Non-additive microbial community responses to environmental complexity

The study investigates how increasing environmental complexity—defined as the number of distinct available nutrients—affects microbial community properties, specifically growth yield and taxonomic diversity. Prior observations from natural and synthetic systems conflict on whether more complex environments enhance productivity and biodiversity. The authors ask whether general, predictive principles govern community responses as resources are combined, and whether outcomes are deterministic across nutrient sets. They frame the problem using an epistasis-inspired approach to quantify non-additivity in community phenotypes when combining environmental resource pools, and use consumer resource models to set expectations and interpret deviations. The work aims to disentangle how niche partitioning, resource-use capabilities, species similarity, and metabolic interactions shape community assembly in increasingly complex environments.

Classical ecological theory (competitive exclusion, niche partitioning) suggests more nutrient types can enable more niches and higher diversity. However, organism-specific resource-use capabilities, ecological niche overlap, and interspecies interactions can cause departures from this expectation. Studies show mixed evidence linking environmental complexity to community productivity and diversity in natural ecosystems and synthetic microcosms. Determinism vs stochasticity in assembly has been debated, with some environments yielding stable communities. The concept of epistasis from genetics provides a quantitative framework to assess non-additive effects of combining perturbations, here adapted to combining environments to evaluate community phenotypes. Consumer resource models (MacArthur-type and extensions) have been successful in reproducing ecological patterns and are appropriate for studying resource-driven dynamics. Prior work highlights roles for generalists vs specialists, niche overlap, cross-feeding, and environmental regimes in shaping community structure and function.

Experimental design: The authors assembled synthetic microbial communities with varying initial species richness (notably a 13-species community, "com13"; smaller communities of 3 and 4 species) and exposed them to hierarchical combinations of defined carbon sources to vary environmental complexity while keeping total carbon constant. They selected 13 organisms spanning diverse taxa and metabolic strategies from an initial pool of 15, based on monoculture metabolic profiling (Biolog PM1 plate) and practical culturing criteria. They chose 32 carbon sources (across sugars, sugar alcohols, organic acids, amino acids, polymers) using criteria to maximize between-organism variance and potential for coexistence, and also ran a 5-carbon-source set with all combinations. Culturing: Organisms were grown individually, washed, combined at equal initial OD600 to assemble communities, and inoculated into M9 minimal media supplemented with equimolar combinations of carbon sources at a fixed total concentration (50 mM carbon). Cultures were grown in 96 deep-well plates at 30 °C with shaking, and passaged every 24–48 h for six passages to approach stable compositions. Growth yield was quantified by OD600 (endpoint minus initial). Outliers were removed via robust Z-scores (MAD) with threshold 3.5. Community composition: For com13, endpoint taxonomic structure was measured by 16S rRNA amplicon sequencing (V4 region; QIIME2 and DADA2 processing; naive Bayes classifier) across replicates and environmental complexities. For smaller communities, colony counts on LB agar allowed species identification via morphology. Diversity metrics computed included species richness (S) and Shannon entropy (H). Hierarchical clustering used Spearman correlations and UPGMA. Epistasis metrics: To quantify non-additivity when combining environments A and B into AB, the authors defined yield epistasis EY = Y(AB) − (Y(A) + Y(B))/2, where Y is growth yield. For diversity, they defined ES = S(AB) − max(S(A), S(B)) and EH = H(AB) − max(H(A), H(B)). Distributions of these metrics were compared to consumer resource model (CRM) simulations as baselines. Consumer resource modeling: They implemented a MacArthur-style dynamical CRM with Monod kinetics. Species i have abundance N_i; resources α have abundance R_α. Uptake rates are defined by a stoichiometric consumption matrix C_{iα} scaled by R_α/(k_α + R_α); a fraction f of consumed flux is leaked back as byproducts via a conversion matrix D_{αβ}, available to all species. Growth rate depends on energy uptake minus maintenance m_i, scaled by g_i and resource energy content w_α. Parameters were informed by experimental growth magnitudes (e.g., OD600 ~0.3 in ~20 h at 50 mM C glucose) and literature; typical settings included g_i = 1 (or organism-specific), w_α = 1e-3, k_α = 1e-4 g/mL, leakage f = 0.8. Simulations spanned 288 h with 48 h dilutions to mimic passaging. Communities of sizes S = 13, 4, 3 were simulated across 32 resources and hierarchical combinations (32 singletons, 16 pairs, 8 quartets, 4 octets, 2 sixteens, 1 thirty-two). Two CRM regimes: CRM-A (similar average resource-use across species; lower generalist-specialist contrast, lower niche overlap) and CRM-B (species-specific resource-use fractions matched to measured monoculture capabilities; generalists and specialists; higher niche overlap). Niche overlap ρ was computed as μ^2/σ^2 from the consumption matrix C. Metabolic byproducts (1–10) were included with 25% transition probability in D to test cross-feeding effects. Multiple random instantiations (n=50 per environment) sampled resource preference variability. Statistical analyses: One-sided t-tests evaluated significance of EY distribution shifts vs CRM, and of ES, EH vs CRM-A. Additional tests compared diversity metrics across complexities and yields across species richness levels.

• Growth yield scaling: CRM predicted near-constant community yield with increasing environmental complexity when total carbon is fixed. The 13-species community (com13) matched this additivity: EY distribution centered at 0 and not significantly different from CRM (paired one-sided t-test p=0.84). However, smaller consortia (3- and 4-species) exhibited non-additive increases in yield with complexity, with EY distributions positively skewed and significantly different from CRM (com3 p=3.6×10^-4; com4 p=2.1×10^-4). Example deviations: D-glcNAc + D-galacturonate showed EY = +0.13 (~2σ); D-glucose + D-sorbitol showed EY = −0.19 (~3σ), indicating resource interference.
• Species richness and Shannon diversity do not increase with environmental complexity in com13: No significant increases in S or H from single- to 32-carbon-source conditions (one-tailed paired t-test p=0.107 for S, p=0.180 for H). Some single-carbon environments had higher diversity than more complex ones.
• Diversity epistasis is predominantly negative: Experimental ES and EH distributions were left-skewed relative to CRM baselines centered at 0. Means: ES = −0.65 ± 1.47, EH = −0.50 ± 0.58. Compared with CRM-A, ES and EH were significantly negative (p=0.034 and 1.26×10^-4, respectively).
• Outcome types when combining environments: Four classes identified; most prevalent was Type III (dominance of the least-diverse constituent environment) accounting for ~40% of cases, driving negative ES. Type I (new taxa appearing only in combinations) occurred in ~20% of cases, consistent with beneficial interactions.
• Deterministic assembly and competition: Community compositions were consistent across replicates, indicating deterministic assembly driven by resource identity. Many organisms that grew in monoculture on a given resource did not persist in community due to competition (e.g., on D-glucose, only Pseudomonas aeruginosa remained despite nearly all species growing in monoculture). Some facilitation observed (e.g., P. aeruginosa persisting on maltose likely via organic acids secreted by E. coli).
• Role of community size, specialization, and niche overlap: Smaller communities had many no-growth cases in simple environments (out of 63 environments: com3, 33 no-growth; com4, 15; com13, 4), explaining yield increases with added resource types. Yields increased with initial species richness. CRM with reduced resource-use breadth and lower niche overlap recapitulated yield increases in smaller consortia. For diversity, CRM-B (generalists/specialists with higher niche overlap reflecting measured capabilities) reproduced flat diversity vs complexity and negative ES/EH skew; CRM-A predicted increasing diversity (up to ~6 coexisting species) not seen experimentally. CRM-B predicted a maximum of ~3 coexisting species, aligning with experiments. Increased niche overlap was generally associated with reduced diversity; cross-feeding slightly mitigated diversity loss.
• Additional quantitative notes: In com13, dominant genera were Pseudomonas (76.7% of environments) and Acinetobacter (18.0%). In 70.4% of environments more than one organism persisted. Species abundance distributions exhibited few high-abundance taxa with long tails, similar to natural microbiota.

The findings show that, when total carbon is fixed, community growth yield can be additive across increasing environmental complexity provided sufficient functional redundancy and resource-use coverage across species. Smaller or more specialized communities fail to utilize all resources in simple environments, leading to non-additive yield increases as complexity rises. However, taxonomic diversity does not necessarily scale with the number of resource types; instead, competition and niche overlap in communities with uneven metabolic capabilities often produce negative diversity epistasis, where combined environments support fewer taxa than expected from constituents. Deterministic assembly tied to resource identity, dominance by particular taxa (e.g., Pseudomonas vs Acinetobacter), and occasional facilitation explain observed structures. CRM analyses clarify how generalist-specialist makeup, degree of niche overlap, and cross-feeding shape non-additivity, providing rules for predicting responses to resource combinations. These insights suggest that simply increasing environmental complexity is not sufficient to boost community diversity; rather, careful tuning of species traits and interactions is needed to design or control microbiomes.

This work introduces an epistasis-based framework to quantify non-additive microbial community responses to environmental complexity and demonstrates contrasting behaviors for growth yield and taxonomic diversity. In 13-species consortia, yields are largely additive with complexity, while smaller communities show overyielding-like increases. Diversity frequently exhibits negative epistasis, with combined environments often dominated by the least-diverse constituent due to competition under uneven metabolic capabilities and high niche overlap. Consumer resource models parameterized by measured resource-use recapitulate these patterns and highlight roles for specialization, similarity, and cross-feeding. Practically, environmental complexity alone is insufficient to maintain or enhance diversity; engineering community composition (generalists vs specialists), reducing niche overlap, and fostering beneficial interactions are more effective levers. Future research should explore how dilution regimes and timescales, alternative limiting nutrients (nitrogen, phosphorus), and microbial strategies for sequential vs simultaneous resource use modulate these epistatic patterns, extending to natural and engineered ecosystems.

• Simplified synthetic system: Communities and environments do not capture the full complexity, spatial structure, and fluctuating conditions of natural microbiomes.
• Organism and resource selection bias: Chosen species and carbon sources may influence generality of conclusions; not representative of a specific biome.
• Fixed total carbon and minimal medium: Results depend on constant total carbon and M9 conditions; other nutrient currencies (N, P), micronutrients, or stoichiometries could alter outcomes.
• Temporal and dilution regime: Endpoint measurements after serial passaging may miss transient dynamics; different dilution frequencies may change assembly and diversity.
• Modeling assumptions: CRM parameter choices (e.g., Monod kinetics, leakage fraction, byproduct structure) and simplified cross-feeding may not capture all metabolic interactions; no spatial structure.
• Diversity measurement: 16S-based relative abundances may have resolution limits; agar-based identification for small consortia relies on morphology.
• Epistasis definitions: Alternative baselines for expected phenotypes could yield different quantitative assessments of non-additivity.