Land conversion to agriculture induces taxonomic homogenization of soil microbial communities globally

Human activities and agricultural intensification are altering ecological communities across spatial scales. While most studies have emphasized local (alpha) diversity, agricultural conversion can also cause biotic homogenization, defined as increased taxonomic or functional similarity among communities (decreased beta-diversity). Prior research has focused mainly on aboveground biota, with fewer studies on belowground microbial communities. Land-use change modifies soil conditions (e.g., compaction, acidification, nutrient imbalances), altering microbial assembly, composition, and diversity. Site-specific studies have shown shifts in microbial taxa and functions following conversion, but generalizable patterns across soil and climate types remain unclear. This study asks: (1) Does agricultural conversion lead to taxonomic and functional biotic homogenization of soil microbiomes at large spatial scales? (2) How do soil microbial taxonomic and functional profiles respond across diverse soils and climates, and which lineages/functions are most affected? The authors hypothesize that conversion to cropland causes taxonomic and functional homogenization of soil microbiomes.

The paper situates its work within evidence that land-use intensification reduces biodiversity and drives biotic homogenization across taxa and habitats. Previous studies have documented aboveground homogenization in plants and animals and localized belowground effects (e.g., Amazon rainforest conversion leading to bacterial homogenization). However, responses of soil microbiomes to agricultural conversion are often site-specific and not easily generalized globally. Functional redundancy in microbial communities can decouple taxonomic shifts from functional changes, yet some functions (e.g., nitrogen fixation, pathogen suppression) may be sensitive to land-use change. There is a recognized knowledge gap regarding consistent, large-scale patterns of soil microbial taxonomic and functional responses to agriculture across different ecosystems, soils, and climates, and the environmental and biotic drivers underlying these patterns.

Study design integrated a continental field survey across China with a global meta-analysis of soil microbiome datasets. Continental survey: 44 regions across China were sampled, each including cropland and adjacent natural ecosystems (forests, grasslands, wetlands) located ~2 km apart to control for climate/soil type. In 2019 (July–August), surface soils (~0–20 cm) were collected from 4–10 plots (2 × 2 m) per ecosystem per region. Three 5-cm-diameter cores per plot were composited, sieved (2 mm), and split for DNA (-80 °C) and soil chemistry (air-dried). In total, 1185 soil samples representing 856 paired soils (303 forest-cropland, 275 grassland-cropland, 278 wetland-cropland pairs) were collected. Soil variables measured: pH, soil organic matter, gravimetric moisture, available phosphorus, nitrate-N (NO3-N), ammonium-N (NH4-N). Enzyme assays: extracellular enzymes for C, N, and P acquisition (BG, CBH, BX, NAG, LAP, APP) using microplate fluorometry at 35 °C; activities expressed as nmol g−1 dry soil h−1. Microbial sequencing: Bacterial 16S rRNA gene (V4–V5; primers 515F/907R) and fungal ITS (ITS5-1737F/ITS2-2043R) amplicons sequenced on Illumina MiSeq PE250. DADA2 pipeline used for quality filtering (min length 100 bp, maxEE=2), error learning, chimera removal (consensus), and ASV inference. Taxonomy assigned against SILVA v138 (bacteria) and UNITE v7 (fungi). Rarefaction to 15,000 (16S) and 21,921 (ITS) reads per sample yielded 31,402 bacterial and 77,962 fungal ASVs. Functional profiling from amplicons used FAPROTAX and PICRUSt2 to infer functional groups. Shotgun metagenomics: Subset of 40 samples (10 per ecosystem) sequenced on Illumina NovaSeq 6000 (150 bp PE). Libraries prepared with ALFA-SEQ kit; QC via FastQC and trimming with Trimmomatic (leading/trailing 3, sliding window 4:15, minlen 36). Average clean yield ~12.2 Gbp/sample. HUMAnN v3.7 (DIAMOND 2.1.6, Bowtie2 2.5.1) with ChocoPhlAn (mpa_vJan21_CHOCOPhlAnSGB_202103) and UniRef90 (201901b) to quantify gene/pathway abundances, regrouped to KEGG Orthologues (KOs), COG categories, and MetaCyc pathways; abundances expressed as TPM. Global meta-analysis: Literature search (2013–Feb 2023) in Web of Science using bacterial and land-use change keywords yielded 297 publications; inclusion required one-to-one cropland–natural ecosystem sequencing pairs and accessible raw data/metadata. After screening, 75 studies remained. Raw 16S (various platforms/primers) were processed with QIIME2/USEARCH and vsearch fastq_filter (-fastq_maxee 1), excluding samples with <10,000 reads. Data were harmonized to an ASV table. To ensure coverage and comparability, forest was used as the natural ecosystem benchmark due to broad representation (over 1300 samples across six continents). Final dataset included 2403 samples from croplands and forests globally. Community and statistical analyses: Beta-diversity (Bray–Curtis) computed for taxonomic (16S) and functional profiles (KEGG and COG). Differences among ecosystem types evaluated via PCoA/PCoA1 and PERMANOVA. Linear mixed-effects models (lme4) assessed agricultural effects on relative abundances of taxa and functional groups with site as random intercept; effect sizes as regression coefficients; significance via Wald χ² tests. DESeq2 used to identify ASVs and functional genes with significant log2-fold changes (apeglm shrinkage; padj thresholds). Ecological processes estimated following Stegen et al.: compute βNTI to infer selection (βNTI < -2 homogeneous selection; > 2 variable selection) and RCbray for dispersal-related processes when |βNTI| < 2. Mantel tests related taxonomic/functional composition to environmental filtering (means) and heterogeneity (within-ecosystem SD) in soil variables and to fungal functional groups (saprotrophs, pathogens, litter saprotrophs; FungalTraits). Structural equation modeling (lavaan) quantified direct/indirect effects of land-use change, soil variables, fungal communities on microbial composition and soil enzyme functions; model fit assessed by χ², RMSEA, CFI.

• Large-scale biotic homogenization: In the continental survey, β-diversity of both microbial taxonomic and functional composition (KEGG, COG) was significantly lower in croplands than in paired natural ecosystems, indicating greater similarity among cropland communities. Example taxonomic β-diversity effects: cropland vs forest F1,915 = 6.429, slope = 0.0016, p < 0.05; cropland vs grassland F1,753 = 1532, slope = 0.0276, p < 0.001; cropland vs wetland F1,770 = 6450, slope = 0.0532, p < 0.001. Functional β-diversity (KEGG) was also lower in croplands than grasslands (F1,88 = 9.021, slope = 0.0885, p < 0.01) and wetlands (F1,85 = 6.886, slope = 0.0527, p < 0.05).
• Global evidence: Meta-analysis across >2400 samples showed cropland vs forest communities differed taxonomically (PERMANOVA R2 = 0.026, p < 0.001) with significantly lower β-diversity in croplands globally (Wilcoxon p < 0.001).
• Range expansion and loss of specialists: Phylotypes present in both habitats occurred in significantly more cropland samples (Wilcoxon p < 0.001), indicating increased geographic ranges. Phylotypes unique to natural ecosystems occurred in fewer samples and likely declined post-conversion (Wilcoxon p < 0.001).
• ASV turnover: On average, agriculture significantly altered ~41–45% of phylotypes across forest, grassland, wetland comparisons. About 20% of ASVs decreased (lost or reduced) and ~23% increased after conversion.
• Consistently enriched phyla in croplands: Chloroflexi, Gemmatimonadota, Planctomycetota, Myxococcota, and Latescibacterota showed higher relative abundance than in forests, grasslands, and wetlands. Dominant phylotype shifts were mainly associated with soil pH and moisture.
• Functional composition: KEGG-based functional composition showed no significant differences between croplands and natural ecosystems; COG-based composition showed minor but significant differences in some comparisons. Less than 10% of KEGG KOs and a small fraction of COGs were affected by conversion (KOs affected: 10% forest, 3% grassland, 8% wetland comparisons; COGs affected: 5%, 1%, 15%, respectively).
• Functional groups (FAPROTAX) and metagenomes: Nutrient-cycling functions (nitrogen fixation, phototrophy, aromatic degradation) decreased in croplands. Metagenomics indicated increased translation/ribosomal/cytoskeleton categories and decreased defense mechanisms in croplands.
• Biogeochemical genes: Croplands showed increased nitrification and denitrification potential (elevated nirk, narG, amoB, hao) and decreased nitrogen fixation (nifH). Genes for organic P mineralization/transport (phn, ugp) decreased. Dissimilatory sulfate reduction genes (apr, dsr) were higher in croplands than forests/grasslands but lower than wetlands. Carbon degradation/fixation genes showed mixed changes with no significant shift in overall carbon metabolism, consistent with functional redundancy.
• Community assembly: Homogeneous selection dominated in croplands (relative importance 94.6%), indicating strong environmental filtering and homogenized conditions due to management. Soil pH filtering was the strongest driver of both taxonomic and functional composition; moisture and NH4-N also important. Fungal saprotroph and pathogen composition correlated directly with bacterial taxonomic composition.
• Soil functions: Enzyme activities for C, N, P cycling differed between croplands and natural ecosystems. Taxonomic composition correlated strongly with soil functions, whereas functional gene composition did not show clear relationships. Lineage-specific associations observed (e.g., Bacteroidota positively correlated with multiple enzymes; Gemmatimonadota showed mixed correlations).

The study confirms the hypothesis that agricultural conversion induces biotic homogenization of soil microbiomes at continental and global scales. Croplands exhibit reduced beta-diversity in taxonomic composition and, to a lesser extent, in functional gene composition, reflecting increased similarity among sites. Mechanistically, agricultural practices homogenize abiotic and biotic conditions (notably soil pH and moisture regimes), intensifying environmental filtering and promoting homogeneous selection. Trait-mediated responses also contribute: taxa with traits conferring broad environmental tolerance and dispersal become more widespread, while habitat specialists decline, leading to range expansion for generalists and loss of rare/narrow-ranged taxa. Despite pronounced taxonomic shifts, functional profiles (especially carbon metabolism) are relatively conserved due to functional redundancy, although key nutrient-cycling capacities (N fixation, P mineralization/transport) are diminished, implying altered nutrient cycling and potential service loss. Biotic interactions with fungal pathogens and saprotrophs further structure bacterial communities under land-use change. The findings underscore that focusing solely on alpha diversity can obscure landscape-scale homogenization, with implications for biodiversity conservation, soil health, and ecosystem service provision. Scale dependence remains important; while broad patterns show homogenization, regional contexts may yield different functional turnover patterns depending on environmental heterogeneity and management regimes.

This work provides large-scale evidence that converting natural ecosystems to cropland leads to taxonomic homogenization and detectable, though smaller, functional homogenization of soil microbiomes. Croplands consistently enrich certain bacterial phyla, expand the ranges of widespread taxa, and reduce the prevalence of habitat specialists. Functionally, carbon metabolism appears buffered by redundancy, but genes underpinning nitrogen fixation and phosphorus mineralization/transport decline, suggesting increased reliance on external nutrient inputs and disrupted symbioses. Environmental filtering (especially pH) and biotic interactions with fungi are key drivers, and homogeneous selection dominates assembly in croplands. These insights advocate for policies that limit further reclamation, promote ecological restoration, and manage soils to maintain environmental heterogeneity and microbial functional diversity. Future research should employ functional expression and activity approaches (e.g., metatranscriptomics, qSIP) to resolve the linkage between taxonomic shifts and in situ function, expand balanced global sampling beyond forest-cropland contrasts, and test how management practices modulate homogenization across spatial scales.

• Global meta-analysis coverage was uneven across continents and ecosystem types, with a strong emphasis on forest–cropland contrasts; other natural ecosystems were underrepresented at global scale.
• Methodological heterogeneity among studies (sequencing platforms, primers, sampling schemes) may introduce comparability constraints despite standardized reprocessing.
• Functional inferences from amplicons (FAPROTAX, PICRUSt2) and limited metagenomic sampling (n = 40) constrain resolution of functional changes; expression-level and activity-based measurements were not performed.
• Continental metagenomic analyses aggregated functions and may miss fine-scale pathway dynamics; observed functional redundancy could mask subtle ecosystem process changes.
• Spatial and temporal variability in agricultural management (crop types, rotations, inputs) was not exhaustively controlled beyond focusing on maize fields in the continental survey.