Rapid diversification of Pseudomonas aeruginosa incystic fibrosis lung-like conditionsAlana Schicka,1 and Rees Kassena

aDepartment of Biology, University of Ottawa, Ottawa, ON K1N 6N5, Canada

Edited by E. Peter Greenberg, University of Washington, Seattle, WA, and approved September 4, 2018 (received for review December 7, 2017)

Chronic infection of the cystic fibrosis (CF) airway by the oppor-tunistic pathogen Pseudomonas aeruginosa is the leading cause ofmorbidity and mortality for adult CF patients. Prolonged infectionsare accompanied by adaptation of P. aeruginosa to the uniqueconditions of the CF lung environment, as well as marked diversi-fication of the pathogen into phenotypically and genetically dis-tinct strains that can coexist for years within a patient. Little isknown, however, about the causes of this diversification and itsimpact on patient health. Here, we show experimentally that, con-sistent with ecological theory of diversification, the nutritionalconditions of the CF airway can cause rapid and extensive diver-sification of P. aeruginosa. Mucin, the substance responsible forthe increased viscosity associated with the thick mucus layer in theCF airway, had little impact on within-population diversificationbut did promote divergence among populations. Furthermore,in vitro evolution recapitulated traits thought to be hallmarks ofchronic infection, including reduced motility and increased biofilmformation, and the range of phenotypes observed in a collectionof clinical isolates. Our results suggest that nutritional complexityand reduced dispersal can drive evolutionary diversification ofP. aeruginosa independent of other features of the CF lung suchas an active immune system or the presence of competing micro-bial species. We suggest that diversification, by generating exten-sive phenotypic and genetic variation on which selection can act,may be a key first step in the development of chronic infections.

experimental evolution | adaptation | diversity |pathogen evolution | chronic infection

The bacterium Pseudomonas aeruginosa is a globally ubiqui-tous, metabolically flexible, gram-negative opportunistic

pathogen. While it is an important nosocomial pathogen causinga range of acute infections, P. aeruginosa is also commonly re-covered from the airways of adult cystic fibrosis (CF) patients,where it causes chronic endobronchial infections in the majorityof adult patients and is the leading cause of morbidity andmortality in this population (1–3). The majority of chronic CFlung infections are thought to be the result of colonization byP. aeruginosa from environmental sources, although highlytransmissible epidemic strains are responsible for ∼25% ofinfections in Canadian patients (4). Once established, chronicinfections can remain persistently associated with a host fordecades, being virtually impossible to eradicate with standardantibiotic therapy (5).The transition from environmental strain to one that causes

chronic infection is characterized by a few repeatable phenotypicchanges underlain by a much larger suite of genetic changes.Free-living environmental strains are typically motile, virulent,and nonmucoid, whereas strains isolated from chronically in-fected CF patients are often nonmotile, avirulent, mucoid, andhighly antibiotic-resistant (6–9). These phenotypes representparallel adaptations to a range of CF lung-specific stressors, in-cluding increased viscosity, osmotic stress, low oxygen, highconcentrations of antibiotics, and immune system attack. Thisparallelism notwithstanding, the other striking feature commonto P. aeruginosa isolates from CF patients is their diversity.Phenotypes resembling both environmental and chronically

infected strains can persist in the same patient for long periods oftime, and isolates from within (10–14) and among (15, 16) pa-tients can be highly diverse, both phenotypically and genetically.This diversity makes it difficult to identify reliable markers forthe onset of chronic infection (17) and predict clinical outcomeson the basis of single colony isolates alone. Indeed, the lack ofcorrelation between the prevalence of a particular phenotypeand clinical outcomes (13) hints that the diversity of the pop-ulation, rather than the abundance of any particular phenotype,could be what makes CF lung infections by P. aeruginosa so re-calcitrant to treatment (10).Diversity in the CF lung typically evolves rapidly and de novo

following colonization (10, 13, 18), with genetically and pheno-typically diverse clones often persisting for years within the samehost (8, 19, 20). This dynamic has many of the hallmarks of anadaptive radiation, the rapid diversification of a lineage into arange of niche specialist types. Both theory (21–23) and experi-ment (reviewed in refs. 24–27) suggest that adaptive di-versification is most likely to occur when divergent selection,imposed by ecological opportunities in the form of vacant nichespace or underutilized resources, is strong relative to the rate ofdispersal. The extent of phenotypic divergence may be furtherexaggerated, or its rate accelerated, through ecological interac-tions such as resource competition or predation (28–31).We suspect that the complex ecological conditions of the CF

lung likely promote diversification in colonizing P. aeruginosa.The CF airway contains a rich spectrum of resources and nutrientsthat could provide ample ecological opportunity to drive special-ization (13, 32, 33), and the thick mucus layer, a by-product of the

Significance

Chronic infection with the opportunistic pathogen Pseudomo-nas aeruginosa is the leading cause of lung transplant or deathin cystic fibrosis (CF) patients. P. aeruginosa diversifies in the CFlung, although why this happens remains a mystery. Weallowed P. aeruginosa to evolve in the laboratory under arange of conditions approximating the CF lung. The diversity ofevolved populations was highest, and most closely resembledthe range of phenotypes among clinical isolates, in environ-ments resembling the spectrum of nutritional resources avail-able in the CF lung. Our results point to the nutritionalcomplexity of the CF lung as a major driver of diversification,and they suggest that diversity could be important in thedevelopment of chronic infections.

Author contributions: A.S. and R.K. designed research; A.S. performed research; A.S.analyzed data; and A.S. and R.K. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Published under the PNAS license.

Data deposition: All data and R scripts for data analysis have been deposited in the DryadDigital Repository database, datadryad.org/ (doi:10.5061/dryad.5c38fv8).1To whom correspondence should be addressed. Email: alana.schick@uottawa.ca.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1721270115/-/DCSupplemental.

Published online October 1, 2018.

10714–10719 | PNAS | October 16, 2018 | vol. 115 | no. 42 www.pnas.org/cgi/doi/10.1073/pnas.1721270115

Dow

nloa

ded

by g

uest

on

Nov

embe

r 16

, 202

0

impaired ability to transport chloride ions resulting from muta-tions in the CF transmembrane conductance regulator (CFTR)gene, serves to reduce dispersal among subpopulations. Dispersalwill be further reduced due to the highly compartmentalized na-ture of the lung itself, with distinct patches or habitats, such asright and left lobes, upper and lower respiratory tract, and thealveoli and bronchioles, on different branches of the respiratorytree. The resource profile and spatial structure of the CF lung thusprovide ideal conditions for strong divergent selection to drivediversification. Additional aspects of life in the CF lung, such asthe emergence of hypermutator lineages (20, 34), high levels ofantibiotic use (35), and interactions between P. aeruginosa andcoinfecting species (36, 37), bacteriophages (38, 39), and the hostimmune system (40, 41), may further contribute to the geneticand phenotypic variability seen among isolates from chronicallyinfected patients.Here, we evaluate the contributions of resource complexity

and reduced dispersal to P. aeruginosa diversification by trackingthe extent of phenotypic diversification within and among in-dependently evolved populations descended from P. aeruginosastrain Pa14 after ∼220 generations of selection in environmentsthat varied in how closely they resemble the resource profile andviscosity of the CF lung. The most CF-like environment con-sisted of the nutritionally complex synthetic CF medium(SCFM), a defined medium based on the free amino acid, anion,cation, and carbon-source profiles from CF sputum (32), sup-plemented with mucin, the major protein component of mucus,to mimic the thick mucus layer in the lumen of CF patients.Several studies have shown that the presence of mucin in-creases the viscosity of the medium and reduces the motility ofP. aeruginosa (42, 43) and is therefore very likely to reduce dis-persal. The least CF-like environment was composed of a mini-mal medium supplemented with glucose as the sole carbonsource in the absence of mucin. The experimental conditionsconsisted of factorial combinations of two levels of nutrientcomplexity (minimal medium vs. SCFM) and viscosity (with orwithout mucin). Phenotypic divergence among independentlyevolved replicate populations provided an estimate of the con-tribution of spatial compartmentalization to diversification. Wescored both colony morphology and 10 traits thought to be as-sociated with pathoadaptation tied to chronic infection of the CFlung (growth rate in LB, pyocyanin and pyoverdine production,biofilm formation, swim and twitch motility, and resistance tofour classes of antibiotics: ciprofloxacin, ceftazidime, colistin,and tobramycin) for 12 randomly sampled isolates from each of12 independently evolved populations within each treatment (fora total of 576 isolates) as well as a collection of 24 P. aeruginosaisolates from the lungs of CF patients from across Ontario (15)to gauge the extent to which our experimental conditions re-capitulate the markers of phenotypic divergence associated withclinical strains.

Results and DiscussionMarkers of Chronic Infection Evolve in the Most Lung-Like Environments.Adaptation to the CF lung is often accompanied by a suite ofphenotypic changes that include loss of virulence factors, loss ofmotility, and increased biofilm formation. We determined the ex-tent to which similar phenotypic changes evolved in our in vitroexperiment by scoring a range of traits associated with CF lungadaptation and using a linear mixed-effects model with medium andmucin as fixed effects and population as a random effect to examinethe impact of different treatment combinations on trait evolution.Our results are summarized for all traits in Table 1 (and SI Ap-pendix, Table S1) and for four traits commonly associated with CFlung adaptation in Fig. 1.We observed marked trait evolution, including a reduction in

pyoverdine production (the main siderophore and a measure ofiron-scavenging ability; Fig. 1A), swim motility (Fig. 1C), and

twitch motility (Fig. 1D), and an increase in biofilm production(Fig. 1B), in the most CF-like conditions, all changes that mirrorthose seen during adaptation to the CF lung. These phenotypicchanges persist when colonies are subcultured, implying that theyhave a genetic basis. Given that our experiment was conductedin vitro without competing microflora or an active immune sys-tem, these results suggest that selection driven by the combinedeffects of resource complexity and viscosity is sufficient to ex-plain changes in these putatively pathoadaptive traits charac-teristic of chronic P. aeruginosa infections of the CF lung.This interpretation might be questioned on the grounds that

many of these trait changes, especially those involving a decreasein trait value relative to the common ancestor (pyoverdine pro-duction, swim motility, and twitch motility), reflect general ad-aptation to in vitro culture conditions rather than specificadaptation to CF-like environments. This explanation cannot,however, explain the treatment-specific changes in these andother traits (SI Appendix, Table S1). Pyoverdine production, forexample, declined less in the presence of mucin than in its ab-sence, while swim motility showed the reverse trend. Changes intwitch motility, on the other hand, were independent of mucinand decreased more in the nutrient-rich conditions of SCFMthan in minimal medium. Pyocyanin and biofilm production bothincreased in the most CF lung-like conditions (SCFM+mucin),with either no change (pyocyanin) or decreases (biofilm) in theother treatments. The increase in pyocyanin (SI Appendix, Fig.S3), a virulence factor, was unexpected in light of the commonlyobserved loss of virulence factors among chronic CF isolates.Nonetheless, it has been noted that pyocyanin production is in-creased in isolates from early stage infections and decreases asdisease severity progresses (44), suggesting that reduced pyo-cyanin production in isolates from chronic infections evolves as aby-product of prolonged infection or in response to an aspect ofecology that we did not investigate, with immune system attackor competition from other species being the most obvious can-didates. Taken together, such treatment-specific responses toselection suggest that our results represent trait changes specif-ically associated with CF lung-like conditions. We note that whilethis may not suggest that the laboratory environment recapitu-lates the conditions of the human lung, it does demonstrate thatthe selection pressures driving these parallel phenotypic changesare recapitulated in the laboratory.Intriguingly, our results also suggest that antibiotic resistance

can evolve in these populations as a pleiotropic effect of adap-tation, a phenomenon observed in several bacterial species (45).While a lack of evolutionary change in resistance to antibioticssuch as colistin and tobramycin was unsurprising, as no drugswere used in the selection experiment, it was surprising to see a

Table 1. Summary of phenotypic evolution

Trait SCFM MIN SCFM+mucin MIN+mucin

Growth rate — — — —

Pyoverdine ↓ ↓ ↓ ↓Pyocyanin — — ↑ —

Biofilm formation ↓ ↓ ↑ ↓Swim motility ↓ ↓ ↓ ↓Twitch motility ↓ ↓ ↓ ↓MIC ciprofloxacin ↑ — ↑ ↑MIC ceftazidime — — ↑ ↑MIC colistin — — — ↓MIC tobramycin — — — —

Arrows represent significant deviations from the ancestral value in eachtreatment group, Bonferroni-corrected for multiple comparisons. Antibioticresistance was measured by minimum inhibitory concentration (MIC), thelowest concentration of drug at which growth was inhibited. See SI Appen-dix, Table S1 for test statistics and P values.

Schick and Kassen PNAS | October 16, 2018 | vol. 115 | no. 42 | 10715

EVOLU

TION

Dow

nloa

ded

by g

uest

on

Nov

embe

r 16

, 202

0

marked increase in ciprofloxacin resistance across all conditionsexcept the simplest environment [M9 minimal salts mediumsupplemented with 0.7% glucose (MIN)] and an increase inceftazidime resistance in environments containing mucin (Table1). Although resistance is commonly observed following antibi-otic treatment in clinical settings, our results suggest that bothciprofloxacin and ceftazidime resistance can evolve as a pleio-tropic by-product of adaptation to CF lung-like conditions.Spontaneous resistance to antibiotics has been shown to arise viamutations to the MexGHI-OpmD efflux pump that are also as-sociated with biofilm production (46). This mechanism seemsunlikely to explain the increases in resistance in our experimentbecause we did not observe the expected positive relationshipbetween biofilm formation and resistance for either drug class,with the correlation being significantly negative for ciprofloxacinand indistinguishable from zero for ceftazidime (SI Appendix,Fig. S1). Interestingly, a recent study found that resistance toceftazidime arose in the absence of this antibiotic as a result ofan increase in beta-lactamase production (47), a mechanism thatwarrants further exploration.We did not observe any mucoid individuals in our evolved

populations, despite this being a common phenotype of lung-evolved isolates that can often be associated with aggregation andbiofilm formation. If mucoidy represents an adaptation to pro-longed infection, it is possible that we would have observed it in alonger experiment. An alternative explanation is that mucoidyevolves in response to selection for some other factor not capturedin our experiment. The mucoid colony morphology is often linkedto the overproduction of alginate (48), a compound that can alsoeffectively scavenge reactive oxygen species (49, 50). If mucoidy isa pleiotropic result of adaptation for tolerance to oxidative burstfrommacrophages (15) or neutrophils (51), for example, we wouldnot expect to see it evolving in our experiment.

Within-Population Phenotypic Diversity Is Driven by ResourceComplexity. We observed extensive phenotypic diversity in allevolved populations. To capture the general trends associated

with multivariate phenotypic diversification within and amongpopulations, we used all 10 traits described above to calculate theEuclidean distance between pairs of isolates within a population(Eq. 1) and between the means of pairs of populations withineach treatment (Eq. 2) as follows:

For  Isolates, dðx, yÞ=ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiXni=1

ðxi − yiÞ2s

. [1]

For  Populations, dða, bÞ=ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiXni=1

��xa,i −�yb,i

�2s

. [2]

Here, xi and yi are standardized z scores for trait i of isolates xand y, respectively, and xa,i and yb,i are standardized z scores fortrait i of isolates in populations a and b, respectively. Note thatcalculating among-population distance by using the populationmeans controls for any effect of variation within populations onthe measure of among-population distance.Our results, shown in Fig. 2, are striking. Populations evolved

in the complex nutritional environment of SCFM supportedhigher amounts of within-population diversity than those evolvedin the simpler nutritional environment of minimal medium withglucose as the sole carbon and energy source (Fig. 2A; two-factorANOVA, F1,44 = 10.21, P = 0.003), independent of the presenceof mucin (F1,44 = 0.65, P = 0.423). Moreover, we found no sig-nificant interaction between resource complexity and mucin(F1,44 = 0.79, P = 0.786). Our results contradict the prevailingview that the within-host diversification by P. aeruginosa ac-companying chronic infection is driven primarily by restricteddispersal due to the thick mucus layer in the lung lumen. Inevolutionary terms, the nutritional complexity of the lung likelyrepresents abundant ecological opportunity for colonizingP. aeruginosa that generates strong divergent selection, leadingto phenotypic divergence. Competition for resources among in-cipient niche specialists in the evolving population could further

A B

C D

Fig. 1. Phenotypic adaptation of evolved isolates for pyoverdine production (A), biofilm formation (B), swim motility (C), and twitch motility (D). Within eachtreatment, each column/color represents a replicate evolved population (n = 12 populations for each treatment). Solid lines represent treatment means, anddashed lines represent the value of that phenotype in the ancestral strain (Pa14).

10716 | www.pnas.org/cgi/doi/10.1073/pnas.1721270115 Schick and Kassen

Dow

nloa

ded

by g

uest

on

Nov

embe

r 16

, 202

0

accelerate phenotypic diversification and contribute to coexistence,perhaps through negative frequency-dependent selection (reviewedin refs. 27 and 52), leading to long-term persistence of phenotyp-ically divergent P. aeruginosa variants within a single host.

Nutritional Complexity and Spatial Structure Cause Populations toDiverge. P. aeruginosa isolates from distinct, spatially separatedcompartments within the CF airway are often genetically andphenotypically distinct, a result that has been taken to implyniche-specific adaptive diversification following colonization

(53). An alternative interpretation is that spatial separation ofsubpopulations in distinct lung compartments leads to geneticand phenotypic divergence, independent of niche-specific ad-aptation due to stochastic effects in the timing and order ofmutations contributing to adaptation. Our results allow us toquantify the extent of phenotypic divergence among replicate-evolved populations within each treatment in our experiment asa proxy for the extent of among-compartment divergence. Bothnutritional complexity and mucin had statistically significant ef-fects on among-population divergence (randomization testwhere treatment labels were resampled and a null distribution ofF statistics calculated; P values were <0.0001, <0.0001, and0.0048 for medium, mucin, and the interaction, respectively,from a two-factor ANOVA), with the SCFM+mucin treatmentdisplaying the most extensive between-population diversity, fol-lowed by the SCFM and MIN+mucin treatments and the MINtreatment (Fig. 2B). Since replicate populations within a treat-ment evolved, by design, under the same selective pressure, theseresults lend support to the idea that mutation-order effects inthe most CF-like conditions can support substantial among-population divergence of multivariate phenotypes.

Comparison of Evolved Populations and Clinical Isolates from CFPatients Across Ontario. To what extent do the conditions in ourexperiment recapitulate the range of variation associated withP. aeruginosa isolates from CF patients? To answer this question,we measured the same 10 phenotypic traits in each of 24P. aeruginosa strains isolated from the lungs of CF patients fromacross the Canadian province of Ontario, a collection that in-cluded both nonepidemic and epidemic strains (4). We thencalculated the Mahalanobis distance between each clinical strainand the multivariate distribution of all evolved isolates from eachtreatment group, separately, to obtain a measure of phenotypicsimilarity between the clinical strains and the isolates evolvedunder different conditions. MIN-evolved isolates were the mostdifferent (largest Mahalanobis distance) from the clinical strains(Fig. 3A; ANOVA, F3, 92 = 17.9, P < 0.001). The clinical strainswere most similar to those evolved in SCFM, but not statisticallymore similar than those evolved in SCFM+mucin or MIN+mucin, evaluated by post hoc pairwise t tests, Holm-adjusted formultiple comparisons. Interestingly, strains classified as epidemicwere consistently less similar to laboratory-evolved isolates thanthose classified as nonepidemic (two-factor ANOVA, effect ofepidemic/nonepidemic, F1,88 = 6.21, P = 0.015; Fig. 3A). Thisdifference is largely attributable to levels of antibiotic resistance,as excluding antibiotic resistance traits eliminates the differencein Mahalanobis distance between epidemic and nonepidemicstrains from our in vitro isolates (Fig. 3B; two-factor ANOVA,effect of epidemic/nonepidemic, F1,88 = 1.69, P = 0.197). Thisresult further supports those noted by ref. 15 that the key dis-tinguishing feature of epidemic strains is that they are more re-sistant to antibiotics than nonepidemic strains. Importantly,epidemic and nonepidemic strains are indistinguishable inthe other phenotypic dimensions measured here, making itdifficult to distinguish them on the basis of nonantibioticphenotypes alone.Interestingly, the Mahalanobis distance between the clinical

isolates and those from even the most CF-like conditions in ourexperiment was always significantly greater than zero, whether ornot we included antibiotic resistance traits in the analysis (Hotel-ling’s T2 test, P < 10−7 for all comparisons). There are two, notmutually exclusive, interpretations of this result. The first is that apositive Mahalanobis distance reflects the absence in our experi-ment of additional selective forces commonly experienced byP. aeruginosa in the lungs of CF patients such as an active immunesystem or competition from a diverse microbial community. Thesecond stems from the observation that most chronic CF infectionsarise from de novo colonization by genetically unique strains of

A

B

Fig. 2. Euclidean distance within (A) and among (B) populations. (A) Eachpoint represents a population with the distance within that population de-termined by the average of all pairwise comparisons of individuals within thatpopulation. (B) Each point represents the distance between the mean values oftwo populations. Solid lines represent treatment means. All distance calculationsare Euclidean distance based on all 10 z-score-transformed phenotypic traits.

Schick and Kassen PNAS | October 16, 2018 | vol. 115 | no. 42 | 10717

EVOLU

TION

Dow

nloa

ded

by g

uest

on

Nov

embe

r 16

, 202

0

P. aeruginosa. If so, then some portion of the phenotypic diver-gence among isolates from different patients could be associatedwith phylogenetic diversity among colonizing strains. The relativecontribution of phylogenetic vs. selective forces in driving CF lung-associated adaptation and divergence is unclear and constitutes asubject for future investigation.

Variation in Colony Morphology Is also Linked to NutritionalComplexity. Colony morphologies are often used as indicatorsfor progression to chronic infection, so we assayed colonies fromeach evolved population for eight morphological characteristics(pigmentation, opacity, iridescence, surface texture, margin,halo, autolysis, and small colony variant). We identified a total of18 distinct morphotypes across all 120 populations and an av-erage of ∼2 morphotypes per population (range from 1 to 4; SIAppendix, Fig. S2). Notably, and consistent with the resultspresented above, populations evolved in SCFM contained moredistinct colony morphs on average than those evolved in MIN(ANOVA, medium, F1,116 = 7.36, P = 0.008), while the presenceof mucin had no significant effect on the number of colonymorphs (ANOVA, presence of mucin, F1,116 = 0.64, P = 0.427).Moreover, there was little correspondence between colonymorphology and the changes in the suite of putatively pathoa-daptive traits we measured, as revealed by inspection of Spearman-rank correlations for all pairs of traits (SI Appendix, Fig. S1).These results lend further support to the growing consensus thatcolony morphotypes are unreliable markers of adaptation andthe onset of chronic infection.

Clinical Significance and Implications. Our results provide directexperimental evidence that phenotypic diversification of P. aer-uginosa in the airways of CF patients can be driven by divergentselection imposed by the ecological opportunity associated withthe nutritionally complex lung environment. Divergence amongsubpopulations can be further exaggerated by reduced dispersalresulting from the thick mucus layer associated with the CFTRdefect and colonization of different airway compartments. To-gether, these results suggest that the combination of nutritionalcomplexity and reduced dispersal alone is sufficient to driverapid and repeatable diversification, independently of othersources of selection associated with adaptation such as immuneevasion, resource competition from other microbial species, orredox stress.This interpretation must be qualified, of course, by the fact

that our experiments were done in vitro under conditions thatwere a far cry from the more complex and dynamic environmentof the CF lung itself. The advantage of our approach is that itaffords us the opportunity to construct focused, highly replicatedtests of the role of nutrient complexity and mucin in driving

diversification on scales that would not otherwise be possible.The disadvantage, of course, is that our environments can onlyrepresent a crude simulacrum of the conditions actually experi-enced over the course of an infection. That said, it is worthnoting that many of the phenotypes thought to be hallmarks ofchronic infection, especially loss of motility and biofilm forma-tion, were also observed in the more CF-like conditions in ourexperiment, suggesting that we have done a reasonable job ofrecapitulating some of the selective conditions driving adapta-tion to the CF lung. These changes are likely due to loss-of-function mutations, a result commonly observed in the initialstages of adaptation to novel environments in microbial selectionexperiments (27). While the detailed genetic changes underlyingthese phenotypes will have to await whole-genome sequencing(currently underway), the striking phenotypic parallelism be-tween adaptation in vitro and in vivo suggests that the initialstages of colonization of the CF lung can be understood as aspecific instance of the more general phenomenon of adaptationand diversification to a novel, nutrient-rich environment.There are two important implications of our results for clinical

practice. First, our results provide direct evidence that CF lung-like conditions promote rapid phenotypic and colony morphdiversification, a result that is in line with both longitudinal (14,53) and cross-sectional (11, 13) studies at the level of bothphenotype and genotype. Such rapid diversification, togetherwith the observation that a number of hallmark phenotypesevolved rapidly in vitro, imply that colony morphology or otherphenotypic biomarkers are not reliable diagnostic traits of thetransition to chronic infection. Second, rapid and extensive di-versification both within and among independently evolvedpopulations suggests that the transition to chronic infection canoccur by many different phenotypic and genetic routes. Thepresence of genetically distinct subpopulations within the lung oramong different patients complicates treatment because it meansthat no single therapy targeting P. aeruginosa is likely to be ef-fective at clearing or managing infection for all patients. Rather,a more tailored, patient-specific approach that focuses therapyon the phenotypic and genomic properties of the strain infectinga given host may prove more useful.The long-term fate of diversity and its consequences for pa-

tient health remain unclear. Longitudinal studies suggest thatP. aeruginosa diversification can occur rapidly following colonizationand can persist for decades within a single host (53), likely due tospecialization of subpopulations to different conditions of growth indistinct lung compartments. The comparatively short duration of ourexperiment does not allow us to distinguish whether the diversitywe observed was transient, being the product of high mutationsupply rates generating competition among independently arising

A B

Fig. 3. Comparison of clinical isolates to laboratory-evolved isolates. Each point represents the multivariate (Mahalanobis) distance between a clinical strain andthe distribution of all individual isolates from replicate populations in each treatment. Mahalanobis distances are calculated for all traits (A) and excluding antibioticresistance (AR) traits (B). Black circles are nonepidemic clinical isolates, and gray circles are epidemic isolates. Solid lines represent treatment means.

10718 | www.pnas.org/cgi/doi/10.1073/pnas.1721270115 Schick and Kassen

Dow

nloa

ded

by g

uest

on

Nov

embe

r 16

, 202

0

genotypes, or stable, being supported by negative frequency-dependent selection linked to resource specialization. Neverthe-less, our results do lend support to the idea that the nutritionalconditions of the CF lung provide a substrate that spurs straindiversification and supports higher levels of genetic variation thanwould be otherwise available in a more nutritionally homogenousenvironment. It may be that this diversity provides the raw ma-terial for colonization of different airway compartments, leadingto compartment-based specialization. Under this view, the tran-sition to chronic infection is intimately tied to, and is in fact theresult of, diversification.

Materials and MethodsSources of strains and growth conditions are presented in SI Appendix.Following isolation of individual colonies, 30 replicate populations in each of

four selection environments (SCFM, MIN, SCFM+mucin, and MIN+mucin;conditions are described in SI Appendix) were propagated by daily serialtransfer for ∼220 generations. Following this period of selection, morpho-logical diversity was quantified by classifying a minimum of 100 coloniesfrom each population based on eight characteristics: pigmentation, opacity,iridescence, surface texture, margin, halo, autolysis, and small colony variant(details are in SI Appendix). To quantify diversity, 10 additional phenotypeswere measured for each of 12 random colonies from 12 random populationsof each treatment, for a total of 576 isolates. Phenotyping included growthrate, swimming and twitching motility, pyocyanin and pyoverdine pro-duction, biofilm formation, and resistance to one of four classes of antibi-otic: ciprofloxacin (fluoroquinolone), ceftazidime (beta-lactam), tobramycin(aminoglycoside), and colistin (polymyxin). Methods for assaying all pheno-types above are described in SI Appendix. Statistics for individual traits,quantifying diversity, Mahalanobis distances, and correlations betweentraits are also described in SI Appendix.

1. Rajan S, Saiman L (2002) Pulmonary infections in patients with cystic fibrosis. SeminRespir Infect 17:47–56.

2. Schaedel C, et al. (2002) Predictors of deterioration of lung function in cystic fibrosis.Pediatr Pulmonol 33:483–491.

3. Hoiby N, Pressler T (2006) Emerging pathogens in cystic fibrosis. Eur Respir Monogr 35:66–78.

4. Aaron SD, et al. (2010) Infection with transmissible strains of Pseudomonas aerugi-nosa and clinical outcomes in adults with cystic fibrosis. JAMA 304:2145–2153.

5. Gibson RL, Burns JL, Ramsey BW (2003) Pathophysiology and management of pul-monary infections in cystic fibrosis. Am J Respir Crit Care Med 168:918–951.

6. Poole K (2005) Efflux-mediated antimicrobial resistance. J Antimicrob Chemother 56:20–51.

7. Smith EE, et al. (2006) Genetic adaptation by Pseudomonas aeruginosa to the airwaysof cystic fibrosis patients. Proc Natl Acad Sci USA 103:8487–8492.

8. Mowat E, et al. (2011) Pseudomonas aeruginosa population diversity and turnover incystic fibrosis chronic infections. Am J Respir Crit Care Med 183:1674–1679.

9. Sousa AM, Pereira MO (2014) Pseudomonas aeruginosa diversification during in-fection development in cystic fibrosis lungs—A review. Pathogens 3:680–703.

10. Jorth P, et al. (2015) Regional isolation drives bacterial diversification within cysticfibrosis lungs. Cell Host Microbe 18:307–319.

11. Fothergill JL, Mowat E, Ledson MJ, Walshaw MJ, Winstanley C (2010) Fluctuations inphenotypes and genotypes within populations of Pseudomonas aeruginosa in thecystic fibrosis lung during pulmonary exacerbations. J Med Microbiol 59:472–481.

12. Ashish A, et al. (2013) Extensive diversification is a common feature of Pseudomonasaeruginosa populations during respiratory infections in cystic fibrosis. J Cyst Fibros 12:790–793.

13. Workentine ML, et al. (2013) Phenotypic heterogeneity of Pseudomonas aeruginosapopulations in a cystic fibrosis patient. PLoS One 8:e60225.

14. Clark ST, et al. (2015) Phenotypic diversity within a Pseudomonas aeruginosa pop-ulation infecting an adult with cystic fibrosis. Sci Rep 5:10932.

15. Dettman JR, Rodrigue N, Aaron SD, Kassen R (2013) Evolutionary genomics of epi-demic and nonepidemic strains of Pseudomonas aeruginosa. Proc Natl Acad Sci USA110:21065–21070.

16. Williams D, et al. (2015) Divergent, coexisting Pseudomonas aeruginosa lineages inchronic cystic fibrosis lung infections. Am J Respir Crit Care Med 191:775–785.

17. Winstanley C, O’Brien S, Brockhurst MA (2016) Pseudomonas aeruginosa evolutionaryadaptation and diversification in cystic fibrosis chronic lung infections. TrendsMicrobiol 24:327–337.

18. Folkesson A, et al. (2012) Adaptation of Pseudomonas aeruginosa to the cystic fibrosisairway: An evolutionary perspective. Nat Rev Microbiol 10:841–851.

19. Ciofu O, Mandsberg LF, Wang H, Høiby N (2012) Phenotypes selected during chroniclung infection in cystic fibrosis patients: Implications for the treatment of Pseudo-monas aeruginosa biofilm infections. FEMS Immunol Med Microbiol 65:215–225.

20. Feliziani S, et al. (2014) Coexistence and within-host evolution of diversified lineagesof hypermutable Pseudomonas aeruginosa in long-term cystic fibrosis infections. PLoSGenet 10:e1004651.

21. Lewontin RC (1974) The Genetic Basis of Evolutionary Change (Columbia Univ Press,New York), Vol 560.

22. Whittaker RH, Levin SA (1975) Niche: Theory and Application (Halsted Press, NewYork), Vol 3.

23. Nevo E (1978) Genetic variation in natural populations: Patterns and theory. TheorPopul Biol 13:121–177.

24. Futuyma DJ, Moreno G (1988) The evolution of ecological specialization. Annu RevEcol Syst 19:207–233.

25. Schluter D (2000) The Ecology of Adaptive Radiation (Oxford Univ Press, Oxford).26. Kassen R (2009) Toward a general theory of adaptive radiation: Insights from mi-

crobial experimental evolution. Ann N Y Acad Sci 1168:3–22.27. Kassen R (2014) Experimental Evolution and the Nature of Biodiversity (Roberts,

Greenwood Village, CO).28. Felsenstein J (1981) Skepticism towards Santa Rosalia, or why are there so few kinds

of animals? Evolution 35:124–138.

29. Rozen DE, Lenski RE (2000) Long-term experimental evolution in Escherichia coli. VIII.Dynamics of a balanced polymorphism. Am Nat 155:24–35.

30. Doebeli M, Dieckmann U (2000) Evolutionary branching and sympatric speciationcaused by different types of ecological interactions. Am Nat 156:S77–S101.

31. Friesen ML, Saxer G, Travisano M, Doebeli M (2004) Experimental evidence for sym-patric ecological diversification due to frequency-dependent competition in Escher-ichia coli. Evolution 58:245–260.

32. Palmer KL, Aye LM, Whiteley M (2007) Nutritional cues control Pseudomonas aeru-ginosa multicellular behavior in cystic fibrosis sputum. J Bacteriol 189:8079–8087.

33. Marvig RL, et al. (2014) Within-host evolution of Pseudomonas aeruginosa revealsadaptation toward iron acquisition from hemoglobin. mBio 5:e00966-14.

34. Oliver A, Cantón R, Campo P, Baquero F, Blázquez J (2000) High frequency of hy-permutable Pseudomonas aeruginosa in cystic fibrosis lung infection. Science 288:1251–1254.

35. Wright EA, Fothergill JL, Paterson S, Brockhurst MA, Winstanley C (2013) Sub-inhibitory concentrations of some antibiotics can drive diversification of Pseudomo-nas aeruginosa populations in artificial sputum medium. BMC Microbiol 13:170.

36. Sibley CD, et al. (2008) Discerning the complexity of community interactions using aDrosophila model of polymicrobial infections. PLoS Pathog 4:e1000184.

37. Korgaonkar A, Trivedi U, Rumbaugh KP, Whiteley M (2013) Community surveillanceenhances Pseudomonas aeruginosa virulence during polymicrobial infection. ProcNatl Acad Sci USA 110:1059–1064.

38. Brockhurst MA, Buckling A, Rainey PB (2005) The effect of a bacteriophage on di-versification of the opportunistic bacterial pathogen, Pseudomonas aeruginosa. ProcBiol Sci 272:1385–1391.

39. James CE, et al. (2015) Lytic activity by temperate phages of Pseudomonas aeruginosain long-term cystic fibrosis chronic lung infections. ISME J 9:1391–1398.

40. Jensen PØ, Givskov M, Bjarnsholt T, Moser C (2010) The immune system vs. Pseudo-monas aeruginosa biofilms. FEMS Immunol Med Microbiol 59:292–305.

41. McKeon DJ, et al. (2010) Cystic fibrosis neutrophils have normal intrinsic reactiveoxygen species generation. Eur Respir J 35:1264–1272.

42. Wang J, Lory S, Ramphal R, Jin S (1996) Isolation and characterization of Pseudo-monas aeruginosa genes inducible by respiratory mucus derived from cystic fibrosispatients. Mol Microbiol 22:1005–1012.

43. Landry RM, An D, Hupp JT, Singh PK, Parsek MR (2006) Mucin-Pseudomonas aeru-ginosa interactions promote biofilm formation and antibiotic resistance. MolMicrobiol 59:142–151.

44. Fung C, et al. (2010) Gene expression of Pseudomonas aeruginosa in a mucin-containing synthetic growth medium mimicking cystic fibrosis lung sputum. J MedMicrobiol 59:1089–1100.

45. Hershberg R (2017) Antibiotic-independent adaptive effects of antibiotic resistancemutations. Trends Genet 33:521–528.

46. Sakhtah H, et al. (2016) The Pseudomonas aeruginosa efflux pump MexGHI-OpmDtransports a natural phenazine that controls gene expression and biofilm develop-ment. Proc Natl Acad Sci USA 113:E3538–E3547.

47. Davies EV, James CE, Brockhurst MA, Winstanley C (2017) Evolutionary diversificationof Pseudomonas aeruginosa in an artificial sputum model. BMC Microbiol 17:3.

48. Marvig RL, Sommer LM, Molin S, Johansen HK (2015) Convergent evolution and ad-aptation of Pseudomonas aeruginosa within patients with cystic fibrosis. Nat Genet47:57–64.

49. Simpson JA, Smith SE, Dean RT (1989) Scavenging by alginate of free radicals releasedby macrophages. Free Radic Biol Med 6:347–353.

50. Hassett DJ, et al. (2009) Pseudomonas aeruginosa hypoxic or anaerobic biofilm in-fections within cystic fibrosis airways. Trends Microbiol 17:130–138.

51. Makam M, et al. (2009) Activation of critical, host-induced, metabolic and stresspathways marks neutrophil entry into cystic fibrosis lungs. Proc Natl Acad Sci USA 106:5779–5783.

52. Rainey PB, Cooper TF (2004) Evolution of bacterial diversity and the origins of mod-ularity. Res Microbiol 155:370–375.

53. Markussen T, et al. (2014) Environmental heterogeneity drives within-host di-versification and evolution of Pseudomonas aeruginosa. mBio 5:e01592-14.

Schick and Kassen PNAS | October 16, 2018 | vol. 115 | no. 42 | 10719

EVOLU

TION

Dow

nloa

ded

by g

uest

on

Nov

embe

r 16

, 202

0