Single-cell twitching chemotaxis in developing biofilms

Significance

Bacterial biofilms affect many aspects of our lives, from causing disease to promoting health and shaping many key processes in the environment. Despite this, surface-attached cells in biofilms are often portrayed as static and sluggish, a stark contrast to the energetic swimming they exhibit in liquid. Here we use microfluidic devices and automated cell tracking to challenge this view: We find that individual cells will actively move toward nutrients within a developing biofilm. This ability not only allows cells to seek out favored positions on a surface but our analyses show that they can regulate their movement with remarkable submicron precision. Our findings suggest we can systematically engineer biofilms by manipulating the movement of the cells from which they are founded.

Abstract

Bacteria form surface-attached communities, known as biofilms, which are central to bacterial biology and how they affect us. Although surface-attached bacteria often experience strong chemical gradients, it remains unclear whether single cells can effectively perform chemotaxis on surfaces. Here we use microfluidic chemical gradients and massively parallel automated tracking to study the behavior of the pathogen Pseudomonas aeruginosa during early biofilm development. We show that individual cells can efficiently move toward chemoattractants using pili-based “twitching” motility and the Chp chemosensory system. Moreover, we discovered the behavioral mechanism underlying this surface chemotaxis: Cells reverse direction more frequently when moving away from chemoattractant sources. These corrective maneuvers are triggered rapidly, typically before a wayward cell has ventured a fraction of a micron. Our work shows that single bacteria can direct their motion with submicron precision and reveals the hidden potential for chemotaxis within bacterial biofilms.

Flagella-driven chemotaxis in bacteria has been thoroughly dissected (1–4), leading to its emergence as a paradigm of both signal transduction and cellular decision-making (4, 5). However, many key phenotypes of bacteria occur when cells are not swimming. In particular, planktonic cells commonly attach to surfaces and form communities, known as biofilms, which are central to health, disease, agriculture, industry, and the environment (6, 7). Biofilms often contain steep chemical gradients that result from cell metabolism, hampered diffusion, and the secretion of a wide variety of compounds (8). Although attached bacteria can be highly motile (9–11), remarkably little is known about the potential for these cells to respond to their chemical environment.

Surface-based movement has been studied in both Myxococcus xanthus (12, 13) and Pseudomonas aeruginosa (14–16) by exposing agar-based colonies to gradients of phospholipids and unsaturated long-chain fatty acids. Over time these colonies develop a bulge toward the chemoattractant source. However, the dense packing of cells within these assays makes it difficult to resolve the cause: asymmetric colonies can form because motility is simply enhanced on the side of the colony where chemoattractants are plentiful (chemokinesis), or instead because cells are actively biasing their motility up the chemical gradient (chemotaxis) (14). Moreover, in sharp contrast to swimming bacteria, experiments with M. xanthus have suggested that solitary surface-attached bacteria are incapable of biasing their motility along chemical gradients (13, 17). We, therefore, designed a new assay to test whether individual cells can perform chemotaxis on surfaces (SI Appendix, SI Materials and Methods). Our experiments reveal that not only are single cells capable of chemotaxis, but they can control their position with a striking level of precision.

Results and Discussion

Individual Bacteria Can Navigate Chemical Gradients on Surfaces.

P. aeruginosa is an intensely studied opportunistic pathogen and a canonical model for the study of biofilms (18). After attaching to surfaces, P. aeruginosa cells are highly motile and move by pulling themselves along via the extension and retraction of their type IV pili, a process known as twitching motility (19). Although this form of movement is common throughout biofilm formation (10, 11), here we follow the movement of solitary bacteria in the early stages of biofilm development so that we can readily calculate each cell's chemical environment and resolve how it modifies their behavior. Importantly, the microfluidic assays used here are analogous to those used in classical studies of biofilm development (9–11), and the cells whose movement we analyze subsequently form 3D biofilm structures (Fig. 1 F and G and SI Appendix, Fig. S1). To generate stable chemical gradients, we use two inlet microfluidic devices where flow balances the smoothing effect of molecular diffusion (Fig. 1A and SI Appendix, Fig. S2 and SI Materials and Methods). Many bacterial chemoattractants strongly promote growth (20), which in our experiments leads to a crowded surface and a limited ability to analyze single cell behavior. We therefore began our experiments with DMSO, which is a known chemoeffector of bacteria (21–23) and, importantly, does not strongly affect growth in our experiments.

Fig. 1.

Surface-attached P. aeruginosa cells direct their movement up gradients of DMSO during biofilm formation. (A) A two-inlet microfluidic device generates a stable chemical gradient via molecular diffusion. TB is continuously injected through one of the inlets, whereas DMSO at a concentration of C_MAX = 350 mM and TB are continuously injected through the other inlet. Here C = C′/C_MAX, where C′ is the local concentration of DMSO that was quantified using fluorescein in a separate experiment (SI Appendix, Fig. S2 and SI Materials and Methods). (B) Cell trajectories over the first 5 h of the experiment (t = 0–5 h) show that motility is biased toward increasing concentrations of DMSO. Trajectories with net movement toward larger C (i.e., in –y) are shown in red and those with net movement toward smaller C (i.e., in +y) are shown in blue. The background shows C computed by a mathematical model (SI Appendix, Fig. S2). Inset shows probability density functions of the angle from each trajectory’s origin to final position and red (blue) bins denote movement in the –y (+y) direction (SI Appendix, SI Materials and Methods). This Inset also shows slight preferential movement in the +x direction, which likely results from interactions with flow (39). (C) Control shows there is no bias in the y direction without DMSO. (D and E) A time series of the chemotactic bias, β, defined as the number of cells that move in –y divided by the number moving in +y, peaks at t ≈ 2 h (D; black line; gray line shows control) and then declines as the surface becomes more crowded and cell speed is attenuated (E). (F) A kymograph of the fluorescent intensity, F, of cells that constitutively produce yellow fluorescent protein (SI Appendix, SI Materials and Methods) show how biofilms develop over time in our experiments. Here we average the fluorescent intensity in x, allowing us to continuously resolve both the chemotaxis of cells (light blue streaks moving toward –y) and formation of densely packed biofilms (red). As biofilms produce a much brighter fluorescent signal than single cells we plot the logarithm of F so that both can be visualized. SI Appendix, Fig. S1 shows the raw images from which this kymograph was constructed. (G) A 3D confocal micrograph of the biofilm shown in F, imaged at t = 20 h, shows the biofilm is already many cells thick, extending ≈20 µm from the surface. Dashed lines indicate the positions of the vertical cross sections.

We used automated cell tracking to follow surface-attached P. aeruginosa cells that are exposed to a stable spatial gradient of DMSO (Fig. 1A and SI Appendix, Fig. S2). Our method allows us to follow large numbers of attached cells and quantify their individual responses to the presence, or absence, of a chemical gradient (SI Appendix, Fig. S3). In a DMSO gradient, we found that cell movement is strongly biased in the direction of increasing DMSO concentration (Fig. 1B and Movie S1), which contrasted with random motility in the absence of chemical gradients (Fig. 1 C and D). After 2 h of incubation, the chemotactic bias peaks, with more than three times as many cells moving toward the chemoattractant source than moving away from it (β = 3.1), but biased motility is maintained even as the surface becomes more crowded with cells (Fig. 1 D and E). These data show single cells direct their motility along chemical gradients and suggest that surface-attached cells are capable of chemotaxis [sensu Adler (1)].

In our experiments, both cell division and motility cause cells to preferentially accumulate on the side of the device with the chemoattractant (Fig. 1F and Movie S1). To test whether the biased motility we observe is an artifact of the variation in cell density, which has been shown coordinate surface motility in M. xanthus (24), we exposed cells to a chemoattractant gradient that switched direction every 3 h. By minimizing crowding on one side of the channel, this assay also allows us to study the responses of individual cells to nutrients like succinate, which is a preferred carbon source of P. aeruginosa and a potent chemoattractant of cells in the planktonic state (25). When the chemoattractant gradient was inverted, cells responded by changing the direction of their bias to track the gradient (Fig. 2 A–G). The refinement with which individual cells dynamically track the gradient can be observed in Movie S2. Our data show then that cells are directly responding to the imposed gradient rather than to de novo gradients generated by the cells. Moreover, we observe these responses for gradients of both DMSO and succinate, the latter of which strongly stimulates both growth and cell motility (SI Appendix, Fig. S4). Finally, the increased data available from this second assay reveals that the chemotactic bias, β, increases with the magnitude of the chemoattractant gradient |G| (Fig. 2H), which is a defining feature of chemotaxis across different biological systems (26). Based on the responses seen in our two single-cell assays (Figs. 1 and 2), we conclude that individual surface-attached bacteria are indeed capable of chemotaxis.

Fig. 2.

Changing the direction of the chemoattractant gradient elicits a change in the direction of cell motility and reveals the chemotactic response increases with the strength of the gradient. (A–C) Cells (white spots) were exposed to gradient of succinate (visualized in yellow, C_MAX = 2 mM) whose direction was inverted every 3 h. (D–F) The resulting chemotactic response was measured by calculating the direction of cell movement, θ, over 16-min intervals, and segments of trajectories that moved in the –y (+y) direction are shown in red (blue). (G) The mean movement direction, $\overline{\theta }$, was obtained by averaging θ for all cells (magenta line), revealing that cell motility is maintained in the direction of the instantaneous chemoattractant gradient (yellow regions). (H) Combining data from the entire experiment (t = 0–9 h) reveals that the chemotactic bias, β, increases with the magnitude of the normalized gradient, |G| = 1/C_MAX |∂C/∂y| (magenta line). Here β is the number of trajectories moving in the direction of increasing concentration C, divided by the number moving in the direction of decreasing C. Grayscale colors in D–F show |G| and correspond to the bins used in the analysis presented in H. Green lines in G and H show results of an analogous experiment conducted with a DMSO gradient (C_MAX = 350 mM). In both experiments, the position of chemoattractant was visualized using a dye that does not induce a chemotactic response (SI Appendix, Fig. S15), and |G| was estimated using a mathematical model of diffusion (SI Appendix, SI Materials and Methods).

Attached P. aeruginosa Use Pili-Based Motility and the Chp Chemosensory System to Perform Chemotaxis.

What is the genetic basis of the chemotactic response we documented above? After attachment, P. aeruginosa is well known to use pili-based motility to move on surfaces (9–11, 18, 19). Accordingly, a mutant that cannot generate functional pili (ΔpilB) is immobile in our experiments (Fig. 3 A and C), whereas a mutant that lacks flagella (ΔflgK) exhibits surface motility (SI Appendix, Fig. S5). We find flagella mutant cells stand upright and show increased motility relative to WT, which is in agreement with a previous study (27), and suggests that flagella may function as a stabilizing anchor during twitching motility. To explore if the flagella could play a role in chemotaxis, we studied the behavior of mutant that has a flagellum but lacks the ability to perform flagella-based chemotaxis. Chemotaxis in swimming P. aeruginosa cells is under the regulation of a Che transduction system homologous to that of Escherichia coli (16, 28), and mutants in CheY1, the key response regulator, can swim but cannot actively bias their movement along chemical gradients. In contrast to swimming, however, we find that the swimming chemotaxis cluster (Che cluster I) is not required for chemotaxis on surfaces because CheY1 mutants perform twitching chemotaxis (SI Appendix, Fig. S6).

Fig. 3.

Twitching chemotaxis in P. aeruginosa is mediated by the Chp signal transduction pathway. (A) Trajectories of WT and different Chp mutants in the absence of a chemical gradient where color denotes instantaneous cell speed. (B) Simplified diagram of Chp chemosensory system. By homology to the Che system of E. coli, it is thought that the protein kinase ChpA drives phosphorylation of two CheY-like response regulator proteins, PilG and PilH. These response regulators interact with the ATPases PilB and PilT to drive pilus extension and retraction, respectively, in a manner similar to how phosphorylated CheY controls flagellar rotation in swimming cells (29). Black circles denote phosphate groups. (C) RMSD of different strains in the absence of a chemical gradient as a function of time lag, τ. Analysis of a large number of cells (SI Appendix, Fig. S16A) reveals ΔpilG cells move similarly to the WT, ΔpilH cells move much faster (note separate color bar in A), ΔchpA cells move at an attenuated rate, and ΔpilB cells lack motility altogether (also see SI Appendix, Fig. S16B). (D) Trajectories of ΔpilG and its respective WT in a DMSO gradient reveals that PilG plays an essential role in twitching chemotaxis. Here trajectories with net movement toward larger C are shown in red and those with net movement toward smaller C are shown in blue. (E) These findings are confirmed by measurements of the chemotactic bias, β, which show quantitatively that ΔpilG cells lack the ability to perform chemotaxis. Data used to compose A and C for ΔpilH cells were collected in the first 15 min after inoculation as they move so quickly, whereas data from the remainder of strains in these panels was collected over the first 10 h. D shows trajectories over the first 8 h of the experiment, when chemotaxis is most pronounced (E).

Recent work has shown that twitching motility in P. aeruginosa is under the regulation of a separate signal-transduction pathway (Che cluster IV) known as the Chp system (16, 29) (Fig. 3B), but to date, there is no direct evidence that this system plays an active role in any form of chemotaxis. To examine the effects of the Chp system, we first recapitulated the motility phenotypes recently described using classical agar-based assays (29). Specifically, in-frame deletion mutants ΔpilB, ΔchpA, and ΔpilG do not form the characteristic twitching rings at the plastic–agar interface around colonies, whereas ΔpilH cells exhibit an intermediate phenotype by producing rings that are smaller than that of the WT (SI Appendix, Fig. S7). However, observing the movement of individual cells in our microfluidic system reveals a very different pattern. We find that ΔchpA and ΔpilG cells, which were previously diagnosed as incapable of twitching (29), are in fact both motile (Fig. 3 A and C). While ΔpilH cells show reduced twitching rings on agar plates (SI Appendix, Fig. S7), in our microfluidic assays they actually move ≈30 times faster than the WT (Fig. 3 A and C) and tend to orient themselves vertically on the surface (SI Appendix, Fig. S8). More formally, the root mean squared displacement (RMSD), a combined measure of both movement speed and the persistence in movement direction (30), of the WT and ΔpilG cells are nearly identical, but show strong differences with that from ΔchpA and ΔpilH (Fig. 3C).

The Chp system then has strong effects on twitching motility in our assay but is it involved in chemotaxis? To answer this question, we focused on the response regulator of the Chp system PilG, which controls pili extension (29). We focused on PilG because the motility of ΔpilG is similar to the WT in our microfluidic assay, and we can study the effects on chemotaxis in the absence of strong effects on motility (Fig. 3 A and C and Movies S3 and S4). We exposed ΔpilG cells to a stable DMSO gradient and used our analysis pipeline to track individual cells. The data reveal that, although ΔpilG cells indeed remain motile, they lack the ability to bias their motion up the gradient (Fig. 3 D and E). Our results then suggest that the Chp system is not just involved in the biosynthesis of type IV pili but also in the transduction of chemotactic stimuli into directed movement.

P. aeruginosa Uses a “Pessimistic” Chemotactic Strategy on Surfaces.

We next sought to understand how P. aeruginosa cells bias their movement on surfaces. Chemotaxis in swimming E. coli cells is achieved by cells performing straight runs interspersed by sharp reorientations (tumbles), where tumbles are delayed when moving up a chemoattractant gradient (2, 3). Do twitching bacteria use similar movement strategies in our experiments? Although twitching motility gets its name from the jerky motion that cells exhibit over the timescale of minutes (31), we find that in chemical gradients cells can maintain a consistent movement direction for periods longer than 1 h (Fig. 4A). Moreover, these twitching runs are interspersed by events where a cell reverses by stopping and then moving back in the opposite direction without turning. Twitching P. aeruginosa cells pull themselves along surfaces with pili that cluster at their poles (32, 33), which drives movement parallel to their long axis (SI Appendix, Fig. S9) and allows rapid changes in direction if a cell changes the pole that it is pulling from. To follow this process, we developed an automated algorithm that detects when the cell’s movement switches direction such that a cell’s leading pole becomes its trailing pole. Importantly, we distinguish these active reversals from changes in direction that occur passively due to cell division (Fig. 4 A and B, Movies S3 and S4, and SI Appendix, SI Materials and Methods).

Fig. 4.

Twitching bacteria chemotax by reversing their motility when traveling away from the chemoattractant source. (A) Cells navigate up DMSO gradients by actively performing reversals (magenta squares), although sharp changes in direction can also occur passively during cell division (green circles). Red dots in A show positions of cell centroids at 1-min intervals; the black lines show smoothed trajectories, and the average cell length is shown for reference. (B) Magnified view of a reversal shows a cell whose movement begins to veer toward decreasing C. The cell quickly performs a reversal sending it back up the DMSO gradient. Ellipses (not drawn to scale) show cell orientation and position in 2-min intervals. Contours show modeled DMSO concentration in C_MAX/30 increments where C_MAX = 350 mM. (C) In both the absence and presence of a DMSO gradient ΔpilG cells actively reverse at a much smaller rate than WT cells. For experiments with a DMSO gradient, we calculated the frequency that cells moving toward smaller C reverse direction (labeled correct, white bars) and the frequency that cells moving toward larger C reverse direction (labeled incorrect, black bars). Although the rate of incorrect reversals in the WT is similar to that in the absence of a gradient (gray bar), correct reversals occur more frequently, suggesting that reversals are deployed as a corrective strategy. (D–G) Using a custom microfluidic device (SI Appendix, Fig. S11), we induced reversals by exposing cells to a DMSO gradient that changed direction every eight minutes. Owing to a large flow velocity, C and its gradient G = 1/C_MAX ∂C/∂y were nearly constant along our field of view, allowing us to average them along x and present their spatiotemporal variations using kymographs (D and E) (SI Appendix, SI Materials and Methods). Cell reversals were identified using an automated algorithm and classified as correct (white squares, D) or incorrect (black squares, D). Both types of reversals increased over time as cell division and attachment increased the number of cells on the surface (D). Counting the number of correct and incorrect reversals within equally spaced bins in y reveals that correct reversals (white bars, F) peak along the middle of the device, whereas the number of incorrect reversals (black bars, F) occurs more uniformly in y. The reaction time, t_E, is the time elapsed from the most recent gradient change (red + symbols in E and F) to the time of a correct reversal. A histogram of t_E for cells in middle 100 µm of the device (dashed gray lines; D and E) shows that correct reversals peak ≈1 min after the gradient changes direction (G). Over this period most cells travel a distance along the gradient, y_E, less than 1 µm (G, Inset).

Both ΔpilG and WT cells were found to actively reverse their direction in both the presence and absence of DMSO gradients. However, ΔpilG cells, which do not show chemotaxis (Fig. 3 D and E), reversed at a much lower rate than WT cells in both of these conditions, suggesting that the reversals are indeed important for chemotaxis (Fig. 4C). To examine the potential link between reversal rate and the ability to perform chemotaxis, we analyzed how the reversal rate changes as a function of the direction that cells are moving. This analysis revealed that ΔpilG cells reverse at nearly the same rate whether they are moving toward or away from the chemoattractant (Fig. 4C). In sharp contrast, WT cells almost double their reversal rate when moving away from the source of a chemoattractant but keep their basal reversal rate when moving toward the source (Fig. 4C). This observation is consistent with a “pessimistic” chemotactic strategy where cells respond to a reduction in the chemoattractant concentration (23), which contrasts with the “optimistic” strategy used by swimming E. coli that chemotax by postponing tumbles when moving toward a chemoattractant source (2, 3). Put another way, swimming E. coli respond “if life gets better” (2) and our data show that twitching P. aeruginosa respond if life gets worse. We also find asymmetries in speed: WT cells move ≈25% faster when traveling toward the source compared with away from it (SI Appendix, Fig. S10). In contrast, ΔpilG cells moving toward and away from the chemoattractant source travel at approximately the same rate (SI Appendix, Fig. S10). Taken together, our data suggest that the response regulator PilG enables the movement of surface-attached bacteria up chemical gradients by modulating both reversal rate and cell speed.

Surface-Attached Bacteria Can Control Their Movements with Submicron Precision.

Swimming P. aeruginosa cells navigate chemical gradients by selectively rotating their flagella in the opposite direction, which sharply changes their direction of movement (5). In contrast, we show that surface-attached bacteria can climb chemical gradients by regulating when they switch directions by pulling themselves using pili attached to the previously trailing pole. Our observations suggest that this system is extremely effective at guiding attached cells toward larger concentrations of chemoattractants (Movie S2). However, just how responsive are individual attached cells to changes in their chemical environment? To probe the limits of the cells’ abilities, we designed and built a microfluidic device that allows us to continuously track cells as they respond to a DMSO gradient that changes direction every eight minutes (Fig. 4 D and E and SI Appendix, Fig. S11). Direct observation suggests that twitching P. aeruginosa cells accurately reverse their movement in response to the alternating gradient (Movie S5), and we quantified their response by categorizing all reversal events as correct or incorrect. Correct reversals occurred in cells moving away from the chemoattractant source and incorrect reversals occurred in cells moving toward the source (Fig. 4 D and E and Movie S6). We find that correct reversals are stimulated by the alternating gradient (rather than incorrect reversals being repressed), which is again consistent with a pessimistic chemotactic response (Fig. 4F). In addition, we find that WT cells actively increase their speed after performing correct reversals but not after performing incorrect reversals (SI Appendix, Fig. S12). In contrast, cells lacking PilG reversed very rarely in the alternating chemical gradient (SI Appendix, Fig. S13). Taken together, these observations show that attached P. aeruginosa chemotax by deploying reversals when their motility is directed away from the source of a chemoattractant. Correct reversals occurred most frequently in the middle of the device, where cells are exposed to the strongest spatial gradients in DMSO (Fig. 4 D and F) and these reversals peak t_E ≈ 1 min after the gradient changes direction (Fig. 4G). The rapid response means that the majority of cells sense the gradient has changed direction and respond before moving one-fifth the length of their bodies (SI Appendix, Fig. S14), a distance smaller than a micrometer (Fig. 4G, Inset). This observation reveals that attached P. aeruginosa can regulate their movement in a chemical gradient with submicron precision.

Swimming bacteria sense chemoattractant gradients by measuring changes in chemical concentrations over time, which is consistent with them moving rapidly and thus being able to detect large changes over short time periods (2, 34). However, the slow speed and highly unsteady, oscillatory movement of twitching cells (31) begs the question of whether they too use temporal sensing. If twitching cells do use temporal sensing, it would require the integration of information over much longer time-scales than swimming bacteria. Moreover, a twitching cell would have to extract the slow, weak changes in concentration due to their average movement from the large, high-frequency changes in concentration that arise from a cell jerking back and forth relative to a chemical gradient. An intriguing alternative is that twitching cells detect gradients spatially by directly measuring changes in concentration across the length of their bodies. Indeed, in our experiments (Fig. 4 D–G), the majority of responding cells experience a fivefold larger change in chemoattractant concentration across their length than they experience over time due to their movement relative to the gradient (SI Appendix, Fig. S14). Although the fast movement of swimming cells allow them to measure changes in concentration over distances equivalent to tens of their body lengths (2), the slow movement of twitching cells suggests that they could collect more reliable information by making spatial measurements, over the length of their body. In either case, our data suggest that the molecular mechanisms underlying twitching chemotaxis have very different properties than the canonical mechanisms so intensively studied in swimming cells.

Conclusion

Bacteria often live attached to surfaces where multiple strains and species meet and interact (35). These cells strongly alter their environment by secreting a wide range of compounds and metabolizing others to create a diverse and changing set of chemical gradients (8). The resulting chemical gradients can be very steep, stable, and important for the fitness of cells that lie in different positions within a community (36). Here we showed that single attached bacteria can respond to chemical gradients. Rather like ants moving through a nest, we find that twitching cells are able to act as individuals that navigate their way through their chemical and biological environment. We also show that these surface-attached cells perform chemotaxis on spatial scales much finer than swimming cells, with corrective maneuvers occurring before a cell has moved a small fraction of their body length. The discovery that bacteria can navigate their chemical environment with submicron precision has implications for both the biology of bacterial communities and how we manipulate them.

Materials and Methods

Our experiments use WT P. aeruginosa PAO1 (Kolter collection, ZK2019) as a working model for twitching chemotaxis (Figs. 1 and 2 and SI Appendix, Figs. S3, S9, and S15). The YFP-labeled strain used in Fig. 1 F and G and SI Appendix, Fig. S1 is a PAO1 WT strain with a ZK2019 background (37). ΔflgK (SI Appendix, Fig. S5), which has been published and described before (38), is a deletion strain lacking the hook filament junction protein FlgK. To study the functional role of the Chp chemosensory system, we used in-frame deletion mutants of pilB, chpA, pilG, and pilH, along their respective WT, which have all been published and described elsewhere (29). These strains are used in Figs. 3 and 4 and SI Appendix, Figs. S4, S7, S8, S10, S12–S14, and S16. The ΔcheY1 mutant strain shown in SI Appendix, Fig. S6 was a gift from the group of Caroline Harwood, University of Washington, Seattle. All strains were grown in shaken culture overnight in LB (37 °C) from frozen stocks and subcultured to obtain cells in exponential phase. These were then diluted to an optical density of 0.25 (at 600 nm) in tryptone broth (TB; 10 g Bacto tryptone/1 L water) before they were injected into our microfluidic devices.