18 original research Evolution,Medicine,andPublicHealth[2014]pp.18–29 article doi:10.1093/emph/eou003 Gallium-mediated siderophore quenching as an evolutionarily robust D o w n lo a d antibacterial treatment e d fro m h ttp Adin Ross-Gillespie*1,y,Michael Weigert1,2,y,Sam P.Brown3 andRolf K¨ummerli1,2 s://a c a d e 1InstituteofPlantBiology,UniversityofZu¨rich,Winterthurerstrasse190,8057Zu¨rich,Switzerland;2SwissFederal m InstituteofAquaticScienceandTechnology(Eawag),EnvironmentalMicrobiology,U¨berlandstrasse133,8600 ic.o D¨ubendorf,Switzerland;3InstituteofEvolutionaryBiologyandCentreforImmunity,InfectionandEvolution,Universityof up .c Edinburgh,WestMainsRoad,AshworthLaboratories,EdinburghEH93JT,UK o m *Correspondence address. Institute of Plant Biology, University of Zu¨rich, Winterthurerstrasse 190, 8057 Zu¨rich, /e m Switzerland.Tel:þ41446352905;Fax:þ41446348204;E-mail:[email protected] p yTheseauthorscontributedequallytothiswork. h/a Received19December2013;revisedversionaccepted24January2014 rtic le -a b s ABSTRACT trac Backgroundandobjectives:Conventionalantibioticsselectstronglyforresistanceandareconsequently t/2 0 1 losing efficacy worldwide. Extracellular quenchingofshared virulence factors couldrepresent amore 4 /1 promising strategy because (i) it reduces the available routes to resistance (as extracellular action /1 8 precludes any mutations blocking a drug’s entry into cells or hastening its exit) and (ii) it weakens /1 8 selection for resistance, as fitness benefits to emergent mutants are diluted across all cells in a 43 5 cooperativecollective.Here,wetestedthishypothesisempirically. 8 3 Methodology:Weusedgalliumtoquenchtheiron-scavengingsiderophoressecretedandsharedamong b y pathogenicPseudomonasaeruginosabacteria,andquantitativelymonitoreditseffectsongrowthinvitro. gu e Weassayedvirulenceinacuteinfectionsofcaterpillarhosts(Galleriamellonella),andtrackedresistance st o emergenceovertimeusingexperimentalevolution. n 0 Results: Gallium strongly inhibited bacterial growth in vitro, primarily via its siderophore quenching 5 A activity. Moreover, bacterial siderophore production peaked at intermediate gallium concentrations, p indicating additional metabolic costs in this range. In vivo, gallium attenuated virulence and ril 2 0 1 growth—evenmoresothanininfectionswithsiderophore-deficientstrains.Crucially,whileresistance 9 soonevolvedagainstconventionalantibiotictreatments,galliumtreatmentsretainedtheirefficacyover time. Conclusions:Extracellularquenchingofbacterialpublicgoodscouldofferaneffectiveandevolutionarily robustcontrolstrategy. KEYWORDS: antivirulence therapy; public good quenching; resistance; experimental evolution; Pseudomonas (cid:2)TheAuthor(s)2014.PublishedbyOxfordUniversityPressonbehalfoftheFoundationforEvolution,Medicine,andPublicHealth.ThisisanOpen AccessarticledistributedunderthetermsoftheCreativeCommonsAttributionLicense(http://creativecommons.org/licenses/by/3.0/),whichpermits unrestrictedreuse,distribution,andreproductioninanymedium,providedtheoriginalworkisproperlycited. Quenchingpublicgoodsasarobustantibacterialtreatment Ross-Gillespieetal. | 19 INTRODUCTION Likeall organisms,pathogens acquire geneticmu- frequentlyinvolveintracellularaction,againstwhich tations,and,intime,even‘pure’cultureswillinevit- many potential resistance-conferring adaptations ablycometoharbormutantlineages.Suchgenetic couldarise(e.g.modifiedmembranepropertiesto variability can make some pathogen variants less blockadrug’sentryintoacell,orupregulatedefflux sensitive to therapeutic interventions than others, pumpstohastenitsexit[15]).Second,QSregulates andunderstrongorsustainedtherapy,theseresist- notonlyPGsbutalsocertainessentialprivategoods antvariantswillhaveaselectiveadvantageandwill [16],givingQQresistantssubstantialpersonalbene- come to predominate over more susceptible vari- fits over susceptibles—and therefore a means to ants.Consequently,thetherapywillloseefficacy[1, spread. For maximal evolutionary robustness, we 2].Toavoidthissituation,wecantrytopreventre- needtherapieswhereresistancemutationsareun- D o sistantvariantsfromarisingand/orfromspreading likelytoariseinthefirstplace(e.g.extracellularac- w n [3].Topreventresistancearising,wecouldattempt tionrestrictspotentialroutestoresistance)andare loa d toreducemutationsupply,throughlimitingeffect- alsounlikelytospread,becausefitnessdifferences ed ive population size or by employing interventions between resistant and susceptible pathogens are fro m with specialized modes of action where relatively minimized.Thelattershouldbethecasewhencol- h ttp few ‘routes to resistance’ are possible. To prevent lective traits are targeted, because fitness conse- s spread,meanwhile,wemustaimtominimizefitness quences are shared across many individuals. Of ://ac a differences across individual pathogens. Killing course,theextentandevennessofthissharingwill de m every individual, the conventional antibiotic strat- depend on the relatedness and spatial structure ic .o egy,couldcertainlyquashfitnessevenly,butthisis of the pathogen population and the diffusive u p difficultinpracticeandwheneverincompletegives properties of the environment, and these factors .co m resistantpathogensastrongrelativefitnessadvan- would also need to be considered during therapy /e m tage.‘Antivirulence’treatments,meanwhile,osten- design[3]. p h siblydisarmbutdonotharmpathogens,suchthat Inthisstudy,weinvestigate—inatestcase—the /artic resistantvariantsshouldbenefitlittlerelativetosus- hypothesis that extracellular PG quenching is an le -a ceptibles[4].However,traitsthataffectvirulencebut effectiveandevolutionarilyrobuststrategyforpatho- b s not fitness are rare, and the label ‘antivirulence’ is gencontrol.ThePGtraitwetargetissiderophores, tra c usedliberally,evenforinterventionsthatyieldsub- importantexoproductswhoseregulationisnotlinked t/2 0 1 stantialfitnessdifferencesamongpathogens[4].A toanyexclusivelyprivategoods.Siderophoresaredif- 4 /1 final way to minimize fitness differences is to tar- fusible molecules with a high affinity for ferric iron /1 8 get pathogens’ collective traits, where costs and (Fe3þ)andaresecretedbymostbacteriatoscavenge /1 8 4 benefitsarewidelyshared.Forinstance,manyviru- thisimportantbutgenerallybio-unavailableformof 3 5 8 lence-related bacterial exoproducts are also public ironfromtheirenvironmentor,inthecaseofpatho- 3 b goods(PGs)[5].UnderPG-quenchingtherapy,any gens, from their host’s own iron-chelating com- y g mutations allowing PGs to build up again should pounds [17]. Once loaded with Fe3þ, siderophores ue s benefit both resistant and susceptible individuals aretakenupbyproducercells—orothernearbyindi- t o n alike, which would hinder the spread of resistance viduals equipped with appropriate receptors— 05 A [1,6–8]. stripped oftheiriron, and secreted once again into p To illustrate why this matters, let’s consider a theenvironment[18].Althoughtheirprimaryfunction ril 2 0 1 specific example. Quorum quenching (QQ), which maybetoscavengeiron,siderophoresalsobind,with 9 disrupts the cell-to-cell communication [quorum varyingsuccess,severalothermetals[19,20].Among sensing(QS)][9]underlyingawiderangeofcollect- these,galliumistheclosestmimicofiron.Ga3þand ively expressed virulence traits, is a PG-targeting Fe3þ ions have very similar ionic radii and binding ‘antivirulence’ therapy regarded as a promising propensitiesbut,crucially,whileFe3þreducesreadily, alternativetoconventionalbacteriocidalorbacterio- Ga3þ does not [19]. Ga3þ therefore cannot replace static treatments [10, 11]. However, early enthusi- ironasaco-factorinredox-dependentenzymes.We asmforQQhasbeentemperedrecentlybyreports investigatedtheiron-mimickingeffectsofgalliumon that bacteria can quite readily evolve resistance to pyoverdine,theprimarysiderophoreofPseudomonas suchtreatments[12–14].Setagainstourframework, aeruginosa[21],awidespreadopportunisticpathogen this is unsurprising: first, QQ interventions withabroadhostrangeand,inhumans,thecauseof 20 | Ross-Gillespieetal. Evolution,Medicine,andPublicHealth notoriously persistent infections in immune- supplemented with Ga(NO ) (such that final Ga 3 3 compromisedtissues,cysticfibrosislungsandinas- concentrations ranged from 0 to 200mM), as sociation with implanted devices [22]. Pyoverdine, wellascomplementary amounts ofNaNO tobal- 3 whichplaysanimportantroleinsuchinfections[23, ance nitrate levels across treatments, and 20mM 24],bindsgalliumatleastasreadilyasiron,andgal- FeCl where iron-replete conditions were required 3 lium-bound pyoverdine is of no use to iron-starved (Fig.1A).Growthassayswereperformedwith200ml cells [19, 20]. Thus, even without entering the cell, culturesin96-wellplates,forwhichopticaldensity galliumcanreduceP.aeruginosagrowthandbiofilm (OD)was trackedover 24hat 37(cid:3)Cusing a Tecan formation by quenching local stocks of secreted Infinite M-200 plate reader (Tecan Group Ltd., pyoverdineandchokingoffironsupply[19,25]. Switzerland), with 15min read intervals preceded Below, we report our investigations into (i) ateachreadby10sofagitation.Toassaypyoverdine D gallium’s in vitro interference with siderophore- production,wefirstgrewPAO1pvdA-gfpin2mlCAA o w mediated iron uptake and consequent effects on staticin24-wellplatesina37(cid:3)Cincubatorfor24h, n lo a bacterialgrowth,(ii)gallium’sinvivoeffectsonviru- thencentrifugedtheculturesat7000rpmfor2min d e lence and in-host bacterial growth and (iii) the topelletthecells.Fromeachculture,200mlofsuper- d fro potential for bacteria to evolve resistance against natantand,separately,thecellfractionresuspended m h galliumtreatment. in200ml0.8%saline,weretransferredtoanew96- ttp s wellplateandassayedforODat600nmandfluor- ://a METHODOLOGY escence(GFPincellfraction:exjem¼488j520nm; cad pyoverdine in supernatant: 400j460nm) [28]. Both em Strains and media fluorescence measures were standardized by OD ic.o u at 600nm. In a series of side experiments, we p .c Pseudomonas aeruginosa strains featured in our investigated potential biases associated with the om experiments included the wild-type strain PAO1 use of optical measures as proxies for pyoverdine /em p (ATCC 15692), the siderophore knock-out mutants production (Supplementary Fig. S1). Data pre- h /a PAO1(cid:2)pvdDandPAO1(cid:2)pvdD(cid:2)pchEF[26],provided sentedinFig.1Barecorrectedforthesebiases. rtic byP.Cornelis,FreeUniversityofBrussels,Belgium,as le -a wellasversionsoftheabovestrainsconstitutivelyex- bs pressing GFP (PAO1-gfp, PAO1(cid:2)pvdD-gfp, chromo- Experimental infections trac somal insertion: attTn7::ptac-gfp), and a version of Infection assays were performed with final instar t/20 1 PAO1withapvdA-gfpreporterfusion(PAO1pvdA-gfp, Galleria mellonella larvae, purchased from a local 4/1 chromosomalinsertion:attB::pvdA-gfp)[27],provided supplier,standardizedformassandgeneralcondi- /18 byP.K.Singh,UniversityofWashington,USA.Wealso tion and stored at 4(cid:3)C until use (within 3days). /18 4 used the Rhl-quorum-sensing deficient mutant A Hamilton precision syringe was used to deliver 35 8 PAO1(cid:2)rhlR, provided by S. P. Diggle, University of 10ml inocula via a sterile 26s gauge needle 3 b y Nottingham, UK. For overnight culturing, we used introducedsub-dermallytoasurface-sterilizedarea g u Luria Bertani (LB) medium, while for experimental between the last pair of prolegs. Inoculations con- es assays we used CAA medium, supplemented with tainedGa(NO3)3dilutedtodifferentconcentrations t on 0 FeCl whereindicatedtomanipulateironavailability. in0.8%saline,withcomplementaryconcentrations 5 3 A LB was obtained pre-mixed from Sigma-Aldrich, ofNaNO ,and,wherespecified,bacteriafromover- p Switzerland. Our standard CAA medium contained nightLBc3ultures(37(cid:3)C,180rpm),standardizedfor ril 2 0 5gl(cid:2)1 casamino acids, 1.18gl(cid:2)1 K HPO *3H O, celldensityanddilutedsuchthateach10mlinocula 19 2 4 2 0.25gl(cid:2)1 MgSO *7H O, 100mgml(cid:2)1 human-apo contained (cid:4)25CFU (post hoc counts of 12 inocula 4 2 transferrin,20mMNaHCO and25mMHEPESbuffer platedouttoLBagargave95%CIof19.41–31.76). 3 (allfromSigma-Aldrich). Specifically,wetestedthefollowingGa(NO ) con- 3 3 centrations: 2.5, 10 and 50mM (‘LOW’; pooled to- gethersincetheirresultingvirulencecurveswerenot In vitro assays of growth and pyoverdine significantly different from one another), 500mM production (‘MED’)and2500mM(‘HIGH’).Our‘Galliumonly’ OvernightLBcultures(37(cid:3)C,180rpm),washedand treatment comprised various concentrations be- standardized for cell density, were diluted to 10(cid:2)4 tween 2.5 and 2500mM, which again we pooled thenusedtoseedreplicateculturesinCAAmedium for statistical analyses because of similar effects Quenchingpublicgoodsasarobustantibacterialtreatment Ross-Gillespieetal. | 21 D o w n lo a d e d Figure 1. GalliumaffectsP.aeruginosa’sinvitrogrowthandsiderophoreproduction.(A)Galliumsuppressesgrowthparticularlywhenpyoverdineispresent,as fro m shownherebycomparingconditionswithandwithoutitsproduction.Symbolsandbarsindicatemeansand95%CIsofintegralsofsplinecurvesfittedthrough h 24hgrowthtrajectories(ODat600nm)of12replicatecultures.(B)Pyoverdine,assayedusingcomplementaryapproaches,isineachcaseupregulatedat ttp s intermediategalliumconcentrations.SymbolsanderrorbarsrepresentmeansandSEsoffivereplicates.Measuresofpyoverdinefromsupernatant(filledcircles) ://a c orpvdAexpressionfromcellfractions(opencircles)areineachcasescaledbycelldensity(ODat600nm). a d e m ic .o u p onsurvival.Post-injection,larvaewereplacedindi- infection,weestimatethatinoculagalliumconcen- .c o m viduallyinrandomlyallocatedwellsof24-wellplates trations of 2.5–2500mM would translate to in-host /e and incubated at 37(cid:3)C. Survival was monitored galliumconcentrationsofroughly(cid:4)0.05to(cid:4)50mM. mp h hourly between 10 and 24h, and larvae were con- /a sidereddeadoncetheynolongerrespondedtotact- rticle ilestimulation.Anylarvaethatbegantopupatewhile Experimental evolution -ab s underobservationordiedwithinthefirst10hpost- We compared the growth inhibitory effects of gal- tra c injection(i.e.asaresultofhandling)wereexcluded liumversustheaminoglycoside,gentamicin(Gm), t/2 0 fromanalyses(n¼23,3.6%).Toassayinvivobac- andthefluoroquinolone,ciprofloxacin(Cp)—twoof 14 /1 terialgrowth,wepreparedourinoculawithstrains several antibiotics recommended for clinical use /1 8 engineered to constitutively express GFP (see against P. aeruginosa [29]. Concentrations were /1 8 4 above),havingpreviouslyestablishedthatconstitu- calibrated such that they reduced growth integrals 3 5 8 tivelyexpressedGFPsignalcouldprovideareliable overtheinitial24hto(cid:5)1/3thatofuntreatedPAO1 3 b correlate of bacterial density under the conditions WT cultures under the same growth conditions. y g u ofthisinfectionmodel(SupplementaryFig.S2).In For each of 12days, a 96-well plate was prepared, e s eachofsixseparateexperimentalblocks,andateach comprisingreplicate198mlvolumesofiron-limited t o n offourdiscrete timepoints,3–4randomly selected CAA medium supplemented, according to a 05 A alanrdvaempaenruatrlelyatpmoewndtewriezreed.flaPsohw-fdroezreedn ilnarvliaqluihdoNm2- roarnadnoemquizievadlelanytovuotlsucmheemofes,awliintheg(saelleiukmey,iannFtiibgi.o3tifcosr pril 2 0 ogenateswereresuspendedin1mlsterileH2O,vig- detailsoftreatmentsusedandtheirrespectivesam- 19 orouslyshakenandthencentrifugedat7000rpmfor plesizes).Day1cultureswereinitiatedwith2mlali- 2min, whereafter the sample segregated into dis- quotsofa10(cid:2)3dilutedovernightLBcultureofPAO1 crete phases. About 200ml of the water-soluble WT (37(cid:3)C, 180rpm), while for subsequent days, liquid phase was extracted and assayed for GFP- freshplateswereinoculatedwith10mlofundiluted fluorescentsignalrelativetocontrolreplicates(sa- culturefromthecorrespondingwellsoftheprevious line-injected larvae), using a Tecan Infinite M-200 day’s plate, directly after it completed its growth plate reader. Given total larval volumes of (cid:4)1ml, cycle.Plateswereincubatedat37(cid:3)C,andcelldensity andassumingthat(cid:4)20%ofthisvolumemightbe and pyoverdine fluorescence measures were re- hemolymphaccessibletoparticlesdiffusingfroma cordedat15minintervals(with10sinitialshaking) single injection site during the course of an acute usingaTecanInfiniteM-200platereader. 22 | Ross-Gillespieetal. Evolution,Medicine,andPublicHealth Endpoint phenotypic assays growth (slope±SE: (cid:2)0.067±0.019, 95% CI for drop¼[2.91–15.86%]; difference in slopes Priorobservations[30]andourownreasoning(see 0.368±0.022, F ¼276.41, P<0.001)—particu- Table 1) suggested that pyoverdine and pyocyanin 1,140 larly over the range of concentrations up to and could both affect the costs and benefits of iron including 50mM, which correspond to the concen- uptake under gallium treatment. Anticipating that trationslikelyexperiencedinourinvivoexperiments the experimental evolution described above might (seebelow). haveinducedchangesinthesetraits,weperformed It has been suggested that as the benefit of phenotypicassaystocompareculturesofourances- pyoverdineproductiondrops,bacteriashouldgrad- tral PAO1 WT, its descendent lines experimentally ually scale back their investment in this trait [19]. evolved in CAA with or without supplementa- On the other hand, it has also been shown that tion with 20mM Ga, and also two knock-out mu- D pyoverdine production is upregulated in response o tant strains which served as negative controls: w n PAO1(cid:2)pvdD (deficient for pyoverdine production) tomorestringentironlimitation[28],aspresumably loa and PAO1(cid:2)rhlR (deficient for the Rhl-quorum- inducedbygallium.Here,wesawacombinationof ded sensingsystemwhichregulatespyocyaninproduc- thesetworegulatoryeffects,withinvestmenttore- fro m placequenchedpyoverdineactuallyincreasingfrom tion[31]).Specifically,weinoculated2mlvolumesof h growthmedium(eitherLBorCAA)with20mlof10(cid:2)3 lowtointermediategalliumsupplementationlevels ttps dilutedovernightLBcultureandincubatedat37(cid:3)C andcessationbecomingevidentonlyathighercon- ://a c in static conditions. After 24h, we measured OD centrations (Fig.1B; ANOVAcomparison of quad- ad e ratic versus linear fits: F >15, P<0.001 in m 600, centrifuged at 7000rpm for 2min, then ex- 1,36 ic tracted 200mlaliquots ofsupernatant andassayed eachcase). .ou p these for growth (OD at 600nm) and levels of Givenourinvitroobservationsofgallium’seffects .c o pyocyanin(usingODat691nm)[32]andpyoverdine ongrowthandpyoverdineproduction,weexpected m/e (fluorescenceat400j460nm),usingaTecanInfinite ittoaffectvirulenceandbacterialfitnessinvivotoo. mp h M-200platereader. Wetestedthisinexperimentalinfectionsofgreater /a waxmoth larvae (G. mellonella). Gallium-supple- rtic le mentedP.aeruginosainfectionsindeedshowedsig- -a b Statistical analyses nificantlyattenuatedvirulencecomparedwithnon- stra c All analyses were performed using R 3.0.0 [33]. supplementedinfections(Fig.2A–C;Weibullcurve t/2 0 comparison: z¼3.10–7.82, P<0.001 in all cases). 1 Splinecurveswerefittedtotimecoursegrowthdata 4 usingthe‘grofit’package[34].Survivalanalyseswere Notably,infectionssupplementedwithmediumand /1/1 8 performedusingtheSurvpackage[35].Althoughin high concentrations of gallium (corresponding to /1 8 the main text we compared survival curves using theintermediategalliumconcentrationusedinthe 43 5 parametricWeibullmodels,wealsorepeatedallana- invitroassays,see‘Methodology’section)weresig- 83 b lyses using Cox proportional hazards regressions, nificantlylessvirulent(z¼4.96and2.39,P<0.05in y g andobtainedqualitativelycomparableresultsinall bothcases)thaninfectionswithPAO1(cid:2)pvdD,amu- ue s cases. tant defective for pyoverdine production that itself t o n showedattenuatedvirulenceversusPAO1(z¼3.49, 0 5 P<0.001).Galliumaloneappearedtohavelittleef- A p RESULTS fect on hosts, with levels of virulence not signifi- ril 2 0 Ininvitroassays,wefoundthatgalliumstronglyin- cantly different from those seen in saline-injected 19 hibitedbacterialgrowth, andthatthe inhibitoryef- controls (Fig. 2A: survival curve comparison: fects were mediated primarily via gallium’s z¼(cid:2)0.93, P¼0.35; Fig. 2B: pairwise proportion extracellularquenchingactivityandnotbecausegal- tests for survival rates: X21¼0.43, P¼0.51). lium is toxic per se (Fig. 1A). When siderophores Bacterial growth in vivo was also significantly were required and could be produced, increasing reducedbygallium(Fig.2DandE).Growthintegrals gallium concentration was associated with a steep wereloweringallium-supplementedlarvaethanin decline in growth (slope±SE of regression with WT-injectedlarvae(Fig.2E;Tukey’s95%CIsforthe log [Ga]: (cid:2)0.435±0.011, t¼(cid:2)38.04, P<0.001). difference:16.21–21.44%,t¼17.24,P<0.001)and, 10 In contrast, when siderophores were not required moreover,lowerthaninlarvaeinjectedwiththesid- and not produced, gallium only weakly affected erophore-defective mutant, PA01(cid:2)pvdD (Fig. 2E; Quenchingpublicgoodsasarobustantibacterialtreatment Ross-Gillespieetal. | 23 D o w n lo a d e d fro m h ttp s ://a c a d e m ic .o u p .c o Figure 2. GalliumattenuatesP.aeruginosavirulenceandgrowthinG.mellonellalarvae.(A–C)Virulenceacrosstreatments,as m /e Kaplan–Meier(steppedlines)andWeibull(smoothedlines)survivalcurves;proportionsurviving(with95%binomialCIs);and m p time-to-death(meansand95%CIs).Weestimatethatinoculawith‘LOW’(2.5–50mM),‘MED’(500mM)or‘HIGH’(2500mM) h/a concentrationsofGa(NO3)3gavein-hostconcentrationsof(cid:4)0.05to(cid:4)50mM(see‘Methodology’section).(D)Bacterialdensity rtic invivo(GFPsignalinhosthomogenate;meansand95%CIsfrom(cid:4)24larvae)correctedagainstsaline-injectedcontrolsand le-a scaledrelativetoPAO1at13h.(E)Meanand95%CIsofbacterialgrowthintegralsderivedfrombootstrapreplicatetimeseries bs (24replicatesplines)from(D) trac t/2 0 1 4 /1 Tukey’s 95% CI¼13.58–18.97%, t¼14.45, toward higher growth (Fig. 3E; H0 slopes¼0; /18 P<0.001). t¼(cid:2)0.30,P¼0.76;t¼1.60,P¼0.11andt¼(cid:2)0.11, /1 8 4 To investigate empirically the general potential P¼0.91forcontrol,Ga1andGa2,respectively). 3 5 8 for resistance against gallium, we performed Per-capita pyoverdine output was generally 3 b experimental evolution with serial batch cultures, steady over the course of experimental evolution y g u comparing P. aeruginosa exposed to gallium ver- [Supplementary Fig. S3: H slopes¼0: control: e 0 s sus several single- and mixed-antibiotic regimes z¼0.56,P¼0.58;Ga1:z¼0.45,P¼0.65;allantibi- t o n (Fig.3A–E).Atfirst,alltreatmentswerestronglyre- otic treatments pooled (Day 1 excluded): z¼0.83, 05 A fractorytogrowth,showing24hgrowthintegralsno P¼0.41],withthatofthe20mMgalliumtreatment p morethanathirdthoseofuntreatedcontrols(range: consistentlyaround2-foldhigherthaneithercontrol ril 2 0 5.8–32.3%).Overthecourseofa12-dayexperiment or antibiotic-treated cultures (95% CIs for fold- 19 (atherapydurationthatmatchesclinicalstandards), difference were 1.86–2.13 versus control, and however, the growth in all antibiotic treatments 1.96–2.24versuspooledantibiotictreatments). increasedsignificantly(Fig.3E;H slopes¼0:Cp1: In the endpoint phenotypic assays performed 0 t¼5.54,Cp2:t¼3.86,Gm1:t¼5.43,Gm2:t¼9.12, under standardized test conditions (CAA and Mix1: t¼5.26, Mix2: t¼7.02; P<0.001 in each LB media), lines evolved in the Ga1 treatment case), andbythefinal timepoint theirgrowthinte- showednosignificantchangeinpyoverdineproduc- gralswerecomparabletothoseoftheuntreatedcon- tion (Fig. 4A) relative to their ancestor (CAA: trols at the start of the experiment (Fig. 3A–D). t¼0.81,P¼0.43;LB:t¼0.08,P¼0.94)ortolines Gallium-treated cultures, meanwhile, like the evolved under control conditions (CAA: t¼(cid:2)0.49, untreated control,didnot showasignificant trend P¼0.63;LB:t¼0.95,P¼0.36),suggestingthatthe 24 | Ross-Gillespieetal. Evolution,Medicine,andPublicHealth D o w n lo a d e d fro m h ttp s ://a c a d e m Figure 3. Evolutionarypotentialforresistanceagainstgalliumtreatment.(A–D)Overthecourseofexperimentalevolution,daily ic .o growthintegralsforculturestreatedwithvariousantibioticsrosesignificantly,whilethegrowthofgalliumtreatedculturesdid u p not.(E)Slopecoefficientsforlinearfitsthroughdatain(A–D),expressedas%ofgrowthofcontrolatDay1.Inallcases,symbols .co m anderrorbarsshowmeansand95%CIsofsixreplicatecultures /e m p h high pyoverdine output seen during experimental may increasingly transit across the cell membrane /a evolution was predominantly a plastic response to andbegintointerferedirectlywithironmetabolism, rticle gallium(seeFig.1B).Incontrast,theproductionof causing general toxicity to bacteria and host cells -ab s pyocyanin did appear to be elevated in the Ga1 alike(Fig.1A;[36]).Here,fitnesscostsareimposed tra c endpoint isolates (Fig. 4B) in CAA medium (ver- intracellularly at the individual cell level, and not t/2 0 sus ancestor: t¼3.40, P¼0.004; versus control: extracellularly at the level of the collective, which 14 t¼3.09, P¼0.008) but not in LB medium (versus wouldtakeusbacktoaclassicantibioticscenario, /1/1 8 ancestor:t¼1.69,P¼0.12;versuscontrol:t¼1.56, with more potential ‘routes to resistance’ and /1 8 P¼0.15). greaterpotentialforsteepfitnessgradientsamong 43 5 8 individual cells. At sub-toxic levels, meanwhile, 3 b where gallium acts primarily through siderophore- y DISCUSSION gu quenching, resistance should evolve less readily. e s Theresultsreportedaboveindicatethatgalliumin- Furthermore,wesawthatthecostsandbenefitsof t o n hibitsP.aeruginosagrowthprimarilythroughextra- siderophore investment itself are also non-linear 05 A cellular interference with its primary siderophore, functionsofgalliumconcentration,owingtotheex- p pyoverdine (Fig. 1A); that this growth inhibition istenceofaregulatory‘trap’.Specifically,intermedi- ril 2 0 occursin aninfection contexttoo(Fig.2DandE), ate concentrations of gallium induced the highest 19 along with a significant reduction in virulence levelsofreplacementpyoverdineproductioninbac- (Fig. 2A–C); and that resistance to gallium treat- teria (Fig. 1B), adding further metabolic stress to mentsdoesnotevolveeasily—atleastnotincom- increasingly iron-limited cells. Our in vivo results, parison to two conventional antibiotics we tested whichshowed thatgallium can suppress virulence (Fig.3). tolevelsbeyondthoseseeninpyoverdine-deficient For gallium to be both optimally effective and strains(Fig.2A–C),areconsistentwiththeinterpret- evolutionarily robust as an antibacterial agent, an ation that an appropriate dose of gallium not only appropriately calibrated dose will be key. At lower restrictsbacterialironuptakebutcanalsoimposea concentrations, efficacy should initially increase costly metabolic burden. Given our understanding with dose, but at too high concentrations, gallium of the regulation of pyoverdine production, this Quenchingpublicgoodsasarobustantibacterialtreatment Ross-Gillespieetal. | 25 D o w n lo a d e d Figure 4. Resistance-relatedphenotypicchangesfollowingexperimentalevolutionundergalliumtreatment.Pyoverdine(A)andpyocyanin(B)productionunder fro m standardizedtestconditions(darkbars¼LBmedium,lightbars¼CAAmedium)ofancestralPAO1,knock-outstrains(i.e.negativecontrols),controllines h (evolvedwithoutgallium)andgallium-selectedlines.PyoverdinemeasuresarescaledtothatofPAO1inCAA,whereaspyocyaninisscaledtothatofPAO1inLB. ttp s Asterisks indicate cases where Ga-selected lines were significantly different from their ancestor and unexposed control lines. Error bars give 95% CIs of ://a 3–6replicates ca d e m ic hump-shaped association between pyoverdine in- conceivably confer resistance against gallium- .ou p vestment and gallium is to be expected. Positive mediatedsiderophorequenching,andunderwhich .c o feedbackoccurswhenincomingFe3þ-boundsidero- conditionssuchadaptionscouldspread.Below,we m/e m phoresactviathereceptorFpvAandtheanti-sigma consider several potential evolutionary responses, p h factorFpvRtoactivatemembrane-boundiron-star- whicharediscussedfurtherinTable1. /a vation sigma factor PvdS [37]. High cytoplasmic First, let’s consider pyoverdine loss-of-function rticle Fe2þlevels,meanwhile,cangeneratenegativefeed- mutants,whichareknowntoarisereadilyunderiron -ab s back.Inthiscase,theFe2þinducesFur(ferricuptake limited conditions [39–41]. In co-infection with sid- tra c regulator)-mediatedrepressionofpvdS[38].Atlow erophore producers, non-producing mutants could t/2 0 galliumconcentrations, ironuptakeintothecellis act as cheats—no longer investing in the PG 14 /1 steady, so negative feedback keeps pyoverdine yetstillbenefitingfromtheinvestmentofnearbyan- /1 8 production at some intermediate level, while at cestors[5].Evenasopportunitiestocheatdwindled, /1 8 4 mid-range gallium concentrations, iron uptake be- such mutants could continue to spread, since, 3 5 8 comes increasingly restricted, leading to steady disadvantaged as they would be with respect to 3 b positive feedback, but weaker negative feedback, autonomousironacquisition,theywouldatthesame y g u andconsequently,pyoverdineproductionincreases. timebefreedofthesubstantialextrametabolicbur- e s Finally,athighconcentrations,ironuptakemaybe denofpyoverdineproductionundergalliumregimes t o n soseverelyrestrictedthatthepositivefeedbackloop (see Fig. 1B). Depending on specific conditions 05 A fails, and pyoverdine production stalls completely. withinhosttissues,thenetfitnessofnon-producers p Exploitingmetabolic‘traps’suchasthiscouldsig- could be not far off that of pyoverdine producers ril 2 0 nificantly increase the effectiveness of treatments, (Fig.2E),sothemutantscouldpotentiallycometo 19 butrequiresthattheassociatedregulatorynetworks occupyasubstantialshareofthepopulation.Wesaw should be left intact and functional. This raises nosignificantchangeinmeanpyoverdineproduction another point in favor of extracellular quenching instrainsevolvedundergallium(Fig.4A),suggesting strategies, as opposed to, say, intracellular- thatcheatsdidnotgainprominenceinthesecultures. mediateddeactivationofentiremolecularpathways. However, certain individual lines (three antibiotic Towhatextentshouldgallium’santibacterialac- linesandoneGa1line)wentextinctduringthecourse tivity be evolutionarily robust? In our selection ex- of the experimental evolution, and this extinction periment(Figs3and4A),wesawlittleevidenceof wasineachcaseaccompaniedbyacrashinpercap- adaptationtogallium,althoughperhapswecanstill ita pyoverdine production levels (Supplementary predict what sort of phenotypic changes could Fig.S3),whichwouldbeconsistentwithascenario 26 | Ross-Gillespieetal. Evolution,Medicine,andPublicHealth Table 1. How likely is resistance against gallium-mediated pyoverdine quenching? Mutant phenotype Why resistant? Likelihood for mutant to arise Likelihood for mutant to spread Pyoverdine production No true resistance, as virulence High Low reduced or shut down. is only partly restored. Pyoverdine-negative mutants arise In mixed cultures, gallium re- However, mutants could avoid readily [39, 41]. duces total population density being ‘trapped’ into high and the effective group size at pyoverdine production which pyoverdine can be (Fig. 1B), which can be a shared, and these effects both substantial fitness drain disfavor the mutant [45, 46]. (Fig. 2E). Pyoverdine modified to Iron uptake efficiency, and Low Low D bind iron with greater hence growth, should (a) Pyoverdine has already (a) Pyoverdine molecules are ow n specificity. improve. evolved high iron specificity shared across the local com- lo a [20]. Further improvements munity [47], so producers of d e d (b) aGrae3þunlikaenlyd. Fe3þ remain tphyeovneordveinleantydpethsewaonucledstbreanlefit from fundamentally very similar in similarly. http binding behavior. s Regulatory shift from Although pyochelin is generally High Low ://a c producing pyoverdine to a less effective siderophore Regulatory mechanisms already (a) Like pyoverdine, pyochelin is ad e producing pyochelin, a than pyoverdine, this strategy exist to facilitate facultative also a shared trait, so benefits m ic secondary siderophore could be advantageous under switching between siderophore would go to non-mutants too. .o u (b) Gallium can quench pyochelin p normally deployed in less extreme conditions (e.g. in types in response to changing .c too, and so it still inhibits iron o iron-limited conditions. the presence of gallium). iron stress [48]. Mutations m uptake [49]. /e that alter this switch could m p probably arise easily. h /a Own pyoverdine production Ceasing pyoverdine production Low Low rtic reducedþspecialization would reduce personal costs, Although P. aeruginosa can al- (a) Most siderophores (e.g. le to use heterologous and heterologous sidero- ready take up heterologous desferrioxamine) are still prone -ab s siderophores from other phores could offer siderophores (e.g. to bind gallium [51]. tra co-infecting species. compensatory benefits. enterobactin, desferrioxamine) (b) Wild-type P. aeruginosa can ct/2 also facultatively switch to 0 [50], this route would require 1 4 cthoa-itnpfercotdiouncewsitahnaacbcaecstseirbiulem hwehteenroelvoegrosuuschsidseidroerpohpohroereusse /1/18/1 become available [50]. 8 siderophore. 4 3 Own pyoverdine production Ceasing pyoverdine production High Low 58 3 reducedþspecialization would reduce personal costs, P. aeruginosa already possesses (a) Some host iron chelators b y to take up iron directly while iron from other sources the means to take up iron in might also bind gallium (e.g. g u from the host. could offer compensatory various forms [50], including citrate). es benefits. when it is in complex with (b) Wild-type P. aeruginosa can t on hosts’ iron chelators. A simple also facultatively switch to al- 05 ternative uptake mechanisms A switch in a regulatory pathway p might be all that is required. when such sources become ril 2 available [50]. 0 1 9 Upregulated production of Reducing agents increase avail- High Low reducing agents ability of the more soluble Upregulation of an already (a) Increased production of a me- (e.g. pyocyanin), which ferrous form of iron (Fe2þ), existing trait could be tabolite would induce extra extracellularly reduce which can be taken up with- achieved easily [30]. costs. (b) Like pyoverdine, pyocyanin is ferric to ferrous iron. out the need for siderophores. also a shared trait, so benefits (in the form of ferrous iron) would go to non-mutants too. Here, weconsider various mutant phenotypes that could putatively confer resistance, and propose hypotheses regarding thelikelihood ofemergence and spread in each case. Quenchingpublicgoodsasarobustantibacterialtreatment Ross-Gillespieetal. | 27 of siderophore-non-producing cheats spreading inducedoverexpressionofpyoverdine(Fig.1B),and in these cultures. In any event, the rise of such in some contexts, this could potentially lead to mutants should still lead to less virulent infections higher virulence, given that pyoverdine production (Fig.2A–C;[42–44]). is linked to certain other virulence factors [53, 54]. Alternative scenarios for evolutionary responses Indeed,whilegalliumisgenerallyknowntoreduce to gallium treatment could involve modifying virulence[19],onerecentstudy[55]showedthatin pyoverdine to have substantially greater affinity verydensecultures,galliumsupplementationactu- forFe3þthanforGa3þ,orswitchingto‘backup’sid- allyupregulatedproductionofcertainvirulencefac- erophores relatively less susceptible to gallium tors.Thus,whilegalliumrepresentsapromisingway (Table 1). Such mutations could conceivably arise toreducebacterialload,itsoveralleffectivenessin but in each scenario we would expect attendant reducingdamagetoahostwill,asalways,depend D selectionforthemutationtoberelativelyweakbe- alsoontheparticularcharacteristicsofthehostand o w cause,asPGs,thesealternativeormodifiedsidero- itsinteractionwiththepathogen. nlo a phores’benefitswouldstillbeaccessibletoallcells d e within diffusion range, including those lacking the CONCLUSIONS AND IMPLICATIONS d fro novelmutation.Inaddition,galliumandironremain m h fundamentally similar in their physical properties, Galliumhasseenapplicationinmedicalcontextsfor ttp s suchthatgalliumwillstillbind—tosomeextentat years(e.g.asananti-cancerdrug[56])andhaspre- ://a least—anymodifiedsiderophore. viously been proposed, and tested, as a treatment ca d Further possible evolutionary responses could againstbacterialinfections[19,25,57,58].Gallium em involvemutantsthatspecializeinthedirectuptake canbedirectlytoxicathighconcentrations,buthere, ic.o of Fe3þ-containing compounds produced by other working with concentrations below this toxic up .c competing microbes (i.e. inter-specific cheats), range, we have focused on its capacity to indir- om orpresentaschelatorsinthehosttissues.Suchmu- ectly affect bacteria through disruption of sidero- /em p tationsarealsoconceivable,giventhatbacteriaal- phore-mediated iron uptake. Specifically, gallium h /a readypossessadiversityofiron-uptakemachineries quenchessiderophoresextracellularly,starvingcells rtic [50].However,consideringthatgalliumcandisplace ofironandpushingthemintoametabolicallycostly le-a Fe3þfromothercompoundstoo,itisnotclearthat regulatory trap from whichthere seems tobe little bs such strategies would offer any clear advantages scopeforevolutionaryescape.Inlightofourresults, trac oversiderophore-mediateduptake. wecontendthatthisapproach—andmoregenerally t/20 1 Finally, bacteria could potentially sidestep their the extracellular targeting of PGs—could curb mi- 4/1 dependence on the Fe3þ form of iron (prevalent crobialvirulenceinanevolutionarilyrobustmanner, /18 /1 under oxygen replete and neutral pH conditions) andthereforerepresentsapromisingalternativeto 8 4 byalteringtheirenvironmenttoincreasetheextra- our dwindling succession of traditional antibiotics 35 8 cellular availability of the more bio-available Fe2þ [59–61]. 3 b y ions. Indeed,overproducersofpyocyanin,aredox- g u activemetabolite,haverecentlybeenreportedtobe es refractory to gallium [30], and in our own experi- supplementary data t on 0 ments, we did see a weak but significant mean 5 SupplementarydataareavailableatEMPHonlineandatthe A icnocnrdeaitsioensinampyooncgyacnuinlturpersodeuvcotlivoend uunnddeerr gcaelrlituamin Dryaddepository:doi:10.5061/dryad.8kk36. pril 2 0 1 treatment (Fig. 4B). However, such metabolites 9 acknowledgements arethemselvesPGs,sothespreadofover-producers could be constrained in due course by the free- WethankA.R.Hall,A.Buckling,S.P.DiggleandF.Gorostidi loading behavior of variants that produce less, yet for discussion, and Ramon Weishaupt and Zoe´ Dumas for stillbenefitbytheincreasedavailabilityofFe2þions. helpinthelaboratory. Inourexperimentalinfections,weobservedthat galliumsupplementationreducedboththevirulence funding and the in-host fitness of P. aeruginosa (Fig. 2). However, pathogen fitness and virulence will not ThisworkwasfundedbytwograntsfromtheSwissNational always be strongly positively correlated [43, 52]. ScienceFoundation(R.K.),aMarieCurieReintegrationgrant Forexample,weshowedthatintermediategallium from the European Commission (R.K.) and the Centre for