Gallium-mediated siderophore quenching as an evolutionarily robust antibacterial treatment

We tested the idea that antibacterial therapies targeting shared exoproducts would be more evolutionarily robust than regular antibiotics. Using gallium, we extracellularly quenched the iron-scavenging molecules shared among Pseudomonas aeruginosa bacteria, inhibiting their growth and virulence in caterpillars. During experimental evolution, gallium remained effective over time, while antibiotics did not.


INTRODUCTION
Like all organisms, pathogens acquire genetic mutations, and, in time, even 'pure' cultures will inevitably come to harbor mutant lineages. Such genetic variability can make some pathogen variants less sensitive to therapeutic interventions than others, and under strong or sustained therapy, these resistant variants will have a selective advantage and will come to predominate over more susceptible variants. Consequently, the therapy will lose efficacy [1,2]. To avoid this situation, we can try to prevent resistant variants from arising and/or from spreading [3]. To prevent resistance arising, we could attempt to reduce mutation supply, through limiting effective population size or by employing interventions with specialized modes of action where relatively few 'routes to resistance' are possible. To prevent spread, meanwhile, we must aim to minimize fitness differences across individual pathogens. Killing every individual, the conventional antibiotic strategy, could certainly quash fitness evenly, but this is difficult in practice and whenever incomplete gives resistant pathogens a strong relative fitness advantage. 'Antivirulence' treatments, meanwhile, ostensibly disarm but do not harm pathogens, such that resistant variants should benefit little relative to susceptibles [4]. However, traits that affect virulence but not fitness are rare, and the label 'antivirulence' is used liberally, even for interventions that yield substantial fitness differences among pathogens [4]. A final way to minimize fitness differences is to target pathogens' collective traits, where costs and benefits are widely shared. For instance, many virulence-related bacterial exoproducts are also public goods (PGs) [5]. Under PG-quenching therapy, any mutations allowing PGs to build up again should benefit both resistant and susceptible individuals alike, which would hinder the spread of resistance [1, [6][7][8].
To illustrate why this matters, let's consider a specific example. Quorum quenching (QQ), which disrupts the cell-to-cell communication [quorum sensing (QS)] [9] underlying a wide range of collectively expressed virulence traits, is a PG-targeting 'antivirulence' therapy regarded as a promising alternative to conventional bacteriocidal or bacteriostatic treatments [10,11]. However, early enthusiasm for QQ has been tempered recently by reports that bacteria can quite readily evolve resistance to such treatments [12][13][14]. Set against our framework, this is unsurprising: first, QQ interventions frequently involve intracellular action, against which many potential resistance-conferring adaptations could arise (e.g. modified membrane properties to block a drug's entry into a cell, or upregulated efflux pumps to hasten its exit [15]). Second, QS regulates not only PGs but also certain essential private goods [16], giving QQ resistants substantial personal benefits over susceptibles-and therefore a means to spread. For maximal evolutionary robustness, we need therapies where resistance mutations are unlikely to arise in the first place (e.g. extracellular action restricts potential routes to resistance) and are also unlikely to spread, because fitness differences between resistant and susceptible pathogens are minimized. The latter should be the case when collective traits are targeted, because fitness consequences are shared across many individuals. Of course, the extent and evenness of this sharing will depend on the relatedness and spatial structure of the pathogen population and the diffusive properties of the environment, and these factors would also need to be considered during therapy design [3].
In this study, we investigate-in a test case-the hypothesis that extracellular PG quenching is an effective and evolutionarily robust strategy for pathogen control. The PG trait we target is siderophores, important exoproducts whose regulation is not linked to any exclusively private goods. Siderophores are diffusible molecules with a high affinity for ferric iron (Fe 3þ ) and are secreted by most bacteria to scavenge this important but generally bio-unavailable form of iron from their environment or, in the case of pathogens, from their host's own iron-chelating compounds [17]. Once loaded with Fe 3þ , siderophores are taken up by producer cells-or other nearby individuals equipped with appropriate receptorsstripped of their iron, and secreted once again into the environment [18]. Although their primary function may be to scavenge iron, siderophores also bind, with varying success, several other metals [19,20]. Among these, gallium is the closest mimic of iron. Ga 3þ and Fe 3þ ions have very similar ionic radii and binding propensities but, crucially, while Fe 3þ reduces readily, Ga 3þ does not [19]. Ga 3þ therefore cannot replace iron as a co-factor in redox-dependent enzymes. We investigated the iron-mimicking effects of gallium on pyoverdine, the primary siderophore of Pseudomonas aeruginosa [21], a widespread opportunistic pathogen with a broad host range and, in humans, the cause of notoriously persistent infections in immunecompromised tissues, cystic fibrosis lungs and in association with implanted devices [22]. Pyoverdine, which plays an important role in such infections [23,24], binds gallium at least as readily as iron, and gallium-bound pyoverdine is of no use to iron-starved cells [19,20]. Thus, even without entering the cell, gallium can reduce P. aeruginosa growth and biofilm formation by quenching local stocks of secreted pyoverdine and choking off iron supply [19,25].
Below, we report our investigations into (i) gallium's in vitro interference with siderophoremediated iron uptake and consequent effects on bacterial growth, (ii) gallium's in vivo effects on virulence and in-host bacterial growth and (iii) the potential for bacteria to evolve resistance against gallium treatment.
In vitro assays of growth and pyoverdine production Overnight LB cultures (37 C, 180 rpm), washed and standardized for cell density, were diluted to 10 À4 then used to seed replicate cultures in CAA medium supplemented with Ga(NO 3 ) 3 (such that final Ga concentrations ranged from 0 to 200 mM), as well as complementary amounts of NaNO 3 to balance nitrate levels across treatments, and 20 mM FeCl 3 where iron-replete conditions were required (Fig. 1A). Growth assays were performed with 200 ml cultures in 96-well plates, for which optical density (OD) was tracked over 24 h at 37 C using a Tecan Infinite M-200 plate reader (Tecan Group Ltd., Switzerland), with 15 min read intervals preceded at each read by 10 s of agitation. To assay pyoverdine production, we first grew PAO1pvdA-gfp in 2 ml CAA static in 24-well plates in a 37 C incubator for 24 h, then centrifuged the cultures at 7000 rpm for 2 min to pellet the cells. From each culture, 200 ml of supernatant and, separately, the cell fraction resuspended in 200 ml 0.8% saline, were transferred to a new 96well plate and assayed for OD at 600 nm and fluorescence (GFP in cell fraction: exjem ¼ 488j520 nm; pyoverdine in supernatant: 400j460 nm) [28]. Both fluorescence measures were standardized by OD at 600 nm. In a series of side experiments, we investigated potential biases associated with the use of optical measures as proxies for pyoverdine production ( Supplementary Fig. S1). Data presented in Fig. 1B are corrected for these biases.

Experimental infections
Infection assays were performed with final instar Galleria mellonella larvae, purchased from a local supplier, standardized for mass and general condition and stored at 4 C until use (within 3 days). A Hamilton precision syringe was used to deliver 10 ml inocula via a sterile 26s gauge needle introduced sub-dermally to a surface-sterilized area between the last pair of prolegs. Inoculations contained Ga(NO 3 ) 3 diluted to different concentrations in 0.8% saline, with complementary concentrations of NaNO 3 , and, where specified, bacteria from overnight LB cultures (37 C, 180 rpm), standardized for cell density and diluted such that each 10 ml inocula contained $25 CFU (post hoc counts of 12 inocula plated out to LB agar gave 95% CI of 19. 41-31.76). Specifically, we tested the following Ga(NO 3 ) 3 concentrations: 2.5, 10 and 50 mM ('LOW'; pooled together since their resulting virulence curves were not significantly different from one another), 500 mM ('MED') and 2500 mM ('HIGH'). Our 'Gallium only' treatment comprised various concentrations between 2.5 and 2500 mM, which again we pooled for statistical analyses because of similar effects on survival. Post-injection, larvae were placed individually in randomly allocated wells of 24-well plates and incubated at 37 C. Survival was monitored hourly between 10 and 24 h, and larvae were considered dead once they no longer responded to tactile stimulation. Any larvae that began to pupate while under observation or died within the first 10 h postinjection (i.e. as a result of handling) were excluded from analyses (n ¼ 23, 3.6%). To assay in vivo bacterial growth, we prepared our inocula with strains engineered to constitutively express GFP (see above), having previously established that constitutively expressed GFP signal could provide a reliable correlate of bacterial density under the conditions of this infection model ( Supplementary Fig. S2). In each of six separate experimental blocks, and at each of four discrete timepoints, 3-4 randomly selected larvae per treatment were flash-frozen in liquid N 2 and manually powderized. Powdered larval homogenates were resuspended in 1 ml sterile H 2 O, vigorously shaken and then centrifuged at 7000 rpm for 2 min, whereafter the sample segregated into discrete phases. About 200 ml of the water-soluble liquid phase was extracted and assayed for GFPfluorescent signal relative to control replicates (saline-injected larvae), using a Tecan Infinite M-200 plate reader. Given total larval volumes of $1 ml, and assuming that $20% of this volume might be hemolymph accessible to particles diffusing from a single injection site during the course of an acute infection, we estimate that inocula gallium concentrations of 2.5-2500 mM would translate to in-host gallium concentrations of roughly $0.05 to $50 mM.

Experimental evolution
We compared the growth inhibitory effects of gallium versus the aminoglycoside, gentamicin (Gm), and the fluoroquinolone, ciprofloxacin (Cp)-two of several antibiotics recommended for clinical use against P. aeruginosa [29]. Concentrations were calibrated such that they reduced growth integrals over the initial 24 h to 1/3 that of untreated PAO1 WT cultures under the same growth conditions. For each of 12 days, a 96-well plate was prepared, comprising replicate 198 ml volumes of iron-limited CAA medium supplemented, according to a randomized layout scheme, with gallium, antibiotics or an equivalent volume of saline (see key in Fig. 3 for details of treatments used and their respective sample sizes). Day 1 cultures were initiated with 2 ml aliquots of a 10 À3 diluted overnight LB culture of PAO1 WT (37 C, 180 rpm), while for subsequent days, fresh plates were inoculated with 10 ml of undiluted culture from the corresponding wells of the previous day's plate, directly after it completed its growth cycle. Plates were incubated at 37 C, and cell density and pyoverdine fluorescence measures were recorded at 15 min intervals (with 10 s initial shaking) using a Tecan Infinite M-200 plate reader.

Endpoint phenotypic assays
Prior observations [30] and our own reasoning (see Table 1) suggested that pyoverdine and pyocyanin could both affect the costs and benefits of iron uptake under gallium treatment. Anticipating that the experimental evolution described above might have induced changes in these traits, we performed phenotypic assays to compare cultures of our ancestral PAO1 WT, its descendent lines experimentally evolved in CAA with or without supplementation with 20 mM Ga, and also two knock-out mutant strains which served as negative controls: PAO1ÁpvdD (deficient for pyoverdine production) and PAO1ÁrhlR (deficient for the Rhl-quorumsensing system which regulates pyocyanin production [31]). Specifically, we inoculated 2 ml volumes of growth medium (either LB or CAA) with 20 ml of 10 À3 diluted overnight LB culture and incubated at 37 C in static conditions. After 24 h, we measured OD 600, centrifuged at 7000 rpm for 2 min, then extracted 200 ml aliquots of supernatant and assayed these for growth (OD at 600 nm) and levels of pyocyanin (using OD at 691 nm) [32] and pyoverdine (fluorescence at 400j460 nm), using a Tecan Infinite M-200 plate reader.

Statistical analyses
All analyses were performed using R 3.0.0 [33]. Spline curves were fitted to time course growth data using the 'grofit' package [34]. Survival analyses were performed using the Surv package [35]. Although in the main text we compared survival curves using parametric Weibull models, we also repeated all analyses using Cox proportional hazards regressions, and obtained qualitatively comparable results in all cases.

RESULTS
In in vitro assays, we found that gallium strongly inhibited bacterial growth, and that the inhibitory effects were mediated primarily via gallium's extracellular quenching activity and not because gallium is toxic per se (Fig. 1A). When siderophores were required and could be produced, increasing gallium concentration was associated with a steep decline in growth (slope ± SE of regression with log 10 [Ga]: À0.435 ± 0.011, t ¼ À38.04, P < 0.001). In contrast, when siderophores were not required and not produced, gallium only weakly affected growth (slope ± SE: À0.067 ± 0.019, 95% CI for drop ¼ [2.91-15.86%]; difference in slopes 0.368 ± 0.022, F 1,140 ¼ 276.41, P < 0.001)-particularly over the range of concentrations up to and including 50 mM, which correspond to the concentrations likely experienced in our in vivo experiments (see below).
It has been suggested that as the benefit of pyoverdine production drops, bacteria should gradually scale back their investment in this trait [19]. On the other hand, it has also been shown that pyoverdine production is upregulated in response to more stringent iron limitation [28], as presumably induced by gallium. Here, we saw a combination of these two regulatory effects, with investment to replace quenched pyoverdine actually increasing from low to intermediate gallium supplementation levels and cessation becoming evident only at higher concentrations ( Fig. 1B; ANOVA comparison of quadratic versus linear fits: F 1,36 > 15, P < 0.001 in each case).
Given our in vitro observations of gallium's effects on growth and pyoverdine production, we expected it to affect virulence and bacterial fitness in vivo too. We tested this in experimental infections of greater waxmoth larvae (G. mellonella). Gallium-supplemented P. aeruginosa infections indeed showed significantly attenuated virulence compared with nonsupplemented infections ( Fig. 2A-C; Weibull curve comparison: z ¼ 3.10-7.82, P < 0.001 in all cases). Notably, infections supplemented with medium and high concentrations of gallium (corresponding to the intermediate gallium concentration used in the in vitro assays, see 'Methodology' section) were significantly less virulent (z ¼ 4.96 and 2.39, P < 0.05 in both cases) than infections with PAO1ÁpvdD, a mutant defective for pyoverdine production that itself showed attenuated virulence versus PAO1 (z ¼ 3.49, P < 0.001). Gallium alone appeared to have little effect on hosts, with levels of virulence not significantly different from those seen in saline-injected controls ( Fig. 2A: survival curve comparison: z ¼ À0.93, P ¼ 0.35; Fig. 2B: pairwise proportion tests for survival rates: X 2 1 ¼ 0.43, P ¼ 0.51). Bacterial growth in vivo was also significantly reduced by gallium ( Fig. 2D and E). Growth integrals were lower in gallium-supplemented larvae than in WT-injected larvae ( Fig. 2E; Tukey's 95% CIs for the difference: 16.21-21.44%, t ¼ 17.24, P < 0.001) and, moreover, lower than in larvae injected with the siderophore-defective mutant, PA01ÁpvdD ( Fig. 2E; Tukey's 95% CI ¼ 13.58-18.97%, t ¼ 14.45, P < 0.001).
To investigate empirically the general potential for resistance against gallium, we performed experimental evolution with serial batch cultures, comparing P. aeruginosa exposed to gallium versus several single-and mixed-antibiotic regimes (Fig. 3A-E). At first, all treatments were strongly refractory to growth, showing 24 h growth integrals no more than a third those of untreated controls (range: 5.8-32.3%). Over the course of a 12-day experiment (a therapy duration that matches clinical standards), however, the growth in all antibiotic treatments increased significantly ( Fig. 3E; H 0 slopes ¼ 0: Cp1: t ¼ 5.54, Cp2: t ¼ 3.86, Gm1: t ¼ 5.43, Gm2: t ¼ 9.12, Mix1: t ¼ 5.26, Mix2: t ¼ 7.02; P < 0.001 in each case), and by the final timepoint their growth integrals were comparable to those of the untreated controls at the start of the experiment (Fig. 3A-D). Gallium-treated cultures, meanwhile, like the untreated control, did not show a significant trend toward higher growth ( Fig. 3E; H 0 slopes ¼ 0; t ¼ À0.30, P ¼ 0.76; t ¼ 1.60, P ¼ 0.11 and t ¼ À0.11, P ¼ 0.91 for control, Ga1 and Ga2, respectively).
In the endpoint phenotypic assays performed under standardized test conditions (CAA and LB media), lines evolved in the Ga1 treatment showed no significant change in pyoverdine production (Fig. 4A)   high pyoverdine output seen during experimental evolution was predominantly a plastic response to gallium (see Fig. 1B). In contrast, the production of pyocyanin did appear to be elevated in the Ga1 endpoint isolates (Fig. 4B) in CAA medium (versus ancestor: t ¼ 3.40, P ¼ 0.004; versus control: t ¼ 3.09, P ¼ 0.008) but not in LB medium (versus ancestor: t ¼ 1.69, P ¼ 0.12; versus control: t ¼ 1.56, P ¼ 0.15).

DISCUSSION
The results reported above indicate that gallium inhibits P. aeruginosa growth primarily through extracellular interference with its primary siderophore, pyoverdine (Fig. 1A); that this growth inhibition occurs in an infection context too (Fig. 2D and E), along with a significant reduction in virulence ( Fig. 2A-C); and that resistance to gallium treatments does not evolve easily-at least not in comparison to two conventional antibiotics we tested (Fig. 3).
For gallium to be both optimally effective and evolutionarily robust as an antibacterial agent, an appropriately calibrated dose will be key. At lower concentrations, efficacy should initially increase with dose, but at too high concentrations, gallium may increasingly transit across the cell membrane and begin to interfere directly with iron metabolism, causing general toxicity to bacteria and host cells alike ( Fig. 1A; [36]). Here, fitness costs are imposed intracellularly at the individual cell level, and not extracellularly at the level of the collective, which would take us back to a classic antibiotic scenario, with more potential 'routes to resistance' and greater potential for steep fitness gradients among individual cells. At sub-toxic levels, meanwhile, where gallium acts primarily through siderophorequenching, resistance should evolve less readily. Furthermore, we saw that the costs and benefits of siderophore investment itself are also non-linear functions of gallium concentration, owing to the existence of a regulatory 'trap'. Specifically, intermediate concentrations of gallium induced the highest levels of replacement pyoverdine production in bacteria (Fig. 1B), adding further metabolic stress to increasingly iron-limited cells. Our in vivo results, which showed that gallium can suppress virulence to levels beyond those seen in pyoverdine-deficient strains ( Fig. 2A-C), are consistent with the interpretation that an appropriate dose of gallium not only restricts bacterial iron uptake but can also impose a costly metabolic burden. Given our understanding of the regulation of pyoverdine production, this hump-shaped association between pyoverdine investment and gallium is to be expected. Positive feedback occurs when incoming Fe 3þ -bound siderophores act via the receptor FpvA and the anti-sigma factor FpvR to activate membrane-bound iron-starvation sigma factor PvdS [37]. High cytoplasmic Fe 2þ levels, meanwhile, can generate negative feedback. In this case, the Fe 2þ induces Fur (ferric uptake regulator)-mediated repression of pvdS [38]. At low gallium concentrations, iron uptake into the cell is steady, so negative feedback keeps pyoverdine production at some intermediate level, while at mid-range gallium concentrations, iron uptake becomes increasingly restricted, leading to steady positive feedback, but weaker negative feedback, and consequently, pyoverdine production increases. Finally, at high concentrations, iron uptake may be so severely restricted that the positive feedback loop fails, and pyoverdine production stalls completely. Exploiting metabolic 'traps' such as this could significantly increase the effectiveness of treatments, but requires that the associated regulatory networks should be left intact and functional. This raises another point in favor of extracellular quenching strategies, as opposed to, say, intracellularmediated deactivation of entire molecular pathways.
To what extent should gallium's antibacterial activity be evolutionarily robust? In our selection experiment (Figs 3 and 4A), we saw little evidence of adaptation to gallium, although perhaps we can still predict what sort of phenotypic changes could conceivably confer resistance against galliummediated siderophore quenching, and under which conditions such adaptions could spread. Below, we consider several potential evolutionary responses, which are discussed further in Table 1.
First, let's consider pyoverdine loss-of-function mutants, which are known to arise readily under iron limited conditions [39][40][41]. In co-infection with siderophore producers, non-producing mutants could act as cheats-no longer investing in the PG yet still benefiting from the investment of nearby ancestors [5]. Even as opportunities to cheat dwindled, such mutants could continue to spread, since, disadvantaged as they would be with respect to autonomous iron acquisition, they would at the same time be freed of the substantial extra metabolic burden of pyoverdine production under gallium regimes (see Fig. 1B). Depending on specific conditions within host tissues, the net fitness of non-producers could be not far off that of pyoverdine producers (Fig. 2E), so the mutants could potentially come to occupy a substantial share of the population. We saw no significant change in mean pyoverdine production in strains evolved under gallium (Fig. 4A), suggesting that cheats did not gain prominence in these cultures. However, certain individual lines (three antibiotic lines and one Ga1 line) went extinct during the course of the experimental evolution, and this extinction was in each case accompanied by a crash in per capita pyoverdine production levels ( Supplementary  Fig. S3), which would be consistent with a scenario  Pyoverdine production reduced or shut down.
No true resistance, as virulence is only partly restored. However, mutants could avoid being 'trapped' into high pyoverdine production (Fig. 1B), which can be a substantial fitness drain (Fig. 2E).
In mixed cultures, gallium reduces total population density and the effective group size at which pyoverdine can be shared, and these effects both disfavor the mutant [45,46].
Pyoverdine modified to bind iron with greater specificity.
Iron uptake efficiency, and hence growth, should improve.
Low (a) Pyoverdine has already evolved high iron specificity [20]. Further improvements are unlikely. (b) Ga 3þ and Fe 3þ remain fundamentally very similar in binding behavior.
Low (a) Pyoverdine molecules are shared across the local community [47], so producers of the novel and the ancestral pyoverdine types would benefit similarly.
Regulatory shift from producing pyoverdine to producing pyochelin, a secondary siderophore normally deployed in less iron-limited conditions.
Although pyochelin is generally a less effective siderophore than pyoverdine, this strategy could be advantageous under extreme conditions (e.g. in the presence of gallium).
High Low (a) Like pyoverdine, pyochelin is also a shared trait, so benefits would go to non-mutants too. (b) Gallium can quench pyochelin too, and so it still inhibits iron uptake [49].
Regulatory mechanisms already exist to facilitate facultative switching between siderophore types in response to changing iron stress [48]. Mutations that alter this switch could probably arise easily. Own pyoverdine production reduced þ specialization to use heterologous siderophores from other co-infecting species.
Ceasing pyoverdine production would reduce personal costs, and heterologous siderophores could offer compensatory benefits.
Although P. aeruginosa can already take up heterologous siderophores (e.g. enterobactin, desferrioxamine) [50], this route would require co-infection with a bacterium that produces an accessible siderophore. Own pyoverdine production reduced þ specialization to take up iron directly from the host.
Ceasing pyoverdine production would reduce personal costs, while iron from other sources could offer compensatory benefits.
High Low (a) Some host iron chelators might also bind gallium (e.g. citrate). (b) Wild-type P. aeruginosa can also facultatively switch to alternative uptake mechanisms when such sources become available [50].
P. aeruginosa already possesses the means to take up iron in various forms [50], including when it is in complex with hosts' iron chelators. A simple switch in a regulatory pathway might be all that is required.
Upregulated production of reducing agents (e.g. pyocyanin), which extracellularly reduce ferric to ferrous iron.
Reducing agents increase availability of the more soluble ferrous form of iron (Fe 2þ ), which can be taken up without the need for siderophores.
High Low (a) Increased production of a metabolite would induce extra costs. (b) Like pyoverdine, pyocyanin is also a shared trait, so benefits (in the form of ferrous iron) would go to non-mutants too.
Upregulation of an already existing trait could be achieved easily [30].
of siderophore-non-producing cheats spreading in these cultures. In any event, the rise of such mutants should still lead to less virulent infections ( Fig. 2A-C; [42][43][44]). Alternative scenarios for evolutionary responses to gallium treatment could involve modifying pyoverdine to have substantially greater affinity for Fe 3þ than for Ga 3þ , or switching to 'backup' siderophores relatively less susceptible to gallium (Table 1). Such mutations could conceivably arise but in each scenario we would expect attendant selection for the mutation to be relatively weak because, as PGs, these alternative or modified siderophores' benefits would still be accessible to all cells within diffusion range, including those lacking the novel mutation. In addition, gallium and iron remain fundamentally similar in their physical properties, such that gallium will still bind-to some extent at least-any modified siderophore.
Further possible evolutionary responses could involve mutants that specialize in the direct uptake of Fe 3þ -containing compounds produced by other competing microbes (i.e. inter-specific cheats), or present as chelators in the host tissues. Such mutations are also conceivable, given that bacteria already possess a diversity of iron-uptake machineries [50]. However, considering that gallium can displace Fe 3þ from other compounds too, it is not clear that such strategies would offer any clear advantages over siderophore-mediated uptake.
Finally, bacteria could potentially sidestep their dependence on the Fe 3þ form of iron (prevalent under oxygen replete and neutral pH conditions) by altering their environment to increase the extracellular availability of the more bio-available Fe 2þ ions. Indeed, overproducers of pyocyanin, a redoxactive metabolite, have recently been reported to be refractory to gallium [30], and in our own experiments, we did see a weak but significant mean increase in pyocyanin production under certain conditions among cultures evolved under gallium treatment (Fig. 4B). However, such metabolites are themselves PGs, so the spread of over-producers could be constrained in due course by the freeloading behavior of variants that produce less, yet still benefit by the increased availability of Fe 2þ ions.
In our experimental infections, we observed that gallium supplementation reduced both the virulence and the in-host fitness of P. aeruginosa (Fig. 2). However, pathogen fitness and virulence will not always be strongly positively correlated [43,52]. For example, we showed that intermediate gallium induced overexpression of pyoverdine (Fig. 1B), and in some contexts, this could potentially lead to higher virulence, given that pyoverdine production is linked to certain other virulence factors [53,54]. Indeed, while gallium is generally known to reduce virulence [19], one recent study [55] showed that in very dense cultures, gallium supplementation actually upregulated production of certain virulence factors. Thus, while gallium represents a promising way to reduce bacterial load, its overall effectiveness in reducing damage to a host will, as always, depend also on the particular characteristics of the host and its interaction with the pathogen.

CONCLUSIONS AND IMPLICATIONS
Gallium has seen application in medical contexts for years (e.g. as an anti-cancer drug [56]) and has previously been proposed, and tested, as a treatment against bacterial infections [19,25,57,58]. Gallium can be directly toxic at high concentrations, but here, working with concentrations below this toxic range, we have focused on its capacity to indirectly affect bacteria through disruption of siderophore-mediated iron uptake. Specifically, gallium quenches siderophores extracellularly, starving cells of iron and pushing them into a metabolically costly regulatory trap from which there seems to be little scope for evolutionary escape. In light of our results, we contend that this approach-and more generally the extracellular targeting of PGs-could curb microbial virulence in an evolutionarily robust manner, and therefore represents a promising alternative to our dwindling succession of traditional antibiotics [59][60][61].