The effect of metal remediation on the virulence and 1 antimicrobial resistance of the opportunistic pathogen 2 Pseudomonas aeruginosa 3

Metal contamination poses both a direct threat to human health as well as an indirect 28 threat through its potential to affect bacterial pathogens. Metals can not only co-select 29 for antibiotic resistance, but also might affect pathogen virulence via increased 30 siderophore production. Siderophores are extracellular compounds released to 31 increase ferric iron uptake — a common limiting factor for pathogen growth within 32 hosts – making them an important virulence factor. However, siderophores can also 33 be positively selected for to detoxify non-ferrous metals, and consequently metal 34 stress can potentially increase bacterial virulence. Anthropogenic methods to 35 remediate environmental metal contamination commonly involve amendment with 36 lime-containing materials, but whether this reduces in situ co-selection for antibiotic 37 resistance and virulence remains unknown. Here, using microcosms containing metal- 38 contaminated river water and sediment, we experimentally test whether metal 39 remediation by liming reduces co-selection for these traits in the opportunistic 40 pathogen Pseudomonas aeruginosa embedded within a natural microbial community. 41 To test for the effects of environmental structure, which can impact siderophore 42 production, microcosms were incubated under either static or shaking conditions. 43 Evolved P. aeruginosa populations had greater fitness in the presence of toxic 44 concentrations of copper than the ancestral strain, but this effect was reduced in the 45 limed treatments. Evolved P. aeruginosa populations showed increased resistance to 46 the clinically-relevant antibiotics apramycin, cefotaxime, and trimethoprim, regardless 47 of lime addition or environmental structure. Although we found virulence to be 48 significantly associated with siderophore production, neither virulence nor siderophore 49 production significantly differed between the four treatments. We therefore 50 demonstrate that although remediation via liming reduced the strength of selection for 51 metal resistance mechanisms, it did not mitigate metal-imposed selection for antibiotic resistance or virulence in P. aeruginosa . metal-contaminated environments may select for antibiotic resistance and virulence traits even when lime.


27
Metal contamination poses both a direct threat to human health as well as an indirect 28 threat through its potential to affect bacterial pathogens. Metals can not only co-select 29 for antibiotic resistance, but also might affect pathogen virulence via increased 30 siderophore production. Siderophores are extracellular compounds released to 31 increase ferric iron uptakea common limiting factor for pathogen growth within 32 hostsmaking them an important virulence factor. However, siderophores can also 33 be positively selected for to detoxify non-ferrous metals, and consequently metal 34 stress can potentially increase bacterial virulence. Anthropogenic methods to 35 remediate environmental metal contamination commonly involve amendment with 36 lime-containing materials, but whether this reduces in situ co-selection for antibiotic 37 resistance and virulence remains unknown. Here, using microcosms containing metal-38 contaminated river water and sediment, we experimentally test whether metal 39 remediation by liming reduces co-selection for these traits in the opportunistic 40 pathogen Pseudomonas aeruginosa embedded within a natural microbial community. 41 To test for the effects of environmental structure, which can impact siderophore 42 production, microcosms were incubated under either static or shaking conditions. 43 Evolved P. aeruginosa populations had greater fitness in the presence of toxic 44 concentrations of copper than the ancestral strain, but this effect was reduced in the 45 limed treatments. Evolved P. aeruginosa populations showed increased resistance to 46 the clinically-relevant antibiotics apramycin, cefotaxime, and trimethoprim, regardless 47 of lime addition or environmental structure. Although we found virulence to be 48 significantly associated with siderophore production, neither virulence nor siderophore 49 production significantly differed between the four treatments. We therefore 50 demonstrate that although remediation via liming reduced the strength of selection for 51 metal resistance mechanisms, it did not mitigate metal-imposed selection for antibiotic 52 resistance or virulence in P. aeruginosa. Consequently, metal-contaminated 53 environments may select for antibiotic resistance and virulence traits even when 54 treated with lime. 55 56 57 58 unexplored. A key microbial trait likely to change after liming is the production of 83 siderophore compounds (18). The canonical function of siderophores is to aid iron (Fe) 84 sequestration from the extra-cellular environment (25,26). Fe is vital for microbial 85 growth as a cofactor for a number of essential enzymes (27,28), but is most commonly 86 present as insoluble Fe 3+ and therefore is of limited bioavailability, especially at near-87 neutral pH (27)(28)(29)(30)(31)(32). Siderophores are released by cells where they form extracellular 88 complexes with Fe 3+ , these are then taken up by selective outer-membrane transport 89 proteins before Fe 3+ is reduced to bioavailable Fe 2+ and the siderophore made 90 available for reuse (33). Siderophores are important virulence factors as they allow 91 pathogens to grow within hosts that actively withhold iron (34,35). Apart from iron, 92 siderophores can also chelate toxic metal ions, but these complexes cannot re-enter 93 the cell due to the selectivity of the outer-membrane transport proteins (28,29). This 94 means siderophore production can be selected for as a detoxifying method in the 95 presence of bioavailable toxic metals (26,29,36). Consequently, toxic metal 96 concentrations can select for greater virulence by selecting for increased siderophore 97 production (37). Lime remediation of metal-contaminated environments thus could 98 potentially select either for the upregulation of siderophore production when it 99 predominantly results in decreased bioavailability of Fe, or for the downregulation of 100 siderophore production when it predominantly results in lower metal toxicity, with 101 concomitant expected changes in virulence. Previous work has shown siderophore 102 production to decrease as a consequence of liming at the level of whole microbial 103 communities (18), but whether this also occurs in environmental pathogens that rely 104 on siderophore-mediated iron uptake remains untested. 105

106
It is well established that some mechanisms that bacteria use to resist metal 107 contamination also confer resistance to antibiotics (38). This can occur through cross-108 resistance when a single mechanism provides resistance to both types of stressors 109 (e.g. efflux pumps (38-44)), through co-resistance when metal and antibiotic 110 resistance genes are located on the same genetic element (45, 46)), or through co-111 regulation when transcriptional and translational responses to both stressors are 112 linked (38,43,(47)(48)(49). However, to our knowledge, it remains untested whether metal 113 remediation could decrease such co-selection for antibiotic resistance. 114

115
In this study, we use the opportunistic pathogen Pseudomonas aeruginosa to test 116 whether liming alters virulence by influencing siderophore production, and whether it 117 decreases co-selection by metals for antibiotic resistance. We applied an experimental 118 evolution approach, utilising microcosms containing water and sediment and the 119 resident microbial community from a river heavily contaminated with historical mine 120 waste (50, 51). We embedded P. aeruginosa within this natural microbial community 121 and quantified antibiotic resistance, siderophore production and virulence in this focal 122 species after 14 days. P. aeruginosa is responsible for a significant proportion of 123 nosocomial infections, particularly those in intensive care units and 124 immunocompromised patients (52). This species is of significant clinical importance 125 as it is resistant to many treatments, both intrinsically and due to its ability to rapidly 126 evolve resistance (53). Outside of the clinical setting, P. aeruginosa is commonly found 127 in soil and water (54). The production of siderophores by P. aeruginosa is well-studied 128 as a virulence factor, metal resistance mechanism and public good (25, 27, 29, 55-129 57). Furthermore, the growing interest in its use along with other siderophore 130 producing species to assist phytoremediation of metals using plants (28, 58), makes 131 it an ideal focal species for this study. 132

133
The insect infection model Galleria mellonella (Greater Wax Moth larvae), a low-cost 134 and ethically expedient alternative for mammalian virulence screens (59), is used here 135 to quantify P. aeruginosa virulence (60). We quantified total siderophore production 136 using a CAS assay (61) and pyoverdine productionthe main siderophore produced 137 by P. aeruginosa (62)by measuring fluorescence; and tested whether these are 138 correlated with virulence. Extracellular siderophore-metal complexes offer a fitness 139 advantage not only to the producer but also to neighbouring cells, whether these are 140 fellow-producers or not (63-65). Non-siderophore producing 'cheats' could gain a 141 selective advantage as they benefit from siderophore production but do not carry the 142 cost of production (25,30,31,66). Cheat fitness is increased in spatially unstructured 143 environments because the greater mixing increases the opportunity to take up 144 siderophore-iron complexes and benefit from siderophores detoxifying the area (65). 145 To take into account the effect of spatial structure on siderophore production, and 146 consequently virulence, we performed our experiments in both static and shaken 147 microcosms. We tested whether the addition of lime or a change in spatial structure 148 affects P. aeruginosa resistance to the antibiotics apramycin, cefotaxime and 149 trimethoprim. Both apramycin and cefotaxime have been declared 'critically important' 150 for human medicine, and trimethoprim 'highly important' by the WHO (67). Moreover, 151 apramycin has been shown to be effective against highly drug resistant strains of P. and contains high concentrations of non-ferrous metals. Sediment was collected using 163 a sterile spatula and water was collected by filling a sterile 1000 mL duran bottle 164 (Schott Duran, Munich, Germany). Sediment (3 g +/-0.1 g) and river water (6 mL) was 165 added to each microcosm (25 mL, Kartell, Noviglio, Italy). The combined water and 166 sediment pH was measured using a Jenway 3510 pH meter (Jenway, Essex, UK). 167

Experimental design 169
Two treatmentsliming (lime amendment/no amendment) and spatial structure 170 (shaken/unshaken)were carried out in a full factorial design (Fig. 1); six replicates 171 were used per unique treatment combination resulting in a total of 24 microcosms. All 172 microcosms were incubated at 20°C. To raise the pH from 5.8 to ~7.0 to represent a 173 metal remediation scenario, 30 mg (+/-1.0 mg) of undissolved hydrated lime (Verve 174 Garden lime, Eastleigh, U.K. (18)) was added to each relevant microcosm, then left 175 for 14 days to equilibrate. To observe differences between structured and non-176 structured environments, microcosms were either kept static or were continuously 177 shaken at 210 rpm (Stuart orbital incubator S1600, Staffordshire, UK). Shaking began 178 on day 14 and ended on day 28 (Fig.1). 179 Microcosms were destructively sampled on day 28. Six replicates were used for all 184 treatments (24 microcosms in total). 185 On day 14, 30 µL (7.3 x 10 8 colony forming units: cfu) of Pseudomonas aeruginosa 186 (PAO1 R lacZ: (69)) was added to each microcosm. This lab-strain is both lacZ marked 187 and gentamicin resistant allowing it to be easily distinguished from the rest of the 188 community on agar containing X-gal (5-bromo-4-chloro-3-indolyl-β-D-189 galactopyranoside; 100 µg/L; VWR Chemicals) and gentamicin (30 µg/mL Sigma). P. 190 aeruginosa was grown overnight in shaking microcosms containing 6 mL of King's 191 medium B (KB; 10g glycerol, 20g proteose peptone no. 3, 1.5 g K2HPO4,1.5 g MgSO4,192 per litre). To remove residual nutrients, cultures were centrifuged at 3500 rpm (1233 193 g) for 30 minutes, after which supernatant was decanted, and the pellet resuspended 194 in half the volume of M9 salt buffer (3 g KH2PO4, 6 g Na2HPO4, 5 g NaCl/L) followed 195 by plating on KB agar to calculate the inoculation density. On day 28, all microcosms 196 were destructively sampled by adding sterile glass beads and 12 mL of M9 buffer and 197 vortexing for one minute. Samples were then aliquoted and stored in glycerol (25% 198 final volume) at -80 o C. 199

Iron analysis (ferrozine assay) 200
To determine if liming affected Fe speciation and therefore bioavailability, a ferrozine 201 assay was used to measure relative concentrations of Fe 2+ and total bioavailable iron 202 (70, 71). The first step of this assay quantifies Fe 2+ which is easily obtainable by 203 bacteria and so does not require siderophores. The second step quantifies both Fe 2+ 204 and Fe 3+ and therefore gives a measure of total bioavailable iron including that which 205 requires scavenging mechanisms, such as siderophores. By dividing the first 206 measurement by the second it is possible to estimate the proportion of iron in each 207 treatment that is of relatively high bioavailability to P. aeruginosa (70, 71). The first 208 measurement is given by digesting 100 µL of fresh sample (n=3) in 4.9 mL of 0.5 M 209 hydrochloric acid for 1 hour before 50 µL was mixed with 2.45 mL of ferrozine solution 210 (1 g ferrozine, 11.96 g (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid/L; adjusted 211 to pH 7) in a cuvette (n = 3 per replicate). This was left to stand for exactly one minute 212 before absorbance at 562 nm was measured using a spectrophotometer (Jenway 213 7315, Essex, UK). To quantify total bioavailable Fe (step 2), 200 µL of 6.25 M 214 hydroxylamine hydrochloride was added to the digested samples and left to stand for 215 another hour. This was then added to ferrozine solution in cuvettes and measured as 216 before. Standards of known concentrations of FeSO4.7H2O were measured to allow 217 conversion of absorbance to Fe concentrations. 218

219
Copper growth assay 220 To confirm that metal concentrations in our river water and sediment samples were 221 sufficiently high to select for metal resistance mechanisms, and to test whether liming 222 impacted this selection, we used a copper growth assay. Specifically, we added 20 µL 223 of either the ancestral P. aeruginosa strain or defrosted samples of the evolved 224 populations to a 96-well plate well containing 180 µL of plain Iso-Sensitest broth 225 (Oxoid) and 20 µL to a well containing 180 µL of Iso-Sensitest broth at a concentration 226 of 1 g/L of copper sulphate (CuSO4; Alfa Aesar, Massachusetts, United States). The 227 optical density OD600 was then read every 10 minutes for 18 hours using a Biotek 228 Synergy 2 spectrophotometer. We used 1 g/L of copper sulphate as this equates to a 229 copper concentration (6.26 mM) previously found in highly polluted environments (72, 230 73). 231 232 233

Siderophore (CAS) assay 234
Total siderophore production was quantified using the Chrome Azurol S (CAS) assay 235 (74). Samples were plated onto tryptic soy agar (TSA: Oxoid) supplemented with 236 nystatin (Sigma: 20 μg/mL) to suppress fungal growth and X-gal. After 48 hours, P. 237 aeruginosa colonies were counted to quantify density, before 24 colonies per replicate 238 were randomly picked using sterile toothpicks. Selected colonies were resuspended 239 in 1 mL of KB media in a deep 96-well plate and grown overnight at 28 °C. These were The insect infection model Galleria mellonella was used to quantify P. aeruginosa 255 virulence (59, 60). Defrosted freezer stocks containing the whole sample microbiome 256 were diluted 100-fold using M9 salt buffer, before 10 µL was injected into twenty final 257 instar larvae per replicate using a 50 µL syringe (Hamilton, Nevada, USA). Injected 258 larvae were incubated at 37 °C and mortality was monitored hourly after 13 hours for 259 12 hours with a final check at 42 hours. Larvae were classed as dead when 260 mechanical stimulation of the head caused no response (60). M9-injected and non-261 injected controls were used to confirm mortality was not due to injection trauma or 262 background G. mellonella mortality; >10% control death was the threshold for re-263 injecting (no occurrences). Prior to assays on microcosms containing P. aeruginosa, 264 we confirmed the natural microbial community caused zero mortality by injecting 265 replicates not inoculated with P. aeruginosa as described above. 266 267

Antibiotic resistance assay 268
To test the evolved resistance of P. aeruginosa to the clinically relevant antibiotics 269 apramycin, cefotaxime and trimethoprim, we used the same P. aeruginosa colonies 270 isolated for the siderophore analysis. We first determined the minimum inhibitory 271 concentration of the three antibiotics for our ancestral strain, by growing the ancestral 272 strain for 24 hours (as described above), and plating it on TSA containing a range of 273 concentrations of the antibiotics that increased in 10 µg/mL increments from 0 -60 274 µg/mL. The minimum inhibitory concentrations were found to be 12 µg/mL, 30 µg/mL 275 and 40 µg/mL for apramycin, cefotaxime and trimethoprim, respectively. Next, the 276 individual evolved clones were defrosted before 2 µL of each was plated onto either

Statistical analysis 284
The effect of liming and shaking, plus their interaction, on the final pH, density of P. 285 aeruginosa (log10(cfu mL -1 )), and the proportion of total bioavailable iron (Fe 2+ + Fe 3+ ) 286 that was Fe 2+ (Fe 2+ / (Fe 2+ + Fe 3+ )) was tested using linear models with liming and 287 shaking as explanatory variables. In general, model reduction was carried out by 288 sequentially removing terms from the full model and comparing model fits using F-289 tests; we report parameter estimates of the most parsimonious model. The effect of 290 pH on the density of P. aeruginosa populations was tested using a linear model with 291 density (cfu mL -1 ) log10 transformed. 292 To test whether evolved samples had greater resistance to copper than the ancestral 293 strain, we first calculated the relative fitness, w, of each population by dividing its 294 maximum optical density after 18 hours when grown with copper (ODmaxC) by its 295 maximum optical density when grown without copper (ODmaxWC), i.e. w = ODmaxC / 296 ODmaxP. We then carried out a one way ANOVA with w as the response variable and 297 treatment (including ancestor) as the explanatory variable. Secondly, we carried out a 298 Dunnett's test, using the 'DescTools' R package (76), to test whether each treatment 299 differed to the ancestor. Finally, we tested the effect of liming and shaking on the metal 300 resistance of the final populations in a linear model, with log(w) as the response 301 variable, and liming, shaking and their interaction as the explanatory variables. In all 302 tests w was log transformed to normalise the residuals 303 To test liming and shaking effects on total siderophore and pyoverdine production, 304 linear mixed effects models (LMEM) were carried out using the 'lme4' package (77)  305 with liming and shaking as explanatory variables, and random intercepts fitted for each 306 replicate to control for multiple clones being sampled from the same microcosm. For 307 these LMEMs, we used the 'DHARMa' package (78) to check residual behaviour, after 308 which the most parsimonious model was arrived at by comparing models with and 309 without the liming-shaking interaction using -tests. Two samples had pyoverdine 310 values much lower than the rest, so a Grubbs test ('outliers' package; (79)) was used 311 to check if they were significant outliers. They were therefore were removed from this 312 and all further models to improve model fit. To test the association between copper 313 resistance and both total siderophore and pyoverdine production, we carried out two 314 linear models with log(w) as the dependent variable and either mean total siderophore 315 per microcosm or mean pyoverdine production per microcosm as the explanatory 316

variable. 317
Virulence was analysed in three separate models. First, we tested whether larvae that 318 died before 42 hours were injected with samples containing more siderophores and 319 pyoverdine than those that remained alive after 42 hours. This was done by carrying 320 out two separate binomial generalised linear mixed models (GLMM) using the 'lme4' 321 package (77), with number of G. mellonella dead versus alive as the binomial 322 response variable, and either the production of total siderophores or pyoverdine as 323 the explanatory variable. In this model pyoverdine production was log10 transformed 324 to normalise residuals. Secondly, we tested whether the mean time it took deceased 325 larvae (20 per replicate) to die was associated with total siderophore and pyoverdine 326 production (both values taken from the mean of 24 clones) using a linear model. 327 Finally, we tested whether virulence differed between treatments. To do this, survival 328 curves were fitted using Bayesian regression in the R package 'rstanarm' (80) and the 329 package 'tidybayes' (81) was used to estimate parameters. A proportional hazards 330 model with an M-splines baseline hazard was fitted, with liming, shaking plus their 331 interaction as fixed effects. We additionally included random intercepts for each 332 sample to control for multiple (20) G. mellonella being inoculated with the same 333 sample. Models used three chains with uninformative priors and were run for 3000 334 iterations. Model convergence was assessed using Rhat values (all values were 1), 335 before we manually checked chain mixing. 336 The proportion of apramycin, cefotaxime, and trimethoprim resistance in each 337 treatment (number of resistant colonies out of 24 in total) was compared using Kruskal-338 Wallace non-parametric tests, with resistance proportion as the response variable and 339 treatment as the explanatory variable. All analyses were carried out in R version 3. Here, we tested whether liming of metal-contaminated aquatic environments 346 decreases co-selection for virulence and antibiotic resistance in the opportunistic 347 pathogen P. aeruginosa. To do this, we evolved P. aeruginosa with or without lime in 348 microcosms containing a mixture of metal contaminated river water and sediment in 349 the presence of the natural microbial community. We employed both shaking and 350 static microcosms to represent turbulent and stagnant aquatic environments, in order 351 to test whether liming effects were dependent on environmental structure (Fig. 1). 352 As expected, liming significantly decreased the acidity of sediment and water from the 353 initial pH of 5.8. However, the extent of this effect was significantly greater in the 354 shaking treatments (liming-shaking interaction: F1,20=23.1, p<0.001; Fig. 2), likely due 355 to increased mixing of lime and oxygen throughout the microcosms. The shaken-limed 356 treatment reached a pH of 7.2 (+/-0.11 SD) whereas the static-limed treatment 357 reached a pH of 6.7 (+/-0.25 SD). Both non-limed treatments had a final pH of 5.7 (+/-358 0.19 SD). As pH is often a good predictor of iron speciation (83), we tested how the 359 treatments affected the relative proportions of Fe 2+ and Fe 3+ . We found the proportion 360 of more bioavailable Fe 2+ to not significantly differ as a result of liming, shaking, nor 361 their interaction (lime main effect: F1,9=3.47, p=0.10; shaking main effect: F1,9=3.00, 362 p=0.12; lime-shaking interaction F1,8=0.73, p=0.42; Fig.2), with Fe 2+ making up 82% of 363 the total iron available on average across the treatments. Given that iron speciation 364 remained similar in all treatments, this indicates that the redox potential within the 365 microcosms did not change to become more anaerobic under static conditions (83). 366 Hence iron bioavailability was not significantly influenced by the different experimental 367 conditions and therefore iron limitation was unlikely to represent a significant driver for 368 siderophore production. 369 370 371 Figure 2 The final pH of microcosms containing river water and sediment after 28 days 372 incubation. We used a factorial design with limed and shaken treatments, each with 373 six replicates (each represented by a white circle). The starting pH was 5.8. The 374 significant effect of liming on pH (p<0.001) was increased through an interaction with 375 shaking (p<0.001). 376 377

P. aeruginosa populations incubated without lime had greater tolerance to copper 378
In order to test whether our river water and sediment samples selected for greater 379 metal resistance, we incubated the ancestral P. aeruginosa strain and final 380 populations in a medium containing a high concentration of copper (1 g/mL of copper 381 sulphate). We then compared the maximum optical density of each culture relative to 382 that of cultures grown without copper (w). Confirming that our samples contained toxic 383 metals, the ancestral strain had a lower relative fitness (w) when grown with copper 384 than all final populations (Dunnett's test: p=<0.013 for all contrasts; Fig. 3). Moreover, 385 when comparing the effect of the different treatments on w, we found populations from 386 the non-limed treatments to have greater relative fitness in a toxic copper environment 387 than those from the limed treatments (liming main effect: F1,21=4.44, p=0.047; Fig. 3 Neither liming nor shaking affected P. aeruginosa density or siderophore production 404 Next, we tested the treatment effects on P. aeruginosa density and siderophore 405 production. The final density of P. aeruginosa after two weeks of evolution varied 406 substantially between samples (1.1 x 10 6 ± 1.6 x 10 6 SD cfu/mL), but was not 407 significantly affected by liming, shaking, nor their interaction (liming main effect: 408 F1,21=1.96, p=0.18; shaking main effect: F1,21=2.77, p=0.11; liming-shaking interaction: 409 F1,20=0.70, p=0.41). There was also no significant effect of pH on P. aeruginosa 410 density (F1,22=0.97, p=0.36). Although pH can affect bacterial density (84), our finding 411 of no effect is consistent with previous results demonstrating that P. aeruginosa 412 densities are similar across an equivalent pH range as used here (85). 413

414
To test whether liming and shaking affected siderophore production, both total 415 siderophore production and the production of pyoverdinethe primary siderophore 416 produced by P. aeruginosa (56)were measured for 24 clones per replicate (24 x 417 24 clones). Quantifying pyoverdine production in addition to total siderophores is 418 important, as it is a key virulence factor in P. aeruginosa but its production does not 419 necessarily correlate with that of other siderophores, such as pyochelin (62). We found 420 neither liming, shaking nor their interaction significantly affected mean total 421 siderophore production (liming main effect:

425
However, we note that there was a large variation in production between the 24 clones 426 used to represent each microcosm (mean production: total siderophores = 4.23; 427 pyoverdine = 766; replicate: total siderophores = 1.94; pyoverdine = 69.3), and that two 428 pyoverdine values were significant outliers and consequently were removed from all 429 further analysis in order for model assumptions to be met (these were one from the 430 non-limed shaken treatment (pyoverdine production = 26.9, p<0.001) and one from 431 the limed-static treatment (pyoverdine production = 174, p<0.001); which were lower 432 than the pre-removed mean pyoverdine production of 710.6 and median of 789.8). 433 434 That siderophore production, which is regulated by iron availability and the presence 435 of toxic metals, did not significantly differ between treatments concurs with the non-436 significant differences in Fe 2+ availability between treatments. However, it is surprising 437 that siderophore production was not reduced by liming, given that P. aeruginosa 438 populations from the limed treatments were less tolerant to toxic copper. To explore 439 this further, we tested whether either total siderophore or pyoverdine production was 440 associated with copper tolerance, and found neither of them to be (Total siderophores: 441 F1,20=0.013, p=0.91; Pyoverdine: F1,20=0.294, p=0.59). This suggests other metal 442 resistance mechanisms, such as decreased outer membrane permeability and 443 increased induction of ATPase efflux transporters, could be responsible for the 444 increased copper tolerance of evolved populations (86). Our finding of no significant 445 differences in siderophore production contrasts with that of Hesse and co-workers 446 (18), who found that the addition of lime to soils collected in the near vicinity of our 447 locality significantly reduced community-wide siderophore production. This difference 448 is most likely due to shifts in siderophore production driven by changes in community 449 composition with liming selecting for non-producing isolates (18), whereas here we 450 solely focused on siderophore production by P. aeruginosa. This suggests that 451 although liming reduces community-wide siderophore production in metal-452 contaminated acidic soils, this effect may not be seen in specific species. Interestingly, 453 P. aeruginosa has been proposed as a suitable siderophore-producing bacterium for 454 use in phytoremediation, which relies on the combined of use of microorganisms and 455 plants to aid toxic metal remediation (87, 88). It has been proposed that liming, by 456 reducing siderophore production, may hinder phytoremediation (18) as metal-uptake 457 by plants is often increased when metals are bound to bacterial siderophores. Given 458 that no significant effect of liming on siderophore production by P. aeruginosa, was 459 observed, we suggest that liming and P. aeruginosa-assisted phytoremediation could 460 be used simultaneously without compromise. Virulence did not differ between treatments, but was positively associated with 465 siderophore production 466 As we found a large variation in siderophore production, which is a known virulence 467 factor in P. aeruginosa (64), we tested whether virulence, quantified using the G. 468 mellonella infection assay, differed as a consequence of pyoverdine production, total 469 siderophore production or treatment. Firstly, we tested whether G. mellonella larvae 470 alive at the final time check (42 hours) had been injected with populations producing 471 less total siderophores and pyoverdine compared with larvae that died before this 472 point, and found that they were (total siderophores:  2 =6.11, d.f.=1, p=0.013; 473 pyoverdine:  2 =6.98, d.f.=1, p=0.004). Next, we tested whether increased siderophore 474 and pyoverdine production resulted in increased virulence. We found a significant 475 positive association between virulence (mean time to death per population) and both 476 total siderophore and pyoverdine production (total siderophores: F1,22=8.9, p=0.007; 477 Finally, virulence was compared between treatments using survival curves (Fig. 4C). 485 Virulence did not significantly differ as a function of treatment, with the credible 486 intervals for liming, shaking and their interaction all crossing 1. No significant treatment 487 effect on virulence is concurrent with the finding that the treatments did not significantly 488 affect siderophore production. Finding virulence to not be significantly different 489 between structured (static) and unstructured (shaking) environments contrasts with 490 findings by Granato and co-workers (92), who found that pyoverdine-mediated 491 virulence in P. aeruginosa was greater when grown in solid media than in liquid. The 492 lack of changes detected in siderophore production and virulence between the 493 experimental treatments might be due to the more subtle (and arguably more realistic) 494 conditions under which spatial structure was varied in our study, as well as the 495 presence of a resident microbial community. 496 aquatic communities as a function of (A) mean total siderophore production and (B) 500 mean pyoverdine production. Virulence was quantified using the Galleria mellonella 501 infection model (n = 20 per replicate) and given as the mean time to death. Pyoverdine 502 and total siderophore production were measured in standardised fluorescence units 503 per OD600. Individual circles show the mean production by 24 clones from each 504 replicate. Colours and shapes represent different treatments: grey and □ = static, no 505 lime, blue and + = static, limed, black and △ = shaken, no lime, and red and ✕ = 506 shaken, limed. Panel C shows the change in survival probability of larvae over time 507 within each treatment. These do not significantly differ from one another. Shaded 508 areas represent 95% confidence intervals. 509

Antibiotic resistance evolution 510
As metal pollution has been shown to co-select for antimicrobial resistance (38), we 511 tested whether lime addition altered P. aeruginosa resistance to the clinically relevant 512 antibiotics apramycin (15 µg/mL), cefotaxime (50 µg/mL) and trimethoprim (60 µg/mL) 513 after evolution in metal-contaminated river sediments. Increased resistance was 514 observed in all treatments (Fig. 5), with neither lime nor shaking affecting resistance 515 to any of the antibiotics tested (apramycin: chi-squared=2.35 p=0.50 df=3; cefotaxime: 516 chi-squared=2.98 p=0.40 df=3; trimethoprim: chi-squared=5.25 p=0.16 df=3; Fig. 5). 517 Of note, one sample from the shaken, non-limed treatment consistently had the lowest 518 resistance to all three antibiotics, with no isolates from this population being resistant 519 to cefotaxime or trimethoprim, and fewer than 50% being resistant to apramycin. Our 520 observation of rapid evolution of antibiotic resistance in the other replicates and 521 treatments supports existing evidence that metal contamination can pose an important 522 co-selective pressure for resistance (44,49,93), including in P. aeruginosa (42). That 523 resistance did not differ significantly between treatments in our experiment 524 demonstrates that liming to pH ~7 is not effective at remediating this co-selective 525 effect, and neither was the loss of spatial structure via shaking. A plausible reason for 526 this is that liming reduces metal bioavailability by precipitating ions from solution into 527 the solid phase. This would mean cells in the sediment (the vast majority of the 528 population) would still be exposed to metals where, although at a lower bioavailability, 529 they can still be a cause of co-selection (48). This is supported by P. aeruginosa 530 evolving greater tolerance to copper in the limed treatments. Although we did not 531 determine the mechanistic basis of co-selection, we note that cross-resistance, co-532 resistance and co-regulation mechanisms have all been reported for Pseudomonads, 533 and the altering of cellular targets is a mechanism commonly used by P. aeruginosa 534 to tolerate metal, trimethoprim and beta-lactam antibiotics such as cefotaxime (42, 535 94). We are aware of a single study testing the effects of liming on antimicrobial 536 resistance {Ramos, 1987 #4}. This study found liming decreased the susceptibility of 537 Rhizobium species from soil to multiple antibiotics, and hypothesised this was due to 538 a greater production of natural antibiotics at near-neutral pH selecting for resistance. 539 Although we note that increasing soil pH will generally decreases the bioavailability of 540 any metals present, the authors {Ramos, 1987 #4} stated that metal effects would not 541 be operative in their study, suggesting no metal contamination was present. (60 µg/mL) antibiotics. Clones were tested after two weeks of evolution in microcosms 547 containing metal-contaminated river water and sediment while embedded in the 548 resident microbial community. Circles show individual replicates; those with a red 549 outline are from the same sample, which is the least resistant to all three antibiotics. 550 551 Conclusion 552 553 P. aeruginosa populations evolved metal resistance after two weeks, and liming 554 reduced this effect. However liming and spatial structure (shaking) were observed to 555 have little effect on P. aeruginosa pathogenic traits. Despite finding a positive 556 association between siderophore production and virulence, neither siderophore 557 production nor virulence systematically differed between treatments, suggesting that 558 liming does not alter the effect of metals on siderophore-mediated virulence in P. 559 aeruginosa. This finding also implies that concurrent use of liming and P. aeruginosa-560 assisted phytoremediation techniques is possible in scenarios were this bacterium can 561 persist in a natural community. Moreover, we found P. aeruginosa rapidly evolved 562 resistance to three clinically relevant antibiotics regardless of treatment. We therefore 563 show that a common metal remediation method did not reduce metal pollution-based 564 co-selection for virulence or antibiotic resistance. Importantly, these findings further 565