Long‐Term Warming Reduces Bacterial Diversity and Functional Potential in Temperate Forest Soil

ABSTRACT

Soil microbes are key regulators of forest carbon cycling, yet how their diversity and functional potential respond to long‐term warming remains poorly understood. Here, we report a five‐year in situ warming experiment in a temperate forest, combining ten repeated measurements of microbial diversity and functional gene potential, as well as continuous monitoring of soil CO₂ flux. We found that warming progressively reduced bacterial diversity and induced phylogenetically conserved community reorganization. Under warming, community composition shifted in a phylogenetically conserved manner. Warming generally reduced the abundance of microbial functional genes across most carbon‐, nitrogen‐, and phosphorus‐cycling gene categories, except for genes associated with starch decomposition. Warming also altered the factors associated with soil CO₂ flux: microbial diversity showed a stronger association with soil CO₂ flux under long‐term warming, whereas soil moisture was the dominant predictor in the control treatment. This warming‐enhanced biodiversity control over soil CO₂ flux was associated with shifts in microbial functional potential, particularly increases in starch‐degrading genes and microbial biomass production potential. Together, our results suggest that warming can restructure microbial communities in ways that strengthen biodiversity‐dependent regulation of soil carbon cycling, with implications for climate‐carbon feedbacks.

Long‐term warming restructures forest soil microbial communities and strengthens biodiversity‐dependent regulation of carbon cycling.

1. Introduction

Forest soils store an enormous amount of carbon, representing one of the largest and most dynamic carbon reservoirs in the terrestrial biosphere (Pan et al. 2011). The persistence of this reservoir depends not only on plant carbon inputs but, critically, on microbial decomposition processes that determine whether plant‐derived carbon is stabilized in soils or released back to the atmosphere (Baldrian et al. 2023; Bradford et al. 2016). Because microbial metabolism is inherently temperature sensitive, global warming has the potential to rapidly alter soil CO₂ emissions and destabilize forest soil carbon stocks (Melillo et al. 2017; Sáez‐Sandino et al. 2023). With global mean surface temperature projected to exceed 2°C above pre‐industrial levels this century (IPCC 2023), understanding how forest soil microbiomes respond to sustained warming has become increasingly urgent (Willing et al. 2024).

Studies have shown that warming can reshape soil microbial composition and diversity in forest ecosystems, yet reported responses remain strongly context dependent (DeAngelis et al. 2015; Li et al. 2025; Nottingham et al. 2022; Shinfuku et al. 2024). Conceptually, warming and its indirect effects, particularly soil drying and shifts in substrate availability, are expected to favor taxa with greater thermal tolerance, drought resistance, and competitive ability under resource limitation (Oliverio et al. 2017; Tajmel et al. 2024). Consistent with this expectation, members of Alphaproteobacteria and Acidobacteria show increased relative abundance after simulated warming (DeAngelis et al. 2015), potentially due to their differences in resource‐use efficiency and stress tolerance. In addition, empirical evidence indicates that microbial diversity responses to warming vary markedly in direction and magnitude. At Harvard Forest, bacterial diversity increased after two decades of field warming (DeAngelis et al. 2015), whereas subsequent analyses revealed strong seasonal modulation of these responses (Shinfuku et al. 2024). In contrast, microbial diversity declined rapidly after two years of warming in a tropical forest (Nottingham et al. 2022). A likely reason for these contrasting outcomes is the scarcity of long‐term, repeated observations of microbial communities under warming, as snapshot sampling may fail to capture gradual, cumulative, or seasonally contingent trajectories (DeAngelis et al. 2015; Romero‐Olivares et al. 2017; Sáez‐Sandino et al. 2025; Sun et al. 2025). Indeed, grassland warming experiments suggest that microbial responses often accumulate over time rather than stabilizing rapidly (Yuan et al. 2021), but whether similar temporal dynamics occur in forest soils, where environmental heterogeneity and substrate complexity are greater (Pan et al. 2011), remains unresolved. These uncertainties limit our ability to predict how warming reorganizes microbial functional potential and its control over soil carbon cycling through time.

In addition, changes in microbial diversity and community composition are generally thought to affect soil carbon decomposition (Crowther et al. 2019; Wang et al. 2026; Zhou et al. 2011). Yet the links between microbial diversity and soil CO₂ emissions are inconsistent, particularly in forest soils where decomposition is frequently constrained by abiotic factors such as soil nitrogen or moisture (Hagerty et al. 2014; Schindlbacher et al. 2015). Under ambient conditions, strong environmental limitation may mask biological controls. However, climate warming may alter this balance by restructuring moisture regimes and substrate availability, potentially strengthening microbial regulation of carbon decomposition (García‐Palacios et al. 2021; Nottingham et al. 2022). The extent to which warming reshapes the pathways linking environmental drivers, microbial communities and soil CO₂ fluxes is still unclear. This uncertainty has important implications for carbon‐cycle modeling, as most forest carbon models emphasize abiotic controls while representing microbial effects using static or highly simplified parameters (Allison et al. 2010; Wieder et al. 2013).

Here, we conducted a large‐scale, five‐year in situ soil warming experiment in a temperate forest using replicated 6 × 18 m heated plots (108 m² each; three replicates; Figure S1). This platform provides a rare multi‐year record of microbial compositional and functional dynamics under sustained warming in forest soil. We collected soil once prior to warming and ten times during the five‐year warming period. We focused the analyses on bacterial communities as bacteria are characterized by rapid turnover, high metabolic plasticity, and strong links to carbon decomposition (Oliverio et al. 2017; Wang et al. 2021). We tested two hypotheses: first, that sustained warming would alter microbial diversity, composition, and functional potential, with effects strengthening through time; and second, that moisture limitation emerging under prolonged warming could indirectly regulate soil CO₂ flux through its effects on microbial diversity and functional genes. Together, these hypotheses motivate a long‐term, high‐resolution assessment of forest soil microbial communities, providing empirical evidence needed to link microbial community change with soil carbon cycling under global warming.

2. Materials and Methods

2.1. Site Description

The experimental warming platform was located at the Qingyuan Forest National Observation and Research Station, Chinese Academy of Sciences in Liaoning Province, northeastern China (41°51′ N, 124°54′ E; Figure S1). The region represents a typical temperate mixed coniferous‐broadleaf forest dominated by Juglans mandshurica and Larix kaempferi . The climate is continental monsoon, with a mean annual precipitation of 811 mm, more than 80% of which occurs between May and September. The mean annual temperature is 4.5°C, with daily extremes ranging from −37.6°C to +36.5°C and a frost‐free period of approximately 130 days. The growing season extends from May to late September, with snow typically covering the soil from late November to late March.

2.2. Experimental Design

The field warming experiment consisted of six 18 × 6 m plots (108 m² each), three of which were randomly assigned to the warming treatment (Figure S1). Each warmed plot was equipped with 18 infrared heaters (2 kW h⁻¹, 8 mm × 151 cm, HS‐2420, Kalglo Electronics Co., USA) suspended 2 m above the ground and evenly spaced to ensure uniform heating. Control plots were fitted with identically shaped but non‐heating “dummy” devices to mimic shading. The estimated shading caused by the heaters accounted for < 3.8% of the total plot area, negligible under the closed canopy. More detailed experimental design can be found in our previous study (Duan et al. 2022). The warming system was test‐operated for approximately six months in 2018 to calibrate the infrared heating and control system, and full experimental warming commenced in early 2019. During the snow‐free period (late March–early December) each year, soil temperature in warmed plots was automatically maintained at +2°C above the control at 5 cm depth using a temperature feedback control system (Beijing Comity Measure & Control Co., China).

2.3. Soil Sampling and Analyses

For background characterization, soil samples were collected in 2017 prior to the initiation of the warming experiment. After warming began, soils were sampled annually during the growing season (July–September) from 2019 to 2023. Each plot was divided into two subplots, and four sampling points were randomly selected within each subplot. At each sampling event, the organic layer (O horizon, ~5 cm thick) and the mineral layer (A horizon, 0 ~ 10 cm thick) were collected. All samples were immediately placed on ice, transported to the laboratory, and sieved through a 2 mm mesh to remove roots and stones. For inorganic nitrogen measurements, 10 g of fresh soil were extracted with 40 mL of 2 mol L⁻¹ KCl solution by shaking for 1 h. The extracts were frozen at −20°C until analysis of NO₃ ⁻ and NH₄ ⁺ concentrations using a SmartChem 200 discrete analyzer (Westco Scientific Instruments, Italy). Soil dissolved organic carbon (DOC) and extractable organic nitrogen (DON) were quantified by extracting fresh soil with 0.5 mol L⁻¹ K₂SO₄ at a 1:4 (w/v) soil‐to‐solution ratio, followed by analysis using a TOC/TN analyzer (Shimadzu, Kyoto, Japan).

2.4. Amplicon Sequencing and Analyses

Soil DNA was extracted from 0.25 g of lyophilized soil using the MoBio PowerSoil DNA Extraction Kit (MoBio Laboratories, USA), following the manufacturer's instructions. Microbial community composition was characterized by 16S rRNA gene amplicon sequencing. The V4 region was amplified using the primer pair 515F (5′‐GTGYCAGCMGCCGCGGTAA‐3′) and 806R (5′‐GGACTACNVGGGTWTCTAAT‐3′). PCR products were sequenced on an Illumina MiSeq platform (paired‐end 2 × 300 bp; Majorbio Co., Shanghai, China). Raw reads were merged and quality‐filtered using vsearch, and chimeras were removed. Amplicon sequence variants (ASVs) were inferred using the unoise3 algorithm in USEARCH. Representative ASVs were taxonomically assigned against the SILVA database (version 138).

Alpha‐diversity indices (richness, Shannon, Simpson, and Faith's PD) were calculated using the vegan R package. Beta‐diversity was assessed using Bray–Curtis dissimilarity and visualized by principal coordinates analysis (PCoA). Differences in community composition between treatments were tested by PERMANOVA. Differentially abundant ASVs (DEASVs) were identified using DESeq2 (Love et al. 2014), edgeR (Robinson et al. 2010), and STAMP (Parks et al. 2014), with false discovery rate (FDR) correction. The resulting DEASVs were subjected to fuzzy c‐means clustering using the Mfuzz algorithm on standardized abundance data to identify groups of taxa exhibiting similar temporal response patterns and dynamic trajectories. Cluster‐level temporal patterns were visualized and summarized using the ClusterGVis package (Zhang et al. 2026). For each cluster, we calculated the mean difference in abundance between warming and control treatments at each sampling time point (warming−control), thereby generating cluster‐specific temporal trajectories of warming‐induced divergence across the experimental period.

To investigate the evolutionary relationships among the DEASVs, we conducted a phylogenetic analysis utilizing their representative sequences. Initially, all DEASV sequences were aligned using MUSCLE (Edgar 2004) to produce a high‐quality multiple sequence alignment. This alignment was subsequently analyzed through maximum‐likelihood phylogenetic inference using IQ‐TREE software (Nguyen et al. 2015). The resulting phylogeny was visualized with ggtree (Yu et al. 2017). This methodological approach enabled us to elucidate the phylogenetic structure of responsive taxa and evaluate the distribution of warming‐associated ASVs within the bacterial evolutionary framework.

2.5. Metagenomic Sequencing and Functional Annotation

To assess functional and taxonomic responses at the whole‐community level, shotgun metagenomic sequencing was conducted on an Illumina NovaSeq 6000 platform (paired‐end 150 bp; depth ≈20 Gb per sample; Majorbio Co., Shanghai, China). Quality control was performed using Fastp (v0.20.0), and clean reads were assembled into contigs with MEGAHIT (v1.1.2) (Li et al. 2015) using a de Bruijn graph (k‐mer = 21–121, step = 10). Contigs > 500 bp were retained, and protein‐coding sequences were predicted using Prodigal (v2.6.3).

Predicted protein sequences were annotated by aligning them with the Kyoto Encyclopedia of Genes and Genomes (KEGG v94.2) (Kanehisa and Goto 2000), utilizing DIAMOND BLASTP with an e‐value threshold of less than 1 × 10⁻⁵. To enhance the KEGG annotations, proteins were further characterized using eggNOG‐mapper (version 2.1) for COG and KO functions, and Pfam (version 34.0) (Mistry et al. 2021) with HMMER for identifying conserved domains. Gene abundances were determined by mapping clean reads back to the nonredundant gene catalog, followed by normalization based on gene length and sequencing depth. KEGG metabolic pathways were also reconstructed in parallel using HUMAnN3 with quality‐controlled reads as input. Only pathways with ≥ 50% completeness were retained. The results from this standardized workflow were used to validate the pathway‐level trends observed from the custom assembly‐based gene annotations.

2.6. High‐Throughput Quantitative PCR (QMEC)

To quantify the absolute abundance of functional genes involved in carbon (C), nitrogen (N), phosphorus (P), and sulfur (S) cycling, we used a high‐throughput quantitative PCR‐based chip (QMEC) on the SmartChip Real‐Time PCR system (WaferGen Biosystems, Fremont, USA), following the standard protocol (Zheng et al. 2018). Each gene was amplified in triplicate with an initial denaturation at 95°C for 10 min, followed by 40 cycles of 95°C for 30 s, 58°C for 30 s, and 72°C for 30 s. Reactions with amplification efficiencies < 1.8 or > 2.2, or Ct > 31, were excluded. The 16S rRNA gene was used as a reference to normalize gene abundance, representing total bacterial load.

2.7. Field Soil CO₂ Flux Analysis

Soil CO₂ fluxes were measured using two multichannel automated chamber systems. Each system consisted of a greenhouse gas analyzer (G2508, Picarro, USA) coupled to a channel multiplexer (PRI‐8600, PRI‐ECO, China), which was sequentially connected to 15 automated soil chambers (PRI‐8160, PRI‐ECO, China). In each experimental plot, five chambers were randomly installed to capture within‐plot spatial variability (Huang et al. 2025).

Each measurement cycle lasted 20 min per chamber. However, flux calculations were restricted to the chamber closure phase after equilibration. At the beginning of each cycle, chambers remained open for a short equilibration period to flush residual air from previous measurements and to minimize concentration discontinuities upon closure. Soil CO₂ fluxes were derived from the linear increase in CO₂ concentration during a 5‐min window within the closure period (241–540 s after cycle initiation).

Fluxes were calculated as the slope of the linear regression between CO₂ concentration and time, following standard non‐steady‐state chamber theory. For quality control, each time series was evaluated for linearity by fitting a simple linear regression of CO₂ concentration against time. Flux estimates were discarded when the coefficient of determination (R ²) was < 0.80. This filtering resulted in the removal of approximately 1% of total flux observations. All flux calculations, time‐window selection, and quality‐control procedures were applied uniformly across treatments and sampling dates.

2.8. Statistical Analyses

All statistical analyses were performed in R (v4.5.1). Linear mixed‐effects models (LMMs) were used to evaluate the effects of warming, warming duration (time), and their interaction on microbial diversity, community composition, functional genes and soil physicochemical properties (i.e., soil water content, DOC, NH₄ ⁺, and NO₃ ⁻) using the lme4 and lmerTest packages. In the models, warming, time, and their interaction were treated as fixed effects, while plot was included as a random effect following the model structure: Y ~ warming × time + (1|plot). To test whether warming effects differed between soil horizons, we additionally fitted linear mixed‐effects models including soil horizon (organic vs. mineral) as a fixed factor together with warming, time, and their interactions, following the model structure: Y ~ warming × time × horizon + (1|plot). Model residuals were examined for normality and homoscedasticity. Post hoc pairwise comparisons were conducted using Tukey's HSD test implemented in the emmeans package. Temporal trajectories of dominant bacterial phyla were visualized using relative abundance ratios between warming and control treatments. Linear trends along warming duration were evaluated using linear regression models, and the corresponding coefficients of determination (R ²) and significance levels (p values) were reported. For taxa exhibiting non‐monotonic temporal dynamics, locally weighted scatterplot smoothing (LOESS) was additionally used to visualize nonlinear temporal responses. To reduce the dimensionality of microbial functional gene datasets, principal component analysis (PCA) was performed separately for genes involved in carbon, nitrogen, and phosphorus cycling using the FactoMineR package in R, and the first principal component (PC1), representing the dominant axis of variation within each functional category, was extracted and used as a composite indicator of microbial functional potential.

To guide the selection of predictors for structural equation modeling, we first performed a multiple regression analysis including soil water content (SWC), dissolved organic carbon (DOC), NH₄ ⁺, NO₃ ⁻, and bacterial Shannon diversity as predictors of cumulative growing‐season CO₂ flux (Figure S2). Based on this analysis, we constructed structural equation models (SEMs) using the lavaan package, with soil moisture, bacterial Shannon diversity, and microbial functional potential as the primary predictors of soil CO₂ flux. Microbial functional potential was represented by the first principal component (PC1) scores derived from carbon‐, nitrogen‐, or phosphorus‐cycling genes, which summarized multivariate functional gene information while reducing model complexity and minimizing collinearity. Because the number of observations was limited, we did not include all functional gene groups and microbial biomass simultaneously in a single SEM; instead, we evaluated alternative models using different microbial functional indicators, including carbon‐, nitrogen‐, and phosphorus‐cycling gene groups and microbial biomass nitrogen (MBN), in separate SEMs. Separate SEMs were fitted for control and warming treatments to identify treatment‐specific causal relationships. Nonsignificant pathways (p > 0.05) were retained if they were theoretically justified to represent potential mechanistic linkages among variables. Model fit was evaluated using the chi‐square test (χ ²) and its associated p‐value, as well as the comparative fit index (CFI), standardized root mean square residual (SRMR), root mean square error of approximation (RMSEA), and the coefficient of determination (R ²). These global SEMs were fitted separately for the control and warming treatments using the lavaan package, and all reported fit indices refer to this framework.

To further assess whether alternative environmental predictors could explain soil CO₂ flux, we conducted sensitivity analyses within the global SEM framework by replacing soil moisture with DOC, NH₄ ⁺ or NO₃ ⁻ in separate models. To further evaluate the robustness of the relationships among warming, soil moisture, microbial diversity, and soil CO₂ flux, we additionally conducted a piecewise SEM analysis by combining observations from both control and warming treatments and explicitly incorporated warming treatment as an exogenous predictor. Plot identity was included as a random effect in all component mixed‐effects models to account for repeated measurements through time. Because the piecewise SEM and the global SEMs differ in their underlying assumptions and model structures, and because the piecewise model was used to account for the repeated‐measures structure of the time‐series data, its fit was evaluated using Fisher's C and AIC rather than the global fit indices. In addition, to assess whether the inferred relationships were sensitive to the choice of microbial diversity metric, additional global SEMs were fitted using observed richness, phylogenetic diversity, and Simpson diversity as alternative measures of microbial diversity. All data visualizations were produced using the ggplot2, ggpubr, and ggsci packages.

3. Results

3.1. Warming Reduces Soil Microbial Diversity

We first tested whether warming effects on bacterial communities differed between soil horizons using linear mixed‐effects models including horizon as a fixed factor. These models revealed significant Treatment × Horizon interactions across all alpha‐diversity metrics (Table S1), indicating that warming effects differed substantially between organic and mineral horizons. The Treatment × Time × Horizon interaction was also significant for richness, suggesting that temporal responses to warming varied between soil horizons.

In the organic horizon, experimental warming consistently reduced soil bacterial α‐diversity, including richness, phylogenetic diversity (PD), Shannon, and Simpson indices, shown by linear mixed‐effects models analysis (all p < 0.05; Figure 1a). Furthermore, these negative effects did not remain static but intensified over time, as evidenced by significant Warming × Time interactions (p < 0.05, Figure 1b). Thus, warming progressively eroded microbial diversity rather than inducing an immediate shift (Figure 1c). Warming also led to a gradual reorganization of microbial community composition. Principal coordinates analysis (PCoA) revealed that bacterial communities were initially similar between warmed and control plots (pre‐warming to 2019), but began to diverge after 2020 and became increasingly distinct by 2021–2023 (p < 0.01; Figure 1d). Further, metagenomic analyses conducted in the final sampling year (2023) independently confirmed lower microbial diversity and pronounced compositional differences in warmed soils (Figure S3), corroborating the 16S rRNA gene‐based results.

FIGURE 1.

Long‐term warming gradually reduces bacterial diversity and reshapes community composition. (a, b) The warming and its interaction with time significantly decreased bacterial α‐diversity (Richness, PD, Shannon, and Simpson indices). Significance was assessed using linear mixed‐effects models; ***p < 0.001; **p < 0.01; *p < 0.05. (c) Heatmap showing the effect sizes of warming on microbial diversity across sampling time. (d) Effects of warming on bacterial community structure. Principal coordinates analysis (PCoA) revealed no treatment separation in pre‐warming and the first warming year (2019), with divergence emerging in 2020 and becoming significant by 2021–2023. (e) Warming significantly altered bacterial community abundance. Differentially abundant ASVs (n = 5810) clustered into five groups, with warming progressively suppressing control‐enriched clusters (C1–C3) and promoting stress‐tolerant clusters (C4–C5).

The observed warming‐induced shifts in bacterial diversity and community composition in the organic layer were accompanied by changes in soil environmental conditions (Figure S4). Soil water content (SWC) declined consistently under warming throughout the five‐year experiment (Figure S4a), whereas dissolved organic carbon (DOC) and inorganic nitrogen pools (NH₄ ⁺ and NO₃ ⁻) showed no consistent responses to warming (Figure S4b–d). Correlation analyses further indicated that, under warming conditions, bacterial α‐diversity was significantly negatively associated with SWC, DOC, and inorganic nitrogen, while no such relationships were detected in control plots (Figure S4e,f).

In mineral soils, warming effects on bacterial α‐diversity were smaller relative to organic soils, and no significant Warming × Time interactions were detected (Figure S5a–c). Bacterial community composition in the mineral layer remained largely stable over time, with PCoA analyses showing minimal separation between warmed and control plots across sampling years (Figure S5d). Given the limited microbial responses observed in mineral soils, subsequent analyses focused exclusively on the organic soils.

3.2. Warming Induces Phylogenetically Conserved Taxonomic Shifts

To resolve how taxonomic changes unfolded through time, we identified 5810 amplicon sequence variants (ASVs) that responded significantly to warming across three independent approaches (see Methods). Temporal clustering of these ASVs revealed five distinct response groups (Figure 1e), indicating a clear and directional reorganization of microbial assemblages under sustained warming. ASVs abundant under control conditions (clusters C1–C3), dominated by Proteobacteria and Acidobacteria, declined progressively over the experimental period. In contrast, warming‐associated clusters (C4–C5), characterized by Firmicutes, Actinobacteria, and Chloroflexi, expanded steadily with increasing warming duration (Figures 1e and S6).

Phylogenetic mapping of warming‐responsive ASVs further showed that lineage‐specific responses became increasingly pronounced through time (Figure 2a,b). Although some subclades exhibited transient fluctuations, the overall phylogenetic signal at ASVs level of warming responses remained stable across years (Table S2), suggesting persistent evolutionary constraints on community reorganization. These temporal patterns were also evident at broader taxonomic levels. Notably, these phyla exhibited distinct trajectories along the warming time (Figure 2a,c), including consistent increases (e.g., Firmicutes), progressive declines (e.g., Planctomycetes, Bacteroidetes, Candidatus Saccharibacteria and Acidobacteria), and non‐linear responses (e.g., Actinobacteria, Chloroflexi and Armatimonadetes).

FIGURE 2.

Long‐term warming induces phylogenetically conserved shifts in microbial taxa. (a) Phylogenetic distribution of 5810 biomarkers illustrating non‐random, clade‐level responses to long‐term warming, with the inner and outer rings indicating the effect sizes of warming and warming × time interaction on bacterial abundance, respectively. Line plots depict the temporal trajectories of relative abundance at the phylum level (warming/control) across all eleven sampling events, highlighting significant linear trends and visually apparent nonlinear temporal dynamics over the five‐year experiment. (b, c) Effect sizes of warming and its interaction with time on dominant bacterial phyla. Significance was assessed using linear mixed‐effects models; ***p < 0.001; **p < 0.01; *p < 0.05.

3.3. Warming Reduces Functional Potential but Enhances Starch Degradation

Experimental warming induced widespread but directional shifts in microbial functional gene assemblages (Figures 3 and S7). Across all measured genes, warming significantly reduced 53% of microbial functional genes involved in carbon (C), nitrogen (N), phosphorus (P), and sulfur (S) cycling (51 genes in total), whereas 17% showed significant increases (Figure S7). When functional genes were grouped by biogeochemical processes, a clear pattern of functional reorganization emerged: genes associated with starch degradation and inorganic phosphorus solubilization increased significantly under warming, whereas genes involved in most other C‐, N‐, and P‐related processes declined consistently (Figures 3b,c and S7). Metagenomic analysis of KEGG pathways after five years of experimental warming further corroborated these gene‐level patterns (Figure 3d). Among the 25 most abundant metabolic pathways, 72% declined significantly under warming, indicating coordinated metabolic restructuring at the pathway level. Pathways involved in glycan, galactose, pyruvate, citrate and central carbon metabolism decreased consistently, whereas starch and sucrose metabolism increased (Figure 3d). Together, these results demonstrate that long‐term warming suppresses central C, N and energy assimilation pathways while preferentially stimulating the breakdown of labile carbon compounds.

FIGURE 3.

Long‐term warming alters soil microbial functional genes and major metabolic pathways. (a, b) Effect sizes of warming and its interaction with time on bacterial functional genes, showing that responses intensified over time. (c) Heatmap of effect sizes across sampling time highlights progressive functional shifts. (d) Effects of warming on 25 top‐level KEGG metabolic pathways; red and blue indicate increases and decreases in relative abundance, respectively. The vertical axis shows estimated changes derived from linear mixed‐effects models. Significance was assessed using linear mixed‐effects models. ***p < 0.001; **p < 0.01; *p < 0.05.

3.4. Warming Strengthens Microbial Diversity Control Over Soil CO₂ Flux

Soil daily CO₂ fluxes over five growing seasons were measured and the annual cumulative CO₂ flux did not differ significantly between warming and control treatments (Figure S8). To identify the primary predictors of soil CO₂ flux, we performed a multiple regression analysis including soil water content (SWC), dissolved organic carbon (DOC), NH₄ ⁺, NO₃ ⁻, and bacterial Shannon diversity as predictors. This analysis identified SWC and Shannon diversity as the only significant predictors (Figure S2). We therefore separated control and warming treatments to explore the role of bacterial diversity in controlling soil CO₂ flux. We constructed a series of structural equation models incorporating soil moisture and microbial diversity, and alternatively including functional genes involved in carbon, nitrogen, and phosphorus cycling, or microbial biomass (Figure S9). Among these, the SEM including carbon‐cycling functional genes showed the best model fit. Across all models, we observed a consistent shift in the dominant drivers of soil CO₂ flux between treatments (Figures 4 and S10–S12).

FIGURE 4.

Long‐term warming strengthens microbial diversity control over soil CO₂ flux. (a) Structural equation models illustrating the pathways linking soil water content, microbial diversity (Shannon index), functional genes associated with carbon decomposition, and CO₂ flux. Arrows indicate standardized path coefficients under control and warming treatments, respectively, highlighting shifts in the relative importance of biotic and abiotic controls. (b) Standardized total effects of soil water content (SWC), microbial diversity (MD), and functional genes (MG) on CO₂ flux under control and warming treatments showing an enhanced contribution of microbial diversity under warming. (c) Partial relationships derived from structural equation models showing associations between soil water and CO₂ flux, and between microbial diversity and CO₂ flux. Values represent standardized residual deviations from model‐predicted partial effects.

Using the model including carbon‐decomposition genes as an illustrative example, soil water content exerted a strong positive correlation with CO₂ flux under control conditions, whereas the contributions of microbial diversity and carbon‐cycling genes were comparatively weak (Figure 4a). Under warming, however, the direct influence of soil moisture on CO₂ flux was substantially reduced and became largely indirect, acting primarily through its effects on microbial diversity and functional gene potential (Figure 4a). Consequently, the overall effect of microbial diversity on CO₂ flux was markedly stronger under warming than under control conditions (Figure 4b). Partial regression analyses independently corroborated this shift in regulatory pathways. Microbial diversity exhibited a significant negative relationship with CO₂ flux only under warming (Figure 4c). Together, these results demonstrate that prolonged warming does not necessarily enhance soil CO₂ emissions but instead strengthens microbial diversity as a key regulator of soil respiration by reorganizing the balance between abiotic and biotic controls.

To assess the robustness of the treatment‐specific SEM results, we performed an additional piecewise structural equation model incorporating both warming treatments and repeated measurements through time (Figure S13). The model showed a good overall fit to the data (Fisher's C = 0.265, p = 0.876). Warming significantly reduced soil moisture (standardized path coefficient = −0.52, p = 0.032) and bacterial Shannon diversity (−0.95, p = 0.002), while both soil moisture (0.65, p < 0.001) and bacterial diversity (−0.55, p < 0.001) significantly influenced soil CO₂ flux. In contrast, no direct effect of warming on soil CO₂ flux was detected (p = 0.876). These findings support the interpretation that long‐term warming affects soil CO₂ flux primarily through indirect effects on soil environmental conditions and microbial communities and corroborate the treatment‐specific SEM results presented in Figure 4. Similar results were obtained when richness, phylogenetic diversity, or Simpson diversity were used as alternative measures of microbial diversity, with all models supporting a significant negative relationship between microbial diversity and soil CO₂ flux (Figure S14). These results indicate that the observed diversity–CO₂ relationship was robust to the choice of diversity metric. To test whether the identification of soil moisture as the primary environmental pathway, and the negative relationship between microbial diversity and soil CO₂ flux, were robust to the choice of environmental predictor, we additionally replaced soil moisture with DOC, NH₄ ⁺, or NO₃ ⁻ in separate models. These analyses yielded qualitatively similar negative relationships between microbial diversity and soil CO₂ flux (Figure S13). However, unlike soil moisture, none of these variables exhibited a consistent response to warming, supporting soil moisture as the primary environmental pathway linking long‐term warming, microbial diversity, and soil CO₂ flux.

4. Discussion

A major contribution of this study is the rare, high‐frequency time series capturing how soil microbial communities respond to experimental warming in a temperate forest. By tracking microbial dynamics continuously over five years, we were able to move beyond the snapshot perspective that characterizes most short‐term warming experiments (Brangarí et al. 2025; Nottingham et al. 2022; Schindlbacher et al. 2015). Rather than observing a transient response (DeAngelis et al. 2015), we found a gradual and cumulative restructuring of the microbial community in forest organic soil. Bacterial diversity declined progressively, while community composition shifted in a consistent and directional manner, supporting our hypothesis that microbial responses to warming can amplify through time rather than converge toward a new equilibrium. Global syntheses and meta‐analyses have examined the effects of warming on soil microbial diversity (Wang et al. 2025; Yang et al. 2021), yet they provide limited insight into the temporal trajectories through which these changes unfold. Some studies have attempted to infer the effects of warming duration using space‐for‐time substitutions (Romero‐Olivares et al. 2017; Sun et al. 2025). However, conclusions drawn from such approaches remain uncertain because spatial heterogeneity can obscure true temporal signals. Our findings are broadly consistent with observations from the long‐term soil warming experiment at Harvard Forest (Melillo et al. 2017; Shinfuku et al. 2024), where microbial communities continued to reorganize, and bacterial diversity exhibited significant changes after two decades of warming (DeAngelis et al. 2015). We also examined fungal community responses over the same period and found that fungal diversity and community composition exhibited comparatively weak responses to warming relative to bacterial communities (unpublished data). Although fungal communities showed occasional compositional differences between warming and control plots at individual sampling dates, these responses were neither as consistent nor as progressive as those observed for bacteria. This contrast suggests that bacterial communities were more sensitive to long‐term warming in this forest ecosystem. Overall, our results highlight the importance of long‐term, temporally resolved observations for capturing gradual but directional microbial responses to climate warming. It is important to note that these microbial responses were primarily observed in the organic horizon, whereas bacterial communities in the mineral soil exhibited comparatively weak responses to warming (Figure S5). This depth‐dependent pattern suggests that warming effects on soil microbial communities are not uniform throughout the soil profile. Therefore, the ecological mechanisms identified in this study should be interpreted mainly in the context of organic‐layer microbial communities, which represent the primary zone of litter decomposition, microbial activity, and carbon turnover in this forest ecosystem.

Our results also indicated that microbial responses to warming are non‐randomly distributed across the bacterial phylogeny (Table S2), revealing pronounced lineage‐specific clustering. This pattern suggests that long‐term warming acts as a sustained selective filter on soil microbial communities, consistent with previous conceptual and empirical frameworks (Bradford 2013; Brangarí et al. 2025; Zhou et al. 2016). The observed shifts reflect the persistence of taxa with conserved traits linked to thermal tolerance, resistance to moisture limitation, or competitive strategies, rather than short‐term physiological responses (Brangarí et al. 2025; Oliverio et al. 2017). Such phylogenetic conservatism likely arises because key temperature‐related physiological traits, such as dormancy capacity (Lennon and Jones 2011), growth rate (Wang et al. 2021), and nutrient utilization ability (Isobe et al. 2020) are themselves evolutionarily conserved. Together, these results suggest that phylogenetically structured microbial responses emerge through cumulative environmental filtering under prolonged warming.

A second major finding of this study is the strong negative relationship between microbial diversity and soil carbon decomposition under prolonged warming. This pattern contrasts with the positive biodiversity‐ecosystem functioning relationships commonly reported under ambient conditions (Delgado‐Baquerizo et al. 2016, 2017; Eisenhauer et al. 2026), but is supported by a recent study reporting negative associations between microbial diversity and soil carbon decomposition (Pan et al. 2025). One possible explanation for this negative relationship is that warming shifts microbial communities toward dominance by fast‐respiring taxa with lower carbon‐use efficiency, resulting in higher CO₂ release (Pan et al. 2025; Wang et al. 2026). In addition, warming was accompanied by an enrichment of microbial functional genes involved in starch and sucrose metabolism, together with an increased potential for microbial biomass production (Figure 3d). Because starch and sucrose represent relatively labile carbon substrates, their enrichment may indicate an increased microbial capacity to utilize readily available carbon compounds under warming. These coordinated functional and physiological changes indicate substantial reorganization of microbial carbon‐processing pathways under warming. However, despite these shifts, cumulative soil CO₂ flux remained unchanged, suggesting that potential stimulation of carbon turnover may have been offset by concurrent constraints, particularly the persistent decline in soil moisture observed throughout the experiment (Hagerty et al. 2014; Wang et al. 2026).

More importantly, our results further indicate that, under warming, the regulation of soil carbon emissions may shift from predominantly abiotic controls (i.e., soil moisture) toward stronger biotic regulation mediated by microbial community structure and function. This hidden reorganization of control pathways may precede nonlinear or threshold responses of soil carbon cycling under continued climate warming (Wang et al. 2026). From a modeling perspective, these findings provide empirical support for explicitly incorporating microbial processes into Earth system models (García‐Palacios et al. 2021). Many current carbon‐climate models typically assume that microbial activity responds instantaneously to temperature, ignoring delayed community adaptation, trait‐mediated selection, and the stabilizing roles of rare taxa (Allison et al. 2010; Crowther et al. 2019). Integrating these mechanisms, together with microbial physiological traits and evolutionary adaptation, could substantially improve projections of soil carbon persistence under climate change (Guo et al. 2020; Hagerty et al. 2014).

Overall, our results demonstrate that sustained warming progressively restructures microbial communities in the organic horizon, leading to cumulative diversity loss and phylogenetically organized shifts that strengthen microbial control over soil CO₂ fluxes and soil respiration. In this temporally evolving context, microbial diversity acts as a key regulator of soil CO₂ fluxes, shaping carbon turnover pathways under warming rather than merely responding to abiotic change. These findings highlight the importance of long‐term microbial dynamics in the organic horizon for understanding and predicting forest soil carbon responses to climate warming.

Author Contributions

Yue Liu: data curation, visualization, writing – original draft, writing – review and editing, formal analysis. Lingrui Qu: investigation. Dongwei Liu: investigation. Yunfeng Yang: supervision. Ember M. Morrissey: supervision. Yihang Duan: investigation. Jian Wang: investigation. Xu Wang: investigation. Kai Huang: investigation. Fangying Qu: investigation. Xue Guo: supervision. Peter B. Reich: supervision. Edith Bai: supervision. Jizhong Zhou: supervision. Yunting Fang: conceptualization. Chao Wang: conceptualization, funding acquisition.

Funding

This work was supported by the National Natural Science Foundation of China, 42322306, 32371845, 42430510, 32301388. National Key Research and Development Program of China, 2023YFE0124000. International Partnership Program of Chinese Academy of Sciences, 064GJHZ2022054FN. Youth Innovation Promotion Association of the Chinese Academy of Sciences, Y2022064. Project for Young Scientists in Basic Research of the Chinese Academy of Sciences, YSBR‐108. Liaoning Revitalization Talents Program, XLYC2403021.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Study site and experimental design of the in situ soil warming experiment in a temperate forest of northeast China.

Figure S2: Multiple linear regression identifying key predictors of soil CO₂ flux.

Figure S3: Metagenomic evidence for warming‐induced declines in bacterial diversity and community reassembly.

Figure S4: Effects of soil variables on microbial diversity under long‐term warming.

Figure S5: Weak responses of bacterial communities in the mineral A horizon to long‐term warming.

Figure S6: Taxonomic organization of soil bacterial communities under long‐term warming.

Figure S7: Long‐term warming progressively alters soil microbial functional genes.

Figure S8: Soil CO₂ flux responses to long‐term warming.

Figure S9: Hypothesized structural equation model.

Figure S10: Relationships among soil variables, microbial diversity, and soil CO₂ flux.

Figure S11: Relationships among soil variables, microbial diversity, and soil CO₂ flux.

Figure S12: Relationships among soil variables, microbial biomass, and soil CO₂ flux.

Figure S13: Sensitivity analyses of environmental predictors in piecewise structural equation models.

Figure S14: Robustness of piecewise structural equation model results across alternative bacterial 55 diversity metrics.

Table S1: Individual and interactive effects of warming treatment, warming duration, and soil layer on microbial alpha diversity indices.

Table S2: Phylogenetic patterns of microbial taxa under warming and its interaction with time.

Liu, Y. , Qu L., Liu D., et al. 2026. “Long‐Term Warming Reduces Bacterial Diversity and Functional Potential in Temperate Forest Soil.” Global Change Biology 32, no. 9: e71104. 10.1111/gcb.71104.

Data Availability Statement

All data are available in the main text or the Supporting Information. The raw sequences of bacteria can be accessed at the link: https://www.scidb.cn/en/s/BvUj6f. The code and data that were used for repeating the analysis in this study can be found at the GitHub: https://github.com/cwangiae/Forest‐warming.git. Please contact the corresponding author for additional data not reported in the manuscript or Supporting Information.