Niche partitioning facilitates coexistence of closely related gut bacteria

Ecological processes underlying bacterial coexistence in the gut are not well understood. Here, we disentangled the effect of the host and the diet on the coexistence of four closely related Lactobacillus species colonizing the honey bee gut. We serially passaged the four species through gnotobiotic bees and in liquid cultures in the presence of either pollen (bee diet) or simple sugars. Although the four species engaged in negative interactions, they were able to stably coexist, both in vivo and in vitro. However, coexistence was only possible in the presence of pollen, and not in simple sugars, independent of the environment. Using metatranscriptomics and metabolomics, we found that the four species utilize different pollen-derived carbohydrate substrates indicating resource partitioning as the basis of coexistence. Our results show that despite longstanding host association, gut bacterial interactions can be recapitulated in vitro providing insights about bacterial coexistence when combined with in vivo experiments.


Introduction
factors to transcriptional changes in the four Lactobacillus species. We grew the four 247 species in vitro in either co-culture or mono-culture, and with either pollen extract (PE) 248 or glucose as growth substrate (G) ( Figure 5A). As for the in vivo RNA-Seq analysis, MDS 249 plots of the normalized read counts indicated that the four species exhibit treatment-250 specific transcriptional responses (Supplementary Figure 4). 251 For each species, whether grown alone or in co-culture, we found between 159-252 393 genes to be differentially regulated between the PE and the G treatment ( Figure 5B, 253 log2FC ³ |2| and p-value£0.01). As in vivo, Carbohydrate transport and metabolism (COG 254 'G') was the predominant functional category among the upregulated genes in the 255 presence of pollen ( Figure 5C) and enriched in all eight comparisons (four species , each 256 alone or in co-culture, Fisher's exact test p-value < 0.01, Supplementary Dataset 6). 257 Moreover, 25.3%-36.9% of the genes upregulated in vivo were also upregulated in vitro 258 in the presence of pollen. In particular, the species-specific carbohydrate metabolism 259 functions described above (Figure 4) showed a similar transcriptional response to pollen 260 in vivo and in vitro ( Figure 5D). In contrast, most of the putative adhesin genes 261 upregulated in vivo were not upregulated in vitro during growth in pollen or had 262 relatively low transcripts per million (TPM). This suggests that these genes are either 263 expressed in response to the host environment , or the presence of entire pollen grains 264 or sugar water, both of which were only included in the in vivo but not in the in vitro 265 experiment (Supplementary Dataset 7). It is also noteworthy that fewer genes were 266 downregulated than upregulated in pollen relative to glucose, and that the COG category 267 'G' was not enriched among the downregulated genes, which is concordant with our in 268 vivo transcriptome analysis. (Supplementary Dataset 6). Based on these results, we 269 conclude that each species upregulates specific operons for the transport and utilization of different carbohydrates (e.g. sugar alcohols and glycans) in response to the presence 271 of pollen, independent of the host environment. We found that a large fraction of the genes upregulated in PE relative to G in the 276 mono-cultures were also upregulated in the co-cultures (58.2-87.8%, Figure 5E). In 277 particular, the gene clusters identified to be regulated in a species-specific manner (see 278 above) showed highly concordant gene expression profiles in vitro independent of the 279 presence/absence of the other Lactobacillus species. This was confirmed by the direct 280 comparison of mono-culture and co-culture conditions. In comparison to the nutritional 281 treatments, fewer genes (9-149 genes) were differentially expressed between co-culture 282 and mono-culture treatments (log2FC ³ |2| and p-value£0.01), (Figure 5F). 283 We could not find any consistent pattern across the four species in terms of COG 284 category enrichment (Supplementary Dataset 6). Moreover, only a few genes were 285 differentially expressed in more than one species (6.25-30%), or across both nutrient 286 conditions (1.86-5.33%). Citrate fermentation genes were upregulated in Lkul in co-287 culture relative to mono-culture when grown in pollen, whereas in Lhel the opposite was 288 observed ( Figure 5G). Also of note, the oligopeptide transporter system which was 289 upregulated in vivo in Lkul in the presence of pollen, was also upregulated in vitro in the 290 presence of pollen, but only when other species were present. These two specific 291 examples show that a few metabolic functions are differentially regulated in response to 292 other bacteria, but not always in the same direction across species, or only in a specific 293 nutrient condition. We thus conclude that the main factor driving changes in gene expression in the four strains is the presence of pollen, rather than the presence of other 295 Lactobacillus species. 296

297
Metabolomics analysis reveals differences in flavonoid and sugar metabolism 298 across the four Lactobacillus species 299 Our transcriptome analyses suggest that differences in sugar metabolism may 300 enable the four species to coexist in the presence of pollen in vitro and in vivo. To assess 301 species-specific metabolic changes when grown in pollen, we profiled the metabolome of 302 the pollen extract medium before (t = 0h) and after bacterial growth (t = 16h) using Q-303 TOF-based untargeted metabolomics 27 . We annotated a total of 657 ions of which 406 304 could be reliably categorized as pollen-derived ions, as opposed to ions originating from To corroborate the species-specific utilization of flavonoids, we incubated each of 323 the four species in base culture medium supplemented with rutin. We observed the 324 formation of a yellow insoluble precipitate only in the wells incubated with Lmel ( Figure  325 6B). Metabolomics analysis confirmed that rutin was depleted in these wells and that the 326 yellow precipitate corresponded to an accumulation of quercetin, the water-insoluble, 327 deglycosylated aglycone of rutin ( Figure 6C). These findings are consistent with our 328 transcriptome results which show that Lmel is the only species that upregulated a 329 rhamnosidase gene known to cleave rhamnose residue from rutin 28 (Figure 4). 330 Other ions with species-specific abundance changes included a plant-derived 331 Based on the untargeted metabolomics analysis, we conclude that the four species 340 target different metabolites, in particular secondary plant metabolites present in pollen. 341 In order to assess differences in the utilization of simple sugars and acids in more detail, 342 we analyzed the supernatants of cultures of the four strains after 0, 8, 16 and 24 h of 343 growth using GC-MS. We used a semi-targeted approach, where we identified a subset of metabolites by preparing analytical standards and the others by using a reference library 345 (see methods). We identified 113 metabolites of which 46 showed a significant change in 346 abundance in at least one strain between timepoint 0h and 24h (log2FC ³ |2| and p-value 347 £ 0.01, Student's t-Test, BH correction) (Supplementary Dataset 8). All four species 348 showed mixed substrate utilization, i.e. they utilized several substrates simultaneously. 349 Moreover, most substrates were utilized by all four species, but often at different rates.  We disentangled the effect of the diet and the host on the interactions between the four 381 species by serially passaging them through gnotobiotic bees or in culture tubes, under 382 two nutrient conditions (pollen versus simple sugars). Our results show that the 383 dynamics in the four-species community is governed by negative interactions, because 384 the growth of each member was lower in co-culture than in mono-culture, independent 385 of the environment (host or culture tube) and the nutrient condition (pollen or simple 386 sugars). This is consistent with previous observations that negative interactions 387 predominate in nutrient-rich environments 31,34-37 . Moreover, the four Lactobacillus 388 species harbor relatively small genomes (1.5-2Mb) with a conserved and streamlined 389 core metabolism and similar auxotrophies, suggesting overlapping nutritional 390 The coexistence of bacterial symbionts can be facilitated by the host, e.g. by 392 providing a spatially structured environment that results in the physical separation of 393 competing strains 41-44 , or by secreting metabolites that support niche specialization 45,46 . 394 However, in the case of the four Lactobacillus species, such host-related features seem 395 not to be sufficient to support coexistence, because the four-species community was rapidly dominated by a single species, when passaged through gnotobiotic bees that were 397 fed a simple sugar diet. In contrast, when providing a more diverse nutrition in the form 398 of pollen, we found that the four species were stably maintained both in vivo and in vitro. 399 We thus conclude that the coexistence of the four Lactobacillus species in the honey bee 400 gut primarily depends on the pollen diet of the host and not the host environment itself. 401 The challenges in replicating the native environment such that it is possible to 402 study relevant interactions of host-associated microbes in vitro are formidable. These Although ecological interactions in bacterial communities have been investigated 436 across a wide range of experimental systems, few studies have tackled the molecular 437 mechanisms underlying coexistence. In some cases, cross-feeding of metabolic by-438 products facilitates the maintenance of diversity in bacterial communities, such as after 439 passaging leaf and soil samples in single carbon sources 12 . However, cross-feeding does 440 not seem to play an important role in maintaining coexistence of the four Lactobacillus 441 species in this study. Unlike the above example, feeding a single carbon source led to the 442 extinction of all but one species. Our metabolomics analysis also did not reveal any major 443 metabolites that could potentially be cross-fed, i.e. were produced by one species and 444 utilized by another. Finally, we identified no transcriptional changes that would suggest cross-feeding activities when comparing mono-cultures and co-cultures of the four 446

Lactobacillus species. 447
Instead, our combined transcriptomics and metabolomics analyses suggest that 448 coexistence is facilitated by specialization towards distinct pollen-derived nutrients. We Lactobacillus Firm5 and could, e.g., explain why the Asian honey bee, Apis cerana, harbors 487 only one species of this phylotype in its gut 56 . 488 However, we have only tested a single strain of each of the four species. Therefore, 489 given the extensive genomic diversity within these species 22 , more work is needed to 490 determine if the identified patterns of coexistence reflect stable ecological niches 491 occupied by the four species or are rather the result of the specific strains selected for 492 our experiments. In a recent study on pitcher plant microbiomes it was shown that even 493 strains that differ by only a few base pairs can have different ecological trajectories in 494 communities and coexist over extended period of time 8 . Expanding our approach to 495 strains within species presents an exciting next step to understand at which level discrete 496 ecological niches are defined in the bee gut and how diversity can be maintained in such 497 with 2% w/v fructose and 0.2% w/v L-cysteine-HCl) from glycerol stocks stored at -80˚C. 507

Materials and methods
MRSA plates were incubated for three days in anaerobic conditions at 34˚C to obtain 508 single colonies. Single colonies were inoculated into a liquid carbohydrate-free MRS 509 medium (cfMRS 57 ) supplemented with 4% glucose (w/v), 4% fructose (w/v) and 1% L-510 cysteine-HCl (w/v) and incubated at 34˚C in anaerobic conditions without shaking. 511 512

In vivo transfer experiments 513
Bacterial colonization stocks were prepared from overnight cultures by washing 514 the bacteria in 1xPBS, diluting them to an OD600 = 1, and storing them in 25% glycerol at 515 -80˚C until further use. For colonization stocks containing all four species, cultures 516 adjusted to an OD600 = 1 were mixed at equal proportions. Microbiota-depleted bees were 517 obtained from colonies of Apis mellifera carnica located at the University of Lausanne 518 following the procedure described in Kešnerová et al. (2017). Colonization stocks were 519 diluted ten times in a 1:1 mixture of 1xPBS and sugar water (50% sucrose solution, w/v) 520 and 5 μL were fed to each bee using a pipette. Five days post-colonization, ten rectums 521 were dissected and homogenized in 1xPBS. An aliquot of each homogenized gut was used 522 for CFU plating to enumerate the total bacterial load and for amplicon sequencing to 523 obtain the relative abundance of each community member (see below). To serial passage 524 the community through microbiota-depleted bees, the ten homogenized gut samples 525 from the same treatment were pooled together and stored in 25% glycerol at -80˚C until 526 a new batch of microbiota-depleted bees was available. At the day of colonization, a 527 frozen aliquot of the pooled gut homogenate was thawed, diluted ten times in a 1:1 528 mixture of 1xPBS and sugar water (50% sucrose solution, w/v), and fed to newly 529 emerged microbiota-depleted bee as described above. This was repeated for a total of six 530 serial passages. Throughout the experiments all bees were kept on either a sugar water 531 or a sugar water/pollen diet according to the two dietary treatment. conditions without shaking (300 rpm). After 24 h of incubation, an aliquot of each sample 546 was subjected to CFU plating to enumerate the total bacterial load. Then, 1% of each 547 culture (i.e. 5 μL) was transferred to a plate with fresh medium supplemented with the 548 appropriate carbon sources and incubated again. These transfers were repeated 10, 549 respectively, 20 times for the two independent experiments. After each transfer, cultures 550 were washed once with 1xPBS and stored at -20˚C for amplicon sequencing analysis. 551 CFUs were counted after 24 h and at the final transfer. 552

Amplicon sequencing 554
The relative abundance of the four strains across all transfer experiments was 555 obtained using amplicon sequencing of a 199-bp long fragment of a housekeeping gene 556 encoding a DNA formamidopyrimidine-glycosylase which allows to discriminate the four 557 strains from each other 58 . 558 Phosphatase -1000U/ ml -NEB, 10% Exonuclease I -20000 U/ ml -NEB, 45% glycerol 575 80% and 30% dH2O). and incubated for 30 min at 37˚C and for 15 min at 80˚C. For the 576 second PCR reaction, 5 μL of purified PCR products were mixed with the same reagents 577 as before. The PCR program was the same as above with the exception that the annealing 578 temperature was set to 60˚C and the denaturation-annealing-extension steps were 579 repeated for only 8 times. The barcoded primers are listed in Supplementary Table 1. The  580 amplicons of the second PCR were purified using the Exo-SAP program as described 581

above. 582
To prepare the sequencing of the amplicons, DNA concentrations were measured 583 using Quant-iT™ PicoGreen ® for dsDNA (Invitrogen). Each sample was adjusted to a DNA 584 concentration of 0.5 ng/μL and 5 μL of each sample were pooled together. The pooled 585 sample was loaded on a 0.9% agarose gel and gel-purified using the QIAquick Gel 586 Extraction Kit (Qiagen) following the manufacturer's instructions. The purified DNA was 587 To obtain absolute abundance data for each strain, we combined the relative 597 abundance data from the amplicon sequencing with CFU counts obtained from plating 598