A 28-Day Antibiotic Cocktail Wiped Out Nearly All Gut Bacteria in Mice

Researchers studying what antibiotics do to the gut have long faced a practical problem: the animal models used to replicate microbial collapse are poorly standardized, making results hard to compare across labs. A new preprint by K Lee and S Song, published June 2025 on Research Square, proposes a solution: a 28-day antibiotic regimen in mice that reliably drives gut microbiota into a near-total collapse, leaving behind a two-species bacterial remnant and measurable damage to the intestinal barrier.

Antibiotic cocktail collapses gut microbiota to two species

The core finding is striking, even in a preliminary study. Mice in the Amox group, treated with a combination of ampicillin, amoxicillin, and gentamicin dissolved in drinking water for 28 days, ended the experiment with a gut microbiota composed entirely of Bacteroidota. Every member of Bacillota and Actinomycetota had vanished. More precisely, the surviving community converged on just two species: Bacteroides stercoris at 50.84% relative abundance and Parabacteroides merdae at 49.16%. Healthy control mice, by contrast, maintained the kind of diverse microbiota associated with normal gut function, including beneficial short-chain fatty acid producers like Lactobacillus and Bifidobacterium.

The researchers attribute Bacteroides dominance under antibiotic pressure to two properties: intrinsic resistance to beta-lactam antibiotics and versatile carbohydrate metabolism that allows survival when other organisms are cleared. The co-dominance of the two species likely reflects niche partitioning, with beta-lactamase-mediated cross-protection helping both persist.

Intestinal barrier damage and inflammation markers in the Amox group

The microbial collapse was not just a compositional curiosity; it came with measurable physiological damage. The Amox group showed the highest serum lipopolysaccharide (LPS) levels of any group, reaching 72.04 ng/mL. Serum LPS is a direct indicator that the intestinal barrier has been breached and bacterial endotoxins are leaking into circulation, a pattern implicated in a range of downstream health problems. Antibiotic-driven gut dysbiosis has been linked to chronic diseases including inflammatory bowel disease, metabolic disorders, and neurological conditions, which is part of what makes a reliable animal model for studying that process valuable.

Histological analysis of colon tissue in the Amox group showed pronounced pathological changes: shortened and irregular crypts, lymphoid aggregates, and dense immune cell infiltration. The disease activity index score, calculated from stool consistency and rectal bleeding, climbed progressively over the 28 days. Cecum size more than tripled compared to controls.

Gene expression results added nuance. COX-2, a marker of intestinal inflammation, was significantly reduced in the Amox group rather than elevated, which the authors interpret as a sign of suppressed, dysregulated immune signaling rather than active inflammation. The tight junction protein ZO-1, which holds the gut barrier together, was not upregulated in the Amox group, unlike in the Amp and Van groups where its compensatory induction offered some barrier protection. That absence of compensation, combined with the highest LPS values, led the researchers to select the Amox regimen as their representative dysbiosis model.

Functional prediction using PICRUSt2, a tool that infers metabolic pathway activity from 16S rRNA sequencing data, showed broad losses in the Amox microbiota. Pathways related to nucleotide synthesis, lipid metabolism, and central carbon metabolism were substantially depleted. Only two pathways were enriched: L-histidine degradation I and pyridoxal-5-phosphate biosynthesis I, which the authors describe as possible adaptive shifts in the surviving species.

How the 28-day protocol was designed for reproducibility

The study used 25 female SPF (specific pathogen-free) C57BL/6 mice, five weeks old, divided into four groups: untreated controls, an Amp group (ampicillin alone), the Amox group (ampicillin, amoxicillin, and gentamicin), and a Van group (ampicillin, gentamicin, and vancomycin). The antibiotic mix was chosen to hit both Gram-positive and Gram-negative bacteria simultaneously. Delivery via drinking water rather than gavage was deliberate: it allows continuous low-stress exposure over the full 28 days. From day 15, the gentamicin concentration in the Amox and Van groups was increased from 0.64 g/L to 2.88 g/L to sustain antimicrobial pressure through the second half of the protocol.

Microbiome composition was characterized by 16S rRNA gene sequencing of the V3-V4 region using the Illumina MiSeq platform. Taxonomic assignment used the NCBI 16S rRNA reference database with BLAST+. LEfSe analysis identified differentially abundant taxa using an LDA score threshold of 4.0 (log10).

The choice of mice as a model system carries inherent limitations. Significant differences exist between human and murine microbiota composition, and findings from mouse dysbiosis models do not always translate cleanly to human biology, a consideration that applies to this work as to any preclinical gut microbiome study.

Small sample size limits what the sequencing data can show

The authors are direct about the study's main weakness. Microbiome sequencing was performed on samples from the Amox and Control groups only, and the control group had just two sequenced samples. That small number constrains statistical power and may not capture the natural variation within a healthy mouse gut microbiota. Intra-group microbial variability, which can be substantial even in genetically uniform SPF mice, is difficult to characterize reliably at this sample size.

The functional predictions from PICRUSt2 are also inferred rather than directly measured: the tool estimates metabolic activity from taxon-level sequencing data rather than from direct metagenomic or metabolomic measurements. This means the pathway losses described should be treated as hypotheses to test rather than confirmed functional changes.

What a validated dysbiosis model could enable for gut research

Even with those caveats, the protocol fills a gap. Research into how antibiotic exposure shapes long-term health outcomes needs reproducible animal models to test mechanistic hypotheses before they can be evaluated in people. Studies in large human cohorts have shown that antibiotics can disrupt the gut microbiome for up to eight years, and the mechanisms behind that persistence are not well understood. Animal models with well-characterized dysbiosis phenotypes are one tool for working backward from observed human disruptions to testable biological explanations.

Separately, chronic antibiotic use has been associated with a roughly 13% increase in colorectal cancer risk, with effects lasting months to years; a model that reproducibly depletes SCFA-producing taxa and impairs barrier integrity could support experiments probing that link directly.

Lee and Song call for larger sequencing cohorts to validate the community-level differences their protocol produces. If that reproducibility holds up, the Amox regimen could become a standardized starting point for studies testing probiotics, dietary interventions, or fecal microbiota transplants aimed at reversing antibiotic-induced dysbiosis.

Evidence rating

INVESTIGATIONAL — 75/100

Caveats:

• Animal study only — findings have not been tested in humans.
• Microbiome sequencing was performed on only 2 control animals, severely limiting statistical confidence in community-level comparisons.
• This is a preprint and has not yet completed peer review.
• The study establishes a mouse model; it does not test any therapeutic intervention or translate directly to human health outcomes.

Preclinical mouse study (n=25 female C57BL/6 mice across four groups) evaluating antibiotic-induced gut microbiota dysbiosis. Findings are associative within an animal model and cannot establish causation in humans; notably, microbiome sequencing was conducted on only two control animals, substantially limiting statistical power for community-level comparisons. This work has not been independently replicated and is currently available as a preprint pending peer review.