Ability of a Microbial Supplement to Restore Gut Microbiome Metabolism After Alcohol or Antibiotic Disruption

We conducted a two-phase experiment using a simulated gut ecosystem inoculated with a complex gut microbial community from a healthy donor (see Materials and Methods). In the first phase (Fig. 1A, test 1), a fecal sample was used to inoculate nine parallel reactors (three conditions with three replicates per condition). Fermentors were incubated for 48 h with either 50 μg/mL metronidazole and 30 μg/mL ciprofloxacin to mimic antibiotic treatment or 30% (vol/vol) commercial vodka to mimic alcohol consumption or were left untreated as a control. In the second phase (Fig. 1A, test 2), we assessed the recovery of the three luminal microbial communities (untreated control, antibiotic treated, and alcohol treated) by reinoculation into fresh reactors with and without the addition of a synthetic microbial consortium. In total, five conditions were tested in this second phase, each in triplicate (antibiotic with or without synbiotic, alcohol with or without synbiotic, and control). Mucin-coated beads, meant to simulate a healthy mucosal microbiota (44), were additionally transferred from the test 1 control reactors to all test 2 reactors.

Both antibiotic and alcohol treatment incubations led to substantial changes in community composition and function. The levels of all tested SCFAs dropped compared to those in control technical replicates, which remained stable throughout the 48-h incubation (Fig. 1B). We observed statistically significant (adjusted P value of <0.05 via a generalized linear model [see Table S1 in the supplemental material]) changes in the relative abundances of different clades (Fig. 1C). We additionally recomputed all models with alternative relative-abundance metrics (see Materials and Methods) and saw similar results (Table S2). There was an overall loss of Proteobacteria and Bacteroidota, while Firmicutes were enriched in the antibiotic treatment and Methanobacteria were enriched in the alcohol treatment. Specific genera that were decreased under both conditions included Parabacteroides, Enterococcus, Dialister, Citrobacter, and Escherichia. Antibiotics uniquely reduced the abundances of Flavonifractor and Bacteroides and increased the relative abundances of various genera, including Lactobacillus, Turicibacter, Streptococcus, and Clostridium, the last of which contains a variety of human pathogens (10). Alcohol supplementation also altered a variety of diverse genera, decreasing, for example, Citrobacter and Lactiplantibacillus.

Treatment with the synbiotic over a 48-h window increased SCFA production (P value of <0.05 for all comparisons described below) (Fig. 2A). By the final time point, total SCFA production in samples treated with dysbiotic stressors (Fig. 1A, test 1) was reduced by 10.7% relative to the uninoculated controls, whereas total SCFA production in samples treated with stressors and synbiotic supplementation (Fig. 1A, test 2) was increased by 5.1%. In samples exposed to a dysbiotic stressor with and without a synbiotic present (Fig. 1A, test 2), supplementation increased butyrate levels by 18.3% in the antibiotic treatment and 29.7% in the alcohol treatment. Synbiotic supplementation increased acetate levels by 18.6% in the antibiotic treatment and 13.4% in the alcohol treatment. Finally, synbiotic supplementation increased propionate levels by 30.4% in the antibiotic treatment and 16.2% in the alcohol treatment.

Lactate production after treatment with dysbiotic stressors was also altered as a function of synbiotic supplementation. After 24 h, only samples treated with the synbiotic consortium showed detectable lactate levels (13.1 to 21.9 mM) compared to the control or non-microbial-treated samples (Fig. 2B). By the end of the incubation, lactate levels had decreased in all samples.

We next compared the impacts of synbiotic treatment on dysbiotic microbial communities. By comparing the overall beta diversity values (Fig. 2C) between experimental treatments, we observed that reactors after treatment with stressors and/or a synbiotic had three discrete microbiome signatures. The first cluster contained samples treated only with stressors and incubated for 48 h (Fig. 1A, test 1, and Fig. 2C, top 3 rows). Samples from test 2 formed the other two clusters, which were associated with the presence or absence of synbiotic supplementation. One of these contained samples treated with stressors and supplemented with the synbiotic, sampled after 48 h of incubation (Fig. 2C, middle 2 rows). The other test 2 cluster contained samples treated with stressors and not supplemented with a synbiotic, sampled after 48 h (Fig. 2C, bottom 3 rows). Note that the microbiome composition of the test 2 control reactor diverged from the microbiome composition of the test 1 untreated control due to the presence of precolonized mucin-coated beads in the test 2 control reactor.

This beta diversity analysis did not demonstrate “recovery” as defined by a return to the initial community composition prior to stress induction for either stressor. Instead, both antibiotic- and alcohol-treated communities treated with synbiotic showed similar endpoint signatures distinct from their dysbiotic states in test 1 (Fig. 2C). The observed convergence in microbiome signatures as a function of treatment with a stressor was driven predominantly by increases in SCFA- and lactic acid-producing species, some of which were members of the synbiotic consortium (Fig. 2D). These included a variety of strains of Bifidobacterium, a genus that contains many acetate producers, and species within the genus Lacticaseibacillus. Out of the 12 species (comprising 24 isolates) present in the microbial consortium, the relative abundances of 11 (species in boldface type in Fig. 2D) were still increased after 48 h in the samples previously exposed to either antibiotics or alcohol. Notably, of these, the abundances of acetate-producing species (Bifidobacterium animalis and Bifidobacterium breve) and lactic acid-producing species (Lacticaseibacillus rhamnosus, Lacticaseibacillus casei, Lactobacillus crispatus, Limosilactobacillus fermentum, Limosilactobacillus reuteri, Ligilactobacillus salivarius, and Lactiplantibacillus plantarum) in the probiotic consortium were all significantly increased in the synbiotic-treated samples (45).

We additionally observed statistically significant, shared decreases in the relative abundance of specific genera or species in both the alcohol and antibiotic treatments (Tables S1 and S2). These included members of the genus Enterocloster, which contains many human pathogens, and Bilophila wadsworthia, which is associated with gas and bloating (46). Finally, alcohol and antibiotics additionally yielded compositional effects specific to each stressor following treatment with the synbiotic. For example, the relative abundances of multiple Faecalibacterium species, known butyrate producers present in the healthy gut microbiome, were significantly increased in the samples treated with both alcohol and the synbiotic.

Notably, we observed an enrichment of bacterial species involved in the transformation of dietary polyphenols into beneficial urolithin compounds (47) for the antibiotic-plus-synbiotic treatment relative to antibiotic treatment alone. There was an increased abundance of Gordonibacter pamelaeae, which is known to produce urolithin A (48). Additionally, the abundance of Ellagibacter isourolithinifaciens, an organism shown to metabolize intermediate urolithin compounds into the more bioavailable beneficial metabolite isourolithin A (49), was increased in the synbiotic supplementation group after antibiotic treatment (Fig. 2D). We note that the only source of exogenous polyphenols added to the reactors derives from the polyphenol-rich Indian pomegranate extract that comprised the prebiotic component of the synbiotic.

Finally, we aimed to determine if synbiotic treatment correlated with the proliferation of any potential pathobionts that were initially repressed by either alcohol or antibiotic usage (Fig. 3A and B). The intent here was to search for evidence of potentially adverse compositional effects that could be induced by treatment. We did not observe this effect. Many species that were initially decreased following treatment with a dysbiotic stressor were further decreased in relative abundance following treatment with a synbiotic. Following antibiotic treatment, for example, organisms that were decreased in relative abundance included Bilophila wadsworthia and many Bacteroides strains. For alcohol, the abundances of many Escherichia coli strains were decreased in both experiments.

Certain bacteria that were reduced following initial exposure to a stressor increased in abundance following synbiotic treatment, indicating at least some form of recovery, but not enough for a complete return to baseline abundances (Fig. 3). Some of these were organisms in the synbiotic itself (e.g., Bifidobacterium infantis). Others, however, like Gordonibacter pamelaeae, various Faecalibacterium species, and Bifidobacterium dentium, were not present in the synbiotic but still increased in relative abundance.

Metronidazole and ciprofloxacin were obtained from Sigma-Aldrich (Bornem, Belgium), while vodka was used as the alcohol (40% [vol/vol] EtOH) source. SH-DS01 capsules containing a 24-strain probiotic mixture encased in an Indian pomegranate (Punica granatum) outer envelope were obtained from Seed Health (Los Angeles, CA, USA). SH-DS01 contains the following strains: Bifidobacterium breve HRVD521-US, Lacticaseibacillus rhamnosus HRVD113-US, Bifidobacterium lactis HRVD524-US, Lacticaseibacillus casei HRVD300-US, Bifidobacterium longum SD-BB536-JP, Lactiplantibacillus plantarum SD-LPLDL-UK, Bifidobacterium animalis subsp. lactis SD-MB2409-IT, Bifidobacterium longum SD-CECT7347-SP, Lacticaseibacillus casei SD-CECT9104-SP, Bifidobacterium lactis SD-CECT8145-SP, Lactiplantibacillus plantarum SD-LP1-IT, Bifidobacterium breve SD-BR3-IT, Lacticaseibacillus rhamnosus SD-LR6-IT, Limosilactobacillus reuteri SD-LRE2-IT, Bifidobacterium infantis SD-M63-JP, Bifidobacterium longum HRVD90b-US, Bifidobacterium lactis SD-BS5-IT, Bifidobacterium lactis SD150-BE, Limosilactobacillus reuteri SD-RD830-FR, Lacticaseibacillus rhamnosus SD-GG-BE, Bifidobacterium adolescentis SD-BA5-IT, Lactobacillus crispatus SD-LCR01-IT, Limosilactobacillus fermentum SD-LF8-IT, and Ligilactobacillus salivarius SD-LS1-IT.

A batch fermentation approach was used for this study. An initial test with fecal material from two healthy individuals was conducted to optimize the dysbiotic conditions. Healthy donors, one male and one female, between 29 and 35 years of age, with a body mass index (BMI) of between 19 and 25, without antibiotic consumption in the previous 6 months or alcohol consumption in the previous 3 days, and with a self-selected standard Western diet were used for this research. Ethical approval by the University Hospital Ghent (reference no. B670201836585) was obtained for the collection of fecal samples, and written informed consent was obtained prior to sample collection. In this prescreening step, colonic bioreactors were inoculated with nutritional medium representative of the human colon (1 g/L of starch [Carl Roth, Karlsruhe, Germany], glucose [Merck KGaA, Darmstadt, Germany], pectin, Arabic gum [Keyser & Mackay, Amsterdam, The Netherlands], and fructooligosaccharides [FOSs] [Sigma-Aldrich, Bornem, Belgium]; 1.8 g/L yeast extract and peptone [Oxoid, Aalst, Belgium]; 4.7 g/L K₂HPO₄; 14.7 g/L KH₂PO₄; 1.8 g/L NaHCO₃ [Chem-lab NV, Zedelgem, Belgium]; 0.45 g/L l-cysteine; and 1.8 mL/L Tween 80 [Sigma-Aldrich, Bornem, Belgium] [pH 6.5]) and five mucin-coated carriers (meant to simulate a healthy gut mucosal lining), prepared according to methods described previously by Van den Abbeele et al. (44).

K1 carriers (AnoxKaldnes AB, Lund, Sweden) were submerged in a mucin solution composed of 50 g/L gastric porcine mucin type II (Sigma-Aldrich) and 10 g/L bacteriological agar (Oxoid) and combined in polyethylene netting (Zakkencentrale, Rotterdam, The Netherlands). Subsequently, anaerobic conditions were obtained by flushing bioreactors with N₂ for 10 min, and fecal samples were inoculated with a 10% (vol/vol) fecal slurry obtained by mixing the fecal sample with 7.5% (wt/vol) anaerobic phosphate-buffered saline (PBS), as previously described by Calatayud et al. (59). For both donors, a control condition, 2 doses of vodka (10% and 30% [vol/vol], corresponding to 4% and 12% EtOH, respectively), and 2 doses of an antibiotic mixture (50 μg/mL metronidazole plus 30 μg/mL ciprofloxacin and 250 μg/mL metronidazole plus 164 μg/mL ciprofloxacin) were tested. Different treatments were added to the reactors and incubated at 37°C for 48 h under anaerobic conditions. After this time, luminal samples were obtained to assess the overall microbial fermentation (pH) and microbial metabolic activity in terms of the production of saccharolytic end products (SCFAs) (see Table S1 in the supplemental material). The conditions selected had the highest microbial metabolic activity following treatment. These were low-dose antibiotics and 30% (vol/vol) vodka.

For the final experiment, two subsequent batch incubations were performed, focusing on (i) dysbiosis induction and (ii) recovery assessment. The total volume per reactor was 70 mL. During the first 48-h incubation (“dysbiosis induction”), three test arms were run in parallel: the untreated control, antibiotic treatment (50 μg/mL metronidazole and 30 μg/mL ciprofloxacin), and ethanol treatment (30% [vol/vol] vodka). The nutritional medium and inoculation density were identical to the ones specified above. Subsequently, at the start of the second incubation (“recovery assessment”), 10% (vol/vol) of the preceding 48-h control-, antibiotic-, or ethanol-treated microbiome was added to fresh nutritional medium. Furthermore, to increase the biorelevance of the recovery assessment, precolonized mucin-coated carriers from the untreated control were administered (Fig. 1A). This simulates recolonization by gut microbes from environments (such as the biofilm covering the mucosal surface) where microbes are better protected from external stressors such as antibiotics and ethanol. The test arms during the recovery assessment comprised the following: (i) control plus mucin beads, (ii) antibiotic treatment plus mucin beads, (iii) alcohol treatment plus mucin beads, (iv) antibiotic treatment plus mucin beads and 5.3 × 10¹⁰ arbitrary fluorescence units (AFU) of the 24-strain probiotic, and (v) alcohol treatment plus mucin beads and 5.3 × 10¹⁰ AFU of the 24-strain probiotic. Reactors were incubated anaerobically at 37°C for a further 48 h. All of the above-mentioned test arms were run using a fecal sample from one donor (in three technical replicates). Six milliliters was sampled from each reactor for downstream analyses at different time points.

Samples were collected after further 0-h, 6-h, 24-h, and 48-h colonic fermentations and assessed for metabolic activity by measuring acetate, propionate, and butyrate as previously described by De Weirdt et al. (60). Previous work demonstrated that fecal suspensions are active for up to 48 h in the medium used here (61). Briefly, after the addition of 2-methyl hexanoic acid as an internal standard, SCFAs were extracted from the samples with diethyl ether. Extracts were analyzed using a GC-2014 gas chromatograph (Shimadzu, ‘s-Hertogenbosch, The Netherlands) equipped with a capillary fatty-acid-free EC-1000 Econo-Cap column (dimensions of 25 mm by 0.53 mm, film thickness of 1.2 μm; Alltech, Laarne, Belgium), a flame ionization detector, and a split injector. The production of unbranched and branched SCFAs was calculated by summing the molar concentrations of acetate, propionate, butyrate, valerate, and caproate and summing the molar concentrations of isobutyrate, isovalerate, and isocaproate, respectively. Total SCFA production was defined as the sum of unbranched and branched SCFAs. Lactate was quantified using a commercially available enzymatic assay kit (R-Biopharm, Darmstadt, Germany).

DNA was isolated as previously described (62). Bacterial cell pellets, collected from 1-mL luminal colonic suspensions, were mechanically and chemically lysed using glass beads and a cationic detergent (cetyltrimethylammonium bromide), respectively. DNA was extracted according to a phenol-chloroform-isoamyl alcohol protocol and precipitated with polyethylene glycol. Finally, the DNA was washed with 70% alcohol and diluted in DNase/RNase/protease-free water. Samples were shipped to CosmosID Inc. (Rockville, MD, USA) for shallow shotgun sequencing analysis. For this purpose, DNA libraries were prepared using the Illumina Nextera XT library preparation kit, with a modified protocol. The library quantity was assessed with a Qubit instrument (Thermo Fisher). Libraries were then sequenced on an Illumina HiSeq 2× 150-bp platform at 3 million total reads. Raw reads were quality controlled by CosmosID to remove trim adapters and human contamination.

Relative abundances were computed from the sequencing output. The quality-controlled reads were aligned back to the Genome Taxonomy Database (GTDB) (release 207) (63) with the BURST package (64). The GTDB was formatted with the BURST plug-in xtree with a kmer length of 29 and a compression level set to 2. Alignments were filtered based on coverage, keeping only genomes with either 0.02% unique coverage (across all genomes in the database) or 50% total. Relative abundance was computed via center-log-ratio transformation. As a secondary approach, we computed relative abundances by summing the total numbers of reads aligned to all genomes in a given sample and dividing the reads aligned to a given genome by this sum (Table S2).

The following comparisons were made between the different arms of the experiment. For test 1 (Fig. 1A, top row), comparisons were made for (i) the control versus the antibiotic-treated reactor (Fig. 1C) and (ii) the control versus the EtOH-treated reactor (Fig. 1C). For test 2 (Fig. 1A, bottom row), comparisons were made for (i) the antibiotic-treated reactor with synbiotic treatment versus the antibiotic-treated reactor without synbiotic treatment (Fig. 2D), (ii) the EtOH-treated reactor with synbiotic treatment versus the EtOH-treated reactor without synbiotic treatment (Fig. 2D), and (iii) both antibiotics and ethanol reactors without synbiotic treatment versus both antibiotics and ethanol reactors with synbiotic treatment (Fig. 3).

All comparisons between groups were carried out using generalized linear regression on center-log-ratio-transformed raw counts (figures in the text) or relative abundances (see the supplemental material).

The data that support the findings of this study are openly available in the National Center for Biotechnology Information (NCBI) repository under accession no. PRJNA689370↗.