Fungal community composition predicts forest carbon storage at a continental scale

The study addresses how soil microbiome diversity and composition relate to forest carbon storage at ecosystem scales. Forests contain a major portion of terrestrial biodiversity and act as large carbon sinks, yet the soil microbiome remains poorly understood in terms of its role in whole-forest carbon dynamics. Microbes regulate key carbon-cycle processes including soil respiration, decomposition, and plant growth via mutualisms and pathogens. While tree species composition is known to affect forest growth, albedo, and carbon sequestration, comparable understanding for soil microbial community composition is lacking. Prior studies suggest microbial composition can influence forest carbon pools and fluxes, for example through growth-promoting endophytes and mycorrhizal fungi, and through effects of bacterial and fungal communities on decomposition and soil organic carbon storage. However, direct links to total forest carbon storage, integrating above- and belowground pools, remain unresolved. The authors aim to explicitly examine how soil bacterial and fungal community composition and diversity are associated with tree growth, biomass carbon stocks, and soil organic carbon (SOC) across Europe, controlling for climate, forest type, and other environmental drivers.

Experimental and observational research indicates microbial diversity and composition can regulate ecosystem functions relevant to the carbon cycle: endophytic fungi can increase plant growth by 52–138% depending on taxa; rhizosphere and endophytic bacteria, and ecto- and arbuscular mycorrhizal fungi, can promote growth. Differences in ectomycorrhizal (ECM) fungal composition have been linked to a three-fold variation in tree growth across Europe. Bacterial composition and fungal richness have been associated with decomposition rates and SOC storage. Carbon models (e.g., RothC, CENTURY) typically predict positive links between plant productivity and SOC, although processes like priming, mineralogy, management, disturbance, SOC saturation, and mycorrhizal-driven nutrient mining can modulate this relationship, sometimes even leading to negative correlations between enhanced plant growth and SOC under elevated CO2. ECM fungi can both promote plant growth and, depending on context, slow decomposition (Gadgil effect) or mine nitrogen from organic matter, affecting SOC. Thus, context dependence necessitates simultaneous evaluation of above- and belowground carbon pools when relating microbiomes to total forest carbon storage.

Study design and sites: Soil and forest data were collected from 285 ICP Forests Level II plots across 18 European countries (ultimately 238 plots from 15 countries included due to DNA extraction/amplification failures). Plots (≥0.25 ha) are intensively monitored with tree censuses approximately every five years. Dominant tree type (broadleaf vs conifer, ≥50% cover), tree species membership (21 species total; richness 1–9, mostly 1–5), forest age (<30 to >120 years; mean ~90), and environmental gradients (MAT −2.5 to 15.5 °C; MAP 443–2082 mm yr−1; productivity 0.10–50.11 t C ha−1 yr−1) were represented. Soil sampling: In July–August 2019–2020, nine samples per plot were collected in a 30 × 30 m subplot on a grid. Organic horizons (where present) were removed, and mineral soils sampled to 10 cm depth (5 cm corer). Samples were pooled by horizon, homogenized, dried (40 °C oven or air-dried ≥48 h), shipped to ETH Zürich, and stored at −20 °C. Carbon and site measurements: Tree growth was computed from DBH increments between the first and last census (mean interval 5.5 years; mean initial year 2005; mean final year 2008), removing dead, shrinking, and <5 cm DBH trees. Species-specific allometries were applied; 50% C content assumed. Because not all trees were measured, trees were resampled with replacement to match in situ stem density (1,000 bootstraps) to estimate stand-level tree growth (t C ha−1 yr−1) and live biomass (t C ha−1). Soil C and N stocks (t C ha−1) used elemental C/N (%), bulk density, and depth. Soil pH measured in 1:2 DI water slurries, clay content from in situ measurements (and SoilGrids 250 m estimates where needed; r=0.51 with in situ, P<0.0001, n=321). Climate from WorldClim; N deposition from EMEP (2019, 1 km). Molecular analyses: DNA extracted from 250 mg frozen soil (DNeasy PowerSoil Pro). Prokaryotes targeted with 16S rRNA V4–V5 (515F/926R); fungi with full-length ITS (ITS9munngs/ITS4ng). PCRs run in duplicate; products pooled and size-selected with AMPure beads. 16S libraries sequenced on four Illumina MiSeq v3 (2×300 bp); ITS on four PacBio Sequel IIe SMRT Cell 8M (15 h movies). Bioinformatics: Demultiplexed with Cutadapt, ITS regions extracted with ITSx. 16S paired reads merged (vsearch), ITS HiFi reads single-end. Quality filtering in QIIME2; dereplication and de novo OTU clustering at 97% (16S) and 98% (ITS). Singletons removed in R. Taxonomy assigned with Greengenes (16S; 2019-05) and UNITE (ITS; v8, 2021-10) using a naïve Bayes classifier (confidence 0.7). Fungal functional guilds annotated at genus level with FUNGuild (probable or higher); metrics also computed for 'pure' ECM and 'pure' saprotrophs (single guild assignments). Statistical analyses: Low-depth samples filtered (<5,000 sequences for 16S; <500 for ITS). Datasets rarefied to lowest depth for alpha (richness, Shannon) and beta diversity (Bray–Curtis). Community composition summarized via PCoA; environmental fits with envfit; distance-based redundancy analyses assessed environmental correlations. Generalized additive models (mgcv, REML) predicted tree growth, tree biomass C, and SOC stocks using nitrogen deposition, soil N stocks (for growth), MAT, MAP, pH, clay, stem density, forest age, forest type (broadleaf vs conifer), and one microbiome predictor per model (PCoA1/2 or richness). Variance inflation factors ≤5. Pearson correlations (r) reported for display. Indicator species for continuous responses (tree growth, SOC) identified using DESeq2 on non-rarefied OTU tables (Wald test, parametric fit); significant OTUs required p-value significance and |log2FoldChange| > 0.6.

- Fungal, but not bacterial, community composition and richness are strongly associated with forest tree growth rates and biomass carbon stocks after controlling for climate, forest type, and other covariates. - Tree growth and tree biomass stocks are themselves positively correlated (r = 0.7, P < 0.001). - Fungal composition (PCoA1) correlates with tree growth (r ≈ 0.55) and biomass (r ≈ 0.52); fungal richness correlates with tree growth (r ≈ 0.41) and biomass (r ≈ 0.36). Bacterial composition and richness show no such correlations with growth or biomass. - Links between fungal composition and tree growth are stronger in conifer than broadleaf forests and stronger in mineral than organic horizons; ECM and endophytic guild compositions are most tightly linked to growth. - Endophyte richness shows the strongest positive association with tree growth among fungal guilds; effect size approximately one-third higher than other groups (saprotrophs, wood-decomposers, plant pathogens, ericoid), while ECM richness was not positively associated. - Indicator taxa linked to faster growth include endophytic Trichoderma (T. citrinoviride, T. koningii) and multiple Mortierella species; ECM indicators of fast growth tend to be Russula, whereas Cortinarius and Inocybe indicators are negatively associated with growth. - Environmental drivers: Fungal composition is correlated with dominant tree type, forest age, and tree growth; both bacterial and fungal compositions correlate with soil pH, clay content, and soil carbon stocks, with pH and carbon stock correlations stronger for bacteria. Overall, environmental variables explain ~22.8–28.2% of microbiome compositional variation. - SOC stocks (organic horizon) are more strongly linked to bacterial than fungal composition; both bacterial and fungal richness are negatively correlated with organic horizon SOC. Bacterial composition also shows a weak but significant correlation with mineral horizon SOC (P = 0.006), about one-sixth the effect size of the organic horizon. - Indicator lineages for organic horizon SOC: In conifers, many Proteobacteria OTUs are negatively associated with SOC; two ECM Russulaceae OTUs (Lactifluus vellereus, Russula rhodopus) are positively associated, consistent with potential Gadgil effects. In broadleaf forests, more indicators are detected; top positives are ECM (e.g., Inocybe, Sebacina, Russula), while top negatives are saprotrophs. - Tree biomass and growth are significantly and positively correlated with mineral horizon SOC stocks (0–10 cm), implying that fungal community attributes indirectly predict mineral SOC via their strong association with tree growth and biomass.

The findings directly address the hypothesis that soil microbiome composition is linked to forest carbon storage by showing strong, guild-specific fungal associations with tree growth and biomass across continental scales, independent of major environmental covariates. Fungal endophytes and ECM communities emerge as key biotrophic groups mediating aboveground productivity, with endophyte richness as a particularly strong positive correlate of growth. The mixed positive and negative ECM indicators (e.g., Russula vs Cortinarius/Inocybe) suggest a spectrum of symbiotic strategies with differing carbon costs to hosts and context-dependent effects influenced by pH, nitrogen deposition, drought, succession, and forest type. Belowground, bacterial community composition more strongly predicts organic horizon SOC than fungal composition, and microbial richness (both bacteria and fungi) is negatively associated with organic horizon SOC stocks. This pattern aligns with biodiversity–ecosystem function frameworks in which higher microbial diversity can increase decomposition rates, although species–energy and species–area considerations might predict the opposite; soil pH emerges as a potential confounder warranting experimental disentanglement. The weak bacterial signal in mineral horizon SOC indicates that, within this study system, the microbiome’s strongest pathway to predicting total forest carbon storage is indirect: fungal community features predict tree growth and biomass, which are themselves positively related to mineral SOC stocks. Thus, the soil mycobiome constitutes a sensitive biological indicator of overall forest carbon storage at continental scales.

Across 238 European forest plots, fungal community composition and diversity are robust predictors of tree growth and biomass carbon stocks, while bacterial community composition is more closely tied to organic horizon SOC. Because tree growth and biomass are positively associated with mineral horizon SOC, fungal community structure indirectly predicts total forest carbon storage. Endophytic fungi, in particular, show strong positive associations with tree growth, highlighting a relatively understudied yet influential component of forest microbiomes. These results demonstrate that soil mycobiome metrics can serve as bioindicators of forest carbon storage and suggest opportunities for leveraging endophytes and ECM communities in forestry and biostimulant development. Future research should employ manipulative experiments to establish causal pathways between fungal endophytes/ECM composition and tree growth; disentangle effects of soil pH and other edaphic variables from microbial richness–SOC relationships; refine taxonomic and functional resolution of biotrophic bacteria; and integrate temporal dynamics to assess stability and predictability of microbiome–carbon links under changing climate and deposition regimes.

The study is observational, preventing causal inference regarding microbiome effects on carbon pools. Functional assignments (e.g., endophytes) can encompass mixed trophic strategies, and the approach cannot resolve the active trophic state in situ. Biotrophic bacterial groups could not be confidently isolated using DNA-based community profiling. Temporal mismatches exist between tree growth census periods and soil sampling, although year-to-year microbiome variation is reportedly low. Sequencing depth differs between platforms (shallower for PacBio ITS), necessitating rarefaction and potential loss of rare taxa. Some environmental correlations with microbiome composition (e.g., with SOC) weaken after controlling for covariates, particularly in mineral horizons. Clay content was partly estimated from SoilGrids outside SOC modeling, introducing potential measurement uncertainty. A subset of plots was excluded due to DNA extraction/amplification issues, possibly affecting representativeness.