Experimental biofilm evolution selects for P. aeruginosa hyperbiofilm mutants
In order to examine the evolutionary traits of P. aeruginosa biofilms under antibiotic stress condition, we exposed biofilms of P. aeruginosa PAO1 to different concentrations of imipenem (40, 80 and 160 μg/mL), which is 10, 20 and 40 times of the minimum inhibitory concentration (MIC), in a cyclic manner (Figure 1). Imipenem, a widely used last resort antibiotic, has been chosen as the selective pressure for experimental biofilm evolution owing to its’ commonly prescribed for treatment of P. aeruginosa infections .
The biofilms of six independent lineages in different concentrations of imipenem treated group initiated from a common ancestor PAO1 strain were formed on the surface of 5 mm glass beads  and treated with imipenem for 24 h. Survivor cells on beads were quantitated by CFU counts. Biofilm survivors were collected and reinoculated in fresh LB medium for the 2nd cycle (Figure 1A). At the first cycle, the CFU counts on bead of each lineage was between 6.30-6.90 log10. After 6 cycles, no hyperbiofilm variants observed in 10× and 20× MIC (Figure S1A and B) treated groups, and two lineages were accumulated hyperbiofilm variants in 40× MIC treated group (Figure S1C). The CFU counts within lineage W1 and W6 biofilms on bead reached 9.82 and 10.01 log10 in 40× MIC treated group (Figure 1B). Next, the biofilm formation ability of ancestral, C6W1, C3W6 and C6W6 population were further confirmed by the crystal violet (CV) biofilm assay. Similarly, the CV method revealed that biofilms formed by C6W1, C3W6 and C6W6 population were between 2- and 3-fold higher than the ancestral population (Figures 1C). In order to track when the hyperbiofilm variants have emerged within lineage W1 and W6 population, we picked 6 colonies form each cycle in random and measured the CFU of biofilms on bead. We found that the hyperbiofilm variants of lineage W1 and W6 appeared since cycle 5 and cycle 2 and enriched at cycle 6 (Figure 1D) and cycle 3 (Figure 1E), respectively. These results indicate that the hyperbiofilm variants only accumulated upon higher concentration imipenem treatment, rather than lower concentration of imipenem. The different appearance timing of hyperbiofilm variants in two independent linages - possibly as a result of varying evolutionary rate . More lineages might accumulate hyperbiofilm variants in 40× MIC treated group if we increase the treatment cycles. Based on these results, we chose colonies C6W1C, C3W6F and C6W6F for further study.
Point mutations in rpoSlead to hyperbiofilm phenotype of P. aeruginosa
In order to elucidate the genetic mechanisms underlying the hyperbiofilm phenotype, we sequenced the C6W1 and C6W6 population, and choose C6W5 population as the negative control. Through comparative genomic analysis, we identified only SNPs in one gene, PA3622 (encodes sigma factor RpoS), which was mutated in C6W1 and C6W6 population but not in C6W5 population when compared to the ancestral strain. The sigma factor RpoS is well known as a master regulator that controls the expression of genes involved in stress response and virulence factors production in P. aeruginosa [33, 34]. A previous study used transcript proﬁling to identify that 772 genes were regulated by RpoS in stationary phase and it affects expression of more than 40% quorum-sensing controlled genes .
We next re-sequenced of rpoS in C6W1C, C3W6F and C6W6F, and identified the nonsynonymous mutations in RpoS of C6W1C (P251L), C3W6F (Q266stop) and C6W6F (Q266stop). To further conﬁrm the causality of rpoS mutations for the hyperbiofilm phenotype, we constructed a de novo mutant allele with a SNP on the ancestor PAO1 genome to yielded RpoSP251L and RpoSQ266stop mutant strains. We found that single point mutation in rpoS could produce the hyperbiofilm phenotype (Figure 2A). We also tested the biofilm formation ability of ΔrpoS strain and confirmed that knockout rpoS in P. aeruginosa PAO1 indeed increased the biofilm formation (Figure 3A). Complementing the mutation strains with wild type rpoS reverted the hyperbiofilm phenotype to the wild-type level (Figure 2A).
Protein domain analysis showed that RpoS consist of 4 regions, and region 4 contains a DNA binding domain. P251L located on the end of region 3, knockout this region increased the biofilm formation. Q266stop mutation leads to RpoS lacking region 4, knockout region 4 has the same phenotype of Q266stop mutation. Moreover, we have constructed region 3 and 4, region 2, 3 and 4 and region 1 deletion strains, all of those region deletion mutants produced the hyperbiofilm phenotype (Figure 2B). Together, these results provide deﬁnitive evidence that the SNPs identiﬁed are necessary and sufﬁcient to cause hyperbiofilm phenotype in P. aeruginosa. Moreover, P251L and Q266stop mutations are very likely lead to the inactivation of RpoS.
Hyperbiofilm variants outcompete the ancestral strain during biofilm competitions
The convergent emergement of P. aeruginosa hyperbiofilm variants from independent lineages suggested a competitive advantage for these variants over the ancestor. We have previously showed that P. aeruginosa cells did not share its EPS with its neighboring cells  and thus we hypothesized that the evolved hyperbiofilm variants only gain advantage to the ancestor strain under biofilm growth condition. We then tested the competition of hyperbiofilm variants with the ancestral strain in both planktonic cultures and biofilms. The fluorescently tagged hyperbiofilm variants (tagged with mCherry) were mixed with ancestral strain (tagged with GFP) in different ratio and inoculated into the bead-containing 24 well microplate. After 24 h cultivation, the planktonic and biofilm cells were analyzed with flow cytometry. We found that, the planktonic cells of C6W1C was slightly higher than ancestral strain when inoculated at the same ratio (Figure 3A), while no difference was found between C3W6F and C6W6F with ancestral strain when inoculated at the same ratio (Figure 3B and C). Interestingly, the competitions in biofilms conﬁrmed that the hyperbiofilm variants have a signiﬁcant and predominant selective advantage against the ancestral strain (Figure 3A, B and C). Next, we increased the inoculation ratio of ancestor and hyperbiofilm variants to 5:1, the proportion of ancestor within biofilm was still much less than hyperbiofilm variants (Figure 3D, E and F). These results indicated that the hyperbiofilm variants increased the competition only in biofilms and this phenotype is not related to the change of growth rate, drug resistance and tolerance. Moreover, the competition advantage of hyperbiofilm variants is growth model specific and did not occur in planktonic culture.
Previous studies showed that acquired mutations conferring beneficial traits such as antibiotic resistance will dominate when exposing biofilm bacteria to high concentrations of antibiotic . Therefore, the enrichment of the hyperbiofilm variants could have been achieved by accumulation of mutations that conferred resistance to imipenem. We found, however, that the MIC of imipenem for colonies isolated from the evolved lines (C6W1C, C3W6F and C6W6F) was indistinguishable from that for their ancestor (Figure 3G). One of the most straightforward ways to gain a competitive advantage is increasing the growth rate. To test this point, we measured the growth rates of the evolved hyperbiofilm variants and the ancestor in LB medium. Whereas, there is no significant difference between the hyperbiofilm variants and ancestor (Figure 3H). Next, we measured the biofilm growth curve of PAO1, ΔrpoS and the complementation strain ΔrpoS/p-rpoS, the CFU of planktonic cultures were measured at the same time. We found that, the CFU of planktonic culture of all three strains showed no significant difference (Figure 3I). For the biofilm growth, the ΔrpoS strain indeed produce more biofilm than PAO1 after 6 h incubation and developed 2 log of biofilm CFU after 24 h incubation (Figure 3I). These results indicated that it is not the resistance level and planktonic growth rate select ΔrpoS mutant; whereas ΔrpoS has a prolonged biofilm mode of growth which eventually leads to occupation of the biofilm community.
Mutations in rpoS lead to an elevated intracellular c-di-GMP levels
Quorum-sensing (QS)  and c-di-GMP  have been well documented to play important roles in P. aeruginosa biofilm formation. To assess whether quorum-sensing and c-di-GMP levels were elevated, we introduced the quorum-sensing and c-di-GMP reporter systems [40-43] into the variants isolated from biofilm evolution experiments and ancestral strain to determine the relative level of corresponding signaling pathways. We found that the ﬂuorescent signal of PlasB-gfp and PrhlA-gfp in ancestral strain were higher than that of hyperbiofilm variants (Figure S2A and B), while there were no differences in ﬂuorescent signal of PpqsA-gfp between the ancestral and hyperbiofilm variants (Figure S2C). For the ﬂuorescent signal of PcdrA-gfp, the hyperbiofilm variants showed 2-fold higher in expression level than the ancestral strain, indicating that the hyperbiofilm variants might have elevated intracellular c-di-GMP levels (Figure S2D). We further showed that the PcdrA-gfp expression level were increased in RpoSP251L, RpoSQ266stop and ΔrpoS strain compared to the PAO1 wild-type strain (Figure 4A). The second messenger c-di-GMP is a key regulator of P. aeruginosa biofilm formation, which is synthesized from two GTP molecules by diguanylate cyclases (DGC) and is degraded into 5’-phosphoguanylyl-(3’-5’) guanosine (pGpG) and/or GMP by phosphodiesterases (PDE) . Till now, 43 DGC and PDE proteins have been identified in P. aeruginosa .
To investigate the mechanism underlying rpoS point mutation-induced increasing in intracellular c-di-GMP content, we performed transcriptomic analysis of PAO1, RpoSP251L and RpoSQ266stop strains using RNA-sequencing. Samples were collected after 8.5 h culture owing to the PcdrA-gfp ﬂuorescent intensity (Figure S3) between mutants and wild type PAO1 strain have the biggest difference at this time point. We found that, 15 DGC and PDE proteins were upregulated at least 2-fold in both RpoSP251L and RpoSQ266stop strains compared to PAO1 (Table 1). This result indicated that the c-di-GMP metabolism in rpoS mutant strains were more active than PAO1.
RpoS regulates the expression of small regulatory RNAs rsmY and rsmZ in Legionella pneumophila . Moreover, rsmY/Z participate in the regulation of c-di-GMP production in P. aeruginosa, the c-di-GMP levels were strongly reduced in the rsmY/Z double deletion mutant . Our transcriptomic analysis showed that the expression of rsmY and rsmZ was increased 3.84 and 5.04-fold in RpoSP251L compared to the PAO1 wild-type, respectively. Next, we measured the expression of rsmY/Z in PAO1, RpoSP251L, RpoSQ266stop and ΔrpoS strains using reporter fusions . We found the rsmY/Z expressions were increased in RpoSP251L, RpoSQ266stop and ΔrpoS (Figure 4B and C), which is consistent with the increased level of c-di-GMP of these mutants. These results showed that the mutation of rpoS has led to the increase in rsmY/Z expression and intracellular c-di-GMP content in P. aeruginosa.
rpoS mutation associated hyperbiofilm phenotype in clinical isolates
Our experimental biofilm evolution data has revealed that single-nucleotide mutations on rpoS confer P. aeruginosa hyperbiofilm phenotype and produce a pronounced competitive advantage within the biofilm microenvironment. In order to analyze the preference of rpoS mutation, we downloaded 4000 sequences of rpoS from pseudomonas genome database (www.pseudomonas.com). Through comparative analysis, we have identified 241 non-synonymous mutations (6.03% of total sequence), 8 insertion or deletion mutations (0.2% of total sequence) and 5 stop coding mutations (0.13% of total sequence) compared to the PAO1 wild-type strain. Among those mutations, 123 mutations were located on the inter-region of rpoS and 131 mutations were within 4 regions (Figure 5A). Moreover, we have identified 2 sequences harbored RpoSP251L mutation. We also analyzed the top 5 mutation sites among 4000 sequences, and found L268Q was the top one with 71 sequences (Figure 5B).
Since imipenem has been used for clinical treatment of P. aeruginosa infection, we wondered whether rpoS mutation caused hyperbiofilm strains exist in clinical isolates. Therefore, we examined the biofilm formation capacity of 288 clinical P. aeruginosa isolates obtained from the patients with culture confirmed P. aeruginosa infections (Table S1). Through quantitative analysis of biofilm formation by measuring crystal violet staining at OD550 nm and total bacterial growth at OD600 nm to exclude growth variation, we identified 29 hyperbiofilm isolates (10.07 % of total isolates) in this collection (Figure 5C). Next, we target sequenced rpoS of 29 hyperbiofilm isolates and confirmed that #16 isolate harbored non-synonymous mutation in rpoS. Interestingly, #16 isolate, which is isolated form the peritoneal drainage fluid, has the same mutation RpoSP251L as our experimental evolved variant C6W6F.
The evolved rpoS variants are hypervirulent
Pyocyanin production is one of the major virulence factors of P. aeruginosa, and plays an important role in P. aeruginosa pathogenesis by causes oxidative stress to the host, induces apoptosis in neutrophils and inhibits phagocytosis of macrophages [47, 48]. Previous studies showed that the pyocyanin production was increased in a rpoS-deletion mutant . In order to test the impact of rpoS point mutation on our biofilm evolved variants on pyocyanin production, we compared the production of pyocyanin by P. aeruginosa PAO1 strain, RpoSP251L, RpoSQ266stop and ΔrpoS. As we expected, similar to the ΔrpoS mutant, the P. aeruginosa RpoSP251L and RpoSQ266stop produced higher amounts of pyocyanin than the wild-type PAO1 strain (Figure 6A). This result suggests that point mutations accumulated in the rpoS gene in P. aeruginosa clinical isolates have similar effect as rpoS gene deletion on its physiology.
Next, we further assessed the impact of evolved rpoS point mutations on virulence using the macrophage cytotoxicity model . The RAW264.7 macrophages were infected with P. aeruginosa PAO1, RpoSP251L, RpoSQ266stop and ΔrpoS, and the release of cytosolic lactate dehydrogenase (LDH) was determined. We found that macrophages infected with RpoSP251L, RpoSQ266stop and ΔrpoS released more LDH compared to P. aeruginosa PAO1 after 4 h infection (Fig 6B), suggesting mutation on rpoS can induce the cell death of macrophage. Altogether, there results suggest that mutations on rpoS can enhance the virulence in P. aeruginosa.