The study was conducted in accordance with the ethical principles of the Declaration of Helsinki and ethical approval was granted by the Ethical Committee of Comac Medical, Sofia, Bulgaria (Reference: #246/13/07/2022). The study protocol was registered with clinicalTrials.gov on 19/04/2022: NCT05355571.

A single-centre, double blind, randomized, placebo-controlled study with equal allocation of participants between two parallel study groups was undertaken. Adults (aged 18–65) who had been prescribed an oral antibiotic for non-gastrointestinal disturbance were recruited by Comac Medical (Sofia, Bulgaria). Physicians, located in Sofia, Bulgaria, prescribing the antibiotics offered potential candidates the opportunity to take part in the study. The inclusion criteria were: receiving a 5 to 10-day course of oral antibiotics for a non-GI related condition, refraining from any other probiotic or prebiotic supplements during the study, willing to maintain normal lifestyle and diet throughout the study, willing to refrain from taking any non-GP prescribed antibiotics during the study and willing to provide faecal samples. Exclusion criteria included: antibiotic intake within the previous three months, regular probiotic intake during the month prior to the study, immunodeficient or undergoing immunosuppressive therapy, pregnancy or planning pregnancy, diabetes, having cardiovascular disease or severe systemic disease.

A total of 50 candidates were recruited in November 2022, with the study taking place between 18/11/2022 and 17/01/2023; there were no dropouts or adverse effects reported in either arm of the study. Baseline demographics of the populations are shown in Table 1 . The oral antibiotics used during the study included amoxicillin, cephalosporins, azithromycin, clarithromycin, clindamycin and spiramycin. The average antibiotic course length was 5.76 days in the placebo group, and 5.92 days in the probiotic group.

The eligible participants were allocated to one of the two study arms in a 1:1 ratio according to a computer-generated random sequence (block-size of four) that was generated using SAS PROC PLAN (SAS v9.4). The allocation sequence was not available to any member of the research team until all databases had been completed and locked. Tamper-proof sealed envelopes containing the participant allocation sequence were held at the trial site.

The probiotic product comprised capsules containing a total of 25 billion colony forming units (CFU) of Lactobacillus acidophilus CUL60 (National Collection of Industrial, Food and Marine Bacteria (NCIMB) 30157), Lactobacillus acidophilus CUL21 (NCIMB 30156), Bifidobacterium bifidum CUL20 (NCIMB 30153) and Bifidobacterium animalis subsp. lactis CUL34 (NCIMB 30172) and 10 billion CFU of Saccharomyces boulardii (Collection Nationale de Cultures de Microorganismes (CNCM)-I-1079) on a base of microcrystalline cellulose, silicon dioxide and magnesium stearate. The placebo only contained the base ingredients and was identical in appearance, size and weight to the probiotic product. The probiotic and placebo products were prepared by Cultech Ltd, Port Talbot, UK. One capsule of the probiotic or placebo was taken daily for 10 days. Participants were instructed to refrigerate the study product and take the capsules at least 2 hours after antibiotic ingestion.

The primary study outcome was changes in the composition of the microbiome assessed through viable cell enumeration and shotgun metagenomic sequencing. Secondary outcomes were analysis of ARG content and abundance. The study overview of participant visits to the trial centre and sample collection is shown, along with a flow diagram of enrolment, allocation and follow-up in Figure 1 .

Three faecal samples were collected; the first on day 1 (baseline sample), the second after finishing the course of antibiotics (varying from days 6–11, post antibiotic sample) and the third at day 30 ± 2, (endpoint sample). Faecal samples were collected using the Fe-Col^® Faecal Sample collection kits (Alpha Laboratories, Hampshire, UK) and were transferred into anaerobic Genbags (Sigma Aldrich, UK). Full details of faecal collection procedure can be found in Supplementary Figure S1 . Samples were stored at <10°C for up to 24 hours then stored at −80°C, pending analysis.

Faecal samples were assessed for viable bacterial numbers using a modified version of the Miles Misra (1938) plate count technique. Decimal dilution series were set up in Maximum Recovery Diluent (MRD, Oxoid, UK) and were plated onto a range selective media (Oxoid, UK, Supplementary Table S1 ). Total Bacteria represents the sum of the counts from total aerobes and total anaerobes. Viable cell numbers were expressed as log₁₀ of the number of CFU/g sample.

DNA was extracted from faeces using the QIAamp^® Fast DNA Stool Mini Kit (Qiagen, Germany) as per the manufacturer’s instructions. The eluted genomic DNA samples were quantified using a Qubit^® (Thermo Fischer Scientific, Unite States) and stored at −20°C. Shotgun metagenomic sequencing was performed using the Illumina Novaseq 6000 platform (Novogene, China). Quality control of raw reads was performed using readfq V10 (https://github.com/cjfields/readfq) using default parameters. The specific processing steps were as follows: a) removal of reads comprising low quality bases (quality threshold value ≤38), >40 base pairs (bp); b) removal of reads wherein the N base reached 10 bp; c) removal of reads presenting an overlap >15 bp with Adapter. Host sequences decontamination was performed using Bowtie2 with the default parameters (Langmead and Salzberg, 2012).

Host-filtered metagenomic samples were assembled using MEGAHIT (Li et al., 2015) and open reading frames (ORFs) in contigs were predicted using MetaGeneMark-2 (Gemayel et al., 2022). The predicted genes from each sample were merged and clustered using CD-HIT (Fu et al., 2012) based on the criteria of identity >95% and coverage > 90% to remove redundant genes. The gene abundance profiles were constructed using Bowtie2 and mapped to the Unigenes datasets. Annotation of ARGs was performed using the Resistance Gene Identifier (RGI) software provided by the Comprehensive Antibiotic Resistance Database (CARD 2023) (Alcock et al., 2023).

Baseline demographics and antibiotic usage of the groups were compared using either an unpaired two-way t-test (age, height, weight, Body Mass Index (BMI), Systolic Blood Pressure (SBP), Diastolic Blood Pressure (DBP) and antibiotic course length) or Fisher’s exact test (sex, antibiotic class and antibiotic type). Differences in the number of faecal CFU were assessed using a mixed-effects analysis with a Tukey’s post-hoc for multiple comparisons (GraphPad Prism, Version 10.0.2) where *p<0.05, **p<0.01 and ***p<0.001.

The R package Phyloseq (McMurdie and Holmes, 2013) was used for data importation and diversity analyses. A generalized linear model with antibiotic type and course length as covariates and post-hoc Bonferroni correction was employed using the R package emmeans (Lenth, 2024) to compare Shannon’s diversity index, changes in bacterial abundance and changes in ARG abundance between interventions and within timepoints. Spatial differences of the groups were observed with a Non-Metric Multidimensional Scaling (NMDS) plot based on Bray-Curtis dissimilarity matrix. The R package Vegan was used to perform permutational analysis of variance (PERMANOVA) computed (with 999 permutations) using the Adonis function and homogeneity of dispersion (Oksanen et al., 2022). Differential abundance analysis between interventions was analysed using DESeq2 (Love et al., 2014).

The co-occurrence network was generated based on the Spearman correlation matrix constructed with the WGCNA (weighted correlation network analysis) package (Langfelder and Horvath, 2008). The nodes in this network represent unique taxa at the genera level and the edges connecting these nodes represent correlations between taxa. All p-values were adjusted for multiple testing using the Benjamini and Hochberg false discovery rate (FDR) controlling procedure at a threshold of 0.05. To evaluate connectedness of the networks and highlight keystone genera, betweenness centrality was used to measure the proportion of the shortest paths in a network that pass through a node; a lower average betweenness centrality score represents a more connected network due to more short paths or fewer shortest paths through each node. Node degree was calculated as a measure of sparsity, with a lower degree equalling a sparser network. These metrics were used to compare the networks using Welsh’s unequal variances t-test. Networks were visualized and betweenness centrality calculated using Gephi (Bastian et al., 2009).

Faecal viable numbers are presented in Figure 2 : total bacteria ( Figure 2A ), total anaerobes ( Figure 2B ), total aerobes ( Figure 2C ), Lactobacilli ( Figure 2D ), Bifidobacteria ( Figure 2E ), Enterobacteria ( Figure 2F ), Enterococci ( Figure 2G ), Bacteroides ( Figure 2H ), Staphylococci ( Figure 2I ), Clostridia ( Figure 2J ) and yeast ( Figure 2K ).

Between group analysis showed no significant differences in the microbiota at each time point. Within group analysis for the probiotic group indicated decreases in the numbers of Enterobacteria post antibiotic (−2.50 log₁₀CFU/g, p = 0.0197, Figure 2F ) and between baseline and the end of the study (−1.57 log₁₀CFU/g, p = 0.0156, Figure 2F ), increased Bacteroides numbers in the re-growth population (−1.72 log₁₀CFU/g, p = 0.0104, Figure 2H ) with an antibiotic impact on the clostridial numbers (−1.77 log₁₀CFU/g, p = 0.0096, Figure 2J ); yeast numbers significantly increased from baseline after antibiotics (0.91 log₁₀CFU/g, p = 0.0282) and at end of the study (0.84 log₁₀CFU/g, p = 0.0479). In the placebo group there were significant reductions in the viable numbers of total bacteria (−0.99 log₁₀CFU/g, p = 0.0022, Figure 2A ), total anaerobes (−0.73 log₁₀CFU/g, p = 0.0241, Figure 2B ), aerobes (−1.94 log₁₀CFU/g, p = 0.0017, Figure 2C ) and Lactobacilli (−2.05 log₁₀CFU/g, p = 0.0009, Figure 2D ), alongside antibiotic associated decreases in Enterococci (−1.75 log₁₀CFU/g, p = 0.0302, Figure 2G ) and Clostridia (−1.69 log₁₀CFU/g, p = 0.0034, Figure 2J ) although numbers had returned to baseline levels by the end of the study. Bifidobacterial numbers did not change in either group during the study ( Figure 2E ) and there were no changes in Staphylococcal numbers in either group over the duration of the study ( Figure 2I ).

Faecal samples were extracted and sequenced using a shotgun metagenomics approach on an Illumina NovaSeq 6000 platform. Each sample was subsequently assembled de novo, with the resulting number of contigs ranging from 143,121 to 615,638 and used to classify bacterial and fungal taxonomic composition. Alpha diversity indices (Shannon and Simpson) of the microbiota ( Figure 3A ) displayed no changes within the probiotic group between timepoints; whereas there was a significant drop in Shannon alpha diversity in the placebo group between baseline and post antibiotic treatment (p = 0.0320) and between baseline and the end of the study (p = 0.0014, Figure 3A ). When groups were sub-grouped based on antibiotics administered into β-lactams and macrolides, the Shannon alpha diversity scores significantly decreased at the end of the study compared to the baseline in the placebo group (p = 0.0444 and p = 0.0352, respectively) but no changes were observed in the probiotic group for either antibiotic class ( Supplementary Figure S2 ).

Analysis of beta diversity ( Figure 3B ) showed no significant between group spatial separation at the baseline or post antibiotic treatment, however, after re-growth a significant spatial separation was observed between the endpoints of the probiotic and placebo groups (p = 0.0109, Figure 3B ). There were no significant within group changes observed. Analysis of homogeneity of dispersion showed no differences either between or within groups (data not shown). No differences were observed in either the alpha or beta diversity measures for the mycobiome at any stage of the study ( Supplementary Figure S3 ).

Bacterial reads accounted for approximately 95% of taxa observed, with Viruses, Archaea and Eukaryota making up approximately 5% with Bacillota, Bacteroidota and Actinomycetota dominant in all participants throughout the study ( Supplementary Figure S4 ). At baseline, the microbiota of the probiotic group had a significantly higher proportion of Enterobacteriaceae (p = 0.0125), Akkermansiaceae (p = 0.0247) and Bifidobacteriaceae (p = 0.0089) than the placebo group. Whereas, the placebo group had a higher proportion of Bacteroidaceae (p = 0.0001) than the probiotic group.

Post antibiotic treatment, Akkermansiaceae and Bifidobacteriaceae dropped significantly in both the probiotic (p = 0.0034 and p = 0.0012, respectively) and the placebo (p = 0.0437 and p = 0.0135, respectively) groups. In the placebo group, compared to baseline, the populations of Enterobacteriaceae and Veillonellaceae decreased following antibiotic treatment (p = 0.0255 and p = 0.0012 respectively) and remained suppressed through the re-growth period (p = 0.0491 and p = 0.0357, respectively). No changes occurred in the Veillonellaceae population in the probiotic group ( Figure 3C ). By the end of the study, in the probiotic group the Enterobacteriaceae were significantly lower than at the baseline (p = 0.0226), while the Bifidobacteriaceae had recovered (p = 0.0410) from the post antibiotic treatment levels back to levels comparable to baseline ( Figure 3C ).

Between group analysis of differentially abundant taxa at the species level at the end of the study indicated higher levels in the placebo group of species belonging to Enterobacteriaceae and Bacteroidiaceae, including E. coli (p < 0.0001), Klebsiella pneumoniae (p = 0.0084) and Bacteroides fragilis (p = 0.0021), In the probiotic group, species belonging to Lactobacillaceae and Bifidobacteriaceae (including Lactobacillus acidophilus (p = 0.0003) and Bifidobacterium longum (p < 0.0001) increased post-antibiotic treatment and at were higher at the end of the study than at baseline ( Supplementary Figure S5 ).

ARGs were detected through the use of the Resistance gene identifier from CARD. In total, 218 ARGs were detected ( Figure 4A ), putatively conferring resistances against antimicrobial classes including aminoglycosides, β-lactams, macrolides, fluoroquinolones, tetracyclines and glycopeptides. At the baseline, 217 ARGs were found in the probiotic group, while 203 ARGs were present in the placebo. Post antibiotic treatment, both groups saw a decrease in overall gene content with 211 present in the probiotic and 185 in the placebo, however, at the end of the study the overall gene content in the probiotic group continued to fall to 203, while the placebo group increased to 201 ARGs. Genes that conferred resistance to tetracycline were amongst the most prevalent in both the placebo and probiotic group, with tetW being the overall most abundant ARG.

Between group differences in the total abundance of ARGs were assessed ( Figure 4B ), and showed that the probiotic group had a significantly higher abundance of ARGs compared the placebo group at the baseline (p = 0.0038) and post antibiotic treatment (p = 0.0012) but at the end of the study, there were no significant differences between groups. Within group analysis ( Figure 4B ) found that the overall abundance of ARGs significantly dropped in both groups post antibiotic treatment (p = 0.0125 for the placebo and p = 0.0482 for the probiotic). ARG abundance in the re-growth population of the probiotic group decreased to lower than that post antibiotic (p = 0.0086) and was lower than baseline (p = 0.0002). The ARG abundance for the re-growth population increased in the placebo group (p = 0.0325), with final levels comparable to baseline (p = 0.5235). When the abundance of ARGs was normalized to each group respective baseline, a significant difference (p=0.0340) was found between groups at the end of the study ( Supplementary Figure S6 )

Analysis of individual ARGs between and within group are presented in Supplementary Tables S2 and S3 . Interestingly, there were no increases in ARG abundance within the probiotic group during the study, with genes including rpoB (p = 0.0095), acrA (p = 0.0064), mdtE (p = 0.0010), mdtF (p = 0.0069) and yojl (p = 0.0033), significantly decreased from the baseline to the end of the study along with DHA-16 (p = 0.0001) and CTX-M-95 (p = 0.0001) being absent in the probiotic group at the end of the study. In contrast, in the placebo group, VanE (p = 0.0030), tetW (p = 0.0333), ErmT (p = 0.0111), had increased in abundance by the end of study compared to baseline levels.

The relative abundances of bacterial taxa known to harbour relevant ARGs was assessed ( Figure 4C ). Within the probiotic group, Kluyvera decreased between baseline and post antibiotic treatment, whereas E. coli, Enterobacter and Shigella decreased in the placebo during the same period ( Figure 4C ). Post antibiotic treatment, no changes in microbial abundance occurred in the probiotic group, while E. coli and Shigella were found to be increased in the placebo group at the end of the study. By the end of the study there were significant decreases in E. coli, Shigella, Kluyvera and Morganella in the probiotic group compared to the baseline ( Figure 4C ), whereas, Enterobacter decreased after the antibiotic treatment in the placebo group. At the end of the study, Morganella was higher than at baseline ( Figure 4C ). Full details of all bacterial taxa linked to ARGs are presented in Supplementary Figure S7 .

Co-occurrence networks of bacterial taxa were generated for each group, at each timepoint based on correlation relationships (and FDR adjusted q-values with a cut off of 0.05, Figure 5 ). At the baseline, both groups had similar betweenness centrality and node degree scores despite the different abundances observed, however significant differences were observed post antibiotic (q = 0.0411) and at the end of the study (q = 0.0284). Node degree and betweenness centralities decreased in the placebo group from the baseline to post antibiotic treatment (q = 0.0321) and to the end of the study (q = 0.0105), whereas, the node degree and betweenness centrality scores remained consistent throughout in the probiotic group with no significant differences observed. Keystone analysis at baseline highlighted Faecalibacterium, Latilactobacillus, Limosilactobacillus and Butyricicoccus, in both groups. These organisms continued as keystone taxa within the probiotic group throughout the study, while in the placebo group Coprococcus, Clostridium and Eggerthella, were the keystone taxa identified both post antibiotic treatment and at the end of the study.