Recurrent Reverse Evolution Maintains Polymorphism after Strong Bottlenecks in Commensal Gut Bacteria.

The evolution of new strains within the gut ecosystem is poorly understood. We used a natural but controlled system to follow the emergence of intraspecies diversity of commensal Escherichia coli, during three rounds of adaptation to the mouse gut (∼1,300 generations). We previously showed that, in the first round, a strongly beneficial phenotype (loss-of-function for galactitol consumption; gat-negative) spread to >90% frequency in all colonized mice. Here, we show that this loss-of-function is repeatedly reversed when a gat-negative clone colonizes new mice. The regain of function occurs via compensatory mutation and reversion, the latter leaving no trace of past adaptation. We further show that loss-of-function adaptive mutants reevolve, after colonization with an evolved gat-positive clone. Thus, even under strong bottlenecks a regime of strong-mutation-strong-selection dominates adaptation. Coupling experiments and modeling, we establish that reverse evolution recurrently generates two coexisting phenotypes within the microbiota that can or not consume galactitol (gat-positive and gat-negative, respectively). Although the abundance of the dominant strain, the gat-negative, depends on the microbiota composition, gat-positive abundance is independent of the microbiota composition and can be precisely manipulated by supplementing the diet with galactitol. These results show that a specific diet is able to change the abundance of specific strains. Importantly, we find polymorphism for these phenotypes in indigenous Enterobacteria of mice and man. Our results demonstrate that natural selection can greatly overwhelm genetic drift at structuring the strain diversity of gut commensals and that competition for limiting resources may be a key mechanism for maintaining polymorphism in the gut.

Introduction

The mammalian gut is inhabited by a highly diverse microbial community (including multiple phyla of bacteria, archaea, eukaryotes, and viruses). Current data suggests that maintaining a species rich gut bacterial community is important for multiple aspects of host health (Clemente etal. 2012), including: resistance to pathogens (van Nood etal. 2013), development of the immune system (Faith etal. 2014) and even behavior (Dinan and Cryan 2017). Maintenance of species diversity in the gut may be influenced by several ecological factors: spatial structure (transversal and longitudinal), temporal fluctuations in the environment (e.g., food intake, David etal. 2014) and competitive interactions amongst colonizing species (Coyte etal. 2015). Much less studied is the diversity within each species of the microbiota (Caugant etal. 1981) and the extent to which such intraspecies variation has functional consequences for both bacteria and their hosts (Schloissnig etal. 2013; Greenblum etal. 2015). In particular, the dynamics of emergence and spread of adaptive mutations in commensal gut bacteria are poorly known (Waldor etal. 2015).

Different evolutionary mechanisms can promote genetic diversity within species (Chesson 2000). The recurrent emergence of new adaptive mutations in different clones that compete for fixation—clonal interference—leads to unstable levels of polymorphism (Gerrish and Lenski 1998). On the other hand, resource partitioning and negative-frequency dependent selection leads to stable polymorphisms (Chesson 2000). Evolution by de novo mutation in the gut has recently started to be studied using a mouse model of colonization, the streptomycin-treated mouse. In this model, the antibiotic (streptomycin) is used to deplete some of the microbiota members (the facultative anaerobes) so that specific strains of commensal or pathogenic bacteria can colonize the gut (Krogfelt etal. 2004). Clonal interference (Barroso-Batista etal. 2014; Lourenço etal. 2016), trade-offs between stress resistance and nutritional competence (De Paepe etal. 2011) and negative frequency-dependent selection (Lourenço etal. 2016), have all been shown to affect the adaptive dynamics of newly emerging mutations in the guts of mice.

The long-term success of bacterial lineages also depends on their ability to transmit between hosts. Such transmission can be vertical, for example, from mother to offspring, or horizontal, for example, in the case of fecal transplantation or migration of microbes between hosts inhabiting the same environment. The process of transmission can entail strong population bottlenecks that may reduce genetic diversity, potentially breaking interactions between strains and/or species that were created within the gut of a given host (Coyte etal. 2015). Genetic diversity can then be regained during population reestablishment. However, it is currently poorly understood whether the recovery in diversity involves mutations that reestablish previously existing polymorphisms at the same or at different loci/traits.

Escherichia coli generally associates with humans as a commensal, being one of the first colonizers of our guts. In particular, the strain K12 constitutes one of the best-studied model organisms. Recently, we began to uncover the nature of the adaptive process that E. coli experiences in the gut, by following the evolutionary dynamics of a lineage colonizing mice. Using the streptomycin-treated mouse model of colonization, described above, and a fluorescently labeled strain of E. coli, we unraveled the spread of adaptive mutations, by the signature it leaves in the dynamics of neutral fluorescent markers (Barroso-Batista etal. 2014). In this mouse model, the E. coli that naturally inhabits the mouse gut, is killed by the antibiotic whereas the fluorescently labeled E. coli (which is streptomycin resistant) is able to stably colonize and evolve in the gut. We found that the first step of adaptation to the gut, arising as rapidly as 3 days postcolonization, consisted in the selective inactivation of the operon that allows E. coli to metabolize galactitol (i.e., gat-negative mutant; Barroso-Batista etal. 2014), a sugar alcohol derived from galactose. Distinct, but phenotypically equivalent, knock-out alleles of this operon bearing a similar selective effect (7.5% benefit; Barroso-Batista etal. 2015) were recurrently observed to emerge across all populations adapting to the gut of independent mice, reaching 95–99% frequency after 24 days (∼430 generations). Thus, the fitness effects of mutations comprising the first step of adaptation to the gut could be described by a simple distribution, where all mutations have similar effect (Hegreness etal. 2006).

Here, we use an evolved gat-negative single mutant to colonize a new set of mice (mimicking a strong transmission bottleneck), and ask if the rate of adaptation slows down in this complex ecosystem. A deceleration in the rate of adaptation is expected under simple fitness landscape models (Couce and Tenaillon 2015) and is typically found in bacterial populations adapting to laboratory environments (Chou etal. 2011; Khan etal. 2011; Kryazhimskiy etal. 2014). If the selective benefit of emerging mutations decreases during adaptive walks, then diversity might be maintained for longer periods of time. We find that the rate of adaptation does not seem to decelerate for at least two consecutive bouts of adaptation. Unexpectedly, whole-genome sequencing of the evolving populations of the second round of adaptation, revealed that evolution took a step back, with the emergence of multiple mutations that reenabled galactitol metabolism (gat-positive) by E. coli. These phenotypic reversions occurred by compensatory mutation and genetic reversion. Subsequently, upon a third round of adaptation, using an evolved gat-positive strain to colonize new mice, we find de novo loss-of-function mutations. The results demonstrate that in this system, irrespectively of the starting genotype and transmission involving the strongest of bottlenecks, natural selection can maintain polymorphism through recurrent reverse evolution. By assaying microbiota composition and invivo competitions, we then show that the recurrent reverse evolution leads to the coexistence of two phenotypes whose niches appear to be independent, as they can consume different resources. Furthermore, while the total size of the E. coli species is dependent on microbiota composition,—with a more diverse microbiota being associated with a smaller load of E. coli—its diversification is not. Indeed, the abundance of the strain that can consume the resource galactitol (gat-positive) is relatively constant across mice with distinct microbiota compositions and can be increased by galactitol supplementation, in a concentration dependent manner. Conversely, the abundance of the clones that cannot grow on galactitol (gat-negative) is influenced by microbiota composition. Finally, we find that natural isolates of E. coli, inhabiting the guts of healthy humans show variation for galactitol consumption—a trait that we find to be under strong selection in laboratory mice.

Results and Discussion

Rate of Adaptation to the Mouse Gut during the Second Steps of Adaptation

We colonized 15 mice with a clonal population of E. coli exhibiting the first beneficial phenotype—inability to uptake galactitol—conferred by a single base pair insertion in the coding region of the gatC gene, which encodes a subunit of the galactitol transporter. The colonizing population was made dimorphic for a neutral fluorescent marker to allow determination of the spread of further adaptive changes (Hegreness etal. 2006). The time series dynamics of the fluorescent marker frequency resulting from the second steps of adaptation are shown in figure 1. These dynamics can be summarized in two effective evolutionary parameters (Hegreness etal. 2006; Barrick etal. 2010; Barroso-Batista etal. 2014), the mutation rate (U) and the fitness effects of adaptive mutations (S) that are estimated to be U = 7.1 × 10−7, S = 9% (see Materials and Methods). These quantitatively describe the simplest adaptive scenario where new beneficial mutations emerge at a constant rate and have a constant effect (Hegreness etal. 2006). The estimates of the evolutionary parameters for the second steps of adaptation are remarkably similar to those previously estimated for the first step (U = 7 × 10−7, S = 7.5%; Barroso-Batista etal. 2014). Direct invivo competitions between samples of evolved clones and their ancestor (a gat-negative strain) further corroborate that the fitness benefit of the accumulated mutations is large, ranging from 5% to 14% fitness increase (mean fitness increase of 9 ± 4% [2SE], fig. 1), which is similar to what was observed for the first steps of adaptation where evolved populations were competed against the ancestral gat-positive strain (8 ± 1% [2SE]). Thus, the strength of selection invivo does not seem to decrease much after the first step of adaptation.

. 1.

Rapid adaptation characterizes the evolution of colonizing the mouse gut. (A) Change in frequency of a neutral fluorescent marker (YFP) along 24 days of colonization. (B) Fitness increase (per generation) measured by direct invivo competitive assays of evolved clones against the ancestor (n = 2).

Widespread Reversion of a Strongly Beneficial Mutation upon Transmission

To determine the genetic basis of adaptation underlying the rapid changes in neutral marker frequency (fig. 1), we performed whole genome sequencing (WGS) of samples of two evolving populations along time in two lineages (2.14 and 2.15 from fig. 1). Remarkably, we uncovered the rapid emergence of several mutations in the galactitol operon (fig. 2, see also supplementary table S1, Supplementary Material online), particularly in the previously pseudogenized gatC (fig. 2). These mutations restored the open reading frame of gatC suggesting that an unexpected regain of function could have evolved. Thus, we tested hundreds of evolved clones from these populations for the ability to metabolize galactitol and found that the new mutations in gatC, spreading through the populations (fig. 2), were indeed gain of function mutations. In face of our previous findings (Barroso-Batista etal. 2014), where gat-positive E. coli evolved a gat-negative phenotype that reached >75% frequency in all the mice after 24 days of colonization (fig. 3), the reemergence of the gat-positive phenotype was completely unexpected. Indeed, we had assumed that the gat-negative phenotype would eventually fix in all populations and therefore started a second colonization with a single adaptive gat-negative mutant. This design, mimicking a strong bottleneck or transmission event, serendipitously revealed that evolution could reverse.

. 2.

Reverse evolution leads to the recurrent regain of a previously lost phenotype, by compensatory and back-mutation. (A and B) Muller Plots showing the emergence of polymorphism in two evolving populations (pop 2.14 (A) and pop 2.15 (B)), as numbered in fig. 1) where regain of function (ability to metabolize galactitol) is detected—gat-rev. Yellow and blue shaded areas are proportional to the frequencies of YFP- and CFP-labeled bacteria. Darker tones of yellow and blue denote the accumulation of adaptive mutations in the respective background. Days shown below the graph depict the temporal points where haplotype frequencies were estimated (see supplementary table S1, Supplementary Material online) through a combination of different approaches. Numbers in red represent temporal points where frequencies were estimated by WGS of the population and numbers in black represent time points where frequencies were obtained by target PCR of collections of clones. Dotted areas represent the proportion of the population that reverted the ancestral gat-negative phenotype and was determined by phenotypic testing. (C) The genetic basis of the gat-reversion phenotype (gat-rev) involves both compensatory (represented by alleles comp1–6) and back-mutation (allele rev). Six different alleles were found in pop 2.14 at day 11 (comp1 to comp6). After 6 days (day17) only alleles comp1 (∼1% frequency) and comp2 (∼13% frequency) could be detected by WGS; we note however that the frequency of gat-rev phenotype in the same time-point was considerably higher (∼50%). This suggests that the polymorphism in the gatC locus estimated by WGS was considerably underestimated in this time-point. On day 24, comp2 was the only allele detected. Days 6 and 11 of pop 2.15 revealed two different alleles (comp3 and rev). On day 17 and 24 only the rev allele was found.

. 3.

Reacquisition of the ancestral metabolic ability is pervasive across independently evolving populations. (A–C) The ancestral genotype and phenotype of each colonization is indicated above the respective graph. (A) Massive selection to inactivate the gat-operon characterized the first colonization, yet in half of the populations the frequency of gat-positive bacteria was kept above 1%. (B) In the second colonization, initiated with a gat-negative strain, reacquisition of the ancestral metabolic ability was pervasive across independently evolving populations reestablishing the polymorphism for the gat phenotype. (C) The third colonization, initiated with a gat-positive clone isolated from the second colonization, exhibited the reemergence of gat-negative mutants, as in the first colonization.

To query whether the two populations studied reflect widespread reversion (Porter and Crandall 2003), we tested large samples of clones from all the evolved populations for a gat-positive phenotype. We found that phenotypic reversion was highly probable, as in 80% of the 15 independently colonized mice the bacteria regained the ability to consume galactitol, with this phenotype reaching a detectable frequency of at least 0.3% and up to 77% in the evolving populations (fig. 3). Thus, under a transmission event involving a single clone, reacquisition of polymorphism at the gat locus constituted a highly predictable adaptive event. Interestingly, the power of selection on this regain of function is clearly observed in population 2.14 (fig. 2), where an allele (involving a 1 bp deletion, the only mutation detected by WGS of the population as indicated in supplementary table S1, Supplementary Material online) emerges in this population at day 11, on the YFP background, but will then compete with a CFP clone which acquired another mutation causing the same phenotype. Thus, the emergence of a gat-positive phenotype in both fluorescent backgrounds within the same evolving population (fig. 2) and the parallelism across independent replicate populations (fig. 3) strongly supports the adaptive nature of the regain of function in this ecosystem (de Visser and Krug 2014). To provide further evidence for selection acting on mutations conferring the gat-positive phenotype, in contrast to being hitchhiker alleles, we sequenced two gat-positive clones. Genome sequencing of the two clones (isolated from population 2.15 at day 6) revealed that: a back-mutation (“rev allele” in fig. 2) had occurred in one clone, with no other mutations detected in its genome; a compensatory mutation (“comp3 allele” in fig. 2), that redoes the open reading frame of gatC had occurred in the other clone, whose genome carried no further mutations. These results suggest that the gat-positive phenotype is intrinsically beneficial. Moreover, the rare event of gain-of-function through back-mutation could even be observed in one of the assayed populations (2.15, fig. 2), supporting the conclusion that the ancestral gene can invade a population where loss of function at this locus had previously evolved. Such an event is particularly important since it erases the signature of the strong past selection that had occurred.

Natural Selection Maintains Polymorphism during Multiple Rounds of Transmission

To understand the extent to which selection can overwhelm genetic drift (associated with extreme population bottlenecks at host-to-host transmission) in driving the emergence of intraspecies diversity at this locus, we performed a third colonization. This was initiated with an evolved gat-positive clone (isolated from population 2.14—fig. 2), which carries a 2 bp insertion that redoes the gatC reading frame and two mutations at other genomic loci (see Materials and Methods for the exact genotype). In all six colonized mice, emergence of the gat-negative phenotype, consistent with inactivation of the gat operon, was seen (fig. 3). In this new genetic background, the emergence of polymorphism at detectable frequencies occurred slightly later than in the first colonization: the median time for gat-negative phenotype to reach 5% frequency was 7 days (6.25–7.75, Interquartile range), whereas for the first colonization it was 3 days (2–4, Interquartile range, Mann–Whitney–Wilcoxon Test, W = 1.5, P = 0.0006). The difference in the time of emergence of the polymorphism could be related with the two additional mutations whose phenotype was not characterized.

The three colonizations together (fig. 3) suggest a model of cycling evolution, where a E. coli population can loose genetic diversity due to intense drift, caused by strong bottlenecks during transmission events between hosts, but can readily restore polymorphism at this locus due to strong selection in this ecosystem and high mutation rate (fluctuation tests show mutation rates of: 1.0 × 10−5 per generation [95% CI, {6.7 × 10−6, 4.0 × 10−4}] (Lourenço etal. 2016) for gat-positive to gat-negative and 9.7 × 10−9 [95% CI, {3 × 10−9, 1.9 × 10−8}] for gat-negative to gat-positive [see Materials and Methods]). Such strong bottlenecks may occur during vertical transmission from parent to offspring (Nayfach etal. 2016) and horizontal transmission of pathogenic strains in a hospital setting.

Competition for Limiting Resources Can Help Explaining the Maintenance of Polymorphism at the gat Operon

Competition for limiting resources is one of the mechanisms likely to be important in the gut microbial ecosystem (Coyte etal. 2015). Such a mechanism can lead to strong selection for evolving polymorphisms (Chesson 2000). Given the observed polymorphism for galactitol metabolism, we hypothesized that this sugar alcohol could be a key limiting resource for E. coli in the mouse gut. Though we did not measure directly the levels of galactitol in the gut, we have explored the potential implications of this being a limiting resource. We first explored the conditions under which a simple resource competition model (Hansen and Hubbell 1980; van Opheusden etal. 2015) could explain the invasions observed in the independent colonizations of figure 3 (see Materials and Methods and supplementary fig. S1, Supplementary Material online) and then extended the model to understand the conditions under which a mutant for galactitol metabolism could invade a resident population of bacteria which possesses a regulated operon (see below and supplementary fig. S2, Supplementary Material online). In the simplest model, we assume that the gat-negative and the gat-positive clones consume two different resources. Given that the gat-positive strain (E. coli K-12 MG1655) constitutively expresses the gat operon (it carries an IS1 insertion in the operon repressor, gatR), we further make the simplifying assumption that this strain only consumes one of the resources (galactitol). As expected under these simple assumptions, the theoretical model can reproduce quantitatively the dynamics of emergence of polymorphisms observed in the experiments (supplementary fig. S1A, Supplementary Material online). Under these conditions, we expect that supplementation of the mouse diet with galactitol should lead to an increase in frequency of gat-positive clones (supplementary fig. S1B, Supplementary Material online). To test for this, we colonized new mice with a coculture of gat-positive and gat-negative bacteria (at a ratio 1:10) while supplementing the mice diet with different concentrations of galactitol. Indeed we observed that the frequency of gat-positive bacteria increased in mice drinking water with galactitol (χ22 = 31.1, P < 0.001, from day 0 to day 2), and more so when the concentration of galactitol was higher (fig. 4, also compare with supplementary fig. S1B, Supplementary Material online). After day 3, the frequency of gat-positive bacteria stabilizes at different levels with different galactitol amounts, with all treatments being significantly different from each other (Tukey tests, P < 0.0001). Furthermore, the absolute abundance of the gat-positive strain increases with galactitol concentration (χ21 = 6.4, P = 0.01; log10 of CFUs: 7.2 ± 0.1, 7.5 ± 0.1 and 8.4 ± 0.03, for 0%, 0.01%, and 0.1% galactitol, respectively, with all treatments being significantly different from each other P < 0.05). The total load of E. coli was not significantly different across diets (χ22 = 4.9, P = 0.09), consistent with the main effect of galactitol supplementation being the increase of gat-positive abundance.

. 4.

Diet supplementation with galactitol increases the frequency and abundance of gat-positive bacteria, without affecting total E. coli density and microbiota composition. Temporal dynamics for the frequency of gat-positive E. coli (A). Temporal dynamics for the density of gat-positive (solid lines) and total E. coli (dashed lines); error bars correspond to standard error (B). Analysis of the mouse gut microbiota shows no evidence for microbiota clustering being influenced by galactitol supplementation (C). Dendrogram on the left represents UPGMA clustering of weighted Unifrac (beta diversity), with * indicating when jackknife node support was above 0.9 (100 subsamples of the OTU table). Consistently, alpha diversity is also not influenced by galactitol supplementation (supplementary fig. S5, Supplementary Material online). Bars on the right indicate the relative frequency of the different operational taxonomic units (OTUs; at 97% identity; full legend is provided in supplementary fig. S3, Supplementary Material online). Data from days 3, 4, and 7 was used in this analysis, as after day 3 both the gat-positive frequency and the mouse microbiota had started to stabilize (supplementary fig. S4, Supplementary Material online shows the temporal dynamics of gut microbiota per mouse). Colors correspond to: blue—no galactitol supplementation; yellow—0.01% galactitol; red—0.1% galactitol (galactitol was diluted in the mice drinking water). Different symbols correspond to different mice.

We then assayed if the composition of the other species in the mouse gut could be altered when adding galactitol to the diet, by 16S rDNA metagenomics of the mouse fecal samples. Figure 4 shows that the microbiota of the mice drinking galactitol-supplemented water do not cluster separately from those that undergo a normal diet. The same conclusion was reached after performing a PCoA analysis on the data shown in figure 4, either considering days 3 through 7 (supplementary fig. S6, Supplementary Material online) or each of the 3 days separately (supplementary fig. S7—day 3, supplementary fig. S8—day 4, and supplementary fig. S9—day 7, Supplementary Material online). Overall, these results suggest that galactitol can increase the frequency of specific strains of E. coli (fig. 4) without significantly changing the load of this species (fig. 4) and the relative abundance of other species of the microbiota (fig. 4within the limits of 16S profiling which is unable to detect differences below the genus level). Nevertheless, we cannot rule out the possibility that galactitol affects the background microbiota in the absence of E. coli.

The results so far presented provide a possible explanation for both the emergence and maintenance of polymorphism for galactitol consumption based on the hypothesis of nonoverlapping niches. Yet, they do not exclude other hypotheses underlying the recurrent inactivation of the galactitol operon. The gat-positive strains which seeded the 1st and 3rd colonization constitutively express the gat operon (due to the IS element insertion in the gatR) and thus lack regulation. We hypothesized that a strain that could regulate galactitol consumption would be better adapted to the gut, and therefore, mutations that inactivate the gat operon would not be selected. To test for this, we replaced the pseudogene gatR, in the original strain, by a functional version, which does not impair the ability of consuming galactitol but regulates the operon expression. We then colonized new mice with this strain and followed its evolution in the mouse gut for 19 days (n = 6). In contrast to the colonization initiated with gat-positive (constitutive), we did not observe the emergence of gat-negative mutants. Therefore, this experiment suggests that the ability to regulate the gat operon impedes (or strongly delays) the loss of function previously observed in this ecosystem. Remarkably though, we did observe the emergence of gat-constitutive bacteria in all mice colonized with a regulated strain (fig. 5). Furthermore, targeted PCR of the evolved clones showed that the phenotype could be caused by de novo IS insertions in the coding region of gatR. Interestingly, these emerging mutants are phenotypically equivalent to the starting E. coli clone, which suggests that constitutive mutants might be indeed segregating in natural populations. These results are predicted by the simple model of resource competition (supplementary fig. S2, Supplementary Material online). The observed loss of regulation invivo is also consistent with previous experiments of adaptation to limiting nutrient concentrations invitro, in which the repressors of operons for sugar metabolism were inactivated (Horiuchi etal. 1962; Zhong etal. 2004, 2009).

. 5.

Emergence of gat-constitutive strains in the guts of mice colonized with a clonal population of bacteria where the repressor of the galactitol operon (gatR) has been restored. (A) Frequency of gat-constitutive mutants that emerged de novo in all colonized mice (n = 6) during the first 5 days of colonization (A). (B) Frequency of gat-constitutive mutants at day 19 in the same mice.

Overall the colonization experiments (figs. 3 and 5) imply that irrespectively of the founder clone being constitutive or regulated, in the long run gat-negative and gat-positive polymorphism is likely to emerge and be maintained in the gut.

Microbiota Composition Affects E. coli Abundance but not E. coli Reverse Evolution

The success of E. coli colonization, in terms of population size, is variable among mice. On the second colonization, approximately half of the animals are colonized with ∼108 CFU/g of feces whereas the other half are colonized with ∼109 CFU/g of feces (fig. 6). This variation could be mediated by the composition of the other members of mouse gut microbiota. We thus enquired whether we could observe an association between the loads of E. coli and the microbiota community composition in the gut of the colonized mice (fig. 6). Remarkably, mice with low loads of E. coli (fig. 6) exhibit a microbiota community composition significantly different from mice with high loads of E. coli (fig. 6). Because the great majority of the E. coli population is gat-negative the same applies to this fraction of the population. In contrast, a remarkable constancy is observed for the abundance of gat-positive E. coli clones, whose numbers remain around 107 CFUs/g irrespectively of microbiota composition. We observed two main microbiota compositions (UPGMA clusters) that we term cluster I and cluster II (I in red and II in blue; fig. 6). Cluster I shows significantly higher species richness (Shannon index = 4.7 ± 0.2 [2SE]; χ21 = 13.5, P < 0.001; fig. 6), being enriched for several OTUs (two Firmicutes and one OTU from Bacteroidetes, Actinobacteria and Deferibacteres; supplementary fig. S10, Supplementary Material online). Cluster II shows a lower Shannon index of 3.3 (±0.3), being enriched for a single OTU belonging to Actinobacteria (Coriobacteriaceae), Proteobacteria (Entobacteriaceae), and Bacteroidetes (Bacteroides). Importantly, cluster I is associated with an abundance of gat-negative E. coli that decreases overtime, tending to <108 (per feces gram), which is significantly lower than that observed in cluster II (with average abundance of 109 per feces gram; χ21 = 33.6, P < 0.001). These results suggest that gat-negative E. coli may compete with other members of the microbiota for resources and that the gat-positive E. coli, which acquired mutations leading to increase expression of the gat operon and increased growth on galactitol, may be selected for because they avoid such competition. As galactitol supplementation and the microbiota independently affect the abundances of gat-positive and gat-negative E. coli (respectively), these results appear to suggest that the two phenotypes occupy different niches.

. 6.

Microbiota composition affects the density of gat-negative, but not gat-positive E. coli. Microbiome analysis was carried for mice from the second colonization (figs. 1 and 3B) in which gat-positive revertants had emerged. Moreover, the densities of gat-positive and gat-negative E. coli were estimated. The gut microbiota affects the densities of gat-negative but not gat-positive E. coli (A). Dot plots indicate the densities of E. coli with each phenotype for the corresponding mouse-day combination. Line and shading indicate the mean and standard error. For gat-negative, two lines are shown as the bacterial densities differ significantly between red and blue clusters. Conversely, for gat-positive, only one line is shown as there is no significant difference. (B) Analysis of the mouse gut microbiota divides mice into two main clusters, I (red) and II (blue). Dendrogram on the right represents UPGMA clustering of weighted Unifrac (beta diversity), with * indicating when jackknife node support was above 0.9 (100 subsamples of the OTU table). Bars on the left indicate the relative frequency of the different operational taxonomic units (OTUs; at 97% identity). Labeled taxa are enriched in either cluster I (red labels) or II (blue labels), as identified by LEfSe (full legend is provided in supplementary fig. S3, Supplementary Material online and LefSe results in supplementary fig. S10, Supplementary Material online). (C–D) Temporal dynamics of gat-negative and gat-positive E. coli showing that the densities of gat-negative E. coli are lower in mice with the cluster I microbiota (C), than in mice with the cluster II microbiota (D) (cluster: time interaction: χ21 = 33.6, P < 0.001). Conversely, the densities of gat-positive clones remain constant across all mice (cluster: time interaction: χ21 = 3.3, P = 0.07 and cluster single effect: χ21 = 1.0, P = 0.31). (E) Microbiota diversity (estimated as Shannon Index) is significantly lower in cluster II than in cluster I (similar observations are made for other diversity indices; see supplementary fig. S11, Supplementary Material online).

The theoretical model suggests that the equilibrium ratio of gat-positive to gat-negative is dependent on the relative abundance of galactitol to other undefined resources (see Materials and Methods) and that the abundance of these two phenotypes can vary independently as these consume different resources. Interestingly, in the galactitol supplementation experiment (fig. 4), we see that increasing galactitol concentration leads to an increase in the abundance of gat-positive bacteria (χ21 = 6.4, P = 0.01), without affecting the abundance of E. coli (P > 0.05). Conversely, in the colonization experiment (fig. 3) it is the abundance of gat-negative bacteria that varies across mice, whereas the abundance of gat-positive bacteria remains relatively constant after their emergence (fig. 6). This suggests that there is variation in the concentrations of the unidentified resources that are consumed by the gat-negative E. coli.

Natural Polymorphism for Galactitol Consumption in Indigenous Enterobacteriacea of Mice and Humans

The findings reported here suggest that the gut ecology may favor natural polymorphism for galactitol metabolism in its microbiota. To test for this, we sampled Enterobacteriacea clones from the natural microbiota of our host model organism (fecal samples of mice untreated with antibiotics in the IGC animal house [n = 20]) and from the gut microbiota of healthy humans with no controlled diet (n = 9). Variation for this trait was found to be segregating across mice and man (table 1). In one out of nine human hosts, polymorphism for the ability to consume galactitol was detected, with 58% of the clones being gat-negative. In five humans, the gat-negative phenotype was present at frequencies above 90%, with some showing a frequency of ∼99%. Only three humans showed no gat-negative Enterobacteriacea. From 20 mice assayed, one was monomorphic for the gat-negative phenotype, one showed an intermediate frequency for this phenotype and the rest showed only gat-positive clones. These results indicate that the level of polymorphism at this locus may be influenced by antibiotic treatment or the strain background used for gut colonization. Nevertheless, in more natural conditions trait polymorphisms can also be found. Interestingly, in a comparative genomics survey of genes under positive selection in E. coli, gatC was found to be a target of repeated replacement substitution driven by positive selection (Chattopadhyay etal. 2009).

Table 1.

Polymorphism for Galactitol Metabolism in Natural Fecal Isolates.

Sample #		gat-pos	gat-neg	Total	Frequency of gat-neg	2SE
Mouse	1	1	143	144	0.99	0.01
Mouse	2	125	19	144	0.13	0.06
Mouse	3–20	144	0	144	0.00	0.01
Human	1	189	3	192	0.02	0.02
Human	2	12	180	192	0.94	0.03
Human	3	192	0	192	0.00	0.01
Human	4	2	190	192	0.99	0.01
Human	5	11	181	192	0.94	0.03
Human	6	81	111	192	0.58	0.07
Human	7	1	191	192	0.99	0.01
Human	8	19	173	192	0.90	0.04
Human	9	192	0	192	0.00	0.01

Conclusions

Overall our results highlight how evolutionary and ecological processes interact to shape the genetic and phenotypic composition of commensal bacteria within a natural ecosystem. At the evolutionary level, we find that a strong-mutation-strong-selection-strong-drift regime of adaptation can lead to the maintenance of polymorphism through frequent episodes of reverse evolution, which defy Dollo’s law of evolutionary irreversibility (Gould 1970; Marshall etal. 1994). This regime of adaptation has been poorly studied, at both the theoretical and experimental levels, but may be important for host-associated microbes as their population dynamics often involves transmission bottlenecks (strong-drift) followed by expansion to large population sizes (strong-mutation-strong-selection). Interestingly, reversion has also been shown to rapidly occur for immune escape mutations in HIV (Friedrich etal. 2004). Our study reveals that phenotypic and genetic reversals may be common in bacteria colonizing the mammalian intestine. Such scenario may have been difficult to predict, as bacteria could be expected to respond to changing environments by gene regulation and not through de novo mutation, as we showed here. One context where reversion is pivotal regards antibiotic resistance. Reversion of antibiotic resistant bacteria to sensitivity has been more rarely observed than what one would desire (Andersson and Hughes 2010). However, since most studies have been done invitro, it would be important to determine if the rate of reversion of resistance invivo could be as high as the one we found here for a metabolic trait (Björkman etal. 2000).

At the ecological level, our results indicate that resource specialization may allow new emerging strains to avoid competition with other members of the microbiota (Conway and Cohen 2015), even if to stably colonize the gut at low abundances. Our results further highlight that precise dietary supplementations may allow for controlling the colonization level of specific strains.

In sum, the evolutionary invasion and ecological stability of de novo emerging clones in genetic identical hosts, nurtured in identical conditions, was highly reproducible, despite being accompanied by significant differences in the level of microbiota species richness between hosts.

Materials and Methods

Bacterial Strains and Culture Conditions

All strains used in this study were derived from MG1655, a K12 commensal strain of Escherichia coli (Blattner etal. 1997).

Ancestors of the First Colonization

The two ancestors of the first colonization (fig. 3), described in a previous experiment (Barroso-Batista etal. 2014), were DM08-YFP and DM09-CFP (MG1655, galK:: YFP/CFP ampR, strR [rpsL150, ΔlacIZYA]).

Ancestors of the Second Colonization

Strains JB19-YFP and JB18-CFP (MG1655, galK:: YFP/CFP cmR, strR(rpsL150), ΔlacIZYA, Ins (1 bp) gatC) were used for the second colonization (fig. 3) and were previously described in (Lourenço etal. 2016). These strains differ from the ancestral MG1655 fluorescent strains DM08-YFP and DM09-CFP by a mutation in the gatC gene (1 bp insertion in the coding region), rendering it unable to metabolize galactitol. To construct these strains, the ampicillin resistance cassette in the ancestral strains DM08-YFP and DM09-CFP was replaced with a chloramphenicol resistant cassette using the Datsenko and Wanner method (Datsenko and Wanner 2000). The yellow (yfp) and cyan (cfp) fluorescent genes linked to cmR were then transferred by P1 transduction to a derivative of clone 12YFP (Barroso-Batista etal. 2015), an evolved clone of DM08-YFP, isolated after 24 days of adaptation in the gut of WT mice, that carried an insertion of 1 bp in gatC and a large duplication. During the genetic manipulations the large duplication was lost, confirmed by whole genome sequencing, leaving this clone with a single mutation in gatC (Lourenço etal. 2016).

Ancestor of the Third Colonization

The strain used to start the third colonization was isolated from population 2.4 of the second colonization at day 24 postcolonization. This strain (named 19CFP (Lourenço etal. 2016) has a 2 bp insertion (+TC) in gatC which reconstructed the open reading frame allowing it to recover the ability to consume galactitol (see below details about the phenotype confirmation). Additionally, this strain has the following mutations: an IS2 insertion in the coding region of ykgB and a 6,790 bp deletion from gene intZ to eutA.

Reconstruction of the transcriptional regulator of the galactitol metabolism (gatR): the gatR coding region is interrupted by an IS element in E. coli MG1655, and therefore in all strains used in this study. This insertion inactivates gatR leading to the constitutive expression of the gat-operon. To restore the function of the gene, and therefore the ability to regulate the operon, a copy of the gatR native sequence (obtained from the strain E. coli HS) was used to replace the nonfunctional copy. This was accomplished by the SacB-TetA counter selection method as previously described (Li etal. 2013). Briefly, the SacB-TetA fragment with the overhangs for recombination was obtained by target PCR using the primers gatR_SacB_F/gatR_SacB_R (supplementary table S2, Supplementary Material online) and genomic DNA of E. coli XTL298 as a template. The resulting PCR product was then used to replace the nonfunctional copy of gatR present in strains DM08 and DM09 by the Wanner method (Datsenko and Wanner 2000). The DNA fragment containing the native sequence of gatR was obtained by target PCR using the primers yegS_F/gatD_R and E. coli HS as a template. Finally, this fragment was used to replace the counter selection cassette in the gatR locus of the recipient strains DM08 and DM09 by the Wanner method (Datsenko and Wanner 2000) and plated on double-counter-selective medium. Candidates for successful restoration of gatR were screened by PCR and tested for antibiotic resistances to confirm the loss of the cassette and the plasmid before sequencing.

To distinguish between gat-negative and gat-positive bacteria, we used the differential medium MacConkey agar supplemented with galactitol 1% and streptomycin (100 µg/mL). Plates were incubated at 30 °C. The frequency of gat-negative mutants was estimated by counting the number of white (auxotrophic for galactitol) and red colonies. On average around 1,000 colonies per time point were plated to estimate the frequency of gat-negative mutants, therefore the minimal frequency detected by this method is ∼0.5%.

In Vivo Experimental Evolution

In order to study E. coli’s adaptation to the gut we used the classical streptomycin-treated mouse model of colonization (Krogfelt etal. 2013) and performed the evolution experiments using the same conditions as before (Barroso-Batista etal. 2014; Barroso-Batista etal. 2015; Lourenço etal. 2016). Briefly, 6- to 8-week old C57BL/6 male mice raised in specific pathogen free conditions were given autoclaved drinking water containing streptomycin (5 g/L) for 1 day. After 4 h of starvation for water and food, the animals were gavaged with 100 µL of a suspension of 108 colony forming units of a mixture of YFP- and CFP-labeled bacteria (ratio 1:1) grown at 37 °C in brain heart infusion medium to OD600 of 2. After the gavage, all animals were housed separately and both water with streptomycin and food were returned. Mice fecal samples were collected for 24 days and diluted in PBS, from which a sample was stored in 15% glycerol at −80 °C and the remaining was plated in Luria Broth agar supplemented with streptomycin (LB plates). Plates were incubated overnight at 37 °C after which fluorescent colonies were counted using a fluorescent stereoscope (SteREO Lumar, Carl Ziess) to assess the frequencies of CFP- and YFP-labeled bacteria. These fluorescent proteins are used as neutral markers with which we can follow to detect the emergence of beneficial mutations, since these markers hitchhike with the beneficial mutations that spread in the populations (Hegreness etal. 2006).

In Vivo Competitive Assays

To test the invivo advantage of the evolving populations at the last time point of the evolution experiment (fig. 1), samples of either YFP or CFP clones isolated from day 24 (subpopulations) were competed against the respective ancestor labeled with the opposite fluorescent marker (n = 2 for each population). These subpopulations were composed of mixtures of ∼30 colonies with the same fluorescent marker isolated after plating the appropriate dilution of mice fecal pellets. The mixtures of clones were then grown in 10 mL of LB supplemented with chloramphenicol (100 µg/mL) and streptomycin (100 µg/mL) and stored in 15% glycerol at −80 °C. In vivo competitions of evolved subpopulations against the ancestral were performed at a ratio of 1:1, following the same procedure described above for the evolution experiment. Mice fecal pellets were collected for 3 days, diluted in PBS and frozen in 15% glycerol at −80 °C. Frequencies of CFP- and YFP-labeled bacteria were estimated using a fluorescent stereoscope (SteREO Lumar, Carl Ziess). The selective coefficient (fitness gain) of these mixtures of clones invivo (presented in fig. 1) was estimated as: , where s is the selective advantage of the evolved clone, Rf and Ri are the ratios of evolved to ancestral bacteria in the end (f) or in the beginning (i) of the competition and t is the number of generation per day. We assumed t = 18 generations, in accordance with the 80 min generation time estimated in previous studies on E. coli colonization of streptomycin-treated mouse (Poulsen etal. 1995; Rang etal. 1999).

Whole Genome Resequencing and Mutation Prediction

Clone Analysis

Two gat-positive clones (revertant) isolated from the second colonization (day 6 of population 2.15) were sequenced to test for the presence of additional mutations, besides the one reverting the gat-phenotype. Two independent colonies were used to inoculate 10 mL of LB and incubated at 37 °C with agitation for DNA extraction (following a previously described protocol; Wilson 2001). The DNA library construction and sequencing was carried out at the IGC sequencing facility. Each sample was paired-end sequenced on an Illumina Mi Seq Benchtop Sequencer. Standard procedures produced data sets of Illumina paired-end 250 bp read pairs. Genome sequencing data have been deposited in the NCBI Read Archive, https://www.ncbi.nlm.nih.gov/bioproject/PRJNA295680 (last accessed August 18, 2017). Mutations were identified using the BRESEQ pipeline (Deatherage and Barrick 2014). To detect potential duplication events, we used ssaha2 (Ning etal. 2001) with the paired end information. This is a stringent analysis that maps reads only to their unique match (with <3 mismatches) on the reference genome. Sequence coverage along the genome was assessed with a 250 bp window and corrected for GC% composition by normalizing by the mean coverage of regions with the same GC%. We then looked for regions with high differences (>1.4) in coverage. Large deletions were identified based on the absence of coverage. For additional verification of mutations predicted by BRESEQ, we also used the software IGV (version 2.1) (Robinson etal. 2011).

Population Analysis

DNA isolation, library construction, and sequencing were carried as described above for the clone analysis except that now it derived from a mixture of >1,000 clones per population grown in LB agar. Two populations, from the evolution experiment, were sequenced: 2.14 and 2.15. Those were sequenced for three time points during the adaptive period (generation 198 [day 11], generation 306 [day 17], and generation 432 [day 24]). Mean coverage per sample was between ∼100× and ∼300× for population 2.14 and between ∼70× and ∼115× for population 2.15. Mutations were identified using the BRESEQ pipeline (version 0.26) with the polymorphism option on. The default settings were used except for: 1) requirement of a minimum coverage of three reads on each strand per polymorphism; 2) eliminating polymorphism predictions occurring in homopolymers of length >3; 3) polymorphism predictions with significant (P = 0.05) strand or base quality score bias were discarded. Data presented in supplementary table S1, Supplementary Material online.

Diet Supplementation Experiment

To test the selective pressure exerted by galactitol, when present in the gut, we performed invivo competitions between gat-positive (DM08) and gat-negative (JB18) bacteria, whereas supplementing the diet with this sugar. Competition experiments were performed as described above except that the initial frequency of gat-positive bacteria was ∼10% and the drinking water was supplemented with 0%, 0.01%, or 0.1% of galactitol (n = 3 for each galactitol concentration) together with streptomycin (5 g/L). Fecal samples were collected daily and plated to access the frequencies of both phenotypes during 4 days.

Estimate of Gain-of-Function Mutation Rate from gat-Negative to gat-Positive Phenotype

Strains JB19 (gatC 1 bp Ins) and 4YFP (gatZ IS Ins) were grown overnight in 10 mL of LB at 37 °C with aeration. After growth, the total number of cells in the cultures was measured by flow cytometry (using BD LSR Fortessa; BD Biosciences) and ∼1,000 cells were used to inoculate 1 mL of LB (10 replicates of each strain) and incubated overnight. Aliquots of each replicate tube were plated in LB agar and MM agar supplemented galactitol 1% and incubated overnight at 37 °C. The number of spontaneous gat-positive mutants and total number of cells grown on LB were used to estimate the mutation rate using the maximum likelihood approach as implemented in FALCOR (Hall etal. 2009).

We measured the mutation rate to reacquire the ability to consume galactitol, since this locus was previously demonstrated to be a hotspot for loss of function mutations (Lourenço etal. 2016). The fluctuation test revealed that the spontaneous rate at which the ancestor of the second colonization (gat-negative) recovers the ability to consume galactitol is 9.7 × 10−9 (95% CI, [3 × 10−9, 1.9 × 10−8]) per generation. Considering that the estimated spontaneous rate of small indels in E. coli is ∼2 × 10−11 (Lee etal. 2012), and that about one third of such mutations in a region of ∼30 aminoacids of gatC could restore the open reading frame in the ancestral clone of the second colonization, the expected rate of mutation towards a gat-positive phenotype is ∼6 × 10−10. This estimate is between 5 and 15 times lower than that observed in the fluctuation assay, suggesting that a high mutation rate may have also contributed to the frequent reemergence of gat-positive bacteria in the evolution experiment.

Identification of Adaptive Mutations and Estimate of Haplotype Frequencies in Selected Populations of the Evolution Experiment

In order to estimate the haplotype frequencies depicted in figure 2, two complementary strategies were employed. In addition to the WGS of the populations, targeted PCR of the identified parallel mutations was performed. For the targeted PCR, 20–80 clones from different time points were screened (from populations 2.14 and 2.15) using the same primers and PCR conditions as in (Barroso-Batista etal. 2014; Lourenço etal. 2016). Because all target mutations correspond to IS insertions an increase in size of the PCR band is indicative of the presence of an IS. Frequency of gat-revertant was estimated by plating in differential media (MacConkey supplemented with 1% galactitol). Frequencies are depicted in supplementary table S1, Supplementary Material online.

Microbiota Analysis

We extracted DNA from mice fecal samples from two experiments: the second colonization (fig. 3) and the galactitol supplementation experiment (fig. 4). For the second colonization, we extracted DNA from mice where gat-positive E. coli could be detected (12 out of 15), at days 12 and 17 or 18 postcolonization (i.e., two samples per mouse). For the galactitol supplementation experiment, we extracted DNA at days 1, 2, 3, 4, and 7 postcolonization, for the nine mice involved in this experiment. Moreover, we extracted and sequenced a negative control (extraction without feces).

Fecal DNA was extracted with a QIAamp DNA Stool Mini Kit (Qiagen), according to the manufacturer’s instructions and with an additional step of mechanical disruption (Thompson etal. 2015). 16 S rRNA gene amplification and sequencing was carried out at the Gene Expression Unit from Instituto Gulbenkian de Ciência, following the service protocol. For each sample, the V4 region of the 16 S rRNA gene was amplified in triplicate, using the primer pair F515/R806, under the following PCR cycling conditions: 94 °C for 3 min, 35 cycles of 94 °C for 60 s, 50 °C for 60 s, and 72 °C for 105 s, with an extension step of 72 °C for 10 min (Caporaso etal. 2011; Caporaso etal. 2012). Samples were then pair-end sequenced on an Illumina MiSeq Benchtop Sequencer, following Illumina recommendations.

QIIME (Caporaso etal. 2010) was used to analyze the 16S rRNA sequences by first quality filtering the samples with a phred quality threshold of 20 (with script split_libraries_fastq) and then OTU picking by assigning operational taxonomic units at 97% similarity against the Greengenes database (DeSantis etal. 2006); using script pick_open_reference_otus with default settings, but enabling reverse strand matches). The OTU table generated by the latter script was then filtered in two steps. First, with script remove_low_confidence_OTUs, from the Microbiome helper repository (Comeau etal. 2017), to remove potentially spurious OTUs. Second, to remove contaminants (Salter etal. 2014) identified in the negative control, we filtered out any OTUs that were present in both the negative control and two or less samples at a frequency <0.01. The OTUs that were kept in the final OTU table and that were shared between negative control and samples had a maximum frequency of 0.004 in the negative control, were present at a frequency of >0.01 in at least five samples and had an average frequency in samples that was at least one order of magnitude higher than the observed in the negative control. The OTU table obtained after these two steps was then used for the analysis below. To understand the clustering pattern of the mouse gut microbiota in each experiment, we used the script jackknifed_beta_diversity to build rarefied OTU tables (at 75% of the minimum read depth in each experiment, as advised on QIIME’s website). This built 100 rarefied OTU tables, computed weighted UniFrac (Lozupone and Knight 2005) and generated a consensus tree (per experiment) with jackknife support of tree nodes. Consensus trees were then plotted with R package ggtree together with the OTU relative abundance (Yu etal. 2017).

To calculate alpha diversity, the OTU tables for each experiment were rarefied 100 times to even depth (multiple_rarefactions_even_depth), alpha diversity was estimated and collated (scripts alpha_diversity and collate_alpha), and the mean for each sample-metric combination was calculated (plotted in supplementary figs. S5 and S11, Supplementary Material online). To test whether there was an effect of cluster (second colonization) or galactitol concentration (galactitol supplementation experiment), we used linear mixed effects models with mouse as a random effect.

For the second colonization, UPGMA clustering identified two main clusters (fig. 6). Thus, we then used linear discriminant analysis effect size (LefSe) to identify if there were any taxa that were differentially abundant between the two clusters (Segata etal. 2011). The input data for LefSe (supplementary fig. S10, Supplementary Material online) was obtained after filtering the 97% OTUs for those that were present in at least six samples (i.e., at least three mice) and at a frequency that was above 0.5%, which corresponds to ∼100 reads.

Furthermore, for the galactitol supplementation experiment, we used the R package phyloseq (McMurdie and Holmes 2013) to compute the following beta-diversity metrics: Jaccard index, Bray–Curtis, unweighted and weighted UniFrac. We then tested for an effect of galactitol concentration on each beta-diversity matrix with PERMANOVA by using the adonis function with the vegan R package, with 1,000 permutations (Anderson 2001; Oksanen etal. 2009). These procedures were applied to data from day 3, 4, and 7 postcolonization or each of the last 3 days separately (as shown in supplementary figs. S6–S9, Supplementary Material online).

Natural Fecal Isolates Collection and Phenotyping

Fecal samples were collected from nine specific pathogen free mice of different litters (strain C57BC-67) and from nine healthy humans. Samples were weighted and resuspended in PBS with 15% glycerol before storage at −80 °C. To isolate lactose fermenting Enterobacteriaceae (red colonies), appropriate dilutions of the fecal samples were plated on MacConkey plates supplemented with 0.4% lactose and incubated overnight at 37 °C. The frequency of gat-positive bacteria among the lactose fermenting Enterobacteriaceae was estimated by replica-plating ∼96 isolated clones on M9 minimal medium supplemented with 0.4% galactitol. The M9 agar plates were incubated at 30 °C and bacterial growth was scored after 24, 48, 72, and 96 h. The ability to metabolize galactitol was tested by three independent trials.

Statistical Analysis

To analyze temporal dynamics data, we used linear mixed models, with mouse as a random effect. Where required, data were transformed to meet assumptions made by parametric statistics. All analysis were performed in R (R Core Team 2016).

Supplementary Material

Supplementary data are available at Molecular Biology and Evolution online.

Author Contributions

I.G. and A.S. designed the study with input from R.R. A.S., R.R., J.B., M.L., and D.G. performed the experiments. A.S., R.R., M.L., and I.G. analyzed the data and wrote the manuscript.

Click here for additional data file.