Single cell viral tagging of Faecalibacterium prausnitzii reveals rare bacteriophages omitted by other techniques

ABSTRACT

The associations of the gut microbiome and virome with human health and disease are increasingly numerous and clear. The mechanistic roles of bacteriophages (phages) in the microbiome, however, are especially unclear, as their cultivation is exceedingly difficult and their diversity so immense. We use viral tagging (VT), a technique wherein fluorescently stained uncultivated viruses are allowed to adsorb to host cells and then host cells are singly sorted. This method identifies interacting phage-bacteria pairs to better sample and characterize the phages in human stool samples from healthy and inflammatory bowel disease (IBD)-affected patients. First, we apply VT to uncultivated bacteria from a healthy human sample, demonstrating far-reaching ability to observe diverse bacteria and phages alike. We also use VT with a cultured Faecalibacterium prausnitzii isolate, a bacterial host of interest due to its anti-inflammatory effects and strong negative correlation with IBD. Comparing VT with virome sequencing and phage identification from single amplified genomes shows that it is a practical technique for phage discovery, especially when it is used to focus on individual bacterial cultivars for which genomes have been sequenced. VT can detect phages so rare as to be undetectable in standard virome sequencing, which is biased toward the most abundant phage species even at high sequencing depth. Remarkably, VT also identified novel prophage integration events in F. prausnitzii, demonstrating that VT interactions can extend beyond the level of surface attachment and constitute active infection events. In total, VT identified at least 328 unique and highly diverse phage-host pairs, almost all of which are entirely uncharacterized, and several phages that are differentially abundant in IBD patients compared to healthy controls. Taken together, we show that VT is an extremely powerful tool to move beyond the cultivation and abundance biases inherent to current techniques and suggest that the phage-host pairs identified by VT here are crucial first step to enable future mechanistic studies of phage-bacteria-human interactions.

Graphical Abstract

Introduction

The human gut is colonized by an immensely diverse microbiota, whose most abundant and diverse components are the bacteriome and the virome, with the latter primarily composed of bacteriophages (phages). Bacterial dysbiosis of the microbiota is linked to an increasingly long list of human disease states and is especially well-documented for inflammatory bowel diseases (IBD),¹ an emerging worldwide health challenge characterized by inflammation of the gastrointestinal tract. The IBD bacteriome exhibits increased abundances of pathobionts while symbiotic and commensal bacteria are depleted, though studies disagree on the correlations of many specific bacterial taxa with IBD and health.^2,3

Intriguingly, increased or altered phage populations in the gut have been observed to correlate with flares of intestinal inflammation in IBD patients.⁴ In Crohn’s disease (CD), one type of IBD, the phageome, the totality of phages in the gut, shifts away from core populations of strictly lytic phages, including CrAssphages and microviruses, and toward expanded abundances of diverse lysogenic phages.^4–7 Like most phage-health associations, however, the links between IBD and the phageome are observational in nature,^8,9 leaving the mechanisms driving them poorly understood. Several health-relevant effects of phage–host interactions have been described to various extents, such as lytic phage infection depleting pathogenic hosts,¹⁰ integrated prophages altering bacterial metabolism of secreted bile acids,¹¹ and phage particles altering the physical structure of bacterial communities in the lungs of cystic fibrosis patients.¹²

For gut-associated phages in particular, there is a severe lack of known bacterial hosts. Identifying hosts for the incredible diversity of gut-associated phages is a critical first step to better understand the phageome in general and to develop the experimental systems required to build mechanistic understanding of the complex phage-bacteria-human interactions. Given the complexity of this challenge, both phage culture and current ‘omics methods fall short of defining of phage-host pairs on the broad scale that the gut microbiome requires. Culture of gut-associated phages, usually by plaque assays, is not a practical method to capture the breadth of phage diversity in the gut, and further hinges upon the compatibility of specific bacterial hosts, suitable culture conditions for infection, and phage lytic activity such that phage infection clears bacterial hosts. Metagenome sequencing can detect integrated prophages in bacterial genomes, thus inferring the phage–host relationship, though it neglects non-integrated phages. Conversely, virome sequencing approaches do not detect integrated or intracellular phage genomes, and cannot directly identify hosts for the phages they observe. CRISPR spacers, tRNAs, and altered codon usage can be used to predict phage hosts post hoc. CRISPRs, however, are not present in all bacteria, the number of spacers in an array is limited, and spacer matching can be nonspecific.¹³ Meanwhile, the presence of host-related tRNAs and codon biases are not very host-specific.¹⁴ Further, all ‘omics techniques, and especially those with amplification steps, are biased toward the most abundant viral or bacterial species, even at high sequencing depth.^15,16 Finally, a growing number of techniques can capture phages co-localized with host cells, such as droplet PCR and Hi-C, but phage-host linkage events are spurious without an enrichment step for phage-associated cells.^17,18 Each approach to identifying phage-host pairs is based on the detection of different stages of the phage life cycle, such as lysis, integration, and adhesion.¹⁹ Viral tagging (VT) is an approach based on single-cell sorting of bacterial cells with attached phage particles, thus identifying phage hosts within the phage’s adsorptive host range.²⁰ Though these phage-host pairs are less specific than those defined by other methods, they are more efficient than culture-based methods and capture phage–host interactions missed by other techniques.

VT has been applied to multiple microbial systems, including the human gut.^21–23 Briefly, viral preparations from stool are fluorescently stained and combined with one or more bacteria, to allow phages to bind to their hosts. After a series of washes to remove unbound or loosely bound phages from cells, tagged cells are singly sorted, thus enriching phage-host pairs, prior to sequencing the phage and host genomes together. By virtue of sequencing single cells with few associated phages, the complexity of sequence assembly is limited, and bacterial host genome sequence is minimized. VT can thus more broadly and efficiently identify phage-host pairs than prior techniques.

One common and abundant member of the healthy gut microbiome, the genus Faecalibacterium, is consistently depleted in IBD patients, especially in those with CD.^24,25 F. prausnitzii produces at least three anti-inflammatory compounds: the short-chain fatty acid butyrate, a secreted microbial anti-inflammatory protein, and extracellular matrix, while consuming the inflammatory gas hydrogen sulfide.^26–29 Additionally, F. prausnitzii was recently approved for human clinical trials as a CD treatment.³⁰ Faecalibacterium genomes are highly plastic and contain a high proportion of prophage and other integrated mobile content.³¹ Few Faecalibacterium phages have been characterized, an endeavor limited by the difficulty of its cultivation. Of the few phages identified for F. prausnitzii, all are lysogenic siphoviruses (tailed phages with flexible tail morphology) induced from a handful of cultured host isolates.^32,33 This lack of diversity of Faecalibacterium phages highlights a need for further sampling of phages infecting this keystone species.

Here, we employ VT with uncultivated bacterial hosts from human stool, and with a single F. prausnitzii isolate, as well as phages predicted from single amplified genomes (SAGs) of individual bacterial cells and metaviromes from the same human samples. A comparison of these four methods shows that VT and phage prediction from SAGs are as efficient on a per-read basis in identifying phage sequence as traditional virome sequencing. Further, they have a substantial advantage over viromes by directly linking these phages to their bacterial hosts. From our limited single-cell survey, we identified an enormous diversity of 1758 viral contigs belonging to 1014 vOTUs and 773 viral clusters (VCs), from at least 21 host genera, totaling 328 unique vOTU-host genus pairs. VT of F. prausnitzii further identified several novel phages related to preexisting phage-like regions of the host genome and two newly integrated prophages. Most phages identified during VT are so rare as to not be detectable in viromes, highlighting the utility of VT to more deeply explore the gut virome. Host range analyses enabled by VT and SAG analysis suggest that several of the phage groups observed here have broader host ranges than what is traditionally observed by phage culture alone. The phages identified here, several of which are significantly associated with IBD or health, encompass remarkable viral diversity, spanning morphologies, taxa, and lifestyles.

Methods

Stool sample & metadata collection

Human stool samples were obtained from a cohort of longitudinally sampled Crohn’s disease (CD) patients (n = 8) and matched healthy household control (HHC) individuals (n = 8) (Table S1), collected at Addenbrooke’s Hospital, University of Cambridge, UK between 2016 and 2019. One to two samples from each individual were used, for a total of 9 CD and 10 HHC samples. Of the 9 CD stool samples, six were collected at the time of active inflammation as indicated by fecal calprotectin (≥600 μg/mg) and/or serum C-reactive protein (≥10 mg/L) levels (Table S1). Stool samples were stored at −80°C.

Isolation of VLP & bacterial fractions from human stool

Stool samples were homogenized in a 6X volume of PBS by vortexing for 10 min. Homogenates were incubated at room temperature for 10 min to allow debris to settle. The upper fraction was removed and centrifuged for 10 min at 7,000 xg to pellet uncultivated cells. Supernatants were removed and sequentially filtered through 0.45 μm and 0.22 μm filters to obtain the VLP fraction. Cell pellets were resuspended in an equal volume of PBS. For FpVT equal volumes of VLPs were pooled within the CD and HHC groups.

Cultivation, sequencing, & genome analysis of F. prausnitzii strain 22

F. prausnitzii strain 22 was isolated from the stool of a healthy individual as part of the Microbiome and Asthma Research Study³⁴. F. prausnitzii strain 22 was grown in an anaerobic chamber in Modified Reinforced Clostridial Broth (ATCC Medium 2107) at 37°C.

A single contig reference genome was generated for F. prausnitzii strain 22 by hybrid assembly of Illumina short read and Oxford Nanopore long read sequencing data. Cells were lysed by homogenizing 1 mL of an overnight liquid culture of F. prausnitzii 22 by bead beating with approximately 200 µL of 1 mm zirconia/silica beads (BioSpec 11079110Z) and beaten for 1 min at 3500 oscillations per minute in a BioSpec bead beater. DNA was extracted by washing twice with an equal volume of phenol:chloroform and then cleaned with the DNeasy Blood and Tissue Kit (Qiagen 69,504).

Half the DNA volume was shotgun sequenced after tagmentation using the Nextera DNA Library Preparation kit (Illumina) as previously described.³⁵ This was followed by PCR-mediated adapter ligation using KAPA HiFi PCR master mix (Roche) to uniquely barcode each sample. DNA libraries were pooled and purified using AMPure XP magnetic beads according to the manufacturer’s instructions to select for approximately 200bp size. Short-read sequencing was performed on the Illumina NextSeq platform using a paired end 2 × 150 protocol at the DNA Sequencing Innovation Lab (Center for Genome Sciences and Systems Biology, Washington University School of Medicine in St. Louis).

The other half of the DNA was cleaned again with the Genomic-tip 100/G DNA Purification Kit (Qiagen 10,243), run on a 0.8% TBE gel to determine approximate fragment size, and prepared for Oxford Nanopore sequencing with the Oxford Nanopore EXP-NBD104 native barcoding kit and the Oxford Nanopore SQK-LSK109 sequencing kit. Long-read sequencing was performed on a MinION sequencer with a MinION flowcell (Oxford Nanopore Technologies). Long-read sequences were processed with Guppy v2.3.1 and hybrid genome assembly of long and shotgun reads was performed with Unicycler v0.4.7.³⁶

Faecalibacterium phylogeny was constructed with UBCG2 using default settings³⁷ and visualized with FastTree v2.1.³⁸ Rnammer v1.2³⁹ was used to extract 16S rDNA sequences.

PHASTER⁴⁰ and VirSorter2 v2.2.4⁴¹ were used with default parameters to identify any integrated prophage sequence in the F. prausnitzii strain 22 genome, which were then assessed for quality and completeness with CheckV v0.7.0. Prophage predictions overlapping with each other ≤5 kb apart were merged into regions of interest, annotated for protein coding genes with Prodigal v2.6.3⁴² using default parameters, and those proteins were compared to the NCBI viral protein database (retrieved August 24, 2021) with BLASTp v2.16.0 using default parameters. Prophage regions were aligned with the NCBI viral nucleotide database by BLASTn v2.16.0 and annotations were further manually examined to determine prophage completeness. Functional annotation of F. prausnitzii genomic regions adjacent to integrated prophages was performed using Prokka v1.14.5.⁴³

Viral tagging

Uncultivated cells from human stool (UnVT) and cultured F. prausnitzii cells (FpVT) were each normalized to an optical density of 1.0 with PBS. Cells were stained with a final concentration of 0.5 µM SYTO-9 green fluorescent nucleic acid stain (Invitrogen) and incubated at room temperature in the dark for 30 min prior to being washed three times with PBS by centrifugation at 5,000 xg to remove unbound stain. VLPs were stained with a final concentration of 2.5 µM SYTO-62 red fluorescent nucleic acid stain (Invitrogen) and incubated at room temperature in the dark for 30 min prior to being washed three times by precipitation to remove unbound stain. VLP precipitation was accomplished by adding polyethylene glycol-6000 and sodium chloride to final concentrations of 4% and 0.5 M, respectively, incubating on ice in the dark for 30 min, and centrifuging for 15 min at 21,000 xg at 4°C. Stained cells were stored in PBS with 20% glycerol at −80°C. Stained VLPs were stored in PBS at −80°C.

Prior to VT, stained cells and VLPs were thawed on ice in the dark. Cells were first centrifuged at 5,000 xg and resuspended in an equal volume of PBS to remove glycerol. VT was accomplished by combining equal volumes of stained VLPs and stained cells and incubating on ice in the dark for 20 min. To remove unbound VLPs, cells were washed three times by centrifugation at 5,000 xg at 4°C and resuspended in an equal volume of PBS. Aliquots of unstained and SYTO9-stained cells without added VLPs were prepared similarly as flow cytometry controls. Prior to FACS, all samples were kept on ice, protected from light, and passed through a cell strainer.

Flow cytometry & cell sorting

Single cells were analyzed and sorted on a BD FACSAria II-3 operated by the Flow Cytometry and Fluorescence Activated Cell Sorting Core (Department of Pathology and Immunology, Washington University School of Medicine in St. Louis) with a 70 µm nozzle at 70 psi. The flow rate was selected to achieve approximately 10,000 events/s and samples were diluted with PBS when necessary. The cell sorter was washed between each sample.

The gating strategy is visualized in Figure S1. Negative controls were processed first to gate for bacterial cells, single cells, and SYTO 9-stained cells (FITC area), as well as define the upper limit of autofluorescence in the PE channel for cells not tagged with VLPs. Bacterial cells were gated on forward scatter area (FSC-A) and side scatter area logarithmic-scale bi-plots. Single cells were gated on FSC-A and FSC-height logarithmic-scale bi-plots. SYTO9-stained single cells were gated on FSC-A and FITC area logarithmic-scale bi-plots. The upper limit for autofluorescence was defined using PE area histograms with SYTO9-stained cells in the absence of VLPs. During both UnVT and FpVT, the 10% most fluorescent SYTO9 positive single cells in the PE channel (SYTO62 VLP stain) were sorted into 96 well plates pre-loaded with 8 µL of TE buffer. The first column (8 wells) of each plate was not used for sorting and served as negative controls for 16S sequencing. Sorted cells were transported on dry ice and stored at −80°C prior to DNA isolation.

Single-cell sorting for the 16S survey (sc16S survey) and subsequent SAG sequencing were performed and reported as part of Lawrence et al.

Single cell DNA isolation & amplification

DNA isolation and amplification was performed using the method of Lawrence, et al. After sorting, 1 µL of lysis solution (0.4 M potassium hydroxide, 10 mM EDTA) was added to each well and lysed by incubating at 30°C for 30 min. Lysis was stopped with 1 µL of 10 mM Tris (pH 4.0).

DNA was amplified by multiple displacement amplification (MDA). All reagents except for the Phi29 polymerase and primers were treated with UV light for 10 min in a Stratalinker 2400 prior to creating the master mix. First, 5.2 µL of DNase/RNase free water (Fisher Scientific AM9935), 2 µL of NEB 10X Buffer (NEB B0269S), 1 µL of Exo-resistant random hexamers (MCLAB ERRP-110), 0.5 µL of Phi29 DNA polymerase (NEB M0269L), and 0.40 µL of NEB BSA (B9000S, resuspended to 10 mg/mL) per sample were combined as a master mix and incubated at 30°C for 30 min to degrade any contaminating DNA via the Phi29 exonuclease activity. To the master mix, 0.8 µL of 10 mM dNTPs (NEB N0447L) per sample was added and mixed, inhibiting further exonuclease activity. Finally, 10 µL of master mix was added to each lysed and neutralized sample and mixed. DNA was amplified at 30°C for 12 hr and the reaction stopped by incubating at 65°C for 10 min. After amplification, 40 µL of TE buffer was added to each reaction, incubated at 30°C for 5 min, and DNA was quantified with the Qubit dsDNA HS Assay Kit (Invitrogen Q32854).

16S rRNA gene sequencing & analysis

16S rRNA gene sequencing was performed for the V4 region using barcoded 515F and 806 R primers, as reported previously.⁴⁴ Each PCR reaction contained 2.5 μL of 10X High Fidelity PCR Buffer (Invitrogen), 0.5 μL of 10 mM dNTPs, 1 μL of 50 mM MgSO₄, 0.5 μL of each of the forward and reverse primers (10 μM final concentration), 0.1 μL Platinum High Fidelity Taq (Invitrogen), and 1.0 μL of amplified sample DNA. Reactions were run at 94°C for 2 min, followed by 30 cycles at 94°C for 15 s, 50°C for 30 s, and 68°C for 30 s; and a final extension for 2 min at 68°C. Uniquely barcoded amplicons were pooled and purified with 0.6x Agencourt Ampure XP beads (Beckman Coulter) according to the manufacturer’s instructions. Sequencing was performed at the DNA Sequencing Innovation Lab (Edison Family Center for Genome Sciences and Systems Biology, Washington University School of Medicine in St. Louis) with the 2x250bp protocol on the Illumina MiSeq platform.

The 16S rRNA V4 region from UnVT, FpVT, and SAG cells was accomplished as previously described,⁴⁵ except taxonomy was assigned using the GTDB release 202 database.⁴⁶ The 16S rDNA profile and SAGs published in Lawrence et al., were also reanalyzed using the GTDB release 202 database. For UnVT sorted cells, samples lacking a 100% match to an ASV from the 16S profile were removed. To remove wells with more than one sorted cell or contaminated with a significant amount of other bacterial DNA, only samples with at least 70% of the same ASV were further sequenced and analyzed. For virus-tagged F. prausnitzii strain 22, only samples with a 100% match to one of the F. prausnitzii 22 16S V4 sequences for at least 70% of the attributable ASVs were further sequenced and analyzed.

Single cell shotgun sequencing

MDA-amplified DNA from virus-tagged single cells passing 16S rRNA cutoffs was used for shotgun sequencing using the same method as described for Cultivation, sequencing, & genome analysis of F. prausnitzii strain 22.

Virome sequencing

Virome preparation and sequencing was performed according to a previously reported protocol⁴⁷ with some modifications. Approximately 100–200 mg of frozen stool were resuspended in buffer, centrifuged, and the supernatants filtered through 0.45 μm filters. Filtered supernatants were treated with lysozyme to liberate bacterial nucleic acid followed by DNase treatment to remove non-encapsidated viral nucleic acid. Total nucleic acid (both RNA and DNA) was extracted on a COBAS AmpliPrep instrument (Roche) or MagNa Pure (Roche) kit according to the manufacturer's recommendations. Purified total nucleic acid was reverse-transcribed and PCR amplified using barcoded primers consisting of a base-balanced 16 nucleotide-specific sequence and used for NEBNext library construction (New England BioLabs). Libraries were multiplexed (12 samples per flow-cell) on an Illumina MiSeq instrument (DNA Sequencing Innovation Laboratory at the Edison Family Center for Genome Sciences, Washington University School of Medicine) using the paired-end 2 × 250 protocol.

UnVT shotgun sequencing analysis

Bioinformatic analysis was carried out by first checking read quality using FastQC v0.12.1 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Adapters and low-quality reads were removed using Trimmomatic v0.39 with the ‘ILLUMINACLIP:adapters.fna:2:30:10:2:True SLIDINGWINDOW:4:20 MINLEN:36’ parameters.⁴⁸ Trimmed reads were mapped PhiX (NC_001422.1) and human genomes (NC_000001.11) using Bowtie2 v2.4.2 (–very-sensitive) and SAMtools v1.15.1.^49,50 Read pairs aligned with PhiX or human sequence were discarded. The remaining reads were assembled on an individual cell basis into contigs using SPAdes with the ‘–sc’ flag.^51,52

Contigs with a length ≥1kb were processed to identify viral contigs using VirSorter2 v2.2.4 (–keep-original-seq – include-groups dsDNAphage, ssDNA – min-length 1000 –min-score 0.5),⁴¹ Cenote-Taker2 v2.1.5 (–prune_prophage False – filter_out_plasmids True – virus_domain_db standard – circ_minimum_hallmark_genes 1 –lin_minimum_hallmark_genes 1),⁵³ and geNomad v1.7.1 (–enable-score-calibration – conservative – sensitivity 7.5).⁵⁴ Contigs were considered viral if they were identified by at least one of these methods. CheckV v1.0.1 was used to evaluate the quality of the identified viral contigs.⁵⁵ The host regions from the edges of contigs that were identified as proviruses by geNomad or CheckV were removed, and only proviruses with a length ≥1kb after removing the host regions were kept for further analysis.

FpVT shotgun sequencing analysis

Read trimming and filtering was performed as was done for UnVT, except after removing PhiX and human contamination, the reads were subsequently aligned to the F. prausnitzii strain 22 genome and categorized into prophage and non-prophage regions (Table S5) using Bowtie2 v2.4.2 (–very-sensitive) and SAMtools v1.15.1.^49,50 Reads that matched F. prausnitzii strain 22 prophage regions with 100% identity and those aligned with the non-prophage regions with an identity of ≥98% were removed using the filter mode from CoverM v0.6.1 (https://github.com/wwood/CoverM). Filtered BAM files from prophage and non-prophage regions were merged and subsequently converted to FASTQ files using SAMtools v1.15.1.⁵⁰ The unmapped singleton reads were extracted using the reformat.sh script from BBMap v39.01. The reads were assembled on an individual cell basis as was done for UnVT.

Contigs with a length ≥1 kb were used to conduct a BLASTn v2.14.0 search against the F. prausnitzii strain 22 genome. ANI and AF were calculated from BLASTn results using the anicalc.py script from CheckV v1.0.155. Contigs aligned to the F. prausnitzii strain 22 genome with ANI ≥95% and AF ≥ 85% were removed. Contigs aligned by partial or fragmented BLASTn hits to the F. prausnitzii strain 22 genome were manually screened for genomic rearrangements, including transposon movement, which were also removed. Remaining contigs containing partial BLASTn hits to the F. prausnitzii strain 22 genome with 99%–100% identity and adjacent to novel sequence were considered newly integrated prophages, trimmed of F. prausnitzii genome sequence, and those longer than 1 kb were included in downstream phage contig analyses.

SAG and virome analysis

Single-cell sequencing analysis was carried out as was done for UnVT. The selected viral contigs were subsequently aligned to the PhiX and human genomes, respectively, using BLASTn v2.14.0 with with e-value cutoff 1e-3. ANI and AF were calculated as described above. One contig that aligned to the human genome with ANI ≥ 95% and AF ≥ 75% was removed from further analysis. Metavirome analysis was performed by first assessing reads quality using FastQC v0.12.1 and removing adapters, primers, low quality reads, and contaminants using Hecatomb’s preprocessing module (version 1.2.0) with the parameter “–library roundAB”.⁵⁶

Cutadapt v4.7 was then used to remove additional internal primer sequences.⁵⁷ The trimmed reads were individually assembled and co-assembled into contigs by condition using metaSPAdes v3.15.5.⁵⁸ Contigs with a length ≥1kb were processed to identify viral contigs as was done for UnVT.

Viral genomics analysis

Identified viral contigs were clustered into species-level virus operational taxonomic units (vOTUs) at 95% average nucleotide identity (ANI) over at least 85% alignment fraction (AF) using the scripts anicalc.py and aniclust.py from CheckV.⁵⁵ The longest contig from each cluster was selected as the representative sequence. Approximately genus-level clusters were defined by nucleotide similarity using reciprocal BLASTn v2.16.0 of vOTU-representative contigs. The python2 script Blast_to_MCL.py (https://github.com/IGBIllinois/VICSIN/.; https://doi.org/10.1101/2021.09.02.458807) was used to calculate the nucleotide percent length aligned (PLA) and total length aligned (TLA) for all query-subject pairs. Genus-level clusters were calculated with MCL v14.137⁵⁹ using PLA and an inflation of 20. Phage genome networks were constructed based on PLA and TLA from reciprocal BLASTn. Networks were visualized with Cytoscape v3.10.⁶⁰ Clustering was not performed with vContact2,⁶¹ which does not accommodate short contigs and failed to network or cluster any VT contigs with each other or with known viruses from the VirSorter2 database (accessed April 12, 2023) using either ClusterONE and MCL clustering and otherwise default parameters.

Viral taxonomic classification was carried out using the following methods in order of priority. 1) BLASTn v2.14.0 against the nucleotide sequences from IMG/VR high-confidence virus database v4.1 (-outfmt ‘6 std qlen slen’ -max_target_seqs 10,000 -evalue 1e-5); 2) DIAMOND BLASTx against the protein sequences from IMG/VR high-confidence virus database v4.1 (–more-sensitive – evalue 1e-5);⁶² 3) DIAMOND BLASTx against the viral proteins from NCBI’s NR database with the ‘–more-sensitive – evalue 1e-5’ parameters (retrieved in 2023–09-04).⁶² The anicalc.py script from CheckV v1.0.1 was used to calculate ANI and AF from the BLASTn result.⁵⁵ The reference genome’s taxonomy was assigned to the contig if it aligned to the reference genomes with ≥95% ANI and 85% AF. For methods based on similarity to protein sequences, the following rules were used. References’ taxonomy was assigned to the contig if all the aligned reference sequences belonged to the same taxonomy, otherwise the lowest common rank was used.

Trimmed sequencing reads were mapped back to the vOTUs. Read counts and the RPM values were calculated using CoverM v0.6.1 with the ‘-m count tpm – min-read-percent-identity 0.95 –min-read-aligned-percent 0.75 –min-covered-fraction 0’ parameters (https://github.com/wwood/CoverM). The trimmed reads were excluded if the overall percent identity was less than 95% and the percent aligned read bases was less than 75%.

Alpha diversity (Observed and Shannon diversity indexes) was calculated using the R package vegan v2.6.4 (https://www.researchgate.net/profile/Gavin-Simpson-2/publication/228339454_The_vegan_Package/). Figures were plotted in R using ggplot2 v3.4.2 and pheatmap v1.0.12 (https://cran.ms.unimelb.edu.au/web/packages/pheatmap/).⁶³

All other statistical tests were performed with GNU PSPP 1.4.1 (https://www.gnu.org/software/pspp/).

Viral contigs were annotated with Pharokka 1.3.2⁶⁴ using default parameters. Gene prediction was conducted using Phanotate⁶⁵ using default parameters followed by functional annotation by matching each predicted coding sequence to the PHROGs⁶⁶, CARD⁶⁷, and VFDB⁶⁸ databases using MMseqs2⁶⁹. Protein annotation was performed with hmmscan⁷⁰ (http://hmmer.org.) against the Pfam⁷¹ v33.1 database.

Contigs annotated as integrated prophages by Cenote-Taker2 were assumed to use a lysogenic lifestyle. Lysogenic phages were identified based on having at least one key lysogenic functional gene (integrase, excisionase, and repressor/antirepressor) in PHROGs, CARD, or VFDB annotation. Contigs were further annotated by protein annotation with hmmscan against the Pfam database using the gathering cutoff. Proteins with integrase (PF00589, PF00665, PF02899, PF02920, PF13333, PF13683), excisionase (PF05930, PF07825, PF09035), or repressor/antirepressor (PF03374, PF06543, PF07022, PF08346, PF10547, PF10548) domains were considered indicators of a lysogenic lifestyle, and annotated as lysogenic if they contained at least one of these functions. Contigs not integrated as prophages and lacking key lysogenic functions were labeled as “Unknown lifestyle.”

Rarefaction analyses

The assembled vOTU rarefaction curves were generated by randomly subsampling cells to specified cell numbers using the sample function from R. The number of assembled vOTUs were obtained from the subsampled cells. The subsampling process was repeated for 50 times. The average number of vOTUs was calculated by dividing the total number of assembled vOTUs by the number of repetitions for each sampling depth.

The coverage rarefaction curves were generated following the methods previously described⁴⁵. Briefly, the trimmed reads were subsampled to specified read counts using the reformt.sh script from BBMap v39.01.⁷² The subsampled trimmed reads were then mapped back to the viral contigs using Bowtie2 v2.4.2 with default parameters.⁴⁹ The number of covered bases were obtained using SAMtools v1.15.1.⁵⁰ The average coverage values were calculated by dividing the number of covered bases by the total number of bases of the viral contigs.

Richness rarefaction curves were generated by mapping the subsampled trimmed reads back to the vOTUs. Read counts and the reads per million values were calculated using CoverM as described above (https://github.com/wwood/CoverM). The observed index was calculated using the R package vegan v2.6.4.⁷³

Results

Viral tagging is a powerful tool to complement traditional virome sequencing

With the goal of evaluating different methods for characterizing human enteric viromes with a particular focus toward F. prausnitzii, we used three distinct and complementary approaches. First, we performed a preliminary survey of the viral diversity of the intestinal microbiota from Crohn’s disease (CD) and healthy individuals using a traditional virome sequencing approach. Stool samples (n = 19) were collected from eight individuals with CD (n = 9 samples) and eight healthy household-matched controls (HHC; n = 10 samples) (Table S1). The viral fractions of each sample were prepared as viral filtrates, shotgun sequenced, and co-assembled as two metaviromes: CD and HHC (Figure 1(a), Table S2). The CD and HHC metaviromes yielded an initial 2,055 and 2,457 contigs ≥1kb, respectively, and were filtered by computational viral prediction, resulting in 1,375 CD (66.9%) and 1,621 HHC (65.8%) viral contigs ≥1kb (N50 = 6,156bp, L50 = 355) (Tables S3 and S4).

Figure 1.

Methods for viral discovery from complementary sample sources and their detected viruses. (a) Schematic of virome sequencing and metavirome co-assembly for CD (n = 9) and matched HHC (n = 10) stool samples. VLP nucleic acids were amplified prior to sequencing with the RoundA/B method (enteric virome negatively affects seroconversion following oral rotavirus vaccination in a longitudinally sampled cohort of Ghanaian infant, Kim et al) to capture both DNA and RNA viral genomes. (b) Schematic of the UnVT method. UnVT was performed with cells and VLPs from the same healthy human stool (Sample HHC-1). (c) Uncultured bacterial genera from Sample HHC-1 as measured by community 16S rDNA profiling, sc16S survey, and UnVT. Several bacterial taxa were enriched after sorting as compared to the 16S rDNA profile due to cellular autofluorescence (Lawrence 2022). Detailed bacterial abundance chart is shown in fig. S1A. (d) Relative abundance of virus-positive and virus-negative cells by single cell method. (e) Schematic of sc16S survey and SAG methods. The sc16S survey was performed on the same healthy human stool sample (HHC-1) as UnVT, and was reported previously (Lawrence 2022). After their identification by 16S sequence analysis, 100 cells were selected for SAG sequencing with the aim of surveying the genomic diversity of Bifidobacterium, Faecalibacterium, Lachnospiraceae, and other bacteria. Bacterial taxa sampled for SAG sequencing are shown in fig. S1A. (f) Schematic of the FpVT method using F. prausnitzii strain 22 and pooled VLPs from the same CD (n = 9) and matched HHC (n = 10) stool samples used for metaviromes in fig. 1A.

Next, to assess its viability for discovery of novel phages from the human gut microbiota, we performed VT with uncultivated bacteria (UnVT) and VLPs (Figure 1(b)) from the same healthy human stool sample (Sample HHC-1) (Table S1). Sample HHC-1, one of the samples included in the HHC metavirome, was chosen for its large proportion of Faecalibacterium to better sample the genus’s viral diversity (Figures 1(c) and S1A). The top 10% of virus-tagged cells were singly sorted (Fig. S1B), sequenced, and screened for the presence of a single microbial host species by 16S rRNA gene amplicon sequencing (Table S2). Of the 176 wells analyzed, 67 (38.1%) were found to contain a single bacterial ASV also observed in independent 16S profiling of the same sample (Figures 1(c), S1A and Table S3). Previous single-cell analyses of the human microbiota yielded similar rates of single cells passing these thresholds.^21,45,74 Despite not being an exhaustive single-cell survey of bacterial diversity, we did observe increased abundances of specific bacterial genera, including Bifidobacterium, and Blautia_A, Ruminococcus_E, and other Lachnospiraceae, after sorting compared to the independent community 16S profile (Figure 1(c) and S1A). This enrichment of some bacterial types is caused by enhanced autofluorescence of these genera and the selection of the most fluorescent cells for sorting.⁴⁵ Of the UnVT cells passing 16S filtering, 66 were sequenced, individually assembled, and filtered for viral sequence, yielding 331 viral contigs ≥1kb from 54 single cells (81.8% of cells sequenced) (N50 = 7,985 bp, L50 = 36) (Figure 1(d), S1C, Table S4). Most UnVT cells (71.0%, n = 54) were found to have more than one associated phage contig (2–23 contigs per cell). Owing to their small size, UnVT contigs encode few open reading frames (average of 7.4 ± 10.0 S.D.).

Sample HHC-1 was further characterized with a sc16S survey of 1056 individually sorted bacterial cells, performed in the absence of VT, from which 100 were selected for single-amplified genome (SAG) sequencing (Figure 1(e), S1A). SAGs were manually selected to sample the genomic diversity of Bifidobacterium, Faecalibacterium, and other diverse taxa, primarily from the family Lachnospiraceae and the order Oscillospirales, as part of our previously published study (Table S3).⁴⁵ Viral prediction identified 1,387 viral contigs ≥1kb belonging to 724 vOTUs and 556 VCs from 93 of the 100 SAGs (93%) (N50 = 119,978bp, L50 = 3) (Figure 1(d), Table S4).

Finally, a targeted survey of phages infecting F. prausnitzii, our primary organism of interest, was performed by VT of F. prausnitzii strain 22 (FpVT) (Figure 1(f), S1B, S1D), a cultured isolate from a healthy human stool sample³⁴. FpVT was performed with pooled VLPs from the same CD samples (n = 9) or matched HHC samples (n = 10; including Sample HHC-1) which were used for metavirome assemblies (Table S1). Of 176 FpVT cells sorted, 132 (75%) passed 16S filtering, and 96 were chosen for shotgun sequencing (Fig. S1E, Table S2). Having a pre-established and complete genome for the F. prausnitzii host, as opposed to unknown host genomes as in UnVT, enabled the precise removal of Faecalibacterium-derived reads in addition to those derived from sequencing adapters, PhiX, and human sequence, resulting in the removal of 81.9% of the initial 220 million reads (Figs. S1E, S1F, Table S2). Filtered reads were assembled for each single cell, and viral sequence was predicted, yielding 73 viral contigs ≥1kb (5.0% of initial contigs) from 35 cells (36.5%). An additional 34 contigs were removed which were found to be the result of internal genome rearrangement (e.g., transposon movement) of the F. prausnitzii 22 genome, resulting in 39 final FpVT viral contigs (2.6%) from 35 F. prausnitzii cells (33.3%) (N50 = 29,557 bp, L50 = 3) (Figure 1(d), S1E, Table S4). Unlike UnVT, very few cells (5.2%, n = 5) had more than one associated phage contig (2–3 contigs per cell), which we attribute to the more stringent removal of the host genome which eliminates preexisting prophages. As was observed for UnVT, the limited length of most FpVT contigs results in few annotatable ORFs (average of 11.1 ± 27.4 S.D.).

Viral tagging efficiently samples rare phage diversity

Direct comparison of these methods is made difficult by incongruent sequencing depth and inherent differences necessary in handling of sequences during read filtering, assembly, and post-assembly filtering. Read filtering for single-cell methods removed more reads than for viromes; 6.5% of reads were filtered from all UnVT cells (average of 6.0%, ±11% S.D.), 16.0% from all SAG cells (average of 15.0%, ±16.8% S.D.), and 3.8% from all virome samples (average of 3.9%, ±1.9% S.D.). Likewise, during viral contig filtration 91.9% of the total UnVT assembly length (average of 89.0%, ±13.1% S.D.), 87.9% of the total SAG assembly length (average of 88.7%, ±9.2% S.D.), and 16.4% of the total metavirome assembly length (average of 16.7%, ±1.8% S.D.) were removed (Table S2). This was expected, given that single-cell sequencing is dominated by the host genome, while virome preparation depletes host DNA.

The final viral assemblies for each method used were very different in their total length and number of contigs. Total assembly lengths were 1.3 Mbp for UnVT (average of 0.02 Mbp per assembly, ±0.02 S.D.), 14.13 Mbp (average of 0.14 Mbp per assembly, ±0.14 S.D.) for SAGs, and 11.7 Mbp for metaviromes (average of 5.8 Mbp per assembly, ±1.8 S.D.) (Table S2). However, the overall distribution of viral contig lengths, though dominated by short contigs for all methods, was highly similar across all methods (Fig. S2A), suggesting that single-cell methods for viral identification produce final datasets of similar quality to those from traditional virome sequencing. Furthermore, total assembly lengths normalized to sequencing depth are not significantly different between UnVT and individual virome assemblies, although FpVT and SAGs produced less viral sequence (Figure 2(a)). In fact, UnVT has the largest average normalized viral length of all methods.

Figure 2.

Comparison of efficiency and diversity sampling achieved by UnVT, FpVT, SAGs, and metaviromes. (a) Normalized total viral length was calculated against the number of filtered reads on a per-assembly basis. Only virus-positive cells from UnVT, FpVT, and SAGs were included. In addition to metaviromes, individually assembled viromes were constructed for analysis. Significant differences between methods were calculated by Tukey’s HSD (***, p ≤ 0.001). (b) Venn diagram of vOtus clustered across all methods and observed in assemblies. Viral contigs were clustered into vOtus at ≥ 95% ANI over ≥ 85% AF. The longest contig from each cluster was selected as the vOTU’s representative sequence. (c) Venn diagram of VCs clustered across all methods and observed in assemblies. VCs were assigned by clustering representative vOTU contig nucleotide sequences with MCL v14.137 at ≥ 20% PLA (Campbell VICSIN). (d) Normalized richness was calculated on a per-assembly basis. The number of unique vOtus observed was normalized to the number of filtered reads. Significant differences between methods were calculated by Tukey’s HSD (***, p ≤ 0.001). (e) vOTU richness rarefaction curve by the number of cells sampled was determined by counting the number of vOtus assembled. Each subsample was performed for 50 replicates and averaged. (f-i) Read coverage of metavirome, UnVT, FpVT, and SAG viral contigs. Each sample’s reads were mapped to the full set of viral contigs assembled for all samples or cells used in that method. (j-m) Sampling curves of vOtus mapped by reads sampled for metavirome, UnVT, FpVT, and SAG vOtus. Detection of vOtus was determined by read mapping from each sample or cell to the full set of vOtus across all methods.

To examine their overall diversity, all 3,619 vOTUs and 2,856 VCs assembled in UnVT, SAGs, FpVT, and metaviromes were compared (Figure 2(b,c)). Approximately half (49.7%) of all viral contigs are singletons with no related contigs at the vOTU or VC level across all methods (Table S4). The majority of vOTUs and VCs are unique to a single method (Fig. S2B, S2C). No vOTUs and only 1 VC were shared across all methods. For the UnVT and SAG datasets, both of which used sorted cells from Sample HHC-1, 32 vOTUs (11.0% of UnVT vOTUs) and 55 VCs (21.7% of UnVT VCs) are shared. For UnVT of Sample HHC-1 and the metavirome of all HHC samples (including Sample HHC-1), only 3 vOTUs and 2 VCs are shared. Only a single VC is shared between FpVT and the HHC metavirome; no vOTUs or VCs are shared between FpVT and the CD virome, despite the majority (59.4%) of FpVT cells having been tagged with VLPs pooled from CD VLPs. These patterns demonstrate that viral populations detected by each method are almost entirely distinct.

Although metaviromes captured a much larger amount of phage diversity, normalizing our observations to sequencing depth reveals that single-cell methods are at least as efficient as virome sequencing for surveying viral diversity (Figure 2(d)). Normalized viral richness is greatest on average in UnVT and is significantly greater than other single-cell methods. These single-cell methods are thus highly efficient per read sequenced, although they would require an enormous number of sorted cells to fully sample the viral diversity in the very complex gut virome (Figure 2(e), S2D). The true number and diversity of VLPs associated with a virus-tagged is, as of now and with these methods, unknowable. Our VT surveys likely undersample viruses for two primary reasons: (i) VT sequencing results in highly fragmented assemblies owing to the enrichment of the host genome during MDA⁷⁵ (ii) the fluorophore used to stain VLPs is not specific to DNA, and also stains RNA. Any host-bound RNA phages would cause greater fluorescence without being detectable in later steps.

The distinction of the viral populations observed by each method is further seen in read mapping. Mapping reads from individual viromes to all viral contigs from the CD and HHC metaviromes yields approximately 1–25% total sequence coverage, while those contigs are largely undetectable when mapping reads from UnVT, FpVT, or SAGs (Figure 2(f)), which was expected given that single-cell methods only subsample the virome. Conversely, we expected that mapping virome reads to UnVT, FpVT, and SAG viral contigs would achieve coverage rates similar to what is observed when mapping reads from the source method. Virome read coverage of UnVT, FpVT, and SAG contigs, however, is much lower than that for each source method (Figure 2(g–i)), despite viromes being approximately 4- to 77-times more diverse than single-cell methods by total vOTUs assembled (Figure 2(j–m)). Together, these results show that the distinct viral populations surveyed by VT or SAGs are not readily captured by traditional virome sequencing, likely due to their rarity being beyond the detection limits of virome sequencing.

Specifically for Faecalibacterium hosts across single-cell methods, the observed phages are also method-unique. The 39 FpVT viral contigs belong to 34 vOTUs and 26 VCs (Table S4). Comparing vOTUs shared by FpVT cells, Faecalibacterium UnVT cells, and Faecalibacterium SAGs shows that, although few assembled vOTUs are shared, vOTUs detected by read mapping from only one method are rare (Fig. S2D). Only 27% of the vOTUs detectable in FpVT are unique to the method, and 59 vOTUs (22% of all Faecalibacterium-associated vOTUs) are shared across all datasets. These patterns are suggestive of host specificity, as only one strain of F. prausnitzii was present in FpVT while UnVT and SAGs captured multiple uncultivated Faecalibacterium cells, and supportive of phage-host specificity during VT.

UnVT captures rare & diverse phages

The 331 viral contigs identified by UnVT are diverse in multiple ways. They show remarkable diversity at the nucleotide level, as they span 290 vOTUs and 254 VCs, with 184 (55.6%) being singletons with no related contigs across all assemblies (Figure 3(a)). These contigs have diverse bacterial hosts from at least 21 host genera spanning 10 families, 8 classes, 7 orders, and 6 phyla (Figure 3(a), Table S4). Given the large number of VCs generated, they are predicted to be highly diverse. Taxonomically, 307 contigs (92.7%) were broadly assigned to the viral class of tailed phages Caudoviricetes, though not assignable at the family level, and the 24 contigs remaining contigs being unclassified (Table S4). We postulate that the scarcity of specific taxonomic assignments is partly due to short contig length, as well as the divergence of UnVT phages from known viruses. UnVT contigs include lysogenic phages: 11 integrated prophage contigs (3.3%) and 24 predicted lysogenic phage contigs (7.3%) were found (Figure 3(b), Table S4). Viral contigs encoding no predicted proteins indicative of viral lifestyle and were classified as “unknown” lifestyle, and many likely have strictly lytic lifestyles. Due to incomplete assembly of UnVT contigs, however, these are likely underestimations of the actual numbers of prophages and lysogenic phages captured. This is supported by the observation that 35 (10.6%) and 72 (21.8%) UnVT viral contigs belong to vOTUs or VCs, respectively, with other predicted prophages or lysogenic prophages, despite themselves being classified as unknown lifestyle (Table S4). Finally, this diversity in UnVT viral contigs extends across known phages, as only contig is similar to any previously characterized phage (Figure 3(c)).

Figure 3.

UnVT and SAG sequencing using the same healthy human stool sample capture distinct phage populations. (a) Network of UnVT viral contig nucleotide similarity by host and predicted viral lifestyle. Only relationships with a total length aligned (TLA) ≥500 bp and PLA ≥ 10% are shown. Only VCs with more than one member contig assembled by UnVT are shown. vContact2 failed to network or cluster any VT contigs. (b) Relative number of viral contigs retrieved by lifestyle. Prophages have identifiable phage-host sequence junctions. Lysogenic phages encode at least one marker gene for lysogeny. Unknown contigs encode no proteins indicative of their lifestyle. (c) Faecalibacterium phage UnVT_Faecalibacterium_19 aligns to Faecalibacterium phage Toutatis (NCBI RefSeq GCF_002958115.1) with high nucleotide and protein identity across two regions spanning genes predicted to encode a phage structural protein, the Avd protein associated with diversity generating retroelements, a domain of unknown function (DUF), and hypothetical proteins. UnVT_Faecalibacterium_19 is a singleton contig within its vOTU and VC. Protein-based alignment was created with Clinker (Gilchrist 2021). (d) Comparison of vOtus detected by assembly or read mapping from UnVT cells and SAGs for select host taxa. The number of single cells in each host group are indicated. (e) Nucleotide similarity network of viral contigs identified from UnVT and SAGs. Only VCs with both UnVT- and SAG-derived viral contigs are shown. Only relationships with a TLA ≥ 500 bp and PLA ≥ 10% are shown. Singleton contigs are not shown. vContact2 failed to network or cluster any VT contigs. Full network of all UnVT and SAG contigs is shown in fig. S3. (f) Heatmap of UnVT vOTU abundance in cohort viromes. No significant differences in vOTU abundance between CD and HHC were observed. 265 UnVT vOtus were not detected in any cohort virome and are not shown.

Incomplete assembly masks the true diversity of phages associated with UnVT hosts. More viruses were detected by SAG sequencing (14.9 viral contigs per virus-positive cell; n = 93) than UnVT (6.1 viral contigs per virus-positive single cell; n = 54). This is likely the consequence of (i) increased sequencing depth of SAGs (Figure 3(a–i)) and (ii) MDA of virus-tagged single cells is biased toward longer sequence, such as bacterial host genomes compared to those of the attached viruses⁷⁵. By mapping reads from each UnVT cell to the complete vOTU dataset, we were able to detect additional vOTUs which originate from SAGs derived from the sample but were not assembled during UnVT (Figure 2(k) and 3(d)). Across all UnVT cells sequenced, read mapping detects an average of 30 additional vOTUs (±22 S.D.) not present in assemblies. Even for UnVT cells that failed to yield any assembled viral contigs (n = 13), read mapping detects an average of 18 vOTUs (±16 S.D.). This further suggests that additional viruses are likely present in each single cell sampled, even if they were not assembled, which we predict would be at least partially remedied by deeper sequencing (Figure 2(k)). Together, these results demonstrate that UnVT is capable of capturing a complex mixture of phages from a single human stool sample.

SAG sequencing was performed for 100 cells also from Sample HHC-1. We expected that SAGs would contain phage sequence primarily in the form of integrated prophages. SAGs might also reveal phage genomes from virions bound to the cell surface and/or actively infecting a cell prior to cell sorting; however, we expect such events to be less prevalent than in UnVT, given that SAG cells were not explicitly tagged with viruses. SAG sequencing revealed 120 (8.6%) integrated prophages and 166 (12.0%) lysogenic contigs (Figure 3(b)). The relative rates of prophage and lysogenic phage detection are ~2.6 times and ~1.6 times, respectively, greater than their in UnVT, a pattern which was consistent for most of the taxa most frequently shared between UnVT and SAGs (Fig. S3B). As in UnVT, these counts are confounded by incomplete assemblies, and accurately assessing the true number of integrated prophages requires more complete SAG assemblies. These data, however, do support our hypothesis that SAGs are enriched for integrated prophages compared to UnVT.

We next compared viral diversity between UnVT and SAGs for Faecalibacterium, the primary host of interest, as well as the four most common other host taxa observed: Bifidobacterium, Blautia_A, other Lachnospiraceae (not including Blautia_A) and Oscillospirales (not including Faecalibacterium). In comparing vOTUs detected in assemblies, we see that a small minority of vOTUs are shared between UnVT and SAGs for these hosts, although more overlap is observed by read mapping (Figure 3(d)). Despite this, a large degree of assembled and mapped phage diversity remains unique to each method. For those vOTUs that are shared, UnVT and SAGs complement each other and identify a startling diversity uncultivated phages and their bacterial hosts (Figure 3(e), S3C).

Between the UnVT and SAG viral datasets, 109 UnVT viral contigs (32.9%) are related to 331 SAG viral contigs (23.8%) across 55 VCs (Figure 3(e), Table S4). Of these, 18 VCs (32.7%) were observed at least 5 times across the UnVT and SAG datasets. These were assigned host ranges after excluding singletons, with six VCs restricted to host genus, six to host family, three to host order, and three to host phylum (Table S4). Of the six VCs restricted to a single host genus, three occur in Bifidobacterium (VC8, VC68, and VC937), two in Faecalibacterium (VC187 and VC221) and one in Gemmiger (VC408) (Figure 3(e)). The largest VCs, VC0 and VC2, have host ranges spanning the phylum Firmicutes_A. Specifically, Lachnospirales and Oscillospirales hosts were observed in both datasets and include integrated prophages from both of these host classes (Tables S3 and S4), greatly substantiating their broad host ranges. Although VCs are broad groups of viral sequence, even analysis of vOTUs shows that some phages might span bacterial phyla or the entire bacterial kingdom, though vOTUs have fewer member contigs. Regardless, these data do support that phages with broad host ranges exist within the human microbiota.

Sample HHC-1 was chosen for its high abundance of Faecalibacterium (6.3% of the community by 16S profiling) (Figure 1(c)) in an effort to survey phages infecting the genus. UnVT sampled four Faecalibacterium cells (7.6%), all of which yielded at least one phage contig (Tables S3, S4). A total of 40 Faecalibacterium-associated phage contigs were identified by UnVT (12.1%) across 29 vOTUs and 254 VCs (Figure 3(a)). Four Faecalibacterium-associated UnVT phage contigs belong to VCs observed across the family Lachnospiraceae (Figure 3(a), Table S4). All previously reported phages for F. prausnitzii are lysogenic dsDNA tailed phages³². UnVT identified four predicted lysogenic Faecalibacterium phage contigs based on the presence of lysogenic marker genes and one (Faecalibacterium phage UnVT_Faecalibacterium_19) related to the known lysogenic Faecalibacterium phage Toutatis (GenBank NC_047915.1; Figure 3(c))³². Despite apparent sequence rearrangement, these two phages share significant homology at the nucleotide and amino acid levels, including an avd gene encoding an essential factor for a diversity generating retroelement⁷⁶, and another encoding a Hoc-like phage head protein. Faecalibacterium phage UnVT_Faecalibacterium_19 is further completely unique within our dataset, being the only member of its vOTU and VC (Table S4). Together, these findings highlight the need for deeper sampling and characterization of human gut phages, especially those infecting the important health-associated microbe Faecalibacterium.

We performed read mapping from virome sequencing of our CD and HHC cohort (Tables S1, S2) to UnVT viral contigs to determine their abundances. Only 20 UnVT phage vOTUs (6.9%) are detectable by read mapping from the source virome and an additional 43 (14.8%) are detectable in any of the individual viromes (Figure 3(f)). This suggests that most UnVT contigs are too rare to be observed by virome sequencing alone. This is further evidenced by the uniqueness of the vOTUs and VCs sampled by UnVT and the metaviromes (Figure 2(b, c)) and the low overall coverage of those UnVT viral contigs with mapped virome reads (Figure 2(g)).

Because UnVT hosts lack a pre-established genome, distinguishing integrated and excised phage sequence is impractical. Read mapping from the source virome HHC-1, however, detected four UnVT prophage vOTUs (36.4%) after trimming of predicted host sequence and one UnVT lysogenic vOTU (4.3%) (Figure 3(f)). This supports that these five UnVT viral contigs are active phages within the source sample, as metaviromes are largely free of cellular host sequence. Indeed, only 18 predicted prophage contigs (Table S4) were detected across both metaviromes (0.6% of metavirome contigs). In the other individual viromes from our cohort, one additional prophage-containing vOTU and 3 lysogenic vOTUs were detected (Figure 3(f)), suggesting that these are representative of active lysogenic phages, though their activity cannot be confirmed in the sample from which they originated (HHC-1). Currently, there is no evidence to suggest the activity of the remaining 25 prophage or lysogenic UnVT vOTUs (73.5% of UnVT viral contigs). In the absence of more comprehensive virome sampling, distinguishing active and integrated phages detected in UnVT remains problematic.

Determination of UnVT phage activity in viromes is limited by their overall low abundance. Of the vOTUs with no predicted lifestyle, only 68 (25.6%) are detectable by read mapping from any virome (Figure 3(f)). This strongly suggests that UnVT is capable of detecting rare phage types missed by virome sequencing alone, which is skewed toward the most abundant sequences even at high sequencing depth.

VT of cultured F. prausnitzii identifies novel phages & active infection

A complete genome for F. prausnitzii strain 22 was generated by assembling short- and long-read sequencing data (Figure 4(a)), and its phylogeny using multiple-core marker genes confirms this strain’s identity as a true member of the species (Fig. S1D), even in consideration of recent updates to the taxonomy of the Faecalibacterium genus^77–79. The genome was further annotated to identify any integrated prophages (Figure 4(a), Table S5). Although most of these integrated phage-like regions are incomplete (Table S5), all were included in read removal to ensure that the final viral contigs were not the result of preexisting phages in the host genome and were strictly those introduced by VT.

Figure 4.

VT of F. prausnitzii identifies diverse phages and new prophage integration events. (a) The genome of F. prausnitzii was sequenced by long and short read sequencing. Prophage-like regions were annotated with VirSorter2 and PHASTER and manually annotated for completeness. (b) Nucleotide similarity network of FpVT phage contigs and endogenous prophage-like regions of F. prausnitzii strain 22. Only relationships with TLA ≥ 500 bp and PLA ≥ 10% are shown. vContact2 failed to network or cluster any VT contigs. (c) Nine FpVT phage contigs from 6 VCs are similar to the endogenous phage-like region 4 of the F. prausnitzii strain 22 genome at the nucleotide level (PLA ≥56%). Contig FpVT_26 is additionally similar to endogenous prophage-like region 6 (58% PLA). Sequences were aligned by predicted protein similarity (≥30% identity), showing conservation of many phage-associated proteins. (d) the four FpVT phage contigs from VC 2 are similar at the nucleotide level to endogenous prophage-like region 3 and 7. Only alignment with endogenous prophage-like region 3 is shown. All alignments were created with Clinker (Gilchrist 2021). (e) Two predicted prophage contigs (FpVT_15 and FpVT_26) were identified during filtering of FpVT viral contigs. Alignment of these contigs with their related regions of the F. prausnitzii strain 22 chromosome demonstrates conservation of attachment site-adjacent proteins. Newly integrated phage sequence does not align with the F. prausnitzii 22 genome. Gene abbreviations: ant, phage anti-repressor; dnaB, replicative helicase; dnaC, helicase loader; int, integrase; mazF, endoribonuclease toxin MazF; parB, partition DNA-binding subunit; repA, replication initiation protein A. (f) Network of FpVT_26 attachment site similarity with SAGs and their associated phages. The FpVT_26 attachment site is at center. All associated phages belong to VC75.

FpVT sequencing revealed 39 viral contigs that are highly diverse, belonging to 34 vOTUs and 26 VCs, 16 of which are singletons. Viral taxonomic prediction identified 23 tailed phages from the class Caudoviricetes, 5 filamentous phages from the order Tubulavirales, including four in the family Inoviridae, and 9 that remain unclassified (Figure 4(b), Table S4). No predicted filamentous phages were found by UnVT or in SAGs. An additional 2 FpVT contigs were identified as Cressdnaviricota and Cirlivirales, both non-phage viruses broadly associated with eukaryotes (Table S4). All FpVT phage contigs are of unknown lifestyle, except for 2 that are predicted prophages. No additional lysogenic phage contigs were found, which is in contrast to their detection in Faecalibacterium cells observed from UnVT and SAGs (Fig. S4A) a distinction that is likely the result of filtering preexisting prophage sequence from FpVT cells. Finally, although one Faecalibacterium-associated phage found in UnVT was similar to a previously described Faecalibacterium phage, this was not the case for any FpVT viral contigs,³¹ demonstrating that the FpVT viruses are novel in multiple ways.

Although 64 FpVT cells lacked any assembled viral contigs, read mapping against the full vOTU dataset suggests that all cells were tagged by at least one vOTU (Fig. S4B), suggesting that assembly was likely impacted by low sequence coverage for some cells (Figure 2(h,l)). FpVT viral contigs are largely undetectable in virome sequencing, with only 13 contigs (33.3%) having any reads from viromes mapping to them, and their presence and abundance are highly variable between individual subject (Fig. S4C). This is consistent with UnVT and demonstrates that VT performed in two distinct ways captures rare phage diversity that goes undetected in source viromes.

Among FpVT viral contigs, 14 contigs from 7 VCs have high nucleotide similarity to endogenous prophage-like Regions 4 and 3 of the F. prausnitzii 22 genome (Figure 4(b)). Twelve FpVT contigs from six VCs align with Region 4, two of which also align with Regions 1 or 3, though with lower nucleotide identity (Figure 4(b,c)). Four FpVT contigs from a single VC align with Region 3; three of these contigs also align to Region 7 with lower nucleotide identity (Figure 4(b,d)). Protein alignments of Regions 4 and 7 with their related FpVT phage contigs further bolsters these relationships, as both hypothetical and functional proteins are shared. Functions that are shared include DNA-active proteins (e.g., phage integrase), structural proteins, and regulatory proteins (Figure 4(c,d)). Together, these relationships suggest that F. prausnitzii is a true host of these FpVT-identified phages.

Remarkably, FpVT also identified two newly integrated prophage contigs (FpVT_15 and FpVT_26), which were observed in independent F. prausnitzii 22 cells, each tagged with separate VLP pools (CD and HHC, respectively) (Figure 4(e)). These were the only phage contigs identified for each of these host cells, though. It remains unclear if (i) the high fluorescence of these cells is the consequence of the unavoidable undersampling in VT phages, (ii) integration occurred after cell sorting, or (iii) integrated prophage DNA maintains its fluorescence. At the nucleotide level, both FpVT_15 and FpVT_26 are similar to endogenous prophage-like region 4 within their predicted viral sequence, but not in the adjacent sequence that is ≥99.9% identical to the F. prausnitzii host. Their attachment sites are not within endogenous prophage-like regions and are approximately 1.3 Mb from each other (Figure 4(a,e)). These observations of newly integrated prophages greatly substantiate the ability of host-specific VT to detect true phage-host pairs, as well as active phage infection and lysogenization events.

To determine if the attachment sites of these newly integrated prophages are used by other phages, we compared them to sequenced SAGs. The attachment sites were inferred to be within the 50 bp adjacent to the prophage integration site, at least 20 bp long, and with at least 95% identity. The attachment site of FpVT_15 is conserved with high identity (98% over 50 bp) in one Faecalibacterium SAG (8p5-11B) (Table S3). Contig FpVT_15 is a singleton within its VC (Table S4), and its attachment site in SAG 8p5_11B does not have an adjacent integrated prophage. The attachment site of FpVT_26 is detected in two Faecalibacterium SAGs (2p1-9A and 1p2-12C), neither of which is adjacent to an integrated prophage (Figure 4(f), Tables S3, S4). Though it is not integrated as a prophage at the conserved attachment site, a lysogenic contig from SAG 2p1-9A (contig SAG_Faecalibacterium_44), is also a member of the same VC as FpVT_26 (VC75). VC75 has six additional members across Faecalibacterium and Gemmiger hosts, four of which are predicted to be lysogenic (Table S4). By examining the other VC75 host genomes, three Gemmiger SAGs were found to have a near match to the FpVT_15 attachment site (≥91.7% identity and ≥20 bp) as well as an identified phage contig from VC75 (Figure 4(f)). Together, these findings strongly suggest that the host range of VC75 phages spans at least two genera, Faecalibacterium and Gemmiger, and that they may use these related attachment sites for integration.

Single cell methods detect phages associated with inflammatory bowel disease

To identify vOTUs associated with CD, we first examined UnVT- and FpVT-assembled vOTUs in the context of our cohort of 9 individuals with CD and their 10 HHCs. Reads mapping from CD and HHC viromes to each vOTU, however, did not identify any vOTUs that were significantly different in their abundance between the two conditions (Figure 3(f), S4C). Only 67 UnVT (23.1%) and 13 FpVT (38.2%) vOTUs were detected in at least one virome.

We next examined VT vOTU abundances in a larger, more deeply sequenced set of viromes from a cohort of 37 CD, 66 ulcerative colitis (UC), and 66 HHC healthy individuals⁴, and found 134 vOTUs were present in at least one virome (Figure 5(a)). Still, most VT vOTUs (58.5%) are not detectable in any virome (Figure 5(b)). This is consistent with our prior conclusion that VT primarily samples rare phage species. In fact, most of the vOTUs that were detected were present in four or fewer viromes ( <2% of viromes) and the most common vOTU (UnVT_Blautia_A_73) was present in only 69 (40.8%) viromes, together showing that the phages identified here by VT are very unique to the sample. For those vOTUs that were detected by read mapping, their average abundances were high within those viromes that they were detected (Figure 5(c)).

Figure 5.

VT detects phages associated with IBD. (a) Read mapping from previously published CD, UC, and healthy viromes⁴ to UnVT and FpVT votus was used as a measure of abundance. (b) Distribution of the count of vOtus detectable in viromes. (c) Distribution of the count of vOtus by average RPM within only those viromes were they were detectable. (d) Five VT viral contigs were found to be associated with at least one condition compared to healthy by Fisher’s exact T test (#, p ≤ 0.05; ##, p ≤ 0.01) and significantly different in abundance between at least two conditions by Tukey’s HSD (*, p ≤ 0.05; **, p ≤ 0.01). (e-i) Alignments of significantly different viral contigs with other contigs selected from within the same VC. Gene abbreviations: int, integrase; mazF, endoribonuclease toxin MazF; parA, partition ATPase; parB, partition DNA-binding subunit; repA, replication initiation protein A; terL, terminase large subunit; terS, terminase small subunit.

Despite the rarity of most VT-sampled phages, read mapping did identify 6 VT viral contigs whose presence is associated with at least one condition (healthy, CD, or UC) and whose abundance was significantly different between at least two disease conditions (Figure 5(d)). These contigs were all associated with Ruminococcaceae, which includes Faecalibacterium, or Lachnospiraceae hosts, the two most common families observed in UnVT, and each of these host taxa has been observed previously to correlate with inflammatory bowel disease (IBD)^{24,25,80–83}.

Two Faecalibacterium viral contigs, UnVT_Faecalibacterium_11 and FpVT_28, are positively associated with and more abundant in CD samples than in healthy samples; FpVT_28 is additionally more abundant in CD than in UC samples. Both contigs are short fragments, but they are also highly unique from their closest related viral contigs (Figure 5(e,f)), suggesting that phages within the same VC can have different abundance patterns across healthy individuals and IBD patients. Contig UnVT_Ruminococcus_E_32 was found as a prophage during UnVT. Its presence in viromes, including three of our cohort viromes, as well as its large number of phage-associated genes, suggests it is representative of active and inducible lysogenic phages (Figure 5(g)). UnVT_Ruminococcus_E_32 is the only VT-identified phage which is significantly associated with the health (Figure 5(d)), which may be the result of prophage induction in the healthy gut. Contig UnVT_Blautia_A_65 is associated with and significantly more abundant in CD viromes than in those of healthy and UC individuals, though its average abundance within CD viromes is low (1709 reads per million). UnVT_Blautia_A_65, a predicted lysogenic phage, does encode auxiliary metabolic genes, which may impact these patterns and bacterial host activities (Figure 5(h)). Finally, two contigs from both from VC45 were observed in two different Lachnospiraceae hosts. Both contigs are significantly associated with CD and are significantly more abundant in UC than in healthy samples. Both show similar abundance patterns across disease conditions. These VC45 phages are closely related to predicted tailed phages, and they may have strictly lytic lifestyles as they lack any lysogenic marker genes (Figure 5(l)). In sum, even this limited effort to sample phages by VT (n = 162cells across UnVT and FpVT) was successful for identifying differentially abundant phages in IBD.

Viral tagging identifies diverse and broad host range phages

As is expected for any survey of the human gut virome, the biological diversity found by VT was immense (Figure 6(a), S5). With the exception of one Faecalibacterium-associated phage (Figure 3(c)), this diversity is entirely uncharacterized. Much of the viral sequence from VT remains poorly annotated, owing to its novelty, making it difficult to interpret its functional potential. Many VCs, however, were found to be of particular interest for a variety of reasons. Nearly all of the hosts we observed here have been associated with health or IBD, although the directionality of those associations are sometimes controversial^{24,25,84–86}. VC632 is a cluster of filamentous phages from the family Inoviridae, four of which were found associated with F. prausnitzii 22 during FpVT, and a fifth with a Gemmiger SAG (Figure 6(b)). Like Faecalibacterium ^24,25, Gemmiger is negatively correlated with IBD^84,85. Though most of the predicted proteins for these phages remain hypothetical, the known factors they do encode are characteristic of filamentous phages. With the exception of one contig (FpVT_18), all members of the VC are nearly identical, and 5,959 or 6,049 bp in length. Given the close relatedness of these phages, these data suggest that they can infect at least across these two host genera, and this is the first report of filamentous phages for both Faecalibacterium and Gemmiger hosts.

Figure 6.

VT detects broad phage diversity for IBD-relevant hosts. (a) Nucleotide similarity network of all viral contigs detected across UnVT, FpVT, SAGs, and metaviromes. Only relationships with TLA ≥ 500 bp and PLA ≥ 10% are shown and only VCs with at least one UnVT or FpVT contig are shown. vContact2 failed to network or cluster any VT contigs. (b-g) Alignments of select VCs demonstrate varying amounts of within-cluster diversity and phage-encoded functions. Gene abbreviations: ant, phage anti-repressor; darB, internal packaging protein; dnaC, helicase loader; dnaD, replication initiation protein; exc, excisionase; int, integrase; MTase, methyltransferase; PAPS reductase, phosphoadenosine-phosphosulfate reductase; parA, partition ATPase; parB, partition DNA-binding subunit; recT, recombinase; relE, translational inhibitor toxin RelE; repA, replication initiation protein A; repB, replication initiation protein B; terL, terminase large subunit; Tra, transposition-associated protein.

Similarly, VC75 was observed multiple times for Faecalibacterium and Gemmiger hosts by FpVT or in SAGs (Figure 6(c)). The VC includes contig FpVT_26, one of the two newly integrated phages observed in FpVT (Figure 4(e)), whose attachment site is conserved in Faecalibacterium and Gemmiger hosts (Figure 4(f)). Beyond their essential lysogenic functions, encoded by the integrase and excision genes, very little of these phage genomes is annotatable or interpretable. The longest member of the cluster, SAG_Gemmiger_335, encodes its own chaperonin, a protein function which has been observed before in phage genomes⁸⁷, and a DNA methyltransferase presumed to confer resistance to host restriction⁸⁸. The dearth of identifiable functions within VC75 is not unusual for uncultivated phage genomes and is sometimes interpreted as reason to question the legitimacy of these types of sequences. Here, however, the functionality of VC75 phages is clear, given the novel integration of FpVT_26, and the multiple observations of VC75 across related host cells in the family Ruminococcaceae.

The largest VCs in our dataset, VC0 and VC2, have 58 and 56 members each, which emerged across UnVT, FpVT, and in SAGs (Figure 6(d,e)). In both VCs, many viral contigs are annotated as lysogenic, and three and six prophages occurred in VC0 and VC2, respectively. In both cases, prophages occur in diverse bacterial host SAGs within the Lachnospirales and Oscillospirales, which strongly suggests the host range of both VCs spans the host class Clostridia. Additionally, each VC was observed multiple times associated with UBA11524, a genus within the Christensenellales, suggesting their host range may extend to the entire Firmicutes_A phylum. Beyond their expansive host ranges, many viruses within VC0 and VC2 carry auxiliary metabolic genes, most frequently ABC transporter proteins. Whether these annotated proteins are functional, and what their roles in phage–host and phage–gut interactions might be, remains unknown.

Perhaps one of the best-annotated VCs in our dataset is VC45, a group of predicted tailed phages that span Lachnospiraceae hosts and occurred in both UnVT and SAGs (Figure 6(f)). Though their assemblies are incomplete, they together contain hallmark proteins for phage replication, packaging, virion structure, and lysis. In addition to these core functions, one viral contig (SAG_CAG-317_14), a prophage in CAG-317, encodes a predicted phosphoadenosine-phosphosulfate (PAPS) reductase, which has been observed in numerous phage genomes from a variety of environments^11,89–91. PAPS reductases are essential for sulfate metabolism and have been suggested to be beneficial to their hosts⁹².

Finally, VC349 presents perhaps the most confounding VC identified here (Figure 6(g)). VC349 contains two long viral contigs, FpVT_1 and UnVT_Bifidobacterium_27, which are 78,118 and 38,758 bp, respectively. These contigs are also 100% identical to each other at the nucleic acid level over the entire length of the shorter contig. Despite this, they were observed in association with Faecalibacterium and Bifidobacterium, genera only related at the level of the bacterial kingdom. UnVT_Bifidobacterium_27 was observed as an integrated prophage in its source cell (1-11D), and was observed by read mapping in 11 other Bifidobacterium cells from UnVT. FpVT_1’s lifestyle is unknown, and read mapping found it present in 17 additional F. prausnitzii 22 cells (Fig. S4C). As both of these phage–host interactions are strongly supported, this suggests that the human gut may harbor phages with host ranges spanning all bacteria.

Discussion

In this study, we employ multiple single-cell methods to identify a total of at least 328 phage-host pairs from the human gut microbiome. Our results demonstrate that viral tagging (VT) is synergistic with traditional virome sequencing in multiple ways, especially by providing the crucial context of a given phage’s bacterial host, an essential requirement to interpret phage evolution and ecology. Though our VT surveys were small, they constitute a crucial step forward in enabling future mechanistic studies of phage-bacteria interactions and their roles in health and disease.

VT is a remarkably efficient tool for phage discovery, requiring as many reads as virome sequencing to retrieve equal viral length and richness. The main limitation of VT is at the level of cell sorting, as it would require sorting an impractical number of uncultivated cells to perform whole-virome surveys. There are a number of additional caveats to this VT method, including biased amplification of bacterial genome prior to sequencing⁷⁵, fragmented assemblies, and an inability to detect RNA viruses. We show, however, that VT is still highly valuable when used in conjunction with virome sequences, and it can further be manipulated to subsample hosts of interest. We also show that VT with a single cultured isolate is incredibly powerful, as it makes clear differentiation between host genome and introduced phage sequence possible. Further, our VT of an F. prausnitzii isolate found two newly integrated prophages, which demonstrates that VT is not only detecting surface interactions, but active phage infection events as well.

Here, we use VT and single-amplified genomes to determine phage host ranges. Our results are likely very different from what might be observed when using traditional culture methods (e.g., plaque assays), which suffer from complicated cultivation biases when making these determinations⁹³. We assume that VT primarily detects surface-level interactions, which is one of the several ways to define host range¹⁹. Except in the case of phage integration into a known host genome, as in F. prausnitzii here, VT cannot detect the steps of infection subsequent to surface attachment. Despite this, we show evidence that the phage–host interactions observed by VT are accurate and specific, as we see many closely related phages infecting closely related hosts, instances of relationships to existing prophages. We observed only two non-phage interactions, which may be biologically accurate, as eukaryotic viruses are well-established to directly bind bacteria^94,95. VT did find many instances of broad host range phages spanning host families, phyla, and potentially all bacteria. This departs from phage biology canon, which asserts that phages have highly restricted host ranges based on their infectiveness in culture, although broad host range phages are increasingly being identified both in vitro ^96,97 and in silico ^98,99. VT is yet another methodology that strongly suggests the field needs to reconsider how it defines host range for a more accurate understanding of phage ecology in vivo.

Just like all viromic and phage-host identification methods, the VT approach has its caveats. All VT experiments must consider biases in their enrichment and sorting approaches, as our results show that fluorescent staining of nucleic acid results in the enrichment of specific bacterial taxa. These biases, however, can be leveraged to select for hosts of interest, and the future development of alternative staining or enrichment approaches for VT will allow the study of other hosts and their phages. The method of VLP preparation should also be considered. We used crude VLPs, filtrates of stool supernatants, which likely afford relevant cofactors for attachment to surface receptors^100–102, though other factors, such as human antibodies, may interfere with host attachment^103,104. In the case of VT of individual isolates, future work should carefully consider their culture methods. Cultivation in vitro results in differential expression patterns in bacterial hosts^105,106, especially relevant surface factors that can mediate or inhibit phage attachment.

Finally, VT is capable of finding rare phage species not observable in viromes, as ‘omics techniques are inherently biased toward the most abundant species. Most of our VT-identified phages are extremely rare, but this does not preclude the possibility that they play outsized roles in the ecology of the microbiota as keystone species. Importantly and in contrast to recent results on viral taxa in inflammatory bowel disease¹⁰⁷, VT sampled poorly explored yet highly relevant bacterial hosts. The most notable of these is Faecalibacterium, but we also captured phages for other putative anti-inflammatory taxa, such as Blautia, Ruminococcus_E, and Bifidobacterium.^80,108 Several of these phages exhibit low overall abundance but are strongly associated with health or inflammatory bowel disease, highlighting that the identification of health- and disease-associated phages in the human gut needs to move beyond those that are abundant enough to be found by virome sequencing.

Funding Statement

This study was supported by the National Institutes of Health (NIH) Human Virome Project grant [U01 AT012998] (S.A.H., M.T.B.) and NIH Computation and Experimental Resources for Virome Analysis in Inflammatory Bowel Disease grant [RC2 DK116713] (M.P., S.A.H.). This work and the collection of human stool samples was supported by the NIHR Cambridge Biomedical Research Center. M.T.B. was also supported by the Kenneth Rainin Foundation and the Crohn’s and Colitis Foundation (CCF) Litwin IBD Pioneers Award [#1065897]. D.E.C was supported by the CCF Research Fellowship Award [#935619] and the NIH [T32 DK077653-29]. D.L. was supported by NIH [T32 HG000045]. J.S.W. was supported by NIH T32 AI007172. J.M.M. was supported by NIH [T32 AI106688 and T32 DK077653].

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

All shotgun sequencing reads not included in Lawrence, et al. (2022) and assembled phage sequences for this project are available at European Nucleotide Archive (ENA) at EMBL-EBI under accession number PRJEB86846.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19490976.2025.2526719