2.1 Study sites and sampling

The study was conducted along a 3000-km transect in alpine grasslands across the Tibetan Plateau (79°49′39″–100°15′33″ E, 31°06′37″–32°43′09″ N), with an average altitude of 4424 m, ranging from 3544 to 5016 m. Along this transect, 23 sites were randomly selected with a spatial distance of approximately 120 km between adjacent sampling sites to determine soil δ¹⁵N (Table S1; Figure S1). Longitude, latitude, and altitude were recorded by an outdoor handheld instrument navigator locator with a global positioning system (GPS) at each site. MAP and mean annual temperature (MAT) data were obtained from the China Meteorological Data Service Centre (CMDC; http://data.cma.cn) based on the location information of each site. Climatic results showed that our study sites covered a broad precipitation gradient, varying between 72 and 648 mm, and the mean annual temperature (MAT) ranged from −3.9 to 5.8°C. Furthermore, the aridity index (AI; the ratio of annual precipitation to potential evapotranspiration) (Delgado-Baquerizo et al., 2013) at each site was extracted from the Global-Aridity and Global Potential Evapotranspiration Climate database (Hu et al., 2021), which ranged from 0.06 to 0.73 along the aridity gradient.

Field sampling was conducted between July 30 and August 28, 2020 (each site was visited once over this period). Briefly, at each site, four 1 m × 1 m sampling plots were randomly selected to measure aboveground and belowground biomass from July to August 2020. Specifically, aboveground plants were clipped and belowground roots were collected from three soil samples in each plot with a 7.5 cm diameter auger. All aboveground clipped plants were dried at 60°C to a constant weight and weighed to determine plant aboveground biomass. Living roots were obtained from soil samples and washed thoroughly with deionized water, and then oven-dried at 60°C for 72 h to weigh to estimate belowground biomass of community.

In each plot, three soil cores from 0 to 10, 10 to 20, and 20 to 30 cm were collected using an auger (7.5 cm in diameter). As the majority of roots are located within the 0–30 cm soil depths (Yang et al., 2009), soil samples from 0 to 10 and 20 to 30 cm were categorized as topsoil and subsoil, respectively. The soil cores at each soil depth were combined into one soil sample, sieved through a 2-mm mesh to remove any non-soil materials and then divided into three subsamples. One subsample was air-dried to determine the soil ¹⁵N abundance and soil properties; the second subsample was stored at 4°C for analyzing the inorganic N (NH₄⁺ and NO₃⁻) concentrations, N transformation rates, and microbial biomass; the third subsample was stored at −80°C for measuring soil bacteria, fungi and denitrifying gene abundances. We also sampled the two dominant grass species, Leontopodium nanum, and Stipa purpurea, to determine foliar δ¹⁵N. These two species were selected based on their vast distribution along the aridity gradient (Li et al., 2023).

2.2 Laboratory treatment and chemical analysis

The isotopic dilution method was used to measure the gross N transformation rate (Hart et al., 1994). In brief, four subsamples of fresh soil (equivalent to 20 g dry weight) were labeled with ¹⁵N, to which two were labeled by the (¹⁵NH₄)₂SO₄ solution (5.2 atom%) and the other two were labeled by the K¹⁵NO₃ solution (5.14 atom%). The soils were incubated at 20°C after being adjusted to 60% water-holding capacity. One subsample of each ¹⁵N labeled portion was extracted with 2 M KCl solution after labeling for 0.5 and 24 h. The NH₄⁺ and NO₃⁻ concentrations in the extracts were determined using a continuous flow analyzer. The N isotope compositions of NH₄⁺ and NO₃⁻ were measured by an isotope ratio mass spectrometer after distilling the NH₄⁺ and NO₃⁻ using the diffusion method. Gross N mineralization and nitrification rates were calculated following previous studies (Davidson et al., 2006; Kirkham & Bartholomew, 1954).

We further determined the copy numbers of bacteria and fungi using quantitative real-time polymerase chain reaction (qPCR) (Borneman & Hartin, 2000; Caporaso et al., 2011). The abundances of bacteria and fungi were used to calculate the bacteria-to-fungi ratios. The copy numbers of denitrifying genes (nirK, nirS) were simultaneously measured as gaseous N losses during soil denitrification have strong discrimination against heavier isotopes (¹⁵N), which can strongly affect the ¹⁵N isotope patterns (Feng et al., 2023; Högberg, 1997). Detailed descriptions of the primers used for PCR amplification are presented in Table S2. Basic soil properties were also measured. Soil total organic carbon (TOC) and TN were measured using a LECO macro-CN analyzer (LECO, St. Joseph, MI, USA). The chloroform fumigation-extraction method was applied to measure MBC and MBN with a conversion factor of 0.45 (Vance et al., 1987). Soil-available phosphorus (AP) was first extracted by the NaHCO₃ solution (0.5 M) and then analyzed using the molybdenum-antimony anti-spectrophotometric method. Soil pH was assessed at a water ratio of 2.5 using a pH meter.

2.3 Statistical analyses

All data were carefully checked to ensure quality before the statistical analyses. First, we examined potential linear or quadratic (nonlinear) relationships of soil δ¹⁵N with increasing aridity. The model goodness-of-fit was evaluated using the Akaike information criterion (AIC), where a lower AIC value represents a superior model fit (Cavanaugh & Neath, 2019). Thresholds were detected when the quadratic regressions fit the data better. We fit segmented regressions using the “segmented” package of R (Muggeo, 2008) to identify the tipping point where the soil δ¹⁵N values abruptly changed along the aridity gradient, with the tipping point detected as the aridity threshold as many previous studies (Feng et al., 2019; Hao et al., 2021). To validate the robustness of the threshold, the relationship between aridity and soil δ¹⁵N values on either side of the threshold was analyzed by linear regression (Berdugo et al., 2020). Furthermore, to better distinguish the relative importance of predictors controlling soil δ¹⁵N below and above the aridity thresholds, we performed two separate random forest models using the “randomForest” (Liaw & Wiener, 2002) package in R, and the “rfPermute” package was also used to assess the significance of each predictor's importance (Archer, 2016). To assess the model performance and mitigate overfitting, we conducted a tenfold cross-validation, with 80% of the data for training and 20% for validation following another study (Hu et al., 2024). The estimations of soil δ¹⁵N in wetter and drier regions by random forest regression showed determination coefficients (R²) of .79 and .92 in predicted versus observed regressions, respectively, indicating that the models exhibited reasonable goodness-of-fit. The relationships of soil N-cycling traits with increasing aridity were also examined by linear or quadratic regressions. All analyses were performed in R4.1.3 (R Core Team, 2022), and figures were created by “ggplot2” package with significance at p < .05.

3.1 Soil δ¹⁵N patterns and soil N dynamics along the aridity gradient

We found that the topsoil δ¹⁵N values ranged from 3.48 to 9.78‰ and exhibited a quadratic (upward-concave) pattern along the aridity gradient, with an aridity index (AI) threshold of 0.27 (Figure 1a). Similar to topsoil, the quadratic pattern of soil δ¹⁵N was also detected in the subsoil layer (with an AI threshold of 0.29) (Figure 1b). In contrast to soil δ¹⁵N, in both soil layers, soil N pools (TN and mineral N content), N-cycling processes (GMR and GNR), and microbial properties (MBN, BA, and nirK gene) significantly decreased with increasing aridity (Figure 2; Figure S2), whereas the soil C:N and MBC:MBN ratios significantly increased with increasing aridity (Figure 2; Figure S2).

3.2 Correlations between soil δ¹⁵N values and soil N dynamics along the aridity threshold

Soil δ¹⁵N showed dramatically different correlations with soil N-cycling traits above and below the AI threshold (Figure 3; Figure S3). Specifically, topsoil δ¹⁵N had significantly positive relationships with TN, mineral N content, MBN, nirK and nirS abundances, BA, GMR, and GNR in the wetter sites (AI >0.27), whereas these positive correlations disappeared in the drier sites (AI <0.27) (Figure 3). Similar to topsoil, subsoil δ¹⁵N also showed significantly positive correlations with certain soil N-cycling traits (e.g., TN, MBN, nirK, and nirS abundances, BA, and N transformations) in wetter sites, whereas no significant relationships or negative correlations (such as mineral N content and MBN) were observed in drier sites (Figure S3).

3.3 Regulation factors in soil δ¹⁵N under different aridity levels

Significant effects of soil N-cycling traits and microbial properties and plant δ¹⁵N on soil δ¹⁵N were explained by the random forest model, but the main regulatory factors were different under wetter and drier conditions. Results showed that soil pH, TN, and microbial properties (MBN and bacteria and denitrifying gene abundances) were the factors significantly affecting soil δ¹⁵N in wetter sites. In contrast, in drier sites, most soil N-cycling traits were not significantly correlated with soil δ¹⁵N. Soil AP, mineral N, and foliage δ¹⁵N were significant contributors to the variation of soil ¹⁵N (Figure 4; Figure S4).

4.1 Soil N availability along the aridity gradient

Previous studies have demonstrated that aridity has a negative effect on the soil N supply, likely exacerbating the limitation of soil N availability (Delgado-Baquerizo et al., 2016; Tang et al., 2021). Based on a systematic survey, we found a significant decrease in soil N pools and N processes in both the topsoil and subsoil layers along the aridity gradient, indicating a corresponding decrease in soil N availability with increasing aridity. Decreased soil N availability at arid sites is unsurprising as drought limits the activity of soil microorganisms (Fuchslueger et al., 2014), which play a vital role in soil N cycling. For example, 16S rRNA gene abundance suggests that the nitrifying activity of soil bacteria and archaea is reduced under drought conditions, resulting in a lower N nitrification rate (Hartmann et al., 2013). Similarly, in this study, the decreased MBN and bacterial abundance along the aridity gradient indicated that drought restricted N-associated microorganisms, leading to a decrease in soil N mineralization and nitrification rates (Figure 2; Figure S2), and consequently reduced soil N availability.

4.2 Soil δ¹⁵N patterns along the aridity gradient

Soil δ¹⁵N values, especially in the topsoil layer, are generally used as an indicator of soil N availability (Kou et al., 2020; Robinson, 2001). Considering the reduced soil N availability with increasing aridity, soil δ¹⁵N values are expected to decrease along the aridity gradient. However, contrary to our first hypothesis, we found that δ¹⁵N values in both topsoil and subsoil exhibited an upward-concave pattern along the aridity gradient. The unexpected soil ¹⁵N patterns under different drought conditions challenge the traditional view of a simple linear correlation between ¹⁵N signals and precipitation (or aridity) (Craine et al., 2009). Furthermore, the quadratic pattern of soil δ¹⁵N values also contradicts the decreased soil N pools and processes along the aridity gradient, given the positive correlation between soil δ¹⁵N and N availability in the traditional view owing to significant ¹⁵N isotope fractionation during open N-cycling processes under conditions of high N availability (Högberg, 1997; Mason et al., 2022). These results suggest that under different aridity conditions, different mechanisms underlying ecosystem N cycling may control soil δ¹⁵N values.

Soil δ¹⁵N values are determined only by the input and output pathways of ¹⁵N to and from soils (Robinson, 2001). As the δ¹⁵N value of N input (mainly by N deposition and N₂ fixation) is approximately equal to 0‰ (Bai & Houlton, 2009; Houlton et al., 2007), the enrichment of soil δ¹⁵N values indicates the occurrence of output processes leaving more ¹⁵N in the soil (Liu et al., 2016). Gaseous N loss, net plant N accumulation, and leaching N loss are output processes, wherein gaseous N loss exhibit the most pronounced isotope fractionation effects (ε_G ≈ 16‰–30‰), followed by net plant N accumulation (ε_P ≈ 5‰–10‰) and leaching N loss (ε_L ≈ 1‰) (Feuerstein et al., 1997). In regions with low aridity, we speculated that the pattern of soil δ¹⁵N is likely influenced by gaseous N loss derived from N transformations, as indicated by significantly positive correlations between soil δ¹⁵N values and GMR or GNR (Figure 3; Figure S3). Additionally, the significantly positive correlations between soil δ¹⁵N and denitrifying genes (nirK and nirS) in the wetter sites but not in the drier sites also highlight the role of denitrification in soil δ¹⁵N values under wetter conditions (Figure 3; Figure S3). These positive correlations between soil δ¹⁵N and soil N transformations and associated functional genes indicate that in wetter regions, high soil microbe-driven N cycling with a significant ¹⁵N fractionation significantly contributes to the enrichment pattern of soil δ¹⁵N.

In agreement with this, the random forest result showed that among multiple factors, soil N-cycling traits (e.g., TN, BA, MBN, and denitrifying genes) significantly explained soil δ¹⁵N changes in wetter sites, which provides robust evidence that ¹⁵N enrichment under relative drought conditions may largely result from strong soil N losses through microbe-driven N transformations under high N availability conditions. Additionally, this microbe-driven mechanism of soil ¹⁵N enrichment is further supported by the pronounced contribution of soil pH to soil δ¹⁵N. As a key factor shaping microbial structure and function, soil pH strongly affects soil N-cycling microbial communities (Ontman et al., 2023; Rousk et al., 2010; Scarlett et al., 2021). Our results showed that pH was the most pronounced driver of soil δ¹⁵N in wetter sites, with dramatically different pH values under different aridity conditions (Figure S5), indirectly highlighting the contribution of microbe-driven N cycling to soil δ¹⁵N. Overall, these findings show that in wetter regions, soil ¹⁵N closely tracked with soil N dynamics, suggesting that soil ¹⁵N isotopes can accurately reflect soil N cycling and N availability in low aridity environments (Figure 5).

Conversely, under extremely dry conditions, microbe-driven soil N-cycling processes should not be the main reason for the increase in soil δ¹⁵N owing to the insignificant correlations between soil δ¹⁵N and soil N dynamics in the drier sites (Figure 3; Figure S3), which has not been detected in previous studies. More intense and prolonged droughts owing to climate change are expected to promote open N cycling in drylands through sustained microbial N processing but low plant N uptake (Evans & Burke, 2012; Wu et al., 2022). Microbial processes have been postulated to predominantly drive N gas evasion in arid systems, thereby significantly affecting soil δ¹⁵N (Homyak et al., 2017). Contrary to expectations, our study spanning an aridity transect revealed that the “open” N cycling accompanied by microbe-driven soil N losses and ¹⁵N enrichment was only observed under low aridity conditions. In high aridity sites, pathways other than microbial mechanisms may significantly contribute to ¹⁵N enrichment. This discrepancy highlights an unclear mechanism controlling soil δ¹⁵N with increasing aridity. We propose the following reasons that may be responsible for the ¹⁵N patterns in drier regions.

On the one hand, according to the aforementioned N isotope input and output model, we speculated that increased plant net N accumulation may explain the increased soil δ¹⁵N values under extreme drought conditions. Previous studies have demonstrated that increasing aridity significantly decreases N return from plants to the soil (Wu et al., 2022; Yuan & Chen, 2008). This may be attributed to the fact that under high aridity conditions and low soil N availability, plants are able to alleviate N limitation by enhancing N retention (e.g., by increasing N resorption) (Luo et al., 2021; Yuan & Chen, 2008). This process may result in an increase in plant net N accumulation, thereby increasing the contribution of ¹⁵N fractionation to soil ¹⁵N enrichment during plant net N accumulation. The result of greater correlation between soil and foliage δ¹⁵N values in drier than wetter regions also highlight different N feedback mechanisms between the plant and soil under different aridity conditions (Figure 4).

On the other hand, previous studies have demonstrated that abiotic N loss processes, such as ammonia (NH₃) volatilization (Liu et al., 2020; Soper et al., 2016; Wang et al., 2014) and the loss of oxidized gaseous N (McCulley et al., 2009), are prevalent under high aridity conditions. We speculated that NH₃ volatilization, which has a significant isotope fractionation factor and typically occurs in high pH soils (Heaton, 1986), may be another process attributed to the observed isotopic pattern in extremely dry regions, where soil pH values are significantly higher than those in less dry regions (Figure S5). Overall, our findings highlight different ecosystem N-cycling processes driving the quadratic pattern of soil ¹⁵N isotopes along the aridity gradient, which may help explain ¹⁵N enrichment under severe aridity conditions.

4.3 Implications and limitations

To the best of our knowledge, this is the first study revealing the quadratic pattern of δ¹⁵N values in both topsoil and subsoil layers across an aridity gradient in alpine grasslands. The markedly different correlations between soil δ¹⁵N and soil N dynamics under different aridity conditions in the current study broadens our traditional understanding that enriched soil ¹⁵N signals indicate the “openness” of soil N cycling and high N availability. Numerous studies use natural nitrogen isotopic benchmarks, particularly topsoil δ¹⁵N, to improve the accuracy of nitrogen-based climate change projections (Feng et al., 2023; Kou et al., 2020). Our results highlight the need for caution when estimating soil N availability through δ¹⁵N of soils or plant organs, particularly in drylands where resources, such as water and nitrogen, are extremely limited. These findings advance our understanding of ecosystem N cycling and N limitation of ecosystem productivity with increasing aridity. However, there are still some limitations.

It is important to note that despite we found the different mechanisms driving the quadratic pattern of soil δ¹⁵N along the aridity gradient, a significant portion of the variability in soil δ¹⁵N change in drier regions remains unexplained. This indicates that further studies are required to obtain a more comprehensive understanding of ecosystem N cycling and δ¹⁵N in alpine drylands with severe droughts. Furthermore, although we focused on the correspondence between soil δ¹⁵N and soil N cycling, foliage δ¹⁵N should also be included to explore the ¹⁵N imprint between soils and plants. We only selected the two most widely distributed dominant species on the aridity gradient to explore foliage δ¹⁵N patterns, which may be insufficient to represent all the plant isotope signatures due to significant differences in foliage δ¹⁵N values among different functional types (Craine et al., 2009; Wang et al., 2023). Future studies should pay more attention to the diversification of plant isotopic niches along climatic gradients to better explore the coupling between leaf and soil δ¹⁵N under different drought conditions. Additionally, differences in potential evapotranspiration methods in AI analysis may increase result uncertainties (Tegos et al., 2023). Although the method we used to collect AI data has been widely employed in many related studies (Hou et al., 2020; Wang et al., 2014), our results need validation in other dryland ecosystems to verify the mechanisms underlying these responses. Overall, to better reveal the characteristics of soil δ¹⁵N values and N cycling and their responses to climatic dryness, further observational and experimental studies spanning larger climatic gradients are urgently required across ecosystems beyond alpine drylands.