Introduction

In nature, bacteria often grow on a surface and develop a multi-cellular community [1, 2]. When a community becomes densely populated, cells migrate away to expand their community [3]. This process poses significant industrial and medical challenges because it leads to the spread of biofouling and chronic infections [4,5,6]. However, several fundamental questions remain unanswered. Specifically, dispersal is mediated by phenotypic (or genotypic) “variants” [7]. What are the ecological factors that drive the generation of these migratory variants? What are the signal transduction pathways that are used to sense these ecological factors and trigger the generation of migratory variants?

The importance of understanding migration extends beyond microorganisms because migration is a fundamental process of living organisms [8, 9]. Migration results in a profound change in population density in space and time, altering local ecological interactions, e.g., species competition and spread of invasive species [10,11,12,13,14,15]. A wide variety of organisms exhibit partial migration, where a fraction of a population moves away while the rest remains sessile. This type of migration has been documented for mammals, birds, fish, amphibians, and insects [16]. However, how and why a population diversifies when expanding its territory is poorly understood. While several theoretical studies have proposed potential benefits of partial migration [17,18,19], they have not been experimentally tested.

In this study, we characterized the surface expansion of bacterial communities in the context of partial migration. We chose Proteus mirabilis as a model system because of its superior ability to expand its community [20]. This superior ability underlies the pathogenesis of this bacterium as it enables P. mirabilis cells to reach across the urethra and colonize the bladder and kidneys and leads to rapid surface fouling of newly inserted catheters [21,22,23,24,25]. P. mirabilis’ superior expansion can be visualized on agar surfaces in a laboratory, as its colony can expand on high agar concentrations (>1%), which are too stiff for other model bacteria including Escherichia coli or Bacillus subtilis [9, 26,27,28]. Here, we investigated the expansion dynamics of P. mirabilis to identify a regulatory mechanism and strategy for its surface colonization.

Materials and methods

Bacterial growth conditions

P. mirabilis ATCC 7022 and its derivatives were used (Supplementary Tables 1 and 2). Liquid cultures were prepared in Luria-Bertani (LB) broth at 37 °C with shaking at 200 rpm. Cells were inoculated from an −80 °C frozen stock a day before an experiment and grown overnight. A small volume of an overnight culture was suspended in fresh medium to prepare an experimental culture. Cells were grown in an experimental culture for at least five doublings before measurements were made as described in the text. Cell density was determined by quantifying the optical density at 600 nm wavelength (OD600) using a Genesys20 spectrophotometer (Thermo Fisher) with a standard cuvette (16.100-Q-10/Z8.5; Starna Cells).

Microscopy and image analysis

The cells were imaged using an inverted microscope (Olympus IX83P2Z). The microscope was controlled by the MetaMorph software (Molecular Devices) and housed in a microscope incubator (InVivo Scientific), which maintained the temperature of samples at 37 °C during the experiments. Images were captured using a Neo 5.5 scientific CMOS camera (Andor). Images were analyzed using MicrobeJ, a freely available plug-in for ImageJ, or custom-built MATLAB software. Additional details about the experiments and analyses are provided in Supplementary Information.

Results

Expansion dynamics of a surface-attached P. mirabilis colony

We studied the expansion of a wild-type (WT) P. mirabilis population by depositing an inoculum droplet on a surface of an LB medium solidified with ~2% agar. When deposited on a surface, a WT P. mirabilis population first grows in its original inoculum zone and forms a colony (Supplementary Fig. 1a). Only after a colony reaches a high cell density (~109 cells/cm2), it initiates migration (Supplementary Fig. 1b). Migration is visualized by the appearance of a faint halo beyond the distinct boundary of a densely-packed parental colony (Fig. 1A). Throughout population migration, the boundary of the parent colony remains stationary and distinct, suggesting that a majority of cells remain in the parental colony.

Fig. 1: Different migration pattern of WT and ΔrcsC strains.
figure 1

A WT colony 7 h after inoculation on a LB agar surface. B ΔrcsC colony 6 h after inoculation. Note that a ΔrcsC colony begins colony expansion ~1 h earlier than a WT colony. Purple-line zones indicate the original inoculum area. C Migratory cells as a proportion of all cells in the colony. Colonies expanding on agar surfaces were suspended in liquid LB culture, and the cells (from the colonies) were immediately imaged with a microscope. Cells were counted and hyperelongated cells (swarmers) were identified (Supplementary Fig. 2). p value (from a T-test) = 0.00069. Error bars indicate one standard deviation from two independent experiments. ~800 cells were analyzed in each experiment. Scale bars represent 10 mm.

Full size image

We then directly determined the migrating fraction in a colony. P. mirabilis cells migrate on stiff surfaces by differentiating to an hyperelongated and hyperflagellated phenotype [20]. Cells exhibiting this motile phenotype can migrate across a surface through a process known as swarming [9, 26,27,28,29]. We and others have shown that differentiation occurs in a subpopulation of cells [30, 31]. Here, we quantified the fraction of this subpopulation by harvesting the entire population and determining the number of hyperelongated cells with a microscope (Supplementary Fig. 2). We found that only ~1% of cells exhibited the motile phenotype (Fig. 1C, green).

Severe repression of migration in the inner region of an expanding colony

How and why is the migratory fraction kept at such a low level? We first sought to unravel a regulatory mechanism controlling the migratory fraction by characterizing the spatio-temporal dynamics of the motile phenotype. The flhDC gene encodes the master regulator for differentiation into the motile phenotype [32, 33]. We constructed a transcriptional fusion of the flhDC and green fluorescent protein (gfp) genes and tracked flhDC expression by monitoring GFP intensity in a colony over time. We found that as a colony matures, GFP intensity increases in cells at the colony edge (Supplementary Movie left panel). These GFP-bright cells differentiate to the motile phenotype and migrate away from the parental colony. In the inner region of the colony, however, GFP intensity remained low, and no differentiation occurred (Supplementary Movie right panel). This severe repression of motility in the inner region explains how a large fraction of a population remains sessile in an expanding colony.

Additionally, we found that there was a narrow, distinct band of cells at the edge of the colony which comprised most GFP-bright cells (Supplementary Movie left panel). The formation of this band can be described by the coffee ring effect, a well-known physical phenomenon caused by capillary flow which pushes particles to the edge of a liquid droplet [34]. This effect was also frequently observed in previous studies of bacterial colony formation [35, 36]. We wondered whether the formation of this band is responsible for the preferential increase in GFP intensity at the colony edge. To avoid this effect, we started colonies from single cells by diluting and spreading an inoculum droplet over a large area of agarose surface. Colonies originating from single cells did not exhibit such bands, and yet GFP intensity was still preferentially higher at the colony edge (Fig. 2A top panel, and Supplementary Fig. 3). To quantify this inhomogeneity, we determined the radial distribution of the bright pixels across multiple colonies. We first characterized the minimum fluorescence threshold value for differentiation into the motile phenotype, which revealed that the cells that expressed fluorescence values greater than 560 (a.u.) eventually differentiated and swarmed. We then determined the radial distribution of the pixels whose GFP intensities were above the threshold as a function of both the normalized (Fig. 2B green) and unnormalized (Supplementary Fig. 4 green) radius. The fraction of GFP-bright pixels within a half the colony radius is almost zero, again showing that GFP-bright cells are primarily located at the colony edge.

Fig. 2: Spatially inhomogeneous motility pattern.
figure 2

A Phase contrast and fluorescence images of WT and ΔrcsC micro-colonies. A PflhDC – gfp transcriptional fusion construct was used. The ΔrcsC images show a colony right before the onset of migration. The WT images show a colony in a similar size. Note that WT colonies initiate migration at a larger colony size than ΔrcsC colonies (Supplementary Fig. 9). WT colonies right before the onset of migration (which are larger than what is shown here) exhibit similar, spatially inhomogeneous distribution of GFP intensities (Supplementary Fig. 3). B Radial distribution of GFP-bright pixels right before the onset of migration. Each colony was divided into 125 rings of equal radial width. GFP-bright pixels (whose GFP intensities were above the threshold as discussed in the main text) were identified, and the fraction of these pixels within each ring (i.e., the number of GFP-bright pixels divided by the total number of pixels within each ring) is plotted as a function of radial position. Here, the radius of each ring was normalized to the actual radius of the colony; absolute (unnormalized) ring radius was also considered (Supplementary Fig 4). Here, the distribution extends slightly beyond one in the x-axis because of ruggedness of colony perimeter. 1 pixel corresponds to 0.1083 μm. The filled areas represent the standard deviation from 10 colonies. Scale bars represent 30 μm.

Full size image

The rcs mutant exhibits very different migration dynamics

In the WT strain, the boundary of a parental colony remains stationary and distinct during expansion (Fig. 1A), because a majority of cells do not participate in migration (Fig. 1C). Our microscope experiments attribute this low participation to severe repression of flhDC expression in the colony center (Fig. 2). During our transposon mutagenesis screening, we identified mutants that showed different dynamics of colony expansion: they shifted the entire colony boundary outward during migration (Fig. 1B and Supplementary Fig. 5). We found that these mutants have the transposon insertion in the rcsC or rcsB genes, two components of the Rcs phosphorelay system [37,38,39]. As transposon insertion might be unstable, we performed clean knockouts of these genes (Supplementary Methods). These mutants exhibited the same, outward shift of the colony boundary. We next tracked the radius of the population edge, which showed that the edge moves faster in the mutant colony than in a WT colony (Supplementary Fig. 6).

The rapid expansion of the colony boundary exhibited by the mutants suggests that a large fraction of the population migrates. We confirmed this by quantifying the fraction of cells expressing the motile phenotype in a ΔrcsC population. We found that ~50% of a population participates in migration (Fig. 1C, red), a fraction significantly higher than the ~1% observed in the WT population (Fig. 1C, green). Importantly, this large increase in the migratory fraction indicates that the rcsC knockout relieved the repression of the master regulator gene flhDC expression. If so, because the repression occurred mostly in the colony center in the WT, ΔrcsC colonies would exhibit strong flhDC expression in the colony center. This prediction is supported by our microscope images, which showed high GFP intensities in the center of ΔrcsC colonies (Fig. 2A bottom panel). This can be seen in the radial distribution of GFP-bright pixels as well, which shows that the fraction of GFP-bright pixels is higher in the colony center than the edge (Fig. 2B, red). Our observation of flhDC repression by Rcs is consistent with our and others’ previous findings that flhDC is a member of a large Rcs regulon [37,38,39,40,41,42,43]. However, regulation of flhDC expression is complex, and exactly how Rcs affects flhDC expression was not clear.

Nutrient-dependent repression of cell motility by Rcs

One major difference between cells in the colony center and edge is the access to nutrients; because nutrients diffusing into the colony are first utilized by cells in the edge, less nutrients are accessible to cells in the colony center [44,45,46,47,48]. Therefore, one possible explanation for low flhDC expression in the WT colony center is that its expression is repressed under low-nutrient conditions. Importantly, given our finding above that Rcs mediates the repression, this would mean that the repression of flhDC expression by Rcs is nutrient-dependent. To examine this possibility, we compared the magnitude of flhDC repression under different nutrient levels. We first measured the flhDC expression in WT and ΔrcsC cells at two different nutrient levels (1× and 0.2× LB) in liquid cultures. An advantage of a liquid culture is that it can be shaken and well-mixed, providing better control for a nutrient condition. Also, because cell density in a liquid culture can be easily determined, flhDC expression can be compared at the same cell density. Cells were grown to an OD600 of 0.35 in liquid culture, transferred to a coverslip, and imaged with a microscope. We found that flhDC expression in WT cells was lower than that in ΔrcsC cells in both nutrient conditions, which is consistent with the observed repressive effect of Rcs on flhDC expression. We then calculated the magnitude of repression by dividing the GFP intensity of ΔrcsC by that of WT (Fig. 3B and Supplementary Fig. 7). We found that the repression was much more severe in 0.2× LB than 1× LB.

Fig. 3: PflhDC – gfp expression in liquid cultures and micro-colonies.
figure 3

A flhDC expression in liquid cultures. Cells were grown in liquid culture to an OD600 of 0.35, transferred to a coverslip, and imaged with a microscope. GFP intensities in ~100 individual cells were measured and averaged in each condition. Error bars represent the standard deviation of two independent experiments. p value (from a T-test) = 0.064 for 1× and 0.015 for 0.2× LB. B Magnitude of flhDC repression by Rcs in liquid cultures. The magnitude of repression was determined by dividing the GFP intensity of ΔrcsC by that of WT (Fig. 3A). Repression is more severe in lower nutrient conditions. p value (from a T-test) = 0.075 for 1× and 0.038 for 0.2× LB. C flhDC expression in the colony center right before the onset of colony migration. The GFP intensity within a 100 pixel (~10 µm) circle was measured at the moment of colony expansion and averaged over 10 colonies (Supplementary Fig. 7). Error bars represent the standard deviation of the average GFP intensity from two independent experiments. D Magnitude of flhDC repression by Rcs in colonies right before the onset of colony migration. The magnitude of repression was determined by dividing the GFP intensity of ΔrcsC by that of WT (Fig. 3C). We additionally determined the magnitude of repression at various points during the colony development for 1× and 0.2× LB (Supplementary Fig. 11).

Full size image

Because our other measurements focused on the center of colonies growing on a semi-solid surface (rather than liquid cultures), we repeated our previous agarose experiments with different LB concentrations and compared the magnitude of flhDC repression in the colony center. The flhDC expression can vary from cell-to-cell (Fig. 2A; further discussed in Discussion). Our analysis showed that averaging flhDC expression within a small disk of radius ~10 μm (which corresponds to the width of a few cells) is sufficient to average out this variation (Supplementary Fig. 8). We found that the averaged flhDC expression in the colony center is always lower in the WT than in the ΔrcsC strain in all nutrient conditions tested (1×, 0.5×, 0.2× and 0.1× LB), again showing flhDC repression by Rcs (Fig. 3C). Importantly, the magnitude of repression increased as the nutrient level decreased (Fig. 3D). Here, this comparison was made for the colonies at the onset of colony migration, but this onset occurred at different colony sizes under different nutrient levels (Supplementary Fig. 9). Thus, we also determined the repression for a smaller but uniform colony size (20 μm radius), which again showed stronger flhDC repression at lower nutrient levels (Supplementary Figs. 10 and 11). This finding agrees with our liquid culture data and further supports that Rcs-mediated repression of flhDC expression is nutrient-dependent.

Fitness advantage of the WT (rcsC+) strain

A ΔrcsC colony expands faster than a WT colony (Supplementary Fig. 6), because it has a much higher fraction of migratory cells (Fig. 1C). Then, what is the advantage of keeping the migratory fraction low? To address this question, we compared the fitness of WT and ΔrcsC strains. Specifically, we spread cells on an agar surface and evaluated the biomass yield by measuring OD600 × ml and cell viability by counting colony forming units (CFU) over time. We used 0.2× LB agar because the repression is pronounced in a low-nutrient environment (Fig. 3C). We found that the ΔrcsC strain produced less biomass than WT in a given territory, although the difference is marginal (Fig. 4A, circles and squares). Importantly, the ΔrcsC strain exhibited a noticeably steeper decrease in cell viability than WT (Fig. 4A, upright and inverted triangles). When we characterized cell viability with propidium iodide (PI) staining, we observed a higher fraction of PI-stained cells in ΔrcsC than WT populations (Supplementary Fig. 12), which is in agreement with lower cell viability of ΔrcsC cells (Fig. 4A).

Fig. 4: The fitness effect of Rcs.
figure 4

A Biomass (OD600 × ml) and cell viability (CFU) as a function of time spent on a 0.2× LB agar plate. At time zero, multiple plates were inoculated by spreading either WT or ΔrcsC cells on the surfaces. At the time points indicated, cells were collected from a plate by suspending them in 2 ml of 0.2× LB liquid. The OD600 values of these suspensions were immediately measured and multiplied by the total suspension volume (2 ml) to determine the biomass (OD600 × ml). These suspensions were then used for CFU assays. In the plot, the measured values were normalized by the value of WT at the first time point. Error bars represent standard deviation from two independent experiments. p value (from a T-test) of CFU at 38 h = 0.028. B The fraction of ΔrcsC cells as a function of time spent on a 0.2× LB agar plate in a competition experiment. WT and ΔrcsC cells were mixed in equal ratio (based on OD600). At time zero, multiple plates were inoculated with the mixed culture. At a time point indicated, cells were collected from a plate, and a CFU assay was performed as discussed above. The ΔrcsC strain carries a kanamycin resistance marker. In the CFU assay, we distinguished ΔrcsC from WT cells by plating the liquid suspensions on kanamycin and kanamycin-free LB plates (Supplemental Methods). Error bars represent the standard deviation from three independent experiments. We additionally performed similar competition experiments on agar plates with different agar concentrations (Supplementary Fig. 13).

Full size image

We next performed competition experiments to more directly compare their fitness. The WT and ΔrcsC strains were co-inoculated onto an agar surface in an equal ratio, and the cell viability of each strain was determined with a CFU assay over time (Supplementary Methods), and the fraction of ΔrcsC cells [= NΔrcsC / (NΔrcsC + NWT)] was calculated. We observed a decrease in the ΔrcsC fraction over time (Fig. 4B and Supplementary Fig. 13). These data collectively reveal a trade-off between population expansion and viability.

Discussion

Our quantitative imaging uncovered intricate spatial dynamics of cell motility governing partial migration of a population. We then systematically analyzed the dynamics (colony center vs. edge, WT vs. Δrcs strains, high vs. low-nutrient levels on agar surface and in liquid culture), which revealed an important role of the Rcs system in these dynamics. The Rcs system is widespread in the Enterobacteriaceae family [49] and is known to regulate a wide range of cellular functions including capsular polysaccharide or colanic acid synthesis, cell envelope maintenance, cell division and virulence [50]. However, physiological cues inducing this system and its functional role in colonizing territories were unclear. Our present work reveals that nutrient-dependent repression of flhDC expression by the Rcs system leads to intricate dynamics of cell motility in an expanding population, thereby having a major impact on the surface colonization pattern.

Our work reveals a trade-off underlying population motility. Previous studies of evolution have shown that E. coli cells can increase their swimming speeds at the expense of cell growth [51, 52]. Hyper-swarming Pseudomonas aeruginosa, which moves faster than the parental strain, can be evolved but is outcompeted in biofilm formation [53]. Previous work has shown that swarming P. mirabilis cells are mechanically fragile and more susceptible to cell-wall stress [54]. Our finding on the fitness disadvantage of Δrcs is consistent with these studies. Furthermore, swimming in liquid costs about 2% of metabolic budget in bacteria [55]. Although the cost of surface motility (swarming) has not been measured, it is thought to be much greater [26, 56]. Consistent with this view, we found that P. mirabilis cells that are actively moving on a surface reproduce at a very low rate or do not reproduce (Supplementary Fig. 14), indicating a severe cost of surface motility. This cost is likely more burdensome in nutrient-limiting conditions, where cells already suffer from a limited metabolic budget. Our studies show that, although inducing migration in the face of this cost (Δrcs strain) does lead to faster population expansion, it severely impairs population fitness. Nutrient-dependent repression of cell motility by Rcs can minimize this cost in nutrient-limited cells while enabling migration only for a small number of privileged cells that have access to nutrients. This mode of partial migration can provide one mechanism to mediate a trade-off, helping a population maintain its fitness during expansion, thereby facilitating surface colonization.

Previous studies suggest an additional mechanism for fitness advantage. RpoS is a general stress sigma factor that is widely present in Enterobacteriaceae [57]. In our previous studies with E. coli, we have shown that in nutrient-limiting conditions, RpoS plays a critical role in conserving a limited metabolic budget for starvation survival [58]. Such a protective role of RpoS is well documented for a wide variety of other bacterial species [57]. Recent studies have found that RpoS expression is induced by Rcs [59, 60]. Therefore, Rcs not only represses cell motility in the nutrient-limited region (as discussed in the preceding paragraph) but also is expected to induce the expression of the important stress sigma factor (RpoS). This two-pronged response is likely to contribute to the fitness advantage conferred by Rcs. Importantly, migration is expected to incur a high cost of dispersal in other organisms as well, impairing their reproduction rate and survival [61,62,63]. A similar strategy could be used in these organisms to optimize territory expansion.

Another important finding from other studies of E. coli is that flhDC expression is stimulated by nutrient limitation [64, 65] (possibly through cAMP-CRP [66,67,68]). A similar mechanism could explain a significant increase in flhDC expression at decreasing nutrient levels in the P. mirabilis ΔrcsC strain (Fig. 3A, C, red). The WT strain also exhibited an increase in flhDC expression (Fig. 3A, C, green). However, the increase in the WT strain was minor because Rcs-mediated flhDC repression became stronger and counteracted the stimulating effect of nutrient limitation on flhDC expression (Fig. 3B, D). This strategy ensures that cells in the colony center do not differentiate and swarm (Fig. 2), keeping the migratory subpopulation low.

Our colony images of the flhDC-gfp strain additionally illustrate the cell-to-cell variability in flhDC expression. GFP-bright cells, which eventually differentiate and swarm, are preferentially located at the colony edge in the WT strain, but not all cells at the edge are GFP-bright (Fig. 2A). In the ΔrcsC colony, GFP-bright cells are located throughout the colony, but again, not all cells are GFP-bright (Fig. 2A). This cell-to-cell variability can be explained by our previous finding that FlhDC auto-activates its expression [30]. This auto-activation is an example of positive feedback, which is known to generate bi-stability, i.e., two distinct levels of gene expression [69]. Positive feedback is shown to be a common mechanism for biological systems exhibiting differentiation [69]. Studies have found that positive feedback, coupled with the stochastic gene expression noise, triggers a fraction of a population to differentiate [70]. If the expression of a key regulator stochastically exceeds a threshold value, positive feedback drives the expression to even higher levels, triggering differentiation. In the ΔrcsC strain, without Rcs-mediated repression, the baseline of flhDC expression is higher, making it easier to exceed the threshold value. Therefore, a higher fraction of cells differentiates and contributes to rapid colony expansion.

Our findings also shed new light on a type of surface-attached microbial community, biofilm. Surface motility is known to have a dramatic effect on biofilm formation [27, 53]. We found that motility is strongly repressed by Rcs in a dense region of a colony (Fig. 2A), which ensures that a majority of cells remain sessile within the colony. Due to this repression, the boundary of a parental colony remains distinct and stationary during expansion (Fig. 1A), which enables the maintenance of the original densely-packed community. The same mechanism is likely at play in a biofilm, where Rcs ensures the formation of a robust biofilm by inhibiting premature dispersal. This is consistent with previous studies of E. coli, P. mirabilis, and other bacteria, which have shown that rcs deletion leads to the formation of defective biofilms [38, 71, 72].