Coexistence of E. coli strains in the mouse gut and strain dependent mode of evolution

We used a classical model for E. coli colonization of the mouse gut, the streptomycin treated mouse model [16]. This treatment allows for the successful colonization by E. coli of the mouse gut while maintaining a complex microbiota composition [12], although with a reduced species diversity [17]. Briefly, female mice were simultaneously colonized with two E. coli strains (n = 9) (Fig 1A) after seven days of antibiotic treatment. One E. coli strain belongs to phylogenetic group B1 and is a common resident of the mouse gut [14]. Previous studies [12,18] indicate that this strain is a good donor of MGEs to the other strain used in this study, E. coli strain K12. The latter belongs to phylogenetic group A, is the best characterized E. coli strain and a good recipient of MGEs from the donor strain. Its genome is smaller than that of mouse resident strain. We will refer to the mouse commensal strain as “strain B1” and to K12 as “strain A”. Each E. coli strain is labelled with two different neutral fluorescent markers, and different antibiotic resistances, to facilitate tracking the emergence of adaptive events in each strain during colonization (Fig 1A). In a clonal population, if adaptation is dominated by strong selective sweeps, then when an unconditionally beneficial mutation arises in a given fluorescent background and spreads to fixation (complete sweep) it will drag along the linked neutral fluorescent marker, which will also be fixed (genetic hitchhiking [19]). A more complex scenario can also occur if adaptation is dominated by clonal interference. This will happen when during the spread of a given beneficial mutation another beneficial mutation arises and escapes genetic drift. In this scenario, the frequency of the linked neutral fluorescent markers will fluctuate according to the dynamics of such competing clones. This may result in partial sweeps and maintenance of polymorphism of the neutral fluorescent markers for long periods of time [20]. Another scenario where neutral fluorescent marker polymorphism is expected to be maintained is if adaptation to the gut is characterized by mutations which are beneficial when rare but deleterious when at high frequencies (negative-frequency dependent selection) [21]. In line with our previous observations when these strains coexist in the mouse gut, we also expect the occurrence of events of HGT [12]. This additional mechanism of evolution has the potential to create new evolutionary events (e.g., due to phage transduction and lysogeny events, or transfer of plasmids by conjugation) and potentially introduce new selective pressures (e.g., due to prophage induction and cell lysis or to plasmid costs). Such events of horizontal transfer can further influence the dynamics of polymorphism at a neutral fluorescent marker.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Experimental design and distinct evolutionary dynamics between strains.

(A) Before co-colonization with E. coli, three cohorts of three mice (n = 3x3) were co-housed for 10 days to achieve microbiota homogenization: the first day without antibiotic treatment, followed by seven days of streptomycin treatment and two days without antibiotics in the water to clear the gut of any streptomycin traces [54]. In each cohort, mice were co-colonized with two fluorescent variants of strain A (CFP, blue circles; YFP, yellow circles) and two fluorescent variants of strain B1 (mCherry, red circles; GFP, green circles) E. coli, to follow their evolution in the mouse gut (see S1 Fig for more details). (B) Time series of the abundances of the two E. coli strains carrying one of the 4 fluorescent markers (CFP, YFP, mCherry and sfGFP). Dotted line represents the limit of detection for E. coli. Error bars represent the standard error of the mean (2*S.E.).

https://doi.org/10.1371/journal.pgen.1011777.g001

Temporal series data of the abundance of each E. coli strain in the co-colonized mice shows that both strains can coexist for at least three months in the mouse gut (coexistence observed in eight out of nine mice, strain A failed to colonize in mouse 9, Figs 1B and S1, and S1 Table). Under coexistence, the absolute abundance of strain B1 in the feces was significantly higher than that of strain A (mean Log₁₀CFUs of strain B1: 8.5 (±0.4, S.E.) g^-1 of feces and mean Log₁₀CFUs of strain A 7.4 (±0.3, S.E.) g^-1 of feces, Linear mixed-effects model fit by maximum likelihood, P = 0.0074). Long-term maintenance of polymorphism at the fluorescent marker locus is observed in the strain with the larger population size (strain B1) (Fig 1B), despite strong changes in fluorescent marker frequency (S2 Fig). However, we observed loss of polymorphism at the neutral fluorescent marker in the strain with lower population size (Fig 1B and S1 Table). Thus, under strain coexistence in the mouse gut, the probability of maintaining polymorphism at a neutral locus is strain dependent (Binomial Test for proportions, P = 0.00028).

Given the large differences in the density of each strain and previous theoretical work suggesting that the probability of fixation of beneficial mutations in clonal populations depends on population size [22], we decided to explore conditions under which a model of positive directional selection in strictly clonal populations could explain the experimental data on strain coexistence and the maintenance or loss of polymorphism at a neutral locus. We performed simulations assuming a simple evolutionary model where adaptive mutations are unconditionally beneficial; the distribution of their effects (DFEBM) is similar between the strains and no gene transfer from one strain to another can occur. Simulations across a wide range of values for the mutation rate show that when the DFEBM is exponential (a commonly used assumption [23–29]), the model does not explain the experimentally observed data (see S3 Fig). However, the observed maintenance vs loss of neutral marker polymorphism in each strain could be seen under other DFEBMs, but only under particular conditions. Specifically, the simulations show that a model of positive directional selection is only compatible with the observed data (Fig 1B) if all beneficial mutations have similar effects (~1%) (S3 Fig).

Genomic landscape of adaptive evolution of coexisting E. coli strains

To determine the evolutionary events that occurred in each E. coli strain under coexistence, and to quantify their rates of molecular evolution, we performed pool sequencing of clones from each strain after the 88 days of co-colonization (1584 generations, assuming 18 generations per day [21]).

In the mice where co-colonization was successful, the average number of mutation events detected in the strain B1 was higher than in strain A (10.3 ± 3.8 S.D. for strain B1 and 4.8 ± 1.6 S.D. for strain A, Paired T-test P = 0.007, S2 and S3 Tables). The 82 events that occurred in strain B1 were caused by SNPs (54 events, of which 34 are non-synonymous, four are synonymous and 16 intergenic), IS insertions (eight events), small insertions and deletions (nine events) and large deletions (11 events) (S3 Table). Across the eight mice where co-colonization was successful, the ratio of non-synonymous to synonymous mutations (dN/dS) was > 1, indicating adaptive evolution. Twelve targets of parallel evolution (i.e., mutational events in the same target observed in different mice) were identified in strain B1 (Fig 2B and S3 Table), two of which correspond to large deletions. Consistent with a strong adaptive value, these deletions were detected in the majority of the mice (six and five out of eight co-colonized mice for the 87,305 bp and 85,944 bp deletions, respectively). Both deletions are flanked by an intact tRNA and a partial tRNA (S4 Table), suggesting that these regions are prone to recombination events. Interestingly, both deleted regions have homologous stretches of DNA (S4A Fig). Functionally, many genes within the deletions are predicted to code for integrases/transposases or to encode hypothetical proteins of unknown function, but a considerable portion of the deleted genes are predicted to be involved in virulence and metabolism (15% and 26% for the 87,305 bp and 85,944 bp deletions, respectively, S4B Fig). Despite the high level of parallel evolution observed in strain B1, none of the mutations detected in this strain reached 100%, which is consistent with the maintenance of both fluorescence markers in this strain (S2 Fig and S1 Table) and the absence of complete selective sweeps. On the other hand, evidence for competition between clones with different beneficial mutations is seen in the evolved populations of strain B1 in six out of eight co-colonized mice. In mouse 1 at least three different alleles at the fimH locus were detected (S3 Table), in mouse 2 three alleles occurred in the adhesin precursor YadA, mouse 3 shows two different SNPs at dpiB and in mouse 4 and mouse 6 two different mutations increased in frequency at a predicted membrane protein and in the transcriptional regulator tdcA, respectively. This data is consistent with a model of intense clonal interference in the strain with larger population size. Pool-sequencing of clones from an earlier time point (at day 43, i.e., after 774 generations of evolution), shows the existence of more than one allele competing for fixation at loci ydaA, dpiB, prokka01045 and tdcA (S3 Table), further corroborating that clonal interference shapes the selection dynamics of strain B1.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Genetic basis of adaptation and rate of mutation accumulation in each E. coli strain.

Circular plots indicating the targets of evolution along the genome of strain A (A) and strain B1 (B) obtained from population sequencing at day 88. Targets that are considered adaptive within each strain, due to mutational parallelism, are highlighted in blue. Mutational events in the same gene or pathway observed in both strain A and strain B1, suggesting evolutionary convergence, are highlighted in bold and with the symbol ●. Each circular track corresponds to one mouse, being the outermost ring the mouse 1 in both strains and the innermost ring the mouse 8 in strain A and mouse 9 in strain B1. The blue bars inside each circular track show the frequencies of each evolutionary event. Gene names with unknown function were renamed for clarity (e.g., PROKKA_020290 is P20290 in the figure). The large deletions observed in strain B1 are indicated as ΔP47-P125 and ΔP1034-P1105, corresponding to the 87,305 bp and 85,944 bp deletions, respectively. (C) Rates of mutation accumulation per generation of each strain (strain B1 in black, strain A in grey) in each mouse. Rates were calculated by summing the allele frequencies of the evolutionary events detected in each strain after 88 days of colonization of the mouse gut. In mouse 9 the recipient strain failed to achieve persistent colonization.

https://doi.org/10.1371/journal.pgen.1011777.g002

Molecular evolution of strain A was characterised by a total of 38 mutational events after 88 days of gut colonization (S2 Table). These were caused by SNPs (13 events, of which 10 are non-synonymous, two are synonymous and one is intergenic), IS insertions (11 events), small insertions and deletions (five events) and large deletions (nine events). We detected five targets of parallel evolution in this strain across all the mice, which involved the following events: deletion of the cryptic prophage e14, mutations in the repressor of the fructoselysine operon frlR, IS insertions in the recombinase fimE, and in the intergenic regions focA/ycaO and mngB/cydA (Fig 2B and S2 Table). Consistent with their key role in adaptation to the gut, mutations in most of these targets were found in previous studies, where this strain colonized the gut as the sole E.coli strain [18,21,30–33], suggesting that these targets are independent of the co-colonization of strain A and B1. Across all co-colonized mice, nine selective sweeps (mutations that reached a frequency higher than 95% in an individual mouse) occurred in strain A. Interestingly, there were less detected partial sweeps in strain A, relative to strain B1, during the period in which evolution was followed. Multiple alleles segregating at the same locus were detected only in one out of eight mice for strain A vs. six out of eight for strain B1, after the 88 days of co-colonization. At an intermediate point (43 days of co-colonization) partial sweeps were detected for strain A in two out of eight mice vs. four out of eight mice for strain B1. This suggests strain A experienced less clonal interference than strain B1 throughout the entire evolution period. The complete sweeps experienced by strain A can account for the loss of polymorphism at the neutral fluorescent marker in most of the mice. In mouse 1, a SNP in flrR reached 100% frequency, consistent with the fixation of the CFP fluorescent marker; in mouse 2, a 1 bp insertion in the intergenic region mltB/srlA likely caused the fixation of the CFP fluorescence; in mouse 3 a SNP in frlR likely caused the fixation of the CFP marker; in mouse 4 either a 22Kb deletion or a IS insertion caused the fixation of the marker; in mouse 5 either a 2 bp insertion in frlR or a IS insertion in fimE caused the fixation of the marker; and in mouse 6 a SNP in prfB likely caused the fixation of the YFP marker. In mice 7 and 8 no mutation event reached 100% frequency. This suggests that in these mice, HGT events could be invoked to explain the fixation of one of the markers (see below).

Evolutionary convergence in E. coli strains during coexistence in the mouse gut

The strains used in our experimental evolution have distinct genome sizes, albeit in the same order of magnitude (strain A: 4.5 × 10⁶; strain B:5.2 × 10⁶ base pairs), and some of the parallel adaptations that were detected within a given strain occurred in genes for which there was intra-strain variation. For example, from the eight mice in which co-colonization was successful, in five mice strain A evolved by accumulating mutations in frlR, a gene that does not exist in strain B1, and in five mice the strain B1 evolved by two large deletions of genes that are not present in strain A. Importantly, the extent to which co-existing E. coli strains will evolve by altering similar functions is still poorly understood. Our experimental design offers a unique opportunity to address this question. In the two phylogenetic distinct strains studied we found that several targets shared between strains underwent convergent evolution (i.e., mutational events in the same gene or pathway observed in both strain A and strain B1). Evidence for evolutionary convergence occurred in nine genes or pathways (Fig 2A and 2B). A striking case was the fim operon, in which multiple alleles were detected in both strains (eight alleles in strain A and nine in strain B1) (Fig 2A and 2B). Consistent with an environment where different E. coli strains likely compete for niches and/or resources, the evolutionary convergent targets are related to motility/adhesion through type 1 fimbriae (fimA, fimE, fimH, fimI) [34], with these fim targets found in E. coli populations colonizing the same host (mouse 1 and 7). Convergent targets were also observed in central carbon metabolism (barA and uvrY) [35], fatty acid degradation (fadH and fadL) in mouse 2 [36], and threonine/serine metabolism (tdcA) [37].

Rates of evolution in the mouse gut of phylogenetic distinct strains of E. coli

The different histories of each E. coli strain lead us to expect that their rates of molecular evolution should be very distinct. As strain B1 is a common resident of the mouse gut [14], it could be better adapted to this environment than strain A, which has been propagated in laboratory environments for decades. Thus, it could evolve at a slower pace. However, it is also possible that the environmental shift caused by the perturbation of the microbiome composition by antibiotic treatment could boost the rate of evolution in strain B1 [38]. Following Good et al. [39], and considering the genome sizes to be in the same order of magnitude, we estimated the genomic rate of mutation accumulation (M(t)) by summing the allele frequencies of all the mutations detected in each strain, after three months of co-colonization, and dividing this value by the number of elapsed generations (1584 generations, assuming 18 generations per day similar for each strain). For strain B1, M(t) was 1.8x10^-3 (±0.3x10^-3, S.D.) per generation, while for strain A it was 1.3x10^-3 (±0.5x10^-3, S.D.) per generation (Fig 2C). These estimates are not significantly different between strains (Paired T-test, P = 0.07), with the same observation when normalizing M(t) for each strain genome size (Paired T-test, P = 0.23), and are in contrast with the expectation given their history and respective mean abundances. In fact, no significant correlation was found between strain abundance and rate of evolution (Pearson’s correlation, P = 0.6 for strain B1 and P = 0.3 for strain A, S5A Fig). Furthermore, lack of significant correlation was also observed with the gut microbiota diversity, measured with by the Shannon Index (Pearson’s correlation, P = 0.8 for strain B1 and P = 0.5 for strain A, S5B Fig) and the total number of amplicon sequence variants (ASVs) (Pearson’s correlation, P = 0.9 for strain B1 and P = 0.2 for strain A, S5C Fig). These results support the observation that both strains evolve at a similar pace independently of their population size and the microbiota diversity in the gut environment. Interestingly, the rates of evolution estimated here are very similar to those observed in the initial period of evolution of the in vitro propagation of a E. coli strain in minimal media with glucose [39].

Phenotypic evolution of E. coli strains evolved in the mouse gut

We next assayed the phenotypic diversity of clones sampled from both strains, after they co-evolved for three months in the gut. We compared the growth of the evolved clones with that of their ancestor in media supplemented with mouse food (S6 Fig), an environment for which strain A (E. coli K12) is, a priori, less adapted to than strain B1 (a mouse commensal). Indeed, in these in vitro fitness assays strain A showed a strong signal of metabolic adaptation to the mouse diet (S5 Table), with their evolved clones showing a large fitness increase, both in terms of carrying capacity and maximum growth rate (Fig 3A and 3C). In contrast, for the strain B1 no clear signal of metabolic adaptation to the mouse diet was found in the evolved clones (Fig 3B and 3D). This result is consistent with the fact that strain B1 was isolated from the feces of mice that have experienced this diet for many generations.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Fitness assays in vitro of evolved clones when grown in mouse food.

(A) Relative maximum carrying capacity of clones of strain A sampled after three months of gut colonization of each mouse. (B) Relative maximum carrying capacity of clones of strain B1 sampled after three months of gut colonization of each mouse. (C) Relative maximum growth rate (μ/μ_anc) of clones of strain A sampled after three months of gut colonization of each mouse. The values of μ_anc (ancestor clones of the strain A) are 0.27 h^-1 (S.D. = 0.02) for the CFP-labeled clone and 0.28 h^-1 (S.D. = 0.02) for the YFP-labeled clone when grown in mouse food. (D) Relative maximum growth rate (μ/μ_anc) of clones of strain B1. The values of μ_anc for the ancestor strain B1 in mouse food are 1.08 h^-1 (S.D. = 0.10) for the sfGFP-labeled clone and 1.11 h^-1 (S.D. = 0.13) for the mCherry-labeled clone.

https://doi.org/10.1371/journal.pgen.1011777.g003

Horizontal gene transfer between strains of E. coli repairs an important K12 gene

In a previous study where antibiotic treated mice were colonized solely with the K12 strain, we had observed that events of HGT could drive its evolution whenever another E. coli commensal residing in the gut would survive antibiotic treatment [12]. Here, where both strains simultaneously invade the gut microbiome, we find that, in most mice, the evolution of the E. coli K12 lineage was strongly impacted by several events of HGT, where DNA was acquired from the strain B1 (Fig 4A and S2 Table). In seven out of the eight co-colonized mice, these events were caused by the transfer of two prophages (Nef and KingRac) and two plasmids. While no mutations were detected in the plasmids acquired by strain A (S2 Table), in strain B1 plasmid copy number evolved and mutation events in the plasmids were observed (S3 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Evolutionary dynamics of HGT from strain B1 to strain A.

(A) Frequency of E. coli HGT events from strains B1 to A at days 43 and 88 after colonization. Error bars represent the standard error of the mean (S.E.M). (B) Schematic representation of the massive number of genes gained by the strain A (recipient) from the strain B1 (donor), likely by transduction via the KingRac prophage. Within the genes gained is fnr, a core gene present in the majority of E. coli sequenced isolates. Gene arrow representation is to scale. Note that for the alignment the genomic orientation of the strain B1 is inversed. A phage symbol is shown next to the clones where the prophage is active. ANI, average nucleotide identity. (C) Competitive fitness assay of a clone evolved in mouse 7 (Evolved) that was competed against a double lysogen (with Nef and KingRac prophages) clone (Ancestor) in mice after one week of antibiotic pre-treatment. The y-axis shows the competitive index along time of colonization in mice (n = 2 male and n = 2 female). The selective coefficient, estimated from the slopes of the competitive index along time in each mouse, is 4.0 per day (S.E. = 0.5) or 0.22 per generation.

https://doi.org/10.1371/journal.pgen.1011777.g004

Illumina sequencing of the E. coli populations of strain A and strain B1 at days 43 and 88 suggested a remarkable case of hybridization between coexisting strains in one of the mice (mouse 7). To validate this finding, we performed Nanopore long-read sequencing on two clones of strain A isolated from mouse 7 at day 88 (S6, S7 and S8 Tables), which confirmed the acquisition of a large genomic region in the chromosome of strain A. Specifically, a 93,712 bp region in strain A was replaced by a 120,546 bp region from strain B1 (Fig 4B). The strain A used to colonize the mice is a derivative of K12, and its original 93,712 bp region contains 99 genes (S6 Table), a cryptic prophage Rac, and a deletion of 13,756 bp adjacent to this cryptic prophage (when compared with the K12 reference genome NC_000913.2). In strain B1, the 120,546 bp region contains 135 genes, and is composed of an active prophage (named KingRac) and an adjacent region with the same genomic organization as that in strain A, with an average nucleotide identity (ANI) ranging from 80–100% for each 100 bp fragment. (Fig 4B and S6 Table). This event of DNA transfer led to the replacement of genes of strain A by homologous genes from strain B1, and also to the repair of the 13,756 bp region absent in strain A. This event of gene gain and repair is likely to have occurred via transduction (either specialized or lateral [40,41]) mediated by the prophage KingRac, given its proximity to this prophage in the genome. Importantly, the 13,756 bp region absent in strain A was originally composed of 13 genes (S6 Table) that have a role in key functions for E. coli gut colonization, and that got lost (i.e., likely negatively selected) under in vitro propagation.

The functions repaired in strain A include: motility, adhesion, and biofilms (uspE [42] and ydaM [43]); stress response and DNA repair (uspE [42] and ogt [44]); metabolite transport (abg operon [45] and ydaN [46]); and the switch from aerobic to anaerobic metabolism (fnr [47]). The latter is particularly interesting as fnr is an important gene for the ability of E. coli to colonize the mammalian gut. In fact, the presence of fnr in the genome of this bacterium is known to provide a fitness advantage in the mouse gut, as demonstrated by competitive fitness assays between fnr KO mutants and wild-type clones across different genomic backgrounds of E. coli [48]. In agreement with a fitness advantage conferred by this event, PCR typing revealed a steadily increase in the frequency of this event along time during the colonization of mouse 7 (Figs 4A and S7, and S2 Table). These results show that genes coding for functions not relevant outside the host can be regained due to HGT in the gut.

To directly measure the fitness benefit of a clone with this HGT event, we performed competitive fitness assays in new mice. We competed the evolved clone carrying the horizontally transferred fnr region against a lysogen of Nef and KingRac, given that these prophages confer an advantage by themselves [14]. We find a clear competitive fitness advantage of the evolved clone (Fig 4C and S9 Table), that together with the PCR typing (Fig 4A and S7 Fig) suggest a strong benefit for strain A driven by the genomic repair event.

Overall, our results show that HGT events of core bacterial DNA, even if rare are highly advantageous and able to restore in the gut previously lost genomic regions. We show here that this type of DNA transfer, can be detected in real-time during evolution in the mammalian intestine, underscoring its importance for the adaptation of E. coli to this environment.

In vivo mobilization of a putative phage satellite

The analysis of the genomes of clones that evolved in mouse 7 identified the transfer, from strain B1 to strain A, of a genomic island predicted to correspond to an incomplete prophage (S8 and S10 Tables). Although it contains an integrase, this region lacks any genes – e.g., capsids or tails - that correspond to the assembly of viral structures (S11 Table). On the other hand, we observed the presence of genes that are typically associated with phage satellites, elements that are known to mobilize by exploiting co-infecting phages. Specifically the gene ash, which has a homologue in the P4 satellite and is a core gene of the P4-like family of satellites [49]. Additionally, the gene alpA was also detected, which is a core gene in four different phage satellite families [50]. This island also contains a gene homologous of TerS, that although often found in phages, it is also one of the core genes of phage inducible chromosomal islands (PICI) [51] (S8 Fig, S11 Table). We used SatelliteFinder [50] to understand if this genomic island corresponds to any known phage satellite. Although there were no reliable matches to any of the known families, SatelliteFinder identified this genomic island as an incomplete P4-like or PICI, due to homology with the core genes described above. This island might thus represent a potentially novel family of satellites, likely mobilized by Nef, KingRac, or both, but further work needs to be done to understand its identity and mobilization. In either case, to the best of our knowledge, this would correspond to the first documented transfer of a phage satellite in vivo in the mammalian gut.

Conclusions

The potential for evolutionary change within bacterial species that colonize the mammalian gut is enormous, given the high population densities, large mutational inputs and the various selective pressures that occur in this ecosystem. Gene exchanges should also be a pervasive mechanism of evolution in the gut, given the high level of strain variation within each species, and the level of polymorphism of their MGEs. Using E. coli as a model for understanding the tempo and mode of evolution under strain coexistence, we show that strains with very distinct past histories and genome sizes stably co-exist in the gut and evolve under a strain-specific selection dynamics within the same host. Intra-strain convergent evolution was observed and large amounts of genetic information were exchanged from one strain to the other, via somewhat non-canonical HGT mechanisms, which can be caught within months in vivo. Such rare HGT events can be highly beneficial and can be captured as different co-existing strains co-evolve. Our results show that an unprecedented richness of evolutionary mechanisms and selection modes happens within the same host, when we follow the evolution of more than one strain, and that a typical gut resident strain can repair another strain that had experienced deleterious gene losses when outside the gut. Furthermore, our observations on convergent evolution during co-colonization and the large impact of HGT could be relevant for other bacterial species in natural populations [52,53]. Overall, our data highlights the complex combination of evolutionary forces that underly the genetic diversity of generalist and human commensal species such as E. coli.

Ethics statement

This research project was ethically reviewed and approved by the Ethics Committee of the Instituto Gulbenkian de Ciência (license reference: A009.2018), and by the Portuguese National Entity that regulates the use of laboratory animals (DGAV - Direção Geral de Alimentação e Veterinária (license reference: 008958). All experiments conducted on animals followed the Portuguese (Decreto-Lei nº 113/2013) and European (Directive 2010/63/EU) legislations, concerning housing, husbandry and animal welfare.

Escherichia coli strains

To facilitate the isolation from the mouse feces all the E. coli strains used in this study express fluorescent proteins and carry antibiotic resistance markers. The E. coli strain A express either a Yellow or a Cyan Fluorescent Protein, namely (YFP) or (CFP), respectively. Each one also carries a streptomycin resistance marker, as well as ampicillin (YFP) or chloramphenicol (CFP) resistance markers. Regarding the E. coli strain B1, these express either a Red (mCherry) or a Green Fluorescent Protein (sfGFP) and a chloramphenicol resistance marker. E. coli clones were grown at 37°C under aeration in liquid media Lysogeny Broth (LB) from SIGMA — or lactose supplemented McConkey agar plates and LB agar plates. Media were supplemented with antibiotics streptomycin (100 µg/mL), ampicillin (100 µg/mL) or chloramphenicol (30 µg/mL) when specified. Serial plating of 1X PBS dilutions of feces in LB agar plates supplemented with the appropriate antibiotics were incubated overnight and YFP, CFP, mCherry and sfGFP-labeled bacterial numbers were assessed by counting the fluorescent colonies using a fluorescent stereoscope (Zeiss Stereo Lumar V12). The detection limit for bacterial plating was ~ 300 CFU/g of feces [12]. To exclude that transfer of fluorescent markers between strains occurred, we examined the frequencies of strain-specific mutations unique to each fluorescent strain (S12 Table). The frequency of these mutations was consistent with their respective markers, suggesting that transfer of fluorescent markers between strains did not occur.