The study site (43∘13′ N, 116∘14′ E) is a typical temperate semiarid steppe at the Inner Mongolia Grassland Ecosystem Research Station (IMGERS) in the Xilin River watershed, Inner Mongolia, China. The mean annual air temperature is 0.9 ∘C, varying from -21.4 ∘C in January to 19.7 ∘C in July. The long-term mean annual precipitation is 351.4 mm, about 72.8 % of which is concentrated from May to August, according to the data from 1980 to 2013 (monitored and provided by IMGERS). The steppe is dominated by Leymus chinensis, Stipa grandis, Agropyron cristatum, and Koeleria cristata. The soil was classified as a Haplic Calcisol by the Food and Agriculture Organization of the United Nations (FAO) soil classification system, with a depth of 100–150 cm and a composition of 21 % clay, 60 % sand, and 19 % silt on average (Hao et al., 2013). The site has experienced an uncontrolled heavy sheep grazing since the 1980s and has been fenced since 1999. No fertilizer was applied before the experiment was conducted.

The experiment was set up in 2008 following a randomized block design. There were nine N addition intensities (0, 1, 2, 3, 5, 10, 15, 20, and 50 g N m-2 yr-1) crossed with two N addition frequencies (2 times per year vs.12 times per year) and two mowing regimes (unmown and mown). Currently, infrequent N addition (i.e., one or two times per year) is commonly used in manipulative experiments to mimic N deposition, which is quite different from natural N deposition occurring continuously throughout the year (Smith et al., 2009). Therefore, a more frequent and even way of N addition (i.e., 12 times per year) was set to simulate natural N deposition (Aneja et al., 2001) and to compare whether the changing frequency of N input affects the grassland ecosystem. Higher rates of N addition were used to mimic accumulative N deposition in the long term and/or extreme N inputs in the future of the grassland (Zhang et al., 2017b). NH4NO3 (> 99.5 %) was added in wet and dry forms to simulate the wet and dry N deposition, respectively. For the high-frequency treatments, NH4NO3 was added monthly since 1 September 2008. For the low-frequency treatments, NH4NO3 was added each year on 1 June and November since November 2008. During the growing season (from May to October), N was added in wet form by mixing NH4NO3 with purified water (9.0 L water in total for each plot, either 9.0 L once in June for low frequency or 1.5 L monthly from May to October for high frequency) and then sprayed evenly with a sprayer. To simulate dry N deposition in the nongrowing season, NH4NO3 was broadcast after being mixed with treated sand to ensure even fertilization. Specifically, 500 g of sand was used once in November for low-frequency treatments, and 80 g of sand was used monthly from November to the next April for high-frequency treatments and then broadcast evenly in every plot. The sand used in this experiment was sieved through a 1 mm sieve, dipped in hydrochloric acid, washed in purified water, and then oven dried at 120 ∘C for 48 h to remove acid-soluble S. Annual mowing was conducted at the end of August using a hay mower at 10 cm height to simulate the local hay-cutting management. The aboveground plant residue was taken away immediately after mowing. We also set an unnamed control without any treatment (N addition, mowing, water, or sand addition) to determine the impacts of water and sand addition, which were also compared with mowing treatment without any other treatments. Thus, there were 38 treatments with 10 replicate blocks for every treatment. Each plot is 8 m × 8 m and separated by a 1 m buffer zone.

We assessed aboveground biomass in each plot by clipping a 1 m × 1 m quadrat above soil surface in late August 2014. All the living plants were clipped and sorted by species, dried, and weighed. The plant samples were washed using deionized water and then dried to constant weight at 65 ∘C for 48 h.

Soil samples were collected from each of the 380 plots at the beginning of August 2014 (i.e., after 6 years of treatments). A mixed sample was taken randomly from five cores of the topsoil (0 to 10 cm) within each plot. Then, samples were passed through a 2 mm sieve immediately to remove the plant residues and then air dried for further analysis.

Soil pH was determined in a soil slurry at 2.5:1 (w/v) water / soil ratio by a digital pH meter (Precision and Scientific Instrument Co. Ltd., Shanghai, China) and the data are given in Table 1 reported by Wang et al. (2018). The concentration of soil organic carbon (SOC) in the topsoil was determined by oxidation using a mixture of K2Cr2O7 solution and sulfuric acid and titration using FeSO4 (Nelson and Sommers, 1982). Soil total inorganic nitrogen (TIN) concentration was calculated as the sum of ammonium and nitrate, which was extracted with 2 M KCl and determined using a continuous flowing analyzer (SAN++ segmented flow analyzer, Skalar, the Netherlands). Soil pH, SOC, and TIN were previously reported in Wang et al. (2018). Soil total sulfur (TS) concentration was analyzed with an elemental analyzer (Vario MACRO cube, Elementar Analysensysteme GmbH, Germany).

A schematic diagram is shown to illustrate the transformation processes and extraction and calculation procedures of the S fractions (Fig. 1). Soluble S, available S, and total inorganic S were extracted with 0.01 M CaCl2, 0.01 M Ca(H2PO4)2, and 1 M HCl at a 5:1 (w/v) water / soil ratio, respectively (Roberts and Bettany, 1985). Briefly, 5.0 g of the air-dried soil was mixed with 25 mL extractant and shaken at 400 rpm for 60 min at 25 ∘C. The extracts were filtered and digested with 1 mL of 30 % H2O2 for 20 min to remove the color of soil organic matter. The Ca2+ in extractants of CaCl2 and Ca(H2PO4)2 can effectively depress the solubility of organic matter during the extraction. The cooled solutions were then mixed with 0.5 mL of HCl (1:4) and 1 mL of gum acacia solution (to stabilize the suspension) successively and adjusted to 25 mL. The SO42- in all the extracts was quantified by turbidimetry with 0.5 g of BaCl2 crystals at 440 nm using a UV–VIS spectrophotometer (UV-1700, Shimadzu, Japan) (Tabatabai and Bremner, 1972). Adsorbed S, total inorganic S, and organic S concentrations were calculated as follows: 1adsorbedS=availableS-solubleS,2insolubleS=totalinorganicS-availableS,3organicS=totalS-totalinorganicS.

We measured total S concentration in two dominant species of Leymus chinensis and Stipa grandis in N addition plots of 0 and 15 g N m-2 yr-1 under low and high frequencies of N addition with and without mowing. Briefly, 0.3 g plant samples were acid digested with a 1:2 (v/v) mixture of 65 % nitric acid and 72 % perchloric acid at around 235 ∘C (Soon et al., 1996). The S concentration of digestion solution was quantified by turbidimetry at 440 nm using a UV–VIS spectrophotometer (UV-1700, Shimadzu, Tokyo, Japan). Plant S uptake of the dominant species was calculated as follows: 4Suptake=∑Si×mi, where Si represents S concentration in plant species i and mi represents aboveground biomass of the corresponding species.

The Kolmogorov–Smirnov test and Levene's test were executed to determine the normality of data and homogeneity of variances, respectively. The TIN concentration was log10-transformed for homogeneity. We used three-way ANOVAs to determine the effects of N addition intensity, N addition frequency, mowing, and their interactions with the concentrations of soil inorganic S fractions and total S. A Student's t test was performed to estimate the difference in plant biomass and S uptake between two N frequencies within each N addition intensity with or without mowing. Correlation analyses were conducted to estimate the relationships between soil parameters and concentration of soil S fractions. The proportion of S fractions relative to total S was calculated and Duncan's multiple range test was employed to estimate differences among treatments. All the analyses above were conducted using SPSS 18.0 (SPSS Inc., Chicago, IL, USA) and all statistical significance was accepted at P<0.05. Moreover, we calculated the response ratio of available S concentration to mowing as follows: 5responseratio=SmownSunmown, where Smown and Sunmown represent available S concentration in mown and unmown plots, respectively. Weighted log-response ratio (logeRR) and its 95 % confidence intervals for the effect of mowing were calculated using the metafor package in R software, version 3.5.1 (Viechtbauer, 2010). Confidence intervals not overlapping zero indicated significant mowing effects on available S concentration.

Structural equation modeling (SEM) was conducted to examine the direct and indirect strengths of N addition intensities and frequencies on soil inorganic S fractions through the changes in soil parameters with the AMOS 24.0 program (Amos Development Co., Greene, Maine, USA). Data from all the treatment were pooled for the SEM analysis. An a priori model (Fig. S1a in the Supplement) was constructed by assuming that soil S fractions could be directly affected by N addition frequency, intensity, and mowing, or indirectly by altering soil pH, plant biomass return, and organic S mineralization (see hypotheses in the Introduction). Insignificant pathways and parameters that had no effect on inorganic fractions were excluded from the model sequentially to obtain the final model (Fig. S1b, c). Data were fitted to the model using the maximum likelihood estimation method. We used χ2 test (P>0.05), root-mean-square error of approximation (RMSEA, <0.08), and Akaike information criteria (AIC) to evaluate the adequacy of the model.

Soil pH significantly decreased and the TIN concentration increased along the N gradient under all treatments of addition frequencies and mowing regimes (Table 1). Soil pH exhibited a sharper decrease in low N frequency compared with high N frequency in unmown plots, especially at N addition intensities of 10 and 15 g N m-2 yr-1. Low frequency of N addition increased TIN concentration in both unmown and mown plots at high N levels (P<0.05 at 50 g N m-2 yr-1). The SOC concentration decreased along the N gradient only under low N frequency in mown plots.

Mowing significantly decreased soil-available S concentration by up to 43 % and 40 % across N intensity at low and high N addition frequencies, respectively (Fig. 3a, b; Table 2). For mown plots, high frequency of N addition significantly increased soil-available S concentration by up to 57 % compared with low N frequency across N addition intensity (Fig. 3b; Table 2). The high intensity of N addition increased soil-available S concentration at the lower frequency of N addition for unmown plots from 15.9 to 24.0 mg per kilogram soil (Fig. 3a, P<0.01), and at both low (from 10.7 to 15.6 mg per kilogram soil, P=0.01) and high (from 12.0 to 23.0 mg per kilogram soil, P<0.01) frequencies of N addition for mown plots (Fig. 3b). This resulted in significant interactive effects of M×F, F×N, and M×F×N on soil-available S (Table 2).

Nitrogen addition intensity enhanced the negative effect of mowing on soil-available S concentration at low N frequency (Fig. 3c, P<0.01). However, negative mowing effect on available S concentration decreased along the increasing N addition intensity and even turned into positive at 15, 20, and 50 g N m-2 yr-1 at high frequency of N addition (Fig. 3c, P<0.01). The proportion of available S relative to total S increased with increasing N intensity at low N frequency for unmown plots (Fig. S2a) and at both low and high N frequencies for mown plots (Fig. S2c, d).

Mowing significantly decreased total inorganic S at both low and high frequencies of N addition by up to 38 % (Fig. 4a, b). The high frequency of N addition significantly increased soil total inorganic S concentration by as much as 15 % (unmown plots) and 26 % (mown plots) compared with low N frequency (Fig. 4a, b). Soil total inorganic S concentration decreased with increasing N intensity (except for 20 and 50 g N m-2 yr-1) for both N frequencies in unmown plots, while it showed no response in mown plots. Soil organic S and total S concentrations were not affected by the intensity of N addition, frequency of N addition, and mowing treatment as suggested by three-way ANOVAs (Table 2). However, the mean values of soil organic and total S concentration tended to decrease with increasing N addition intensity as indicated by its linear correlation with N addition intensity at both low and high N addition frequencies in mown plots (Fig. 4d, f). Mowing significantly decreased the proportion of total inorganic S relative to total S and increased organic S proportion (Figs. S2a vs. 2c and Figs. S2b vs. 2d).

The high intensity of N addition increased aboveground biomass regardless of N addition frequency and mowing management (Fig. S3). The low frequency of N addition increased aboveground biomass at 2 and 15 g N m-2 yr-1 compared with high frequency of N addition in unmown plots. Sulfur uptake of Leymus chinensis and Stipa grandis showed no response to mowing management in plots without N addition, but it increased with mowing under both frequencies of N addition. Low frequency of N addition increased S uptake of the dominant species at 15 g N m-2 yr-1 in both mown and unmown plots (Fig. 5). Nitrogen addition intensity increased plant S concentration of two dominant species (Fig. S4) and relative biomass proportion of Stipa grandis (Fig. S5).

Soil pH was negatively correlated with adsorbed S and available S concentrations under both low (Fig. 6a) and high frequencies of N addition (Fig. 6b) (all P<0.01). However, insoluble S was positively correlated with pH under both N frequencies. Soil TIN concentration was positively correlated with adsorbed S and available S but negatively correlated with insoluble S concentration under both low and high frequencies of N addition. Soil organic carbon was positively correlated with soluble S under both N frequencies. Insoluble S was negatively correlated with adsorbed and available S under both N frequencies (all P<0.01, Fig. 6a, b). Organic S concentration was negatively correlated with adsorbed S and available S at high frequency of N addition (P<0.01, Fig. 6b).

Results of SEM showed that N addition intensity affected soil inorganic S fractions through altering soil pH, plant biomass, and SOC concentration (Fig. 7a, b). Plant biomass, representing plant S uptake, showed a negative effect on adsorbed S concentration. With increasing intensity of N addition, higher plant biomass increased soil organic carbon, which elevated soil-soluble S concentration (Fig. 7b). The frequency of N addition directly and positively affected soil-soluble S (standardized total effect size: 0.26) but indirectly influenced adsorbed S (effect size: 0.07) and inorganic S (effect size: 0.05) concentrations by changing soil pH (Fig. 7c). Consistent with the results of simple regressions, soil pH had a negative effect on soil adsorbed S, which was positive for soil-insoluble S. Mowing decreased soluble S, adsorbed S, and insoluble S concentrations (Fig. 7d).

Consistent with the first hypothesis, increasing N addition intensity enhanced soil-available S fractions, especially adsorbed S concentrations. In this study, higher adsorbed S under N addition was due to the fact that the level of sulfate adsorption increased with the decrease in soil pH (Fig. 7b), which was in line with Nodvin et al. (1986). Adsorption of SO42- is pH dependent as anionic groups from SOM compete with SO42- for adsorption sites on iron and aluminum hydroxides (Johnson and Todd, 1983). Under acidic conditions, soil matrix can provide adsorption sites with positive charges to attract the negatively charged SO42- (Tabatabai, 2005). Therefore, lower soil pH contributed to higher adsorbed S concentration via enhancing adsorption strength and increasing electrostatic potential of the adsorption sites under higher intensities of N addition (Scherer et al., 2012). However, unaffected adsorbed S in unmown plots could be possibly due to the fact that soil pH tended to be higher at high N frequency compared with low N frequency at the same N addition level (significant at 10, 15, and 20 g N m-2 yr-1; see Table 1). Moreover, soil pH decreased at a much lower rate along with increasing N addition intensity (significant decreases only detected at 20 and 50 g N m-2 yr-1) under high N addition frequency compared with low N addition frequency. This resulted in weaker S adsorption strength, less SO42- release from insoluble S dissolution, and consequently unchanged adsorbed S concentration with increasing N intensity at high N frequency.

Increasing N addition intensity enhanced soil-available S fractions partially due to the dissolution of soil-insoluble S as suggested by the negative correlation between the two S fractions in our study (Fig. 6). Without N deposition, free sulfate in soil could precipitate as calcium-, magnesium-, or sodium sulfate and co-crystallize or co-precipitate with CaCO3 (Tisdale et al., 1993), especially in calcareous soil rich in exchangeable Ca2+, Mg2+, and Na+ (Wang et al., 2018). Soil acidification induced by N addition can reverse the precipitation process by dissolving CaCO3–sulfate complexes (Chen et al., 1997; Zhang et al., 2016). Therefore, insoluble S dissolution sequentially enhanced SO42- mobility due to soil acidification under excessive N input. This postulation was further confirmed by the positive relationships of soil pH and insoluble S at both N frequencies (Fig. 6a, b). For the insoluble S fraction, cautions should be paid to the fact that a small part of organic S might be oxidized into inorganic S during H2O2 (1 mL) digestion (Williams and Steinbergs, 1959). At low N frequency of N addition, the non-significant effect of N intensity on soluble S concentration could probably attribute to the balance between S adsorption and dissolution processes under soil acidification thus keeping the concentration of soluble S stable. Soluble S is the most active and mobile S fraction in topsoil for it can be easily utilized by plants and leached along with soil pore water (Tabatabai, 2005). The increase in soluble S with a high frequency of N addition indicates that N deposition might increase the risk of S losses via leaching and/or surface runoff.

Our results demonstrate that N input increased soil-available S potentially by increasing organic S mineralization in mown plots as indicated by the decreased organic S concentration (Fig. 4d). This was consistent with previous observations that N addition enhanced mineralization of organic S to increase S availability by elevating microbial activity (Ghani et al., 1992). Soil N availability has considerable impacts on the mineralization of organic S (Niknahad-Gharmakher et al., 2009). The increase in soil TIN concentration following N input possibly accelerated organic S mineralization in this study. Changes in SOC concentration also contributed to the increase in soluble S (Fig. 7a) as biological mineralization of C-bonded S was positively related to SOC concentration (Niknahad-Gharmakher et al., 2009). Moreover, decrease in soil pH and higher plant S uptake under N input (Fig. 5) could promote biochemical mineralization of organic S via the enhancement of arylsulfatase secretion by soil microorganisms (McGill and Cole, 1981). This was further confirmed by Niknahad-Gharmakher et al. (2009), who reported the upregulation of soil organic S mineralization by the decrease in soil pH. Elevation of soil S availability via the two pathways of insoluble S dissolution and organic S mineralization implies that N deposition would substantially promote transportation of soil S into plant biomass in the semiarid grassland. This finding is relevant to understand the fates of S under ecosystem N enrichment with the presence of other disturbances such as mowing and livestock grazing of grasslands.

Partially contrary to our second hypothesis, low frequency of N addition decreased soluble S, adsorbed S, available S, and total inorganic S concentrations compared with high-frequency addition. A change in soil-available S concentration was mainly determined by the input from dissolution of insoluble S and output to plant uptake and leaching. A sharper decrease in soil pH with low N frequency relative to high N frequency would result in higher soil-available S concentration via the enhancement of insoluble S dissolution. On the contrary, lower available S concentration was found under low N frequency compared with high N frequency (Fig. 3), which could be possibly driven by higher plant S uptake (Fig. 5), surpassing the amount of S dissolution. Leaching loss of SO42- was evident with infrequent and extreme rainfall pulses in sandy soils (Eriksen and Askegaard, 2000). Therefore, another potential explanation could be that large-pulse water input at low N frequency resulted in higher leaching loss of available S than the high N frequency of adding a small amount of water each time. The results in the control plots supported this explanation, with lower concentrations of adsorbed S, available S, and total inorganic S under low N frequency compared with high-frequency (especially in unmown plots). With the increasing N intensity, leaching loss of available S was exacerbated due to enhanced insoluble S dissolution (Fig. 2e, f). Compared with high N frequency, organic S mineralization did not contribute to lower total inorganic S and available S concentrations at low N frequency for the same N intensity as no difference was detected for organic S concentration between the two N frequencies. These results suggest that using low frequency of N addition to mimic N deposition may overestimate insoluble S dissolution (especially in unmown plots) and plant S uptake in temperate grasslands. However, some of the S fractions responded differently to both N intensity and frequency with or without mowing treatment, suggesting that the effects of N addition strongly depended on mowing.

Mowing decreased inorganic S fractions in all treatments, which supports our third hypothesis. Mowing could alter plant community composition (Lü et al., 2012) and ecosystem nutrient cycling (Koncz et al., 2015). Decreased soil nutrient availability was found with mowing in a similar grassland ecosystem resulting from reduced plant residue return (Lü et al., 2012). Similarly, mowing resulted in significant decreases in inorganic S fractions and the proportion of total inorganic S relative to soil total S (Figs. 4a, b, S2). In contrast, relative organic S proportion increased with mowing treatment (Fig. S2), which indicates that mowing management removes soil S out of the ecosystem and has a larger impact on inorganic S transformation than on the organic S mineralization (Fig. 4a, b vs. Fig. 4c, d). This is contrary to the findings of Li et al. (2001), which suggested that over 70 % of S removal by harvests was derived from organic S mineralization in an agroecosystem. The difference could probably be attributed to the relatively lower intensity of plant S uptake and fertility of soil of the grassland ecosystem than the cropland.

The negative mowing effect on the soil-available S was suggested to be exacerbated due to enhanced plant S uptake coincident with higher plant biomass (Jackson, 2000). We found that combined mowing and N addition treatments tended to increase S uptake of the dominant species (Fig. 5). This was probably due to higher S concentration in Stipa grandis (Fig. S4) and increases in its biomass proportion after mowing (Fig. S5). However, the exacerbation of the negative mowing effect on available S was only detected at low frequency of N addition in this study, and high N frequency alleviated the negative mowing effect and even turned it into a positive effect (> 15 g N m-2 yr-1) with the increasing N addition intensity (Fig. 3c). The discrepancy might be recognized as the tradeoff between the available S output and input processes, which are regulated by differential responses of soil pH, plant biomass, and S uptake in the two N addition frequencies. Therefore, it was reasonable to detect the decrease in soil total S but the increase in soluble S and adsorbed S concentrations under N addition with mowing. Our results suggest that the increase in S removal from the grassland ecosystem could temporally stimulate available S formation via abiotic dissolution of insoluble S and mineralization of organic S. In the long term, however, N enrichment together with continuous mowing would cause depletion of the S pool.