Tad Pili Play a Dynamic Role in Caulobacter crescentus Surface Colonization

Bacteria are able to colonize surfaces in environmental, industrial, and medical settings, where they form resilient communities called biofilms. In order to control bacterial surface colonization, microbiologists need to gain a detailed understanding of the processes that bacteria use to live at the liquid-surface interface and that allow them to adhere to and move on surfaces and eventually grow and persist on solid media. To facilitate these processes, bacteria are equipped with adhesive structures such as flagella and pili and with matrix components such as exopolysaccharides. How these cellular organelles are coordinated to optimize surface processes is currently subject to intense investigations. Here we used the model organism Caulobacter crescentus to demonstrate that polar pili are highly dynamic structures that are functionally interconnected with the flagellar motor to mediate surface sensing, thereby enforcing rapid and permanent surface attachment. These studies provide an entry point for an in-depth molecular analysis of bacterial surface colonization.

and specific host tissues and promote pathogen spreading and cellular invasion (4)(5)(6)(7). The highly corrugated surface of the extended pilus filaments mediates attachment to hydrophobic and hydrophilic surfaces through reversible, nonspecific interactions (8,9). Type IV pili represent the most sophisticated class of these filaments. The members of the best-studied subgroup, type IVa, are dynamic machineries that undergo cycles of extension and retraction through the rapid assembly and disassembly of pilin subunits at the proximal end of the structure (6,10). Extension and retraction are powered by specific cytoplasmic ATPases, which generate rotational movements of the assembly platform in the inner membrane to incorporate pilin subunits into or extract them from the helical filaments (4,11). Through the coordinated extension and retraction of multiple polar pili, single cells are able to move on surfaces and explore their environments (1,12,13). Two types of pilus-mediated movements have been described (14). Crawling movements of horizontally positioned cells, called twitching, can result in traversal of large distances with high directional persistence (15)(16)(17). In contrast, walking movements of orthogonal upright cells facilitate rapid exploration of smaller areas (18).
Type IV pili are widespread in bacteria and archaea (11,12). Distinctive features divide these structures into two classes (type IVa and type IVb). Type IVa represents a uniform class that is found in important human pathogens such as Pseudomonas aeruginosa, Vibrio cholerae, and Neisseria spp. and in environmental bacteria such as Myxococcus xanthus, Shewanella putrefaciens, and Bdellovibrio bacteriovorus. The type IVb subclass is less homogenous and is best characterized for enteropathogenic Escherichia coli or V. cholerae (19). A subclass of type IVb pili is represented by the tight adherence (Tad) or Flp (fimbrial low-molecular-weight protein) pili ( Fig. 1A) (20)(21)(22). Sometimes classified as a separate type IVc group, Tad pilin subunits are smaller than the pilins of other type IV systems but show similar hydrophobic intermolecular interactions providing the main force holding the fibers together (13,19,21). Tad pili promote surface colonization, cell-to-cell aggregation, and biofilm cohesion and promote virulence of different bacterial pathogens (13,19,21,(23)(24)(25). In contrast to type IVa and other type IVb pilus systems, Tad clusters seem to lack a dedicated retraction ATPase and most Tad pili do not seem to be dynamic (19). However, recent studies indicated that toxin-coregulated pili (TCP) of V. cholerae and Tad pili of C. crescentus are retractable (26,27). It was proposed for TCP that incorporation of minor pilin subunits (TcpB) into a growing pilus can block assembly and trigger a spontaneous disassembly and retraction event.
In C. crescentus, polar Tad pili facilitate the attachment of planktonic cells to surfaces (20,27,28). During C. crescentus division, a polarized sessile stalked (ST) cell produces a motile offspring, the swarmer (SW) cell, which is equipped with a single flagellar propeller and multiple pili (Fig. 1B) (20,29). The newborn SW cell remains in a motile, nonreplicating state for a defined period called G 1 . After this period, chromosome replication resumes coincident with cell differentiation, during which flagellum and pili are replaced by an adhesive exopolysaccharide (EPS), the holdfast, and the stalk. While the developmental program defines an extended time window of motility, SW cells that are challenged with surface are able to transit to the sessile state within seconds (28,30,31). This process is executed by a surface recognition program that involves sensing of mechanical cues and triggering of a burst of c-di-GMP, a second messenger that controls the motile-sessile transition in a wide range of bacteria (32)(33)(34). In turn, c-di-GMP allosterically activates a preassembled holdfast synthesis machinery to irreversibly anchor cells that encounter a surface (28,35).
Although the initial surface contact and adherence are indisputably mediated by pili, different views have been put forward for the role of polar Tad pili in C. crescentus surface sensing. In one model, dynamic action of pili positions the flagellar pole in close contact with the surface to allow mechanosensation by the membrane-integral rotary motor, triggering a spike of c-di-GMP through the activation of a motor-associated diguanylate cyclase (28). A second model proposed that surface-bound Tad pili themselves serve as surface sensors, mediating an internal upshift of c-di-GMP levels after experiencing resistance upon retracting (27). In line with the first model, recent studies demonstrated that flagellar motors, but not type IVa pili, are required for surface sensing in P. aeruginosa (36) and that a flagellar motor-coupled diguanylate cyclase increases levels of c-di-GMP in this organism (37). Moreover, studies in V. cholerae and P. aeruginosa had shown that c-di-GMP is positioned upstream of type IV pili and that this second messenger regulates pilus assembly and activity (36,(38)(39)(40)(41). In line with the second model, type IV pili were recently shown to be required for a surface-mediated increase of cAMP and virulence gene expression in P. aeruginosa (42).
To more closely scrutinize the role of Tad pili in C. crescentus surface recognition and surface colonization, we carefully analyzed their dynamic behavior and regulation. We used microfluidics to perform experiments under controlled flow conditions. We demonstrate that under conditions of steady medium flow, Tad-mediated cell attachment is transient, offering motile bacteria a short window of opportunity during which they can sense a surface and trigger holdfast biogenesis. Tad pili are already highly dynamic before the motile SW cell separates from its ST mother, explaining the observed ultrarapid surface recognition capability of newborn SW cells (28 (19,22). PilA is the major pilin subunit that matures upon removal of the signal peptide by the prepilin peptidase CpaA and is then assembled into a filament by the inner membrane platform that is located at the base of the filament (CpaG and CpaH). The CpaF ATPase is the functional motor protein, while the CpaE ATPase is required for polar localization of the pilus machinery (19,22). Minor pilin subunits (CpaJ and CpaK) and envelope-spanning components of the pilus machinery are indicated. OM, outer membrane; PG, peptidoglycan; IM, inner membrane. Role of Tad Pili in C. crescentus Surface Colonization ® can go through multiple rounds of extension and retraction, mediating walking-like motility on surfaces. Finally, we present data indicating that the flagellar motor and Tad pili functionally interact and that an increase in the c-di-GMP concentration results in retraction of pili. Together, these data led to a proposal of a model in which highly dynamic Tad pilus structures are integrated as part of a complex mechanism that senses mechanical stimuli upon C. crescentus surface encounter to promote and accelerate surface anchoring.

RESULTS
Tad pili mediate transient surface attachment under conditions of medium flow. Surface adhesion of C. crescentus via its polar pili is transient and weaker than the strong and long-lasting attachment via the adhesive holdfast (43). To investigate the overall contribution of pili to surface attachment without interference of the holdfast, we analyzed the behavior of a C. crescentus holdfast mutant (NA1000) (44) and scored attachment efficiency in simple microfluidic channels with a single inlet supplying a culture with a constant flow of medium. Microscopy time-lapse images were recorded to determine the rate of surface attachment. Importantly, mutants lacking pili were unable to adhere to the glass surface in such an assay (Fig. 1C).
The number of attached cells scaled with the flow velocity of the medium (Fig. 1C). We observed a plateau in the density of attached cells per unit surface area, arguing that the attachment was transient and that, under these conditions, attachment and detachment of bacteria reached an equilibrium. A plateau was generally reached 5 to 10 min after the start of the experiment. Colonization density at equilibrium showed a strong dependency on the flow velocity. The highest density of attached cells was observed with a fluid flow velocity in the microchannel of 0.75 mm/s (maximal velocity in the middle of the channel). This velocity was chosen as the standard for further experiments. At lower flow speed, we measured a lower plateau value of colonization density (Fig. 1C) due to a decreased rate of attachment of cells, while the detachment frequency was unchanged compared to the optimal flow velocity of 0.75 mm/s ( Fig. 1D and E). Similarly, with increasing flow velocities, colonization densities also decreased, with attachment being completely abolished above 5 mm/s/cell (Fig. 1C). The decreased plateau levels were primarily due to a 2-fold-to-10-fold decrease in attachment rates, while the detachment frequencies were 2-fold-to-3-fold lower, meaning that the attached cells were less likely to leave ( Fig. 1D and E). As a consequence, the average residence time of piliated SW cells on the glass surface was increased at higher flow (see Fig. S1A in the supplemental material). At a flow rate of 0.75 mm/s, 30% of the cells were retained on the surface for more than 2 min, whereas the fraction increased to 50% to 60% at higher flow rates.
Surface-bound cells subjected to fluid flow experience a drag proportional to the flow velocities (Fig. S1C). The observed increase in residence time at higher flow velocities may result from conformational changes in surface-attached pilus fibers that strengthen the interaction with the surface at higher drag forces (45,46). Together, these observations demonstrated that pili mediate transient surface attachment of C. crescentus SW cells under conditions of medium flow. Thus, while pilus-mediated surface adherence provides a short time window for tactile sensing and holdfast production, a strain lacking holdfast machinery is unable to attach permanently (28). In contrast, cells harboring intact holdfast machinery (hfsA ϩ ) rapidly attach to and permanently colonize glass surfaces (Fig. 1C).
Tad pili are dynamic and active before cell division. Surface-attached C. crescentus SW cells were most often found standing upright in an orientation perpendicular to the surface, irrespective of their ability to synthesize an adhesive holdfast (28). This is an unfavorable position, considering the constant drag force from the medium flow. Also, after landing, cells occasionally moved a few microns against the medium flow in an upright position. Since only an active force could move cells against or position them in an orientation perpendicular to the flow, we speculated that surface-bound Tad pili are able to retract under these conditions (18). To investigate pilus dynamics, we first analyzed strains that are capable of secreting an adhesive holdfast in flow channels mimicking conditions that C. crescentus encounters in its natural environment (31). Under such conditions, offspring of attached dividing mothers are exposed to surface before division as a consequence of medium flow over the crescentoid dividing cells (Fig. 2) (31).
To investigate if pili are already able to retract before cells divide, movements of surface-attached dividing cells were carefully analyzed. We observed that the piliated pole of late predivisional cells was pulled away from the ST pole, thereby stretching the typical crescentoid cell shape into a straight line ( Fig. 2A). To quantify this behavior, we determined the angle (␣) between the two cell bodies 5 min and 1 min prior to cell separation. C-shaped C. crescentus cells with pili showed discrete peaks of angle ␣ at 180°and at 150° (Fig. 2B). In comparison, a ΔcreS mutant that lacks the characteristic crescentoid cell curvature (47) showed only a single peak at 180°. This indicated that the peak at 150°represents cells with their natural, unstrained crescentin-mediated curvature in the late predivisional stage. A mutant lacking pili did not show a peak at 180°but retained peaks at 150°, arguing that cell stretching before division is mediated by the action of polar pili bound to a surface. Consistent with this, cells lacking the polar flagellum retained their stretching ability (Fig. 2B). Moreover, cell stretching was more prominent at very late stages of division, with the peak at 180°increasing at the expense of the peak at 150° (Fig. 2B). Finally, we observed a striking difference between the behavior of predivisional cells that generated offspring able to attach after separation (Fig. 2B, attach) and that of predivisional cells producing SW cells that were washed out after cell division (Fig. 2B, wash out). While the former showed a prominent peak at 180°, this peak was missing in dividing cells destined to produce offspring unable to attach. Instead, the latter showed a distribution of angle ␣ resembling the pilus-deficient strain. The correlation of pilus-mediated predivisional cell stretching and successful surface attachment of SW offspring suggested that pili play an active role in surface attachment. This is consistent with the observation that SW offspring, which were freely rotating before detachment from their mothers, generally failed to remain attached to the surface (28). Of note, the orientation of the concave side of attached predivisional cells showed a strong bias to one side with respect to the flow direction ( Fig. 2B; see also Fig. S2A to C). This suggested that the "C" shape of C. crescentus cells actually has a small helical twist (Fig. S2). While the concavity distribution is expected to be random for cells with a straight C shape, we found that the cells were more likely to orient to the left with respect to the flow direction, irrespective of their position within the channel or the nature of the surface.
Together, these experiments indicated that pili are assembled and active at the flagellate pole before cell division takes place and that pilus activity increases as dividing cells approach the separation stage. Moreover, dynamic pili are instrumental for surface attachment. We had proposed earlier that pilus retraction at this stage of cell division is critical to position the flagellar mechanosensor in close proximity to the surface in order to successfully initiate biogenesis of the adhesive holdfast, therefore preventing cells from being washed out (28).
Dynamic Tad pili pull attached swarmer cells into an upright position. We observed that within a few seconds after cell division, newborn SW cells were standing up against the medium flow. To quantify these movements, we measured the twodimensional (2D) projections of individual cells in the xz plane and used this information to infer the cells' 3D orientation and tilt angles () over time (Fig. 2C). Newborn SW cells were unable to change their position as long as they were physically connected to their ST mothers. However, upon separation, SW cells rapidly changed their tilt, moving into an upright position of about 60 to 75°degrees. In contrast, ST cells retained their low value after cell division ( Fig. 2D and E). Because mutants unable to assemble pilus fibers (ΔpilA mutants) showed very low levels of attachment, we analyzed the cell movements of a strain lacking both the major pilin subunit and the external parts of the flagellum (mutant ΔflgDE) as a control. We had shown earlier that A 1 min before div. 5 min before div. C. crescentus cells lacking the outer parts of the polar flagellum (ΔflgDE cells) show a hypersensitive surface response that partially alleviates the strict requirement for pili (28). In agreement with this, SW cells of the ΔflgDE ΔpilA mutant sporadically remained surface attached after separation from their mothers. Importantly, all cells invariably retained the same low angle value (Fig. 2D). In contrast, a strain lacking only the flagellar structures (strain ΔflgDE) showed wild-type (wt)-like standing-up behavior upon cell separation ( Fig. 2D and E). Deletion of the crescentin cytoskeleton (ΔcreS) did not influence the ability of attached cells to stand up, arguing that cell curvature does not influence dynamic cell movements after division. Taking the data together, this indicated that force-generating pili are required for newborn SW cells to immediately move into a vertical (upright) position. Despite the relatively strong medium flow, daughter cells were able to keep their upright position for about 10 to 15 min before angle gradually decreased. This coincides well with the timing of the differentiation of swarmer cells to stalked cells that occurs when pili disappear and stalk biogenesis is initiated (Fig. 1B). Tad pili mediate walking movements against the medium flow. Generally, pilus-mediated cell movements into an upright position or predivisional cell stretching were observed only once whereas repeated cycles of pilus extension and retraction were observed occasionally. Moreover, we occasionally observed cells moving for very short (1-to-3-m) distances in holdfast-deficient cells after landing on a surface. We reasoned that the prevailing activity of pili might be masked in attached SW cells by the rapid synthesis of the holdfast adhesin, which immobilizes cells in an upright position (28).
To address if C. crescentus pili are capable of undergoing multiple cycles of extension and retraction, we made use of a hfsK mutant (35). HfsK is a c-di-GMP effector protein, the activity of which increases holdfast cohesion. Mutants lacking this protein secrete holdfast material that is strong enough to glue cells to the surface in microfluidic devices with medium flow but that fails to firmly anchor cells at the place of initial attachment (35). To tune the amount of holdfast generated, a ΔhfsK strain was engineered that allowed modulating intracellular levels of c-di-GMP (rcdG 0 ::P lac -dgcZ), the primary allosteric regulator of the holdfast secretion machinery, via the controlled expression of the exogenous diguanylate cyclase gene dgcZ (28). In flow chambers, cells of this strain were dragged across the surface by the medium flow, leaving trails of stained holdfast material behind (35). Intriguingly, we observed that newborn SW cells, after standing up, were able to move against the medium flow ( Fig. 3A and B). These results indicated that pili can retain their dynamic behavior to dislodge cells that are weakly attached to the surface. Tracking the trajectories of individual cells identified the time period, step size, and speed of pilus-driven movements (   movements, SW cells, although unable to remain standing for longer periods of time, repeatedly moved back into an upright position (Fig. 3D). This behavior was particularly pronounced at the conclusion of each step event, arguing that full retraction of pili forces cells into an upright position ( Fig. 3D; see also Fig. S3B). Together, these results indicated that C. crescentus pili remain highly active over longer periods of time and that, under these conditions, they engage in repetitive cycles of extension and retraction. This is highly reminiscent of the twitching or walking movements observed for other bacteria possessing type IV pili (18,48).
Determining the force and energy of pilus retraction. To measure the forces generated by pilus retraction, we made use of optical tweezers. Because C. crescentus cells are highly susceptible to the phototoxicity elicited by the strong light source used in optical traps, it was not practical to directly manipulate cells with this method. To avoid this problem, we exploited the ability of C. crescentus predivisional cells to permanently adhere to surfaces via their adhesive holdfast. By coupling predivisional cells to polystyrene beads and immobilizing the beads in the optical trap, phototoxicity was eliminated, as cells could now be kept in the optical trap for at least 1 h without losing their ability to grow and divide (Fig. 4A). By moving the beads carrying predivisional cells toward the glass surface, pili were able to attach and, upon retraction, displace the beads from the trap (Fig. 4A).
To avoid potential interference from the flagellum, these measurements were carried out with a strain lacking the external parts of the rotary motor (strain ΔflgDE). A representative example of an optical trap measurement is shown in Fig. 4B. Initially, the bead rests in the center of the optical trap, with some noise due to Brownian motion. Upon pilus attachment to the surface, Brownian motion decreases followed by bead displacement of up to 1 m from the center before a rapid movement back into its original position (Fig. 4B). Bead displacement (d) and trap stiffness (⌲ trap ) values allowed calculation of the maximum retraction force (F ϭ ⌲ trap ·d). We tested different ⌲ trap values and found a maximum retraction force applied by pili during retraction of Step  Role of Tad Pili in C. crescentus Surface Colonization ® approximately 8 pN (Fig. 4C). These values are slightly lower than those reported earlier, in which pilus forces were assessed by measuring the displacement of elastic micropillars (27). These values are similar to the drag force of around 10 pN that surfacebound cells experience in a microchannel with a medium flow speed of 0.75 mm/s (Fig. S1B). Other bacteria such as P. aeruginosa or N. gonorrhoeae with type IVa pili display higher pilus retraction forces (30 to 100 pN) (4,5,49), possibly reflecting lower drag forces experienced by C. crescentus in natural settings and differential functionalities of type VIc pili related to rapid holdfast-mediated attachment.
The mean speed of C. crescentus Tad pilus retraction was 100 nm/s (Fig. 4D). In contrast, the speed at which beads backtracked into the trap center was considerably higher (Fig. S4E and F), arguing that these events resulted from pilus detachment rather than reelongation of the filaments. From this, we concluded that elongation events are unlikely to interrupt retraction phases, suggesting that pilus retraction is processive, showing disassembly without interruption until all pilin subunits are internalized.
C-di-GMP regulates Tad pilus activity. To better understand the role of pili during surface attachment and their functional interaction with the flagellum, we scored the colonization efficiency of nonmotile flagellar mutants in microfluidic channels, using the same setup as described above (Fig. 2). Both the ΔmotB and the ΔflgDE mutants showed very low levels of colonization densities, suggesting that active swimming is important for cells to efficiently reach the surface in this experimental setup (Fig. 5A). Interestingly, nonmotile strains also showed higher detachment frequencies and shorter residence times than the wild-type strain ( Fig. S4A and B), behaviors that were most pronounced for the ΔmotB strain. Since the MotA and MotB stator units of the flagellar motor are involved in surface sensing (28), this indicated that the flagellar motor and pili may be connected through a regulatory-feedback mechanism.
It was recently shown that C. crescentus surface sensing leads to a rapid increase in the level of the second messenger c-di-GMP, which in turn mediates biogenesis of the adhesive holdfast (27,28). One of the enzymes implicated in this process is the diguanylate cyclase DgcB (28). However, it has remained unclear if Tad pili are positioned upstream of c-di-GMP and if they contribute to the increase in the level of the second messenger during mechanotransduction or if Tad pili are a target of c-di-GMP control during this process. To clarify this, we monitored the behavior of pili in response to changing levels of c-di-GMP. Intriguingly, a ΔdgcB mutant showed a significant delay in pilus-mediated surface colonization (Fig. 5A) and a minor defect in standing up against the medium flow (Fig. 5B) compared to wild type. The fact that pilus-mediated behavior was affected only partially in the ΔdgcB mutant in these experiments indicated that an additional diguanylate cyclase(s) may be involved in this process (28). This is in line with the observation that newborn SW cells of a ⌬motB mutant that were unable to sense surface showed a considerably stronger defect in reorienting into an upright position against the medium flow (Fig. 5B), while a ΔflgDE mutant (without a surface-sensing defect) was not affected (Fig. 2E). In agreement with the view that Tad pili are able to respond to changes in c-di-GMP, we found that the level of pilusmediated attachment was strongly reduced in a mutant lacking several CheY-like Cle proteins (Fig. 5A). Cle proteins bind c-di-GMP and then can interact with the flagellum and are an integral part of C. crescentus surface recognition (50).
To analyze the role of global c-di-GMP levels in pilus dynamics, we tested a strain harboring an inducible copy of the dgcZ diguanylate cyclase gene (rcdG 0 ::P lac -dgcZ). While low-level induction of dgcZ increased pilus-mediated surface attachment, surface colonization gradually decreased upon stronger induction of dgcZ (Fig. 5C). This effect was the consequence of decreased attachment frequencies (Fig. S4A). Low-level induction of dgcZ also decreased detachment rates and increased the cells' residence time on surface; however, these effects were reversed at higher levels of induction of dgcZ. Furthermore, increased induction of dgcZ also strongly interfered with the ability of newborn SW cells to stand up ( Fig. 5D; see also Fig. S4C). These observations are in line with an earlier report demonstrating that a low level of c-di-GMP is strictly required for Sangermani et al. C. crescentus Tad pilus expression and assembly (32). Transmission electron microscopy (TEM) revealed that the number of pili per SW cell increased upon surface exposure and that piliation was indeed dependent on the intracellular c-di-GMP concentration. While low-level induction of dgcZ increased piliation, the number of pili per cell decreased at higher dgcZ expression levels (Fig. S4D). Together, these results indicated that both the flagellar motor and c-di-GMP control Tad pilus-mediated behavior, arguing that c-di-GMP is positioned upstream of pili in the mechanotransduction pathway and is able to regulate pilus dynamics. High concentrations of c-di-GMP promote Tad pilus retraction. The results described above suggested that high levels of c-di-GMP promote Tad pilus retraction. To directly visualize Tad pilus dynamics, we used a pilA T36C gene to fluorescently label pilus subunits with maleimide-based fluorescent dyes. This technique was recently used to visualize Tad pilus retraction in real time (27). Although we were able to visualize sporadic events of pilus retraction, pilus filaments are fragile and are prone to breaking during the labeling process. Because only a very few cells showed dynamic pili after labeling, it was difficult to obtain statistically solid data sets to compare the behaviors of different mutant strains. To overcome this problem, we adopted a more static fluorescence assay, with pilA T36C pilus labeled externally by fluorescence maleimide dyes. Because the fluorescent dyes are unable to penetrate the cell envelope, PilA subunits are labeled only if they are assembled into a pilus filament, thereby crossing the outer membrane (27) (Fig. 5E). Upon pilus retraction, disassembled, now fluorescent PilA subunits diffuse back into the cytoplasmic membrane, thereby increasing the fluorescence signal in this compartment (Fig. 5E) (27). Thus, to enable quantification of pilus retraction, we determined the fluorescence intensities of cell membranes rather than directly monitoring the fragile exterior pilus filaments. Importantly, cells expressing wild-type pilA or cells expressing pilA T36C but lacking the motor subunit of the pilus machinery (ΔcpaE) did not accumulate fluorescence. In contrast, cells expressing pilA T36C showed a weak but robust fluorescence signal ( Fig. 5E and F; see also Fig. S4E). Similarly, a strain harboring dgcZ (strain rcdG 0 ::P lac -dgcZ) showed weak fluorescence when expression of the exogenous diguanylate cyclase was not induced (Fig. 5F; see also Fig. S4E). To monitor Tad pilus dynamics at increasing c-di-GMP concentrations, dgcZ expression was induced for increasing amounts of time before external pili were labeled by adding the dye. When dgcZ expression was induced for 5 min, the average fluorescence intensity of cells slightly increased compared to uninduced cells. Longer induction of dgcZ for 10 or 20 min gradually increased the amount of fluorescence accumulating in the cytoplasmic membrane in a high proportion of cells ( Fig. 5F; see also Fig. S4E). Thus, as c-di-GMP levels built up, Tad pilus activity increased. However, when dgcZ was induced for more than 20 min before pili were labeled externally, fluorescence in the cytoplasmic membrane decreased, indicating reduced pilus dynamics under these conditions. Importantly, dgcZ induction did not affect the overall concentration of pilin subunits (Fig. S4F). These results support a model where moderate c-di-GMP levels increase pilus dynamics whereas higher c-di-GMP levels lead to the retraction of Tad pili (Fig. 5G).

DISCUSSION
Bacteria have evolved complex mechanisms that enable them to effectively colonize surfaces through a combination of tactile sensing and the exposure of surface adhesins. We showed earlier that C. crescentus SW cells are able to sense surface encounter with their polar flagellar motor and, in response, to deploy an adhesive holdfast to remain irreversibly anchored on the surface. On the basis of the results presented here, we propose that the spatially associated flagellum and Tad pili synergistically optimize the C. crescentus surface response. The succession of events leading from temporary to long-term attachment is summarized in the model in Fig. 5G. SW cells swimming in close proximity to a surface are able to attach via preexisting Tad pili (Fig. 5G, step 1), thereby creating opportunities for initial surface sensing by the flagellar motor (Fig. 5G,  step 2). Activation of the flagellar-motor-coupled diguanylate cyclase DgcB (and possibly other diguanylate cyclases) then generates an increase in the level of c-di-GMP, which boosts the activity of Tad pili. The assembly of additional polar pili increases the cell's probability of remaining attached to and of forming tight connections with the surface (Fig. 5G, step 3). As c-di-GMP levels increase, processive pilus retraction nudges the flagellate pole into close proximity to and contact with the surface. In particular, reorientation of cells into an upright position optimally positions the flagellar motor, strengthening its tactile sensing and further increasing c-di-GMP to peak levels required for the allosteric activation of holdfast biogenesis (Fig. 5G, step 4) (28). Our model proposes that the role of pili goes beyond that of passive adhesins promoting temporary attachment and that, instead, Tad pili and the flagellar motor together promote surface adherence in a highly dynamic and coordinated fashion. The model predicts that Tad pili actively guide the efficient surface sensing by the flagellar motor and thus contribute to the critical upshift of c-di-GMP levels mediated by DgcB and by other, as-yet-unknown diguanylate cyclases. This model is compatible with the notion that Tad pili themselves may contribute to surface sensing during this process (27). Evidence for our model comes from direct observation of Tad pili, which increase in numbers when cells are surface exposed or when they experience a moderate increase of c-di-GMP levels but a decrease in numbers when c-di-GMP reaches peak levels. Moreover, incorporation of fluorescently labeled pilin subunits indicated high Tad pilus activity at intermediate levels of c-di-GMP but reduced pilus activity when levels of c-di-GMP further increased. Finally, pilus-mediated surface attachment and vertical positioning of cells were abrogated when c-di-GMP levels were artificially increased to higher levels. On the basis of these observations, we speculate that at moderate levels of c-di-GMP, the overall number of polar pili per cell increases through a boost in assembly or through slowed disassembly or a combination thereof. In contrast, peak levels of c-di-GMP, which are reached upon sustained surface sensing or during development of the motile SW cell into a sessile ST cell (32,51), signal pilus retraction and PilA subunits are internalized. Thus, the functional interdependence of the motor, c-di-GMP, and pilus retraction generates a positive-feedback loop that imposes a directional process that progressively drives cells toward holdfast-mediated permanent surface attachment (Fig. 5G). We had shown earlier that pili are not required per se for the motor-mediated surface response and that bacteria grown in very narrow microfluidic chambers where they constantly encounter surface are able to rapidly attach via their holdfast even if they lack pili (28). Under normal conditions, however, bacteria swimming close to a substratum experience surface contact only transiently, a situation that may fashion the need for a highly dynamic and efficient process to stabilize and reinforce this interaction on short time scales. Pili colocalize with the flagellum at one cell pole and are thus optimally positioned to direct the motor toward the surface and to incite collisions that optimize the strength and duration of mechanosensing. Tad pili are present (20) and active (Fig. 2) in the predivisional cell immediately before cell division takes place at a time when the flagellum becomes fully operational (28). In line with the idea that pili are available to reinforce the motor-mediated surface program, their activity in the predivisional cell strictly correlates with the ability of SW offspring to permanently attach next to their mothers in strong flow (Fig. 2). Likewise, SW cells that freely rotate around their long axis during the process of separating from their stalked mothers, presumably because they fail to be dragged toward the surface by Tad pili, rarely manage to remain attached to the surface. In contrast, cells that stopped their rotation generally remained attached after budding off their mothers (28). These observations directly link pilus activity with flagellar obstruction and surface sensing.
The notion that c-di-GMP, depending on its concentration, influences pilus activity in distinct ways is supported by observations indicating that c-di-GMP promotes type IV pilus assembly in several bacterial species, including C. crescentus (32), V. cholerae (40,41), and P. aeruginosa (36,39). Moreover, increased and decreased levels of c-di-GMP were shown to impact type IV pilus-based motility in Myxococcus xanthus, arguing that the second messenger is required for but interferes with pilus function at increased concentrations (17). Similarly, polar Tad pili are retracted during the C. crescentus SW-to-ST transition coincident with c-di-GMP reaching peak levels in ST cells (5,32,51). Mechanistically, this complex regulation could result from the antagonistic activities of two effector proteins that bind c-di-GMP with different affinities to promote the assembly and disassembly of pilus filaments, respectively. Similar mechanisms were described in P. aeruginosa and X. campestris, where two c-di-GMP binding proteins, FimX and FimW, localize to the cell poles to modulate pilus formation. While the mode of action of FimW is still unknown, FimX facilitates pilus elongation during twitching at the leading cell pole by interacting with the assembly motor ATPase PilB (38,39). Direct interaction of c-di-GMP with the pilus machinery was also shown in V. cholerae, where c-di-GMP binding by the motor assembly protein MshE promotes polymerization of pili in a dose-dependent manner (40,41). Exactly how c-di-GMP influences both pilus assembly and retraction in C. crescentus remains to be shown.
Our model for C. crescentus surface attachment proposes tight functional cooperation of the flagellar motor and Tad pili. Evidence for this stems from the observation that pilus activity was reduced in mutants lacking the MotB stator unit or lacking all five Cle proteins (CleA to CleE) ( Fig. 5A and B). C. crescentus Cle (CheY-like) proteins were recently shown to bind c-di-GMP and, in response, to interact with the flagellar motor to impede flagellar activity and promote surface adaption (50). It was proposed that one or several of these proteins are part of a positive-feedback loop that reinforces the motor response during surface sensing, thereby facilitating a rapid upshift of c-di-GMP levels and production of holdfast material, which takes over the requirement for pilus-mediated attachment. The reduced pilus-mediated attachment observed for the strain lacking cleA to cleE (ΔcleA-E) could thus be due to an insufficient raise in c-di-GMP or, alternatively, may implicate one of the Cle proteins in the regulation of pili directly, providing a mechanistic basis for the coordination of both organelles. Further studies are needed to clarify the role of Cle components in this process.
Experiments in flow devices demonstrated that C. crescentus Tad pili are highly dynamic and are able to promote twitching-or walking-like movements of C. crescentus SW cells (Fig. 3). For these experiments, we used a mutant lacking the HfsK N-acetyltransferase, an enzyme that was proposed to chemically modify holdfast material. A ΔhfsK mutant forms malleable holdfast structures that can facilitate the clinging of cells to surfaces but lack the adhesive strength required to hold cells in place under flow conditions or when pulled by other forces such as pilus retraction. It was recently proposed that HfsK acylates the EPS component of the holdfast and that this modification is necessary for proper holdfast cohesion and anchoring (35). The observation that HfsK activity is itself regulated by c-di-GMP indicated that holdfast material may be formed under certain conditions but remains in a nonacylated form. If so, limited cohesive strength of the holdfast could attach C. crescentus cells to surfaces without restricting their pilus-mediated motility. This would allow this aquatic organism to explore liquid exposed surfaces with their pilus motors in a manner similar to the well-known twitching and walking motility of soil bacteria.

MATERIALS AND METHODS
Bacterial strains and growth conditions. The wild-type (wt) strain of reference was either NA1000, a laboratory-adapted strain with a point mutation in gene hfsA, which makes it unable to secrete the holdfast, or NA1000 hfsA ϩ , where the functional gene was reintroduced. The strain rcdG 0 is a c-di-GMPfree strain where all major endogenous GGDEF and EAL domain-encoding genes were deleted. Introduction in rcdG 0 of the exogenous diguanylate cyclase dgcZ gene under the control of the IPTG (isopropyl-␤-D-thiogalactopyranoside)-inducible lac promoter (P lac -dgcZ) allows tuning of the expression of dgcZ and, in turn, control of the intracellular levels of c-di-GMP, as previously reported (32). A full list of the strains used in this study is provided in Table S1 in the supplemental material.
Fabrication of microfluidic devices. Masters were fabricated via standard photolithography protocols (52). PDMS (polymethyldisiloxane [Sylgards 184]; Dow Corning) devices were created via replica molding and were then aged via heat treatment on a hotplate at 150°C for 30 min (53). Holes were drilled at the inlet(s) and the outlet(s), and the devices were treated in oxygen plasma and covalently bound onto round, 50-mm diameter borosilicate glass cover slides (VWR) (thickness no. 1). All microflow experiments were conducted in single microchannels with dimensions of either 100 m in width and 25 m in height or of 200 m in width and 50 m in height.
Microfluidics and microscopy setup. For attachment assays, overnight cultures (generally NA1000) were diluted 1:50 and grown at 30°C in peptone-yeast extract (PYE) medium with agitation. For the rcdG 0 ::P lac -dgcZ strain, we added the desired IPTG concentration during the dilution step. Once the culture reached an optical density at 600 nm (OD 660 ) of 0.15, it was loaded in a 1-ml plastic syringe (Soft-ject; Henke-Sass, Wolf). The syringe was then plugged into a needle (BraunMelsungen AG) (23 gauge, 0.6 by 30 mm) which was connected in turn to one end of a polytetrafluoroethylene (PTFE) microtube (Fisher Scientific) (0.56 by 1.07 mm). The tubing was filled with the culture, and the terminal end of the tubing was connected to the inlet of a microfluidic device (channel cross section, 50-m width by 200-m height). The syringe was mounted on a syringe pump (neMESYS low-pressure module V2; Cetoni GMBH) (14:1 gear). With the microfluidic setup ready, the device was placed on the stage of an inverted microscope (IX81; Olympus GMBH). The syringe pump was initially set to create a strong flow of 25 mm/s for 1 min to ensure that the microchannel surface was devoid of cells. The flow was then set to the desired target velocity at room temperature for the duration of the experiment (30 to 45 min). Time-lapse images were recorded at 0.1 to 0.16 frames/s (fps), using a 40ϫ oil immersion objective (UPlanFLN 40ϫ oil; Olympus). We used a new batch of culture at an OD 660 of 0.15 for each experiment.
For observations of single-cell division events, a culture of the desired strain (generally hfsA ϩ ) was loaded in the microfluidic channel of a device (channel cross section, 25-m width by 100-m height). Cells were left to colonize the surface for a short period, resulting in an average distance between cells of about 20 to 30 m. This low density ensured that single-cell division events were not disturbed or influenced by neighboring cells. A 1-ml plastic syringe loaded with fresh PYE medium (plus the desired IPTG concentration for strain rcdG 0 ::P lac -dgcZ) was plugged into a needle and connected to one end of a PTFE microtube. The syringe was then mounted on a syringe pump and the terminal end of the tubing plugged into the device's inlet. A steady flow of 1 mm/s was then set for the duration of the entire experiment. Cells were left to grow for 1 to 2 h before recording started with image sequences taken at 1 or 5 fps, using a 100ϫ oil immersion objective (CFI Plan Apo DM100ϫ oil; Nikon). The experiments were conducted at room temperature for no more than 10 h. This procedure ensured steady growth conditions and no overgrowth/clogging in the inlet.
Optical tweezer setup and force measurements. The optical tweezer experiments were performed on a custom-built bright-field microscope, complemented with a laser diode setup (LD830-MA1W; Thorlabs) ( ϭ 830 nm). A water immersion, high-aperture objective (UPlanSApo 60ϫ water; Olympus) was used to focus the laser beam, trap the beads with attached bacteria, and image the fluctuations of the bead in the trap and the attached bacteria. The experiments were carried out in position clamp mode at a constant laser power setting. The images of the beads and the attached cells were recorded at 50 Hz and 75 Hz using a fast camera (Phantom Miro EX4; Vision Research Inc.). We measured the change in position of the bead during the experiment, and we observed the active cell.
Calibration of the optical tweezer was carried out via fluctuation calibration. For each laser power setting used, an image sequence of a bead in the trap was recorded at 1,000 Hz. The variance () was determined from the fluctuation of the bead in the trap. The trap stiffness (⌲ trap ) was calculated as follows: ⌲ trap ϭ k B T/ 2 , with k B representing the Boltzmann constant and T the room temperature.
An exponential culture of strain NA1000 hfsA ϩ ΔflgDE maintained at an OD 660 of 0.15 was mixed with polystyrene beads (Fluoresbrite YG Carboxylate Microspheres; Polysciences) (3.0-m diameter) to reach a final concentration of 1.7·ϫ 10 8 beads per milliliter of cell suspension. The mix was incubated for 2 min to allow cells to attach to the beads. Then, a 1:1 dilution with fresh PYE medium was injected into the device and experiments were conducted under no-flow conditions. The devices used for optical tweezer measurements had chambers attached to a main channel (52,54) connected by an opening of less than 10 m. Single bead-carrying predivisional cells were chosen and placed inside the chambers where the optical tweezer measurements could be performed undisturbed from other cells.
Pilus staining and fluorescence. The labeling of pili was done following a protocol published recently by Ellison et al. (27). Briefly, an overnight culture was diluted 1:50 or 1:100 and grown to an OD 660 of 0.10, with agitation. A 1-ml volume of the culture was transferred into an Eppendorf vial, 25 g/ml of maleimide-reactive dye AF647-mal (Sigma-Aldrich) was added, and the reaction mixture was gently mixed by inversion. The sample was incubated for exactly 5 min and was then centrifuged (4,000 ϫ g for 1 min), washed with 1 ml fresh PYE medium, and centrifuged again. The final pellet was resuspended in 50 l fresh PYE medium. Finally, a 2-l volume was spotted onto a 1% agarose pad and imaging at the microscope was conducted immediately. For strain rcdG 0 ::P lac -dgcZ, we added 100 M IPTG for an exact time window before the washing step. The total duration of IPTG induction included the 5-min maleimide-labeling step. For induction times longer than 5 min, the culture was kept in incubation until the labeling step.
The elapsed time between the washing step and the beginning of imaging was minimized, on average taking 4 to 5 min. Each sample was imaged for no longer than 7 to 8 min after the labeling step. Imaging was done at an inverted microscope (Eclipse Ti2; Nikon Instruments Europe B.V.) using a 100ϫ oil immersion objective (CFI Plan Apo DM100ϫ oil; Nikon). We imaged 100 to 150 random positions for each pad, and we acquired a phase-contrast image (exposure, 50 ms) and a fluorescent image (excitation [Ex], laser wavelength of 640 nm at 25% intensity for 150 ms; emission [Em], mCherry 592 nm to 667 nm) for each position.
To analyze the fluorescent signal inside cells' bodies, we used the MATLAB-based program micro-beTracker for cell detection (55) and an in-house-built MATLAB-based program to analyze the fluorescent channel. We selected SW cells by using as the criterion a body length Ͻ2.5 m.
3D-SIM superresolution microscopy. Three-dimensional structured illumination microscopy (3D-SIM) was performed on a DeltaVision OMX-Blaze V4 system (GE Healthcare). Images were acquired using a Plan Apo N 60ϫ, 1.42-numerical-aperture (NA) oil immersion lens objective (Olympus) and 4 liquidcooled scientific complementary metal-oxide-semiconductor (sCMOS) cameras (pco.edge 5.5; PCO) (full frame, 2,560 by 2,160 pixels). Exiting light was directed through a movable optical grating to generate a fine-striped interference pattern on the sample plane. The pattern was shifted laterally through five phases and three angular rotations of 60°for each z-section. The 488-nm-wavelength laser lines were used during acquisition, and the optical z-sections were separated by 0.125 m. Laser power was attenuated to 10%, and exposure times were typically 80 ms. The laser power was adjusted to achieve optimal intensities at between 2,000 and 3,000 counts in a raw image of 15-bit dynamic range at the lowest laser power possible to minimize photobleaching. Raw 3D-SIM images were processed and reconstructed using the DeltaVision OMX SoftWoRx software package (v6.1.3; GE Healthcare).
Image analysis. To quantify the attachment efficiency of C. crescentus SW cells in microfluidic devices, we used the MATLAB-based program microbeTracker for identification of surface-attached cells (55). The number of cells was assessed for every recorded frame of each experiment and normalized to the imaged area to obtain the cell density over time. To evaluate cell movements and orientations, we determined the cell outline, as detected by the MATLAB-based program microbeTracker (55). Further tracking, analysis, and statistics determinations were carried out with in-house-developed MATLAB scripts. Cell perimeters were fitted to an ellipse, and the eccentricities of the ellipses (), representing the ratio between the major and minor axes, were monitored. We used to calculate the inclination of cells with respect to the surface (angle ). A value of 1 was set to correspond to a tilt angle () of 90°and a value of 0.1 to correspond to a tilt of 0°. Eccentricity values below 0.1 could be excluded as they represent highly elongated ellipses, with a length-to-width ratio that does not occur for C. crescentus SW cells.
To track the trajectories of cells, we choose to follow the position of the holdfast. This position lies at a point between a vertex and a focus of the ellipse which fits the 2D projection of the cell. The position was dependent on the value of , such that when the cell is upright, it coincides with the focus, and when the cell lies flat, it is close to the vertex (see Fig. 2C).
Step events (see Fig. 3B) were defined as representing fast movement against the flow, with speeds of Ն0.1 m/s and movement sustained for Ն1 s.
Electron microscopy. To examine surface induced cells, 5-l volumes of sample were spotted on a 400-mesh copper grid covered with a Parlodion and carbon film for 5 to 20 min. The grid was then gently washed using a micropipette, and cells were fixed with 0.1% glutaraldehyde. Alternatively, to examine planktonic samples, glutaraldehyde was added to a bacterial liquid culture to reach a final concentration of 0.1%. Then, 5-l volumes were spotted on the grid. The samples were washed with water 4 times and subjected twice to negative staining with 0.5% uranyl acetate. Images taken at the TEM (Morgagni 268D FEI [80 kV] and FEI T12 [120 kV] TVIPS F416) were visually inspected and SW cells identified by the absence of a stalk and a cell length of Ͻ2.5 m.
Immunoblot analysis. Proteins were separated by electrophoresis on 20% SDS-polyacrylamide gels and transferred onto a 0.2-m-pore-size nitrocellulose blotting membrane (Amersham Protran 0.2-m NC; GE Healthcare). A PageRuler prestained protein ladder (Thermo Fisher) was used to mark protein sizes. The primary antibody for detection was polyclonal rabbit anti-PilA antibody (56) (diluted 1:4,000). The secondary antibody was swine anti-rabbit antibody coupled to horseradish peroxidase (Dako), used at a dilution of 1:10,000. Antibody-treated blots were incubated with LumiGLO (KPL) for exposure of super RX-N films (Fujifilm).