The vertebrate gastrointestinal tract (GIT) is the habitat to a collection of microbes that is enormous in density and diversity, most of them bacteria. This community exerts an important influence on the development and performance of the vertebrate host through extensive immunological and metabolic interactions (Hooper and Gordon, 2001; Rakoff-Nahoum et al., 2004; Peterson et al., 2007; Li et al., 2008; Round and Mazmanian, 2009). Despite the importance of gut microbes for vertebrate biology, little is known about how this symbiosis evolved and the selective pressures that shaped it. In general, animals evolved in a ‘microbial soup’ with pathogenic infection as a constant selection pressure, leading to the emergence of multiple immune strategies to confine microbes to the intestinal lumen (Cerutti and Rescigno, 2008). Beyond merely containing microbes, however, vertebrates developed mechanisms to preserve community composition as they gained benefits from the microbes (nutrients, defense, regulation of immune responses), and interestingly, some host responses to the gut microbiota have evolved early as they are conserved in mice and zebrafish (Rawls et al., 2004). From the microbial perspective, inference from comparative genomic analysis of representative genera (Bacteroides, Bifidobacterium) revealed that gut bacteria possess features highly adapted for life in the digestive tract, having evolved sophisticated systems of energy utilization and coexistence with the host's immune system (Schell et al., 2002; Xu et al., 2003, 2007; Peterson et al., 2007). An open question is whether vertebrate gut microbes share long-term evolutionary histories with particular host species or if they have evolved to become generalists for gut ecosystems.

Despite lack of definitive evidence, coevolution, the reciprocal adaptation occurring between interacting species (Moran, 2006), is now generally postulated to account for the development of the microbial symbiosis between humans and their gut microbes (Dethlefsen et al., 2007; Claus et al., 2008; Round and Mazmanian, 2009). Coevolution has largely been inferred by 16S rRNA surveys that revealed that different host species sustain microbial communities containing lineages that are host specific (Dethlefsen et al., 2007). However, this does not preclude the possibility that the specific environmental conditions in individual vertebrates simply select for phylotypes that are generalists for gut environments. In addition, many phylotypes are associated with several host species, and these lineages could be environmentally acquired. Nevertheless, Ley et al. (2008a), who analyzed microbial populations of 59 mammalian species by a 16S rRNA survey, provided evidence that the patterns of community similarity matched the mammal phylogeny and that some lineages co-diversified with their mammalian hosts. However, as many phylotypes were shared across host species, the authors concluded that gut microbes may be fairly promiscuous between hosts. This view contrasts with phenotypic characterizations of gut microbes that clearly showed that bacterial colonization and epithelial adherence can be host specific (Lin and Savage, 1984; Tannock et al., 1984). An important question is whether host specialization, when observed, is the outcome of coevolution or just the result of short-term adaptation by a promiscuous gut microbe after invading a new vertebrate species.

The majority of studies that infer the evolution of gut microbes used 16S rRNA surveys of whole bacterial communities (Dethlefsen et al., 2007; Ley et al., 2008a, 2008b). Their disadvantage is that they underestimate the real diversity of the vertebrate microbiome as 16S rRNA sequences possess insufficient genetic resolution to differentiate all bacterial groups that have ecologically distinct roles (Koeppel et al., 2008). Therefore, many evolutionary events will remain undetected, especially if lineage diversification and specialization occurred more recently. A more sensitive and systematic strategy is, therefore, to examine the phylogeny of a single bacterial species that inhabits multiple vertebrate hosts by population genetic approaches. In this instance, the signatures of evolution can be compared directly because of the shared ancestry of the genomic content. If gut microbes have coevolved with their host species, one would hypothesize to find distinct subpopulations, each represented by a monophyletic clade associated with a specific vertebrate host species.

The Gram-positive bacterium Lactobacillus reuteri is an excellent model organism to study evolutionary mechanisms of a vertebrate gut symbiont and test the hypothesis of coevolution as this species stably inhabits the GIT of mammals as diverse as humans, pigs, mice and rats as well as different species of birds. In rodents, pigs and chickens, L. reuteri is one of the dominant species in the GIT (Leser et al., 2002; Salzman et al., 2002; Brooks et al., 2003; Abbas Hilmi et al., 2007). In these animals, L. reuteri colonizes the surfaces of the stratified squamous epithelial lining of the proximal regions of the digestive tract (Tannock, 1992; Walter et al., 2007). The ecology of L. reuteri in the human gut is less clear and the organism is only detected in some subjects (Walter, 2008). Nevertheless, the type-strain of the species, DSM 20016T, was detected in a human subject over several months, and L. reuteri has been considered autochthonous to the human gut (Reuter, 2001). In this study, we have investigated the population structure of a large collection of L. reuteri strains from multiple vertebrate hosts to gain an understanding of the evolution of this vertebrate symbiont. Our data provide remarkable evidence for the diversification of the L. reuteri population into ecotypes that are highly host-specific.

Materials and methods

Bacterial isolates and growth conditions

A total of 165 L. reuteri strains isolated from different animal hosts from global geographic locations were used. Information on all strains is provided in the Supplementary Table S1. Isolates were obtained from humans (n=35), pigs (n=41), chickens (n=28), turkeys (n=5), mice (n=35) and rats (n=26). Human isolates were obtained mainly from fecal samples but also from breast milk, vagina, stomach and the oral cavity. Animal isolates were obtained from GIT or fecal samples. Bacteria were grown anaerobically using modified MRS (mMRS) medium at 37 °C for 48 h. mMRS is MRS medium (Difco, BD, Franklin Lakes, NJ, USA) supplemented with maltose (10 g per liter) and fructose (5 g per liter). DNA isolation and bacterial identification by 16S rRNA sequence analysis was performed as described in the Supplementary materials.

Amplified-Fragment Length Polymorphism (AFLP)

AFLP was performed using the AFLP template preparation kit available from LI-COR Biosciences (Lincoln, Nebraska, USA), with slight modifications to the manufacturer's protocol (details are given in Supplementary materials). For each strain, two individual AFLP electropherograms were generated with primers that had different specific nucleotide overhangs (Supplementary Table S4). AFLP images analysis and PCA was performed using Bionumerics software Version 5.0 (Applied Maths, Kortrijk, Belgium). AFLP dendrogram was constructed using the neighbor-joining (NJ) algorithm implemented in PAUP* 4.0b10.

Multi-Locus Sequence Analysis (MLSA)

MLSA was performed with seven housekeeping genes: D-alanine-D-alanine ligase (ddl), phosphoketolase (pkt), leucyl-tRNA synthetase (leuS), DNA gyrase B subunit (gyrB), D-alanine-D-alanyl carrier protein ligase (dltA), RNA polymerase alpha subunit (rpoA) and recombinase (recA) using standard methods (details are given in Supplementary materials). The nucleotide sequences of all MLSA loci used during this study have been submitted to Genbank under the following database accession numbers: ddl (GQ378230—GQ378345); pkt (GQ378694—GQ378809); leuS (GQ378578—GQ378693); gyrB (GQ378462—GQ378577); dltA (GQ378346—GQ378461); rpoA (GQ378926—GQ379041); and recA (GQ378810—GQ378925). Descriptive and phylogenetic analysis of MLSA data and the test for recombination using SplitsTree4 (version 4.10) was performed as described in the Supplementary materials.

Reconstruction of genealogies using the ClonalFrame software

Sequences from all seven MLSA loci were analyzed using ClonalFrame (Didelot and Falush, 2007). ClonalFrame uses a Bayesian framework assuming a neutral coalescent model. Five independent runs were performed, each consisting of 300 000 iterations (100 000 burn-in iterations and 200 000 post-burn-in iterations). Phylogenies were sampled every 100 iterations after the burn-in, resulting in a total of 10 000 sampled trees. Analyses from five independent runs were concatenated to construct 50% majority-rule consensus tree. Genealogies were drawn using the radial tree option in MEGA4 software package. ClonalFrame analyses were performed with all strains and clade-by-clade to obtain evolutionary parameters such as the relative impact of recombination as compared with point mutation (r m–1).

Competition experiment in ex-germ-free (GF) mice

Equal volumes of cell solutions of L. reuteri strains were administered to germ-free Swiss Webster mice housed under GF conditions at the University of Nebraska gnotobiotic facility (details are given in Supplementary materials). Fecal samples of individual mice were taken at days 1, 4, 8, 11, and mice were killed on day 11. The strain composition in fecal samples as well as on the forestomach epithelium and the cecum was determined by pyrosequencing of PCR amplicons of the leucyl-tRNA synthetase (leuS) gene using a 454 GS-FLX sequencer (Roche, Basel, Switzerland). The sequencing resulted in an average of 12 739 sequence tags per sample. A local nucleotide database was established in Bioedit v7.0.9 for each sample, and the BLASTn algorithm was used to determine the number of sequences that account for each individual strain.


AFLP and MLSA genotyping revealed highly diverged, host-confined subpopulations of L. reuteri

To gain insight into the population structure of L. reuteri, we investigated 165 strains of L. reuteri originating from humans, pigs, mice, rats, chickens and turkeys from global geographic locations (Supplementary Table S1). Comparisons of partial 16S rRNA gene sequences (700 bp) of all strains revealed an identity of at least 99.4% to the sequence of the type strain (DSM20016T), indicating that these strains fall within what is currently accepted as the species L. reuteri. As a first step to inferring the population structure within this species, AFLP analysis was performed using two independent primer pairs (the electropherograms are shown in Supplementary Figure S1). Subsequent phylogenetic analysis by the neighbor-joining (NJ) method revealed significant genetic diversity within the L. reuteri population (Figure 1a). Remarkably, the phylogeny revealed several distinct clusters that were almost completely host specific. Three rodent (B, D, E), two pig (A, Cii), and one human cluster (Ciii) were detected, and the poultry strains formed a large cluster together with a second group of human isolates (Ci). Isolates from rodents (mice and rats) and poultry (chicken and turkey) grouped in the same clades (Supplementary Table S1). Principal component analysis (PCA) of AFLP data revealed that the genetically distinct subpopulations of L. reuteri group according to host origin but not by geographic origin (Figure 1b).

Figure 1
figure 1

AFLP analysis of a global collection of Lactobacillus reuteri strains from humans, rodents (mice and rats), pigs and poultry (chicken and turkey). (a) The neighbor-joining phylogeny of 165 strains based on 439 markers obtained from two AFLP primer pairs. Branches are colored according to host origin. Lactobacillus coryniformis (Li146:1, not shown in the figure) was used as the outgroup to root the L. reuteri phylogeny. (b) Principal component analysis of AFLP data. Plots are based on the values of the first three principal components with each circle corresponding to an individual strain colored according to host origin or geographic origin (provenance).

To validate the phylogenetic relationships inferred from the AFLP data, 116 strains were examined by MLSA. The strains were selected to be equally distributed in the AFLP dendrogram while maintaining maximal geographic diversity (Supplementary Table S1). Seven conserved gene loci (ddl, pkt, leuS, gyrB, dltA, rpoA and recA) were amplified by PCR, sequenced and analyzed. The descriptive analysis of the MLSA data is summarized in Supplementary Table S2. Significant sequence diversity within the L. reuteri population was detected. The number of alleles per locus ranged from 16 to 39 and the number of polymorphic nucleotide sites from 16 to 88. The low dN/dS ratios indicated that all genes are under purifying selection as expected from highly conserved housekeeping genes.

MLSA revealed a significant association of multi-locus genotypes with particular host species. The 116 strains represented 60 genotypes (or sequence types, STs), and virtually all STs were specific for a particular host (Supplementary Table S1). This indicates that specific clones of L. reuteri are dominant in particular host species. To infer the evolutionary relationships within L. reuteri, we reconstructed a phylogenetic tree with the concatenated sequence of the seven housekeeping genes (in total 3822 bp) using the maximum-likelihood (ML) method. The ML tree is shown in Figure 2a (the same ML tree with strain names is provided in Supplementary Figure S2a). The analysis revealed six subpopulations that were almost completely host specific and were referred to as I (rodents 2), II (human), III (rodents 1), IV (pig 1), V (pig 2) and VI (poultry/human). Despite the excellent reflection of host origin, the clusters identified by ML showed some degree of ‘fuzziness’, meaning that not all strains of the subpopulations formed discrete phylogenetic groups. Nevertheless, there was excellent correlation between the MLSA and AFLP genotypes, as both techniques identified subpopulations that contained virtually the same strains (Supplementary Table S1). For instance, the composition of strains in MLSA clusters IV, V and VI was identical with the composition of the corresponding AFLP clusters A, Cii and Ci. In accordance with the AFLP analysis, isolates from rodents (mice and rats) and poultry (chicken and turkey) fall within the same MLSA clusters.

Figure 2
figure 2

Phylogenetic analysis of 116 strains of Lactobacillus reuteri based on the concatenated sequences of seven loci. (a) The maximum-likelihood tree was reconstructed with PhyML using the GTR+I+G model of nucleotide substitution and 100 bootstrap replicates. Lactobacillus vaginalis ATCC49540 (not shown in the figure) was used as the outgroup to root the L. reuteri phylogeny. Bootstrap values above 50% are shown at the nodes. Clonal complexes (CC) are labeled with a number corresponding to its predicted founder genotype (as identified by eBURST). Only CCs that contain more than four strains are shown. Isolates from rats are labeled with a circle, and isolates from turkey are labeled with a triangle. (b) Phylogenetic network of the 60 sequence types (STs) are present in the L. reuteri population. Conflicting phylogenetic signals are visualized by the network structure in the center.

Host-specific clusters differ markedly in genetic heterogeneity

A remarkable feature of the L. reuteri population structure inferred by MLSA was the substantial difference in the genetic heterogeneity observed among individual host-specific clades. For example, clusters II (human), IV (Pig 1), V (Pig 2), and VI (poultry/human) had very low sequence diversity, whereas the two rodent-specific clusters I and III were highly heterogeneous (Supplementary Table S3). These differences were confirmed using the eBURST algorithm, which infers patterns of evolutionary descent by grouping strains into clonal complexes (groups of strains that share a minimum of five out of the seven alleles). This analysis classified the 60 STs into 9 clonal complexes (CCs) and 29 singletons. As shown in Figure 2a, all strains within MLSA cluster IV (pig 1) and most strains of cluster II (human) fell within the CC-3 and CC-47, respectively. Several clonal complexes and one ST (containing 13 isolates) could also be identified in clade VI (poultry/human). As expected, many of the identified clonal complexes correspond to the MLSA clusters supported by high bootstrap values (>94%) in the phylogeny. In contrast, most strains of the rodent-specific clusters I and III did not fall within CCs but represented ‘singletons’.

The diversification of L. reuteri took millions of years

To estimate the time that was required for L. reuteri diversification, we determined the age of the last common ancestor using a molecular clock with the MLSA data set. The mean pairwise difference at synonymous sites (Ds) derived from the concatenated sequences of all seven loci from 116 L. reuteri strains was 0.122. Using the previously reported synonymous substitution rate of 4.7 × 10−9 per site per year for Escherichia coli/Salmonella typhimurium and 8.2 × 10−9 per site per year for obligate symbiotic bacteria, Buchnera aphidicola (Lawrence and Ochman, 1998; Ochman et al., 1999), the average age of the cenancestor of the L. reuteri strains studied here was estimated to be 7.6–12.95 million years. Although time scales obtained with molecular clocks should be taken with caution as accelerated rates of evolution have been detected for some bacteria (Wilson et al., 2009), our data nevertheless indicates a long-term association of L. reuteri lineages with particular vertebrate species.

Recombination is an important factor in the evolution of L. reuteri

It is now well established that bacteria do not necessarily conform to the clonal model of evolution (Spratt et al., 2001). Recombination leads to the replacement of short pieces of DNA with the homologous segments from other strains, and in some species, it contributes more to diversification than point mutations. It can often confound the reconstruction of phylogenetic relationships, and the importance of recombination in L. reuteri was made manifest in the highly incongruent topologies of phylogenetic trees reconstructed from individual housekeeping genes (although host specificity was still observed for all individual loci) (Supplementary Figure S3). It is therefore likely that the ‘fuzzy’ topology of the ML tree (Figure 2a) was caused by recombination, as described previously for other recombinogenic bacteria (Hanage et al., 2005). We used the split-decomposition method, implemented in the software SplitsTree4, to detect the signatures of recombination in the MLSA data set (Huson, 1998). Split-decomposition does not force sequence information into a tree-like phylogeny, and the presence of conflicting phylogenetic signals results in a net-like structure. This analysis revealed extensive networked evolution for L. reuteri, which is indicative of recombination, for the concatenated sequence (Figure 2b) and most of the individual genes (Supplementary Figure S4). The Pairwise Homoplasy Index test revealed statistically significant evidence for intragenic recombination for the following genes: ddl, leuS, gyrB and dltA (Supplementary Table S2). Remarkably, despite extensive recombination, host-specific groupings were still detected in the tree reconstructed with SplitsTree4 (Figure 2b).

Reconstruction of a more accurate genealogy

In an attempt to deduce a more accurate evolutionary history of L. reuteri, we reconstructed an ML phylogenetic tree with the concatenated sequence of the three loci that, according to the PHI statistical test, were not significantly affected by recombination (pkt, rpoA and recA). This tree showed a similar topology to the tree reconstructed with all seven loci (Supplementary Figure S2). We further applied the ClonalFrame software, which uses a coalescent-based Bayesian method to infer strained relationships. ClonalFrame excludes putative recombinant regions in its analysis, thus estimating real genealogies more accurately (Didelot and Falush, 2007). As shown in Figure 3, the genealogy inferred by ClonalFrame featured host-specific branches with clear demarcations with little reflection of geographic origin. A determination of relative recombination rates (r m–1) with ClonalFrame revealed an overall r m–1 value of 3.0, and variable levels of recombination in individual subpopulations (Table 1). Despite clear evidence for recombination in L. reuteri, the clusters identified by ML phylogeny and ClonalFrame were in almost complete agreement (Figures 2a and 3). Therefore, we conclude that although recombination is prominent in L. reuteri, it does not confound the ancestry of individual lineages.

Figure 3
figure 3

L. reuteri genealogy inferred by ClonalFrame using a 50% majority-rule consensus tree from five independent runs. Scale bar in the ClonalFrame genealogy represents time in coalescent units. Isolates from rats are labeled with a circle, and isolates from turkey are labeled with a triangle. Trees are color coded by host (a) and geographic origin (b).

Table 1 Relative recombination rate of L. reuteri population and individual lineages as inferred by ClonalFramea

Ecological fitness of L. reuteri strains reflects host origin

To gain insight into the evolutionary factors that led to the development of host-confined subpopulations of L. reuteri, we performed a competition experiment in mice. The objective of this experiment was to determine if evolution of L. reuteri was adaptive, which would result in specialization and fitness benefits, or merely neutral. We introduced an equal mixture of nine L. reuteri strains originating from a mouse (1), rats (2), humans (2), chickens (2) and pigs (2) together with a mouse isolate Lactobacillus johnsonii DB-C4 into germ-free mice. L. johnsonii was added as this species co-exists with L. reuteri in natural populations of the rodent forestomach (Salzman et al., 2002; Brooks et al., 2003). Each of the L. reuteri strains used in the experiment carried a distinct allele of the leuS gene, allowing their differentiation. The composition of the L. reuteri population in the inoculum, fecal samples (days 1, 4, 8 and 11) and the forestomach and cecum (day 11) was characterized by 454 pyrosequencing of leuS amplicons obtained by PCR. This analysis showed that two of the rodent strains possessed a very obvious competitive advantage (Figure 4). The rodent isolates 6799jm-1 (mouse) and R2LC (rat) outgrew all other strains and comprised around 97% of the total population on the forestomach epithelium, the natural niche of L. reuteri. This competition experiment confirmed earlier studies on the host specificity of L. reuteri strains in gut colonization (Molin et al., 1992; Carbajal et al., 1999; Schreiber et al., 2009), and together, the results clearly show that the phylogenetic relationships discovered for L. reuteri do reflect not only host origin but also ecological fitness.

Figure 4
figure 4

Test of ecological fitness of L. reuteri strains in the digestive tract of ex-germ-free mice. Germ-free mice were colonized with an equal mixture of L. reuteri strains 6799jm-1 (mouse), R2LC (rat), 100-23 (rat), DSM20016T (human), ATCC55730 (human), 1063 (pig), CP395 (pig), 11284 (chicken), and JCM1081 (chicken) and a mouse isolate Lactobacillus johnsonii DB-C4. Relative abundance of individual L. reuteri strains in populations was determined by pyrosequencing of the leucyl-tRNA synthetase (leuS) gene. (a) Abundance (%) of strains in fecal samples taken during the experiment (11 days). Means with s.e.m. are shown (n=5). (b) Abundance after 11 days on the forestomach epithelium and cecum of mice. Means with s.e.m. are shown (n=5).


Symbiotic microorganisms form intimate associations with most members of the animal kingdom (Moran, 2006). These partnerships vary in how they evolved, their mechanisms for achieving and maintaining associations, and the roles played by the microbe in host biology. To date, the evolution of microbial symbiosis is best understood in invertebrate endosymbionts, where evolutionary mechanisms have been characterized and coevolution has been convincingly proven in many cases (Dale and Moran, 2006; Moran et al., 2008). The phylogenetic analysis of a global L. reuteri population presented here now allows, for the first time, insight into the evolutionary history of a vertebrate symbiont. The most striking finding with both AFLP and MLSA was the clear reflection of host origin within the L. reuteri population structure and phylogeny. As provenance was not reflected, it appears that the distinct ecological conditions within the digestive tract of vertebrate species were the major forces that drove diversification of the L. reuteri ancestral population into host-specific lineages. The results also predict a stable relationship of L. reuteri with vertebrate hosts, as the deep divisions between host-specific lineages indicate a long period of close association. One would assume that the maternal care of animals such as rodents and pigs would facilitate parental–offspring transmission, and for pigs and humans, maternal transmission of L. reuteri has been empirically shown (Tannock et al., 1990; Casas and Dobrogosz, 2000). Although the data does not allow us to pinpoint the exact duration of stable symbiotic relationships, the high average sequence divergence between subpopulations (Supplementary Table S3) and the time span determined by a molecular clock suggested that L. reuteri lineages have evolved with particular hosts for millions of years.

How host specific is the evolution of L. reuteri?

The phylogeny of L. reuteri suggests that its long-term evolutionary relationships with vertebrates resulted in host-specific subpopulations. However, they do not appear to be absolutely constrained to individual host species. Isolates from rodents (mice and rats) and poultry (chicken and turkey) group in the same phylogenetic clusters. This suggests that the time for the divergence of truly host-specific lineages was not yet sufficient, or that L. reuteri evolved with groups of closely related host species. Such a ‘joint’ evolution would result in the development of generalists, which are promiscuous in terms of their ability to colonize the GITs of all members of these groups. In fact, this can be observed in L. reuteri isolates from mice and rats, which show comparable ecological performance in the mouse digestive tract (Figure 4 and references: Molin et al., 1992; Carbajal et al., 1999; Schreiber et al., 2009). In addition, murine Lactobacillus isolates show no host specificity in their adherence to epithelial tissues of mice and rats, whereas they generally do not adhere to epithelial cells of chickens and pigs (Savage, 1972; Lin and Savage, 1984; Tannock, 1992). Based on these findings, we hypothesize that some vertebrate gut symbionts evolve with groups of closely related host species that possess similar niches (that is, shared anatomy, host glycan composition, diet composition and innate immune receptors) in the GITs and whose social behavior allows horizontal transfer of bacteria.

Although the population analysis revealed clusters that reflect host origin, the clade VI contained not only poultry strains but also human strains. In addition, other essentially host-specific clusters contained some ‘outliers’ originating from unrelated hosts. These findings could indicate that some strains might be promiscuous and have recently switched hosts. However, we argue that a promiscuous lifestyle that entails common host jumps would not have resulted in the mostly host-specific subpopulations detected during this study. It is more likely that these ‘outliers’ are allochthonous to the particular host animal. Allochthonous bacteria are often present in the intestinal tract as they get introduced through the diet (Dal Bello et al., 2003; Walter, 2008), but they are only transient and hence do not share an evolutionary history with the host (Savage, 1972). For example, L. reuteri strains originating from poultry might become transiently associated with humans just as Salmonella that often originate from chicken (Callaway et al., 2008). Nevertheless, the detection of allochthonous strains originating from other species indicates that there is an opportunity for a horizontal transfer between hosts, and host switches are likely to have had an important role when L. reuteri became associated with a variety of vertebrate species. However, the population structure detected here indicates that such jumps between unrelated hosts have been rare events during the evolution of L. reuteri.

The evolution of L. reuteri is adaptive

As the genetic clusters of L. reuteri showed high cohesion and could be robustly linked to specific ecological settings (the gut of particular vertebrate species), we postulate that they represent ecotypes. Ecotypes are populations adapted to a particular set of environmental conditions whose members are both genetically and ecologically similar (Cohan and Perry, 2007). The population structure predicts the host environment as the main driver of diversification, although further unknown factors led to the diversification of rodent and pig lineages into two ecotypes. But what evolutionary process led to diversification of L. reuteri? Recent theoretical and empirical studies showed that diversification and genetic coherence within ecotypes can be facilitated by both natural selection and genetic drift (Fraser et al., 2005, 2007; Cohan and Perry, 2007). It has been argued that genetic drift could be especially important for commensals, as host transmission and life in attached populations might reduce the effective population size (Ne) and generate population bottlenecks that favor neutral diversification (Cohan and Perry, 2007; Fraser et al., 2007). Here, we provide evidence that there is more than neutral evolution and genetic drift in L. reuteri diversification. First, ecological fitness of L. reuteri reflects host origin in competition experiments. Second, the average ratio of homologous recombination to mutation rates (r m–1) in L. reuteri is 3.0 (Table 1), which exceeds the clonal–sexual threshold allowing formation of clusters in the absence of selection (Fraser et al., 2007). These findings are in agreement with conclusions made by Ley et al. (2008b) that selection and competition might have a significant role in gastrointestinal ecosystems.

Selective forces vary in different vertebrate hosts

The physiological, immunological and behavioral differences of host species are likely to constitute distinct challenges and selective pressures for their gut inhabitants. This is reflected in the L. reuteri population structure. The allelic diversity of housekeeping genes within the two rodent clades (1.72% and 1.53%) was much higher than the diversity within clusters of human, pig, and poultry origin (<0.87%), which were also much more clonal (Figure 2a). Diversity might have been purged in the human, porcine and poultry clusters through recent events of periodic selection in accordance with the stable ecotype model (Cohan and Perry, 2007). Periodic selection refers to a mutation event that improves the fitness of one strain such that its descendants out-compete all other strains within the ecotype. However, the rodent clusters were clearly too diverse to support a recent selective sweep, and processes such as clonal interference or the adaptation of individual strains to micro-niches might have contributed to the diversity within genetic clusters (Doolittle and Zhaxybayeva, 2009). Why diversity is much higher in the two rodent ecotypes remains to be evaluated. In Helicobacter pylori, genetic variability generated by both high mutation rates and recombination has long been thought to contribute to the adaptation to individual human hosts and bacterial transmission (Suerbaum and Josenhans, 2007). It is possible that the ability to introduce genetic variations, especially through a high rate of intrastrain recombination (Table 1), is an important mechanism by which rodent L. reuteri lineages adapt to physiological differences of individual animals and, probably, mice and rats.

Mutualism as a result of coevolution?

A fascinating aspect of the evolution of vertebrate symbionts is that the host itself is under selective pressure (Ley et al., 2006). Therefore, beneficial traits of gut bacteria that share long evolutionary history with their host species could have been shaped by natural selection as they promote host fitness, which in turn facilitates the microbe's dissemination (Dethlefsen et al., 2007; Ley et al., 2008b; Round and Mazmanian, 2009). L. reuteri is documented to benefit humans and animals and can therefore be considered a true symbiont of vertebrates (Shornikova et al., 1997; Madsen et al., 1999; Casas and Dobrogosz, 2000; Rosenfeldt et al., 2003; Moller et al., 2005; Pena et al., 2005; Weizman et al., 2005; Savino et al., 2007). The phylogenetic relationships discovered in this study suggest that the beneficial attributes of L. reuteri strains could be the outcome of a long-term evolutionary process that resulted in a mutualistic relationship between microbe and host. An important practical value of the phylogeny established here is its use for the selection of strains for probiotics. It is a logical working hypothesis that strains that share a long evolutionary history with particular vertebrate species possess adaptive traits that benefit their hosts (Walter, 2008). Clearly, our findings establish a framework for future phylogenomic investigations to identify adaptive traits of L. reuteri important for both the microbe's ecological performance and its effects on the host.

In conclusion, this study provided crucial knowledge for our understanding of microbial symbiosis in the vertebrate gut, as it revealed that a gut microbe can share a long-term evolutionary history with particular vertebrate host species. By theory, such an evolutionary process would favor the development of a mutualistic relationship and might account for the important beneficial attributes of the gut microbiota (Ley et al., 2006; Dethlefsen et al., 2007). The findings obtained during this work strengthen the notion that a disruption of the ancient host–microbial partnership through a modern lifestyle (antibiotics, diet, formula feeding, hospital deliveries, hygiene, etc.) could have significant consequences for health and might have contributed to the recent increase in diseases that have been linked to the aberration of the gut microbiome (Ley et al., 2005; Nicholson et al., 2005; Cani et al., 2007). Clearly, future research should be targeted to identify evolutionary strategies for other members of the vertebrate gut microbiota.