Soil microbial community assembly drives ecosystem multifunctionality under grazing disturbance by regulating diversity and network structure

Abstract

In the context of global climate change, overgrazing has impacted ecosystem multifunctionality (EMF). However, the influence of different characteristics of microbial communities on EMF under grazing pressures remains unclear. In this study, we examined how microbial community diversity, co-occurrence network structure, and assembly processes influence ecosystem function and multifunctionality under three grazing pressures: light grazing, heavy grazing, and grazing exclusion. Our results show that light grazing significantly increased microbial community diversity, richness, network complexity, and stability compared to heavy grazing. Soil bacterial and fungal communities were predominantly shaped by stochastic processes. As grazing pressure increased, the proportion of deterministic processes in both bacterial and fungal communities also grew. Furthermore, light grazing significantly reduced both ecosystem functions and multifunctionality compared to heavy grazing. We observed that microbial community assembly processes under grazing disturbance can directly or indirectly regulate EMF by shaping community diversity and network structure. The relationship with EMF appears to depend more on diversity than on network structure. Moreover, we emphasize that the complexity and stability of community networks are stronger predictors of EMF changes than diversity alone. In conclusion, the protection and careful management of microbial communities are crucial for enhancing ecosystem resilience and supporting sustainable development.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-35462-0.

Introduction

Interactions among ecosystem functions play a critical role in maintaining the stability of material cycles, energy flows, and information transmission, thereby enhancing the human living environment and quality of life^1,2. Ecosystem multifunctionality (EMF) encompasses key processes and functions essential for ecosystem operation and serves as a comprehensive indicator of multiple functions, including nutrient cycling, water regulation, pollution reduction, climate regulation, and carbon storage³. EMF is widely regarded as a key indicator of ecosystem health and stability^3,4, reflecting the ecosystem’s ability to resist and recover from environmental disturbances^5,6. Consequently, in-depth research into changes in EMF and the underlying mechanisms is essential for optimizing sustainable ecosystem management strategies. This research will contribute to mitigating global climate change, protecting ecological environments, and fostering green social development^7,8.

Plant diversity is widely recognized as a key factor in maintaining variation in EMF, as demonstrated by numerous studies^4,8. Subsequent studies confirm that plant community composition, structure, and the abundance of key species are critical drivers of EMF^5,8. In contrast, soil microbial communities play a significant role in influencing aboveground biodiversity and ecosystem functions^3,9. Their role in sustaining EMF is particularly prominent in arid regions^10–12. Moreover, global data analyses reveal a consistent and significant positive correlation between microbial diversity and multifunctionality, independent of abiotic factors such as climate, soil, and spatial scale. Specifically, the loss of microbial diversity can impair microorganisms’ capacity to maintain multifunctionality^9,13. Current research primarily focuses on the mechanisms by which microbial community co-occurrence network complexity and stability influence EMF. It remains unclear whether different microbial community characteristics—such as diversity, richness, network structure, and assembly processes—differentially drive multifunctionality.

Environmental disturbances can actively promote the aggregation of microbial communities. Microbial communities shaped by environmental disturbances often exhibit enhanced ecological resilience, enabling better adaptation to environmental stress¹⁴. However, excessive environmental disturbances can induce an “elimination effect” in soil microbial communities, leading to the loss of ecological resilience and diminished community resistance¹⁵. Microbial community assembly determines species composition and abundance distribution and is central to understanding community formation and succession mechanisms^16–18. This process is influenced by both deterministic and stochastic factors. Deterministic factors, including both biotic and abiotic influences, shape microbial communities through ecological selection¹⁶. Stochastic factors, such as birth, death, dispersal, and ecological drift, are emphasized for their dominant influence on community assembly. Notably, microbial community assembly directly regulates changes in diversity¹⁶, suggesting that it may also drive changes in EMF^19,20. However, our understanding of the relationship between microbial community assembly and EMF under environmental disturbances is still limited.

Climate change and human activities have led to various environmental disturbances on the central northern slope of the Tianshan Mountains, resulting in increasingly severe grassland degradation^21–25. The inherent vulnerability and complexity of grassland ecosystems further exacerbate this problem. Grazing plays a crucial role in driving the material cycle and energy flow within this ecosystem^26,27. Grazing activities, including foraging, trampling, and excretion, directly affect plant communities, soil structure, and nutrient dynamics²⁸, ultimately driving the plant-soil feedback process^29,30. As a key summer pasture in the region, the mountain meadows of northern Tianshan Mountains provide essential grazing land for livestock during the peak growing season, play an irreplaceable role in maintaining regional ecological security. They are not only core areas for water conservation, soil and water conservation, and biodiversity protection, but also serve as a natural barrier against natural disasters due to their powerful carbon sequestration, oxygen release, and climate regulation functions. However, under the combined pressures of climate change and human activities, the inherent vulnerability of this ecosystem poses a severe risk of degradation. In this context, grazing, as a key process driving the material cycle and energy flow within the system, profoundly impacts meadow health directly through the plant-soil feedback mechanism. Therefore, scientifically determining and managing grazing intensity based on in-depth research into this feedback mechanism has become a core strategy for balancing regional pastoral development and ecological protection, and for ensuring ecological security in northern Xinjiang.

We selected this region for our study, constructed a microbial community co-occurrence network, assessed community assembly, and examined how various microbial community characteristics regulate EMF. Based on the habitat filtering and niche differentiation hypotheses, alongside relevant research^4,9,20, our results indicate that EMF may be regulated by various microbial community characteristics. We hypothesize that increasing grazing intensity may promote cooperation among soil microbial communities, while reducing their complexity and stability. Moreover, under grazing disturbance, community assembly regulates EMF by influencing its diversity and co-occurrence network structure.

Materials and methods

Study area

This study was conducted in pastures located in the southern part of Changji City, within the Xinjiang Uyghur Autonomous Region. The region is situated on the central northern slopes of the Tianshan Mountains, at an altitude ranging from 1800 to 2400 m (Fig. 1). The climate is temperate continental, with an average annual precipitation of 400–500 mm, peaking between May and September, and lacking a fixed frost-free period³¹. All studied grasslands were mountain meadows with mountain chernozem soils, characteristic of the region’s ecological environment. The dominant species in the grassland community include Poa pratensis, Stipa capillata, Carex liparocarpos, and Bromus inermis. Associated species include Alchemilla tianschanica, Potentilla bifurca, Heteropappus altaicus, Myosotis silvatica, Geranium pratense, Polygonum viviparum, Trollius chinensis, and Leontopodium campestre.

Fig. 1.

Experimental site in Xinjiang Uyghur Autonomous Region, China. (a) Map of the study site; (b) Mountain meadows in the grazing experiment; (c) Design of grazing experimental plots. Map was generated using ArcGIS Desktop version 10.8 (Esri, Redlands, CA, USA; available at https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview).

Experimental design and soil sampling

Grazing in the study area typically occurs over 5 months, from May to September. During the remaining months, the grasslands are fenced and protected³¹. Herders confine their livestock to settlements, with seasonal shifts allowing for free grazing. Xinjiang fine-wool sheep are the primary livestock species grazed. We selected six pastures within the study area that exhibited similar topography, slope, grazing history, and grazing patterns to the grazing plots. Each plot covered 0.33 km², and the number of livestock per household was consistently maintained at 300 ± 20 sheep units year-round.

This study quantified grazing intensity differences using metrics such as vegetation cover and grassland utilization, based on site conditions, herders’ grazing practices, and previous research^31–33. As water sources and supplementary feed are located near settlements, livestock are typically confined to these areas. Specifically, we used each herder’s settlement as the reference point and radiated grazing intensity effects outward. Plots situated 500 m from settlements were classified as heavily grazed (HG) plots. The dominant plant species in these plots included Achnatherum inebrians, with grassland cover below 70% and utilization exceeding 90% (Table S1). Plots situated 2000 m from settlements were classified as lightly grazed (LG) plots, with grassland cover above 90% and utilization below 50%. The dominant species in these plots included Carex liparocarpos and Stipa capillata. The exclusion grazing plot (GE) was located in an enclosed grassland 2500 m from a settlement. This area has been enclosed since 2015 and is regularly used for collective grazing by herders. Mowing is performed annually from early to late August. It is important to note that differences between plots are primarily attributed to herders’ practices. Differences in dominant plant species within the plots are more likely attributed to variations in grazing intensity rather than grassland type itself.

Parallel transects were established within the grazing plots (Fig. 1), with seven 1 m × 1 m plots set 50 m apart within each grazing plot. Plant communities in each plot were surveyed and recorded between July and August in 2022 and 2023. After the survey, plants were cut at ground level, placed in labeled envelopes, dried, and weighed to determine aboveground biomass. The resulting weight was used to calculate the aboveground biomass (AG) of the plant community. Soil samples (0–10 cm) were collected three times per plot using soil augers. After collection, seven soil samples from each grazing plot were combined into one composite sample. Roots were collected by removing residual roots and gravel, then placed in fine nylon mesh bags. The attached soil was removed, dried, and weighed to calculate the belowground biomass (BG). Finally, the sieved soil was divided into two equal aliquots: one for physical and chemical property analysis and the other stored at − 80 °C for microbial community sequencing.

Soil properties analysis and high-throughput sequencing

Soil physical and chemical indices measured included soil organic carbon (SOC, g kg⁻¹), total nitrogen (TN, g kg⁻¹), total phosphorus (TP, g kg⁻¹), total potassium (TK, g kg⁻¹), available nitrogen (AN, mg kg⁻¹), available phosphorus (AP, mg kg⁻¹), available potassium (AK, mg kg⁻¹), soil pH, bulk density (BD, g cm⁻³), and moisture (SM, %). Detailed measurement methods are provided in Supplementary Table S2³⁴.

Total DNA from soil samples was extracted using the MP-Fast DNA kit (Omega Bio-Tek, Norcross, GA, U.S.). For bacterial detection, primers 338F (5′-ACTCCTACGGGAGGCAGCAG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′) were used. For fungal detection, primers ITS1F (5′-CTTGGTCATTTAGAGGAAGTAA-3′) and ITS2R (5′-GCTGCGTTCATCGATGC-3′) were used. PCR amplification was performed with an initial pre-denaturation step at 95 °C for 3 min, followed by 27 cycles. Each cycle included DNA denaturation at 95 °C, primer annealing at 55 °C, and DNA chain extension at 72 °C, with a final extension at 72 °C for 10 min to ensure product integrity. Purified PCR products were sequenced on the Illumina MiSeq platform (Meiji Inc, Shanghai, China). Using FLASH software (version 1.2.7, https://ccb.jhu.edu/software/FLASH/) and Trimmomatic (version 0.39, http://www.usadellab.org/cms/index.php?page=trimmomatic), raw MiSeq data were assembled to obtain valid sequences. Sequences were annotated using the SILVA database (release 128) and the UNITE database (release 8.0).

Microbial community diversity analysis

We characterized the foundational structure of the microbial communities using standard diversity metrics. Alpha diversity analysis was used to assess community species richness and evenness. To this end, we calculated several complementary indices. The Chao1 and Ace indices were used specifically to estimate species richness (i.e., the inferred total number of potential species), while the Shannon and Simpson indices were employed to comprehensively assess community richness and evenness. All calculations were based on data leveled to a uniform depth (33,042 sequences) and performed using the vegan package (version 2.6.10, https://github.com/vegandevs/vegan) to ensure comparability. To quantify and visualize compositional differences between grazing treatments, β-diversity (BDtotal, or dissimilarity) was quantified using the betapart (version 1.6, https://cran.r-project.org/package=betapart) and adespatia (version 0.3.28, https://github.com/adeverse/adespatial) packages³⁵, further decomposed into replacement diversity (Repl, turnover) and species richness difference diversity (RichDif, nestedness). That is, BDtotal = RichDif + Repl, and Similarity = 1 − BDtotal. Nonmetric multidimensional scaling (NMDS) analysis was used as an ordination method to examine differences in microbial communities between plots, with significance assessed using ADONIS and ANOSIM.

Network complexity and stability analysis

To infer potential ecological interactions and assess community stability, we constructed species co-occurrence networks. Prior to constructing the network, microorganisms were screened based on OTUs with relative abundance below 0.01% and a frequency of fewer than 5 occurrences per treatment. Weakly correlated, insignificant edges, and spurious associations that may be caused by species autocorrelation were excluded based on Spearman correlation coefficient (< 0.7) and P value (> 0.01) to obtain robust species co-occurrence patterns³⁶. Microbial community co-occurrence networks were constructed using the R software (version 4.3.2, https://www.r-project.org/) with the igraph package (version 2.1.4, https://r.igraph.org/). Topological properties of bacterial and fungal community co-occurrence networks were calculated using the ggClusterNet (version 0.1.0, https://github.com/taowenmicro/ggClusterNet) package, including the number of nodes, edges, modularity, average degree, clustering coefficient, average path length, network diameter, and graph density. A composite index reflecting network complexity was calculated by averaging the standardized scores of bacterial and fungal community topological properties.

Multiple metrics were used to evaluate the stability index based on the ggClusterNet, phyloseq (version 1.50.0, 10.1371/journal.pone.0061217), igraph, and sna (version 2.8, https://statnet.org) packages, including network robustness³⁶, which is often used to characterize natural connectivity, network vulnerability, and cohesion, which is often used to characterize positive and negative interactions within a community. Positive and negative cohesion were weighted based on significant associations, with cohesion calculated as the quotient of their absolute values. Finally, similar to network complexity, these metrics were standardized and averaged to derive the stability index. Subsequently, One-way analysis of variance (ANOVA) was performed using the vegan package to compare changes in soil microbial community diversity, network complexity, and stability across different grazing plots.

Assembly process analysis

We quantitatively evaluated the microbial community assembly process in various grazing plots³⁰. Microorganisms were classified into different bins based on phylogenetic distance. The β-net relatedness index (βNRI) and Raup-Crick metrics (RCbray) of each bin were analyzed using the null model to identify different assembly processes based on the iCAMP (version 1.5.12, https://github.com/DaliangNing/iCAMP1) and ape (version 5.8.1, https://github.com/emmanuelparadis/ape) packages^17,20,30. Specifically, βNRI < − 1.96 indicates homogeneous selection, βNRI > 1.96 indicates heterogeneous selection, |βNRI|≤ 1.96 and RCbray < − 0.95 indicates homogeneous diffusion, |βNRI|≤ 1.96 and RCbray > 0.95 indicates diffusion limitation, and |βNTI|≤ 1.96 and |RCbray|≤ 0.95 indicates drift^16,30. Finally, we integrated and weighted the relative abundance of each bin to further assess the importance of each assembly process in the community. A neutral community model (NCM) was constructed using the Stats4 (version 4.4.2, https://cran.r-project.org/package=spaa) and Hmisc (version 5.2.3, https://hbiostat.org/R/Hmisc/) packages to evaluate the relative importance of stochastic processes in assembly. Additionally, Levin’s niche breadth index was calculated using the spaa (version 0.2.5, https://github.com/helixcn/spaa) package to examine the contributions of deterministic and stochastic processes to community assembly under grazing disturbance. To link this assembly process to potential drivers, we performed the Mantel test using the linkET (version 0.0.7.4, https://github.com/Hy4m/linkET) package to assess the driving effect of soil physicochemical properties on the community assembly process, and random forest analysis was used to identify microbial community features (diversity and network attributes) driving the variation of the assembly process (βNTI) using the randomForest (version 4.7.1.2, https://www.stat.berkeley.edu/~breiman/RandomForests/) and rfPermute (version 2.5.2, https://github.com/EricArcher/rfPermute) packages.

Ecosystem multifunctionality

To quantify the integrated ecosystem response to grazing, we evaluated EMF. This study focused on evaluating EMF in relation to belowground nutrient cycling, transformation, and closely associated aboveground grassland productivity. Nine indicators were selected to evaluate EMF: SOC, TN, TP, TK, AN, AP, AK, AG, and BG (Table S2). EMF was evaluated using the average method, and Z-score standardization was performed on all nine indicators to ensure uniform magnitude^3,5,7. The mean of all standardized variables was calculated to represent EMF. A weighted average method based on cluster analysis was used to validate the accuracy of the mean method. One-way analysis of variance (ANOVA) was performed using the vegan package to compare changes in ecosystem function and multifunctionality across different grazing plots.

Linking microbial communities to ecosystem function

To determine the association between EMF and microbial community under grazing disturbance, we employed a two-step analysis approach. First, random forest regression was used to identify the most important microbial community attributes predicting EMF (e.g., diversity indices, network topology attributes, and assembly processes). Subsequently, linear models were applied to quantify the significant relationships between these key predictors and EMF based on the lme4 (version 1.1.36, https://github.com/lme4/lme4/) package. To examine the pathway by which grazing influences EMF through the microbial community, we constructed a structural equation model (SEM) using the piecewiseSEM (version 2.3.0.1, https://github.com/jslefche/) package. All variables were standardized before analysis, and model fit was assessed using the chi-square test (χ²), root mean square error of approximation (RMSEA), comparison fit index (CFI), and P-value.

Results

Diversity of fungal and bacterial communities

The α diversity of bacterial and fungal communities was highest in the LG plot, followed by the GE plot, and lowest in the HG plot (Fig. 2). The α diversity of bacterial communities in the GE and LG plots was 108.49% and 191.21% higher, respectively (P < 0.05, Fig. 2a), and the α diversity of fungal communities was 91.67% and 208.33% higher, respectively (P < 0.05, Fig. 2b), than in the HG plot. Similarly, bacterial richness in the LG plot was 205.32% higher (P < 0.05, Fig. 2c), and fungal richness was 199.74% higher (P < 0.05, Fig. 2d), than in the HG plot.

Fig. 2.

Diversity of bacterial and fungal communities in different grazing plots. (a, b) α diversity of bacterial and fungal communities. (c, d) richness of bacterial and fungal communities. (e, h) β diversity components of bacterial and fungal communities. (g, j) β diversity of bacterial and fungal communities. (g, j) non-metric multidimensional scaling analysis of bacterial and fungal communities. (*) indicate significant differences between grazing intensities (P < 0.05, Duncan’s multiple range test). Grazing categories are defined as follows: GE, Grazing exclusion; HG, Heavy grazing; LG, Light grazing. Analysis of variance (ADONIS) and analysis of similarity (ANOSIM) were used to examine differences in plant communities between different grazing intensities.

We analyzed the β diversity (BDtotal, or dissimilarity) of bacterial and fungal communities, dividing it into replacement diversity (Repl) and species richness difference diversity (RichDif). BDtotal was dominated by species replacement (bacteria: 0.9057/0.9278, 97.62%; fungi: 0.8155/0.8435, 96.68%) (Fig. 2e, h). Additionally, the BDtotal of both fungal and bacterial communities exhibited a clear trend of increasing and then decreasing (Fig. 2g, j) as grazing intensity increased. Specifically, the BDtotal in the LG plot was the highest (Fig. 2b), significantly greater than in the GE and HG plots (P < 0.05). Furthermore, fungal and bacterial communities in the different grazing plots exhibited significant differentiation along the ordination axes, as shown by NMDS results (Fig. 2f, i). We found that changes in grazing intensity were the primary factor driving alterations in fungal and bacterial community structure, as confirmed by ANOSIM analysis based on the Bray–Curtis distance (P < 0.05).

Network structure of bacterial and fungal communities

Species in the bacterial and fungal community co-occurrence networks at the LG plot were more tightly connected than those at the GE and HG plots (Fig. 3). The number of nodes and edges in the bacterial and fungal community co-occurrence networks at the LG plot was higher than that at the GE and HG plots. The average degree of the bacterial community co-occurrence network at the LG plot was 8.72–44.09% higher than that at the HG plot (Table S4), and the average degree of the fungal community network was 42.56–95.99% higher (Table S5). Similarly, the robustness, positive cohesion, and negative cohesion of the bacterial and fungal community networks at the LG plot were significantly higher than those at the GE and HG plots (P < 0.05, Fig. S1a–c). However, vulnerability at the HG plot was significantly higher than at the GE and LG plots (P < 0.05, Fig. S1d).

Fig. 3.

Network structures of fungal and bacterial communities differed in grazed plots. (a–f) Co-occurrence networks of bacterial and fungal communities. Nodes of different colors represent different modules. (g, h) Network complexity and stability of bacterial communities. (i, j) Network complexity and stability of fungal communities. Grazing categories are defined as follows: GE, Grazing exclusion; HG, Heavy grazing; LG, Light grazing. Asterisks (*) indicate significant differences between grazing intensities (P < 0.05, Duncan’s multiple range test).

In particular, the network complexity and stability of bacterial and fungal communities at the LG plot were significantly higher than those at the HG plot (P < 0.05). Correlation analysis revealed a significant relationship between network complexity and stability indicators for both bacterial and fungal communities (Fig. S2a). Linear regression results also showed a highly significant positive correlation between network complexity and stability (P < 0.001, Fig. S2b).

Assembly processes of bacterial and fungal communities

Stochastic processes dominated the assembly of bacterial and fungal communities across different grazing plots, with grazing altering their relative proportions (Fig. 4). Quantification of βNTI and RCbray revealed that drift in bacterial and fungal community assembly increased with grazing intensity. As grazing intensity increased, dispersal limitation became the dominant factor in bacterial community assembly, surpassing the influence of drift and homogeneous dispersal. In fungal communities, drift accounted for more than 50% of the stochastic processes in all grazed plots. Homogeneous selection consistently dominated the deterministic processes of bacterial and fungal communities, initially increasing and then decreasing with grazing intensity. The niche breadth of bacterial and fungal communities in the LG plots was greater than that in the GE and HG plots, as indicated by the habitat niche breadth index (Fig. S3). Results from the NCM indicated that bacterial communities fit the neutral community model more closely. As grazing intensity increased, the goodness of fit decreased for both bacterial communities (GE: 38.6%, LG: 32.9%, HG: 26.3%) and fungal communities (GE: 22.5%, LG: 14.8%, HG: 11.1%) (Fig. S4).

Fig. 4.

Assembly processes of bacterial and fungal communities in different grazing plots. (a, d) Bacterial and fungal community assembly processes during grazing exclusion. (b, e) Bacterial and fungal community assembly processes during light grazing. (c, f) Bacterial and fungal community assembly processes during heavy grazing. According to the values of βNTI and RCbray. Including homogeneous selection (βNTI <  − 1.96), heterogeneous selection (βNTI > 1.96), dispersal limitation (|βNTI|< 1.96 and RCbray > 0.95), homogenous dispersal (|βNTI|< 1.96 and RCbray <  − 0.95), and drift (|βNTI|< 1.96 and |RCbray|< 0.95).

Soil factors were significantly associated with the βNTI of both bacterial and fungal communities, as confirmed by the Mantel test (Fig. S5a). Changes in fungal communities were more strongly influenced by soil factors than changes in bacterial communities. Random forest analysis showed that while α diversity, richness, complexity, and stability influenced βNTI, β diversity was the greatest contributor to bacterial and fungal community assembly (Fig. S5b, c).

Ecosystem multifunctionality and driving force

Ecosystem functions across the different grazing plots exhibited consistent trends, initially increasing and then decreasing with grazing intensity (Fig. S6a). EMF values were lowest in the HG plot, followed by the GE plot, and highest in the LG plot (P < 0.05, Fig. 5a). We assigned different weights to each evaluation metric through hierarchical clustering (Fig. S6b) and re-estimated the EMF using a weighted average method (Fig. S6c). General linear regression showed a high correlation between the weighted and average EMF (R² = 0.94, P < 0.001; Fig. S4d).

Fig. 5.

Relationship between bacterial and fungal communities and ecosystem multifunctionality in different grazing plots. (a) Ecosystem multifunctionality in different grazing plots. (b) Associations of different characteristics of bacterial communities with ecosystem multifunctionality. (c) Associations of different characteristics of fungal communities with ecosystem multifunctionality. (d, f) Regulation of ecosystem multifunctionality by bacterial and fungal community assembly processes under grazing disturbance. (e, g) Effect size based on partial least squares path model. Grazing categories are defined as follows: GE, Grazing exclusion; HG, Heavy grazing; LG, Light grazing. Asterisks (*) indicate significant differences between grazing intensities (P < 0.05, Duncan’s multiple range test).

All ecosystem functions were positively correlated with bacterial and fungal community α diversity, β diversity, richness, complexity, and stability, as indicated by the linear regression results (Fig. S7). Similarly, network complexity (Fig. 5b) and stability of bacterial and fungal communities exhibited stronger predictive power, as demonstrated by the EMF fitting results (Fig. 5b, c). Furthermore, grazing disturbance not only directly reduced EMF but also modulated it by altering microbial community diversity and co-occurrence network structure (Fig. 5d, f). Specifically, although community assembly processes may not directly regulate EMF, they can maintain grassland EMF by modulating microbial community diversity. Furthermore, community diversity explained less variation in EMF than network structure. β diversity contributed more to EMF than α diversity, and network complexity contributed more than stability, a trend consistent across both bacterial and fungal communities.

Discussion

Grazing affects network structure

We observed that positive correlations between bacterial and fungal community networks intensified with grazing pressure (Fig. 3a–f). This suggests enhanced cooperation among microbial species, supporting our first hypothesis. According to the pressure gradient hypothesis, in areas with limited resources or severe disturbances, microbial communities respond to environmental stress by strengthening cooperation³⁷. In contrast, in areas with abundant resources and stable environments, cooperation among microbial species may decrease, while competition increases³⁸. Furthermore, under moderate grazing conditions, favorable soil nutrients and water availability promote high diversity and richness in microbial communities. This life history strategy maintains a balance between cooperation and competition among species, enhancing the community’s resistance and resilience to external disturbances^37,39. In contrast, resource scarcity and harsh environments under heavy grazing inevitably lead to increased cooperation and higher functional redundancy in microbial communities⁴⁰.

We found that the network complexity and stability of bacterial and fungal communities followed a unimodal curve, initially increasing and then decreasing with increasing grazing intensity (Figs. S1 and 3g–j), consistent with previous studies⁴¹. The mechanisms driving the grazing network structure can be explained from multiple perspectives, including soil structure, chemical fertility, nutrient uptake, and plant communities. First, moderate grazing improves soil physical structure, promoting the formation of more stable aggregates²⁶. This optimizes the soil microenvironment, enhancing nutrient circulation and migration, thus providing a solid foundation for microbial survival and reproduction^5,31,42. Second, under moderate grazing, abundant root and leaf litter in the soil creates a fertile microenvironment, supporting the growth and development of microbial communities²⁶. Furthermore, moderate grazing stimulates grassland plant communities, increasing intraspecific variation and enhancing the coordination of environmental resources⁴³. This improves adaptability to grazing disturbance by expanding ecological niches both spatially and temporally⁴⁴. This shift affects nutrient availability for microbes and promotes the emergence of microbial species with diverse metabolic strategies⁴⁵, driving changes in microbial community composition.

Furthermore, we found a significant positive correlation between network complexity and stability across different grazing plots (Fig. S2), indicating that increased complexity often results in increased stability³⁹. This finding confirms the intrinsic link between complexity and stability in ecosystems and aligns with MacArthur’s⁴⁶ view in macroecology that “stability in natural ecosystems stems from complexity”. However, this result contradicts May’s⁴⁷ view that high complexity may hinder ecosystem stability. In summary, research on the complexity-stability trade-off remains controversial, hindering our understanding of the relationship between microbial community complexity and stability. Future research should address this issue through in-depth exploration using replicated experiments across multiple regions, ecosystem types, and methodologies.

Grazing affects community assembly

Our quantification of βTNI revealed that stochastic processes dominated bacterial and fungal community assembly across grazed plots (Fig. 4), indicating that grazing disturbance acts as a selective force, influencing microbial community assembly. Grazing disturbance increased the proportion of deterministic processes in both bacterial and fungal communities. Our results align with previous studies demonstrating that disturbed environments favor the deterministic assembly of soil microbial communities⁴⁸. Changes in βTNI can be explained from the perspective of niche breadth, with the index exhibiting a trend of initially increasing and then decreasing with increasing grazing intensity (Fig. S3). Generally, in the absence of environmental stress and with abundant resources, microbial communities selectively utilize resources⁴⁹. The low level of interactions between species suggests that microbial communities in this environment exhibit fewer deterministic processes, with a relative increase in stochastic effects. Conversely, under specific environmental stresses, soil microbial communities utilize resources more efficiently and exhibit enhanced cooperative and competitive behaviors^30,40. As a result, the proportion of deterministic processes increases with grazing disturbance intensity.

Furthermore, we found that the fit and migration dispersal of bacterial and fungal communities decreased with increasing grazing intensity, indicating that grazing significantly disrupts the dispersal capacity of microbial communities (Fig. S4). More importantly, we found that the fit and migration dispersal of fungal communities across grazed plots were lower than those of bacterial communities, as indicated by NCM models. This finding was corroborated by Mantel test and random forest models (Fig. S5). This suggests that environmental filtering plays a key role in the assembly of fungal communities in mountain meadows on the central northern slope of the Tianshan Mountains^50,51. This may be explained by the size plasticity hypothesis. Bacterial communities are smaller than fungal communities, making them less susceptible to environmental filtering⁵². The niche breadth index of bacteria is higher than that of fungi (Fig. S3). A broader niche enables bacteria to exhibit a higher potential migration rate⁵³. Additionally, differences in nutrient utilization between fungi and bacteria further elucidate the distinct assembly strategies of the two in varying environmental contexts⁵⁴. In summary, this study confirms that environmental disturbance plays a critical role in microbial community assembly and suggests that fungi can serve as reliable indicators of the impact of grazing activities.

Grazing affects ecosystem multifunctionality

We observed that both ecosystem function and EMF initially increased and then decreased with grazing intensity, peaking under light grazing. This is consistent with the moderate disturbance and grazing optimization hypotheses, which propose that moderate grazing stimulates compensatory growth in plant communities⁴⁴, enhancing their adaptability and resistance to grazing⁵⁵, thereby increasing grassland productivity. Additionally, improved plant communities under moderate grazing feedback into the soil and microbiome through their roots. This process enhances soil structure and fertility while improving the stability and complexity of the microbiome^39,56. Notably, the grazing and trampling behavior of livestock actively promote this feedback mechanism^44,45. Ultimately, when the plant community, soil microbial community, and soil ecosystem improve under moderate grazing conditions, the grassland ecosystem enters a virtuous cycle, and various grassland ecological functions and multifunctionality steadily improve.

Linear fitting results revealed significant positive correlations between alpha diversity, beta diversity, richness, network complexity, stability of bacterial and fungal communities, and ecosystem function. In particular, network complexity and stability explained more variation in ecosystem function and EMF than diversity and richness, indicating that they play a critical role in maintaining EMF. This finding was confirmed by PLS-PM (Fig. 5). Network complexity and stability are not only products of microbial interactions but also key drivers of nutrient cycling and energy flow in ecosystems⁵⁷. High levels of network complexity and stability imply more frequent and intense cooperation or synergy among microbial communities. This enhances resource utilization efficiency and accelerates metabolic rates in biochemical cycles^58,59, ultimately contributing to the stable maintenance of high EMF levels. Interestingly, across all linear fitting results, microbial communities explained more variation in EMF than ecosystem function, especially in the results for complexity and stability. This suggests that the maintenance and driving of ecosystem functions by microbial communities are not merely the sum of coexisting individuals but rely on interactions among microbial groups to sustain EMF. Furthermore, when multiple ecosystem functions coexist, their relationships with fungal and bacterial communities may be more profound. In other words, incorporating more ecosystem functions can more comprehensively reflect the diverse nature and complex structure of ecosystems^3,60.

To test our hypothesis, we used PLS-PM and found that grazing drives EMF by altering the diversity and network structure of microbial communities. This result aligns with previous studies^9,11. According to the niche complementarity hypothesis, grazing promotes soil microbial community adaptation to environmental stresses through resource diversification and microhabitat selection⁶¹, which subsequently alters ecosystem function, resilience, and stability. Interestingly, this finding more robustly explained EMF in fungal communities. Notably, the direct effect of community assembly on EMF was not significant, and the relationship between the two was more dependent on community diversity, a finding also evident in fungal communities. This difference is reflected in several aspects: first, fungi reproduce slowly, resulting in the slow and incomplete decomposition of complex organic matter. In contrast, bacteria reproduce more rapidly and excel at breaking down small, simple organic matter, including products of fungal decomposition⁴⁹. Furthermore, the community assembly processes of the two groups differ. Variations in deterministic and stochastic processes reflect the strength of species competition and the influence of non-stochastic factors^16,20. As grazing pressure increases, external resources available to microbial communities decrease, resulting in more interactions between microbial groups (Fig. 2). This is particularly true for fungal communities, significantly strengthening their association with EMF.

Previous research and our findings indicate that grazing disturbance plays a key role in grassland ecosystems. Specifically, grazing affects soil microbial communities by altering plant and soil systems^5,28,29,62, thereby regulating and driving EMF. Our findings align with previous studies^10,12, demonstrating that microbial assembly processes actively regulate their diversity and network structure, thereby altering EMF. However, our results differ from previous studies, likely due to variations in soil properties among the study areas. Additionally, researchers have differed in their criteria for defining changes in grazing intensity⁶³. Furthermore, the varying responses of dominant plant species to grazing may also contribute to these discrepancies^31,64. Furthermore, although microbial communities are essential decomposers in ecosystems, the bottom-up trophic effects they exert on multifunctionality remain poorly understood. Therefore, future research should investigate the mechanisms by which diversity at multiple trophic levels drives multifunctionality and explore these mechanisms across various ecological scales.

Conclusion

In summary, our study revealed that the diversity, richness, network complexity, and stability of bacterial and fungal communities were significantly lower under heavy grazing compared to light grazing. Similarly, ecosystem function and multifunctionality were significantly reduced in heavily grazed plots. Grazing drives bacterial and fungal community assembly toward deterministic processes, directly influencing the diversity and network complexity of microbial communities. Additionally, grazing drives ecosystem multifunctionality by directly or indirectly altering the diversity and network structure of microbial communities. The association between community assembly processes and ecosystem multifunctionality is primarily driven by diversity construction under grazing disturbance. We emphasize that network complexity and stability play a more critical regulatory role than diversity in maintaining ecosystem multifunctionality, which is essential for sustaining high levels of multifunctionality. These findings highlight the importance of protecting and optimizing microbial communities to maintain grassland ecosystem stability and provide new perspectives on the mechanisms by which microbial communities drive multifunctionality under climate change.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

K.W. Jiang, Y.F. Wang, and Q.Q. Zhang conceived the study; K.W. Jiang, C. Lv, and H. Li performed the field experiments; H. Li and K.W. Jiang performed the data analysis. K.W. Jiang, Q.Q. Zhang, R. Tuerxunnayi, and P. J. Yu wrote the first draft of the manuscript. Y.F. Wang, C. Lv, P. J. Yu, and K.W. Jiang edited the manuscript. All authors commented on drafts of the manuscript and approved the final version.

Funding

This work was supported by the National Natural Science Foundation of China (32260311 and 31960338) and Xinjiang Uygur Autonomous Region Natural Science Foundation (2022D01A189).

Data availability

All raw sequencing data have been deposited in the NCBI GenBank Short Read Archive under accession number PRJNA1322087. The datasets generated and/or analyzed during this study are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.