Unprecedented Parallel Selection for rpfR Mutations during Evolution Experiments.

From previous evolution experiments (15, 18, 21, 22) with B. cenocepacia grown in our bead model that simulates the biofilm life cycle, we identified at least 72 rpfR mutations in 32 independent populations that affected multiple protein domains (Fig. 1B and SI Appendix, Table S1). Additional mutations were identified from an analysis of mutations selected during evolution in media made of macerated onion (23). Mutations in rpfR were always among the first mutations to rise to high frequency (>25%) in each laboratory experiment and were associated with increased competitive fitness and biofilm production (18, 19). The mutation spectrum demonstrates strong selection for altered or eliminated protein function: 45 out of 46 nucleotide substitutions were nonsynonymous, and 26 additional mutations produced deletions or premature stop codons. All but two deletions removed both rpfR and the adjacent rpfF gene, suggesting that selection acted upon interactions between these two gene products. The distribution of single nucleotide polymorphisms (SNPs) was also nonrandom and significantly enriched in linker regions between the four domains rather than in the catalytic or sensory sites themselves (X² = 10.47, degrees of freedom [df] = 1, P = 0.0012, Table 1 and Fig. 1B). This result suggests selection for altered interactions between functional domains rather than for disrupting BDSF sensing or c-di-GMP catalysis. Among the 13 mutations in the DGC domain, 8 occurred at Y355 or R377, pointing to the functional importance of these residues. Further, 10 mutations affecting the phosphodiesterase EAL domain occurred in just two positions, S570 (n = 3) and F589 (n = 7). In total, selection acted on the rpfR sequence with remarkable precision that prompted further study of their functional roles.

In the long-term evolution experiment, three rpfR mutants arose in the same population, coexisted during long-term biofilm selection, and associated with different ecology, which suggested that these mutations were not functionally equivalent (15). These mutants, A106P in the region linking the FI and PAS domains, Y355D in the DGC domain, and a deletion mutant of both rpfR and rpfF (or a functionally equivalent de novo evolved mutant), also evolved in parallel among replicate populations and became a major focus of this study. Together, these findings suggested that selection could produce multiple, discrete phenotypes by altering different domains of a dominant c-di-GMP regulator.

Biofilm and c-di-GMP Levels Vary with Mutated rpfR Domains.

We introduced the evolved point mutations or targeted deletions into the ancestral HI2424 strain (SI Appendix, Tables S2 and S3) and confirmed that they were otherwise isogenic by whole-genome sequencing. Hereafter, we refer to these engineered genotypes as evolved mutants. Further, to test the contributions of each sensor and enzymatic domain, we constructed deletions of the FI domain (1-95aa) and alanine replacements predicted to eliminate diguanylate cyclase activity (GGDAF, equivalent to E319A) or phosphodiesterase activity (AAL or E443A). We also deleted rpfR in its entirety, the adjacent BDSF synthase rpfF, or both these genes. Because a 95-gene deletion removing both rpfR and rpfF repeatedly evolved during our experiments and was available before successful construction of the ΔrpfFR genotype, some experiments were conducted with this ΔrpfRF+93 genotype (SI Appendix, Table S2). We subsequently competed these two genotypes and found their fitness to be statistically indistinguishable (t = 0.38, df = 10, P = 0.12).

A handy screen for elevated c-di-GMP is rugose colony morphology or increased uptake of Congo red dye, both of which result from the increased polysaccharide production often associated with high c-di-GMP (24). The evolved point mutants (A106P and Y355D) displayed increased uptake of Congo red dye on morphology plates (Fig. 2A), and all evolved colonies produced a characteristic studded center and smooth periphery in contrast to the smooth phenotype of wild type (WT) (SI Appendix, Fig. S1A). These colony phenotypes correlated with increased biofilm production and reduced motility (Fig. 2B and SI Appendix, Fig. S1B). Similar phenotypes were observed in the engineered AAL and ΔrpfF mutants (SI Appendix, Fig. S1), which should eliminate the PDE domain activity and BDSF production that activates the PDE domain, respectively, increasing c-di-GMP levels. To test these predictions, we quantified in vivo levels of intracellular c-di-GMP at both 12 and 24 h from planktonic and biofilm cultures of each mutant (Fig. 2 C and D and SI Appendix, Table S4). First, we learned that absolute values of the signal were generally much greater at 24 h in denser biofilms, but relative differences (values divided by WT value) were greater at 12 h when colonization of the plastic beads accelerates in our model (SI Appendix, Table S4) (18, 25). Second, the A106P mutant of the FI-PAS linker region produces modest but consistent increases in c-di-GMP across conditions, suggesting this mutant interferes with PAS-mediated activation of the PDE. Third, as predicted, the AAL mutant that should disable the PDE domain and the ΔrpfF mutant that produces no BDSF to activate the PDE domain both increase c-di-GMP. Fourth, the evolved ΔrpfRF+93 mutant produced elevated c-di-GMP in biofilms at 12 h and in planktonic cultures at 24 h, which suggests that losing the PDE activity of RpfR unmasks contributions of other DGCs. Interestingly, deleting only rpfR did not significantly alter c-di-GMP levels in biofilms but did increase levels in planktonic cultures, suggesting that functional RpfF in the absence of RpfR affects the c-di-GMP pool in an unknown manner. Finally, the evolved Y355D mutant of the DGC domain produced the highest levels of c-di-GMP, suggesting this is a gain-of-function mutation in a domain thought to be nonfunctional (26, 27). To test this prediction, we constructed a GGDAF mutation that should disable the DGC domain in the Y355D mutant (Y355D-GGDAF) and found, as expected, it produced WT levels of c-di-GMP (SI Appendix, Fig. S2A). This result demonstrated that the RpfR DGC domain is directly responsible for the high c-di-GMP levels in Y355D. Together, these results indicate that evolved genotypes produce different basal levels of c-di-GMP depending on the affected domain and alter production depending on their environment. In broader terms, growth of B. cenocepacia in a cycle of biofilm attachment, formation, and dispersal can select for differentiated phenotypes caused by single rpfR mutations.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Evolved and engineered rpfR genotypes produce diverse c-di-GMP-regulated phenotypes. (A) Colony characteristics of different evolved and engineered mutants on Congo red, tryptone agar plates (colony backgrounds are different due to Congo red gradients). (Scale bar, 5 mm.) (B) Biofilm productivity (attachment) at 24 h measured by crystal violet staining. Evolved mutants are represented in different colors, and engineered mutants are in gray. Relative levels of c-di-GMP to WT measured at (C) 12 and (D) 24 h in biofilm and planktonic conditions. Error bars indicate the 95% CI. Different letters in the same graph indicate significant differences between mutants (one-way ANOVA and post hoc comparisons via the Benjamini and Hochberg method (28), q value < 0.05). For C and D, letters in blue and black are for planktonic and biofilm conditions, respectively.

Fitness in the Biofilm Model Relates to c-di-GMP Levels.

We predicted that varied c-di-GMP levels and associated differences in biofilm matrix production contributed to fitness. Evolved and engineered mutants were competed against the WT strain in equal ratios and demonstrated significant variation in fitness, with the Y355D mutant the most fit (Fig. 3). Overall, fitness in the biofilm model at 24 h, when development matures, positively correlated with c-di-GMP levels at 12 h, when rates of attachment accelerate (Fig. 3A). However, the rate of fitness increase decelerates with increasing c-di-GMP levels, especially among evolved mutants, suggesting diminishing returns (Fig. 3A). The A106P mutant was disproportionately more fit at 24 h, implying additional advantages of this genotype affecting the FI-PAS linker region, yet fitness of ΔrpfF was equivalent to WT in biofilm despite very high c-di-GMP levels (Fig. 3B). Further, the evolved ΔrpfRF+93 and the engineered ΔrpfR genotypes were more fit against WT despite modest increases in c-di-GMP. Many of the mutants were also more fit against the WT under planktonic growth conditions, which is a necessary component of our bead model that requires dispersal, but fitness benefits were lower and less variable among mutants than those in biofilm conditions (SI Appendix, Fig. S3). The loss of PDE activity (AAL) increased c-di-GMP levels, as predicted, and also greatly increased fitness (Fig. 3 A and B). This strong benefit suggests that the dominant role of RpfR is its PDE activity, as has been shown in orthologs of other species (3, 29)

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Fitness of rpfR genotypes as a function of c-di-GMP levels. Fitness is calculated as selection rate, with neutral fitness = 0. Colored symbols are evolved genotypes; gray are engineered. (A) Nonlinear relationship (segmental linear regression, r² = 0.77) between c-di-GMP production at 12 h and fitness in biofilm at 24 h for evolved mutants. Engineered mutants shown in gray are not included in function. (B) Relative fitness vs. WT at 24 and 48 h in biofilm growth conditions. The experiment encompasses the four distinct biofilm stages: attachment, assembly (measured at 24 h), reattachment, and reassembly measured at 48 h. Different letters indicate significant differences between genotypes by post hoc testing following ANOVA. (C) Fitness differences between rpfR mutants after 24 h of competition starting from different starting frequencies. The intersection between the regression lines and the x axis is the predicted mutant frequencies at equilibrium. The slopes for competitions between A106P and Y355D and between A106P and ΔrpfRF+93 differ significantly from zero (P < 0.001). Error bars are 95% CIs.

Biofilm Ecology: Coexistence of rpfR Mutants.

The sustained coexistence of different rpfR mutants in evolving biofilm populations (19) could be explained by niche differentiation within the biofilm life cycle. If these niches support populations of different sizes, the fitness of different genotypes should depend on their relative frequencies and genotypes should be able to invade one another when rare, also known as negative-frequency-dependent selection (NFDS) (30). We tested this hypothesis by competing each evolved genotype against the others after 24 h in the biofilm and found support for this model (Fig. 3C). Both A106P and Y355D can invade one another when introduced at low frequency, with a predicted equilibrium ratio of 1:4 A106P:Y355D (Fig. 3C, linear regression analysis y = −0.0149 × x + 0.3599, r² = 0.98). Further, the A106P and ΔrpfRF+93 or ΔrpfR mutants show comparably high fitness in competition with WT (Fig. 2A) but may coexist via NFDS when cocultured (Fig. 3C, linear regression y = −0.01164 × x + 0.8477, r² = 0.94 and SI Appendix, Fig. S4B, y = −0.02265 × x + 0.945, r² = 0.64). However, the Y355D mutant was significantly fitter than the ΔrpfRF+93 genotype that ultimately displaced it during the long-term evolution experiment (Fig. 3C, yellow, linear regression y = −0.0029 × x + 1.074, r² = 0.03). High Y355D fitness is consistent with its sweep to high frequency (15) and parallel evolution (Table 1), but this cannot explain why ΔrpfRF+93 ultimately displaced Y355D. Prior studies indicated that the spread of other mutations within the ΔrpfRF+93 lineage increased its relative fitness and excluded other rpfR lineages (19), and we explore other explanations below. In contrast, ΔrpfF and WT fail to invade each other when rare (SI Appendix, Fig. S4A), which is consistent with complementation of the ΔrpfF phenotype by WT BSDF production. In summary, different rpfR genotypes that avoid BDSF-mediated dispersion in various ways readily displace the WT ancestor in our biofilm model and can coexist for hundreds of generations by NFDS.

Biofilm Ecology: Coaggregation and Synergistic Interactions.

Sustained coexistence of different genotypes in biofilms could be enabled by forming aggregates of different composition and form. We tested this potential mechanism of niche differentiation using fluorescently labeled genotypes to measure their colocalization and total volume by confocal microscopy (SI Appendix, Table S5). When cultured separately, both A106P and Y355D formed large, thick aggregates that were well dispersed (Fig. 4A), whereas ΔrpfRF+93 produced thinner, more uniform biofilms arranged in small clusters. This result shows that the loss of the RpfRF complex and/or BDSF production alters the form of biofilm development, whereas the point mutants appear to produce larger clusters than those produced by WT (Fig. 4 A and B and SI Appendix, Table S5 and Fig. S4A). Different genotype combinations produced aggregates of varying size and biofilm thickness (Fig. 4B). The difference in biofilm development by ΔrpfRF+93 was even more apparent when this mutant was cocultured with either Y355D or A106P, resulting in thinner, more uniform structures and indicating a dominant effect of ΔrpfRF+93 on biofilm development (Fig. 4 A and B and SI Appendix, Table S5). Incidentally, we observed that ΔrpfF and ΔrpfR formed small clusters when mixed with other mutants but formed larger aggregates when grown together, which is consistent with cross complementation (SI Appendix, Fig. S4A).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Cocultures of evolved mutants exhibit complementary interactions. (A) Confocal images depict specific structural differences between single-strain and coculture biofilms. Large aggregates with spaces are observed in Y355D, A106P, and Y355D x A106P mutant biofilms, whereas monocultures of ΔrpfRF+93 and cocultures with this mutant (A106P x ΔrpfRF+93 and Y355D x ΔrpfRF+93) exhibit small clusters and uniform thickness. To improve viewing, RFP-labeled cells are false colored in magenta and YFP-labeled cells are in yellow. White spots indicate coaggregation of differently labeled strains. (Scale bar, 10 μm.). (B) Correlation between the average aggregate size of the attached aggregate and biofilm thickness (symbols represent the genotypes, and colors represent the partnered genotype). (C) Total biofilm productivity as CFU of individual strains in coculture. Stacked bars with the same color indicate competition between oppositely labeled cells of the same mutant, while bars with different colors represent coculture. The mutants in the graph are color-coded (green, A106P; blue, Y355D; and red, ΔrpfRF+93). The expected (exp) values are obtained from the individual competitions, while the observed (obs) values are experimentally determined. Letters denote different pairwise statistical groupings, as described in Extended Methods (SI Appendix). (D) Coaggregation in biofilms, where positive coefficients indicate the extent of overlap between two channels (*values significantly different from zero).

Interactions between genotypes can range from antagonistic, which would reduce net productivity of both types, to synergistic, which would increase productivity of both. We measured productivity as attached colony forming units (CFU) per milliliter and microscopic biovolume for all genotype combinations. In most cases, cocultures of evolved rpfR mutants grown on polystyrene beads were significantly more productive than mutants grown alone (Fig. 4C and SI Appendix, Table S5). This indicates that different rpfR genotypes facilitate attachment and growth of one another, which supports conclusions from prior studies of long-term evolved biofilm populations involving more complex genotypes (19). Notably, the biofilm productivity of cocultures of Y355D and A106P is higher than that of the individual genotypes but is lower than either cocultured with ΔrpfRF+93, which increased coaggregation with both point mutants (Pearson’s r > 0.5) (Fig. 4 C and D and SI Appendix, Table S5). These results demonstrate that mixtures of rpfR mutants that vary in c-di-GMP levels and BDSF signaling capacity are more productive than when the mutants are grown alone and produce more uniform biofilm structures together. We speculate that the increased evenness of the mixed biofilm architecture may be an adaptation to maintain attachment to the polystyrene beads, which collide frequently in the test tubes.

Biofilm Ecology: Polysaccharide Composition.

B. cenocepacia encodes the capacity to produce various polysaccharides. The best known of these is cepacian (composed of rhamnose, mannose, glucose, galactose, and glucuronic acid) (31), but others include Bep (Burkholderia extracellular polysaccharide) and galactan-deoxy-d-manno-octulosonic acid (32⇓⇓–35). We hypothesized that the different binding and aggregation properties of rpfR mutants related to the production of the components in these exopolysaccharides of varied composition. We used fluorescein-tagged lectins that bind different sugars to visualize and quantify differences in the exopolysaccharide (EPS) composition of evolved mutants (36). All genotypes, including WT, produced a matrix composed of mannose, and this sugar was particularly elevated in the ΔrpfRF+93 genotype. However, fucose was only detected in rpfR mutants and not ΔrpfF (Fig. 5B). Galactose, N-acetyl glucosamine, and N-acetyl galactosamine were not detected in the EPS produced by any genotype. We then used calcofluor white to stain cellulose and found that Y355D produces much more cellulose than any other mutant (Fig. 5C and SI Appendix, Fig. S5). Thus, the varied biofilm phenotypes of rpfR mutants may result from different genotypes secreting different polymers that could serve as shared products that benefit collective attachment.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Varied EPS composition of evolved rpfR mutants. Biovolumes are calculated from labelled cells and from the fluorescent intensities of bound fluorescently tagged lectins for (A) mannose and (B) fucose or from calcofluor for (C) cellulose using IMARIS 9.0. The biovolume of the EPS-stained channel was divided by the biovolume of the bacteria to approximate EPS/bacterial cluster. Different letters are used to indicate significant differences (Tukey’s post hoc tests following ANOVA) between the mutants.

Transcriptomic Differences among rpfR Mutants.

Mutations in rpfR are clearly pleiotropic, so to examine the extent of their altered regulation we conducted RNA-seq of six genotypes (A106P, Y355D, ΔrpfR, ΔrpfF, ΔrpfRF, and WT) grown under selective biofilm conditions. Hundreds of genes distinguished mutant expression from WT (at q values < 0.05), with Y355D recording the greatest number (∼930 genes at fold change < 1.5), and dozens of genes separated mutants from one another (SI Appendix, Fig. S7). As expected from the elevated c-di-GMP levels of mutants, motility and chemotaxis processes were down-regulated (except in ΔrpfR, which also produced near-WT levels of c-di-GMP in biofilm), and in the mutant with the highest c-di-GMP levels, Y355D, other PDEs (e.g., Bcen2424_5027) were up-regulated (Fig. 6). One gene cluster encoding the synthesis of Bep exhibited the greatest increase in expression across all mutants, providing strong evidence that this polymer is responsible for increased biofilm production. Further, the berA gene (Bcen2424_4216), which binds c-di-GMP and activates Bep production and cellulose synthesis, was up-regulated in all mutants but ΔrpfRF (32, 37). Notably, expression of genes within the Bep cluster varied among mutants, with the most up-regulated being Bcen2424_4206 (a manC homolog), which encodes mannose-1-phosphate guanylyltransferase. This enzyme plays dual roles, acting as a transferase to convert mannose-1-phosphate to GDP-mannose, a precursor for other sugar nucleotides such as GDP-fucose and GDP-rhamnose, and as an isomerase on mannose-6-phosphate to produce fructose-6-phosphate for gluconeogenesis (38). We hypothesize that increased expression of this gene may activate fucose synthesis (Fig. 5) via the intermediate GDP-mannose. Interestingly, both berA and manC are most active in Y355D, which could explain the high fucose and cellulose in the EPS of this mutant. Another up-regulated gene in Y355D is predicted to encode Flp/Fap pilin (Bcen2424_5868, SI Appendix, Fig. S7), which is known to initiate surface attachment in many bacteria (39). These differences strongly suggest genetic pathways of functional differentiation among rpfR mutants via c-di-GMP-responsive transcription.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

Global changes in expression in evolved and engineered rpfR and rpfF mutants grown in biofilms. Genes that differentiated four or five mutants from WT are shown and categorized by function (q value < 0.05). Up-regulated and down-regulated processes are plotted in shades of blue and orange, respectively. Results are from three biological replicates and were analyzed as described in Extended Methods (SI Appendix).

The gene cluster most consistently down-regulated among rpfR/F mutants encodes three fucose-binding lectins (40). These lectins reportedly have high affinity for galactose and fucose and bind carbohydrates in mucus or glycoconjugates at epithelial cell surfaces, which enables them to adhere specifically to host surfaces as single cells (41), but this form of attachment is unavailable in our laboratory system. This result also indicates that rpfR/F balances solitary lectin-based attachment against aggregate formation via polysaccharide synthesis. Another cluster that was down-regulated among rpfR/F mutants putatively encodes fatty acid biosynthesis (Fig. 6). Overall, selection appears to have favored these rpfR mutants because of their global regulatory effects that produce a variety of phenotypes related to attachment and biofilm production, as well as dispersal and reattachment. While many of them can be explained generally as classic outcomes of high c-di-GMP, mutants are also differentiated in their patterns of expression. Lastly, in a potent demonstration of the power of experimental evolution as a forward genetic screen, the rpfR deletion produced the smallest number of expression changes (Fig. 6); this deletion did not evolve in our experiments, and this mutant was less beneficial than the evolved SNPs that changed but did not eliminate RpfR function.