Experimental model and composition of pooled viral and bacterial inoculums
In order to determine the effects of healthy and UC VLPs on bacterial and DSS colitis severity, we performed in vivo “cross-infection” experiments in HMA mice (Experiments “A–C,” outlined in Fig. 1). Germ-free (GF) mice were first colonized with pooled bacterial communities from 3 healthy volunteers or 3 UC patients (healthy-HMA mice, UC-HMA mice). Following bacterial colonization, mice were given single or multiple doses of VLPs, followed by 2% DSS (Fig. 1). The use of 2% DSS to induce colitis allowed for the temporal control of mild inflammation, thus enabling us to study VLP-mediated effects on bacterial communities both independent of, and in the presence of, intestinal inflammation. In addition, as DSS-induced inflammation is largely restricted to the colon, our model mimics pathology similar to that observed in UC patients [43].
To first characterize the virome of the pooled healthy and UC VLP inoculums given to HMA mice, we performed shotgun sequencing on VLP fractions from fecal samples. Quality-filtered reads from each sample were assembled into scaffolds, and viral scaffolds were detected and annotated (see the “Methods” section, Supplementary Fig. S1). Of the 2063 assembled scaffolds greater than 3 kb in length assembled from healthy and UC inoculum samples, 1791 (86.8%) were classified as viral using our methodology. In total, 718 and 1261 viral scaffolds were found in the pooled healthy and UC VLP inoculums, respectively (Fig. 2A, Supplementary Table S1). In agreement with the reported high inter-individuality of virome samples and differences in virome composition between disease states [30, 37], only 45 scaffolds were shared between these inoculums (Fig. 2A). We also used vConTACT2 to form viral clusters (VCs) based on shared protein-coding genetic content in order to account for the high-inter individuality of these viromes [44]. Using this approach, 523 and 941 VCs (including singletons) were found in the healthy VLP and UC VLP inoculums, respectively (Fig. 2A, Supplementary Table S1). Of these VCs, only 86 were shared between the two pooled inoculums (Fig. 2A), suggesting that a substantial portion of these viral inoculums remained unique at this high taxonomic level. Using a vote-based approach to assign viral taxonomy to the VLP scaffolds [30], we also observed differences in the viral families present in each inoculum (Fig. 2B). The healthy inoculum virome was predominately composed of dsDNA phages belonging to the order Caudovirales and the families Myoviridae and Siphoviridae and unclassified viral scaffolds (Fig. 2B). In contrast, the pooled UC inoculum virome was dominated (77.4% relative abundance) by scaffolds belonging to the (ssDNA) Microviridae family, along with crAss-like phages and phages belonging to the Siphoviridae and Podoviridae families (Fig. 2B). These data are consistent with previous studies, showing that individual gut viromes can be dominated by ssDNA Microviridae [30, 32, 34, 37]. Using the Random Forest Classifier, BACPHLIP, we were also able to classify viral scaffolds in our dataset as temperate [45]. The pooled UC inoculum contained both higher absolute numbers of unique scaffolds identified as temperate and a higher proportion of temperate scaffolds (Fig. 2C, Supplementary Table S1), in line with previous associations between temperate phages and IBD [37].
In addition to these differences in virome composition, we also used 16S rRNA gene sequencing of the V4 region to determine the composition of the pooled healthy and UC bacterial inoculums used to colonize GF mice. Phylum-level analysis revealed decreased relative abundance of Bacteroidetes (2.13% UC, 36.32% healthy), increased Actinobacteria (18.80% UC, 4.63% healthy), and increased Proteobacteria (0.80% UC, 0.15% healthy) in the pooled UC inoculum (Fig. 2D, Supplementary Table S1), consistent with previous observations of UC bacterial communities [46,47,48]. We also compared genus-level differences between the bacterial inoculums with differences in genus-level bacterial host predictions of VLP scaffolds in our dataset using clustered regularly interspaced short palindromic repeats (CRISPR) spacer homology [49]. In total, 196/718 (27.92%) of healthy VLP inoculum scaffolds were successfully assigned CRISPR spacer-based genus predictions and 296/1261 (23.47%) of UC VLP inoculum scaffolds were assigned genus-level predictions (Supplementary Table S1). Some genus-level differences in relative abundance in the pooled bacterial inoculum were consistent with differences in genus-level bacterial host predictions of VLP scaffolds, indicative of concordance between the viral and bacterial fractions of these inoculums. For instance, increased relative abundance of Bifidobacterium in the UC bacterial inoculums (9.87% UC, 1.38% healthy) was consistent with a higher percentage of VLP scaffolds predicted to infect Bifidobacterium (10.47% UC, 3.57% healthy) (Fig. 2E, Supplementary Table S1). Additionally, Prevotella was present at high relative abundances (21.10%) in the healthy bacterial inoculum and was not detected in the UC bacterial inoculum (Supplementary Table S1). This disparity in Prevotella abundance was reflected in a high proportion of Prevotella-infecting VLPs in the healthy inoculum (10.20%) and zero scaffolds in the UC VLP inoculum predicted to infect Prevotella (Fig. 2E, Supplementary Table S1). Interestingly, Bacteroides-infecting VLPs made up 45.61% of UC VLP inoculum scaffolds with predicted CRISPR spacer hosts (Fig. 2E, Supplementary Table S1), despite low Bacteroides relative abundance (1.79%) in the UC bacterial inoculum (Supplementary Table S1), which could be reflective of an expansion of phages targeting and depleting Bacteroides in these UC patients. Together, our data highlight UC-specific alterations in both the viral and bacterial fractions of the pooled fecal samples used for VLP cross-infection experiments in HMA mice.
UC bacterial communities enhance DSS colitis severity in comparison to bacterial communities from healthy controls
To first determine the effects of single healthy and UC VLPs doses on bacterial diversity and DSS colitis severity, we colonized GF mice with the pooled healthy and UC bacterial communities described above (Fig. 2D). After 21 days of bacterial colonization, mice were given a single dose of either healthy or UC VLPs, followed by 2% DSS on day 30 (Fig. 1A). After a single dose of VLPs, we did not observe significant differences in bacterial beta-diversity by weighted UniFrac distance between mice given healthy or UC VLPs in either healthy or UC-HMA mice at any sampling point (Supplementary Tables S2 and S3). Using ANCOM II, a statistical framework that accounts for the underlying structure of microbial communities [50], we were also unable to identify any species that were differentially abundant between mice given healthy and UC VLPs in healthy or UC-HMA mice during the VLP gavage period and DSS/washout periods. Similarly, there were no significant differences in DSS colitis severity in HMA mice given healthy or UC VLPs (Supplementary Fig. S2), suggesting that single doses of VLPs did not alter bacterial community composition or regulate intestinal inflammation.
However, regardless of whether healthy or UC VLPs were administered, UC-HMA mice had increased colitis severity compared to healthy-HMA mice as determined by innate immune cellular infiltration, inflammatory cytokine secretion in colonic explants, and tissue histology (Fig. 3A–F). Overall, these data are consistent with previous studies [51, 52] indicating that gut bacteria from UC patients predisposes HMA mice to an enhanced form of colitis. To determine the differences between the gut bacterial communities in mice humanized with microbial communities from UC patients or healthy volunteers, we performed 16S rRNA gene sequencing on mouse fecal pellets. Principal coordinate analysis (PCoA) on weighted UniFrac distances including all sampling points revealed differences in bacterial beta-diversity between healthy and UC-HMA mice (Fig. 3G). PERMANOVA analyzes revealed significant differences in beta-diversity between these groups at 8/9 of the sampling points sampled (Supplementary Table S4). We next tested for treatment-specific differences in bacterial species previously associated with human IBD or experimental colitis severity. Using ANCOM II, we identified 68/142 species that were differentially abundant between mice given bacterial communities from healthy volunteers and UC patients (ANCOM cutoff W > 0.6, Supplementary Table S5). HMA mice humanized with UC bacteria showed reduced proportions of Akkermansia sp. across all sampling points (Fig. 3H), a bacterial genus typically reduced in IBD patients and shown to ameliorate DSS colitis [12, 53]. It is also well established that bacteria from the Enterobacteriaceae family increase in abundance in IBD and exacerbate experimental colitis severity [13, 14]. Accordingly, we found an expansion of Escherichia-Shigella sp. during DSS colitis, only in mice humanized with UC bacteria (Fig. 3I). Importantly, in the bacterial inoculums used to gavage the HMA mice, there were similar increases in Escherichia-Shigella sp. (0.45% UC, 0.069% healthy) and decreases in Akkermansia sp. (0% UC, 2.14% healthy) (Supplementary Table S1) in the UC samples compared to healthy volunteers. Together, these colonization-specific differences in bacterial taxa may explain the exacerbation of DSS colitis in UC-HMA mice.
Administration of healthy and UC VLPs increases fecal viral abundance and virus-to-bacteria ratio (VBR) in HMA mice
We next wanted to determine whether multiple VLP doses from healthy volunteers could alter the gut bacteriome and prevent the exacerbation of DSS colitis. As we did not observe noticeable changes in bacterial community composition (Supplementary Tables S2 and S3) or DSS colitis severity (Supplementary Fig. S2) following a single dose of healthy or UC VLPs, we explored whether multiple doses of healthy or UC VLPs would alter the gut microbiota of UC-HMA mice (Fig. 1B, C). One common approach in phage therapy to increase treatment efficacy is to add multiples doses instead of one single dose to sustain high phage densities [54]. Thus, for the remaining experiments, we proceeded to give HMA mice multiple repeated doses of VLPs.
In the first experiment (experiment “B,” Fig. 1B), UC-HMA mice were given four doses of healthy VLPs, UC VLPs, or PBS over the course of 9–10 days. Following the VLP gavage period, mice were given 2% DSS followed by a washout period. This first experiment was performed independently twice (trial #1, trial #2). In a second experiment (experiment “C,” Fig. 1C), we tested the impact of phage viability as well as bacteria-independent effects of multiple doses of VLPs on intestinal inflammation using UC-HMA mice administered intact or heat-killed UC VLPs and GF mice given UC VLPs alone (experiment “C,” Fig. 1C). We first determined whether repeated dosing of VLPs would increase viral abundance and virus-to-bacteria ratio (VBR) in HMA mice. Low levels of VLPs (8.12 × 108 virus mL−1) could be detected in HMA mice during the bacterial colonization period (Fig. 4A; experiment “B,” trial #1) before the addition of inoculum VLPs, likely a result of prophage induction of newly colonized bacteria and/or VLPs that we were unable to remove during the separation of the bacterial and VLP fractions in the inoculums. Still, these levels of VLPs during the bacterial colonization period were lower compared to after VLP gavage (3.93-fold increase in UC VLP treated mice and 3.91-fold increase in healthy VLP treated mice) (Fig. 4A). Additionally, regardless of the source (healthy or UC samples), repeating VLP dosing generally increased viral abundance and virus-to-bacteria ratio (VBR) during the VLP gavage period relative to PBS (Fig. 4A–D, Supplementary Fig. 3A–D) or heat-killed UC VLP controls (Supplementary Fig. 4A–D).
Despite the general increase in viral abundance and VBR post-VLP gavage, there was no associated decrease in total bacterial abundance compared to PBS (Fig. 4B, Supplementary Fig. S3B) or to heat-killed controls (Supplementary Fig. S4B), which suggests that bacteria killed by phage-mediated lysis may be rapidly replaced by genetically or phenotypically resistant bacterial taxa [19, 55]. Notably, there was no detectable increase in viral abundance in GF mice given UC VLPs alone, suggesting that phage persistence required bacterial hosts (Supplementary Fig. S4A).
Since some ecological models predict that more metabolically active bacteria are less susceptible to lytic phage infection [56], we next determined if healthy and UC VLPs could alter the proportion of active bacterial cells. Using SYBRGreen I staining and flow cytometry, we determined the proportion of metabolically active fecal bacteria in HMA mice, as described previously [57] (Supplementary Fig. S5A). While not necessarily specific to measuring bacterial growth rate, SYBRGreen I staining is commonly used as a broad marker for bacterial metabolic activity [57, 58]. Compared to PBS and heat-killed UC VLP controls, healthy and UC VLPs did not drastically shift the proportion of active bacteria (Supplementary Fig. S5B–D), except at one sampling point where the proportion of active cells significantly increased in both groups of mice given UC VLPs compared to the heat-killed UC VLP control (Supplementary Fig. S5D). We also used Propidium Iodide (PI) staining to determine the proportion of damaged bacterial cells in HMA mice (Supplementary Fig. S6A). Healthy and UC VLPs also did not dramatically shift the proportion of damaged cells compared to PBS or heat-killed controls (Supplementary Fig. S6B–D), with the exception of a single time-point, where UC VLPs decreased the proportion of damaged bacteria (Supplementary Fig. S6C). Together, these data suggest that donor VLPs can interact with HMA mice but are not able to shift the proportions of active or damaged gut bacterial cells.
Fecal VLPs derived from healthy volunteers and UC patients drive unique changes in the virome of HMA mice
Shotgun sequencing was also performed on the VLP fraction of mouse fecal pellets in experiment “B” (trial #1) to follow changes in virome composition of HMA mice following healthy or UC VLP gavage. Of the 6742 scaffolds greater than 3 kb in length assembled from mouse samples, 4219 (62.6%) were identified as viral using our methodology (Supplementary Fig. S1). To assess the effectiveness of the transfer of VLPs from the pooled inoculums to the HMA mice, we determined the proportion of scaffolds found in the VLP inoculums that were only found in HMA mice post-VLP gavage. In total, 76/718 (10.58%) of healthy VLP inoculum scaffolds were also found in HMA mice post-healthy VLP gavage and not pre-VLP gavage (Supplementary Table S6). In contrast, a higher number and proportion of UC VLP inoculum scaffolds, reaching 203/1261 (16.10%), were found in HMA mice post-UC VLP gavage and not found pre-VLP gavage, possibly reflective of autologous transfer of UC VLPs to UC-HMA mice (Supplementary Table S6). In line with these data, post-VLP gavage, we observed increased richness at the scaffold and VC level in HMA-mice given healthy and UC VLPs compared to the PBS control (Supplementary Fig. S7). To further assess the effectiveness of VLP transfer, we measured the Jaccard distance of VCs between each inoculum and HMA samples before and after transfer to HMA mice. While not significant, there was a reduction in Jaccard distance between the viromes of pooled healthy inoculum and the UC-HMA mice given healthy VLPs, and between the pooled UC VLP inoculum and UC-HMA mice given UC VLPs (Supplementary Fig. S8), suggesting increased virome similarity to the viral inoculum over time.
We next performed NMDS on Bray-Curtis dissimilarity of viral scaffolds and VCs to follow differences in viral beta-diversity over time in UC-HMA mice. Grouping sampling points together within each time period, there were no significant differences in Bray-Curtis dissimilarity during the bacterial colonization period at the scaffold or VC level by PERMANOVA (Fig. 5A, Supplementary Fig. S9). However, after VLP gavage, VLP treatment had a significant effect on the viral scaffold and VC diversity, further indicating successful transfer and replication of donor VLPs (Fig. 5A, Supplementary Fig. S9). Interestingly, the effect size of VLP treatment was greatest during the DSS/washout period, suggesting heightened virome divergence after inflammation (Fig. 5A, Supplementary Fig. S9). In agreement with these data, VLP treatment had a significant effect on Bray-Curtis dissimilarity by PERMANOVA, at each sampling point following VLP gavage (Supplementary Table S7). It should also be noted that there was a close to significant difference in Bray-Curtis dissimilarity before VLP gavage (viral scaffolds: p = 0.053, viral clusters: p = 0.064), which could indicate that isolator effects during viral colonization may have influenced the virome composition in these mice (Supplementary Table S7).
At the viral family level, following VLP gavage, there was an expansion of Myoviridae in mice given healthy (7.20-fold increase) and UC VLPs (3.80-fold increase), which persisted after DSS exposure (Fig. 5B). Interestingly, we also observed a large increase in crAss-like phage relative abundance (21.61-fold) in the DSS/washout period only in mice given UC VLPs (Fig. 5B). To further assess differences between VLP-treated mice and PBS controls, we used DESeq2 to determine viral scaffolds that significantly changed after VLP gavage between treatment groups, while controlling for repeated sampling [59]. In total, there were 332 differentially abundant viral scaffolds between UC VLP-treated mice controls and PBS controls (260 over-abundant, 72 under-abundant) and 262 differentially abundant scaffolds between healthy VLP-treated mice and PBS controls (250 over-abundant, 12 under-abundant). Of these significantly upregulated scaffolds, 50/260 in UC VLP-treated mice and 18/250 in healthy VLP-treated mice were also found in their respective pooled inoculums. Of all the over-abundant scaffolds, mice given healthy or UC VLPs were both enriched with phages predicted to infect Bacteroides and Eubacterium phages (Supplementary Table S8). However, only UC VLP-treated mice were enriched with crAss-like phages and phages predicted to infect Erysipelatoclostridium (Supplementary Table S8). In contrast, healthy VLP-treated mice were enriched with Myoviridae phages and phages predicted to infect Parabacteroides (Supplementary Table S8). Together, these data indicate that VLPs were transferred from the pooled inoculums to HMA mice, with distinct changes in viral beta-diversity, phage families, and potential phage-bacterial interactions following VLP gavage.
VLPs derived from healthy individuals or UC patients have distinct effects on the UC gut bacterial communities in vivo
In order to identify whether multiple healthy and UC VLPs doses had distinct effects on bacterial community composition, we performed 16S rRNA gene sequencing on mouse fecal pellets before and after VLP administration, as well as during the DSS/washout period. Using ANCOM II, we identified bacterial species differentially abundant between UC-HMA mice given multiple doses of either UC VLPs, healthy VLPs, or PBS (experiment “B,” summarized in Supplementary Tables S9 and S10). To account for variation due to isolator-specific differences during colonization, bacterial species that were differentially abundant during the baseline bacterial colonization period were not included. In addition, the remaining differentially abundant species were each ranked based on the likelihood that treatment-specific differences were due to VLP treatment or isolator effect (1: likely due to VLP treatment, 2: possibly due to VLP treatment, 3: likely due to isolator effect, see methods for ranking criteria).
Across the two trials where UC-HMA mice were given multiple doses of healthy VLPs, UC VLPs, or PBS, 2 differentially abundant species (trial #1; Anaerotruncus sp., trial #2; Eubacterium] fissicatena group sp.) were identified in the VLP gavage period before DSS administration, and 12 species (3 species trial #1; 9 species trial #2) were identified in the DSS/washout period (ANCOM cutoff W > 0.6, Supplementary Table S9). Between mice given multiple doses of UC VLPs (+/− DSS) or heat-killed UC VLPs (experiment C), 1 differentially abundant bacterial species was identified in the VLP gavage period (Blautia hydrogenotrophica), and 4 differentially abundant species were identified in the DSS/washout period (Supplementary Table S10). Some of these differentially abundant bacterial species have been shown to influence experimental colitis severity or to be differentially abundant in CD or UC patients. For example, Eubacterium limosum was significantly reduced in mice given UC VLPs (Fig. 6A, left), consistent with its ability in mouse models to ameliorate DSS colitis severity [60]. In addition, compared to heat-killed UC VLPs, mice given intact UC VLPs showed a significant increase in the proportion of Escherichia-Shigella sp. after VLP gavage (Fig. 6A, right). Given that Enterobacteriaceae are increased in IBD patients and are thought to exacerbate dysregulated immune responses in IBD [7, 13, 61], these data suggest that UC VLPs may allow for the expansion of this species.
Grouping samples within time periods together, there were no significant differences in bacterial beta-diversity of HMA mice based on the VLP treatment group during the bacterial colonization period and during the following 9-day period where 4 doses of VLPs (healthy, UC, or PBS) were given to these mice (Fig. 6B). In contrast, we observed significant differences in bacterial community composition between treatment groups, and the largest effect size following DSS administration, supporting the idea that these phages have distinct effects on bacterial communities, which are amplified during DSS colitis (Fig. 6B). Similar trends were observed in a second independent trial (Supplementary Fig. S10) and between UC-VLP treated mice and heat-killed controls (Supplementary Fig. S11). In line with these data, performing PERMANOVA analyses on each sampling point in each experiment revealed that significant differences between treatment groups and the largest effect sizes were only observed during the DSS/washout period (Supplementary Tables S11–S13).
In order to investigate whether these greater changes in bacterial diversity following DSS administration were in part due to prophage induction as a result of intestinal inflammation, we also determined the proportion of VLP scaffolds identified as temperate following DSS. Within each treatment group, there was only a modest increase in the relative abundance of VLP scaffolds identified as temperate following DSS (Supplementary Fig. S12). Additionally, to determine whether there was experimental evidence for prophage induction due to direct interactions between DSS and UC bacterial communities, we performed an in vitro prophage induction assay [62,63,64]. In UC bacterial cultures grown anaerobically in the presence of DSS, we did not detect an increase in VLPs in the bacterial supernatant (Supplementary Fig. S13), suggesting that any DSS-mediated prophage induction occurring in HMA mice is likely not due to direct DSS toxicity to bacterial cells and is likely through DSS-mediated induction of downstream immune responses.
Together, these data indicate that healthy and UC VLPs drive distinct changes in the relative abundance of bacterial species in UC-HMA mice, some of which have been shown to influence experimental colitis severity and IBD disease progression. In addition, while some bacterial species were found to be differentially abundant during the VLP gavage period, differences in bacterial diversity between VLP-treated mice are greatest during DSS colitis, suggesting that intestinal inflammation may provide a more conducive environment for phage-mediated changes in these gut bacterial communities.
UC VLPs increase colitis severity in HMA mice
Given that the healthy and UC VLPs are distinct in their composition and contribute differently to bacterial community composition, we next tested if they had distinct effects on colitis outcomes. We first determined whether there were differences in colitis severity between UC-HMA mice given multiple doses of healthy VLPs, UC VLPs, or PBS. Upon DSS administration, a slight decrease of body weight could be seen only in mice administered UC VLPs (Fig. 7A). Importantly, at day 10 post-DSS challenge, only the weight of UC-HMA mice given UC VLPs was declining (Fig. 7A). HMA mice given UC VLPs also showed a significant shortening of colon length (Fig. 7B) and an increase of pro-inflammatory cytokine production (Fig. 7C), indicating that UC VLPs can exacerbate the severity of DSS colitis compared to healthy VLPs and the PBS control.
To further investigate the function of UC VLPs on the pathology of DSS-induced colitis, we next assessed DSS colitis severity between UC-HMA mice given (i) UC VLPs with DSS; (ii) UC VLPs without DSS to determine if UC VLPs can cause spontaneous colitis; (iii) heat-killed UC VLPs to determine if intact VLPs are necessary for increased colitis severity; and (iv) GF mice given UC VLPs to determine if increased colitis severity was a result of direct UC VLP-immune interactions (Fig. 7C).
Consistent with our previous results (Fig. 7C), UC-HMA mice given UC VLPs had a significantly shortened colon and produced the most TNF-⍺ and IL-1β from colon explants compared to the other UC-HMA groups (Fig. 7D–F). Notably, the heat-killed UC VLP group mice showed significantly less severe colitis compared to the intact UC VLPs group, suggesting the changes we observe are due to viral infection rather than the presence of phage components alone (Fig. 7D–F). Finally, we found that GF mice given UC VLPs prior to DSS treatment exhibited the greatest colon shortening, yet least inflammatory cytokine production (Fig. 7D–F). These results are similar to our observations of increased colonic shortening in GF mice compared to HMA mice (Supplementary Fig. S14) and suggest that, without bacterial hosts, phages cannot modulate colitis outcomes.