The Nutritional Environment Is Sufficient To Select Coexisting Biofilm and Quorum Sensing Mutants of Pseudomonas aeruginosa

ABSTRACT

The evolution of bacterial populations during infections can be influenced by various factors including available nutrients, the immune system, and competing microbes, rendering it difficult to identify the specific forces that select on evolved traits. The genomes of Pseudomonas aeruginosa isolated from the airways of people with cystic fibrosis (CF), for example, have revealed commonly mutated genes, but which phenotypes led to their prevalence is often uncertain. Here, we focus on effects of nutritional components of the CF airway on genetic adaptations by P. aeruginosa grown in either well-mixed (planktonic) or biofilm-associated conditions. After only 80 generations of experimental evolution in a simple medium with glucose, lactate, and amino acids, all planktonic populations diversified into lineages with mutated genes common to CF infections: morA, encoding a regulator of biofilm formation, or lasR, encoding a quorum sensing regulator that modulates the expression of virulence factors. Although mutated quorum sensing is often thought to be selected in vivo due to altered virulence phenotypes or social cheating, isolates with lasR mutations demonstrated increased fitness when grown alone and outcompeted the ancestral PA14 strain. Nonsynonymous SNPs in morA increased fitness in a nutrient concentration-dependent manner during planktonic growth and surprisingly also increased biofilm production. Populations propagated in biofilm conditions also acquired mutations in loci associated with chronic infections, including lasR and cyclic di-GMP regulators roeA and wspF. These findings demonstrate that nutrient conditions and biofilm selection are sufficient to select mutants with problematic clinical phenotypes including increased biofilm and altered quorum sensing.

IMPORTANCE Pseudomonas aeruginosa produces dangerous chronic infections that are known for their rapid diversification and recalcitrance to treatment. We performed evolution experiments to identify adaptations selected by two specific aspects of the CF respiratory environment: nutrient levels and surface attachment. Propagation of P. aeruginosa in nutrients present within the CF airway was sufficient to drive diversification into subpopulations with identical mutations in regulators of biofilm and quorum sensing to those arising during infection. Thus, the adaptation of opportunistic pathogens to nutrients found in the host may select mutants with phenotypes that complicate treatment and clearance of infection.

INTRODUCTION

During infection, a subset of spontaneous mutations in the pathogen population may produce dangerous clinical consequences including resistance to antimicrobial treatment and immune clearance (1–4). These become more probable with increasing generations of growth as populations expand or persist (5–7). It is therefore critical to characterize adaptations that are selected in vivo, the causative mutations, and the environmental factors that contribute to their selection. Longitudinal Pseudomonas aeruginosa isolates from many people with cystic fibrosis (CF) commonly share several evolved traits, including altered production of virulence factors (8), increased antimicrobial resistance (9), loss of O-antigen (10), and increased alginate production (11). These traits also often vary among isolates within patients and this diversity is thought to contribute to ineffective clearing by treatment (12, 13) and ultimately increased morbidity and mortality (14, 15). Whole-genome sequencing (WGS) of these isolates has identified convergent mutations in certain genes that persist within the CF respiratory environment (2, 8, 16). One of the most commonly mutated genes is lasR, encoding a transcriptional regulator of quorum sensing, and mutations in regulators of biofilm production are also common, among others (2, 17, 18). However, numerous factors could select for phenotypic and genotypic diversity in the CF airway (12, 19), making it challenging to infer causes of the prevalence of these mutations and phenotypes.

Evolution experiments in models of host systems can clarify which factors most influence pathogen fitness in vivo. The nutritional environment is increasingly appreciated as a major selective pressure, which motivated development of an artificial sputum medium that approximates the nutrient concentrations found in CF sputum and produces similar growth phenotypes and transcriptional responses as actual sputum (20–22). Evolution experiments with P. aeruginosa have been conducted in this medium to identify beneficial mutations in particular conditions, including antibiotic pressure (23), biofilm lifestyle (24), and presence of mucins (23, 25). Several observations are consistent across these studies, particularly the rapid diversification of biofilm, motility, and colony phenotypes. Mutations in certain genes have also been identified repeatedly including lasR, wsp, and flagellar synthesis genes (23, 24, 26, 27).

Because these common mutations produce broad pleiotropic effects, determining which phenotypes were initially adaptive within the host remains unresolved. For instance, the lasR gene encodes the transcriptional regulator of the LasRI acyl-homoserine lactone (AHL) quorum sensing system which regulates the expression of hundreds of genes, including other quorum sensing systems (RhlRI and PQS). Loss-of-function lasR mutations reduce production of extracellular proteases (28), improve fitness in microoxia (29), increase resistance to ceftazidime (24), improve growth in amino acids (30), and produce a “lysis and sheen” colony morphology (30). LasR mutants are frequently described as social cheaters because they can use public goods such as proteases produced by competitors without undergoing the cost of producing these products themselves (31, 32), although recent findings suggest that lasR mutants actually overproduce these public goods when cocultured with a LasR⁺ strain (33). In addition, mutations in regulators of cyclic di-GMP, a second messenger that promotes biofilm and suppresses motility when upregulated, may also produce a wide range of effects on cell cycle, virulence, and motility (34, 35). These mutations may be selected for the complete suite of new phenotypes they cause or perhaps only one of them. Modeling specific subsets of the infection environment is therefore required to quantify contributions of different selective pressures to pathogen evolution.

We sought to identify which adaptations to the CF airway may be related to the nutrient environment by performing an evolution experiment of P. aeruginosa strain PA14 in a defined medium containing carbon sources prevalent in CF sputum. Replicate populations were propagated either by serial broth dilution or in a biofilm model using beads to study the influence of lifestyle on mutant selection (36). We performed whole population genome sequencing of each lineage and sequenced isolated clones to infer population structure and to link evolved phenotypes to genotypes. We also characterized the consequences of these mutations on clinically relevant phenotypes including biofilm production and motility to examine how these adaptations could influence pathogenesis.

RESULTS

To examine the role of nutrients in P. aeruginosa adaptation to the CF airway, strain PA14 was propagated in substrates found to be prevalent in the cystic fibrosis respiratory tract (11 mM glucose, 10 mM LD-lactate, and 4 mM amino acids) for 12 days. The concentrations of these carbon sources differ from those in an established synthetic CF medium (SCFM; 3.2 mM glucose, 9.3 mM lactate, and 19 mM amino acids) but are within the range previously reported for CF sputum with the exception of increased glucose (21). Populations were propagated with either planktonic selection through a 1:100 dilution into fresh medium every 24 h or biofilm selection by transferring a colonized polystyrene bead to fresh medium every 24 h as previously described (37). Although other factors associated with our culture conditions also influence selection, we can identify mutations associated with growth in CF nutrients by comparing adaptations in this medium with previous experiments from our laboratory, which used different medium but an identical ancestral strain and transfer protocol (38). Population transfer sizes were ∼10⁸ cells in both planktonic and biofilm transfer methods, producing approximately 6.7 generations/day and 80 generations over the course of the experiment. We estimate that approximately 10⁶ new mutations occur each day in this experiment based on reported mutation rates of this and related strains (38–41), but these large effective population sizes ensure that only the fittest mutants within each population will rise to a detectable frequency within this short time frame (42). Thus, the mutations reported here almost certainly produce adaptations.

Genetic targets of selection in CF nutrients mimic in vivo adaptation.

We performed whole genome sequencing of five evolved planktonic and five evolved biofilm populations to read depths of 86-254X to identify selected genotypes (Fig. 1). After only 12 days, mutations in lasR (PA14_45960) and morA (PA14_60870) evolved within four of five planktonic populations. Mutations in lasR included large deletions encompassing both lasR and lasI genes (PA14_45920 to PA14_46440, Δ49,133 bp; and PA14_45800 to PA14_46240, Δ40,707 bp), nonsynonymous SNPs, and nonsense mutations (Fig. 2A). The lasR and lasI genes encode the regulator and autoinducer synthesis proteins, respectively, of the Las quorum sensing system (28, 43). The morA gene encodes a protein with both diguanylate cyclase (DGC) and phosphodiesterase (PDE) domains, which produce and degrade second messenger cyclic di-GMP, respectively (35). Nonsynonymous substitutions occurred at five different residues within morA (A1109V, K1123E, N1124K, E1153K, and L1155Q), indicating that they arose independently and were selected in parallel (Fig. 2B). Mutations in morA and lasR are two of the most frequently mutated genes in CF clinical isolates (2, 8, 30), and strikingly, this experiment selected mutations in some of the same residues altered during CF infections, including lasR P117 and A231 (8). We also identifed multiple instances of residue and domain-level parallelism between morA mutations detected in this experiment, other evolution experiments in SCFM or rich media, and clinical isolates (Fig. 2B; see Fig. S1 in the supplemental material) (2, 23, 44). This convergent evolution demonstrates that the nutrients found in CF sputum are sufficient to select for similar mutations as identified in clinical isolates.

FIG 1.

Propagation of P. aeruginosa in nutrients present in the CF airway rapidly selects for mutations in regulators of quorum sensing and cyclic di-GMP. Mutations were inferred from whole genome sequencing of evolved populations after 12 days of selection. All mutations detected by population WGS are indicated and genes that acquired multiple mutations are labeled. Details of detected variants are in the supplemental material. Allele frequency for each mutation is indicated by symbol size, mutation type by symbol shape, and function of the impacted loci by symbol color. Mutations detected by new junction evidence, which include insertions, large deletions, and structural rearrangements, are indicated at the position of the most upstream side of the junction.

FIG 2.

Convergent evolution of mutations in lasR and morA reveal sites under strong selection. (A) Two large deletions that include the lasI and lasR genes were detected as well as many nonsynonymous SNPs and missense mutations. (Top) The extent and overlapping region of these deletions; (bottom) evolved lasR mutations by relative position. (B) Mutations in morA were also frequently selected, primarily within the diguanylate cyclase (DGC) domain and linker region between the DGC and phosphodiesterase (PDE) domains (This Study). We also identified nonsynonynous morA mutations in other evolution experiments in our laboratory and other studies (Laboratory Isolates) (23, 44). Identical mutated sites have been identified in clinical isolates from patients with cystic fibrosis (Clinical Isolates) (2). Mutations are shaded by domain in which they occur.

Biofilm-adapted populations also nearly exclusively selected mutations associated with cystic fibrosis infection including genes regulating cyclic di-GMP, quorum sensing, and type IV pili (2, 4). All biofilm populations acquired at least two mutations in regulators of cyclic di-GMP (wspAF, roeA, and PA14_64050), often on different lineages at high frequencies (Fig. 1). These findings add to evidence that modulating levels of cyclic di-GMP is advantageous in a biofilm environment (45, 46). Yet, the PA14 genome harbors ∼40 genes containing DGC domains, PDE domains, or both that could alter cyclic di-GMP levels (47), so the repeated selection of mutations in only these three loci suggests that these regulators of cyclic di-GMP are not redundant but rather are environment specific, as shown elsewhere (48). Biofilm lineages 1 and 2 also acquired mutations in lasR, suggesting that altered quorum sensing was beneficial in both planktonic and biofilm environments. However, these regulators of quorum sensing and cyclic di-GMP were not among the most frequently mutated genes in identical evolution experiments from our laboratory using growth medium containing arginine as sole carbon source, indicating that their benefit is linked to available nutrients and concentrations (38). Most importantly, the almost exclusive selection of mutations known to arise during chronic infection in CF suggests that our simple nutritional model closely recapitulates several major selective forces acting in the CF airway.

Mutations in lasR, morA, and wspF underlie rapid morphological and phenotypic diversification.

Colony morphologies became conspicuously diverse in both biofilm and planktonic treatments after 12 days of selection, so we picked clones with representative phenotypes and sequenced their genomes to identify the causative mutations (Fig. S2 in the supplemental material). These genotypes demonstrated that lasR, morA, and wspF lineages arose independently and identified additional lasR and wspF mutations that were undetected by population WGS. Mutations in wspF or between the DGC and PDE domains of morA produced wrinkly or rugose small colony variants (RSCVs). We measured the motility and biofilm production of isolated mutants and found that lasR and morA mutants from the same population were functionally distinct (Fig. 3). All morA mutations decreased swimming motility and increased biofilm, whereas lasR mutants showed no change in biofilm or swimming motility, except for one that also had disrupted flagellar genes (PA14_45800 to PA14_46240 Δ40,707 bp), which lost swimming motility. Unsurprisingly, most mutants isolated from biofilm-evolved populations produced more biofilm and showed decreased swimming motility (Fig. 3) and swarming motility compared to the ancestor (Fig. S3 in the supplemental material). Therefore, similar phenotypes evolved in replicate populations associated with parallel genotypes.

FIG 3.

Mutations in lasR and in multiple cyclic di-GMP regulators produce coexisting subpopulations with distinct biofilm and motility phenotypes (A and B). Biofilm production by isolates from planktonic lineages (A) and biofilm lineages (B); (C and D) swimming motility by isolates from planktonic lineages (C) and biofilm lineages (D). Strains are labeled by genotype. Data points represent the average of technical replicates from at least three independent experiments and are shown with mean and 95% CI. Data were analyzed by one-way ANOVA [biofilm, F(16,34) = 32.76, P < 0.0001; swimming motility, F(16,41) = 72.67, P < 0.0001] with Tukey’s multiple-comparison test. Groups labeled with the same letter are not statistically different (P < 0.05).

morA mutations produce adaptations specific to high nutrient concentration.

The repeated selection of morA and lasR mutations in planktonic populations indicates that independently of any advantages of increased biofilm or altered virulence factor production in a host, they produce growth advantages. Prior studies have shown that morA mutant phenotypes depend on carbon sources (48) and that lasR mutants enhance growth in amino acids (30). To build upon these findings and to explore possible mechanisms explaining their coexistence, we measured mutant fitness in each of the carbon sources in the evolution medium: glucose, amino acids, and lactate (Fig. 4). A mutant with a nonsynonymous SNP in the linker domain of morA outcompeted the ancestor in medium containing all carbon sources but was less fit in media containing only a subset of the carbon sources (Fig. 4B). Other isolates with SNPs in morA also grew worse than the ancestor when cultured in only amino acids or only lactate (Fig. S4). These findings suggest that fitness of morA mutants may be influenced by nutrient identity, nutrient level, or both. To distinguish between these possibilities, we performed competition assays in various nutrient levels and found that morA mutant fitness was indeed dependent on nutrient concentration (Fig. 4B). For instance, halving the concentration of each of the carbon sources from net 25 mM to 12.5 mM in the evolution medium decreased morA mutant fitness. Therefore, nutrient abundance alone is sufficient for the selection of morA mutants, though nutrient composition may modulate the strength of this selection.

FIG 4.

Environment-specific fitness advantages of lasR and morA mutants. A marked ancestral strain was competed against the ancestral strain (A), an isolate with a morA SNP (E1153K) and an intergenic SNP between a tRNA and the tufB gene (B), an isolate with a SNP in lasR (R216Q) (C), and an isolate with a large deletion encompassing lasR and lasI (D). Competitions were performed in the medium used for the evolution experiment (All) as well as medium containing only a subset of the carbon sources (Glucose, Amino Acids and Lactate). The morA mutant was also competed in media in which the concentration of nutrients was doubled or halved to determine the effect of nutrient concentration on fitness (0.5X All, 0.5X Glucose and 2X Amino Acids). Data points represent fitness measurements from independent competitions spread across at least three batches of assays. Mean and 95% CI are shown. Within each genotype, statistical differences in fitness in each media were determined using ANOVA [ancestor, F (3, 41) = 2.680, P = 0.0594; morA tRNA/tufB, F (6, 92) = 15.20, P < 0.0001; lasR R216Q, F (3, 35) = 8.064, P = 0.0003; ΔPA14_45920.PA14_46440, F (3, 38) = 10.07, P < 0.0001] with Tukey’s multiple-comparison test. Groups labeled with the same letter are not statistically different.

lasR mutations are beneficial in the absence of a competitor.

The fitness advantage of lasR mutants in chronic infections has been proposed to derive from multiple features, including social cheating, growth advantages in amino acids or in microoxic environments, and increased growth at high cell densities (28–31, 50–53). Social cheating refers to the ability of a lasR mutant to reap the benefits of public goods secreted by LasR+ cells within the population, including proteases, pyocyanin, and hydrogen cyanide, without undergoing the cost of producing those factors themselves (31, 54, 55). However, we observed that in the absence of a competitor, lasR mutants demonstrated greater area under growth curves (AUC) than the ancestor (Fig. S4 in the supplemental material) and similar or greater growth yields (Fig. S5 in the supplemental material). This result shows that exploitation of a competitor is unnecessary for increased lasR mutant fitness.

Prior studies of lasR mutants have implicated altered carbon catabolite repression as the source of their growth advantages in aromatic amino acids (30). Yet we found that lasR mutants attained similar fitness advantages in media containing only glucose or only lactate as in only amino acids (Fig. 4C and D). Nonetheless, we tested whether lasR mutations altered carbon catabolite repression by analyzing the nutrient levels in the spent medium of a lasR mutant compared to spent medium of the ancestor. We cultured strains separately in the evolution medium and used HPLC to quantify nutrient levels every hour for the first 8 h of growth. We hypothesized that lasR mutations may alter the rate or order of nutrient consumption when growing in medium containing multiple carbon sources. We noticed lasR mutants produced a spike in OD-normalized amino acid levels at 1 h postinoculation that we cannot explain, but no subsequent significant differences in nutrient consumption rate or order were found (Fig. S6 in the supplemental material). Therefore, we cannot attribute the selective advantage of lasR mutants in this complex medium to altered carbon catabolite repression, but it is possible that some relevant metabolic changes were too subtle to be detected using this approach. The specific metabolic source of the growth advantage of lasR mutations therefore remains unclear in this environment and merits further study.

Ecological interactions between morA and lasR mutants facilitate the maintenance of diversity.

We found that morA and lasR genotypes defined coexisting subpopulations in four out of five planktonic lineages, prompting us to ask if their relative fitness advantages derived from ecological interactions between them. We tested their ability to facilitate each other’s coexistence by mixing genotypes at different starting concentrations (Fig. 5). Each mutant was more fit when introduced at a lower proportion, consistent with a negative frequency-dependent interaction, or a relationship in which the fitness of a competitor increases as it becomes rarer (56, 57). We repeated this experiment using different morA and lasR mutants and found this frequency-dependent interaction was consistent (Fig. S7 in the supplemental material). Therefore, nutrients present in the cystic fibrosis lung environment are sufficient to rapidly drive stable, functional diversification, even in the absence of spatial heterogeneity. This finding is relevant to clinical settings as diversity within infections may increase recalcitrance to treatment (13).

FIG 5.

Relative fitness of lasR and morA mutants is greater when in the minority. We performed direct competitions at the starting concentrations indicated and measured relative fitness at 48 h. Genotypes were differentiated using a lac-marked ancestral strain for the ancestral competitions and by colony morphology for the lasR mutant versus morA mutant competitions. The first competitor listed is genotype 1; the second is genotype 2. Data was analyzed by linear regression, shaded area represents 95% CI of the regression. Ancestor versus ancestor, slope = −0.2060; lasR mutant versus morA mutant, slope = −0.6760; difference between the slopes was statistically significant F = 4.195, DFn = 1, DFd = 83, P = 0.0437.

DISCUSSION

The realization that chronic infections of the CF airway by P. aeruginosa not only involve adaptive evolution, but also a process of diversification, has motivated numerous explanations (19). Most of these understandably focus on aspects of ourselves as hosts. However, one of the most basic and essential components of host-microbe interactions is nutrition. Here we show that both adaptation and diversification of P. aeruginosa occur during propagation in the nutrients found in the cystic fibrosis respiratory environment. Moreover, selected genotypes acquire mutations identical to those recovered from infections, including mutations in cyclic di-GMP regulators (morA, wspF and roeA) and quorum sensing regulators (lasR). Further, the phenotypes caused by these mutations are of potential clinical importance. Loss-of-function lasR mutations have been shown to alter the production of virulence factors and increase resistance to beta-lactam antibiotics (24, 30), and are associated with lung disease progression (58). Similarly, high biofilm phenotypes such as those produced by morA mutants (RSCVs) have been shown to increase persistence under stresses like the immune system and antibiotics, making these infections more difficult to treat (26, 59–61). Selection for either of these phenotypes alone merits concern, but here we show that morA and lasR mutants frequently are coselected in vitro and facilitate one another’s coexistence. Though the mechanism underlying this ecological relationship is yet to be identified, the maintenance of this diversity poses its own risk for population persistence in the face of treatment.

One defining phenotype of many P. aeruginosa infections is increased secreted polysaccharides such as those produced by morA mutant genotypes. Their selection during planktonic serial transfer was surprising because biofilm matrix components ought to be costly to produce. While the mechanism underlying this paradox is unclear, our results suggest that evolved morA mutations were likely selected due to their strong fitness advantage when nutrients are abundant. In recent work, Katharios et al. present a model by which the phosphodiesterase activity of morA that degrades cyclic di-GMP is induced during nutrient limitation to repress costly biofilm production and thus prevent cell death (62). Consequently, disrupted morA signaling results in increased biofilm when nutrients are abundant, but cell death during nutrient limitation. Our findings are consistent with this report: nonsynonymous substitutions in morA increased biofilm and relative fitness but were costly when nutrient levels were reduced. Abundant nutrients within the cystic fibrosis airway may therefore play an important role in selecting high biofilm mutants.

The specific types and locations of evolved mutations in morA and other genes can improve understanding of the selected phenotypes and even how the gene functions. In clinical isolates, both nonsynonymous SNPs and deletion mutations in morA are prevalent (2), but MorA harbors catalytically active DGC and PDE domains, as well as multiple sensor (PAS) domains, rendering the effect of different types and locations of mutations within the gene unclear. Further, most prior studies of effects of morA on motility (48, 63), biofilm (48, 62), and colony morphology (26) in Pseudomonas species used deletion or nonsense mutants. We found that phenotypes of missense mutations in morA largely coincide with these deletion mutants, but with some nuance. For instance, missense mutations in morA decrease swimming and swarming motility, like morA deletions (48), and this is sensible given that morA modulates timing of flagellar biosynthesis and both forms of motility are flagellum dependent (63, 64). However, we identified varied levels of swimming motility among missense morA mutants, suggesting that the location of mutations within this protein alters mutant phenotype. This subtlety mirrors prior findings from our laboratory focused on a distant homolog in Burkholderia cenocepacia (65). Likewise, we saw that missense morA mutations increased biofilm formation in agreement with increased biofilm formation during early growth of morA deletions (62). Certain missense mutations in morA also produced wrinkled colony morphologies, which is consistent with the RCSV phenotype reported for a nonsense mutation (26). Altogether, we find that missense mutations in morA produce similar increases in biofilm and decreases in motility as caused by gene deletions, both consistent with disrupted PDE function, but to varying degrees.

A similar analysis can be applied to the wide spectrum of lasR mutants we observed during propagation in CF nutrients, including large deletions, nonsense mutations, and missense mutations, which is consistent with the range of mutations observed in clinical isolates (8, 17). In fact, we identified several missense mutations at identical residues (P117L and A231) or adjacent residues (C79, A189 and R216) to those detected in clinical samples. Several of these missense variants reduce, but do not eliminate, LasR activity in PAO1 (17). If these mutations confer similar phenotypes in PA14, the selection of lasR mutants in this experiment suggests that both complete loss and reduction of LasR signaling are advantageous in this environment; however, the effect of lasR mutation is known to vary based on genetic background (17). In addition, previous studies indicate that P. aeruginosa may induce the RhlR-RhlI quorum sensing system to overcome defective LasR (17), therefore lasR deletion mutants identified in our experiment are not necessarily completely defective in quorum sensing. But the substantial overlap between genotypes and phenotypes from this study and from prior work on laboratory and clinical mutants indicates their prevalence resulted from shared selective forces. Specifically, altered LasR signaling is advantageous during growth on CF nutrients either with or without biofilm selection, but some mutants may maintain some level of LasR activity or quorum sensing through RhlR.

We investigated the contributions of social cheating and improved growth on amino acids to the fitness of lasR mutants in CF nutrients but were unable to completely attribute lasR mutant fitness to either variable. In the absence of a clear explanation for selection of lasR mutants in this environment, we suggest a few plausible ones. First, lasR mutations de-repress growth and provide resistance to autolysis at high cell densities (52, 53), which is consistent with our finding that lasR mutants produce greater net growth than LasR+ strains. Second, altered LasR signaling could reduce expression of metabolically costly products and select lasR mutants even in the absence of competitors. Third, lasR mutants exhibit advantages in microoxia, which likely occurs during late-phase growth in our cultures (29). These explanations are all consistent with our observations that lasR mutants are advantageous in the absence of competitors or cooperators as well as conditions lacking amino acids. We might reconcile these explanations by noting that growth in CF nutrients is sufficient to select lasR mutants, but their advantages may not be specific to these conditions. Rather, lasR mutants are frequently selected in a broad range of environments including CF (2, 8), models of CF infection (24, 27), and laboratory media (27). Still, the physiological and regulatory mechanisms underlying the fitness advantages of lasR mutants demand further study.

This evolution experiment also reveals at least two curious findings regarding the genetics of adaptation by P. aeruginosa. First, adaptation to CF nutrients is produced by mutated global regulators of multicellular behavior (quorum sensing, biofilm formation) with remarkable consistency. Although many other ways of increasing fitness in these nutrients and propagation methods surely exist, it is notable that the most beneficial mutants in this environment were those which alter massive and complex regulons, rather than downstream effectors of cell behavior (34, 66). Second, we detected several large deletions and structural variants in prevalent genotypes, including those that result in the loss of both lasI and lasR. In clinical isolates, SNPs within lasR are frequently detected, however we hypothesize that large deletions may be underreported due to the computational challenge of identifying them in draft genomes produced by short-read sequencing. One-fourth of mutations in this evolution experiment were indels or structural variants, so sequencing efforts of clinical isolates ought to analyze these possible alterations whenever feasible.

In summary, extensive genetic adaptation and diversification is a well-known feature of P. aeruginosa populations growing in the cystic fibrosis respiratory tract (12) and experimental models of infection but the specific selective causes have been elusive (21, 24, 25, 67). Studying adaptation in vivo is challenging because obtaining an early infecting strain that serves as an appropriate internal reference for subsequent isolates is often not possible. Further, in more established infections, evolved mutations may be too numerous to infer the phenotypic impact of any single change. Thus, linking mutations known to occur in vivo with the conditions that selected them and the phenotypes they confer is critical to understand how P. aeruginosa adapts to the CF respiratory environment. Remarkably, the evolution of genotypes producing more biofilm and altered quorum sensing can be selected by a subset of nutrients found in the cystic fibrosis airway and coexist by reciprocal ecological facilitation. Although the presence of other microbes, host factors, or antibiotic pressure may only add to the potential diversification of opportunists like P. aeruginosa, they may not be required to promote increased persistence in vivo.

MATERIALS AND METHODS

Evolution experiment.

Pseudomonas aeruginosa strain UCBPP-PA14 (68) was propagated for 12 days in minimal medium supplemented with the nutrients shown to be abundant in the cystic fibrosis respiratory environment (21). The evolution experiment was described previously in the work of Scribner et al. (37), in which these lineages were used to test whether mutations in lineages exposed to tobramycin occurred in the absence of antibiotic selection, but were not analyzed further. Briefly, the minimal medium consisted of 11.1 mM glucose, 10 mM dl-lactate (Sigma-Aldrich 72-17-3), 20 ml/liter MEM essential amino acids, 10 ml/liter MEM nonessential amino acids (Thermofisher 11130051, 11140050), an M9 salt base (0.1 mM CaCl₂, 1.0 mM MgSO₄, 42.2 mM Na₂HPO₄, 22 mM KH2PO₄, 21.7 mM NaCl, 18.7 mM NH₄Cl), and 1 ml/liter each of Trace Elements A, B, and C (Corning 99182CL, 99175CL, 99176CL). We propagated cultures in 18 by 150 mm glass tubes containing 5 ml of medium and incubated in a roller drum at 150 rpm at 37°C. Lineages were initiated by resuspending a single ancestral clone in PBS and using this suspension to inoculate each population. Five lineages each were propagated with either planktonic and biofilm selection: planktonic through a dilution of 50 μl into 5 ml of fresh medium every 24 h, and biofilm through transfer of a marked colonized bead to a tube with fresh media and 3 new, oppositely marked beads. Lineages were randomly numbered at the initiation of the experiment. The evolution experiment continued for 12 days with sampling for freezing in 25% glycerol at −80°C on days 6 and 12. Planktonic populations were sampled by freezing 1 ml aliquots and biofilm populations were sampled by sonicating a bead in PBS and freezing a 1 ml aliquot. We performed population WGS by inoculating frozen populations into the evolution medium and incubating for 24 h, then removing an aliquot for DNA extraction for planktonic populations and sonicating a colonized bead and removing an aliquot for biofilm populations. We performed population WGS at day 12 for each lineage. We also isolated clones with colony morphologies and phenotypes distinct from the ancestor and sequenced these clones by WGS. Genotypes of evolved clones are shown in Table S1 in the supplemental material.

Whole genome sequencing and analysis.

We extracted DNA using the DNeasy blood and tissue kit (Qiagen, Hiden, Germany) and prepared the sequencing library as previously described (69, 70) using the Illumina Nextera kit (Illumina Inc., San Diego, CA). We sequenced populations to an average read depth of 86-254x and clones to an average read depth of 10-58x using an Illumina NextSeq500. Sequences were trimmed using the trimmomatic software v0.36 with the following criteria: LEADING:20 TRAILING:20 SLIDINGWINDOW:4:20 MINLEN:70 (71). The breseq software v0.35.0 was used to call variants using the -p polymorphism flag when analyzing population sequences (72). These parameters detect variants at >5% frequency within a population. To minimize false positive variant calls, we implemented breseq parameters requiring that variants be supported by at least three reads on each strand. For our reference genome, we used the P. aeruginosa UCBPP-PA14 109 genome (NC_008463) with annotations from the Pseudomonas Genome DB (73). Mutations that were detected between our ancestral strain and reference genome were removed from subsequent analysis as they were not selected by the experimental conditions but are listed in the supplemental material. Variant calls indicative of poor read mapping were also removed, including variants that occurred at only the ends of reads, only within reads with many other mutations, or only at low total read depth. The breseq software also reports new junctions within the genome which may occur due to mobile element insertions, large deletions, structural rearrangements, and prophage excision and circularization, which we included in our analysis. Unfiltered breseq output and code documenting all analysis steps can be accessed at https://github.com/michellescribner/pa_nutrient. Filtering and plotting was performed in R v4.0.5 using packages tidyverse and ggrepel (74–76).

Analysis of morA mutations detected in laboratory and clinical environments.

To identify instances of residue and domain-level parallelism, we analyzed nonsynonymous mutations in morA detected in this experiment, other laboratory experiments (23, 44), and clinical isolates (2). We determined the locations of morA mutations using the supplementary data of two studies (2, 23) and by analyzing sequences of populations propagated in the absence of antibiotic using breseq for Sanz-García et al. (44).

We also visualized morA mutations detected in other evolution experiments from our laboratory. These mutations were detected in an evolution experiment identical to this study but with the following alterations: populations were cultured in deep well plates with daily transfer through either 1:100 dilution or transfer of a colonized pipette tip into a well containing 5 ml fresh medium. Deep well plates were incubated with shaking at 100 rpm. Three evolved lineages from each treatment were sequenced following 12 days of transfer. Whole population genome sequence data for these evolved populations are available in NCBI BioProject PRJNA692838.

The domain positions within the MorA protein were determined from the Pfam database (77). All nonsynonymous mutations detected in clinical and laboratory studies were visualized on the protein structure of PAO1 MorA published by Phippen et al. in Fig. S1 in the supplemental material (78, 79). Mutations were visualized based on their relative position within the PAO1 MorA protein using UCSF Chimera with color corresponding to domain (80).

Colony morphology.

Morphologies of evolved mutants were visualized on agar plates containing 1% tryptone, 20 μg/L Coomassie brilliant blue, and 40 μg/ml Congo red. Strains were grown overnight in the evolution media and 5 μl was spotted onto plates. Plates were then incubated for 24 h at 37°C followed by 72 h at room temperature before photographing.

Motility and biofilm assays.

Motility and biofilm assays were performed similarly to as previously described with the alterations noted below (64, 81, 82). For swimming motility assays, evolved mutants were inoculated in the evolution medium and incubated overnight. A sterile pipette tip was dipped into cultures and used to stab 0.3% agar supplemented with an M8 base, glucose, Casamino Acids, and MgSO₄. Plates were incubated for 24 h at 37°C and diameter of the resulting growth was measured. Swarming motility assays were performed analogously to swimming assays, except plates contained 0.5% agar and 2.5 μl of culture was spotted onto each plate. Plates were incubated for 24 h at 37°C then photographed. We estimated biofilm production of evolved mutants by crystal violet assay (82). We diluted overnight cultures of evolved mutants 1:100 in fresh minimal medium to a volume of 200 μl in a 96-well plate. After incubation for 24 h at 37°C with shaking for 30 s every 15 min, we gently rinsed plates twice with water. We stained wells with 250 μl of 0.1% crystal violet, incubated for 15 min, rinsed three times with water, then allowed them to dry overnight. Crystal violet was solubilized by adding 250 μl 95% ethanol (EtOH) solution (95% EtOH, 4.95% distilled H₂O, 0.05% Triton X-100) to each well for 15 min. Biofilm formation was then visualized by measuring Abs OD₆₀₀. Datapoints are the average of technical replicates from at least three independent experiments.

Fitness assays.

Evolved mutants were incubated for 24 h in the evolution medium, then inoculated into fresh tubes at the indicated starting proportions with a competitor (25 μl of competitor A and 25 μl of competitor B in 5 ml fresh evolution medium to generate a starting proportion of approximately 0.5). Cultures were serially diluted in PBS and plated onto ½ concentration tryptic soy agar at the starting time point to estimate precise initial proportions. Cultures were then incubated for 24 h, diluted 1:100 into fresh medium, and incubated for another 24 h. At the 48-h time point, cultures were again plated. Fitness was calculated as selection rate as described by Turner et al. (56) where A and B represent competitors A and B and t = 0 and t = 2 denote CFU/ml at time points 0 and 48 h. Strains were differentiated by competing against a Lac⁺ marked ancestor that appears blue on agar containing X-Gal or differentiated by colony morphology in competitions of morA versus lasR mutants.

$$r=\frac{\text{ln}\frac{A_{t=2}}{A_{t=0}}-\text{ln}\frac{B_{t=2}}{B_{t=0}}}{2}$$

Growth curves.

We measured growth curves of evolved mutants by growing cultures overnight in the evolution medium, then diluting 1:100 into 200 μl of fresh medium in a 96-well plate. Cultures were incubated for 24 h at 37°C with shaking for 30 s every 15 min, and OD₆₀₀ was measured at 15-minute intervals. We analyzed growth curves from at least three independent experiments with three replicates each. Growth curves were subsequently analyzed in R using the Growthcurver package (83).

Nutrient analysis.

The ancestral strain and lasR R216Q mutant were grown from freezer stocks in the evolution medium for 24 h at 37°C. Cultures were diluted 1:100 into fresh medium supplemented with all carbon sources (glucose, lactate and amino acids) and incubated at 37°C. We removed 1 ml aliquots from these cultures every hour for 8 h and collected supernatants by centrifugation at 13,000 × g for 1 min which were then filtered using a 0.2 mm filter. Two biological replicates for the ancestor and three for the lasR mutant were analyzed for nutrient content. We measured cell density by Abs OD₆₀₀ for each aliquot prior to centrifugation.

Amino acid levels in the supernatants were analyzed using the Waters AccQ-Tag chemistry package. Samples were hydrolyzed using trichloroacetic acid precipitation. A 1:4 ratio of TCA was added to the samples, which were chilled on ice for 10 min and then centrifuged at 13,000 × g for 10 min. The supernatant from the TCA precipitation was removed and pH balanced to pH 8.2 to 10 using KOH. The sample was derivatized using the AccQ-Tag methodology (Waters) and analyzed on a Waters HPLC system consisting of an e2985 Separations Module and a 2475 FLR Detector as per manufacturer’s instructions. For glucose and lactate consumption, samples were enzymatically analyzed using a d-Lactic Acid/L-Lactic Acid Enzymatic Bioanalysis UV-Test kit and a d-Glucose Enzymatic Bioanalysis UV-Test kit (Roche Diagnostics) with a modified manufacturer’s protocol to reduce total sample size to 300 μl. Results were read at 340 nm on a BioTek Synergy HTX plate reader.

Statistical analysis.

Data was analyzed using GraphPad Prism 9 and using R where noted.

Data availability.

Data and code used for data analysis can be accessed at https://github.com/michellescribner/pa_nutrient. All sequencing reads were deposited in NCBI under BioProject accession numbers PRJNA595915 and PRJNA692838.