Sequential laxative-probiotic usage for treatment of irritable bowel syndrome: a novel method inspired by mathematical modelling of the microbiome

The intestinal microbiota is central to human health, and dysbiosis has been implicated in multiple diseases. While the microbiota is a promising therapeutic target, its dynamic and largely unknown ecological laws complicate precise manipulation. Probiotic interventions in IBS show variable efficacy, with optimal strain, dose, duration and combinations unsettled. Mathematical models can describe ecosystem dynamics and may streamline microbiome management, yet the gut differs from classic ecosystems (e.g., periodic defecation, multiple niches), necessitating tailored models. The authors hypothesized that a new model could better simulate gut dynamics and guide probiotic regimens in IBS. They developed a non-extinction and defecation-normalized (NEDN) model adapted from generalized Lotka-Volterra to reflect daily bacterial growth with inter-genus interactions, non-extinction of low-abundance taxa, and normalization mimicking defecation. They then used simulations to predict how probiotics and laxatives affect stability and transitions of microbiome states and conducted a preliminary clinical trial testing model-inspired sequencing of laxative and C. butyricum in IBS.

Prior work indicates gut microbiome resilience and daily-scale dynamics, with composition at a given time strongly influencing future states. Classical ecological models (e.g., gLV) have been applied to microbiomes, but may not fully capture gut-specific processes like defecation. Clinical probiotic trials in IBS show mixed results and lack consensus on regimen parameters. Diet-based interventions (e.g., low-FODMAP) show that baseline fecal profiles and volatile organic compounds can predict responses, highlighting baseline-dependent treatment effects. These findings motivate model-guided tailoring of interventions and attention to timing relative to perturbations such as bowel cleansing.

Mathematical modeling and simulations: The NEDN model modifies the gLV framework to better reflect gut dynamics and sequencing outputs by modeling daily steps: (1) biological growth where each genus increases according to a positive inherent growth rate (vector α) and inter-genus interactions (matrix β), with a non-extinction floor (abundances below a minimal threshold are reset to the threshold), and (2) defecation normalization where, if total abundance exceeds 1, all genera are proportionally scaled so total equals 1, representing biomass discharge. The model ingests relative abundance data and produces next-day predictions. Parameters α (13 values) and β (13×13 zero-diagonal; 156 values) were fit to dense time-series 16S data (Caporaso et al.) restricted to 332 fecal samples from one individual (M3). Of these, 269 day0–day1 pairs were used for training. Parameter ranges were constrained to [-3, 3]. Optimization minimized squared error between predicted next-day genus abundances and observed values using NLopt (NLOPT_GN_CRS2_LM) with stringent tolerances (xtol_rel and ftol_rel 1e-12), requiring 57,705,618 iterations. Model stability and predictive accuracy were compared to a similarly parameterized gLV model using day0–day2 (n=266) and day0–day3 (n=266) pairs; cumulative deviation (L value) and error accumulation over 3 days were assessed. Static-state search: To explore attractors, simulations iteratively predicted day-to-day changes for 400 days starting from (a) 332 real microbiomes and (b) 1000 synthetic compositions sampled from a 13-dimensional Dirichlet distribution (shape 0.2). A microbiome was considered static if variance over the last 100 days summed to <1e-20. NMDS and clustering characterized initial and stabilized states, identifying five stable statuses. Intervention simulations: Probiotic-like additions were modeled by increasing Clostridium cluster XIVa by specified increments (e.g., single dose +5% or +20%; daily +1% or +5% for 14 days), followed by renormalization. Laxative-like treatment reduced all genera to 1% of original abundance at day 50. Combinations tested included single or 14-day probiotic courses immediately after laxative. Alternative laxative intensities (5%, 10%, 0.1% remaining biomass) were explored for sensitivity. Clinical trial: An open-label, three-arm, randomized exploratory trial enrolled adult IBS patients (Rome III) at a single center (Qilu Hospital, Shandong University) from July 2014 to February 2015. Exclusions included recent antibiotics/probiotics/laxatives, organic GI disease, major comorbidities, pregnancy/lactation, major abdominal surgery, severe endometriosis/dementia. Participants were randomized by Excel-generated numbers to: (1) LP: 2 L bowel prep with colonoscopy, then immediate C. butyricum MIYA (C.B.) for 2 weeks; (2) L2P: same bowel prep and colonoscopy, but C.B. started 2 weeks later; (3) P: no bowel prep/colonoscopy, immediate C.B. for 2 weeks. C.B. dosing: 2 tablets, three times daily, for 2 weeks. No other medications allowed; diet maintained. Assessments at baseline, week 2, and week 4 included IBS Likert-7 symptom scores, IBS-QoL, daily diaries (symptoms, stool frequency, Bristol score, side effects). Primary endpoint: patient global assessment of relief during 2-week probiotic usage and overall after 4 weeks. Secondary endpoints: changes in individual and composite symptoms, QoL, and fecal microbiome composition. Fecal sampling at baseline, week 2, and week 4. Microbiome methods: V3–V4 16S rRNA sequencing (Illumina PE300) with defined quality filters, read merging, and demultiplexing. Sequences from IBS and a healthy reference cohort (44 individuals; PRJNA544721) were processed with usearch v10 (dereplication, OTU clustering at 97% with cluster_otus, taxonomy via RDP sintax cutoff 0.8). OTU abundances summarized at genus and phylum; alpha and beta diversity computed (usearch alpha_div, cluster_agg, beta_div). Sequencing data archived: IBS PRJNA573815; healthy PRJNA544721. Statistics: R (v3.5.1) in RStudio. Bray distances (vegan::vegdist), hierarchical clustering (hclust ward.D). Group comparisons: Kruskal–Wallis with Dunn’s post hoc (Bonferroni; significance threshold p<0.025); Wilcoxon for two-group comparisons; Fisher’s exact for categorical variables. Beta diversity visualized by Euclidean NMDS with 95% confidence ellipses (vegan::ordiellipse). Adonis PERMANOVA tested group differences. Multiple testing correction for taxonomic comparisons via FDR.

Model performance and dynamics: • The NEDN model, incorporating non-extinction and defecation normalization, fit time-series data better than gLV (lower cumulative deviation; L=10.17 vs 11.19 under same constraints) and maintained stable prediction errors from day+1 to day+3, unlike gLV whose errors accumulated rapidly. • Simulations revealed the microbiome self-stabilizes into five stable statuses; the most common was Prevotella-dominant (48.9%, 643/1332). In silico interventions: • Single-dose addition of Clostridium cluster XIVa at +5% caused only transient minor disturbance; +20% (unrealistically high) triggered a prolonged Bacteroides-dominant phase before stabilizing to Prevotella-dominant status 1. • Repeated daily additions for 14 days were more effective than single dosing and showed dose dependence: +5% daily for 14 days shifted to Prevotella-dominant status 1; +1% daily failed to shift. • Laxative-like reduction to 1% of original biomass allowed the microbiome to regain baseline composition within ~20 days if untreated. However, a much smaller probiotic dose (+0.5% Clostridium XIVa) given immediately after laxative could shift the microbiome via a transient Bacteroides-dominant phase to Prevotella-dominant status. Fourteen days of +0.5% dosing post-laxative had similar effects. Weaker laxatives (to 5% or 10%) did not promote shifts; extremely strong (to 0.1%) could induce shifts alone. • The same laxative-probiotic regimen did not shift all starting statuses (e.g., failed from status 4), underscoring baseline dependence. Clinical trial outcomes (n analyzed=56; LP=21, L2P=22, P=13): • Baseline demographics and symptom scores were comparable across groups. • During the 2 weeks of probiotic intake, LP showed the greatest improvements: abdominal pain, discomfort, bloating/distention, urgency, and summed symptom score improved most in LP (overall Kruskal–Wallis p<0.05 across outcomes). Pairwise Dunn’s tests showed LP improved more than L2P for abdominal discomfort, bloating/distention, urgency, and summed score (adjusted p<0.025). • By week 4, reductions in abdominal pain favored LP vs P but were not statistically significant (overall KW p=0.062; adjusted p<0.0299 threshold not met). Median self-reported relief was highest in LP, not significant (overall KW p=0.18). Microbiome changes in vivo: • At baseline, IBS had lower alpha diversity than healthy (Wilcoxon p=2.64e-6) and higher within-group beta diversity (weighted UniFrac; p=2.09e-12). Adonis showed IBS vs healthy significantly different (p=0.035). • After laxative, microbiomes changed markedly (Adonis p=0.001). • After 2 weeks of C.B., LP’s community composition became less heterogeneous and similar to healthy: Adonis LP vs healthy non-significant (p=0.266); within-group UniFrac distances similar to healthy (post hoc p=1). L2P and P did not show comparable convergence during the same interval. • At week 4, L2P remained significantly different from healthy (Adonis p=0.017), while LP and P were not (p=0.143 and p=0.109, respectively). • Taxonomically, IBS had lower Proteobacteria and higher Actinobacteria (FDR-adjusted p=0.040 and p=0.000, respectively). Bifidobacterium was higher in IBS (FDR-adjusted p=0.038). No significant in vivo increase in Bacteroides was observed post-treatment despite in silico transient boosts.

The study demonstrates that a gut-specific dynamical model (NEDN) can capture core features of fecal microbiome behavior—self-stabilization, resilience, and dose-dependent responses—and can generate actionable hypotheses about intervention timing and sequencing. Simulations predicted that a strong, transient perturbation (laxative) lowers the dose threshold required for a single-genus probiotic to shift community states, creating a window of opportunity for modulation. The exploratory clinical trial supports this prediction: initiating C. butyricum immediately after bowel cleansing produced the most pronounced short-term symptom relief and shifted fecal community structure toward a healthy-like configuration, whereas delaying probiotic initiation by 2 weeks attenuated these benefits. These findings address the clinical question of how to optimize probiotic efficacy in IBS by highlighting the importance of timing relative to perturbations and the dependence on baseline microbiome status. The work underscores a complementary role for mathematical models to guide regimen design, while recognizing that symptom outcomes also depend on host factors (barrier, immune, neural, psychological) beyond compositional shifts. Integrating modeling with clinical testing can refine microbiome-targeted therapies and move toward precision interventions tailored to initial community states and perturbation history.

A novel non-extinction and defecation-normalized model more accurately simulates human fecal microbiome dynamics than a classical gLV model and predicts that a laxative-induced low-biomass state enables lower-dose probiotic interventions to shift community structure. In a randomized exploratory trial, starting C. butyricum immediately after bowel cleansing yielded the greatest short-term symptom improvements and drove the fecal microbiome toward a healthy-like state, compared with delayed initiation or probiotic alone. These results illustrate the utility of integrating mathematical simulation with clinical strategy and suggest that sequencing laxative followed by probiotic may optimize IBS management. Future work should validate these findings in larger, blinded, multicenter trials, refine model realism (e.g., multi-compartment gut, empirically measured perturbation strengths), and explore generalization to other probiotics and patient subtypes.

Modeling limitations include the simplified single-chamber representation of the gut, enforced non-extinction threshold, and arbitrary parameterization of laxative intensity (set to 1% remaining biomass without empirical measurement). The NEDN model was not suited to simulate complex multi-intervention scenarios and was limited to 13 aggregated genera due to data constraints. Clinical limitations include open-label design without placebo, small sample size from a single center, short 4-week follow-up, and potential biases inherent to exploratory trials. Generalizability to broader IBS populations, other probiotic strains, or different bowel preparation protocols remains uncertain.