Bacterial Swarming Reduces Proteus mirabilis and Vibrio parahaemolyticus Cell Stiffness and Increases β-Lactam Susceptibility

Proteus mirabilis and Vibrio parahaemolyticus are bacteria that infect humans. To adapt to environmental changes, these bacteria alter their cell morphology and move collectively to access new sources of nutrients in a process referred to as “swarming.” We found that changes in the composition and thickness of the peptidoglycan layer of the cell wall make swarmer cells of P. mirabilis and V. parahaemolyticus more flexible (i.e., reduce cell stiffness) and that they become more sensitive to osmotic pressure and cell wall-targeting antibiotics (e.g., β-lactams). These results highlight the importance of assessing the extracellular environment in determining antibiotic doses and the use of β-lactam antibiotics for treating infections caused by swarmer cells of P. mirabilis and V. parahaemolyticus.

B acteria have evolved a variety of mechanisms to adapt to their physical environment. For example, in response to fluctuating environmental conditions, changes in biochemistry and gene regulation can alter bacterial shape and increase cell fitness. Cell filamentation is a commonly observed change in bacterial cell shape (1,2) and has been described as a mechanism that enables bacteria to evade predation by the innate immune system during host infections (1).
In close proximity to surfaces, many bacteria alter their morphology and leverage cell-cell physical contact to move collectively to access new sources of nutrients and growth factors (3,4). Referred to as "swarming," this process is common among motile bacteria and has been connected to bacterial pathogenesis and infections (3,4). Swarmer cells of Salmonella enterica, Pseudomonas aeruginosa, Serratia marcescens, and Bacillus subtilis have reduced susceptibility-compared to vegetative cells-to a variety of antibiotic drugs that alter protein translation, DNA transcription, and the bacterial cell membrane and cell wall (5)(6)(7)(8). The specific biochemical and biophysical mechanisms underlying these observations are unknown.
Here, we describe physical changes in swarmer cells of the Gram-negative pathogenic bacteria Proteus mirabilis and Vibrio parahaemolyticus that have the opposite effect: they increase the susceptibility of cells to cell wall-targeting clinical antibiotics. We found that large changes in the length of P. mirabilis and V. parahaemolyticus swarmer cells are accompanied by an increase in flexibility (i.e., a reduction in cell stiffness) that enables long cells to pack together tightly and form cell-cell interactions; maximizing cell-cell interactions promotes surface motility (9). Using biophysical, biochemical, and structural techniques, we quantified changes in the structure and composition of the cell wall of P. mirabilis and V. parahaemolyticus in swarmer and vegetative cells and characterized their susceptibility to osmotic changes and cell wall-modifying antibiotics. Our results indicate that morphological changes that enable these bacteria to adapt to new physical environments come at a significant fitness cost, as cells become more susceptible to their chemical environment. In particular, changes in the composition and thickness of P. mirabilis and V. parahaemolyticus swarmer cells may make them more sensitive to osmotic changes and to cell wall-modifying antibiotics, thereby suggesting that these classes of drugs may be useful in treating infections of these bacteria (e.g., in urinary tract infections [UTIs]).

RESULTS
The bending rigidity of P. mirabilis and V. parahaemolyticus cells decreases during swarming. During surface motility, P. mirabilis and V. parahaemolyticus cells grow into swarmers that are characteristically long (10 to 100 m) and present flagella at a high surface density that enables them to translate through viscous environments (3,10). We found that these swarmer cells display an unusual phenotype that is rarely observed among Gram-negative bacteria: remarkable flexibility and a shape that is dynamically altered by adjacent cell motion and collisions (Fig. 1). The ability of P. mirabilis swarmer cells to maximize cell-cell contacts plays a role in their cooperative motility (10); our observations indicate that flexibility enables these long cells to optimize packing into multicellular structures that move cooperatively across surfaces.
Bacterial cell mechanics is generally attributed to the peptidoglycan (PG) layer of the cell wall, which has a thickness of ϳ3 to ϳ50 nm and surrounds the cytoplasmic membrane (11). Very little is known about mechanical regulation in bacteria (12)(13)(14)(15)(16)(17), and we are unaware of studies connecting swarming to changes in cell mechanics. We quantified changes in swarmer-cell stiffness using cell-bending assays in a reloadable, poly(dimethylsiloxane) microfluidic system ( Fig. 2; see also Fig. S1, available at https:// osf.io/hvawp) that is related to a method developed previously (18). In bending assays, we applied a shear fluid force to multiple filamentous cells, resulting in horizontal deflection of their cell tips (Fig. 2); fitting the deflection data to a mechanical model provided us with a value of the (flexural) bending rigidity of cells (Fig. S2). After a thorough comparison of several models of variable complexity, we found our results to provide a reasonable semiquantitative estimate of bending rigidity. Introducing a reloadable mechanism enabled us to perform rapid bending measurements of P. mirabilis and V. parahaemolyticus swarmer cells after isolating them from swarm plates. Once removed from a surface, P. mirabilis and V. parahaemolyticus swarmer cells dedifferentiate, grow, and divide to form vegetative cells that resemble wild-type cells with respect to length, requiring us to rapidly perform assays with swarmer cells after their isolation from surfaces. As a point of comparison, we filamented vegetative cells of P. mirabilis and V. parahaemolyticus using aztreonam-an inhibitor of the divisionspecific transpeptidase PBP3-to match the length of swarmer cells (22.2 Ϯ 12.5 m and 12.4 Ϯ 8.2 m, respectively) and compared their bending rigidity values to those determined for swarmer cells. As a control, we measured the bending rigidity of cells of Escherichia coli strain MG1655, which we filamented using aztreonam, and determined the value to be 3.7 ϫ 10 Ϫ20 N m 2 (Fig. 3); using a value for the thickness of the PG of 4 nm (19) yields a Young's modulus of 23 MPa, which is close to values that have been reported previously and supports the choice of using aztreonam to filament cells, as it apparently has no effect on the bending rigidity of cells (12,18). We assume that the effect of aztreonam on P. mirabilis and V. parahaemolyticus cells is similar to that which we measured for E. coli. We found a substantial decrease in the bending rigidity of swarmer cells of both P. mirabilis (ϳ26-fold) and V. parahaemolyticus (2.1-fold) compared to vegetative cells (Fig. 3), which is consistent with our observations of the

Proteus swarmer
Vibrio swarmer flexibility of swarming cells by microscopy. V. parahaemolyticus vegetative cells were remarkably flexible: ϳ154-fold more flexible than E. coli cells and ϳ58-fold more flexible than P. mirabilis swarmer cells (Fig. 3).
To confirm that using aztreonam to inhibit PBP3 and produce filamentous cells does not change the cross-linking density of PG at the division plane or alter cell mechanics,  we compared bending rigidity values of cells treated with aztreonam and cells filamented by overexpressing SulA, a protein that prevents polymerization of the division protein FtsZ and blocks cell division. The two methods of filamenting E. coli cells produced similar bending rigidity values: 3.8 ϫ 10 Ϫ20 N m 2 (SulA) and 3.7 ϫ 10 Ϫ20 N m 2 (aztreonam) (Fig. S3A).
During swarming, P. mirabilis and V. parahaemolyticus cells dramatically increase their density of flagella. Since flagella are attached to these cells through a protein complex that extends through the PG (referred to as the basal body), it is possible that these structures create local defects in the PG that alter its mechanical properties. To test whether the density of flagella on swarming cells is responsible for the changes in cell mechanics that we observed, we performed bending rigidity measurements (Fig.  S3B) of two K-12-derived strains of E. coli that have substantially different densities of flagella (Fig. S3C). Our results indicated no appreciable change in stiffness values (bending rigidity of 4.1 ϫ 10 Ϫ20 N m 2 at high flagellar density and 3.7 ϫ 10 Ϫ20 N m 2 at low flagellar density; Fig. S3B).
Overexpressing FlhDC-the heterohexameric activator that is important for swarming-in vegetative cells growing in liquid produces a phenotype that replicates many of the characteristics of swarmer cells, including increased cell length and flagellar density (20). Whether there is a relationship between FlhDC production and cell mechanics is untested. To test whether FlhDC is connected to changes in swarmer cell stiffness, we overexpressed proteins encoded by the plasmid-carried pflhDC genes in filamentous cells of P. mirabilis and measured their bending rigidity. We detected an ϳ1.8-fold difference in bending rigidity between the wild-type cells (1.4 ϫ 10 Ϫ20 N m 2 ) and the pflhDC-containing P. mirabilis vegetative cells (7.8 ϫ 10 Ϫ21 N m 2 ) and approximately equal levels of bending rigidity for the wild-type swarmer cells and pflhDCcontaining P. mirabilis swarmer cells (5.5 ϫ 10 Ϫ22 and 7.4 ϫ 10 Ϫ22 N m 2 , respectively) The values for each cell type represent an average of the data from two fitting models (see the supplemental material), and the bracket bars represent the upper and lower limits of the two models. More than 100 cells were used for each cell type from at least 3 independent experiments. The 95% confidence intervals associated with the fits are shown in Fig. S19. The plot has a logarithmic y-axis scale.
Bacterial Swarming Reduces Cell Stiffness ® (Fig. 3). These results indicate that FlhDC overexpression during swarming is a minor contributor to the mechanical phenotype of swarming cells; the majority of the mechanical changes that we observed arose from another regulatory pathway(s). P. mirabilis and V. parahaemolyticus swarmer cells are more sensitive to osmotic stress than vegetative cells. To complement the cell stiffness measurements, we also studied changes in the mechanics of swarmer cells and filamented vegetative cells by measuring their changes in length and width in response to changes in osmotic pressure, which cause cells to shrink (in a NaCl solution) and swell (in H 2 O) (Fig. S4). We filamented P. mirabilis, V. parahaemolyticus, and control E. coli cells, using aztreonam to create a range of cell lengths that matched the lengths of P. mirabilis and V. parahaemolyticus swarmer cells. We expected that low values of bending rigidity and compositional and structural changes in PG would cause swarmer cells to elongate (in comparison to filamented, vegetative cells) in response to changes in osmotic pressure. We found that the amount of P. mirabilis and V. parahaemolyticus swarmer cell extension in response to osmotic shock was significantly greater than that seen with filamented, vegetative cells (Fig. 4). To circumvent the cellular production of osmolytes to protect cells from large changes in osmotic pressure during shock (typically pro- duced within 1 min) (21), we used a microfluidic device to switch rapidly (Ͻ5 s) between H 2 O and a NaCl solution and measured the changes in cell length that occurred before the cells adapted (Fig. S4).
Osmotic shifts produced similar changes in the length (Fig. 4A) and width (Fig. 4B) of vegetative filamented P. mirabilis cells compared to E. coli cells ( Fig. 4A and B). In contrast, V. parahaemolyticus vegetative cells substantially increased in cell length ( Fig. 4C) compared to E. coli cells, with no observable change in cell width (Fig. 4D). In response to osmotic downshifts (i.e., transitioning from NaCl to H 2 O), P. mirabilis and V. parahaemolyticus swarmer cells dramatically increased in length ( Fig. 4A and C) (and, to a lesser extent, in width for P. mirabilis but not for V. parahaemolyticus; Fig. 4B and D) compared to vegetative cells. These results suggest that the changes in P. mirabilis and V. parahaemolyticus swarmer cell stiffness made them more responsive to osmotic changes.
Changes in PG composition of P. mirabilis and V. parahaemolyticus swarmer cells. We next studied the structural and biochemical mechanisms underlying the change in the mechanical properties of swarmer cells. The connection between the structure of PG and cell mechanical properties-discovered in early antibiotic studies that revealed the penicillin binding proteins-suggested the use of PG as a starting point for these studies.
PG consists of a disaccharide building block, ␤-(1,4)-N-acetylmuramic acid/Nacetylglucosamine (MurNAc-GlcNAc), in which a pentapeptide is attached to each 3=-OH group on MurNAc. Cross-linking between adjacent pentapeptides creates a mesh-like polymeric layer, and altering its structure and composition affects cell mechanical properties (14,15). To determine whether the PG composition of P. mirabilis and V. parahaemolyticus cells changes during swarming, we isolated PG sacculi from vegetative and swarmer cells and used ultra-high-performance liquid chromatographymass spectrometry (UPLC-MS) to quantify their chemical composition (Fig. S5). As the PG composition of V. parahaemolyticus has not yet been reported, we characterized its muropeptide stem peptide using UPLC-tandem MS (UPLC-MS/MS) ( Fig. S6 and S1). Similarly to E. coli (22) and P. mirabilis (23), V. parahaemolyticus has a PG structure like that conserved across other Gram-negative bacteria, in which the peptide stem consists of L-Ala-D-Glu-meso-diaminopimelic acid (meso-DAP)-D-Ala-D-Ala (Fig. S6).
P. mirabilis swarmer cells contained fewer monomers (MurNAc-GlcNAc), more dimers, and more anhydrous-substance-containing saccharides, which are found at the terminating end of glycan polymers, than are contained by P. mirabilis vegetative cells ( Fig. 5A) (11). We detected no differences between swarmer and vegetative cells of P. mirabilis in the relative abundances of trimers (Fig. 5A). The increase in the levels of anhydrous-substance-containing saccharides that we observed in P. mirabilis swarmer cells was correlated with a decrease in polysaccharide length (Fig. 5B). A similar increase in the levels of anhydrous-substance-containing saccharides and decrease in the length of polysaccharides occurred in V. parahaemolyticus swarmer cells ( Fig. 5A and B). We found no differences in cross-linking densities between vegetative and swarmer cells of either P. mirabilis or V. parahaemolyticus (Fig. 5C).
Swarmer cells have reduced PG thickness and display membrane defects. Changes in the thickness of the PG layer and structure of the cell envelope may also explain the observed decrease in swarmer cell stiffness. To identify changes in the PG thickness of swarmer cells, we isolated intact sacculi from P. mirabilis vegetative and swarmer cells and measured the thickness of dried sacculi using tapping-mode atomic force microscopy (AFM) (Fig. 6A). No differences in the nanoscopic appearance of the sacculi of different cells were observed by AFM (Fig. S7). The thickness of isolated, dry P. mirabilis swarmer cell sacculi (1.0 Ϯ 0.2 nm) was reduced ϳ1.5-fold compared to those of vegetative cells (1.5 Ϯ 0.2 nm) (Fig. 6A). V. parahaemolyticus swarmer cells (0.6 Ϯ 0.1 nm) exhibited a similar ϳ1.2-fold decrease in thickness compared to vegetative cells (0.8 Ϯ 0.2 nm). Earlier AFM measurements of isolated sacculi indicated that dehydration reduced the thickness of Escherichia coli PG by ϳ2ϫ, which we used to estimate the dimensions of hydrated PG from P. mirabilis (3.1 and 2.1 nm for vegetative and swarmer cells, respectively) and V. parahaemolyticus (1.7 and 1.4 nm, respectively) (24). Comparisons of PG thickness and cell-bending rigidity data suggested that the relationships between these data are approximately exponential (R 2 ϭ 0.9874) (Fig. 6B).
A caveat with respect to conversions between dried and hydrated values is that they are most accurate for examples of PG that best mimic the structure and composition (e.g., cross-linking and polysaccharide composition) of the reference material: E. coli PG. Alterations in the polysaccharide length of PG from P. mirabilis and V. parahaemolyticus may make it more elastic and make it stretch out during drying, thereby appearing to have a thickness that is reduced compared to that of E. coli PG. Our control measurements of isolated, dry E. coli sacculi yielded a thickness value of 2.0 nm, which differs from the value of 3.0 nm published previously by Yao et al. (24), who used the same measurement technique. A difference of ϳ30% between these E. coli measurements might arise for several reasons, including variations in physical conditions that impact AFM measurements, improvement in the resolution of AFMs, and/or the precision of fitting force curves.
The relatively low range of variability (ϳ16% to ϳ25%) in the values that we measured for isolated sacculi from P. mirabilis and V. parahaemolyticus cells using AFM demonstrated a consistent reduction in PG thickness for swarmer cells, suggesting that our measurements are sufficient for comparing PG from vegetative and swarming cells UPLC-MS data reveal that the muropeptide compositions of P. mirabilis and V. parahaemolyticus vegetative and swarmer cells differed slightly in the abundances of monomers, dimers, and anhydrous-substance-terminated saccharides. (B) We observed a relative increase in the amount of anhydrous-substance-containing saccharides in swarmers consistent with a decrease in polysaccharide strand length. (C) There was no change in PG cross-linking of P. mirabilis and V. parahaemolyticus vegetative and swarmer cells, although V. parahaemolyticus did display a lower level of cross-linking. n ϭ 3 biological replicates. Error bars represent the standard deviations of the means. For (panels A to C, significance was determined via two-way analysis of variance: *, P Յ 0.05; **, P Յ 0.01; ***, P Յ 0.001; ****, P Յ 0.0001; ns, not significant (P Ͼ 0.05). and for demonstrating a connection between PG and changes in cell mechanics. In contrast, the variability in AFM measurements and analysis makes us less comfortable with respect to comparing the absolute values of PG thickness measured in our studies to those reported for bacteria in other papers.
To complement the AFM measurements, we attempted to measure the thickness of native PG using electron cryotomography (ECT) on intact vegetative and swarmer cells ( Fig. 6C and D; see also Fig. S8 and S9). Although we were unable to resolve the thickness of native PG by direct ECT measurements, we noticed that the subtomogramaveraged ECT volumes of the P. mirabilis cell wall (Fig. 6C) indicated that the distance between the inner and outer membranes of swarmer cells was smaller than in vegetative cells (Fig. 6C and D). P. mirabilis vegetative cells had a characteristically smooth membrane ( Fig. S8A and B) that was similar to the presentation of the membrane found along the lateral, cylindrical walls of swarmer cells ( Fig. S8C and D). In contrast, the polar regions of both P. mirabilis and V. parahaemolyticus swarmer cells had an undulating outer membrane suggestive of an altered structure ( Fig. S8E and F and S9D to F), and V. parahaemolyticus cells had significant defects in their cell envelopes, including membrane budding, vesicle formation, and ruptured cell walls (Fig. S9D to F). Bacterial Swarming Reduces Cell Stiffness ® Increased susceptibility of P. mirabilis and V. parahaemolyticus swarmers to the cell wall-targeting antibiotics. Although swarming colonies of bacteria display resistance to many antibiotics (5), our experiments suggest that P. mirabilis and V. parahaemolyticus swarmer cells have changes in PG structure, composition, and properties that may increase their susceptibility to cell wall-targeting antibiotics. To determine the sensitivity of P. mirabilis and V. parahaemolyticus swarmers to the cell wall-targeting antibiotics cephalexin (which inhibits PBP3 and PBP4) (25) and penicillin G (inhibits PBP3, PBP4, and PBP6), we measured swarmer single-cell growth in the presence of the MIC of antibiotics in a microfluidic device. We determined MICs of cephalexin against vegetative cells of E. coli MG1655 (1ϫ MIC ϭ 6.25 g/ml; 32ϫ MIC ϭ 200 g/ml), P. mirabilis (100 g/ml), and V. parahaemolyticus (50 g/ml). The MIC of penicillin G against vegetative cells of P. mirabilis was 12.5 g/ml. We were unable to measure MICs for these antibiotics against swarmer cells because they dedifferentiate at shorter time scales than are used for the measurements in the MIC protocol. Instead, we chose 3 h of incubation for our measurement for three key reasons: (i) swarmer cells do not dedifferentiate over this time period; (ii) a past study demonstrated that treating cells with cephalexin for 3 h was sufficient to kill ϳ100% of a population of E. coli cells (26); and (iii) ϳ2 h of cephalexin treatment was sufficient to observe cell lysis of E. coli cells measured by single-cell growth (27).
We constructed a simple microfluidic flow cell to maintain the swarming phenotype of cells during experiments and to supply cells with a source of continuously replenished fresh growth media to ensure exponential cell growth and a constant concentration of antibiotics. The channel walls consisted of the biocompatible polymer poly(dimethylsiloxane) (PDMS). Cells were in contact with the channel floor, which consisted of a layer of a 2% agarose gel. (See Fig. S12 for a cross section of the device.) We found that the survival rates of P. mirabilis (66%) and V. parahaemolyticus (64%) vegetative cells treated with 1ϫ MIC of cephalexin were slightly higher than that of E. coli cells (55%) (  (Fig. 7A). We characterized dead cells using microscopy to measure membrane blebbing, cell lysis, and changes in the refractive index of cells.
Cell wall-targeting antibiotics are most effective against actively growing cells (28,29); therefore, increases in growth rates reduce cell survival. To determine whether a growth rate phenotype explains our observations, we normalized data for cell length and were unable to detect a change in the growth rate of swarmer cells under conditions of treatment with these two antibiotics (Fig. S11). We measured the mean survival time of cells (the amount of time elapsed after treatment with drugs before cell death) for P. mirabilis and V. parahaemolyticus swarmer cells treated with cephalexin and penicillin G (Fig. 7B) and found that the survival time for these swarmer cells was lower than that for vegetative cells (Fig. 7B), which is consistent with alterations in the cell wall.

DISCUSSION
P. mirabilis is commonly associated with complicated urinary tract infections and increased mortality in cases of bacteremia (30). Swarming is hypothesized to enable P. mirabilis cell movement from the ureter to the kidney; in support of this hypothesis, swarming-deficient mutants have lower rates of host infection (31). We found that during swarming, P. mirabilis and V. parahaemolyticus cells become more flexible due to changes in the thickness and composition of the PG. Although we were unable to precisely measure the thickness of isolated PG, the decrease in the distance between the inner and outer membranes of swarmers may reflect a reduction in PG thickness. Several recent studies suggested how changes in the cell wall influence cell stiffness. A recent study revealed the connection between changes in the biochemical regulation of D-Ala-D-Ala and Pseudomonas aeruginosa cell stiffness in cells in which this biochem-ical machinery is altered (32). The current study and the P. aeruginosa paper (32) demonstrate that it is still difficult to quantitatively understand the scaling between the magnitude of the biochemical changes in cells and changes in cell stiffness. The range of techniques now available for quantitatively measuring single-cell stiffness will enable the development of this relationship.
We found that P. mirabilis and V. parahaemolyticus swarmer cells were more susceptible to large changes in cell length than vegetative cells during osmotic shift experiments. Although these changes are consistent with what would be expected from a reduction in PG thickness in swarming cells, they are also relevant to other cell wall changes. We also observed that the length of polysaccharides was reduced in swarming cells, which would not obviously lead to large changes in cell elongation during osmotic shifts if the organization of PG in these cells follows what is known about the structure of PG in E. coli cells (33): polysaccharides reported to be arranged circumferentially around cells with the peptide cross-links oriented parallel to the long axis of cells. Instead, we expect that a decrease in polysaccharide length would make swarmer cells more prone to changes in diameter. P. mirabilis swarmer cells appear to increase in diameter during osmotic shifts (with no change occurring for vegetative Survival of cells treated with 1ϫ MIC of cephalexin after 3 h of incubation. We define percent survival as follows: (cell count no lysis /cell count total ) ϫ 100. P. mirabilis and V. parahaemolyticus swarmer cells exhibited levels of survival that were decreased (ϳ30%) compared to those seen with vegetative cells (n ϭ Ն90 cells from at least two independent experiments). A similar decrease occurred when P. mirabilis was treated with penicillin G (n ϭ Ն77 cells from at least two independent experiments). Error bars represent the standard deviations of the means. (B) After exposure to cephalexin or penicillin G, the survival time of P. mirabilis and V. parahaemolyticus swarmers was ϳ2-fold to ϳ3-fold lower than that of vegetative cells. Survival time was determined for Ն49 cells that lysed from at least 2 independent experiments. The statistical significance of the data presented in panels A and B was determined using a two-tailed t test. *, P Յ 0.05; **, P Յ 0.01; ****, P Յ 0.0001.
Bacterial Swarming Reduces Cell Stiffness ® cells); however, the data are noisy and do not accurately fit with standard models. We were unable to observe a trend with V. parahaemolyticus swarmer cells (Fig. 5).
Another recent study demonstrated the role of the outer membrane in the cell stiffness of Gram-negative bacteria (34). Chemical and genetic changes in the outer membrane increased the deformation of the cell envelope in response to stretching, bending, and indentation forces and increased the amount of cell lysis arising during osmotic shock. It is possible that the morphological changes in swarmer cell membranes that we observed during ECT experiments indicate membrane perturbations that contribute significantly to changes in cell stiffness. Swarming in Proteus and Vibrio is well known to occur in parallel with the secretion of lipopolysaccharides (3,4), the source of which is presumably the outer membrane of these bacteria, which may alter cell stiffness.
This paper reports for the first time the ability of bacteria to modulate their cell stiffness (beyond L-forms). This phenotype (i.e., greater flexibility) presumably conveys an adaptive advantage for cells that is supported by observations from many laboratories reporting that the motility of communities of swarmer cells is enhanced when cell-cell contacts are increased. The adaptive advantage of swarming in P. mirabilis and V. parahaemolyticus, however, appears to be offset by a decrease in their fitness, as cells become more sensitive to osmotic changes and cell wall-targeting antibiotics, thereby creating an "Achilles heel" with respect to targeting this phenotype in infectious diseases. This report highlights the plasticity of bacteria (1), the continuing need for methods of evaluating chemotherapies that measure efficacy against cells in a physiological state relevant to specific infections, and the reconsideration of cell walltargeting antibiotics for treating UTIs and other infections in which P. mirabilis and V. parahaemolyticus may be present.
Immunostaining of flagella on E. coli cells to determine flagellar surface density. As a control to determine the influence of the surface density of bacterial flagella on measurements of bending rigidity, we imaged two E. coli MG1655 strains with different densities of flagella using a polyclonal flagellin antibody and an immunostaining procedure for visualization of flagella (10).
Calculating vegetative and swarmer cell division times. We prepared vegetative cells by diluting an overnight culture 1:200 in fresh medium and grew the cells at 30°C with shaking at 200 rpm to an optical density (OD; ϭ 600 nm) of 0.6. We prepared swarmer cells as described previously (10). We monitored the growth of P. mirabilis, V. parahaemolyticus, and E. coli vegetative cells over 120 min at 30°C in the microfluidic flow chamber device described in the next section. We performed a similar experiment with P. mirabilis and V. parahaemolyticus swarmers growing over 65 min at 30°C to determine the amount of time that elapsed before the cells divided. We collected images of cells at 1-min intervals and determined the division time for a maximum of 10 generations (for vegetative cells) and 4 division events (for swarmer cells).
Measuring the sensitivity of cells to osmotic shock in a microfluidic device. We prepared filamentous vegetative cells of P. mirabilis and V. parahaemolyticus by diluting an overnight culture of cells 1:200 into fresh medium. We then grew cultures at 30°C with shaking for 1 h, added aztreonam (MP Biomedicals) to reach a final concentration of 10 g/ml, and grew the cells for an additional 70 min. Swarmer cells were prepared as described previously (10). To maximize the number of cells attached to a surface in the microfluidic device (flow chamber construction described below), we concentrated cells to an OD 600 of 8 in 100 l of liquid nutrient medium.
We prepared the microfluidic device for experiments by flowing 10 l of undiluted Cell-Tak (Corning) into the device and incubating it for 10 min at 25°C. Next, we flowed 20 l of a suspension of cells (OD 600 of 8) through the device and then repeated the procedure with another 20-l aliquot of cell suspension. To aid the adherence of the highly motile swarmer cells to the Cell-Tak-coated surface, we centrifuged the device for 5 min at 300 ϫ g in a centrifuge (Beckman Coulter) equipped with a swinging bucket rotor.
For osmotic shock experiments, we filled separate 1-ml syringes (BD) with a solution of UltraPure distilled water (Invitrogen) or a 1 M concentration of a NaCl solution prepared in double-distilled water (ddH 2 O) and connected the syringes using a three-way valve. We mounted the microfluidic device on a TE2000-E inverted microscope and imaged cells in the chamber at 25°C (decreasing the temperature resulted in reduced growth and cell division). Prior to the osmotic shock step, we flowed 200 l of fresh To determine changes in cell size and width, we extracted individual cell contours from the images taken for each treatment using MATLAB 2014a (MathWorks) and used MicrobeTracker (35) to determine cell lengths (L) and widths (W). From these measurements we calculated ΔL (L hypotonic -L hypertonic ) and ΔW (W hypotonic -W hypertonic ) for vegetative and swarmer cells of P. mirabilis and V. parahaemolyticus.
Determining the MIC of vegetative cells of P. mirabilis and V. parahaemolyticus. We used the microdilution protocol (36) to determine the MIC of cephalexin in accordance with Clinical and Laboratory Standards Institute guidelines. Briefly, we added 400 g/ml of cephalexin (BP Biomedicals) or penicillin G (BP Biomedicals) to the first well of a 96-well microplate (Nunc) and diluted these antibiotics 2-fold across adjacent wells (wells 1 to 11; well 12 was a no-drug control). We determined the MIC after 16 h of growth at 30°C with shaking by identifying by visual inspection the lowest concentration of cephalexin and penicillin G that inhibited cell growth. The MIC was determined from three replicate plates. We determined the MIC of cephalexin against E. coli MG1655 (1ϫ MIC ϭ 6.25 g/ml; 32ϫ MIC ϭ 200 g/ml), P. mirabilis (100 g/ml), and V. parahaemolyticus (50 g/ml). The MIC of penicillin G against P. mirabilis was 12.5 g/ml.
Antibiotic treatment of vegetative and swarmer cells and measurement of their growth in a microfluidic device. We prepared vegetative and swarmer cells of P. mirabilis and V. parahaemolyticus as described above. Prior to use, we diluted cells 1:100 in fresh medium to a cell density that enabled us to image many individual cells simultaneously in the microfluidic device described in the section above.
To prepare the microfluidic device for monitoring growth, we applied a 250-m-thick layer of PDMS prepolymer to the surface of cover glass (Fisherbrand) (no. 1.5; 35 by 50 mm) by the use of a Spin Coater (Laurell Technologies) and polymerized the polymer overnight at 60°C. We subsequently removed a 6-mm-by-4-mm rectangle of PDMS from the center of the cover glass using a scalpel. We applied transparent tape to both sides of this rectangular well and pipetted 100 l of media containing a 2% (wt/vol) solution of UltraPure agarose (Invitrogen) into the well. We placed a cover glass (Fisherbrand) (no. 1.5; 22 by 30 mm) on top of the liquid agarose (to flatten the agarose surface), pressed the cover glass against the transparent tape, and solidified the agarose at 25°C. Next, we removed the no. 1.5 cover glass and tape and any residual agarose from the PDMS surface. We pipetted 2 l of a suspension of cells on the agarose pad surface, waited until the excess liquid had evaporated or been absorbed by the agarose, carefully removed the agarose layer, inverted it, and placed it back into the well such that the cells were positioned between the agarose surface and the cover glass. We placed the PDMS flow chamber used in the construction of the osmotic shock microfluidic device on the PDMS-coated cover glass and aligned it such that the agarose pad was centered in the flow channel (see Fig. S12, available at https://osf.io/hvawp).
For antibiotic treatment of cells in the microfluidic cell growth device, we filled a 6-ml syringe (Norm-Ject) with cephalexin or penicillin G at a concentration corresponding to 1ϫ MIC dissolved in nutrient medium. We supplied a constant flow of 20 l/min to the device using a syringe pump (Harvard Apparatus) and monitored the growth of individual cells during antibiotic exposure using a TE2000-E inverted microscope. The stage and objective heater were maintained at 30°C. Images were collected every 1 min for 3 h.
Due to the aberrant shape of cells treated with cephalexin and penicillin G (Fig. S10), we were unable to use an automated script (e.g., MicrobeTracker) to determine cell death. Instead, we visually determined the time of death for individual cells when they exhibited three phenotypes that collectively indicate cell death: blebbing (membrane swelling), lysis (bleb rupture), and disappearance of cells in phase-contrast microscopy (loss of cytoplasmic material).
Determining cell growth rates in the presence of antibiotics. We prepared vegetative and swarmer cells of P. mirabilis and V. parahaemolyticus as described above. We monitored individual cell growth at 30°C in the presence of 1ϫ MIC of cephalexin and pencillin G in our microfluidic growth device (described above) by collecting an image every 1 min for 15 min. To calculate the growth rate, we first extracted individual cell contours at each time point using MicrobeTracker and determined cell length. To take into account the differences in starting length between P. mirabilis and V. parahaemolyticus cells, we normalized the change in length to the initial cell length (ΔL/L 0 ) at each time point. To determine the growth rate for individual cells, we fitted their relative length over time to an exponential function using GraphPad Prism 6.0 (GraphPad Software).
Measuring cell envelope architecture and peptidoglycan thickness using electron cryotomography. We prepared vegetative and swarmer cells of P. mirabilis and V. parahaemolyticus and concentrated them to an OD 600 of 10. For ECT, we mixed vegetative and swarmer cells with bovine serum albumin-treated 10-nm-diameter gold particles that served as fiducial markers, applied them to electron microscopy grids, and subjected them to plunge-freezing in a mixture of liquid ethane and propane, as described previously (37). Grids were stored in liquid nitrogen until imaging was performed.
We acquired images on a 300-KeV Polara transmission electron microscope (FEI) with a GIF energy filter (Gatan) and a K2 Summit direct detector (Gatan). We collected tilt series from Ϫ60°to ϩ60°with 1°increments using UCSFtomo software with a defocus of Ϫ10 m and a total dosage of 190 e Ϫ /Å 2 (38) at a magnification of ϫ27,500. Tomograms were calculated using IMOD software (39). For subtomogram averaging, smooth and flat membrane regions were chosen by eye; a volume of 40-by-70-by-12 voxels (62 nm by 109 nm by 19 nm) was centered using the outer membrane and extracted. We aligned 38 extracted "membrane fragments" from four P. mirabilis vegetative cells and 42 fragments from nine P. mirabilis swarmer cells and averaged them in PEET (40). The densities from two averaged membranes were scaled to match each other using IMOD (39), cross-averaged density profiles were measured using ImageJ 1.50c (41), and figures were generated in OriginPro (OriginLab).

Bacterial Swarming Reduces Cell Stiffness
Determining the thickness of P. mirabilis and V. parahaemolyticus sacculi (isolated peptidoglycan) by atomic force microscopy (AFM) in ambient conditions. We isolated P. mirabilis and V. parahaemolyticus swarmer cells, concentrated them at 800 ϫ g for 10 min, removed the supernatant, subjected the cell pellet to flash-freezing in liquid nitrogen, and stored it at -80°C. Swarmer cell pellets were thawed at 4°C and pooled for isolation of sacculi (intact peptidoglycan). To increase the efficiency of cell lysis prior to isolating sacculi, we resuspended vegetative and swarmer cell pellets in 3 ml of cold 1ϫ phosphate-buffered saline (Thermo Scientific) and then lysed cells with a tip sonicator (Qsonica) for ϳ10 s at a power setting of 75%. We confirmed cell lysis using optical microscopy. We isolated sonicated cells, resuspended sacculi in 20 l of ddH 2 O (V. parahaemolyticus swarmers) or 200 l of ddH 2 O (all other cells), immediately subjected the sacculi to flash-freezing in liquid nitrogen, and stored them at -80°C.
To prepare sacculi for AFM, we transferred 10 l of sacculi thawed at 4°C to a new microcentrifuge tube placed in a bath sonicator (Branson) that was cooled with ice for 10 min to aid in the dispersal of sacculi without affecting the sacculus architecture (42). After sonication, we pipetted 10 l of the sacculi onto freshly cleaved mica (Ted Pella), dried the mica under nitrogen gas, washed it three times with 1 ml of ddH 2 O (filtered through a 0.2-m-pore-diameter filter [Corning]), dried the sacculi under nitrogen gas, and performed imaging immediately after preparation.
We performed AFM using a Catalyst AFM (Bruker) operating in tapping mode under ambient conditions (air) with an aluminum reflex-coated silicon AFM probe (Ted Pella; k ϭ 40 N/m). Before imaging, AFM probes were autotuned using Nanoscope 8.15 (Bruker). We collected all images at high resolution (512 by 512 pixels) with a scan speed of 1 Hz and analyzed images using NanoScope Analysis 1.4 (Bruker). Prior to determining the thickness of the sacculi, we flattened (0th order) all images to remove variations in surface thickness. Thickness was determined perpendicularly to the long axis; we avoided surface debris, folds in the sacculi, and trapped material in sacculi (Fig. S7).
Determining the composition of peptidoglycan in P. mirabilis and V. parahaemolyticus vegetative and swarmer cells using ultra-high-performance liquid chromatography/mass spectrometry (UPLC/MS). We prepared vegetative and swarmer cells of P. mirabilis and V. parahaemolyticus as described above. To purify peptidoglycan for UPLC/MS, we carried out a previously reported isolation technique for Gram-negative bacteria (43) with the following modifications. Briefly, after trypsin (Sigma) inactivation, we incubated the sacculi in 1 M HCl solution (Fluka) for 4 h at 37°C to remove any O-acetylation from the peptidoglycan, which is present in P. mirabilis (23). Then, the sacculi were washed three times in ddH 2 O, resuspended in 500 mM boric acid (Sigma) [pH 9] to an OD 600 of 3, and mixed with 1/10 the volume of mutanolysin (Sigma). The sample was incubated 16 h at 37°C with shaking at 200 rpm. The next day, the samples were centrifuged for 10 min at 9,500 ϫ g, pelleting the remaining insoluble material. The supernatant was removed and put into a glass vial. To reduce the isolated muropeptide fragments, we added 50 l of 20 g/ml sodium borohydride (Sigma)-500 mM boric acid (pH 9) and incubated the mixture for 30 min at 25°C. We adjusted the pH of the solution to 2 to 3 by adding 50% phosphoric acid (Fluka) and then filtered the muropeptide solution through a Durapore polyvinylidene fluoride filter (Millex) (0.22-m pores) into a clean vial. Vials were immediately stored at -80°C until use within 1 week of muropeptide isolation.
For UPLC/MS, we injected 7.5 l of purified muropeptides into a Cortecs C 18 column (Waters) (2.1 by 100 mm) packed with 1.6-M-diameter particles and equipped with a Cortecs C 18 guard column (Waters). The column temperature was maintained at 52°C using an Acquity standard flow UPLC system equipped with an inline photodiode array (Waters). For muropeptide separation by UPLC, we used solvent A (Optima liquid chromatography-mass spectrometry [LCMS]-grade water with 0.05% trifluoroacetic acid) and solvent B (30% [vol/vol] Optima LCMS-grade methanol, Optima LCMS-grade water, and 0.05% trifluoroacetic acid) (Fisher Scientific). Muropeptides were eluted from the column with a gradient of increasing levels of solvent B (1-min hold at 1% solvent B, ramp to 99% solvent B over 59 min, hold at 99% solvent B for 5 min, followed by a decrease to 1% solvent B over 1.5 min and then a hold at 1% solvent B for 4.5 min) at a flow rate of 0.2 ml/min. We analyzed the eluent from the column using a Bruker MaXis ultra-high-resolution time-of-flight 4G mass spectrometer (Bruker Daltonic) with either an MS method or a data-dependent, top 3 MS/MS method. For both methods, capillary voltage was set to 4,100 V, the nebulizer pressure was 2.0 bar, and the drying gas was set to 6.0 liters/min at 220°C. Muropeptides were detected at a level of 205 nm by MS.
Fabrication of microfluidic devices for measuring cell bending and cell growth. Masters for the cell-bending device and the cell growth/osmotic shock device were fabricated on separate 3-in 2 silicon wafers using SU8 photoresist that was exposed on a Heidelberg PG101 mask writer (Heidelberg Instruments, Heidelberg, Germany) and developed. The bending device (Fig. S1) used was a 2-layer device. The first SU8 layer consisted of ϳ1-m-tall channels for capturing cells. The wafer was then coated with the second SU8 layer (ϳ25 m tall), which formed a central flow channel. The flow chamber (Fig. S12) consisted of a single channel with a length of 10 mm, a width of 5 mm, and a height of 50 m. The volume of the flow chamber was ϳ10 l. After developing the masters for both devices, we used the masters to emboss layers of poly(dimethylsiloxane) (PDMS), punched holes for inlets and outlets, and cleaned the surfaces with tape. If the device was attached to a glass coverslip, it was treated with oxygen plasma and immediately sealed against a plasma-treated no. 1.5 cover glass (Fisherbrand) (24 by 50 mm) to form a permanent seal.
Measuring the bending rigidity of P. mirabilis, V. parahaemolyticus, and E. coli cells. We used streak velocimetry to determine the profile of fluid flow rates in the central channel of the microfluidic device driven by gravity flow. We added fluorescently labeled 0.22-m-diameter microspheres (Polysciences) diluted ϳ1:10,000 in ddH 2 O containing 0.01% Brij 35 (Sigma) to the microfluidic channel and collected videos of the fluorescent beads moving through the channel at focal planes 2 m apart. We analyzed the movies with custom-written code in Igor Pro 6.37 (WaveMetrics). Briefly, we applied a Gaussian blur and threshold to each frame and used the thresholded image to establish a region that was fitted to a function based on a two-dimensional (2D) Gaussian function. We used the image exposure time and the length of the streaks-taking into account the size of the microspheres-to calculate microsphere velocity. We mapped the velocity profile within the channel by analyzing several hundred microspheres. We binned the velocity data into a 3D matrix and fitted it to a Poiseuille function, letting the velocity coefficient, height, and width float. The calculated velocity coefficient was used as an input to the cell-bending fitting function.
The system for gravity flow pumping (Fig. S13) consisted of two 6-ml syringes (Norm-Ject), one 60-ml syringe (BD), one 1-ml syringe (Norm-Ject), a ruler, and a VC-6 channel valve controller (Warner Instruments) connected to a VC-8 minivalve system (Warner Instruments) that drives a three-way solenoid valve (The Lee Company). Syringe no. 1 (6 ml) was mounted to an immobile post, positioned 40 cm above the table surface; this syringe was used to load cells. Syringe no. 2 (6 ml) was positioned 75 cm above the table surface, was connected to a stage that could be raised and lowered vertically, and was used to apply flow force in the device. Syringe no. 3 (60 ml) was mounted to an immobile post, was positioned 75 cm above the surface of the table, and was used to apply an opposing flow force at the device outlet and to collect spent media/cells. We attached two-way valves, a blunt-end needle, and tubing to each syringe, including syringe no. 4 (1 ml). We joined the two 6-ml syringes using a Y-junction connector that led to the VC-8 minivalve system inlet. The outlet of the valve system was connected to the microfluidic system. A ruler on an immobile post indicated the zero position (no pressure).
For cell-bending experiments, we prepared filamentous vegetative cells and swarmer cells of P. mirabilis and V. parahaemolyticus; cells were normalized to an optical density (OD; ϭ 600 nm) of 1. Prior to starting a measurement, syringes 1 and 2 were flushed with ddH 2 O and fresh medium. Syringe 2 was filled with 4 ml of medium, syringe 3 was filled with 30 ml of ddH 2 O, and syringe 4 was filled with 0.8 ml of ddH 2 O. We added cells to syringe 1 and allowed them to flow them through the tubing until they reached the device outlet. To load the cells into the side channels of the device (from the central channel), we applied a suction force using the 1-ml syringe. After the side channels were loaded with cells, we adjusted the height of syringe 2 until no flow occurred in the device. Syringe 2 was then raised 7 cm from the no-flow position (a height of 0 cm). We collected images of cells in the channel when syringe 2 was at positions 0 cm and 7 cm using a Zeiss Axiovert 100 inverted microscope (Zeiss) equipped with an iXon3 charge-coupled-device (CCD) camera (Andor), a 63ϫ Plan-Apochromat oil objective (Zeiss), and Micro-Manager 1.4.16. After collecting cell-bending deflections for all loaded cells, we expelled cells from the side channels, allowed the liquid to flow through the device for 15 s, and reloaded the side channels with new cells.
We analyzed images to determine cell deflection under flow conditions using custom image analysis software written in Igor Pro 6.37. The cell-bending model (see the supplemental material) is a differential equation that lacks an analytical solution and thus requires calculation of a numerical solution. To determine the bending rigidity of cells, we wrote custom fitting code in Igor Pro 6.37 that uses a variety of input parameters, including channel dimensions, fluid velocity, bending rigidity of the cell, cell radius, and cell length. The function numerically calculates the maximum deflection based on our model. This function is integrated into a fitting algorithm to find a least-squares solution to a data set of maximum deflections versus cell lengths with bending rigidity as the fitting parameter.
Our model for bending a cell under flow conditions is based on the mechanics model of a suspended rod or cantilever bending under its own weight (see "Derivations" in the supplemental material). Our experimental system was similar to that of Amir et al. (18). Although the laminar flow is perpendicular to the long axis of cells in this system, the lateral deformations of cells that we measured were substantially larger (in our case, as large as 10 m compared to the value of Ͻ1 m reported previously by Amir et al. [18]). For this reason, many of the assumptions in the model presented by Amir et al. are not valid for our data set. The model that we developed to extract the bending rigidity of the bacterium takes into account the shape of the laminar flow profile, the angle of the cell against the flow profile, and the arc length of the cell (longer cells tend to fold over and do not penetrate as deeply into the flow profile). Additional data and mathematical derivations are described in the supplemental material.