L. pneumophila deploys a self-active inhibitor for inter-Legionella competition

The bacterial pathogen Legionella pneumophila alternates between intracellular replication within host eukaryotes and extracellular residence in multi-species biofilms. To persist in the extracellular state, L. pneumophila must withstand competition from neighboring bacteria, including other Legionella. Here, we find that L. pneumophila can exclude other Legionella species from its local environment using a secreted inhibitor: HGA (homogentisic acid), the unstable, redox-active precursor molecule to L. pneumophila’s brown-black pigment. Unexpectedly, we find that L. pneumophila is itself inhibited by HGA secreted from neighboring isogenic strains. This HGA sensitivity is density-dependent and HGA secretion is linked to growth phase, suggesting that production of – and resistance to – this inhibitor are functionally linked. Our genetic approaches further identify lpg1681 as a gene that modulates Legionella susceptibility to HGA. We propose that HGA behaves as an unusual public good, allowing established Legionella communities to locally protect against invasion by low-density competitors.


Introduction 16
Inter-bacterial conflict is ubiquitous in nature, particularly in the dense and highly 17 competitive microenvironments of biofilms (Davey & O'toole, 2000;Foster & Bell, 2012;18 Ghigo & Rendueles, 2015). In these settings, bacteria must battle for space and 19 nutrients while evading antagonism by neighboring cells. One strategy for managing 20 these environments is for bacteria to cooperate with their kin cells, sharing secreted 21 molecules as public goods (Abisado, Benomar, Klaus, Dandekar, & Chandler, 2018;22 Nadell, Drescher, & Foster, 2016). However, these public goods are vulnerable to 23 exploitation by other species or by 'cheater' bacterial strains that benefit from public 24 goods but do not contribute to their production. For this reason, many bacteria 25 participate in both cooperative and antagonistic behaviors to survive in multispecies 26 biofilms. Bacterial antagonistic factors can range from small molecules to large proteins, 27 delivered directly or by diffusion, and can either act on a broad spectrum of bacterial 28 taxa or narrowly target only a few species. Although narrowly targeted mechanisms may 29 seem to be of less utility than those that enable antagonism against diverse bacterial 30 competitors, targeted strategies can be critical for bacterial success because they tend 31 to mediate competition between closely-related organisms that are most likely to 32 overlap in their requirements for restricted nutrients and niches (Hibbing, Fuqua, Parsek, 33 & Peterson, 2010). 34

35
The bacterium Legionella pneumophila (Lp) naturally inhabits nutrient-poor 36 aquatic environments where it undergoes a bi-phasic lifestyle, alternating between 37 replication in host eukaryotes and residence in multi-species biofilms (Declerck, 2010;38 Declerck, Behets, van Hoef, & Ollevier, 2007;Lau & Ashbolt, 2009;Taylor, Ross, & 39 Bentham, 2013). If Lp undergoes this lifecycle within manmade structures such as 40 cooling towers or showers, the bacterium can become aerosolized and cause outbreaks 41 of a severe, pneumonia-like disease in humans, called Legionnaires' disease (Fields, 42 Benson, & Besser, 2002;Fraser et al., 1977;McDade et al., 1977). Because of the 43 serious consequences of Lp colonization, the persistence and growth of Legionella in 44 aquatic environments has been the subject of numerous studies. These studies have 45 examined replication within protozoan hosts (Hoffmann, Harrison, & Hilbi, 2014;Isberg, 46 water under nutrient stress (L. Li, Mendis, Trigui, & Faucher, 2015 potentially via a secreted inhibitor ( Figure 1A). We empirically found that this inhibition 94 was most robust when we plated the Lp strain on low-cysteine media 3-4 days prior to 95 plating Lm, allowing time for the inhibitory molecule to be produced and spread across 96 the plate. Previous work (Stewart et al., 2011) proposed that inter-Legionella inhibition 97 could be caused by Lp's secreted surfactant, which is produced by Lp but not Lm 98 (Stewart et al., 2009). However, we observed that the zone of inhibition surrounding Lp 99 did not always co-occur with the spread of the surfactant front (Supplemental figure 1A), 100 suggesting that Lp may secrete a separate, undescribed inhibitor. To test this 101 hypothesis, we deleted a surfactant biosynthesis gene, bbcB, from the Lp genome 102 (Stewart et al., 2011). The resulting ∆bbcB strain did not produce surfactant 103 (Supplemental figure 1B), yet it still inhibited adjacent Lm, demonstrating that surfactant 104 production is not required for inter-Legionella inhibition ( Figure 1A, Supplemental Figure  105 1C). To quantify this inhibition, we recovered Lm grown at different distances from Lp. 106 After 48h incubation, we found a 10,000-fold difference in growth between Lm 107 antagonized by Lp and those plated outside of the zone of inhibition ( Figure 1B). We 108 next asked if surfactant enhanced inhibition by quantifying inhibition from the ∆bbcB Lp 109 strain on Lm. Again, surfactant production had little impact and we observed nearly 110 identical inhibition from both wild type Lp and ∆bbcB Lp ( Figure 1C). We therefore 111 conclude that L. pneumophila can cause strong growth inhibition of neighboring 112 Legionella using an unknown molecule that is distinct from surfactant. To determine which molecule(s) might be responsible for inter-Legionella 117 inhibition, we performed an unbiased genetic screen in Lp. Briefly, we generated Lp 118 mutants using a drug-marked Mariner transposon that randomly and efficiently 119 integrates into the Legionella genome (O'Connor, Adepoju, Boyd, & Isberg, 2011). To 120 identify mutants that were defective in producing the inhibitor, we transferred each 121 mutant onto a lawn of L. micdadei on low-cysteine plates, and examined the resulting 122 zone of inhibition surrounding each Lp mutant ( Figure 2A). After screening 2870 clones, 123 we isolated 19 mutants that produced a smaller zone of inhibition than wild type Lp, as 125 well as 5 mutants that showed a complete loss of inhibition ( Figure 2B, Supplemental 126 Table 1). We refer to these as "small zone" and "no zone" mutants, respectively. 127

128
We focused on the "no zone" mutants, as these had the strongest defects in 129 inhibition. These 5 mutants carried transposon insertions in two separate operons 130 ( Figure 2C). The first operon had two insertions in the hisC2 gene (lpg1998), which 131 breaks down tyrosine as part of a metabolic pathway called the HGA-melanin pathway 132 ( Figure 2D). Its downstream gene, pphA, converts phenylalanine to tyrosine in the same 133 pathway. To validate the role of hisC2 in inhibition, we overexpressed this gene in the 134 hisC2 transposon mutant background, and found that hisC2 alone was sufficient to 135 complement the mutant phenotype (Supplemental Figure 2A). Having confirmed the 136 role of hisC2, we turned to the second operon, where we had recovered mutations in 137 two uncharacterized genes, lpg2276 and lpg2277 ( Figure 2C). These two genes lie 138 immediately upstream of hpd (lpg2278), which is known to act with hisC2 in the HGA-139 melanin pathway (Gu, Song, Bonner, Xie, & Jensen, 1998;Steinert et al., 2001) (Figure  140 2D). Because transposon insertions at the beginning of an operon can disrupt the 141 expression of downstream genes via polar effects, we hypothesized that the insertions 142 we recovered in lpg2276 and lpg2277 altered inter-Legionella inhibition via disruption of 143 hpd expression. Indeed, we were able to complement insertions in both genes by 144 overexpressing hpd, despite the fact that hpd overexpression caused a growth defect 145 (Supplemental Figure 2A). In conclusion, all five "no zone" isolates had mutations that 146 disrupted the same metabolic pathway. 147

148
The HGA-melanin pathway is found in diverse bacteria and eukaryotes (Liu & 149 Nizet, 2009;Nosanchuk & Casadevall, 2003), where it produces homogentisic acid 150 (HGA) from the catabolism of phenylalanine or tyrosine (Fang, Yu, & Vickers, 1989;151 Steinert et al., 2001) ( Figure 2D). Once made, HGA can either be further metabolized 152 and recycled within the cell via HmgA-C, or it can be secreted outside of the cell, where 153 it auto-oxidizes and polymerizes to form a black-brown pigment called HGA-melanin, or 154 pyomelanin (Kotob, Coon, Quintero, & Weiner, 1995) (Figure 2D). To our knowledge, 155 the HGA-melanin pathway has not previously been implicated in inter-bacterial 156 competition. To test whether intracellular metabolites downstream of HGA are 157 necessary for inhibition, we deleted hmgA, the first gene in the pathway that returns 158 HGA into central metabolism. The ∆hmgA strain produced a zone of inhibition that was 159 similar or perhaps slightly larger than wild type (Supplemental Figure 2B), suggesting 160 that the intracellular processing of HGA is not important for inhibition. Furthermore, we 161 observed defects in HGA-melanin pigmentation in all of the "no zone" mutants as well 162 as some of the "small zone" mutants (Supplemental Figure 3A). We therefore inferred 163 that synthesis of secreted HGA and/or HGA-melanin is required for Lp inhibition of Lm. light (Steinert, 186 Engelhard, Flügel, Wintermeyer, & Hacker, 1995). We therefore asked whether the 187 active inhibitor produced by the pathway was HGA-melanin, or alternatively if it could be 188 a transient precursor molecule ( Figure 3A). In repeated experiments testing the activity 189 of HGA-melanin pigment from Lp conditioned media, we never observed any inhibition 190 of Lm. We hypothesized that perhaps the pigment secreted into rich media was too 191 dilute to be active, or alternatively that other nutrients in the media might interfere with 192 inhibition. To address these concerns, we isolated a crude extract of HGA-melanin from 193 Lp conditioned media via acid precipitation (as in (Chatfield & Cianciotto, 2007)), 194 washing and concentrating the pigment approximately 10-fold. In multiple assays, the 195 concentrated pigment also showed no inhibitory activity ( Figure 3B The first metabolite secreted by the HGA-melanin pathway is HGA, which auto-199 oxidizes and polymerizes to form HGA-melanin (Steinert et al., 2001). In contrast to 200 HGA-melanin, synthetic HGA robustly inhibited Lm growth, both when spotted onto a 201 lawn of Lm and when titrated into AYE rich media ( Figure  We next asked which molecular features of HGA were relevant for its inhibitory 217 activity. Two HGA-related molecules-2-hydroxyphenylacetic acid and 3-218 hydroxyphenylacetic acid-differ from HGA by only the removal of a single -OH group. 219 Neither compound inhibited Lm growth at any concentration tested (Supplemental figure 220 6), suggesting that inhibition of Lm by HGA is relatively specific at the molecular level. 221 We also considered the possibility that HGA as a weak acid could inhibit Lm indirectly 222 by altering the local pH, but we observed that adding HGA at 1mM into AYE media or 223 PBS caused little to no change in pH. 224

225
Another well-studied feature of HGA is its redox potential (Eslami, Namazian, & 226 Zare, 2013). If HGA can auto-oxidize, it may cause growth inhibition by oxidizing other 227 nearby molecules, either in the media or on bacterial cells. This scenario is consistent 228 with our observation that oxidized HGA-melanin is inactive ( Figure 3B, Supplemental Figure 4A). To test the specific hypothesis that HGA oxidizes and depletes nutrients in 230 the media, we allowed synthetic HGA to oxidize completely for 24h in AYE media before 231 adding Lm (Supplemental Figure 5A). If HGA acts by causing nutrient depletion, we 232 expected that media pre-incubated with HGA would be unable to support normal Lm 233 growth. Contrary to this hypothesis, we observed that Lm grew normally in all conditions 234 and that the pre-oxidation completely abolished synthetic HGA activity ( Figure 3D, 235 compare to 3C). We can therefore reject the hypothesis that HGA activity results from 236 indirect nutrient limitation, or any other modifications of the media. Instead, we infer that 237 Lm inhibition results from direct interactions between bacterial cells and either HGA 238 itself or unstable, reactive intermediates produced during HGA oxidation. 239

240
If the redox state of HGA is critical for inhibition, we reasoned that it should be 241 possible to modulate HGA activity by altering the redox state of the media with reducing 242 agents. We accomplished this by titrating L-cysteine from 25% to 200% of the levels in 243 standard AYE media. In the absence of HGA, these altered cysteine concentrations had 244 little impact on Lm growth (Supplemental Figure 4B). However, lower levels of cysteine 245 greatly sensitized Lm to HGA, while excess cysteine was completely protective 246 (Supplemental Figure 4B). In related experiments, incubation of HGA with reduced 247 glutathione or DTT (dithiothreitol), two other reducing agents, similarly quenched HGA's 248 inhibitory activity (Supplementary Figure 4C). We conclude that HGA is less potent in 249 rich media because it reacts with and oxidizes the excess cysteine (or other bystander 250 molecules) before it can interact with Lm. Taken together, these results implicate HGA's 251 oxidizing activity and/or its oxidative intermediates in inter-Legionella inhibition. 252

L. pneumophila can be susceptible to its own inhibitor 254 255
Our results so far indicated that, while HGA can be a potent inhibitor, its activity 256 appears to be volatile and capable of reacting with many types of molecules. If Lp uses 257 HGA to compete with neighboring Legionella spp., we hypothesized that Lp would have 258 some form of resistance to its secreted inhibitor. Therefore, we next tested Lp 259 susceptibility to inhibition in low-cysteine conditions, as we had previously done for Lm. 260 Surprisingly, we found that Lp was quite sensitive to inhibition by neighboring Lp that was already growing on the plate ( Figure 4A). Indeed, Lp susceptibility closely mirrored 262 the susceptibility of Lm to inhibition (compare to Figure 1A), even though the bacteria 263 secreting the inhibitor were genetically identical to the inhibited Lp. In both cases, we 264 observed a sharp boundary at the edge of the zone of inhibition. In contrast, the "no HGA was restricted to low-cysteine conditions, we grew Lp in the presence of synthetic 269 HGA in rich media. Again, we found that the HGA was inhibitory, causing a growth 270 delay in liquid cultures ( Figure 4B). Notably, synthetic HGA was active against Lp at the 271 same concentrations that were inhibitory to Lm (compare Figure 4B and 3C). These 272 results are consistent with a potential role for HGA in both interspecies and intraspecies 273 Legionella inhibition. 274

275
We initially found these results to be somewhat perplexing, as we had expected 276 Lp to show some resistance to its own inhibitor. Why should Lp secrete large quantities 277 of HGA and potentially inhibit its own growth if this process is non-essential (as 278 evidenced by the robust growth of the hisC2::Tn mutant)? Although we had already 279 estimated that wild type Lp can secrete enough HGA to generate abundant pigment 280 (equivalent to 1.7mM, Supplemental Figure 5B), we did not know if this HGA was 281 secreted all at once, or alternatively if it was released slowly over time. Under the latter 282 scenario, the HGA might oxidize before it accumulated to inhibitory concentrations. To 283 differentiate between these possibilities, we tracked HGA secretion across a growth 284 curve of Lp in rich media, using HGA-melanin levels in fully oxidized media to infer the 285 amount of secreted HGA. It has long been known that Lp produces abundant HGA-286 melanin pigment in stationary phase, when the bacteria are undergoing very slow or no 287 growth ( Figure 4C) (Berg, Hoff, Roberts, & Matin, 1985;Pine, George, Reeves, & 288 Harrell, 1979;Wiater, Sadosky, & Shuman, 1994). By comparing to a synthetic HGA 289 standard curve (Supplemental Figure 5), we estimate that Lp produces a burst of HGA 290 in stationary phase, secreting over 250µM within 5 hours ( Figure 4C). HGA secretion 291 then continues after the population has ceased growing. These quantities of HGA 292 should be more than enough to be inhibitory to Lp ( Figure 4B). 293 294 Based on the timing of HGA secretion, we hypothesized that Lp might be more 295 resistant to this inhibitor when at high density and/or when the cells are not actively 296 growing. To test this hypothesis, we washed and diluted stationary-phase Lp in nutrient-297 free PBS with 125µM HGA and measured viability by CFUs at 0 and 24 hours. No 298 measurable darkening of the HGA was detected in this assay, suggesting that the 299 oxidation and de-activation of HGA was considerably slowed in low-nutrient conditions. 300 We found that Lp incubated at high density with HGA (above 10^8 CFU/mL) were 301 largely protected from inhibition ( Figure 4D). In contrast, when we diluted the same 302 culture to lower density (below 10^7 CFU/mL), bacteria were extremely sensitive to 303 HGA, with at least a 100,000-fold reduction in CFUs. Similarly, HGA did not inhibit Lp 304 growth rich media when bacteria were inoculated at high density (Supplemental figure 305 7). Based on the substantial, density-dependent difference in HGA susceptibility, we 306 wondered if Lp might regulate HGA resistance via quorum sensing. However, when we 307 deleted lqsR, the putative quorum sensing response regulator from Lp (Tiaden et al., 308 2007), this had no detectable impact on HGA susceptibility (Supplemental figure 7). 309 Therefore, the density-dependent susceptibility of Lp to HGA must be independent of 310 this pathway. 311 Taken together, these results show that Lp can be highly susceptible to inhibition 313 by adjacent, genetically identical bacteria via HGA. Similar to our findings in Lm, the 314 potency of HGA-mediated inhibition is exacerbated in low nutrient conditions. We also 315 observe that Lp secretes abundant HGA when in stationary phase and at high density. 316 Because Lp at high density also appear to be protected from HGA, this strategy may 317 restrict the potential for self-harm. 318 319 Non-essential gene lpg1681 sensitizes L. pneumophila to HGA 320

321
Our results suggested that Lp bacteria are sensitive to HGA at low density but 322 resistant at high density. We next investigated the genetic basis of HGA susceptibility 323 and resistance in Lp. First, we hypothesized that the HmgA-C proteins, which break 324 down intracellular HGA and recycle it back into central metabolism, might also process 325 and deactivate extracellular HGA (Rodriguez-Rojas et al., 2009) ( Figure 2D). In growth 326 curves of the ∆hmgA mutant with increasing concentrations of synthetic HGA, we found 327 that its response to HGA was nearly identical to that of wild type Lp (Supplemental 328 Figure 8). These results suggest that the intracellular recycling pathway does not play 329 an appreciable role in Lp responses to HGA. 330 Having excluded the obvious candidate pathway for HGA resistance, we pursued 332 an unbiased genetic approach. Because HGA is strongly inhibitory to low density 333 bacteria, we performed a selection for spontaneous, HGA-resistant mutants of Lp and 334 Lm using a high HGA concentration that normally prevents almost all growth for both 335 species. To prevent HGA 336 from reacting with media 337 components and becoming 338 inactive (as in Figure 3D,  lpg1681. A) Growth of HGA-selected spontaneous mutants (*) compared to wild type Lp. All isolates grew better than wild type in selection conditions (HGA + low cysteine), as well in low cysteine conditions. B) Predicted structure of lpg1681-containing operon in Lp strains. lpg1681 is a hypothetical gene that lies downstream of lpg1682, a predicted oxidoreductase/dehydrogenase, and upstream of dsbD2, a thiol:disulfide interchange protein. C) In rich media, a spontaneous lpg1681 mutant (lpg1681*) and the lpg1681 deletion strain (∆lpg1681) are less sensitive to growth inhibition by HGA than wild type Lp, as seen by a shorter growth delay at each concentration of HGA. Overexpression of lpg1681 (OE) heightens sensitivity to HGA (longer growth delay).  better than wild type Lp on low-cysteine plates without HGA, raising the possibility that 363 HGA susceptibility is related to cysteine metabolism and/or that the strains had adapted 364 to low-cysteine conditions. 365 366

368
To determine the underlying genetic basis of these phenotypes, we sequenced 369 the genomes of all 29 strains plus our wild type strain of Lp to a median depth of 118x 370 and identified mutations genome-wide. Each mutant strain carried 1 to 3 unique point 371 mutations relative to the starting strain, and most of these mutations were found in a few 372 shared loci (Table 1). Nineteen of the twenty-nine strains carried mutations in 373 translation-related machinery, 5 had missense mutations in secY or secD (lpg0349 and 374 lpg2001), members of the translocon complex that moves polypeptides across the 375 cytosolic membrane, 1 had a mutation in aceE pyruvate dehydrogenase (lpg1504), and 376 4 had mutations in a hypothetical gene, lpg1681 ( Figure 5B). We suspect that the 377 translation-related mutations had highly pleiotropic effects on protein expression by 378 disrupting elongation factor P function and thereby enhancing ribosome stalling at 379 polyproline tracts (Doerfel et al., 2013 frequency of polyprolines in the Lp proteome, disruptions to elongation factor P function 382 have the potential to impact the expression of about 33% of Lp proteins. The 383 phenotypes observed in these mutants could therefore result from either a large scale 384 shift in Lp gene expression, or from the altered expression of specific susceptibility 385

genes. 386
Given that all the HGA-selected mutants grew better than wild type Lp on both 388 low-cysteine plates and on low-cysteine plates with HGA overlays, we asked whether 389 the mutants had specific resistance to HGA, if they had primarily adapted to low-390 cysteine conditions, or if they had pleiotropic mutations that broadly impacted cell 391 growth. Growth curves of representative HGA-selected mutants with or without HGA in 392 cysteine-rich AYE media showed that lpg0288 and aceE mutants exhibited some 393 growth defects in cysteine-rich media, consistent with potential pleiotropic impacts on 394 the cell (Supplemental Figure 8E). Mutants in lpg1681, secD, and secY had no growth 395 defects and also showed a reproducible decrease in HGA susceptibility ( Figure 5C, 396 Supplemental Figure 8E). Of these, lpg1681 mutants showed the largest improvement 397 in growth in the presence of HGA. 398

399
The hypothetical gene lpg1681 is predicted to encode a small, 105 amino acid 400 protein with no predicted domains apart from two transmembrane helices. This gene is 401 adjacent in the genome to lpg1682, a predicted oxidoreductase/dehydrogenase, and 402 lpg1680, the thiol:disulfide exchange protein DsbD2 ( Figure 5B). Functional studies of 403 DsbD2 (aka DiSulfide Bond reductase D2) have demonstrated that it interacts with 404 thioredoxin to regulate disulfide bond remodeling in the periplasm (Inaba, 2009;Kpadeh, 405 Day, Mills, & Hoffman, 2015). If lpg1681 has a redox function related to its neighboring 406 genes, we expected its syntenic locus to be conserved across bacterial strains and 407 species. We found that the lpg1680-1682 locus is present and conserved among over 408 of lpg1681 is also found in the draft genome of Piscirickettsia litoralis, a gamma proteobacterium and fish pathogen (Wan et al., 2016). In all cases, lpg1681 resides 413 upstream of dsbD2, suggesting a functional link between these proteins (Supplemental 414 Figure 9). 415 416 Because lpg1681 appears to reside within an operon with redox-related genes 417 and HGA activity is linked to its redox potential, we viewed lpg1681 as a promising 418 candidate for a gene involved in HGA susceptibility. We constructed lpg1681 419 overexpression and deletion strains in Lp and tested the susceptibility of these strains to 420 HGA. Similar to the spontaneous mutants we recovered, the ∆lpg1681 strain was more 421 resistant to HGA in rich media ( Figure 5C). Conversely, overexpression of lpg1681 422 increased Lp sensitivity, resulting in longer growth delays than wild type at high 423 concentrations of HGA. We therefore conclude that wild type lpg1681 expression 424 sensitizes Lp to inhibition by extracellular HGA. 425 426 427

Discussion 428
In this study, we identify HGA as a mediator of inter-Legionella inhibition. 429 Furthermore, we find that inter-Legionella inhibition and surfactant production are 430 separate, but co-occurring, phenomena. Although surfactant is not required for 431 inhibition, we hypothesize that surfactant may enhance the spread of HGA and/or its 432 reactive intermediates away from the secreting Lp bacteria, which may explain previous 433 results (Stewart et al., 2011). 434

435
The HGA-melanin pathway is well-studied and widespread among bacteria and 436 eukaryotes, so we were surprised to find a previously undescribed role for HGA in 437 interbacterial competition. Somewhat paradoxically, HGA-melanin production in bacteria 438 has previously been implicated both in the production of and protection from reactive 439 oxygen species (production: (Noorian et al., 2017); protection: (Keith, Killip, He, Moran, 440 & Valvano, 2007;Orlandi, Bolognese, Chiodaroli, Tolker-Nielsen, & Barbieri, 2015)). 441 Based on our results, we hypothesize that HGA may be responsible for oxidative 442 damage, while the HGA-melanin pigment serves other, beneficial roles. The redox-based inhibition that we observe for HGA is analogous to that of pyocyanin from 444 Pseudomonas aeruginosa and similar phenazine pigments, which are also produced at 445 high cell density (Baron, Terranova, & Rowe, 1989;Hassan & Fridovich, 1980). One 446 major difference is that pyocyanin impacts an extremely broad range of species, 447 including diverse bacteria and eukaryotes (Baron et al., 1989;Kerr et al., 1999;Noto, 448 Burns, Beavers, & Skaar, 2017). In contrast, previous work found that Lp secretions can 449 inhibit Legionella spp., but have little impact on other bacterial taxa (Stewart et al., 450 2011). We speculate that the observed specificity arises simply because Legionella spp. 451 are quite sensitive to oxidative stress relative to other bacteria (Domingue, Tyndall, 452 Mayberry, & Pancorbo, 1988;Hoffman, Pine, & Bell, 1983). Nevertheless, because the 453 ecological niche of Lp appears to overlap with other Legionella strains and species 454 (Pereira et al., 2017;Wery et al., 2008), HGA-mediated inter-Legionella inhibition has a 455 strong potential to impact the success of Lp in both natural and manmade 456 environments. 457

458
Another major difference between phenazines and HGA is that phenazines have 459 not been reported to be self-active, because the phenazine-producing bacteria are 460 protected by their antioxidant defenses (Hassett, Charniga, Bean, Ohman, & Cohen, 461 1992). However, self-active effects can be difficult to observe, particularly if bacteria at 462 high cell density are resistant. In our hands, intraspecies inhibition by HGA was 463 apparent only in assays competing high-density populations with low-density, invading 464 competitors. Furthermore, we note that the inhibitory activity of molecules like HGA is 465 transient and is quenched by the cysteine-rich media typically used to grow Legionella 466 in the lab ( Figure 3D, Supplemental Figure 4). This may explain why HGA-mediated 467 inhibition of Legionella has not been previously reported. Still, our complementary 468 inhibition assays performed in rich media, low-cysteine media, and nutrient-free media 469 suggest that HGA inhibition can occur across environmental settings, and only becomes 470 more potent as conditions approach the nutrient-limited, slow growth conditions that 471 Legionella likely experience within biofilms. 472 In many modes of interbacterial competition, organisms that produce toxic 474 molecules also possess resistance genes to avoid self-targeting. It was therefore initially 475 perplexing that Lp can produce a self-active inhibitor, particularly given that genes for 476 HGA production and the HGA-sensitizing gene lpg1681 are both non-essential under 477 laboratory conditions. One possible explanation is that the benefits of HGA or HGA-478 melanin for Lp outweigh the self-harm of HGA. These benefits could include antagonism 479 against other Legionella as demonstrated here, or previously described benefits of 480 HGA-melanin, including UV protection and iron reduction (Chatfield & Cianciotto, 2007;481 Steinert et al., 1995). Antagonism against eukaryotic cells may be another benefit; the 482 hpd gene in the Lp HGA synthesis pathway was originally named lly, or legiolysin, 483 because expression of the gene in E. coli caused erythrocyte lysis on blood agar plates 484 (Wintermeyer et al., 1994). Finally, the redox activity of HGA may play a beneficial role 485 in Legionella gene expression and/or anaerobic respiration within biofilms, as is the 486 case for pyocyanin and other phenazines in Pseudomonas (Mavrodi et al., 2013). 487 488 Beyond these benefits, the context and timing of HGA secretion may constrain 489 the potential for self-harm. In laboratory conditions, HGA secretion occurs primarily in 490 late stationary phase, when cells are at high density (~10^9 CFU/mL) ( Figure 4C). 491 Because Lp at high density are also more resistant to HGA ( Figure 4D), Lp self-492 protection apparently involves a definition of "self" based on growth phase, density, 493 and/or regulatory state, rather than one based on the presence of genetically-encoded 494 resistance genes. Consistent with this idea, our selection scheme to identify genes 495 involved in HGA resistance ( Figure 5) recovered mostly pleiotropic defects in 496 translation, in addition to lpg1681, which enhances Lp susceptibility to HGA ( cooperate in the production of shared, secreted public goods (Nadell et al., 2016;512 Steenackers, Parijs, Foster, & Vanderleyden, 2016). For Lp, such public goods include 513 the production of siderophores, surfactant, and quorum sensing signals (Liles, Scheel, & 514 Cianciotto, 2000;Spirig et al., 2008;Stewart et al., 2011). However, such public goods 515 can be vulnerable to cheating if other bacteria exploit these extracellular resources. In 516 this context, HGA is an unusual public good ( Figure 6). Because high-density, 517 established bacterial communities are largely resistant to inhibition, they can use HGA 518 to protect against low-density, invading Legionella with little harm to themselves. Due to 519 these dynamics, we propose that HGA can act as a niche-protective public good. 520 Moreover, because HGA activity can be felt well outside the spread of other Lp public 521 goods such as surfactant (Supplemental Figure 1), HGA may simultaneously protect a 522 number of other public goods from being exploited by invading bacteria. Across many 523 bacterial species, the production and use of public goods is regulated by quorum 524 sensing (Abisado et al., 2018). Although our results suggest that the known quorum 525 sensing pathway in Lp is not involved in HGA production or resistance (Supplemental 526 Figure 7), future studies on the regulation of HGA resistance may identify additional 527 molecular pathways that are critical to the social behaviors of Legionella in bacterial 528 communities. 529 Figure 6: Model for HGA activity in biofilms. Lp (purple) colonizes a surface and grows to form a microcolony. Once cells are at high density, they secrete abundant HGA (yellow). Through unknown mechanisms, high density Lp are resistant to HGA's effects (bold outline), while low density Lp or other Legionella species (blue) are inhibited by HGA and cannot invade the microcolony's niche. The territorial protection via HGA allows for the established Lp community to secrete other public goods such as surfactant, which are protected from exploitation by other bacteria.

Bacterial strains and growth conditions
The bacterial strains and plasmids used for 533 this study are listed in Table 3. As our wild type Legionella pneumophila (Lp) strain, we 534 used KS79, which was derived from JR32 and ultimately from isolate Philadelphia-1 (de 535 Felipe et al., 2008;Rao, Benhabib, & Ensminger, 2013;Sadosky, Wiater, & Shuman, 536 1993). Compared to JR32, the KS79 strain has a comR deletion to enable genetic 537 manipulation (de Felipe et al., 2008). We used Legionella micdadei (Lm) tatlock as our 538 susceptible strain (Garrity, Brown, & Vickers, 1980). Liquid cultures of Legionella were 539 grown shaking in AYE liquid media at 37˚C (De Jesús, O'Connor, & Isberg, 2013). To 540 manipulate the redox state of AYE, we altered the amount of cysteine added to the 541 media from 0.4 g/L in standard AYE to 0.1, 0.2, and 0.8 g/L. On solid media, Legionella 542 were grown either on BCYE agar plates either containing the standard cysteine 543 concentration (0.4g/L) or in "low cysteine" conditions (0.05g/L) (Feeley et al., 1979;544 Solomon & Isberg, 2000). E. coli strains used for cloning were grown in LB media. 545 Where indicated, antibiotics were used at the following concentrations in solid and liquid 546 media, respectively; chloramphenicol (5 µg/mL and 2.5 µg/mL), kanamycin (40 µg/mL) 547 and ampicillin (50 µg/mL and 25 µg/mL). For counter-selection steps while generating 548 deletion strains, 5% sucrose was added to BCYE plates. For top agar and overlay 549 experiments, we used 0.7% agar dissolved in water, which was kept liquid at 50˚C 550 before pouring over BCYE plates. 551 552 Gene deletions and complementations Genomic knockouts in L. pneumophila were 553 generated as previously described (Wiater et al., 1994). Briefly, we used an allelic 554 exchange plasmid (pLAW344) harboring chloramphenicol and ampicillin selection 555 cassettes and the counter-selection marker SacB, which confers sensitivity to sucrose. 556 Into this plasmid, we cloned ~1kb regions upstream and downstream of the gene of 557 interest to enable homologous recombination. Following electroporation and selection 558 on chloramphenicol, we used colony PCR to verify insertion of the plasmid into the 559 chromosome, before counter-selection on sucrose media. From the resulting colonies, 560 we performed PCR and Sanger sequencing to verify clean gene deletion. For complementation, the coding region of a candidate gene was cloned into a plasmid 562 (pMMB207c) following a ptac promoter (J. Chen et al., 2004). To induce gene 563 expression, strains carrying pMMB207c-derived plasmids were exposed to 1mM IPTG. 564 All constructs were assembled using Gibson cloning (NEB Catalog #E2621). 565

Inhibition assays on agar plates To visualize inhibition between neighboring 567
Legionella on solid media, a streak of approximately 5 x 10^6 CFU of the inhibitory 568 strain of Lp was plated across the center of a low-cysteine BCYE plate. After 3 days 569 growth at 37˚C, dilutions of susceptible Lp or Lm were plated as 10µLspots 570 approximately 1 cm and >2 cm from the central line. Once spots were dry, plates were 571 then incubated for an additional 3 days before scoring for inhibition. This assay was also 572 used to quantify the bactericidal inhibition of Lm, with slight modifications. Here, all Lm 573 was plated in 20µLspots at 10^6 CFU/mL. The time of plating susceptible Lm was 574 treated as t=0. Once spots were dry, plugs were extracted from within the Lm spots 575 using the narrow end of a Pasteur pipette. These plugs were transferred into media, 576 vortexed, and plated to quantify CFU. This procedure was repeated after 48h at 37˚C to 577 compare Lm viability and growth within ("near", Figure  were mixed in equimolar ratios with HGA, and incubated shaking at room temperature 588 for 1 hour before spotting onto bacterial lawns. HGA-melanin pigment was prepared 589 from Lp conditioned media as previously described (Zheng,Chatfield,Liles,& 590 Cianciotto, 2013) from KS79, the unpigmented hisC2::Tn mutant, and the 591 hyperpigmented ∆hmgA mutant. Briefly, conditioned media was collected and sterile 592 filtered from 100 mL cultures of Lp in AYE media grown shaking at 37˚C for 3 days. The 593 conditioned media was acidified to a pH of 1.5 and transferred to 4˚C for 2 hours to 594 precipitate. Precipitated pigment was collected by centrifugation at 4000 x g for 15 595 minutes and then washed with 10 mM HCl. Pelleted pigment was then returned to 596 neutral pH and resuspended in PBS at 10X before testing. 597 598 Transposon mutagenesis screen For random transposon insertion mutagenesis, we 599 used a Mariner transposon from the pTO100 plasmid (O'Connor et al., 2011). We 600 electroporated this plasmid into the KS79 strain and allowed cells to recover at 37˚C for 601 5 hours. To select for cells with integrated transposons, cultures were plated on 602 BCYE/Kan/sucrose plates and incubated at 37˚C for 3 days before screening individual 603 mutant colonies. 604 To identify transposon mutants with defects in Lm inhibition, we transferred each 605 Lp mutant onto a low-cysteine plate with a lawn of 5 x 10^7 CFU Lm and visually 606 screened for those with either small zones of inhibition or no zone of inhibition. Mutants 607 were added on top of the Lm, either by replica plating with sterile Whatman paper 608 (Whatman: #1001150) or by transferring with a sterile toothpick. Plates were then 609 incubated at 37˚C for 3 days and scored. All putative mutants underwent clonal re-610 isolation, were diluted to OD 600 of 0.1, and spotted on Lm lawns to retest their 611 phenotypes. To map the sites of transposon integration, we used arbitrary PCR as 612 described in (T. Chen, Yong, Dong, & Duncan, 1999), with primers redesigned to work 613 with the pTO100 transposon (Table 3). Briefly, this protocol involved two PCR steps to 614 amplify the DNA flanking the transposon. The first step used low annealing 615 temperatures to allow the arb1 randomized primer to bind many sites in the flanking 616 DNA while the pTO100_F or pTO100_R primer annealed within the transposon, 617 generating multiple products that overlapped the flanking DNA. These products were 618 amplified in the second step PCR using the arb2 and pTO100_Rd2 primers, and we 619 used the pTO100_Rd2 primer for Sanger sequencing. PCR programs and conditions 620 were as in (T. Chen et al., 1999). 621 4B), overnight cultures of Legionella were diluted to 10^8 CFU/mL in AYE, mixed with 624 synthetic HGA or pigment in 96 well plates, and grown shaking at 425 cpm at 37˚C. The 625 cytation 3 imaging reader (BioTek™ CYT3MV) was used to monitor growth by OD 600 626 measurements. Because oxidized pigment from synthetic HGA is detected at OD 600, 627 each experiment included bacteria-free control wells containing media and each 628 concentration of HGA. To correct OD 600 readings for pigment development, at each 629 time point we subtracted the control well reading from bacterial wells that received the 630 same concentration of synthetic HGA. For experiments with HGA "pre-oxidation" (Figure  631 3D), we diluted HGA in AYE media and incubated this solution shaking at 37˚C for 24 632 hours in the plate reader before adding bacteria. Complete oxidation of HGA during the 633 24 hours was monitored using OD400 to track the accumulation of HGA-melanin 634 pigment (Supplemental figure 5). In Lp, HGA inhibition in AYE rich media resulted in a 635 growth delay, similar to an extended lag phase ( Figure 4B). To compare sensitivity to 636 HGA among Lp strains, we calculated the lag phase from the growth curve of each well 637 using the GrowthRates program (Hall, Acar, Nandipati, & Barlow, 2014). We excluded a 638 small number of samples where the growth curve was not well fit (R<0.99), and then for 639 each strain used the difference in lag time between the samples with and without HGA 640 to calculate the growth delay due to HGA (Figures 5C). 641 642 As a complementary assay, we evaluated Legionella viability when exposed to 643 HGA in nutrient-free PBS at different cell densities. Stationary phase cultures were 644 washed once and re-suspended in PBS. We diluted these bacteria to estimated starting 645 concentrations of 10^9, 10^8, and 10^7 CFU/mL and plated for CFU at t=0. We 646 distributed the remaining bacteria into 96 well plates with or without 125 µM HGA. 647 Plates were incubated shaking in a plate reader at 425 cpm at 37˚C for 24 hours before 648 plating to quantify CFU on BCYE plates. CFU were counted after 3-4 days growth at 649

37˚C. 650
Estimation of amount of HGA secreted by Lp HGA-melanin is a black-brown pigment 652 that is easily detected at OD 400. We took advantage of this coloration to estimate the amount of HGA that had been secreted by Lp by comparing the color of conditioned 654 media to a standard curve of oxidized synthetic HGA. To isolate conditioned media from 655 pigment mutant strains, cultures of KS79, ∆hmgA, and HisC::Tn were inoculated with 656 fresh colonies from a BCYE plate into 5 mL AYE and were grown shaking at 37˚C for 48 657 hours. We then collected conditioned media by pelleting the bacteria and passing the 658 supernatant through a 0.2µm filter. To harvest conditioned media for a time course, 659 cultures of Lp were inoculated into 5 ml AYE and grown shaking at 37˚C. After 15, 20, 660 24, 39, 44, and 48 hours, we measured the OD 600 of the culture and collected 661 conditioned media. To create a standard curve, we diluted synthetic HGA into AYE at 662 the following concentrations: 62.5 µM, 125 µM, 250 µM, 500 µM, and 1 mM. The 663 conditioned media and standard curve samples were incubated in a 96 well plate in a 664 plate reader shaking at 37˚C for 24 hours to allow the HGA to oxidize. We used OD 400 665 data from the 24 hour time point to generate a standard curve for each HGA 666 concentration and calculated a line of best fit using linear regression. This equation was 667 used to estimate the amount of secreted HGA that corresponded to the OD 400 of each 668 conditioned media sample. (Supplemental Figure 5). 669 670 HGA-resistant mutants Because the inhibitory activity of HGA is quenched through 671 interactions with rich media (Supplemental Figure 4), it was not possible to select for 672 HGA-resistant mutants by mixing HGA into BCYE agar. Instead, to reduce the potential 673 for HGA to react with media components while allowing sufficient access to nutrients for 674 mutant cells to grow, we selected for HGA-resistant mutants by mixing 4 x 10 ^7 CFU 675 Legionella with 1mM HGA in 4mL of 0.7% water agar and pouring this solution as an 676 overlay on a low-cysteine BCYE plate. Plates were incubated at 37˚C for 6 days, before 677 candidate resistant colonies were picked and clonally isolated. The HGA resistance and 678 growth of each isolate was re-tested on overlays with or without 1mM HGA on both 679 regular and low-cysteine BCYE. 680 681 Twenty-nine isolates were more resistant to HGA than wild type Lp upon 682 retesting. We sequenced and analyzed genomic DNA from these isolates and a 683 matched wild type strain as follows. DNA was prepared from each strain using a Purelink genomic DNA mini kit (Invitrogen, #K1820). DNA concentrations were 685 quantified using Qubit and normalized to 0.5 ng/uL. Barcoded libraries were prepared 686 using tagmentation according to Baym et al. 2015(Baym et al., 2015, analyzed with 687 QuantIT DNA quantification, pooled, and sequenced with 50 bp paired-end reads on an 688 Illumina HiSeq 2500 machine. Reads were trimmed for quality and to remove Nextera 689 indices with Trimmomatic (Bolger, Lohse, & Usadel, 2014) and mapped to the 690 Philadelphia-1 genome (Chien et al., 2004) using Bowtie2 with default parameters 691 (Langmead, Trapnell, Pop, & Salzberg, 2009). Coverage plots were generated for each 692 strain using bamcoverage (Ramírez et al., 2016) and visually examined for evidence of 693 large genomic deletions and amplifications. None were observed, apart from a 694 prophage that was present in the reference genome but missing from all sequenced 695 strains, including our wild type KS79 strain. Variants were detected for each mutant 696 using Naive Variant Caller (Blankenberg D, et al. In preparation). Those variants that 697 were detected in mutant strains but not the wild type strain were considered as putative 698 causative mutations. For each of these mutations, we inspected the mapped reads and 699 excluded faulty variant calls that either were adjacent to coverage gaps or that did not 700 appear to be fixed in the clonal mutant and/or wild type sequences, likely due to errors 701 in read mapping. After this manual filtering, 1-3 well-supported mutations remained for 702 each mutant genome. Nine of the mutants had undergone selection on a different day 703 from the other mutants: in addition to various unshared mutations, these nine strains 704 each carried exactly the same missense mutation in rplX, which we disregarded as a 705 background mutation that likely arose before selection. Following this exclusion, each 706 mutant carried only a single well-supported mutation in a coding region. Most often this 707 coding mutation was the only mutation we detected, although one mutant carried two 708 additional intergenic point mutations. The coding mutations were point mutations or 709 small deletions that resulted in non-synonymous changes, frame shifts, or gene 710 extensions. Across different mutants, the mutations we uncovered were repeatedly 711 found in the same, few loci (Table 1). 712