**Background:** In vitro models of the human colon are valuable tools for studying the gut microbiome (GM) and its responses to external factors like antibiotics. Multi-stage chemostat models are clinically reflective but are resource-intensive and difficult to scale, limiting experimental throughput and replication. The authors developed the Mini Gut (MiGut) platform to address this gap by miniaturizing a validated triple-stage human gut model (HGM) while preserving physiological relevance.
**Methods:** MiGut consists of four independent triple-stage models, each with three vessels (V1, V2, V3) representing proximal, medial, and distal colon, with a reduced fluid volume of 45 mL per vessel (vs. 300 mL in HGM). Temperature was maintained with heat mats, and pH was controlled at 5.5±0.1 (V1), 6.25±0.1 (V2), and 6.75±0.1 (V3) using 0.1 M HCl or NaOH. Growth medium flow rate was 0.045 mL/min to maintain a 48-h retention time. A preliminary study used four different faecal slurries (pooled from five healthy volunteers each) to inoculate MiGut models, with 16S rRNA sequencing of V3 contents after 2 weeks. The main validation study ran for 9 weeks: 2 weeks equilibration, 3 weeks of sequential antibiotic dosing (amoxicillin 24 mg/mL 2× daily for 5 days, ciprofloxacin 278 mg/mL 3× daily for 5 days, piperacillin/tazobactam 828 mg/mL 2× daily for 5 days, each with 2-day washout), and 4 weeks post-antibiotic recovery. One HGM and four MiGut models were run in parallel (one MiGut model was lost due to technical fault). Bacterial populations were monitored thrice weekly by rtPCR targeting eight key groups: Eubacteria, Bacteroides spp., Bifidobacteria, Clostridium coccoides group, Clostridium leptum group, Enterobacteriaceae, Lactobacillus spp., and Prevotella spp. Data were analyzed using Bray-Curtis (BC) dissimilarity, non-metric multidimensional scaling (NMDS), analysis of similarities (ANOSIM), and product-moment correlation coefficients (PMCC).
**Key Results:** In the preliminary study, PCoA of 16S rRNA sequencing data showed that MiGut V3 contents clustered closely with their respective faecal inocula, and taxonomic profiles confirmed recapture of major taxa. During the 9-week study, equilibration was achieved by Day 6 for V1 and Day 8 for V2 and V3 (defined by gradient <0.002/day). BC dissimilarity from faecal inoculum at equilibrium was 9.67% (V1), 8.15% (V2), and 7.85% (V3) for MiGut, compared with 9.68% (V2) and 7.18% (V3) for HGM (V1 of HGM did not equilibrate). Standard deviations of BC dissimilarity for MiGut were 1.61 (V1), 1.53 (V2), and 0.81 (V3). ANOSIM showed high similarity between replicate MiGut models (R<0.1 for all vessels). Comparison between MiGut and HGM showed V3 was highly similar (R<0.1), while V1 and V2 showed some differences (0.1<R<0.25). Antibiotic dosing caused pronounced disruption: BC dissimilarity relative to faecal slurry increased during dosing and did not fully return to equilibrium even after 4 weeks post-antibiotic. BC dissimilarity between MiGut and HGM correlated extremely well (PMCC>0.98). Per-population PMCC values were: Prevotella spp. 0.99, most populations >0.9, Bifidobacteria 0.78, Enterobacteriaceae 0.65. Maximum deviation between MiGut and HGM during antibiotic stages was 1.54 log10 copies/μL (Bifidobacteria), which reduced to 0.31 log10 copies/μL by the end of the study. By post-antibiotic Week 4, all populations showed ≤0.65 log10 copies/μL difference. NMDS ordination showed that post-antibiotic samples formed a distinct cluster from initial equilibrium (ANOSIM R=0.92, p=0.0001), indicating incomplete recovery.
**Clinical Implications:** MiGut provides a scalable, reproducible platform for studying gut microbiome dynamics that closely matches a clinically validated HGM. The platform's ability to run multiple models simultaneously enables complex study designs with biological replicates, which was previously impractical with multi-stage systems. The demonstration that antibiotic-induced dysbiosis persists for at least 4 weeks post-exposure aligns with clinical observations of long-term microbiome disruption. MiGut's throughput and scalability make it suitable for investigating multiple interacting factors (e.g., age, diet, medications) that affect the GM, potentially informing more targeted clinical interventions. Limitations include the absence of host cells, immune interactions, and absorption of metabolites, though the authors note potential for future integration with co-culture systems.