The cystic fibrosis (CF) lung is a spatially complex and inflamed environment that provides beneficial growth conditions for a number of bacterial species, including the opportunistic pathogen Pseudomonas aeruginosa [1,2,3,4,5]. Chronic infection of CF lungs with highly adapted and antibiotic-resistant biofilms of P. aeruginosa is a major cause of lung function decline, which results in a concomitant increase in morbidity and mortality in individuals with CF [3, 4, 6,7,8,9,10,11]. Longitudinal studies of chronic CF infections with P. aeruginosa have revealed that patients become infected at a young age with an environmental or transmissible isolate that evolves and adapts over time to the lung environment [12,13,14,15,16,17,18]. Studies on P. aeruginosa populations isolated from individual lungs, have demonstrated divergent evolution, resulting in heterogeneous populations of P. aeruginosa within patients [16, 17, 19]. This genetic adaptation and diversification are likely to impact on levels of pathogenicity and the efficacy of antibiotic treatment [16, 20], and could potentially impact how other species of microbes colonize the CF lung [4, 21,22,23,24,25,26,27].

Studies on explanted CF lungs have shown that the spatial structure found within lungs, and physical separation of infecting isolates, plays a role in generating the vast phenotypic and genotypic heterogeneity seen within P. aeruginosa populations in individual patients [16, 17, 28, 29]. Major adaptations of P. aeruginosa to the CF lung include alginate production, loss of quorum sensing (QS), hypermutability, and increased resistance to antimicrobials [7, 8, 14, 30, 31]. Heterogeneity in P. aeruginosa populations has also been explained by early divergent evolution and adaptation to differential ecological niches [32] and recombination between isolates residing in the airways [16, 33]; however, the exact mechanisms leading to heterogeneity have yet to be fully elucidated. Further, while it is accepted that genomic heterogeneity arises in CF chronic lung infections, it remains unknown how genotypic changes shape community functions within whole P. aeruginosa populations. Understanding what drives community structure and function remains a key goal in microbial ecology, because the overall community function is determined by all the individuals in the population.

In this study, we hypothesized that the combination of wild-type and mutated alleles in populations shapes community functions, which could result in clinically relevant outcomes such as increased or decreased antibiotic tolerance and traits important for virulence. To test this, we first evolved P. aeruginosa PAO1 in biofilms on plastic beads [34] for up to 50 days. This bead biofilm system has previously been successfully used to study genetic adaptation and phenotypic diversity of Burkholderia cenocepacia and P. aeruginosa to different environmental conditions [35]. Our study differs from this previous work, in that our major focus was on community function within populations rather than on specific isolates and defining ecological niches. We grew our bead-associated biofilms in a synthetic CF sputum media (SCFM), which recapitulates the chemical environment found in CF sputum [36,37,38]. Our study therefore generated phenotypically and genotypically heterogeneous populations of P. aeruginosa in a spatially structured environment chemically relevant to CF sputum. We utilized these heterogeneous evolved populations to study the changes in the functional community phenotypes instead of the traditional approach of working with single-evolved isolates. We used a metagenomic approach to assess genetic alterations within evolving populations, and we monitored fluctuations in allele frequency during the selection process. To determine the impact of genomic heterogeneity within populations on various phenotypes, we assessed collective phenotypic traits (community function) of the evolved populations.

One of the most commonly described adaptations of P. aeruginosa to the CF lung is the loss of the las QS system, predominantly through point mutations, frameshifts, and deletions in the lasR gene [17, 39,40,41]. We used our evolved populations to specifically focus on the impact of lasR mutation frequency on QS phenotypes. The lasR gene encodes the LasR transcriptional regulator, which binds the QS signal N-(3-oxo-dodecanoyl)-l-homoserine lactone (3O-C12-HSL) [42,43,44,45]. LasR-bound 3O-C12-HSL controls the transcription of ~10% of the P. aeruginosa genome, including a number of genes involved in social behaviors, pathogenesis, antibiotic resistance, and biofilm formation [46,47,48,49]. Despite a number of previous studies that describe the changes and adaptation of various lineages of P. aeruginosa in CF lungs [14, 50,51,52,53], it remains unclear how polymorphisms in the lasR gene impact on community function within evolving heterogeneous P. aeruginosa populations. This is because most previous studies that focused on within-host adaptation of P. aeruginosa used single colonies isolated from temporal CF sputum samples and not whole populations [12, 14, 15, 19, 54].

Overall, we found that (i) evolutionary trajectories were reproducible between independently evolving populations, and that over 60% of genomic changes in populations occurred within the first 10 days of selection; (ii) after 30 days of evolution in SCFM, the evolved communities displayed an increase in lasR mutant frequency and a decrease in QS-dependent traits; (iii) there was a significant correlation between lasR mutant frequency, the loss of social traits, and an increase in tolerance to β-lactam antibiotics. Our findings provide insights into how allelic polymorphism and population heterogeneity, in general, can impact on phenotypes and community functions within evolving P. aeruginosa populations. Further, we demonstrate that changes in P. aeruginosa population dynamics can alter factors associated with virulence and provide explanations for increased antibiotic tolerance, even in situations when antibiotics have not been used.


Genomic variation in evolving biofilm populations over 50 days of selection in SCFM

We evolved the P. aeruginosa strain PAO1 for 50 days (≈800 generations) in biofilms using a previously described biofilm bead method [34], and a growth medium that chemically mimics CF sputum (SCFM), and where the physiology of P. aeruginosa is similar to when grown in human sputum [36,37,38]. Our experimental evolution approach contained four independent replicate lines (Fig. S1). We collected and stored biofilm-evolved populations after 10, 20, 30, 40, and 50 days of evolution (Rounds 1–5: R1–R5). We used the Illumina MiSeq platform to deep-sequence evolved populations in order to determine genomic changes through time. We also sequenced our laboratory PAO1 ancestral strain, and after de novo genome assembly of this strain, we mapped the sequence reads of the evolved populations to the ancestor in order to detect SNPs [55]. Our SNP calling analysis, combined with an analysis of allele frequency, revealed that in all four independent replicate lines, an average of 282 ± 13 SNPs occurred in the populations after 10 days of selection (Fig. 1a). We found that around 60% of these SNPs were present through all other rounds of selection (Figs. 1b and S2). This suggests that the evolutionary trajectories of biofilm growth in SCFM are similar in independently evolving populations, and that the major genetic heterogeneity in evolving populations occurs during the early phases of selection.

Fig. 1: The evolutionary trajectory of P. aeruginosa during biofilm growth in SCFM.
figure 1

The evolution trajectories were similar in four independent replicate lines, and the major genetic heterogeneity in the evolved populations occurred during the first 10 days of selection. a SNP calling analysis showed that the four independently evolved populations had more than 80% similarity (267 shared SNPs) in genomic changes in the populations over the 50 days of selection. b The Venn diagram shows the shared SNPs between all four replicate lines, over 50 days of selection. Numbers in brackets (i.e., 1354 for 10 days) represent the combined number of SNPs in each of the four evolved lines. The majority of genetic variation within the populations occurred within the first 10 days of selection. Each circle represents the total number of SNPs for each independently evolved line.

SNP frequency in genes involved in social traits fluctuates over time

In our evolution experiment, we found emergence of polymorphisms in 45 genes involved in various physiological functions (Fig. S3 and Table S1). We found that between 10 and 25% of SNPs were fixed (frequency of 1) in the populations over 50 days of selection across all four replicate evolution lines (Fig. S4). When we focused on the allele frequency and not the number of positions altered in each coding region, we found that the frequency of SNPs in genes involved in different traits changed during the course of the experiment (Figs. 2S3S4, and S5). The genes highlighted in Fig. 2 are genes that have previously been shown to be commonly mutated in P. aeruginosa CF isolates [9, 15, 16, 29, 30, 56].

Fig. 2: Allele frequency of SNPs changes over the course of selection.
figure 2

Emergence of nonsynonymous SNPs in genes involved in social traits (QS: lasR), oxidative respiration and DNA mismatch repair, motility, and iron chelation occurred after 20 and 30 days of selection, and fixed SNPs in genes such as elongation factor tufA, Cytochrome c oxidase subunit (ccoN2), and pyoverdine biosynthesis protein pvdJ emerged after 10 days of selection, and were fixed in the populations over 50 days (ad represent replicate evolving lines).

The allele frequencies of nonsynonymous SNPs in ccoN2 (PA1557), and synonymous SNPs in pvdJ (PA2400) and tufA (PA4265) became fixed in the population at a frequency of 1, while the frequency of nonsynonymous SNPs in phzC2 (PA1901) and pvdD (PA2399) fluctuated between 0.4 and 0.5 in different rounds of selection. We detected a number of SNPs occurring in mutS (PA3620), pilQ (PA5040), and pilN (PA5043) at 10, 20, and 30 days of selection in all four independent evolved lines. We observed an increase in lasR (PA1430) mutant allele frequency between 30 and 40 days of selection (Figs. 2, S3, and S5).

Accumulation of SNPs shapes community functions in evolved P. aeruginosa populations

We next examined the production of phenotypic social traits in evolved populations in order to determine changes in community function of the genetically heterogeneous evolving populations. We measured the levels of biofilm formation, QS signals, total protease, and the siderophores pyoverdine and pyochelin. We observed a small but significant increase in biofilm formation by evolved populations when compared with the PAO1 ancestor (Fig. 3a). After 30 days of selection, the production of total protease (Fig. 3b) and the 3O-C12-HSL QS signal (Fig. 3c) decreased in evolved populations; however, the levels of C4-HSL signal (Fig. 3c), and pyochelin and pyoverdine (Fig. 3d), did not follow this trend, and any changes were generally not significantly different from the values of the ancestor strain. We also observed an increase in colony morphology types (morphotypes) in the evolved populations starting after 20 days of selection (Fig. S6).

Fig. 3: Phenotypes of evolved populations of PAO1.
figure 3

a Increase in biofilm formation by evolved populations of PAO1 (Kruskal–Wallis, Dunn’s multiple comparison with PAO1, p < 0.05, n = 5. Errors bars indicate the standard error of the mean, SEM). b The total protease activity was reduced at 30 days of selection. c Production of 3O-C12-HSL significantly decreased at 30 days of selection (Kruskal–Wallis, uncorrected Dunn’s test multiple comparison to PAO1, p < 0.05, n = 5. Errors bars indicate the SEM), and there were no significant changes in the levels of C4-HSL. d There were no significant changes in siderophore production in evolved populations compared with the PAO1 ancestor (Kruskal–Wallis, uncorrected Dunn’s test multiple comparison to PAO1, p > 0.05, n = 5). Each dot represents the mean value of one independent experiment for each evolved population at each round of selection. All experiments were performed in three technical replicates.

To determine whether the emergence and increase in the frequency of SNPs in the populations impact upon community function, we used a linear regression model. We focused on changes in the allele frequency detected in the QS regulator LasR. We chose lasR because the genes and phenotypes it regulates in P. aeruginosa are well understood [30, 40, 44, 47, 57] and also, lasR mutants are regularly isolated from CF sputum [13, 50, 58]. We detected a nonsynonymous SNP (V208G) in the DNA-binding domain of LasR between 30 and 40 days of selection (Fig. S5). Interestingly, SNPs in the same position (V208) have been identified in P. aeruginosa isolates collected from CF sputum samples [40]. V208 is adjacent to the D209 residue in the LasR DNA-binding domain [59, 60], suggesting an impact on the structure of LasR and its DNA-binding affinity. We assessed the impact of lasR V208G SNP frequency on changes on two QS-dependent phenotypic traits: production of 3O-C12-HSL signal and total protease. We found a significant negative correlation between the frequency of the V208G lasR SNP in the whole-evolved populations and the total protease activity of evolved populations. We found that 87% (R2 = 0.8704, F = 20.15; p = 0.0206) of the decreased protease activity was correlated with the accumulation of the lasR mutation in the populations (Fig. 4a). We used the same analysis to determine whether lasR mutant accumulation impacts on 3O-C12-HSL production within populations. We found that only 53% of the decreased 3O-C12-HSL levels (R2 = 0.5363, F = 3.469; p = 0.1594) can be explained by the accumulation of lasR mutation in the populations (Fig. 4b). However, when we only included the last 30 days of selection in our analysis, the decrease in 3O-C12-HSL levels could be fully (100%) explained by the accumulation of lasR mutants (R2 = 1.0, p = 0.0034) (Fig. 4b). Our analysis did not show any correlation between the frequency of the lasR SNP and changes in the production of C4-HSL (Fig. 4c) or biofilm formation (Fig. 4d).

Fig. 4: Loss of protease activity and a decrease in QS signal production in evolved populations can be explained by an accumulation of lasR SNPs.
figure 4

A linear regression model showed that the loss of total protease activity and the decrease in 3O-C12-HSL levels can be explained by the accumulation of lasR SNPs. a There is a significant negative correlation between lasR SNP frequency and total protease activity of the evolved populations (R2 = 0.7298, p = 0.04; n = 8). b There is no significant correlation between 3O-C12-HSL levels and accumulation of the lasR SNP in the evolved populations (R2 = 0.5363, p = 0.1594; n = 8). However, if only changes in 3O-C12-HSL levels after 30 days of selection are considered, there is a strong correlation between emergence and accumulation of lasR SNP in evolved populations (small inset: R2 = 1.0; p = 0.0034; n = 8). c There are no significant correlations between accumulation of lasR SNPs in the evolved populations and levels of C4-HSL (R2 = 0.1187, p = 0.5702; n = 8). d There is no significant correlation between lasR frequency and biofilm formation (R2 = 0.0004, p = 0.9062; n = 8).

Accumulation of lasR mutants in evolved populations leads to increased tolerance to β-lactam antibiotics

Previously, it has been shown that lasR mutants display increased β-lactamase activity, and therefore increased resistance to β-lactam antibiotics (such as ceftazidime) [61] that are routinely used in CF clinics. To determine possible links between the loss of lasR function and changes in tolerance to routinely used antibiotics, we first assessed the antimicrobial susceptibility levels of the evolved populations. When we tested the levels of antimicrobial susceptibility to six routinely used antibiotics for chronic CF lung infection, after 30 days of selection, the evolved populations showed an increased tolerance (indicated by a decrease in the zone of inhibition) to three antibiotics: ceftazidime, piperacillin/tazobactam, and meropenem that are all β-lactam-class antibiotics (Figs. 5 and S7). We then tested for correlations between the frequencies of the lasR V208G SNP and resistance to ceftazidime and piperacillin/tazobactam. We observed a positive and significant correlation between accumulation of lasR mutants and the increased resistance to both ceftazidime and piperacillin/tazobactam (Fig. 5a, b). We also tested whether the increase in tolerance could be explained by an increase in biofilm formation. We found that there was no significant correlation, suggesting that increases in biofilm formation do not necessarily translate to an increase in drug tolerance (Fig. S7C).

Fig. 5: Increased resistance to ceftazidime and piperacillin/tazobactam in evolved populations is correlated with lasR SNP frequency.
figure 5

There is an increased resistance, indicated by a decrease in the zone of inhibition (ZOI) to ceftazidime (a) and piperacillin/tazobactam (b) in evolved populations after 30 days of selection (R3). The red lines represent the ZOI breakpoint (mm) for each antibiotic (BSAC guidelines). Below, the lines indicate resistance. There is a positive correlation between increased resistance and the frequency of the lasR SNP in the evolved populations (n = 8, Pearson correlation. Each dot represents the mean value for the zone of inhibition at each round of selection in all eight independent experiments). c Complementation of a 50-day evolved isolate (BB8) containing a SNP in lasR with an intact lasR in trans, restores sensitivity to ceftazidime and piperacillin/tazobactam. Differential frequencies of B8 mixed with ancestral PAO1 significantly increased tolerance to ceftazidime (n = 5, two-way ANOVA, multiple comparison to PAO1, Fisher’s LSD test, p < 0.05. Errors bars indicate SEM).

To confirm that our evolved populations contained lasR mutants with increased β-lactam tolerance, and that lasR frequency affects community function, we first selected and sequenced an evolved isolate (BB8) from the final round of selection (50 days) with no protease activity, and which demonstrated increased antibiotic tolerance compared with the ancestor PAO1 (Fig. 5c). The sequencing of isolate BB8 revealed the same SNP in the lasR gene that we detected in the evolved populations (V208G) (Table S2). We complemented BB8 with an intact copy of the PAO1 lasR gene in trans on a plasmid and then assessed β-lactam tolerance levels. We observed a decrease in tolerance to ceftazidime and piperacillin/tazobactam in the lasR-complemented BB8 isolate (Fig. 5c). Finally, we tested whether changes in the starting frequency of lasR mutants in mixed populations impacted on community functions. We found that when BB8 was mixed with wild-type cells, this resulted in an increased tolerance to ceftazidime (Fig. 5c) and a reduction in 3O-C12-HSL but not C4-HSL in mixed populations (Fig. S8).


Despite a number of recent studies focused on adaptive changes in evolved populations of P. aeruginosa in environments designed to mimic CF sputum [35, 62,63,64,65,66], there remain significant gaps in knowledge about how genomic and phenotypic diversity impacts upon interactions within populations and on community-level phenotypes. In our current study, we focused on understanding how the evolution and coexistence of multiple lineages of P. aeruginosa shape community functions. To generate diversity, we performed a 50-day selection experiment, evolving PAO1 as biofilms in an artificial sputum medium (SCFM) [36,37,38]. We observed (i) a rise in genetic diversity after 10 days of selection in biofilms grown in SCFM; (ii) up to 25% of SNPs became fixed in the population and carried in evolved populations through 50 days of selection; (iii) emergence and accumulation of SNPs in genes involved in motility, respiration, DNA mismatch repair, transcription, and QS emerged between 10 and 40 days of selection; (iv) accumulation of lasR SNPs correlated with decreased protease activity, 3O-C12-HSL production, and increased tolerance to β-lactam antibiotics, despite no prior treatment with any antibiotic.

The divergent evolution of P. aeruginosa during chronic infection of CF lungs has been the focus of numerous studies on longitudinal collections of sputum samples. Analysis of single P. aeruginosa isolates from these collections has identified genomic signatures for adaptation to the CF lung environment [9, 14, 29, 50, 53, 67, 68]. Similar studies on collections of P. aeruginosa isolates sourced from single-sputum samples, have shown considerable phenotypic and genetic diversification of P. aeruginosa strains that group into clades [16, 17], and which may colonize different ecological niches in CF lungs [28, 69]. All of these studies focused on the diversification, adaptation, and heterogeneity of P. aeruginosa during chronic CF lung infection. However, the impact that this heterogeneity may have on collective phenotypes has been largely overlooked. We hypothesized that the combination of wild-type and mutated alleles in populations shapes collective phenotypes. We observed that the accumulation and frequency of various SNPs in heterogeneous populations of P. aeruginosa, significantly impact the collective phenotypes of evolved populations (Figs. 2, 3). Although we observed that the majority of genomic changes occurred during the first 10 days of selection, the functional outcome was only impactful after 20–30 days, where we observed an increase in biofilm formation and colony morphotypes, and a significant decrease in total protease activity and production of 3O-C12-HSL by evolved populations.

The biofilm bead system has previously been used to examine the evolution of P. aeruginosa in biofilms [35], although there are significant differences between this and our study. First, the environmental conditions were different, with the previous study growing in a minimal media (M63) while we used SCFM. Second, we evolved PAO1, whereas the previous study evolved PA14 [35]. While both are popular model strains for P. aeruginosa study, they differ in a number of factors including virulence [5]. Third, we focused primarily on the emergent properties of the whole community rather than individual isolates, which we show have important implications for signaling and antimicrobial tolerance. Finally, the previous study studied the emergence of mutL and mutS genes that may have contributed to the emergence of different individual morphotypes [35]. We observed low-frequency mutations in the mutS gene after 30 days (Fig. S3). As these were synonymous mutations, they did not appear to significantly increase the mutation rate after 30 days of selection.

Our study also shows similarities with another recent study. Here the authors evolved PA14 in SCFM supplemented with mucin to create a structured environment [62]. The main findings of this study were that mucin promoted diversification of P. aeruginosa between and not within populations. The authors found increases in tolerance to antibiotics in evolved populations, and changes in growth, motility, pyocyanin production, and biofilm formation. They did not present genomic evidence that underpinned these changes [62].

Our study identified genomic changes often observed in P. aeruginosa isolates taken from CF lungs, including changes in secretion genes (hcpA, hcpB, vgrG2a, and vgrG4b), motility genes (pilQ and pilN), iron aquisition genes (pvdD and pvdJ), and biofilm formation (cdrA). We also observed a consistent change in all lines and rounds of selection in anaerobic respiratory pathways (Figs. 2 and S3). ccoN2 (PA1557) encodes the cytochrome C oxidase subunit (ccb3 type). Increased expression of ccoN2 in a small colony variant has been previously noted during microaerophilic conditions [70, 71].

One of the best-known signatures of P. aeruginosa adaptation to the CF lung environment is mutation in the QS regulator LasR. Several studies on P. aeruginosa strains isolated from CF sputum samples found significant genomic changes in lasR, including truncation, deletion, frameshifts, and SNPs, with many resulting in a loss of function [30, 40, 41, 50, 72]. Although there can be a high percentage of lasR-deficient isolates collected from patients, studies have also shown differential frequencies of functional and intact lasR alleles within patients [13, 73]. Considering the importance of lasR-dependent social interactions for the fitness of P. aeruginosa, mutation frequency of lasR within populations of P. aeruginosa could significantly impact on QS-dependent phenotypes and fitness at the population level [74,75,76]. Previously, we have demonstrated that an increase in the frequency of QS social cheats (lasR mutants) in defined populations of wild-type and lasR mutants, leads to a reduction in cooperation and virulence in mouse models of infection [74]. These studies showed that even simple mixed genotype populations of P. aeruginosa can have a significant impact on community function, virulence, and the outcome of infection.

In our current study, we observed a negative correlation between an increase in the frequency of the lasR V208G SNP in evolved populations and levels of protease activity and 3O-C12-HSL production (Fig. 4). Mutation of lasR in P. aeruginosa isolates from CF lungs has also previously been shown to be important for increased tolerance to β-lactam antibiotics such as ceftazidime [19, 56, 61]. Here we examined whether the increase of lasR mutants in our evolved populations had an impact on the levels of antimicrobial tolerance. Despite no prior treatments with antibiotics, we observed an increased tolerance to three antibiotics from the β-lactam family (Fig. 5). In contrast, we found no correlation between increased biofilm production by evolved populations and antibiotic resistance (Fig. S7). Resistance of P. aeruginosa to antibiotics in chronic infections such as CF or wounds, is often thought to be due to specific mechanisms such as efflux pumps, or via production of excess polysaccharides such as alginate. Our findings suggest that the accumulation and frequency of genetic variants that might not traditionally be associated with resistance to drugs (e.g., QS mutants) within a heterogeneous population, can alter phenotypes within populations that can result in important clinical repercussions.

Our work suggests that in the future, we should consider metagenomic and metaphenotypic assessments of P. aeruginosa populations collected from CF patients, rather than focusing on single colonies. This is because the phenotype of populations is dictated by the frequencies of various alleles in the populations. Focusing on just single isolates sourced from infections or long-term evolution experiments, results in particular strains being characterized with certain phenotypes, which misrepresents what is found in the population as a whole. It becomes particularly problematic in studies focusing on single colonies from longitudinal samples, and when genomic sequencing predicts how a strain genetically evolves over time during an infection. Our findings may also be particularly relevant when considering whether a P. aeruginosa infection is resistant or sensitive to antibiotic treatments. Our findings may extend to other infections caused by P. aeruginosa, such as nonhealing chronic wounds, and they may also be relevant to other species of bacteria.

Materials and methods

Bacterial strains and growth conditions

For our experimental evolution, we used the PAO1 (University of Nottingham) strain of P. aeruginosa. For SCFM, we followed the protocol provided in refs. [36, 38]. Briefly for the buffer base, we prepared NaH2PO4 (1.3 mM), Na2HPO4 (1.2 mM), KNO3 (0.348 mM), K2SO4 (0.271 mM), NH4Cl (2.28 mM), and KCl (14.9 mM); NaCl (51.8 mM) was prepared in 10 mM of MOPS at pH = 6.8; then the amino acids were added [36, 38]. Dextrose (3 mM), l-lactic acid (9.3 mM), CaCl2*2 H2O (1.75 mM), MgCl2* 6H2O (0.606 mM), and Fe.SO4*7H2O (0.0036 mM) were added fresh every time the media was prepared.

Long-term experimental evolution

To assess how genomic diversity impacts on P. aeruginosa populations, we generated a diverse population using a long-term evolution experimental approach. We evolved the P. aeruginosa strain PAO1 in biofilms using plastic beads [34, 35, 77] (9 × 6-mm width) suspended in SCFM, in order to mimic a biofilm life cycle and a chemical environment similar to that found in CF lung sputum. To start the experimental evolution process, we first grew PAO1 on an LB agar plate. Then we inoculated a single colony of PAO1 into 3 ml of fresh SCFM [38], and incubated for up to 6 h to grow up to the mid-log phase. We stored this mid-log-phase cells as the ancestral PAO1 and compared all further phenotypic and genomic properties of the evolved populations with this. We inoculated the mid-log-phase cells to OD600 ≈ 0.05 into four tubes (in order to evolve separate independent lines named A–D) containing 3 ml of SCFM and a plastic bead. We incubated the cultures for 24 h at 37 °C/200 rpm. After 24 h of incubation, we transferred bacterial covered beads into fresh tubes containing 3 ml of SCFM and a new bead. After each round, the biofilms that formed on beads were composed of ~108 cells. We then incubated again for 24 h at 37 °C/200 rpm. We continued the bead transfers for 50 days and stored the biofilm portion of populations (attached to plastic beads) every 10 days (Rounds 1–5: R1–R5) (Fig. S1). We did not transfer or store the planktonic fractions at any point during the selection experiment.

Deep sequencing of evolved populations

We extracted genomic DNA from evolved populations after 18 h of growth in SCFM, using DNeasy® Blood & Tissue Kit (QIAGEN) by following the manufacturer’s instructions. We prepared sequencing libraries using the NexteraXT protocol (Illumina), and sequenced in 24-plex on the Illumina MiSeq platform to obtain an approximate calculated level of coverage of 220–600× for each evolved population. A de novo assembly of the ancestral strain genome (P. aeruginosa PAO1 ancestor) was obtained using Spades with the –careful flag, and annotated using Prokka. We mapped reads of the evolved populations against the ancestral PAO1 genome using BWA [78]. For SNP calling, the sequences were summarized using a MATLAB script for base quality, genomic position, and mapping-quality [55] script using bowtie2 [79], SAMtools [80], and BAMtools. To determine the allele frequency, we applied the breseq  polymorphism mode [81] to each of the samples collected. Later, for the variant sites, we used the composition in each nucleotide as the surrogacy for the allele frequency, and we excluded any allele frequency below 10% in each evolved population. Our deep-sequence analysis was not designed to detect any insertions or deletions (INDELS), only SNPs. It is likely that INDELs were present in genes in our evolved populations.

Measurement of biofilms formed on beads

To determine the levels of biofilm formation by each evolved population, we grew biofilms on plastic beads as previously described [34, 77]. For each set of biofilm assays, we directly inoculated a 10-μl loop of frozen evolved populations (~106 cells) into 3 ml of SCFM and incubated at 37 °C/200 rpm for 16 h. Then we measured the OD600, and diluted in 3 ml of SCFM to OD600 ≈ 0.05. Then three plastic beads were added to each tube. Growth was monitored at 24 h at 37 °C/200 rpm and biofilm formation. The beads were then washed 3× with 10 ml of PBS to remove any residual planktonic cells not bound to the plastic beads. Then we transferred each plastic bead into 1 ml of PBS and sonicated the beads for 10 min, using a bath sonicator to detach biofilm-forming cells from the beads. We then serially diluted the cells and plated out onto LB agar plates for colony-forming unit (CFU) calculations. To assess other phenotypic traits, a cell-free supernatant of biofilm-forming cells was prepared from the liquid part of the cultures and corrected values for OD600 ≈ 1.

Preparation of cell-free supernatants

To assess the levels of protease, QS signals, and siderophore production during biofilm formation, we collected 3 ml of the SCFM used in the biofilm assays. We used the media surrounding the bead biofilms, and measured the OD600 and adjusted it to 1 for all the cultures with SCFM. We then filtered the supernatants using 0.22-µm filters and used these cell-free supernatants to measure phenotypic traits.

Total protease activity

To assess the total protease activity of evolved populations, we used skimmed milk agar plates. We inoculated 10 µl of cell-free supernatant from each evolved population onto skimmed milk agar plates (1.2% Bacto Agar, 0.015% of skimmed milk) alongside 10 μl of 10 μg/ml of proteinase K and supernatant of PAO1 as controls. The zone of clearance was scored based on appearance and measured with a ruler (in mm). We then imaged each plate using an Epson scanner at 800 dpi. We then compared it with the zone of clearance produced by the PAO1 ancestor [82].

Siderophore production

To measure the levels of the two main siderophores produced by P. aeruginosa, we used the cell-free supernatants. In total, 100-μl aliquots of cell-free supernatant from evolved populations and the PAO1 ancestor was transferred into a black clear-bottom 96-well plate (Corning). We measured the emission as relative fluorescent units (RFU) using a multi microplate reader (Tecan Infinite® M200 Pro). We measured the wavelengths at excitation of 400 nm/emission 460 nm for pyoverdine and 350/430 nm for pyochelin [30, 83, 84]. We corrected the values for pyoverdine and pyochelin to the absorption at (OD600) of the original cultures.

Measurement of C4-HSL and 3O-C12-HSL produced by evolved populations

The cell-free supernatants were used to determine the concentration of QS signals. We used two E. coli bioreporter strains to measure the production of the two main signal molecules by the evolved population. The E. coli reporters pSB536 and pSB1142 were used to detect C4-HSL and 3O-C12-HSL, respectively [85]. We calculated signal levels based on standard curves fitted to the concentrations of synthetic 3O-C12-HSL and C4-HSL standards (Sigma) [86, 87].

Mixed isolate experiments

For mixed constructed populations, we grew PAO1, the BB8-evolved isolate, and BB8 complemented with lasR (BB8:lasR) and empty vector (BB8:vector) in SCFM to OD600 ≈ 0.5. We then mixed PAO1 and BB8 at two different starting frequencies (50:50 and 30:70) in 3 ml of SCFM. The cultures were then incubated for 16 h at 37 °C/200 rpm. We measured and adjusted the OD600 of each culture to OD600 ≈ 1, and prepared a cell-free supernatant for QS signal activity measurements.

Antibiotic susceptibility assay

To determine the antibiotic susceptibility in evolved populations, we followed the British Society for Antimicrobial Chemotherapy (BSAC) guidelines (Version 14, 05-01-2015) using Isosensitest agar plates (Oxoid). We tested the susceptibility of evolved populations, PAO1 ancestral strain and the NCTC (10662) PAO1 strain to Gentamicin (10 μg), Meropenem (10 μg), Ciprofloxacin (1 μg), Ceftazidime (30 μg), Piperacillin/Tazobactam (85 μg), and Amikacin (30 μg) (Oxoid). The zone of inhibition and clearance in this method was compared with the available zone of inhibition breakpoints for susceptibility (mm) for each tested antibiotic based on BSAC guidelines.

Determining colony morphology diversity in evolved populations

To determine the diversity in colony morphology in the biofilm-evolved population, we used a Congo Red-based agar media (1% agar, 1×M63 salts (3 g of monobasic KHPO4, 7 g of K2PO4, and 2 g of NH4.2SO4, pH adjusted to 7.4), 2.5 mM magnesium chloride, 0.4 mM calcium chloride, 0.1% casamino acids, 0.1% yeast extracts, 40 mg/l Congo red solution, 100 μM ferrous ammonium sulfate, and 0.4% glycerol) [82]. We recovered the evolved populations from beads and serially diluted the populations and then inoculated onto CRA plates alongside the PAO1 ancestor. We incubated the plates overnight at 37 °C, and for a further 4 days at 22 °C. The colonies were imaged using an Epson scanner at 800 dpi.

Complementation of lasR

To complement an evolved isolate (BB8) containing a lasR mutation with a functional lasR allele, we amplified a 920-bp product comprising lasR and 200 bp upstream of the lasR start codon that includes its native promoter, from genomic DNA isolated from wild-type ancestral PAO1. We cloned this PCR product into the shuttle vector pME6032, which replicates in both E. coli and P. aeruginosa [88] by Gibson assembly [89] using the commercially available NEBuilder HiFi DNA Assembly Cloning Kit (New England Biolabs, Ipswich, MA). We introduced the lasR complementation construct into BB8 by electroporation [90] and plating on selective media containing 300 µg/ml tetracycline. We then tested the susceptibility of the lasR-complemented strains to ceftazidime and piperacillin/tazobactam using the BSAC method. For mixed constructed populations, we grew PAO1, the BB8-evolved isolate, and BB8 complemented with lasR (BB8:lasR) and empty vector (BB8:vector) in SCFM to OD600 ≈ 0.5. We then mixed PAO1 and BB8 at two different starting frequencies (50:50 and 30:70) in 1 ml of PBS, and tested for antibiotic tolerance by following the BSAC guidelines (Version 14, 05-01-2015) using Isosensitest agar plates (Oxoid).

Statistical analysis

For statistical analysis of the phenotypic assays, we used GraphPad Prism 8.0. For analysis of SNP frequency, we used R package 3.6. We used the Interactive Venn [91] to analyze shared SNPs within and between evolved populations.

Publication of genome sequencing

All sequences described in this paper have been uploaded to the NCBI SRA database (accession number PRJNA613708).