stained as previously described [23]. Briefly, P. aeruginosa overnight cultures grown at 37˚C in LB were diluted 1:100 into fresh medium the next morning, and grown with aeration for 2h at 37˚C. Cells were harvested by centrifugation (3,000 g for 3 min), fixed in 4% paraformaldehyde for 20 min at room temperature and washed twice with phosphate-buffered saline (PBS). Flagella were stained with anti-FliC antibodies labeled with Alexa Fluor 488 or Alexa Fluor 594 (1 μg/ml in PBS) for 30 min at room temperature, then washed with PBS. Stained flagella were visualized using a Nikon Eclipse TS100 microscope (100x objective) equipped with a fluorescein isothiocyanate filter and a monochrome Spot camera (Diagnostic Instruments) running Spot 4.0.1 software. Flagella were directly stained in live, unfixed bacteria expressing FliC(T394C) with AlexaFluor 488 C5 maleimide as previously described [45,75].

FliC tethering assay
FliC binding assays were performed as previously described [23]. Briefly, P. aeruginosa was grown overnight at 37˚C in LB medium, diluted 1:50 into fresh medium the next morning, and incubated for 1h at 37˚C. Slides were incubated with PBS containing BSA (100 μg/ml) plus anti-FliC antibodies (2 μg/ml) for 30 min at RT. Antibody-coated slides were incubated with bacteria (and with 20 mM cAMP dissolved in 50 mM Tris-HCl pH 8.0, as indicated) for 5 min at RT prior to imaging.
For the analysis of the cell fate over time, movies were taken immediately using a Nikon Eclipse TS100 microscope (100x objective) equipped with a Canon Vixia HFS200 camera. To monitor the same cells over time, 30 sec movies were taken at 2-15 minute intervals. All cells rotating at t = 0 were tracked in each subsequent movie and scored for one of three fates: attached/rotating, attached/not rotating, detached.

Measurement of intracellular cAMP
Intracellular cAMP concentrations were measured using the Cayman cAMP EIA assay kit. P. aeruginosa cultures of indicated strains were incubated 15-16h with aeration at 37˚C in LB, diluted 1:100 into fresh medium the next morning, and grown with aeration to OD 600 0.4-0.6. Cells were harvested by centrifugation (20,000 xg for 2 min at 4˚C) and washed twice with 1 ml cold 0.9M NaCl. Pellets were suspended in 150 μl 0.1N HCl and incubated on ice for 10 min with occasional vortexing. The lysates were centrifuged at 20,000 xg for 5 min at 4˚C and supernatants were used to measure intracellular cAMP concentrations via the Cayman cAMP EIA assay kit, following the manufacturer's protocol for sample acetylation.

Swimming analysis
Bacteria were grown overnight at 37˚C in LB, diluted 1:100 into fresh medium the next morning, and incubated for 2h at 37˚C. Cells were spotted onto glass slides and observed by darkfield microscopy with a Zeiss Axiostar plus microscope (10x objective). Video clips were obtained using a Canon Vixia HFS200 camera. For each strain, the movement of at least 100 bacteria was analyzed for 1 min (1800 consecutive images with 30 frames per 1 sec) to obtain data for the swimming analysis.
To identify objects, the movies were background-subtracted by averaging over six-second windows and subtracting that average from each frame in that window. To analyze individual cell speed and angular velocity, each cell's trajectory was reconstructed from the movie image sequence using a custom MATLAB (Mathworks) code [71]. This tracking software uses particle detection via a radial symmetry method [72] and particle tracking adapted from u-track 2.1 [73]. The cell speed at each time point was calculated according to v i ¼ ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi and the cell angular velocity was calculated according to normalizing for the frame rate of the camera. To remove non-cell tracks, we filtered the tracks to have a mean speed between 15 and 80 μm/s (except for ΔmotAB, where a mean speed filter of 10-80 μm/s was used), a � 3 sec trajectory time, and a maximum mean squared displacement of � 400 μm 2 . Reversal analysis, swimming cells. To define reversals in free-swimming bacteria, a modification of the algorithm described by Theves et al. was written in MATLAB [74]. All changes in angular velocity ja i À a iÀ 1 j > p=2 were identified and considered potential reversals, and the maximum cell speed measured between all potential reversals was extracted. For a potential reversal x to be scored a true reversal, the following criteria all needed to be met: (i) the maximum cell speed measured during the interval between (x-1, x) was >20 μm/s (15 μm/s for ΔmotAB); (ii) the maximum cell speed measured during the interval between (x, x+1) was >20 μm/s (15 μm/s for ΔmotAB); (iii) the minimum speed within 0.1s of the potential reversal x was less than half of the maximum cell speed during the interval before or after the reversal; (iv) no other potential reversals were identified within 0.1s before or after x. These parameters filtered out "pausing" or "jiggling" events, where bacteria exhibit repeated large changes in angular velocity without moving [74]. Speed, angular velocity, and true reversals were plotted for each track as a function of time and manually examined to confirm expected behavior of our algorithm.
To calculate reversal frequency, the number of true reversals per track was divided by the track length. In cases of zero reversals, we calculated an upper limit for reversal frequency by assuming that one reversal would be observed if the tracking time was extended by 0.3s. The percentage of ΔmotAB tracks affected by this correction is 10.7-11.3%; for ΔcheR1, it is 39-46%.

Reversal frequency analysis, tethered cells
For the analysis of reversal frequency, bacteria and anti-FliC coated slides were prepared as described above. Tethered cells were recorded for 60 seconds at 100 frames per second with a digital scientific CMOS camera (Hamamatsu ORCA-Flash4.0 V2, 2x2 pixel binning, 1024 x 1024 array, 10 ms exposure) mounted on an inverted microscope (Nikon Eclipse TI-U) with a 100x oil immersion objective (Nikon CFI Plan Fluor, N.A. 1.30, W.D 0.2 mm) and LED white light diascopic illumination (Thorburn Illumination Systems).
Movies of tethered cells were analyzed frame by frame using a custom MATLAB (Mathworks) script, which fits an ellipse over the cell body, measures the length of the ellipse's major axis, and calculates the centroid (center of mass) of the ellipse. These values were then used to obtain the cell body's angle (in degrees) with respect to the center of the image region. This step was repeated for each subsequent frame of the movie. Once all angles were known for each movie frame, the angular velocity and rotation frequency was calculated by � � � frame rate and reversal frequency was obtained from the switching frequency from clockwise to counterclockwise rotation (and vice-versa).

Statistical analysis
Curve fitting and data analysis were carried out using Prism 5.0 (GraphPad) software. Data are expressed as means ± SD. For data sets in which two conditions were varied, p values were calculated by analysis of variance (two way ANOVA) followed by Bonferroni's posttest. Normally distributed data sets were compared by one-way ANOVA followed by Bonferroni's posttest, while data sets that were not normally distributed were analyzed by Kruskal-Wallis and Dunn's Multiple Comparison posttest. For statistical analysis of survival curves the Mantel-Cox test was performed. p values < 0.05 were considered significant.
Supporting information S1 Fig. Tethering assay protocol and scoring. Bacteria were incubated with an anti-FliC antibody-coated slide for 5 minutes prior to initial imaging. After 5 minutes (t = 0), bacteria were scored as either attached/spinning or attached/not spinning; in this example, we scored 66% spinning, 33% not spinning. All bacteria identified as spinning were then followed by obtaining 1 minute movies at the indicated time points. Every cell scored as spinning at t = 0 was categorized as still rotating ("R"), no longer rotating and horizontally attached to the surface We propose that FlhF interacts with an unknown protein ("X") to stop flagellar rotation when bacteria attach to a surface via their flagellum. In the case of a FlhF(R251G) mutation, the mutant FlhF(R251G) protein adopts a conformation that allows it to interact with protein X even when bacteria are not tethered at a surface. We also observe that ΔflhF bacteria fail to stop flagellar rotation when tethered at a surface and hypothesize that this results from an absence of the FlhF-Protein X interaction. We predict that we can identify suppressors that disrupt the FlhF-Protein X interaction by finding bacteria that can swim in liquid despite expression of FlhF(R251G). Our model also predicts that if these suppressor mutations are introduced into the wild-type background, they will phenocopy a ΔflhF mutant and fail to stop flagellar rotation after bacterial tethering to a surface. (TIF)