Soil conditions linked to predicted biosynthetic diversity in Streptomyces

A large-scale ecological-genomic analysis found that distinct combinations of soil and climatic conditions were associated with differences in predicted biosynthetic gene cluster richness among terrestrial soil-associated Streptomyces genomes. Researchers screened 14,597 publicly available assemblies, linked a subset to environmental measurements, and ultimately compared 208 genomes across 89 independent ecological sites. Multivariate analyses resolved three environmentally coherent groups whose genomes differed significantly in the total number of predicted biosynthetic regions.

The findings are computational associations, not proof that soil chemistry directly determines bacterial metabolism or that any predicted cluster produces a particular compound. The study, published in Frontiers in Microbiology, focused on Streptomyces populations associated with terrestrial habitats. Assemblies were retrieved from the National Center for Biotechnology Information RefSeq database, and environmental information was drawn from the corresponding BioSample records.

Investigators examined fields describing geographic origin, isolation source, biome, environmental feature, medium and host association. Marine, freshwater, aquatic, clinical, human-associated, animal-associated, laboratory, industrial and other non-terrestrial records were excluded, as were records with incomplete or ambiguous habitat descriptions. Agricultural soils, forest soils, rhizosphere and plant-associated soils, compost, desert soils and mangrove soils were retained.

Environmental filtering initially left 895 annotated strains. Additional quality checks reduced the collection to 364 genomes, and 208 had complete measurements for all five selected environmental variables: soil pH, soil organic carbon, clay content, mean annual temperature and annual rainfall. To avoid pseudoreplication, strains collected at the same geographic coordinates were combined into one ecological site, with median values used for their environmental measurements.

This produced 89 independent sites. They covered a wide environmental range, including annual rainfall from 20.1 to 2,967.5 millimetres, mean temperatures from minus 5.2 to 27.74 degrees Celsius, soil pH from 4.5 to 8.3, soil organic carbon values from 8.0 to 273.5 and clay content from 75 to 500. Before ordination and clustering, soil organic carbon and rainfall were log-transformed because their distributions were strongly right-skewed, while all five variables were standardized to reduce scale-dependent bias.

Pairwise analysis showed that soil pH was negatively correlated with rainfall, with Pearson’s r of minus 0.67, and with soil organic carbon, with r of minus 0.53. Rainfall and soil organic carbon had a positive correlation of r equal to 0.37. Temperature was moderately inversely related to soil organic carbon, and clay content showed partial covariance with temperature-related gradients.

None of the variable pairs approached complete collinearity. Principal component analysis showed pronounced environmental structure among the sampling sites. PC1 accounted for 44.3% of the total variance and PC2 contributed 31.1%, together explaining 75.4%.

Rainfall and soil organic carbon pointed toward positive PC1 values, while soil pH lay in the opposite direction. Temperature and clay content contributed mainly along positive PC2, identifying an additional axis of differentiation. The sites occupied different positions in this multivariate space rather than forming one homogeneous assemblage, although some occupied intermediate positions between more distinct environmental extremes.

K-means clustering, with the number of groups selected through silhouette analysis, divided the sites into three clusters. Cluster 0 was associated mainly with higher clay content, greater rainfall and comparatively warm conditions. Cluster 1 was characterized by higher soil organic carbon, lower temperatures and a reduced clay-associated profile.

Cluster 2 differed most clearly from the other groups and was associated primarily with alkaline soils and lower rainfall. Kruskal–Wallis tests with false-discovery-rate adjustment confirmed differentiation for every retained variable, with adjusted p-values below 0.001. The largest reported separation involved temperature, with a statistic of 56.61, followed by soil organic carbon at 44.58 and clay content at 44.10.

The researchers then applied GECCO, a genome-mining framework based on conditional random fields, to predict biosynthetic gene clusters and their broad classes. Of the 363 genomes that completed the analysis, the tool identified 15,839 predicted clusters. These included polyketide, non-ribosomal peptide, RiPP, terpene, saccharide and alkaloid-associated regions, together with hybrid architectures.

Polyketide-associated clusters were the most abundant class across the dataset, followed by non-ribosomal peptide and RiPP-associated clusters. Some genomes contained fewer than 30 predicted clusters, while others contained more than 70; the maximum reported was 71. Genome assemblies ranged from approximately 5.8 to 13.6 megabases, and larger genomes generally contained more predicted biosynthetic regions.

GECCO’s confidence metrics were high, although model confidence is not experimental verification. The mean average prediction probability was 0.946, and the mean maximum probability was 0.980. Among predicted clusters, 85.6% had an average probability of at least 0.90 and 59.2% exceeded 0.95.

A total of 70.4% reached a maximum prediction probability of at least 0.99. These values support the consistency of the computational calls, but they do not demonstrate that a predicted region is expressed in soil, produces a detectable metabolite or has a particular biological activity. The study itself notes that comparisons with complementary genome-mining tools and experimental validation would be needed to refine the predictions.

When the 208 genomes with complete environmental records were assigned to the three clusters, total predicted biosynthetic richness differed significantly among groups. The Kruskal–Wallis statistic was 7.28, with p below 0.05. Cluster 0 generally contained both larger genomes and more predicted biosynthetic regions, whereas Cluster 2 more often combined smaller assemblies with lower richness.

The distributions nevertheless overlapped substantially. Class composition was more consistent across the groups than total richness: polyketide-associated clusters dominated in all three ecological settings, non-ribosomal peptide clusters were widespread, and RiPP and terpene-associated regions were less abundant. Cluster 2 showed slightly greater non-ribosomal peptide abundance, while RiPP distributions varied modestly among groups.

The results provide a framework for examining associations between habitat conditions and the organization of specialized-metabolism genes in natural Streptomyces populations, but they do not identify a direct environmental cause of those differences. Variation in lineage composition, genome evolution and the acquisition or retention of biosynthetic regions could also contribute to the observed pattern. Environmental conditions may additionally influence when cryptic pathways are activated, with nutrient availability, microbial competition, temperature and pH acting as possible regulatory signals.

Nevertheless, combining environmental context with genome-scale biosynthetic prediction revealed distinctions that would not be apparent from cluster counts alone and offers an approach for guiding subsequent functional characterization and natural-product research.

Subject of Research: Associations between environmental conditions and predicted biosynthetic gene cluster diversity in terrestrial soil-associated Streptomyces genomes

Article Title: Ecological structuring of biosynthetic potential across terrestrial soil-associated Streptomyces genomes

Article References: Mhammedi, H., Mhammedi, S., Bouda, S., Alaoui, Y. E. H., & Essarioui, A. (2026). Ecological structuring of biosynthetic potential across terrestrial soil-associated Streptomyces genomes. Frontiers in Microbiology, 17, Article 1884766. https://doi.org/10.3389/fmicb.2026.1884766

Image Credits: AI Generated

DOI: 10.3389/fmicb.2026.1884766

Keywords: Streptomyces, biosynthetic gene clusters, ecological genomics, soil microbiology, genome mining, secondary metabolism, environmental metadata, polyketides, non-ribosomal peptides, natural products, Ecological, structuring