Gnotobiotic mouse model of phage–bacterial host dynamics in the human gut

Significance

A consortium of sequenced human gut bacteria was introduced into germ-free mice followed by a “staged” phage attack with virus-like particles purified from the fecal microbiota of five healthy adult humans. Unique phages were identified attacking microbiota members in nonsimultaneous fashion. Some host bacterial species acquired resistance to phage attack through ecological or epigenetic mechanisms. Changes in community structure observed after attack were transient. Spontaneous induction of prophages present in seven bacterial taxa was modest, occurring independently of the phage attack. Together, these results reveal a largely temperate phage–bacterial host dynamic and illustrate how gnotobiotic mouse models can help characterize ecological relationships in the gut by taking into account its most abundant but least understood component, viruses.

Abstract

Bacterial viruses (phages) are the most abundant biological group on Earth and are more genetically diverse than their bacterial prey/hosts. To characterize their role as agents shaping gut microbial community structure, adult germ-free mice were colonized with a consortium of 15 sequenced human bacterial symbionts, 13 of which harbored one or more predicted prophages. One member, Bacteroides cellulosilyticus WH2, was represented by a library of isogenic transposon mutants that covered 90% of its genes. Once assembled, the community was subjected to a staged phage attack with a pool of live or heat-killed virus-like particles (VLPs) purified from the fecal microbiota of five healthy humans. Shotgun sequencing of DNA from the input pooled VLP preparation plus shotgun sequencing of gut microbiota samples and purified fecal VLPs from the gnotobiotic mice revealed a reproducible nonsimultaneous pattern of attack extending over a 25-d period that involved five phages, none described previously. This system allowed us to (i) correlate increases in specific phages present in the pooled VLPs with reductions in the representation of particular bacterial taxa, (ii) provide evidence that phage resistance occurred because of ecological or epigenetic factors, (iii) track the origin of each of the five phages among the five human donors plus the extent of their genome variation between and within recipient mice, and (iv) establish the dramatic in vivo fitness advantage that a locus within a B. cellulosilyticus prophage confers upon its host. Together, these results provide a defined community-wide view of phage–bacterial host dynamics in the gut.

The human gut is home to tens of trillions of microbial cells representing all three domains of life, although most are bacteria. These organisms collaborate and compete for functional niches and physical locations (habitats). Together, they form a continuously functioning microbial metabolic “organ.” The microbial diversity, interpersonal variation, and dynamism of the human gut microbiota make the task of identifying the factors that define community configurations extremely challenging.

In some ecosystems, phages maintain high bacterial strain level diversity through lysis of their host strains (constant diversity dynamics model; refs 1, 2). The resulting emptied niche is filled with either an evolved resistant bacterial strain or a taxonomically closely related bacterial species. These dynamics have been observed in open marine environments (1). In contrast, a recent study of 37 healthy adults indicated that a person’s fecal microbiota was remarkably stable, with 60% of bacterial strains retained over the course of 5 y (3). Stability followed a power law dynamic that when extrapolated suggests that most strains in an individual’s gut community are retained for decades (3). In a metagenomic analysis of virus-like particles (VLPs) purified from the fecal microbiota of healthy adult monozygotic twins and their mothers, sampled over the course of a year, viral community structure exhibited high interpersonal variation. In contrast, the viral (phage) population within an individual was very stable over time, both at the level of sequence conservation and relative abundance (4). These observations, as well as other reports (5–7), suggest that temperate lifestyles, rather than a predator–prey relationship, dominate the phage–host bacterial cell dynamic in the distal guts of healthy humans.

To improve our understanding of viral–bacterial host dynamics, we constructed a gnotobiotic mouse model containing a simplified defined artificial community composed of 15 prominent human gut-derived bacterial taxa whose genomes had been sequenced (Dataset S1). This 15-member artificial community was used as bait for a staged attack that involved oral gavage of VLPs purified from human fecal samples. This system allowed us to (i) test whether phage populations would mount a simultaneous attack on susceptible members of the microbial community or whether such an attack would be nonsimultaneous (i.e., have an identifiable sequence), (ii) document the capture of previously unknown viruses present in the VLP preparations by members of the artificial community, and (iii) track induction of native prophages.

Results and Discussion

Attacking a 15-Member Artificial Human Gut Microbiota with VLPs Isolated from the Fecal Microbiota of Healthy Adult Humans.

Our experimental design consisted of three groups of germ-free C57BL/6J mice (n = 5 per group). Each group was kept in a separate gnotobiotic isolator, where each mouse was individually caged. The first group was gavaged with the 15-member artificial community at 8 wk of age. Three weeks later, they were each gavaged with a pool of VLPs (p-VLP) isolated from fecal samples obtained from five healthy humans (“live p-VLP group”). A “heat-killed p-VLP group” was also colonized with the artificial community but 3 wk later received a heat-killed version of the VLPs used for the first group. The third group did not receive a gavage of bacteria (“germ-free group”) but was gavaged with the same live p-VLP pool given to the first group. Fecal samples were collected from members of each treatment group at frequent intervals (Fig. S1).

Neither the bacterial gavage nor the p-VLP inoculum contained components that appeared to compromise gut barrier/immune function or perturb overall health status. At the time of sacrifice, none of the treatment groups exhibited any significant differences in total body weight or adiposity (epididymal fat pad weight as a percentage of total body weight; P = 0.3957 and P = 0.4794, respectively; Kruskal–Wallis test; Dataset S1). FACS analysis did not reveal any differences between the groups in the CD4⁺ and CD8⁺ T-cell compartments of their spleens or mesenteric lymph nodes (MLNs), as judged by CD44 and CD62L T-cell activation markers, Ki-67⁺ (proliferation marker), and FoxP3⁺ (Treg cells marker). For the germ-free group, Illumina shotgun sequencing of ileal and colonic contents obtained at the time of sacrifice disclosed that the p-VLP inoculum did not contain bacterial taxa (or bacterial spores) that could establish themselves in the guts of recipient animals (Dataset S1).

Microbial biomass, defined as nanograms DNA/milligram wet weight of feces, increased linearly and abruptly in the 4 d following introduction of the artificial community in both the live and heat-killed p-VLP groups (Fig. S2A). Fecal DNA concentrations correlated significantly with fecal bacterial cell counts (determined by flow cytometry; Spearman correlation, 0.843; Fig. S2B).

Because the genome sequences of the 15 bacterial taxa were known, we used community profiling by sequencing (COPRO-Seq; ref. 8), a method based on short read (50 nt) shotgun sequencing of total fecal community DNA, to quantify the relative abundance of each taxon as a function of time after initial colonization and after the staged VLP attack [2,544,433 ± 96,255 (mean ± SEM) reads per sample; Dataset S1]. Principal coordinates analysis of a Hellinger distance matrix constructed from the COPRO-Seq datasets showed that most of the variation in composition over time occurred during the period of initial community assembly (Fig. S3A). Changes in the relative abundance of community members also occurred following gavage of the live but not the heat-killed p-VLP preparation (Fig. 1 A–C and Fig. S3 B–N).

Fig. 1.

Sequential changes in the relative abundance of two members of the 15-member artificial human gut microbiota and correlation with the appearance of two previously undescribed phages. (A) Relative abundance plot for each bacterial species as a function of time for either the live p-VLP or the heat-killed p-VLP treatment groups. Mean values ± SEM are shown (n = 5 mice). The color key next to the plot indicates the identity of each bacterial species. (B and C) Plots of the relative abundance (fraction of the total community; mean ± SEM; n = 5 animals per treatment group) of B. caccae and B. ovatus in the fecal microbiota of gnotobiotic mice as a function of time before and after gavage with live purified VLPs pooled from the fecal microbiota of five human donors or a control heat-killed version of the same p-VLP preparation (time of gavage indicated by the upward pointing arrow; t = 0 on the x axis refers to the time of introduction of the 15-member artificial community into germ-free animals). The change in abundance of these Bacteroides spp. occurs in a reproducible sequence among individually caged mice that received live but not heat-killed p-VLPs. (D and E) Changes in the abundance of two phages, derived from the p-VLP sample, in the fecal microbiota of recipient gnotobiotic mice. Differences in the time course of change in bacterial and viral abundances are highlighted by the gray shading (lighter for B. caccae and ϕHSC01). Insets in D and E are assembled genome sequences for ϕHSC01 and ϕHSC02. The location of genes on the positive strand (green) and negative strand (red) are shown; those that have significant sequence similarity to known viral genes are colored blue (blastp E-value <10⁻⁵; Dataset S1). The inner plot represents GC skew based on 200-bp windows (yellow, G/C ratio is greater than the average for the genome; purple, ratio is lower than the average).

A Nonsimultaneous Pattern of Change in the Abundances of Five Phages.

To identify which exogenously administered VLP-associated viruses might be causing the observed structural rearrangements in community configuration, we modified our previously reported method for purifying VLPs (4) so that it could be applied to mouse fecal samples. We then sequenced DNA isolated from the purified VLP preparations [n = 27; two fecal pellets/VLP preparation, each amplified by multiple displacement amplification (MDA); 49,819 ± 6,983 (mean ± SEM) pyrosequencer reads per sample; Fig. S1 and Dataset S1]. To discriminate between activation of endogenous prophage in members of the artificial community versus exogenous viruses derived from the p-VLP preparation, reads generated from either the input VLPs or total mouse fecal DNA were mapped to the sequenced genomes of bacterial community members and to the mouse genome. We used reads without significant matches to either dataset to characterize viral genomes not represented in the starting 15-member artificial community.

In total, five viral genomes, none of which have been described previously, were assembled and annotated from these analyses. These viruses were detected in the gut communities of mice that had received the live p-VLP preparation but not in the heat-killed p-VLP group (Fig. 1 D and E, Fig. S4 A–H, and Dataset S1). Rather than finding a concurrent attack on all susceptible members of the model human gut microbiota, we observed a nonsimultaneous pattern of change in the abundances of these viruses with corresponding changes in the representation of community members.

A DNA virus with a circular 37-kb genome, human synthetic community phage 01 (ϕHSC01), was the first to significantly increase in abundance. It not only encodes typical phage proteins (e.g., terminase, tail protein, DNA polymerase; Dataset S1), but also a protein containing a Bacteroidetes-associated carbohydrate-binding often N-terminal (BACON) domain (Profam ID: PF13004) postulated to target glycoproteins and possibly host mucin (9). From the time of its first detection in feces 24 h after animals were gavaged with the live p-VLP preparation, the marked increase in abundance of ϕHSC01 over the course of the next 2 d correlated with a decrease in the abundance of Bacteroides caccae (R² = −0.446; P = 3.2 × 10⁻⁸ after Bonferroni correction; Fig. 1 B and D). No other community member showed a statistically significant inverse correlation, suggesting that this bacterium is a host for ϕHSC01. The drop in the relative abundance of B. caccae was abrupt, occurring over the course of 1 d between days 2 and 3 after VLP gavage. The 74.5 ± 3.7% decrease relative to pretreatment levels was followed by a recovery to 75.8 ± 5.2% of the pre-VLP gavage values within 3–6 d (Fig. 1). This fourfold decrease was independently validated using quantitative (q)PCR (from 3.4 × 10⁵ to 7.8 × 10⁴ genome equivalents/mg of fecal pellet; see SI Methods). The spike in viral abundance, just like the coincident reduction in B. caccae abundance, was remarkably consistent in terms of its magnitude and time course among the individually caged members of this treatment group.

Evidence That Phage Resistance Occurs Because of Ecological or Epigenetic Mechanisms.

To determine whether B. caccae’s recovery after viral attack was based on acquisition of identifiable fixed changes in its genome, we performed deep shotgun sequencing of total fecal community DNA isolated from samples obtained 9–19 d after bacterial gavage and 9–25 d after gavage of live and heat-killed p-VLPs. Pooling sequencing reads from these four groups of samples allowed us to assemble the B. caccae genome at an average coverage of 30-fold per treatment group, giving us enough resolution to identify mutations that could be responsible for conferring viral resistance. The results did not reveal deletions, insertions, or SNPs that were unique to the live p-VLP treatment group after the viral gavage and fixed in more than 10% of the B. caccae population (SI Results). Incorporation of short fragments of viral DNA within a locus flanked by short repeats [Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR, elements)] leads to bacterial resistance to viruses whose genomes have sequence similarity to the incorporated fragments (10). B. caccae does not contain any discernible CRISPR loci or associated proteins (Dataset S1). Moreover, no other prominent community member accumulated new spacers during the experiment (SI Results).

One interpretation of these findings is that phage “resistance” occurred because of ecological rather than genetic mechanisms. In this conceptualization, the gut environment consists of a number of microhabitats, some of which are occupied by B. caccae in ways that make it inaccessible to viral attack; the rise of B. caccae following the staged phage attack is not due to emerging resistance to the virus but rather an expansion of an unexposed population after the virus is washed out of the ecosystem. An alternative but not mutually exclusive possibility is that resistance is acquired due to phase variants or down-regulation of a phage receptor, making B. caccae resistant to viral attack without any discernible mutations in its genome (epigenetic mechanism). Another explanation is that mutations occur in one or more regions of the genome that are difficult to sequence and/or assemble.

Nonsimultaneous Detection of Viruses and Community Rearrangements.

The second virus to show an increase in abundance was ϕHSC02. All four predicted proteins encoded by this phage with a 6.2-kb genome (Fig. 1E) exhibit significant similarity to the Alpavirinae, a recently described subfamily identified from Bacteroidetes prophages, that belongs to the Microviridae (previously considered to be exclusively lytic) (11, 12). The changes in relative abundance of ϕHSC02 best correlated with a change (reduction) in the abundance of Bacteroides ovatus. Expansion of this virus and the attendant decrease in B. ovatus were first detected 2 d after the “crash” of B. caccae (i.e., within 5 d of gavage with the live p-VLP preparation) and coincided with the onset of recovery of the B. caccae population (Fig. 1 B–E).

The six- to eightfold reductions in relative representations of these two Bacteroides species took place during the 7-d period after gavage with live p-VLPs and were followed by a rise in abundance over a 7- to 8-d interval (Fig. 1 B and D). As these organisms increased their representation, we documented transient decreases in Bacteroides cellulosilyticus, as well as the two Bacteroides thetaiotaomicron strains present in the community. At the same time, levels of Parabacteroides distasonis, Clostridium symbiosum, Clostridium scindens, and Ruminococcus obeum rose. These successional changes in bacterial abundance were limited to the group of mice that had received the live p-VLPs (Fig. S3).

The rise and fall of these organisms in the live p-VLP group occurred during a time period when three other previously undescribed viruses appeared: ϕHSC03 (153.4 kb), ϕHSC04 (104.2 kb), and ϕHSC05 (95.7 kb). These viruses, initially detected 7 d after gavage of the live p-VLP preparation, subsequently increased in abundance to approximately equivalent levels and persisted during the remaining 14 d of the experiment (Fig. S4 C–H). Unlike the distinctive negative correlation between ϕHSC01 and B. caccae abundance, and subsequently ϕHSC02 and B. ovatus abundance, the simultaneous appearance and rise of ϕHSC03, ϕHSC04, and ϕHSC05, their subsequent persistence, and the coincident complex patterns of change in the abundances of bacterial community members during this later period of the experiment made it difficult for us to assign candidate bacterial hosts to these three previously undescribed phages. Therefore, the same live p-VLPs used for the in vivo staged phage attack were also used for in vitro “attacks” of monocultures of each of the species present in the 15-member artificial community (SI Methods). There was no enrichment of any of the five previously undescribed viruses in any of the cultures (n = 2 independent experiments), suggesting that the bacterial host susceptibilities and requirements for infection with these phages are not recapitulated under these in vitro conditions (SI Results and Dataset S1). This observation highlights the value of gnotobiotic animal models for isolation of previously undescribed gut viruses.

Tracking the Origin of Each of the Five Phages Among the Human Donors as Well as Their Genome Variation Between and Within Recipient Mice.

To determine whether ϕHSC01–ϕHSC05 were distributed among all of the human VLP donors or whether they were unique to particular individuals, we generated a hybrid assembly using reads from the original VLP-derived viromes from each of the five donors as well as from the p-VLP preparation used for gavage (SI Methods). The hybrid assembly yielded 159 contigs greater than 2 kb. Based on read distribution and percent identity in contigs, we concluded that ϕHSC01 and ϕHSC04 originated from a twin in family 2 (F2T1.2), whereas ϕHSC02 and ϕHSC03 originated from the twin in family 4 (F4T1.2) (Fig. S5A). ϕHSC05 was observed in four of the five individuals used to construct the VLP pool. Mapping reads from each human donor fecal virome to ϕHSC05 revealed that the virus recovered from mice likely came from individual F3T1.2 because the average percent identity of reads from this person’s virome mapping to the ϕHSC05 genome was equivalent to the percent identity obtained from VLPs isolated from mouse fecal samples (Fig. S5B). Nonetheless, analyzing VLP DNA purified from fecal samples obtained from each recipient mouse 9 d and 17 d after gavage of the live p-VLPs, and from cecal samples obtained at sacrifice, revealed animal-to-animal variations in ϕHSC05 genome structure (Fig. S5C). These genomic variations could have been present in the human donor fecal virome and distributed to different mouse recipients. However, we also observed variations in ϕHSC05 genome structure within mice over time (Fig. S5C), raising the possibility of a red queen dynamic (evolution of the viral genome in response to evolution of the bacterial host over time). It is important to note that none of the other four phages (ϕHSC01–4) showed this type of variation between animals, or within individual animals as a function of time (Fig. S5 D–G), indicating that ϕHSC05’s genome evolution is not a general feature in mice harboring this artificial community. Together, our findings not only illustrate the utility of gnotobiotic mice for identifying candidate bacterial hosts for human gut-associated phage, but also for identifying differences in properties among related virotypes derived from different human gut viromes.

Distribution and Persistence of Phages Within the Gut.

Intestinal transit time in mice is on the order of several hours (13). We collected intestinal contents from the proximal and distal small intestine, cecum, and colon, as well as a fecal sample from each mouse in the live and heat-killed p-VLP treatment groups at the time of their sacrifice (25 d after the staged p-VLP attack). All gut samples were processed for COPRO-Seq analysis, and an aliquot of cecal contents was used to isolate VLPs for subsequent shotgun sequencing of viral DNA (Dataset S1). The results revealed no detectable phages in any of the gut segments from mice that received the heat-killed p-VLP preparation (Fig. S6A). In the live p-VLP treatment group, neither ϕHSC03, ϕHSC04, nor ϕHSC05 exhibited significant differences in their relative abundances between the distal small intestine and distal colon, and between luminal contents and feces (Fig. S6A). Moreover, at the time of sacrifice, there were no significant biogeographical differences in the relative abundance of bacterial species within or between members of the two treatment groups (Fig. S6B).

The fact that ϕHSC03, ϕHSC04, and ϕHSC05 first appeared in members of the live p-VLP treatment group 7 d after the single gavage of p-VLPs suggests that an intra- and/or extracellular compartment/reservoir exists that harbors components of the administered human fecal phage population. Pseudolysogeny, a state where phages exist in a host bacterial cell without multiplying or synchronizing their replication with the host (14) could represent one potential mechanism for persistence. Hypervariable domains, including C-type lectins, have been identified in gut-associated phages (15), suggesting that extracellular sequestration with binding to mucus or epithelial cell surface glycans could represent another potential mechanism. Barr et al. found that enrichment of phages in mucus occurs through binding of Ig-like domains exposed on phage capsids to carbohydrate residues present in the mucin glycoprotein component of mucus, thereby creating a form of antimicrobial defense that could protect mucosal surfaces (16). COPRO-Seq analysis of cecal samples obtained from mice in the germ-free treatment group that lacked the 15-member artificial community and were gavaged with the live p-VLP preparation alone revealed no detectable phages in the cecum at the time of sacrifice (Dataset S1), supporting the notion that persistence of ϕHSC03–ϕHSC05 may be dependent upon the presence of bacteria. [Note that members of the gut microbiota, including Bacteroides spp. represented in the artificial community, are known to impact the mucus layer and mucosal glycans through a variety of means (17–19).] Although we cannot completely rule out the possibility that ϕHSC03 and ϕHSC04 were first detected at later time points because there had been a selection for mutants that could replicate better, this seems unlikely; their late appearance was a common feature in mice receiving the live p-VLP preparation, and, as noted above, there was no obvious variation in their viral genomes between animals and within a given mouse over time (Fig. S5 F and G).

Prophage Activation in the 15-Member Artificial Community.

Thirteen of the 15 bacterial taxa in the artificial community had predicted prophages in their genomes (Dataset S1). To verify these predictions and to assess the capacity of these prophages to undergo induction, we used reads obtained from shotgun pyrosequencing of DNA isolated from two sources: (i) VLPs purified from fecal samples collected at weekly intervals from animals gavaged with live p-VLP preparation and (ii) VLPs purified from cecal samples obtained at sacrifice. Instead of mapping randomly throughout bacterial genomes (implying a background level of bacterial DNA contamination in the purified VLPs), VLP reads mapped to one or more of the predicted prophages. In this way, we identified 10 prophages derived from seven bacterial genomes that had the capacity to undergo induction in vivo (Fig. S7).

B. cellulosilyticus WH2 has two prophages, one of which (prophage 1) exhibited the greatest fold-induction among these 10 prophages. Prophage 1 has a lambdoid genome architecture (Fig. 2A) with syntenic arrangements identified in other Bacteroides genomes (Fig. S8A). Its induction was observed in all mice 5–9 d after initial gavage of the 15-member artificial community prior to introduction of either live or heat-killed p-VLPs. Induction occurred at the end of the period of initial bacterial community assembly, right after microbial biomass reached its peak (Fig. S2A), suggesting a potential role of bacterial density in the induction process. Induction of prophage 1 correlated with a decrease in the relative abundance of its bacterial host (Fig. 2 B and C). The other B. cellulosilyticus prophage (prophage 2) did not exhibit significant levels of induction at any time point surveyed during the experiment (Fig. 2D and Fig. S8B; also see SI Results showing that prophage 2 can be induced in vitro).

Fig. 2.

Prophage induction in B. cellulosilyticus WH2. (A) VLP-derived sequencing reads from mouse fecal samples mapped to a 150-kbp fragment of the B. cellulosilyticus WH2 genome containing prophage 1. The y axis corresponds to the log (10) of the read coverage (blue) for a given position in the prophage genome. Mapping VLP reads to the bacterial genome identified the prophage insertion site at an arg-tRNA gene, with the corresponding duplicated region generating the attachment (att) sites. No reads were obtained from potential cos sites (red arrow, zoomed-in fragment). (B–D) Relative abundance (mean ± SEM) of the bacterial host and its prophage in the fecal microbiota of individually caged mice. Relative abundance was measured based on the ratio of COPRO-Seq reads mapping to each prophage and elsewhere in the bacterial genome. Bacterial relative abundance was scaled to its community relative abundance (Fig. S3B); prophage abundance was also scaled accordingly. Equivalent abundances correspond to phage in an uninduced state, whereas relative increases in prophage abundance indicate induction.

B. cellulosilyticus WH2 was represented in the 15-member artificial community by a library of 93,458 isogenic mutants, with each mutant strain containing a single randomly inserted modified mariner transposon (Tn) (91.5% of predicted ORFs had insertions covering the first 80% of each gene with an average of 13.9 insertions per ORF). Because the modified Tn had engineered recognition sites for the type II restriction endonuclease MmeI at its ends, 16 bp of flanking chromosomal DNA could be excised together with the Tn after MmeI digestion of community DNA and sequenced (20). This makes it possible to use high throughput sequencing to define the precise location and abundance of each transposon mutant in the library (SI Methods). Comparing the number of reads for each mutant in an “output” population after a given selection to the number of reads generated from an “input” population provides information about the effect each transposon insertion had on the fitness of the organism under the selection condition applied (20, 21).

Applying this Tn INsertion Sequencing (INSeq) analysis to DNA prepared from fecal samples collected before, during, and after prophage 1 induction showed a dramatic enrichment for transposons located within a ∼600bp intergenic region positioned between the ORFs encoding the prophage’s putative Rha protein (22) and cI repressor at the time of its induction 5–9 d after introduction of the 15-member artificial community (Fig. S9 A and B and SI Results). The intergenic region upstream of cI in phage lambda is an extremely well-studied transcriptional regulatory region; the right operator (O_R) with its three sites that competitively bind the repressor and Cro proteins, constitutes a carefully regulated switch between lysogenic and lytic cycles (23, 24). Upon RecA activation, the repressor is cleaved and the prophage is induced (25). Thus, accumulation of mutations in this region could have important consequences on the regulation and lifecycle of the lambdoid prophage 1 and its bacterial host.

In some mice, enrichment of strains with Tn inserts in the cI-rha intergenic region was observed as early as 2 d after gavage of the 15-member artificial community (before prophage induction). Enrichment did not reflect clonal expansion of a single mutant strain within a given animal, but rather expansion of 1 or more of 10 independent mutants, each harboring a single transposon insertion within this intergenic region. The number and sites of these insertion mutants varied between animals (Fig. S9A). Moreover, no Tn insertions were observed within the ORF encoding the putative cI repressor or in the region 100 bp immediately upstream of the ORF in the input library, nor in any of the output fecal samples (Fig. S9A), suggesting an essential role for the repressor and the upstream O_R region in bacterial host fitness (Fig. S9 C and D).

Control experiments were carried out for 20 d (the time before p-VLP gavage) using the same 15-member artificial community but where B. cellulosilyticus WH2 was represented by the wild-type strain rather than by a library of Tn mutants. Although community assembly and structure were highly similar to that observed when the artificial community contained the Tn mutant library (compare Fig. S8C with Fig. 1A), neither prophage 1 nor prophage 2 were induced, and no drop in B. cellulosilyticus WH2 abundance was observed (Fig. S8D). These findings suggest that Tn mutations in the cI-rha intergenic region facilitate prophage induction in the mouse model.

We defined the time course of clonal expansion of bacteria containing Tn inserts in this intergenic region to quantify the fitness effects of disrupting this part of the prophage genome. Reads mapping to this locus represented 77.9 ± 4.6% (mean ± SEM) of all Tn reads in fecal samples collected between 5 d and 9 d after bacterial gavage (range, 61.3–96.9%). A sliding window analysis was performed to determine if any other 600-bp region of the B. cellulosilyticus WH2 genome containing a Tn insertion went through a clonal expansion analogous to that documented for the cI-rha intergenic region during the first 31 d of the experiment in mice belonging to the live and heat-killed p-VLP treatment groups. The results revealed that on average any given 600-bp window with a Tn decreased its abundance over time, usually to less than 0.001% of the B. cellulosilyticus WH2 population. Only 5–10% of the windows other than the cI-rha intergenic region exhibited any enrichment over time, with less than 0.1% of the windows reaching levels >1% of the population (Fig. S9B). However, all of these other enrichments occurred 11 d or more following gavage when the bacterial host population was recovering from prophage induction (Fig. S9B). Importantly, strains with the Tn-containing cI-rha intergenic region selected for before and during prophage 1 induction subsequently maintained high relative abundance (∼4%) in the B. cellulosilyticus population in both the live and heat-killed p-VLP treatment groups (Fig. S9B).

These results indicate that prophage 1 induction is restricted in time (i.e., nonrecurring over the course of the experiment) and insensitive to the attack of other members of the artificial community by exogenous human fecal phage. Together, the data demonstrate how disruption of an intergenic region, located just upstream of the predicted O_R region that functions as a switch between lysogenic and lytic cycles in other lambda phages, and between a putative repressor and antiterminator, is capable of conferring a fitness advantage to its bacterial host strain before and independently of prophage induction.

Prospectus.

Gnotobiotic mice containing defined consortia of sequenced human gut bacterial symbionts provide a tractable system that is more realistic than in vitro approaches for characterizing phage–bacterial host dynamics. These mice disclosed that: (i) a deliberately executed phage attack with a mixture of diverse human phages did not result in a simultaneous attack of all susceptible members of the artificial human gut microbiota, but rather was manifested by a succession of changes in the abundance of a subset of its bacterial taxa; (ii) phage resistance can occur through an ecological or epigenetic mechanism (i.e., without changes in bacterial CRISPR elements or bacterial genes encoding cell surface markers); (iii) one phage that was widely distributed among the five human donors but only reached low abundance in mice exhibited variations in its genome sequence in gnotobiotic animals over time, raising the possibility of red queen dynamics in this case; and (iv) prophages contain important in vivo fitness determinants for their host strains; these determinants can reside in regulatory regions responsible for prophage induction (as illustrated by B. cellulosilyticus WH2).

The model gut microbial community that these animals harbored was remarkably resilient with several fold changes in the relative abundance of different taxa occurring for only brief periods of time. Nonetheless, the identified targets of phage attack (B. caccae and B. ovatus) did not fully return to preattack levels in the artificial community or to levels seen in mice receiving the heat-killed p-VLPs that were sampled at corresponding time points, suggesting modest long-lasting effects. The transient nature of the changes in community structure observed after p-VLP attack, and the fact that prophage induction was modest in most bacterial hosts, support the view that the phage–bacterial host dynamic in this simplified defined gut ecosystem is predominately temperate rather than lytic (4).

Our findings extend previous work analyzing the dynamics of well-characterized T4 and T7 phages in gnotobiotic mice monocolonized with Escherichia coli, where their in vitro behavior was a very limited predictor of their in vivo behavior (26). Using a more complex defined microbiota, we also observe a diverse range of viral dynamics: (i) exogenous viruses rising in abundance soon after their introduction (with a corresponding decrease in abundance of their putative bacterial hosts) followed by depletion of these viruses from the community without any obvious trace of genetic resistance or adaptation; (ii) exogenous viruses that survive at undetectable levels for almost a week before increasing their abundances, and in one case, displaying genetic variability over time; and (iii) basal levels of induction of 10 prophages with only one prophage achieving a level that produced detectable alteration in the abundance of its bacterial host. Duerkop et al. also showed that prophage induction differed in vitro and in vivo, and how under the appropriate conditions induction could provide a fitness benefit to the bacterial host (27), further highlighting the importance of a temperate lifestyle in the gut, and why prophages are so widely distributed in gut bacterial genomes without necessarily being induced at significant levels. Our findings are also consistent with the previous finding that T7 phages are capable of surviving at undetectable levels for 1 wk in germ-free animals before they rise in abundance after gavage of a bacterial host (26); this capacity to maintain infectivity has potential implications for preventive phage therapy.

We envision a future where gnotobiotic mouse models allow “personalized” assessment of phage–bacterial interactions. Complex mixtures of VLPs, isolated from previously frozen fecal samples obtained from human donors representing various ages, physiological or disease states, or geographical regions/cultural traditions of interest, can be introduced into recipient mice harboring a defined collection of human gut community members. The community can be used as a “filter” to identify and assemble the genomes of phages present in the human donor viromes and link them to bacterial hosts. Miniaturization of methods for preparing VLPs from mouse fecal pellets provides a way for purifying these phages and at the same time verifying that they have lytic activity. The system is capable of distinguishing very closely related virotypes present in multiple human gut microbiota based on their differential ability to establish themselves in recipient gnotobiotic mice. These attributes not only provide a discovery pipeline that complements metagenomic surveys of the human gut virome by identifying phages “buried” in large gut microbiome datasets, but facilitate identification of phages that can be used as experimental tools to deliberately manipulate model microbial communities or that can be considered as candidate therapeutic agents.

Methods

Protocols for the recruitment of human subjects and sampling of their fecal microbiota were approved by the Washington University Human Research Protection Office. All experiments involving mice were performed with protocols approved by the Washington University Animal Studies Committee. Procedures for gnotobiotic mouse husbandry, introduction of VLPs purified from human fecal samples into mice, sampling fecal microbiota from gnotobiotic mice, isolation of total DNA from mouse feces and intestinal contents, quantification of microbial cells in fecal samples by flow cytometry, preparation of DNA libraries for Illumina HiSeq or MiSeq sequencing, preparation of VLP DNA from mouse fecal samples, 454 pyrosequencing of VLP-derived DNA, COPRO-Seq analysis, assembly and annotation of viral genomes, cross-contig comparisons, INSeq analysis of fitness determinants present in the B. cellulosilyticus WH2 prophage, PCR quantification of B. caccae abundance in the fecal microbiota of gnotobiotic mice, CRISPR analysis, PCR determination of Tn insertions in uninduced and induced B. cellulosilyticus WH2 prophage 1, in vitro induction of B. cellulosilyticus WH2 prophages, and in vitro assays for bacterial host tropism of phages represented in the pooled human fecal VLP preparation are described in SI Methods.