henhao
Sign in
MicrobiologyScience
Peer-reviewed journal article

Imbalance in gut microbial interactions as a marker of health and disease

Roberto Corral LópeziD, Juan A. BonachelaiD, Maria G. Dominguez-BelloiD, Michael ManhartiD, Simon A. LeviniD, Martin J. BlaseriD, Miguel A. MuñoziD

Abstract

Imbalances in the human gut microbiome (dysbioses) are linked to multiple diseases but remain poorly understood. Current biomarkers to identify dysbiosis are inconsistent and fail to capture the ecological mechanisms differentiating healthy from diseased microbiomes. We propose a general dysbiosis biomarker, inspired by phenomenology observed in a gut-microbiome theoretical model introduced here. The emergent communities show complex interaction networks and two distinct collective states, corresponding to healthy and dysbiotic microbiomes. Our robust metric for dysbiosis, by quantifying the balance between cooperation and competition, differentiates these states in both simulated and real datasets across diverse diseases. Moreover, it reveals that dysbiosis results from a shift toward greater cooperation in the community. Our metric further correlates with disease progression, highlighting its potential as a diagnostic tool.

PDFDOIData
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
34
FWCI
70.27
Field percentile
top 1%
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (3 May 2025) · CC BY 4.0

A gut consumer-resource model with explicit metabolic pathways yields competition-dominated healthy and cooperation-dominated dysbiotic states; an interaction-balance index (ENBI) detects this shift in human disease cohorts.

Common dysbiosis markers such as reduced alpha-diversity behave inconsistently across diseases, and few gut models produce alternative stable states that could explain mechanistically how healthy and diseased communities differ. The work couples a metabolism-based model with emergent community states to a size-normalized cooperation-versus-competition metric that can be computed from co-abundance networks inferred from stool metagenomes.

Key results

  1. ShownDiseased gut communities shift toward cooperation: inferred net interaction exceeds controls in inflammatory bowel disease (IBD), C. difficile infection (CDI), irritable bowel syndrome (IBS) and colorectal cancer (CRC), so the Ecological Network Balance Index (ENBI) is positive in each.

    FlashWeave ρinferred over 500 bootstrapped subsamples per dataset (Fig. 3 uses subsample fraction 0.8) is significantly higher in disease for IBD, CDI, IBS and CRC (two-sided Mann-Whitney U); BEEM-Static reproduces this for the model and all diseases; LIMITS on longitudinal data does so for the model and IBD. · 4 The balance of community interactions is…; p-value; Method 1. FlashWeave; Figure legends

  2. SuggestedIn the model, emergent healthy-like states (many types, fast turnover) are competition-dominated (ρ < 0) and dysbiotic-like states (few dominant, quasi-stationary types) are cooperation-dominated (ρ > 0); both states emerge robustly across trade-off forms and wide parameter changes.

    Replicate 95% confidence intervals capture abundance-fluctuation, mean-abundance, Taylor's law and prevalence patterns without parameter fitting; both states persist under trade-offs g1, g2, g3 and under raised or lowered δ, h, γ, rmax, K, emax, R, M, U and extinction threshold; FlashWeave ρinferred is higher for dysbiotic than healthy snapshots and rises with true ρ among dysbiotic ones. · 2 A new model for the dynamics of the gut…; 3 Alternative states in the gut microbiome model…; 4 The balance of community interactions is…; Figure legends

  3. SuggestedWithin colorectal cancer, the cooperative shift grows with disease stage, so ENBI might track progression; the trend holds whether taxa are profiled with MetaPhlAn2 or mOTUs.

    CRC ENBI and ρinferred rise with more advanced disease stages at every subsampling fraction tested (Figs. 3C, S9B; 500 bootstrap samples); positive ENBI and the stage trend are reproduced with both MetaPhlAn2 and mOTUs profiles (Fig. S12); CRC is the only disease with progression data. · 4 The balance of community interactions is…; 5 Discussion; Figure legends

  4. ShownAlpha-diversity fails as a universal dysbiosis marker. In the model, Shannon diversity is only loosely tied to interaction balance, even though its mean trends downward as ρ increases.

    Fig. 1: significant Shannon decrease for IBD (Yarur, Ning, Human Microbiome Project data) and CDI (Berkell, Crobach, Ferreti), but mixed or non-significant shifts for IBS (Mars, Kim, Barandouzi) and CRC (Zeller, Yachida, Vogtmann); in the model a Shannon value of 1.5 occurs for ρ anywhere in [−0.1, 0.7]. · 1 The search for a universal indicator of…; 4 The balance of community interactions is…; Figure legends

  5. ShownModel dysbiotic states have lower Shannon diversity, fewer enzymes and fewer secreted metabolites, and appear more efficient at turning available energy into biomass. IBD data follow the same pattern, but the enzyme and metabolite markers are inconsistent for IBS and CRC.

    Fig. 2C-E: these markers are lower in model dysbiosis and in IBD patients vs healthy individuals; Fig. S5: dysbiotic states show higher mean biomass with greater energy depletion; Fig. S7: enzyme counts and metabolite Shannon index not consistently reduced in IBS or CRC; Bray-Curtis β-diversity is higher in disease except for CRC (Fig. S6). · 3 Alternative states in the gut microbiome model…; Figure legends

Methods, data and reproducibility

Consumer-resource model with nutrient inflow, dilution, immigration, cross-feeding and random strain-specific catabolic pathways penalized by an enzymatic-cost trade-off; states are labeled by net interaction sign, and FlashWeave, BEEM-Static and LIMITS networks from model and disease metagenomes give ENBI (diseased minus control inferred ρ).

  • Simulated communities from the consumer-resource model (healthy vs dysbiotic snapshots; weekly sampling for temporal and macroecological analyses) · n = 1000 samples per state per realization; ~100 realizations
  • Inflammatory Bowel Disease Multi'omics Database: stool metagenomic, enzyme and metabolomic profiles from IBD patients and healthy controls · human
  • Published irritable bowel syndrome metagenomic, enzyme and metabolomic datasets (Fig. 1 diversity: Mars et al., Kim et al., Barandouzi et al.) · human
  • Published C. difficile infection stool metagenomes (Fig. 1 diversity: Berkell et al., Crobach et al., Ferreti et al.) · human
  • Colorectal cancer metagenomic, enzyme and metabolomic data spanning disease stages, profiled with MetaPhlAn2 and mOTUs (Fig. 1: Zeller, Yachida, Vogtmann et al.) · human
  • Further sources for the Fig. 1A IBD diversity comparison: Human Microbiome Project samples plus IBD cohorts published by Yarur and by Ning · human
  • F4 and M3 gut metagenomic data from the MicrobeDS repository
Methods & tools
FlashWeave (co-abundance network inference; sensitive non-heterogeneous mode for disease data, heterogeneous mode for model data), BEEM-Static (generalized Lotka-Volterra inference from cross-sectional data), LIMITS (discrete-time Lotka-Volterra inference from longitudinal data), Python solve_ivp (multi-step variable-order integrator), MetaPhlAn2, mOTUs, Mann-Whitney U test (two-sided), Shannon index, Bray-Curtis dissimilarity, BioCyc ontology (pathway grouping), Ecological Network Balance Index (ENBI)
Data & accessions
https://ibdmdb.org/https://github.com/twbattaglia/MicrobeDS
Reagents & environment
Model integrated in Python with solve_ivp (multi-step, variable-order) using adapted step sizes to avoid negative abundances; trade-off parameters b = 8, c = 6, v = 1.1; no package versions or containers given.
Reproducibility gaps
Code not yet released (authors say it will go to Zenodo or GitHub); Table S1 parameter values and Supplementary Text equations are absent from the provided text; IBS, CDI and CRC datasets are cited by reference only, without URLs or accessions.
Limitations (6)
  • Model simplifications: anabolism enters only indirectly via the cost trade-off, the host acts solely through nutrient inflow and dilution, energy and enzyme-cost matrices are minimally parametrized (cost set equal to energy), and phages and bacterial chemical warfare are absent. · stated by the authors
  • Empirical conclusions rest on statistically inferred interaction networks that depend on data quality, quantity and method assumptions, while differing sampling and taxonomic profiling across source studies prevent quantitative cross-disease comparison of ENBI. · stated by the authors
  • ENBI flags dysbiosis but does not indicate which disease is responsible, so diagnostic use would require pairing with disease-specific tools such as colonography, endoscopy or biochemical markers. · stated by the authors
  • Association with progression was examined only in CRC, the sole condition with progression data, and generalization to other diseases is untested; whether microbiome shifts cause or follow disease is left open. · stated by the authors
  • ENBI contrasts networks inferred from pooled cohort subsamples; no per-individual classification accuracy or held-out validation is reported, and p-values from 500 overlapping bootstrap subsamples may overstate certainty, so patient-level monitoring is untested. · noted in this summary
  • Model healthy and dysbiotic states are defined by the sign of true ρ, so their separation by ρ is partly by construction; the informative tests are the inferred-ρ comparisons, and pathways are randomly assigned with no evolutionary adaptation. · noted in this summary

Who should read:Read this if you study gut dysbiosis biomarkers or microbial community ecology and want a mechanistic, interaction-network alternative to diversity metrics for IBD, IBS, CDI or colorectal cancer.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISLam '22Pasqualini '23Soverini '26Armour '19Liu '20Gupta '20
70% semantic matchBioinformatics
Tony J. Lam, Yuzhen Ye·Scientific Reports·Oct 19, 2022· 36 citations