Long-term persistence of crAss-like phage crAss001 is associated with phase variation in Bacteroides intestinalis

Background The crAss-like phages are ubiquitous and highly abundant members of the human gut virome that infect commensal bacteria of the order Bacteroidales. Although incapable of lysogeny, these viruses demonstrate long-term persistence in the human gut microbiome, dominating the virome in some individuals. Results Here we show that rapid phase variation of alternate capsular polysaccharides in Bacteroides intestinalis cultures plays an important role in a dynamic equilibrium between phage sensitivity and resistance, allowing phage and bacteria to multiply in parallel. The data also suggests the role of a concomitant phage persistence mechanism associated with delayed lysis of infected cells, similar to carrier state infection. From an ecological and evolutionary standpoint, this type of phage-host interaction is consistent with the Piggyback-the-Winner model, which suggests a preference towards lysogenic or other “benign” forms of phage infection when the host is stably present at high abundance. Conclusion Long-term persistence of bacteriophage and host could result from mutually beneficial mechanisms driving bacterial strain-level diversity and phage survival in complex environments. Supplementary Information The online version contains supplementary material available at 10.1186/s12915-021-01084-3.


Background
The crAss-like bacteriophages are a recently described family of dsDNA-tailed bacterial viruses of the order Caudovirales that are predicted to infect bacteria of the phylum Bacteroidetes [1]. They are present in a variety of host-associated, aquatic, and terrestrial habitats, but are especially predominant in the human gut [2][3][4].
The current provisional classification of crAss-like phages based on gene sharing networks lists ten genera in four subfamilies (Alphacrassvirinae, Betacrassvirinae, Gammacrassvirinae, and Deltacrassvirinae) [5]. Subfamily Alphacrassvirinae includes the prototypical crAssphage (p-crAssphage), discovered in silico in 2014 through cross-assembly of metagenomic reads from multiple unrelated human gut microbiota samples [6]. The presence of p-crAssphage in >70% of human gut metagenomes together with its exceptional abundance in some of the samples (>90% of the gut virome and >20% of total fecal DNA) led to its description as "the most abundant virus of the human body" [4]. Furthermore, p-crAssphage was found to be widespread in the global human population with sequence variants correlating with geographic location [7]. Based on the observation of related genomes in the non-human primate gut, it was proposed that p-crAssphage shares a long co-evolutionary history with the human species. In addition, limited dietary associations, but no link to disease was observed in crAssphage colonization patterns [7]. Despite best efforts, p-crAssphage and other closely related viruses could not be isolated in pure culture. At the same time, these viruses could be grown in vivo after transfer into oligoxenic mice or after the introduction of crAssphage-rich microbiota transplants into human patients, and ex vivo as part of a native fecal community in a bioreactor [5,[8][9][10].
Other members of the family belonging to subfamilies Alphaand Betacrassvirinae, moderately abundant in the human gut, appear to be cultivable. The exemplar strains were isolated from human feces (ΦcrAss001 and ΦcrAss002), wastewater effluent (DAC15 and DAC17), and seawater (Φ14:2), infecting Bacteroides intestinalis [11], Bacteroides xylanisolvens [12], Bacteroides thetaiotaomicron [13], and Cellulophaga baltica, respectively [14]. The genus Bacteroides, predicted or demonstrated to serve as host to many crAss-like phages [6,11,12], comprises strictly anaerobic Gram-negative non-spore-forming bacteria, many of which are predominant members of the human gut microbiome and are highly adapted to the gut environment [15]. Genome plasticity, extensive regulation of gene expression, and high adaptability to changes in their environment are characteristic features of this bacterial group [16]. Phase variation of surface structures resulting from DNA inversions catalyzed by serine and tyrosine recombinases has been implicated as an important mechanism in Bacteroides for adaptation to environmental change, improved fitness, and efficient colonization of the mammalian gut [17][18][19][20][21].
A recent longitudinal study of the human gut virome highlighted long-term (months and even years) highlevel persistence of different combinations of crAss-like phages across individuals [10]. When vertically transmitted from mothers to their infants or transplanted in the course of fecal microbial transplantation (FMT), crAsslike phage engraft successfully and persist stably [8,22]. CrAss-like phages can also persist in pure bacterial culture in vitro over many serial passages without any apparent effect on the density of the host population [11]. The overall effect of stably persisting crAss-like phages, sometimes reaching absolute concentrations of 10 11 genome copies per gram of feces [10], on the structure (composition) and function of the gut microbiome is yet to be determined. Despite a lack of disease associations with p-crAssphage, correlations with other candidate genera and individual strains within the family of crAsslike phages have not been thoroughly examined. Certain recent studies hinted at the depletion of certain crAsslike viral clusters in an IBD patient cohort [23], while at the same time a member of Betacrassvirinae, termed IAS virus, was prevalent in HIV patients with unexplained diarrhea [24].
It remains unclear what mechanisms are responsible for the long-term persistence and high prevalence of crAss-like phages in the human gut. Several plausible mechanisms of phage persistence are possible, including spatial heterogeneity of microbial habitats in the gut [25], physiological stochasticity of the phage sensitivity phenotype in their host bacteria [26], reversible genetic switching (phase variation) of surface receptor expression [27], constant sweeps of new mutations leading to arms race co-evolution with their hosts [28], or perhaps, an unusual life cycle (carrier state infection, pseudolysogeny) of the viruses themselves. Establishing these mechanism(s) is a challenging task due to the lack of tools for genetic manipulation, and the low similarity of crAsslike phage genomes and proteins to well-characterized model viruses [6]. Our limited understanding of the biological properties of crAss-like phages derives from ΦcrAss001, DAC15, and DAC17 [11,13].
Here, we report the first comprehensive effort to characterize the interaction of crAss-like phages with their hosts, focusing primarily on crAss001. We demonstrate that phase variation of capsular polysaccharides creates dynamic sub-populations of sensitive and resistant Bacteroides cells and drives long-term phage persistence. A yet unexplained mechanism is apparently responsible for a paused infection cycle and delayed release of phage progeny from a significant fraction of infected cells, a phenomenon that may represent a novel type of carrier state infection.

Results
CrAss-like phages are capable of long-term stable persistence in the mammalian gut and select for resistance in their bacterial host We have previously demonstrated the long-term persistence of crAss-like phages in the human gut microbiome [10]. The levels of crAss-like phage colonization in ten subjects varied from <10% of the total fecal virome to nearly 100%. Eight of the ten subjects were stably colonized by one or several individual crAss-like phage strains. To reinforce this observation, we followed the dynamics of crAss-like phage persistence in a monoxenic mouse model using a cultivable representative of the family, ΦcrAss001. Germ-free C57BL/6N mice (n = 6) were colonized by gavage on three consecutive days with a mixture of 10 9 cfu of B. intestinalis APC919/174, and 2x10 8 pfu of ΦcrAss001 per gavage. Mice (n = 6) colonized with B. intestinalis alone served as a control. Within the first few days upon bacterial gavage, fecal B. intestinalis reached levels of 10 11 cfu g -1 in both the presence or absence of phage (p>0.05 in Kruskal-Wallis test with FDR correction for all time points except day 31). Between days 7 and 17, ΦcrAss001 reached stable levels of 10 6 -10 8 pfu g -1 (no viable phage could be recovered from control group feces). With some fluctuation over time, the phage persisted until the termination of the experiment at 136 days (Fig. 1a).
Long-term persistence of the phage-host pair quickly results in the emergence of a ΦcrAss001-resistant subpopulation in B. intestinalis. By day three from the first gavage, 11/17 (65%) isolates of B. intestinalis from murine feces were completely resistant to phage in spot and plaque assays, compared to 2% of spontaneously resistant cells in the original culture used for the inoculation of animals (see below). By week 12, 15/29 (52%) clones were resistant. The sensitive clones showed varied degrees of partial resistance that was manifested by the formation of spots of variable turbidity. Unexpectedly, isolates from the phage-free animals also showed a switch towards the phage-resistant phenotype. While day three isolates were all sensitive (17/17), 8/29 (27%) of clones from week 12 were resistant to a varying extent, showing that a switch between sensitive and resistant phenotypes can occur in the host strain independent of the presence of phage. This persistence can be recapitulated in vitro in a B. intestinalis culture infected in its early logarithmic phase at a multiplicity of infection (MOI) of 1, followed by overnight incubation and daily passages of the resulting phage-bacterial co-culture. Initial adsorption of phage resulted in 96.5% reduction of bacterial CFU after 30 min (Fig. 1b). This was followed by the first round of phage genome replication, which was largely completed by 90 min after infection, resulting in a calculated virus-to-microbe ratio (VMR) of Fig. 1 Long-term persistence of crAss-like phages in the gut microbiome and in vitro. a Persistence of ΦcrAss001 in the gut of C57BL/6NTac mice (n = 6) mice colonized with ΦcrAss001/B. intestinalis APC919/174 mixture at an MOI=1; the control group of mice received oral gavage of B. intestinalis only; b replication of ΦcrAss001 in an exponentially growing culture of B. intestinalis (OD600=0.3 at infection with an MOI=1) followed by four daily transfers of phage/host co-culture in fresh broth; bacterial and free phage counts were obtained by plating and plaque assays respectively; c VMR determined using shotgun metagenomic sequencing of total DNA in the same phage/host co-cultures as in b, with additional samples collected after 18 and 23 daily transfers. A lag between completed DNA replication in the first infection cycle (90 min, [b]) and release of phages progeny (120-180 min, [c]) is noticeable ~22.8±1.2 (median±IQR, Fig. 1c). Parallel propagation of bacteria and phage for the first 24 h maintained roughly the same level of VMR and resulted in a high titer of phage particles 3.86x10 10 ±7x10 9 pfu ml -1 (Fig. 1b). Subsequent daily transfers led to a drop of VMR and phage titers, but resulted in increased viable bacterial titers and a phage/host dynamic that is stably maintained for as many as 23 daily transfers (Fig. 1c). Agreeing with previous observations in non-crAss-like phages and B. thetaiotaomicron [27], this evidence again points towards a rapid shift to dominance by resistant cells with retention of a sufficient sensitive sub-population to allow for limited phage reproduction and persistence.
Resistance of B. intestinalis cells to ΦcrAss001 develops at a high rate in vitro, but is reversible and associated with inhibition of phage adsorption ΦcrAss001 forms turbid spots in soft agar overlays containing B. intestinalis APC919/174, confirming that bacterial growth can occur in the presence of phage (Fig. 2a). This could be a result of a high rate of lysogeny or could be a consequence of a pre-existing resistant subpopulation. To distinguish between these possibilities, transmission electron microscopy (TEM) was conducted on material collected from the center of the spot shown in Fig. 2a. Consistent with the hypothesis that a pre-existing resistant subpopulation of bacteria are present in the spot material, the electron micrographs revealed the presence of both uninfected and infected cells among the cell debris, including many instances in which cells in the process of the division had phage progeny restricted to only one of the two nascent daughter cells (Fig. 2b-f).
In the presence of an excess of phage in soft agar overlays (efficiency of plating assay, EOP), clonal cultures of B. intestinalis APC919/174 had an immediate resistance rate of 2.06±0.14% (mean±SD, Fig. 2g, h), which is in close agreement with 96.5% cells being infected. None of the tested colonies (n = 100) contained ΦcrAss001 DNA detectable by PCR after three passages. When ten resistant clones were each sub-cultured into 12 separate clones, four of the 120 (3.3%) reverted to a sensitive phenotype. Resistant clones demonstrated a dramatically reduced ability to adsorb phage, while the spontaneously sensitive "revertants" had restored phage adsorption (Fig.  2j). Some of the sub-clones had altered colony morphologies, with resistant colonies being more transparent, while sensitive ones were opaque (e.g., sub-clones 8T and 8W in Fig. 2i). When a resistant clone 8T was subjected to an EOP assay the rate of survival was raised to 88.7±3.3%, but in the sensitive sub-clone 8W it had returned closer to the original level of 11.3±0.9%. TEM of 8T cells revealed an altered "hairy" appearance, with a loose and bulky polymeric substance indicative of production of a different type of capsular polysaccharide (CPS) (Fig. 2k). In contrast, the wild-type (WT) culture demonstrates a much more compact thinner layer of capsular polysaccharide intimately attached to the outer membrane (Fig. 2l).
Together, these data suggest that cultures of B. intestinalis contain mixed populations of cells. Approximately 2% of cells in naïve cultures are resistant to phage. Exposure to phage in continuous culture in vitro or in the murine gut drives the bacterial population towards phage resistance, but this does not reach 100% or lead to the extinction of the phage under the conditions tested. We hypothesize that the transient phage-resistant phenotype results from constantly ongoing reversible switching of expression of cell surface structures through some genetic or epigenetic mechanism Phase variation of multiple surface loci and concomitant capsular polysaccharide switching in B. intestinalis In order to identify the mechanism of resistance, we subjected ten resistant clones and two revertant sensitive sub-clones ( Fig. 2j) to whole genome sequencing. We also performed shotgun sequencing of samples collected from the long-term in vitro persistence experiment 5 min-28 days following the inoculation and some of the fecal samples collected in the long-term murine colonization experiment. For the wild-type strain, resistant clone 8T ("transparent"), and its corresponding revertant sensitive clone 8W ("white"), complete circular genomes were assembled. A comparison of the three assemblies revealed that the only detectable differences consist of the inversion and rearrangement of several genomic loci (Additional file 1: Fig. S1a). Similar to previous observations with B. thetaiotaomicron [27], three types of events were detectable, including (a) large inversions involving entire gene clusters comprised of genes coding for TonB-dependent transporters [29] and/or RagB/SusD-like nutrient uptake proteins [30], often flanked with a dedicated integrase/recombinase gene; (b) a more complex type of reshuffling of clusters containing ABC-transporter genes (Additional file 1: Fig. S1b); and (c) the inversion of small, <200 bp regions adjacent to capsular polysaccharide biosynthesis operons that contain potential promoters. Remarkably, all these genomic loci, designated as "phase variable regions" (PVRs), can be directly implicated in phage sensitivity. Their encoded products are either components of cell surface structures with a potential to serve as receptors or barriers for phage adsorption or potential phage defense molecules (e.g., restriction/modification system).
Transcriptomic analysis of serial daily broth cocultures of B. intestinalis and ΦcrAss001 1-5 days after the initial inoculation revealed an array of genes showing divergent responses to the presence of phage (p<0.01, DESeq2). The strongest transcriptional response in the Fig. 2 Resistance of B. intestinalis APC919/174 to ΦcrAss001 is reversible and associated with capsular polysaccharide alterations and loss of phage adsorption. a Spotting of 4x10 7 pfu of ΦcrAss001 on B. intestinalis APC919/174 agar overlay results in turbid clearing zone; b-f TEM (x9900-105,000) of ultra-thin sections (80 nm) of samples taken from the center of the spot shows the presence of un-infected cells among lysed cell debris, infected cells with empty virions ("ghosts") still attached, and cells in the process of division with phage progeny visible inside of them; g, h soft agar overlays inoculated with the same number of B. intestinalis CFU in the absence and presence of the excess of phage, respectively; i phenotypic dissociation of colonies of phage-resistant clone 8, (8W, "white"phage-sensitive subclone; 8T, "transparent"phageresistant subclone); j kinetics of ΦcrAss001 adsorption to B. intestinalis phage resistant and sensitive clones; vertical axis, concentration of phage in the supernatant; horizontal axis, time passed after addition of phage; R, resistant clones (n=5); Rev, revertant sensitive sub-clones derived from one of the resistant clones; values are mean±SD from three independent experiments; k, l ultrathin section TEM visualization (x60,000) of altered cell surface morphology in phage-resistant clone 8T, compared to phage-sensitive WT strain; notable is the production of either loose or compact capsule; cm, cytoplasmic membrane; om, outer membrane host cells, already visible after the first 24 h, was associated with PVRs and consisted in the strong~100-fold repression of the PVR9 capsular polysaccharide (CPS) biosynthesis operon; the upregulation of similar CPS operons associated with loci PVR7, PVR8, and PVR11 (Fig. 3a); and the upregulation of a group of nutrient uptake genes in a complete inversion of the integrase/recombinase gene-flanked operon PVR3. Importantly, similar alterations of gene expression patterns relative to the phagenaïve WT culture of B. intestinalis can be observed in the resistant clone 8T (in the absence of phage). Its phagesensitive derivative 8W displays restored levels of PVR9 expression; downregulated (compared to WT) PVR7, PVR8, and PVR11; and upregulated expression of PVR3.
The CPS-encoding loci PVR7-11 have broadly differing gene content with limited levels of amino acid similarity between certain shared protein orthologs (e.g., oligosaccharide flippases, dehydrogenases, dehydratases, and glycosyltransferases; Fig. 3b). All of these loci share a putative invertible, moderately conserved promoter sequence of 183-186 bp ( Fig. 3c) flanked by a perfect, completely conserved 10 bp inverted repeat GTTCGT TTAA at which recombination most likely occurs. This sequence partially overlaps with inverted repeats ARAC GTTCGTN flanking phase variable promoters controlled by a serine recombinase family, master DNA invertase (mpi) in Bacteroides fragilis [17]. At least two homologs of this enzyme are encoded in the genome of B.
intestinalis APC919/174 (NCBI accession QDO70032.1 and QDO69715.1). A search for similar patterns in the B. intestinalis APC919/174 genome revealed the presence of an additional, fifth, CPS locus (denoted PVR12) with identical inverted repeat flanking its promoter region. This locus, however, did not show significant upregulation or downregulation of expression in our RNAseq assays. By mapping the RNAseq reads, we located the approximate transcription start in PVR7-11 inside the proximal copy of the inverted repeat (Fig. 3c), confirming that the invertible regions are capable of acting as promoters.
To explore the rate and frequency of phase variation on the complete genome-scale and to identify other potential PVRs, we performed local alignment of individual long DNA reads (obtained from WT culture, resistant clones 8T and PhR5, and corresponding revertant subclones 8W and PhR5-1) to the assembled circular genome reference, searching for cases where parts of the same long read align inconsistently, indicating either inversion or translocation. This analysis revealed a series of strong recombination hotspots present in every culture tested. The majority were associated with already known PVRs, with exception of a few novel finds, one of which is the previously mentioned fifth CPS locus PVR12 (Additional file 1: Fig. S2). We then used mapping of short shotgun reads to allow the quantitation of inversions in the long-term in vitro and in vivo persistence experiments, ten resistant clones, and one revertant sub-clone. By aligning paired-end short reads, we calculated the relative proportions of cells with two opposite orientations of the invertible promoters of PVR7-12 in different cultures of B. intestinalis APC919/ 174 (Fig. 3d, e). We found that the sensitive WT culture is characterized by PVR9 promoter mainly in the ON orientation, while PVR7, PVR8, PVR11, and PVR12 are mainly OFF. In agreement with the RNAseq data, the resistant clone 8T shows a switch of CPS expression from PVR9 towards PVR7, while the WT genotype is largely restored in sensitive revertant sub-clone 8W. Interestingly, other resistant clones showed the PVR9 promoter in mostly the ON orientation, combined with an ON state in one of the three other CPS-associated PVRs. Variable patterns of promoter activation with PVR9 being ON were detectable in the long term in vitro and in vivo persistence experiments (Fig. 3d, Additional file 1: Fig. S3).
We suggest that in the B. intestinalis-ΦcrAss001 phage-host pair some of the CPSs can be permissive to phage infection, and even serve as phage receptors (possibly PVR9), while some others can be protective (PVR7, PVR8, PVR11, and PVR12) or neutral. This is similar to what was observed for a large panel of non-crAss-like phages [27], as well as crAss-like-phages DAC15 and DAC17 in B. thetaiotaomicron [13]. We conclude that phase variation of CPS expression leading to phage resistance occurs spontaneously, even in the absence of phage. As a result, a dynamic equilibrium between sensitive and resistant sub-populations is maintained, which can be shifted towards the resistance phenotype by phage-driven selection. However, a constant switching back to sensitivity provides the phage with a constant supply of sensitive cells, thus assuring its long-term persistence.
Delayed release of ΦcrAss001 progeny leads to multiple echo phage bursts and potentially provides an infection escape mechanism We previously reported a remarkably low apparent burst size of only 2.5 pfu per cell infected with ΦcrAss001 [11]. This cannot be explained by a failure of phage to infect the majority of cells as only~2% cells in naïve B. intestinalis cultures were phage-resistant due to phase variation. At the same time, >50 particles per cell are visible in TEM images from ultrathin (80 nm) cell sections (Fig. 2b, d, f) and at least 20 new copies of phage genome are produced in an infected culture per copy of bacterial genome after 90 min of infection (as shown by metagenomic sequencing, Fig. 1c), ruling out the possibility of extremely low progeny counts per infected cell. Further to that, the efficiency of the center of infection tests (EOCI) showed that 56.6±12.7% of cells infected at an MOI=1 and 59.16±27.9% infected at an MOI=10 were capable of giving rise to infection centers (plaques) in B. intestinalis lawns. Electron microscopic observations of cultures infected at an MOI=1 reveal that >90% of cells show signs (discussed below) of the early stages of the virion assembly process 40 and 90 min after infection (Additional file 1: Fig. S4). Therefore, the observed low burst size cannot be explained by phase variationbased resistance or low progeny per cell counts and intestinalis APC919/174. a Host transcriptional response in a B. intestinalis APC919/174-ΦcrAss001 serial broth co-culture (24-120h in vitro persistence) experiment, analyzed using RNAseq (values are log 2 fold gene transcript expression change relative to the un-infected control from two independent experimental runs); clone-8T, phage-resistant derivative of the parental strain; clone-8W, spontaneous phage-sensitive revertant clone; only genes with significantly changed expression (p < 0.01 in DESeq2, n = 826) are shown; genes, associated with phase-variable genomic regions (PVR) are marked with color bars on top; b structure of phase variable operons, encoding five different CPS; protein sequence homologies (tBLASTx) between gene products are shown as colored parallelograms; two-sided arrows and ON/OFF labels mark positions of invertible promoters; c consensus sequence of the regions flanking invertible promoters (ON orientation) in CPS biosynthesis loci; approximate transcription start was identified though analysis of RNAseq data; stacked area charts show RNAseq read coverage of regions adjacent to the promoter in four CPS loci; d, e fraction of concordantly-aligned Illumina read pairs supporting orientation of invertible promoters in ON direction; WT, wild type strain; clones 8T, PhR1-9, phage-resistant derivatives; 8W, phage-sensitive revertant; P18d/P23d, samples in e were taken from the long-term in vitro persistence experiment (also see Additional file 1: Fig. S3 for similar data on the long term in vivo persistence) could have resulted from the action of an unknown mechanism limiting replication and/or release of fully assembled phage progeny from a significant fraction of infected cells.
In order to gain insights into this mechanism, we performed strand-specific RNAseq analysis of a single phage replication cycle in B. intestinalis cells infected at an MOI of 1. Three transcriptional modules can be observed in the ΦcrAss001 genome, which also correlate with ORF orientation and the putative operon organization of the genome (Fig. 4a). The expression of the early genes, located on the right end of the linear phage genome, is initiated between 0 and 10 min after phage infection. The predicted products of these genes include a transcriptional regulator, two conserved domain proteins with unknown function and a TROVE domain protein (a possible ribonucleoprotein component). A recent study in a related marine crAss-like phage Φ14:2 infecting C. baltica indicated that a similarly organized operon of early genes is transcribed by a giant virion-associated multi-subunit RNA polymerase [RNAP, [31]], ejected into the cell in the course of infection [14]. A homolog of this giant polymerase is present in all crAss-like phages and is also encoded by three large ORFs located in the central part of ΦcrAss001 genome [4,5,11]. Proteomic analysis of the crAss001 virion also confirmed the presence of the subunits of the giant polymerase in the assembled phage particle [11].
Transcription of the middle and late genes is initiated between 10-30 and 30-60 min, respectively, and is likely to depend on the action of the host RNAP [as suggested for Φ14:2 in [31]]. The middle genes are largely composed of functions associated with DNA replication and recombination, as well as nucleotide metabolism. Other genes of the same class can be involved in the degradation of host DNA and mRNA (HKD family nuclease, HicA mRNAse). The late genes, occupying the left-hand half of the genome, mainly encode structural proteins of the virion head and tail, as well as proteins participating in the virion assembly (terminase, chaperone) and cell lysis (holin, endolysin). Large ORFs coding for virionassociated RNAP subunits are located in the approximate center of the linear genome and are transcribed during the middle (gp49, predicted catalytic subunit, and gp50) or late (gp47) transcriptional stages. It is unclear why the temporal separation in RNAP subunit production might be needed. It may be that gp49 and gp50 are involved in transcription of the late genes in the same infection cycle in which they are produced, whereas gp47 is needed solely for the early gene transcription in the next cycle or plays a role in the packaging, ejection, or activation of the RNAP holoenzyme.
Interestingly, there was a significant overlap in the transcription of early, middle, and late genes: once activated, transcripts of each of the three modules persisted through the rest of the replication cycle (Fig. 4a,  b). The host transcriptional response was largely confined to strong genome-wide inhibition of transcription, occurring between 10 and 30 min preceded by the activation of certain energy metabolism functions at 10 min (most notably citrate synthase and NADP-dependent iso-citrate dehydrogenase, Fig. 4b, Additional file 2: Supplementary Dataset).
Morphological changes in the infected cells were visible as early as after 45 min as the formation of variablesized electron-light spots (Fig. 4c). Ninety minutes into the infection cycle, well after the onset of expression of structural genes, the first complete phage particles (electron-dense spots) begin to appear among the electronlight spots. Large clusters of viral particles are visible inside many of the infected cells at 120 min. Agreeing with the one-step growth curves (Fig.4d, e), cell lysis becomes noticeable at 150 min and widespread at 180 min. Interestingly, however, a fraction of cells remain intact at 150 min, and only show signs of the early stage of infection with no properly formed phages particles (Fig. 4f). This observation can be interpreted as de-synchronization of the phage replication cycle with some of the cells falling behind the others in terms of infection stage. Supporting this hypothesis is the continued expression of early and middle operon genes, overlapping in timing with late genes. This is further reflected in the one-step growth curves, where at an MOI of 1 or 10, a series of discrete, evenly spaced bursts are apparent with a low total progeny output (~3-6 pfu per capita).
Recently, bacteriophages similar to ΦcrAss001 in morphology and genome organization (Additional file 1: Fig. S5a-c) were isolated against B. thetaiotaomicron [13]. Phages DAC15 and DAC17 are also capable of long-term persistence in serial broth cultures of B. thetaiotaomicron VPI-5482 (Additional file 1: Fig. S5d). While the wild-type strain can produce eight different phase-variable CPS, a panel of engineered mutants has been previously developed that include an acapsular version as well as eight strains expressing only one of the eight different CPS's, referred to hereafter as CPS1-8 [21,32]. Phages DAC15 and DAC17 were reported to infect the cps3 + mutant efficiently, while only showing weak replication in WT VPI-5482 and other single-CPS expressing derivatives, indicating that CPS3 plays a permissive role in phage infection.
In our hands, both DAC15 and DAC17 were able to persist stably in serial daily broth cultures of cps3 + , as well as in WT (albeit at lower levels, Additional file 1: Fig. S5d). This is despite the fact that an inability to switch to the expression of alternative CPSs leads to a very low fraction of pre-existing resistant cells (< 10 -5 ) in populations of cps3 + , compared to 52.8±1.2-64.2±3.0% of cells resistant to DAC15/DAC17 in the WT strain. This indicates that the removal of CPS variability leads to greatly reduced resistance to phage but does not eliminate the ability of phage to persist long-term. In onestep growth experiments at an MOI=1, DAC15 and DAC17 behaved very similarly to ΦcrAss001, producing a series of bursts and a low apparent output of progeny (Additional file 1: Fig. S5e-f). However, unlike in B. intestinalis-ΦcrAss001 system, infection of cps3 + mutant at MOI=10 resulted in a single burst of~160 pfu per capita, a value which is in a much closer agreement with the progeny size observed microscopically (Additional file 1: Fig. S5g).
These data suggest that a mechanism of gradual lysis and release of phage progeny operates in Bacteroides cells infected with crAss-like phages ΦcrAss001, DAC15, and DAC17. This mechanism is largely independent of the dynamic phase variation mediated resistance discussed above. It appears that at high MOI, the mechanism might become overwhelmed or that the stress response in cells simultaneously infected by multiple phage particles switches it off. This leads to abrupt cell lysis and release of large phage progeny simultaneously, more consistent with reproduction of a virulent phage.

Discussion
Since their discovery in 2014, crAss-like phages have become one of the most intriguing elements of the human virome [3,6,10,11]. Of special interest is their unexplained, overwhelming prevalence in the gut microbial community, their ability for a long-term persistence and seemingly benign interaction with their hosts [10]. Observations of a long-term persistence of virulent bacteriophages in the mammalian gut are not unprecedented. Experiments in monoxenic mice revealed that administration of virulent phage fails to lower the levels of target E. coli bacteria and often leads to continuous shedding of high phage levels in feces [33,34]. Alternative mechanisms of non-lethal, persistent forms of infection were described in virulent phages, including the carrier state and unstable pseudolysogeny in Campylobacter jejuni, Salmonella typhimurium (phage P22), and Pseudomonas aeruginosa [35][36][37]. In C. jejuni, virulent bacteriophages CP8 and CP30A have been shown to maintain equilibrium with their hosts for many generations and to persist intracellularly in an episomal form without integration into the host chromosome. It has also been demonstrated that the carrier state can have a profound effect on bacterial host physiology, such as a major acquisition of self-targeting CRISPR spacers, decreased fitness, loss of colonization potential, and motility in C. jejuni [38]. More recently, a carrier state was described in P. aeruginosa infected with ssRNA levivirus LeviOr01 [39]. Infection with LeviOr01 caused dissociation of the sensitive host into two sub-populations: a small subset of phage-carrying superinfection immune cells, spontaneously and continuously releasing phage progeny, and a larger subset of uninfected non-carrier cells. Together, previous observations suggest that unconventional forms of phage infection span a wide taxonomic variety of bacteria and levels of complexity of phages (from Leviviridae containing only four genes in their genomes to Myoviridae with hundreds of genes).
In this study, we shed some light on the mechanisms underpinning the persistence of crAss-like phages using the ΦcrAss001-B. intestinalis pair as a model. We observed that, similarly to metagenomic findings from the human gut [10,22], a monoxenic mouse model was able to support long-term and stable persistence of ΦcrAss001. Bacterial clones recovered from colonized mice showed varied degrees of resistance to phage, but no obvious cost with regard to bacterial population density was associated with phage colonization. We propose that at least two separate mechanisms operating simultaneously are responsible for observed persistence in vitro: (i) dynamic and reversible acquisition of phage resistance in the host population, dependent on phase variation of CPS; (ii) delayed release or progeny from infected cells, resulting in a pseudolysogenic-like or carrier state phenotype (Fig. 5).
Sustained phage attack resulted in either switching off the PVR9 CPS locus, implicating this structure in phage adsorption, or increased expression of alternative CPSs (PVR7, PVR8, PVR11, and PVR12) with possible protective effects. We propose that the dynamic switching between different CPS types maintains an equilibrium of sensitive and resistant cells, allowing for restricted ("quarantined") phage replication and enabling longterm phage persistence and herd immunity in the bacterial population. While phage attack provides an obvious selective force driving switching to resistance states, the force driving a return to sensitivity remains unidentified. In complex microbiota conditions, expression of CPS protective against one phage could make cells sensitive to another phage present in the same environment [27], creating an equilibrium of CPS expression states driven by opposing phages. In addition to that, the expression of particular types of CPS can affect bacterial fitness in various ways, for example, by shielding the strain from host adaptive immune responses and therefore increasing its competitiveness in the complex gut microbiota [21]. In monoculture, however, such diverse selective forces are absent, making the metabolic cost of expression of a particular protective CPS (and availability of precursors [40]) an obvious reason for switching it off. The availability of a single-CPS mutant panel in B. thetaiotaomicron prompted us to check whether the persistence of DAC15 and DAC17 is still possible despite the inability of the strain to evade phage by swapping its CPS types. Indeed, not only was persistence in such a strain still possible, it actually occurred at higher levels than in the WT.
Another recent study has also implicated variation of multiple surface structures including CPS, S-layer lipoproteins, TonB-dependent nutrient receptors, and OmpA-like proteins, in dynamic switching of diverse phage resistance/sensitivity patterns in B.
thetaiotaomicron [27]. Furthermore, crAss-like phages DAC15 and DAC17 were shown to preferentially infect a phase variant of B. thetaiotaomicron expressing a specific type of CPS [13]. Modeling with phase variable phage receptor in Haemophilus influenzae demonstrated the role of phase variation in bacterial "herd immunity", a phenomenon in phage-host population dynamics where "protective quarantining" of one third of the host cells already confers significant protection against phage attack to the whole population [41]. We hypothesize that the second mechanism of phage persistence operates in parallel with herd immunity. This is based on the low apparent burst size (despite progeny counts of >50 per cell under the EM), multiple sequential smaller bursts and cells carrying phage progeny without lysing in a timely manner, the high expression of mainly early genes (regulatory proteins), and some of the middle genes (replication functions) in broth cultures of B. intestinalis in the presence of phage. Several explanations are possible: (a) carrier state, persistence in the form of fully formed or incompletely formed viral particles in the host cytoplasm; (b) unstable pseudolysogeny, persistence of phage genomes in an episomal form; and (c) inhibition of phage reproduction by cellular mechanisms and curing of infected cells. Interestingly, overwhelming the cells with high MOI was enough to overcome this mechanism of gradual release of progeny in B. thetaiotaomicron but not in B. intestinalis.
Future research will be needed to answer a number of key questions about this hypothetical mechanism: How does the carrier/pseudolysogeny state affect the viability and fitness of bacteria? Can bacteria continue with their normal lifecycle while in a carrier state? What are the signals that trigger entering and exiting of the carrier/ pseudolysogeny state? We suggest that the interplay between phase variation of phage receptors and the possible carrier state infection/pseudolysogeny enable highly synchronized replication of a crAss-like phage with its host at a population level and lead to an equilibrium of phage/host ratio maintained over multiple generations.
It has been proposed that the high availability of bacterial prey in aquatic environments and at mucosal surfaces selects for temperate bacteriophage lifestyles and lysogenic mode of replication. Such persistent infection is highly beneficial for the phage and is at least "affordable" (if not also beneficial) to the host [42,43]. This ecological model has become known as "piggyback-thewinner", as opposed to the "kill-the-winner" model [43]. The mammalian gut presents an example of one of the most densely populated microbial communities, with Bacteroides being one of its most abundant and temporally stable bacterial genera [10,44,45]. Our expectation was that the majority of bacteriophages in such an environment would be temperate, "piggybacking" on the ecological success of their bacterial hosts [46][47][48]. At the same time, our recent metagenomic analysis highlighted the prevalence of apparently virulent bacteriophages in the healthy core virome [10,23]. One could hypothesize that crAss-like phages, incapable of lysogeny and comprising a large part of these metagenomic sequences, employ alternative life cycle strategies, such as the one highlighted in this study in order to benefit from "piggyback-the-winner" dynamics.
We hypothesize that crAss-like phage-host systems encompass a range of behaviors, from one resembling virulent phages that produce plaques but whose replication in pure host cultures and complex microbial communities is typically supported at lower levels; to one similar to temperate phage that produce no plaques but persist in host populations at very high levels. It is notable that, regardless of the degree of virulence of a Fig. 5 Model of perpetuated replication of crAss-like phage in Bacteroides cultures. At least two separate mechanisms operating simultaneously are responsible for observed persistence in vitro: (i) dynamic and reversible acquisition of phage resistance in the host population, dependent on phase variation of CPS; (ii) delayed release or progeny from infected cells, resulting in a pseudolysogeny or a carrier state phenotype. Simultaneous action of the two mechanisms results in a steady supply of phages and phage-sensitive cells to maintain phage persistence at constant VMR, while allowing the majority of bacterial cells to proliferate unaffected by phage crAss-like phage, their stable and high-level persistence in vitro and in vivo has no obvious detrimental effect on the population densities or fitness of their bacterial host. This phenomenon is generally consistent with a temperate mode of replication, but we conclude from our analysis of ΦcrAss001, DAC15, and DAC17 that these viruses are non-lysogenic. Specifically, (i) no cases of related prophages have ever been observed in public databases of bacterial genomes, (ii) no clear lysogeny modules could be identified their genomes, and (iii) lysogens or pseudolysogens could not be obtained experimentally.
While the ecological benefits of carrier state infection are obvious for the phage, we can expect that the bacterial host party should also derive certain benefits from this interaction. In addition to exerting a constant selective pressure leading to host diversification through phase variation and point mutations and an improved overall fitness, the persistent phage can participate in lateral gene transfer and provide protection from incoming competitor strains and superinfection immunity against cognate phages.

Conclusions
The relationship between lytic bacteriophage and their bacterial hosts is often characterized as parasitic, but there are many instances, including what are shown here, of phage and host co-existing at high densities. The long-term benefits for the phage in maintaining a stable mixed population of resistant and sensitive hosts are obvious, but there may also be long-term advantages for the bacterial host. These may come in the form of phage acting as a driver of strain-level diversity in a complex environment, ensuring that the bacterial population does not simply adopt the most immediately advantageous genotype driving its short-term competitiveness, but maintains genotypic diversity, cycling between alternative surface structures and ensuring its competitiveness in a constantly changing environment like the gastrointestinal tract. In this scenario, the relationship between phage and bacterial host could be regarded as a successful partnership that ensures mutual persistence in complex environments.

Bacteriophage ΦcrAss001 and related viruses
The host strain B. intestinalis APC919/174 was propagated anaerobically on Fastidious Anaerobe Broth (FAB, Neogen, UK) and agar (FAA) at 37°C as described before [11]. To generate high titer phage lysates, early log phase cultures (OD 600 =0.2;~2x10 8 cfu ml -1 ) were infected with ΦcrAss001 at an MOI of 1. Cultures were left to grow overnight at 37°C in anaerobic jars, centrifuged for 15 min at 5200 g and 4°C, and filtered through 0.45 μm polyethersulfone (PES) syringe-mounted filter membranes. Plaque and spot assays throughout different experiments were performed after filtering 1 mL of phage/ bacteria co-cultures, performing serial dilutions and combining 100 μl of them with 300 μl of a fresh APC919/174 overnight culture in 0.4% bacto agar overlays. Incubations were performed anaerobically at 37°C. Bacteriophages DAC15 and DAC17 and their host strains B. thetaiotaomicron VPI-5482 (WT and cps3 + ) were propagated in the same conditions as B. intestinalis/ΦcrAss001.

In vitro experiments with crAss-like phages
Phage adsorption experiments (Fig. 2j) were conducted in aerobic conditions at room temperature on cells from an overnight Bacteroides culture (10 mL, optical density adjusted to OD 600 =0.2), triple washed with growth medium, and resuspended in the same media volume before the experiment. Phage was applied at an MOI=1, aliquots were removed, filtered as described above, and subjected to plaque assays 5 and 20 min after addition of phage.
The efficiency of bacterial plating (EOP) in the presence of phage (Fig. 2g, h) was conducted as follows: Serial tenfold dilutions of overnight Bacteroides cultures were inoculated into 3 mL of semi-solid FAA (0.4% agar) with or without the addition of 100 μL high-titer phage (~10 10 pfu mL -1 ) and poured onto 100 mm diameter FAA agar plates. The efficiency of lysogeny was determined as a percentage of colonies observed on phagecontaining overlays relative to the total counts on negative control overlays after 48 h of anaerobic incubation at 37°C.
In vitro phage persistence experiment (Figs. 1b, c, 3a, d, S5d) was performed as follows: 10 ml liquid cultures (n=3) of Bacteroides were infected at OD 600 =0.2 and MOI=1 and incubated overnight. The co-culture was then transferred daily, for 5 days in fresh FAB medium at 1:50 ratio.
One-step growth curves (Fig. 4d, e, S5e-g) were built as follows: early logarithmic phase (OD 600 =0.2) cultures of Bacteroides (20 mL) were infected at an MOI of 1 or 10 for 5 min at room temperature, followed by centrifugation at 5200 g, 4°C for 15 min, removal of supernatant, and re-suspending of the infected cells in fresh FAB medium. Incubation was continued anaerobically at 37°C for further 225 min with removal of 1 mL samples every 15 min. Samples were filtered through 0.45 μm pore PES filters and subjected to standard plaque assays with appropriate dilutions. Efficiency of center of infection (EOCI) was determined by doing plaque assays with infected cells at the same time intervals but without filtration step.

Transmission electron microscopy
Ultrathin section TEM of Bacteroides cells grown in broth cultures (Fig. 2k-l) or embedded in semi-solid agar (Figs. 2b-f, 4c, f, S4) were performed as follows. Pellets from 1 mL broth cultures or agar overlay fragments (~100μL volume) were immediately fixed and stored in 1 mL of 2% (v/v) glutaraldehyde, 1.5% (w/v) paraformaldehyde, and 75 mM Tris-HCl pH 7.5 solution. Samples were post-fixed with 1% Osmium tetroxide in 0.1 M Sorenson's phosphate buffer for 1 h at room temperature. A series of graded dehydrations were performed with increasing ethanol concentrations (30%, 50%, 70%, 100%; cells from broth cultures were pelleted by centrifugation between steps). Dehydrated samples were embedded in Epon resin. Briefly, dehydrated samples were transferred into 100% acetone and then 50:50 acetone:Epon resin for a minimum of 1 h. Samples were then put in 100% Epon resin at 37°C for 2 h. The resin was polymerized at 60°C for a minimum of 24 h. Embedded samples were sectioned at 80-nm thickness onto copper grids using a Leica UC6 ultramicrotome. Samples were stained with 2% uranyl acetate for 20 min and 3% lead citrate for 5 min. Grids were then imaged on a FEI Tecnai 120 at 120 kV accelerating voltage.

Shotgun sequencing and analysis of bacterial genomes
Shotgun sequencing on Illumina platform was performed for parental strain APC919/174 (Figs. 3b, e, S1, S2), resistant clones PhR1-9, 8T and spontaneous sensitive revertant derivative (8W) of the latter (Figs. 3e, S1, S2), as well as samples from in vitro phage persistence experiment and fecal samples from mice colonized with ΦcrAss001/B. intestinalis phage-host pair (Figs. 3d, S3). BioSample accession numbers are SAMN16802447-SAMN16802559 and SAMN16803214. Genomic DNA was extracted from bacterial cultures and phage-host cocultures using DNeasy Blood & Tissue Kit (Qiagen) in accordance with the manufacturer's Gram-negative bacteria protocol. Fecal DNA was extracted using QIAamp Fast DNA Stool Mini Kit (Qiagen). The concentration of DNA was determined using the Qubit dsDNA HS kit and the Qubit 3 fluorometer (Invitrogen/ThermoFisher Scientific). One hundred nanograms of purified DNA sample was sheared with M220 Focused-Ultrasonicator (Covaris) applying the 350 bp DNA fragment length settings (peak power 50 W, duty factor 20%, 200 cycles per burst, total duration of 65 s). Random shotgun libraries were prepared from genomic DNA using TruSeq Nano Library Preparation Kit (Illumina, cat #20015964) with dual-indexing following the standard manufacturer's protocol. Paired-end sequencing with 2×150 nt chemistry was performed on HiSeq 4000 platform at Eurofins Genomics.
For long-read Oxford Nanopore sequencing, genomic DNA from APC919/174, 8T, 8W, Phr5, and Phr5-1 (a sensitive derivative of the resistant clone Phr5) was extracted as follows: Cells from two mL of overnight cultures were collected by centrifugation at 5000 g for 10 min, washed by resuspending in 1 mL of TES (50 mM NaCl, 100 mM Tris-HCl, 70 mM disodium EDTA, pH 8.0), and re-centrifuged. After resuspending cell pellets in 0.4 mL of TES supplemented with 25% w/v sucrose and 30 mg mL -1 of lysozyme, 50 U of RNase I (Fisher Scientific) were added and samples were incubated at 42°C for 1 h. Cells were then lysed by addition of 20 μL of 10% (w/v) sarkosyl and 5 μL of 20 mg mL -1 proteinase K solutions with incubation at 37°C until complete clearing of the lysate (30-60 min). The obtained lysates were diluted by adding 350 μL of TES and 16 μL of 5M NaCl and gently extracted twice with an equal volume of 25: 24:1 phenol:chloroform:isoamyl alcohol mix with room temperature centrifugation at 8000 g for 5 min. This was followed by a similarly done extraction by an equal volume of chloroform. Finally, the aqueous phase was gently mixed with 2 volumes of 96% ethanol and DNA was pelleted by centrifugation at 8000 g, 4°C for 15 min. Pellets were rinsed with 1 mL of 70% ethanol, air-dried, and resuspended in 50 μL TE (10 mM Tris-HCl, 0.1 mM disodium EDTA, pH 8.0) overnight incubation at 4°C.
For detection of PVR promoter inversions and variant analysis in the in vitro co-culture and mouse colonization experiments (Figs. 3d-e, S3b) paired Illumina reads were mapped to the assembled APC919/174 as a reference. Reads were aligned using bowtie2 v2.3.4.1 in the end-to-end mode. Concordantly aligned read pairs were selected, sorted, indexed, and converted into count tables using samtools v1.7.
For detection of genomic recombination on a single read level, sufficiently long Oxford Nanopore reads (> 1000 nt) were aligned against APC919/174 reference genome using BLASTn v2.10.0 with e value cut-off of 1e −20. Individual alignments were retained only if the length was >200 nt with an identity of >90%. Next, reads were identified containing internally inverted alignments or blocks of the aligned sequence corresponding to coordinates in the reference genome shifted >200 nt relative to other aligned blocks inside the same read. Such blocks of the inconsistently aligned sequence were deemed as evidence of recombination events dynamically unfolding inside an individual clonal culture. Starting coordinates of these misaligned blocks were then plotted against the APC919/174 reference genome in a histogram with 1000 nt long bins and hotspots of recombination were identified (Additional file 1: Fig. S2).
One milliliter of culture aliquots was pelleted by centrifugation for 3 min at 17,000 g, anaerobically. Pellets were immediately lysed using 1 mL of TRIzol reagent (ThermoFisher Scientific). Total bacterial RNA was extracted using the standard manufacturer's protocol. Extracted RNA (20 μL) was treated with 2U TURBO DNase (ThermoFisher Scientific), further purified using RNeasy Mini Kit (Qiagen), and assayed for RNA integrity and quantity on Agilent Bioanalyzer using 6000 RNA Nano Kit.
Stranded RNA-seq libraries were prepared from 1 μg total RNA using ScriptSeq v2 Complete kit for bacteria (epicenter) and sequenced on Illumina NovaSeq 6000 platform at GENEWIZ. Read quality checks, filtration, and trimming were performed as described above in Shotgun sequencing and analysis of bacterial genomes.
Reads were then mapped to ΦcrAss001 and B. intestinalis APC919/174 reference genomes (NCBI accession numbers MH675552 and CP041379, respectively) using bowtie2 v2.3.4.1 in the end-to-end mode. Alignments were sorted by DNA strand and indexed using samtools v1.7. Samtools "mpileup" command was used to calculate read coverage per nucleotide position in both genomes. HTSeq framework v0.6.1.p1 was used to calculate aligned read counts per feature with default parameters. Read counts and coverage tables were then imported into R environment using data.table library v1.12.8 and compared between time points using DESeq2, using Wald test and parametric fit.

Quantification and statistical analysis
Bacterial colonization levels in vivo in the presence or absence of phage (Fig. 1a) were compared using Kruskal-Wallis test with FDR correction and were all found to be insignificantly different, except for days 31 and 38 (p < 0.05). Adsorption of crAss001 to sensitive and resistant bacterial variants (Fig. 1a) was compared using Kruskal-Wallis test with Bonferroni correction and was found significantly different (p < 0.05). Analysis of differentially abundant transcripts (Figs. 3a, 4b) was performed using DESeq2. Only genes showing significant change (p < 0.001 for host genes in in vitro persistence experiment, Fig. 3a, and p < 0.05 for both phage and host genes in one-step growth experiment, Fig. 4b) between any of the consecutive time points are displayed.
Additional file 1: Figure S1. Phase variation associated with major rearrangements in B. intestinalis APC919/174 genome. Figure S2. Discovery of additional phase variable regions n B. intestinalis APC919/174 genome using Oxford Nanopore MinION long read sequencing platform. Figure S3. Phase variation in Bacteroides intestinalis APC919/174 CPS operon expression in mouse colonisation experiment in the presence or absence of phage crAss001. Figure S4. TEM of B. intestinalis cultures infected with crAss001 at an MOI=1. Figure S5. Morphology, genome structure and biological properties of B. thetatiotaomicron phages DAC15 and DAC17 in comparison with ΦcrAss001.
Additional file 2:. Supplementary Dataset. Spreadsheets of numerical data and raw unedited photographs required for reproduction of Figs.1, 2, 3 and 4