Abstract
Bacteria often live in biofilms, which are dense microbial communities surrounded by a secreted extracellular matrix. Here, we demonstrate that hydrodynamic flow and matrix organization interact to shape competitive dynamics in Pseudomonas aeruginosa biofilms. Irrespective of initial frequency, in competition with matrix mutants, wild type cells always increase in relative abundance in flat-surfaced, straight-tunnel microfluidic devices under simple flow regimes. By contrast, in microenvironments with complex, irregular flow profiles – which are common in natural environments – wild type matrix-producing and isogenic non-producing strains can coexist. This result stems from local obstruction of flow by wild-type matrix producers, which generates regions of near-zero flow speed that favor occupation by cells that do not invest in matrix secretion. Our findings connect the evolutionary stability of matrix production with the hydrodynamics and spatial structure of the surrounding environment, providing a potential explanation for the variation in biofilm matrix secretion observed among bacteria in natural environments.
Introduction
In nature, bacteria predominantly exist in biofilms, which are surface-attached or free-floating communities of cells held together by a secreted matrix [1–3]. The extracellular matrix defines biofilm structure, promoting cell-cell and cell-surface adhesion and conferring resistance to chemical and physical insults [4–6]. The matrix also plays a role in the population dynamics of biofilm-dwelling bacteria [2, 7, 8]; simulations and experiments using Vibrio cholerae, Pseudomonas fluorescens, and Pseudomonas aeruginosa show that matrix-secreting cell lineages can physically smother or laterally displace other cell lineages and, in so doing, outcompete neighboring matrix non-producing cells [9–16]. However, not all wild and clinical isolates of these species produce biofilm matrices, despite the clear ecological and competitive benefits of possessing a matrix [17–19]. Fitness trade-offs between the benefits of being adhered to surfaces and the ability to disperse to new locations can cause variability in matrix production [10, 18]. However, it is not known how selective forces within the biofilm environment itself might drive the coexistence of strains that make matrix with strains that do not. Here we explore how selection for matrix production occurs in biofilms on different surface geometries and flow regimes, including those that are relevant inside host organisms and in abiotic environments such as soil.
The local hydrodynamics associated with natural environments can have dramatic effects on biofilm matrix organization. This phenomenon has been particularly well established for P. aeruginosa, a common soil bacterium [20] and opportunistic pathogen that thrives in open wounds [21, 22], on sub-epithelial medical devices [23], and in the lungs of cystic fibrosis patients [24–26]. Under laminar flow in simple microfluidic channels, P. aeruginosa forms biofilms with intermittent mushroom-shaped tower structures [24, 27–29]. Under irregular flow regimes in more complex environments, however, P. aeruginosa also produces sieve-like biofilm streamers that protrude into the liquid phase above the substratum [30–33]. These streamers – whose structure depends on the secreted matrix – are proficient at catching cells, nutrients, and debris that pass by, leading to clogging and termination of local flow [34]. The spatial and temporal characteristics of flow thus combine to alter matrix morphology, which, in turn, feeds back to alter local hydrodynamics and nutrient advection and diffusion. By modifying community structure and solute transport in and around biofilms [35], this feedback could have a significant influence on the evolutionary dynamics of matrix secretion in natural environments.
Here, we examined the relationship between flow regime and matrix-mediated biofilm competition using strains of P. aeruginosa PA14 that differ only in their production of Pel, a visco-elastic matrix polysaccharide that is central to biofilm and streamer formation [27, 34, 36, 37]. Using a combination of fluid flow visualization and population dynamics analyses, we reveal a novel interaction between hydrodynamic conditions, biofilm architecture, and competition within bacterial communities.
Results / Discussion
We performed competition experiments with wild type PA14 and an otherwise isogenic strain deleted for pelA, which is required for synthesis of Pel [38]. Deletion of pelA significantly impairs biofilm formation in PA14, which does not naturally produce Psl, an additional matrix polysaccharide secreted by other P. aeruginosa isolates [39, 40]. Wild type cells produced GFP, and ∆pelA mutants produced mCherry. Experiments in shaken liquid culture using genetically identical cells producing GFP or mCherry confirmed that fluorescent protein expression constructs had no measurable effect on growth rate (Figure S1). Our first goal was to compare the population dynamics of the wild type and ∆pelA strains in typical straight-chamber microfluidic devices, which have simple parabolic flow regimes, and in porous environments containing turns and corners, which have irregular flow profiles and better reflect the packed soil environments that P. aeruginosa often occupies [20, 41, 42]. To approximate the latter environment, we used microfluidic chambers containing column obstacles; the size and spacing distributions of the column obstacles were designed specifically to simulate soil or sand. Analogous methods have been used previously to study bacterial growth [43] and the behavior of Caenorhabditis elegans [44] in realistic environments while maintaining accessibility to microscopy. In our setup, flow was maintained through these chambers at rates comparable to those experienced by P. aeruginosa in a soil environment [45].
The flow regime alters selection for matrix production in biofilms
Several approaches are available to study how competitive dynamics differ in particular flow environments. Most commonly, one would monitor biofilm co-cultures of wild type and ∆pelA PA14 cells over time until their strain compositions reached steady state. Performing such time-series experiments was not possible here due to a combination of low fluorescence output in the early phases of biofilm growth coupled with phototoxicity incurred by cells during epifluorescence imaging. To circumvent this issue, we measured the change in frequency of wild type and ∆pelA cells as a function of their initial ratio in both straight-tunnel chambers and soil-mimicking chambers containing column obstacles. From these measurements, we could infer the final stable states of Pel-producing and non-producing cells as a function of surface topography and flow conditions. This method is commonly used to measure the behavior of dynamical systems, and it has been employed successfully in a variety of analogous experimental applications [10, 11, 16, 46–48].
In simple flows, wild type PA14 increased in relative abundance regardless of initial population composition, indicating uniform positive selection for Pel secretion. This result is consistent with recent studies of V. cholerae and Pseudomonas spp. demonstrating that the secreted matrix cannot be readily exploited by non-producing mutants (Figure 1A,B) [10, 12, 14, 16]. Wild type cells obtained their competitive advantage in biofilms despite a reduction in maximum growth rate relative to ∆pelA mutants in shaken liquid culture (Figure S1). Confocal microscopy revealed that Pel-producers mostly excluded non-Pel-producing cells from biofilm clusters in straight tunnel chambers (Figure 2A,B), although some ∆pelA mutants resided on the periphery of wild type biofilms. The liquid effluent from these chambers contained an over-representation of the ∆pelA mutant relative to wild type, consistent with the interpretation that ∆pelA strain was displaced from the substratum over time (Figure 2C). When wild type and ∆pelA cells competed in microfluidic devices simulating porous microenvironments, there was a pronounced shift to negative frequency-dependent selection for Pel production (Figure 1A,C). Wild type PA14 was selectively favored at initial frequencies below ~0.6. Above this critical frequency, the ∆pelA mutant was favored. From this result, we can infer that in this porous environment, ∆pelA null mutants can grow and stably coexist with wild type Pel-producers.
Wild type P. aeruginosa obstructs porous environments, generating shear-free regions that Pel-deficient mutants can occupy
Previous work has shown that in environments with corners, wild type P. aeruginosa produces biofilm streamers that catch cells and other debris in the surrounding flow [34]. Such streamers were produced by the wild type in our system. Although ∆pelA cells could be found in these structures, they were not abundant, so cell capture by streamers cannot account for the observed coexistence of the two strains. This result suggests that streamers do indeed catch debris, consistent with prior studies [31, 34], but that in our system, ∆pelA cells are not abundant enough in the passing liquid phase to accumulate substantial population sizes by this manner (Figure S2).
Our microscopy-based observation of soil-mimicking chambers suggested that wild type biofilms gradually obstructed some of the regions located between columns over time. We hypothesized that partial clogging could render those portions of the chambers more suitable for growth of the ∆pelA strain, which was previously shown to be sensitive to removal by shear [39, 40]. This hypothesis predicts that ∆pelA cells should be found predominantly in regions of the chamber that have been clogged by wild type biofilms, or immediately downstream of such regions. We tested this prediction by repeating our co-culture competition experiment with wild type and ∆pelA cells in chambers containing columns, measuring the distribution of each strain as above, and then re-inserting new influent tubing through which fluorescent beads were introduced to the chambers (Figure S3). By tracking the beads with high frame-rate microscopy, we could trace the presence and absence of fluid flow and then superimpose this information onto the spatial distributions of wild type and ∆pelA mutant cells. Wild type biofilms accumulated primarily between columns parallel to the long axis of the chamber, with clusters of ∆pelA cells in close proximity. Importantly, in support of our prediction, ∆pelA cells grew most successfully in regions that were locally obstructed by wild type biofilms (Figure 3A,B). ∆pelA cells did not clog chambers when grown in isolation, supporting the interpretation that ∆pelA accumulation relies on and only occurs after clogging by the wild type. This result supports prior indications that Pel-producers do indeed have a higher tolerance for shear than Pel-deficient cells [39, 40]. We directly confirmed this prediction in our system by independently growing each strain in isolation under varying shear stress in simple straight-tunnel chambers (Figure 3C).
Collectively, our findings suggest that when P. aeruginosa wild type and ΔpelA mutants experience irregular flow in heterogeneous environments, wild type biofilm formation causes partial clogging, regionally reducing local flow speed. The lack of flow generates favorable conditions for the ∆pelA strain, whose biofilm would otherwise be removed by shear forces, enabling it to proliferate locally using nutrients diffusing from the bulk liquid phase [49].
Conclusions
Our results show that a feedback process operates between hydrodynamic flow conditions, biofilm spatial architecture, and competition in P. aeruginosa biofilms. Pel non-producers do not appear to directly exploit the matrix material produced by wild type cells, but they can take advantage of the low-shear conditions that wild type matrix producers generate in complex flow environments. P. aeruginosa is notorious as an opportunistic pathogen of plants and animals, including humans. It also thrives outside of hosts, for example, in porous niches such as soil. Despite the well-documented ecological benefits of matrix secretion during biofilm formation, environmental and clinical isolates of P. aeruginosa exhibit considerable variation in their production of matrix components, including loss or overexpression of Pel [17]. Our results offer an explanation for natural variation in the ability of P. aeruginosa to produce extracellular matrix, particularly among bacteria in porous environments: the evolutionary stable states of extracellular matrix secretion vary with the topographical complexity of the flow environment in which bacteria reside.
Materials and Methods
Strains
All strains are derivatives of Pseudomonas aeruginosa PA14. Wild type PA14 strains constitutively producing fluorescent proteins [34] were provided by Albert Siryaporn (UC Irvine), and they harbor genes encoding either EGFP or mCherry under the control of the PA1/04/03 promoter in single copy on the chromosome [50]. The ∆pelA strain was constructed using the lambda red system modified for P. aeruginosa [51].
Liquid growth rate experiments
To determine maximum growth rates and the potential for fluorescent protein production to cause fitness differences, bacterial strains were grown overnight in Luria Bertani (LB) medium at 37º C. To acclimatize the cultures to the metabolic shift that occurs in minimal medium, cultures were back-diluted into minimal M9 medium with 0.5% glucose and monitored until their optical densities at 600 nm were ~0.6, corresponding to mid-logarithmic phase. Cultures were back diluted again into minimal M9 medium with 0.5% glucose and transferred to 96 well plates. Hourly measurements of culture optical density were taken until saturation, corresponding to stationary phase. MatLab curve fitting software was used to calculate the maximum growth rate of each strain. These experiments confirmed that the fluorescent protein markers had no measurable effect on growth rates and thus did not contribute to competitive outcomes in our experiments.
Microfluidics and competition experiments
Microfluidic devices consisting of poly(dimethylsiloxane) (PDMS) bonded to 36 mm × 60 mm glass slides were constructed using standard soft photolithography techniques [52]. We used flat, straight-tunnel microfluidic devices with no obstacles to simulate environments with simple parabolic flow profiles, and we used devices with PDMS pillars interspersed throughout the chamber volume to simulate environments with complex (i.e. irregular, non-parabolic) flow profiles. The size and spatial distributions of these column obstacles were determined by taking a cross-section through a simulated volume of packed beads mimicking a simple soil environment. Flow rates through these two chamber types were adjusted to equalize the average initial flow velocities, although the local flow velocity varied as biofilms grew during experiments (see main text).
For all competition experiments, bacterial strains were grown overnight. The following morning, aliquots of the overnight cultures were added to Eppendorf tubes, and their optical densities were equalized prior to preparation of defined mixtures of wild type and ∆pelA cells. 100 µL volumes of the wild type strain alone, the ∆pelA strain alone, or mixtures of the two strains (for competition experiments), were introduced into microfluidic chambers using 1 mL syringes and Cole-Parmer polytetrafluoroethylene tubing (inner diameter = 0.30 mm; outer diameter = 0.76 mm). After 3 h, fresh tubing connected to syringes containing minimal M9 medium with 0.5% glucose were inserted into the inlet channels. Syringes (3 mL BD Syringe, 27G) were mounted onto high precision syringe pumps (Harvard Apparatus), which were used to tune flow rate according to empirical measurements of flow speeds in soil [45]. It should be noted that microfluidic experiments in the chambers containing columns experience a high failure rate, in which no biofilms appear to grow after the 72 h period of the experiment. This problem was overcome by performing the experiment at high replication in parallel. Sufficient data were thus collected to populate the relevant panels in Figures 1 and 3 of the main text.
Microscopy and image analysis
Mature biofilms were imaged using a Nikon Ti-E inverted microscope via a widefield epifluorescence light path (using a 10x objective) or a Borealis-modified Yokogawa CSU-X1 spinning disk confocal scanner (using a 60x TIRF objective). A 488 nm laser line was used to excite EGFP, and a 594 nm laser line was used to excite mCherry. For all competition experiments, 3-5 biological replicates were prepared on different days in different microfluidic chambers. Within each chamber, 4-6 non-overlapping, randomly selected areas (technical replicates) were selected and imaged for quantification. Quantification of biofilm composition was performed using custom MatLab scripts for image analysis [48]. Imaging of biofilms could only be performed once for each experiment, precluding time-series analyses, due to phototoxicity effects after multiple rounds of imaging. Phototoxicity was a particularly notable issue here due to dimness of the fluorescent proteins in P. aeruginosa, which made long exposures necessary to capture images of sufficient quality for later analysis. For this reason we opted for inferential population dynamics analysis as described in the main text.
Effluent Measurements
To measure strain frequencies in the biofilm effluent of straight-tunnel chambers (Figure S2), 1:1 strain mixtures of wild type and ∆pelA cells were prepared and inoculated into simple flow chambers according to the procedure outlined above for competition experiments. At 0, 24, and 48 h, 5 µL samples were collected from the microfluidic chamber outlet tubing, mixed vigorously by vortex, and plated onto agar in serial dilution. After overnight growth at 37º C, plates were imaged with an Image Quant LAS 4000. Cy3 and Cy5 fluorescence settings were used for EGFP and mCherry excitation, respectively. Image Quant TL Colony Counting software was used to measure the relative abundance of each strain.
Flow Tracking Experiments
1:1 mixtures of the wild type and the ∆pelA mutant were prepared and introduced into obstacle-containing flow chambers according to the procedure described above. Minimal M9 medium with 0.5% glucose was introduced into the chambers for 72 h as described above. The entire chamber was then imaged using widefield epifluorescence microscopy to document the locations of wild type and ∆pelA cell clusters. Subsequently, the influent syringes were replaced with syringes containing yellow-green fluorescent beads (Invitrogen; sulfate-modified, diameter = 2 µm) at a concentration of 0.3%. Bead suspensions were flowed into the microfluidic chambers. To determine the presence or absence of flow with respect to the spatial distributions of wild type and ∆pelA cells, and to obtain large images for statistics, the entire chamber was imaged with a 1 s exposure time, over which traveling beads were captured as streaks. It should be noted that this experiment also has a high failure rate do to the sensitivity of the microfluidic chambers to removal and re-insertion of tubing, and required optimization to execute successfully. Custom MatLab code was written to correlate the presence or absence of fluid flow with the accumulation of wild type and ∆pelA cells. In brief, the positions of the columns were first identified and used to divide the chamber into triangular sampling areas using a network structure in which columns served as nodes and straight lines between column centers served as vertices. Within each sampling triangle, the area covered by columns was first removed, and subsequently, the averaged ∆pelA and wild type fluorescence intensities in the remaining area were used to determine if a particular region had wild type and/or ∆pelA accumulation. In parallel, each sampling area was scored for the presence of flow in the corresponding bead tracking images Figure S3).
Author Contributions
CDN and BB conceived the project. CDN, DR, JY, and KD performed experiments and analyzed data. KD and JY contributed new analytical tools. All authors wrote the manuscript.
Conflict of Interest
The authors declare no conflict of interest.
Supplemental Information
Supplemental Materials and Methods
Acknowledgments
We thank members of the B.L.B. laboratory, as well as Thomas Bartlett and Alvaro Banderas for helpful discussions. This work was supported by the Alexander von Humboldt Foundation (CDN), the Human Frontier Science Program Grant CDA00084/2015-C (KD), the Max Planck Society (KD), the Howard Hughes Medical Institute (BLB), NIH Grant 2R37GM065859 (BLB), National Science Foundation Grant MCB-0948112 (BLB), and National Science Foundation Grant MCB-1344191 (BLB).
Footnotes
↵* Co-first authorship