The study area is located in Tunken Forest, Kangbao County, Zhangjiakou City, Hebei Province, China (114.47°E, 41.54°N), which belongs to a typical agricultural pastoral ecotone. The area has a semi-arid monsoon climate in the north temperate zone. It has high temperature and little rain in summer and is cold and dry in winter. The average annual temperature is 2.1°C, and a frost-free period lasts for 90–105 days. The average annual precipitation is about 300 mm, with rain and heat concentrated in the same season, primarily from July to September. The main afforestation species are ulmus, populus, pinus, and larch. The regional soil is chestnut soil and meadow soil, with medium soil fertility.

In May 2023, natural grassland (GL), larch plantation forest (LS), pinus plantation forest (ZS), populus plantation forest (PS), and ulmus plantation forest (YS) with similar elevation, flat terrain, and basically the same restoration age were selected within the Tunken Forest Farm for the study. The selected sites all had similar soil characteristics prior to afforestation. The plantation forests had fewer understory plants and lower surface biomass than natural grassland (Table 1). Three sample plots (1 m × 1 m) were randomly selected in each sample site. The five-point sampling method was used to collect soil samples of 0–20 cm, 20–40 cm, and 40–60 cm using a soil auger with an inner diameter of 5 cm around the sample points and at the center. Visible matter, such as plant stubs, stones and gravels in the samples, were removed. Then soil samples from the same soil layer were immediately mixed to form a single soil sample. The mixed soil samples were divided into two parts, which were placed into sealed bags and sterilized tubes for microbiological sampling. The soil sample in the sealed bag was air-dried and used to determine soil physical and chemical properties; the soil sample in the sterilized tubes for microbiological sampling was returned to the laboratory with an ice pack and stored at −80°C for high-throughput sequencing analysis.

Soil pH was measured by a pH meter (a 1:2.5 soil/water mixture), soil organic carbon (SOC) was measured by the potassium dichromate volumetric method, total carbon (TC) was measured by the high-temperature scorching method, total nitrogen (TN) was measured by the semi-micro Kjeldahl method (Bao, 2000), and soil particle size was measured using a Mastersizer 3000 laser particle sizer. Soil particle size grading was based on the American system of soil grain size: clay (<2 μm), silt (2–50 μm), and sand (50–2000 μm) (Chen et al., 2020).

Total DNA extraction from 0.5 g soil samples was performed according to the FastDNA^® Spin Kit for Soil (MP Biomedicals, United States) instructions, and DNA extraction quality was checked using 1% agarose gel electrophoresis. DNA concentration and purity were determined using a NanoDrop2000.

PCR was performed on a GeneAmp 9700 PCR system. Primers 338F (5′-ACTCCTACGGGAGGCAGCAG-3′) -806R (5′-GGACTACHVGGGTWTCTAAT-3′) and ITS1F (5′-CTTGGTCATTTAGAGGAAGTAA-3′) -ITS2R (5′-GCTGCGTTCTTCATCGATGC-3′) were used for soil bacterial 16S rRNA genes in the V3-V4 hypervariable regions and soil fungal internal transcribed spacer (ITS) region, respectively (Sun et al., 2017). Amplification products were detected by 2% agarose gel electrophoresis and recovered from the gel using the AxyPrep DNA Gel Extraction Kit, washed with Tris–HCl and verified by 2% agarose gel electrophoresis (Hu et al., 2017). PCR products were quantified using a QuantiFluorTM-ST fluorometer, and samples were adjusted according to sequencing needs.

Sequencing was performed on the Illumina MiSeq platform by Shanghai Major Biomedical Technology Company. The high-throughput sequencing data were deposited in the NCBI Sequence Read Archive (BioProject ID PRJNA1015983, study accession number SRP460000).

Sequence analysis was performed using Uparse software (v7.0.1090) to cluster valid data from soil samples. Sequences shorter than 200 bp, ambiguous bases, and sequences with an average mass less than 25 were removed (Hu et al., 2017). Chimeric sequences were removed using USEARCH (v11). Unique sequences >97% similarity levels were clustered into an operational taxonomic units (OTU) (Nguyen et al., 2016). Bacterial and fungal sequences were classified by comparing the representative sequences of OTUs with Silva version 138 16S rRNA and UNITE version 8.0 databases, respectively (Zhou et al., 2020). Diversity indices Chao and Shannon were calculated using the Qiime (v1.9.1) platform.

Soil physicochemical properties and microbial diversity were analyzed and plotted using R (4.2.1). We used microbial OTU richness as the metrics of, and calculated, the microbial α-diversity. Soil physicochemical properties and microbial α-diversity were compared by one-way ANOVA, and significance of differences was compared by the LSD method. The data of soil physicochemical properties and microbial α-diversity are presented as mean values ± standard deviation (SD). Beta diversity of soil microbial communities was resolved by principal co-ordinates analysis (PCoA) based on the weighted unifrac distance algorithm, and differences between groups were tested using ANOSIM similarity analysis (number of permutations = 999) with the R “vegan” package. Linear regression analysis was used to assess the relationship between microbial community composition and environmental factors.

The construction of microbial co-occurrence networks used with Pearson’s correlation coefficients matrix the R “WGCNA” package (Langfelder and Horvath, 2008). Pearson’s correlation (p > 0.8) and significance values (p < 0.01) were used for microbial network building. Topological features of the overall network (e.g., nodes, edges, average degree, modularity, density, average path length and average clustering coefficient) were calculated using the R “igraph” package (Wang et al., 2023). To obtain the microbial network complexity index, we calculated the Z-scores for some topological features (nodes, edges, average degree, average clustering coefficient, density and the reciprocal of average path length) of the microbial network individually. The Z-scores for all measured variables were averaged to obtain the complexity index of the microbial network for each locality (Wang et al., 2023). The co-occurrence networks were plotted using Gephi (0.10.1) software and visualized by the Frucherman Reingold algorithm.

Significant differences were observed in the physicochemical properties (Table 2). Soil pH was slightly alkaline and varied from 8.29 to 8.57. The soil was dominated by sand, followed by silt, and the content of clay particles was the lowest. In 0–20, 20–40, and 40–60 cm layers, total carbon (TC) in PS was significantly higher than that in GL (p < 0.05), with increases of 90.23, 86.73, and 109.01%, respectively. In 40–60 cm layer, the TC in YS was significantly higher than that in GL (p < 0.05), with an increase of 114.15%. In 0–20 cm layer, total nitrogen in PS and YS was significantly higher than that in GL (p < 0.05), with increases of 41.54 and 60.00%, respectively. In 20–40 cm layer, TN in PS and YS was significantly higher than that in GL (p < 0.05), with increases of 101.35 and 120.27%, respectively. In 40–60 cm layer, TN in the PS and ZS was significantly higher than that in GL (p < 0.05), with increases of 292.31 and 315.38%, respectively. In 0–20, 20–40, and 40–60 cm layers, soil organic carbon (SOC) in PS was significantly higher than that in GL (p < 0.05), with increases of 53.95, 82.61, and 360.00%, respectively. In 0–20 and 20–40 cm layers, SOC in YS was significantly higher than that in GL (p < 0.05), with increases of 60.90 and 107.41%, respectively. Therefore, compared with GL, plantation forest increased soil carbon and nitrogen, especially in PS.

A total of 2,636,522 (ranging from 34,619 to 63,468 sequences per sample) and 3,975,388 (ranging from 50,899 to 166,964 sequences per sample) quality sequences of the bacterial 16S rRNA and fungal ITS genes were obtained, respectively. In total, 19,957 bacterial OTUs and 5,957 fungal OTUs were obtained in the five vegetation restoration types.

The bacterial Shannon of LS and ZS were significantly 7.27 and 7.77% higher than those of GL in the 20–40 cm soil layer, respectively (Figure 1). The bacterial Chao indices of LS and ZS were significantly 21.06 and 22.39% higher than those of GL in the 20–40 cm soil layer, respectively. The fungal Shannon and Chao indices of PS were significantly 29.47 and 30.17% lower than those of GL in the 0–20 cm soil layer. The fungal Chao indices of YS were significantly 85.70 and 146.86% higher than those of GL in the 20–40 cm and 40–60 cm soil layer, respectively. Overall, Shannon and Chao indices decreased with increasing depth. Generally, compared to GL, plantation forests vegetation restoration types increased microbial alpha diversity.

The dominant phylum in the bacterial and fungal communities were generally consistent across the five vegetation restoration types (Figure 2). Actinobacteria, Proteobacteria, and Acidobacteriota were the primary bacterial phyla in the examined soil samples (Figure 2A), while Ascomycota and Basidiomycota were the primary fungal phyla (Figure 2B). Notably, the abundance of Ascomycota was higher in GL, LS, and YS soils with 69–88%, 59–64%, and 59–77%, respectively, while the abundance of Basidiomycota was higher in PS soils with 59–89%. The abundance of dominant phyla of fungi in the surface layer of the ZS and the lower layer of the ZS varied, with the abundance of Ascomycota in the 0–20 cm layer being high with 56%; its abundance in the 20–40 cm and 40 cm layers was also high. The abundance of Ascomycota was higher in the 0–20 cm soil layer with 56%, while the abundance of Ascomycota was higher in the 20–40 cm and 40–60 cm soil layers with 54 and 53%, respectively.

Principal Coordinate Analysis was used to resolve the variability in the composition of soil bacterial and fungal communities in the five vegetation restoration types (Figure 3). The results demonstrated that, at the phylum level, the first two axes (PC1 and PC2) explained 63.17 and 88.80%, respectively, of the total variance in bacteria and fungi in the vegetation restoration types (Figures 3A,B). By contrast, at the genus level, the first two axes (PC1 and PC2) explained 56.01 and 50.82%, respectively, of the total variation in bacteria and fungi (Figures 3C,D).

Tests of intergroup differences using ANOSIM similarity analysis (number of permutations = 999) showed that, at the phylum level, no significant difference was found in the composition of soil bacterial communities (p = 0.108) (Figure 3A), but significant differences were observed in the composition of fungal communities (p = 0.001) among the five patterns (Figure 3B). On the contrary, at the genus level, a significant difference was found in the composition of soil bacterial communities (p = 0.001) and soil fungal communities (p = 0.001) (Figures 3C,D).

In this study, after removing redundant variables, four environmental parameters were selected. The results demonstrated that the microbial community structure was driven by various soil environmental factors. Specifically, SOC (p = 0.003) and TN (p < 0.001) significantly influenced the bacterial community β-diversity (Figures 4A,C), while SOC (p = 0.030), clay content (p = 0.004) and silt content (p < 0.001) significantly influenced fungal community β-diversity (Figures 4B,F,H). TN did not affect the fungal community β-diversity (p > 0.05) (Figure 4D). Clay content and silt content did not affect the bacterial community β-diversity (both p > 0.05) (Figures 4E,G). Bacterial and fungal community structure was influenced by different environmental factors.

The results of the analyses showed that soil bacterial and fungal community networks in various soil layers of different vegetation types exhibited different co-occurrence patterns (Figure 5). Among the five vegetation restoration types, the bacterial and fungal network complexity indices of GL were the lowest, with the sum total of each soil layer being −0.601 and −1.425, respectively (Tables 3, 4). The bacterial network complexity index of PS was the highest (1.108) and the fungal network complexity index of LS was the highest (0.802). Broadleaf forests (PS, YS) having higher bacterial network complexity than coniferous forests (ZS, LS), and coniferous forests having higher bacterial network complexity than natural grassland (GL). This trend is similar to that in soil carbon and nitrogen.

In this study, compared with natural grassland, after about 50 years of vegetation restoration, four plantation forests had a increase of soil nutrients (TC, TN, SOC). Especially, populus had significantly higher TC, TN and SOC than natural grassland in each soil layer (p < 0.05). From the view of improving soil quality (carbon and nitrogen), populus are the better choices of tree species for plantation forest restoration types in the agricultural pastoral ecotone of Zhangjiakou. Populus are considered the king of the world’s fast-growing tree species, with a rapid growth rate and large annual increase (Łukaszkiewicz et al., 2024). It is able to absorb and store large amounts of carbon dioxide in a relatively short period of time (Bahrian et al., 2024; Mulyana et al., 2024). In addition, populus have a symbiotic relationship with certain nitrogen-fixing bacteria that can help them fix more nitrogen, thereby increasing the nitrogen content of the soil and improving its fertility. This study revealed that broadleaf forests (PS, YS) had higher TC than coniferous forests (ZS, LS), and coniferous forests had higher TC than natural grassland (GL). This trend was attributed to the fact that broadleaf had a thicker layer of litter that usually decomposes faster than that in coniferous forests, which, in turn, increases the carbon input to the soil (Wei et al., 2021). Previous studies have revealed higher soil carbon accumulation in broadleaf forests than in coniferous forests in northern area, which is similar to the results of our study (Geng et al., 2009; Wei et al., 2021). Particle size analysis revealed that the soil in the study area had a higher content of sand particles and a lower content of silt and clay particles, which are more susceptible to erosion and transport by monsoon winds due to their smaller size and lighter mass (Chen et al., 2020). Compared to grassland, plantation forests play a role in reducing wind speed and slowing down wind erosion. Thus, fine particles carried by the atmosphere are intercepted, which leads to the gradual accumulation of more fine materials in the soil surface layer (Guo et al., 2020). This phenomenon is reflected in the higher content of clay and powder particles in plantation forests.

Our results showed that bacterial α-diversity was significantly higher in ZS and LS than that in GL in the 20–40 cm soil layer. Plantation forests have a large and complex root systems, which provides a comfortable habitat for soil microbial communities. This condition promotes soil microbial massive uptake of soil nutrients and have an increase of microbial α-diversity (Yang et al., 2020b). However, fungal α-diversity was significantly lower in PS than that in GL in the 0–20 cm soil layer. The plantation forest vegetation in this study was all monospecific, with low litter diversity in the top soil layer (0–20 cm), which also had a negative impact on microbial α-diversity (Otsing et al., 2018; Xu et al., 2022). Regression analyses showed that SOC and TN were significantly correlated with soil bacterial community composition; SOC, clay content and silt content were significantly correlated with soil fungal community composition. This study revealed that SOC were both significantly correlated with soil bacterial and fungal community composition. Since SOC serves as the primary source of energy and nutrients for soil microorganisms, and different microorganisms have different efficiencies and preferences in utilizing SOC, changes in SOC can significantly influence soil microbial community composition (Sokol et al., 2022). Furthermore, these findings implied that the bacterial community composition was regulated by soil nutrients to a certain extent in the process of vegetation restoration in agricultural pastoral ecotone, while the fungal community structure is mainly regulated by soil texture. This result was similar to findings by Xia et al. that soil texture has a greater effect on fungi than bacteria (Xia et al., 2020).

Our results showed that the microbial network complexity increased with soil nutrients. Overall, the microbial network complexity of GL was lowest in five vegetation types, which is similar to the findings of Yang et al. (2020a, 2020b). A rise in carbon and nitrogen content can increase the number of soil microorganisms, which will lead to the formation of a more complex soil microbial network (Zhou et al., 2020; Jiao et al., 2021). In this study, the complexity of soil microbial network in plantation forest vegetation gradually increased with the improvement in soil nutrient status. The restoration of plantation forest vegetation enhanced microbial community stability, which is important for slowing down soil wind erosion, resisting environmental disturbances, and promoting the cycling of soil nutrients (Wang et al., 2023). Especially, the bacterial network complexity of PS was highest in five vegetation types. We mentioned above that populus are the better choices of tree species to improve soil quality. Now we do believe that populus are very suitable for planting in the agricultural pastoral ecotone of Zhangjiakou. Complex microbial networks not only enhance plant growth and resilience, but also improve ecosystem stability and sustainability and promote the long-term benefits of environmental restoration (Jagadesh et al., 2024).

We also found that the modularity index and average path length were higher for bacteria than for fungi. The network modularity index usually characterizes the resistance of the system to external disturbances, with higher modularity indexes being associated with greater resistance to external disturbances (Deng et al., 2012). A smaller average path length indicates a closer interaction among network nodes, which also implies a faster diffusion of external disturbances in the microbial network and a more sensitive response to disturbances (Carpenter et al., 2012). This observation suggests that bacterial networks are more stable than fungal networks in responding to disturbances, which is consistent with previous studies (Zhao et al., 2019). Bacteria are able to cope with drought through rapid reproduction, dormancy and metabolic adjustments (Lebre et al., 2017), while fungi are more dependent on water and moist environments, rendering them susceptible to greater impacts under drought conditions (Logan et al., 2021). After drought or other disturbances, bacteria are able to restore ecological functions more quickly and promote ecosystem restoration. In contrast, fungi are slower to recover and may slow down the ecosystem restoration process. In conclusion, more attention should be paid to bacterial network characteristics during dryland vegetation restoration.