Engineering natural microbiomes toward enhanced bioremediation by microbiome modeling

Microbes drive key biogeochemical processes and underpin applications in agriculture, food fermentation, bioenergy, and pollutant degradation. Synthetic microbiomes can outperform single strains and natural communities by distributing metabolic tasks and alleviating burdens across members. Despite advances in synthetic biology at the strain level, practical principles and tools for engineering natural microbiomes are scarce. Two broad strategies exist: top-down (selective pressures and inoculation steer community assembly) and bottom-up (rational assembly of strains based on known interactions). This work aims to (1) demonstrate that herbicide application combined with targeted inoculation can steer different soil microbiomes toward functionally convergent, high-degradation communities; (2) develop a scalable microbiome modeling framework (SuperCC) to characterize metabolic interactions and predict performance; and (3) identify keystone species to build simplified synthetic microbiomes with enhanced and stable bioremediation capacity.

The paper situates its approach within emerging microbiome engineering strategies that leverage community-level functions. Prior studies highlight that synthetic consortia can share metabolic loads and achieve complex transformations but lack generalizable design rules. Top-down assembly under selective pressures (e.g., nutrients, host cues) and bioaugmentation can shape community function; however, outcomes are variable and dose-dependent. Community metabolic modeling has been recognized as a powerful tool to infer interactions and optimize consortia, yet scalable, practical pipelines for natural-to-synthetic translation remain limited. The authors build on LEfSe- and random forest-based marker discovery, strain isolation, and genome-scale modeling to link compositional shifts to function, addressing gaps in principled keystone selection and in silico optimization.

Top-down phase: Three distinct soils (red, yellow cinnamon, purple; pH 5.0–8.1) were subjected to herbicide application and bioaugmentation with single strains and consortia. Two pollutants were used: a complex herbicide derivative (BO; degradable only by a synergistic consortium of Pseudomonas X, Y and Comamonas sp. 7D-2) and a simpler pollutant (DBH/DBHB/DBH8; degradable by single strains Comamonas sp. 7D-2 or Pseudomonas sp. H8). Inoculation strategies included single strain (e.g., 7D-2), synergistic (X-1 and 7D-2), or competitive/complete consortia (e.g., H8 and 7D-2). Single vs repeated and low- vs high-dose inoculations were compared; repeated high-dose inoculation was selected for efficiency. Soil microcosms were sampled over 30 days (Days 0, 3, 9, 18, 30) to monitor degradation and community dynamics. Analytical assays: Degradation efficacy was quantified via HPLC/RSLC for BO, bromoxynil, and DBH in media after soil inoculation (BO/DBH at 50 mg/L; 0.5 g soil; 10 h incubation). BO extraction used dichloromethane; bromoxynil and DBH were analyzed in acetonitrile/water/acetic acid mobile phase. Community profiling: 16S rRNA gene amplicon sequencing (V4–V5; Illumina MiSeq) was performed on 320 of 348 collected samples (post-QC). Sequences were processed with QIIME2/DADA2 and assigned via SILVA. Metagenomics (Illumina NovaSeq) on 181 samples (Days 0 and 30 in selected treatments) used MEGAHIT assembly, Prodigal ORF prediction, CD-HIT for non-redundant catalogs, DIAMOND taxonomy (NCBI NR), and annotations via eggNOG and KEGG. Diversity analyses included α-diversity (Shannon/Simpson), β-diversity (Bray-Curtis NMDS), and PCA on KEGG modules. Keystone identification: Differentially abundant genera were detected using LEfSe (LDA > 3.0). Random forest classifiers identified key ASVs. Parallel strain isolation from soils (290 isolates; 16S-based taxonomy) provided candidate keystones. Integrated phylogenetic mapping of isolates with key ASVs guided species selection. Modeling (SuperCC): The authors developed SuperCC, a scalable multi-compartment community metabolic modeling framework integrating single-species models into a shared extracellular compartment with transport and exchange reactions. Simulations used parsimonious FBA (pFBA) to maximize community biomass while minimizing exchange fluxes; FVA identified key reactions. Scenarios included equal abundances, unconstrained abundances, target-organism maximization, and defined abundances. All combinations of keystone sets were simulated under MM medium with BO as sole C/N source and BO supplemented with glucose, NH4+, or NO3−. Predictions focused on growth, degradation, and metabolite exchanges. Experimental validation of modeling: Pairwise and three-member co-cultures assessed growth and degradation in BO/DBH media and in soils. LC–MS detected predicted exchanged metabolites (e.g., succinate, L-glutamate, hypoxanthine, xanthine, D-mannose, fumarate, D-glucosamine, L-proline) from co-cultures (e.g., 7D-2×X-1; 7D-2×H8). DNA-SIP with 13C-labeled 4-hydroxybenzoic acid (an intermediate of BO degradation) assessed assimilation across an 8-strain consortium. RNA-seq compared single vs co-culture gene expression for strains X-1 and 7D-2 to validate pathway activation. Additional pot experiments tested predicted keystone-enhanced consortia in three soils. Data and statistics: Rarefaction, ANOVA with Tukey’s HSD, and two-sided Student’s t-tests were used. Random forest models trained on 80% of data with 100,000 trees, repeated 20 times.

• Top-down assembly enhanced bioremediation: Herbicide application plus inoculation (especially repeated, high-dose) significantly improved degradation of both complex (BO) and simple (DBH/DBHB) pollutants across three disparate soils. Inoculating synergistic or competitive consortia outperformed single strains for both pollutant types.
• Convergent functional succession: α-diversity decreased post-treatment; functional gene profiles converged across soils. Proteobacteria and Actinobacteria dominated among degradation-associated genera. In treated microbiomes, 52%–100% of the BO-degradation key enzyme signal originated from the inoculated Comamonas sp. 7D-2, particularly notable in yellow cinnamon soil.
• Differential abundance across treatments: LEfSe (LDA > 3.0) identified 133, 34, and 52 genera with differential abundance between early and late phases in BO7D-2×X-1 treatments for purple, yellow cinnamon, and red soils, respectively; of these, 27, 6, and 17 showed consistent contrasting changes. For DBHB7D-2×8H treatments, 67, 71, and 35 genera were differentially abundant across the three soils. Only subsets increased over time (e.g., 11, 11, and 40 enriched genera in BO7D-2×X-1 for purple, red, and yellow cinnamon; 11, 8, and 12 enriched in DBHB7D-2×8H). Over 40% of significantly changed genera were shared among soils under the same treatments, exceeding 65% in yellow cinnamon soils. Bacillus and Sphingobacterium increased significantly in all soils in BO groups.
• Keystone identification and synthetic communities: Eighteen species (including inoculants) were selected as keystones based on abundance shifts, random forest key ASVs, and isolation. Modeling and experiments showed that not all keystones improved performance; four species—Bacillus sp. (P56), Lysinibacillus sp. (LMS), Acinetobacter sp. (A6), and Bradyrhizobium sp. (BR1)—consistently enhanced growth and degradation of inoculated consortia in BO contexts.
• Metabolic interactions validated: SuperCC predicted cross-feeding among keystones and degraders. In 7D-2×X-1, 7D-2 secreted succinate and L-glutamate consumed by a partner; X-1 secreted hypoxanthine. LC–MS detected all predicted exchange metabolites in co-cultures. Supplementing predicted metabolites enhanced growth and degradation in monocultures and pairs. RNA-seq showed upregulation of genes encoding enzymes involved in the exchanged-metabolite pathways in co-culture versus single culture.
• Environmental modulation: Adding glucose and/or ammonium increased community growth as predicted. BR1 enabled nitrate utilization benefitting communities with 7D-2, consistent with SuperCC predictions.
• Dynamics with substrate depletion: As BO concentrations decreased, the biomass ratio 7D-2:X-1 increased in co-culture and in soils, matching model predictions.
• DBH group: Simulations and experiments largely agreed; only limited enhancement (notably by P56) was observed, likely due to already high DBH degradation rates in well-mixed conditions (~95%).

The study demonstrates that applying selective pressures (herbicide) alongside targeted bioaugmentation reliably steers different natural soil microbiomes toward functionally convergent states with enhanced degradation. This validates the top-down premise that community assembly can be guided under defined pressures. However, fully assembled functional microbiomes remain complex and potentially unstable for applications; thus, a bottom-up reduction using identified keystones is valuable. By integrating statistical biomarker discovery, strain isolation, and scalable community metabolic modeling (SuperCC), the authors bridge compositional shifts with mechanistic interaction maps, enabling rational selection of degraders and helper strains. Experimental validations—metabolite detection, supplementation assays, DNA-SIP, and transcriptomics—corroborate predicted cross-feeding, emphasizing that mutualistic exchanges can be pivotal to community function, potentially underestimated in natural settings. Environmental amendments (e.g., glucose, NH4+, NO3− via BR1) can further tune performance, suggesting biostimulation avenues. The convergence across diverse soils suggests generalizability of the framework for bioremediation-focused microbiome engineering.

This work introduces a practical, combined top-down/bottom-up framework for converting natural soil microbiomes into function-enhanced synthetic microbiomes for herbicide bioremediation. The authors develop SuperCC, a scalable modeling platform that predicts community performance and metabolic exchanges, enabling informed keystone selection and community optimization. Eighteen keystone species were identified, with four helper species (Bacillus sp. P56, Lysinibacillus sp. LMS, Acinetobacter sp. A6, Bradyrhizobium sp. BR1) validated to boost degrader consortia. The study also proposes computational synthetic cell design by transferring essential reactions learned from community models to single strains, offering a new direction for efficient strain engineering. Future work could extend SuperCC-guided design to other pollutants and environments, refine keystone selection with additional omics layers and longitudinal dynamics, and translate synthetic communities to field-scale applications.

• Incomplete functionalization of all candidate keystones: Several taxa identified by abundance shifts did not enhance community performance upon isolation and testing, indicating that differential abundance does not always equate to functional contribution.
• DBH systems showed limited improvement: For DBH-treated groups, enhancements beyond baseline were minimal (except P56), potentially due to already high degradation rates, limiting detectable gains.
• Proxy substrate in SIP: 13C-labeled 4-hydroxybenzoic acid was used as a BO pathway intermediate due to unavailability of labeled BO, which may not fully capture all BO-specific assimilation dynamics.
• Context dependence and resource competition: Some negative or neutral interactions were observed when combining strains, underscoring potential resource competition and environmental dependency that may constrain generalization.
• Taxonomic and nomenclature inconsistencies in strain labels across sections could affect exact replication without consulting Supplementary materials.