Phage-microbe dynamics after sterile faecal filtrate transplantation in individuals with metabolic syndrome: a double-blind, randomised, placebo-controlled clinical trial assessing efficacy and safety – Nature Communications

Study design

We set up a prospective, double-blinded, randomised, placebo-controlled intervention study that was performed in our academic hospital, the Amsterdam University Medical Centres location AMC in the Netherlands. After passing screening, 24 subjects with MetSyn were randomised to receive a sterile FFT from a lean healthy donor or a placebo transplant. Prior to the intervention and after 28 days at follow-up, subjects underwent an OGTT to assess their glucose metabolism. In addition, a week prior to one week after intervention, subjects monitored their blood glucose using a flash glucose monitoring device (Freestyle Libre). Faecal samples were collected at multiple timepoints between baseline and follow-up to study dynamic changes in the microbiome. Finally, during every study visit a medical exam was conducted in addition to blood plasma collection to assess the safety of the intervention. Figure 1a provides a schematic overview of the study.

Study subjects

Study participants were all European Dutch, overweight (body mass index (BMI) ≥ 25 kg/m²) subjects between 18 and 65 years of age and had to meet the National Cholesterol Education Program (NCEP) criteria for the metabolic syndrome³¹. Both male and female participants were included in the study and sex was self-reported. Main exclusion criteria were the use of any medication, illicit drug use, smoking, or alcohol abuse in the past 3 months, as well as a history of cardiovascular, gastrointestinal, or immunological disease. Supplementary Table 3 summarises all in- and exclusion criteria and Supplementary Fig. 1 provides and overview of the recruitment of study participants.

Donor screening

Faeces donors were lean healthy European Dutch subjects who were thoroughly screened according to the guidelines of the European FMT Working Group⁵⁵. Screening of potential donors was performed in a stepwise manner as previously published³². Briefly, potential donors first completed an extensive screening questionnaire. If they passed this stage, their faeces were screened for pathogenic parasites. When negative, several faecal samples were screened for presence of pathogenic bacteria, viruses and multidrug resistant organisms (MDROs), as well as the level of calprotectin. Donors screened after May 2020 were additionally screened for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2)⁵⁶. In addition, blood was collected for serological testing and to screen for an abnormal liver or renal function or an impaired immunity. When donors passed this screening, they were allowed to donate faeces for a period of 6 months. Supplementary Table 4 lists the specific in- and exclusion criteria for faeces donors. Every two months, active donors underwent a short rescreening, which included, among others, screening for MDROs and SARS-CoV-2. In addition, before every donation, donors had to complete a shortened questionnaire to confirm their eligibility. We matched donors and recipients based on their sex and whether they have had a prior infection with cytomegalovirus or Epstein–Barr virus.

Sterile faecal filtrate production and administration

Production of the sterile faecal filtrate started the day before administration to the MetSyn subjects. First, 50 g of stool was collected from a screened donor, which was homogenised with 500 ml sterile saline. Large particles were filtered from the faecal suspension using double sterile gauzes. Most of the bacteria were removed in two subsequent centrifugation steps, in which the suspension was spun for 1 h at 10.000×g. Finally, the supernatant was filtered through a sterile 0.2 μm membrane using a tangential flow filtration device (Vivaflow 50). Production of the filtrate from donor stool was performed within 6 h and took, on average, 334 min (SD = 27). The filtrate was stored overnight in a fridge until administration. The production is depicted in Supplementary Fig. 3A.

The sterile faecal filtrate was administered to the patient via a nasoduodenal tube. The day prior to the administration, subjects were asked to clean their bowel using a laxative (Klean-Prep®, Norgine B.V.), which is a standard pre-treatment for FMT procedures in our hospital. Nasoduodenal tubes were placed with the help of a Cortrak®2 enteral access system (Avanos Medical Inc.), making sure the nasoduodenal tube was correctly positioned. The faecal filtrate was slowly infused with a 60 ml syringe, on average 300 ml during a 15-20 min period. Supplementary Fig. 3B provides a schematic overview of the FFT procedure.

During the optimisation of the tangential flow filtration, we quantified the VLP numbers of the faecal filtrates from four different donors, as previously described^57,58. Briefly, faecal filtrates were concentrated, from which VLPs were isolated with caesium chloride density gradient centrifugation, stained with SYBR Gold and counted by epifluorescence microscopy. Faecal filtrates contained on average 1.25E + 08 VLPs/ml (SD 0.45E + 08), which is in line with previous publications^59,60. We confirmed the absence of bacteria from the faecal filtrate with a qPCR for the bacterial 16 S rRNA gene as previously described⁶¹, showing a 10⁵-fold decrease in bacterial DNA (Supplementary Fig. 3C). We further confirmed this by culturing of the faecal filtrate using Biosart® 100 monitors (Sartorius). 100 ml of faecal filtrate was filtered and the cellulose nitrate membranes were incubated on petri dishes with Columbia agar + 5% sheep blood (bioMérieux) for two days at 37 °C under both aerobic and anaerobic conditions. We did not observe any colony-forming units in 100 ml of faecal filtrate.

Outcomes

The primary outcome was change in glucose metabolism, as determined by the AUC for glucose excursion during the OGTT. Secondary outcomes related to glucose metabolism were changes in fasting glucose, insulin, HOMA-IR and HbA1c between baseline and follow-up after 28 days, as well as changes in glucose variability measured by CGM a week before and after intervention. Other secondary outcomes were the dynamic changes in gut bacteriome and virome populations following FFT or placebo intervention and the comparison of phage composition between lean donors and subjects with MetSyn. Finally, we assessed the safety of the FFT as determined by the occurrence of (serious) adverse events, physical exam and several blood parameters for renal and liver function and inflammation.

Sample size calculation

Based on previous data from our group in which individuals with MetSyn received an FMT^28,29, and the hypothesis that a faecal phage transplant can be equally effective as a traditional FMT^25,26,27,30, we assumed a 15% improvement in glucose tolerance upon FFT. With a two-sided 5% significance level and a power of 80%, a sample size of 12 patients per group was necessary, given an anticipated dropout rate of 10%. To recruit 24 individuals with MetSyn, we anticipated a 12-month inclusion period.

Randomisation

Data were captured with electronic case report forms build in Castor EDC (v2019.2.0-2020.2.25)⁶². In CASTOR, subjects were randomly assigned to an intervention by block randomisation with stratification for age and sex, and block sizes of 4, 6 and 8. The day prior to the intervention, both the faecal filtrate and placebo (sterile saline with brown colour) were prepared and stored overnight. Both the faecal filtrate and placebo looked identical. A randomisation assistant unblinded for the treatment allocation prepared the correct solution for administration and destroyed the other. The investigator administered the allocated treatment in blinded syringes and through an opaque nasoduodenal tube, making sure both participants and the investigator were blinded for the intervention throughout the study.

Oral glucose tolerance test and biochemical measurements

For the OGTT, overnight fasted subjects ingested a standardised glucose solution (75 g). Blood was drawn from an intravenous catheter at baseline and 15, 30, 45, 60, 90 and 120 min after ingestion. Both blood serum and plasma were aliquoted and stored at −80 °C. From these aliquots we measured glucose and C-peptide, which was performed by the Endocrinology department of the Amsterdam UMC. In addition, blood samples collected at baseline and follow-up were used to measure fasted glucose, insulin, HbA1c and the clinical safety parameters for renal/liver function and inflammation, all measured by the Central Diagnostics Laboratory of the Amsterdam UMC. Results were reported in the electronic medical record software from EPIC (versions August 2019-November 2020).

Continuous glucose monitoring

To reduce the study burden and prevent daily finger pricks, we used a continuous glucose monitoring device (Freestyle Libre 1) to monitor blood glucose, which allowed subjects to perform all normal activities while wearing the sensor. Subjects were taught to subcutaneously implant the CGM sensor and were instructed to extract the data from the sensor at least every 8 h. One week prior to the intervention subjects started to monitor their glucose until one week after the intervention. Compliance among participants was good, with a median 100% (range 76–100%) of data correctly collected, during a median period of 14 (range 11–27) days with a median 1350 (range 1043–2617) sensor readings. During that same period, participants were asked to record their diet using an online food diary (Eetmeter v2019-2020) from the Voedingscentrum)⁶³. At the follow-up visit, data from the CGM scanner were exported and analysed with the previously published CGDA package v0.8.2 for CGM data analysis⁶⁴.

Faeces collection

The day before the intervention and 2, 4, 7, 14 and 28 days thereafter, subjects were asked to collect several faecal samples. Faeces were collected by participant in stool collection tubes, which were directly stored in a freezer at home inside a safety bag. In addition, participants registered the time, date and consistency of the collected faeces according to the Bristol Stool Chart. At the baseline and follow-up visits, these faecal samples were transported to the hospital frozen, where they were directly stored at −80 °C until the end of the study.

Bacteriome and virome sequencing

To study the bacteriome and virome, we performed whole genome shotgun (WGS) sequencing. From the stored frozen faeces samples, total genomic DNA was extracted using a repeated bead beating method as described previously³³. Briefly, 250 mg of faeces were weighed in bead-beat tubes, 700 µl of S.T.A.R. buffer (Roche Diagnostics Cat# 03335208001) was added, and samples were homogenised three times using a bead-beater (FastPrep-24™, MP Biomedicals™) set to 5.5 ms for 1 min. Lysates were incubated at 95 °C for 15 min centrifuged for 5 min at full speed (14,000×g) at 4 °C, and supernatant was transferred to nuclease-free tubes. The above was repeated once with 300 µl of S.T.A.R. buffer to extract any remaining DNA. DNA was further cleaned from the lysates using the Maxwell® RSC Blood DNA Kit (Promega Cat#ASB1520) according to manufacturer’s instructions. Extracted DNA was stored at −20 °C and shipped on dry ice to Novogene (Cambridge, United Kingdom). Libraries for shotgun metagenomic sequencing were prepared using the NEBNext Ultra II Library prep kit (New England Biolabs Cat#E7645L) and sequenced on an Illumina HiSeq instrument with 150 bp paired-end reads and 6 Gb data/sample. Supplementary Figure 3D summarises the sequencing and bioinformatics pipeline used. For both the WGS and VLP sequencing (see below) negative controls were included to check for contamination during DNA extraction and library prep. These negative controls did not yield any measurable DNA after library prep and were therefore not sequenced. No mock communities were included as positive controls in the current sequencing pipeline.

VLP sequencing

To study phage virions, we isolated the faecal VLP fraction and sequenced dsDNA phages as previously described¹⁹. Briefly, the VLPs were extracted from 500 mg of faeces using high-speed centrifugation followed by filtration through a 0.45 µm membrane. Any free-DNA debris was digested prior to lysing the VLPs, whereafter the DNA was purified using a two-step phenol/chloroform extraction protocol. Finally, the DNA was purified using the DNeasy Blood&Tissue kit (Qiagen Cat#69506) according to the manufacturer’s protocol. Library preparation was done with the NEBNext Ultra II FS DNA library prep kit (New England Biolabs Cat#E7805L) and the NEBNext Multiplex Oligos for Illumina dual indexes (New England Biolabs Cat#E7600S) according to manufacturer’s instructions. Quality and concentration of the VLP libraries were assessed with the Qubit dsDNA HS kit (ThermoFisher Cat#Q32854) and with the Agilent High Sensitivity D5000 ScreenTape system (Agilent Technologies). Libraries were sequenced using 2×150 bp paired-end chemistry on an Illumina NovaSeq 6000 platform with the S4 Reagent Kit v1.5, 300 cycles (Illumina Cat#20028312) at the Core Facility Genomics of the Amsterdam UMC.

Sequence assembly

Sequencing resulted in an average of 21.7 ± 3.5 M reads per WGS sample (median: 22.4 M reads), and 23.6 ± 18.3 M per VLP sample (median: 18.1 M reads). Before assembly, reads belonging to the same participant were concatenated. Adapter sequence removal and read trimming were performed with fastp v0.23.2 (option –detect_adapter_for_pe)⁶⁵. As previously recommended⁶⁶, reads were then error corrected with tadpole (options mode=correct, ecc=t, prefilter=2), and deduplicated with clumpify (options dedupe=t, optical=t, dupedist=12000), both from bbmap v38.90 [https://jgi.doe.gov/data-and-tools/bbtools]. High-quality reads from WGS samples were then cross-assembled per participant using metaSPAdes v3.15.5⁶⁷ (option–only-assembler). Due to their great complexity, we were unable to assemble some of the VLP samples. We thus assembled these with MEGAHIT v1.2.9⁶⁸, which we did for all VLP samples to keep methodological consistency.

Viral sequence recognition and clustering

To identify viral sequences among the WGS and VLP assemblies, contigs longer than 5000 bp were analysed with virsorter v2.2.3⁶⁹ (option –exclude-lt2gene) and checkv v1.0.1⁷⁰. Contigs were taken to be of viral origin if at least one of the following criteria was true: checkv identified at least one viral gene, VirSorter2 gave a score of at least 0.95, VirSorter2 identified at least 2 viral hallmark genes, checkv identified no viral or bacterial genes. In total, we selected 53,204 contigs with at least 1 viral gene, 782 with a virsorter2 score of > 0.95, and 1 with at least 2 viral hallmark genes. The resulting viral sequences were then deduplicated at 100% with bbdupe from bbmap v38.90 (option minidentity=100). This resulted in a non-redundant database of 50,724 viral contigs, which were subsequently clustered at 90% average nucleotide identity (ANI) into viral populations (VPs) using blastn all-vs-all searches with BLAST v2.12.0+⁷¹. The longest contigs in each VP were further clustered into viral clusters (VCs) by vContact2 v0.11.3⁷². Since the conclusions of the analyses were identical regardless of whether they were performed with VPs or VCs, only VP-level analyses were reported.

Viral read depth determination

Viral relative abundance was determined by mapping high-quality reads from each sample (i.e., one mapping per participant and time-point) against non-redundant viral sequences with bowtie2 v2.4.2⁷³. Following earlier recommendations⁷⁴, contigs were considered to be present if at least 75% of their bases were covered by at least 1 read mapped with over 90% ANI. To determine this, reads mapping with less than 90% ANI were removed from alignments with coverm filter v0.6.1 (option –min-read-percent-identity 90 [https://github.com/wwood/CoverM]), and coverage was determined with bedtools genomecov v2.27.1⁷⁵ (option -max 1). Read counts per contigs were then determined with samtools idxstats v1.15.1⁷⁶, and those with a horizontal coverage of <75% were set to zero. Read counts and contig lengths were summed per VP, and reads per kilobase per million mapped reads (RPKM) values were calculated to take differential contig lengths.

Bacterial community profiling and binning

Bacterial population compositions of WGS samples were profiled per participant and time point with mOTUs v3.0.3⁷⁷. Binning contigs into metagenome assembled genomes (MAGs) was done per participant. First, high quality reads from each time-point were mapped to cross-assembled contigs of at least 2500 bp with bowtie2 v2.4.2. Read depth tables were then constructed with jgi_summarize_bam_contig_depths v2.15, and contigs were binned with metabat2 v2:2.15⁷⁸. Completion and contamination of putative MAGs were then determined using checkm v1.2.1⁷⁹ and, like was previously done⁸⁰, MAGs were considered for further analysis if completeness – (5 x contamination) was at least 50. Taxonomy of such MAGs was determined with GTDB-Tk v2.1.1⁸¹ using the R207-v2 database package. This resulted in a database of 3011 MAGs with an assigned taxonomy.

Determining phage-host links

Viral sequences were linked to bacterial MAGs in two ways. Firstly, if a viral contig was contained within a MAG, it was considered to be a prophage. Secondly, viral contigs were linked to MAGs using CRISPR spacer hits. For this, CRISPR spacer arrays were identified among MAGs using CRISPCasFinder v4.2.20⁸². CRISPR spacers between 20 and 30 bp in length were then matched to viral contigs through a blastn search with BLAST v2.12.0+ (options -task blastn-short). Spacer hits were finally filtered for those with 2 or fewer mismatches, minimising the risk of spurious hits.

Statistical analyses

All statistical analyses were performed in R v4.2.1. Richness, α-diversities, principal component analysis (PCA), and principal response curves (PRC) were all calculated with the vegan R package v2.6-4⁸³. For richness and α-diversity RPKM values were used, while PCAs and PRCs used centred log ratio (clr)-transformed data so as to account for the compositionality of the data⁸⁴. Before clr-transformation, VPs of low abundance and prevalence were removed by removing those with total RPKM of <100 over all samples, as well as those with RPKM values of >20 in less than 10% of samples. Significance levels of PCAs were calculated with a permutational analysis of variance (PERMANOVA) test, as implemented in the vegan R package v2.6-4 and were controlled for age and sex. For the PRC-analysis, the permutest function was used to calculate significance. Both PERMANOVA and permutest used 1000 permutations. p-values were adjusted for multiple testing using the Benjamini–Hochberg approach where necessary. General linear models were constructed with the glmmPQL function from the MASS R package v7.3-58.1 with the age, sex, day, group and day:group as fixed effects and participants as random effect.

Differential abundance

Differential abundance of VPs among VLP samples on day 2 was determined with ANCOM-BC v1.2.2⁸⁵. Input of ANCOM-BC consisted of the raw read counts summed per VP in each sample, because this method has its own internal data normalisations to account estimated sample fractions. ANCOM-BC was run on VPs with at least 20 reads reported in at least 10% of samples. To account for the relatively small sample sizes, structural zero discovery was turned on but the usage of the asymptotic lower bound turned off⁸⁵. Differential abundance was corrected for the effects of age and sex. ANCOM-BC used multiple-testing correction according to the Benjamini-Hochberg method, with a significance cutoff of 0.05. The number of differentially abundant (DA) VPs was then determined per host species. Enrichment of host species among DA VPs was calculated using a hypergeometric test as implemented in the phyper R function, with the number of DA VPs infecting a given species-1 as q, the total number of VPs in the dataset infecting the same species as m, the total number of VPs with host-m as n, the total number of DA VPs as k, and lower.tail set to FALSE.

Phage-host interactions

To determine the dynamics of phage-bacterium interaction across the entire population, the change in relative abundance between days 0 and 2, 2 and 28 and 0 and 28 were determined for all VPs with a host and all MAGs with a known phage. The resulting values were then averaged for both VPs and MAGs at the species level, after which Spearman correlation coefficients were calculated.

Patient and public involvement statement

Patients were involved in the assessment of the grant proposals for this study by the Dutch Diabetes Research Foundation (Diabetes II Breakthrough grant (459001008) and Senior Fellowship (2019.82.004)). Moreover, the patient panel advised on the patient burden of the clinical study. In addition, patients were involved in the ethical approval of this study (as part of the ethics committee). Once the trial results became available, participants were informed of the results with a letter suitable for a non-specialist audience.

Ethics approval and informed consent statement

This study involves human participants and was approved by the Medical Research Ethics Committee Academic Medical Center Amsterdam (METC 2018_231). Both participants and faeces donors gave informed consent to participate in the study before taking part. The study was registered at the Dutch National Trial Register (NTR) under NL8289 on the 15^th of January 2020, while the first patient was included in October 2019. The delay in registration was due to a miscommunication between investigators. When this mistake came to light during the first monitor visit after the first three patients had been included, the study was directly registered at the NTR. This registry does not exist anymore and all data has been added unaltered to the Dutch Trial Register (LTR) under https://clinicaltrialregister.nl/en/trial/26916. While these data are automatically included in the International Clinical Trial Registry Platform (ICTRP), thereby fulfilling the requirement of prospective registration as required by the International Committee of Medical Journal Editors (ICMJE), it was unfortunately no longer possible to adjust the data.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.