A field experiment was conducted at the Ansai Research Station (ARS) of the Chinese Academy of Sciences (36∘51′30′′ N, 109∘19′23′′ E; 1068–1309 m a.s.l.), northern Loess Plateau, China. The mean annual temperature of the study area is 8.8 ∘C, and the mean annual precipitation is approximately 505 mm, with >70 % concentrated from July to September. Annual evaporation ranges from 1500 to 1800 mm. To control soil erosion and improve the ecological environment, the Chinese government has implemented the policy of converting sloping cropland to grassland in the region in the 1990s. Synchronously, restoration of the local grassland is mainly dependent on abandoned farmland. In order to study the effect of slope aspect on the soil microbial community in the restored grassland, three grassland areas abandoned in the same year were selected for the experiment. The main vegetation in the region includes woods such as Robinia pseudoacacia and Platycladus orientalis; shrubs such as Caragana korshinskii, Hippophae rhamnoides, Syzygium aromaticum, and Ostryopsis davidiana; and herbage such as Artemisia sacrorum, Bothriochloa ischaemum, Setaria viridis, Artemisia giraldii, and Artemisia capillaris. Details of the soil properties and a map of sampling sites were described by Ai et al. (2017a).

The representative slopes of the three grassland areas were south-facing (S15∘ W), northwest-facing (N75∘ W), and north-facing (N57∘ E), which had the same site conditions (all with Artemisia sacrorum as the dominant species, same rehabilitation age, geographical proximity) and represented sunny slope, half-sunny slope, and shady slope, respectively. The three study areas were selected in September 2014 after consultation with ARS researchers and reviewing relevant land documents. The basic characteristics are shown in Table 1.

Three replicate 10 m×10 m plots were established at each site (A. sacrorum was the dominant plant at each site). The distance between sampling plots within each sampling site was not less than 20 m. Each plot was first surveyed for latitude, longitude, elevation, slope aspect, and slope gradient. Three 1 m×1 m quadrats were then randomly set in each plot to characterize the vegetation, e.g. plant species, coverage, and number. The plants were removed, and the soil strongly adhering to the roots, i.e. RS, was collected (0–20 cm soil layer). Soil was also sampled from the same layer at locations approximately 15 cm from the plant roots (i.e. NRS). Each NRS sample was a composite of subsamples collected at five points (the four corners and the centre of the plot). A total of 18 soil samples (3 sites × 3 plots per site × 2 soil types) were collected, and each was divided into two subsamples: one subsample was placed in a cool container, and the other was placed into a cloth bag. The samples were then taken to the laboratory, and gravel and coarse fragments were removed. The container samples were homogenized and sieved to 2 mm and were also divided into two subsamples: one subsample was stored at -80 ∘C, and the other was stored at 4 ∘C until analysis. The samples in the cloth bags were air-dried and sieved to 0.25 and 1 mm prior to analysis.

The samples stored at 4 ∘C were used for determining MBC content (mg kg-1), basal respiration (BR; mg kg-1 h-1), and substrate-induced respiration (SIR; mg kg-1 h-1). Microbial biomass was measured by chloroform fumigation (Vance et al., 1987). The soil samples were fumigated for 24 h at 25.8 ∘C with CHCl3 (ethanol free) after the fumigation or non-fumigation treatments and then were extracted with 100 mL of 0.5 M K2SO4 by horizontal shaking for 1 h at 200 rpm and then filtered. The amount of K2SO4-extracted organic C was determined by a liquiTOCII analyser (Elementar, Hanau, Germany), and MBC content was calculated using a kEC (coefficient for extracting microbial carbon from the soil) factor of 0.38 (Vance et al., 1987). The soil BR was estimated by measuring the CO2 evolution from 10.0 g of field-fresh soils. The homogenized soil samples were first placed in a polyethylene bottle with rubber stopper (the soil water content was adjusted to 50 % of field water-holding capacity). The polyethylene bottle was then incubated at 28 ∘C for 2 h, and the CO2 evolution was measured by an infrared gas analyser (QGS-08B, Beijing, China; Hueso et al., 2011). Soil SIR was determined using the same method as for BR but with the addition of 0.06 g glucose to the soil, after the glucose and soil were fully compounded, they were then incubated at 28 ∘C for 1 h.

The soil stored at -80 ∘C was used for the determination of PLFA contents. The structures of the microbial communities were determined using a method (Bligh and Dyer, 1959) modified by Bardgett et al. (1996). Briefly, fatty acids were extracted from 3.0 g of freeze-dried soil using a solution containing citrate buffer, chloroform, and methanol. The PLFAs were separated from neutral and glycolipid fatty acids by solid-phase-extraction chromatography. After mild alkaline methanolysis, the PLFAs were analyzed using a gas chromatograph (GC7890A, Agilent Technologies Inc., Wilmington, USA) equipped with MIDI Sherlock software (version 4.5; MIDI Inc., Newark, USA).

An external standard of 19 : 0 methyl ester was used for quantification (Frostegård et al., 1993), and the amounts were expressed as nmol g-1 for dry soil. Zelles (1999) reported that specific PLFA signatures could serve as indicators of specific microbial groups. Total PLFAs were obtained by summing the contents of all fatty acids detected in each sample. The classification of the PLFAs are shown in Table 2.

The concentrations of soil organic carbon (SOC), total nitrogen (TN), and total phosphorus at the sites have been reported by Ai et al. (2017a). Soil pH and available phosphorus (SAP), ammonium N (NH4), nitrate N (NO3), water-soluble organic C (WSOC), water-soluble NH4 (WNH4), and water-soluble NO3 (WNO3) contents were measured as described by Ai et al. (2017b). The soil moisture contents of sampling sites during the investigation were determined gravimetrically by drying the samples to a constant weight at 105 ∘C, and then the water content was expressed as a percentage of the soil dry weight. The characteristics of the rhizospheric and non-rhizospheric soils are shown in Table 3.

The metabolic quotient (103 h-1) was calculated as BR per unit MBC: metabolic quotient =103×BR/MBC=103×(mgkg-1h-1)/(mgkg-1) (Anderson and Domsch, 1993). RE was calculated as RE = Rs / NRs, where Rs is a microbial property in RS, and NRs is a microbial property in NRS (Mukhopadhyay et al., 2016). For example, the RE for MBC: RE = RS MBC / NRS MBC = (mg kg-1) / (mg kg-1). All data were analyzed using one-way ANOVAs, followed by Duncan's tests at a probability level of P<0.05 for multiple comparisons. All statistical analyses were performed using SPSS 20.0 (SPSS Inc., Chicago, USA), and structural equation models were analyzed using the AMOS SPSS expansion pack. A redundancy analysis (RDA) was performed using CANOCO 5.0 (Biometris, Wageningen, the Netherlands). The graphs were plotted using SigmaPlot 12.5 (Systat Software, San Jose, USA).

RS MBC content did not differ significantly among the slope aspects, but NRS MBC content in the north-facing slope was higher than those in the south- and northwest-facing slopes (Fig. 1a). The RE for MBC in the south-facing slope was highest among the slope aspects (Fig. 2a). Slope aspect did not affect BR, BR / MBC, or SIR in either RS or NRS (Fig. 1b and Table 4). The RE for BR did not differ significantly among the slope aspects (Fig. 2a). The RE for SIR in the south-facing slope was higher than that in the north-facing slope.

The microbial PLFA contents in RS differed significantly among the slope aspects. Total PLFA contents in the north- and northwest-facing slopes were 115 % and 88 % higher, respectively, than that in the south-facing slope (Fig. 3a). Fungal PLFA content did not differ significantly among the slope aspects (Fig. 3a). Bacterial PLFA content was similar to the trend for total PLFA content, with the lowest content in the south-facing slope (Fig. 3b). In contrast to total PLFA content, the ratio of fungal PLFA content to bacterial PLFA content (F/B ratio) in the south-facing slope was significantly higher than those in the north- and northwest-facing slopes (Table 4). Both G+ and G- PLFA contents had trends similar to that of the bacterial PLFA content, with the lowest contents in the south-facing slope (Fig. 3b). The ratio of G+ PLFA content to G- PLFA content (G+ / G- ratio) did not differ significantly among the slope aspects (Table 4). Actinomycete PLFA content in the north-facing slope was 102 % higher than that in the south-facing slope, similar to that of G- PLFA content (Fig. 3a).

The composition of the NRS PLFA contents also differed significantly among the slope aspects. Total PLFA content in the north-facing slope was 50 % and 62 % higher than those in the south- and northwest-facing slopes, respectively (Fig. 3c). Bacterial PLFA content had a trend similar to that of total PLFA content, with the highest content in the north-facing slope (Fig. 3d). Fungal PLFA content in the south- and north-facing slopes was significantly higher than that in the northwest-facing slope (Fig. 3c). The F/B ratio in the south-facing was substantially higher than that in the north-facing slope (Table 4). G- PLFA content had a trend similar to that of bacterial PLFA content, and G+ PLFA content did not differ significantly among the slope aspects (Fig. 3d). The G+ / G- ratio did not differ significantly among the slope aspects (Table 4). Actinomycete PLFA content had a trend similar to that of fungal PLFA content, with higher contents in the south- and north-facing slopes, which were 149 % and 117 % higher, respectively, than that in the northwest-facing slope (Fig. 3d).

The REs for total, bacterial, G+, G-, and actinomycete PLFA contents differed significantly among the slope aspects, but the RE for fungal PLFA content did not (Fig. 2b and c). The REs for total, G+, G-, bacterial, and actinomycete PLFA contents in the northwest-facing slope were highest among the slope aspects.

The constrained RDAs indicated that environmental factors affected RS microbial characteristics (Fig. 4a). The total variation was 6.10, and the explanatory variables accounted for 96.8 %. The first two axes (RDA1 and RDA2) explained 89.6 % of the total variance, wherein 84.1 % was attributed to RDA1 and 5.5 % to RDA2. WSOC content was the most significant of the seven environmental factors and explained 63.6 % (P=0.006) of the total variance. The slope aspect was the next most significant environmental variable and explained 62.8 % (P=0.004), followed by NH4 (58.6 %, P=0.004), SAP (45.7 %, P=0.022), and WNH4 (45.2 %, P=0.032) contents.

The constrained RDAs indicated that environmental factors affected NRS microbial characteristics (Fig. 4b). The total variation was 2.97, and the explanatory variables accounted for 94.2 %. RDA1 and RDA2 explained 81.6 % of the total variance, 68.3 % for RDA1 and 13.3 % for RDA2. Among the seven environmental factors, WNO3 content was the most significant and explained 34.7 % (P=0.04) of the total variance.

The final structural equation model based on all indices adequately fitted the data to describe the effects of the environmental factors on RS microbial characteristics (x2=0.506; P=0.918; RMSEA, P<0.001; Fig. 5a). The final model accounted for 99 % of the variation in RS WSOC content, with 71 % of the variation in bacterial PLFA content, 78 % of the variation in G+ PLFA content, and 72 % of the variation in total PLFA content. Slope aspect was positively correlated with WSOC content (P<0.001). WSOC content was negatively correlated with bacterial PLFA (P<0.001), G+ PLFA (P<0.001), and total PLFA (P<0.001) contents.

All indices adequately fitted the data to describe the effects of the environmental factors on NRS microbial characteristics (x2=3.222; P=0.521; RMSEA, P<0.001; Fig. 5b). The model was able to explain 59 % of the variation in WNH4 content, 58 % of the variation in MBC content, 55 % of the variation in G- PLFA content, and 45 % of variation in total PLFA content. Slope aspect was strongly positively correlated with WNH4 content (P<0.001). WNH4 content was strongly negatively correlated with MBC (P<0.001), G- PLFA (P<0.05), and total PLFA (P<0.05) contents.

Soil microbial biomass is closely associated with soil-moisture content (Zhang et al., 2005; Drenovsky et al., 2010; Ma et al., 2015). The north-facing slope contained more moisture than the south-facing slope (Sariyildiz et al., 2005), so microbial activity in the north-facing slope was higher than that in the south-facing slope. NRS MBC content in the north-facing slope was significantly higher than that in the south-facing slope in our study, supporting our hypothesis 1 and in agreement with other studies (Huang et al., 2015; Sidari et al., 2008). Carletti et al. (2009), however, reported an opposite trend: soil MBC content was higher in a south-facing slope. This disparity may have been due to the differences in plant species, soil type, and regional climate (Gilliam et al., 2014). NRS MBC content in the northwest-facing slope was lower than that in the north-facing slope, inconsistent with Huang et al. (2015), whose study area had the same soil and climatic conditions as ours. The different result may mainly be due to the different plant species: the effect of shrubland plants (Huang et al., 2015) on NRS may be different from the effect of grassland plants (our study). Plant shade can affect soil microbial activity (Blok et al., 2010), so different shading can lead to different MBC contents.

Plant roots release a high amount of exudates, such as sugars, amino acids, organic acids, hormones, and enzymes (Zhang et al., 2012; Grayston et al., 1997). In contrast, soil with a low amount of shading is prone to desiccation (Wang et al., 2008), and the quantity of exudates released by plant roots is low, which may lead to lower activities of the microorganisms. RS MBC content therefore should be higher than NRS MBC content, consistent with our results. In our study, the RE for MBC content in the south-facing slope was significantly higher than that in the north-facing slope. The light in south-facing slopes would have a greater impact on the soil micro-environment than that in north-facing slopes, because south-facing slopes in the Northern Hemisphere receive more sunlight than north-facing slopes, leading to a larger difference between RS and NRS in south-facing slopes. The RDA and path analysis found that NRS WSOC, WNO3, and WNH4 contents were well correlated with MBC content (Figs. 4b and 5b), supporting our hypothesis that soil C and N are the main soil nutrient factors that affect RS and NRS microbial communities, and in agreement with other studies (Haynes, 2000; Huang et al., 2014).

Neither RS nor NRS BR, SIR, and BR / MBC differed significantly among the slope aspects, indicating that the actual microbial activities, potential microbial activities, and bioenergetic status of the microbial biomass (Nannipieri et al., 1990; Wardle, 1995; Sinha et al., 2009) were similar among the slope aspects in the study area. The RE of SIR in the south-facing slope was 96 % higher than that in the north-facing slope, indicating that the effect of slope aspect on the RS and NRS SIRs was more evident in the south-facing slope than that in the north-facing slope, even though the influence of slope aspect on SIR was not significant either in RS or NRS.