Identification and analysis of the signaling pathways, matrix-digestion enzymes, and motility components controlling Vibrio cholerae biofilm dispersal

(200 words) Bacteria alternate between being free-swimming and existing as members of sessile 2 multicellular communities called biofilms. The biofilm lifecycle occurs in three stages: cell 3 attachment, biofilm maturation, and biofilm dispersal. Vibrio cholerae biofilms are hyper-infectious 4 and biofilm formation and dispersal are considered central to disease transmission. While biofilm 5 formation is well-studied, almost nothing is known about biofilm dispersal. Here, we conduct an 6 imaging screen for V. cholerae mutants that fail to disperse, revealing three classes of dispersal 7 components: signal transduction proteins, matrix-degradation enzymes, and motility factors. 8 Signaling proteins dominated the screen and among them, we focused on an uncharacterized 9 two-component sensory system that we name DbfS/DbfR for Dispersal of Biofilm 10 Sensor/Regulator. Phospho-DbfR represses biofilm dispersal. DbfS dephosphorylates and 11 thereby inactivates DbfR, which permits dispersal. Matrix degradation requires two enzymes: 12 LapG, which cleaves adhesins, and RbmB, which digests matrix polysaccharide. Reorientations 13 in swimming direction, mediated by CheY3, are necessary for cells to escape from the porous 14 biofilm matrix. We suggest that these components act sequentially: signaling launches dispersal 15 by terminating matrix production and triggering matrix digestion and, subsequently, cell motility 16 permits escape from biofilms. This study lays the groundwork for interventions that modulate V. cholerae biofilm dispersal to ameliorate disease.


Main 19
Bacteria transition between existing in the biofilm state, in which cells are members of 20 surface-associated multicellular collectives, and living as free-swimming, exploratory individuals. 21 Biofilms consist of cells surrounded by a self-secreted extracellular matrix that protects the 22 resident cells from threats including predation, antimicrobials, and dislocation due to flow. 1-3 23 Biofilms are relevant to human health because beneficial microbiome bacteria exist in biofilms, 24 and, during disease, because pathogens in biofilms evade host immune defenses, thwart medical 25 intervention, and exhibit virulence. [4][5][6][7] The biofilm lifecycle consists of three stages: cell 26 attachment, biofilm maturation, and dispersal (Fig. 1A). 8 Cells liberated during the dispersal step 27 can disseminate and found new biofilms. 8 The environmental stimuli and the components 28 facilitating biofilm attachment and maturation have been defined for many bacterial species. 9 In 29 contrast, little is known about the biofilm dispersal stage. 30 The model pathogen Vibrio cholerae forms biofilms in its aquatic habitat, biofilm cells are 31 especially virulent in mouse models of cholera disease, and biofilms are thought to be critical for 32 cholera transmission. [10][11][12][13][14] Studies of V. cholerae biofilms have predominantly focused on matrix 33 overproducing strains that constitutively exist in the biofilm mode and that do not disperse. This 34 research strategy has propelled understanding of V. cholerae biofilm attachment and maturation, 35 revealing that the second messenger cyclic diguanylate (c-di-GMP) is a master regulator of biofilm 36 formation, and that expression of vibrio polysaccharide (vps) biosynthetic genes are required. 15-37 17 The strategy of characterizing constitutive biofilm formers, while successful for uncovering 38 factors that promote biofilm formation, has necessarily precluded studies of biofilm dispersal. 39 Here, we employed a microscopy assay that allowed us to monitor the full wild-type (WT) V. 40 cholerae biofilm lifecycle. We combined this assay with high-content imaging of randomly 41 mutagenized WT V. cholerae to identify genes required for biofilm dispersal. Investigation of the 42 proteins encoded by the genes allowed us to characterize the signaling relays, matrix-digestion 43 enzymes, and motility components required for biofilm dispersal, a key stage in the lifecycle of 44 the global pathogen V. cholerae. 45

Results 46
Previously, we developed a brightfield microscopy assay that allows us to monitor the full 47 WT V. cholerae biofilm lifecycle in real time. 18 In our approach, V. cholerae cells are inoculated 48 onto glass coverslips at low cell density and brightfield time-lapse microscopy is used to monitor 49 biofilm progression. WT biofilms reach peak biomass after 8-9 h of incubation and subsequently 50 dispersal occurs and is completed by 12-13 h (Fig. 1B, C). To identify genes required for biofilm 51 dispersal, we combined mutagenesis with high-content imaging of the output of this assay. 52 Specifically, WT V. cholerae was mutagenized with Tn5 yielding ~7000 mutants that were arrayed 53 in 96-well plates. Following overnight growth, the mutants were diluted to low cell density in 54 minimal medium, a condition that drives initiation of the biofilm lifecycle. Brightfield images of each 55 well were captured 8 h post-inoculation to assess biofilm maturation and at 13 h to evaluate biofilm 56 dispersal. Mutants that showed no defects in biofilm maturation as judged by the 8 h images but 57 displayed significant remaining biofilm biomass at the 13 h timepoint were identified. To verify 58 phenotypes, candidate mutants were individually reevaluated by time-lapse microscopy. Mutants 59 that accumulated at the bottom of wells due to aggregation or that failed to attach to surfaces 60 were excluded from further analysis, eliminating strains harboring insertions in O-antigen and 61 flagellar genes, respectively. The locations of transposon insertions in the 47 mutants that met 62 our criteria were defined and corresponded to 10 loci. The new genes from the screen fell into 63 three classes: signal transduction (blue), matrix degradation (green), and motility (red) (Fig. 1A,  64 C). In-frame deletions of each gene were constructed, and the biofilm lifecycles of the deletion 65 mutants were imaged to confirm that the genes are required for biofilm dispersal (Table 1,  66 Supplementary Video 1). We also identified insertions in genes encoding proteins with known 67 roles in biofilm dispersal (i.e., RpoS, quorum sensing), which we excluded from further 68 analysis. 18  Proteins involved in signal transduction dominated the screen (7 of 10 loci) and included the  70  ribosome-associated GTPase, BipA, multiple cyclic diguanylate (c-di-GMP) signaling proteins,  71  polyamine signaling proteins, and a putative two-component histidine kinase, Vc1639. The signal  72 transduction mutants displayed different severities in their biofilm dispersal phenotypes. The 73 ∆bipA displayed a modest defect: ~19% of its biofilm biomass remained at 16 h, the final timepoint 74 of our data acquisition, while the WT showed ~6% biomass remaining. By contrast, the ∆vc1639 75 mutant underwent no appreciable dispersal (Table 1). In the category of matrix degradation, two 76 enzymes were identified, LapG a periplasmic protease, and RbmB, a putative polysaccharide 77 lyase (Table 1). A single motility mutant was identified with an insertion in the gene encoding the 78 chemotaxis response regulator cheY3 (Table 1). Below, we carry out mechanistic studies on 79 select mutants from each category to define the functions of the components. Other mutants will 80 be characterized in separate reports. 81 The mutant from our screen that exhibited the most extreme dispersal phenotype had a 83 transposon in a gene encoding an uncharacterized putative histidine kinase (designated HK), 84 Vc1639 (Table 1). A screen for factors required for V. cholerae colonization of the suckling mouse  85 intestine repeatedly identified Vc1639, suggesting that this HK is core to the cholera disease. 20 86 HKs typically contain periplasmic ligand binding domains and internal catalytic domains that 87 switch between kinase and phosphatase activities based on ligand detection. 21 HKs transmit 88 sensory information to cognate response regulators (RR) by altering RR phosphorylation. 22 RRs,89 in turn, control gene expression and/or behavior depending on their phosphorylation states. 90 Deletion of vc1639 in V. cholerae resulted in an 80% increase in peak biofilm biomass relative to 91 WT and nearly all the biofilm biomass remained at 16 h demonstrating that Vc1639 is essential 92 for biofilm dispersal ( Fig. 2A and induction of DbfS production caused the phospho-DbfR species to disappear (Fig. 2F). Thus, 124 under our experimental conditions, DbfS functions as a DbfR phosphatase. We infer that some 125 other unknown kinase must exist and phosphorylate DbfR (Fig. 2G). We propose that phospho-126 DbfR is active, and it drives expression of matrix biosynthetic genes, and increased 127  matrix production prevents biofilm dispersal. It is possible that phospho-DbfR also controls other 128 genes involved in suppressing biofilm dispersal. 129 BLAST analysis of the DbfS protein sequence against the Escherichia coli K-12 genome 130 revealed limited homology to the cation regulated HK, PhoQ, with 32% sequence identity (E 131 value=1e -41 ), with the lowest region of similarity in the predicted ligand binding domain. We tested 132 whether the ligands that control PhoQ signal transduction also regulate DbfS-DbfR signaling 133 (Extended Data Fig. 2A a yet-to-be defined stimulus to regulate biofilm dispersal. 136

Matrix disassembly mediates V. cholerae exit from biofilms 137
The second group of mutants in our screen harbored insertions in the gene encoding the 138 calcium-dependent periplasmic protease LapG that degrades outer-membrane spanning 139 adhesive proteins and in the gene specifying the extracellular polysaccharide lyase RbmB that 140 degrades the VPS component of the biofilm matrix. 23,24 The ∆lapG strain exhibited slightly lower 141 peak biofilm biomass compared to WT, with a short delay in the onset of dispersal, and ~55% of 142 its biomass remained at 16 h ( Fig. 3A, Table 1). The ∆lapG and the WT strains had similar vpsL-143 lux expression patterns ( Fig. 3B) consistent with LapG playing no role in repression of matrix 144 production, but rather functioning downstream in matrix degradation. The LapG mechanism is 145 known: When c-di-GMP concentrations are high, the FrhA and CraA adhesins are localized to 146 the outer membrane where they facilitate attachments that are important for biofilm formation (Fig.  147 3C). 25,26 Under this condition, LapG is sequestered and inactivated by the inner membrane c-di-148 GMP sensing protein LapD. 25 When c-di-GMP levels fall, LapD releases LapG, and LapG cleaves 149 FrhA and CraA facilitating cell detachment from biofilms. 25 Our results are consistent with this 150 mechanism; in the absence of LapG, FrhA and CraA remain intact, and V. cholerae cells cannot 151 properly exit the biofilm state. To verify that the established c-di-GMP-dependent regulatory 152 mechanism controls LapG activity in our assay, we deleted lapD (Fig. 3C). Indeed, in the ∆lapD 153 strain, biofilm dispersal occurred prematurely indicating that, without LapD, LapG is not 154 sequestered, and unchecked LapG activity promotes premature adhesin degradation, and, as a 155 consequence, early biofilm disassembly (Fig. 3D). The ∆lapD ∆lapG double mutant had the same 156 dispersal phenotype as the ∆lapG single mutant confirming that LapG functions downstream of 157 LapD (Fig. 3D). Lastly, in a reciprocal arrangement, overexpression of lapG from an ectopic locus 158 caused peak biofilm formation to decrease by ~65% (Extended Data Fig. 3A) suggesting that 159 enhanced LapG-mediated cleavage of adhesins prematurely released cells from the biofilm. 160 Thus, the conserved Lap pathway, which responds to changes in c-di-GMP levels, facilitates 161 biofilm dispersal in V. cholerae. 162 Regarding the RbmB polysaccharide lyase, the ∆rbmB strain formed biofilms to roughly 163 the same peak biomass as WT, however, it exhibited a 2 h delay in dispersal onset and most of 164 its biomass (~70%) remained at 16 h (Fig. 3E, Table 1). The level of vpsL-lux expression in the 165 ∆rbmB mutant was similar to the WT, showing that the RbmB dispersal function does not concern 166 production of VPS (Fig. 3F). Complementation with inducible rbmB expressed from an ectopic 167 locus in the ∆rbmB strain caused a ~40% reduction in peak biofilm formation, confirming that 168 RbmB negatively regulates biofilm formation, however the complemented strain retained a 169 modest biofilm dispersal defect, suggesting that the timing or level of rbmB expression is critical 170 for WT biofilm disassembly (Extended Data Fig. 3B). To verify that the ∆rbmB dispersal defect 171 stems from the lack of vps degradation, we grew ∆rbmB biofilms for 16 h (i.e., post WT biofilm 172  To account for differences in biomass, the WGA-txRed signal was divided by the 4', 6-diamidino-2-phenylindole (DAPI) signal in each biofilm. Values were normalized to the mean signal for the ∆lapG strain. >100 individual biofilms were quantified for each strain. An unpaired t-test was performed for statistical analysis, with **** denoting p < 0.0001. (H) Proposed model for the role of RbmB in biofilm dispersal. Gray lines represent the polysaccharide matrix. In all cases, N = 3 biological and N = 3 technical replicates, ± SD (shaded). a.u., arbitrary unit. For vpsL-lux measurements, N = 3 biological replicates, ± SD (shaded). RLU, relative light units. OM, outer membrane; IM, inner membrane. dispersal completion), and subsequently fixed and stained the non-dispersed biofilms with wheat 173 germ agglutinin conjugated to Texas Red (WGA-txRed), which binds to N-acetylglucosamine 174 sugars in the VPS matrix. 27 We used the ∆lapG mutant as our control since its biofilm dispersal 175 phenotype should not involve changes in VPS. On average, the ∆rbmB mutant exhibited ~6x 176 more WGA-txRed signal than the ∆lapG mutant (Fig. 3G). Collectively, our results show that the 177 non-dispersed ∆lapG biofilms contain little VPS, consistent with possession of functional RbmB, 178 while non-dispersed ∆rbmB biofilms contain excess VPS due to the lack of RbmB-mediated 179 polysaccharide digestion. Thus, we suggest that RbmB-directed VPS disassembly is critical for 180 proper biofilm disassembly (Fig. 3H). 181 Extracellular DNA (eDNA) is a component of the V. cholerae biofilm matrix and two 182 DNAses secreted by V. cholerae, Dns and Xds, digest eDNA. 28 Although we did not identify dns 183 and xds in our screen, we nonetheless investigated whether they contributed to biofilm dispersal. 184 Neither the ∆dns and the ∆xds single mutants, nor the ∆dns ∆xds double mutant displayed a 185 biofilm dispersal defect in our assay (Extended Data Fig. 3C), suggesting that eDNA digestion is 186 not required for dispersal. In a similar vein, we did not identify genes encoding the eight V. 187 cholerae extracellular proteases that could degrade matrix proteins. Consistent with this finding, 188 measurement of the phenotypes of mutants deleted for each extracellular protease gene showed 189 that none exhibited a dispersal defect. Thus, no single extracellular protease is required for biofilm 190 dispersal (Extended Data Fig. 3D). It remains possible that proteases contribute to biofilm 191 dispersal by functioning redundantly. Together, our results indicate that two enzymes, LapG and 192 RbmB, are the primary matrix degrading components that enable biofilm dispersal. 193

Reorientations in swimming direction are required for biofilm dispersal. 194
The final category of genes identified in our screen are involved in cell motility. As noted 195 above, non-motile mutants were excluded from analysis because they are known to be impaired 196 in surface attachment. Nonetheless, we identified a mutant containing a transposon insertion in 197 cheY3 as defective for biofilm dispersal. cheY3 is one of the five V. cholerae cheY genes 198 specifying chemotaxis RR proteins. 29 Notably, cheY3 is the only V. cholerae cheY homolog 199 required for chemotaxis. 29 The ∆cheY3 mutant exhibited similar peak biofilm timing and biomass 200 as WT V. cholerae, however, ~21% biomass remained at 16 h (Fig. 4A, Table 1). Expression of 201 vpsL-lux in the mutant was identical to the WT indicating that the dispersal phenotype was not 202 due to elevated matrix production (Fig. 4B). 203 The V. cholerae default motor rotation direction is counterclockwise (CCW), which fosters 204 smooth, straight swimming. 30 Transition to clockwise (CW) motor rotation causes reorientations 205 in swimming direction. 30 Phospho-CheY3 binds to the flagellar motor switch complex to mediate 206 the change from CCW to CW rotation. Thus, the ∆cheY3 mutant is non-chemotactic and the cells 207 are locked in the CCW, straight swimming mode (Fig. 4C). We reasoned that the ∆cheY3 mutant 208 dispersal defect could stem from an inability to chemotact or from an inability to reorient swimming 209 direction. To distinguish between these possibilities, we examined biofilm dispersal in a V. 210 cholerae mutant carrying a cheY3 allele, cheY3 D16K, Y109W (henceforth, cheY3*) that locks the 211 motor into CW rotation and so also disrupts chemotaxis. cheY3* cells undergo frequent 212 reorientations and are unable to swim in smooth straight runs (Fig. 4C). 29,31 The cheY3* strain 213 had WT biofilm dispersal capability. Thus, being chemotactic is not required for V. cholerae to exit 214 biofilms (Fig. 4A). 215 We reasoned that analysis of the unique motility characteristics of our strains could reveal 216 the underlying causes of the ∆cheY3 biofilm dispersal defect. We measured the turning 217 frequencies and swimming velocities of the WT, ∆cheY3, and cheY3* V. cholerae strains. 218 Consistent with previous reports, these three mutants exhibited notable differences: on average, 219 the WT turned once every 3 s, the ∆cheY3 mutant turned less than once every 40 s, and the 220 cheY3* strain turned once every 0.5 s ( Fig. 4C and D). 29,31 The cheY3* strain displayed slightly 221 lower average swimming velocity than the WT and ∆cheY3 strains, due to its high turning 222 frequency as turning necessarily involves a decrease in velocity (Fig. 4E). 32 Together, these 223 results suggest that the low turning frequency of the ∆cheY3 mutant is responsible for the biofilm 224 dispersal defect. We propose that if cells do not frequently change their direction of motion, they 225 become trapped by the biofilm matrix mesh which compromises their ability to escape (Fig. 4F). 226 Indeed, in other bacteria, straight-swimming mutants are deficient in traversing fluid-filled porous 227 media compared to WT organisms that can reorient. 33 Together, these results indicate that 228 chemotaxis itself is not required for biofilm dispersal, but, rather, that the chemotaxis machinery 229 facilitates random reorientation events that allow V. cholerae cells to navigate a porous biofilm 230 matrix. The same non-chemotactic mutants used here exhibit stark differences in competition 231 experiments in animal models of cholera infection, showing that their differences in motility and, 232 possibly, their differences in biofilm dispersal capabilities, are pertinent to colonization. 31 233 Finally, we determined whether the ability to locomote was required for biofilm dispersal 234 or, by contrast, if non-motile cells could escape the digested matrix via Brownian motion. As 235 mentioned above, we could not simply study dispersal of non-flagellated and non-motile mutants 236 because of their confounding surface attachment defects and feedback on biofilm regulatory 237 components. 34,35 To circumvent this problem, we employed phenamil, an inhibitor of the Na + -238 driven V. cholerae flagellar motor, which, as expected, dramatically reduced planktonic cell 239 motility (Extended Data Fig. 4). 36 To assess the role of swimming motility in biofilm dispersal, we 240 first allowed WT V. cholerae cells to undergo biofilm formation for 5 h, at which point we perfused 241 DMSO or phenamil into the incubation chamber (Fig. 4G)   For biofilm biomass assays, N = 3 biological and N = 3 technical replicates, ± SD (shaded). a.u., arbitrary unit. For vpsL-lux measurements, N = 3 biological replicates, ± SD (shaded). RLU, relative light units. For motility measurements, 45-125 individual cells of each strain were tracked. In panels D and E, unpaired ttests were performed for statistical analysis, with P values denoted as *P < 0.05; **P < 0.01; *** P < 0.001; ****P < 0.0001; n.s., P > 0.05. frequency. DNA fragments that were not native to V. cholerae were synthesized as g-blocks (IDT,  279 Coralville, IA, USA). 280

Bacterial Strains and Reagents
All V. cholerae strains generated in this work were constructed by replacing genomic DNA 281 with DNA introduced by natural transformation as previously described. 18,37 PCR and Sanger 282 sequencing were used to verify correct integration events. Genomic DNA from recombinant 283 strains was used for future co-transformations and as templates for PCR to generate DNA 284 fragments, when necessary. Deletions were constructed in frame and eliminated the entire coding 285 sequences. The exceptions were mbaA, dbfS, and dbfR, which each overlap with another gene 286 in their operons. In these cases, portions of the genes were deleted ensuring that adjacent genes 287 were not perturbed. For tagA, the first 103 base pairs, including the nucleotides specifying the 288 start codon, were deleted. All strains constructed in this study were verified by sequencing at 289 Genewiz. 290

Microscopy and Mutant Screening 291
The biofilm lifecycle was measured using time-lapse microscopy as described 292 previously. 18 All plots were generated using ggplot2 in R. replicates were averaged and plotted using ggplot2 in R. 323

VPS Quantitation 324
To assess VPS levels in non-dispersed biofilms using WGA-txRED, biofilms were grown for 16  single-cell locomotion trajectories were calculated as described. 38 Curvature of less than 0.3 μm -359 1 was used to identify the turning events. MSD was calculated as described previously. 39 360

Phos-tag Gel Analysis 361
To monitor DbfR and phospho-DbfR via SDS-PAGE, the endogenous dfbR gene was 362 replaced with dbfR-SNAP in the ∆dbfS strain, and PBAD-dbfS was introduced at the ectopic locus, 363 vc1807. To assess DbfR-SNAP phosphorylation in the absence and presence of DbfS, overnight 364 cultures of the strain were diluted 1:1000 and subsequently grown for 4 h at 30 o C with shaking 365 to an OD600 ~ 0.6. To each culture, 1 µM SNAP-Cell TMR Star (New England Biolabs) was added 366 to label the SNAP tag, and the culture was subsequently divided into two tubes. To one tube, 367 0.2% D-fucose was added, and to the other, 0.2% L-arabinose was added to repress and induce 368 DbfS production, respectively. The cultures were returned to 30 o C with shaking. After 1 h, the Healthcare) using a Cy3 filter set. 377 Extended Data Fig. 1. Complementation, functional tagging, and mutagenesis  PvpsL-lux outputs for strains and growth conditions in B over the growth curve. (D) As in B except following the addition of water or 5 µg/mL C18G. In all cases, N = 3 biological and N = 3 technical replicates, ± SD (shaded). a.u., arbitrary unit. For vpsL-lux measurements, N = 3 biological replicates, ± SD (shaded). RLU, relative light units. Fig. 3 Extended Data Fig. 3. Introduction of lapG and rbmB complements the ∆lapG and ∆rbmB  biofilm defects, respectively, and

DbfS is not equivalent to PhoQ
In E. coli, low Mg 2+ and cationic peptides activate PhoQ kinase activity. 40 Sequence alignment of the DbfS sensory domain with that from PhoQ of E. coli, Salmonella enterica, and Pseudomonas aeruginosa revealed that DbfS lacks all of the key residues involved in Mg 2+ binding (Extended Data Fig. 2A). 41 To test if Mg 2+ alters DfbS activity, we measured the V. cholerae biofilm lifecycle in response to low Mg 2+ conditions in WT V. cholerae and in the ∆dbfR mutant. If, analogous to PhoQ, DfbS kinase activity is activated by low Mg 2+ , when Mg 2+ is limiting, WT V. cholerae should exhibit an altered biofilm dispersal phenotype while the ∆dbfR mutant would be impervious to Mg 2+ changes. 40 Extended Data Fig. 2B shows that Mg 2+ limitation does indeed inhibit V. cholerae biofilm dispersal, however, inhibition occurs in both the WT and the ∆dbfR strains. Mg 2+ limitation did not alter vpsL-lux expression in either strain (Extended Data Fig. 2C). Thus, Mg 2+ does not control DfbS activity. We obtained the same results following exogenous addition of the cationic peptide C18G (Extended Data Fig. 2D). Together, these results demonstrate that DfbS does not respond to the ligands that control PhoQ activity.

Supplementary Table 1
Strains used in this study.