The research area encompasses Wenquan County of the Boltala Mongolian Autonomous Prefecture in the western region, as well as Fukang City and Qitai County, situated in the Changji Hui Autonomous Prefecture, to the east of the northern slope of the Tianshan Mountain. The fundamental details of the sample points are outlined in Figure 1 and Supplementary Table S1. Specifically, within the study area of Fukang City, the grassland type is classified as temperate desert, featuring sierozem soil and a sandy soil texture. The dominant flora comprises Haloxylon ammodendron and Seriphidium santolinum, with Kali collinum as an associated species. On the other hand, the grassland type in the Wenquan County research area is temperate steppe, characterized by chestnut soil and a loam soil texture. The primary flora includes Stipa capillata, Festuca ovina, and Artemisia frigida, with Aster altaicus as an associated species. Furthermore, the Qitai County research area is classified as mountain meadow, with chernozem soil and a loam soil texture. The dominant plant species is Alchemilla tianschanica, while the main associated species encompass Bromus inermis, Polygonum viviparum, Trifolium lupinaster, Carex liparocarpos, Astragalus membranaceus, and Geranium carolinianum, among others.

The test sites were strategically positioned within the designated national fixed monitoring sites, each encompassing an area of approximately 3 × 10⁴ m². Two treatments are set in temperate desert (TD), temperate steppe (TS), and mountain meadow (MM), respectively: grazing exclusion plots (GE) and freely grazing plots (FG). Notably, all these exclusion plots have been enclosed since 2012 and remained enclosed for a duration of 9 years by the sampling year of 2021. The overall stocking rate for the three grassland sites ranged from 0.90 to 4.50 sheep units per hectare (hm²). The grazing intensities at these three sites fell within the category of moderate grazing in the local context, aligning with the carrying capacity of livestock in the area. Before enclosure, the enclosed areas exhibited comparable vegetation composition, community characteristics, and landforms to the control grazing areas.

Vegetation surveys were conducted during peak biomass season (July 2021) across enclosure and grazed sites. Three 50-m spaced transects per treatment contained three 1 × 1 m quadrats (50-m intervals; 54 quadrats total). Plant species composition was recorded, with coverage (%) assessed by the point-intercept method, height (cm) by tape measure, density (plants·m⁻²) by direct counting, and above-ground biomass (AGB) by harvesting, oven-drying (105 °C/30 min → 80 °C/24 h), and weighing. Below-ground biomass (BGB) was determined from intact soil blocks (10 × 20 cm) after washing and drying [methods follow Liu et al. (2022)].

Soil sampling occurred concurrently within vegetation quadrats. Stratified samples (0–5 and 5–10 cm depths) were homogenized per layer, with subsamples stored at 4 °C (microbial assays) or air-dried (physicochemical analysis). Standard protocols quantified: (1) pH (soil:water = 5:1); (2) SOC (dichromate oxidation); (3) N/P fractions (Kjeldahl N; Mo-Sb colorimetric P); and (4) Microbial biomass (fumigation-extraction for MBC/MBN/MBP) (Liu et al., 2022) (Full analytical details in Appendix A). All datasets are original. Plant/soil data will be analyzed in a companion study currently in preparation, with no content overlap between manuscripts.

In July 2021, the soil microorganism sample collection was carried out, and three typical samples were set up in each of the treatment area and the control area, with a sample line of 50 m, and the sample of three 1 m × 1 m on each sample line, and the sample space was about 50 m (Figure 1). First, the soil microorganism samples were collected from the soil 0–5 cm and the 5–10 cm layer, and each sample line was mixed evenly and then put in the car refrigerator (−20 °C) in a sealed bag and brought back to the laboratory. The study of soil microorganism samples by high-pass sequencing analysis of community structure and diversity, the research steps are: (1) to extract the total DNA in the sample using the Hongkey group DNA extraction kit. (2) the V3 ~ V4 variable zone sequence of the bacterium 16 s rRNA gene is the ITS1-1f zone of its rRNA gene, with the 338f-806r of the barcode sequence (the fungal derivative is the ITS1-1f-fdon ITS1-1f-r), and the PCR amplification is used to obtain the PCR product (Ju et al., 2024; Caporaso et al., 2011). (3) After the PCR product was built by quantitative and library, the Illumina MiSeq PE 300 platform was used to sequence the high pass, and to obtain the information of the base sequence of bacterial and fungal variable areas. (4) using the QIIME2.0 package, the OTU (taxonomic operational unit) cluster, the OTU represents the sequence and the SILVA database (Fungi for UNITE database) for the comparison analysis, and obtain the OTU corresponding classification unit and its corresponding quantity of abundance information (Jiang et al., 2022). (5) through the analysis of the microbial sequence, the abundance information of the microbiogenic cloud platform of the micro group, the discovery of α diversity, and so on. The analysis of the microbiological sequencing of the study was completed in the city of Shenzhen Microleague Technology Group Co., LTD (Jiang et al., 2022). The soil microbial respiration (SMR) was measured using an automatic soil aerobic culture system (He et al., 2013; Zhang et al., 2025).

The raw microbial sequence data from this study have been deposited in the Figshare repository and are publicly available under the DOI https://doi.org/10.6084/m9.figshare.30024817.v1 and at NCBI BioProject, accession PRJNA1338829 (Figshare permanent link: https://figshare.com/articles/figure/microbial_raw_sequence_data/30024817). These data are publicly available as of the date of publication.

Primary data preprocessing was performed using Microsoft Excel 2020 (Microsoft Corp.). Microbial α-diversity indices and soil microbial respiration metrics across grassland types were statistically analyzed through IBM SPSS Statistics 25.0 (IBM Corp.) employing a hierarchical analytical framework: (1) Before parametric tests, homogeneity of variances was assessed using Levene’s test (α = 0.05). For variables violating homogeneity (p < 0.05), Welch’s ANOVA or non-parametric Kruskal-Wallis tests were applied. (2) One-way ANOVA with post-hoc Tukey tests for inter-biome comparisons; (3) Multifactorial ANOVA to disentangle enclosure effects from ecosystem-type interactions; and (4) Independent-sample t-tests for paired enclosure vs. grazed plot analyzes. Microbial community assembly processes were quantified using the β-nearest taxon index (βNTI) and normalized stochasticity ratio (NST) computed through the “picante” (v1.8.2) and “NST” (v3.0.3) packages in R 4.2.1 (R Core Team). Deterministic versus stochastic dominance was assessed using established ecological thresholds (βNTI > 2 or βNTI < −2 indicates deterministic assembly; −2 < βNTI < 2 reflects stochastic dominance).

Partial Least Squares Path Modeling (PLS-PM) was performed using the plspm package (v0.4.9) in R 4.2.1 to elucidate the direct and indirect effects of environmental covariates on microbial respiration. This method was selected for its robustness with small sample sizes and its suitability for exploratory research and complex model structures. The analysis was designed to: (1) Decouple the direct and indirect effects along hypothesized pathways linking enclosure measures, soil properties (e.g., pH, SOC, TN), plant characteristics (e.g., above-ground biomass, below-ground biomass, richness), microbial properties (e.g., microbial biomass carbon, nitrogen), and microbial respiration; and (2) Test the proposed ecological mechanisms governing respiration dynamics in each grassland type. The model’s predictive accuracy and explanatory power were evaluated using the coefficient of determination (R²) for endogenous latent variables. The overall model quality was assessed by the Goodness-of-Fit (GoF) index. The significance of path coefficients was determined through a bootstrap validation procedure with 5,000 iterations to generate robust standard errors and confidence intervals.

All statistical graphics were programmatically generated using “ggplot2” (v3.4.2) in R 4.2.1, with subsequent vector refinement and compositional optimization performed in Adobe Illustrator 2021 (Adobe Inc.). Data are presented as mean ± standard error (SE).

Bacterial phyla (Figure 2a): Actinobacteria, Proteobacteria, Acidobacteria, and Bacteroidetes dominated all grassland types and soil layers. After enclosure, Actinobacteria and Acidobacteria increased by 0.66–8.44% and 0.41–2.03% respectively, in the 0–5 cm layer of all grasslands, while Proteobacteria showed a 1.23% increase exclusively in a temperate steppe. Conversely, Bacteroidetes, Gemmatimonadetes, and Firmicutes decreased by 1.29–5.64%, 0.04–3.50%, and 0.05–0.53% in this layer. In the 5–10 cm layer, Actinobacteria increased by 2.60–9.41%, whereas Bacteroidetes, Verrucomicrobia, and Gemmatimonadetes decreased by 0.17–2.86%, 0.19–1.37%, and 0.28–4.09%, respectively.

Fungal phyla (Figure 2b): Ascomycota and Basidiomycota dominated all grasslands and soil layers, with other phyla (excluding Glomeromycota) accounting for <1% relative abundance. After enclosure, Basidiomycota increased by 7.54–12.64% in the 0–5 cm layer of all grasslands, while Ascomycota showed a 6.62% increase exclusively in a temperate steppe. In the 5–10 cm layer, Basidiomycota increased by 0.13–10.58% across all grasslands, while Ascomycota increased by 1.82–9.95% in temperate steppe and mountain meadow.

Enclosure-induced divergent bacterial alpha-diversity responses across grassland ecosystems and soil depths. After enclosure, in the 0–5 cm layer, the Chao1 index increased by 14.0% (p < 0.05) in the temperate desert but decreased by 10.0% (p < 0.05) in the mountain meadow, whereas no significant changes occurred in the temperate steppe [Figure 3A(a)]. At 5–10 cm depth, after enclosure, only the temperate desert exhibited a response, with a 2.2% increase in Shannon index [p < 0.05, Figure 3A(b)]. Beta diversity analysis (Bray-Curtis PCoA) demonstrated strong grassland-type-driven segregation (p < 0.001), with bacterial communities of temperate desert and steppe separated along PCoA axis 1 (32.49–32.51% variance explained), independent of soil depth [Figures 3A(c,d)].

Fungal communities displayed contrasting layer-specific enclosure effects. After enclosure, in surface soil (0–5 cm), the Shannon index increased by 12.4% (p < 0.05) in the temperate desert, while subsurface soil (5–10 cm) showed a 27.2% Chao1 index rise (p < 0.05) in the temperate steppe [Figures 3B(a,b)]. Principal coordinates analysis (PCoA) demonstrated distinct stratification patterns: in 0–5 cm soil layer, temperate desert, and steppe ecosystems exhibited primary fungal community segregation along PCoA axis 1 [26.04% variance, Figure 3B(c)], and in 5–10 cm soil layer, temperate steppe communities showed predominant separation along PCoA axis 2 [15.5% variance, Figure 3B(d)].

As shown in Figure 4, after grassland enclosure, the assembly of soil bacterial communities (βNTI) in both soil layers (0–5 cm and 5–10 cm) across all three grassland types shifted to complete dominance (100%) by deterministic processes [Figure 4A(a)]. Critically, in the 0–5 cm layer before enclosure, bacterial assembly exhibited a mixed process (55.6% deterministic + 44.4% stochastic), which transitioned entirely to deterministic processes (100%) following enclosure [Figure 4A(b)].

For fungal communities under enclosure conditions, stochastic processes dominated assembly in both soil layers of the temperate desert and the temperate steppe, whereas deterministic processes prevailed in the mountain meadow [Figure 4B(a)]. In the 0–5 cm layer, pre-enclosure fungal assembly was predominantly deterministic (88.9%), but shifted to stochastic dominance (55.6%) post-enclosure [Figure 4B(b)]. This directional reversal in the topsoil, from mixed to deterministic for bacteria versus deterministic to stochastic for fungi, highlights an opposing response to enclosure management. These assembly shifts provide the mechanistic basis for the respiration patterns analyzed in Section 3.5.

By analyzing the linear relationships between plant properties, soil characteristics, microbial diversity, and soil microbial assembly processes, our results demonstrate that grassland enclosure management shifted microbial community assembly from predominantly stochastic toward deterministic processes; specifically, for bacterial βNTI, aboveground biomass (AGB) was the dominant driver [R² = 0.52, p < 0.001; Figure 4A(c)], while for fungal βNTI, AGB showed the strongest correlation (R² = 0.58, p < 0.001), with significant but weaker associations observed for belowground biomass (BGB: R² = 0.23, p < 0.05), soil organic carbon (SOC: R² = 0.28, p < 0.05), total nitrogen (TN: R² = 0.33, p < 0.05), and total phosphorus (TP: R² = 0.22, p < 0.05) [Figure 4B(c)].

Through the Figure 5 analysis, in the 0–5 cm layer, microbial respiration decreased by 71.3% (p < 0.01) in the temperate steppe and 31.8% (p < 0.05) in mountain meadow compared to pre-enclosure levels (Figure 5a). At 5–10 cm depth, only the temperate desert showed a marked decrease of 74.7% (p < 0.05) in microbial respiration after enclosure, while other grassland types remained statistically unchanged (Figure 5b).

Pearson correlation analysis showed that soil microbial respiration was significantly correlated with NH₄⁺-N, NO₃⁻-N, MBC, MBN, and fungal Shannon index in the temperate desert after enclosure (p < 0.05 and p < 0.01). However, there were significant and extremely significant correlations with BGB, TN, and bacterial-βNTI in the mountain meadow (p < 0.05 and p < 0.01), but no significant correlations with each index in temperate steppe (Figure 5c).

The results of the structural equation modeling revealed that the factors influencing soil microbial respiration varied across grassland types (Figure 6). In the temperate desert, microbial respiration was primarily driven by the direct effects of soil properties and microbial biomass (p < 0.05, Figure 6a). Together with microbial diversity and assembly processes, these factors accounted for 95.1% of the variation in microbial respiration (Figure 6a). However, the total effects indicated that grazing exclusion had the strongest influence, suggesting that enclosure measures altered the habitat conditions of the temperate desert grassland, thereby indirectly affecting microbial respiration (Figure 6b). In the temperate steppe, microbial respiration was mainly influenced by plant traits (p < 0.05), followed by microbial assembly processes (Figures 6c,d). These factors, along with microbial diversity, collectively explained 45.5% of the variation in microbial respiration (Figure 6c). This pattern contrasted sharply with that observed in the temperate desert. In the mountain meadow, microbial respiration was directly affected by soil properties and microbial assembly processes (p < 0.01, Figure 6e), which, together with microbial biomass and diversity, explained 88.3% of the variation in microbial respiration (Figures 6e,f).

High-throughput sequencing revealed conserved phylum-level dominance patterns in grassland microbial communities, with bacterial assemblages consistently dominated by Actinobacteria, Proteobacteria, Acidobacteria, Bacteroidetes, Verrucomicrobia, and Chloroflexi, while fungal communities were primarily composed of Ascomycota and Basidiomycota (Figures 2a,b). These patterns align with established grassland microbial ecology frameworks, suggesting evolutionary conservation of core taxa with multifunctional adaptations to environmental fluctuations (Wei et al., 2020; Bachar et al., 2010; Chu et al., 2016; Zhang et al., 2013). Specifically, these conserved microbial guilds exhibit high soil niche plasticity, enabled by traits such as xerotolerance, hydric stress resilience, and metabolic versatility across moisture gradients. This adaptive capacity allows them to maintain functional dominance despite grassland-type variations (Li et al., 2019; Ren et al., 2018; Li et al., 2018). Following enclosure implementation, we observed systematic shifts in bacterial functional guilds: oligotrophic groups (Actinobacteria, Acidobacteria) increased in relative abundance, whereas copiotrophic phyla (Bacteroidetes, Firmicutes) declined. This successional transition from fast-growing to slow-growing microbial strategists likely reflects enhanced soil nutrient availability and vegetation diversity post-grazing exclusion (Zhou et al., 2017; Zhang et al., 2018), particularly through increased organic matter inputs that favored resource-competitive Actinobacteria and Acidobacteria. Concurrently, improved soil aeration under enclosure reduced the viability of anaerobic Firmicutes and Bacteroidetes, demonstrating how management-induced physicochemical changes cascade through microbial communities. Fungal communities demonstrated Basidiomycota enrichment across soil profiles, aligning with their lignin-degrading specialization (Lienhard et al., 2014; Ferreira de Araujo et al., 2017) and enclosure-induced accumulation of lignin-rich graminoid litter (He et al., 2019; Li et al., 2019).

Ecosystem-specific α-diversity trajectories emerged, with the xeric temperate desert exhibiting the strongest microbial re-assembly after enclosure, while mesic systems remained comparatively stable (Figure 3). New correlative evidence (Supplementary Figure S1) attributes these patterns to management-dependent plant–soil–microbe feedbacks: In the temperate desert, enclosure reversed bacterial linkages-from positive correlations with plant richness/SOC/TP/AP under grazing to negative associations—signaling a resource-driven regime shift where elevated plant inputs intensified competitive exclusion (Supplementary Figures S1a,b) (Cheng et al., 2016); simultaneously, fungi shifted from neutral to positive correlations with BGB/TN/AP, indicating nutrient-mediated functional specialization toward saprotrophic pathways under enrichment (Supplementary Figures S1a,b) (Zhou et al., 2024). In the temperate steppe, bacteria tightened phosphorus coupling (TP/AP, post-enclosure), revealing copiotrophic exploitation of newly available P; fungi decoupled from plant diversity (for Chao1), yet gained positive Shannon links with microbial biomass, demonstrating taxon-specific N-limitation persistence (Supplementary Figures S1c,d). In the mountain meadow, bacterial diversity strengthened ties to plant richness and MBC, confirming edaphic niche creation via litter-root synergy; fungal diversity maintained negative plant correlations, reflecting unwavering competitive exclusion by vegetation (Supplementary Figures S1e,f). Collectively, these data-anchored linkages establish that enclosure duration and ecosystem type jointly determine feedback loop directionality—from resource competition to niche optimization—refining grassland microbial models (Zhang et al., 2018; Zhang and Fu, 2022).

Our null model analysis demonstrated that enclosure management induced divergent microbial assembly pathways: bacterial communities shifted toward deterministic selection (βNTI > − 2.0) in arid surface soils, while fungal assembly maintained stochastic dominance (+2.0 < βNTI < −2.0) (Figure 4). Crucially, these patterns were driven by distinct environmental factors: (1) Bacterial deterministic assembly was primarily governed by aboveground biomass (AGB) [R² = 0.52, p < 0.001; Figure 4A(c)], where enhanced plant inputs intensified niche competition through litter-root synergy, optimizing selective pressures for competitive taxa (Wu et al., 2022; Fierer, 2008; Thuiller et al., 2005; Mason et al., 2011); (2) Fungal stochastic persistence occurred under multi-resource constraints involving belowground biomass (BGB: R² = 0.23), soil organic carbon (SOC: R² = 0.28), total nitrogen (TN: R² = 0.33), and total phosphorus (TP: R² = 0.22) [p < 0.05; Figure 4B(c)]. Consistent with Dini-Andreote et al. (2015), such heterogeneous resource availability amplifies stochasticity under moderate selection, enabling species-specific physiological adaptations that sustain biodiversity in resource-abundant environments (Chase, 2010).

These results refine classical niche theory (Bahram et al., 2018; Tripathi et al., 2012; Jiao et al., 2022): enclosure promotes deterministic bacterial assembly via plant-derived filters (AGB), aligning with observed competitive exclusion patterns (Section 4.1), while fungi respond to localized resource mosaics (BGB/SOC/TN/TP) through stochastic dynamics that facilitate functional diversification. This establishes taxon-specific environmental drivers as key mediators of microbial stabilization strategies under perturbation.

Our investigation demonstrates that grazing exclusion consistently suppressed soil microbial respiration across three distinct grassland ecosystems (Li et al., 2013), indicating its universal effectiveness in reducing carbon mineralization and potentially enhancing ecosystem carbon sequestration (Figures 5, 6). However, the underlying mechanisms driving this common outcome were strikingly divergent, highlighting the paramount importance of ecosystem-specific context in predicting the response to management interventions.

In the temperate desert grassland, the reduction in respiration was primarily governed by microbial physiological adaptation to nutrient limitation (Figures 6a,b) (Esch et al., 2017). Enclosure improved soil nutrient conditions and promoted microbial biomass accumulation, but it also induced strong stoichiometric imbalances (Supplementary Table S2). This triggered a fundamental metabolic trade-off, where microbes prioritized survival and biomass formation over respiratory activity. Consequently, the community shifted from a fast-growing, high-respiration state to a maintenance-oriented, energy-conserving state, a process consistent with the “microbial carbon pump” concept (Liang et al., 2017; Liang et al., 2019). Here, the direct negative effect of soil properties and microbial nutrient limitation overwhelmed the positive effect of increased biomass, leading to enhanced carbon retention efficiency.

Conversely, in the temperate steppe, the plant community emerged as the central mediator. Enclosure led to a decline in plant diversity, which itself exerted a strong negative effect on microbial respiration, likely through the production of poorer quality litter (Supplementary Table S3; Figures 6c,d) (Wang et al., 2014). Paradoxically, the net respiration decreased, a outcome resolved by a cascade of changes: enclosure fostered a more deterministic microbial assembly process, which selected for a community with a higher carbon use efficiency, thereby enhancing carbon retention—a finding that aligns with the newly identified primacy of microbial CUE in controlling soil carbon storage at a global scale (Figure 4) (Zhang et al., 2022; Tao et al., 2022). This efficiently assembled community effectively counteracted the upward pressure on respiration from both increased plant biomass and soil nutrients. Thus, the microbial response to environmental filtering, rather than plant traits directly, ultimately governed the carbon cycle.

In contrast to both, the mountain meadow exhibited a clear top-down control mechanism (Figures 6e,f) (Aqeel et al., 2023; Reynolds et al., 2003). Enclosure promoted plant growth, which directly and negatively impacted microbial biomass (Supplementary Tables S2, S3). This sharp decline in microbial abundance was the paramount factor suppressing respiration, as evidenced by a strong positive relationship between them (Total effect = 0.53). The reduction in biomass further weakened the microbial assembly process, which itself had a strong positive effect on respiration. This uncoupled the shift toward deterministic assembly from an increase in overall metabolic activity. Despite strong positive direct effects from soil properties, their influence was insufficient to offset the overwhelming suppressive effect of a diminished microbial pool.

Synthesis and Implications: Despite a common outcome, our study reveals three unique pathways through which enclosure moderates the soil carbon cycle: (1) Microbial Metabolic Trade-off in deserts, (2) Microbial Assembly-Mediated Efficiency in steppes, and (3) Plant-Induced Biomass Reduction in meadows. This spectrum of mechanisms—from primarily abiotic (soil nutrients) to biotic (plants and microbes)—underscores that effective ecosystem management must be context-dependent. Our findings move beyond a simple narrative of enclosure increasing carbon stocks and instead illuminate the complex biotic interactions that ultimately determine ecosystem carbon cycling efficiency. Future assessments of grazing exclusion policies should integrate these ecosystem-specific mechanisms to accurately predict their long-term impacts on grassland carbon sequestration.