Knockout of a labeled strain of Escherichia coli in the mouse gut using native phages

Background: There is a lack of methodological investigation of the in situ functions of a bacterial species in microecosystems such as the animal gut, although the microbiome has become a focus in the microbiology eld worldwide. Results: We used native mixed phages containing Escherichia phages T1 and T4 as a microbial editing tool for eliminating Escherichia coli MG1655 labeled with green uorescent protein in the mouse gut. The phages possessed rigorous host specicity at both the genus and species levels, resulting in an 8.8-log10 decrease in the titer of viable bacteria after 12 h of phage treatment in an in vitro test. In vivo, they knocked out strain MG1655 not only at concentrations of 10 6 -10 8 CFU g -1 colonizing the mouse gut but also even in mice fed with feedstuff containing the bacterium. In addition, the impact of phage treatment on the microbial community structure of the mouse gut was not signicantly ( p (cid:0) 0.05) based on a 16S rRNA amplicon gene sequencing analysis, although the richness of some bacteria changed signicantly. Conclusions: We provide a feasible microbial editing technique for the animal gut. Native phages with strict host specicity can effectively knock out a target bacterium by single or continuous gastric perfusion, with limited perturbation of microbial diversity, which is benecial for studies of the function of a specic bacterial species colonizing a complicated microecosystem.


Background
The microbiome, de ned as the totality of microorganisms and their collective genetic materials in a wellrecognized ecosystem, has become a current focus of scientists and governments. It contains information including the composition and structure of the microbiome, genetic and physiological functions, and even interactions between microbes and their hosts or environments. A series of national plans/initiatives have been launched to support research on the microbiome, with the aim to expand understanding about nature and humans themselves [1][2][3]. It is believed that the outcome of microbiome research will provide new resources and innovative technology for social and economic developments, as well as improve the quality of life of human beings. The human microbiome, including the gastrointestinal microbiome, respiratory microbiome, reproductive tract microbiome, oral microbiome and skin microbiome, is closely related to human health, as an integral part of human body [4]. The gut ora has particularly important roles in the nutritional, immunological, and physiologic processes of the host.
The human gut harbors at least 100 trillion microbial cells, and the quantity is 10 times that of the human body's own cells. Approximately 1,000 to 1,150 species of bacteria colonize the human intestine, with an average of approximately 160 dominant species in each body [5][6][7]. The gut ora is correlated with intestinal in ammatory bowel disease, Crohn's disease, diabetes, asthma, liver disease, obesity, mental disease and cancer, making it an important metabolic organ [8][9][10][11][12].
There is currently a lack of knowledge of how to methodologically study the in situ function of a speci c species in a complicated ecosystem. Generally, knowledge about microorganisms obtained by pure cultivation cannot accurately represent the role they play in the ecosystem. Current sequencing-based analyses of the microbiota, including DNA chips and metagenomic and 16S rRNA gene sequencing, only permit evaluation of the correlation between a species, genus or core microbiota and the phenotype of the host by determining the variation in microbial abundance and diversity, and even further comparative analyses of relative microbiome data cannot provide information about the extent or directionality of changes in abundance or metabolic potential due to nonquantitative pro ling [13][14][15]. Thus, these relative approaches are limited in revealing the interplay between the speci c microorganism or microbiota constituent and host phenotypes or health [16,17]. In addition, various gnotobiotic and germ-free (GF) animals, including zebra sh and mice, are also used to study the function of the gut ora [18,19].
However, these animals can only be used to investigate the function of the speci c microorganism or microbiota under a simple background setting, and how a species works in a complicated community cannot be uncovered in this manner. This methodological de ciency is a major bottleneck for studies of the function of a speci c microorganism in the microbiota, although a few novel techniques are emerging, including isotope probes or quantitative microbiome pro ling [17,20]. Theoretically, similar to gene function revealed using gene knockout and complementation techniques, the function of a speci c microorganism in a microbiota should be illuminated by microbial knockout and complementation. In fact, the identi cation of pathogenic bacteria in infectious diseases by Koch's law follows the principle of microbial complementation since it is technically easy and operable for a culturable microorganism.
Conversely, strategies to knock out a microbial species in a complicated microsystem and ensure that its neighbors are not mistakenly removed due to off-target effects is a substantial technical challenge.
Phages, the viruses that infect bacteria, exist widely in the environment. As antibacterial agents, they have been applied in the prevention and control of bacterial infection or contamination in animals, crops and foods [21][22][23]. In the present work, we used isolated phages with high host speci city to knock out Escherichia coli MG1655 in the mouse gut, and the feasibility of phage as a bacterial knockout tool for the gut microbiota was evaluated. This study will contribute to the development of a simple and e cient technique for investigations of the function of a bacterial species in ecosystems.

Results
Morphology and sequencing analysis of phages Two phages, W1 and W3, were isolated from mouse fecal samples and selected for further knockout testing due to their superior lysis ability against strain MG1655. Their plaques are shown in Additional le 1: Fig. S1a. Phage W1 had an icosahedral head (65±2 nm in diameter) and a soft, but exible curved tail with a length of approximately 100±20 nm under TEM (Additional le 1: Fig. S1b). Compared with phage W1, phage W3 was larger, with a 100±10-nm-wide icosahedral head and 150±10-nm-long tail composed of the tail tube, base plate, short spines and long tail bers (Additional le 1: Fig. S1c). Phage W1 contained a double-stranded DNA, and its complete genome consisted of 48,737 nucleotides with an average GC content of 46.13%. Gene annotation analysis showed that the phage contained 72 genes, and 8 genes encoded structural proteins of the capsid and tail (Fig. 1a). The genome of phage W1 showed the highest similarity to Escherichia virus T1 (GenBank accession numbers of MK213796 and AY216660), with 100% identity on the basis of the 100% query coverage [24], indicating that it is an Escherichia phage T1 (Fig. 1b). The double-stranded DNAgenome of phage W3 is 168,040 bp with an average GC content of 35.54%, and it encodes 173 proteins, including structural proteins of the capsid and tail, as well as cytolysis, DNA repair, transcription and nucleic acid metabolism enzymes (Fig. 1c). The complete genome of phage W3 showed the highest similarity to Escherichia virus T4 (GenBank accession numbers of KJ477685, KJ477684 and KJ477686), indicating that the phage is an Escherichia phage T4 (Fig. 1d).
Host range of phages Forty-two strains belonging to 21 genera were collected (Additional le 2: Table S1) for host-speci c analysis of phages W1 and W3. No plaques appeared on LB plates, which indicated that neither phage infected these strains a liated with twenty genera of Enterobacteriaceae (Fig. 2). Thus, the phages possessed high host speci city at the genus level. At the species level of Escherichia, phage W1 was not infectious to all six strains tested, while phage W3 could lyse E. fergusonii, an emerging species within the genus Escherichia proposed by Farmer et al [25], showing a wider host range than phage W1. In addition, phage W3 lysed all subspecies of E. coli tested. Interestingly, phage W1 was not suited for infecting strains DH5α, Scarabxpress and zzy7 isolated. Therefore, the phages could distinguish the different genera under Enterobacteriaceae and species under Escherichia (except E. fergusonii), although they showed little host speci city at the subspecies level.

Transformation of pGFPuv and stability analysis
A strain of E. coli MG1655 tagged by green uorescence protein (GFP) was prepared to differentiate the tested strain with the aboriginal E. coli strains in mouse gut. Plasmid extracted from the obtained transformants (T GFP ) was digested with the Mlu I enzyme, and electrophoresis showed that pGFPuv was successfully introduced into strain MG1655 (Additional le 3: Fig. S2). Compared with the untagged strain, T GFP with green uorescence is easily observed under uorescence microscopy (Figs. 3a and 3b).
Moreover, it also exhibits blue and green uorescence on LB plates when exposed to UV light (Figs. 3c and 3d), facilitating enumeration due to the green uorescent phenotype. Theoretically, plasmid loss for T GFP occurs since the plasmid is not integrated into the bacterial genome [26], and the enumeration of T GFP requires plasmid stability. Following passages 1 to 11 in the absence of ampicillin, less than 0.05% of plasmid loss was obtained, indicating that the plasmid was relatively stable even after 11 days of propagation (Fig. 3e). Thus, the low plasmid loss guarantees the reliability of bacterial enumeration based on uorescence labeling.

Infection test in vitro
Infection of T GFP with phages W1 and W3 at various multiplicities of infection (MOIs) was monitored for 48 hours, and e ciencies of the different phages in reducing the titer of viable T GFP varied. In the phagefree control, there was a rapid increase in the T GFP concentration at hour 12. Compared with the control, infection of phage W1 resulted in a 0.2-to 3.5 log 10 decrease at hour 24 at MOIs ranging from 10 -3 to Page 5/24 reduction of viable bacteria was observed at both hour 12 and hour 24 with an MOI of 10 3 PFU/CFU (Figs. 3f and g). The maximum decline in bacterial quantity reached 10 6 -fold less than that in the control at hour 12. However, it should be noted that neither phage was able to eliminate all bacteria, although W3 was more e cient than W1 in reducing viable bacteria.
During the infection of T GFP with mixed phages (MP, W1 and W3 at a ratio of 1:1) at an MOI of 10 3 PFU/CFU, an 8.8-order of magnitude decrease in the titer of viable bacteria was obtained at hour 12 compared with that in the phage-free control, and the culture suspension in the test tube was clear. MP almost eliminated T GFP at this point, although the culture nally recovered to a 10 8 CFU ml -1 bacterial density at hour 48 [27,28].

Knockout test in vivo
The sampling sites of the mouse gut in vivo test are shown in Fig. 4a. A 3-day low-concentration gastric perfusion (LCGP) with 10 6 CFU ml -1 T GFP suspension led to bacterial colonization in mouse intestine. The T GFP quantity varied greatly in different intestinal sections. The caecum and colon are major sites of T GFP colonization, and they harbor 9.0×10 6 CFU cm -1 and 1.4×10 7 CFU cm -1 T GFP , respectively (Fig. 4b). A highconcentration gastric perfusion (HCGP) with 10 11 CFU ml -1 T GFP suspension resulted in the appearance of more T GFP in the mouse gut, and after 3 days of treatment, the bacteria in the caecum and colon reached up to 0.9×10 8 CFU cm -1 and 1.38×10 8 CFU cm -1 , respectively (Fig. 4c). In both the LCGP and HCGP, T GFP quantities in feces showed no signi cant difference (p values 0.05) with those in the caecum and colon (Fig. 4d), and these three sites were suitable to evaluate the knockout effect of T GFP in the mouse gut in the following studies considering that they harbored more T GFP than the other sites.
Subsequently, the results of the microbial knockout test showed that bacteria in the mouse gut were almost eliminated by MP in LCGP groups. After gastric perfusion of MP, the T GFP quantity in feces was 6.16×10 2 CFU g -1 at hour 24 ( Fig. 4e), and at hour 48, T GFP was undetectable in the caecum, colon and feces (Figs. 4f-4h). However, in the control gavaged with inactivated mixed phages (IMP), bacteria in the caecum, colon, and feces remained at 6 orders of magnitude in CFU cm -1 or CFU g -1 . In addition, T4 showed better removal than T1, although neither of them could eliminate T GFP . Therefore, only MP was selected for T GFP knockout in subsequent experiments.
In the HCGP groups, after the rst phage infusion (PI), T GFP in feces still reached up to 0.77×10 4 CFU g-1 at hour 24. Thus, the second PI was conducted at hour 48. At hour 72, the bacteria in feces decreased to 0.35×10 2 CFU g -1 , and until hour 120, T GFP in the feces did not recover (Fig. 4i). Simultaneously, T GFP quantities in the caecum and colon were 0.40×10 2 CFU cm -1 and 0.20×10 2 CFU cm -1 , respectively (Fig. 4j), and the remnants were almost undetectable by the plate colony-counting method. Thus, T GFP in the HCGP groups also decreased to a 10 2 CFU cm -1 level after two gastric perfusions.
We investigated the knockout effect of MP when the mouse gut was constantly supplemented with T GFP by feeding feedstuff containing T GFP . In the control, the T GFP quantity in feces was stable at the level of 10 9 -10 10 CFU g -1 (Fig. 4k), showing that feeding increased the T GFP density in mouse feces. After MP treatment at an interval of 48 hours, the T GFP quantity decreased by 4.5 to 6 orders of magnitudes compared with the initial content, indicating that MP could knock down the bacteria but could not knock out them when they were supplemented every day. Moreover, T GFP enumeration in the caecum (6.30×10 4 CFU cm -1 ) and colon (3.17×10 3 CFU cm -1 ) also proved that T GFP was not eliminated (Fig. 4l).

Discussion
De nition of microbial knockout Microbial knockout is a novel issue in the microbiome eld and is signi cantly different from gene knockout, which is used extensively for investigation of gene function. Generally, gene knockout is performed to remove one or several speci c gene(s), while microbial knockout means that millions of microbial individuals are eliminated. Considering the insu cient contact between phages and host bacteria in the mouse gut, the resistance against phages developed by bacteria and other causes [29], it is very di cult to achieve complete elimination, especially regarding the high density of bacteria. Thus, we should consider how to de ne microbial knockout within a broader range. In the knockout experiments in both the LCGP and HCGP group, we obtained a microbial knockout effect considering that the T GFP population in feces was 10 2 CFU g -1 , which is almost undetectable by the plate colony-counting method due to the inevitable dilution of solid samples. With constant T GFP supplementation in the mouse gut, T GFP could not be eliminated, although marked decreases of 4.5 to 6 orders of magnitude of viable bacteria were obtained. Considering that there were still thousands of surviving bacteria (10 4 CFU g -1 ) in feces, the reduction in the bacteria titer was clearly not a knockout but a knockdown. Normally growing animals live in a poor-hygiene environment, and their intestines are open systems. They always have an opportunity to obtain bacteria from the environment, including the air, food and water, even if the bacteria have already been knocked out in the gut, which probably leads to the failure of microbial knockout due to bacterial population recovery. Therefore, it is much more di cult to completely knock out a speci c microorganism in an open environment, and prevention of the continuous input of exogenous microorganisms is a key issue during microbial knockout.

Off-target analysis
The possibility of off-target effects should be evaluated during bacterial knockout by phages. Host speci city analysis of phages showed that although the two phages possessed strict host speci city, phage W3 might mistakenly knock out E. fergusonii at the species level. E. fergusonii, formerly known as enteric group 10, is an infrequently occurring but emerging animal and human pathogen [30], and it can be differentiated from E. coli by sorbitol and lactose fermentation negativity but adonitol, amygdalin and cellobiose fermentation positivity. E. coli and E. fergusonii are the closest relatives of one each other under the genus Escherichia, and E. fergusonii isolates possess genotypic and phenotypic features found in known pathotypes of E. coli, which leads to the following argument: is E. fergusonii in fact another E. coli [31]? This question remains unanswered, and thus we retested the bacterial classi cation based on the 16S rRNA gene sequence. A 1421-bp gene sequence of E. fergusonii was determined, and the sequence comparison showed that the strain (GenBank accession number MK168572.1) shared the highest similarity (100%) with E. coli (CP020516.1). According to the current opinion of molecular identi cation based on the 16S rRNA gene, this strain is an E. coli in nature by phylogenetic analysis (Additional le 4: Fig. S3). Off-target effects did not occur at the species level if the taxonomic deviation was considered. However, this does not mean that off-target effects will never occur during future microbial knockouts because we only tested the reported strains under the Escherichia genus.

Impact of MP on the bacterial community
The design of experiments focused on the impact of MP on the bacterial community is shown in Fig. 5a. After 16S rRNA gene sequencing, trimmed sequences with an average sequence length of 398 bp were obtained, and statistical analysis of fecal samples, including the identi ed operational taxonomic units (OTUs) and alpha diversity indexes, are shown in Additional le 5: Table S2. Good's coverage indicated that 99.76-99.81% of the species in the samples were recovered at a cutoff of 97% sequence similarity.
The Chao1 indexes of the PBS buffer (PBS), IMP, and MP groups were 399.2, 407.6, and 340.2, showing that the bacterial richness was signi cantly different after the phage treatments (Fig. 5b). The Shannon indexes of the three groups showed that phage treatment had no signi cant impact on bacterial diversity in the mouse gut (Fig. 5c). Dissimilarity comparison tests between the different groups at the genus level (Fig. 5d) also indicated that the addition of MP did not change the main microbial community structure (p values 0.05).
We analyzed the RA changes of bacteria in different treatments at the genus level because of the limited accuracy of molecular identi cation based on 16S rRNA gene sequencing at the species level. There were 42 and 43 genera with signi cant RA differences in the MP and IMP group (Fig. 5e) compared with the PBS group, indicating that the disturbance, de ned as the quantity of bacterial genus with a signi cant RA change, caused by MP was not signi cantly different from that in the IMP treatment (p 0.05). In addition, RAs of 34 genera changed signi cantly between the MP and IMP group. Co-occurrence network analysis showed that the three groups shared the great majority of the species. Twelve genera including Faecalibaculum, Butyricimonas and Psychrobacter appeared in both the PBS and IMP group but lost in the MP group (Fig. 5f). According to the previous phage host analysis, the disappearance of these bacteria was obviously not caused by off-target effect of phages. Therefore, treatment by live phages resulted in remodeling of the bacterial community in the mouse gut. However, whether in the IMP or MP group, the core top 30 genera (total RA 98.5%) did not disappear (Additional le 6: Fig. S4), and a similar conclusion could be drawn based on the OTU analysis. Among 483 OTUs in total, 58 OTUs present in the controls (MP and PBS groups) were absent in the MP group (Fig. 5g). However, the core 70 OTUs accounting for the total RA of 93.9% were not lost, although the RAs of 30 OTUs showed a signi cant difference (Fig. 5h). Therefore, it is feasible to knock out a core member of the gut ora with native phages. However, the disappearance of noncore microorganisms or OTUs showed the reconstruction of the bacterial community structure. It is necessary to evaluate their interference during veri cation of the function of T GFP knockout, and thus, further microbial supplementation experiments should be considered.

Microbial interactions upon knockout
RAs of the Escherichia genus in the MP group were signi cantly lower than in the PBS group (Fig. 5d), and enumeration results on day 8 showed that E. coli was reduced by approximately 2 orders of magnitudes (Fig. 6a) in the MP group. However, neither Escherichia nor E. coli was knocked out completely because in addition to E. coli, the genus might contain species that are not sensitive to phages T1 and T4, and the MP we used targeted strain MG1655 rather than the entirety of strains of E. coli in the mouse gut. We isolated 1000 strains of E. coli from the fecal samples, and 161 strains were insensitive to phage W1 or W3. The Spearman correlation network analysis revealed that Escherichia was negatively correlated with 5 genera and positively correlated with 12 genera (r values≥0.6), indicating that it might affect the RA change of other genera through various cascade reactions (Fig. 6b). Moreover, it should be noted that knockout of a speci c bacterium should not be attributed entirely to the phages used. In the mouse gut, T GFP faces complicated ecological competition from various microorganisms.
During coculture of E. coli and two common gut bacteria, Proteus vulgaris and Sphingobacterium lactis in visual biomimetic reactors (Fig. 6c), the bacterial presence signi cantly reduced the E. coli population in both the E+P (E. coli +P. vulgaris) and E+S (E. coli + S. lactis) groups (Fig. 6d). The addition of MP resulted in the clearance of T GFP during bacterial coculture, while this phenomenon was not observed in the control group (E+MP) (Fig. 6e). This nding indicates that the interspecies competition improved phage control of the bacterial population [32]. Therefore, in addition to phages, bacterial interactions in the mouse gut contribute to the knockout of T GFP .

Conclusion
Overall, the results presented herein demonstrate that it is feasible and effective to knock out a speci c bacterium by its corresponding phages. The phages with rigorous host speci city not only knocked out E. coli MG1655 at a concentration of 10 6 -10 8 CFU/g in the mouse gut but also knocked down the strain even in the mice fed with feedstuff containing strain MG1655. The changes in both the metabonomic and immune system caused by phage treatment were not evaluated in this work considering that the phages isolated from mouse feces are endogenous and the impact of MP on the bacterial community was not signi cant. However, these factors should be considered when assessing the function of a bacterial species in its host in future work.

Microorganisms and chemicals
All chemicals used were spectral or analytical grade. Forty-two bacteria including E. coli subspecies, Escherichia species, and strains belonging to twenty different genera of Enterobacteriaceae were purchased from the China Center for Type Culture Collection (CCTCC), China General Microbiological Culture Collection Center (CGMCC), or China Center of Industrial Culture Collection (CICC) or provided by the Key Laboratory of Microbial Resources and Functional Molecules of Henan Province, China. The detailed information about these bacteria is listed in Additional les 2. Table S1.

Phage isolation and identi cation
Phages W1 and W3 were isolated from mouse feces collected from the Animal Breeding Base in Henan Normal University, China. Isolation, propagation, and titration measurement of phages were conducted as described by Adams [33]. After the staining of phage particles by 2.0% aqueous uranyl acetate (pH 4.5-5.5) on a carbon-coated grid, phages were observed using transmission electron microscopy (TEM, JEM-1400, JEOL Ltd., Japan) at an accelerating voltage of 80 kV. Phage DNA was extracted using a phage DNA extraction kit (Aidlab Biotech, China) [34]. DNA sequencing was conducted using the Illumina HiSeq (PE250) platform at Hangzhou Lianchuan Biological Information Co., Ltd., China. The paired-end reads were assembled using SOAP denovo v2.04 (http://soap.genomics.org.cn/), and the potential open reading frames (ORFs) were predicted using GeneMarkS 4.6b (http://topaz.gatech.edu/GeneMark/). Possible tRNAs in the genome were determined using tRNAscan-SE (http://lowelab.ucsc.edu//tRNAscan-SE/). Comparisons of nucleic acid and predicted protein sequences with other known sequences were performed by BLAST analysis (http://blast.ncbi.nlm.nih.gov/Blast.cgi). Maps of a circular representation of phage genomes were generated using DNAMAN (version 6.0; Lynnon Biosoft). Neighbor-joining trees were drawn using MEGA 5.05 [35]. The complete genome sequences of phages W1 and W3 were deposited in GenBank under accession numbers PRJNA494624 and PRJNA494627, respectively.
Host range analysis of phages Strains were cultivated in LB liquid medium (LB, 10 g l -1 tryptone, 5.0 g l -1 yeast extract, 10 g l -1 NaCl) for 24 hours, and then bacterial lawns were prepared by pouring 3 ml of LB soft agar (LB SA , LB liquid medium supplemented with 10 g l -1 agar) containing 0.1 ml of broth culture onto LB agar plates (LB A , 10 g l -1 tryptone, 5.0 g l -1 yeast extract, 10 g l -1 NaCl and 15 g l -1 agar). After solidi cation of the soft agar, the plates were spotted with a phage suspension (10 μl, 10 5 -10 6 PFU μl -1 ) on triplicate plates. The plates were dried for 15 minutes at room temperature (25±2°C) before incubation. After 18-22 hours of incubation at 37°C, the effects of phages on bacterial lawns were observed and recorded. LB broth culture without phage was used as a negative control, and a positive response was de ned as a number of ≥20 plaques or full lysis on the plates.

coli MG1655 labeled with GFP
The plasmid pGFPuv, including a pUC origin of replication (ColE1-like), TEM β-lactamase as a selectable marker, and a cycle GFP reporter gene under control of the Plac promoter (Clontech, CA, USA), was used to obtain a GFP-tagged transformant of E. coli MG1655. The plasmid was transformed into strain MG1655 by the heat-shock method [36]. Brie y, 100 μl of chemically competent cells and 100 ng of plasmid DNA were mixed and incubated on ice for 30 min; then, they were heat-shocked in a water bath at 42°C for 90 seconds, allowed to recover for 1 hour, and plated in the presence of 100 µg ml -1 ampicillin. T GFP was veri ed by PCR ampli cation after digestion with the Mlu I enzyme (Takara Biotechnology Dalian Co., Ltd.) and uorescence microscopy observation.
To evaluate the stability of the transformant in the lab, T GFP was picked from plates and incubated in a 250-ml ask containing 50 ml of nonselective LB medium at 37°C and 120 rpm. After 24 hours of cultivation, the obtained cultures were set as passage 1 (P1). Subsequently, 2 ml of culture broth from P1 was transferred to 50 ml of fresh LB medium every 24 hours, and this procedure was repeated ten times [37]. To determine plasmid loss, 2-ml cell cultures were collected at different passages, serially diluted, and plated on LB A . Colonies with green uorescence under UV light were counted, and plasmid loss was expressed as the percentage of viable cells with abolished GFP uorescence to the total cells.

Preparation of phage stocks
Each phage suspension prepared using sterile salt-magnesium (SM) buffer was plated onto LB A along with LB SA containing 10 7 CFU ml -1 T GFP overlay. After an overnight incubation at 37°C, SM buffer was added to the plates, and the top soft agar slurry was harvested and centrifuged twice at 12,000 ×g for 20 minutes to collect phage-rich supernatant (lysate). The lysate was added with an equal volume of chloroform and then ltered using a 0.22-μm lter to remove cells and debris. After titer measurement, the ltrate was serially diluted to obtain 10 3 -10 6 PFU ml -1 phage stocks.
High-titer phage stocks were prepared from the lysates by liquid infection. For each phage, the low-titer lysate (1.0 ml; 10 6 PFU ml -1 ) mixed with T GFP (1.0 ml; 10 7 CFU ml -1 ) was added to 200 ml of LB and incubated for 24 hours at 37°C with aeration. The cultures were each treated with an equal volume of chloroform, and the lysates were harvested twice by centrifugation (12,000 ×g, 20 minutes) as well as by ltration with a 0.22-μm lter. After titer measurement, the ltrate was serially diluted to obtain 10 7 -10 11 PFU ml -1 phage stocks.

Knockout test in vitro
T GFP was cultivated in LB for 24 hours, and 100 μl of cultures with a concentration of 4.0×10 7 CFU ml -1 were transferred to a test tube containing 5 ml of LB. Then, phage suspensions (100 μl) of W1, W3, and MP with different titers (10 4 -10 10 PFU ml -1 ) were seeded in different tubes. The tubes were incubated at 37°C, and T GFP was enumerated every 12 hours by the plate colony-counting method. The reproducibility of the tests was confirmed in three independent continuous cultures. A phage-free culture containing only bacteria was used as a control to demonstrate the absence of contamination.

Knockout tests in vivo
Ten-week-old male Kunming species mice with an average body weight of 21 ± 2 g (Laboratory Animal Center of Henan Province, China) were used for knockout tests in vivo. All the mice were maintained in a 12-hour light/dark cycle, supplied with water and a standard diet (65% carbohydrate, 11% fat and 24% protein), and housed at 20-25°C. The mouse experiments were not performed in a blinded manner, and the experimental groups were randomly allocated.
We rst investigated the colonization of T GFP in the mouse gut. After 3 days of gastric perfusion with 5.0 g l -1 ampicillin [38], 120 mice were divided into 2 groups, the LCGP and HCGP group, and gavaged with 200 μl of different concentrations of the T GFP suspension (1.4×10 6 CFU ml -1 and 4.3×10 11 CFU ml -1 ) once per day for six consecutive days. After the treatments, T GFP in mouse feces was enumerated daily. On day 12 and 15, ve mice from the LCGP and HCGP groups were dissected, and T GFP in different sections of the gut was collected and enumerated to validate T GFP colonization.
Forty-eight mice in the LCGP group were divided into four groups (n=12) and were gavaged with MP, phage T1, phage T4, or IMP. As a control, IMP was prepared by sterilization of MP at 121°C and 0.1-0.15 MPa of steam pressure for 30 min in an autoclave (Hiclave HVE-50, Hirayama, Japan). After gastric perfusion, T GFP in mouse feces was enumerated every 24 hours.
Twenty mice in the HCGP group were divided into two groups (n=10) to quantitatively evaluate the knockout e ciency of MP against the high titers of T GFP. The two groups were gavaged with MP or IMP.
After gastric perfusion, T GFP in mouse feces was enumerated every 24 hours, and T GFP in the caecum and colon was evaluated at hour 120.
Thirty mice in the HCGP group were used to quantitatively evaluate the knockout effect of MP when T GFP was constantly supplemented in the mouse gut. Mice used in this test received feedstuff with T GFP (2.0×10 9 CFU g -1 ) daily. They were divided into two groups (n=10) and gavaged with MP (10 11 PFU ml -1 ) or IMP. Moreover, ten mice without any treatment were used as a control to demonstrate the absence of T GFP contamination between different groups, i.e., no naturally green-uorescent bacteria in the mouse gut.
Phylogenetic analysis of E. fergusonii Genomic DNA of E. fergusonii (Access number CICC24137) was extracted, and the 16S rRNA gene sequence was ampli ed using PCR with the primer pair 27F and 1492R [39]. The puri ed PCR product was cloned into the vector pMD19-T and sequenced. Sequences of related taxa were obtained from the GenBank and EzTaxon-e databases [40]. Phylogenetic analysis was performed using MEGA software version 5.05 after multiple alignment of data by DNAMAN. Evolutionary distances and clustering were constructed by the neighbor-joining method.
Experimental design and 16S rRNA gene sequencing Forty-ve Kunming species mice were used to test the impact of phage treatment on the microbial community structure. The detailed experimental design is described in Fig. 6a. Mice in the MP group were orally gavaged with MP once per day for 3 days, and as controls, the PBS and IMP groups were gavaged with PBS buffer and inactivated MP, respectively. During the tests, changes in movement, appetite, and dejecta were recorded. Fecal pellets of mice (20 mg per mouse) from the three groups were collected each day. Samples were shaken su ciently for 30 min in a 50-ml sterile centrifuge tube. Subsequently, both sterile gauze and 5-μm lter membranes were used to remove large particles, and the lter liquor was used for both enumeration of E. coli using CHROMagar™ E. coli chromogenic medium (CHROMagar, France) [41] and further genomic DNA extraction.
Bacterial genomic DNA in fecal samples was extracted using an Omega Bio-Tek, OMEGA-soil DNA Kit (Qiagen, Germany). After veri cation of the purity and concentration, PCR ampli cations of the highly variable V3-V4 regions of the bacterial 16S rRNA gene were conducted based on the universal primer pair 338F (ACTCCTACGGGAGGCAGCAG) and 806R (GGACTACHVGGGTWTCTA AT).
The thermocycling procedure consisted of an initial denaturation step at 95°C for 2 minutes, followed by 25 cycles each consisting of 94°C for 30 seconds (denaturation), 55°C for 30 seconds (annealing), and 72°C for 30 seconds (extension), and a nal extension at 72°C for 5 minutes. Each reaction was conducted in a 20-μl reaction mixture containing 10 ng of template DNA, 5 μM each primer, 2.5 mM deoxynucleoside triphosphate mix, and 1 unit of FastPFU Polymerase (TransGen Biotech, China). PCR cycling reactions were performed in a GeneAmp® 9700DNA thermocycler (ABI, USA), and the ampli ed products were visualized on agarose gel containing EB and puri ed with a DNA gel extraction kit (Axygen Inc., USA).
Prior to sequencing, the DNA concentration of each PCR product was determined, and the amplicons from each PCR were pooled together in equimolar ratios to reduce biases of each individual reaction and subjected to emulsion PCR to generate amplicon libraries [42]. Deep sequencing was performed on the Illumina MiSeq platform at the Majorbio Bio-pharm Technology Co., Ltd. (Shanghai, China). Any sequence with more than two base mismatches was discarded by Seqcln software analysis. The lowquality sequences and redundant reads were further trimmed using Mothur software. The "dist.seqs" command was performed to identify OTUs by 97% similarity. The obtained sequences were subjected to Megablast and searched against SILVA, aligning to the 16S small subunit rRNA sequence database (version 111), to acquire high taxonomic resolution. The rarefaction curves, Chao1 richness, and Shannon diversity index were determined by Mothur analysis [43]. The 16S rRNA gene sequencing data were deposited in the National Center for Biotechnology Information (NCBI) Sequence Read Archive under accession number PRJNA578836.
Coculture experiments T GFP was cocultivated with Proteus vulgaris and Sphingobacterium lactis in visual biomimetic reactors that can simulate peristalsis of the human intestinal tract [44], and a monoculture of T GFP was used as the control. In addition, MP was added in the reactors to validate the clearance effect of T GFP during bacterial coculture. The medium used contained corn our (20.0 g l -1 ), protein powder (20 g l -1 ), glucose (5.0 g l -1 ), Ox-gall salt (10.0 g l -1 ), Na 2 HPO 4 (10.0 g l -1 ), KH 2 PO 4 (10.0 g l -1 ), and NaCl (1.0 g l -1 ), and the initial pH value was 7.2-7.5. During the cocultures, the compression frequency of the peristaltic pump was 10 times min -1 , the compression range was 1 cm, and the T GFP quantity was measured every 12 hours.

Statistical methods
Statistical analyses to identify signi cant differences were performed using SPSS software. Unless otherwise speci ed, all data are presented as the mean±SEM. When three or more means were compared for statistical signi cance, one or two-way ANOVA was conducted with treatments as independent factors. When two groups of measurements were examined for statistical signi cance, the two-sided Student's t-test was conducted, and a p-value <0.05 was considered statistically signi cant. A dissimilarity test of fecal samples was performed in R based on the Bray-Curtis dissimilarity index using analysis of similarities [45]. Co-occurrence and correlation network analysis were performed by Networkx software [46], and only Spearman correlations with an r > 0.6 (p < 0.05) were considered to indicate a valid interactive event.
Declarations Figure 1 Genome characteristics and phylogenetic analysis of phages W1 and W3. a Genome characteristics of phage W1. b Phylogenetic tree of phage W1 based on the complete genome sequence. c Genome characteristics of phage W3. d Phylogenetic tree of phage W3 based on the complete genome sequence (con dence values above 50% obtained from 1000-replicate bootstraps are indicated at branch nodes. The scale bar indicates the number of base substitutions per site).

Figure 2
Host range analysis of phages W1 and W3.  TGFP Knockout test in vivo. a Sampling sites. TGFP quantity at different intestinal sections in LCGP (b) and HCGP (c). d TGFP quantity in fecal samples in LCGP and HCGP. e TGFP quantity in feces after phage treatments at hour 24 in LCGP. TGFP quantity in the caecum (f), colon (g), and feces (h) after phage treatment at 48 hours in LCGP. i Variation in TGFP quantity with time after two MP gastric perfusions in HCGP. j TGFP quantity in the caecum and colon at hour 120 after two MP treatments in HCGP. k Variation in TGFP quantity in feces with time after continuous MP treatment when TGFP was constantly supplemented in the mouse gut. l TGFP quantity in the caecum and colon at hour 168 after continuous MP treatment when TGFP was constantly supplemented in the mouse gut. ns, nonsigni cant; *p < 0.05; **p < 0.01.  Enumeration of E. coli as well as correlation network analysis of bacteria in knockout, and coculture tests.