adapted to colonize chitin surfaces, exploit the resources embedded in them, and spread to other particles are thus likely to be more frequently represented in estuarine conditions (20,21). The marine natural history of V. cholerae makes this bacterium an ideal candidate for studying the effects of different surfaceoccupation strategies that might occur in estuarine and oceanic environments (16,18,20,22,23).

Many strains of V. cholerae and other Vibrio spp. can be found in the marine environment, with varying propensities for biofilm formation (16,21). Recent work from several groups has characterized the spatial and temporal patterns of biofilm development in remarkable detail for model strains of V. cholerae O1 El Tor, the predominant agent of the modern cholera pandemic. These studies have revealed the localization of different polysaccharide and protein subcomponents of the secreted matrix (24)(25)(26)(27)(28)(29), the key cell group architecture transitions and origins of cell-cell orientation (30,31), and the cell-cell interaction mechanics underlying collective structure in response to fluid flow (32). V. cholerae cellular morphology has been carefully studied in laboratory conditions (33)(34)(35); however, the degree of V. cholerae morphological variation in the environment, the impact of cell shape on biofilm architecture, and the consequences for interstrain competition are less well understood. To explore these questions, we developed a microfluidic assay in which chitin particles are embedded in flow devices perfused with

The human pathogen Vibrio cholerae, when not inside of a host, grows in cell clusters (biofilms) on pieces of detritus in aquatic environments. Here we discovered that some isolates of V. cholerae can change their shape from small commashaped cells to long filaments in seawater. This altered cell shape allows cells to make new types of biofilms, and provides an advantage in quickly colonizing particles in seawater, at the expense of longer-term competitive ability. The filamentous cell-shape strategy is particularly effective at competing in environments with quick turnover of chitin particles. This result showcases how bacterial cell shape can be coupled to environmental success during surface occupation, competition within biofilms, and dispersal to new resource patches.

artificial seawater and onto which cells could be readily inoculated to monitor colonization and biofilm growth.

Surveying samples of V. cholerae pandemic isolates from the two major extant serogroups, O1 El Tor and O139, we observed that some variants produce subpopulations of filamented cells under nutrient-limited conditions, including on particles of chitin in seawater. We show that filamentation confers markedly altered chitin colonization and biofilm architecture relative to shorter cells. Differences in chitin colonization and biofilm architecture, in turn, influence competition for space and resources, suggesting that normal-length and filamentous morphotypes are advantageous in different regimes of chitin particle turnover in the water column. Overall, our results highlight a novel mode of biofilm assembly and yield new insights into the potential roles of bacterial filamentation in microbial ecology.

Filamentation in Conditions Mimicking a Marine Habitat. Using simple microfluidic devices in which cells grow on glass substrata under the flow of M9 minimal media, we first explored the cell shape and biofilm architecture of different isolates of V. cholerae from the two pandemic serogroups O1 El Tor and O139 (for a full strain list, see SI Appendix, Table S1). At least one representative from both serogroups produces subpopulations of filamentous cells in minimal media, and we sought to isolate this morphotype to further understand its consequences for growth and biofilm development. V. cholerae AI-1837 from the O139 serogroup and its avirulent derivative, CVD112 (36), filament especially aggressively (SI Appendix, Figs. S1 andS5), and we therefore took CVD112 as a model to explore the consequences of filamentation for biofilm growth dynamics. For comparison throughout the study, we use the model strain N16961 (serogroup O1, El Tor biotype), which did not produce filaments under any conditions tested. To allow for visualization of these strains via confocal microscopy, the far-red fluorescent protein-encoding locus mKate2 was inserted in single copy under the control of a strong constitutive promoter into the chromosomal lacZ locus of CVD112 and its derivatives. Analogously, a construct encoding the orange fluorescent protein mKO-κ was inserted into the chromosomal lacZ of N16961 and its derivatives. We have shown previously that insertion of these fluorescent protein constructs into the chromosome does not substantially influence growth rate or other biofilm-associated phenotypes (19,37).

In preliminary experiments, CVD112 and N16961 were inoculated into separate microfluidic devices and perfused with M9 minimal media containing 0.5% glucose. After 24 h, confocal imaging revealed dramatic differences in the biofilm architecture of the two strains (Fig. 1). Consistent with prior reports, N16961 grew in well-separated microcolonies of cells about 2 μm long, with each hemispherical biofilm microcolony appearing to descend from one progenitor on the glass surface (Fig. 1A) (30,37,38). In contrast, CVD112 formed large groups of filamented cells entangled with one another (Fig. 1B).

We found that CVD112 filamentation occurs most strongly under nutrient-limited medium environments, as CVD112 and N16961 cells were both observed exclusively as cells ∼2 μm long when cultivated in shaken culture with nutrient-rich lysogeny broth (LB) (SI Appendix, Fig. S1). In other circumstances, filamentation has been associated with an acute stress response before cell death, and since CVD112 filaments were seen in M9 minimal media, but not LB, we first suspected that they might be in a disturbed physiological state. However, CVD112 showed no growth defect in either LB or M9 media relative to N16961 as measured by colony-forming unit (CFU) count (Fig. 1C).

In natural seawater environments, V. cholerae colonizes the exoskeletons of arthropods, where it secretes chitinases to digest and consume N-acetyl glucosamine (GlcNAc) (18). We found that CVD112, but not N16961, filaments in artificial seawater with 0.5% GlcNAc as the sole source of carbon and nitrogen, identical to the morphology shown in Fig. 1B. As was the case in LB and M9 media, CVD112 and N16961 growth curves are indistinguishable by CFU count in artificial seawater with GlcNAc, indicating similar rates of cell division (Fig. 1C). Following this observation, we suspected that CVD112 produces more total biomass than N16961 per unit time in low-nutrient media (where CVD112 produces filaments, while N16961 produces cells of normal length) but not in high-nutrient media (where both strains produce cells of normal length). We tested this possibility by measuring the rate of total biomass production in liquid culture by optical density instead of CFU count. As predicted, CVD112 produces biomass more quickly and to higher final density than N16961 in M9 media with glucose, and in artificial seawater with GlcNAc, but not in LB (Fig. 1D). This result was further confirmed with dry biomass measurements of each strain after reaching stationary phase (SI Appendix, Fig. S2). We visualized cells at regular time intervals over the course of their growth and confirmed that CVD112 begins filamenting in midexponential phase in seawater with GlcNAc, but not LB (SI Appendix, Fig. S1).

Filamentation of CVD112 in nutrient-limited conditions must derive in part from changes in cell-elongation rate relative to cell-division rate. By simultaneously staining DNA and the bacterial membrane, we found that chromosome replication continues-visible as multiple separated nucleoids-as filaments elongate (SI Appendix, Fig. S3). Motility assays on lowconcentration agar plates (SI Appendix, Fig. S4), as well as direct observation by microscopy (Movie S1), showed that filamentous CVD112 cells were actively motile, though at lower speed than cells of shorter length. The relative rates of elongation and division of CVD112 can be reverted under nutrient-rich conditions, such that filamentous cells divide into multiple ∼2μm-long cells similar to strain N16961. We could observe this directly by following the growth of CVD112 cells prefilamented in low-nutrient media, which were then transferred under an agar pad made from LB (Movie S2). From these results, we infer overall that CVD112 has a decreased rate of division, but not elongation, in low-nutrient media, which leads to filamentation in a reversible manner. As our focus will be on the biofilm ecological consequences of filamentation, we emphasize that there is no growth-related fitness defect accompanying this morphology under the conditions tested. Beyond these observations, understanding the regulatory and mechanistic basis for filamentation, and its broader physiological relevance, are notable areas for future work. After inspecting our set of isolates of V. cholerae, we found that at least one strain of O1 El Tor and O139 produces subpopulations of filamented cells on chitin in seawater, while others of each serogroup do not [SI Appendix, Fig. S5; SI Appendix, Table S2 provides a list of the differences between these strains as identified by whole-genome sequencing (39)]. Other work has noted natural variation in the propensity to filament among nontoxigenic strains of V. cholerae as well (40). Filamentation is thus observed in a variety of strain backgrounds and is not specific to toxigenic strains, or to one serogroup among pandemic isolates. This is the pattern one would expect to see if natural selection for filamentation varies with local environmental conditions. To understand potential reasons why this morphotype could be maintained by selection, we next set out to determine how filamentous cells interact and compete with those of shorter length in a naturalistic context.

Matrix-Independent Biofilm Formation. We next explored the biofilm morphologies of filamenting CVD112 and nonfilamenting N16961 V. cholerae, simulating the natural conditions they experience in marine environments. We inoculated these strains in microfluidic channels containing PDMS traps to immobilize chitin particles (in this case, shrimp shell), a natural substrate of V. cholerae for biofilm growth and nutrient consumption (41,42) (SI Appendix, Materials and Methods). The chambers were perfused with artificial seawater at a flow rate similar to what would be experienced by a descending marine snow particle (19). Other carbon and nitrogen sources were omitted from the liquid medium, in order to minimize growth in areas of the chamber other than the chitin particles.

In the process of inoculating chitin with V. cholerae under flow, we noticed that CVD112 filaments were particularly adept at colonizing shrimp shell surface, with single cells often wound around the contours of individual chitin particles (SI Appendix, Fig. S6). This suggests that single filaments can undergo large deformation in shear flow, despite the stiffness of the cell wall, and that these flexible filaments can therefore wrap around objects in the flow path. To determine if these properties give CVD112 a colonization advantage on chitin, we measured the attachment rates of filamentous CVD112 and nonfilamenting N16961 by flowing a 1:1 mixture of the two strains (normalized by total biomass per volume) onto chitin particles in artificial seawater. Filamentous V. cholerae does indeed colonize chitin more rapidly than short cells (Fig. 2A). This result holds when the two strains' colonization rates are tested in monoculture (SI Appendix, Fig. S7A), and when cells dispersing directly from previously occupied chitin particles are flowed into new chambers containing fresh chitin (SI Appendix, Fig. S7B). Though N16961 does not produce filaments on its own, we could artificially induce this strain to do so by treating it with subinhibitory concentrations of cefalexin for 1 h, which blocks cell division with minimal impact on cell viability or other aspects of V. cholerae morphology (34). N16961 pretreated in this manner exhibited similar cell elongation and a chitin colonization rate statistically equivalent to that observed for CVD112 filaments (Fig. 2B). Below, we assess the contribution of other adhesion factors that could influence CVD112 attachment and subsequent growth on chitin.

Following colonization, CVD112 produces biofilms that differ dramatically from those of N16961; they are composed of enmeshed cell filaments and lack the typical cell-cell packing and radial orientation associated with N16961 biofilm microcolonies (Fig. 2C) (26,30). The relative absence of tight cell-cell association prompted us to ask whether CVD112 was producing biofilm matrix, including Vibrio polysaccharide (VPS) and the adhesin RbmA, which interacts with the cell exterior and with VPS to hold neighboring cells in close proximity (24,26,29,30,37). To assess the contribution of matrix production to filamentous biofilms on chitin, we tested CVD112 and its isogenic ΔvpsL null mutant, which is unable to synthesize VPS or to accumulate any of the major matrix proteins (24). Biofilm production of wild-type N16961, whose structure depends on matrix secretion (24,25,31), and its isogenic ΔvpsL null mutant were also measured for comparison. All strains contained a FLAG epitope inserted at the C terminus of the native rbmA locus (24,37). Fusion of a FLAG epitope to RbmA has been shown not to interfere with its function (24,26), and allowed us to localize and quantify RbmA by immunostaining as a proxy for general matrix accumulation.

As expected, N16961 produced biofilms with abundant RbmA, while its isogenic ΔvpsL mutant was impaired for biofilm growth relative to the wild-type parent and showed no matrix accumulation by RbmA staining (Fig. 2 D-F). The biomass and visible biofilm architecture of CVD112 and its ΔvpsL null mutant, on the other hand, were equivalent, and CVD112 biofilms showed no detectable RbmA accumulation, even in areas of dense growth (Fig. 2 D, G, andH). Previous reports have suggested the involvement of Oantigen and capsule polysaccharide in calcium-dependent biofilms of (nonfilamentous) V. cholerae O139 (43), but here we observed that filamentous biofilms did not rely on these factors: Their biomass on chitin was unchanged in the absence of calcium, and in a ΔwbfR background, which cannot synthesize O-antigen or capsule polysaccharide (SI Appendix, Fig. S8).

Recent work has demonstrated an important role for competence pili in biofilm structure and clonal autoaggregation of V. cholerae O1 El Tor with normal cell length on chitin (44). For filamentous biofilms of CVD112 in our experimental flow conditions, a ΔpilA deletion mutant lacking competence pili did not differ from CVD112 in biomass accumulation (SI Appendix, Fig. S8). Likewise, deletion mutants (ΔtcpA and ΔgbpA, respectively) lacking toxin-coregulated pili or chitin-binding protein were no different from the parental CVD112 (SI Appendix, Fig. S8). An immotile mutant (ΔflaA) lacking the polar flagellum showed a small but statistically significant defect in biomass accumulation, consistent with a role for the flagellum in surface colonization or exploration (25,45,46). On the other hand, chitin attachment and subsequent biofilm growth were highly dependent on MSHA pili (18,46,47), as a ΔmshA deletion mutant was severely defective for biomass accumulation relative to the parental CVD112 (SI Appendix, Fig. S8). Importantly, quantitative RT-PCR analysis of mshA transcripts per unit biomass indicated lower expression of mshA in CVD112 than in N16961 (SI Appendix, Fig. S9), so differential production of MSHA adhesin on its own cannot explain the augmented colonization ability of filamentous CVD112.

Overall, these results suggest that filamentation, in conjunction with other key adhesion factors including MSHA pili, yields an advantage in chitin attachment and early-stage biofilm accumulation in V. cholerae CVD112 relative to N16961. The filamentous biofilm structures of CVD112 are independent of the primary matrix polysaccharide VPS and cell-cell adhesin RbmA. We infer from our matrix staining and deletion mutant analysis that filamentous biofilm architecture derives at least in part from entanglement of the cells themselves, which serve as both the actively growing biomass and a foundation of the biofilm's structure.

Filamentation Is Advantageous in Frequently Disturbed Environments.

Once we had found that filamentation provides an advantage during chitin attachment, we sought to understand the relative fitness effects of this cell morphology on chitin in coculture with cells of normal length that produce biofilm matrix. To do this, we inoculated CVD112 and N16961 together on chitin in artificial seawater and measured their subsequent biofilm compositions by confocal microscopy daily for 12 d. Our goal in the following experiments was to identify when and how filamentation might provide a fitness advantage relative to conventional biofilm architecture in conditions closely mimicking the marine environment V. cholerae naturally occupies.

When biofilms were left unperturbed for the full duration of the experiment, filamenting CVD112 cells had an initial advantage consistent with our colonization and growth rate experiments, but nonfilamenting N16961 eventually increased in frequency to become the overwhelming majority of the population (Fig. 3 A andC). Time series of single locations on the chitin surface show the progressive displacement of filamentous CVD112 biofilms by the matrix-replete biofilms of N16961 (SI Appendix, Fig. S10). As noted above, filamentous CVD112 biofilms are independent of VPS and RbmA matrix components, while N16961 secretes copious extracellular matrix within chitinbound biofilms. Though CVD112 is superior in its initial surface occupation, its eventual displacement by N16961 is consistent with our prior work demonstrating that matrix secretion and cellcell packing confer a pronounced local competitive advantage within biofilms of V. cholerae, as well as resistance to spatial invasion by other bacteria (13,14,37,48,49). As opposed to matrixreplete biofilms of V. cholerae, biofilms of filamentous CVD112 permit competing strains to invade their interior volume: They do not maintain a grip over the space they initially claim during the colonization and early growth phases of the experiment. Direct proximity to chitin is a crucial determinant of fitness in this simulated marine environment, as chitin surfaces are the sole source of carbon and nitrogen, and active flow causes any excess chitin digestion products to be swept out of the system before they can be exploited by cells at a distance from the chitin particles (19).

Although matrix-producing N16961 outcompetes CVD112 within a patch of chitin-attached biofilm over time, the latter's colonization advantage led us to speculate that it could be successful when residence times on a given chitin particle are shorter. This could be the case, for example, under frequently agitated water column conditions or when chitin particle sizes are smaller, such that they are depleted quickly. To assess this idea, we repeated our competition experiments, but instead of tracking biofilm growth within a single microfluidic chamber over 12 d, we periodically used the liquid effluent exiting the chitin chamber to inoculate a new chamber containing fresh chitin, where a new competition would resume (SI Appendix, Fig. S11). This process simulates a disturbance event in which dispersal increases representation on new chitin particles elsewhere in the water column. We implemented two disturbance regimes, recolonizing fresh chitin chambers once every 72 h, or once every 24 h.

When dispersed cells were taken from the effluent and allowed to recolonize fresh chitin every 72 h, filamenting CVD112 cells showed a protracted early increase in frequency, but once again were eventually outcompeted by matrix-secreting, nonfilamented N16961 cells (Fig. 3 A, B, andD). However, when effluent collection and recolonization of fresh chitin occurred every 24 h, the CVD112 strain dominated cocultures for the full duration of the 12-d experiment (Fig. 3 A, B, andE). The trajectories of these population dynamics were remarkably consistent from one run of the experiment to the next in all dispersal/recolonization regimes, despite variation in one replicate to the next in the particular arrangement of chitin particles in the chamber and the locations of initial attachment by the two strains. This demonstrates the dominating strength of the effects of competition during colonization, biofilm growth, and dispersal, relative to the influence of stochastic factors such as orientation of chitin particles in the chambers, initial strain colonization patterns, or local variation in flow regime.

Individual cell morphology varies widely within and across bacterial species (50,51), but in most cases it is not clear how cell shape relates to emergent structure and ecology of cell collectives such as biofilms. Here we have found that some isolates of V. cholerae produce long cell filaments under nutrient-limited media, including conditions closely matched to the natural marine environment. This cell morphology generates a pronounced advantage in chitin surface colonization and a matrix-independent biofilm architecture that permits rapid surface occupation and high dispersal rates. However, this advantage comes at a cost: Filamentous biofilms are ultimately displaced by matrix-secreting strains in long-term competition experiments. Filamentation is thus advantageous when patches of chitin turn over quickly, such that faster colonization and reversible attachment are more important. Our results demonstrate that the shape of V. cholerae variants can tune the relative investment into surface colonization, long-term biofilm robustness, and dispersal back into the planktonic phase. These are fundamental elements of fitness for any biofilm-producing microbe (13,52), and they are especially important for the marine ecology of V. cholerae as it colonizes, consumes, disperses from, and then recolonizes chitin in its marine environment.

Bacterial cell shape can serve a broad range of adaptive functions (49)(50)(51). The curved shape of Caulobacter crescentus, for example, promotes the formation of biofilms as hydrodynamic forces reorient single cells to optimize daughter cell attachment (53); this process also nucleates clonal clusters under strong flow (54,55). Simulations and experiments with engineered variants of Escherichia coli suggest that rod-shaped bacteria can obtain a competitive advantage over spherical cells in colonies on agar plates, because rodshaped cells burrow underneath spherical cells and spread more effectively to access fresh nutrients on the colony periphery (56). Filamentation has been observed in a wide variety of bacteria and eukaryotic microbes; this morphology is implicated in assisting spatial spread through soil or host tissue and defense against phagocytosing amoeboid predators (50,57,58). In this work, we have shown that filamentation can alter surface colonization, biofilm architecture, and, as a result, the relative investment into rapid surface occupation versus long-term competitive success in biofilms.

The biophysical bases of our results are a topic of future work, but here we note that single filaments of V. cholerae can rapidly bend in shear flow, despite the stiffness of the bacterial cell wall (59). Our results demonstrate that sufficiently long bacteria can behave as elastic filaments (60). In analogy with the stretching behavior of polymers in flows that approach and split at the interface with an obstacle (i.e., extensional flows), we expect that filamentous bacteria experience shear that stretches them into alignment with stationary surfaces in the flow path (61)(62)(63). This phenomenon presumably increases the dwell time and contact area of filaments in proximity to obstacles in flow (SI Appendix, Fig. S12). We speculate that this process, in conjunction with previously known surface-adhesion mechanisms such as MSHA pili (46,47), promotes rapid attachment and wrapping of filaments around chitin particles (SI Appendix, Fig. S6).

Following surface attachment, filamentation allows the construction of biofilms in which cell-cell contacts generate a mesh network that is not dependent on the major polysaccharide or cell-cell adhesion components of the V. cholerae biofilm matrix. In this respect, filamentous biofilms may be analogous to a polymeric gel in which elongated cell bodies are directly associated through physical entanglement (64). Here, this cell network is more natively inclined to fast surface spreading and subsequent dispersal but also porous and prone to physical invasion by competing strains or species (SI Appendix, Fig. S10). This strategy of rapid colonization and biomass accumulation but high reversibility of surface association is particularly well suited to fluctuating environments in which resource patches are shortlived. This could be the case when particles are small and quickly consumed, or when disturbance events are common and destroy or disrupt particles with high frequency.

Understanding the transitions between individual cell behavior and collective properties that emerge among biofilm-dwelling bacteria is a topic of increasingly intense interest in the microbiology community. Our results here suggest that changes in cell shape are a fundamental element of this individual-to-collective transition. For V. cholerae, and potentially other microbes producing biofilms on and dispersing from particle to particle, filamentation may offer a surface-occupation strategy better suited to transient environments by shifting a biofilm's ability to spread over surfaces, at the expense of long-term competitive ability against matrix-replete, more highly adhesive cell groups. The ecological scope and impact of this cell morphology for V. cholerae and its relatives in realistic multispecies communities remain open questions for future work.

Methods for microfluidic device assembly, biological experiments, microscopy, image analysis, and statistical analysis can be found further described in SI Appendix, Materials and Methods.