Nonlinear response of soil respiration to increasing nitrogen additions in a Tibetan alpine steppe

Nitrogen (N) availability is a key regulator of carbon (C) cycling in terrestrial ecosystems. Anthropogenic N input, such as N deposition and fertilization, increases N availability in soil, which has important implications for an ecosystem’s C storage and loss. Soil respiration (Rs), which is the second largest C flux from terrestrial ecosystems to the atmosphere, plays an important role in terrestrial C cycles. The direction and magnitude of the responses of Rs and its components to N addition have been widely evaluated, but it remains unclear how these processes change across multiple N addition levels. Here we conducted a two-year field experiment to examine the changes of Rs and its autotrophic respiration (Ra) and heterotrophic respiration (Rh) components along a gradient of eight N levels (0, 1 2, 4, 8, 16, 24, 32 g m−2 yr−1) in a Tibetan alpine steppe, and used structural equation modeling (SEM) to explore the relative contributions of biotic and abiotic variables and their direct and indirect pathways regulating the Ra and Rh. Our results indicated that both Rs and Ra exhibited first increasing and then subsequent decreasing trends at the threshold of 8 g N m−2 yr−1. In contrast, the Rh declined linearly with the N addition rate continuously increasing. SEM analysis revealed that, among various environmental factors, soil temperature was the most important one modulating Rs, which not only had a direct effect on the two Rs components, but also indirectly regulated the Ra and Rh via root and microbial biomass. These findings suggest that the nonlinear response patterns of Rs should be considered for better predicting terrestrial C balance, given that anthropogenic N input to the terrestrial ecosystems is increasing continuously.


Introduction
Nutrient availability is recognized as a key regulator of terrestrial carbon (C) balance (Fernández-Martínez et al 2014, Campioli et al 2015, Wieder et al 2015. Nitrogen (N), a very important nutrient for plants and soil organisms, usually limits the productivity of most terrestrial ecosystems (LeBauer and Treseder 2008). Due to accelerating anthropogenic activities over the past few decades, reactive N entering terrestrial ecosystems has increased continuously (Galloway et al 2008, Bodirsky et al 2014, which substantially alters ecosystem C cycling processes (Janssens et al 2010, Liu and Greaver 2010, Zhou et al 2014. Numerous studies using N addition experiments have examined how changes in N availability affect diverse C processes (Janssens et al 2010, Liu and Greaver 2010, Yue et al 2015. Among those studies, most focus on the direction and magnitude of C dynamic responses to elevated N (Janssens et al 2010, Liu and Greaver 2010).
However, as N continues to increase, ecosystems generally experience limited to saturated N (Aber et al 1989N (Aber et al , 1998, and the nonlinearity of the N-induced changes in ecosystem processes has always been observed (Arens et al 2008, Bradford et al 2008, Vivanco et al 2015. Thus, better understanding the nonlinear responses of various C processes to continuously increasing N inputs is of paramount importance for accurately predicting future terrestrial C balance.
Soil respiration (Rs), as the second largest C flux between terrestrial ecosystems and the atmosphere, plays an important role in terrestrial C cycles Zhou 2006, Janssens et al 2010). Though some prior studies examined the nonlinear responses of aboveground C processes to multiple N loads, such as gross primary productivity (GPP) (Arens et al 2008) and above-ground net primary productivity (ANPP) (Vivanco et al 2015), how Rs, which links to many below-ground biotic and abiotic processes, responds to multiple N levels remains unexplored. Rs usually consists of two major components: autotrophic respiration (Ra) in live roots and their symbionts, and heterotrophic respiration (Rh) from soil microorganisms during soil organic matter and litter decomposition (Boone et al 1998, Luo andZhou 2006). It has been reported that root and microbial growth respond distinctly to changing soil N availability compared with above-ground, especially under high N addition levels (Wei et al 2013, Liu et al 2014. Thus, it is interesting to know how Rs and its components respond to gradually increasing N availability, particularly experiencing the N limitation to saturation scenarios.
Numerous biotic and abiotic factors are assumed to regulate Rs and its autotrophic and heterotrophic components and N addition will likely affect Rs through impacts on these drivers. For example, additional N inputs may alter soil temperature and moisture due to the changes in the interception of solar radiation and ecosystem evapotranspiration (Luo and Zhou 2006), which would substantially influence soil CO 2 flux (Davidson andJanssens 2006, Bond-Lamberty andThomson 2010). Meanwhile, the N-induced decrease in soil pH has a great effect on microbial growth and activities (Treseder 2008, Chen et al 2016, usually leading to decreases in Rh (Wei et al 2013, Chen et al 2016. Besides these environmental factors, plant above and below-ground C partitioning also play an important role in regulating Rs, due to the variation in substrate supplies (Janssens et al 2010, Chen et al 2016. However, to date, we know little about how these factors directly and indirectly regulate Ra and Rh across a range of N additions. To address the above issues, we conducted a twoyear field experiment with eight N addition levels (0,1,2,4,8,16,24, 32 g m À2 yr À1 ) in an alpine steppe on the Tibetan Plateau. This area has experienced longterm N limitation due to the low N input and slow N mineralization rate (Zhou 2001), which provides an ideal platform to simulate ecosystems experiencing limited to saturated N scenarios. The objectives of the present study are to i) reveal the general response patterns of Rs and its components to multiple N addition levels, and ii) untangle the relative contributions of biotic and abiotic variables and their direct and indirect influences on Ra and Rh along the N addition gradient.

Site description
The study was conducted on the Sanjiaocheng Sheep Breeding Farm (37°18 0 N, 100°15 0 E, 3290 m a.s.l), an alpine steppe located in Qinghai Province, China. The long-term mean annual temperature is 0.08°C and the mean annual precipitation is 387 mm . The native vegetation is a typical alpine steppe community, characterized by the dominant species of Stipa purpurea, Carex rigescens, Poa crymophila Keng, Agropyron cristatum, Heteropappus altaicus (Willd.) Novopokr, etc. The soil type at the study site is classified as chestnut according to Chinese classification or Haplic Calcisol according to the FAO (Food and Agriculture Organization) classification, with 61.0% sand, 33.4% silt, and 4.9% clay. The chemical properties of the 0-30 cm soil layer at the beginning of the experiment were as follows: soil organic C 20.5 g kg À1 , total-N 2.5 g kg À1 , total phosphorous (P) 0.6 g kg À1 , pH 9.5, NO 3 À -N 11.5 mg kg À1 , NH 4 þ -N 0.1 mg kg À1 and available P 2.2 mg kg À1 .

Experimental design
The experimental site was fenced and the N enrichment experiment started in May 2013. Other than grazing, no other management had been applied on the site before the experiment. A gradient of eight N levels (0,1,2,4,8,16,24, and 32 g m À2 yr À1 , labeled N0, N1, N2, N4, N8, N16, N24 and N32, respectively) was established and arranged in randomized complete blocks with five replications. The plot size was 6 m Â 6 m and a total of forty plots was created. The distance between any two adjacent plots within a block was 1 m and each block was separated by 2 m buffer strips. N fertilizer, in the form of ammonium nitrate, was applied at the beginning of each month during the growing season from May to September every year. At each plot, a specific amount of fertilizer was dissolved in 10 L of water and sprinkled by a backpack sprayer. Two passes were made across each plot to ensure an even distribution of the fertilizer. The control plot received the same amount of water without N.
2.3. Measurement protocols 2.3.1. Rs and its components In May 2013, two PVC collars (20 cm in diameter) were inserted into the soil permanently at least 50 cm away from the plot edge. One collar (5 cm in height) was inserted into the soil to a depth of 3 cm for Rs Environ. Res. Lett. 12 (2017) 024018 measurement, and another (62 cm in height) was inserted to a depth of 60 cm to detect Rh; these collars cut off old roots and prevented new roots from growing inside the collars. Ra was then calculated as the difference between Rs and Rh. This method has been successfully used to separate Ra and Rh in previous studies (Hasselquist et al 2012, Suseela and Dukes 2013. Soil CO 2 flux was measured using a LI-8100A Automated Soil CO 2 Flux System (Li-Cor Inc., Lincoln, NE, USA) with ca. every 10 d between 9:00-12:00 am (local time) from May 11 to October 22 in 2014, and from April 24 to October 13 in 2015. Living plants inside the collars were removed by hand at least one day before each measurement. During the measurements, the flux chamber was placed on the short collar (5 cm in height) first and subsequently moved to the long collar (62 cm in height) in each plot. At the same time the soil CO 2 efflux was measured, the soil temperature and volumetric moisture at a 0-10 cm soil depth were monitored adjacent to the PVC collar using a digital thermometer and a portable TDR-100 soil moisture probe (Spectrum Technologies Inc., Plainfield, IL, USA), respectively.

Plant and soil sampling
Plant and soil samples were collected in the middle of August in both experimental years. Within each plot, we randomly selected three 25 cm Â 25 cm quadrats, and harvested green above-ground plants, which represented the current year's ANPP. After aboveground sampling, three root cores (8 cm in diameter and 10 cm in depth) were collected in the corresponding sites and then mixed as one sample. Root samples were immediately placed in a cooler and then cleaned of residual soil using a 0.5 mm sieve soaked in water. Live and dead roots were distinguished by their color, resilience and the attached fine roots (Vogt andPersson 1991, Yang et al 2010), and the live root biomass was presented. All plant samples were ovendried at 65°C until they reached a constant weight, and were then weighed to estimate dry mass. After root collection, soil cores (3 cm in diameter) were collected from the surface layer (0-10 cm) adjacent to root cores in each quadrat, and combined to gain one composite sample. After the roots and organic debris had been removed manually, soil samples were immediately sieved through a 2 mm screen in the field and then stored in a cooler. These samples were then transported to the laboratory for further analysis.

Laboratory analyses
The soil's inorganic N (NO 3 − -N and NH 4 þ -N) was extracted with a 1 M KCl solution and analyzed using a flow injection analyzer (Autoanalyzer 3 SEAL, Bran and Luebbe, Norderstedt, Germany). The soil's microbial biomass C (MBC) was estimated using the chloroform fumigation extraction method (Vance et al 1987). An aliquot of 20 g of fresh soil was fumigated with chloroform for 24 h and then extracted by 0.5 M K 2 SO 4 . Another aliquot of 20 g unfumigated fresh soil was directly extracted by a K 2 SO 4 solution. MBC was calculated from the differences between extractable C concentrations in the fumigated and unfumigated samples by a conversion coefficient of 0.45 (Joegensen andMueller 1996, Jenkinson et al 2004). Soil pH was determined with a glass electrode in a 1:2.5 soil: water solution (w/v).

Statistical analysis
To examine the nonlinear response patterns of Rs, Ra and Rh along the N addition gradient, an empirical function (y = a þ bx c /e dx , where y is Rs, Ra or Rh, x is the N addition rate, and a, b, c and d are the coefficients, Yang et al 2010) was employed to characterize the relationships of these variables with N addition rates during each growing season. One-way ANOVAwas used to compare the treatment difference for Rs, Ra and Rh as well as biotic and abiotic variables. Ordinary least square regressions were conducted to explore the relationships of Rs, Ra and Rh with biotic and abiotic parameters. All these analyses were performed with SPSS statistic 16.0 (IBM SPSS, Chicago, Illinois, USA).
Structural equation modeling (SEM) was used to examine the relative importance and direct and indirect pathways of various factors regulating Ra and Rh along the N addition gradient. Before SEM analysis, we constructed an a priori conceptual model of hypothetical relationships (figure S1 available at stacks.iop.org/ERL/12/024018/mmedia). Theoretically, a number of biotic and abiotic factors are assumed to influence soil CO 2 flux, but our model only adopted those factors showing significant correlations with Ra or Rh along this N addition gradient, at least in 2015 (see figures 3 and 4). Therefore, we hypothesized that, soil temperature directly influenced Ra and Rh, and also indirectly influenced them through regulating root and microbial biomass. Meanwhile, soil inorganic N and pH affected Rh directly and also indirectly via MBC (figure S1). During SEM analysis, data were fitted to the model using the maximum likelihood estimation method. This procedure compared the model-implied variance-covariance matrix with the observed variance-covariance matrix, and used the chi-square goodness-of-fit statistic and its associated P value to test the model fit to the data. The covariance structure of the data was considered not to be significantly different from the expected model if the P value was larger than 0.05 (Grace 2006). SEM analyses were conducted using Amos version 21.0 (Amos Development Corporation, Chicago, IL, USA).

Changes of Rs and its components along the N addition gradient
With the N addition rate continuously increasing, the seasonal means of Rs and Ra exhibited first increasing Environ. Res. Lett. 12 (2017) 024018 and subsequent decreasing trends at a threshold of 8 g N m À2 yr À1 (figures 1(a)-(d)). In contrast, a linear negative relationship was fit to Rh data under increasing N additions (figures 1(e) and (f)). During the two experimental years, except for Rh in 2014, additional N inputs significantly altered Rs and its components (table S1). The annual budgets for the three parameters were also influenced by N additions, showing similar patterns as seasonal averages along the N addition gradient (table S1).
In general, the seasonal dynamics of Rs and its autotrophic and heterotrophic components followed a bell-like pattern in both experimental years, with higher values in the mid-growing season and lower values in the early and late seasons (figure 2). Interannual variations of Rs, Ra and Rh were observed across the two experimental years. The maximum values of Rs and its components appeared in mid-August in 2014 but in mid-July in 2015. In addition, N input exaggerated their inter-annual variability by strengthening the maximum values (figure 2).

Effects of increasing N additions on biotic and abiotic variables
Increasing N additions had significant influences on the biotic and abiotic factors (table 1). ANPP consistently displayed a saturation trend with increasing N additions regardless of inter-annual variation, with a saturation point occurring at 8 g N m À2 yr À1 . The relationship between root biomass and the N addition rate was characterized by a hump-shaped pattern similar to Rs and Ra, with a peak appearing in the N8 treatment. In 2014 no significant response to the N addition was observed in MBC, whereas in 2015, it increased gradually until 4 g N m À2 yr À1 and thereafter declined.
Soil temperature increased with increasing N additions up to 4 g N m À2 yr À1 and then decreased after this N level in 2014, while in 2015, no obvious change was found among the N0 to N4 treatments, but the N8 to N32 treatments significantly reduced soil temperature. In both years, soil moisture generally showed a funnel-like pattern along the N gradient, with the minimum at 8 g N m À2 yr À1 . Soil inorganic N increased monotonously with increasing N additions in 2015, but the increasing trend leveled off after the N8 treatment in 2014. Soil pH was also negatively related to the N addition rate in 2015, while no obvious trends were observed along the eight N levels in 2014.

Effects of biotic and abiotic factors on Rs and its components
Across the 40 experimental plots, Rs, Ra and Rh increased linearly with increasing soil temperature in both experimental years (figures 3(d), (k) and (r); figures 4(d), (k) and (r)). Meanwhile, Rs and Ra were also positively related to root biomass (figures 3(b) and (i); figures 4(b) and (i)). In addition, Rh increased with MBC in 2015 ( figure 4(q)). Soil inorganic N also

Partitioning of Rs components
In the current study, we used the deep-collar method to distinguish Ra and Rh, a method which has been widely applied to partition Rs in other studies (Zhou et al 2007, Hasselquist et al 2012, Suseela and Dukes 2013. Given that roots in the top 30 cm of soil accounted for more than 90% of the biomass of those in the 0-50 cm soil profile at our experimental site (data not shown), the 60 cm long collar was long enough to exclude most of the living roots. It has been reported that the deep-collar method generally results in a transient increase of Rs due to the decomposition of dead roots and disturbance (Hanson et al 2000, Zhou et al 2007. It has also been suggested that the transient increase of Rs lasted for about five months after collar insertion in a prairie in Southern Great Pains, USA (Zhou et al 2007). Given that collars were inserted into the soil in May 2013, the effects of dead root decomposition on the Rh measured after a whole year should be minimized. More importantly, our results revealed that the contribution of Rh to Rs was 46%-63% and 52%-70% in 2014 and 2015, respectively, and these values are within the range reported in grassland ecosystems (25%-90%, Subke et al 2006). Therefore, despite the fact that some uncertainties may be inevitable, the root exclusion method is a feasible way to partition the two components of Rs in this alpine steppe.
Environ. Res. Lett. 12 (2017) 024018 term N enrichment would cause some differences. For example, the N-induced shifts in plant community composition and structure exert great influences on ecosystem C cycling (Niu et al 2010, Chen et al 2016, but we know little about how these changes will affect Rs after a long time period, particularly under multiple N loads. Thus, experiments with long duration and multiple N loads (at least more than five N addition levels) are called for to better depict the Rs/Ra/Rh-N rate relationships in future studies. The seasonal dynamics of Ra and Rh displayed temporal variability across the two experimental years (figure 1). Inter-annual variation of Rs, Ra and Rh have been frequently reported in grasslands and other ecosystems (Liu et al 2002, Concilio et al 2009, which may result from temporal variations in various biotic and abiotic factors (Luo and Zhou 2006). First, the precipitation distribution varied significantly during the two growing seasons (figure S2). In particular, for the peak plant growth in August, the precipitation amount in 2015 (79 mm) was significantly lower than in 2014 (186 mm (table 1). Larger root biomass indicates more C was allocated below ground (Peng and Yang 2016), which is usually associated with higher root respiration (Bai et al 2010). Meanwhile, increased microbial biomass was generally observed to have higher Rh (Zhou et al 2014, Chen et al 2016). These differences between the two experimental years may lead to greater soil CO 2 fluxes during the peak growth stage in 2014 (figure 2).

Regulations of biotic and abiotic factors in Ra and Rh
Several biotic and abiotic factors regulate the change of Ra and Rh along this N addition gradient. The SEM analysis indicated that root biomass and MBC were direct controllers of Ra and Rh, and further, that soil temperature also directly and indirectly explained their variations ( figure 4). Root biomass had a direct effect on Ra, because larger root biomass means greater root surface area for respiration (Zhou et al 2014). Consistent with this point, root biomass had a similar hump-shape response pattern to Ra along this N gradient (table 1). Before the N saturation (8 g N m À2 yr À1 ), root biomass increased in both years (table 1). This is because during this time, albeit with the N addition reducing growth limitation to a certain degree, plants are still under the limited-N condition. Thus, they need to allocate a portion of C to roots for additional N uptake to maintain plant growth (Peng and Yang 2016). Meanwhile after the saturation point, root biomass decreased; this was likely due to 1) decreased root production. In cold regions, such as the tundra or alpine steppe, experimental N addition generally decreased C input to roots (Peng et al 2017), especially under high levels of N addition (Sullivan et al 2007). This is likely because temperature is usually the major growth-limiting factor in these sites, and plants cannot absorb too much N due to the low growth rate even under N addition. Meanwhile, high N accumulation in mineral soils may be harmful for root growth (Marschner 2011), thus resulting in low production of roots. 2) increased root turnover rate, or in other words, reduced root life span. For instance, Bai et al (2008) reported that an addition of 32 g N m À2 yr À1 significantly reduced root life span from 81 d to 64 d in a 0-30 cm soil depth in a temperate steppe on the Inner Mongolian Plateau. Taken together, the decreased root production and increased turnover rate may cause reduced root standing biomass at high N addition levels.
Microbial biomass directly explained the variance in Rh along the gradient of N additions in 2015 (figure 4). Low N addition levels (from 1 to 4 g N m À2 yr À1 ) gradually increased microbial biomass, but high levels of N addition suppressed it (table 1). The stimulation of microbes by low N may be due to the reduced C and N limitation after increased litter input from the first two years; furthermore, the decrease in the microbial biomass under the high N supply is potentially because the high dose of N addition reduced the C transfer to the rhizosphere, and induced shifts in the microbial community and decomposing enzymes; this would enhance the formation of recalcitrant compounds, thus causing decreased microbial respiration (DeForest et al 2004, Janssens et al 2010, de Vries et al 2014. However, under low N levels, the changes in the microbial biomass did not match well with the changes in Rh, and the underlining mechanisms need further investigation. Soil temperature was the only environmental factor entering the final SEM (figure 5), likely indicating that it is the most important abiotic factor regulating Rh, while the effects of pH and inorganic N are relatively weaker along this N-addition gradient. Soil temperature not only directly explained the variations of Ra and Rh by regulating respiration enzymes (Atkin and Tjoelker 2003), but also indirectly influenced them via affecting root and microbial growth (Janssens et al 2010). We observed that soil temperature displayed a decreasing trend with increasing N additions, especially for those plots receiving a N rate of more than 8 g m À2 yr À1 (table 1). This is likely because N addition increased aboveground plant growth and the more closed canopy Numbers at arrows are standardized path coefficients or covariation coefficients, and the thickness of the arrows is proportional to the strength of the relationship. The significance of each path coefficient or covariation coefficient follows the numbers ( Ã P < 0.05, ÃÃ P < 0.01, ÃÃÃ P < 0.001). The pathways without significant effects (P ≥ 0.1) are indicated by dashed arrows. r 2 values above the variable box indicate the proportion of variance explained by endogenous variables. Goodness-of-fit statistics for the model are shown below the model. Abbreviations. ST: soil temperature; RB: root biomass; MBC: microbial biomass C.
Environ. Res. Lett. 12 (2017) 024018 prevented the incoming solar radiation to the soil surface (Luo and Zhou 2006). Consistent with this deduction, we observed a negative relationship between soil temperature and light interception in 2015 ( figure S4). Thus, the CO 2 respired from roots and microbes could be reduced at high-level N input due to the decreased soil temperature, especially in cold regions where temperature is usually considered a limiting factor for living beings (Lloyd and Taylor 1994).
In grassland ecosystems, water availability is usually observed to constrain CO 2 flux with or without N additions (Harpole et al 2007, Xia et al 2009, Yan et al 2010. However, in our present experiment, we did not find any significant correlations of Rs, Ra and Rh with soil moisture along the N addition gradient (figures 3(e), (i), (s); figures 4(e), (i), (s)). This is likely because of the large moisture variation during both growing seasons ( figure S5). Previous studies indicated that there might be a threshold of soil moisture in grasslands (8%-14% V/V%, Zhou et al 2007, Yan et al 2010), above which water is not the constraint of soil C process responses to climate change. Due to this point, it may be that, in aggregate, there were no detectable influences of the N-induced soil moisture changes on soil CO 2 fluxes since the soil moisture was above the critical level at most of our measurement times (figure S5). Nevertheless, as we mentioned above, precipitation-associated water changes are still very important in this alpine steppe because they regulate the seasonal dynamics and inter-annual variation of Rs and its components (figure 2).

Conclusions and implications
The present study revealed that, Rs and its autotrophic component displayed a nonlinear (hump-shaped) response to increasing N additions, while Rh declined linearly along the N levels. Soil temperature as well as root and microbial biomass explained the variations in Rs and its components. These findings have two important implications for understanding C-climate feedback in terrestrial ecosystems: 1) Many studies predicting the future terrestrial C balance have assumed a linear relationship between ecosystem C fluxes and N additions (e.g. Janssens et al 2010, Zhou et al 2014, Greaver et al 2016. However, the results observed in the current study, together with the nonlinear GPP and ANPP responses to increasing N inputs reported in previous studies (Arens et al 2008, Vivanco et al 2015), indicate that the linear relationship of the above and below-ground C fluxes with increasing N inputs assumed by previous studies may be inaccurate. Furthermore, nonlinear responses of various ecosystem components to increasing N additions are suggested when modeling the global C cycling, given that anthropogenic N input to the terrestrial ecosystems is continuously increasing (Penuelas et al 2013). Meanwhile, the hump-shaped response pattern of Rs to continuous N addition are different from GPP and ANPP, which generally increased first and then plateaued after saturation level (see Arens et al 2008, Vivanco et al 2015. Therefore, Earth System Models also need to consider the different response patterns in C fluxes from various ecosystem components. 2) Since significant climate warming has been widely reported on the Tibetan Plateau during recent decades (Wang et al 2008), the negative N effect on soil temperature as revealed in our study highlights that increased N input might mitigate the positive C feedback, not only via the enhanced plant C accumulation but also through reduced warming effects on Rs. Overall, the nonlinear responses of Rs and its components to multiple N loads merit further attention in future experimental and modeling analyses. Yue K, Peng Y, Peng C H, Yang W Q, Peng X and Wu F Z 2015 Stimulation of terrestrial ecosystem carbon storage by nitrogen addition: a meta-analysis Sci.