**1. Introduction**

Soil organic matter (SOM) consists of a heterogeneous mixture of substances in various stages of decay, mainly including plant and animal residues, microbial necromass, and new substances synthesized and released by microbes into the soil [1,2]. The global SOM pool in the surface meter stores approximately 1500 Pg carbon (C) [3] and 95 Pg nitrogen (N) [4] as well as other essential elements for plants and microbes. Therefore, SOM is critical for soil quality and ecosystem dynamics [5,6]. At the same time, SOM plays an important role in global climate change because soils could act as a potential sink for C [7,8]. Therefore, a large number of studies have investigated the effects of various climatic, edaphic and biotic factors on the dynamics of C and N in soils to better understand their turnover and SOM destabilization as well as their role in climate change [9–13].

It is difficult to explore the dynamics of SOM by direct measurement of the change in C and N stocks due to their large size [14]. With the rapid development of isotope ratio mass spectrometry [15], the analysis of stable isotope composition of soil C (δ13C) and N (δ15N) has become a powerful tool to explore the stability and dynamics of SOM [12,16–18] and soil development [12,19]. Soils δ13C and δ15N are ideally suited to provide wider insights into C and N cycles in soil ecosystems because they are primarily based on either an isotopic fractionation during microbial degradation and transformation (e.g., ammonification, nitrification and denitrification) or the preferential decomposition of the substrates depleted in 13C and 15N [20]. Generally, older and more microbially-processed SOM is enriched in 13C and 15N compared to less-decomposed substrates [18,21].

Additionally, variation in δ13C and δ15N content of SOM in natural ecosystems is largely controlled by the input of new plant residues and overall isotopic fractionation during microbial decomposition [18]. The signature of δ13C and δ15N in SOM is closely related to vegetation changes and microbial decomposition as well as anthropogenic N input [22–24]. Moreover, climatic and edaphic factors, including temperature, precipitation, pH, and contents of soil C, N and phosphorus (P) as well as soil texture, greatly impact δ13C and δ15N content of SOM [13,25,26]. As a result, the signature of 13C and 15N in SOM can be used as a valid proxy for SOM dynamics and provide integrated information about the ecosystem N cycling [9,27–32].

To understand the factors controlling δ13C and δ15N in SOM, numerous studies have investigated the patterns of soil δ13C and δ15N on regional and global scales [9–11,14,31,33,34]. It has been shown that climate controls forest soil δ13C in the southern Appalachian Mountains [13]. Climate can likewise have an effect on soil δ15N, with values increasing in response to rain events, which enhance the processes that cause the loss of N but discriminate against 15N loss [35]. Further evidence shows that aridity can nonlinearly alter soil δ15N values in arid and semi-arid grasslands [34]. Consequently, soil δ15N values along precipitation gradients can reflect the pattern of N losses relative to turnover [36–38]. On a global scale, soil δ15N converges across climate and latitudinal gradients [11]. In addition to climatic factors, substrate age, soil texture and litter input as well as land-use change also can affect soil δ13C and δ15N [25,31,39,40]. Nonetheless, the controls on C and N isotope ratios in soil still remain unclear [27].

Grasslands are an interesting ecosystem to study in this context because they store large amounts of C and N in soil [41,42] and have grea<sup>t</sup> potential to affect CO2 concentrations in the atmosphere. Additionally, grasslands are widely distributed over the world and account for 26% of the ice-free land [43]. As a result, grassland soils play an important role in the context of global climate change and regulate biogeochemical cycles [44]. Among the various types, temperate grasslands are widely distributed across the Eurasian continent and form the Eurasian steppe [45]. Recent studies of temperate grasslands showed that climatic variables control approximately 50% of the variation in soil δ15N along an east–west transect in Northern China. Soil δ15N was found to decrease with increasing mean annual precipitation (MAP) and mean annual temperature (MAT) [10]. Further studies demonstrated that the aridity can nonlinearly alter soil δ15N values [34]. Nonetheless, it remains unclear how co-varying climatic, edaphic and biotic factors control soil δ13C and δ15N in such temperate grasslands. We hypothesize that distinct factors control the soil δ13C versus δ15N signature: (1) biotic factors such as microbial biomass C (MBC) and N (MBN) as well as plant belowground biomass could

exert more impact on soil 15N than edaphic and climatic factors since 15N fractionation is largely controlled by biological processes [11]; and (2) climatic and edaphic factors have more e ffects on soil δ13C than biological factors because water can strongly a ffect 13C in plant tissues [46]. To test the above hypotheses, we collected soil and plant samples from temperate meadow steppes, temperate steppes and temperate deserts along a vegetation transect in Inner Mongolia.

#### **2. Materials and Methods**

### *2.1. Study Sites*

This study was conducted along a 1280 km transect across Inner Mongolia from west to east in northern China (Table 1). The longitude of the transect ranged from 107◦15 to 122◦17 and the latitude ranged from 38◦44 to 50◦12. The region was characterized predominantly by an arid and semi-arid continental climate. MAP ranged from 154 to 517 mm and MAT ranged from 1 to 4 ºC. The MAP and MAT of each sampling site were calculated from the NMIC (China National Meteorological Information Center). The main vegetation types across this transect were temperate meadow steppe, temperate steppe, and temperate desert. All the soils were classified as chestnut soil, corresponding to Calcicorthic Aridisol according to USDA Soil Taxonomy [47].

In 2014, a field campaign was conducted to collect soil and plant samples along this transect. In total, 22 sampling sites including six temperate meadow steppes, nine temperate steppes and seven temperate deserts were selected. At each site, four plots (1 m × 1 m) were randomly selected for soil and plant sampling. The aboveground plant parts were harvested for the estimation of aboveground net primary production (ANPP). Additionally, five soil cores (2.5 cm diameter, 5 cm depth) were collected randomly using a soil corer from each plot and mixed thoroughly as one composite sample. Living roots were carefully collected from the soil and washed by water and then dried for the estimation of belowground biomass. Soil samples were sieved through a 2.0 mm sieve and then separated into two parts: one was stored in a plastic bag and frozen at −20 ◦C for measurement of soil moisture, MBC and MBN, and the other was air-dried for measurements under natural conditions.

#### *2.2. Analysis of Soil* δ*13C and* δ*<sup>15</sup> N*

Dried soil samples were ground into powder using a ball mill (Retsch MM2; Retsch, Haan, Germany). Approximately 1 g soil was put into 5 mL centrifuge tube. To remove carbonate, 3 mL of 0.5 M HCl was added to the tubes overnight. Afterwards, samples were freeze-dried and washed with H2O until a pH of 7.0 was reached. The soil was weighed into tin capsules for analysis of total N (Nt), soil organic C (SOC), δ13C and δ15N by continuous flow gas isotope ratio mass spectrometry (isoprime precisION, Elementar, Germany). The isotope results of soil C or N were calculated as follows: δ<sup>13</sup> C/δ15N ( ‰) = (Rsample/Rstandard − 1) \* 1000, where Rsample and Rstandard are the ratios of 13 C/12C or 15 N/14N in the sample and standard. The standards for δ13C and δ15N are Pee Dee Belemnite and atmospheric molecular N, respectively. The standard deviation of repeated measurements of laboratory standards was ±0.15‰ for these isotope analyses.

#### *2.3. Analysis of Soil Properties and Microbial Biomass*

Total phosphorus (Pt) in the soil was measured using optical emission spectrometry (Optima 5300DV; PerkinElmer, Shelton, USA) after nitric-perchloric acid digestion [48,49]. Soil pH was measured by a dry soil-water ratio of 1:2. Soil MBC and MBN were determined using the chloroform fumigation–extraction method [50,51]. Briefly, 10 g fresh soil was extracted with 24 mL of 0.5 M K2SO4. An additional 10 g soil was fumigated with ethanol-free chloroform for 24 h and then extracted again in the same manner. All extracts were shaken for 1 h and filtered through 5895 paper. Total organic C and N concentrations in the K2SO4 extracts were measured with a Dimatec-100 TOC/TIC analyzer (Liqui TOCII, Elementar, Germany).



The values are means ± standard errors of 4 replicates. MAP = mean annual precipitation, MAT = mean annual temperature, SOC = soil organic carbon, N = nitrogen, P = phosphorous, MBC = microbial biomass C, MBN = microbial biomass N.

#### *2.4. Calculations and Statistics*

MBC and MBN were calculated as the di fference between the total C or total N content in fumigated and non-fumigated soils, divided by a kEC factor of 0.45 [52] and a kEN factor of 0.54, respectively [50,51].

The standard errors of means are presented in figures and tables as a variability parameter. The normality of soil δ15N and δ13C, as well as, other edaphic and biological data were tested. A one-way analysis of variance was performed with SPSS 21.0 (SPSS Inc., Chicago, IL, USA) to evaluate the e ffects of grassland type on soil δ15N and δ13C values. Correlations between soil δ15N and δ13C and climatic (MAP, MAT), edaphic (SOC, Nt, Pt) and biotic factors (MBC, MBN and belowground biomass) were analyzed with SPSS 21.0 (SPSS Inc., Chicago, IL, USA). To identify how all the factors a ffect soils δ15N and δ13C, we conducted a data analysis in two steps using R version 3.5.2 (R Development Core Team 2019). The first step was to generate a series of all possible multiple linear models based on the information-theoretic method. To avoid overfitting our models, a Pearson correlation test was conducted to identify and remove highly correlated factors (r > 0.6 or < −0.6, Table 2) within one model. The second step was to calculate estimates and the relative importance of predictors considering changes to the models' Akaike's information criterion (AIC) changes of less than 2 (model.avg function in MuMIn package) with the model averaging method [53]. Information-theoretic AIC corrected for small samples sizes (AICc), ΔAIC (di fference between AICc of one model and the model with the lowest AICc), and AICc weight (wAICc) were calculated for model ranking. All di fferences were tested for significance ( *P* < 0.05).

**Table 2.** Pearson's correlation matrix for raw input variables in explaining change in soil 15N and 13 C along the vegetation transect across Inner Mongolian temperate grasslands. The asterisks indicate a significant relationship between variables at *P* < 0.05.


MAP = mean annual precipitation, MAT = mean annual temperature, SOC = soil organic carbon, N = nitrogen, P = phosphorous, MBC = microbial biomass C, MBN = microbial biomass N, BB = belowground biomass.
