Pre-experimental test of fecal fractionating process
Mock community
We used an exogenous community of five known phages (Φ29, T4, c2, Φ6, ΦX174) to estimate the active phage concentration in each step of the fractionating process prior to fecal separation of the intervention material. The community was made by propagating the phages on their respective hosts [32] to a concentration of more than 107 plaque-forming units (PFU) per mL, then diluting in SM buffer (200 mM NaCl, 10 mM MgSO4, 50 mM Tris-HCl, pH 7.5) with 20% glycerol and mixing to get one mock solution with a concentration between 104 and 107 PFU/mL for each phage.
Fractionating process
The separation of viruses from the residual fluid was done by; 1) pelleting larger particles and bacteria, 2) filtering off remaining bacteria, and 3) concentrating the viral part using an ultrafiltration device (Fig. 1A). This was based on previous protocols [32, 33]. 2 g porcine feces were dissolved in 17 mL SM buffer along with 1 mL of mock community in a sterile 100 mL filter bag (BagPage, Interscience, Saint Nom la Brétèche, France) and homogenized for 2 minutes using a stomacher (Stomacher 80, Seward, Worthing, UK). The 20 mL slurry was transferred to a centrifuge tube and centrifuged at 4,200 × g for 30 minutes at 4°C. The supernatant was filtered through a 0.45 µm PES membrane filter (Filtropur S, Sarstedt, Helsingborg, Sverige) into a clean tube and diluted to the original volume of 20 mL with SM buffer (FFT, fecal filtrate). Viruses in the filtrate were then concentrated by repeatedly adding sample to a 30 kDa ultrafiltration device (Vivaspin 20, PES, Sartorius; Göttingen, Germany), centrifuging at 1,500 × g, and collecting the virus-free residual (RES). Once the entire sample had passed through the filter, the retention chamber was filled with 20 mL SM buffer and stored at 4°C to release virus particles from the filter membrane (VIR).
Phage enumeration
The fractionating process with mock phages was reproduced in triplicates. The next day, spot and plaque assays were performed to estimate the concentration of infective viruses in each fraction. The mock phages in FFT and VIR were enumerated using spot assays of dilutions from 100 to 10− 5. The mock phages in RES were enumerated using plaque assays without dilution. The laboratory methods are described in detail by Larsen et al. [32].
Donor collection and fecal separation
Two batches of donor material were investigated. Batch 1 was collected in 2014 and originated from five littermate pigs from a conventional herd and was stored at -80˚C. Batch 2 was collected in 2021 from four pigs from four different herds, including the herd from which Batch 1 was taken. All pigs were 10-day-old, sow-reared, and only received antibiotics on the first day of life. The pigs were euthanized and content was collected from the cecum, colon, and rectum. Gut content from the pigs within each batch was pooled and diluted 1:1 in 20% sterile glycerol.
The fecal separations were conducted in 2021. Batch 1 was thawed and Batch 2 was processed directly after collection. The feces were diluted in SM buffer to a working concentration of 0.05 g/mL. After homogenization and centrifugation (4,200 × g for 30 min at 4˚C), the above-described fractionating process was done to produce FFT, VIR, and RES fractions for the in vivo experiment. All three fecal solutions from each batch were divided into aliquots and stored at -80˚C until use.
Count of virus-like particles
To estimate the concentration of virus-like particles (VLPs) in each fecal solution, aliquots of 10 µl were stained with SYBR™ Gold (Thermo 328 Scientific, cat. no. S11494) as described previously [34]. Hereafter, the samples were mounted on a 0.02 µm filter (Whatman Anodisc, Cytiva, USA) and assessed at 1000 × magnification under an epifluorescence microscope fitted with a 490 nm filter block. Ten representative pictures were captured with a Photometrics CoolSNAP camera and VLPs enumerated using the ImageJ software [35]. The number of VLPs per mL was calculated using the equation: mean number of VLPs per image × number of images on filter / sample volume on filter. The detection limit of the method was 2.5 × 105 VLPs/mL.
Preterm pig experiment
Animal experimental procedures and monitoring
Sixty-eight crossbred piglets ((Landrace × Yorkshire) × Duroc) were delivered by cesarean section from four sows at 90% gestation (106 days) as described previously [36]. After delivery, the pigs received respiratory and circulatory support while recovering from anesthesia. When stabilized, the pigs were fitted with orogastric tubes for enteral feeding and umbilical arterial catheters for parenteral nutritional support. Hereafter, the pigs were stratified by sex and birth weight and randomly allocated to four groups (Fig. 1B) receiving fecal filtrate transfer (FFT), isolated viruses (VIR), virus-depleted filtrate residual (RES), or SM buffer (control, CON). All pigs were housed individually in preheated incubators and monitored closely.
As a substitute for the normal passive immunization with colostrum, the pigs received maternal plasma by arterial infusion (16 mL/kg) on the first day of life. Hereafter, the pigs were supplemented with declining levels of intraarterial parenteral nutrition (PN) during the study period (2–4 mL/kg0.75/d, Kabiven, Vamin, Fresenius-Kabi; Bad Homburg, Germany). Every third hour, the pigs were fed enteral boluses of infant formula with increasing volumes (24, 40, 64, and 96 mL/kg0.75/d on days 1, 2, 3, and 4–5, respectively). The enteral formula composition is presented in Additional file 1: Table S1. The pigs were weighed once daily and every third hour, they were assessed by experienced personnel for clinical status and fecal output. Fecal characteristics were scored as 1) firm feces, 2) pasty feces, 3) droplets of watery feces, 4) moderate amounts of watery feces, and 5) large amounts of watery feces. Score 3–5 were considered diarrhea and the exact time of first diarrhea episode was registered. If pigs presented with clinical signs of NEC, they were euthanized.
Transfer of fecal solutions
Pigs from the first litter were inoculated with donor material from Batch 1 and the following three litters were inoculated with Batch 2. The FFT and VIR pigs were inoculated with a total dose of 1010 VLPs/pig, divided into administrations twice daily on days 1 and 2 after birth. The remaining two groups received equivalent volumes of RES or CON solution (SM buffer). All were administered via the orogastric tube and flushed with 1 mL of sterile water.
Euthanasia, tissue collection, and scoring of intestinal lesions
On day five, all animals were tissue collected. The piglets were anesthetized, and blood samples were taken by cardiac puncture followed by euthanasia with an intracardiac injection of pentobarbital. Blood samples were subjected to routine hematology analysis (Advia 2120i Hematology System, Siemens). Visceral organs were removed and weighed. The stomach, small intestine, and colon were assessed for macroscopic lesions of hyperemia, hemorrhage, necrosis, and pneumatosis intestinalis (transmural gas) by an experienced person blinded to the investigation. A gross pathology score was given for each intestinal segment according to the weighted and cumulative point system described in [37]. NEC was defined as the presence of moderate hemorrhage or presence of pneumatosis intestinalis, and/or necrosis. Colon tissue with luminal content was snap-frozen for analysis of bacterial and viral composition. A 10 cm section of distal ileum was inverted, washed in sterile saline, and blotted to remove remaining fluid. Mucosa tissue was then scraped off and snap-frozen for later determination of brush-border enzyme activity (lactase, maltase, sucrase, aminopeptidase N, aminopeptidase A, and dipeptidyl peptidase IV), as previously described [38]. Lastly, the worst macroscopic lesions from the SI and colon were snap-frozen or immersed in 4% paraformaldehyde for later cytokine analysis and histological assessment, respectively.
Lesion histological analyses and cytokine measurements
Formalin-fixed SI- and colon lesion biopsies were embedded in paraffin and two serial 3–5 µm sections were made. One was hematoxylin and eosin stained prior to histopathological assessment by an experienced investigator blinded to the investigation. Each segment received a microscopic lesion score of 0–8 as described previously [37]. Another was used for goblet cell visualization by Alcian blue and Periodic acid-Schiff (PAS) staining. The goblet cell density was calculated as the area of PAS staining relative to the total area of the tunica mucosae using the Fiji software [39]. Interleukin (IL) 8, IL-1β, and tumor necrosis factor (TNF) α levels were analyzed in SI- and colon lesion homogenates by commercial porcine ELISA kits (R&D Systems, Abingdon, Oxfordshire, United Kingdom) according to the manufacturer's instructions and expressed as picograms per milligram of wet tissue [40]. TNF-α levels were below the detection limit in all samples.
Pathogen detection in donor material
The donor material was subjected to a wide range of pathogen detection by multiplex high-throughput real-time qPCR amplification, using the BioMark HD and 192.24 dynamic array integrated fluidic circuit system (Fluidigm, CA, USA) as previously described by [41] and [42]. The following bacterial and viral pathogens were measured in Batch 1 and 2 pooled feces as well as single donors and fecal fractions from Batch 2: E. coli F4 and F18, Lawsonia intercellularis, Brachyspira pilosicoli, porcine circovirus 2 and 3, porcine parvovirus, influenza A virus, and rotavirus group A, B, C, and H.
Bacterial DNA extraction, sequencing, and raw data processing
Fecal filtrate (FFT) and fractions (VIR, RES) from Batch 1 and Batch 2, and single donor fecal aliquots were thawed, mixed with 5 mL SM buffer, and homogenized on a shaking board at 300 rpm for 10 minutes. Due to NEC, it was not possible to extract content from several recipient colons. Instead, a section of colon tissue with content was collected from each recipient and homogenized with 5 mL SM buffer in Stomacher 80 Biomaster Lab Blender (Seward, Worthing, UK) for two minutes to get as much content out as possible. The donor and recipient slurries were then centrifuged at 5,000 × g at 4 ºC for 30 minutes. The pellets were collected for bacterial DNA extraction and the supernatants for viral DNA/RNA extraction. The bacterial DNA were extracted from pellets with the DNeasy PowerSoil Pro Kit (Qiagen, Germany) following the manufacturer's instructions. The final DNA products were stored a -80 ºC and their concentrations were measured with Qubit™ 1x dsDNA high sensitivity kit on Varioskan Flash (Thermo Fisher Scientific, USA). Gut microbial composition was assessed by sequencing a near full-length 16S ribosomal RNA (rRNA) gene amplicon using GridION (Oxford Nanopore Technologies, Oxford, UK), as described previously [43]. The data was generated using GridION and collected using MinKNOW software v22.10.7 (Oxford Nanopore Technologies, Oxford, UK). Raw FAST5 was base-called to FASTQ (Oxford Nanopore Technologies, Oxford, UK) using the Guppy v6.2.8 base-calling toolkit. The Long Amplicon Consensus Analysis pipeline generated the abundance table from the raw FASTQ files [44]. The SILVA database was used to assign taxonomy to the quality corrected reads.
Viral DNA/RNA extraction, and viral metagenomic sequencing
The donor and recipient supernatants were filtered through a 0.45 µm PES membrane filter (Filtropur S, Sarstedt, Helsingborg, Sverige) and then concentrated by centrifugation at 1,500 × g at 4ºC using Centrisart ultrafiltration devices with a filter cut-off at 100 kDa (Sartorius, UK) [45]. Viral DNA/RNA was extracted from fecal supernatant with Viral RNA mini Kit (Qiagen, Germany) as previously described [46]. Reverse transcription was performed using SuperScript VILO Master Mix according to the manufacturer's instructions, followed by purification using DNeasy Blood and Tissue Kit (Qiagen, Germany), only following steps 3 to 8. Multiple displacement amplification (MDA, to include ssDNA viruses) using the GenomiPhi V3 DNA Amplification Kit (Cytiva, USA). The sequencing library preparation used the Nextera XT Kit performed as previously described [46], sequenced using the NovaSeq platform (NovoGene, China). Viral operational taxonomic unit (OTU) tables were generated using a modified version of the Vapline v2.0 [47]. In short, raw reads were quality controlled, trimmed and dereplicated using trimmomatic, BBMap, and seqkit [48–50]. Reads were then paired and assembled within-samples using SPAdes v3.13.0 using both paired and unpaired reads [51]. Contigs shorter than 2200 bp were removed and the remaining contigs were assessed by checkV, VirSorter2, and VIBRANT[52–54]. Contigs identified as viral in at least one of the three softwares were classified as viral contigs and used for the continued analysis, while the rest were discarded. Viral contigs were assigned taxonomies using mmseqs2 to align sequences with a custom database based on the VOGDB (v217), and hosts were predicted using iPHoP [55]. The vOTU table is generated by assigning raw reads from each sample to viral contigs using samtools [56].
Cell stimulation assay
The human monocyte cell line THP-1 was used to characterize the cellular innate immune response following challenge with an Escherichia coli bacterium, an isolated phage cocktail, and fecal fractions from Batch 2. The E. coli strain (ST-2064) was cultured in a liquid Luria-Bertani medium (BD Difco™ LB Broth Lennox, Becton Dickinson, USA) until it reached the exponential phase and adjusted to a concentration of 108 colony forming units (CFU) per mL. The phage cocktail was composed of 10 lytic phages, targeting Enterobacter and E. coli bacteria (further described in Additional file 1: Table S2). Each phage was propagated on its respective host followed by 0.45 µm filtration (Filtropur S, Sarstedt, Germany). The phage solutions were mixed to reach a concentration of 5 × 109 plaque forming units (PFU) per mL total, as described by Larsen et al. [32].
Before the challenge, endotoxins were removed from all bacterial, phage, and fecal solutions by running them on Pierce™ High Capacity Endotoxin Removal Spin Columns (0.5 mL, ThermoFisher Scientific, USA), according to the manufacturer's instructions. The cell cultivation procedure adhered to the method described by Spiegelhauer et al. [57], though, the cells were differentiated with 5 ng/mL phorbol 12-myristate 13-acetate (PMA) (ThermoFisher Scientific, USA) instead of 50 ng/mL. After differentiation, the medium was replaced with RPMI Glutamax and 10% FBS, and the cells were further incubated for 1 hour before challenge. 100 µL of FFT, VIR, RES, CON, phage cocktail (Phages), or E. coli bacterial solution were added to the respective well. The plate was then centrifuged for 5 min at 300 rpm and incubated for 20 hours. Seeding and challenges were performed in triplicates (cell passages 17, 19, and 20).
RNA was extracted from the cells with Qiagen RNeasy Mini Kit (Qiagen, Germany) and 10 ng RNA was reverse transcribed to cDNA with the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Massachusetts, USA) according to the manufacturer's protocol (MultiGene Optimax, LabNet International, USA). Quantitative PCR (qPCR) was performed in duplicates in 10 µL volumes (LightCycler 480SYBR Green Master, Roche Holding AG, Switzerland) with 1 × SYBR green and 0.4 µM forward primer and 0.4 µM reverse primer (TAG Copenhagen, Denmark). The cycling conditions were as follows: 95°C for 15 minutes, 45 cycles of (94°C for 15 seconds, 58°C for 30 seconds, 72°C for 30 seconds), 72°C for 10 minutes (LightCycler 480 II from Roche, LightCycler 480 Software release 1.5.1.62 SP3). Genes and primers are listed in Additional file 1: Table S3.
Statistical analysis
Statistical analyses were performed in R version 4.3.2. Gross pathology and microscopic lesion scores were analyzed using a proportional ordered logistic regression model (R package MASS) with group, sex, litter, and birth weight as fixed effects. The group effect was analyzed using the likelihood ratio test with Benjamin-Hochberg (BH) correction. Remaining continuous data were analyzed using linear mixed models with group, sex, and birth weight as fixed effects and litter as random effect. The group effect was analyzed using ANOVA and Tukey’s test for pairwise comparisons. Tissue cytokines were analyzed with a Kruskall-Wallis test followed by Dunn's test with BH correction. Time to the first diarrhea episode was analyzed with pairwise Log-Rank tests with BH adjustment (R package survminer). Correlation between time to first diarrhea and intestinal gross pathology (log1p tansformed) was done using Pearson's correlation coefficient. Changes in body weight over time were analyzed using a linear mixed model, which included day, group, sex, birth weight, and litter as fixed effects and which further assumed an unstructured covariance pattern to account for repeated measurements on the same subjects (R package LMMstar). Body weights were transformed to natural logarithms before analysis and hence estimated differences are expressed as the relative difference from the median birth weight across groups. P-values below 0.1 were considered tendencies and P-values below 0.05 were considered statistically significant.
Microbiome statistical analyses
Microbiome results were visualized using the R package ggplot2, unless other stated [58]. R package Decontam performed the decontamination with a threshold of 0.5 [59]. R package metagenomeSeq performed cumulative sum scaling (CSS) normalization [60]. PERMANOVA (adjusted by False Discovery Rate (FDR)), in the Adonis function in the R package vegan, was used for beta diversity comparisons [61]. The Wilcoxon test was used for both alpha diversity and Bray-Curtis dissimilarity within groups adjusted by FDR. DESeq2 was used to identify differentially enriched microorganisms at summarized species level for bacteria and at OTU level for viruses [62]. This was visualized in volcano plots using the R package EnhanceVolcano with P-value cut-off 0.05 and log2 fold change cut-off 0.6 [63]. Spearman correlation analysis was used to determine the correlation among viral OTUs and clinical variables with P-values adjusted by FDR. The correlation network was visualized by R packages igraph and ggraph [64, 65].