Expansion of bacteriophages is linked to aggravated intestinal inflammation and colitis

SUMMARY

Bacteriophages are the most abundant members of the microbiota and have the potential to shape gut bacterial communities. Changes to bacteriophage composition is associated with disease but how phages impact mammalian health remains unclear. We noted an induction of host immunity when experimentally treating bacterially-driven cancer, leading us to test whether bacteriophages alter immune responses. Treating germfree mice with bacteriophages lead to immune cell expansion in the gut. Lactobacillus, Escherichia, and Bacteroides bacteriophages and phage DNA stimulated IFN-γ via the nucleotide-sensing receptor TLR9. The resultant immune responses were both phage-and bacteria-specific. Additionally, increasing bacteriophage levels exacerbated colitis via TLR9 and IFN-γ. Similarly, Ulcerative Colitis (UC) patients responsive to fecal-microbiota-transfer (FMT) have reduced phages compared to non-responders and mucosal IFN-γ positively correlates with bacteriophage levels. Bacteriophages from active UC patients induced more IFN-γ compared to healthy individuals. Collectively, these results indicate that bacteriophages can alter mucosal immunity to impact mammalian health.

Graphical Abstract

In Brief

Bacteriophages are abundant components of the gut microbiota but how they impact health and immunity is unknown. Gogokhia et al. report that bacteriophages activate IFN-γ through a TLR9-dependent pathway and exacerbate colitis. Supporting this, increased abundance of bacteriophages in patients with ulcerative colitis correlates with mucosal IFN-γ responses.

INTRODUCTION

Mammals are colonized with microorganisms from all three domains of life that have significant impacts on mammalian immunity and health (Gensollen et al., 2016). A large component of our microbiota is not classified within these domains as they are unable to self-replicate. Intestinal viruses, while containing members that directly infect eukaryotic cells, are largely composed of bacteriophages that target bacteria (Cadwell, 2015; Reyes et al., 2010; Virgin, 2014). In most ecosystems, bacteriophages outnumber bacteria by a factor of 10, and likely represent the most abundant foreign microorganism on the mammalian body (Suttle, 2005). While many bacteriophages reside as prophages within their bacterial host, many are found as free phage virions embedded within the mucus layer and have been proposed to function to maintain the intestinal barrier by controlling invasive bacterial populations (Barr et al., 2013).

Phages have been utilized as curative elements in a variety of infections, however the success of these treatments has been variable, and therefore antibiotics have dominated the therapeutic landscape (Kutter et al., 2010; Tsonos et al., 2014). With the rapid evolution of bacterial resistance to antibiotics, modulation of bacterial responses by phage has regained interest and is currently being tested in a variety of clinical and experimental settings(Bragg et al., 2014). Many diseases such as inflammatory bowel disease (IBD) and colorectal cancer are hypothesized to arise from alterations in the resident microbiota, with specific causal organisms identified (Arthur et al., 2012; Palm et al., 2014; Rubinstein et al., 2013). Based on this, bacteriophage-based therapies may emerge as a unique treatment option for diseases associated with the microbiota.

Bacteriophages are members of the resident gastro-intestinal microbiota and their composition is significantly different in individuals with IBD when compared to healthy controls (Lepage et al., 2008; Norman et al., 2015). The most notable change is an increase in the order of bacteriophages, Caudovirales, within the intestine of individuals with Crohn’s disease (CD) and ulcerative colitis (UC)(Norman et al., 2015; Wagner et al., 2013). While these viruses do not directly infect the mammalian host, they possess many molecules that could potentially stimulate the immune system (Duerkop and Hooper, 2013;Gorski et al., 2012; Kurzepa et al., 2009;. Hodyra-Stefaniak et al., 2015; Tetz et al., 2017;(Roach et al., 2017). However, it remains unclear whether bacteriophage stimulation of immunity could have consequences on mammalian disease.

RESULTS

Isolation of human associated bacteriophages against adherent invasive E. coli

Perturbations to commensal microbes that result in an outgrowth of potentially harmful organisms are associated with colorectal cancer (Levy et al., 2017). Adherent invasive Escherichia coli (AIEC) and Fusobacterium nucleatum are found on tumors in individuals with intestinal cancer (Kostic et al., 2013; Winter et al., 2013). E. coli NC101, an AIEC, enhances the growth of intestinal tumors in animals that are prone to inflammation, such as IL-10^−/− mice (Arthur et al., 2012), or in genetically susceptible, non-inflammatory models, like the APC^min mouse(Dejea et al., 2018). Mutations in tumor suppressor adenomatous polyposis coli (APC) is a major initiating factor in the etiology of colorectal cancer in humans(Barker et al., 2009). Disease in this model is significantly worsened by colonization with AIEC strains (Dejea et al., 2018). To determine whether bacteriophages could be employed to target AIEC in colorectal cancer, we sought to isolate phages against AIEC from the human microbiota.

We identified and sequenced several AIEC-specific bacteriophages from an individual with IBD, and chose three that were easily propagated and purified (Figures 1A and S1). Phage NC-A has Podoviridae short tailed morphology and closely resembles E. coli phage T7. Phage NC-B is a temperate phage, also a member of the Podoviridae with relatively rare elongated head morphology and a close relative of E. coli phages øEco32 and ECPB2 (Nho et al., 2012; Savalia et al., 2008). Phage NC-G has Myoviridae long-contractile tailed morphology and is related to the well-studied T4-like phages (Figure S2)(Grose and Casjens, 2014). All of these phages belong to the order Caudovirales, consistent with this order of bacteriophages being increased in individuals with Crohn’s disease (Norman et al., 2015; Wagner et al., 2013).

Figure 1. Bacteriophages against AIEC can reduce bacterial and tumor burden.

(A) Electron micrograph of negatively stained, purified bacteriophages used in this study.

(b) Germfree animals were colonized with a contrived community of five different organisms including E. coli NC101, Bacteriodes fragilis, Lactobacillus johnsonii, Bifidobacterium longum and Clostridium symbiosum and treated with either heat killed or live E. coli phages within their drinking water for 24 hours (indicated by arrows) and E.coli colonization was monitored overtime from the feces by plating.

(C-E) Germfree animals were placed on drinking water containing phages or vehicle control and colonized with E.coli NC101. Bacterial titers in feces were determined by plating on selective media n= 6 animals per group. Levels of E. coli NC101 from animals treated in C were determined by q-RT-PCR in small intestinal luminal contents and tissue. n= 6 animals per group. Data are represented as mean ± SEM. * p<0.05, **p<0.01, ***p<0.005, ****p<0.0001 as determined by a student’s t-test. See also Figure S1.

To verify that each phage, or the combination of phages was able to reduce levels of AIEC in vivo, we associated germfree mice with a small community of bacteria that included Bacteroides fragilis, Lactobacillus johnsonii, Bifidobacterium longum and Clostridium symbiosum with E. coli NC101 and subsequently treated with phages. A single delivery of the phage cocktail reduced the colonization of E. coli NC101 for a single day, at which point the bacteria quickly rebounded and became resistant to subsequent phage treatment (Figure 1B). However, continuous administration of phages within the drinking water was sufficient to persistently suppress bacterial growth in the lumen and at tissue sites (Figure 1C–E).

A bacteriophage cocktail can prevent exacerbated intestinal tumor growth

Use of bacteriophages to target specific bacteria and alleviate disease has been successful in multiple settings (Chadha et al., 2016; Kutter et al., 2010; Roach et al., 2017), however it has yet to be tested in experimental cancer models. Therefore, we sought to use purified phage to prevent tumor growth in a mouse model of bacteria aggravated-colorectal cancer. Bacteriophages were extensively purified using several methods to remove the residual contaminating lipopolysaccharide (LPS), that might be present from propagating the phage in its host (Szermer-Olearnik and Boratynski, 2015). The LPS concentration in phage preparations was monitored using multiple assays and was consistently just above background levels found within normal drinking water (Figure S3A). This residual level of LPS was not sufficient to stimulate bone marrow derived dendritic cells (BMDCs), which are known to detect very low levels of LPS (Figure S3B). Regardless, similar concentrations of LPS were added as a vehicle control throughout all experiments. Multiple reports have estimated that the relative bacteriophage load within the intestine is between 5×10⁷-10¹² PFU per gram of feces (Breitbart et al., 2003; Lepage et al., 2008; Mills et al., 2013). Based on this, we administered approximately total 3 ×10⁷ PFU particles per day. Composite E. coli phage or vehicle control was added to the drinking water of specific pathogen free (SPF) and animals were colonized with E.coli NC101. Colonization of APC^min mice with E. coli NC101 leads to worsened disease outcome (Figure 2A and B). While APC^min mice displayed no overall difference in the number of tumors that formed within the small intestine, colonization with E. coli NC101 resulted in a significant increase in large tumor formation (Figure 2A to C). Composite bacteriophage treatment in the absence of AIEC had no effect on tumor growth (Figure 2B). Importantly, bacteriophage treatment of AIEC colonized APC^min animals reduced E. coli colonization (Figure 2D). This reduction in E. coli NC101 colonization was associated with decreased tumor size and enhanced survival (Figure 2A to C). Supporting these results, transcriptomic analysis of the small intestinal tissues revealed a drastic down-regulation of genes associated with colorectal cancer, tumor growth, metastasis and invasion of GI cancer in animals treated with E. coli bacteriophages (Figure 2E). These results demonstrate that continuous treatment with bacteriophages isolated from the human GI tract can effectively reduce target bacteria in the intestine and protect from an invasive bacteria-exacerbated colorectal cancer.

Figure 2. Bacteriophages isolated from the human virobiota can control AIEC colonization preventing tumor growth and mortality.

(A) Survival of APC^min mice colonized with E. coli NC101 and either treated for 70 days with vehicle or a cocktail of 3 phages that infect E. coli NC101. (n=13, 6, 20 and18 animals per group, respectively).

(B-C) Tumors over 8mm diameter throughout the small intestine (SI) were counted 70 days’ post-treatment (n= 8 for non-E. coli NC101 colonized controls, n=5 for phage colonized control, n=16 for vehicle treated and n=20 animals for E. coli phage cocktail treated). The results of five independent experiments are summed.

(D) E. coli NC101 levels in the SI tissue determined by qRT-PCR n=6/6 animals for vehicle treated and phage treated animals.

(E-F) RNAseq analysis of animals treated as in (A) (n=5 and 4 animals per group).

G-J) Percentage and number of CD4+ cells and activated T cells within the MLNs in vehicle treated (n=7) and in phage treated animals (n=9). The data is representative of five different experiments. Data are represented as mean ± SEM. *p<0.05, **p<0.01, ***p<0.005, ****p<0.001 as determined by student’s t test (D, H, J), Mantel-Cox test (A) and One-way ANOVA with Holm-Sidak Multiple Comparison test (B).

See also Figure S3

Bacteriophage treated animals display heightened immune responses

Analysis of the gene expression profiling data showed that despite the global down-regulation of genes associated with cancer, a large set of immune system transcripts and pathways were up-regulated in phage treated animals compared to control. Specifically, induction of both innate and adaptive immunity was upregulated in bacteriophage treated animals (Figure 2F). Supporting this, there were small but significant increases in the percentage of CD4+ and CD8+ T cells in the mesenteric lymph nodes (MLNs) of bacteriophage treated animals (Figure 2G, H and Figure S3C and D). Additionally, there were fewer naive and increased activated CD4+ T cells in both the spleen and MLNs of APC^min mice that underwent bacteriophage treatment (Figure 2I, J and Figure S3E to H). Two possibilities exist for the increased immune responses observed in bacteriophage treated mice. The first, is that bacteriophage mediated lysis of E. coli releases multiple bacterial components that could stimulate immunity, while the second, is that bacteriophages directly stimulate the mammalian immune system. While the latter has been speculated by multiple investigators, it has yet to be directly tested in vivo.

Bacteriophages can stimulate the mammalian immune system

A few reports have demonstrated that bacteriophage preparations can elicit the induction of mammalian cytokines in vitro. In contrast, a more recent study has demonstrated that neutrophils are required to prevent the outgrowth of phage resistant bacteria during phage therapy of a lung Pseudomonas infection, but reported that a single pulse of phage alone did not elicit immune induction using RNA-seq (Roach et al., 2017). Thus, whether bacteriophages can directly stimulate the immune response remains contentious and has yet to be definitively tested. Based on this, we administered the purified E. coli bacteriophage cocktail or a vehicle control to germfree (GF) mice that are devoid of bacterial colonization. While temperate phages dominate the microbial landscape, free bacteriophage are found to be embedded within the intestinal mucus during the steady state and have been reported to be transcytosed by intestinal epithelial cell lines (Barr et al., 2013; Nguyen et al., 2017). Additionally, during intestinal colitis, there is an increase in the abundance of phage that is not associated with its bacterial host (Duerkop et al., 2018; Norman et al., 2015). These represent situations where bacteriophages residing outside of bacteria would be available to potentially stimulate immunity, making this a physiologically relevant experiment. Continuous phage treatment resulted in a steady level of live phage within the gut, indicating that bacteriophages survive the transition through the gastrointestinal tract (Figure S4A and B). Consistent with previous reports, GF mice had significantly fewer CD4+ T cells within the Peyer’s patches (PP) when compared to SPF mice (Mazmanian et al., 2005), while GF animals treated with bacteriophages had similar proportions and numbers of CD4+ T cells as found in SPF mice (Figure 3A to C). Heat-killing of bacteriophages dissociates intact phage virions, disrupting the capsid structure and denaturing DNA complexes present within the phage, while any residual LPS within the sample is heat stabile(Hecker et al., 1994). Therefore, heat killed phage preparations were used to control for LPS and to determine whether intact, infectious particles were required for immune activation. Heat-killed phage preparations did not cause expansion of CD4+ T cells, which remained similar to the levels of CD4+ T cells in GF mice (Figure 3A to C). Specific microbes within the gut have been demonstrated to impact CD4+ T cell development leading to differential induction of Th1, Th17 or Treg responses (Belkaid and Hand, 2014). Bacteriophage treatment did not impact the development of Th17 or Treg cells in GF mice (Figure 3D, G and H and Figure S4C to E), but significantly increased the proportion and number of IFN-γ producing T cells within the PP, but not the spleen (Figure 3D–F). This was observed in both outbred (Swiss Webster; Figure 3E) and inbred animals (C57Bl/6; Figure 3I), indicating that this stimulation is not a nuanced phenotype of a particular mouse strain. A statistically significant increase in CD8+ T cells was also seen in bacteriophage treated GF animals (Figure 3J). Similar results were attained for gut CD4+ and CD8+ T cells using the well-characterized T4 phage, supporting that induction of mammalian immune responses is not exclusive to the bacteriophages that we isolated from human stool (Figure 3K–M and Figure S4F–H). Norovirus replicates within eukaryotic cells and has been shown to correct the defective intestinal architecture in GF mice in a type I interferon dependent manner (Kernbauer et al., 2014). However, we find that bacteriophages do not correct these deficiencies (Figure S5A and B). Additionally, there was no induction of type I interferon responsive genes, IFN-β or MX-1 during phage treatment, suggesting that bacteriophages do not stimulate this pathway (Figure S5C and D). Collectively, these results argue that bacteriophages can directly stimulate host immunity.

Figure 3. Bacteriophages directly stimulate gut immune system development.

(A-J) Swiss Webster germfree animals were treated with either vehicle control or the cocktail of three E. coli phages in their drinking water for four weeks. GF indicates germfree mice, SPF indicates specific pathogen free mice, HK indicates heat-killed.

(A-C) The percentage and number of CD4+ T cells determined by flow cytometry within the Peyer’s patches in the phage treated animals (n=23), the vehicle animals (n= 19), the germfree animals (n=12) and the SPF animals (n=5). These data are combined from at least four independent experiments.

(D-H) The proportion of Th1 or Th17 producing cells within the PP and spleen determined by flow cytometry. Each dot indicates an individual mouse. n=8 animals per group for vehicle and phage treated groups from two independent experiments.

(I) Th1 cell analysis in germfree B6 animals were treated with vehicle or the cocktail of E. coli phages within their drinking water for four weeks (n=5/group).

(J) CD8+CD3+ T cells in the PPs of animals treated as described in A by flow cytometry (n=13/group).

(K-M) The percentage and number of CD4+ T cells determined by flow cytometry within the Peyer’s patches of germfree animals treated with either vehicle control or T4 phage in their drinking water for four weeks. The percentage and number of CD4+ or CD8+ T cells was determined within the PP. Data are represented as mean ± SEM.*p<0.05, **p<0.01, ****p<0.001, ns-not significant as determined by student’s t test.

See also Figure S4 and S5A–D.

Bacteriophages activate IFN-γ responses through TLR9

Since bacteriophages do not infect eukaryotic cells, we sought to understand the mechanism by which they could stimulate IFN-γ using in vitro approaches. While multiple cell types produce IFN-γ in vivo, such as ILCs and NKT cells, for these in vitro experiments we chose to focus on CD4+ T cell produced IFN-γ as these cells are readily isolated and there are well established methods to work with them in culture. Bacteriophages do not directly stimulate IFN-γ production from T cells (Figure S5E). Generally, T cell responses are initiated by interactions with dendritic cells (DCs) in the gut, and DCs are known to directly sample and present antigens found within the intestine (Rescigno et al., 2001). As additional controls, we also included heat-killed phage, and phages against the prominent gut commensals, Bacteroides thetaiotaomicron and the gram-positive commensal Lactobacillus plantarum. DCs pre-incubated with any type of bacteriophage significantly increased the production of IFN-γ by CD4+ T cells (Figure 4A), indicating that bacteriophage recognition by DCs could stimulate IFN-γ production. Intact bacteriophages harbor various capsid proteins and DNA within the capsid head that could mediate stimulation of the immune system. Bacteriophage treatment with EDTA causes release of DNA from the phage head, leaving the capsid empty but intact (Bauer et al., 2015). Treatment of DCs with DNAase treated empty capsid did not stimulate IFN-γ in culture, however an over 40 fold induction of IFN-γ was initiated upon treatment of DCs with purified phage DNA (Figure 4B). This is consistent with a recent report that failed to detect immune responses against three T4 capsid proteins (Miernikiewicz et al., 2013). Therefore, our results suggest that DC recognition of bacteriophage DNA can stimulate IFN-γ.

Figure 4. Bacteriophages directly stimulate commensal and phage specific Th1 responses via TLR9.

(A-B) IFN-γ measured by ELISA in the supernatant of CD4+ cells co-cultured with phage primed WT, TLR3^−/−, or TLR9^−/− DCs derived from 3 animals.

(C) IL12 measured by ELISA after 24 hours of BMDC incubation with phages and controls.

(D) BMDC incubated with SYBR-green pre-stained phage (green) (bottom), or stained directly with SYBR green (top) fixed and stained with Alexa Fluor 555 Phalloidin for cytoplasm staining (red).

(E) BMDCs incubated with SYBR green pre-stained phage and subsequently stained with Alexa Fluor 680 Phallodin (blue) or a Histone 2B-RFP dye (red). Each stain is shown individually and the overlay of these is shown in the bottom right. Data are represented as mean ± SEM. *p<0.05, **p<0.01, ***p<0.005, ****p<0.001, ns-not significant as determined by student’s t test (C) and One-way ANOVA with Bonferroni’s Multiple Comparison Test (A, B).

See also Figure S5E and S6

Several pattern recognition receptors (PRRs) function to recognize viruses and foreign nucleic acids (Medzhitov and Janeway, 2000). These include the RIG-I like receptors, Rig-I, MDA5, and LGP2, that act as sensors of viral replication within cells, TLR3, which is stimulated by double stranded RNA, and TLR9 which recognizes unmethylated CpG dinucleotides, a motif associated with microbial DNA (Kawai and Akira, 2008). Bacteriophages cannot infect or replicate within eukaryotic hosts and therefore would not be recognized by the RIG-I machinery. Based on this, we focused on nucleic acid sensing receptors. TLR3 and TLR9 deficient DCs were pulsed with the indicated bacteriophages and subsequently incubated with T cells. While induction of Th1 responses were readily induced in TLR3^−/− DCs in response to bacteriophage treatment, IFN-γ production from T cells was completely lost when stimulated with TLR9^−/− DCs (Figure 4A). Similarly, incubation of purified phage DNA with TLR9^−/− deficient DCs failed to stimulate IFN-γ production (Figure 4B). The cytokine IL-12 is produced by DCs and is known to stimulate IFN-γ from multiple cell types, therefore, we analyzed DC activation in response to bacteriophage treatment. E. coli phages induced the expression of several cytokines including IL-12, IL-6, IL-10, and IP-10 but not IL-1α, TNF-α, or the surface markers CD80 or CD86 over vehicle control (Figure 4C and Figure S6A to G). Induction of these genes was dependent on Myd88, an adaptor molecule downstream of most TLRs, consistent with recognition of bacteriophages through TLR9 (Figure S6H). Moreover, DCs were still activated in the absence of CD14, a molecule critical to the recognition of LPS, further supporting that residual LPS contamination in bacteriophage preparations was not responsible for the induction of immune responses (Dobrovolskaia and Vogel, 2002) (Figure S5I). Intact E. coli and L. plantarum phages potently induced IL-12 production in DCs that was blocked by the addition of a TLR9 chemical antagonist (Figure 4C). Moreover, phage DNA and UV treated phage, but not DNA purified from DCs, heat killed phage or capsid, stimulated IL-12 production from DCs (Figure 4C). Consistent with our results, UV treated DNA has been reported to maintain the ability to activate TLR signaling (Figure 4C)(Harberts and Gaspari, 2013).

TLR9 is known to be an intracellular receptor that signals within endosomes. Based on this, intact bacteriophages or phage DNA would have to be taken up to stimulate host immune responses. Consistent with this, phages stained with a DNA-binding dye and subsequently incubated with DCs can be found within the cytoplasm of dendritic cells (Figure 4D). These puncta were never detected in dendritic cells incubated in the absence of phage and stained with the DNA dye, nor was histone-containing DNA (that would indicate DNA from the DC) ever detected as such puncta (Figure 4D, E and Figure S6J). Importantly, bacteriophage mediated IL-12 production from DCs is lost in the presence of Cytochalasin D, an inhibitor of actin polymerization which prevents endocytosis (Figure 4C). Therefore, our data demonstrate that bacteriophage DNA is required for activation of DCs through the intracellular receptor TLR9.

Bacteriophages stimulate commensal and phage specific immunity.

T cells are known to mount antigen-specific responses against bacterial, viral, commensal or even self-antigens. Based on this, we tested whether bacteriophage recognition by DCs would lead to phage-specific responses or if bacteriophages could function as an adjuvant against other antigens within the gut. MHCII expression on DCs is required for bacteriophage mediated induction of IFN-γ producing cells (Figure 5A). OT-II T cells, which express the TCR that recognizes ovalbumin (OVA), was used as a source of T cells in our co-culture system. DCs were pre-pulsed with OVA and B. thetaiotaomicron, L. plantarum, or E. coli phages and subsequently incubated with OT-II T cells. Treatment with phages potently stimulated OVA-specific IFN-γ production in a TLR9, but not a TLR3 dependent fashion, suggesting that bacteriophages directly stimulate immunity (Figure 5B). To directly test this in vivo, GF mice were mono-associated with B. fragilis expressing the OVA protein, allowing us to track antigen specific responses against a commensal (Figure 5C)(Kubinak et al., 2015). T cells were then isolated from animals treated with either vehicle or phage containing drinking water and subsequently re-stimulated with DCs pulsed with either OVA antigen or bacteriophage antigens (Figure 5C). Phage treatment did not influence the colonization levels of B. fragilis-OVA, further indicating the inability of these E. coli phages to lyse off-target bacteria (Figure S7A). Bacteriophage treated animals had significantly increased OVA-specific IFN-γ production that was not present in vehicle treated animals (Figure 5D). Phage-specific IFN-γ production was also significantly elevated in bacteriophage treated mice (Figure 5D). Vehicle treated mice exhibit a four-fold induction of phage specific IFN-γ over media controls, consistent with levels of IFN-γ observed in the previous in vitro experiments. However, bacteriophage treated GF animals had an even greater induction of IFN-γ in response to phage antigens, indicating that phage specific responses are primed in vivo. Supporting this, anti-phage IgA can also be detected within the feces (Figure 5E). Collectively, these data demonstrate that bacteriophages within the gut can stimulate both phage specific as well as non-specific immune responses and can thus enhance immunity against others organisms present within the gut.

Figure 5. Bacteriophages act to induce both commensal and phage specific immunity.

(A) IFN-γ measured by ELISA in the supernatant of BMDCs from MHCII^−/− mice pulsed with phage for 24 hours and co-cultured with WT T cells for 72 hours.

(B) IFN-γ measured by ELISA in the supernatant of BMDCs from WT mice pulsed with antigens and ovalbumin (100 μg/mL) and co-cultured with CD4+ T cells from OTII mice.

(C, D) Experimental set-up of Phage-Ovalbumin co-culture experiments. IFN-γ was measured in the supernatant by ELISA.

(E) Feces from bacteriophage cocktail treated germfree mice assayed for phage specific IgA by ELISA. Data are represented as mean ± SEM. *p<0.05, **p<0.01, ***p<0.005, p<0.001, ns-not significant as determined by student’s t test and One-way ANOVA with Bonferroni’s Multiple Comparison Test (A, B, D).

See also Figure S7A

Enhanced bacteriophage abundance exacerbates intestinal colitis through TLR9 and IFN-γ

Inflammatory bowel disease is a complex interaction between genetic susceptibility, environment and microbiota (Manichanh et al., 2012; Vermeire et al., 2011). Individuals with IBD possess a significant loss of bacterial diversity and a concomitant increase in bacteriophage diversity and abundance (Gevers et al., 2014; Norman et al., 2015). More recent studies have shown that the bacteriophage community transitions to a stochastic dysbiosis during murine colitis (Duerkop et al., 2018). Whether these changes to the bacteriophage community matter during intestinal colitis has yet to be tested. Individuals with IBD have a specific expansion in the richness and diversity of Caudovirales bacteriophages. Based on this, we utilized our bacteriophage cocktail, that contains Caudovirales phages, to artificially increase the levels of free bacteriophage within the gut to test a role for bacteriophages during intestinal colitis.

Conventionally colonized animals (specific pathogen free-SPF) were provided drinking water containing the bacteriophage cocktail and subsequently induced for DSS colitis. Of note, our mouse colony is not associated with any detectable levels of E. coli as measured by colony formation on plates and 16s sequencing, and therefore our phage cocktail should not induce lysis of any endogenous microbes (Fig. S7B). Animals with colitis that had previously received the cocktail of E. coli bacteriophages lost significantly more weight than vehicle treated mice and had worse histology scores (Figure 6AC–c). Additionally, there was an increase in the proportion and number of CD4+ and CD8+ T cells in the MLNs and significantly more of these cells exhibited an activated phenotype and more IFN-γ and IL-17a producing T cells (Figure 6D to J and Figure S7C–G). To determine whether TLR9 sensing within the host serves to initiate bacteriophage mediated exacerbation of colitis, we treated TLR9^−/− animals with the E. coli bacteriophage cocktail and induced DSS colitis. While disease severity was significantly worsened by bacteriophage treatment in WT mice, TLR9^−/− animals were protected from bacteriophage enhanced colitis and had reduced immune responses (Figure 6K–M). Similarly, induction of IFN-γ by phage is required for exacerbated colitis as phage treated IFN-γ^−/− animals do not develop as severe colitis as WT animals as assayed by histology and colonic shortening (Figure 6N and O and Figure S7H and I). Collectively, these data indicate that local increases in bacteriophage abundance, as seen in patients with IBD, might also function to exacerbate the inflammation within the intestine and contribute to disease severity.

Figure 6. Increased abundance of host-free bacteriophage within the gut exacerbates intestinal colitis.

SPF WT (A-J), TLR9^−/− (K-M) IFN-γ^−/− (N-O) mice were treated with the E. coli bacteriophage cocktail for 3-4 weeks prior to induction of DSS colitis.

(A) Weight loss,

(B, M, O) clinical score,

(C, L, N) histology by H&E staining of colons from indicated animals.

(D-K) Percent and number of total CD4 and CD8 T cells and Th1 and Th17 cells within the MLN of SPF WT mice was determined by flow cytometry. Data shown is from one representative experiment with 6–8 animals per group. Similar results were seen in three independent experiments with total n=23 for vehicle treated animal and n=21 for bacteriophage treated animals. Data are represented as mean ± SEM. *p<0.05, **p<0.01, ***p<0.005, ****p<0.001, ns-not significant as determined by student’s t test (B, E, F, G, I, J), Mantel-Cox test (A) and One-way ANOVA with Bonferroni’s Multiple Comparison Test (K, M, O).

See also Figure S7B–G

Caudovirales bacteriophages are increased in individuals that do not respond to FMT

Fecal microbiota transplants (FMTs) are emerging as a mechanism to restore healthy microbial communities and have proven clinically successful for the treatment of Clostridium difficile infection. More recently, FMT for IBD has begun to be trialed to restore microbial diversity that is frequently associated with disease severity in both CD and UC patients. Several recent studies have demonstrated clinical success of FMT in patients with UC(Jacob et al., 2017; Moayyedi et al., 2015). FMT increases microbial diversity in recipients irrespective of the clinical response and it remains unclear if diversity or specific bacterial taxa are responsible for the clinical efficacy (Jacob et al IBD, 2017). We examined if differences in the intestinal bacteriophage community impact the clinical success of FMT in active UC patients. To this end, we performed total nucleic acid sequencing to detect the viral community in 20 individuals with active UC prior to and 4 weeks after receiving FMT (Figure 7A). Comparison of bacteriophage communities detected by VirMAP (Ajami et al., 2018) between healthy donors and UC recipients revealed a significant increase in the relative abundance of Caudovirales bacteriophages in individuals with UC, consistent with recent findings (Figure 7B) (Frank et al., 2007). Patients that had a clinical response to FMT had a lower relative abundance of Caudovirales bacteriophages at the time of transplant compared to patients that did not respond to therapy (Fig. 7C). Furthermore, the relative abundance of Caudovirales in non-responders increased after FMT while no change was observed in responders (Figure 7C). Additionally, consistent with our mechanistic findings in mice, CD4+ T cell production of IFN-γ from rectal mucosal biopsies show a positive correlation between IFN-γ and the relative abundance of total viral reads as well as Caudiovirales specifically (Figure 7D, Figure S7J and K). To validate our in vivo findings that colitis associated bacteriophages can induce IFN-γ producing T cells, we isolated virus-like particles (VLPs) from healthy patients, active UC and inactive UC in remission. DCs pulsed with UC-associated and healthy VLPs were co-cultured with naive CD4+ T cells. VLPs from active UC, but not inactive UC or healthy controls caused activation of naive CD4+ T cells and potent induction of IFN-γ (Figure 7E). Collectively, our studies have identified the functional evidence that commensal bacteriophage communities can directly exacerbate intestinal disease and their abundance can correlate with success of FMT.

Figure 7. Caudovirales are increased in individuals that do not respond to FMT.

(A) Prospective, open label pilot study design to assess the safety and efficacy of FMT in patients (n=20) with active Ulcerative Colitis.

(B, C) Average viral relative abundance per group determined on a per-sample basis by the ratio of viral reads per taxon relative to total viral reads as classified by VirMAP.

D) Correlation of total number of viral reads called by VirMAP per sample against percentage of CD4+ T cells producing IFN-γ in total lamina propria mononuclear cells (LPMC) from rectal biopsies of patients with active UC as measured by flow cytometry. Significance determined by spearman correlation.

(E) IFN-γ measured by ELISA in the supernatant of BMDCs from WT mice pulsed with VLPs isolated from healthy, inactive and active UC patients for 24 hours and co-cultured with WT T cells for 72 hours. Data are represented as mean ± SEM. *p<0.05, **p<0.01, ***p<0.005, ****p<0.001, ns-not significant as determined by student’s t test.

See also Figure S7J,K

DISCUSSION

There is mounting evidence that disturbances to the gut microbiota can lead to diseases such as IBD, and identification of bacterial members that influence disease outcome has been the focus of much research over the last decade. However, how fungal, archaeal and viral members of the microbiota might affect IBD is less explored (Duerkop et al., 2018; Norman et al., 2015; Sokol et al., 2016). Bacteriophage recognition by the mammalian immune system has been considered by many but only studied in a few limited settings (Dabrowska et al., 2005; Eriksson et al., 2009; Gorski et al., 2012; Khan Mirzaei et al., 2014). More importantly, mechanistic insight of phage recognition by mammalian immune systems and determination of the relevance of phages during disease has yet to be trialed in in vivo settings. Here we dissect the molecular pathway by which bacteriophages influence mammalian immunity within the gastro-intestinal tract, where these organisms are found in the highest abundance. This is highly relevant to human health, as bacteriophage virions have been identified to be abundant within the intestinal mucus where they could be sampled by immune cells (Barr et al., 2013). Additionally, recent reports have demonstrated that intestinal epithelial cell lines are capable of transcytosis of phage (Nguyen et al., 2017). Either by direct uptake or transcytosis, bacteriophages translocate from barrier surfaces to systemic circulation (Breitbart et al., 2003; Duerr et al., 2004; Moustafa et al., 2017) and likely directly interact with immune cells.

While our data demonstrate that IFN-γ induction is required for bacteriophage mediated exacerbation of colitis, multiple cells types are able to produce this cytokine. Here we show that stimulation of DCs can influence IFN-γ production from CD4+T cells in vitro, however, NKT, NK, CD8+ T cells, and ILCs can all produce IFN-γ. Indeed, we see expansion of CD8+ T cells in phage treated GF mice (Fig. 3), suggesting that bacteriophages can influence multiple cell types. As most of these cell types also express TLR9, it is likely that bacteriophages can activate these cells to contribute to exacerbation of colitis. Additionally, just as specific bacteria can be recognized by multiple immune pathways, bacteriophages likely stimulate several other mammalian receptors such as cGAS-Sting, which function as sensors of cytosolic DNA. Therefore, it likely that bacteriophages will have a multi-faceted influence on mammalian immunity through a variety of pathways that should be an area of continued investigation.

Immune activation by bacteriophages could be physiologically relevant and context-specific. As reported in many animal models, specific targeting of pathogenic bacteria by bacteriophages in the infectious setting can successfully reduce the bacterial burden and ameliorate disease. Here we report the use of E. coli specific bacteriophages to significantly reduce colonization by carcinogenic bacteria and increase survival of animals predisposed to colorectal cancer development. Bacteriophage treatment of APC^min animals in the absence of E. coli colonization did not reduce tumor burden, suggesting that bacteriophage stimulated immunity is not sufficient to mount effective anti-tumor immunity in this setting. However, our data demonstrate that during chronic inflammatory diseases, expansion of bacteriophages, could be detrimental. Supporting this, recent reports have indicated that individuals with Crohn’s disease display a significant increase in the abundance of bacteriophages within the order Caudovirales (Norman et al., 2015). Given that bacteriophages can confer differential functions to bacterial populations (i.e. antibiotic resistance), it is interesting to speculate that unique bacteriophages are present in diseased individuals. A recent study supports this, as they identified that Enterobacteriaceae specific phages were enriched in animals during colitis (Duerkop et al. 2018). Our data indicates that these organisms can be functionally relevant during intestinal disease as the presence of bacteriophages significantly enhance intestinal immune responses in a TLR9 dependent manner. Consistent with our findings in mice, we identify a positive correlation between mucosal IFN-γ production and total viral abundance. Additionally, we show, that VLPs isolated from UC patients with active disease, but not inactive disease or healthy controls are capable of inducing potent IFN-γ responses in vitro. Interestingly, Caudovirales bacteriophages are even more significantly enriched in individuals that fail to respond to a fecal transplant which support the notion that elevated Caudovirales phages might predict FMT failure and need for additional maintenance FMT delivery or escalation of treatment. Therefore, bacteriophages may act as double-edged sword within the inflammatory environment of the gut. As immunity stimulated by bacteriophages is in part dependent on TLR9, antagonizing this pathway during phage therapy may ameliorate inflammatory responses against the phage itself. Given the specificity of bacteriophages for their bacterial targets and their influence on the mammalian intestine, future work is warranted to dissect how these complex interactions can be manipulated for therapeutic benefit.

STAR METHODS

CONTACT FOR REAGENTS AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, June L. Round (june.round@path.utah.edu)

EXPERIMENTAL MODEL AND SUBJECT DETAILS

In vivo animal studies:

All experiments were performed with animals that were between 8 and 12 weeks of age. Both C57Bl/6 and Swiss Webster germfree mice were maintained in sterile flexible film isolators (CBC) at the University of Utah germfree facility. Sterility of the germfree isolators was monitored every month by microbiological plating under aerobic and anaerobic conditions and PCR amplification of conserved 16S rDNA using DNA extracted from the feces of multiple germfree mice in each chamber. The germfree colony was established using animals that were originally purchased from Taconic or University of Michigan. Breeder pairs of C57BL/6J-SPF animals were bred in our facility and age, gender matched littermate controls were used throughout the experiment. APC^min, TLR3^−/−, TLR9^−/−, MHCII^−/−, OT II and IFNγ^−/− mice were purchased from The Jackson Laboratory. Animal procedures performed in this study were in compliance with federal regulations and guidelines set by the University of Utah’s Institutional Animal Care and Use Committee (Protocol# 1704009). Experiments were replicated and either representative plots or combined data is presented in the manuscript where indicated. Simple randomization technique was used to select animals for each experimental groups. No specific sample sizes estimation was used, animals were allocated randomly and equally in the groups by the availability. Mice were assessed by staff veterinarian daily. Any animal that was reported for health status by the veterinarian was excluded from the experiments.

Human IBD Subjects:

This prospective, open-label pilot study was registered with ClinicalTrials.gov (, IND 15988) and approved by IRB protocol 1404014982. Patients with active Ulcerative colitis were recruited from Jill Roberts Center for Inflammatory Bowel Disease. Informed consent was obtained from all study participants. Study population included male and female participants of >18 years of age with biopsy proven (Mayo score ≧ 3 and endoscopic subscore ≧ 1), inadequately controlled UC as defined by steroid dependence or the need for escalation of medical care. Patients were screened and those with Crohn’s disease, infectious diarrhea, clinical conditions requiring emergency management, primary sclerosing cholangitis, pregnancy, history of FMT, antibiotic use within previous 2 months and recent malignancy, as well as other conditions deemed FMT unsafe for the patients were excluded from the trial. Donors were recruited by universal stool bank (OpenBiome) and screened rigorously. Two donor FMT material was prepared from individual frozen donor samples prior to the procedure. All patients meeting study enrollment criteria received standard PEG based colonoscopy bowel preparation and FMF material was delivered in the terminal ileum. 2 biopsy samples were obtained prior the FMT administration for cytokine and cellular analysis. Patients were followed post FMT by medical interview and physical exam at 2, 4 and 12 weeks. Fecal samples were collected at screening (Pre-FMT), week 2 and week 4 post FMT and flexible sigmoidoscopy was performed at the end of the study (week 4) to determine the response and remission, as well as obtain biopsy samples of rectal mucosa for cellular analysis. Primary endpoint of safety was determined by obtaining information about adverse events using open ended questions. Secondary endpoints included clinical response ΔMayo score ≧3 and bleeding subscore ≦ 1), clinical remission (Mayo score ≦ 2 and no subscore > 1) and progression of disease (measured by initiation of biologics, escalation of dosage or colectomy).

Primary murine cells.

Bone marrow was isolated from the femur and tibia of male C57Bl/6 and knockout animals by removing soft tissue from the bone using 70% ethanol-soaked paper towels and tweezers. The proximal end of the bone was exposed and the bone marrow was flushed using a 10mL syringe and 29G needle filled with complete RPMI. Single cell suspensions were centrifuged for 5 min at 1350 RPM followed by one-minute incubation with 5mL of RBC lysis buffer (Biolegend). 5mL of RPMI were added to inactivate the RBC lysis buffer and cells were subsequently centrifuged at 1350 rpm for 5 min followed by two washes in complete RPMI. 3x10⁶ cells were plated in tissue culture dishes in complete RPMI media supplemented with 20ng/mL GM-CSF and incubated at 37°C. Cells in suspension were collected and placed in fresh media containing GM-CSF after 72 hours and incubated for an additional four days for a total of eight days. T cells were isolated from the spleen of male C57bl/6 mice by disrupting spleen tissue and passing through 100um cell strainer. Single cell suspension was incubated in RBC lysis buffer and washed twice in complete RPMI. CD4+ T cells were purified using a MACS column with negative selection according to manufacturer’s instructions (Miltenyi Biotech).

Bacterial strains and plasmids.

Bacteroides fragilis OVA expressing strain (Kubinak et al., 2015) was grown at in anaerobic chamber at 37°C for all experiments. Escherichia coli NC101 strain was provided by Dr. R. Balfour Sartor and was grown at 250rpm and 37°C shaker for all experiments. pACYC177 plasmid containing Ampicillin and Kanamycin resistance cassettes (provided by Matt Mulvey, University of Utah School of Medicine).

METHOD DETAILS

Bacteriophage isolation.

Fecal samples from individuals diagnosed with Crohn’s disease were obtained by Dr. John Valentine in the IBD clinic at the University of Utah hospital under approval by the IRB_00059476. Informed consent was obtained from the study patients. Upon arrival, samples were resuspended in 5mL of phosphate buffered saline (PBS) and stored at −80°C until use. The bacterial strain, Escherichia coli NC101, was used as an enrichment host to isolate bacteriophages, provided by Dr. R. Balfour Sartor. For phage isolation, the samples were thawed and an equal volume of HBSS supplemented with Ca²⁺ and Mg²⁺ (Corning Cellgro) was added. After incubation at room temperature for 2 hours, the samples were centrifuged for 20 minutes at 6000 rpm. Supernatant was collected and filtered through a 5μm filter followed by a 0.5μm filter. Five mLs of 10X LB broth (Fisher Bioreagents) 45 mL of fecal filtrate and a 1mL overnight culture of E. coli NC101 were combined and incubated with shaking at 37°C overnight. The following day 500uL chloroform (Fisher Chemical) was added to the overnight culture and shaken for 5 minutes. Cell debris was removed by centrifugation for 20 minutes at 6000 rpm. The supernatant was collected and centrifuged again overnight at 20°C 8000 rpm to pellet viral particles. The resulting supernatant was discarded and the pellet was resuspended in 5mL of HBSS supplemented with 1 mM Ca²⁺ and Mg²⁺ (Corning Cellgro). To remove lipopolysaccharide (LPS) and other bacterial products, the viral particles were further purified by a CsCl step density gradient centrifugation. For this, a gradient was produced using 10% sucrose, 1.4 g/cc and 1.6 g/cc CsCl into 14×89 mm Quickseal tubes (Beckmann). A 5mL viral suspension was carefully overlaid onto this gradient and subsequently ultracentrifuged for 2 hours at 4°C in a SW41 6236 rotor at 38,000 RPM (Model L8M, Beckman, Germany). The lower band at the 1.4/1.6 g/cc interface was collected using a syringe and hollow needle. To remove CsCl from phage preps, viral suspension was then dialyzed through 10K MWCO Snakeskin tubes (ThermoFisher Scientific) in HBSS supplemented with 1 mM Ca²⁺ and Mg²⁺ overnight at 4°C. Plaque assays were used to determine the concentration of phages. Phage solution was serially diluted and 100uL of each dilution was mixed with bacteria, 0.7% LB agar along with 100uL overnight culture of the host. The suspension was carefully distributed on the top of the LB agar plates and incubated overnight at 37°C. Phage titer was calculated by multiplying number of plaques by dilution factor.

Measurement of LPS concentration.

LPS concentration was measured using the LAL Chromogenic Endotoxin Quantitation Kit (Pierce) and EndoZyme (Hyglos HmbH,) according to the manufacturers protocol. The fluorescent signal and optical density were measured in a Synergy H1 microplate reader (Biotech). We chose LPS from E. coli O111:B4 as control since the bacteriophages used in the experiment were specific for E. coli. LPS in vehicle control was 0.5ng/mL of standard LPS from E. coli 0111:b4 (Invivogen).

Phage sequencing.

Phage DNA was isolated using Phage DNA Isolation Kit (Norgen) to prepare individually barcoded, MiSeq compatible, sequencing libraries was prepared using Illumina’s (San Diego, CA) TruSeq DNA PCR-free sample prep kit. These libraries were pooled and run on a single MiSeq paired-end 300 cycle sequencing run. Raw reads were quality filtered and trimmed and TruSeq adapter filtered, using BBDuk (https://sourceforge.net/projects/bbmap/) with a minimum base quality threshold of 20 and a minimum trimmed read retention length of 100. Filtered reads were subsequently assembled using velvet assembler with a hash length of 121 and minimum contig length of 200 nucleotides. Assembled contigs were submitted for gene calling and annotation to PHANTOME and RAST, using the GeneMarkS gene caller. Sequencing was performed at the DNA Sequencing Core Facility, University of Utah.

Electron microscopy.

Negative stain transmission electron microscopy (TEM) was used for imaging purified phage preparations. A concentration of 10¹³ PFU/mL of indicated phages were placed on a carbon-coated copper grid, washed with deionized H₂O, stained with 1% uranyl acetate for 20 seconds and subsequently air dried. The specimen was blotted with filter paper between each step. Electron images were recorded by David Belnap on a FEI Tecnai 12 (Hillsboro, Oregon, USA) and JEOL JEM-1400 Plus (Tokyo, Japan) transmission electron microscope at the University of Utah Electron Microscopy Core laboratory.

Flow Cytometry.

The spleen, MLNs and Peyer’s patches (PPs) were mashed through 40μm cell strainers (Fisher Scientific) using Roswell Park Memorial Institute Medium (RPMI) without Ca²⁺ or Mg²⁺ (Corning Cellgro) supplemented with 10% FBS (HyClone), Pen/Strep (1%), sodium pyruvate (1%), 2.05mM L-glutamine (1%), non-essential amino acids (1%) and β-MercaptoEthanol (0.05mM) (complete RPMI) to obtain single cell suspensions. Cells were centrifuged at 1350 rpm for 5 minutes and subsequently resuspended in complete RPMI solution. For spleen lymphocyte isolation, cells were first incubated in RBC lysis buffer (TONBO Biosciences) for 3 minutes followed by the addition of 5mL of RPMI solution to deactivate the lysis buffer. For surface antibody staining, the cells were plated at 0.5×10⁶ – 1×10⁶ cells per well in a 96-well plate and washed twice with sterile HBSS buffer supplemented with 2.5% FBS (FACS buffer). Cells were stained in 100uL of FACS buffer containing fluorescent antibody stains for 20 minutes in the dark at 4°C. Antibody concentrations are provided below. Cells were washed twice in FACS buffer to remove unbound antibodies and subsequently analyzed on a BD LSRFortessa flow cytometer. For intracellular stains, the cells were stimulated with ionomycin (500 ng/ml) (Sigma Aldrich), PMA (5 ng/ml) (Fisher Biosciences), and Brefeldin A (5 mg/ml) (Biolegend) for 4 hr at 37°C prior to staining. Stimulated cells were surface stained as described above and subsequently permeabilized and fixed in 100μl Perm/Fix buffer (eBiosciences) overnight. Cells were then washed one time with Perm/Wash buffer (eBiosciences) and stained with fluorescent antibodies diluted in Perm/Wash buffer (eBiosciences) for 30 min at 4°C. Cells were washed twice with 1× BD Perm/Wash buffer. Single, unstained and/or isotype controls were used in all experiments to set the appropriate gates. The following antibodies were used: anti-CD4 (eBioscience: clone RM4-5 FITC; eBioscience: clone GK1.5 APC; Biolegend: clone GK1.5 PE) 1:250, anti-CD3 (TONBO-biosciences: clone: 17A2 VioletFluor 450) 1:200, anti-CD8 (eBioscience: clone 53-6.7 PerCP-Cyanine5.5) 1:500, anti-CD62L (Biolegend: clone MEL-14 PE) 1:250, anti-CD69 (eBioscience: H1.2F3 PE/Cyanine7) 1:250. Following antibodies were used for intracellular cytokine staining: anti-Foxp3 (eBioscience: clone FJK-16 s APC/PerCP-Cy5.5) 1:100, anti-IL-17A (eBioscience: clone eBio17B7 eFluor 660; eBioscience: clone: eBio17B7 PE-Cyanine7) 1:100, or anti-IFN-γ (Biolegend: clone XMG1.2 PE) 1:100.

Germ-free animal experiments.

Eight-week-old germ free Swiss Webster animals were removed from the isolators and placed into sterile hepa filtered cages. Animals were provided drinking water supplemented with 50 μg/mL ampicillin (Fisher Scientific), neomycin (Fisher Scientific), erythromycin (Fisher Scientific) and gentamycin (GoldBio) and purified NC phages (3×10⁷ PFU/mL) or vehicle control LPS. While animals were maintained under sterile conditions, addition of antibiotics was used as a precaution to prevent against bacterial contamination as has been described in (Round et al., 2011). Animals maintained sterility throughout the entire experiment.

Determination of NC phage killing efficiency in germfree mice

Eight-week-old germfree mice were transferred to SPF conditions and provided 50μg/mL ampicillin and kanamycin within the drinking water supplemented with phage NC-A, NC-B, NC-G or a mixture of the three. Mice were subsequently orally gavaged with 100 μL 10⁸ E. coli NC101 (Amp^RKan^R) bacterial suspension that was prepared by overnight culture in medium supplemented with 50μg/mL ampicillin and kanamycin. The next day the cells were washed twice with HBSS supplemented with Ca²⁺ and Mg²⁺. Fecal pellets from infected mice were collected each day for 5 days and plated on MacConkey agar (BD Difco) containing 50μg/mL Ampicillin and Kanamycin to determine the level of colonization. DNA was isolated from small intestinal luminal content and 5 mm SI tissue sections to determine colonization of E. coli NC101 at the mucosa. DNA was extracted for bacterial quantification from fecal pellets using the PowerFecal DNA Isolation Kit (Qiagen). Bacteria were quantified with q-PCR using the protocol described in the section “Quantitative real-time PCR” using the following primers: E. coli F 5-CATGCCGCGTGTATGAAGAA-3 and E. coli R 5-CGGGTAACGTCAATGAGCAAA-3.

APC^min cancer model experiment.

4 weeks old APC^min mice were provided the mixture of NC phages in drinking water without antibiotics as described above. Mice were maintained in this condition for 4 weeks, after which mice were challenged with 100 μL E. coli NC101 bacterial suspension via oral gavage biweekly for 2 weeks. Bacterial suspension was prepared with overnight culture of 10mL E. coli NC101. Next day the cells were washed twice with HBSS supplemented with Ca²⁺ and Mg²⁺. Stabile colonization was confirmed by plating the fecal pellets on MacConckey agar. Mice were maintained on drinking water containing phage NC mixture for 2 months. Mice survival was monitored throughout the experiment. At the end of the experiment, mice organs were harvested as described above, as well as the number and size of intestinal polyps were counted. 5mm small intestinal snippets were collected for RNA extraction and bacterial quantification. Immune phenotyping of MLNs and Spleens were performed as described above.

Bone Marrow Dendritic Cell stimulation.

For stimulation of BMDCs as shown in Figure 3, BMDCs were isolated as described above and incubated the phage cocktail at a MOI of 10 for 24 hours followed by analysis of the cells by ELISA or q-RTPCR.

Quantitative real-time PCR.

Expression of various cytokines was determined either with quantitative real-time PCR (Q-PCR) or ELISA. For RNA extraction, cells were resuspended in RiboZol RNA extraction reagent (Amresco) and frozen immediately at −20°C until further use. RNA isolation was performed with RNeasy kit (Qiagen) using manufacturer’s instructions. Reverse transcription using the qScript cDNA superMix (Quanta Biosciences) was performed in accordance with the manufacturer’s instructions. RT-qPCR was performed with LightCycler 480 SYBR Green I Master on a LightCycler 480 machine (Roche). The PCR protocol consisted of one cycle at 95’C (7.5 min) followed by 45 cycles of 95°C (10 sec) and 60°C (20 sec). Expression of ribosomal protein L32 was used as a standard. All primers used for the entire study are listed in Key Resources Table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
CD4 FITC	eBiosciences	clone RM4-5 Cat# 11-0042-82
CD4 APC	eBiosciences	clone GK1.5 Cat# 17-0041-82
CD4 PE	Biolegend	clone GK1.5 Cat# 12-0041-82
CD3 VioletFluor 450	Tonbo	17A2 Cat# 75-0032
CD8 PerCp-Cyanine5.5	eBiosciences	clone 53-6.7 Cat# 45-0081-82
CD62L PE	Biolegend	clone MEL-14 Cat# 104407
CD69 PE/Cyanine 7	TONBO	H1.2F3 Cat# 60-0691-U025
Foxp3 APC	eBiosciences	FJK-16 s Cat# 17-5773-82
Foxp3 PercP	eBiosciences	FJK-16 s Cat# 45-5773-82
IL17 eFIuor 660	eBiosciences	17B7 Cat# 50-7177-82
IL17 PE-Cy7	eBiosciences	17B7 Cat# 25-7177-82
IFN-γ PE	Biolegend	XMG1.2 Cat# 505807
Bacterial and Virus Strains
Esherichia Coli NC101	Balfour Sartor	N/A
Bacteroides fragilis overexpressing OVA	Kubinak et al, 2015	N/A
Biological Samples
Human stool samples	Weill Cornell Medicine	, IND 15988
Human stool samples	University of Utah IBD clinic	IRB_00059476
Chemicals, Peptides, and Recombinant Proteins
HyClone™ RPMI 1640 Media	Thomas Scientific	Cat# C838R34
RBC lysis buffer	TONBO Biosciences	Cat# TNB-4300
Recombinant Mouse GM-CSF (carrier-free)	Biolegend	Cat# 576306
HBSS supplemented with Ca2+ and Mg2+	Corning Cellgro	Cat# 21-023-CV
LB broth	Fisher Bioreagents	Cat# BP1426-500
Chloroform	Fisher Chemical	Cat #C298-500
Cesium Chloride	Fisher BioReagents	Cat #BP210-100
E. coli 0111:b4	Invivogen	Cat# tlrl-eblps
FBS	HyClone	Cat# SH30071.03
Sodium pyruvate	Corning	Cat# 25000CI
Penicillin-Streptomycin Solution	Corning	Cat# 30001CI
L-glutamine Solution	Corning	Cat# 25005CI
MEM Nonessential Amino Acid Solution	Corning	Cat# 25025CI
2-Mercaptoethanol	Sigma Aldrich	Cat# M314825ML
Ionomycin	Sigma Aldrich	Cat#I3909-1ML
PMA	Fisher Biosciences	Cat#BP685-1
Brefeldin A	Biolegend	Cat#420601
Perm/Fix buffer	eBiosciences	Cat# 00512343
Perm/Wash buffer	eBiosciences	Cat# 00833356
Ampicillin	Fisher Scientific	Cat# BP1760-25
Neomycin	Fisher Scientific	Cat# BP266925
Erythromycin	Fisher Scientific	Cat# BP920-25
Gentamycin	GoldBio	Cat# G-400-25
MacConkey Agar	BD Difco	Cat# 212122
Ribozol Rna Extraction Reagent	Quality Biological Inc	Cat# N580200ML
SYBR Green	Thermo-Fisher	Cat# S33102
Phalloidin-AlexaFluor680	Thermo-Fisher	Cat# A22286
Histone 2B-RFP	Thermo-Fisher	Cat# C10595
Dextran sulfate sodium salt	MP Biomedicals	Cat# 0216011010
Critical Commercial Assays
CD4+ T Cell Isolation Kit mouse	Miltenyi Biotech	Cat# 130-104-454
LAL Chromogenic Endotoxin Quantitation Kit	Pierce	Cat#88282
EndoZyme	Hyglos HmbH	Cat#890030
Phage DNA Isolation Kit	Norgen	Cat# 46800
PowerFecal DNA Isolation Kit	Qiagen	Cat# 12830-50
RNeasy Mini Kit	Qiagen	Cat# 74106
qScript™ cDNA SuperMix	QuantaBio	Cat# 101414-106
LightCycler 480 SYBR Green I Master	Roche	Cat# 04707516001
IL-1 beta Mouse ELISA Kit	Invitrogen	Cat# BMS6002
IL-10 Mouse Uncoated ELISA Kit with Plates	Invitrogen	Cat# 88-7105-22
IL-6 Mouse ELISA Kit	Invitrogen	Cat# KMC0061
IP-10 (CXCL10) Mouse ELISA Kit	Invitrogen	Cat# BMS6018
TNF alpha Mouse Uncoated ELISA Kit with Plates	Invitrogen	Cat# 88-7324-22
IL-12/IL-23 p40 (Total) Mouse Uncoated ELISA Kit	Invitrogen	Cat# 88-7120-88
IFN gamma Mouse ELISA Kit	Invitrogen	Cat# KMC4022
IgA Mouse ELISA Kit	Invitrogen	Cat# EMIGA
MagMax Viral RNA Isolation Kit	Thermo Fisher	Cat # AM1939
ChargeSwitch	Thermo Fisher	Cat# CS12000
AccuPrime™ Taq DNA Polymerase System	Thermo Fisher	Cat# 12339016
SuperScript II RT	Thermo Fisher	Cat# 18064014
Deposited Data
Bacteriophage Genomes	Genbank	NC-A MK310182 NC-B MK310183 NC-G MK310184
Experimental Models: Organisms/Strains
Mouse C57BL/6J	Jackson Laboratories	Cat# 000664
Mouse C57BL/6J-Tlr9M7Btlr/Mmjax	MMRRC	Cat# 034329-JAX
Mouse B6N.129S1-Tlr3tm1Flv/J	Jackson Laboratories	Cat# 009675
Mouse MHCII−/−	From Peter Jensen	University of Utah
Mouse B6.129S7-Ifngtm1Ts/J	Jackson Laboratories	Cat# 002287
B6.Cg-Tg(TcraTcrb)425Cbn/J	Jackson Laboratories	Cat# 004194
Oligonucleotides	See Table S1
Software and Algorithms
GENOME PAIR RAPID DOTTER (GEPARD)	http://cube.univie.ac.at/gepard	Krumsiek J, Arnold R, Rattei T. Gepard: A rapid and sensitive tool for creating dotplots on genome scale. Bioinformatics 2007; 23(8): 1026-8. PMID: 17309896
USeq’s (v8.8.8)		http://useq.sourceforge.net/
novoindex (v2.8)		http://www.novocraft.com/
Novoalign (v2.08.03)		http://www.novocraft.com/
Ingenuity Pathway Analysis	Qiagen	www.qiagenbioinformatics.com/products/ingenuity-pathway-analysis/
VirMAP	N/A	Ajami, et al 2018
Graphpad Prism 7	GraphPad Software	N/A
FlowJo LLC v8.7	Becton Dickenson	N/A
Other
Quickseal tubes	Beckmann	Cat# 344059
10K MWCO Snakeskin tubes	ThermoFisher Scientific	Cat #68035
40μm cell strainers	Fisher Scientific	Cat# 22-363-547

Confocal Microscopy.

Purified phage particles were stained with 1:1000 SYBR Green (ThermoFisher) at 4°C for 15 min, followed by an incubatio n with PEG NaCl (5x) in the dark for 1 hour at 4°C to precipitate stained phage particles. Phages were centrifuged at 13,000 RPM for 3 min at 4°C and resuspended in sterile PBS; pellet was wash ed with sterile PBS twice to remove residual dye. BMDCs were isolated and stained with either Phalloidin-AlexaFluor680 (ThermoFisher) or Histone 2B-RFP (Thermo-Fisher) in blocking buffer (2% BSA, 0.1% Triton X-100, 0.1 M phosphate buffer, pH 7.4). Labeled BMDCs were mixed with stained phages at 1:100. Images were acquired using a Zeiss LSM800 confocal microscope.

ELISAs.

IL-1β, IL-10, IL-6, IP-10 and TNFα secretion were measured in cell culture supernatants using ELISA kits (eBioscience) according to the manufacturer’s protocol.

Quantification of phage specific antibodies.

To quantify phage specific IgA antibodies, fecal pellets were collected and resuspended in 500μl sterile HBSS and spun at 2000 RPM for 5 minutes. Supernatants were placed in a new tube and spun again until samples were clear of bacterial pellet. Final supernatants were used as samples for an IgA specific ELISA kit (eBioscience). Phage specific IgA quantification was performed according to kit instructions with slight modification. Instead of coating plates with capture antibody specific for IgA, plates were coated overnight at 4°C with sonicated 10¹² phage particles in HBSS. Absorbance was read at 450nm and normalized to fecal weight.

Dendritic cell – T cell co-culture experiments.

Dendritic cells were isolated as described above and incubated with the phage cocktail at a MOI of 10 or 100 ug/mL OVA for 24 hours and subsequently washed with complete RPMI prior to incubation with T cells. CD4+T cells were isolated from the spleen of C57bl/6 mice and purified using a MACS columns with negative selection according to manufacturer’s instructions (Miltenyi Biotech). 1×10⁵ dendritic cells were incubated with 5×10⁵ CD4+ T cells for 72 hours in a 24 well plate and cytokines were analyzed in the supernatant from these cultures by ELISA according to the manufacturer’s protocols (eBioscience).

B. fragilis mono-association experiments.

Eight-week-old germ free B6 mice were removed from their flexible film isolators and transferred to sterile hepa filtered animal cages. To maintain sterility as described in (Round et al., 2011; Round and Mazmanian, 2010), animals were provided 50μg/mL Erythromycin and Gentamycin in the drinking water supplemented with the NC phage mixture. Concurrently, mice were challenged with 100 μL 10⁷ Bacteroides fragilis OVA expressing strain via oral gavage(Kubinak et al., 2015). Equal colonization of mice with the bacteria was confirmed by plating the fecal samples. Mice were maintained in this condition for 4 weeks to ensure the mucosal immune response to the presence of bacteriophage particles. T cells were then isolated using the Miltenyi CD4 negative selection kit according to manufacturer’s protocol (Miltenyi Biotech) and incubated with BMDCs that were primed for 24 hours with either phage or OVA peptide. Co-cultures were analyzed for IFN-γ secretion by ELISA after 72 hours of co-culture.

DSS colitis experiments.

Eight-week-old SPF C57Bl/6, TLR9^−/−, IFN-γ^−/− mice were provided a mixture of NC phages within the drinking water as described above. Mice were provided 2.5% DSS in the drinking water four weeks later. Animals were monitored for weight loss at the same time of day, each day for seven days. Animals were analyzed for histology and induction of immune responses within the MLN upon necropsy. Disease severity was scored on H&E stained colon samples by a blinded pathologist using the following scoring system: Both crypt loss and inflammation at the site was given its own score according to severity. 0, none; 1, very mild, 2, mild; 3, medium; 4 severe. The amount of colon affected by crypt loss or cellular infiltration was also taken into consideration: 1, 1 to 10%; 2, 10 to 20%; 3, 20 to 40%; 4, 40 to 60%; and 5, 60%-80%, 6, >80% for a maximum score of 20.

Viral nucleic acid sequencing.

Viral nucleic acids were extracted and sequencing libraries consisted of DNA and transcribed RNA. After extraction using the MagMax Viral RNA Isolation Kit (Thermo Fisher Scientific). Viral RNA was reverse transcribed using SuperScript II RT (Thermo Fisher) and random hexamers. After short molecule and random hexamer removal with ChargeSwitch (Thermo Fisher), molecules were amplified and tagged with a BC₁₂-V₈A₂ construct using AccuPrime™ Taq polymerase and cleaned with ChargeSwitch kit. Viral amplicons were normalized, pooled, and made into an Illumina library without shearing. The library (150-600bp) was loaded in an Illumina HiSeq2000 (Illumina, Carlsbad, CA) and sequenced using the 2×100bp chemistry.

Human VLP isolation.

VLPs from fecal samples were purified as described previously (Reyes, et al, 2010) with minor modification. Briefly, 5g of fecal samples were resuspended in 25mL PBS and centrifuged three times at 2500 g for 10 min. Final supernatant was sequentially passed through 40μm, 10μm, 1μm and 0.45μm filters to remove residual cells and byproduct and refilled to 25mL with sterile PBS. 4.14g CsCl was added to the filtrate to make the 1.12 g/ml⁻¹ density solution. 5mL of the filtrate solution was deposited on the top of a 15-ml step gradient prepared using 5 ml CsCl solutions with respective densities of 1.7 g ml⁻¹, 1.5 g ml⁻¹ and 1.35 g ml⁻¹ SM buffer. Samples were centrifuged for 3 hours at 60 000 g at 4C using SW32 swinging bucket rotor (Beckman). 1.5g ml⁻¹ layer was recovered from the tube and dialyzed against PBS overnight.

QUANTIFICATION AND STATISTICAL ANALYSIS

Bacteriophage sequencing analysis.

Phages were classified according to the “cluster” method of Grose and Casjens (2014) where the strength of dot plot diagonal lines indicates the extent of overall genome sequence similarity. Dot plots comparing phage NC-A, NC-B and NC-G genome sequences to related phages were created with the computer program gepard and are shown in parts a, b and c of the figure, respectively. In each plot phage names are shown above and at the left, and colored circles indicate the genus of each phage’s bacterial host. Red stars highlight the positions of phages NC-A, NC-B and NC-G. Thin red lines separate phage genome sequences, and thick red lines separate groups or “subclusters” of highly related phages. Representative phages from each of the subclusters are included in each plot; the phage names are sufficient to retrieve the sequences from GenBank database, and host species are given in the GenBank annotations. Each of the three phages characterized here falls unambiguously into a previously defined phage subcluster; NC-A lies in subcluster A of the T7-like phages, NC-B lies in subcluster A of the øEco32-like phages, and NC-G lies in subcluster C of the T4-like phages. These subclusters correspond approximately to the following genera defined by the International Committee of Taxonomy of Viruses (http://www.ictvonline.org/virustaxonomy.asp): T7virus, PhiEco32virus and Js98virus, respectively.

RNAseq.

RNA was isolated using a Qiagen RNeasy kit (Qiagen) and sequenced by the University of Utah Sequencing Core Facility. Mouse ensemble annotations (build 74) were downloaded and converted to genePred format. Splice junction sequences were generated using USeq’s (v8.8.8). Transcriptome application were made using a radius of 46. The splice junction sequences were added to the mouse chromosome sequences (mm10) and run through novoindex (v2.8) to create the transcriptome index. Reads were aligned to the transcriptome index described above using Novoalign (v2.08.03), allowing up to 50 alignments for each read. USeq’s SamTranscriptomeParser application was used to select the best alignment for each read and convert the coordinates of reads aligning to splices back to genomic space. Read counts for each gene were generated using USeq’s DefinedRegionDifferentialSeq application. These counts were used in DESeq2 to measure the differential expression between each condition. After RNAseq analysis, genes upregulated and downregulated by the treatment were subjected to Ingenuity Pathway Analysis (IPA, Qiagen). The Z scores produced by IPA quantify the bias between the number of significantly upregulated and downregulated genes in each pathway, weighing each gene according to the strength of its association with that pathway based on curated experimental findings. Genes used for this analysis were significantly (p<0.05) up or downregulated at least 1.5 fold.

VFMT virome analysis with VirMAP.

Reads were demultiplexed into a sample bin using the barcode prefixing read-1 and read-2, allowing zero mismatches. Demultiplexed reads were further processed by trimming off barcodes, semi-random primer sequences, and Illumina adapters. This process utilized a custom demultiplexer and the BBDuk algorithm included in BBMap. The resulting trimmed dataset was analyzed using VirMAP. Briefly, VirMAP employs minimum set cover-based clustering algorithm and tiered mapping assembly using protein and nucleotide alignments to create putative viral genomes. VirMAP assigns taxonomies to reconstructed viral genomes using a scoring system that incorporates nucleotide and translated nucleotide alignment results in a per base fashion and optimizes for the highest resolution taxonomic rank with the widest breadth of database support.

Statistical analysis.

Statistical analysis was performed using Graphpad Prism 7. Results represent mean ± SEM and were analyzed by unpaired Students t-test, One-way ANOVA with Mantel Cox or Bonferroni’s Multiple Comparison Test as indicated in the figure legends. Number of animals used for each experiments is described in the corresponding methods or figure legends section. Significance was determined as p<0.05, **p<0.01, ***p<0.005, ****p<0.001

DATA AND SOFTWARE AVAILABILITY

RNAseq data has been deposited in GEO under GSE124123 . The complete genome sequence of phages NC-A accession number MK310182, NC-B accession number MK310183 and NC-G accession number MK310184 is available in the GenBank database.

Highlights.

• Bacteriophages target specific bacteria and mitigate bacterially-driven colon cancer

• Phages activate phage-specific and non-specific IFN-γ mediated immune responses via TLR9

• Phages exacerbate colitis and TLR9/IFNγ blockade abrogates phage-mediated inflammation

• UC patient responses to fecal microbiota therapy correlate with Caurovirales abundance