Distinct Assembly Processes and Determinants of Soil Microbial Communities between Farmland and Grassland in Arid and Semiarid Areas

ABSTRACT

It is critical to identify the assembly processes and determinants of soil microbial communities to better predict soil microbial responses to environmental change in arid and semiarid areas. Here, soils from 16 grassland-only, 9 paired grassland and farmland, and 16 farmland-only sites were collected across the central Inner Mongolia Plateau, covering a steep environmental gradient. Through analyzing the paired samples, we discovered that land uses had strong effects on soil microbial communities but weak effects on their assembly processes. For all samples, although no environmental variables were significantly correlated with the net relatedness index (NRI), both the nearest taxon index (NTI) and the β-nearest taxon index (βNTI) were most related to mean annual precipitation (MAP). With the increase of MAP, soil microbial taxa at the tips of the phylogenetic tree were more clustered, and the contribution of determinism increased. Determinism (48.6%), especially variable selection (46.3%), and stochasticity (51.4%) were almost equal in farmland, while stochasticity (75.0%) was dominant in grassland. Additionally, Mantel tests and redundancy analyses (RDA) revealed that the main determinants of soil microbial community structure were MAP in grassland but mean annual temperature (MAT) in farmland. MAP and MAT were also good predictors of the community composition (the top 200 dominant operational taxonomic units) in grassland and farmland, respectively. Collectively, in arid and semiarid areas, soil microbial communities were more sensitive to environmental change in farmland than in grassland, and unlike the major impact of MAP on grassland microbial communities, MAT was the primary driver of farmland microbial communities.

IMPORTANCE As one of the most diverse organisms, soil microbes play indispensable roles in many ecological processes in arid and semiarid areas with limited macrofaunal and plant diversity, yet the mechanisms underpinning soil microbial community are not fully understood. In this study, soil microbial communities were investigated along a 500-km transect covering a steep environmental gradient across farmland and grassland in the areas. The results showed that precipitation was the main factor mediating the assembly processes. Determinism was more influential in farmland, and variable selection of farmland was twice that of grassland. Temperature mainly drove farmland microbial communities, while precipitation mainly affected grassland microbial communities. These findings provide new information about the assembly processes and determinants of soil microbial communities in arid and semiarid areas, consequently improving the predictability of the community dynamics, which have implications for sustaining soil microbial diversity and ecosystem functioning, particularly under global climate change conditions.

INTRODUCTION

Arid and semiarid regions cover about 41% of the earth’s land surface and support over 2 billion people (1). Although the conditions vary considerably in different regions of the world, all of them are characterized by water deficiency, extreme temperature, low nutrient levels, and limited macrofaunal and plant diversity, which lead to a highly fragile ecosystem susceptible to human disturbance and climate change (2, 3). Soil microbial communities represent the majority of terrestrial ecosystem diversity and play pivotal roles in maintaining ecosystem functions and services in arid and semiarid areas (4). Therefore, it is crucial to unravel the mechanisms that underpin soil microbial community to predict soil microbial responses to environmental change in the areas (5).

Community assembly seeks to explain the evolutionary and ecological mechanisms that govern community and is a key topic in ecology (6, 7). It can not only reflect the historical processes of community formation but also be used to predict the future development of community. Microbial community assembly is commonly divided into deterministic and stochastic processes (8). Deterministic processes involve environmental filters and biotic interactions, which result in community dissimilarities reflecting differences in the adaptability of biological traits (9). Stochastic processes include random birth, death, immigration, and emigration, which lead to microbial communities indistinguishable from those caused by random chance alone (10). Recently, it has been generally accepted that determinism and stochasticity simultaneously govern soil microbial communities, and their relative contributions may depend on environmental conditions (11). Tripathi et al. (12) showed that determinism was more important in extremely acidic or alkaline soils than in near-neutral soils. In addition, determinism dominated in harsh environments and low-productivity systems, whereas stochasticity prevailed in more benign habitats and higher-productivity systems (13, 14). However, the relative contribution of determinism and stochasticity in soil microbial communities in arid and semiarid areas has not been clarified.

Numerous studies have demonstrated that precipitation plays a major role in structuring soil microbial communities in natural ecosystems of arid and semiarid areas, mainly grassland (15–17). Similarly, precipitation is identified as the first contributor to vegetation distribution, particularly at large spatial extents (18, 19). Given that different microbial taxa have distinct inherent tolerances to water stress and nutrient use strategies, precipitation may affect soil microbial communities by altering the water and substrate availabilities, which subsequently induce microbial physiological stress, growth, and metabolic activity responses (20–22).

Furthermore, agriculture has been widely developed in arid and semiarid areas and will continue to extend to feed the growing population (2, 23). There are still large knowledge gaps in the drivers of soil microbial communities of farmland in these areas, even though the communities of farmland in other areas have received extensive attention. Along the black soil zone of northeast China, bacterial communities of agricultural soils were mostly affected by soil pH and total carbon (24). Across the typical wheat planting fields on the Northern China Plain, soil pH described the greatest variance in microbial community structures in both bulk and rhizosphere soils (25). Soil microbial communities along the Yangtze River were primarily defined by soil pH in the flooded paddies, and the communities from the rained croplands between North and South America were similar despite a great geographic distance, implying that edaphic variables rather than climatic factors predominantly affected soil microbial communities (26). Collectively, soil properties, especially pH, drove soil microbial communities in farmland of more humid areas, whereas the driving factors of the communities in farmland of more arid areas need further investigation.

Here, we present a regional-scale field study along a 500-km transect with a steep climatic and edaphic gradient across the central Inner Mongolia Plateau (Fig. 1). From the northwest to southeast of the transect, the major land use type shifted gradually from grassland to farmland along the gradient, and there was a small overlap between the two land uses in the middle of the transect, which is a prevalent phenomenon in arid and semiarid areas (27). We examined soil microbes using high-throughput sequencing and aimed to determine soil microbial community assembly processes along the transect and the main drivers of the communities, especially in farmland. This study will provide important insights into the maintenance mechanism of the soil microbial community in arid and semiarid areas.

FIG 1.

Sampling sites across the central Inner Mongolia Plateau in arid and semiarid areas. Blue crosses represent grassland sites, and black triangles represent farmland sites. Different colors represent different mean annual precipitation (MAP). (Images created using ArcGIS v10.3.)

RESULTS

Environmental variables across the transect.

The major soil properties, climatic factors, and vegetation conditions across the transect are summarized in Table 1. Significant differences between farmland and grassland were observed in all the soil properties measured except for total nitrogen (TN) and soil organic matter (SOM). Soil pH and sand were higher in grassland, while electrical conductivity (EC), total phosphorus (TP), clay, and silt were higher in farmland. Furthermore, mean annual precipitation (MAP) ranged from 285 to 470 mm in farmland and 179 to 341 mm in grassland, mean annual temperature (MAT) ranged from 3.40 to 9.90°C in farmland and 3.40 to 5.24°C in grassland, and the normalized difference vegetation index (NDVI) was greater in farmland than in grassland, suggesting that farmland sites had a more benign climate and a better vegetation condition. The above-described results implied that the transect covered a great environmental gradient.

TABLE 1.

Environmental variables in farmland and grassland across the transect^a

Parameter	Farmland (N = 25)		Grassland (N = 25)
	Mean	CV	Mean	CV
pH	8.34 ± 0.14b	0.02	8.56 ± 0.19a	0.02
EC	156 ± 31a	0.20	100 ± 21b	0.22
TN (g/kg)	1.04 ± 0.34a	0.33	1.04 ± 0.45a	0.43
SOM (g/kg)	19.98 ± 6.61a	0.33	17.25 ± 8.38a	0.49
TP (g/kg)	1.91 ± 0.52a	0.27	0.88 ± 0.42b	0.48
Clay (%)	8.70 ± 3.02a	0.35	6.67 ± 2.24b	0.34
Silt (%)	38.40 ± 11.47a	0.30	24.64 ± 10.52b	0.43
Sand (%)	52.89 ± 13.95b	0.26	68.71 ± 12.22a	0.18
MAP (mm)	378 ± 59a	0.16	249 ± 64b	0.26
MAT (°C)	7.25 ± 2.46a	0.34	4.17 ± 0.47b	0.11
NDVI	0.60 ± 0.14a	0.24	0.26 ± 0.12b	0.48

^aValues are given as means ± standard deviations, and values within a column followed by different letters are significantly different (t tests). CV, coefficient of variation; EC, electrical conductivity; TN, total nitrogen; SOM, soil organic matter; TP, total phosphorus; MAP, mean annual precipitation; MAT, mean annual temperature; NDVI, normalized difference vegetation index.

Overview of the high-throughput sequencing data and microbial taxa.

After randomly subsampling 19,000 sequences from each sample based on the minimum sequencing number among all samples, operational taxonomic units (OTUs) were clustered at a 97% similarity level. The ends of rarefaction curves flattened out at this sequencing depth, indicating that the sequencing data were large enough to reflect the majority of soil microbial taxa (see Fig. S1A at http://vdb3.soil.csdb.cn/resources/myfiles/Supplementary_material.pdf). A total of 7,078 OTUs were identified from all samples, and farmland and grassland soils possessed 6,330 and 6,329 OTUs, respectively (Fig. S1B).

Of all sequences, 99.6% were annotated at the phylum level and grouped into 42 phyla. There were 11 phyla with relative abundances of >1%. Among these dominant phyla, Proteobacteria, Chloroflexi, Planctomycetes, Bacteroidetes, and Nitrospirae were more abundant in farmland, Actinobacteria, Thaumarchaeota, and Tectomicrobia were more abundant in grassland, and Acidobacteria, Gemmatimonadetes, and Firmicutes did not change significantly (Table S1 at http://vdb3.soil.csdb.cn/resources/myfiles/Supplementary_material.pdf). These results demonstrated that the relative abundances of most soil microbial taxa were markedly different between the two land uses.

Soil microbial communities and their assembly processes in farmland and grassland for paired samples.

The effects of land use on soil microbial communities and their assembly processes were evaluated using 9 paired samples from adjacent farmland and grassland. The results showed that soil microbial communities were separated obviously by the two land uses based on nonmetric multidimensional scaling (NMDS) and the analysis of similarities (ANOSIM) (Fig. S2 at http://vdb3.soil.csdb.cn/resources/myfiles/Supplementary_material.pdf), suggesting that land uses had a major impact on soil microbial communities.

The net relatedness index (NRI), the nearest taxon index (NTI), and the β-nearest taxon index (βNTI) were used to assess the assembly processes. Paired t tests revealed that the NRI was significantly higher in farmland than in grassland, yet the NTI did not change remarkably (Fig. 2A and B). The t tests showed that the differences in the βNTI between farmland and grassland were also not significant (Fig. 2C). Of all the environmental variables prone to vary with land uses, only TN was significantly positively correlated with the NRI, and none were significantly correlated with the NTI and βNTI (Table S2). These results displayed that land uses had a minor impact on the assembly processes.

FIG 2.

Comparing the net relatedness index (NRI) and the nearest taxon index (NTI) by paired t tests and the β-nearest taxon index (βNTI) by t tests between farmland and grassland for the paired samples. P values are the statistical significances of paired t tests or t tests.

Determinants of soil microbial communities in farmland and grassland for all samples.

For all samples, the NMDS analyses showed that soil microbial communities in farmland were closely associated with MAT (Fig. 3A), while the communities in grassland were tightly bound to MAP (Fig. 3B). Similarly, Mantel tests and redundancy analyses (RDA) suggested that MAT and MAP exerted the greatest effects on soil microbial community structure in farmland and grassland, respectively (Table 2, Fig. S3 at http://vdb3.soil.csdb.cn/resources/myfiles/Supplementary_material.pdf). Moreover, the community structure of farmland was also significantly affected by NDVI, TN, MAP, SOM, TP, and EC. The community structure of grassland was significantly related to NDVI, TP, SOM, TN, EC, and silt. Notably, as primary determinants of the community structure, the r value of Mantel tests was only 0.341 for MAP in grassland, while it was 0.522 for MAT in farmland.

FIG 3.

Nonmetric multidimensional scaling (NMDS) analyses of soil microbial communities in farmland and grassland based on unweighted UniFrac distance (stress = 0.126). Sites have been colored according to mean annual temperature (MAT) (A) and mean annual precipitation (MAP) (B).

TABLE 2.

Mantel tests showing the effects of environmental variables on soil microbial community structures in farmland and grassland^a

Parameter	Farmland		Grassland
	r	P	r	P
pH	0.067	0.202	0.064	0.285
EC	0.164	0.020	0.235	0.027
TN	0.340	<0.001	0.291	0.002
SOM	0.300	<0.001	0.300	0.002
TP	0.258	0.001	0.320	0.013
Clay	0.091	0.132	−0.104	0.829
Silt	0.039	0.309	0.184	0.024
Sand	0.046	0.294	0.103	0.148
MAP	0.304	<0.001	0.341	<0.001
MAT	0.522	<0.001	0.044	0.315
NDVI	0.432	<0.001	0.332	<0.001

^aFor abbreviations, see Table 1.

Additionally, we explored the relationships between the relative abundances of the top 200 dominant OTUs and environmental variables in both farmland and grassland. The most OTUs were markedly associated with MAT in farmland and NDVI and MAP in grassland (Pearson’s P value <0.05 after Benjamini-Hochberg correction) (Table 3), indicating that MAT was the most important factor influencing soil microbial community composition of farmland, while NDVI and MAP strongly affected the community composition of grassland.

TABLE 3.

Number of OTUs significantly correlated with environmental variables in farmland and grassland in the top 200 dominant OTUs^a

Soil	pH	EC	TN	SOM	TP	Clay	Silt	Sand	MAP	MAT	NDVI
Farmland	6	29	47	38	39	13	12	13	43	84	60
Grassland	2	43	44	38	21	5	22	21	50	0	53

^aPearson’s P value is adjusted by the Benjamini-Hochberg correction. When the corrected P is less than 0.05, the correlations between relative abundances of OTUs and environmental variables are significant. For abbreviations, see Table 1.

Soil microbial community assembly processes in farmland and grassland for all samples.

Across all samples, both the NRI and NTI were above 2, and the NTI was considerably higher than the NRI (Fig. S4). The correlations between the NRI, NTI, and βNTI and environmental variables are presented in Table 4. None of the environmental variables, including TN, were significantly correlated with the NRI, which might result from the insignificant change of TN between farmland and grassland (Table 1). The NTI had the strongest positive correlation with MAP, followed by TP, MAT, EC, and NDVI. The βNTI was most related to the distance of MAP, followed by the distance of NDVI, pH, EC, TP, silt, sand, MAT, and clay. All samples were divided into subgroups according to different levels of MAP. The βNTI increased significantly with the level of MAP (Fig. S5 at http://vdb3.soil.csdb.cn/resources/myfiles/Supplementary_material.pdf), which indicated a shift from stochastic processes to deterministic processes. Collectively, the assembly processes were strongly mediated by MAP across all samples.

TABLE 4.

Pearson correlation coefficients between the NRI and NTI and environmental variables and between the βNTI and environmental distance for all samples^a

Index	pH	EC	TN	SOM	TP	Clay	Silt	Sand	MAP	MAT	NDVI
NRI	0.100	−0.134	−0.046	−0.035	0.077	−0.115	−0.140	0.139	0.068	0.159	0.032
NTI	−0.108	0.298*	0.033	0.111	0.324*	0.169	0.166	−0.173	0.341*	0.311*	0.280*
βNTI	−0.192***	−0.162***	0.036	−0.007	−0.142***	−0.071*	−0.127***	−0.114***	−0.234***	−0.081**	−0.233***

^aAsterisks represent significance of correlation (*, P < 0.05; **, P < 0.01; ***, P < 0.001). For abbreviations, see Table 1.

According to the βNTI and the Bray-Curtis-based Raup-Crick (RC_bray), the assembly processes of soil microbial communities in farmland and grassland were quantified for all samples. As t tests revealed, the βNTI values were significantly higher in farmland than in grassland, and most of the βNTI values in grassland were between −2 and 2 (Fig. 4A). Determinism (48.6%) and stochasticity (51.4%) were almost equal in the communities of farmland, whereas stochasticity (75.0%) was more important than determinism (25.0%) in the communities of grassland (Fig. 4B). More specifically, variable selection (46.3%) contributed most to farmland microbial communities, and undominated processes (39.0%) were dominant in grassland microbial communities. Furthermore, the distance-decay relationship between soil microbial community similarity and environmental distance was estimated by the ordinary least-squares regression. The slope of farmland was significantly steeper than that of grassland through the matrix permutations (P = 0.002), and the R² value was higher in farmland (Fig. 5), implying that compared to grassland, environmental change better explained the variation in the communities of farmland.

FIG 4.

Soil microbial community assembly processes between farmland and grassland. (A) The values of βNTI for soil microbial communities (t test, P < 0.001). Horizontal dashed gray lines indicate the thresholds at βNTI = +2 and −2, respectively. (B) The fraction of soil microbial community assembly governed principally by deterministic processes (variable selection and homogeneous selection) or stochastic processes (dispersal limitation, homogenizing dispersal, and undominated processes).

FIG 5.

Distance-decay curves between soil microbial community similarity (1 − unweighted UniFrac distance) and environmental distance. The lines denote the ordinary least-squares linear regression in farmland and grassland. Statistics are derived from regression analyses. Asterisks represent significance of correlation (***, P < 0.001).

DISCUSSION

Land uses had strong effects on soil microbial communities but weak effects on their assembly processes.

In the present study, we assessed the effects of land use on soil microbial communities and their assembly processes using the paired samples. Considerable differences in the communities were observed between farmland and grassland (see Fig. S2 at http://vdb3.soil.csdb.cn/resources/myfiles/Supplementary_material.pdf). Our findings supported many studies reporting that soil microbial communities were strongly affected by land uses (28–30). Land uses and soil microbial communities are linked through a variety of interactions. Land uses alter soil properties, such as pH, organic matter, and soil texture, and consequently influence the communities (31–33). Litter and root exudates provided by different vegetation types also differ in quality and quantity and may drive the change of the communities (34, 35).

In contrast, land uses had a minor impact on soil microbial community assembly in this study. Land uses altered the NRI but not NTI for the paired samples (Fig. 2A and B), suggesting that land uses affected species clustering with a tree-wide metric but not at the tips of the phylogenetic tree within a community. A significant relationship was discovered between the NRI and TN (Table S2), implying that land uses influence the NRI by affecting TN. Furthermore, both the NRI and NTI were positive across all samples, and the NTI was substantially higher than the NRI (Fig. S4), indicating that species cooccurring were generally more closely related than expected by chance, and species clustering was more obvious at the tips than across a phylogenetic tree. This result agrees with previous studies revealing that the phylogenetic signal for many ecological optima was mainly located near the tips of the tree (36, 37). Because of this, the βNTI is generally used to evaluate the assembly processes between communities. However, the βNTI did not change with the two land uses (Fig. 2C). Meanwhile, no environmental variables susceptible to land uses were significantly related to the NTI and βNTI (Table S2). These results implied that the NTI and βNTI are mainly mediated by environmental variables independent of land uses in arid and semiarid areas, such as climatic factors.

MAT and MAP were the main determinants of soil microbial communities in farmland and grassland, respectively.

In farmland, above 40% of the dominant OTUs were influenced by MAT (Table 3), and soil microbial community structure was predominantly influenced by MAT (Table 2), suggesting that MAT was the main determinant of soil microbial communities in agricultural ecosystems in arid and semiarid areas. The communities in farmland of more humid areas are generally controlled mainly by soil properties, especially pH (24–26). However, the relationship between the communities and pH was weak in farmland in this study. This discrepancy might be attributed to the high pH of the areas (Table 1), and the importance of pH as a driver of the communities was mostly displayed in acidic soils (3, 38).

The sensitive response of soil microbial communities to temperature change has been observed at large spatial scales. Although the rates of soil microbial diversity turnover across the global temperature gradient were substantially lower than those recorded for plant and animal, temperature was still a good predictor of soil microbial community structure in forest soils (39). Through analyzing soil, sediment, and biomat samples from 36 geothermal areas covering a temperature range of 7.5 to 99°C, Sharp et al. (40) reported that microbial diversity was strongly correlated with temperature. A recent study also revealed that the variation in soil bacterial community structure was mainly explained by temperature in forest ecosystems spanning the latitude of 21.6 to 50.8°N (22). At such a small spatial scale (∼300 km) and MAT span (3.40 to 9.90°C), it was rare that temperature as the main driver of soil microbial communities stood out from all the environmental variables measured. However, this study area was located in the fringe of the Asian monsoon region, and temperature fluctuated dramatically (41). Meanwhile, it was an ecological transitional zone between farmland and grassland, with small usable microhabitats (patches). The interspecific and intraspecific competition was extremely intense, because only a few individuals might occupy any given patch (42, 43). Many microbial taxa in this area might survive at the limit of the environment that they could tolerate and therefore responded sensitively to temperature change.

In grassland, not only soil microbial community composition but also structure was strongly affected by MAP (Tables 2 and 3). In this study, grassland sites were located in more arid areas, and MAP ranged from 179 to 341 mm. Under water limitation, the importance of precipitation on soil microbial communities has been widely documented in natural ecosystems (15–17). Wang et al. (44) discovered that a small amount of precipitation could trigger dramatic responses of soil bacterial communities under extreme drought conditions, and then the responses weakened as precipitation continued to increase. In this study, farmland sites with MAP ranging from 285 to 470 mm were located in more humid areas and equipped with irrigation facilities. Hence, although MAP still significantly affected soil microbial communities of farmland, it became less limiting. Moreover, the communities of grassland were also sensitive to the change of NDVI, probably because NDVI was tightly associated with MAP in the areas (Pearson’s r = 0.97, P < 0.001).

Soil microbial community assembly processes were distinct between farmland and grassland.

In this study, determinism, mainly variable selection, and stochasticity had almost equal shares in soil microbial communities of farmland, whereas stochasticity was dominant in the communities of grassland (Fig. 4B). This result is consistent with the research of Feng et al. (45), who found that variable selection (52.9%) mainly governed soil bacterial communities in maize fields of northeast China. Ning et al. (46) also reported that soil microbial communities were more stochastic (∼60%) than deterministic in grassland in the U.S. Great Plains.

MAP strongly mediated the assembly processes in arid and semiarid areas. Given the minor impact of land uses on the assembly processes, we evaluated the relationships between the NRI, NTI, and βNTI and environmental variables across all samples without regard to land uses. Although the NRI did not have significant correlations with all environmental variables, both the NTI and βNTI were most related to MAP (Table 4). Within a community, soil microbial taxa at the tips of the tree were more phylogenetically clustered with increasing MAP. Tripathi et al. (47) documented that more phylogenetic clustering could indicate greater influences of determinism. Similarly, the βNTI increased with the level of MAP (Fig. S5 at http://vdb3.soil.csdb.cn/resources/myfiles/Supplementary_material.pdf), suggesting that determinism became progressively more important between communities as MAP increased. It has been reported that the complex carbon substrates released by the roots of drought-tolerant plants, mainly natural vegetation, provided a resource-rich environment that decreased competitive pressures and increased microbial compositional stochasticity (37, 48). Therefore, the divergence of precipitation might be responsible for the distinct assembly processes between farmland and grassland in the areas.

Since determinism is characterized by a coupling between environmental parameters and soil microbes, understanding the assembly processes is essential to determine where and when to study soil microbial communities (49). Variable selection in soil microbial communities of farmland (46.3%) was twice that of grassland (23.3%) (Fig. 4B), and the community similarity decreased more rapidly with the environmental distance in farmland (Fig. 5), implying that the communities of farmland might be more sensitive to environmental change (50). Thus, also as the primary drivers, MAP explained a fraction of variation in the community structure in grassland, while the impact of MAT in farmland was relatively large (Table 2), which was reasonable from the view of the community assembly. Nevertheless, the resistance of soil microbial communities to environmental change is significant to maintain ecosystem functions (51). Especially in agricultural ecosystems, a stable environment and soil microbial community are of great necessity to sustain a high crop yield. These findings highlight that we should pay more attention to environmental change in farmland than in grassland in arid and semiarid areas.

Conclusions.

This study revealed that the determinants and assembly processes of soil microbial communities were different between farmland and grassland in arid and semiarid areas. Land uses had a major impact on soil microbial communities but a minor impact on their assembly processes through investigating the paired samples. Across the transect, the assembly processes were mainly mediated by MAP, and the contribution of determinism increased with MAP. Soil microbial communities of grassland were strongly affected by MAP. More interestingly, the communities of farmland were susceptible to MAT. Furthermore, determinism and stochasticity were almost equal in farmland, while stochasticity dominated in grassland, implying that the communities in farmland were more sensitive to environmental change. Overall, this study not only broadens our understanding of the maintenance mechanism of current soil microbial community but also improves the predictability of future soil microbial community in arid and semiarid areas.

MATERIALS AND METHODS

Study area and soil sampling.

The study was conducted along a 500-km transect across the central Inner Mongolia Plateau in arid and semiarid areas, extending from 40.50 to 43.56°N and from 112.04 to 116.04°E (Fig. 1). From the northwest to southeast of the transect, mean annual precipitation (MAP) ranges from 179 to 470 mm, mean annual temperature (MAT) ranges from 3.40 to 9.90°C, and soil texture varies from sand to loam. Correspondingly, grassland is the major land use type under harsh climatic and edaphic conditions, and farmland begins to appear and eventually dominates with increasing MAP and MAT and decreasing content of sand. Along the transect, 50 sampling sites were set up, including 16 grassland-only, 9 paired farmland and grassland, and 16 farmland-only sites. The paired sites were always within 2 km of each other. In the grasslands, plant species are dominated by Ceratoides latens (J. F. Gmel.) Reveal et Holmgren, Stipa klemenzii Roshev., and Allium polyrhizum Turcz. ex Regel. All the farmlands cropped maize and grew one crop a year. The growth period of maize is from May to August, irrigation facilities are equipped, and nitrogen and phosphorus are the main fertilizers used in the areas. Soil samples were collected in August 2017, when plant biomass almost peaked for a year. At each site, 10 bulk soil cores (3 cm in diameter and 15-cm depth) were randomly collected in a plot with an area of approximately 100 m² and subsequently pooled to form a composite sample. Each sample was divided into two parts: one was transported on ice packs (4°C) to the laboratory and stored at −80°C until DNA extraction, and the other was air dried for later soil property analyses.

Climatic factors, vegetation data, and soil properties.

MAP and MAT were obtained from the WorldClim global climate data set (https://www.worldclim.com) with a spatial resolution of 30 s (52). Vegetation conditions were assessed by the normalized difference vegetation index (NDVI). The NDVI data were from the moderate-resolution imaging spectroradiometer (MODIS) aboard NASA’s Terra satellites (https://modis.gsfc.nasa.gov/), which had a time resolution of 16 days and a spatial resolution of 250 m. We calculated the mean values of NDVI during the sampling period, and the mean NDVI was equal or close to the peak value of annual NDVI.

Soil organic matter (SOM), total nitrogen (TN), and total phosphorus (TP) were measured through the K₂Cr₂O₇-H₂SO₄ oxidation method, the semimicro-Kjeldahl digestion method, and the acid-digestion molybdate colorimetric method, respectively. These analyses were performed according to the procedures described by ISSCAS (53). Soil particle sizes were determined using a laser diffraction method (54). Soil pH was measured with a pH monitor in a 1:2.5 soil-water slurry. Electrical conductivity (EC) was measured with a conductivity meter in a 1:5 soil-water suspension.

DNA extraction, amplification, and sequencing.

DNA was extracted from 0.5 g fresh soil using the FastDNA Spin kit (MP Biomedicals, Solon, OH, USA) according to the manufacturer’s protocol. The DNA concentration and quality were assessed by a NanoDrop spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, USA). Each DNA sample was amplified at bacterial and archaeal 16S rRNA V4 and V5 regions using the forward and reverse primers 515F (5′-GTGCCAGCMGCCGCGG-3′) and 907R (5′-CCGTCAATTCMTTTRAGTT-3′) (55). The 12-bp barcoded oligonucleotides were fused to the forward primer. PCR was carried out under the following procedures: initial denaturation at 94°C for 3 min, 30 cycles (94°C for 40 s, 56°C for 60 s, 72°C for 60 s), and a final extension at 72°C for 10 min. For each sample, the PCR was performed in triplicate, and the PCR products were pooled and purified with the QIAquick PCR purification kit (Qiagen). High-throughput sequencing was performed on an Illumina MiSeq platform using paired-end reads (2 × 250 bp) (MiSeq reagent kits v2; Illumina Inc., San Diego, CA, USA).

Bioinformatics processing.

The raw sequencing data were processed using Quantitative Insights into Microbial Ecology (QIIME) ver. 1.9.0 (56). The paired-end reads were assembled with the Fast Length Adjustment of SHort reads (FLASH) software (57). To minimize the effects of random sequencing errors, we discarded the reads with a length shorter than 300 bp, a quality score lower than 20, ambiguous bases, or improper primers. Chimeric sequencings were identified and eliminated using USEARCH and the UCHIME algorithm (58). The remaining sequences were clustered into OTUs using UPARSE with a 97% sequence similarity (59). The taxonomic identity of representative sequences for each OTU was determined based on the SILVA 128 database (http://www.arb-silva.de) by the Ribosomal Database Project (RDP) classifier (60). To reduce the impact of sequencing depth, each sample was rarefied to 19,000 sequences based on the minimum sequencing number of all samples, and soil microbial beta diversity was indicated by unweighted UniFrac distance.

Statistical analysis.

The differences in environmental variables and the relative abundances of dominant phyla between farmland and grassland were assessed by t tests. Pearson correlation analyses were conducted to test the relationships between the relative abundances of the top 200 most abundant OTUs and environmental variables in both farmland and grassland, and Pearson’s P value was adjusted by the Benjamini-Hochberg correction using the “fdrtool” package in R. Rarefaction curves of OTUs were used to assess whether sequencing depth covered the majority of soil microbial taxa. Soil microbial communities were ordinated by the two-dimensional nonmetric multidimensional scaling (NMDS) analysis based on unweighted UniFrac distance, and the differences in the communities were evaluated through analysis of similarity (ANOSIM). Mantel tests based on unweighted UniFrac distance and redundancy analyses (RDA) based on OTUs were performed to identify the correlations between soil microbial community structures and environmental variables in both farmland and grassland. Rarefaction curves, NMDS analyses, ANOSIM, Mantel tests, and RDA were conducted using the “vegan” package in R. Furthermore, the distance-decay relationship was calculated by an ordinary least-squares regression between environmental distance and soil microbial community similarity (1 − unweighted UniFrac distance). To test whether the slopes of the distance-decay curve were significantly different between farmland and grassland, we used the matrix permutations, based on 9,999 permutations (61).

To evaluate within-community assembly processes, the net relatedness index (NRI) and the nearest taxon index (NTI) were calculated by the null model “taxa.labels” (999 randomizations) in the “ses.mpd” and “ses.mntd” functions of the “picante” R package, respectively (62). The NRI refers to the mean phylogenetic distance of all pairwise combinations of species, and the NTI refers to the mean phylogenetic distance of the nearest relative for all species. Positive NRI and NTI values indicate phylogenetic clustering, while the negative values represent phylogenetic dispersal (63). To quantify between-community assembly processes, the β-nearest taxon index (βNTI) and the Bray-Curtis-based Raup-Crick (RC_bray) were measured to estimate the relative contributions of deterministic and stochastic processes according to the framework proposed by Stegen et al. (64). Specifically, |βNTI| > 2 represents the influence of deterministic processes: βNTI > 2 and βNTI < −2 indicate variable selection and homogeneous selection. When βNTI is between −2 and 2, RC_bray is used to partition stochastic processes: RC_bray < −0.95, RC_bray > 0.95, and |RC_bray| < 0.95 suggests homogenizing dispersal, dispersal limitation, and undominated processes. For paired samples, the NRI and NTI between farmland and grassland were compared by paired t tests, and the differences in the βNTI were evaluated by t tests. The relationships between the NRI, NTI, and βNTI and environmental variables prone to change with land uses were analyzed by Pearson correlation analyses. Moreover, analysis of variance (ANOVA) (least significant difference method) was used to compare the βNTI values in different levels of MAP. Pearson correlation analyses, t tests, paired t tests, and ANOVA were conducted by SPSS 20.0.

Data availability.

The raw sequence data used in this study have been deposited in the DDBJ Sequence Read Archive under accession number PRJDB12153.