A model Roseobacter employs a diffusible killing mechanism to eliminate competitors

The roseobacter clade is a group of α-proteobacteria that have diverse metabolic and regulatory capabilities. They are abundant in marine environments and have a substantial role in marine ecology and biogeochemistry. However, interactions between roseobacters and other bacterioplankton have not been extensively explored. In this study, we identify a killing mechanism in the model Roseobacter Ruegeria pomeroyi DSS-3 by competing it against a group of phylogenetically diverse bacteria. The killing mechanism involves an unidentified antimicrobial compound that is produced when cells are grown on both surfaces and in suspension and is dependent on cell density. A screen of random transposon mutants revealed the killing phenotype, as well as resistance to the antimicrobial, require genes within an ~8 kb putative γ-butyrolactone synthesis gene cluster, which resembles similar pheromone-sensing systems in actinomycetes that regulate secondary metabolite production. Transcriptomics revealed the gene cluster is highly upregulated in wild-type DSS-3 compared to a non-killer mutant when grown in liquid coculture with a roseobacter target. Our findings show that R. pomeroyi has the capability to eliminate closely- and distantly-related competitors, providing a mechanism to alter the community structure and function in its native habitats.

One strategy to increase fitness within a community is interference competition, in which a 48 molecular mechanism is employed to kill or inhibit competitors. Some roseobacter strains have the 49 capacity to produce antimicrobials and behave antagonistically toward other bacteria (14, 15). 50 Leisingera sp. JC-1, isolated from the surface of Hawaiian bobtail squid egg clutches, produces the 51 antimicrobial indigoidine, which differentially inhibited various Vibrio species and was hypothesized to 52 protect the eggs from fouling microorganisms (16). Phaeobacter gallaeciensis produces the 53 antimicrobial tropodithietic acid (TDA) when grown in coculture with the coccolithophore Emiliana 54 huxleyi. TDA protects the algae from pathogens, and P. gallaeciensis receives nutrients in return. When 55 the algal bloom begins to senesce, P. gallaeciensis produces anti-algal compounds known as 56 roseobacticides that cause the mutualistic relationship to transition to parasitism (17). Finally, extracts 57 from 14 roseobacter strains were found to produce compounds that inhibit the γ-proteobacterium 58 Vibrio anguillarium (15). As these examples demonstrate, antimicrobial and anti-algal production by 59 roseobacters can support beneficial relationships with their symbiotic partners and prevent competitor 60 bacteria from gaining a foothold in preferred niches. Given that roseobacters are known to make up a 61 large part of marine microbial assemblages and represent a mostly unexplored source of antimicrobials, 62 further study of roseobacter interference competition mechanisms is important to understand how this 63 diverse group shapes the communities they inhabit, which ultimately impacts the ecological services 64 that these communities provide. 65 In this study, we used coculture assays to determine the potential for model Roseobacter 66 Ruegeria pomeroyi DSS-3 to kill competitors and then employed a random mutagenesis approach to 67 and spotted onto ½ YTSS agar. The percent recovery of each strain was calculated by dividing the CFU 169 count after 24 hours of coincubation by the initial CFU count at the start of the experiment. Recoveries 170 >100% indicate the strain's CFU count has increased over 24 hours, whereas <100% recovery indicates a 171 strain has decreased. To determine whether a roseobacter strain is inhibited by DSS-3, the percent 172 recovery of each strain with DSS-3 was compared to the recovery of the strain when incubated alone. 173 After coincubation with DSS-3 at an initial 1:1 ratio, the recovery of every competing roseobacter was 174 significantly lower when compared to its growth in monoculture, except for the control where DSS-3 175 was coincubated with itself, and for P. daeponensis (Student t-test, P<0.05, Fig. 1). However, when DSS-176 3 was mixed with P. daeponensis at a 9:1 ratio, this resulted in a similar reduction of P. daeponensis 177 recoveries as observed with the other roseobacter strains (Fig. 1). 178 To determine whether DSS-3 can inhibit more distantly related bacteria, several marine γ-179 proteobacteria and actinobacteria were selected as competitors: Alteromonas sp. RAM1611, 180 Saccharospirillium sp. RAM1647, Idiomarina sp. RAM1191, Vibrio fischeri ES114, Escherichia coli DH5α, 181 Microbacterium phyllosphaerae RAM275, and Micrococcus sp. RAM1600. These strains were initially 182 mixed at a 1:1 ratio (DSS-3: competitor) as done with the roseobacter coculture assays, but 9:1 ratios 183 were also performed to determine if inhibition occurs at higher starting DSS-3 cell densities. In the 1:1 9 ratio experiments, the Idiomarina, Saccharospirillium, and Microbacterium strains did not show 185 statistically significant decreases in recovery when coincubated with DSS-3 Kn, whereas the 186 Alteromonas, Vibrio, Escherichia, and Micrococcus strains did (Student t-test, P<0.05). In the 9:1 ratio 187 experiments, all y-proteobacteria and actinobacteria strains tested had significantly lower recoveries 188 when cocultured with DSS-3 Kn versus in monoculture (Fig. 1). Taken together, these data suggest that 189 DSS-3 uses an unknown mechanism to kill competitor strains during growth on surfaces. CFUs were reduced by four orders of magnitude when coincubated with DSS-3 in liquid medium, and by 196 4 hours the TM1035 abundance was reduced below the limit of detection (200 CFUs/ml) ( Fig. 2A). 197 Given that some killing mechanisms are dependent on cell density (27-34), and we sometimes 198 only observed inhibition when DSS-3 initially outnumbered its competitor (Fig 1), we next examined 199 whether the DSS-3 killing phenotype might be correlated with population density in culture. Liquid 200 competition assays were conducted as above, except the starting densities of both DSS-3 and TM1035 in 201 the coculture were diluted by 2-, 4-, 6-, and 8-fold. If diluting the initial coculture cell density delays DSS-202 3 killing, then killing activity requires a particular cell density to promote the killing function. For each 203 dilution tested, a statistically significant reduction of TM1035 target cells did not occur until DSS-3 204 reached densities of ~10 8 CFU mL -1 (Fig. 2B). This result suggests that, under the conditions used here, 205 DSS-3 must achieve a cellular concentration threshold of >10 8 cells mL -1 before killing is detected and 206 that its antimicrobial function may be controlled in a density-dependent manner. 207 10 208

R. pomeroyi uses a diffusible killing mechanism 209
Antimicrobials can function as diffusible molecules, as is the case for most conventional 210 antibiotics, or they can require direct cell-cell contact for transfer from killer to target cells (35-38) . To 211 determine whether DSS-3's killing phenotype is contact dependent or diffusible, DSS-3 was cocultured 212 with TM1035 on agar plates as above, except the two strains were separated by a 0.22 µm nitrocellulose 213 filter, which prevents physical contact between the two strains while allowing diffusible molecules to be 214 exchanged. When tagged TM1035 was spotted on a filter with nothing below, or with untagged 215 TM1035 below the filter, the percent recovery of tagged TM1035 was above 100% ( Fig 2C). However, if 216 tagged TM1035 was spotted onto a filter above DSS-3 or concentrated kanamycin antibiotic, TM1035 217 CFUs were reduced beyond the limit of detection (Fig. 2C). These results suggest that DSS-3 employs a 218 diffusible killing mechanism that does not require direct contact with target cells. 219 220 Random transposon mutagenesis yields non-killer DSS-3 mutants. To identify the genes and possible 221 mechanism(s) required for DSS-3 killing, we generated a random transposon library and screened it for 222 DSS-3 mutants that can no longer kill a competitor strain. The screen was based on the observation that 223 when wild-type DSS-3 is grown on a lawn of fluorescently-tagged TM1035 target cells, a distinct zone of 224 killing is observed around the DSS-3 colony ( Fig 3A). If the transposon disrupts a gene required for 225 killing, then the DSS-3 mutant colony will not produce a zone of killing. We screened 10 000 DSS-3 226 mutants and isolated seven non-killer mutants that were either unable to produce a zone of killing of 227 TM1035, or displayed an intermediate zone of killing (Fig. 3A). 228 To confirm these mutants could no longer kill, DSS-3 mutants were coincubated with TM1035 229 on agar surfaces, as described above using both a 1:1 and 9:1 starting ratio of DSS-3 to TM1035 target. 230 Of the seven mutants, six were unable to kill TM1035 to a statistically significant degree after 24 hours 231 at either starting ratio (Fig. 3B). However, one mutant (GCS134), which exhibited only partial killing on 232 agar overlay plates (Fig. 3A), was not as efficient at killing target at a 1:1 starting ratio (Fig. 3B, filled 233 circles), but was able to kill when initially outnumbering the target at a 9:1 starting ratio (Fig 3B, empty  234 circles). These results suggest that six of the isolated mutants have lost the ability to kill, and one mutant 235 (GCS134) displayed reduced killing ability that could be restored by increasing its population size at the 236 beginning of the coincubation experiment. 237

238
The genes required for killing are located in a putative γ-butyrolactone biosynthesis gene cluster. The 239 locations of the transposon insertions were mapped using inverse PCR. For mutant GCS134, whose 240 killing ability was restored at 9:1 starting ratio, the transposon insertion site mapped to a sulfate 241 adynylyltransferase (ATP sulfurylase) gene (SPO0900), which is predicted to mediate the cellular 242 assimilation of inorganic sulfur (39-41). We hypothesized that this mutant, which could not kill TM1035 243 cells at a 1:1 starting ratio (Fig 3B), has a reduced growth rate and could not achieve the threshold 244 density required for killing. Indeed, the growth rate of GCS134 was reduced compared to the tagged 245 DSS-3 Kn strain (Supplemental Fig 1A). Moreover, when the CFUs were calculated for both the mutant 246 and TM1035 target strains in coculture, GCS134 was unable to reach the 10 8 CFU ml -1 threshold when a 247 1:1 starting ratio was used, and did not eliminate TM1035 (Supplemental Fig 1B). However, when 248 GSC134 was coincubated with TM1035 at a 9:1 starting ratio, the mutant could achieve the necessary 249 cell density and eliminate TM1035 target after a 24 hour coincubation (Supplemental Fig. 1B). Because 250 GCS134 retained its ability to kill when coincubation conditions permitted it to achieve a sufficiently high 251 cell density, we did not consider this mutant in further analysis. 252

12
The six remaining mutants contained transposon insertions in a single gene cluster located on 253 DSS-3's megaplasmid (Fig. 3C). According to an antiSMASH analysis (42), this gene cluster encodes a 254 potential γ-butyrolactone (GBL) biosynthesis gene cluster, including a predicted A-factor synthesis (AfsA) 255 domain protein that is essential for GBL production in other bacteria (43, 44), additional biosynthetic 256 proteins, three predicted transcriptional regulators, a predicted transporter, and two hypothetical 257 proteins (Fig. 3C). GBLs, such as A-factor, are quorum sensing molecules with a similar structure to the 258 well-studied quorum sensing molecule C-4 homoserine lactone (Fig. 3D) and have been found to 259 regulate antibiotic production and cell differentiation in streptomycetes (45-49). Together, these data 260 revealed two important findings: 1) they indicate our screen reached saturation because we obtained 261 multiple, independent insertions in one gene cluster, with two genes having multiple, independent 262 transposon insertions, and 2) the predicted GBL biosynthesis cluster is required for DSS-3 to kill a 263 competitor strain.  Table S2). Of the 35 organisms found to encode a 271 homolog of SP0A0342, 29 encode all three genes and 6 encode SPOA0342 and one flanking gene. Only 272 five of these organisms are alphaproteobacteria, two of which are roseobacters (Ruegeria sp. EL01 and 273 Shimia marina), and the majority of organisms that encode homologs to at least two of these three GBL 274 genes belonged to the γ-proteobacteria and Actinobacteria (Fig. 4, Supplemental Table S2). We next 275 searched more broadly for proteins containing the key A-factor domain (Pfam domain PF03756) using 276 13 AnnoTree (50). Similar to the Blastx search results, the AnnoTree results revealed that ~93% of the total 277 identified A-factor domain containing proteins were found in γ-proteobacteria and actinobacteria, and 278 less than 2% were found in roseobacters (Fig. 4). Together, these results suggest that DSS-3's GBL gene 279 cluster may have been acquired horizontally from either a γ-proteobacteria or Actinobacteria lineage. 280 281 GBL genes are required to protect against self-killing. Bacteria will often produce immunity factors to 282 prevent self-killing while employing antimicrobials (51). Given that antimicrobial and immunity genes 283 are often encoded near one another on the genome and coexpressed, our transposon insertions may 284 have also disrupted DSS-3's genetic factors for immunity. To determine whether the non-killer mutants 285 had become sensitive to killing, we coincubated the mutant DSS-3 strains with a differentially-tagged 286 parental DSS-3 strain using a 1:1 starting ratio and assayed percent recovery of each mutant after 24 hrs 287 on surfaces. Although some mutants grew in the presence of the parent strain (>100% recovery), others 288 were significantly inhibited or nearly eliminated. All three afsA (SPOA0342) mutants (GCS64, GCS124, 289 and GCS141) showed no statistically significant reduction in percent recovery when coincubated with 290 the parent strain, suggesting these strains retained immunity to the antimicrobial (Fig 5A). By contrast, 291 both mutants with transposon insertions in the hypothetical protein SPOA0341 (GCS121 and GCS122) 292 had reduced percent recoveries when coincubated with the wild-type (Fig. 5A), suggesting they are 293 sensitive to antimicrobial production by the parent strain. Interestingly, we recovered ~100-fold less of 294 SPOA0341 mutant GCS122 compared to SPOA0341 mutant GCS121, suggesting that although both 295 mutants have a transposon insertion in the same gene, the insertion site and/or orientation of the 296 transposon may also influence expression of immunity factors. Finally, the predicted regulator mutant 297 (SPOA0344, GCS140) showed the highest sensitivity to killing, with the recovery of this strain reduced 298 below the limit of detection after 24 hrs coincubation with the parent; a result similar to what we 299 observed when coincubating other sensitive roseobacter isolates with differentially-tagged DSS-3. 300 14 Taken together, these findings indicate that in addition to encoding the factors necessary for killing, the 301 gene cluster also encodes genes required for immunity. 302

303
The GBL gene cluster is energetically costly. In an analysis of the DSS-3 proteome, Christie-Oleza et al. 304 (2012) found five proteins encoded in the GBL gene cluster (SPOA0339-0343) comprise 1-6% of the 305 entire DSS-3 proteome under various conditions (52). Therefore, we hypothesized that the non-killer 306 mutants would have a higher growth rate, either due to their reduced GBL protein synthesis or lack of 307 antimicrobial production, which may also be energetically costly. To determine whether the mutants 308 grew on surfaces more quickly than the wild-type, the DSS-3 Kn, SPOA0342 afsA mutant GCS64, and the 309 SPOA0344 regulator mutant GCS140 were plated onto ½ YTSS agar and the diameters of the colonies for 310 each strain were measured after 48 hours of incubation at 29˚C. The colony diameter for two 311 representative non-killer mutants (one that is immune and one that is no longer immune) were 312 approximately twice that of the wild-type strain (Fig 5B). Taken together, these data suggest that killing 313 ability, but not immunity function, may be a fitness cost for DSS-3. 314 315 afsA is required for transcription of the GBL gene cluster. Because A-factor is known to regulate 316 antimicrobial production in actinomycetes, we hypothesized that the afsA gene product may have a 317 similar role in regulation of DSS-3 antimicrobial production. To determine the AfsA-dependent regulon 318 in DSS-3, we compared the transcriptomes of the afsA mutant GCS64 and DSS-3 Kn when coincubated 319 with Roseovarius sp. TM1035 in liquid suspension. The cocultured cells used for transcriptome 320 sequencing were collected at 1.5 hours when killing of TM1035 by wild-type DSS-3 begins to occur in a 321 1:1 liquid coculture (Fig. 6AB). Of the 4 252 genes in the DSS-3 genome, only 21 genes were significantly 322 differentially transcribed between DSS-3 Kn and the afsA mutant GCS64, 10 of which corresponded to 323 the putative GBL cluster (Fig 6C). Transcripts of these ten genes were enriched by 4 to 60 fold in the 324 wild-type strain compared to the afsA mutant. The only gene encoded in the GBL gene cluster that was 325 not differentially expressed was the predicted regulator encoded in SPOA0344. Taken together, these 326 data show 1) the afsA gene is required for expression of most of the genes in the GBL operon, and 2) this 327 regulatory mechanism does not significantly impact expression of other genes in the DSS-3 genome. 328

329
The GBL gene cluster is required for DSS-3 to outcompete phylogenetically diverse marine bacteria. To 330 test the GBL gene cluster's role in killing other bacterial types, DSS-3 Kn and the afsA mutant GCS64 331 were used in competition assays with the full taxonomic range of competitor strains described in Figure  332 1. For the roseobacter competitions, a 1:1 starting ratio was used for all the roseobacter competitions 333 except for P. daeponensis, in which case a 9:1 DSS-3 to P. daeponensis ratio was used. For all γ-334 proteobacteria and actinobacteria competitions, 9:1 ratios (DSS-3:competitor) were used. For these 335 experiments, log relative competitive indexes (log RCI) were calculated for each coincubation. A positive 336 log RCI indicates that DSS-3 had a competitive advantage after coincubation, while a negative log RCI 337 indicates that the competitor strain had an advantage. A log RCI of zero indicates that neither strain has 338 a competitive advantage. After 24 hours, all coincubations of competitor strains with DSS-3 Kn, had 339 significantly higher log RCI values compared to that of coincubations with the non-killer DSS-3 mutant 340 GCS64 (Fig. 7), suggesting the GBL cluster is required for DSS-3 to outcompete these isolates in 341 coculture. The one exception was Microbacterium sp. RAM275, which when coincubated with either 342 DSS-3 Kn or GCS64 had log RCI values near zero, suggesting the DSS-3 killing mechanism does not convey 343 a competitive advantage against this Actinobacterium under these conditions. 344

345
Discussion 346 Based on the data presented here, we propose the following model for how the GBL gene 347 cluster may enable DSS-3 to eliminate competitors. DSS-3 grows together in the presence of competitor 348 bacteria until a cell density threshold is reached, at which point the antimicrobial can eliminate 349 phylogenetically-diverse competing bacteria, significantly reducing their population sizes and enabling 350 DSS-3 to dominate the niche space. 351 Both killing and immunity functions require genes encoded in the GBL gene cluster. Given that 352 the killing phenotype is density-dependent and requires GBL biosynthesis genes, we hypothesize that a 353 GBL-like molecule may be the antimicrobial and a high cell density is needed to achieve cytotoxic levels 354 sufficient to kill competitor cells, or GBL may act as a signaling molecule that combines with one or more 355 regulators to activate production of an unknown, diffusible antimicrobial whose synthesis is regulated in 356 a density-and GBL-dependent manner. Future work should focus on identifying the antimicrobial 357 molecule, the type of GBL produced, and its specific role in mediating interbacterial killing. 358 Because competition can impact microbial community structure and function, it is critical to 359 identify the ecologically-relevant habitats and conditions that support and restrict DSS-3's killing activity. 360 The density dependent requirement of this killing mechanism limits the environments and micro-361 habitats where such a competitive mechanism may be advantageous. In our study, DSS-3 needed to 362 achieve a cell density of ~10 8 CFU/ml to kill a competitor. However, the cell density threshold required 363 for killing in the marine environment may be different because environmental viscosity and cellular 364 metabolism can influence fluid flow of the local environment, which may promote or prevent the 365 accumulation of signaling and/or antimicrobial molecules (53, 54). For example, the phycosphere or 366 organic particles are habitats with low diffusibility and high nutrients (55), and therefore may support 367 growth of DSS-3 microcolonies and allow local concentrations of these molecules to activate killing. (black) or in coculture with DSS-3 (white) on ½ YTSS agar, where less than 100% recovery corresponds to 537 a decrease from the initial colony forming unit (CFU) count, and greater than 100% recovery 538 corresponds to an increase from the initial CFU count. Asterisks denote a P<0.05 using a students t-test 539 comparing percent recovery of each species alone to percent recovery when coincubated with DSS-3. 540 Error bars indicate standard deviation.