From shared genes to staged decay: ecosystem-specific genomic potential for litter decomposition in soils

Technical Summary: From shared genes to staged decay: ecosystem-specific genomic potential for litter decomposition in soils

Problem Statement
Plant litter decomposition is a dominant terrestrial carbon flux, returning approximately 60 Pg C yr⁻¹ to the atmosphere as CO₂. While it is established that decomposition rates vary significantly across climates and ecosystems, the specific microbial traits driving these ecosystem-scale differences remain unresolved. Previous research has utilized metagenomics to quantify genes encoding plant cell-wall-degrading enzymes (CAZymes), yet it remains unclear whether ecosystem differences in decay rates reflect distinct enzymatic repertoires or merely quantitative shifts in shared genes. Furthermore, it is unknown whether genomic data improves the prediction of decomposition rates beyond established environmental and soil property covariates. Current analyses often lack resolution, grouping CAZymes at coarse functional levels or failing to organize them along the biochemical stages of plant cell-wall decomposition (initial, mid, and last stages).

Methodology
The study paired standardized litter decomposition assays with soil metagenomics across a global scale.

• Experimental Design: The authors utilized the "TeaComposition" initiative, employing green tea (labile) and rooibos tea (more recalcitrant) bags incubated for up to 24 months across 264 sites worldwide (295 soil samples).
• Sampling: Soil samples were collected at the 24-month endpoint. Mass loss was quantified at 3, 12, and 24 months.
• Genomic Analysis: Shotgun metagenomes were generated from the soil samples. The study focused primary inference on 196 European plots (grasslands, broadleaf forests, and coniferous forests) while utilizing the full global dataset (n=295) for predictive modeling and cross-ecosystem transfer tests.
• Bioinformatics:
  • Assembled contigs were taxonomically classified, with a focus on bacteria, fungi, and archaea.
  • CAZyme genes were annotated using dbCAN3 and organized by target substrates (cellulose, hemicelluloses, pectin, lignin) and biochemical decomposition stages (initial, mid, last, broad).
  • The study constructed a stage-resolved catalogue comprising 17.6 million CAZyme annotations (European subset) spanning 379 families and 11,865 subfamilies.
  • Analyses included assessing gene abundance, subfamily enrichment, phylogenetic diversity within families, and directional selection on protein sequences.
• Statistical Modeling:
  • Variation partitioning was used to determine the unique and shared explanatory power of environmental variables, spatial factors, soil functional indicators (respiration, biomass), and CAZyme profiles on 24-month mass loss.
  • Predictive models (elastic-net regression and random forests) were trained to predict decomposition trajectories (mass loss slopes and Area Under the Curve) under four validation regimes: within-ecosystem, leave-one-ecosystem-out, leave-one-continent-out, and out-of-Europe.

Key Results

1. Decomposition Patterns: Forest soils consistently exhibited higher tea litter mass loss than grasslands across both tea types and time points. Broadleaf and coniferous forests did not differ significantly from each other in overall mass loss, though specific contrasts existed at certain time points.
2. CAZyme Richness vs. Abundance: Ecosystems showed similar CAZyme family richness across substrates (e.g., cellulose, xylan, lignin). The primary difference was not the presence of specific gene families but their relative abundance. For instance, conifer forests had higher abundances of cellulose- and hemicellulose-associated CAZymes, while broadleaf forests showed higher lignin-associated abundances compared to grasslands.
3. Stage-Resolved Divergence:
   • Taxonomic Shifts: The taxonomic composition of CAZyme-encoding communities shifted more strongly across biochemical stages than across ecosystems. However, ecosystem context modulated these shifts (e.g., higher Acidobacteriota contribution in conifer forests).
   • Subfamily and Sequence Variation: Ecosystems differed in the allocation of CAZyme subfamilies and protein sequences. Conifer forests showed enrichment in mid-stage depolymerization families, while grasslands were enriched in terminal processing families.
   • Directional Selection: Evidence of directional selection was found in 42 CAZyme families, with ecosystem-specific residue preferences (e.g., broadleaf forests favoring polar residues in pectin-related families, conifer forests favoring aromatic/hydrophobic residues).
4. Predictive Power:
   • CAZyme profiles explained a unique, non-redundant fraction of variation in 24-month mass loss (6% for green tea, 1% for rooibos), adding to the explanatory power of environmental and spatial covariates.
   • Within-Ecosystem Prediction: Including CAZyme profiles significantly improved out-of-sample prediction (R²) for decomposition trajectories within the same ecosystem type, particularly for random forest models.
   • Cross-Ecosystem Generalization: Models failed to generalize well across different ecosystems or continents. Predictive performance often dropped to R² ≤ 0 when extrapolating, indicating that the relationship between CAZyme profiles and decomposition rates is highly context-dependent.

Significance and Claims
The paper claims to shift the paradigm of microbial trait inference from simple gene inventories to a mechanistic framework based on the stage-specific distribution of shared enzymatic functions.

• Mechanistic Insight: The study demonstrates that ecosystem differences in decomposition arise not from the presence or absence of specific CAZyme families, but from how a largely shared repertoire is quantitatively allocated across biochemical stages of decay.
• Modeling Implications: The authors argue that CAZyme profiles should be used as stage-resolved indicators of ecosystem-specific depolymerization capacity rather than as universal rate constants. They suggest that while genomic data improves within-ecosystem predictions, its utility for global carbon-cycle modeling is limited unless the strong context dependence of gene-rate relationships is explicitly acknowledged.
• Future Directions: The paper posits that improving model portability requires denser metagenomic sampling of underrepresented biomes (e.g., tropical forests, wetlands) to separate region-specific associations and link sequence adaptation more directly to functional consequences. It emphasizes the need to move beyond gene inventories to understand how conserved functions are allocated across substrates and stages to translate into measurable differences in litter decay.