This study was conducted at the Agro-pastoral Ecotone Experimental Station, Grassland Research Institute of the Chinese Academy of Agricultural Sciences (40°35′N, 111°46′E) in Hohhot, Inner Mongolia, China. The study site has a mesothermal continental monsoon climate, with a mean annual temperature of 6.7°C and mean annual precipitation of 400 mm. Rainfall occurs mainly in July and August. The soil type is predominantly tidal soil.

Land use was classified into five types: shrublands dominated by Caragana korshinskii Kom, woodlands dominated by PopulusL, artificially managed grasslands dominated by Leymus chinensis, natural grasslands dominated by Stipa capillata, and agricultural fields consisting of maize (Zea mays) cropland. Of these, only artificial grasslands and farmlands involved water and fertilizer addition and tillage management (see Supplementary Table S1 for more details). On July 25, 2023, we established ten sample plots (1 m × 1 m) as biological replicates of each land use type, with a spacing of ≥10 m. The litter and humus layers were removed, and the upper soil layer (0–20 cm) was sampled using a soil auger at three randomly selected locations per plot; these samples were mixed to obtain a single composite sample. Thus, a total of ten composite samples were obtained per land use type. We removed visible stones, animal and plant residues, roots, and other substances from each sample, and then sieved the sample through a 2-mm mesh. The sieved soil samples were sealed in sterile plastic bags, placed in an ice box, and transported to the laboratory. Each soil sample was divided into two parts, for soil physicochemical property analyses and DNA extraction, respectively.

Soil moisture content was measured gravimetrically by drying at 105°C until a constant weight was achieved. Soil pH and electrical conductivity were measured in a soil–water slurry (1:2.5, w/v) as described previously (Widdig et al., 2020). Soil organic C (SOC) and total N (TN) were quantified using the dichromate oxidation and Kjeldahl digestion methods, respectively (Qiu et al., 2018). Alkaline N (AN) decomposition was measured colorimetrically using ultraviolet spectrophotometry and the indophenol blue method (Hu et al., 2021). Soil available P (AP) and available K (AK) were extracted with 0.5 M sodium bicarbonate and 1 M ammonium acetate, and measured by molybdenum blue spectrophotometry and flame photometry, respectively (Cui et al., 2023). Soil microbial biomass C (MBC) and microbial biomass N (MBN) were determined using the chloroform fumigation–extraction method.

Soil genomic DNA was extracted using the FastDNA Spin Kit for Soil (MP Biomedicals, Irvine, CA, United States), following the manufacturer’s instructions. All DNA in 0.5 g of soil was extracted according to the recommended amount for the instrument. DNA quality was evaluated via 1% (w/v) agarose gel electrophoresis. Soil bacterial and fungal community compositions were determined by high-throughput sequencing on a cloud-based platform at Majorbio (Shanghai, China). The V3–V4 hypervariable regions of bacterial 16S rRNA genes were amplified using the 338F (5′- ACT CCT ACG GGA GGC AGC AG-3′) and 806R (5′-GGA CTA CHV GGG TWT CTA AT-3′) primer set (Xu et al., 2016). In fungi, the internal transcribed spacer 2 (ITS2) region was amplified using the ITS1F (5′-CTT GGT CAT TTA GAG GAA GTA A-3′) and ITS2R (5′-GCT GCG TTC TTC ATC GAT GC-3′) primer set (Karlsson et al., 2014). Polymerase chain reaction (PCR) amplification was performed using a GeneAmp PCR system (Model 9,700; Thermo Fisher Scientific, Waltham, MA, USA).

The raw data sequences were processed and analyzed using QIIME2 (Bolyen et al., 2019) based on the workflow provided at https://qiime2.org. Briefly, to obtain the amplicon sequence variant (ASV) table, quality control of the raw sequencing data was performed using the DADA2 (Callahan et al., 2016) plug-in and clustered based on 100% shared identity. The taxonomy of bacterial and fungal phylotypes was identified using the silva138/16s_bacteria (Quast et al., 2013) and unite8.0/its_fungi (Nilsson et al., 2019) databases, respectively. Finally, we obtained 3,318,073 bacterial sequences and 4,373,406 fungal sequences, which were classified into 40,525 and 7,910 distinct ASVs in bacteria and fungi, respectively.

We analyzed α-diversity parameters such as the Chao1 and Shannon indices using the mothur ver. 1.30 software. Differences in mean α-diversity values between two independent groups were analyzed using the Wilcoxon rank-sum test with the stats package in the R ver. 4.3.2 software (R Core Team, Vienna, Austria). The Bray–Curtis dissimilarity index was used to assess soil bacterial and fungal beta diversity levels, and the results were visualized through principal coordinates analysis (PCoA) using the vegan package in R. Analysis of similarities was performed to quantitatively estimate community similarities among sample groups (Yan et al., 2020). Stacked bar charts illustrating the relative abundance of microorganisms in each group were drawn using Python ver. 2.7. The relative abundances of bacteria and fungi in soil samples from each land use type were compared using one-way analysis of variance, followed by Tukey’s multiple comparison test to detect significant differences. Spearman’s correlation analysis was conducted using the psych package in R, and a heatmap of the results was plotted using Python ver. 3.7. Variance partitioning analysis was performed and visualized using the vegan package in R. Soil environmental factors were categorized as basic physicochemical properties (pH, soil moisture, and electrical conductivity), soil nutrients (SOC, TN, AP, and AK content and AN decomposition) and microorganism nutrients (MBC and MBN).

A soil microbial community co-occurrence network was constructed based on ASV levels using the Spearman correlation method (r ≥ 0.7, false discovery rate-adjusted p < 0.05), implemented in the igraph R package (Zhou et al., 2011; Yuan et al., 2021). To minimize potential spurious correlations, we selected only ASVs occurring in ≥30% of all samples and accounting for >0.01% of the total for the correlation calculation. The Gephi ver. 0.10.1 software was used for network visualization. To interpret the effects of land use types on the complexity of ecological networks, we extracted network topological characteristics, including node number, edge number, average degree, average weighted degree, network diameter, network density, modularity, average clustering coefficient, and average path length, for each soil sample using the “subgraph” function in the igraph R package (Jiao et al., 2022).

We calculated changes in bacterial and fungal α-diversity for different land use types (Figures 1A,B). The results showed that bacterial richness (Chao1 index) and diversity (Shannon index) were highest in shrublands and lowest in natural grasslands. Woodlands showed the highest fungal abundance, and there were no significant differences in fungal diversity among groups.

PCoA analyses of β-diversity based on the Bray–Curtis difference matrix showed clear separation of bacterial (R = 0.4783, p = 0.001) and fungal (R = 0.5379, p = 0.001) community structures across land use types (Figures 2A,B).

Figure 3 shows bacterial and fungal community compositions at the phylum level under different land use types. Actinobacteriota had the highest relative bacterial abundance in natural grasslands, at 29%. In the other four land use types, Proteobacteria had the highest relative bacterial abundance, at 30–35%. Ascomycota had highest relative fungal abundance across all five land use types. The relative abundances of Proteobacteria and Bacteroidota were significantly lower in natural grasslands than in all other land use types. Actinobacteria had significantly higher relative abundance in natural grasslands than in shrublands, woodlands, and farmlands, and Chloroflexi had significantly higher relative abundance in natural grasslands than in woodlands. Among fungi, Ascomycota had significantly lower relative abundance in woodlands than in all other land use types, and the relative abundance of Basidomycota was significantly higher in woodlands than in natural grasslands, shrublands, and farmlands.

We constructed microbial community networks for each of the five land use types and found that their structures varied greatly (Figure 4; Supplementary Table S2). Artificial grasslands and farmlands had simpler network structures, with fewer nodes and edges, and lower average degree, average weighted degree, mesh diameter, and average path length; by contrast, these metrics were higher and the network structures were more complex in natural grasslands, shrublands, and woodlands. There were no clear trends among land use types for indicators such as network density, degree of modularity, and average degree of clustering.

At the phylum level, Proteobacteria and Chloroflexi, which represented a large proportion of bacteria in our soil samples, were significantly correlated with various environmental factors, showing positive and negative correlations with the majority of environmental factors, respectively (Figure 5A). Gemmatimonadota, Methylomirabilota, Dadabacteria, and candidate phyla GAL15 and MBNT15 were also strongly (mainly negatively) correlated with environmental factors.

However, at the fungal genus level, AP and AK were positively or negatively correlated with most taxa (Figure 5B). We attempted to explain these differences in bacterial and fungal community structure according to environmental factors by dividing these into three categories: basic physicochemical properties (pH, soil moisture, and electrical conductivity), soil nutrients (SOC, TN, AP and AK content, and AN decomposition), and microorganism nutrients (MBC and MBN), but the results were not conclusive, with residuals of 84.73 and 93.58% for bacterial and fungal communities, respectively.

We detected significant differences in bacterial diversity and abundance among the five land use types, which may have been caused by differences in the dominant plants and degree of anthropogenic interference among land use types (Szoboszlay et al., 2017; Nkuekam et al., 2018). Bacterial richness and diversity index values were highest in shrublands, perhaps because shrubland vegetation has higher biomass, water status, and soil permeability, as well as denser root systems than the other land use types examined in this study (Michelsen et al., 1996; Guanghua et al., 2006). In addition, shrubland is also the least anthropogenic land use among the five land use types (no ploughing, watering, grazing, recreation, etc.), which may be one of the reasons for its higher diversity (Tomazelli et al., 2023). Fungal richness was also highest in woodlands, probably because tall trees have strong xylem root systems that can act as hosts for fungi and can provide them with available nutrients (Urbanova et al., 2015; de Vries et al., 2018). The α-diversity indices of both bacteria and fungi were lower in natural grasslands, presumably due to poorer soil conditions and lower moisture content (Supplementary Table S3), and because herbaceous root systems are less able to retain water (Torsvik and Ovreås, 2002; Deng, 2012). The PCoA results showed that soil bacterial and fungal β-diversity values were highly variable across land use types (Figures 2A,B; p = 0.001). We hypothesise that this is the result of large differences in anthropogenic activities under different land-use types, and that actions such as grazing (Wang and Tang, 2019), fertilizing (Zhong and Cai, 2004), and tilling (Fan et al., 2010) are all important measures to modify heterogeneity between microbiomes.

The phylum Actinobacteria had the highest bacterial relative abundance in natural grasslands, whereas Proteobacteria had the highest bacterial relative abundance in all other land use types. This result may be explained by the co-trophic hypothesis (Fierer et al., 2012). Actinomycetes acquire water and limited nutrients through hypha modification, which allows them to adapt to harsh conditions such as drought (Xu et al., 2021). Among the land use types examined in this study, natural grassland had the lowest soil water content and therefore the highest relative abundance of actinomycetes due to the shallowest root system and lowest height of the vegetation, and therefore poor soil water fixation and high evapotranspiration. Many members of the Proteobacteria are N-fixing; these bacteria can increase nutrient use efficiency by dissolving phosphates, fixing N, and degrading residues (Yang et al., 2023). Compared to natural grasslands, the other land use types exhibited higher N demands (Cornell et al., 2023), resulting in a significant increase in the relative abundance of Proteobacteria. In summary, differences among land use types within the agro-pastoral ecotone of northern China led to differences in microbial community composition and diversity.

We detected fewer nodes and edges and a lower average degree of network structure in artificial grasslands and farmlands, indicative of a simpler network structure (Lan et al., 2022). However, the mean path length and network diameter were also smaller in these land use types, suggesting that microorganisms in these networks were more closely connected to each other (Liu L. X. et al., 2023). In contrast to our results, a recent Brazilian study found more nodes and edges in the network structures of more intensively managed rangelands than in natural grasslands (Tomazelli et al., 2023). This discrepancy may be largely attributed to differences in climatic conditions between Brazil and the Inner Mongolian Plateau (Goss-Souza et al., 2017). Microbial network structural complexity is not only reflected in the numbers of nodes and edges but also closely related to the network diameter and average path length (Lan et al., 2022). However, we also found smaller network diameters and average path lengths in both artificial grasslands and farmlands, suggesting that microorganisms are more tightly connected and that materials and energy are less lost in the transfer process, making their efficient transfer possible (Liu L.X. et al., 2023). Consistent with the results of previous studies (Goss-Souza et al., 2017; Pedrinho et al., 2019; Costa et al., 2022; Cornell et al., 2023), we found that intensive land use can increase network structural complexity through reducing the network diameter and average path length, possibly because the land causes the death of many microorganisms during and intense use, and the remaining microorganisms need to co-operate better in order to adapt to the new environment (Cornell et al., 2023). However, the overall complexity did not change significantly due to decreased node and edge numbers.

In shrublands and woodlands, which are characterized by less anthropogenic disturbance, we observed the opposite trend, with more nodes and edges and smaller network diameters and average path lengths. Recent studies have also shown increases in the numbers of microbial network nodes and edges following land use shifts from grassland to woodland (Yang et al., 2022). This phenomenon coincides with the highest microbial diversity and richness in scrub and woodland observed in this study, with the highest number of nodes and edges due to more microbes involved in the network structure.

Numerous studies have shown that soil-based physicochemical factors such as soil pH, moisture, and electrical conductivity have important effects on microbial communities (Wu et al., 2020; Chen et al., 2021; Bai et al., 2023). Soil microbial communities may also be affected by nutrients such as C, N, P, and K (Moro et al., 2014; Huang et al., 2016; Batista and Dixon, 2019; Ali et al., 2022), which regulate soil physicochemical properties and alter the inter-root microecological environment by influencing plant root secretions. We predicted that MBC and MBN would influence microbial structures. However, our results showed that although these metrics showed some correlation with the relative bacterial and fungal abundance (Figures 5A,B), the soil base physicochemical traits did not explain structural differences in bacterial and fungal communities (Figures 5C,D), perhaps because anthropogenic disturbance levels varied greatly between the different land use types examined in this study. Examples include occasional grazing on natural grasslands, compared to fertilizer and water application on artificial grasslands and farmlands, and recreation by villagers on woodlands.

Tillage can also impact soil fungal diversity (Liu W. S. et al., 2023). Artificial grasslands and farmlands with different tillage practices and higher tillage intensity levels due to increasing demand for higher agricultural yields may be important contributors to microbial community structural changes (Zhang et al., 2023). Thus, water and fertilizer management in artificial grasslands and farmlands may be an important reason for microbial community structural differences compared to natural grasslands, shrublands, and woodlands. Grazing affects the structure of subsurface microbial communities through animal feeding, trampling, and the return of feces and urine to soil (Ingram et al., 2008; Aldezabal et al., 2015; Zhao et al., 2017); therefore, grazing behaviors may explain microbial community differences between natural grasslands and other land use types. To summarize, our results showed that soil properties had a limited influence on microbial community structure in the agro-pastoral ecotone of northern China. Thus, we speculate that differences in anthropogenic activity levels among the land uses types examined in this study had the most important influence on microbial community structure in the study area, in contradiction to our second hypothesis.