Higher-order microbial interactions revealed by comparative metabolic modeling of synthetic communities with varying species composition

Abstract

Understanding how microbial interactions scale with community complexity is key to microbiome engineering and ecological theory. This study investigates emergent metabolic behaviors in controlled in vitro synthetic anaerobic communities of two, three, or four species: cellulolytic bacterium (Ruminiclostridium cellulolyticum), a hydrogenotrophic methanogen (Methanospirillum hungatei), an acetoclastic methanogen (Methanosaeta concilii), and a sulfate-reducing bacterium (Desulfovibrio vulgaris), representing core metabolic guilds in cellulose degradation and carbon conversion. We applied a systems biology framework combining proteogenomics, stoichiometric flux modeling, and SMETANA (Species Metabolic Coupling Analysis) to quantify syntrophic cooperation and competition across configurations. Cooperation peaked in tri-cultures and declined nonlinearly in more complex assemblies. Species roles shifted contextually. Ruminiclostridium cellulolyticum was the dominant donor, adjusting cellulase and hydrogenase expression by partner. Methanosaeta concilii became fully metabolite-dependent while enhancing methanogenesis. Desulfovibrio vulgaris improved syntrophic efficiency via redox and hydrogen turnover. In contrast, Methanospirillum hungatei’s metabolic centrality declined despite higher CH₄ output, suggesting interaction strength depends more on compatibility than richness. Reduced interactions in the four-species community stemmed from a single configuration and need further validation. This study moves beyond descriptive work by quantitatively resolving how metabolic networks rewire across defined communities. By characterizing context-dependent flux shifts at multiple layers, we provide a framework for interpreting and engineering stable, functionally interdependent microbial ecosystems.

Introduction

Microbial transformation of cellulosic biomass into methane (CH₄) and carbon dioxide (CO₂) drives global biogeochemical cycles, shapes microbial communities in animal and human gastrointestinal tracts, and supports industrial applications. These processes influence carbon retention and release [1–3]. Within these communities, diverse species assemble into complex networks involving multiple levels of trophic interactions [4–7], resulting in cascades of cross-feeding. Cellulosic biomass is hydrolyzed into simpler compounds by primary degraders. Secondary consumers, such as methanogens and sulfate-reducing bacteria (SRBs), metabolize these intermediates into CH₄ and CO₂ [7, 8]. Interactions, both cooperative and competitive, within and among these groups shape microbial community structure, stability, and function, influencing ecosystem- and host-level processes [9–12].

The complexity and variability of natural microbial communities complicate the mechanistic understanding of biogeochemical processes. In particular, context-dependent cooperation and competition among microbial guilds hinder accurate prediction of community behavior [9, 13, 14]. While synthetic microbial communities (SynComs) provide tractable models, most prior research has focused on pairwise combinations or uncontrolled multispecies assemblies. As a result, the metabolic consequences of more complex community compositions remain poorly understood [15, 16]. Although omics studies enhance our understanding of microbial diversity and metabolic potential, they often fail to translate static taxonomic composition into predictive models of metabolic function [17–23].

In earlier work, we developed a synthetic microbial system composed of a cellulose hydrolyzer (Ruminiclostridium cellulolyticum), a hydrogenotrophic methanogen (Methanospirillum hungatei), an acetoclastic methanogen (Methanosaeta concilii), and a sulfate-reducing bacterium (Desulfovibrio vulgaris) [24]. These well-characterized anaerobes participate in cellulose degradation and carbon conversion in environments such as wetlands, the gut, and anaerobic digesters [25–28]. Their defined metabolic roles and genome-scale models made them ideal for constructing a minimal and mechanistically interpretable system. The design enabled complete combinatorial testing across bi-, tri-, and quad-species assemblies, allowing high-resolution analysis of interspecies metabolic interactions. This system revealed emergent synergies, such as increased CH₄ and CO₂ production in quad-cultures, as well as negative effects where increased species richness led to reduced cellulose degradation. These results highlight the nonlinear nature of microbial cooperation and competition, although the mechanisms driving these effects remain unclear [29–30].

In this study, we propose that cooperation and competition are context-dependent, shaped by species-specific metabolic capacities and requirements in each configuration. Capturing this variability is essential for predicting microbial community succession. We combined the SMETANA (Species Metabolic Coupling Analysis) framework with proteogenomics and stoichiometric modeling to quantify the strength and direction of interspecies interactions [31, 32]. SMETANA provides two quantitative metrics: Metabolic Interaction Potential (MIP), which reflects cooperative metabolic exchange, and Metabolic Resource Overlap (MRO), which quantifies competition for shared substrates. These computational predictions were validated using metaproteomics and metabolic flux analysis, linking protein expression patterns to carbon partitioning from cellulose hydrolysis to CH₄ and CO₂ production (Fig. 1A). This work advances SynCom-based research by resolving how community composition alters interspecies cooperation and metabolic structure through higher-order interaction dynamics.

Figure 1.

Experimental design and metabolic interactions in synthetic microbial communities. (A) Schematics of the combinatorial experimental setup. (B) Metabolic interaction potential (MIP) and metabolic resource overlap (MRO) across seven synthetic microbial communities. (C) Cumulative production of fermentation products, quantified by carbon and electron molarity, across different microbial assemblies. Heatmaps depict the transferred carbon and electron, with darker colors indicating higher transformation. Arcs illustrate the correlation between MRO and MIP for each microbial assembly, line thickness represents the strength of the correlation. (D) Fluxes of key metabolic reactions contributing to methane (CH₄) and carbon dioxide (CO₂) production. The circles represent seven days of reaction fluxes, and error bars indicate the range of fluxes observed in CH₄ and CO₂ production across eight microbial assemblies. Rc: Ruminiclostridium cellulolyticum, Mh: Methanospirillum hungatei, Mc: Methanosaeta concilii, Dv: Desulfovibrio vulgaris.

Materials and Methods

Genome-scale metabolic model building and community simulation

We used Ruminiclostridium cellulolyticum H10 (ATCC 35319), Desulfovibrio vulgaris Hildenborough (ATCC 29579), M. hungatei JF1 (txid323259), and Methanosaeta concilii (txid990316). Genomes and annotations were downloaded from NCBI. Genome-scale metabolic models (GEMs) were reconstructed using CarveMe, which generates draft models from genomic content [33]. Gap-filling was done using modified VM medium to ensure feasibility, and models were refined with BiGG and KEGG annotations. Community simulations were performed using SMETANA to estimate cross-feeding, metabolic dependencies, and interaction strengths [32]. Simulations used a virtual medium mimicking experimental condition. SMETANA ran with default settings, yielding three metrics: Metabolic Interaction Potential (MIP) indicates cooperative metabolite acquisition. MRO quantifies competition by shared metabolite import, and the SMETANA Score estimates syntrophic likelihood. Mathematical definitions are available in the original SMETANA publication [1].

Experimental conditions and metabolite measurements

Experiments followed protocols from prior work [24]. Cultures were grown in 160 ml anaerobic bottles with 20 ml modified VM medium plus 10 g/L cellulose. Ruminiclostridium cellulolyticum and D. vulgaris were revived in VM and LS4D media at 34°C. Methanospirillum hungatei and M. concilii were pre-cultured in RST medium with H₂/CO₂ (80:20) and 50 mM acetate at 37°C. All strains were acclimated in VM at OD₆₀₀ of 0.5 and inoculated at ~5 × 10⁷ cells per species. Cultures were incubated at 34°C, 200 rpm for 7 days under 3% H₂/97% N₂. Daily, 2 ml samples were centrifuged at 14 000 rpm for 15 min, supernatants were stored at −80°C. Headspace gas (5 ml) was collected in vacuum-sealed vials. Medium (2 ml) was replenished daily and pH reset to 7.3. Sulfate treatments received 2 ml of 100 mM K₂SO₄ daily, controls received water. Metabolites (acetate, lactate, ethanol, formate, glucose, cellobiose) were measured by HPLC with refractive index detection. Gases (H₂, CH₄, CO₂) were analyzed by GC with thermal conductivity detection, pressure was digitally recorded. All assays were performed in technical duplicates using standard curves. Detailed metabolite data are provided in Table S1.

Calculation of Gibbs free energy change

Gibbs free energy under biological conditions (ΔG′) was calculated as:

where ΔG°′ is standard Gibbs energy at pH 7 and 25°C, R = 8.314 J/mol·K, T = 298.15 K, and Q is the reaction quotient from metabolite concentrations or gas pressures.

Protein identification and quantification

Samples collected postincubation were processed using established protocols [34, 35]. Pellets were washed with nanopure water and lysed in buffer (10 mM Tris–HCl, 1% SDS, 0.1 M DTT) at 60°C for 1 h. Proteins were precipitated with trichloroacetic acid overnight at 4°C, washed with cold acetone, and resuspended in guanidine buffer. Concentrations were quantified via the Bicinchoninic Acid assay [36]. 20 mg protein samples were processed using filter-aided sample preparation, digested with Trypsin/LysC, and labeled using the TMTpro™ 16plex set. Peptides were fractionated into 46 groups and consolidated into 28 super-fractions. Each was further separated via UHPLC and analyzed on an Orbitrap Eclipse Tribrid MS with multi-notch MS3. MS1 scans were performed at 120000 resolution (375–1500 m/z). MS2 used CID (NCE 35) and ion trap detection. Synchronous precursor selection allowed HCD (NCE 65) and MS3 detection at 50000 resolution (100–500 m/z).

Spectra were searched using Proteome Discoverer v3.1 with SEQUEST against a custom four-species database. Peptides were filtered to 1% FDR [37, 38]. Protein groups were excluded from downstream analysis. Abundances were calculated from TMT intensity, normalized to dataset totals, and adjusted relative to each species’ total proteome abundance [39–41].

Construction of the stoichiometric model

A set of overall metabolic reactions per species was constructed and fitted to observed metabolite levels (Figs S1 and S2). Adenosine Triphosphate (ATP) yields were sourced from KEGG, MetaCyc, and prior literature [42–45]. The model included pathways for cellulose degradation, hydrogenotrophic and acetoclastic methanogenesis, sulfate reduction, and fermentation. Protein abundances from metaproteomics were used to estimate pathway activity, assuming a positive correlation with reaction rates and scaled by ATP yield. This provided a semi-quantitative estimate of species-specific energy contributions. Deviations between model and measurements were attributed to biological variation or analytical limits. Code and data are available at: https://github.com/thepanlab/ProStoichiometric.

Statistical analysis

Metabolite data were analyzed using Student’s t-test. Protein abundance differences were analyzed using the DESeq R package [46], with q-values calculated by the Benjamini–Hochberg method [46]. Proteins with q < 0.05 were considered significant.

Results

Microbial complexity enhances cooperation and methane output

Different combinations of R. cellulolyticum, M. hungatei, M. concilii, and D. vulgaris, including one mono-culture, three bi-cultures, three tri-cultures, and one quad-culture, were co-cultured in a medium with cellulose as the sole carbon source (Fig. 1A). These synthetic communities were analyzed using the SMETANA framework to compute MIP and MRO as proxies for cooperative metabolism and resource competition, respectively (Fig. 1B) [32].

Among bi-cultures, R. cellulolyticum with M. concilii had the highest MIP, indicating strong cooperation. In contrast, combinations with M. hungatei or D. vulgaris showed higher MRO, reflecting greater competition (Fig. 1B). Although these microbes utilize different primary substrates (cellulose, hydrogen, lactate), SMETANA inferred competition for shared imports like ammonium, phosphate, glycerol, and ferric iron. These micronutrients are essential for biosynthesis, so MRO values reflect auxiliary, not primary substrate, competition.

Adding a third species to form tri-cultures significantly altered MIP-MRO dynamics. The R. cellulolyticum–M. concilii–M. hungatei tri-culture showed 60% and 300% MIP increases over respective bi-cultures, and a 45% MRO decrease. Similarly, the R. cellulolyticum–M. concilii–D. vulgaris tri-culture increased MIP by 20% and 200% and reduced MRO by 48%. The R. cellulolyticum–M. hungatei–D. vulgaris tri-culture showed a 50% MIP increase and 53% MRO decrease. These results suggest that tri-cultures enhance cooperation and reduce competition.

In the quad-culture, MIP dropped by 22% compared to cumulative bi-cultures, but MRO declined by 76% (Fig. 1B), suggesting that while cooperation may diminish slightly, reduced competition stabilizes metabolism in the full community.

To assess functional outcomes, carbon and electron transfers were measured (Fig. 1C). The R. cellulolyticum–M. hungatei bi-culture transferred 0.42 ± 0.007 mmol carbon and 3.33 ± 0.06 mmol electrons to CH₄. Adding D. vulgaris increased both values by 112%. In the quad-culture, carbon and electron transfer reached 1.37 ± 0.03 mmol and 10.99 ± 0.21 mmol, a 226% increase over the corresponding tri-culture and combined bi-cultures. These findings suggest D. vulgaris functions as a syntrophic lactate oxidizer that facilitates hydrogen transfer and supports methanogenesis. Acetate production was negatively correlated with MIP (r = −0.69, P = .002) and positively with MRO (r = 0.76, P = .005), indicating that acetate accumulates under competition and declines with cooperation.

Metaproteomics identified 1296 proteins across all eight communities using species-specific peptides (Table S2). A refined stoichiometric model incorporated six key reactions [24]: lactate fermentation, hydrogenic acetogenesis, mixed-acid fermentation, hydrogenic lactate oxidation, hydrogenotrophic methanogenesis, and acetoclastic methanogenesis (Fig. S1), with ATP yields incorporated. Reaction contributions were fitted to cumulative measurements of lactate, acetate, ethanol, H₂, CO₂, CH₄, and biomass. Model performance was high (R² > 0.9, NRMSE = 0.084) (Fig. S2), validating calculated carbon and energy fluxes. Flux analysis identified acetoclastic methanogenesis as the primary CH₄ and CO₂ source, especially in multi-species cultures. This pathway was dominant for methane production and energy conservation. Total carbon and electron fluxes to CH₄ and CO₂ in tri- and quad-cultures exceeded the sum of bi-cultures, revealing synergistic, non-additive interactions. These results, consistent with previous findings [24], emphasize acetoclastic methanogenesis as a central driver of emergent cooperation in complex communities (Fig. 1D).

Community context shapes cellulose degradation and fermentation in R. cellulolyticum

Cellulose degradation by R. cellulolyticum increased in the presence of other species (Fig. 2A), with cocultures showing higher cellulase protein abundance than the mono-culture, as revealed by metaproteomics (Fig. 2C). However, the enhancement showed weak correlations with MIP (r = 0.30, P = .51) and MRO (r = 0.35, P = .40), suggesting that predicted metabolic interactions alone cannot explain cellulolytic activity. MIP and MRO capture community-level exchange potential and competition but do not account for species-specific regulation, spatial interactions, or local conditions that influence cellulase expression.

Figure 2.

Metabolic activities of R. cellulolyticum in synthetic communities. (A) Comparison of cellulose consumption by R. cellulolyticum among different cultures. Significant differences with P-values determined using Student’s t-test, and adjusted for the false discovery rate, are marked by ^* for P-value < .05, ^** for P-value < .01, and ^*** for P-value < .001. The error bars are defined as standard deviation. (B) Cross-feeding interactions between R. cellulolyticum and other species in all the synthetic communities. (C) Protein expression profile of the major carbon metabolism pathways in R. cellulolyticum. (D) Fluxes of the key reactions in R. cellulolyticum modeled in all cultures. Rc: Ruminiclostridium cellulolyticum, Mh: Methanospirillum hungatei, Mc: Methanosaeta concilii, Dv: Desulfovibrio vulgaris. Egl: β-endoglucanase, Exl: Β-exoglucanase, Bgl: β- glucosidase, CE: cellobie 2-epimerase, Ats: ABC transporter, pgm: phosphoglucomutase, HK: hexokinase, Pfk: 6-phosphofructokinase, Gpm: 2,3-bisphosphoglycerate-dependent phosphoglycerate mutase, Eno: Enolase, Ldh: lactate dehydrogenase, Por: pyruvate:ferredoxin oxidoreductase, Hyd: periplasmic [NiFeSe] hydrogenase, Nhd: NADH-fd reductase, Pta: phosphate acetyltransferase, Ack: acetate kinase, Aad: acetaldehyde dehydrogenase, His: alcohol dehydrogenase.

Cellulose degradation efficiency varied with species composition. Over 7 days, the R. cellulolyticum mono-culture degraded 184.4 ± 0.5 mg of cellulose. Coculture with D. vulgaris increased degradation by 10.5% (P = .002), but adding M. concilii reduced it by 3% (P = .03), and M. hungatei had no significant effect (P = .2). Including both methanogens with R. cellulolyticum and D. vulgaris in a quad-culture led to a 3% increase over the bi-culture (P = .008) and a 5% increase compared to the tri-cultures with M. concilii (P = .003) or M. hungatei (P = .02). These results suggest that D. vulgaris enhances degradation, while methanogens have context-dependent effects.

SMETANA-based simulations revealed metabolic interdependencies in these communities (Tables 1 and 2, Fig. 2B) [31]. The SMETANA score quantifies essential metabolite exchange between species pairs. Higher scores indicate stronger dependence, while lower scores suggest greater autonomy. Each species may act as a donor or a receiver. In the R. cellulolyticum–D. vulgaris bi-culture, mutualism was evident; the score was 2.12 when R. cellulolyticum was the donor and 2.46 as the receiver. This likely reflects facultative syntrophy, where R. cellulolyticum supplies lactate and hydrogen to D. vulgaris, and receives ammonium, sulfur compounds, and small organics in return. In contrast, in bi-cultures with M. concilii or M. hungatei, R. cellulolyticum had high donor scores (4.60 and 4.85), but receiver scores dropped by 47.8% and 83.1%, respectively. This asymmetry suggests R. cellulolyticum contributes more than it gains from methanogens, unlike its mutualistic relationship with D. vulgaris. In tri- and quad-cultures, R. cellulolyticum remained the dominant donor, reinforcing its ecological role in cellulose breakdown and provision of intermediates to syntrophic partners.

Table 1.

Summary of metabolic interactions, stoichiometric modeling, and proteomic analyses within tri-cultures.

Microbial assembly	Species pair		SMETANA	Stoichiometric model			Proteomics
	Donor	Receiver	Fold change	Metabolite	Reaction	Fold change (P-value)	Protein name	Fold change (P-adj)
Rc & Mc & Mh			1.55 ▲	Acetate	Hydrogenic acetogenesis	1.08 (*) ▲	Acetate kinase	ns
	Rc	MC			Mixed-acid fermentation	> 100 (***) ▼	Aldehyde-alcohol dehydrogenase	1.89 (***) ▼
Rc & Mc & Dv			1.55 ▲		Hydrogenic acetogenesis	1.13 (***) ▼	Acetate kinase	ns
					Mixed-acid fermentation	ns	Aldehyde-alcohol dehydrogenase	1.79 (*) ▼
Rc & Mh & Mc			1.13 ▼	H2, CO2	Hydrogenic acetogenesis	1.10 (**) ▼	Pyruvate flavodoxin/ferredoxin oxidoreductase	1.49 (**) ▼
	Rc	Mh			Mixed-acid fermentation	> 100 (***) ▼	Aldehyde-alcohol dehydrogenase	1.89 (***) ▼
Rc & Mh & Dv			1.06 ▼	H2, CO2	Hydrogenic acetogenesis	1.26 (**) ▲	Pyruvate flavodoxin/ferredoxin oxidoreductase	1.56 (*) ▼
					Mixed-acid fermentation	> 100 (***) ▼	Aldehyde-alcohol dehydrogenase	2.28 (*) ▼
Rc & Dv & Mc			1.34 ▲	Lactate	Lactate fermentation	1.56 (*) ▼	Lactate dehydrogenase	1.16 (*) ▲
Rc & Dv & Mh	Rc	Dv	2.02 ▲			1.53 (*) ▼		ns
Rc & Mc & Mh			>100 ▼	Acetate	Acetoclastic methanogenesis	ns	Methyl-coenzyme M reductase	2.80 (***) ▲
Rc & Mc & Dv	Mc	Rc	>100 ▼			1.38 (**) ▼		2.08 (***) ▲
Rc & Mh & Mc			>100 ▼	H2, CO2	Hydrogenotrophic methanogenesis	3.29 (*) ▼	Formylmethanofuran dehydrogenase	> 30 (***) ▲
Rc & Mh & Dv	Mh	Rc	>100 ▼			1.81 (*) ▲	Formylmethanofuran dehydrogenase	> 30 (***) ▲
Rc & Dv & Mc	Dv	Rc	>100 ▼	Lactate	Hydrogenic lactate oxidation	9.30 (***) ▼	Phosphate acetyltransferase	1.36 (*) ▼
Rc & Dv & Mh			>100 ▼			2.83 (***) ▼		1.31 (*) ▼

▲, significant increase; ▼, significant decrease; -, no significant changes were observed.

^*For Padj < .05, ^**for Padj < .01, and ^***for Padj < .001.

Table 2.

Summary of metabolic interactions, stoichiometric modeling, and proteomic analyses within quad-culture.

Species Pair		SMETANA	Stoichiometric model			Proteomics
Donor	Receiver	Fold change	Metabolite	Reaction	Fold change (P-value)	Protein name	Fold change (P-adj)
Rc	Mc	1.73 ▲	Acetate	Hydrogenic acetogenesis	1.53 (**) ▲	Acetate kinase	2.15 (***) ▲
				Mixed-acid fermentation	2.72 (*) ▼	Aldehyde-alcohol dehydrogenase	ns
Rc	Mh	1.20 ▼	H2, CO2	Hydrogenic acetogenesis	1.30 (***) ▲	Acetate kinase	ns
				Mixed-acid fermentation	1.47 (*) ▼	Aldehyde-alcohol dehydrogenase	14.2 (***) ▲
Rc	Dv	1.42 ▲	H2, CO2	Lactate fermentation	5.18 (***) ▼	Lactate dehydrogenase	1.18 (*) ▼
Mc	Rc	>100 ▼	Acetate	Acetoclastic methanogenesis	2.55 (***) ▲	Methyl-coenzyme M reductase	> 30 (***) ▲
Mh	Rc	>100 ▼	H2, CO2	Hydrogenotrophic methanogenesis	1.83 (**) ▲	Formylmethanofuran dehydrogenase	>30 (***) ▲
Dv	Rc	>100 ▼	Lactate	Hydrogenic lactate oxidation	9.97 (***) ▼	Phosphate acetyltransferase	> 30 (***) ▲
Mc	Mh	NA	CO2	Acetoclastic methanogenesis	2.53 (***) ▲	Methyl-coenzyme M reductase	2.98 (***) ▲
Mh	Mc	1.22 ▼	CO2	Hydrogenotrophic methanogenesis	6.01 (***) ▲	Formylmethanofuran dehydrogenase	ns
Dv	Mc	1.27 ▼	Acetate	Hydrogenic lactate oxidation	ns	Phosphate acetyltransferase	15.11 (***) ▲
Dv	Mh	NA	H2	Hydrogenic lactate oxidation	3.52 (*) ▼	Phosphate acetyltransferase	ns

▲, significant increase; ▼, significant decrease; -, no significant changes were observed.

^*For Padj < .05, ^** for Padj < .01, and ^*** for Padj < .001.

Metaproteomics revealed changes in R. cellulolyticum’s fermentation enzyme profiles across configurations (Fig. 2C). Compared to mono-culture, sugar-related ABC transporter (Ats) protein levels were similar in bi-cultures but decreased over 30-fold in tri- and quad-cultures (q < 0.0001), suggesting reduced substrate uptake in complex communities. Protein levels of NADH-fd reductase (Nhd), acetaldehyde dehydrogenase (Aad), and alcohol dehydrogenase (His) increased significantly in bi-cultures (fold change >30, q < 0.0001), but declined in tri- and quad-cultures (q < 0.0001), indicating community-driven restructuring of fermentation pathways. Hydrogenase (Hyd) protein abundance, linked to H₂ production, dropped by 1.37-fold in the R. cellulolyticum–M. concilii bi-culture and by over 30-fold in tri- and quad-cultures (q < 0.0001). This suggests active suppression of H₂ production to maintain low hydrogen levels favorable for syntrophy. In simpler cultures, R. cellulolyticum likely increases H₂ production to boost ATP yield. In more complex communities, efficient H₂-consuming partners relieve this burden, allowing metabolic resources to shift. This highlights cooperative adaptation and metabolic flexibility.

To quantify these transitions, we modeled flux through R. cellulolyticum’s key fermentation pathways—lactate fermentation, mixed-acid fermentation, and hydrogenic acetogenesis—across all community types (Fig. 2D). Lactate fermentation flux dropped significantly with D. vulgaris, declining by 88% in the quad-culture relative to mono-culture (P < .0001). Although exergonic (ΔG′ ≈ −32 kJ/mol) (Table 3), this flux likely decreased due to lactate consumption reducing its thermodynamic drive. Mixed-acid fermentation also declined in the presence of M. hungatei and was undetectable in the R. cellulolyticum–M. hungatei–D. vulgaris tri-culture. Despite being moderately exergonic (ΔG′ ≈ −26 kJ/mol), reduced activity may reflect a redirection of reducing equivalents toward more favorable reactions under syntrophy. In contrast, hydrogenic acetogenesis flux increased in the presence of other species. It rose by 58.8% in the tri-culture with M. hungatei and D. vulgaris (P = .0004), and by 61.8% in the quad-culture (P = .0009), relative to mono-culture. This pathway showed the greatest ΔG′ improvement, from −7.5 kJ/mol in mono-culture to nearly −29 kJ/mol in quad-culture, driven by H₂ removal. These findings show that M. hungatei and D. vulgaris promote a thermodynamically favorable shift from classical fermentation to hydrogenic acetogenesis, supporting cooperative energy flow and enhanced facultative syntrophy.

Table 3.

Gibbs free energy (ΔG°, in kJ/mol) of key metabolic reactions across different microbial community assemblies.

Assemblies	Lactate fermentation	Hydrogenic acetogenesis	Mixed-acid fermentation	Hydrogenic lactate oxidation	Hydrogenic methanogenesis	Acetoclastic methanogenesis
Rc	−106.54	−76.74	−134.71	−24.02	−191.83	−105.06
Rc&Dv	−103.15	−75.54	−131.81	−22.89	−133.41	−44.69
Rc&Mc	−103.24	−76.49	−131.78	−23.62	−129.84	−40.83
Rc&Mc&Dv	−104.63	−74.89	−131.68	−22.80	−130.82	−41.36
Rc&Mc&Mh	−101.84	−72.78	−128.93	−22.07	−129.17	−38.46
Rc&Mc&Mh&Dv	−102.17	−71.53	−127.41	−21.06	−128.12	−35.52
Rc&Mh	−101.41	−76.55	−131.62	−23.97	−126.93	−40.27
Rc&Mh&Dv	−101.65	−72.62	−128.45	−21.67	−128.36	−37.24

Rc, R. cellulolyticum, Mc, M. concilii, Mh, M. hungatei, Dv, D. vulgaris.

Community complexity increases syntrophic dependency and methanogenic activity in M. concilii

Metanosaeta concilii showed increased dependency on other species as community complexity rose. In tri- and quad-cultures, SMETANA scores were zero when M. concilii acted as a donor, indicating no detectable metabolic contribution. Instead, it functioned solely as a receiver, relying on partner-derived metabolites. Compared to the R. cellulolyticum–M. concilii bi-culture, the SMETANA score between these two species increased by 72.6% in the quad-culture (Table 2 and Fig. 3A), indicating stronger metabolic interdependence with complexity.

Figure 3.

Metabolic activities of M. concilii and of M. hungatei in synthetic communities. (A) Cross-feeding interactions between M. concilii and other species in all synthetic communities. (B) Functional proteins profiling of acetoclastic methanogenesis. (C) Flux of acetoclastic methanogenesis modeled in M. concilii. (D) Cross-feeding interactions between M. hungatei and other species in all synthetic communities. (E) Functional proteins profiling of hydrogenotrophic methanogenesis. (F) Flux of hydrogenotrophic methanogenesis modeled in M. hungatei. Rc: Ruminiclostridium cellulolyticum, Mc: Methanosaeta concilii, Mh: Methanospirillum hungatei, Dv: Desulfovibrio vulgaris, CdhC: acetyl-CoA decarboxylase/synthase, CdhA: anaerobic carbon-monoxide dehydrogenase, Hdr: heterodisulfide reductase, Mcr: methyl-coenzyme M reductase, Fmd: formylmethanofuran dehydrogenase, Fdh: formate dehydrogenase.

Metaproteomic analysis supported this trend. The abundance of four enzymes associated with acetoclastic methanogenesis in M. concilii—acetyl-CoA decarboxylase/synthase (CdhC), carbon monoxide dehydrogenase (CdhA), heterodisulfide reductase (Hdr), and methyl-coenzyme M reductase (Mcr)—significantly increased in tri- and quad-cultures compared to the bi-culture (fold change >2.0, q < 0.0001) (Fig. 3B). This upregulation likely reflects increased acetate availability, consistent with R. cellulolyticum shifting toward hydrogenic acetogenesis in more complex cultures (Fig. 2D). These results suggest that M. concilii’s methanogenic activity is primarily substrate-driven, while community complexity may indirectly influence its role by modulating upstream fluxes.

Acetoclastic methanogenesis flux was modeled in all cultures containing M. concilii (Fig. 3C). In the bi-culture with R. cellulolyticum, flux was 0.32 ± 0.03 mmol. Adding M. hungatei had no significant effect (P = .87), while D. vulgaris reduced flux by 28% (P = .005). The quad-culture showed a 153% increase compared to the tri-culture of R. cellulolyticum, M. concilii, and M. hungatei (P < .0001), and a 47% increase relative to the combined fluxes of the bi-culture and R. cellulolyticum–M. concilii–D. vulgaris tri-culture (P = .006). These trends align with thermodynamic predictions. Although acetoclastic methanogenesis is highly exergonic (ΔG′ ≈ −36 kJ/mol) (Table 3), its flux depends on acetate availability and substrate competition. Reduced flux with D. vulgaris likely reflects competition, while the increase in the quad-culture indicates enhanced acetate production and sharing. These findings suggest that M. concilii’s methanogenesis is driven by acetate supply and strengthened in complex communities through improved cooperative exchange.

Reduced metabolic contribution of M. hungatei and enhanced methanogenesis

The metabolic contribution of M. hungatei decreased as community complexity increased, particularly with D. vulgaris and M. concilii present (Fig. 3D). In the bi-culture with R. cellulolyticum, M. hungatei acted as a donor with an SMETANA score of 0.82. Adding D. vulgaris to form a tri-culture reduced the donor score to zero. A similar reduction occurred in more complex assemblies. In the tri-culture of R. cellulolyticum, M. hungatei, and M. concilii, the donor score was 2.54, which decreased by 18.3% upon addition of D. vulgaris to form the quad-culture. Along with reduced donor contribution, M. hungatei showed decreased dependence on community-derived metabolites with greater complexity. As a receiver in the R. cellulolyticum–M. hungatei bi-culture, its SMETANA score was 4.85, which dropped by 11.5% with M. concilii added. Similarly, in the tri-culture of R. cellulolyticum, M. hungatei, and D. vulgaris, the receiver score was 4.59, decreasing to 4.03 when M. concilii was included in the quad-culture. These findings suggest that M. hungatei becomes metabolically marginalized in higher-order consortia, with both contribution and reliance constrained by more competitive or dominant species like D. vulgaris and M. concilii.

Two key enzymes in hydrogenotrophic methanogenesis, formylmethanofuran dehydrogenase (Fmd) and formate dehydrogenase (Fdh), were detected across co-cultures (Fig. 3E). Compared to the R. cellulolyticum–M. hungatei bi-culture, their abundance significantly increased in tri- and quad-cultures (fold change >30, q < 0.0001), indicating elevated methanogenesis activity driven by more complex interactions.

Desulfovibrio vulgaris significantly enhanced hydrogenotrophic methanogenesis efficiency (Fig. 3F). Over seven days, its presence led to a 78.6% increase in flux (P = .01) compared to the R. cellulolyticum–M. hungatei bi-culture. The flux also rose by 525% in the quad-culture compared to the tri-culture of R. cellulolyticum, M. concilii, and M. hungatei (P < .0001). These findings align with thermodynamics, as hydrogenotrophic methanogenesis is strongly exergonic (ΔG′ ≈ −131 kJ/mol) under low H₂ conditions (Table 3). Desulfovibrio vulgaris likely promotes hydrogen turnover, maintaining low partial pressures that favor this pathway. This supports efficient electron transfer and explains the enhanced methanogenesis seen in complex communities.

Desulfovibrio vulgaris modulates hydrogen metabolism for facultative syntrophy

Desulfovibrio vulgaris oxidizes lactate to acetate, CO_2, and H₂ in the absence of sulfate [47]. The accumulation of H₂ was differentially influenced by two methanogens—M. concilii and M. hungatei. Co-culturing D. vulgaris with M. concilii resulted in lower H₂ accumulation compared to co-culturing with M. hungatei (Fig. 4A). Introducing M. hungatei to the bi-culture of R. cellulolyticum and D. vulgaris led to a 13.2% decrease in H₂ accumulation (P-value = .005), which is comparable to the 13.0% decrease observed in the tri-culture of R. cellulolyticum, D. vulgaris, and M. hungatei compared to the mono-culture (P-value = .0008). Adding M. concilii to the same bi-culture resulted in a 14.5% reduction in H₂ accumulation (P-value < .0001). However, the subsequent addition of M. hungatei to this tri-culture did not lead to significant changes in H₂ accumulation compared to the tri-culture baseline. Although M. concilii does not metabolize hydrogen, metaproteomic analysis showed increased expression of acetoclastic methanogenesis enzymes in multi-species cultures (Fig. 3B), while hydrogenase expression in R. cellulolyticum decreased. These results indicate that the presence of M. concilii is associated with lower hydrogen accumulation and shifts in fermentative metabolism across the community.

Figure 4.

Metabolic activities of D. vulgaris in synthetic communities. (A) Comparison of H₂ accumulation in the cocultures with D. vulgaris. Significant differences with P-values determined using Student’s t-test, and adjusted for the false discovery rate, are marked by ^* for P-value < 0.05, ^** for P-value < 0.01, and ^*** for P-value < 0.001. (B) Cross-feeding interactions between M. hungatei and other species in all synthetic communities. (C) Functional proteins profiling of hydrogenic lactate oxidation. (D) Flux of hydrogenic lactate oxidation modeled in D. vulgaris. Rc: Ruminiclostridium cellulolyticum, Mh: Methanospirillum hungatei, Mc: Methanosaeta concilii, Dv: Desulfovibrio vulgaris, Cyt c3: cytochrome C3, Hmc: high-molecular cytochrome, Hyd: hydrogenase, Lpe: L-lactate permease, Ldh: L-lactate dehydrogenase, Coo: membrane-bound Coo hydrogenase, Por: pyruvate:ferredoxin oxidoreductase, Pta: phosphate acetyltransferase.

The presence of M. concilii or M. hungatei affected the metabolic contribution of D. vulgaris to community function in contrasting ways—M. concilii increased the contribution, while M. hungatei decreased it (Fig. 4B). When D. vulgaris acted as the metabolic donor, the addition of M. concilii to the R. cellulolyticum–D. vulgaris bi-culture increased its SMETANA score by 114.6%. In contrast, the addition of M. hungatei decreased the SMETANA score of D. vulgaris to zero. Furthermore, in the tri-culture of R. cellulolyticum, M. concilii, and D. vulgaris, adding M. hungatei reduced the SMETANA score of D. vulgaris by 21.4%. When D. vulgaris acted as receiver, R. cellulolyticum was its primary metabolic donor, its SMETANA score increased by 34.0% with M. concilii and by 102.4% with M. hungatei, compared to bi-culture. However, the combined presence of both M. concilii and M. hungatei in the quad-culture resulted in only a 42.5% increase, which is less than the additive effect of their individual contributions. This suggests a negative synergistic interaction between M. concilii and M. hungatei in shaping the metabolic dependency of D. vulgaris.

Metaproteomics identified six key enzymes in the hydrogenic lactate oxidation pathway of D. vulgaris, L-lactate permease (Lpe), L-lactate dehydrogenase (Ldh), pyruvate:ferredoxin oxidoreductase (Por), phosphate acetyltransferase (Pta), periplasmic [NiFeSe] hydrogenase (Hyd), and membrane-bound Coo hydrogenase (Coo) (Fig. 4C). Compared to the bi-culture of R. cellulolyticum and D. vulgaris, the abundance of L-lactate permease, pyruvate:ferredoxin oxidoreductase, and phosphate acetyltransferase—enzymes involved in lactate fermentation—significantly increased in tri-cultures and quad-cultures, correlating with decreased lactate accumulation. Conversely, the abundance of periplasmic [NiFeSe] hydrogenase, involved in H₂ production, decreased by up to 92.3-fold (q-value < 0.0001) in the tri-culture of R. cellulolyticum, M. concilii, and D. vulgaris, and by 30-fold (q-value <0.0001) in the tri-culture of R. cellulolyticum, M. hungatei, and D. vulgaris, as well as in the quad-culture. These findings were supported by flux modeling of hydrogenic lactate oxidation across all community combinations (Fig. 4D). In the bi-culture of R. cellulolyticum and D. vulgaris, the flux of hydrogenic lactate oxidation was 0.18 ± 0.01 mmol. This flux decreased significantly in the tri-cultures and quad-culture, with an 89.9% reduction in the quad-culture. Although hydrogenic lactate oxidation is exergonic (ΔG′ ≈ −36 kJ/mol), the thermodynamic benefit is maximized under low H₂ conditions (Table 3). The presence of hydrogenotrophic methanogens likely maintains low hydrogen partial pressure, improving the energy yield of lactate oxidation and facilitating interspecies hydrogen transfer. Thus, the observed reduction in H₂ accumulation reflects tighter syntrophic coupling with hydrogen consumers, not suppression of lactate metabolism. This shift enables more efficient energy conservation and contributes to overall metabolic stability in complex microbial communities.

Microbial interactions from modeling carbon and energy flow

The stoichiometric model estimated metabolic exchange fluxes to infer interactions among community members (Fig. 5A). Interactions between R. cellulolyticum and M. concilii were linked to acetate fluxes, those with M. hungatei to H₂ and CO₂, and those with D. vulgaris to lactate and H₂. The model also inferred interactions between M. concilii and M. hungatei via CO₂, M. concilii and D. vulgaris via acetate, and M. hungatei and D. vulgaris via lactate and H₂. Variations in metabolic pathways and partner efficiency likely explain community-dependent interaction differences.

Figure 5.

Stoichiometric modeling of synthetic communities. (A) Conceptual metabolic network among four microbial species in the synthetic community. (B) Exchanging fluxes estimating the metabolic interactions between species. The edges represent the exchange flux, and the circles represent the biomass. Rc: Ruminiclostridium cellulolyticum, Mh: Methanospirillum hungatei, Mc: Methanosaeta concilii, Dv: Desulfovibrio vulgaris.

Quantitative analysis of the bi-cultures revealed substantial variability in the strength of these metabolic exchanges (Fig. 5B). The strongest interaction occurred between R. cellulolyticum and M. hungatei, with H₂ and CO₂ exchange flux of 0.68 ± 0.003 mmol. This was followed by the acetate exchange flux between R. cellulolyticum and M. concilii, measured at 0.32 ± 0.03 mmol. The weakest interaction was observed between R. cellulolyticum and D. vulgaris, with a lactate and H₂ exchange flux of only 0.18 ± 0.001 mmol. These findings suggest that the strength of metabolic exchange was highly dependent on metabolic complementation.

The nature of microbial relationships—such as cooperation, inhibition, or competition—and their intensity were significantly altered by community complexity, impacting overall metabolic dynamics and efficiency (Fig. 5B). Compared to the bi-culture of R. cellulolyticum and M. hungatei, the addition of M. concilii reduced the H₂ and CO₂ exchange flux of H₂ and CO₂ by 72.1% (P-value < .0001). Conversely, the inclusion of D. vulgaris increased the exchange flux by 72.1% (P-value = .006). Subsequent addition of M. concilii to this tri-culture caused no significant change in the exchange flux (P-value = .76), suggesting that M. concilii weakened the interaction between R. cellulolyticum and M. hungatei in the absence of D. vulgaris. This pointed to a dynamic equilibrium in the quaternary interaction between R. cellulolyticum and M. hungatei. In the bi-culture of R. cellulolyticum and M. concilii, the acetate exchange flux remained unaffected with the addition of M. hungatei (P-value = .87) but decreased by 31.2% (P-value = .004) upon the inclusion of D. vulgaris. However, the subsequent addition of M. hungatei to this tri-culture led to a 146.9% increase in exchange flux (P-value < .0001), indicating that M. hungatei mitigated the inhibitory effect of D. vulgaris on the interaction between R. cellulolyticum and M. concilii in the quad-culture. Similar trends were observed for the bi-culture of R. cellulolyticum and D. vulgaris, where the addition of M. concilii reduced the H₂ and lactate exchange flux by 88.9% (P-value < .0001), with a further reduction of 61% (P-value = .005) upon subsequent addition of M. hungatei. Remarkably, the simultaneous presence of both M. concilii and M. hungatei in the quad-culture increased the exchange flux by 83.3% (P-value < .0001), indicating a synergistic enhancement of microbial interactions, where the inhibitory effects of M. concilii and M. hungatei on the interaction between R. cellulolyticum and D. vulgaris were mitigated.

Beyond the common pairwise exchanges, three emergent fluxes were identified in the tri-cultures and the quad-culture. In the tri-cultures, a CO₂ flux of 0.012 ± 0.001 mmol was observed between M. concilii and M. hungatei, while an acetate flux of 0.005 ± 0.001 mmol was detected between M. concilii and D. vulgaris. Both fluxes increased by 75%–88% (P-value < .0001) in the quad-culture. Meanwhile, an H₂ flux of 0.07 ± 0.02 mmol between M. hungatei and D. vulgaris remained unchanged in the quad-culture (P-value = .78). These observations underscored the complexity and dynamic nature of metabolic interactions in multi-species communities, characterized by emergent fluxes and altered interactions.

These fluxes aligned with thermodynamic favorability. Hydrogenotrophic methanogenesis (ΔG′ ≈ −131 kJ/mol) likely drove the strong R. cellulolyticum–M. hungatei interaction. The acetate flux with M. concilii reflected favorable acetoclastic methanogenesis (ΔG′ ≈ −36 kJ/mol). Weaker exchanges with D. vulgaris likely result from tighter energy margins in lactate fermentation (ΔG′ ≈ −32 kJ/mol) and substrate competition (Table 3). Overall, the thermodynamic context helps explain cooperative dynamics in increasingly complex microbial communities.

Discussion

Anaerobic microbial interactions in carbon and electron transfer have been well characterized in natural and simplified lab systems. Prior studies defined key functions of hydrogenotrophic and acetoclastic methanogens, sulfate reducers, and cellulolytic fermenters, highlighting cross-feeding’s role in methane production and redox balance. However, most insights stem from pairwise cultures or descriptive omics, lacking resolution on how interaction networks change in complex communities [24, 48–50]. In this study, we address this limitation by using a fully combinatorial four-species synthetic system to explore how complexity reshapes cooperation and competition during cellulose degradation to CH₄. The model organisms used have defined metabolic roles, genomic resources, and ecological relevance, providing a tractable, interpretable platform. Unlike studies testing hypotheses or reporting correlations, our approach integrates SMETANA modeling, stoichiometric flux analysis, and proteogenomics to resolve how metabolic roles shift with community configuration. The aim was not pathway discovery but to build a quantitative foundation for understanding context-dependent syntrophic dependencies.

Our results demonstrate that community complexity alters both the magnitude and structure of metabolic interactions in nonlinear, context-specific ways. Tri-cultures often had stronger cooperation than bi-cultures, with increased MIP and reduced MRO. While the four-species community had lower MIP than some tri-cultures, this result reflects a single configuration and cannot be generalized. Rather than species richness suppressing cooperation, our data suggest that richness reshapes interaction networks based on partner compatibility. These results align with recent studies showing that interaction strength is driven more by functional traits and network topology than community size [51–53].

Ruminiclostridium cellulolyticum consistently functioned as the central metabolic donor. In more complex configurations, it showed regulatory changes that supported community-level syntrophy. For example, it decreased hydrogen and cellulase production while shifting carbon flux toward hydrogenic acetogenesis. Thermodynamically, this pathway is only mildly favorable in monoculture (ΔG′ ≈ −7.5 kJ/mol) but becomes more favorable under syntrophic conditions (ΔG′ ≈ −29 kJ/mol in the quad-culture). These adjustments minimized product accumulation, facilitated electron flow, and prioritized community-level energy balance over individual energy yield, consistent with other syntrophic fermenters [54, 55].

Methanogens showed divergent responses. Methanosaeta concilii became a fully dependent receiver in tri- and quad-cultures while increasing its methanogenic output. This was marked by elevated expression of acetoclastic pathway enzymes and increased flux, especially when D. vulgaris was present, likely enhancing acetate availability through cross-feeding. Acetoclastic methanogenesis, being highly exergonic (ΔG′ ≈ −36 kJ/mol), serves as an efficient sink for carbon and electrons under cooperative conditions (Table 3) [56–58]. In contrast, M. hungatei showed decreased donor and receiver roles with increasing complexity but displayed substantially enhanced hydrogenotrophic methanogenesis in the quad-culture. This decoupling of exchange roles and functional output mirrors observations in facultative syntrophs, where metabolic roles persist even with diminished network centrality [32]. Given that hydrogenotrophic methanogenesis is highly favorable (ΔG′ ≈ −131 kJ/mol), it provides a robust electron sink for the community (Table 3).

Desulfovibrio vulgaris emerged as a key modulator of community metabolism. While its donor role varied by species context, its presence consistently improved both acetoclastic and hydrogenotrophic methanogeneses. Proteomic and flux data showed increased lactate oxidation alongside reduced hydrogenase expression and hydrogen flux. This indicates that hydrogen was still produced but consumed rapidly by methanogens such as M. hungatei, maintaining low partial pressures and thermodynamic favorability for lactate oxidation (ΔG′ ≈ −26 kJ/mol) (Table 3) [59, 60]. Rather than eliminating hydrogen production, D. vulgaris likely benefits from interspecies hydrogen transfer, using methanogens as efficient electron sinks. This redox balance improves energy efficiency and supports metabolic stability in complex communities.

While SMETANA effectively predicted cooperation and competition trends, combining it with proteomic and flux data yielded deeper insight. For instance, the predicted reduction in competition in the quad-culture aligned with proteomic evidence of niche specialization and reduced fermentation redundancy. This multi-layered approach supports systems biology perspectives that emphasize integrating models with omics to uncover emergent community properties beyond what sequence data alone can reveal [61, 62].

It is important to emphasize that these findings derive from a defined four-species system. Although the taxonomic composition is not novel, the full combinatorial design and simplicity allowed precise mapping of metabolic interactions, carbon flows, and species-specific contributions. This framework makes it possible to isolate and study higher-order interactions that are difficult to access in natural microbiomes, offering a foundation for hypothesis-driven analysis of community metabolic dynamics [63].

In summary, this study demonstrates that microbial cooperation and competition are shaped not merely by species richness but by context-dependent metabolic compatibility and partner-specific interactions. By leveraging a fully combinatorial and mechanistically resolved synthetic system, we show that microbial communities dynamically reorganize their metabolic networks in response to changing community context, producing emergent behaviors that are not predictable from pairwise associations or genomic potential alone. These findings advance beyond previous studies by providing a quantitative framework that links species identity, flux distribution, and functional outcomes. This approach offers a foundation for predictive modeling of microbial ecosystems and informs the rational design of stable consortia for applications in bioconversion, waste treatment, and synthetic ecology.

Author contributions

D.W. conceptualized and designed the study, conducted the metaproteomics analysis, and performed the modeling with assistance from K.A.H., B.A., and D.A.S. The biological interpretation was a collaborative effort involving D.W., C.P., P.C., B.A., F.Z., R.S.T., X.T., A.Z., J.Z., K.B.D., M.J.M., M.H.W., and D.A.S. C.P. and M.H.W. served as the principal investigators for the funding sources. D.W. drafted the manuscript, with all authors contributing to its revision and approving the final version.

Conflicts of interest

None declared.

Funding

This study is primarily funded by the Genome Sciences Program of the Office of Biological and Environmental Research under project FWP SCW1677. Part of the proteomics and modeling methodology development was supported by National Institutes of Health grant (R01AT011618).

Data availability

The proteomics datasets generated during the current study are available in ProteomeXchange Consortium via the PRIDE (Proteomics Identification Database) partner repository with the dataset identifier PXD056517.