Gut microbe Lactiplantibacillus plantarum undergoes different evolutionary trajectories between insects and mammals

Background Animals form complex symbiotic associations with their gut microbes, whose evolution is determined by an intricate network of host and environmental factors. In many insects, such as Drosophila melanogaster, the microbiome is flexible, environmentally determined, and less diverse than in mammals. In contrast, mammals maintain complex multispecies consortia that are able to colonize and persist in the gastrointestinal tract. Understanding the evolutionary and ecological dynamics of gut microbes in different hosts is challenging. This requires disentangling the ecological factors of selection, determining the timescales over which evolution occurs, and elucidating the architecture of such evolutionary patterns. Results We employ experimental evolution to track the pace of the evolution of a common gut commensal, Lactiplantibacillus plantarum, within invertebrate (Drosophila melanogaster) and vertebrate (Mus musculus) hosts and their respective diets. We show that in Drosophila, the nutritional environment dictates microbial evolution, while the host benefits L. plantarum growth only over short ecological timescales. By contrast, in a mammalian animal model, L. plantarum evolution results to be divergent between the host intestine and its diet, both phenotypically (i.e., host-evolved populations show higher adaptation to the host intestinal environment) and genomically. Here, both the emergence of hypermutators and the high persistence of mutated genes within the host’s environment strongly differed from the low variation observed in the host’s nutritional environment alone. Conclusions Our results demonstrate that L. plantarum evolution diverges between insects and mammals. While the symbiosis between Drosophila and L. plantarum is mainly determined by the host diet, in mammals, the host and its intrinsic factors play a critical role in selection and influence both the phenotypic and genomic evolution of its gut microbes, as well as the outcome of their symbiosis. Supplementary Information The online version contains supplementary material available at 10.1186/s12915-022-01477-y.

In this light, several studies have sought to dissect the relative contributions of these factors in shaping the gut microbiota of insects and mammals. Among these, most have stressed the importance of the host's diet as a key force in determining the microbiota configuration in both invertebrates [65][66][67] and vertebrates [40,47,64,[68][69][70][71][72][73]. In mice, switching from a low-fat, plant polysaccharide-rich diet to a high-fat/high-sugar "Western" diet can shift the microbiota structure within a single day [47], causing a progressive loss of species diversity over generations [64,72]. Importantly, such diet-mediated microbiota alterations can ultimately result in specific microbiota-host layouts, which, in turn, affect host health and disease [69,74].
Although these investigations have undoubtedly broadened our understanding of the diversity, resilience, and complexity of the gut microbiota across animal hosts, most of them have focused on characterizing how these external and internal factors shape microbiome compositional and functional features, eventually linking the resulting microbial pattern with a specific host trait (i.e., health or disease condition). Furthermore, by using common biomarkers (i.e., 16S rRNA), most of these studies have profiled the bacterial diversity at the genus or species level, therefore masking the potential presence of dynamic and rapidly evolving sub-populations. As a consequence, much less is known about the microbial evolutionary processes in the gut across animals. In vivo experiments, combined with deeper genome analyses, have recently brought forth a new appreciation of the gut microbes' capability of rapidly diversifying and adapting in a newly colonized environment over short [75][76][77] and long timescales [78], providing insights into the ecological and molecular mechanisms underlying such evolutionary paths [79][80][81][82][83][84]. In this context, Escherichia coli is the most widely used model bacterium for studying bacterial evolution in the mammalian gut [48,76,77,80,85]. The reasons are manifold and linked to its ecological and clinical relevance, together with ease of experimental and genetic manipulation and tractability. However, further progress into understanding the drivers of microbial evolution in the gut of different animal hosts requires us to move beyond focusing on this particular species and to look at evolution in real time across a broader range of species. This is crucial for determining the main factors governing the evolution of gut microbes.
Here, we use Lactiplantibacillus plantarum, a common inhabitant of the gastrointestinal tract of different animals [86], as model species to explore the evolutionary trajectories of gut microbes across animal hosts, and particularly if and how they differ between insects and mammals.
By using Drosophila melanogaster as an animal model, we previously demonstrated that the host's diet, rather than the host environment per se, is the predominant force in driving the emergence of the symbiosis between L. plantarum and the fruit fly [87]. Here, we hypothesize that, given the higher persistence and colonization ability of gut microbes in the gastrointestinal tract of mammals, the mammalian host and its intrinsic factors represent key agents of selection in the evolution of gut microbiota as compared to the mammalian host's diet. To explore this, we performed a parallel experimental evolution of the bacterial strain L. plantarum NIZO2877 , which was previously shown to moderately promote growth both in Drosophila and mice [88]. We evolved L. plantarum in mono-association with germ-free C57Bl6 mice and in the mouse laboratory diet, separately. At the same time, we experimentally evolved the same strain in association with Drosophila and its nutritional environment to test if and how the presence of the invertebrate host affects the tempo and mode of evolution of its gut microbiota.
Our results indicate that the evolution of the same gut bacterium diverges between insects and mammals, pointing out the effects of the different host-derived selection pressures. While, in Drosophila, the host seems to benefit the fitness of L. plantarum in a short timescale, without significantly affecting its evolutionary trajectory, microbiome evolution in mammals follows a completely different path. Here, host factors represent a crucial agent of selection, shaping the evolution of gut microbes both on a phenotypic and genomic level. This results in an increased bacterial adaptation toward the host's intestinal environment, revealing new insights into the symbiosis between L. plantarum and mammalian hosts.

L. plantarum evolution within a mammalian host leads to higher adaptation to the host intestinal environment
With the aim of investigating the evolutionary dynamics of L. plantarum in a mammalian host and to dissect the role of the mammalian host's diet in the evolution of gut microbes, we experimentally evolved the bacterial strain L. plantarum NIZO2877 (Lp NIZO2877 ) in the intestine of germ-free (GF) C57Bl6 mice (Host setup) and in the mouse laboratory diet (Diet setup) separately (Fig. 1A), following the same experimental setup that we have previously applied to Lp evolution in Drosophila [87]. Specifically, in the Diet setup, we mono-associated the ancestral strain with the mouse diet in the absence of the host and monitored microbial evolution for 20 transfers (T) in five independent replicates (i.e., 20 days, corresponding to ~400 bacterial generations). In the Host setup, we monocolonized 7 GF mice housed in cages in one Trexler-type isolator with the ancestral strain by intragastric gavage; once the mono-association had been performed, 4 female mice were bred together in one cage to generate F0 generation for the next 10 months and 2 females and 1 male were the F0 founders of the subsequent mouse generation. Bacterial evolution was followed across F0 and the subsequent generations of mice (F1, F2, F3, and F4) for 10 months (i.e., ~286 Lp bacterial generations; Fig. 1A, B). Evolving bacteria were horizontally dispersed and vertically transmitted across generations with no further artificial inoculation. Over this period, the mice were fed with the same diet used for the Diet setup.
Overall, L. plantarum growth showed a similar trend across the two evolutionary setups. In the mouse diet, we found that the microbial load increased extremely fast, as it shifted from the initial inoculum of 10 2 total CFU/mL to 3.34 × 10 8 CFU/mL in less than 24 h (Fig. 1B). Such a growth trend was maintained over the course of the evolution experiment, during which the bacterial load ranged between 10 7 and 10 9 CFU/mL, with the lowest value at T9 (i.e., after ~180 Lp generations, mean T9 = 2.06 × 10 7 CFU/mL) and the highest at T2 (i.e., ~40 Lp generations, mean T2 = 2.12 × 10 9 CFU/mL) (Fig. 1B). Microbial growth within the mouse intestine was approximately lower of 1 log CFU than that observed in the mouse diet and showed comparable loads between 50 and 200 bacterial generations, reaching the highest peak after ~110 Lp generations. Moreover, within each mouse generation, no significant differences were detected when comparing the initial and final bacterial concentrations (Fig. 1B). These results suggest that L. plantarum was able to reach and maintain high abundance both in the mouse diet and in the mouse intestine.
To compare L. plantarum phenotypic evolution with and without the mammalian host, we conducted a morphological analysis of the Lp-evolved colonies over time in both experimental setups. Interestingly, we detected the appearance of an evolved L. plantarum sub-population showing a different colony morphology compared to the ancestral strain in both evolutionary setups. Specifically, while L. plantarum typically forms rounded, smooth colonies on MRS agar, the newly evolved colonies showed a less-defined, rough morphology, which looked more transparent than the ancestral one (Additional file 1: Fig. S1). Such a population appeared after ~57 Lp generations in the Host setup (F0) and after ~280 Lp generations in the Diet setup (T14). In the mouse intestine, it initially affected 48% of the whole bacterial population, but it tended to decrease over time, reaching 38% of the population during mice Generation 3 (i.e., ~140 Lp generations, Additional file 1: Fig. S1). To further characterize L. plantarum phenotypic change, we analyzed the two bacterial morphotypes with an electron microscope. Interestingly, while the ancestral-like colonies showed a compact bacillary structure with a smooth surface, the newly evolved rough colonies had an irregular surface and looked elongated and filamentous (Additional file 1: Fig. S1B, C). However, it was unclear whether such longer chains resulted from undivided bacterial cells or whether they corresponded to single cells. Notably, when a rough colony was re-streaked on MRS agar or cultured in MRS broth, the morphology reversed to a smooth, ancestrallike one, showing that the newly evolved phenotype was transient and reversible (data not shown).
We next sought to investigate the mechanisms underlying the emergence of the new microbial morphotype. Bacterial morphological changes commonly occur as a result of stress responses to a wide range of factors [89][90][91][92]. In the mammalian intestine, one of the most stressful conditions is due to the activity of bile acids (BA) [89,[93][94][95][96]. We thus hypothesized that the presence of bile acids encountered during transit through the mouse gastrointestinal tract might have contributed to the appearance of the newly evolved rough morphology. On the contrary, the appearance of the rough morphology in the Diet-evolved populations could not have resulted from a stress response to bile acids, since they are absent in the mouse diet. As a consequence, we expected that both Diet-evolved colonies (rough and smooth morphotypes) would be equally affected by the presence of bile acids.
To verify these hypotheses, we first tested whether and how bile acids were able to affect L. plantarum growth. We serially propagated the ancestral strain Lp NIZO2877 in MRS broth and MRS broth with the addition of 0.3% BA for 7 days and measured microbial growth. As expected, bacterial loads were significantly lower in presence of 0.3% BA compared to the control (MRS) already after one transfer (Lp T1-MRS+0.3%BA = 4.07 L. plantarum evolution in mice leads to higher adaptation to the host intestinal environment. A Schematic representation of the Lp NIZO2877 experimental evolution (EE) protocols in the mouse diet (Diet setup) and mouse intestine (Host setup). The ancestral strain was inoculated in the mouse laboratory diet and in the mouse intestine by intragastric gavage. Five and two replicates of EE were performed for the Diet and the Host setup, respectively. An evolved bacterial subpopulation was plated out on MRS agar plates for colony counting at different time points throughout the EE in both setups. The EE in the Diet setup was conducted up to 20 transfers (i.e., ~20 days, corresponding to approximately 400 Lp generations). In the mouse intestine, Lp EE was carried out from generation 0 (F0) to 4 (F4) of mice (i.e., 10 months after the mono-association, ~300 days, corresponding to ~286 Lp generations) for one replicate and for one generation (F0) (3 months after the mono-association) for the second replicate. Four female mice were housed in a single cage. One male and two female mice were housed in a separate single cage. BLp NIZO2877 growth trend monitored over the course of the EE in both experimental setups (Host and Diet). C Final absorbance values (OD = 600) reached by the Lp strains under standard growth conditions (MRS broth) and in MRS broth added to with 0.3% bile acids (BA). S indicates smooth colonies, while R indicates rough colonies. Each dot represents the mean of at least three experimental replicates per condition, with bars indicating the respective SD (standard deviation). For both experimental conditions (MRS broth and MRS broth + 0.3% BA), asterisks indicate significance between the final absorbance values of each strain against those of the Lp NIZO2877 ancestor (unpaired t test; *p ≤ 0.05, **p < 0.01, and ***p < 0.001). D Quantitative PCR analysis of Lp NIZO2877 ancestral strain (Anc) and Lp NIZO2877 -derived population (Evo) evolved in the mouse intestine. Each bar represents the standard error of the mean (SEM) of normalized ∆C T ratios (1/∆C T ) of 3-4 mice/group. Statistical significance of the results is included (unpaired t test, **p≤0.01) × 10 6 CFUs; Lp T1-MRS = 1.75 × 10 8 CFUs; unpaired t test ***p<0.001) throughout the experiment (Lp T7-MRS+0.3%BA = 8.06 × 10 7 CFUs; Lp T7-MRS = 3.14 × 10 11 CFUs) (Additional file 2: Fig. S2). By monitoring the bacterial morphology on agar plates, we noticed that, although the Lp colonies grown in the presence of BA looked slightly smaller and more transparent, the colonies maintained the ancestral smooth morphotype over the seven serial transfers (data not shown). These results demonstrate that bile acids are able to impair L. plantarum growth.
Next, we analyzed the fitness of Diet-and Hostevolved bacteria in the presence of bile acids and detected a Lp-specific response to BA depending on potential differences in the genetic background and the evolutionary history. In detail, all rough colonies reached the lowest final absorbance values in the presence of bile acids, regardless of their evolutionary history (i.e., Host-or Diet-evolved) (Fig. 1D, Additional file 3: Fig. S3 and Additional file 4: Fig. S4). They also showed a slower and delayed growth in standard MRS broth compared to the ancestor (Additional file 3: Fig. S3). Remarkably, while the Diet-evolved smooth morphotypes reached a significantly lower absorbance value compared to the ancestor (Additional file 3: Fig. S3), we detected signatures of adaptation toward bile acids among the Host-evolved smooth colonies (F0, F3). Such colonies reached, overall, the highest absorbance values in the presence of bile acids (Fig. 1C), suggesting higher tolerance to the stress compound.
To investigate whether L. plantarum populations evolved in the mouse intestine exhibited signs of adaptation to the animal host, mice bearing a conventional gut microbiota were gavaged either with the Lp NIZO2877 ancestral strain or with the Lp NIZO2877derived population evolved in the mouse intestine for 10 months (sample F0-10). Whole bacterial DNA was isolated from feces and real-time PCR was performed to track L. plantarum persistence over time (i.e., up to 72 h). Notably, we observed a significantly higher persistence of the L. plantarum-evolved population compared to the ancestral one until 24 h after gavage, while no L. plantarum was detected after 36 h (Fig. 1D).
Altogether, our data demonstrate that L. plantarum populations evolved in the mouse intestine showed adaptation towards the animal host. This, among other factors, may result from the increased tolerance to intestinal stress (i.e., presence of bile acids) and leads to higher persistence in the host intestinal environment.

Bacterial evolution within the mammalian host intestine is shaped by the emergence of hypermutators and a higher number of mutations compared to the evolution in the host diet
To investigate the influence of the mammalian animal host and its nutritional environment on the genomic evolution of its symbiotic bacteria, we sequenced the genomes of whole bacterial populations evolved in the mouse diet and in the mouse intestine at different time points. Specifically, we sequenced the genomes of fifteen L. plantarum-evolved populations isolated from feces pooled from 3-4 individual mice across the five generations from one replicate of experimental evolution (EE) (Additional file 8: Table S1). As for the Diet setup, we sequenced the genomes of six evolved bacterial populations isolated during transfers 2 and 14 from three independent replicates out of five (Additional file 8: Table S1). In addition, we sequenced the genomes of two single bacterial colonies: one showing the smooth, ancestrallike phenotype and the second one showing the newly evolved rough morphology (replicate 2, transfer 14, Additional file 8: Table S1).
Bacteria propagated in the mouse diet showed overall a low number of mutations, with the highest value in T14 (N= 10 mutations) ( Table 1, Fig. 2A, Additional file 9: Table S2).
During this time point, the single smooth and rough colonies showed the emergence of only 1 and 3 mutations per strain, respectively (replicate 2, Additional file 10: Table S3). On the contrary, bacteria evolved in the mouse intestine revealed a marked increase in the total number of mutations compared to the Diet-evolved ones ( Table 1, Fig. 2A, B, Additional file 11: Table S4). The high number of genetic variants detected in the Host-evolved Lp populations correlated with the appearance of novel mutations in the genes encoding for DNA replication and repair proteins mutS and mutL (methyl-directed mismatch repair complex subunits), dnaE (DNA polymerase III alpha subunit), and dinB (DNA polymerase IV) (Additional file 11: Table S4).
Mutations in the mouse host populations were overwhelmingly A:T to G:C base pair substitutions, whereas these substitutions accounted for less than 10% of mutations in the mouse diet (Fig. 2C). An elevated A:T to G:C mutational bias is consistent with mutations in mutS or mutL [97] but not dinB [98] or dnaE [99], so we concluded that defects in methyl-directed mismatch repair are probably largely responsible for the hypermutator phenotype. To test whether hypermutation of L. plantarum in mouse was repeatable, we sequenced the whole genome of two Lp populations evolved in the second replicate of experimental evolution in the mouse intestine (F0-sequenced time points: 1, 2 months). Here, we detected 94 and 240 mutations, respectively, which included five additional variants in the mutS gene (Additional file 12: Table S5). This strongly suggests that hypermutation is a common evolutionary strategy of L. plantarum to adapt to the mouse intestine.
Whole genome sequencing of the first replicate of Lp EE in mice detected five separate mutations in the mutS gene and one in the mutL gene (Fig. 2D, Additional file 11: Table S4). The earliest of these mutations are a base substitution that causes an A41T amino acid substitution in mutS and a deletion of base pair 1303 of the mutS coding sequence that results in a frameshift. Secondary mutations affecting methyl-directed mismatch repair and mutations in genes involved in other DNA replication and repair processes appeared later in lineages that already had one of these two mutS mutations. These later mutations may have further modulated mutation rates in some cases, but we expect that these initial mutS mutations are responsible for most of the changes in mutation rates and spectra in the evolution experiment. In at least one case, we can predict that a secondary mutation had no effect. A point mutation in mutS that would lead to a D761Y amino acid substitution tracked with that of the Δ1303 mutation as soon as they both reached observable frequencies (Fig. 2D). Since this D761Y mutation is located after a new stop codon in the mutS gene created by the Δ1303 mutation at codon 456, it would not affect the mutS gene product in the context of this mutation. Therefore, we considered the A41T and Δ1303 mutations as defining two distinct mutS hypermutator lineages.
The two mutS A41T and Δ1303 hypermutator lineages and a nonmutator lineage that maintained the ancestral mutation rate competed throughout the history of the evolution experiment in different animals (Fig. 2D). The mutS A41T lineage appeared first and increased in frequency over the initial three months (63 generations) in the F0 mouse, reaching up to a 90% frequency. Between the third month (F0-3) and the tenth month (F0-10) however, this lineage fell below the level of detection and the mutS Δ1303 lineage rose to near 100% frequency. In the F3 and F4 mouse generations, the mutS Δ1303 lineage also ultimately swept to 100% frequency. In the F1 mouse however, neither lineage exceeded 20% frequency while, in the F2 mouse generation, both mutS lineages increased to ~40-50% frequency between generations 82 (F2-1) and 104 (F2-2) but then declined to less than 3% frequency by generation 123 (F2-3).
By fitting a model that assumed both mutS alleles evolved near the beginning of the experiment and accounted for a certain fraction of the population at each time point, we were able to estimate that the mutS A41T lineage accumulated mutations at a rate of 2.453± 0.146 per generation and the mutS Δ1303 lineage accumulated mutations at a rate of 0.310 ± 0.060 per generation (± values are standard errors of fit values). These rates are 323-and 41-fold the rate in the mouse diet treatment. The roughly 8-fold factor by which the two mutS lineage rates differ from one another is significant (F-test, p = 2.0 × 10 −8 ). It may indicate that the mutS alleles have different effects on their own or in combination with mutations in other genes. For example, the dnaE mutation tracks with the mutS A41T allele throughout its entire history, and the dinB mutation appeared in this genetic background later in the F4 mouse (Fig. 2D).
By comparing the genetic variants detected in the Host-and Diet-evolved populations, we identified 5 genes that mutated in both evolutionary setups ( Fig. 2A, Additional file 9: Table S2 and Additional  file 11: Table S4). However, none of them persisted across generations of both conditions. We then turned our attention to those targets that exclusively mutated in the Host and Diet setups and that showed persistence across EE cycles. While within the Diet-evolved populations, we only detected one mutation (i.e., gene: dipeptidase, Additional file 9: Table S2), the Lp populations evolved in the mouse intestine showed a significantly higher number of mutations (n = 1169; Additional file 11: Table S4), which were detected in 41 genes (and six intergenic regions). To understand whether such variants were linked to specific functions, we clustered them according to their predicted Generations Total mutation frequency Mouse generation   Table S6). Notably, four mutational targets belonging to this functional category included genes encoding for phosphotransferases system (PTS) transporters (PTS-mannose-specific IIC, PTSsucrose-specific IIA/IIB/IIC, phosphocarrier protein of PTS system, and PTS system-beta-glucoside-specific IIB/IIC/IIA component). In addition, the genes belonging to the signal transduction and metabolism (n = 5) and transcription (n = 5) categories resulted to be affected by mutations in the Host-evolved populations (Fig. 3, Additional file 13: Table S6). Altogether, our results demonstrate that L. plantarum follows divergent evolutionary paths in the mouse intestine compared to the mouse diet, both genomically and phenotypically. While L. plantarum evolution in the mouse diet was characterized by low mutational load, hypermutators were detected during evolution in the mouse intestine. In addition, L. plantarum populations evolved inside the host exhibited higher adaptation to host intrinsic factors (i.e., presence of bile salts).

Drosophila melanogaster benefits L. plantarum growth on a short timescale
The ecological and evolutionary dynamics of gut microbes vary greatly between mammals and insects. Among many factors, this is largely due to the different extent through which microbes colonize the gut of animals and, as a consequence, the different degrees of dependence between animals and microbes across hosts [10, 20-26, 28, 86, 100]. By using Drosophila melanogaster as an animal model, we have previously demonstrated that, in Drosophila/L. plantarum symbiosis, the host nutritional environment, rather than the host per se, is the predominant force in driving the emergence of such symbiosis [87]. The experimental setup was comparable to the one used in the present study to experimentally evolve L. plantarum in the mouse intestine and in the mouse diet. Specifically, in the Host setup, bacteria were horizontally and vertically transmitted among individuals and across generations, while in the Diet setup, artificial passages of evolving bacterial populations were conducted through experimental generations [87]. We then asked whether such differences in the selection regime between setups (natural transmission of bacteria in the Host setup versus artificial inoculation in the Diet setup) might have influenced the microbial evolutionary dynamics both in the fruit fly and in mice. To test this, we decided to replicate the Lp experimental evolution with and without Drosophila (Host and Diet setups, respectively) for a total of 20 cycles (220 days, corresponding to ~1760 bacterial generations) by applying the same transfer and sampling time for both setups (five independent replicates per setup, Fig. 4A). In this way, we were able to minimize differences between setups so that the microbial evolutionary trajectories relied exclusively on the presence/absence of the animal host.
We first investigated whether the presence of Drosophila affected Lp NIZO2877 evolution by conducting morphological evaluations of the evolved bacterial populations and analyzing the microbial growth dynamics throughout the two experimental protocols. Specifically, the evolved bacterial populations in the ten . For the first EE cycle, the ancestor strain (Lp NIZO2877 ) was inoculated into tubes containing a poor-nutrient Drosophila diet (Diet setup) or a poor-nutrient diet containing 40 germ-free Drosophila embryos (Host setup). No further inoculation of the ancestor was performed until the end of the experimental evolution. As soon as at least 15 pupae emerged from all host tubes (i.e., after ~11 days, corresponding to ~88 bacterial generations), 150 μl of food was collected from both setups using a sterile loop, homogenized, and plated out to isolate bacteria (frozen "fossil" records of EE cycle 1). This bacterial population was used as the inoculum for the following generation/ transfer. Subsequent EE cycles followed the same experimental procedure as cycle 1 and started from the fossil records belonging to the previous generation/transfer. The EE lasted 20 cycles (220 days, corresponding to ~1760 bacterial generations). BLp growth over the course of the Host and Diet EE protocols across 20 total EE cycles (i.e., 1760 bacterial generations). Each point represents the mean of the five experimental replicates, with bars indicating the standard error of the mean (SEM). ANCOVA ***p < 0.0001). C Microbial load obtained by mono-associating each of the five replicates of both setups (Host-and Diet-evolved bacteria) isolated from cycle 17, both with and without the host. Each bar represents the standard error of the mean (SEM) obtained by analyzing five replicate populations for each condition. Ordinary one-way ANOVA (*p ≤ 0.05, **p <0.01, and ***p < 0.001) independent replicates have been plated out at the end of each experimental cycle (total of 20 cycles), macromorphological evaluation of bacterial colonies was routinely performed and microbial load was measured at the end of each cycle (~11 days). Contrary to what we observed during L. plantarum evolution in the mouse intestine, no differences in colony morphology were observed across the Lp populations evolved with and without Drosophila. Remarkably, the microbial load was significantly higher overall in the absence of the host, except for in the first generation (i.e., 88 bacterial generations) (Fig. 4B). Since the difference in microbial growth between the two setups was detected at a specific time point (i.e., the end of each experimental generation-11 days), we asked whether the higher microbial load observed in the Diet setup resulted from a delayed growth dynamic compared to that of the Host-evolved populations. To address this question, we analyzed the microbial concentration at an earlier time point (i.e., 7 days after the mono-association). The Lp concentration was confirmed to be overall significantly higher in the absence of the host, except for generations 6, 19, and 20 (i.e., after ~528, 1672, and 1760 bacterial generations, respectively) (Additional file 6: Fig. S6).
Interestingly, throughout our sampling period, we also noticed an unexpected correlation between the growth dynamics of the Host-and Diet-evolved populations, which was particularly pronounced from generations 6 to 12 (i.e., from 528 to 1056 Lp NIZO2877 generations) (Fig. 4B). To test the repeatability of our findings, we replayed L. plantarum EE from cycles 5 to 12 in both setups and analyzed the microbial concentration after 7 and 11 days. Our results further confirmed the correlation in microbial growth dynamics between the Host and Diet setups (Additional file 6: Fig. S6). This demonstrates that Lp growth dynamics are repeatable and did not result from experimental artifacts or external variables.
We then asked whether and how Lp evolutionary history (i.e., evolving in the presence or absence of its host) could generate evolutionary tradeoffs in a different environment. To this end, we mono-associated Lp Host-and Diet-evolved populations isolated at the end of EE cycle 17 from each of the ten independent replicates (five replicates per setup), both in the presence and absence of Drosophila and analyzed the microbial load in both conditions after 11 days. Lp concentration was always significantly higher in the presence of Drosophila (Fig. 4C). This was also visible after cycle 1 of Lp EE (i.e., ~88 Lp generations; Fig. 4B). However, when comparing the microbial evolutionary backgrounds, the Diet-evolved populations reached significantly higher loads compared to the Host-evolved populations when associated with the fruit fly (Fig. 4C). Taken together, our results show that, although Drosophila benefits L. plantarum growth on a short timescale, bacterial evolution ultimately leads L. plantarum to grow to a higher extent in the absence of Drosophila.

Genome sequencing reveals parallel genomic evolution between Lp populations evolved with and without Drosophila
We next investigated if and how Drosophila influences Lp evolution on a genomic level. To do this, we performed metagenomic sequencing of bacterial populations isolated from both experimental setups during EE cycles 2, 8, 14, and 20 (three independent replicates sequenced per time point and setup, Additional file 8: Table S1). Across the two evolutionary setups, we detected a similar mutational trend in terms of the number of genetic variants ( Table 2, Fig. 5A, B) and identified signatures of strong parallel genomic evolution. Bacteria evolved in the fly host populations accumulated 0.00177 ± 0.00018 mutations per generation and fly diet populations accumulated 0.00150 ± 0.00010 mutations per generation (± values are standard errors of fit values). These rates did not differ significantly from one another (F-test, p = 0.059). Mutations in the fly host and diet populations were mainly C:G to A:T and C:G to T:A substitutions (Fig. 5C).
Three genes mutated in both setups (i.e., ackA, rho, and gltL, Fig. 5D, Additional file 14: Table S7 and Additional  file 15: Table S8). Specifically, all Lp experimental replicates isolated from the Host and Diet setups showed at least one genomic change within the acetate kinase A gene (ackA), which was the first gene to be affected by mutation and exhibited the highest number of mutations per gene (Fig. 5D, Additional file 14: Table S7 and Additional file 15: Table S8). This result confirms our previous work showing that mutations of L. plantarum ackA occur both in the presence and in the absence of Drosophila. We identified 12 non-synonymous ackA mutations, one of which was shared between the Host and Diet setups. The shared variant appeared during EE cycle 2 (i.e., after 176 bacterial generations), reaching fixation in all Dietevolved populations (Fig. 5D, Additional file 14: Table S7 and Additional file 15: Table S8), while it disappeared in the Host-evolved populations. Here, it was replaced by multiple additional variants within the same gene. At the same time, we detected mutations that exclusively occurred in each evolutionary setup. Specifically, 13 genes mutated in the Host-evolved populations at least once, while 9 exclusive mutational targets were detected in the Diet-evolved populations (Fig. 5A, D, Additional file 14: Table S7 and Additional file 15: Table S8). Altogether, our results show that L. plantarum undergoes parallel genomic evolution with and without its invertebrate animal host, further demonstrating that, contrarily to what we observed in L. plantarum/mouse symbiosis, in the relationship between L. plantarum and the fruitfly, the host nutritional environment largely dictates the microbial evolutionary trajectories both genomically and phenotypically.

Discussion
Given the complexity of host-microbial symbioses and the high variability that characterizes such associations in natural environments, understanding the evolutionary trajectories of gut microbes across different symbiotic relationships is of particular interest. In this work, we coupled experimental microbial evolution with genome sequencing and phenotypic characterization to explore the evolutionary path of Lactiplantibacillus plantarum, a common animal gut commensal, in vertebrate (mouse) and invertebrate (Drosophila melanogaster) animal models. We previously demonstrated that, in the symbiosis between Drosophila and L. plantarum, the nutritional environment is the main selective agent in the evolution and adaptation of the commensal bacterium [87]. Here, we asked if and how L. plantarum evolution varied in the context of its symbiosis with mammalian hosts, where gut commensals exhibit higher levels of colonization and mutualistic interactions are reported as more persistent than those observed in insects [23,32]. Although diet is well known to largely dictate gut microbes' evolution in mammals [47,84], disentangling the respective roles of nutritional and host factors in such processes is challenging. To address this, we tracked L. plantarum phenotypic and genomic evolution in the mouse intestine and in their diet. L. plantarum growth dynamics in the presence and absence of the mammalian animal host were similar between the two evolutionary setups (Fig. 1B). We believe that, in the Diet setup, the fast and high microbial persistence can be explained by considering the structure of our experimental setting (i.e., bacterial reinoculation at the beginning of each EE cycle), a result that was not expected in the presence of the host. Here, bacterial administration was performed at the beginning of the experimental evolution and bacteria were transferred vertically and horizontally among mice, without any further external input. In addition, the complex physical and chemical conditions encountered during transit through the mammalian gastrointestinal tract usually provide a challenge to the commensal microbiota attempting to colonize the host GI niche [101,102]. Here, the innate immune system of the host, the food transit, and the plethora of antimicrobial compounds secreted into the gut (including hydrochloric acid, bile, hydrolytic enzymes, and antibodies), as well as other ecological factors, such as the intense microbial competition for space and nutrient resources, represent obstacles to both bacterial colonization and survival [103,104]. We speculate that the high bacterial concentration observed in the Host setup is directly linked to the use of monocolonized mice. It will be interesting to assess the Lp evolutionary growth dynamics in mice harboring a complex microbial community. L. plantarum adaptation to the mammalian host was also visible by the increased tolerance under bile acid stress and the higher persistence in the mouse intestine exhibited by the Lp populations evolved in the Host setup (Fig. 1C, D, Additional file 3: Fig. S3 and Additional file 4: Fig. S4). This suggests that, while facing the adverse conditions along the mouse GI tract, some bacterial subpopulations gradually adapted to the host intrinsic factors, resulting in improved growth, fitness, and colonization ability.
During Lp experimental evolution in the mouse setup, we also detected a morphological transition of bacterial colonies, characterized by the emergence of a rough phenotype that occurred both in the Host and Diet setups. Morphological transitions within a single strain population, known as pleomorphism [105], often occur in bacteria evolution and have been reported by several studies as an adaptive strategy for survival in response to fluctuating environments, especially to limiting or changing growth conditions. Moreover, L. plantarum had already been shown to exhibit a rough surface and cell elongation under a wide range of stress environments, including heat and ethanol shocks [106][107][108], presence of nitrite [109], low pH [91,107], lactic acid stress [110], nutrient stress [111], and bile stress [89]. Interestingly, when observed with an electron microscope, the rough colonies exhibited a filamentous and chaining phenotype (Additional file 1: Fig. S1B, C). In this regard, it is worth noticing that the Diet-evolved strains exhibiting the rough morphology showed a mutation in a gene encoding for the cell division protein DivIVA (Additional file 10: Table S3). DivIVA is a coiled-coil protein first discovered in Bacillus subtilis and highly conserved among Gram-positive bacteria. It clusters at curved membrane areas such as the cell poles and invaginations that occur during cell division, where it serves as a scaffold protein for the recruitment of Min proteins, which spatially regulate the division process [112,113]. In contrast to B. subtilis, where the deletion of the divIVA gene was responsible for the formation of a filamentous and mini-cell phenotype [114], ΔdivIVA mutants of Listeria monocytogenes exhibited a pronounced chaining phenotype [115]. Although these cells had clearly completed cell division, they remained attached even after completion of crosswall synthesis. This indicates that the deletion of divIVA, although not affecting cell division per se, might affect the post-divisional separation of daughter cells. However, further evidence of divIVA functioning within L. plantarum species is needed to assess our hypothesis, considering that the functionality of divIVA has been shown to be species-specific [113]. We next investigated the cause of the emergence of the rough morphotype. Even if we cannot assess whether the pleomorphism that emerged from both the Diet-and Host-evolved rough colonies was due to the same stress factor, it is interesting to note that all rough colonies had reduced vitality and increased susceptibility to bile acids stress compared to the smooth ones, regardless of their evolutionary history (i.e., Hostor Diet-evolved) (Fig. 1C, Additional file 3: Fig. S3 and Additional file 4: Fig. S4). This is consistent with other findings, showing that the gradually increasing severity of changes in L. plantarum morphology coincide with a respective decrease in the bacterial growth rate [89].
Genome sequencing of Lp evolved with and without its mammalian host revealed weak parallelism. The Dietevolved populations showed a significantly lower number of mutations, among which only one persisted over time (Dipeptidase gene). On the contrary, we detected several mammalian host-specific mutational targets, most of which are involved in carbohydrate transport and metabolism (Fig. 3, Additional file 4: Fig. S4). Among these, four genes belong to the phosphotransferases system (PTS), a highly conserved bacterial phosphotransferase cascade whose components modulate many cellular functions in response to carbohydrate availability [116] and which has already been observed to be over-expressed or mutated in other mouse gut colonization experiments [76,86,117]. This is in line with a recent study showing that L. plantarum convergently evolves across vertebrate animal hosts (i.e., human, mouse, zebrafish) by acquiring mutations primarily modulating carbohydrate utilization and acid tolerance [118].
Notably, Lp genomic evolution within the mouse intestine was repeatedly characterized by the emergence of hypermutators carrying multiple mutations in the mutS, dnaE, and dinB genes (Fig. 2B, D), which were not observed in any of the Diet-evolved populations. These genes are involved in the DNA mismatch repair system and DNA replication accuracy [119][120][121]. Mutations in these regions can lead to up to a 100-fold increase in the spontaneous mutational rate compared to their wild-type counterparts [122][123][124][125] and have already been observed in clinical, environmental, and laboratory microbial populations, suggesting that the evolutionary strategies of bacteria include systems for increasing mutability [126,127]. In the context of the mammalian gut, most of the experimental research has been carried out using E. coli. With this species, spontaneously arising mutator bacteria can also quickly become dominant during the course of gut colonization [76,80,83,85,128]. Such an advantage depends on the ability of the hypermutators to generate adaptive mutations rather than on the beneficial pleiotropic effects of the mutator allele, suggesting that adaptive mutations are fixed rapidly in mutator populations. Indeed, while in stable environments the maintenance of a low mutational rate is fundamental to avoid deleterious mutations that might lead to species loss, several experimental studies, performed both in vitro [129,130] and in vivo [85,131], have shown that increased mutational rates can be beneficial to bacterial populations facing unpredictable adverse conditions, where mutations might help cells to overcome selective pressures [126,132]. At the same time, it has been shown that a high mutation rate can be initially beneficial because it allows faster adaptation, but this benefit disappears once adaptation is achieved [80,85]. Although we did not assess whether the high mutation rate observed within the Host-evolved Lp populations is directly responsible for adaptive mutations in the mouse intestine, our results showed that the hypermutator lineages persist longer in the mouse intestine compared to the ancestral population (Fig. 1D). Nevertheless, the elevated mutation rate, as well as other events commonly observed in studies of gut microbe evolution (e.g., clonal interference, epistasis, horizontal gene transfer) might mask signatures of adaptation, making it difficult to separate selection from drift [77,80,85,133,134]. From this standpoint, understanding the role of persisting mutations and their effect on gut microbiota fitness and physiology might be of great interest to further dissect their consequences on the evolution of the symbiosis between bacteria and its mammalian animal host.
The divergent trajectories detected by comparing L. plantarum evolution in a mammalian host and in its diet are in contrast with L. plantarum evolutionary dynamics observed in Drosophila, where we have previously demonstrated that gut microbes undergo parallel evolution in the presence and absence of the fruit fly [87]. This indicates that selection regimes were comparable in the two environments. Specifically, in the Host setups, bacteria were horizontally and vertically transmitted among hosts (both in mice and Drosophila), while in the Diet setups, artificial passages of the evolving bacterial population were performed. However, we acknowledge that the experimental differences in bacterial propagation between the Host and Diet setups might not allow a direct comparison of the bacterial evolutionary dynamics.
To address this issue, in this work, we replayed L. plantarum experimental evolution in Drosophila by applying the same transferring and sampling time in both setups. This allowed us not only to demonstrate that the parallelism between Lp populations evolved in the presence of Drosophila and in its nutritional environment is repeatable, regardless of the experimental selection regimes used, but also to provide new insights into the respective role of the invertebrate host in the evolutionary path of L. plantarum. Indeed, we detected host-specific mutations that occurred in a later stage of bacterial evolution and showed persistence across EE cycles (Additional file 15: Table S8.). These genes mainly belong to bacterial immune evasion pathways, amino acid transport, and metabolism. Specifically, Lp Host-evolved populations were repeatedly affected by non-synonymous mutations of the mprF gene, which encodes for an L-Olysylphosphatidylglycerol synthase, an enzyme that is present in both Gram-positive and Gram-negative bacteria [135] and catalyzes the transfer of a lysyl group to the negatively charged phosphatidylglycerol (PG), a major component of the cytoplasmatic membrane [136,137]. This reaction modifies the net charge of PG, neutralizing the membrane surface and thus significantly impacting the interactions with cationic antimicrobial peptides (CAMPs) produced by the host's immune system (defensins and cathelicidins). Accordingly, the loss of Lys-PG in mprF mutants has been shown to lead to an increase in bacterial susceptibility to a broad variety of cationic antimicrobial peptides [138] in different bacterial species [139][140][141], thereby demonstrating a general role of mprF in bacterial immune evasion. In addition, non-synonymous Host-specific mutational targets also include glnQ, encoding a glutamate transport ATP-binding protein, which is involved in glutamate uptake in other Gram-positive bacteria [142], and glnA (encoding for a glutamine amidotransferase), involved in amino acid transport and metabolism. Specifically, glutamine amidotransferases (GATase) are enzymes that catalyze the removal of the ammonia group from a glutamine molecule and its subsequent transfer to a specific substrate, thus creating a new carbon-nitrogen group on the substrate. It is important to notice that, although such mutations were detected only in Host-evolved Lp populations, they ultimately did not confer a fitness advantage in the presence of the host, as Diet-evolved populations ultimately reached higher loads when associated with the fruit fly (Fig. 4B). These results differ from other previous findings, according to which Drosophila has a positive impact on the growth of its gut microbiota [26]. To further investigate this point, we compared the growth of the two Lp-evolved populations in both experimental setups (Host and Diet). Notably, we found that the Lp concentration was always significantly higher in the presence of Drosophila. However, the Diet-evolved bacteria had an overall growth advantage compared to the Host-evolved populations when associated with the fruit fly (Fig. 4C). Interestingly, this result was already visible from the Lp growth dynamics monitored during the experimental evolution (Fig. 4B), where Lp was retrieved in a higher concentration in the presence of the fly only in the early stages (~88 Lp generations, EE cycle 1).
Taken together, our results suggest that the fly improves Lp growth in a short ecological timescale, that is in the absence of evolution, regardless of the microbial evolutionary background. On the other hand, bacterial growth is favored in the absence of its fly host in the longer term. We speculate that such different growth dynamics might be due to a combination of factors. On the one hand, the growth advantage initially conferred by the fly host to its microbiota might result in an evolutionary "relaxed" selective environment, which in turn affects bacterial growth and adaptation. On the other hand, the stronger selection occurring in the absence of the fly host might ultimately result in higher bacterial fitness in a longer timescale. Indeed, it is commonly expected that the rate of adaptation is higher when selection is stronger [143]. At the same time, it is worth noticing that the lower microbial load detected in the presence of the host might be partly due to the continuous ingestion of bacteria by the fruit fly, as it is known that a large fraction of ingested bacteria gets killed while passing through the stomachlike region of the Drosophila gut [26,144,145]. However, this does not fully explain the higher Lp loads retrieved in the absence of the host, as it should have been visible already during the first EE cycle (Fig. 4B).

Conclusions
Our results demonstrate that L. plantarum evolution diverges between insects and mammals. Specifically, we show that in Drosophila, the nutritional environment dictates microbial evolution, while the host benefits L. plantarum growth only over short ecological timescales. By contrast, in a mammalian animal model, L. plantarum evolution results to be divergent between the host intestine and its diet, both phenotypically (i.e., Host-evolved populations show higher adaptation to the host intestinal environment) and genomically. Here, both the emergence of hypermutators and the high persistence of mutated genes within the host's environment strongly differed from the low variation observed in the host's nutritional environment alone. This indicates that the mammalian animal host, together with host's intrinsic factors, represent crucial agents of selection for the evolutionary path of gut microbes. In addition, we believe that ecological factors need to be considered in the interpretation of our findings. Specifically, gut bacteria in flies are largely environmentally determined, as flies live in their nutritional environment. On the contrary, mammalian gut microbes are known to stably colonize the host intestine and their dispersal strongly relies on cohabitation and host-mediated transmission. Altogether, such ecological drivers contribute to the divergent evolutionary trajectory of gut microbes across animals, beyond the selective pressure exerted by the host per se. Furthermore, we believe that increasing the number of independent replicates of bacterial experimental evolution in the mouse intestine would be needed to further demonstrate the replicability of our findings. From this standpoint, the key questions we need to address in future studies should center on the characterization of the targets of selection, as well as the factors driving the gut microbes' adaptation from a subspecies level to higher levels of microbial interactions. Addressing this will allow us to better understand the relationship among evolution, adaptation, and microbial function and will reveal the principles that govern the colonization success, persistence, and resilience of gut microbes and how they vary across animals and humans.

Bacterial strains and culture conditions
All strains used in the present study were derived from the ancestor L. plantarum NIZO2877 that was originally isolated from a sausage in Vietnam [146]. At the end of each experimental evolution transfer or generation, the evolved strains were stored at −80°C in 1 mL of phosphate-buffered saline (PBS, Sigma) by adding 200 μL of 80% glycerol.

Drosophila stocks and breeding
Drosophila yw flies were used as the reference strain in this work. Drosophila stocks were cultured at 25°C with 12/12-h dark/light cycles on a yeast/cornmeal medium containing 50g/L inactivated yeast (rich diet) as described by Storelli et al. [10]. Poor-nutrient diet was obtained by reducing the amount of yeast extract to 8 g/L. Germ-free (GF) stocks were established and maintained as described in Storelli et al. [10].

Drosophila diet
The fly diet used in the present study was a poor yeast diet containing 8 g inactivated dried yeast, 80 g cornmeal, 7.2 g agar, 5.2 g methyl 4-hydroxybenzoate sodium salt, and 4 mL 99% propionic acid per 1 L. After preparation, fly food was poured in 50-mL tubes by adding 10 mL of food to each tube.

Mouse diet
For mouse breeding and for in vitro Lp evolution we use mouse breeding extrudate diet V1126-000 (Ssniff, Soest, Germany). The diet was vacuum packed and sterilized by gamma-irradiation (25 kGy, Bioster, Czech Republic). It is a grain-based diet consisting of wheat, soybean products, corn (maize) products, oat middlings, minerals, soybean oil, sugar beet puLp, vitamins and trace elements, L-lysine HCl, and DL-methionine. For the experimental transfer of Lp NIZO2877 in the mouse diet, the food was manually crushed, as it was initially provided in the form of pellets.

L. plantarum experimental evolution in mice
Germ-free (GF) C57Bl6 mice were kept under axenic conditions in Trexler-type plastic isolators, and the absence of aerobic and anaerobic bacteria, molds, and yeast was confirmed every 2 weeks by standard microbiological methodology [88]. The mice were kept in a room with a 12-h light-dark cycle at 22°C, fed an irradiated sterile diet V1126-000 (Ssniff, Soest, Germany), and provided sterile autoclaved water ad libitum. L. plantarum NIZO2877 was grown in De Man, Rogosa and Sharpe (Oxoid) in static culture overnight at 37°C for the monocolonization of GF mice. Set up of the first replicate of EE: Seven 12-week-old GF mice (1 male and 6 females) were colonized with a single dose (2 × 10 8 CFU/200 μL PBS) by intragastric gavage to create F0 generation. Four female mice were housed in a single cage. One male and two female mice were housed in a separate single cage. Evolving bacteria were horizontally dispersed and vertically transmitted with no further artificial inoculation. The stability and level of colonization was checked periodically by plating of appropriate feces dilution collected from 3-4 mice on MRS agar and counting after aerobic cultivation for 48 h at 37°C (CFU/mL). After verification of stable colonization, mice were mated and colonization of F0 and subsequent generations (F1, F2, F3, F4) was followed for 10 months. Fecal pellets were collected and pooled from 3-4 mice for the duration of the entire experimental evolution (10 months), diluted in PBS 1X, and plated in MRS agar to isolate the evolving L. plantarum colonies in order to follow their evolution along time or stored with 20% glycerol at −80°C for future analysis. Set up of the second replicate of EE: Two 12-week-old GF female mice were colonized with a single dose (2 × 10 8 CFU/200 μL PBS) by intragastric gavage. Each mouse was housed in a single cage inside an isolator. There was no further artificial inoculation. The stability and level of colonization was checked periodically by plating of appropriate feces dilution collected from each mouse on MRS agar and counting after aerobic cultivation for 48 h at 37°C (CFU/mL). Fecal pellets were collected and stored with 20% glycerol at −80°C for future analysis. The animal experiments were approved by the Committee for the Protection and Use of Experimental Animals of the Institute of Microbiology of the Czech Academy of Sciences.

L. plantarum experimental evolution in the mouse diet
The experimental evolution of L. plantarum NIZO2877 in the mouse laboratory diet was designed as follows: Lp NIZO2877 (ancestor) was cultivated at 37°C overnight in 10 mL of MRS Broth. On the following day (day 0), 1 mL of the overnight culture was centrifugated at 4000 rpm for 10 min and washed in sterile PBS. After proper dilutions, 10 μL of PBS-washed culture of L. plantarum (corresponding to 10 2 total CFUs) was inoculated in microtubes (five total technical replicates) containing 150 mg of the crushed mouse laboratory food supplemented with 100 μL of sterile deionized water. At the same time, 100 μL of the bacterial inoculum was plated out on MRS Agar and grown at 37°C for 48 h as a control. To mimic the host's intestine condition, bacteria were incubated at 37°C for 24 h. On the following day (day 1), the evolved bacteria of transfer 1 (T1) were isolated from each of the five replicates of the mouse diet. Specifically, the medium was crushed using the Tissue Lyser II (Qiagen) (frequency of 30 Hz for 40") in 1 mL of PBS microtubes containing 0.75/1-mm glass beads. A total of 10 μL of the crushed medium (10 2 total CFUs) was used to inoculate five novel sterile medium microtubes (day 0 of T2). This allowed the propagation of an evolving bacterial subpopulation derived from the ancestor on the new medium.
To determine the microbial load reached at the end of the first transfer (day 1 of T1), 100 μL of the crushed medium was plated out on MRS agar at 37°C for 48 h. Each experimental transfer followed the same experimental setup as the one described above, with the exception that, since bacteria were propagated along with the food, no further inoculation of the ancestor strain Lp NIZO2877 was performed. The EE in the mouse diet was conducted for a total of 20 transfers.

L. plantarum experimental evolution in Drosophila
Two EE protocols were performed simultaneously to evolve L. plantarum NIZO2877 in the presence of both Drosophila and its diet (Host setup) or just with its diet (Diet setup). For the first generation of both setups, L. plantarum NIZO2877 was cultured overnight in 10 mL of MRS broth at 37°C. At the same time, GF female flies had been placed inside cages containing poor-nutrient GF medium to lay eggs. On the following day, 40 embryos were transferred to 5 replicate tubes containing poornutrient diet (Host setup). Five tubes containing poornutrient diet were also used for the Diet setup, with the exception that no Drosophila eggs were added in this case. Lp NIZO2877 overnight culture was washed in sterile PBS and after proper dilutions 1 mL of PBS-washed culture was added directly on the eggs and the fly food (bacterial inoculum = 10 5 CFU/mL). No further inoculation of the ancestor strain Lp NIZO2877 was performed after the beginning of the first generation until the end of the experimental evolution. Once the mono-association had been performed, Host and Diet tubes were incubated at 25°C. As soon as at least 15 pupae emerged from all Host tubes, 150 mg of food was transferred from each of the ten tubes into as many new microtubes where 0.75/1mm of glass beads were previously introduced. One milliliter of MRS broth was added in each microtube and the content was dissolved by using Tissue Lyser II (30 Hz for 1 min). After proper dilutions, 100 μL deriving from each microtube was plated out on MRS agar plates, which were incubated at 37°C for 48h for colonies counting. Finally, 200 μL of sterile glycerol (80%) was added in each microtube to store the bacteria at −80°C. The preparation of the subsequent host generations (G) or diet transfers (T) reflected the described procedure and started from the frozen microtubes obtained in the previous generation. Depending on their bacterial concentration, an adequate number of dilutions was performed in order to inoculate 10 5 CFU/mL to the new Diet and Host tubes for each of the following transfers/generations.

Generation time of L. plantarum Generation time of L. plantarum in the Drosophila experimental setup
To determine the generation time of L. plantarum strains in the Drosophila Host and Diet experimental setup, we used a modified version of a method that reported the correlation between bacterial growth rate and 16S rRNA content [147]. L. plantarum was cultured to stationary phase (18h) and washed in sterile PBS. Serial dilutions have been prepared and 5 μL containing a total of 10 3 colony-forming units (CFUs) were added to 100 μL of GF poor-nutrient diet with and without Drosophila larvae (Diet and Host setups, respectively) and kept at 25°C. Samples were snap-frozen in liquid nitrogen at different time points across 5 days of growth. Bacterial RNA was extracted using NucleoSpin RNA Isolation kit (Macherey-Nagel, Germany) following manufacturer's instructions. Reverse transcription of total extracted RNA into cDNA has been performed using Superscript II (Invitrogen, USA) according to the manufacturer's instructions. Quantitative PCR was performed in a total of 20 μL on a Biorad CFX96 apparatus (Biorad) using SYBR GreenER qPCR Supermix (Invitrogen, USA). The reaction mixture consisted of 0.5 μL of each primer (10 μM each), 12.5 μL of SYBR GreenER mix, 10 μL of water, and 1.5 μL of template cDNA. The PCR conditions included 1 cycle of initial denaturation at 95°C for 2 min, followed by 45 cycles of 95°C for 10 s and 60°C for 40 s. Absolute quantification of 16S rRNA was conducted as follows: five 1:10 serial dilutions of the standard sample (100 ng/μL of cDNA extracted from L. plantarum NIZO2877 culture) were quantified by real-time PCR using universal 16S primers (forward primer, UniF 5′-GTGSTGC AYG GYT GTC GTCA-3′ and reverse primer, UniR 5′-ACG TCR TCCMCAC CTT CCTC-3′) [148]. Each dilution has been tested in triplicate. Melting curves of the detected amplicons were analyzed to ensure specific and unique amplification. Standard curves were generated plotting threshold cycle (Ct) values against the log of the standard sample amount. Based on the data obtained from the standard curve, the Ct values of the Host and Diet samples have been used to obtain the log of their 16S rRNA concentration at each time point. The 16S rRNA values during exponential phase have been used to infer the bacterial generation time following the equation reported by Widdel et al. [149].

Generation time of L. plantarum in the mouse experimental setup
Generation time of L. plantarum in the mouse intestine was estimated in the jejunal loops of germ-free mice. L. plantarum was grown overnight in MRS broth at 37°C, centrifuged (4500rpm × 10 min), washed with sterile PBS and adjusted to 10 7 CFU/mL. Four 8-week-old germ-free female C57Bl6 mice were anesthetized by intraperitoneal injection of ketamine/xylazine mixture. Mice were shaved on the abdomen, laparotomy was performed, and 2 jejunal loops were created with nylon ligatures. 10 6 CFU of L. plantarum in total volume of 100 μL was applied directly into the loops using gauge needle. Intestines with loops were put back inside the abdominal cavity; mice were placed into individual boxes on heated pad (37°C) saturated with 0.5% isofluoran. Mice were euthanized at T0 (immediate loops harvest), T1 (1 h), and T2 (2 h); loops were taken out from the re-opened cavity; and each loop was homogenized in 1 mL sterile PBS using Tissue-Lyser LT (Qiagen) and stainless-steel beads. Serial dilutions in PBS were plated on MRS agar and colonies were counted after 48 h at 37 °C. 10 7 CFU/ml of L. plantarum in PBS and the same aliquot after incubation for 2 h at 37 °C were plated out as controls (Additional file 7: Fig.  S7). The experiments were approved by the Animal Care Committee of the Czech Academy of Sciences (protocol n. 18/2019) and were in accordance with the EU and NIH Guide for the Care and Use of Laboratory Animals.
L. plantarum generation time in the mouse intestine and diet was estimated by the following formula (as reported by Kushkevych et al. [150]): Lp NIZO2877 serial passages in MRS broth and MRS broth +

% bile acids (BA)
The ancestral strain Lp NIZO2877 was cultured overnight at 37°C in 10 mL of MRS broth and MRS broth added with 0.3% sterilized bovine bile acids (Sigma) by using three replicates per condition. On the following day, 100 μL of the overnight culture were plated out on MRS agar plates or MRS agar plates containing 0.3% of sterile bile acids and incubated at 37°C for 48 h for colony counting. At the same time, 10 μL of the overnight culture were transferred into a new MRS broth or MRS broth + 0.3% bile acid medium. The same procedure has been followed for 7 days, thus allowing to determine the bacterial growth (CFU/mL) overtime.

Spectrophotometer assays
Spectrophotometer assays were carried out in the presence of MRS broth or MRS broth added with 0.3% bile acids to investigate the bacterial growth by mimicking the stress conditions found in the mouse gastrointestinal tract. One hundred microliters from the −80°C stock of each strain were cultured on MRS agar plates at 37°C for 48 h. Next, one colony was resuspended in 60 μL of PBS and 10 μL was then tested in triplicates in a 96-well plate containing 100 μL of MRS broth (control), or MRS broth with the addition of 0.3% sterilized bovine bile acids (Sigma), respectively. The bacterial growth was assessed turbidimetrically by measuring optical density at 600 nm every 5 min for 24 h using the Multiskan ™ GO Microplate Spectrophotometer (Thermo Scientific).

Electron microscopy
Bacterial colonies transferred from cellophane into cacodylate or phosphate buffer (pH 7.2-7.4) were carefully resuspended and fixed in buffered 3% glutaraldehyde overnight at 4 °C. Thoroughly washed cells were sedimented onto poly-L-lysine coated round 12-mm coverslips for 48 h at 4 °C. Coverslips were then washed with ddH 2 O and postfixed in 1% OsO4 at room temperature for 1 h. After LpG − Time = log 2 · t 2 −t 1 log X 2 −log X 1 t = time; X = colony − forming units per milliliter post-fixation, the coverslips were washed three times with ddH 2 O and dehydrated in graded alcohol series (25,50,75,90,96,100, and 100%), followed with 100% acetone, each step for 20 min. Finally, the coverslips were critical-point dried (K850, Quorum Technologies Ltd, Ringmer, UK) and sputter-coated with 3 nm of platinum (Q150T ES, Quorum Technologies Ltd, Ringmer, UK). Alternatively, pieces of cellophane with bacterial colonies were mounted onto glass slides with Scotch tape. The mounts were then put into a Petri dish with a small container filled with 2% OsO 4 in ddH 2 O. Fixation in osmium vapor was then performed in closed Petri dishes in the desiccator for several days at room temperature. Pieces of cellophane with fixed colonies were mounted onto standard aluminum stubs and sputter-coated with 3 nm of platinum (Q150T ES, Quorum Technologies Ltd, Ringmer, UK). All samples were examined in an FEI Nova NanoSEM scanning electron microscope (FEI, Brno, Czech Republic) at 3 to 5 kV using ETD, CBS, and TLD detectors.

Adaptation of L. plantarum to the mouse intestine
Three C57Bl6 mice (2 month-old) were colonized with a single dose (2 × 10 8 CFU/200 μL PBS) of the Lp NIZO2877 ancestral strain and Lp NIZO2877 -derived population evolved in the mouse intestine (sample: F0-10 months) by intragastric gavage. Fecal pellets were sampled every 12 h (until 72 h after gavage) and immediately freezed at −80°C. Bacterial DNA was extracted from mice stool using NucleoSpin DNA Isolation kit (740472.50 Macherey-Nagel, Germany) following manufacturer's instructions. Real-time PCR amplifications were performed on a LightCycler 480 thermal cycler (Roche Diagnostic, Mannheim, Germany) in a final volume of 10 μL, which included 2.5 μL of DNA. The PowerUp ™ SYBR ™ Green Master Mix (Applied Biosystems ™ , USA) was used together with 0.25 μL of each primer. Primers designed on L. plantarum ackA gene were used to specifically amplify L. plantarum DNA, while universal 16S primers were used to amplify the total bacterial DNA (Additional file 16: Table S9). The cycling conditions were as follows: 50 °C for 2 min, followed by 2 min at 95 °C, and 45 cycles at 95 °C for 10 s and 60°C for 1 min. Outputs of real-time amplifications were analyzed by means of the LightCycler 480 Basic Software Version 1.2 (Roche Diagnostic, Mannheim, Germany). The amount of L. plantarum DNA detected was normalized to the total bacterial DNA values to account for DNA extraction efficiency according to cycle threshold analysis (∆C T ).

Fitness assessment of L. plantarum-evolved populations in Drosophila Host and Diet setups
Lp Host-and Diet-evolved populations belonging to Drosophila generation/transfer17 were tested. Ten independent replicate populations (Host: five replicates, Diet: five replicates) were analyzed. Specifically, 10 4 CFU/mL of bacteria taken from each Lp-evolved population were inoculated into new microtubes (N = 10 microtubes per evolutionary background) containing 250 μL of Drosophila food. One-day-old Drosophila larvae were added to five out of the 10 microtubes (Host setup). The remaining five microtubes, which included only the fly diet, represented the Diet setup. After 11 days of growth at 25°C, the whole content was transferred from each sample into novel microtubes where 0.75/1mm of glass beads and 1 mL of MRS broth were previously introduced and the content was dissolved by using Tissue Lyser II (30 Hz for 1 min). After proper dilutions, 100 μL from each microtube was plated out on MRS agar plates and cultured at 37°C for 48 h for colony counting.

DNA extraction and whole-genome sequencing
The bacterial samples processed for whole genome sequencing are listened in Additional file 1: Table S1. For the bacterial populations evolved in the mouse intestine, fecal pellets pooled from 3-4 mice were analyzed per time point. For the DNA extraction, 100 μL of each bacterial population or strain has been plated out in two MRS agar plates and incubated for 48 h at 37°C. Genomic DNA was extracted from a mixture of > 1000 colonies per sample by using the Power Soil DNA extraction Kit (Qiagen) by following manufacturer's instructions. DNA library construction and sequencing were carried out by the EMBL Genomics Core Facilities (Heidelberg, Germany). Each sample was pair-end sequenced on an Illumina MiSeq Benchtop Sequencer. Standard procedures produced data sets of Illumina paired-end 250-bp read pairs.

Mutation identification
Raw reads were trimmed and filtered using the parameter SLIDINGWINDOW in Trimmomatic [151], with a 4-base wide sliding window and cutting when the average quality per base drops below 20. The mean coverage per population was between 124 and 193. Processed reads were aligned and analyzed against their respective reference strain (ancestor) genome (Lp NIZO2877 ) (accession number LKHZ01000000). Candidate mutations were identified by running two passes of the breseq pipeline in polymorphism mode [152]. Initially, each sample was analyzed individually. Mutation predictions that passed default filtering cutoffs in any one sample were merged into one overall list of candidates. Then, breseq was rerun a second time on each sample with the combined list as a user input file so that it output the counts of reads supporting the mutant and reference alleles for each mutation in all samples, even when a potential mutation was at a low frequency or did not pass other default filtering thresholds in a given sample. We further filtered this list of candidate mutations using a combination of Python and R scripts to remove false-positive calls, including those caused by reads that were mapped incorrectly due to incomplete assembly of the reference genome, and to distinguish lowfrequency mutations from sequencing errors. For each candidate mutation, we first examined a Poisson model of the counts of reads supporting the mutant allele with the total counts of reads supporting either the mutant allele or the reference allele in each sample as an offset. Real mutations that are sweeping through populations should have frequencies that significantly deviate in some samples from the average rate across all samples. To test for this signal, we used a likelihood ratio test to calculate the p value for the significance of adding sample as a fixed factor to the Poisson model for each mutation. Candidate mutations with p values that were not significant after correcting for multiple testing using the Benjamini-Hochberg procedure with a false-discovery rate of 1% were removed. We next eliminated candidate mutations with frequencies that were ≥5% in at least half the samples from all treatments. Then, we kept only mutations that had a >20% range in their predicted frequencies among the samples in a given evolution treatment or that both reached a frequency of >10% and appeared in less than or equal to half of these samples to arrive at the final lists of mutations that were analyzed.

Mutation accumulation rates and spectra
We fit linear models to the summed frequencies of all mutations observed in each sample to estimate average rates of mutation accumulation per generation. These models were constrained to have no initial mutations (i.e., an intercept term of zero). For the fly diet, fly host, and mouse diet treatments, we fit models with one rate of mutation accumulation. For the mouse host treatment, we fit three rates: one for ancestral (nonmutator) lineages, one for mutS A41T lineages, and one for mutS Δ1303 lineages. We modeled the total mutation frequency in a sample as the sum of each of these three population's rates multiplied by its frequency in the population and the number of generations that had elapsed from the beginning of the experiment up to that sample. This procedure assumes that both mutator lineages evolved close to the beginning of the experiment and experienced one constant rate of mutation accumulation throughout their history. The frequencies of the two mutS alleles in each sample were used to estimate the fraction of the population in each of the three categories. The mutS lineage frequencies were normalized to a total of 100% when they slightly exceeded this value when added together due to experimental and/ or sampling errors in the estimates of allele frequencies from sequencing reads.

Muller plots
We included only mutations that reached at least a 10% frequency in a population when constructing Muller plots. Three mutations associated with tRNA and rRNA genes that met this criterion were excluded because they appeared to result from a single structural variant that could not be fully resolved to ensure that the inferred mutation frequencies were accurate. Since whole-population (metagenomic) sequencing does not provide information about linkage between mutations, we inferred which mutations were likely in the same genetic backgrounds from how their frequencies changed over time under the assumption that there was no recombination during the experiment (i.e., purely asexual reproduction). We also assumed that once a lineage had a mutation in a certain gene that it was unlikely to sustain a second mutation in the same gene since these populations retained the low ancestral mutation rate. These rules disambiguated how most mutations could be ordered. We then corrected the genotype frequencies observed at each time point for two types of errors resulting from how the frequencies of all mutations in a sample are estimated independently from the sequencing reads overlapping each genomic site. First, we reduced the frequencies of genotypes with new mutations that exceeded the frequencies of their ancestral genotypes to fit within the earlier group. Second, we normalized the total frequencies of all genotypes at a given time point to 100% if it exceeded this value. Both types of corrections changed the inferred genotype frequencies by <10% in all cases. We used the ggmuller R package to plot the resulting dynamics and then manually adjusted the locations of the curves between the time points with measured values to improve the visibility of mutations and lineages.