In summary, by integrating direct measurements of warming-induced changes in soil N stocks with analyses of N-cycling processes, we elucidated long-term response patterns and the underlying mechanisms in a Tibetan permafrost ecosystem. Our results show a substantial decline in the surface soil N stock after eight years of warming (Fig. 6). A comprehensive N budget, conducted through the integration of field-based observations and DeNitrification-DeComposition (DNDC) model simulation, revealed that plant N retention (29.6%), N leaching (14.8%) and gaseous N emissions (13.3%) were the three predominant processes responsible for the decline in soil N stocks (Extended Data Fig. 6; see Supplementary Note 2 for detailed discussions). The observed declines in soil N stocks after eight years of warming could be the result of cumulative changes in these processes over time under warming (Extended Data Fig. 7; see Supplementary Note 3 for detailed discussions). More broadly, a meta-analysis of 38 warming experiments across all permafrost areas, including Arctic, subarctic and Tibetan alpine regions, demonstrated that the decline in soil N stocks observed in our study might also be observed in subarctic regions, although soil N dynamics in response to warming did exhibit regional differences (Extended Data Fig. 8), possibly associated with the heterogeneous climatic conditions and the context-specific soil properties (Extended Data Fig. 9; see Supplementary Note 4 for detailed discussions). In agreement with our experimental results, the DNDC model simulations also demonstrated that declines in the soil N stock would continue to occur in the studied permafrost ecosystem under future climate warming (Extended Data Fig. 10; see Supplementary Note 5 for details). Such a decrease in soil N stocks could limit the N supply to plants, potentially offsetting the positive effects of warming on plant productivity. This limitation could, in turn, weaken the role of vegetation in C fixation, thereby diminishing its capacity to mitigate the permafrost C-climate feedback24,28. Continuous warming could also increase N load in surrounding water bodies and amplify N2O emissions, potentially intensifying the non-C permafrost-climate feedback16,46. To gain further insights for understanding mechanisms behind soil N dynamics and the subsequent consequences on permafrost-climate feedback, in situ observations are required to quantify several key N-cycling processes not included in the present study, particularly regarding N losses through lateral flow, other gaseous forms with varying climatic relevance (for example, gaseous nitrous acid) and rodent and spider foraging47,48,49,50. Overall, this study advances our understanding of warming effects on soil N dynamics in permafrost regions, and contributes to improving model predictions of C-N interactions and permafrost-climate feedbacks under future warming.
The manipulative warming experiment is situated at the base of Wayan Mountain (37° 45' N, 100° 05' E; altitude 3,800 m) in Gangca County, Qinghai Province, China. The site, characterized by swamp meadow vegetation primarily dominated by sedges such as Kobresia tibetica, Kobresia kansuensis and Carex atrofusca, experiences a mean annual temperature of -3.4 °C and mean annual precipitation of 466 mm, with approximately 90% of the precipitation occurring during the growing season (May to September). The soil, classified as Cambisol according to Food and Agriculture Organization of the United Nations (http://www.fao.org/), comprises 5.0% clay, 49.0% silt and 46.0% sand. The study area is underlain by discontinuous permafrost, with an active-layer thickness of approximately 4 m (ref. ).
In June 2013, ten 4 × 4 m blocks were randomly established within a 50 × 50 m fenced area (Extended Data Fig. 1). The warming experiment followed a paired design, with warming and control plots arranged diagonally within each block. The warming treatment was applied using hexagonal open-top chambers, 0.5-m tall and 2.08-m in basal diameter, made from polymethyl methacrylate panels. Soil temperature and volumetric water content at 5-cm depth were logged using ECHO sensors connected to EM50 data loggers (Decagon Devices). The warming device resulted in an average annual soil temperature increase of 1.3 °C between 2014 and 2023, but did not significantly affect soil volumetric water content during the growing season (Supplementary Fig. 1). Detailed experimental design and set-up are described in ref. .
To assess warming effects on soil N stocks over a ten-year period, we measured total soil N content and bulk density in both warmed and control plots. Between 2014 and 2023, soil samples were collected annually in late August using a 3-cm auger at the depths of 0-10, 10-20 and 20-30 cm, but samples of deeper layers (30-40 and 40-50 cm) were sampled from 2018 onwards. Three soil cores from the same depth were composited into a single soil sample per plot. Soil samples were sieved through a 2-mm mesh to remove roots and debris, then each divided into two parts: one part was frozen at -20 °C for further analysis, while the other part was air dried and ground to measure total N content using an elemental analyser (vario EL Ш; Elementar). Soil samples were also collected using 100-cm steel cylinders, dried at 105 °C for 48 h and weighed to calculate bulk density. On the basis of these analyses, soil N stocks for each layer were calculated according to ref. :
where SND, T, BD, TN and C are soil N density (kg N m), soil thickness (cm), bulk density (g cm), soil total N content (g kg) and >2-mm rock content (%) in layer i, respectively. Notably, warming did not alter soil bulk density over the course of the experiment (Supplementary Fig. 12). To explore underlying mechanisms for warming-induced changes in topsoil N stocks, we measured a suite of N-cycling processes and microbial variables, including N inputs, internal N-cycling processes and N outputs. These variables were determined using plant and soil samples collected after eight years of warming, when significant declines in topsoil N stocks became apparent compared with control plots (Fig. 1).
To evaluate warming effects on external N inputs, we quantified biological N fixation and litter N release rates in 2023. Biological N fixation was measured in the field using a N-N pulse labelling technique. Specifically, on 11 August, two cores (5-cm depth, intact vegetation) were carefully extracted from each plot, and placed into transparent polycarbonate cylinders (10-cm diameter, 20-cm height and 1,200-ml volume), and then returned to the soil (Supplementary Fig. 13a). Subsequently, 120 ml of air (~10% volume) was taken from the headspace of one cylinder in each plot and replaced with 120 ml of N-N (98 at.%), while the air in the other cylinder was replaced with 120 ml of unlabelled N. All cylinders were incubated in the field for 7 d. On 18 August, soil mesocosms were collected by separating the aboveground plants from the soil. Soil samples were weighed, sieved through a 2-mm mesh and then air dried. Roots were gently picked from the soil and carefully rinsed with water. The leaf, root and soil samples were oven dried at 65 °C for 48 h and their dry mass was subsequently determined. Plant and soil samples were ground for total N content analysis (vario EL Ш) and δN values measured (EA-Isolink-253 Plus). Finally, we calculated the rate of biological N fixation using the following equations.
where at.%N and at.%N are the at.%N excess in the labelled and unlabelled samples (‰), BNF is the biological N fixation rate in the leaf, root and soil (mg m d), total N is the N content in the sample (%), t is the incubation time (days), %N is the volume of N-N as a percentage of the total amount of N in the cylinder and W is the dry mass of the leaf, root and soil (g m).
We further determined above- and belowground litter decomposition and litter N release using buried nylon mesh bags. Briefly, fresh leaf litter and fine roots (that is, ≤1 mm in diameter) were collected from the warming and control plots. After plant samples were selected and dried at 65 °C, we placed ~1.5 g of leaf litter and roots into 10 × 15 cm nylon litterbags (with a 0.1-mm mesh size). Twenty leaf litterbags were fixed to the soil surface, and 20 root litterbags were buried at 0-10 cm depth. In late October, litterbags were collected and dried to a constant mass at 65 °C. After the remaining dry mass was weighed, we ground the initial and final litter samples and determined their N concentrations with an elemental analyser (vario EL Ш). Finally, the litter mass loss and the N release were calculated according to ref. (details in Supplementary Note 6).
To quantify the effects of warming on soil N availability, we measured a suite of microbial N transformation rates indicative of the major N-cycling processes. Of these transformations, gross protein depolymerization and amino-acid consumption rates were measured using the N pool dilution technique according to ref. . Briefly, as shown in Supplementary Fig. 14, duplicate fresh soil samples (~4 g) were pre-incubated, amended with N-labelled amino acids (10 at.%), and incubated in the dark at 15 °C. After 15 and 45 min, soil samples were extracted with KCl solution. The suspensions were filtered and separated into two parts: one part was used to determine the concentration of the amino acid, while the other part was used to analyse the N abundance of amino acids. Total free amino-acid concentrations were quantified using a mix of o-phthaldialdehyde and 3-mercaptopropionic acid by fluorometric determination. The N abundance of amino acids was measured by converting ammonium-free extract into NO using ClO under vacuum, following a modified protocol from ref. (see Supplementary Note 7 for more details).
Gross rates of soil ammonification and nitrification were measured in the field with a N pool dilution assay. Briefly, we collected five soil cores with a cylinder of known volume. Of these, one soil core was used to determine soil abiotic and biotic variables (that is, soil water content, inorganic N content and functional gene abundance), two cores were labelled with (NH)SO solution and the remaining two cores were labelled with KNO (Supplementary Fig. 13b). These four labelled soil cores were then placed back into their original locations in the soil for incubation. The two soil cores labelled with (NH)SO and KNO were collected and extracted with KCl solution after 15 min of incubation, while the other two cores were collected and extracted after 48 h. The concentrations of NH and NO in soil extracts were analysed using a flow injection analyser (AutoAnalyzer 3 SEAL, Bran and Luebbe). Soil extracts were immediately processed for diffusion of NH and NO on acid filter traps to analyse their N abundance (see Supplementary Note 8 for details).
On the basis of these measurements, the gross production (that is, protein depolymerization, ammonification and nitrification) and consumption (that is, microbial amino acid consumption, NH and NO immobilization) rates were calculated using the following equations:
where gross production and consumption are the N production and consumption rates (mg N kg d), t and t are the two incubation times (d), C and C denote the concentrations of total free amino acid, NH or NO at the two time points (mg N kg), and APE and APE represent the respective N at.% excess values (at.%N sample - at.%N background) of amino acid, NH or NO at the two time points (at.% N). The microbial immobilization rates were obtained from the sum of the rates of NH and NO immobilization.
Soil potential denitrification rate was measured by the acetylene inhibition method. Although this technique may underestimate denitrification by inhibiting microbial activity, it is widely used to determine the denitrification potential. Briefly, fresh soil (~5 g) from the top 10 cm was placed in 120-ml brown glass bottles. The bottles were sealed and purged with pure N (99.999% purity), and injected with 12 ml of acetylene to prevent NO to N reduction. Samples were incubated at 15 °C in the dark, with gas samples collected every 2 h over 6 h. NO concentrations in these gas samples were determined with a gas chromatograph (Agilent 7890A, Agilent Technologies). Finally, soil potential denitrification rate was obtained on the basis of the linear regression between the NO concentration and incubation time (0, 2, 4 and 6 h) for each sample. In addition, we estimated the denitrification rate using a N bulk soil model, and found that the denitrification fluxes were 0.29 ± 0.04 and 0.75 ± 0.14 g N m yr in control and warming plots, respectively (see Supplementary Note 9 for more details).
To characterize the effects of warming on plant growth and N uptake, we measured a series of plant N-related parameters. We first determined the N pools of leaf, leaf litter, root and root litter in each plot. Briefly, in August 2021, we randomly selected a 20 × 20 cm quadrat in each plot and collected aboveground litter and green leaves. After aboveground sampling, we collected three soil cores in the top 50 cm within the quadrat to collect root samples. Roots were rinsed and separated into live and dead parts by their colour and tensile strength. In the laboratory, leaf, leaf litter, root and root litter samples were processed and their N concentrations analysed (vario EL Ш). Finally, the plant N pools of green leaves, leaf litter, root and root litter were calculated on the basis of their biomass and N concentration.
The preferential uptake of different N forms was quantified with an in situ labelling experiment. In early August 2021, five 10 × 10 cm quadrats were randomly selected in each plot, with quadrats spaced at least 40 cm apart (Supplementary Fig. 13c). Analysis of root δN values in the unlabelled quadrats showed no outliers, indicating that these procedures could avoid potential N pollution among quadrats (Supplementary Fig. 15). Four N-enriched compounds were used in the field: (NH)SO, KNO, 2-CN-glycine and 4-CN-aspartic acid. Deionized water was applied to replace the labelling solution in unlabelled quadrats. Each quadrat was divided into nine subplots, and 1 ml of tracers was injected at the centre of each subplot using 10-cm needles (9 ml total). The amount of N added to soils was 40 mg N m. After 24-h labelling, all five whole quadrats in each plot were destructively sampled to 10-cm depth. Plants with attached roots were carefully excavated, gently rinsed to remove soil and then separated by species. Root samples were submerged in 0.5-mM CaCl for 3 min to remove root surface N. Leaves and roots were separated, dried and weighed. Both leaf and root samples were ground and analysed for their N concentration (vario EL Ш), and δC and δN values determined (EA-Isolink-253 Plus). The amount and rate of plant N uptake were calculated as follows:
where APE (%) represents the N atom excess, which is the difference in at.%N between labelled and unlabelled plant samples, and biomass and N are leaf or root biomass (g m) and their N concentration (%).
We measured foliar N resorption efficiency for three dominant species in 2023: Kobresia tibetica, Kobresia kansuensis and Carex atrofusca, which together accounted for over 85% of cumulative relative abundance (Supplementary Fig. 16). Due to the inaccessibility of roots and the complexity of the root structure, we randomly dug out roots to a depth of 10 cm to quantify root exudation rates and fine-root traits of these three dominant species in 2021 (see Supplementary Note 10 for details). Additionally, we measured root enzyme activities (that is, β-1,4-N-acetylglucosaminidase and leucine aminopeptidase) using a microplate fluorometric assay according to ref. (see details in Supplementary Note 11).
To quantify warming effects on ecosystem N losses, we monitored N leaching and gaseous losses as NO during the growing season in 2023. We determined the soil N leaching rate using in situ soil core incubations following ref. . Before incubation, a polyvinyl chloride (PVC) tube (5-cm diameter) was inserted to 10-cm depth in each plot, and the tubes were subsequently pulled out together with intact soil and plants. Then, a nylon mesh bag containing ~5 g of ion exchange resin (IRN-150) was buried beneath the soil core in the tube, after which the PVC tube was returned to its original position. After incubation, the PVC tubes were dug out and the nylon mesh bags were collected, and then extracted with 1 mol l KCl solution to analyse the concentration of NH and NO with a flow injection analyser (AutoAnalyzer 3 SEAL). In each plot, one soil core with the resin bag set was incubated from July to August, while the other set was incubated from August to September. Finally, soil N leaching rate was calculated as the average of the July and August values.
We monitored ecosystem NO fluxes using a laser-based analyser (LI-7820 NO/HO Trace Gas Analyzer, LI-COR Biosciences) connected to a smart chamber (LI-8200-01S, LI-COR Biosciences) for the entire growing season in 2023 (May-October). In late April, a PVC collar (5-cm height, 20-cm diameter) was inserted to 3-cm depth in each plot. During each measurement, the chamber was placed on the PVC collars, and NO concentrations were recorded at 1-s intervals throughout a 150-s period for each collar. Ecosystem NO flux measurements were performed between 09:00 a.m. and 12:00 p.m. (local time) at 15-d intervals. To further assess the relative contributions of nitrification and denitrification to field NO fluxes, gas samples were collected to determine their NO concentration and natural abundance isotopic composition of NO (refs. ), and site preference (difference in δN at the central and terminal positions) was used in a two-end-member isotopic balance equation (see Supplementary Note 12 for more details).
To assess the responses of topsoil N-cycling parameters to warming, we measured a series of soil chemical variables, N-cycling-related extracellular enzymes and functional genes. For these, the dissolved organic N, NH and NO concentrations of fresh soil were extracted with 1 mol-l KCl solution and analysed as described above. Microbial biomass N content was measured with a Multi-N/C 3100 TOC/TN analyzer (Analytik Jena) after carrying out a chloroform-fumigation extraction procedure. We calculated the microbial biomass N content as the difference in the organic N content between fumigated and unfumigated extracts with a conversion coefficient of 0.54 (ref. ). The activities of two soil hydrolase enzymes, β-1,4-N-acetylglucosaminidase and leucine aminopeptidase, were determined using fluorometric techniques (see details in Supplementary Note 11). Additionally, we quantified the abundance of functional genes involved in nitrogen cycling, including nifH for biological N fixation, ureC for ammonification, bacterial and archaeal amoA for nitrification and nirS, nirK and nosZ for denitrification (see Supplementary Note 13 for more details).
The data were checked for normality and log-transformed when necessary, and analysed in the following four steps. First, to capture the cumulative response of soil N stocks to decadal warming, we conducted paired-samples t-tests to compare the means of soil N stocks and soil N content between warming and control treatments for each year. Second, to assess the response of N-cycling variables to warming, we used paired-samples t-tests to examine the differences in the N-cycling variables between warming and control plots. We then calculated the response values (%) to evaluate the effects of warming on soil and root microbial properties as well as root exudation relative to the control plots as follows:
where T is the value of variables in warming plots, and T is the value in control plots.
Third, to test the effects of warming on the seasonal dynamics of NO fluxes, we conducted a repeated-measures analysis of variance with the warming treatment as a between-subject effect and the sampling date as a within-subject effect. Finally, linear mixed models were used to quantify the associations of N-cycling parameters with edaphic factors (that is, soil temperature, soil NO content and functional gene abundance). In the linear mixed-effects models, biotic and abiotic variables were included as fixed factors, and block was treated as a random effect. The linear mixed-effects models were constructed using the lme4 R package. All analyses were conducted in R v.4.3.2 (ref. ).