Topographic decoupling of tree-mortality disturbances drives contrasting soil carbon fates via divergent microbial strategies

Abstract

Climate change is exacerbating tree mortality worldwide, threatening the stability of forest soil organic carbon (SOC) stocks. However, how SOC responds to tree mortality amid co-occurring disturbances remains mechanistically unresolved. Here, we demonstrate that topography spatially decouples the canopy-opening and detrital-legacy effects of tree mortality and transforms these co-occurring disturbances into distinct selective pressures on soil microbial communities, thereby driving SOC toward contrasting fates through divergent microbial strategies for carbon and energy acquisition. On ridges, canopy-opening effects co-select for exo-enzyme and aerobic-respiration traits, alongside SOC loss associated with reduced heavy-fraction organic carbon. In valleys, detrital-legacy effects promote light-fraction-associated SOC gain while co-enriching endo-enzyme and anaerobic-respiration traits, potentially strengthening SOC retention. By integrating amplicon sequencing with genome-informed functional traits, our study provides genomic evidence for the microbial mechanisms underlying divergent SOC responses to tree mortality and highlights the potential of genome-based microbial traits for predicting ecosystem-scale biogeochemical processes.

Topography spatially decouples the canopy-opening and detrital-legacy effects of tree mortality, creating distinct pressures on soil microbes that drive soil carbon toward contrasting fates through divergent microbial strategies.

Introduction

Soil organic carbon (SOC) constitutes a crucial reservoir that underpins the persistence of forest carbon sinks^1,2, owing to its slow turnover and consequent long residence time in forest ecosystems³. Under ongoing global warming, climate-driven droughts, storms, and insect outbreaks are exacerbating tree mortality worldwide⁴, raising concern that this slow-cycling carbon reservoir may be destabilized and contribute to long-term carbon release^5,6. Yet the mechanisms governing SOC fate following tree mortality remain poorly resolved, given that multiple associated disturbances often co-occur and vary substantially in intensity and duration across diverse mortality scenarios⁷. Clarifying how these disturbances regulate SOC fate is therefore essential for improving predictions of forest-climate feedbacks and guiding mechanistically informed forest carbon management⁸.

Canopy opening and detrital legacies are two common mortality-induced ecological consequences^9,10. They often modify soil hydrothermal conditions and substrate supply, thereby altering SOC dynamics^11,12. Although their effects on soil environments vary widely in magnitude across mortality modes, stand structures, and ecosystem types, they tend to follow consistent patterns at broader scales^13–15. Canopy opening typically increases light exposure^16–18, soil temperature¹⁹, and evaporative water loss^20–22, whereas detrital legacies generally dampen soil hydrothermal fluctuations^23–25, increase organic polymer inputs^13,26, and promote anaerobic processes¹⁵. In montane forests, their relative dominance may be modulated by topography^27,28. Specifically, on convex ridges, greater radiation exposure²⁹, more active air exchange³⁰, and limited detritus retention^31,32 may allow canopy opening to dominate soil carbon dynamics. In concave valleys, by contrast, detritus tends to accumulate, while topographic shading²⁹ and cold-air pooling³³ may attenuate canopy-opening effects, making detrital legacies more likely to dominate soil biogeochemistry. These contrasts suggest that topography may act as a physical filter, spatially decoupling mortality-associated canopy-opening and detrital-legacy effects by modulating their relative dominance, and drive SOC dynamics towards divergent trajectories.

Current conceptual frameworks posit that SOC stabilization or destabilization is largely governed by its susceptibility to microbial decomposition^34,35, through which microorganisms acquire carbon and energy for growth and maintenance^36,37. Accordingly, microbial shifts in carbon- and energy-acquisition strategies may be closely coupled with SOC fate following tree mortality. For carbon acquisition, microorganisms generate assimilable monomers primarily through two complementary pathways³⁸: (i) terminal cleavage of polymer chains by exo-enzymes (exo-hydrolases), enabling the immediate release of monomers, which is more advantageous in environments rich in short-chain substrates as its efficiency depends largely on the number of polymer chains³⁹; and (ii) internal cleavage by endo-enzymes (endo-hydrolases or auxiliary oxidases), promoting generation of chain ends. Despite delayed monomer release, this strategy can substantially improve monomer release efficiency in environments dominated by long-chain substrates³⁹. Microbial energy acquisition is achieved mainly through respiration, which depends on the terminal electron acceptor available in their surroundings⁴⁰. Under well-aerated conditions, environmental selection favors microbial taxa with the capacity for aerobic respiration (using O₂)⁴¹, whereas oxygen limitation promotes taxa capable of anaerobic respiration using alternative terminal electron acceptors (e.g., NO₃⁻ and SO₄²⁻)⁴².

Given that most microorganisms possess constrained metabolic repertoires and enzyme complements, they tend to occupy distinct niches along the SOC decomposition cascade^35,43. Mortality-associated disturbances may shift the rate-limiting step within this cascade, thereby driving microbial community reorganization towards a newly adapted state⁴⁴. We thus hypothesize that mortality-associated shifts in microbial taxon abundance would be related to their functional potential for carbon and energy acquisition, and that such relationships may reflect microbial carbon-processing pathways underlying SOC fate. Moreover, the ongoing expansion of microbial genomic resources, particularly metagenome-assembled genomes (MAGs) and genome-resolved reference databases^45,46, has provided an important foundation for inferring taxon-specific functional potential⁴⁷. Although horizontal gene transfer, gene gain, and loss may obscure links between microbial phylogeny and function⁴⁸, many traits associated with carbon source preference⁴⁹, carbon use efficiency⁵⁰, and respiratory metabolism^51,52 have been shown to be phylogenetically clustered. Accordingly, we hypothesize that closely related taxa may retain comparable traits in these functions, and genomic information generated across diverse locations may therefore provide a useful reference framework for inferring taxon-specific functional potential for carbon and energy acquisition.

To elucidate the microbial mechanisms governing SOC fate following tree mortality, we conduct a field survey across two contrasting topographic settings within a 20-ha forest dynamics plot in Dinghushan, Guangdong, China. We pair large dead trees with adjacent living conspecific trees of comparable size, and sample soils from the corresponding canopy gaps and canopy-covered microsites. We separate SOC by density fractionation into two fractions differing in source and molecular chain characteristics, quantify the activities of carbon-degrading extracellular enzymes as indices of SOC decomposition potential, and use amplicon sequencing to profile shifts in microbial taxon abundance. To examine whether mortality-associated shifts in microbial taxon abundance correspond to their functional potential for carbon and energy acquisition, we assign genome-informed functional traits to taxa from target-gene counts and use interpretable machine-learning models to assess how these traits contribute to abundance shifts. Finally, we use shotgun metagenomic sequencing to evaluate whether inferred trait selection is consistent with changes in functional gene abundance.

Results

Contrasting SOC responses in two topographic settings

Linear mixed-effects models showed that the effect of tree mortality on SOC depended significantly on topographic setting (χ² = 13.67, P < 0.001; Supplementary Table 1). Tree mortality significantly reduced SOC on ridges (Fig. 1a), whereas it significantly increased SOC in valleys (Fig. 1b). Following tree mortality, light-fraction organic carbon (LFOC, mainly plant-derived long-chain polymers) increased in both ridges and valleys (Supplementary Fig. 1a). However, the increase was significant only in valleys, where it was 5.3 times that on ridges (P = 0.042; Supplementary Fig. 1a), consistent with stronger detrital-legacy effects in valleys than on ridges. By contrast, heavy-fraction organic carbon (HFOC, mainly microbially derived compounds) slightly decreased on ridges and slightly increased in valleys, with neither change reaching statistical significance (P > 0.05; Supplementary Fig. 1b).

Fig. 1. Tree mortality effects on soil organic carbon.

Paired comparisons of soil organic carbon (SOC) beneath live and dead trees in ridge (a) and valley (b) settings. Gray lines connect spatially paired samples (a dead tree and a nearby live conspecific). Error bars represent mean ± standard deviation (n = 18). Statistical significance was assessed using Tukey-adjusted pairwise comparisons of estimated marginal means from a linear mixed-effects model (n = 72 observations; lme4), with tree status (live, dead), topography (ridge, valley), and their interaction as fixed effects, and pair identity as a random intercept. Model-estimated marginal relationships of mortality-associated changes in SOC (ΔSOC = dead − live) with changes in density-separated carbon fractions (ΔLFOC and ΔHFOC) in ridge (c) and valley (d) settings. For each topographic setting, ordinary least squares (OLS) models were fit with ΔSOC as the response and ΔLFOC, ΔHFOC, and their interaction as predictors (n = 18). Lines depict the estimated marginal relationship for each density fraction, with the other fraction held at its mean value (solid when two-sided P < 0.05; dashed otherwise); shaded ribbons denote 95% confidence intervals from the OLS fits. Reported statistics include the conditional slope (β) and its two-sided P value estimated from the corresponding OLS model. LFOC, light-fraction organic carbon; HFOC, heavy-fraction organic carbon. Source data are provided as a Source Data file.

Marginal relationships estimated from ordinary least-squares regression models further showed that SOC loss on ridges was significantly associated with HFOC decline (Fig. 1c), whereas SOC increase in valleys was significantly associated with LFOC accumulation (Fig. 1d). Soil moisture responses to tree mortality also differed significantly between ridges and valleys (χ² = 25.96, P < 0.001; Supplementary Table 1), with a significant decline on ridges (P < 0.001; Supplementary Fig. 1c) and no significant change in valleys (P > 0.05; Supplementary Fig. 1c). These patterns were consistent with a stronger canopy-opening effect on ridges than in valleys.

Effect of tree mortality on carbon-degrading extracellular enzyme activities

We measured the potential activities of six soil extracellular enzymes involved in SOC degradation: four hydrolases responsible for hydrolyzing glycosidic bonds at the chain ends of polymers or oligomers, and two oxidases that catalyze the oxidative transformation of phenolic and lignin-related aromatic compounds (Fig. 2). We then expressed enzyme activities as mass-normalized activities (per gram dry soil) and SOC-normalized activities (per unit SOC). Principal component analysis (PCA) showed that the first two principal components (PC1 and PC2) explained 65.6% of the variance in mass-normalized activities and 74.6% of the variance in SOC-normalized activities (Fig. 2a, b). For both normalization approaches, all six enzymes loaded positively on PC1, whereas PC2 primarily reflected variation in oxidase activities (Supplementary Table 2 and Fig. 2a, b).

Fig. 2. Tree-mortality effects on carbon-degrading extracellular enzyme activities.

Principal component analysis (PCA) of six extracellular enzyme activities after soil mass normalization (a) and soil organic carbon (SOC) normalization (b). Each point represents one soil sample. Colors indicate topographic setting and tree status. Arrows represent loadings of individual enzymes on the first two principal components (PC1 and PC2) and were uniformly scaled twofold for visualization. Paired comparisons of sample scores along PC1 and PC2 from the mass-normalized PCA (c) and SOC-normalized PCA (d). Gray lines connect spatially paired samples (a dead tree and a nearby live conspecific). Error bars represent mean ± standard deviation (n = 18). Statistical significance was assessed using Tukey-adjusted pairwise comparisons of estimated marginal means from linear mixed-effects models (n = 72 observations; lme4), with tree status (live, dead), topography (ridge, valley), and their interaction as fixed effects, and pair identity as a random intercept. AG, α-glucosidase; BG, β-glucosidase; BX, β-xylosidase; NAG, N-acetyl-β-glucosaminidase; PPO, polyphenol oxidase; PER, peroxidase. Source data are provided as a Source Data file.

Tree mortality had topography-dependent effects on PC1 scores for both mass-normalized (χ² = 7.74, P = 0.005) and SOC-normalized activities (χ² = 4.53, P = 0.033; Supplementary Table 3). On ridges, tree mortality significantly decreased PC1 scores for mass-normalized activities (Fig. 2c), but did not significantly affect those for SOC-normalized activities (Fig. 2d). In valleys, the opposite pattern occurred, with no significant change in mass-normalized PC1 scores (Fig. 2c) but a significant decrease in SOC-normalized PC1 scores (Fig. 2d). PC2 scores showed no significant response to tree mortality in either topographic setting (Fig. 2c, d, and Supplementary Table 3). These patterns indicated that tree mortality had different effects on the dominant enzyme-activity axis in the two topographical settings. On ridges, carbon-degrading enzyme activities tended to decline with SOC loss, while showing little change per unit SOC. In valleys, SOC accumulation outpaced the enzyme-activity response, tending to reduce carbon-degrading enzyme activities per unit SOC.

Mortality-associated shifts in microbial taxon abundance

We characterized soil microbial community composition by amplicon sequencing of the bacterial 16S ribosomal RNA (16S rRNA) gene and the fungal internal transcribed spacer (ITS) region. Bacterial communities were dominated by Proteobacteria, Acidobacteria, Planctomycetes, and Actinobacteria (Supplementary Fig. 2a), whereas fungal communities were dominated by Basidiomycota and Ascomycota (Supplementary Fig. 2b). Permutational multivariate analysis of variance (PERMANOVA) showed that mortality-associated compositional shifts differed significantly between topographic settings for bacterial communities (F = 2.31, P = 0.019) and fungal communities (F = 2.14, P = 0.004; Supplementary Table 4). Standardized major axis (SMA) regression further showed that, for bacteria and fungi, taxon-specific abundance shifts were both inversely related between ridges and valleys (Fig. 3a, b). These patterns indicate that mortality-associated disturbances acted as opposing selective pressures in the two topographic settings, thereby driving bacterial and fungal composition shifts in divergent directions.

Fig. 3. Relationships between mortality-associated shifts in microbial taxon abundance on ridges versus in valleys.

Scatterplots depict mortality-associated log₂ fold changes (log₂FC) in microbial amplicon sequence variant (ASV) abundance in valleys versus those on ridges for bacteria (a) and fungi (b). Each point represents an ASV that meets the filters (mean relative abundance ≥0.01% across samples and detected in ≥3 samples). Lines show standardized major axis (SMA) fits with 95% confidence bands (n = 417 ASVs for bacteria; n = 415 ASVs for fungi). Reported statistics are the SMA slope (β), squared Pearson correlation coefficient (R²), and the two-sided P value for the correlation.

Genome-informed functional traits for carbon and energy acquisition

To test whether mortality-associated shifts in microbial taxon abundance were related to functional potential for carbon and energy acquisition, we used the Genome Taxonomy Database (GTDB) and the National Center for Biotechnology Information (NCBI) databases to retrieve high-quality genomes that matched bacterial species and fungal genera inferred from amplicon sequencing (Supplementary Note 1). In total, 5783 high-quality bacterial reference genomes and 217 fungal nuclear genomes were obtained (Supplementary Table 5). Phylogenetic placement showed that these genomes covered the major lineages represented in the bacterial and fungal communities of the study area (Fig. 4a, b). The relative abundance of amplicon taxa with matched genomes decreased with increasing taxonomic resolution. For bacteria, genome-matched representation declined from 94.6% ± 1.5% at the phylum level to 35.3% ± 4.8% at the species level, with 187 amplicon taxa retaining species-level matches to reference genomes (Supplementary Table 5). For fungi, this decline was more pronounced, falling from 89.6% ± 5.3% at the phylum level to 9.9% ± 6.5% at the species level, with only nine amplicon taxa matched to reference genomes at the species level (Supplementary Table 5). Matching at the genus level increased fungal representation to 34 amplicon taxa, which accounted for 26.1% ± 10.2% of fungal community relative abundance (Supplementary Table 5).

Fig. 4. Phylogenetic positions and functional trait profiles of amplicon taxa with matched reference genomes.

Phylogenetic trees show the taxonomic placement of bacterial species (a) and fungal genera (b) represented among amplicon taxa with matched reference genomes. Circle size and branch width indicate the number of bacterial species or fungal genera within each lineage. c Histogram of the number of target functions inferred from Kyoto Encyclopedia of Genes and Genomes Orthology (KO) annotations per bacterial or fungal species. d Kernel density estimates show pairwise Jaccard dissimilarities calculated from presence-absence profiles of KO-inferred target functions.

We assigned Kyoto Encyclopedia of Genes and Genomes Orthology (KO) annotations to the protein sequences of each genome, and then curated target KO sets representing 12 target functions associated with carbon acquisition (exo-enzymes, endo-enzymes, and lignin-targeting oxidases) and energy acquisition (genes involved in the use of terminal electron acceptors, including O₂, NO₃⁻, and SO₄²⁻). For each genome, the counts of protein sequences assigned to KOs within each target function (hereafter, target gene counts) were used to quantify genome-informed functional traits associated with carbon and energy acquisition (Supplementary Note 1). The presence-absence profiles of these target functions at the species level showed that most bacterial species encoded only a subset of the target functions and rarely carried target genes associated with lignin degradation (Supplementary Fig. 3a and Fig. 4c). In comparison, most fungal species concurrently encoded enzymes targeting cellulose, hemicellulose, chitin and lignin, whereas their nuclear genomes lacked the target genes for terminal electron acceptor use examined in this study (Supplementary Fig. 3b and Fig. 4c). Pairwise interspecific Jaccard dissimilarities calculated from target-function presence-absence profiles indicated greater heterogeneity in carbon- and energy-acquisition functions among bacterial species, whereas fungal species showed more similar functional combinations (Fig. 4d).

For bacterial groups that exhibited high interspecific functional heterogeneity, we calculated the clade depth (τ_D) of each functional trait (excluding traits related to lignin degradation) from a curated dataset of 8385 high-quality genomes (representing 274 species) (Supplementary Notes 2 and 3). Although τ_D values were generally low, these traits (except sulfate respiration) showed significantly greater τ_D than expected by chance (Supplementary Fig. 4), indicating that they were predominantly shared within shallow clades and remained constrained by vertical inheritance. Furthermore, we applied nested random-intercept mixed models to these bacterial genomes and 217 fungal nuclear genomes to estimate the proportion of variance in target gene counts that remained within taxa at successive bacterial and fungal taxonomic ranks (hereafter, within-taxon variance proportion, WVP; Supplementary Note 4). The results showed that WVP values generally declined towards finer taxonomic ranks for all traits, reaching their lowest levels at the bacterial species level (median WVP range: 4.19–9.90%) and the fungal genus level (4.66–29.20%), where they were significantly lower than expected by chance (P < 0.05; Supplementary Figs. 5 and 6). Accordingly, based on reference genome coverage and within-taxon stability of functional traits, we selected the bacterial species level and the fungal genus level as the taxonomic resolutions for subsequent functional trait analyses.

For each amplicon-derived taxon with available reference genomes, we calculated mean target gene counts at the bacterial species level and the fungal genus level, and used them as predictors in extreme gradient boosting (XGBoost) regression models of mortality-associated shifts in microbial taxon abundance. Model evaluation on the held-out test set showed that predictions based on carbon- and energy-acquisition traits were significantly associated with observed shifts in bacterial taxon abundance, whereas carbon-acquisition functions encoded in fungal nuclear genomes were insufficient as standalone predictors of shifts in fungal taxon abundance (Supplementary Fig. 7).

Divergent microbial strategies in two topographic settings

For bacterial XGBoost models with significant predictive ability, we used Shapley additive explanations (SHAP) to quantify the contribution of each functional trait to predicted shifts in bacterial taxon abundance (Supplementary Fig. 8). We then generated SHAP dependence plots for the two highest-ranking traits by importance within each of the three functional categories (exo-enzymes, endo-enzymes, and terminal oxidoreductases; Fig. 5a). On ridges, SHAP values (contributions to predicted log₂ fold changes in abundance, dead versus live trees) increased with exo-enzymes and aerobic respiration gene counts, whereas they decreased with endo-enzymes and anaerobic respiration gene counts (Fig. 5a). In valleys, these relationships were reversed, with SHAP values decreasing with exo-enzymes and aerobic respiration gene counts and increasing with endo-enzymes and anaerobic respiration gene counts (Fig. 5a).

Fig. 5. Topography-dependent microbial functional selection for carbon and energy acquisition following tree mortality.

a Shapley additive explanations (SHAP) dependence plots showing how bacterial functional traits, represented by target gene counts, contributed to extreme gradient boosting predictions of mortality-associated shifts in bacterial species abundance. For each topographic model, we show the two highest-ranking traits by importance within each of the three functional categories, ranked by mean absolute SHAP values. Each point represents one bacterial species. Solid curves indicate smoothed trends, shaded areas indicate 95% confidence intervals, and dashed horizontal lines mark zero contribution. b Mortality-associated changes in metagenomic abundances of target functional genes associated with carbon- and energy-acquisition in ridge and valley soils. Heatmaps show log₂ fold changes (log₂FC) in gene abundance beneath dead trees relative to living trees. Statistical significance was tested using negative binomial generalized linear models followed by likelihood-ratio tests, with Benjamini–Hochberg correction to control the false discovery rate (FDR). Asterisks indicate significant differences after correction at FDR < 0.05. TEA denotes terminal electron acceptor.

We further performed shotgun metagenomic sequencing to profile target genes spanning bacteria, fungi, and other microbial lineages. To maximize consistency with the genome-informed functional traits, we constructed a non-redundant metagenome-derived gene catalogue, annotated its encoded protein products using the same KO annotation framework, and quantified mortality-associated shifts in target gene abundance (Supplementary Note 5). The results are consistent with the trait co-selection patterns inferred from functional traits. Tree mortality on ridges increased exo-enzymes and aerobic respiration gene abundances, but enriched endo-enzymes and anaerobic respiration gene abundances in valleys (Fig. 5b). Together, these results indicate that, following tree mortality, microbial communities adopted distinct strategies for carbon and energy acquisition in the two topographic settings. On ridges, tree mortality co-selected for a functional combination involving carbon acquisition through exo-enzymes and energy acquisition through aerobic respiration. In valleys, it favored a contrasting functional combination involving carbon acquisition through endo-enzymes and energy acquisition through anaerobic respiration.

Discussion

This study revealed a topographic decoupling mechanism in which two co-occurring ecological consequences of tree mortality, canopy opening and detrital legacies, are spatially separated by topography and converted into opposing ecological selection pressures, driving SOC towards contrasting fates through distinct microbial strategies for carbon and energy acquisition (Fig. 6). On ridges, where canopy-opening effects prevail, tree mortality favors acquisition-oriented microbial strategies, reflected in the concurrent promotion of exo-enzyme traits for carbon acquisition and aerobic-respiration traits for energy acquisition, alongside SOC loss that was associated with HFOC reduction. In valleys, where detrital-legacy effects tend to predominate, tree mortality drives LFOC-associated SOC accumulation and shifts microbial communities towards conservation-oriented strategies, reflected in the concurrent promotion of endo-enzyme traits for carbon acquisition and anaerobic-respiration traits for energy acquisition. Under such conditions, SOC retention may be enhanced.

Fig. 6. Conceptual schematic of the topographic decoupling mechanism underlying the effect of tree mortality on SOC fate.

This schematic illustrates a topographic decoupling mechanism, in which topography modulates the relative dominance of two mortality-associated ecological consequences, converting them into opposing ecological selection pressures on microbial communities. Consequently, microbial communities shift toward two divergent carbon- and energy-acquisition strategies and drive soil organic carbon (SOC) toward contrasting fates through distinct microbial carbon processes. On ridges, where greater exposure coincides with weaker detrital retention, canopy-opening effects are more likely to prevail. Mortality-associated soil warming and moisture loss favor acquisition-oriented microbial communities, characterized by the concurrent promotion of carbon acquisition through exo-enzymes and energy acquisition through aerobic respiration. This would drive SOC loss and may link to the disproportionate depletion of short-chain substrates. In valleys, where topographic shading coincides with stronger litter retention, detrital-legacy effects are more likely to dominate. The accumulation of long-chain polymers and the expansion of anaerobic microsites favor conservation-oriented microbial communities, reflected in the concurrent promotion of carbon acquisition through endo-enzymes and energy acquisition through anaerobic respiration. In this context, SOC retention would be enhanced.

Mortality-associated canopy opening generally increases soil temperature¹⁹ and promotes evaporative water loss^20–22. It is widely recognized that soil warming can accelerate SOC turnover by enhancing microbial metabolic activity⁵³, whereas evaporation-induced soil moisture loss may act as a selective filter regulating microbial strategies for carbon and energy acquisition^54–56. Soil moisture loss is generally accompanied by increased aeration⁵⁷, which would promote the enrichment of aerobic microbial taxa⁵⁸, and by disruption of water-film continuity, which would constrain substrate diffusion and enzyme-substrate encounters^54,55. Under these conditions, exo-enzyme strategies may confer an advantage by directly releasing assimilable monomers³⁹. Short-chain substrates, owing to their greater solubility and mobility⁵⁹, are more likely to be encountered by microorganisms, which would facilitate their disproportionate consumption. Consistent with this interpretation, we observed that, on ridges, mortality-induced canopy opening co-selected for exo-enzymes and aerobic respiration traits (Fig. 5a, b), and that SOC loss was associated with HFOC reduction (Fig. 1c). These results suggest that microbial communities shift toward an acquisition-oriented strategy for carbon and energy acquisition, and resulting SOC loss may be driven by disproportionate depletion of short-chain substrates.

In valleys, SOC was observed to increase (Fig. 1b), and the magnitude of this increase was strongly associated with LFOC accumulation (Fig. 1d). This pattern can be explained by the fact that mortality-associated detrital inputs directly promote SOC accumulation by supplying additional carbon sources. Detrital inputs have been widely shown to increase the proportion of long-chain polymers²⁶, and would increase microbial demand for polymer depolymerization³⁶. At the same time, detrital legacies can enhance soil water retention²³ and oxygen depletion, together promoting the expansion of anaerobic microsites⁵⁷. Under these conditions, even if exo-enzyme strategies can immediately release monomers, their ecological advantage would be reduced as these monomers cannot be rapidly oxidized to generate energy^39,40. Therefore, tree mortality in valleys was observed to co-select for functional traits associated with endo-enzymes and anaerobic respiration (Fig. 5), accompanied by a downward tendency in carbon-degrading enzyme activities per unit SOC (Fig. 2d). These shifts indicate that mortality-associated detrital legacies promote SOC accumulation, while driving microbial communities toward a more conservative strategy for carbon and energy acquisition. Under such conditions, SOC retention may be enhanced.

Notably, although bacterial and fungal taxon abundances were both observed to shift in opposite directions between ridges and valleys following tree mortality (Fig. 3), only bacterial taxon shifts were predicted by genome-informed functional traits (Supplementary Fig. 7). Fungal nuclear genomes lacked the target genes for terminal electron acceptor use examined in this study, and carbon acquisition traits alone are insufficient to predict mortality-associated shifts in fungal taxon abundance. This discrepancy may be partly attributable to greater similarity in carbon acquisition functional combinations among fungal taxa (Fig. 4c, d, and Supplementary Fig. 3), which would weaken the selective effects of substrate changes on fungal communities. Moreover, fungal responses to tree mortality may be regulated by additional traits, including hyphal expansion and spatial foraging ability, as well as tolerance to environmental stresses (e.g., drought or acidification)⁶⁰. Nevertheless, metagenome-inferred shifts in abundance of target functional genes spanning bacterial, fungal, and other microbial lineages were consistent with the trait co-selection pattern inferred from bacterial reference genomes (Fig. 5a, b). These results further indicate that microbial communities in the two topographic settings adopted divergent strategies for carbon and energy acquisition following tree mortality.

Canopy opening and detrital legacies are two prevalent ecological consequences derived from tree mortality^9,10, and their effects on the soil environment follow relatively consistent patterns across scenarios^13–15. Accordingly, tree mortality across diverse climatic, vegetational, edaphic, and topographic settings can be conceptualized as the composite effects of canopy opening and detrital legacies, with variation in their intensity, duration, and spatial distribution, which regulates microbial strategies for carbon and energy acquisition and drives varied SOC fates. This framework provides a transferable process-based perspective on divergent SOC responses under intensified climate-driven tree mortality. Additionally, this study leveraged genomic resources compiled from public databases to infer taxon-specific functional traits, which remain relatively stable at fine taxonomic levels (Supplementary Figs. 4–6) and show discernible predictive power for shifts in the abundances of bacterial taxa (Supplementary Fig. 7). These findings indicate that large-scale genomic resources across diverse sources can provide a useful reference framework for linking changes in microbial community composition to ecosystem-scale biogeochemical processes.

Although this study identified a topographic decoupling mechanism linking tree mortality to SOC dynamics, the current field evidence was derived from two contrasting topographic settings within a 20-ha humid subtropical forest plot. The magnitude and duration of canopy-opening and detrital-legacy effects, their interaction, and microbial responses to the resulting environmental changes may be modified by vegetation context (e.g., community composition and stand density), mortality mode (e.g., insect-induced mortality or windthrow), climate regime, and initial soil conditions (e.g., SOC content, soil moisture, and texture). Cross-site validation is therefore needed to assess the broader generality of these mechanisms. Root-derived carbon represents another pathway regulating SOC dynamics⁶¹. Although its contribution to the 0–30 cm soil layer may be limited here, deeper SOC pools might be more sensitive to cessation of root-derived carbon inputs. Genome-matched taxa accounted for only about one-third of bacterial species-level and one-quarter of fungal genus-level abundances in this study, warranting caution when inferring abundance shifts in rare or poorly characterized taxa. Although metagenomic gene-abundance analysis supported the trait co-selection patterns inferred from reference genomes, the potential associations among these traits remain model-based predictions. Recovering MAGs from local metagenomes would improve the representation of site-specific microbial lineages and enable more direct assessment of both trait-trait associations and the phylogenetic conservation of these traits. The applicability of functional traits inferred from target gene counts depends on within-taxon stability, which should therefore be assessed before these genome-informed traits are applied. Finally, this field study provides correlational support for the proposed microbial pathways, and the causal contribution of trait shifts to SOC change remains inferred from these patterns and has not been directly tested. Microcosm manipulations of litter addition, warming, and drought, coupled with isotopic tracing, could test whether inferred shifts in carbon and energy acquisition contribute to changes in SOC loss and retention.

Our findings show that mortality-associated canopy opening and detrital legacies exert contrasting effects on microbial communities and SOC fates, which also provide a mechanistic basis for forest management aimed at maintaining persistent carbon sinks under increasingly climate-driven tree mortality. Accelerating canopy closure through enrichment planting or assisted natural regeneration may buffer microclimatic fluctuations that favor aerobic, fast-cycling microbial activity and associated SOC losses. Retaining woody detritus from dead trees can sustain carbon inputs, increase the supply of complex compounds, and promote anaerobic conditions, which may further promote SOC retention.

Methods

Study site

This study was conducted in a 20‑ha forest dynamics plot at the Dinghushan Biosphere Reserve, Guangdong, China (23°10′N, 112°30′E). The region has a humid subtropical monsoon climate (mean annual temperature: 20.9 °C; mean annual precipitation: 1985 mm). The plot (400 m × 500 m; elevation 230–476 m) was divided into 500 quadrats (20 m × 20 m). For each quadrat, we surveyed topographic variables, including elevation, slope, aspect, concavity, and convexity, which were further classified as ridge, slope, or valley using multivariate regression trees⁶². Soils are lateritic red soils from shale and sandstone (typical depth 40–80 cm).

Sampling design and sample collection

In July 2022, we sampled six canopy species: Castanopsis chinensis, Schima superba, Engelhardia roxburghiana, Machilus chinensis, Syzygium acuminatissimum, and Pinus massoniana, at two contrasting topographic settings (ridge, valley). For each species-topography combination, we selected three large dead trees that had died between 2015 and 2020, each with a diameter at breast height (DBH) of at least 30 cm. Their mortality created canopy gaps, which remained unrecovered during the sampling period. Concurrently, for each dead tree, we selected an adjacent living tree of the same species and comparable DBH as a control (Supplementary Fig. 9). From the microsite beneath each tree’s canopy cover (for living trees) or within its canopy gap (for dead trees), we collected mineral soil samples from depths 0 to 30 cm. Five cores were composited into a single homogenized sample for each individual. The samples were then sealed with dry ice and transported to the laboratory for further analysis.

Soil properties and density fractionation

Gravimetric water content was determined from mass loss after drying at 105 °C for 24 h. SOC was measured by the potassium dichromate-sulfuric acid wet oxidation followed by ferrous sulfate titration. LFOC and HFOC were obtained from density fractions separated with sodium polytungstate (SPT; density, 1.7 g mL⁻¹). Briefly, 15 g of air-dried soil sieved through a 2-mm mesh was suspended in 75 mL of SPT solution (1:5, w/v), shaken for 60 min, and centrifuged at approximately 1500 × g for 10 min. The supernatant was filtered through a 0.45-µm filter to recover the light fraction, and the remaining pellet was retained as the heavy fraction. Both fractions were washed with 0.01 M CaCl₂ and treated dropwise with 0.5 M HCl as a precautionary step to remove potential carbonate carbon. The fractions were subsequently dried at 60 °C to constant mass and ground to a fine powder (<0.15 mm). Carbon concentrations of the light and heavy fractions were measured using an elemental analyzer (Vario MAX cube, Elemental, Germany) and used to calculate LFOC and HFOC, respectively.

Extracellular enzyme activities

We assayed the potential activities of six carbon-degrading extracellular enzymes, including four hydrolases and two oxidative enzymes. Fresh soil (1 g, maintained at its original moisture content) was suspended in 125 mL of 50 mM sodium acetate buffer (pH 5.0) and stirred magnetically for 1 min to obtain a soil suspension. The resulting suspension was used for all subsequent extracellular enzyme assays.

Potential activities of α-glucosidase (AG), β-glucosidase (BG), β-xylosidase (BX), and N-acetyl-β-glucosaminidase (NAG) were measured fluorometrically in black flat-bottom 96-well microplates using 4-methylumbelliferone (MUB)-conjugated substrates. Each reaction contained 200 μL of soil suspension and 50 μL of 200 μM enzyme substrate solution (Supplementary Table 6). Reactions were incubated at 25°C for 3 h in the dark, after which 10 μL of 1 M NaOH was added to each well prior to fluorescence measuremen. Fluorescence was recorded at excitation and emission wavelengths of 365 and 450 nm, respectively. Fluorescence signals were corrected using substrate controls and sample blanks and quantified using matrix-matched MUB calibration standards. Activities were expressed as nmol MUB released g⁻¹ dry soil h⁻¹.

Polyphenol oxidase (PPO) and peroxidase (PER) activities were measured colorimetrically in clear 96-well plates using L-3,4-dihydroxyphenylalanine (L-DOPA) as the substrate (Supplementary Table 6). Reaction mixtures contained 200 µL of soil suspension and 50 µL of 25 mM L-DOPA; PER reactions additionally received 10 µL of 0.3% H₂O₂. Plates were incubated at 20 °C for 18 h in the dark, and absorbance was measured at 450 nm to quantify L-DOPA oxidation. Activities were expressed as μmol L-DOPA oxidized g⁻¹ dry soil h⁻¹.

Enzyme activities expressed per gram of dry soil were used as mass-normalized activities, whereas SOC-normalized activities were obtained by further normalizing these values to SOC content and were expressed per gram of SOC.

DNA extraction, amplicon sequencing, and analysis

Total DNA was extracted from fresh soil using the DNeasy PowerSoil Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions and further purified with AMPure XP paramagnetic beads (Beckman Coulter, Brea, CA, USA) at a 0.8× bead-to-sample ratio. DNA concentration and purity were assessed using a Qubit 4.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA) and a NanoDrop One spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, USA), respectively. DNA integrity was checked by 1% (w/v) agarose gel electrophoresis. Amplicons of the full-length bacterial 16S rRNA gene and the fungal ITS region were generated by polymerase chain reaction (PCR) using primer pairs 27F/1492R^63,64 and ITS1F/ITS4R^65,66, respectively [27F, 5′-AGAGTTTGATCMTGGCTCAG-3′; 1492R, 5′-TACGGYTACCTTGTTACGACTT-3′; ITS1F, 5′-CTTGGTCATTTAGAGGAAGTAA-3′; ITS4R, 5′-TCCTCCGCTTATTGATATGC-3′]. After verification on 2% (w/v) agarose gels, PCR products were purified with AMPure XP beads and quantified using a Qubit 4.0 Fluorometer. Sequencing libraries were prepared with the Ligation Sequencing Kit v14 (Oxford Nanopore Technologies, Oxford, UK) according to the manufacturer’s instructions. The purified library was loaded onto R10.4.1 flow cells and sequenced on a PromethION sequencer (Oxford Nanopore Technologies, Oxford, UK).

Oxford Nanopore reads were basecalled and demultiplexed using Dorado v7.1.4 on the sequencing platform. For 16S rRNA gene amplicons, reads were filtered with NanoFilt v2.5.0 to retain sequences with a mean Q-score ≥10 and lengths of 1200–1800 bp. Primers were trimmed using cutadapt v4.6 with a maximum error rate of 0.08 and a minimum overlap of 12 bp, and reads containing ambiguous bases were discarded. For ITS amplicons, reads were retained only if they had a mean Q-score ≥10 and a length of 300–1000 bp. Primers were trimmed using cutadapt v4.6 with a maximum error rate of 0.10 and a minimum overlap of 18 bp. Cleaned reads were then aligned against the SILVA 138.1 for 16S rRNA genes and the UNITE v10.0 for ITS sequences using LAST. Taxonomy was assigned based on the best hit with ≥90% identity; otherwise, the lowest common ancestor among the top matches was used.

Genome-informed functional profiling of amplicon-detected taxa

We used the NCBI and the GTDB as reference resources to retrieve reference genomes corresponding to the detected bacterial species and fungal genera. Predicted protein sequences from these genomes were annotated using KofamScan v1.3.0 and assigned to target functions based on their KO annotations. For each genome, we counted the protein sequences assigned to each target function, and the resulting counts were used as genome-informed estimates of functional traits (Supplementary Note 1). We then evaluated clade depth (τ_D) for bacterial functional traits and quantified the within-taxon variance proportion (WVP) of target gene counts across bacterial and fungal taxonomic levels (Supplementary Notes 2–4).

Metagenome sequencing and analysis

Detailed procedures for library preparation, sequencing, and bioinformatic processing are provided in Supplementary Note 5. Briefly, quality-controlled reads from all samples were co-assembled, and open reading frames were predicted from the resulting contigs. Predicted open reading frames were clustered to generate a non-redundant gene catalogue. Representative sequences from this catalogue were then functionally annotated and assigned to KOs. Reads from each sample were mapped back to the non-redundant gene catalogue to estimate sample-specific gene abundances.

Statistical analysis

All tests were two‑tailed unless stated. Biological replicates were tree‑level composite soil samples (n = 72). Exact sample size, error‑bar definitions, and test statistics are reported in text and figure captions. Analyses were run in R v4.5.1.

Mixed‑effects models

We used linear mixed-effects models to assess the effects of tree mortality across topographic settings on SOC, LFOC, HFOC, soil moisture, and principal component scores (PC1 and PC2) derived from mass-normalized and SOC-normalized extracellular enzyme activities (lme4 R package). Models included tree status (live, dead), topographic setting (ridge, valley), and their interaction as fixed effects, with a random intercept for each sampling pair. Each pair consisted of a dead tree and a nearby live tree of the same species and comparable size. Residual normality and homoscedasticity were evaluated, and when model assumptions were not met, statistical testing was based on a CR2 cluster-robust variance-covariance estimator clustered on Pairs.

Ordinary least-squares models

Within each topographic position (ridge, valley), we fit ordinary least-squares models with ΔSOC (dead minus live) as the response and ΔLFOC, ΔHFOC, and their interaction as predictors to identify carbon-fraction drivers of mortality-associated shifts in SOC.

Principal component analysis

PCA was performed separately for mass-normalized and SOC-normalized extracellular enzyme activities (psych R package). Each analysis included six extracellular enzymes (AG, BG, BX, NAG, PER, and PPO) and was based on the correlation matrix without rotation. Sample scores for PC1 and PC2 were retained for downstream statistical analyses.

Permutational multivariate analysis of variance

For bacterial and fungal communities, we retained amplicon sequence variants (ASVs) detected from 16S rRNA gene profiles for bacteria and ITS profiles for fungi that had a relative abundance of ≥0.01% and occurred in at least three samples. We calculated Bray–Curtis dissimilarities from ASV-level abundance matrices and performed PERMANOVA separately for bacteria and fungi with 9999 permutations (vegan R package). Each model included tree status (live, dead), topographic setting (ridge, valley), and their interaction as fixed effects.

Standardized major axis analysis

For bacteria and fungi, mortality-associated ASV abundance shifts were calculated separately for ridge and valley samples and expressed as the log₂ fold change in abundance (dead versus live). Standardized major axis regressions were then fitted separately for bacteria and fungi, with valley log₂ fold change as the response variable and ridge log₂ fold change as the explanatory variable (smatr R package).

Machine learning model and Shapley additive explanations

We trained XGBoost regressors to model mortality-associated shifts in bacterial and fungal taxon abundance, using mortality-associated log₂ fold change in abundance as the response variable and the retained taxon-specific target gene counts from genome-informed functional profiling as predictors. Data were randomly split into training and held-out test sets at an 80:20 ratio. Hyperparameters were tuned using a Cartesian grid of 2916 configurations covering the learning rate, maximum tree depth, minimum child weight, minimum loss reduction, subsampling rate, column subsampling rate, and L1 and L2 regularization strengths. Each configuration was evaluated by 10-fold cross-validation with early stopping, using a patience of 50 rounds and a maximum of 500 boosting rounds. The final hyperparameter set was selected based on the lowest cross-validated root mean squared error (RMSE), and the corresponding best boosting iteration was used to train the final model. Model performance was evaluated on the held-out test set using RMSE, mean absolute error, and the squared Pearson correlation coefficient (R²).

For bacterial models with predictive skill, we used SHAP to quantify the contribution of each functional trait to model predictions. Global feature importance was summarized as the mean absolute SHAP value. Based on this ranking, we selected the two most important traits within each of the three functional trait groups (exo-enzymes, endo-enzymes, and terminal oxidoreductases associated with terminal electron acceptor use), and used the selected traits to generate SHAP dependence plots.

Negative binomial generalized linear models

Metagenome-derived counts for functional genes associated with microbial carbon and energy acquisition were analyzed using the edgeR R package. Models were fitted separately within each topographic setting. The design matrix included tree status (dead, live) and matched-pair identity as a blocking factor. P values were adjusted using the Benjamini–Hochberg (BH) procedure to control the false discovery rate, and genes with BH-adjusted P < 0.05 were considered significant.

Author contributions

D.Z., Y.Z., J.L., H.C., and W.Y. conceived the study. D.Z., Z.W., J.Y., Y.F., and Y.B. collected field samples and curated the metadata. D.Z., Y.L., B.Z., and J.Z. performed the laboratory assays and sequencing. D.Z. wrote the first draft of the manuscript. J.L., J.Y., Y.F., and Q.Y. contributed to the interpretation of the results and revised the manuscript. J.L. acquired funding and supervised the project.

Peer review

Peer review information

Nature Communications thanks Eoin Brodie and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the National Natural Science Foundation of China-Guangdong Joint Fund (U23A20156) and the National Natural Science Foundation of China (32571866).

Data availability

The raw amplicon reads generated in this study have been deposited in the China National GeneBank Sequence Archive (CNSA) under accession code CNP0008410 https://doi.org/10.26036/CNP0008410⁶⁷. The soil metagenomic sequence data generated in this study have been deposited in CNSA under accession code CNP0008455 https://doi.org/10.26036/CNP0008455⁶⁸. The processed datasets generated in this study, including soil-property data, microbial community profiles, genome-informed functional traits, and associated analysis outputs, are publicly available in Zenodo https://doi.org/10.5281/zenodo.20849331⁶⁹. The accompanying Source Data file contains the sampling-location, soil-property, and extracellular-enzyme-activity data underlying the relevant main and Supplementary Figs. All datasets are publicly accessible without restrictions. Source data are provided with this paper.

Code availability

The R code used to generate the statistical analyses, figures, and numerical outputs reported in this study is publicly available through GitHub at https://github.com/zhangdx2025/tree-mortality-soil-carbon. A version of the repository corresponding to this study has been permanently archived in Zenodo https://doi.org/10.5281/zenodo.21740253⁷⁰.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77237-1.