How Genome-Scale Models Redefine Marine Heterotrophs Beyond the Copiotrophic‑Oligotrophic Dichotomy

Background and Motivation

Marine heterotrophic microbes drive organic‑matter degradation and carbon cycling, yet the long‑standing classification into copiotrophic (fast‑growing) and oligotrophic (slow‑growing) groups fails to capture substrate‑specific metabolic niches. The authors set out to replace this binary framework with a finer‑grained system based on genome‑scale metabolic capabilities.

Data Collection and Curation

The study leveraged the Ocean Microbial Database (OMD) hosted on microbiomics.io, which contains ~35,000 microbial genomes (metagenome‑assembled, single‑cell amplified, and cultured isolates). Only genomes with completeness > 80 % and contamination < 5 % (assessed by CheckM and Anvi’o) were retained. Using dRep with a 95 % ANI threshold, redundant genomes were removed, yielding 3,738 high‑quality heterotrophic bacterial genomes after excluding 180 photosynthetic cyanobacteria.

For phylogenetic analysis, an additional 66 reference genomes from the BiGG database were added, resulting in 3,976 genomes; eight were omitted due to insufficient single‑copy marker genes, leaving 3,976 for tree construction with GTDB‑Tk v2.1.0 and GTDB r214.

Metabolic Modeling and Quality Control

Each of the 3,738 genomes was modeled with CarveMe v1.5.1, generating 60 independent models per genome (total 224,280 models). The authors observed that model diversity plateaued around 60 models, indicating sufficient coverage of possible metabolic configurations.

To assess model consistency, a novel consistency score C was introduced (see image). Only genomes with C ≥ 0.8 (1,578 genomes) were kept for downstream analysis.

Quantifying Metabolic Strategies

Using COBRApy v0.25.0, flux balance analysis (FBA) was performed under two conditions: substrate‑sufficient and substrate‑limited (flux reduced to 50 % of the sufficient uptake). A sensitivity coefficient S was defined (see image) to quantify how growth rate μ changes with substrate limitation f = 0.5. Values S ≥ 0.8 indicated strong growth sensitivity to that substrate.

Unsupervised Clustering of Metabolic Niches

The 1,578 genomes were filtered further by removing 100 genomes with total variance > 0.1 across the 60 models, leaving 1,478 genomes (88,680 model instances). Each instance contributed 11 substrate‑sensitivity features.

These features were standardized and fed into a Self‑Organizing Map (SOM) using kohonen v3.0.12 on a 20 × 20 toroidal hexagonal grid, with 1,500 iterations, learning rates (0.025, 0.01), and default neighborhood radius. After training, K‑means clustering partitioned the SOM into eight distinct clusters.

Growth rate categories were assigned using gRodon‑derived dCUB values, and the geographic distribution of each cluster was validated against Tara Oceans, BioGeoTraces, Malaspina metagenomes, and the Global ASV dataset.

Validation of Model Predictions

Experimental validation involved 186 marine microbes cultured on 11 carbon sources. CarveMe‑derived predictions matched literature data with 75.5 % accuracy and 87.4 % precision. Bootstrap analysis confirmed that these performance metrics significantly exceeded random expectations.

Eight Metabolic Clusters

Cluster 6 – Fast Copiotrophic : 79.5 % of genomes predict dCUB < ‑0.08 (fast growth). Representative orders include Enterobacterales, Flavobacteriales, Rhodobacterales, and Pseudomonadales. Growth is insensitive to removal of any of the 11 substrates.

Clusters 1, 5, 8 – Substrate‑Specific Slow Oligotrophs : dCUB ≈ ‑0.111. Cluster 5 (61.8 % of this group) includes Opitutales (Verrucomicrobiota) and Pelagibacterales, highly enriched (435 % and 362 % respectively). These taxa show strong growth limitation by specific substrates.

Clusters 2, 3, 4, 7 – Intermediate Growers with Specialized Substrate Preferences : Growth rates lie between fast and slow groups. Each cluster responds primarily to a single substrate class: amino acids (cluster 2), carboxylic acids (cluster 3), carbohydrates (cluster 4), and B‑vitamins (cluster 7). Recent studies corroborate that dominant surface‑layer heterotrophs are often slow‑growing copiotrophs, consistent with the observed carbohydrate preference of cluster 4.

Implications and Future Directions

The new eight‑cluster framework replaces the decades‑old copiotrophic/oligotrophic dichotomy with a substrate‑driven ecological classification. By reducing the complexity of marine heterotrophic communities to eight functional groups, the system can be integrated into global biogeochemical models, enabling predictions of organic‑matter degradation and carbon fluxes under climate‑change scenarios without enumerating thousands of individual species.

Overall, the combination of genome‑scale metabolic modeling, quantitative substrate‑sensitivity analysis, and unsupervised machine learning provides a reproducible pipeline for dissecting microbial metabolic niches in any large‑scale genomic dataset.