A Genome-Wide Analysis of Adhesion in Caulobacter crescentus Identifies New Regulatory and Biosynthetic Components for Holdfast Assembly

Bacteria routinely encounter biotic and abiotic materials in their surrounding environments, and they often enlist specific behavioral programs to colonize these materials. Adhesion is an early step in colonizing a surface. Caulobacter crescentus produces a structure called the holdfast which allows this organism to attach to and colonize surfaces. To understand how the holdfast is produced, we performed a genome-wide search for genes that contribute to adhesion by selecting for mutants that could not attach to cheesecloth. We discovered complex interactions between genes that mediate surface contact and genes that contribute to holdfast development. Our genetic selection identified what likely represents a comprehensive set of genes required to generate a holdfast, laying the groundwork for a detailed characterization of the enzymes that build this specialized adhesin.

ing environment (2). The roles of surface polysaccharides such as capsules, exopolysaccharides, and O-antigens in promoting colonization of preferred niches are well established for both free-living and host-associated bacteria (3)(4)(5). However, the enzymes that synthesize and export these polysaccharides have been difficult to characterize due to the chemical complexity of the metabolic intermediates (6). Defining the molecular details of how extracellular carbohydrates are produced is critical to understanding bacterial colonization and how it can be controlled.
The aquatic bacterium Caulobacter crescentus has a dimorphic lifestyle characterized by an association with exogenous surfaces. Division in C. crescentus is asymmetric and produces two distinct cell types: a chemotactic swarmer cell and a sessile, replicationcompetent stalked cell (7). In response to environmental and developmental signals, the swarmer cell sheds its flagellum, disassembles its pili, and transitions into its stalked-cell form before dividing (8). Stalked cells are named for a specialized envelope extension called the stalk that emerges from the old pole after disassembly of the flagellum and pili. During the swarmer-cell-to-stalked-cell transition, cells often produce a polysaccharide-rich matrix called the holdfast at the site of stalk development (9). This highly adhesive material allows C. crescentus to form essentially permanent interactions with exogenous surfaces (Fig. 1A) (10).
Due to the irreversible nature of surface attachment in C. crescentus, the timing of holdfast production is tightly controlled. Only small proportions of cells produce a holdfast under conditions of growth in defined medium, whereas nearly all cells produce a holdfast under conditions of growth in complex medium (11). This effect is due to elevated expression of the holdfast inhibitor A (hfiA) gene in defined medium (Fig. 1B). hfiA expression is also coordinated with the cell cycle. Its transcript levels drop during the swarmer to stalked transition, which corresponds to the developmental stage at which holdfasts begin to appear. Numerous signaling pathways target the hfiA promoter, allowing the cell to integrate environmental, nutritional, and developmental cues into a single output that regulates adhesion (11,12). In yet another regulatory regime, holdfast synthesis can be induced when a swarmer cell encounters a surface (13). Physical disruption of flagellar rotation or pilus retraction upon surface contact stimulates the production of a holdfast (14,15). How the numerous regulatory pathways converge to control holdfast development remains unclear, but the complexity of these networks reflects the significance of committing to a surface-associated lifestyle. Genetic analysis of nonadhesive mutants indicates that the holdfast is a polysaccharide-based material. The holdfast synthesis (hfs) genes include those encoding predicted glycosyltransferases, carbohydrate modification factors, and components of a wzy-type polysaccharide assembly pathway (16)(17)(18)(19). wzy-dependent carbohydrate assembly utilizes a lipid carrier known as undecaprenylpyrophosphate (UPP) on which glycosyltransferases assemble an oligosaccharide repeating unit in the cytoplasm (20). The resulting glycolipid is flipped from the cytoplasmic face of the inner membrane to the periplasmic face where the oligosaccharide is polymerized and exported to the cell surface (21). The wzy mechanism is used to produce an impressive diversity of polysaccharides and is broadly conserved among bacteria (22). Thus, characterization of enzymes involved in the biosynthesis of the holdfast has the potential to uncover broadly applicable principles about how bacteria produce carbohydrate polymers.
The chemical nature of the holdfast matrix remains poorly characterized. The holdfast binds to N-acetylglucosamine (GlcNAc)-specific lectin wheat germ agglutinin (WGA) and is sensitive to the GlcNAc-specific hydrolases chitinase and lysozyme, indicating that GlcNAc is a component of the matrix (23). Little other information about the carbohydrate content has been reported. Extracellular DNA and an unidentified protein component(s) contribute to the stiffness of the holdfast, but only mutations in polysaccharide biosynthesis genes or genes encoding pleiotropic regulators of cell polarity abolish holdfast production (24). Currently, three glycosyltransferase steps are known to be required for holdfast synthesis. An initial reaction carried out by a genetically redundant HfsE, PssY, or PssZ enzyme is thought to be followed by the activities of HfsG and HfsJ, suggesting that the polysaccharide may be composed of a trisaccharide repeat (11,18). However, new hfs genes continue to be discovered, hinting that additional glycosyltransferases may remain unidentified (11,25). Uncertainty about both the composition of the holdfast and the saturation of screens for hfs genes presents a major obstacle to characterizing enzymatic reactions in the pathway.
Here, we utilized saturating transposon mutagenesis to probe holdfast production at the genome scale. We developed a barcoded transposon library in C. crescentus and enriched for nonadhesive mutants by performing passaging across multiple days in the presence of cheesecloth. We discovered a surprising number of genes with distinct adhesion phenotypes that ranged from hyperadhesive to nonadhesive. We found that disrupting the smooth lipopolysaccharide (SLPS) leads to a holdfast-independent form of ectopic adhesion that is not restricted to the cell pole but is instead mediated throughout the cell surface. The temporal adhesion profiles of known SLPS mutants were used to identify and characterize new genes in the SLPS pathway. The same fitness correlation approach was used to place previously uncharacterized genes in the holdfast pathway. We further demonstrated that disrupting the assembly of polar surface appendages modulates the activity of the hfiA holdfast inhibitor. In particular, individual mutations in the pilus machinery had a range of adhesion phenotypes, suggesting that distinct intermediates in the pilus assembly pathway have opposing effects on hfiA. On the basis of our comprehensive analysis of holdfast regulation, biosynthesis, and assembly, we propose a model that outlines the sequence of enzymatic steps required to produce the holdfast polysaccharide.

RESULTS
A screen for mutants with altered adhesion characteristics. The holdfast promotes adhesion of C. crescentus cells to a variety of surfaces (26). We reasoned that adhesive cells could be depleted from liquid cultures by adding an attachment substrate with a sufficiently large surface area. Cheesecloth has been used in this manner to enrich for holdfast mutants in both C. crescentus and Asticcacaulis biprosthecum, another stalked bacterium in the Caulobacteraceae clade (27,28). Adding sterile cheesecloth to wild-type C. crescentus cultures decreased the turbidity of the medium by titrating adhesive cells from the broth. This effect was amplified in the hyperadhesive ΔhfiA strain but not observed in the holdfast-deficient ΔhfsJ strain, demonstrating the effectiveness of cheesecloth at capturing cells with a holdfast (Fig. 1B). We concluded that growth in the presence of cheesecloth could be used as the basis of a selection to identify mutants defective in adhesion.
Saturating transposon mutagenesis coupled with transposon insertion sequencing (TnSeq) offers the advantage of scoring phenotypes for all nonessential genes in the genome simultaneously (29). Thus, combining TnSeq-based mutant profiling with cheesecloth depletion seemed appropriate to perform a saturating screen for holdfast biosynthesis genes and to identify missing biosynthesis factors. We developed a randomly barcoded transposon library in C. crescentus to enable the use of BarSeq (30) for profiling mutant fitness. Adhesive cells were depleted by passaging the library in cheesecloth for five cycles. During each passage, the library was cultured for 24 h in the presence of cheesecloth, after which the unattached cells in the medium were used to reinoculate a fresh culture containing cheesecloth. An aliquot of unattached cells in the medium was also harvested for BarSeq analysis. Three passaging experiments with cheesecloth were performed in parallel. To discriminate mutants with adhesion defects from those with growth defects, we also performed three passaging experiments without cheesecloth for comparison (see Table S1 in the supplemental material).
The abundance of each mutant during the passaging steps was assessed using BarSeq, providing a temporal fitness profile for each gene over the course of the experiment (30). Genes with positive fitness scores reflect mutants with adhesion defects that were enriched in medium that had been depleted by the use of cheesecloth. Genes with negative fitness scores represent hyperadhesive mutants that are depleted more efficiently by cheesecloth than is the wild type. As expected, most genes had inconsequential effects on adhesion. However, there were a significant number of genes whose mutation caused strong cheesecloth-dependent changes in abundance over the course of the multiday experiment. The 250 genes with the highest fitness values (above or below the baseline level) are shown in Fig. 1C. The time-resolved nature of the experiment allowed us to group mutants with similar fitness profiles into distinct classes. Mutants with similar temporal adhesion profiles often mapped to genes with similar or complementary annotations, indicating that they represented groups of functionally related genes. We predicted the functions of uncharacterized genes using known functions for genes with similar fitness profiles. For example, a cluster of genes whose mutants showed a continuous increase in abundance after each passage contained many known hfs genes, and any uncharacterized genes that shared this fitness profile would be predicted to contribute to holdfast synthesis as well. We identified four clusters containing mutants that displayed distinct fitness profiles for the cheesecloth passaging experiment. Each cluster is described in detail below.
Mutants defective in smooth lipopolysaccharide display ectopic adhesion. We identified a cluster of mutants with strong fitness decreases (i.e., with increased adhesion to cheesecloth) in early passages with relative abundances that recovered to nearly neutral or even positive fitness values as passaging proceeded ( Fig. 2A). Many of the genes in this "recovery" cluster had annotations associated with polysaccharide biosynthesis but had no known cellular function. However, the wbq genes that are required for the production of smooth lipopolysaccharide (SLPS) comprised a subset of the recovery cluster (31). This suggested that mutants sharing this fitness profile might also be defective in the biosynthesis of SLPS.
We focused on three uncharacterized genes in the recovery cluster. C. crescentus CCNA_00497 is annotated as encoding a putative rhamnosyl transferase, CCNA_02386 is annotated as encoding an O-antigen ligase, and CCNA_03744 is homologous to rfbB, a gene required for the biosynthesis of dTDP-L-rhamnose (32). Mutations in rfbB were previously shown to suppress the holdfast attachment defect observed in a ΔhfaD mutant, but SLPS was not examined in these mutants (33). We created in-frame deletions of CCNA_00497, CCNA_02386, and rfbB and analyzed SLPS production by immunoblotting. A deletion of wbqP, which is thought to encode the initial glycosyltransferase step in the O-polysaccharide biosynthesis pathway, was used as a positive control. Disruption of CCNA_02386, rfbB, or wbqP led to the loss of detectable SLPS, and ΔCCNA_00497 cells showed a reduction in SLPS levels (Fig. 2C). Additionally, all four mutants released the S-layer protein RsaA into the spent medium, an additional hallmark of SLPS defects in C. crescentus (Fig. 2C) (34). None of the mutants displayed  Table S5. wbqP was not characterized in our library due to low insertion density in this region. (B) Surface attachment of SLPS mutants measured by CV staining. Cultures were grown for 24 h in M2X medium before staining of surface-attached cells. Disruption of SLPS led to increased adhesion in a holdfastindependent fashion. The graph shows averages Ϯ standard deviations of results from five biological replicates. Statistical significance was assessed by analysis of variance (ANOVA) with a pairwise Dunnett's posttest to determine which samples differed from the wild type. **, P Ͻ 0.01; ****, P Ͻ 0.0001. (C) Smooth LPS production was disrupted in mutants ΔCCNA_00497, ΔCCNA_02386, and ΔrfbB. (Top) Western blot (WB) to detect SLPS. (Middle) Total protein stained with Ponceau S as a loading control for SLPS blotting. (Bottom) Coomassie staining of spent medium from the cultures. Each mutant showed a loss of or a decrease in SLPS production by Western blotting and released the S-layer protein RsaA into the spent medium. The cell-surface defects can be complemented by ectopic expression of the appropriate gene. "Empty" refers to plasmid control strains.
Genome-Wide Analysis of Adhesion ® observable changes in rough LPS, demonstrating that they were not defective in the production of lipid A or the core oligosaccharide (see Fig. S1 in the supplemental material). All of the defects could be complemented by ectopic expression of the target gene, confirming their roles in the production of SLPS ( Fig. 2C; see also Table S2).
The fitness profiles for early stages of cheesecloth passaging suggested that disrupting SLPS led to production of hyperadhesive cells that were rapidly depleted by cheesecloth. In complex medium, nearly all of the wild-type cells produce a holdfast, making the dynamic range for detecting increased adhesion quite small. Thus, we chose to investigate potential hyperadhesive phenotypes by examining mutants using a defined medium in which fewer cells produce a holdfast. We found that the SLPS mutants were indeed hyperadhesive, producing crystal violet (CV) staining values ranging from two to four times the level seen with the wild type (Fig. 2B). To understand the basis of hyperadhesion, ΔwbqP cells were imaged after staining with fluorescently labeled wheat germ agglutinin (fWGA) to label holdfasts. Most wild-type cells displayed a fluorescent focus at the tip of the stalk, and stalks from multiple cells often aggregated around a single focus to form the rosette structures that are characteristic of holdfast production. In the ΔwbqP background, a comparable number of cells produced a holdfast, but the structure of the rosettes was altered. Cells that assembled around a holdfast were more tightly packed, and not all of them adhered to the aggregates through the tip of the stalk (Fig. 2D).
The unusual rosette structures in the ΔwbqP mutant suggested that cells with disrupted SLPS might have a second mode of adhesion that did not require a holdfast. We compared fWGA staining in the holdfast-deficient ΔhfsJ strain to fWGA staining of a ΔhfsJ ΔwbqP double mutant that lacks both holdfast and SLPS. ΔhfsJ cells did not stain with fWGA and did not form aggregates. ΔhfsJ ΔwbqP cells did not stain with fWGA, but, in contrast to the ΔhfsJ strain, the cells still formed aggregates (Fig. 2D). These aggregates appeared not to be mediated by stalk-stalk interactions but rather through interactions with the cell body. This further supported the idea that a holdfastindependent mode of ectopic adhesion operates in SLPS mutants. Consistent with this model, bulk adhesion in the ΔhfsJ ΔwbqP double mutant was not abolished; the level was, in fact, higher than that seen with the wild type (Fig. 2B). We conclude that disrupting SLPS production caused defects in the cell surface leading to a holdfastindependent mode of adhesion that represented the dominant mode of adhesion for these mutants early in our experimental time course.
The flagellum and type IV pili regulate holdfast production. A second cluster of mutants primarily contained genes known to participate in chemotaxis and flagellar motility as well as genes corresponding to components of the type IV pilus machinery. The fitness profiles for these mutants suggested that disrupting the assembly of polar appendages, either pili or flagella, leads to hyperadhesion. To study the effects of polar appendages on adhesion, we deleted the genes for the flagellar basal body component FlgH and the pilus assembly protein CpaH (35,36). We confirmed that the ΔflgH mutant showed the expected loss in motility and that the ΔcpaH mutant was resistant to the type IV pilus-specific phage ⌽CBK ( Fig. S2 and S3).
Both the ΔflgH and the ΔcpaH mutants showed adhesion defects in complex medium (Table S2). However, in defined medium, both mutants displayed increased adhesion relative to the wild type, indicating that disrupting the pilus or the flagellum causes hyperadhesion under these conditions (Fig. 3B). To reconcile these differences, we used fWGA staining to measure the proportion of cells that produced a holdfast. The ΔflgH and ΔcpaH mutants produced more holdfasts than the wild type in both complex medium and defined medium ( Fig. S3; see also Table S3). We conclude that flagellum and pilus mutations increase holdfast production but that loss of either appendage also leads to holdfast-independent defects in surface colonization. Pili and flagella often have similar effects on surface colonization in other systems (37,38). Because the baseline level of holdfast production is low in defined medium, the significance of the enhanced holdfast production seen in the ΔflgH and ΔcpaH backgrounds appears to outweigh that of surface colonization defects under these conditions (Fig. S3).
A recent report showed that flagellar hook mutants displayed decreased transcription from the hfiA promoter (P hfiA ) in defined medium and that this effect did not occur in the absence of the pleiotropic cell cycle regulator PleD (39)(40)(41). This led us to examine the relationships between our polar appendage mutants, hfiA and pleD. We used a P hfiA -lacZ reporter to measure transcription from the hfiA promoter. Because expression from P hfiA is low in complex medium, the dynamic range for measuring decreased activity is small. Therefore, we focused on transcriptional changes that occurred in defined medium where the baseline activity of P hfiA was high. The ΔflgH and ΔcpaH mutants showed reduced hfiA transcription in the reporter assay (Fig. 3D). This modest reduction in P hfiA -lacZ reporter signal was statistically significant, and changes of this magnitude are known to affect holdfast development (11). The decrease in P hfiA -lacZ signal was abrogated in the ΔflgH ΔpleD and ΔcpaH ΔpleD double mutants (Fig. 3D). Likewise, bulk adhesion in defined medium reverted to near wild-type levels in the ΔflgH ΔpleD and ΔcpaH ΔpleD mutants, confirming that pleD contributed to the modulation of P hfiA in the pilus and flagellar mutants (Fig. 3C). We note, however, that  Table S5. (B) Surface attachment of motility mutants measured by CV staining. Cultures were grown for 17 h in M2X medium before staining of surface-attached cells. Deletion of the genes for either the outer membrane flagellar base protein FlgH or the inner membrane type IV pilus component CpaH caused increased adhesion. In both mutant backgrounds, hfsJ is required for attachment. The graph shows averages Ϯ standard deviations of results from five biological replicates. (C) Effect of hfiA and pleD deletions on surface attachment of cpaH and flgH mutants. The ΔhfiA mutation did not affect adhesion in the ΔflgH background and increased adhesion in the ΔcpaH background. The increased adhesion in both the ΔflgH and ΔcpaH mutants can be eliminated by deletion of pleD. The graph shows averages Ϯ standard deviations of results from four biological replicates. (D) P hfiA -lacZ reporter activity in polar appendage mutants. The chart shows averages Ϯ standard deviations of results from four biological replicates. Statistical significance was assessed by ANOVA with a pairwise Dunnett's posttest to determine which samples differed from the wild type. nd, not detected; ns, not significant; *, P Ͻ 0.05; ****, P Ͻ 0.0001. Where appropriate, P values for additional pairwise comparisons pertinent to interpretation are indicated in the text.
Genome-Wide Analysis of Adhesion ® a full reversion of the hyperadhesive-holdfast phenotype would be predicted to display bulk adhesion levels below that of the wild type due to the holdfast-independent attachment defects seen in pilus and flagellum mutant backgrounds. Thus, while pleD did contribute to the enhanced adhesion seen in the ΔflgH and ΔcpaH mutants, the effect was not completely dependent on the presence of this gene.
To test whether the hyperadhesive phenotypes in the polar appendage mutants could be explained by repression of hfiA, we created ΔflgH ΔhfiA and ΔcpaH ΔhfiA double mutants. Bulk attachment in the ΔflgH ΔhfiA strain was not significantly increased relative to the level seen with the ΔflgH single deletion (Fig. 3C). This suggests that the ΔflgH mutation effectively inactivated the effects of hfiA and that holdfastindependent adhesion defects lowered the maximum level of surface attachment that was achievable in a ΔflgH background. The enhanced attachment seen in the ΔcpaH background was further increased in a ΔcpaH ΔhfiA double mutant (P Ͻ 0.0001; Fig. 3C). Thus, the ΔcpaH mutation had an intermediate effect on hfiA activity by dampening but not completely masking its activity. Consistent with this, the fraction of ΔcpaH cells that produced a holdfast in defined medium was intermediate between those of the wild-type and ΔhfiA mutant strains, supporting the idea that the ΔcpaH mutation caused both intermediate enhancement of holdfast production and holdfastindependent surface-attachment defects ( Fig. S3; see also Table S3). Finally, the level of bulk adhesion in a ΔflgH ΔcpaH double mutant was indistinguishable from that seen with the ΔflgH mutant, and P hfiA transcription was lower in the ΔflgH ΔcpaH mutant than in either the ΔflgH single mutant (P Ͻ 0.0001) or the ΔcpaH single mutant (P Ͻ 0.0001) ( Fig. 3B and C). These results indicate that holdfast production was likely maximized in the ΔflgH mutant and that the ΔflgH and ΔcpaH mutants modulated P hfiA through separate pathways.
A complex role for the pilus in regulating adhesion. The ΔcpaH phenotype suggested that disruption of pilus assembly leads to increased holdfast production via the repression of hfiA. However, a closer examination of the fitness profiles for genes involved in type IV pilus assembly revealed a range of phenotypes for various components of the apparatus (Fig. 4A). Most of the genes encoding components of the pilus secretion machinery, including cpaH, had fitness profiles consistent with increased adhesion. However, mutations in the gene coding for the main pilin subunit, PilA, displayed the opposite trend. pilA mutants had fitness profiles that would be expected for mutants with adhesion defects. We confirmed that, indeed, the ΔpilA strain was defective in surface attachment in both complex and defined medium ( Fig. 4B; see also Table S2).
To examine the relationship between pilA-dependent loss of adhesion and the activation of adhesion observed in mutants that disrupt pilus and flagellum assembly, we created ΔflgH ΔpilA and ΔcpaH ΔpilA double mutants. The phenotypes for these mutants were similar in both complex medium and defined medium (Table S4). Surface attachment levels in the ΔflgH ΔpilA mutant were intermediate with respect to those of the ΔpilA mutant (P Ͻ 0.0001) and the ΔflgH mutant (P Ͻ 0.0001), suggesting that flgH and pilA regulate adhesion through independent, additive pathways ( Fig. 4C; see also Table S4). In contrast, adhesion in the ΔcpaH ΔpilA mutant was indistinguishable from that in the ΔpilA mutant, demonstrating that the effects of pilA on adhesion were epistatic with respect to those of cpaH ( Fig. 4C; see also Table S4). We conclude that pilin subunit PilA is required for the holdfast-promoting effect caused by disruption of the pilus assembly apparatus.
We further explored the model that pilA contributed to the modulation of adhesion by cpaH by examining the relationships between pilA, hfiA, and pleD. The decreased adhesion observed in the ΔpilA mutant was not affected by the subsequent deletion of pleD, indicating that the effect of pilA on adhesion is pleD independent. Adhesion in a ΔpilA ΔhfiA double mutant was elevated to a level slightly below that of the ΔhfiA strain (P Ͻ 0.0001; Fig. 4B). The difference between the ΔpilA ΔhfiA and ΔhfiA mutants with respect to the levels of surface attachment likely reflects the holdfast-independent surface attachment defects caused by the loss of a functional pilus. Expression from P hfiA was slightly elevated in the ΔpilA mutant (Fig. 4D). However, because hfiA was already highly expressed under those conditions, it is difficult to determine if P hfiA is activated further in the ΔpilA mutant. Finally, the ␤-galactosidase activity shown by the P hfiA reporter was restored to wild-type levels in both the ΔflgH ΔpilA and ΔcpaH ΔpilA backgrounds, demonstrating that pilA is required to lower hfiA expression in the ΔflgH and ΔcpaH mutants (Fig. 4D).
New factors in the holdfast biosynthesis pathway. The final two clusters of mutants presented in Fig. 1C displayed fitness profiles consistent with adhesion defects. We used the magnitude of the measured fitness changes to separate these genes into (i) a cluster with higher fitness changes that contained all of the hfs genes known to be required for robust adhesion and (i) a cluster displaying lower fitness changes which contained the hfsK gene. hfsK encodes a putative N-acyltransferase thought to modify the holdfast polysaccharide in order to produce a fully adhesive holdfast (25). We chose three uncharacterized genes from these clusters of mutants for detailed examination of holdfast defects.
Disruption of CCNA_01242, which encodes a predicted amino acid permease, led to the strongest nonadhesion fitness profile of any gene in the cheesecloth passaging Staining is slightly lower for the ΔhfiA ΔpilA mutant than for the ΔhfiA mutant, reflecting the holdfast-independent defect in surface attachment that occurred when the pilus was disrupted. pleD has no effect on adhesion in the ΔpilA mutant. The graph shows averages Ϯ standard deviations of results from six biological replicates. (D) P hfiA -lacZ reporter activity in various pilA mutants. The chart shows averages Ϯ standard deviations of results from four biological replicates. Statistical significance was assessed by ANOVA with a pairwise Dunnett's posttest to determine which samples differed from the wild type. ns, not significant; **, P Ͻ 0.01; ****, P Ͻ 0.0001. Where appropriate, P values for additional pairwise comparisons pertinent to interpretation are indicated in the text.
Genome-Wide Analysis of Adhesion ® experiment (Fig. 5A). However, the ΔCCNA_01242 strain had only a modest defect in surface attachment (Fig. 5B). There were no obvious holdfast defects in the mutant, and we could not detect a significant adhesion defect under any conditions tested ( Fig. 5C; see also Table S2). Instead, ΔCCNA_01242 had an unusual, biphasic growth profile. In complex medium, the log phase was shorter than that shown by the wild type, leading to a lower level of optical density (OD) as growth began to slow prematurely. Growth of this strain continued slowly over the next 24 h and eventually plateaued at an OD similar to that seen with the wild type (Fig. S4). Such biphasic growth seems to confound fitness calculations for samples collected during the sequential passaging  Table S5. (B) Surface attachment of putative holdfast mutants measured by CV staining. Cultures were grown for 24 h in PYE medium before being stained. The ΔCCNA_01242 and ΔCCNA_02722 mutants displayed reduced staining, and the ΔCCNA_02360 mutant was nonadhesive. The graph shows averages Ϯ standard deviations of results from five biological replicates. Statistical significance was assessed by ANOVA with a pairwise Dunnett's posttest to determine which samples differed from the wild type. nd, not detected; ns, not significant; ****, P Ͻ 0.0001. (C) Analysis of holdfast phenotypes by fWGA staining. The top panels show overlays of phase-contrast and fluorescence images after staining of planktonic cells was performed as described in Materials and Methods. Adherent cells from the slide attachment assay are shown as phase-contrast images in the middle set of panels, and the fluorescence channel showing attached holdfast material from the same slides is represented in the bottom panels. The ΔCCNA_01242 mutant did not have an apparent holdfast defect, the ΔCCNA_02722 mutant had a holdfast attachment defect, and the ΔCCNA_02360 mutant did not produce holdfasts. The scale bars represent 5 m. experiment. It is not clear why CCNA_01242 mutants were more enriched when cheesecloth was included in the medium, but we conclude nonetheless that this gene does not contribute to holdfast production.
CCNA_02360 is a predicted member of the GT2 family of glycosyltransferases. Its disruptions have a fitness profile closely resembling the profiles of many of the known hfs genes (Fig. 5A). A ΔCCNA_02360 mutant was nonadhesive in surface attachment assays and did not stain with fWGA under any conditions tested ( Fig. 5B and C). Given the lack of holdfast production in ΔCCNA_02360 cells, we predict that CCNA_02360 encodes a glycosyltransferase that contributes one or more monosaccharides to repeating unit of the holdfast polysaccharide and have named this gene hfsL. Previous studies have described mutants in CCNA_02360 as holdfast-deficient control strains (25,40), but, curiously, identification of the gene and characterization of its phenotype have not been reported. Closely related genes could be identified in many stalked bacteria within the Caulobacterales clade, suggesting that HfsL carries out a conserved step in holdfast biosynthesis. Identifying close homologs in more distantly related Alphaproteobacteria was difficult due to the abundance of GT2 family glycosyltransferases in bacterial genomes. Nevertheless, HfsL represents a fourth glycosyltransferase that is required for holdfast biosynthesis in C. crescentus.
CCNA_02722 encodes a hypothetical protein that does not show homology to any known protein families. It has a predicted N-terminal signal peptide for export to the periplasm. The fitness profile is similar to that of hfsK, which is indicative of modest defects in adhesion (Fig. 5A). The ΔCCNA_02722 mutant was significantly impaired in surface-attachment (Fig. 5B). When planktonic ΔCCNA_02722 cells were stained with fWGA, holdfast staining was apparent (Fig. 5C). However, using adhesion assays in which cells were grown in the presence of a glass slide that was then washed, stained with fWGA, and imaged, we detected a holdfast anchoring defect. Wild-type cells normally coat the surface of the slide and show fWGA foci at the site of attachment, but we observed very few attached ΔCCNA_02722 cells. Instead, the slide was coated with an abundance of fWGA-reactive material (Fig. 5C). This holdfast shedding phenotype is characteristic of mutants with defects in anchoring the holdfast matrix to the surface of the cell (28), and the gene has been named hfaE accordingly. Like hfsL, hfaE could be identified in the genomes of many other stalked bacteria, suggesting a conserved role in holdfast anchoring.

DISCUSSION
Holdfast production in C. crescentus presents an attractive model system to interrogate the assembly of polysaccharides in bacteria. Lectin staining, enzymatic sensitivity, and functional annotations for the hfs genes indicate that the holdfast contains a polysaccharide (17,18,23). However, despite decades of work, surprisingly little else is known about the chemical structure of the holdfast. As part of our efforts to characterize the biosynthetic pathway, we sought a complete list of enzymes required for holdfast biosynthesis. In order to saturate the search for hfs genes, we developed a TnSeq-based method to measure the adhesion phenotype conferred by each nonessential gene in the genome.
The genome-wide approach provided a surprisingly rich set of insights into adhesion in C. crescentus. Not only did the screen identify nonadhesive mutants representing missing components of the holdfast pathway, but it also resolved mutants that displayed increased adhesion. One class of such mutants was defective in the production of smooth LPS. Disrupting SLPS resulted in elevated adhesion levels in both holdfast-producing and holdfast-deficient backgrounds. SLPS mutants no longer adhered exclusively at the stalks of holdfast-producing cells but instead displayed a generalized form of adhesion throughout the cell surface. Our analysis of these mutants supports a model in which the C. crescentus envelope is structured to ensure that the cell surface is nonadhesive, maximizing the opportunity for polar adhesion via the holdfast. It is still unclear why disrupting the cell surface led to the depletion and recovery profile seen during cheesecloth passaging. Regardless of the mechanism, simply identifying mutants that shared this temporal profile allowed the characterization of new SLPS biosynthesis genes. Such cofitness approaches have been useful in other contexts (42) and allowed us to greatly expand the number of genes identified in the C. crescentus SLPS pathway (see Table S5 in the supplemental material).
Genes with predicted functions in motility, flagellar biosynthesis, and type IV pilus assembly displayed a hyperadhesive profile that could be distinguished from the profile of SLPS mutants by the lack of a recovery phase. Mutations in components of the flagellar basal body were recently shown to enhance holdfast production by inhibiting the expression of hfiA, a result that we confirmed here in our examination of flgH (41). Cofitness analysis indicated that the cpa genes, which code for components of the type IV pilus, have a similar phenotype, and we showed that mutation of the inner membrane pilus assembly component gene cpaH also increased holdfast production by repressing hfiA. However, mutation of pilA, which codes for the main subunit of the pilus filament, reduced adhesion. The adhesion defect in ΔpilA cells was partially restored in a ΔpilA ΔflgH background but remained unchanged in ΔpilA ΔcpaH cells. Thus, although both the ΔflgH and ΔcpaH mutations enhanced holdfast production, the two pathways can be distinguished by their requirement for pilA. Disentangling the specific routes by which the various polar appendage mutants modulate hfiA activity will require identifying intermediate factors in the signaling pathways, but our results underscore the interconnectedness of the flagellum, pilus, and holdfast.
Numerous reports have debated the roles of pili and the flagellum in surface attachment (13-15, 43, 44). Our unbiased, genome-wide analysis of adhesion unambiguously identified both appendages as determinants of attachment. Two recent studies, in particular, showed that mutating the flagellar basal body represses hfiA and that disruption of flagellar rotation upon surface contact stimulates holdfast production (15,41). In our cheesecloth passaging experiment, the flagellar motor (mot), flagellin glycosylation (flm), and chemotaxis (che) genes shared the same hyperadhesive fitness profile as components of the flagellar basal body (Table S5). Some of these mutants would be predicted to disrupt flagellar rotation without affecting assembly per se (45). We propose that disrupting flagellar function, either through physical interaction with a surface or through mutation of motility genes, stimulates holdfast production. It will be interesting to test this model by determining whether the repression of hfiA seen in the flagellar mutants is required to activate holdfast production after surface contact. We also note that future studies should take into account the finding that mutating flgH reduces biofilm formation in a holdfast-independent manner (see Fig. S3 in the supplemental material), suggesting that flagellar motility likely also promotes productive interactions with a surface that lead to permanent attachment.
Much like flagellar mutants, disrupting components of the type IV pilus causes both increased holdfast production and holdfast-independent surface colonization defects. A recent report proposed that contact with a surface inhibits the retraction of PilA filaments, leading to a stimulation of holdfast production (14). One might initially conclude that pilus assembly defects in the cpa mutants mimic the obstruction of pilus filament retraction. However, the situation is more complex because pilA was required for increased holdfast production in the ΔcpaH mutant. These findings can be reconciled in a model in which the disruption of filament oscillation caused either by surface contact or upon mutation of the cpa genes leads to increases in the pool of unassembled PilA proteins that serve as a signal to stimulate holdfast production. A similar model was proposed for the regulation of biofilm formation by the Agrobacterium tumefaciens pilus (38). Furthermore, unassembled pilin subunits in Pseudomonas aeruginosa directly activate the sensor kinase PilS, leading to the repression of pilA transcription (46). Although no clear PilS homologs are found in C. crescentus, this example demonstrates that the membrane-associated pilin pool can serve as an input that activates signaling cascades.
An important aspect of both the flagellar and pilus pathways for holdfast regulation is their partial dependence on pleD. pleD is a pleiotropic regulator of cell polarity that is required for flagellar ejection and stalk synthesis during the swarmer-cell-to-stalked-cell transition (47). It functions as a diguanylate cyclase that is activated by phosphorylation at distinct stages of the cell cycle to produce the bacterial second messenger cyclic di-GMP (cdG) (39,48). Though pleD has been shown to regulate the timing of holdfast synthesis during the cell cycle, ΔpleD mutants do not have significant bulk adhesion defects (Fig. 3C) (40). More broadly, placement of pleD within a signaling cascade that regulates holdfast synthesis is confounded by the fact that both pleD and cdG contribute to numerous processes that intersect with holdfast synthesis, including flagellar function, pilus assembly, stalk biogenesis, and cell cycle progression (39,49). Importantly, it has been shown that the holdfast glycosyltransferase HfsJ, which is the target of inhibition by HfiA, is also directly activated by cdG (11,15). Thus, the roles of PleD in promoting holdfast synthesis upon disruption of the pilus or flagellum are likely twofold. It activates HfsJ by producing cdG and relieves inhibition by lowering transcription of hfiA through an unknown mechanism. PleD-dependent increases in cdG concentrations may account for the disparity between the large changes in adhesion and the modest decreases in hfiA expression in the ΔflgH and ΔcpaH backgrounds (Fig. 3).
Finally, the initial goal of this study was to saturate the search for holdfast production factors. We used the cheesecloth passages to define a temporal fitness profile that was shared by genes known to be required for holdfast biosynthesis and used this pattern to search for missing genes in the pathway. In addition to a new holdfast anchoring factor, hfaE, we identified a glycosyltransferase, hfsL, and showed that it is required for holdfast production. Due to the saturating nature of our experiment, we believe that HfsE (along with the redundant PssY and PssZ), HfsJ, HfsG, and HfsL carry out the only glycosyltransferase steps required for holdfast biosynthesis (Fig. 6). Thus, we predict that a repeating unit of four or more monosaccharides makes up the holdfast polysaccharide because each of these enzymes likely contributes at least one sugar to the glycolipid intermediate that serves as a substrate for polymerization. Implicit in defining the complete set of genes in the holdfast pathway is knowledge of genes that do not contribute. Many bacterial polysaccharides contain specialized monosaccharide components (50,51), and these intermediates are often synthesized as nucleotide-activated precursors that are directly utilized by glycosyltransferases (52,53). We did not identify any genes involved in nucleotide sugar metabolism that had significant adhesion defects when disrupted. The modification factors HfsH and HfsK do likely convert certain monosaccharide components into more specialized residues. However, functional annotations for these enzymes predict that they act on lipid-linked or polymerized substrates downstream of the glycosyltransferases, and neither enzyme is explicitly required for holdfast synthesis (19,25). Our results suggest that specialized sugar precursors are not needed to produce the holdfast and that it is instead built using "housekeeping" sugars that are shared by other cellular processes. The ability to utilize standard nucleotide sugars as substrates without a requirement for specialized chemical syntheses will make the holdfast biosynthetic enzymes useful models for probing the catalytic mechanisms of bacterial polysaccharide biosynthesis.
The findings reported here highlight the advantages of probing mutant phenotypes in parallel. Classical genetic selection methods in which mutants are enriched and then isolated and analyzed separately inherently favor the most extreme phenotypes. The single-strain approach also favors longer genes that contribute more individual mutants to the pool. Screening mutants in parallel using TnSeq reduces these biases, allowing detection of a range of phenotypes and capturing phenotypes for lessabundant mutants in the pool. In this study, we discovered new regulatory networks that modulate holdfast synthesis by identifying mutants that display modest adhesion increases under conditions in which cells are already adhesive. Such mutants would be extremely difficult to isolate using a classical approach. Additionally, the saturating nature of these experiments makes them ideal for outlining complete biosynthetic pathways. Having a reasonable measure of saturation allowed us to leverage the results of the genome-wide screen to propose a model for the enzymatic steps in the holdfast pathway that will inform efforts to reconstitute the biosynthetic reactions in vitro (Fig. 6).

MATERIALS AND METHODS
Strains, growth conditions, and genetic manipulation. Strains and plasmids used in this study are listed in Table S6 in the supplemental material. Escherichia coli strains were grown in LB at 37°C with diaminopimelic acid (60 mM), kanamycin (50 g/ml), or tetracycline (12 g/ml) included as needed. C. crescentus CB15 was grown in either peptone-yeast extract (PYE; complex) medium or M2 salts containing 0.15% xylose (M2X, defined) at 30°C (54). When necessary, sucrose (3%), kanamycin (25 g/ml), or tetracycline (2 g/ml) was included in solid medium. Kanamycin (5 g/ml) or tetracycline (1 g/ml) was included in liquid medium when necessary. Standard techniques were used for Gibson assembly-based cloning and sequencing plasmids (55). The primer sequences used for cloning specific constructs are available on request. Plasmids were introduced by electroporation or, in the case of the pFC1948 reporter plasmid, by triparental mating. Unmarked mutations were created through two-step deletion using SacB counterselection. Mutants were complemented by insertion of target genes into pXGFPC-2, which allows the integration of the plasmid at the xyl locus and transcriptional control from P xyl (56). When necessary, target genes were inserted into pXGFPC-2 in reverse orientation under the control of their native promoters.
Library development and mapping. The barcoded HiMar transposon pool APA_752, developed by Wetmore et al. (30), was used to create a barcoded Tn library in C. crescentus CB15. Construction of the library was reported previously, along with its associated statistics (57). Briefly, the transposon pool was introduced into C. crescentus by conjugation. Transconjugants appearing on selective plates were pooled, used to inoculate a liquid culture with kanamycin, and grown for 3 doublings. Glycerol was added to achieve a final concentration of 15%, and 1 ml aliquots were frozen and stored at Ϫ80°C. TnSeq also followed the method of Wetmore et al. (30). A 1-ml library aliquot was centrifuged, and genomic DNA was extracted from the pellet. The DNA was sheared, size selected for ϳ300-bp fragments, and end repaired. A custom Y-adapter (Mod2_TS_Univ annealed to Mod2_TruSeq) was ligated, and transposon junctions were amplified by PCR using the Nspacer_BarSeq_pHIMAR and P7_mod_TS_index1 primers. An Illumina HiSeq 2500 system was used to generate 150-bp single-end reads of the library. The genomic positions of each barcoded insertion were determined with BLAT. The barcode corresponding to each insertion site was determined using MapTnSeq.pl. This information was used to develop a list of barcodes that mapped to unique insertion sites using DesignRandomPool.pl (available at https://bitbucket.org/ berkeleylab/feba).
Passaging in cheesecloth. Aliquots (1 ml) of the transposon library were thawed at room temperature, a 300-l volume of the library was inoculated into a well of a 12-well microtiter plate containing 1.2 ml PYE, and a stack of 5 squares (ϳ10 mm by ϳ10 mm) of sterile cheesecloth was added. The culture was grown with shaking at 155 rpm at 30°C for 24 h, after which 100 l of the planktonic culture was used to inoculate a fresh well containing 1.4 ml PYE and a fresh piece of sterile cheesecloth. An additional 500 l of the planktonic culture medium was centrifuged, and the pellet was stored at Ϫ20°C for BarSeq analysis. The process was repeated for a total of five passages, and each passaging experiment was performed in triplicate. The same procedure was used to perform passaging experiments in which no cheesecloth was added.
Fitness determination with BarSeq. Cell pellets from 0.5 ml of planktonic culture medium that had been frozen after each passage were used as templates for PCRs that simultaneously amplified the barcode region of the transposon insertions and added TruSeq indexed Illumina adaptors (30). The PCR products were purified, pooled, and multiplexed on a single Illumina 4000 lane for sequencing. Fitness values for each gene were determined using the pipeline described by Wetmore et al. (30). Barcodes for each read were mapped using MultiCodes.pl and correlated with their associated insertion positions using combineBarSeq.pl. The data were used to calculate fitness using FEBA.R. This analysis determines strain fitness as the log 2 ratio of barcode counts in a sample to the barcode counts determined under the reference conditions, for which we used the first passage of the library in PYE without cheesecloth. Gene fitness was calculated by determining the weighted average of the insertion mutants with a given gene, excluding the first and last 10% of the open reading frame (ORF). The scripts used for fitness determination can be found at https://bitbucket.org/berkeleylab/feba.
Analysis of fitness data. We focused on identifying the genes with the highest absolute (positive or negative) fitness scores. For each passaging step (with and without cheesecloth), an average and a standard deviation for the three replicate samples were calculated. Genes for which the largest standard deviation across the 10 conditions was greater than the largest absolute fitness score were eliminated from further analysis. Fitness scores determined for each passage performed without cheesecloth were subtracted from those for the corresponding passage performed with cheesecloth to normalize for growth defects. These normalized values were used to rank each gene according to its largest absolute fitness score at any stage of cheesecloth passaging. The top 250 genes were then sorted by hierarchical clustering (58) to identify related fitness patterns. Groups derived from the clustering analysis were curated manually to produce the gene data shown in Fig. 1C. The "Pilus Assembly" subsection of Table S5 was created by manually identifying known pilus assembly genes, some of which do not have phenotypes that are strong enough to meet the cutoff used to identify the genes in Fig. 1C.
Crystal violet (CV) staining of adherent cells. Overnight cultures of C. crescentus grown in PYE were diluted to an OD at 600 nm (OD 660 ) of 0.4 with PYE, 1 l was inoculated into the wells of 48-well microtiter plates containing 450 l medium, and the cultures were shaken for 24 h. For the cultures used to measure the effects of polar appendage mutants on adhesion, the incubation was shortened to 17 h. The cultures were then discarded, and the plates were washed thoroughly with a steady stream of tap water. The surface-attached cells remaining in the wells were stained for 5 min with 0.01% crystal violet and washed again with tap water. Stain was extracted for 5 min in 100% ethanol (EtOH) and quantified by reading absorbance at 575 nm.
Fluorescent wheat germ agglutinin (fWGA) staining. For staining of planktonic cells, 400 l of liquid culture was added to an Eppendorf tube and Alexa594-conjugated WGA (Thermo Fisher) was added to achieve a final concentration of 0.5 g/ml. After incubation of the cells for 5 min in the dark, 1 ml of sterile water was added and the cells were centrifuged at 6,000 ϫ g for 2 min. The pellet was resuspended in 5 l of medium, spotted on a glass slide, and imaged. Cells were imaged with a Leica DM500 microscope equipped with a HCX PL APO 63ϫ/1.4-numerical-aperture (NA) Ph3 objective. fWGA staining was visualized using a red fluorescent protein (RFP) fluorescence filter (Chroma set 41043). For quantifying the number of holdfast-producing cells, a culture was inoculated to reach an OD 660 of 0.002 and harvested when the OD 660 reached 0.05 to 0.1 to minimize the number of rosettes.
For staining attached cells, a glass coverslip that had been washed with ethanol was added to 1 ml of PYE in a 12-well plate, and 100 l of a starter culture (diluted to an OD 660 of 0.4) was added to the well. The cultures were grown for 6 to 8 h. Nonattached cells were removed from the coverslip by washing under a stream of distilled water. One side of the glass was covered with a solution of 0.5 g/ml fWGA-PYE, incubated for 5 min in the dark, and washed under a stream of distilled water. The coverslip was placed stain side down on a glass slide and imaged with phase contrast and fluorescence as described above.
Smooth lipopolysaccharide (SLPS) immunoblotting. Lysates were prepared for immunoblotting by the method of Walker et al. (59). Pellets from saturated cultures were treated with DNase and lysozyme, mixed with SDS-running buffer, and digested with proteinase K. Samples were separated on a Tris-glycine SDS-PAGE gel with 12% acrylamide and transferred to nitrocellulose. The amount of sample loaded was normalized to the final optical density of the culture at harvest. The membrane was blocked in 5% milk, probed with a 1-in-20,000 dilution of anti-SLPS serum raised in rabbit (34), washed, probed with horseradish peroxidase (HRP)-conjugated goat anti-rabbit antibody (Invitrogen), washed again, and visualized with peroxidase substrate. The top half of the membrane was removed before blocking and stained with Ponceau S (0.25% [wt/vol] Ponceau S-1% acetic acid) as a loading control.
Analysis of rough LPS. LPS was extracted by the method of Darveau and Hancock (60). Cells were isolated from saturated 50-ml C. crescentus cultures grown in M2X by centrifugation, resuspended in 2 ml of 10 mM Tris-HCl (pH 8.0) containing 2 mM MgCl 2 , and sonicated. DNase I and RNase A were added to achieve concentrations of 100 g/ml and 25 g/ml, respectively, and the lysate was incubated for 1 h at 37°C. Additional DNase I and RNase A were added to achieve concentrations of 200 g/ml and 50 g/ml, respectively, and the lysate was incubated for an additional 1 h at 37°C. SDS and EDTA were added to achieve concentrations of 2% and 100 mM, respectively, and the lysate was incubated for 2 h at 37°C. The solution was then centrifuged for 30 min at 50,000 ϫ g. Proteinase K was added to achieve a concentration of 50 g/ml in the supernatant, and the solution was incubated for 2 h at 60°C, after which the LPS was precipitated with 2 volumes of ice-cold 0.375 M MgCl 2 -95% EtOH and collected by centrifugation at 12,000 ϫ g. The precipitate was resuspended in 2% SDS containing 100 mM EDTA, incubated overnight at 37°C, and reprecipitated with 0.375 M MgCl 2 -95% EtOH. The precipitate was then suspended in 10 mM Tris-HCl (pH 8.0) and centrifuged for 2 h at 200,000 ϫ g. The LPS pellets from each strain were suspended in SDS loading dye and separated by Tris-Tricine SDS-PAGE on an 18% acrylamide gel containing 6 M urea. LPS was stained using the periodate-silver method of Kittelberger and Hilbink (61).
Bacteriophage ⌽CBK sensitivity. Saturated cultures of C. crescentus grown in PYE were diluted to an OD 660 of 0.4 with PYE. Volumes of 4 l of these suspensions, along with six 10-fold serial dilutions (prepared in PYE), were spotted on PYE plates containing 0.15% xylose that had been spread to 10 7 PFU/ml (assuming a volume of 60 ml for each 150-mm-diameter plate) with ⌽CBK or on PYE plates with 0.15% xylose alone. The plates were incubated for 48 h at 30°C and photographed.
Soft-agar swarming assay. A 1.5-l volume from a saturated culture of the appropriate C. crescentus strain grown in PYE was spotted in plates of PYE containing 0.3% agar and 0.15% xylose. The plates were incubated for 4 days at 30°C and photographed. lacZ reporter assay. Cultures for measuring reporter activity were grown in M2X medium, and the amount of culture required to achieve an OD 660 of 0.0005 to 0.00075 was added to fresh M2X medium. These cultures were grown to an OD 660 of 0.05 to 0.15, and the ␤-galactosidase activity was measured as previously described (11,62).
Data availability. The raw sequence data for each BarSeq sample and a table describing the barcode abundances and their genomic insertion sites have been uploaded to the NCBI Gene Expression Omnibus (GEO) at https://www.ncbi.nlm.nih.gov/geo/ under accession no. GSE119738.