Simultaneous tracking of Pseudomonas aeruginosa motility in liquid and at the solid-liquid interface reveals differential roles for the flagellar stators

Bacteria sense chemicals, surfaces and other cells and move toward some and away from others. Studying how single bacterial cells in a population move requires sophisticated tracking and imaging techniques. We have established quantitative methodology for label-free imaging and tracking of individual bacterial cells simultaneously within the bulk liquid and at solid-liquid interfaces by utilizing the imaging modes of digital holographic microscopy (DHM) in 3D, differential interference contrast (DIC) and total internal reflectance microscopy (TIRM) in 2D combined with analysis protocols employing bespoke software. To exemplify and validate this methodology, we investigated the swimming behavior of Pseudomonas aeruginosa wild type and isogenic flagellar stator mutants (motAB and motCD) respectively within the bulk liquid and at the surface at the single cell and population levels. Multiple motile behaviours were observed that could be differentiated by speed and directionality. Both stator mutants swam slower and were unable to adjust to the near surface environment as effectively as the wildtype highlighting differential roles for the stators in adapting to near surface environments. A significant reduction in run speed was observed for the P. aeruginosa mot mutants, which decreased further on entering the near-surface environment. These results are consistent with the mot stators playing key roles in responding to the near-surface environment. Importance We have established a methodology to enable the movement of individual bacterial cells to be followed within a 3D space without requiring any labelling. Such an approach is important to observe and understand how bacteria interact with surfaces and form biofilm. We investigated the swimming behavior of Pseudomonas aeruginosa, which has two flagellar stators that drive its swimming motion. Mutants that only had either one of the two stators swam slower and were unable to adjust to the near surface environment as effectively as the wildtype. These results are consistent with the mot stators playing key roles in responding to the near-surface environment, and could be used by bacteria to sense when it is near a surface.

Introduction 51 52 Flagella and type IV pili (TFP) mediated motility enable bacteria to migrate towards nutrients or away from 53 toxic substances (5, 49) and play key roles in bacterial biofilm formation and host-pathogen interactions 54 (25,46). In addition, these bacterial appendages also play a role in surface sensing (30). The mechanisms 55 by which individual cell behaviors drive social phenomena such as swarming, twitching and biofilm 56 development have been the subject of intense investigations (1, 13-14, 16, 23-25, 29, 31, 34, 40, 47). The 57 ability to collect data simultaneously on individual bacterial cells in a population that includes both 58 only been demonstrated for the analysis of single bacterial cells released near a surface using optical 108 tweezers (8). 109   110 To gain novel insights into the differences in cell behavior within the bulk and at the surface, it is necessary 111 to image both environments concurrently. To achieve this, we developed a novel multimode 2/3D 112 microscope with interlaced acquisition of 2D surface differential interference contrast (DIC) or total 113 internal reflectance microscopy (TIRM) images with 3D DHM. The spatial and temporal resolution of the 114 DIC images at the surface were sufficient to allow the bacterial cell orientation to be determined, whilst 115 TIRM imaging confirmed the close proximity to the surface (<200 nm) of near surface swimming cells. In 116 addition, to enable imaging of a higher density bacterial population than is possible for DHM, we 117 incorporated an optical relay (10) to recreate the sampling area virtually, enabling it to be rapidly scanned 118 in the z-dimension using a remote piezo-driven objective to avoid sample disturbance. A schematic of the 119 multimode 2/3D microscope setup is shown in Fig. 1A-B. Using this methodology we were able to 120 simultaneously determine the individual cell orientation and proximity to the surface as well as the 3D 121 location of cells at low and high densities over time. To exemplify the utility of this system, the microscope 122 was used to explore the respective contributions of the P. aeruginosa MotAB and MotCD stators to bulk 123 liquid movement and interactions at the solid-liquid interface. 124 125

Materials and Methods 126
Bacterial strains and growth conditions 127 the downstream nucleotide regions of motAB or motCD were generated using the primer pair 1FW/1RW, 133 and 2FW/2RW, respectively (Table S1). Both PCR products were fused by overlapping PCR to create a 134 deletion in the corresponding gene. The resulting fragment was cloned into the suicide plasmid pME3087 135 (45). Following transformation into the target strain by conjugation, single cross-overs were selected on 136 tetracycline (125 μg.ml -1 ). The double crossover mutants were selected by carbenicillin enrichment (28). 137 After 3 rounds of counter-selection, the resulting P. aeruginosa colonies were screened for the loss of 138 antibiotic resistance by plating on LB agar supplemented with or without tetracycline. The in-frame 139 deletions were confirmed by PCR and DNA sequence analysis. 140 141 For genetic complementation of the in-frame deletion mutants the motAB and motCD genomic regions were 142 PCR-amplified using P. aeruginosa chromosomal DNA as a template and the primer pairs indicated in 143   Table S1. The purified fragments were cloned into the shuttle vector pME6032 (19). The insertions were 144 verified by restriction enzyme and sequence analysis and introduced into the motAB and motCD mutants by 145 electroporation. The swarming ability of the mot mutants was verified by swarming plate assays (data not 146 shown). 147 148

Sample Chamber 149
Two strips of double-sided tape, approximately 2 × 15 × 0.1 mm, were placed on a borosilicate glass 150 coverslip (VWR) in parallel with a gap of ~4 mm. A second glass coverslip was placed on top of the 151 double-sided tape to create a chamber with a floor and ceiling of glass and walls of double-sided tape. The 152 second coverslip was rotated 45° to allow ease of loading the channel with growth medium (Fig. 1C). After 153 inoculation with bacteria, the two ends of the chamber were sealed using silicone free grease (Apiezon). 154 155

Microscopy 156
Imaging was achieved using a bespoke multimode microscope ( Fig. 1A; Cairn Ltd.). Samples were 157 analyzed at 37°C using a Nikon Eclipse Ti inverted microscope using a 60×, NA=1.49, WD=0.13 mm oil 158 objective. The microscope was fitted with an environmental chamber (Okolab) to regulate temperature, 159 relative humidity and CO2. Images were acquired using an Orca-Flash 4.0 digital CMOS camera 160 (Hamamatsu) at a typical acquisition rate of 41.6 Hz. DIC imaging was achieved using a single channel 161 white MonoLED (Cairns) light source. A polariser was inserted above the condenser, and Wollaston prisms 162 were inserted between the condenser and polariser and below the objective. Use of the prisms did not 163 adversely affect DHM image quality. Inline DHM imaging was acquired using a 685 nm LX laser (Obis). 164 TIRM was conducted using an Obis 488 nm LX laser (Cairn) controlled using an illumination system 165 (iLas2). The sample was illuminated through the objective by use of a 50 % mirror, controlled to achieve 166 total internal reflectance by irradiating the sample at an incidence angle (θ) greater than the critical angle 167 (2) 174 A total of 5 biological repeats for each mutant was prepared and imaged by interlaced capture of DHM with 175 either DIC or TIRM. Each image sequence consisted of 2000 frames (1000 frames for each acquisition 176 mode) for a total duration of 48 s. 177

178
The optical requirements of DHM, DIC and TIRM were met without interfering with the other imaging 179 modes, thus, multimode acquisition could be achieved by modulating the intensity of the different 180 illumination sources. Different modes of acquisition could not be acquired simultaneously but required interlaced capture achieved through computer control. The limiting factor in the time resolution of bacterial 182 motion capture was the frame rate of the camera, which in this case was 100 Hz. 183 184 Z-stack images were acquired using a bespoke remote focusing assembly as described previously (10). 185

186
To sample the early stages of bacterial cell surface attachment, P. aeruginosa was inoculated into the 187 sample chamber at OD600 0.015 in LB. Bacteria were observed to undergo cell division within the chamber 188 for 4-6 h until reaching stationary phase. 189

190
To observe flagella orientation during reversals, PAO1 was fluorescently stained using Alexa Fluor 191 carboxylic acid succinimidyl esters (Alexa Fluor 488, ThermoFisher;Friedlander et al. 2013;Turner, Ryu, 192 and Berg 2000). PAO1 cells were grown overnight in LB medium at 37 °C and 200 rpm, resuspended and 193 diluted in fresh LB medium at OD600 = 0.01. When cells reached exponential phase, they were centrifuged 194 at 2,000 x g for 10 min and the growth medium removed. The cell pellet was gently resuspended in a wash 195 buffer of 10 −2 M KPO4, 6.7 x 10 −2 M NaCl, 10−4 M EDTA, pH-adjusted to 7.0 with HCl and centrifuged. 196 After two additional rinses, cells were incubated with 0.5 mg/mL Alexa Fluor 488 carboxylic acid 197 succinimidyl ester for 1 h at room temperature with gentle rocking. Residual dye was removed after 198 washing twice and the cells resuspended in PBS for imaging. 199

Image processing 201
After acquisition, holograms were processed using a bespoke Matlab script (ImageProcess). Image intensity 202 across a stack was normalised, the median image over the stack was subtracted to remove background 203 signal including attached cells. DHM image output sequences were reconstructed using a bespoke Python 204 package (See-Through Scientific) (17-51) making use of Rayleigh-Sommerfeld formalism to determine 205 bacterial X-Y-Z coordinates. Bacterial trajectories were determined using a bespoke Matlab script 206 (DHMTracking). To exclude noise a voxel limit was set and applied to each image such that a similar 207 number of objects were identified in each frame. Visual inspection of interferogram image sequences 208 enabled an estimate of the number of bacteria per frame to ensure the voxel limit was set so as not to 209 exclude bacterial cells. Bacterial trajectories were determined using the bespoke Matlab script (tracking). 210 Objects were matched with their nearest neighbour within the next frame, applying a distance limit based 211 upon the maximum speed of bacteria (100 μm/s). The script looked up to 4 frames ahead for a matching 212 object. All tracks were also visually inspected using a bespoke Matlab script (Graphing) to ensure 213 trajectories skipping more than 4 frames could be joined. All Matlab scripts used for processing the data are 214 (3) 221 DIC images and Z-stacks were processed using a bespoke Matlab script (StackMaster). Each image was 222 flattened by a line-by-line polynomial fit. A threshold was then applied to binarise the image to segment 223 pixels with bacteria. Threshold was set as signal 2 :noise 2 > 10. The square of the signal was measured so as 224 to identify both bright and dark regions associated with the DIC image. Objects less than 20 pixels were 225 excluded as noise. Holes within objects were filled using the Matlab function 'imfill'. Bright and dark 226 regions were paired based upon proximity and a common directional vector between objects within the 227 same frame. The centre of mass of objects was calculated giving equal weighting to bright and dark regions. 228 The length and width of the smallest rectangle around an object and the orientation of the bacterial cells 229 was determined using a bespoke Matlab script (ParticleAnalysisDIC). For DIC Z-stacks, objects on 230 consecutive frames were analyzed to identify overlapping pixels considering the X-Y dimensions. If 231 common pixels were identified objects were grouped to form a single object and the centre of mass was 232 determined for this object based upon the combined pixels using a bespoke Matlab script 233 (DICZStack_ImageProcessing). Bacterial trajectories were determined as described for DHM data. 234 235 TIRM images were processed using ImageJ 1.50b. When processing interlaced DHM and TIRM images the 236 centre of mass of moving cells as determined by DHM was shifted to the average position between two 237 adjacent frames to account for the 20 ms offset between the capture of TIRM and DHM images. 238 239

Statistical analysis 240
Differences between datasets were assessed using unpaired t-tests or one way ANOVA analysis as 241 appropriate to determine the value p. The value N indicates the number of biological replicates. Error bars 242 indicate ± 1 standard deviation unit. 243 244

Imaging of individual P. aeruginosa cells in the bulk liquid and at the solid-liquid interface 246
To characterize P. aeruginosa motility using the multimode microscope, interlaced capture of either 2D 247 DIC or TIRM at the surface and 3D inline DHM images in the bulk liquid respectively was conducted 248 within 5 min of motile cells being added to the sample chamber ( Fig. 1C) at 41.7 Hz (total frame rate). At 249 this rate, bacterial fluctuating motion was under-sampled (body and Brownian motion) but directional 250 motion was oversampled (Fig. S1). After data acquisition, bacterial trajectory generation was achieved 251 using bespoke Matlab scripts (DHMTracking and StackMaster). Using this approach, trajectories could be 252 captured within the bulk medium ( Fig. 2A) and at the glass surface-liquid interface ( Fig. 2C-D). Single cell bacterial trajectories were determined using DHM ( Fig. 2A) at cell densities of up to 1.7 × 10 4 265 cells/μL, but above this, overlapping interference patterns prevented digital holographic reconstruction. To 266 measure the 3D distribution of cells at higher populations we used the remote image relay to acquire a series 267 of DIC images separated in the z-axis, referred to as z-stacks. Analysis of the z-stack data enabled the position 268 of each bacterium to be determined at a population density of 3.7 × 10 5 cells/μL, greater than an order of 269 magnitude improvement over DHM, with a maximum population density within any one image of 8.0 × 10 5 270 cells/μL. This enabled the growth of a P. aeruginosa cell culture to be followed over 4 h (Fig. 2B). Objects 271 that were vertically contiguous for adjacent frames were attributed to a single bacterium and the position of 272 each bacterial cell was determined as the centre of mass calculated from all pixels taken for combined objects. 273 We employed a z-spacing of 1 μm in order to ensure that we did not omit bacterial cells when imaging (typical 274 cell dimensions = 2 Î 5 μm, depth of field at 60× magnification NA = 1.4 ≈ 0.6 μm). Over a z-range of 100 275 µm and a maximum imaging rate of 100 Hz, this meant that the maximum acquisition rate using the DIC z-276 stack was 1 Hz. Although useful for determining cell population density and distribution, the lower 277 acquisition rate of this approach compared to DHM made it impractical for tracking fast swimming bacteria 278 since, for accurate tracking, bacteria should not travel further than their dimensions per frame, thus, limiting 279 this approach to cell movement slower than 5 μm/s. As such, DIC z-stacks were not used for tracking in this 280 study but rather to assess cell distribution at high cell density.

Characterization of bacterial trajectories 283
A number of different bacterial swimming trajectories were readily visually discernible within P. aeruginosa 284 populations (Fig 2A and Fig 4). Relatively straight trajectories with cells travelling at high speeds are referred 285 to as runs after the convention of Berg et al (5, 6) ( Fig. 4A). Cells swimming parallel to the surface, were 286 termed near surface runs (Fig. 4B), a phenomenon that has been assigned to various mechanisms including 287 hydrodynamic interactions, Brownian motion and surface contact (7,27,35). Visual inspection of the tracks 288 indicated that near surface runs had a greater incidence of directional change events than those observed in 289 the bulk where trajectories were straighter for longer. Moreover, we observed circular trajectories where 290 bacteria were travelling at the surface consistent with previous observations ( where all cells were reported to swim with the front pole down to the surface as estimated using 3 colour 305 DHM (8).
Away from the solid-liquid interface, an alternative class of swimming trajectory was observed, where the 308 movement of the bacteria was slower than the near surface runs, and characterized by frequent reversals of 309 swimming direction, a phenomenon assumed to be caused by reversals of flagella rotation (Fig. 4C), termed 310 oscillating after the classification applied by Rosenhahn et al (44). These oscillating bacterial cells were 311 travelling at a reduced speeds of 12 ± 5 μm/s, substantially slower than the cells exhibiting run trajectories at 312 59 ± 4 μm/s. Notably, we did not observe oscillating cells at the surface, likely because the constraints of the 313 proximal surface restricted the directional change of the cells upon flagella reversal. Near surface runs which 314 we term rambling were slower and exhibited more frequent changes in direction than bulk runs. This rambling 315 motion has not previously been reported. Rambling at the surface did not include the high frequency of 316 directional reversals seen in the bulk (>90 degree) for oscillating cells. 317 318 Some cases were observed where P. aeruginosa trajectories were reminiscent of peritrichously flagellated 319 Escherichia coli run-and-tumble patterns (5, 6), whereby a straight 'run' path was interspersed by a high 320 frequency (2 Hz) of reversal events before the bacteria recommenced a run trajectory (Fig. S2B). The 321 average reversal rate observed in the bulk population was 0.5 Hz. We observed no changes in the 322 orientation of the bacterial body for surface reversal events (where the change in direction exceeded 120° -323  (37). To observe this more closely, we imaged cells using a third of the 326 camera's chip size enabling a higher acquisition rate of 333 Hz. At this higher frame rate we were still 327 unable to observe any reorientation of the bacteria (Movie S3). Moreover, equipping the microscope with a 328 fluorescence filter set enabled the use of total internal reflectance fluorescence (TIRF) imaging in order to 329 visualize flagella through the use of fluorescent staining. Using this approach, flagella were readily 330 observed at both the leading and trailing end of bacteria before and after surface reversal events (Fig. S2C, 331 Movie S4), showing that P. aeruginosa cells are able to use their flagella to both push and pull, consistent 332 with previous observations for monotrichous bacteria (12, 20, 32). We observed no statistical differences in 333 the speed of bacterial trajectories before or after a reversal event (Fig. S2D). 334 335 Stationary cells at the surface were found attached either horizontally or vertically (long-axis or pole 336 attached). Long-axis attached cells were identified from DIC images where the distinct rod-shape of the 337 bacteria was observed and remained stationary throughout the duration of an image sequence with no change 338 in the centre of mass of the cell (Fig. 2C). In contrast, pole-attached bacteria were observed to be smaller and 339 circular in appearance and small deviations in the cell's centre of mass were observed as the cell wobbled on 340 a single attachment point (Fig. 2C). In some cases, polar attached cells were observed to be spinning due to 341 the flagella being fixed to the surface whilst the bacteria body remained free and able to rotate, a phenomenon 342 observed previously (14) (Movie S5). 343 344

Exploring the respective contributions of the MotAB and MotCD stators to swimming motility 345
To exemplify the applications of the multimode microscope, we investigated the relative contributions of 346 the MotAB and MotCD stators to swimming motility by comparing the trajectories of the P. aeruginosa 347 wild type with the corresponding motAB, motCD and motABCD deletion mutants in order to assess cell 348 behavior both within the bulk and at the surface. By tracking bacteria over a 2.4 s interval (Fig. 5), we first 349 characterised the trajectories of each strain to determine whether the different forms of cellular movement 350 apparent on visual inspection of cell tracking videos could be quantitatively categorized. As anticipated, the 351 motABCD mutant was non-motile (15, 39), with all trajectories observed from this strain having a speed 352 below 10 μm/s and a KMSD below 1.5 (Fig. S3P). The average speed and KMSD observed for this strain was 353 5.3 μm/s and 0.86, respectively. Therefore, the motABCD mutant was not analyzed further. 354 355 Cell tracks were quantitatively characterized by instantaneous speed, total displacement, and their mean 356 square displacement (MSD) over the time range (δt) of 50 to 1000 ms (Fig. S3A-I). The slope of the log 357 plot (KMSD) was used to assess the directionality of a track, whereby a slope < 1 suggests constrained 358 motion, a slope = 1 suggests Brownian diffusion and a slope > 1 suggests active motion (Fig. S3A-I) (11, 359 38, 48). 360 361 Considering the mean speed and KMSD together, it was evident that two trajectory types were apparent in 362 the P .aeruginosa wildtype population at both the surface and in the bulk (Fig. S3J-O). A cluster of tracks 363 was observed for cells with a mean velocity above 30 μm/s and a KMSD above 1.5, with the remaining 364 trajectories having a mean velocity below 20 μm/s or a KMSD below 1. A similar grouping of trajectories 365 was observed for the motAB and motCD mutants, although in both cases the separation of the two groups 366 was less evident due to a reduction in the velocity observed for the directionally swimming bacteria (Fig.  367 S3K-L). Trajectories were separated into two classes, the first included bacterial tracks with a high velocity 368 (> 25 μm/s) and a high KMSD (>1.5), whilst the remaining were grouped into the second. Plots of the 369 resulting tracks divided into the two classes are shown in Fig. 5. It was evident that based upon the 370 selection criteria described, the trajectories separated into two clearly distinct motility types; in the first, the 371 bacteria moved over large distances in a highly directional manner which have previously been described as 372 runs, (Fig. 4A) and in the second, frequent reversals in direction were evident, termed oscillating (Fig 4C). 373 Trajectories that remained at the surface were assigned as rambling as opposed to oscillating. We assigned 374 runs that remained near the surface (within the focal plane of DIC images) for greater than 1 s as near 375 surface runs (Fig. 4B). We were thus able to automatically assign the trajectory type and exclude the 376 possibility of operator bias. The different bacterial trajectories were individually assessed to determine 377 frequency, average speed and KMSD (Table 1). 378 379 Using the analysis described, trajectory types were compared at the surface and within the bulk. The bulk 380 bacterial wild-type trajectories were predominantly (64 %) runs, at an average speed of 59 ± 4 μm/s (mean 381 ± standard deviation). In comparison both motAB and motCD mutants were predominately oscillating 382 (≈54%) and the average run speed was significantly (p<0.0001) reduced in both cases to 29 ± 9 μm/s and 45 383 ± 18 μm/s, respectively ( Fig. 6A; Table 1). This average measure of single-cell bacterial swimming 384 contrasts with conventional measurements of population swimming speeds in low-viscosity agar whereby 385 under these conditions no change in speed was observed for P. aeruginosa motAB and motCD mutants (40). 386 In this prior study different classes of motility were not taken into account. 387 388 Increased variance in the speed within a single run trajectory was observed for the motCD mutants in the 389 bulk (average standard deviation = 32 ± 36 μm/s compared with 18 ± 4 μm/s and 9 ± 1 μm/s for the 390 wildtype and motAB strains, respectively. Fig. 6A, Table 1). Thus, removal of the motCD stator reduced 391 the ability of the bacteria to swim at a constant speed. 392

393
The average speed of near surface runs was significantly (p<0.01) reduced compared to the average run 394 speed in the bulk for all strains by 7 %, 39 % and 51 % for the wildtype, motAB and motCD mutants, 395 respectively ( Fig. 6A-B, Table 1). Thus, strains lacking either stator reduced their speed more significantly 396 at the surface than the wildtype strain and, therefore, either reduced their power input or were unable to 397 achieve the same speed for a certain power input within the fluid dynamics experienced at the liquid-solid 398 interface (43, 50). This demonstrates a difference in the response to the surface between the stator mutants 399 and the WT, consistent with a key role for the stators in responding to the surface environment as bacterial 400 cells shift from the bulk environment to the near surface environment. Here, observation of the differential 401 behaviour of the bacteria at the surface and bulk was achieved through the interlaced capture of cells both 402 within the bulk and at the surface. The directionality of the run trajectories for all strains (including the 403 wildtype) was unaltered either in the bulk or at the surface, with a KMSD of 1.8 to 1.9 observed in all cases 404 (Table 1). 405

406
In an oscillating trajectory the bacteria rapidly changed direction (Fig. 4C) as a result of the reversal of 407 bacterial flagella rotation. (Fig. 4E). Both the wildtype and motCD mutant had high oscillating frequencies 408 of 3.1 ± 1.0 Hz and 2.6 ± 0.6 Hz, respectively, whereas the motAB mutant exhibited a reduced oscillating 409 frequency of 1.5 ± 1.0 Hz (significant at p<0.05) ( Table 1). The average instantaneous speed of the 410 oscillating motAB mutant was also higher (19 ± 5 μm/s) than the two other strains (≈11 μm/s; Table 1. As 411 oscillating trajectories were observed for both mutants, either stator appears to be sufficient to support this 412 type of trajectory. However, the altered phenotype of the motAB mutant suggests a possible role for MotAB 413 in controlling the oscillatory trajectory. The directionality of the oscillating trajectories over a δt of 50 to 414 1000 ms also varied between the two mutants. The KMSD values of 1.2-1.6 were lower than the run 415 trajectories, thus this trajectory type allow the bacteria to move in a more diffusive manner. A statistically 416 significant increase in KMSD was observed for the motAB mutant compared with both the wildtype (p=0.04) 417 and motCD (p=0.001) that is consistent with the oscillating trajectory requiring a contribution from MotAB. 418 419 A statistically significant (p<0.001) reduction in average speed was observed between the run and rambling 420 trajectories at the surface for the wildtype and motAB strains, but not for the motCD mutant due to the high 421 variability in the speed of surface runs for different trajectories observed for this strain ( Table 1). The 422 reduction in average speed from 22 to 12 µm/s for surface runs and rambling, respectively, for the motCD 423 mutant was consistent with the other strains. For all strains a statistically significant (p<0.05) reduction in 424 directionality from a KMSD of 1.8-1.9 to 1.5-1.7 was observed. No significant difference in the KMSD values 425 for the rambling trajectories of the different strains was observed. 426 In addition, surface attachment, detachment and reversals for the mot mutants were also compared. After 5 428 min a larger proportion (72 %) of the wildtype cells were attached to the surface compared with either of 429 the mot mutants (≈22 %) ( Table 1). Of the attached cells, variance was also observed in the ratio of pole-430 attached to long axis attached cells, with values of 0.42, 2.7 and 1.2 observed for the wildtype, motAB and 431 motCD mutants respectively ( Table 1). Switching from long axis to polar attachment has previously been 432 associated with cell surface departure (13). Consistent with this, although no statistically significant 433 differences were observed between the attachment rates, the detachment rates for both mutants was higher 434 (≈ 2.3 % of attached cells per second) than the wildtype (0.2 % of attached cells per s; (Table 1). Path 435 deviations greater than 120 degrees, classified as reversals, were observed very infrequently for runs or 436 rambling tracks. The highest reversal rates were observed for the wildtype bacteria (0.5 ± 0.1 Hz) whilst 437 both mot mutants had similar reversal rates of 0.1 ± 0.1 Hz. The reversal rates for all strains increased when 438 the bacteria were involved in near surface runs. Whereas for both wild type and the motAB mutant, reversal 439 rates increased 2-fold upon exposure to the surface, the motCD mutant increased reversals by 6-fold (Table  440 1). This result contrasts with reversal rates observed previously for P. aeruginosa strain PA14 (39), where 441 the reversal rates observed for the respective motAB and motCD mutants were 2-3 fold higher than the 442 wildtype. This is likely due to different P. aeruginosa strain or culture conditions used, specifically the 443 PA14 experiments were conducted using stationary phase bacteria grown in M63 medium supplemented 444 with glucose and with the addition of 3% Ficoll to increase growth medium viscosity and so slow motility 445 to enable imaging. The load responses of MotAB and MotCD stators have also been well studied (2) Fig. 7. A higher association of the wildtype and the motAB mutant with the two glass 466 surfaces compared to the bulk was observed, both at the top and bottom (Fig. 7). For the wildtype bacteria 467 the cell fraction (per μm) was more than 15×  Characterization of bacterial trajectories revealed two types of motility that were consistent with previous 497 studies (37, 44), and could be categorized by consideration of speed and directionality. Notably, although 498 runs were observed both at the surface and in the bulk, oscillatory trajectories were observed only within 499 the bulk whilst rambling trajectories were observed only at the surface. Deletion of either stator did not 500 inhibit any trajectory type, demonstrating that a specific stator complex was not associated with a specific 501 trajectory type. However, the oscillatory frequency for the motAB mutant was reduced compared to the WT, 502 suggesting a role for this stator in regulating this particular trajectory type. It was not clear what caused the 503 bacteria to exhibit a particular trajectory type, however, the lower speed and decreased directionality of the 504 rambling and oscillatory trajectory types suggest an explorative behavior in contrast to runs, which allow 505 bacteria to travel more rapidly over larger differences. Adopting the rambling or oscillatory trajectories also 506 enable bacterial cells to alter direction in a similar manner to the tumbling action observed for the 507 peritrichous E. coli (5). Whereas the WT strain swimming speed was statistically indistinguishable between bacteria swimming at 517 the surface and those within the bulk, the motAB and motCD mutants swimming speed at the surface was 518 significantly (p<0.0001) slower than cells within the bulk. It is possible that either the WT is able to adjust 519 to the altered physico-chemical near-surface environment or to obstruction of flagella rotation by the 520 surface due to the exchange of the two stator complexes to modulate flagella torque. Stator exchange has 521 been proposed to act as a signaling mechanism in concert with c-di-GMP within P. aeruginosa (1, 3, 24). 522 The absence of either stator within the motAB and motCD mutants removes the option for stator exchange, 523 preventing the cells from being able to either sense or respond to the near surface environment using this 524 mechanism. 525 526 The involvement of both bacterial appendages, flagella and pili, in bacterial swarming, biofilm formation 527 and, ultimately, surface sensing has been reported (3,23,25,29,33). In P. aeruginosa, flagella signaling 528 involves c-di-GMP and the diguanylate cyclase SadC (3), whilst pili, particularly TFP, are able to 529 mechanically interact with a surface through successive pili extension and retraction that enables signal 530 transduction through the Chp system to regulate cyclic adenosine monophosphate (c-AMP) levels (31). 531 Whilst the mechanosensing role of pili for regulating the behavior of surface associated bacteria is clear, 532 our results, and those of others (25) also support a role for the flagellum in surface sensing. 533

534
Here we have also shown that both the motAB and motCD mutant populations included a higher proportion 535 of detached cells, and that attached cells were more likely to detach, further demonstrating the differential 536 surface response of P. aeruginosa cause by the deletion of either of the flagella stators. This is likely caused 537 by the inability of the bacteria to sense the surface effectively and so they fail to trigger the upregulation of 538 c-di-GMP and pili required for cells to become irreversibly surface attached leading to biofilm formation. 539 The similar behaviors of both mot mutants suggest that both stators are involved in the cell signaling, 540 potentially through stator exchange. The absence of surface accumulation noted only for the motCD mutant 541 however, indicates a greater role for the MotCD stator in surface sensing and subsequent biofilm formation. 542

543
In summary, we have developed a multimode 2/3D microscope that combines DIC, DHM, TIRM and TIRF 544 imaging modes to achieve simultaneous bulk and surface label-free imaging of single cells in a motile 545 bacterial population. The microscope was used to investigate the role of the P. aeruginosa motAB and 546 motCD flagella stators on motility and surface interactions, enabling observations of altered phenotypic 547 behavior of cells located within the bulk or at the surface. 548    849 TABLE S1. List of oligonucleotides used in this study for the construction of motAB and motCD in frame 850 deletion mutants and for genetic complementation. 851

MotCD FW 5'-TATGAATTCATGGATGTGCTCAGCCTG-3'
MotCD RW 5'-TATCTCGAGTCATGGCGAAGGCGACGG-3' 852 1 KMSD has widely been used to characterise the movement of particles, enabling directed, Brownian and confined movement to be defined. A number of different examples of bacterial tracks and the associated KMSD measurements are shown in Fig. SI5. Tracks travelling directionally were observed to produce a linear relationship between the log of MSD over the log of time intervals with a slope approaching 2 (Fig. S5A). A track characterised by frequent oscillations had a reduced KMSD of 1.56 (Fig. S5B), whilst the drifting of a non-motile bacterium had a KMSD closer to 1 (Fig. S5C). A bacterium attached to the surface and therefore confined has a KMSD of approximately 0 (Fig. S5D). A KMSD below 2 was also observed for a bacterium travelling in circular trajectories (Fig. S5E) or spinning (Fig.  S5F). In this case a linear relationship between the ln of MSD and ln of dt was observed up to the time interval associated with a complete oscillation, whereupon a plateau in the log MSD was observed at a position determined by the diameter of the curved path (Fig. S5F). In cases where the bacterium experienced oscillating movement over short time scales (fluctuating motion) and directional movement over larger times scales (mean motion) two slopes were observed on the loglog plot of MSD and dt (Fig. S5G), whereupon the two slopes were indicative of the directionality of the two movement types. As directional movement of the bacteria was of interest for comparing trajectories and the oscillating movement of the bacteria over short time frames was undersampled (Fig. S1) the KMSD value was calculated over dt values of 50 to 1000 ms. Changes in the bacterial movement during a single track caused by reversal events or attachment or detachment events resulted in spurious measurements of KMSD (Fig. S5H-I). For this reason tracks were split when attachment, detachment or reversal events were observed prior to KMSD analysis.