Soil moisture shifts under warming are associated with divergent topsoil nitrogen responses in permafrost-affected ecosystems

Abstract

Understanding how soil nitrogen (N) responds to climate warming is critical for predicting permafrost carbon–climate feedbacks, yet long-term N responses and their underlying mechanisms remain elusive. Here we show that divergent topsoil N responses to warming are associated with warming-induced shifts in soil moisture. Twelve-years of high-level warming increases topsoil N stocks in mesic alpine meadows but triggers substantial topsoil N losses in waterlogged swamp meadows. Enhanced topsoil N retention in the alpine meadow coincides with warming-driven soil drying, reduced gross protein depolymerization, and elevated microbial N immobilization. By contrast, declining topsoil N stocks in the swamp meadow coincide with elevated soil moisture, accelerated N transformation rates, greater plant N accumulation, and elevated potential N losses. A meta-analysis further indicates that such moisture-associated divergent N responses are widespread across permafrost-affected ecosystems. These findings highlight the necessity of hydrological shifts when projecting N cycling and permafrost-climate feedbacks under climate warming.

A 12-year warming experiment shows divergent responses of topsoil nitrogen stocks to warming in alpine and swamp meadows, with these contrasting responses associated with warming-induced shifts in soil moisture.

Introduction

Permafrost regions cover approximately 15% of the Earth’s land surface yet store more than 1300 Pg of soil organic carbon (C), making them one of the largest terrestrial C reservoirs^1,2. Over the last four decades, warming in permafrost regions has proceeded at two to four times the global average³, a pronounced amplification that threatens to remobilize previously frozen soil C through enhanced decomposition, and thereby accelerate positive C–climate feedbacks. Under a warming climate, the direction and magnitude of permafrost C–climate feedbacks may partly depend on soil nitrogen (N)^4–6, because N regulates plant growth, organic matter decomposition, and greenhouse gas emissions⁷. Therefore, understanding how soil N responds to warming and which mechanisms are involved is essential for improving predictions of permafrost C cycling and its feedbacks to climate warming.

In contrast to the well-documented fate and vulnerability of soil C under climate warming, far less is known about how soil N responds to warming and which mechanisms are involved in permafrost ecosystems^5,8. Previous field experiments and global meta-analyses have examined soil N responses to warming, yet they report highly inconsistent outcomes, ranging from no detectable change to either increases or declines in soil N pools^9–11. A key reason for this inconsistency is that many existing warming experiments are short-term (<5 years), limiting their ability to capture the cumulative, delayed, or non-linear responses of soil N pools to long-term warming. More recently, ref. ¹². demonstrated that soil N stocks declined under decadal warming in a swamp meadow on the Tibetan Plateau, with this decline attributed to cumulative N losses via leaching and gaseous emissions, as well as enhanced plant uptake. However, whether long-term warming leads to soil N depletion or retention likely also depends on other factors, such as soil hydrological conditions, which may modulate microbial N transformations and plant N uptake. How warming-induced changes in soil moisture are linked to soil N responses remain poorly understood in permafrost-affected ecosystems.

Active-layer in permafrost regions encompass a wide range of soil hydrological regimes, from persistently waterlogged to relatively well-drained conditions¹³, which likely contribute to the high variability observed in soil N responses to warming. Although warming can accelerate microbial N transformation rates by increasing enzyme activity^9,14, its effect on soil N cycling is regulated by ambient soil moisture conditions. Under moist or waterlogged conditions, warming may enhance soil N cycling by relieving temperature constraints on microbial activity¹⁵. In contrast, under relatively dry conditions, warming can further reduce soil moisture, thereby suppressing microbial activity and limiting N transformation rates^16,17. Because soil moisture governs microbial community structure, enzyme activities, and soil oxygen availability¹⁸, warming-associated shifts in moisture regimes can alter soil N cycling pathways and, ultimately, soil N stocks. However, how warming and associated changes in soil moisture interact to regulate soil N cycling and stocks remains poorly constrained. Understanding these coupled effects is therefore essential for predicting future N availability in permafrost ecosystems and their feedbacks to the C–climate system.

To address these knowledge gaps, we conducted a 12-year warming experiment in alpine meadow and swamp meadow ecosystems on the Tibetan Plateau. These ecosystems share a similar climatic and topographic background but differ markedly in soil hydrological regimes. Alpine meadows are relatively dry and prone to warming-associated drying, whereas swamp meadows are waterlogged and buffered against soil moisture declines¹⁹. Such hydrological contrasts have been linked to contrasting responses of ecosystem functioning, including ecosystem productivity and C dynamics, to warming, a pattern likely underpinned partly by differences in soil N dynamics. Here, we tested the hypothesis that contrasting moisture responses under warming are associated with divergent topsoil N responses in permafrost regions. Through soil and plant sampling after long-term warming, we examined whether warming alters topsoil N stocks and the partitioning of N among soil and plant pools in these ecosystems. Specifically, we quantified changes in topsoil N pools and evaluated the processes potentially associated with these responses by measuring (i) plant and soil N stocks within the measured 0–10 cm plant–soil system, (ii) soil microbial N transformations using ¹⁵N isotope pool-dilution techniques to estimate gross protein depolymerization, gross mineralization, gross nitrification, and microbial inorganic N immobilization, and (iii) N allocation and retention pathways within the measured 0–10 cm soil–plant system. Concurrently, we conducted a meta-analysis to examine how soil N responses to warming vary across permafrost regions with contrasting soil moisture responses.

Results

Warming effects on topsoil N stocks in the 0–10 cm soil–plant system across meadow types

Within the 12-yr warming cohort initiated in 2012, topsoil N stocks in the 0–10 cm soil layer changed significantly, with divergent responses between the alpine and swamp meadow ecosystems (Fig. 1). In the alpine meadow, the high warming treatment significantly increased topsoil N stocks compared with low warming (p < 0.05; Fig. 1a), with an increase of approximately 56.0 ± 14.3 g N m⁻² relative to the control (Fig. 1a). In contrast, the swamp meadow exhibited a decline in topsoil N stocks under warming. Relative to the control, topsoil N stocks decreased by 39.7 ± 15.1 g N m⁻² under low warming and by 106.5 ± 10.4 g N m⁻² under high warming (p < 0.01; Fig. 1a). In the 9-yr warming cohort initiated in 2015, warming effects on topsoil N stocks were relatively small and showed greater uncertainty in both alpine and swamp meadows. Warming did not significantly alter soil bulk density, except for the low-warming treatment in the 9-yr warming cohort of the swamp meadow, suggesting that the observed changes in topsoil N stocks were more closely associated with changes in topsoil N concentration than with changes in soil bulk density (Supplementary Fig. 5).

Fig. 1. Divergent responses of nitrogen (N) stocks in the measured 0–10 cm soil–plant system to experimental warming in the alpine and swamp meadow.

Topsoil N stocks (a), plant N stocks (b), and combined soil–plant N stocks (c) are shown for the 0–10 cm soil layer under different warming treatments. Measurements were made in 2024. The warming treatments established in 2012 and 2015 are referred to as the 12-yr and 9-yr warming cohorts, respectively, based on the time elapsed between treatment establishment and sampling. These cohorts were initiated in different years and represent separate experimental cohorts rather than a continuous temporal warming gradient. In the swamp meadow, the 2015 cohort included only ambient and low-warming treatments because the high-warming OTCs were accidentally damaged. Green and orange indicate the nominal low- and high-warming levels, corresponding to intended +2 and +4 °C, respectively. Bars represent means, error bars indicate ± SE (n = 3), and overlaid points show a single measurement under different warming treatments. Asterisks indicate significant differences from the control († 0.05 ≤ p < 0.1;*p < 0.05; **p < 0.01).

Warming also altered N stocks in the measured 0–10 cm soil–plant system, with contrasting responses between the two ecosystems (Fig. 1b, c). In the 12-yr alpine meadow warming cohort, high warming increased plant N stocks by 43.6 ± 10.7 g N m⁻² and increased combined 0–10 cm soil–plant N stocks by 98.6 ± 21.2 g N m⁻² relative to the control. In contrast, although plant N stocks also increased under warming in the swamp meadow, combined soil–plant N stocks declined. In the 12-yr swamp meadow warming cohort, high warming decreased N stocks in measured 0–10 cm soil–plant system decreased by 63.0 ± 10.5 g N m⁻² relative to the control.

Divergent N cycling and transformation pathways across meadow types

Warming altered multiple N-cycling proxies in the alpine meadow but had weaker effects in the swamp meadow. Specifically, warming significantly increased leaf C:N ratio in the alpine meadow, with no significant change in root C:N ratio. In the swamp meadow, neither leaf C:N nor root C:N ratios were significantly affected by warming (Fig. 2a).

Fig. 2. Divergent responses of soil N cycling proxies and microbial functional genes to warming in the alpine and swamp meadows in 2024.

a Plant C:N ratios and soil N transformation rates, including leaf C:N ratio, root C:N ratio, gross N mineralization, gross nitrification, gross protein depolymerization and microbial inorganic N immobilization, in control and warming plots of the alpine and swamp meadows. Bars represent means, error bars indicate ± SE (n = 3), and overlaid points show a single measurement under different warming treatments. b Soil N cycling functional genes and enzyme activities under different warming treatments in the alpine and swamp meadows. Points indicate the log response ratio (LnRR) of microbial functional gene abundance and LAP enzyme activity under different warming treatments relative to the control. Horizontal error bars denote 95% confidence intervals. The shaded backgrounds indicate the two warming durations represented by the 9-yr and 12-yr warming experiments. Green and orange indicate nominal low- and high-warming levels, corresponding to intended +2 and +4 °C, respectively. Functional genes represent key N-cycling processes: amoA for ammonia oxidation, gdh for microbial N assimilation and ammonification-related processes, ChiA for chitin degradation and organic N depolymerization, and ureC for urea hydrolysis. Asterisks indicate significant differences from the control († 0.05 ≤ p < 0.1; *p < 0.05; **p < 0.01).

Warming also induced distinct responses in soil N transformation processes and microbial functional potential across meadow types. In the alpine meadow, warming enhanced gross nitrification and microbial inorganic N immobilization, but reduced gross protein depolymerization, a key step in soil organic N mobilization (Fig. 2a). Conversely, several measured N transformation processes, including gross mineralization, gross protein depolymerization, and microbial inorganic N immobilization, increased significantly under warming in the swamp meadow (Fig. 2a). Consistent with these divergent responses in gross N transformation rates, microbial functional markers also showed contrasting responses between meadow types. Twelve years of high warming in 2012 cohorts significantly reduced the relative abundances of amoA, gdh, ChiA, and ureC, as well as LAP activity, in the alpine meadow, but significantly increased these variables in the swamp meadow (Fig. 2b).

Soil moisture responses in relation to topsoil N stock and process changes

When soil moisture was included as a covariate in the linear mixed-effects models, the estimated warming effects on topsoil N stocks and gross protein depolymerization rate (GPDR) were attenuated and, in some cases, became non-significant (Supplementary Table 6 and Supplementary Fig. 3). This suggests that soil moisture variation was associated with part of the variation in these variables under warming. We further analyzed the relationships between changes in soil moisture and changes in topsoil N stocks and GPDR to evaluate whether soil moisture responses were associated with these variables under warming. The LnRR of topsoil N stock (LnRRNstock) tended to decline with increasing LnRR of soil moisture (LnRRSM) in the alpine meadow (n = 12, p = 0.07, R² = 0.27, slope = −1.09) and declined significantly with increasing LnRRSM in the swamp meadow (n = 9, p < 0.05, R² = 0.57, slope = −1.21; Fig. 3a). In contrast, the LnRR of gross protein depolymerization (LnRRGPDR) increased significantly with increasing LnRRSM in both the alpine meadow (n = 12, p < 0.05, R² = 0.29, slope = 8.84) and swamp meadow (n = 9, p < 0.05, R² = 0.45, slope = 2.57; Fig. 3b). Across the two meadow ecosystems, LnRRSM was consistently associated with changes in both LnRRNstock and LnRRGPDR (Supplementary Fig. 6). Together, these results indicate that warming-related variation in soil moisture responses was associated with variation in both topsoil N stock changes and GPDR responses under warming.

Fig. 3. Relationships between warming-induced soil moisture shifts and soil N responses.

Relationships between the log response ratio of soil moisture and the log response ratio of topsoil N stock (a) or gross protein depolymerization rate (b). Points represent response ratios for individual warming treatments calculated relative to the corresponding control within the same meadow type and block. Colors indicate treatment categories, and symbols indicate meadow types (squares, alpine meadow; circles, swamp meadow). In both panels, regressions were fitted separately for each meadow type using 12 observations for the alpine meadow and 9 observations for the swamp meadow. For each meadow type, the colored line and shaded areas represent the fitted regression line and its 95% confidence interval, respectively.

Global meta-analysis insights from long-term warming experiments

Bayesian meta-analysis revealed no significant overall effect of warming on soil N stocks and N concentrations across all observations, with 95% credible intervals overlapping zero (Fig. 4a). However, when the data were partitioned according to soil moisture responses, contrasting patterns emerged between subgroups. For soil N stocks, warming significantly decreased soil N stocks in the subgroup with a positive or neutral response ratio of soil moisture (RRSM ≥ 0), with a posterior mean effect size of −0.59 and a 95% credible interval of −1.13 to −0.07. In contrast, no significant response was detected in the subgroup with RRSM < 0, where the posterior mean effect size was 0.35. A similar pattern was observed when the analysis was restricted to studies measuring soil N stocks within the upper 10 cm of soil (Supplementary Fig. 7). The soil N concentrations showed substantially greater uncertainty, with 95% credible intervals overlapping zero in both RRSM subgroups (Fig. 4a). Consistently, linear regression showed a negative relationship between RRSM and the effect size of soil N stocks under warming (slope = −3.06, p = 0.04; Fig. 4b). Because of the limited number of available observations, no significant warming effects were detected for either soil inorganic N, including NH₄⁺-N and NO₃⁻-N, or soil organic N indicators, including microbial biomass N and dissolved organic N (Supplementary Fig. 8).

Fig. 4. Effects of experimental warming on soil nitrogen across permafrost-affected ecosystems and their relationships with the response of soil moisture.

a Bayesian posterior distributions of warming effects on soil N stocks and concentrations across all observations and under different soil moisture response categories. Overall effect includes all observations, while RRSM ≥ 0 and RRSM < 0 represent observations in which soil moisture increased/remained stable or decreased under warming, respectively. Numbers indicate sample sizes. Filled black points with horizontal lines represent posterior means and 95% credible intervals for soil N stocks, and open diamonds with horizontal lines indicate posterior means and 95% credible intervals for soil N concentrations, respectively. The vertical dashed line indicates zero effect. b Regression showing the relationship between the response ratio of soil moisture and the warming effect size of soil N stocks and concentrations under warming. Filled circles represent soil N stock, and open diamonds represent soil N concentrations. The solid red line and shaded band show the fitted regression and 95% confidence interval for soil N stock, whereas the dashed orange line shows the fitted regression for soil N concentrations.

Discussion

Previous short-term warming studies have often reported limited or nonsignificant changes in soil N stocks^9,11,20. By contrast, our long-term warming experiment revealed contrasting topsoil N stock responses in the 12-yr warming cohort established in 2012, with the magnitude and direction of responses differing between meadow types and associated with contrasting soil moisture responses under warming. Previous studies showed that short-term warming can alter soil N turnover processes because microbes respond rapidly, but detectable shifts in soil N stocks, plant N uptake, and ecosystem N losses generally require prolonged exposure¹². Our results, therefore, underscore the importance of long-term experiments for understanding topsoil N stock responses to warming. Based on the contrasting topsoil N stock responses observed in the 12-yr warming cohort in alpine and swamp meadows (Fig. 1a), we quantified several key soil N-cycling processes and variables in 2024 to explore the potential mechanisms associated with the observed changes in topsoil N stocks.

Divergent responses of topsoil N stocks to warming across meadow types

The swamp meadow and alpine meadow exhibited contrasting responses of topsoil N stocks in the 12-yr warming cohort initiated in 2012. In the swamp meadow, topsoil N stocks declined significantly, with the strongest reductions observed under high warming. These results align with a recent study conducted in a Tibetan Plateau swamp meadow¹², which also reported topsoil N losses after 10 years of warming. In our swamp meadow, topsoil N stock declines under warming were accompanied by faster gross protein depolymerization, faster gross mineralization, and greater plant N accumulation (Figs. 1c and 2a). First, the observed responses of soil N transformation processes indicated enhanced internal soil N cycling under warming. Specifically, processes involved in organic N mobilization and inorganic N transformation were significantly elevated in the 12-yr warming cohort, suggesting that warming may have stimulated microbial activity^11,21. The observed responses of soil N cycling under warming were accompanied by increased relative abundances of microbial functional genes, such as ChiA and ureC and higher enzyme activities (LAP), consistent with greater microbial functional potential for organic N depolymerization and urea hydrolysis. Second, enhanced transformation of both organic and inorganic N may have increased the supply of inorganic N²², which is also supported by the observed increase in net N mineralization rates (Supplementary Fig. 9). This may have favored plant N uptake as well as potential soil N losses through leaching and gaseous emissions²³. We also observed an increase in plant N stock rather than plant N concentration, suggesting that the higher plant N stock was primarily driven by increased plant biomass (Supplementary Fig. 10). This pattern further suggests that warming may have promoted the transfer and accumulation of soil-derived N into perennial plant biomass, a pattern also reported in similar swamp meadow ecosystems, where increased plant N uptake contributed to approximately 30% of the decline in topsoil N stock¹². However, the increase in plant N stocks was much smaller than the decrease in topsoil N stocks (Supplementary Table 8), suggesting an apparent imbalance within the measured 0–10 cm soil–plant system. This imbalance indicates that part of the mobilized N may have been redistributed or lost through pathways not fully quantified in this study. In line with possible changes in N losses pathway, the isotope-derived f_gas index, calculated from bulk soil–plant δ¹⁵N model²⁴ (Supplementary Note 3), was significantly higher under high warming in the 12-yr cohort (Supplementary Table 9). However, this index should be interpreted as an indirect isotopic pattern compatible with altered gaseous N loss pathways, rather than as a direct measure of denitrification, N₂O flux, or N₂ flux. This interpretation is also consistent with recent studies showing that, in swamp meadows, gaseous N losses pathways may contribute to soil N loss¹². It should be noted that hydrological N losses, including nitrate leaching and lateral N export, were not directly quantified in this study. However, they may also have contributed to the decline in topsoil N stocks in the swamp meadow, given the high soil moisture and strong hydrological connectivity of this ecosystem²⁵, as also suggested by increased soil NO₃⁻-N concentrations (Supplementary Fig. 11). These pathways and in-situ fluxes should be directly monitored and quantified in future work.

In contrast, warming in the alpine meadow triggered the opposite pattern: topsoil N stocks increased under high warming in 12-yr warming cohorts but remained largely unchanged under low warming. The combined topsoil and plant N stocks in the measured 0–10 cm soil–plant system increased by approximately 98.6 g N m⁻² under 12-yr high warming, indicating N accumulation within the upper soil–plant compartment. At least three non-exclusive mechanisms may have contributed to this increase in topsoil N stocks. First, GPDRs were significantly lower under 12-yr high warming (Fig. 2a), suggesting reduced organic N mobilization at a key step in soil N cycling²⁶. Reduced protein depolymerization may have constrained organic N activation²¹ and, in turn, reduced the production and mobility of inorganic N²⁷. LAP enzyme activity and the relative abundance of ChiA, both associated with macromolecular organic N degradation, also decreased under warming. This suggests that reduced extracellular enzyme activity and microbial functional potential may have contributed to the observed changes in soil N-cycling processes in the alpine meadow. Consistent with this pattern, net N mineralization rates also declined significantly in the alpine meadow under warming (Supplementary Fig. 9). Second, enhanced microbial inorganic N immobilization and organic matter stabilization may have further contributed to topsoil N retention in the alpine meadow. Increased microbial N demand, together with reduced organic N mobilization, may have promoted the retention of N in microbial necromass, which could subsequently be stabilized as mineral-associated organic N (MAON)^28,29. In our study, MAON increased in the alpine meadow under warming (Supplementary Fig. 12). Third, isotope-based evidence provided an additional, but indirect, indication of altered N loss pathways in the alpine meadow. In particular, 12-yr high warming significantly decreased f_gas. This pattern is also consistent with observations from other recent warming experiments in high-altitude forests¹⁷, although the magnitude of gaseous N losses remains uncertain in our study. Although we did not directly quantify hydrological N losses, the relatively low soil moisture in this ecosystem, together with further moisture reductions under warming, suggests that leaching losses were likely limited. Taken together, the observed patterns suggest that long-term high warming increased topsoil N stocks in the alpine meadows in association with reduced gross protein depolymerization, enhanced microbial inorganic N immobilization, increased MAON, and isotope patterns compatible with altered gaseous N loss pathways. This pattern contrasts sharply with the response observed in the swamp meadow.

Across both alpine and swamp meadows, topsoil N stock responses under the 9‑year warming treatment followed directional patterns comparable to those of the 12‑year cohort, albeit with smaller effect magnitudes. Nevertheless, the disparities between the two warming cohorts cannot be interpreted as a progressive temporal trajectory of warming impacts unfolding between year 9 and year 12. These two experimental cohorts were established in different years and constitute independent experimental units, instead of sequential repeated measurements on identical plots. Accordingly, contrasts between the cohorts reflect divergences across distinct long‑term warming experiments, and do not constitute evidence for time‑dependent accumulation of warming effects.

Changes in soil moisture were associated with soil N responses to warming

Warming by OTCs often reduces soil moisture, as observed in our alpine meadow, likely because OTCs alter the energy balance and enhance evapotranspiration^30,31. However, in some high-elevation and high-latitude ecosystems, especially permafrost-affected peatlands, tundra, and wet meadows, warming does not always lead to drying and may instead have neutral or positive effects on soil moisture^32,33 (Supplementary Fig. 13). In the swamp meadow, soil moisture increased under warming. One possible explanation is that this poorly drained ecosystem has a high groundwater table (Supplementary Fig. 14). Under warming, warming-related thaw and hydrological changes could help maintain groundwater closer to the soil surface, and this groundwater supply may partly offset or even exceed warming-induced evaporative losses^32,33. This explanation requires further validation through future groundwater and hydrological monitoring. In addition, the OTC designs differed in chamber height and ground area. These structural differences may also influence wind conditions, radiation, precipitation interception, and snow accumulation, thereby potentially influence soil moisture dynamics and contribute to variation in topsoil N responses under warming.

The divergent responses of topsoil N stocks between the two meadow types were associated with differences in soil moisture responses to warming. Although our experiment did not independently manipulate soil moisture and therefore cannot fully separate the effects of warming and soil moisture, the observed soil moisture responses to warming were associated with part of the variation in topsoil N stock responses (Fig. 3a). This pattern was further supported by our meta-analysis, which showed that soil N stocks tended to increase when warming reduced soil moisture, but tended to decline when warming increased or maintained soil moisture (Fig. 4). However, differences in vegetation, soil properties, and hydrological conditions between alpine and swamp meadows may also contribute to these contrasting responses.

Changes in soil moisture under warming may influence soil N dynamics through two potential pathways: (i) by regulating microbial activity, functional potential, and metabolism^34,35; and (ii) by affecting potential N losses via leaching and gaseous emissions^7,8,12 (Fig. 5). Under high warming in the 12-yr cohort, reduced soil moisture in the alpine meadow may have imposed drought stress on microbial communities and constrained some soil N-cycling rates. Soil moisture can directly affect the diffusion of extracellular enzymes, substrates, and decomposition products among plant roots, microbial cells, and the surrounding soil environment^35,36, thereby potentially constraining organic N mobilization and subsequent inorganic N production. By contrast, in the swamp meadow, higher soil moisture and greater buffering against warming-induced drying may have allowed microbial N cycling to remain active or even increase²⁵. These findings are supported by several global meta-analyses showing that ecosystems with higher precipitation tend to sustain higher rates of soil N cycling^27,37. Second, soil moisture can affect hydrological connectivity and soil aeration, thereby regulating the potential for hydrological and gaseous N losses. Reduced soil moisture in the alpine meadow may have limited hydrological N transport and constrained microbial processes associated with gaseous N production by reducing anaerobic microsites, thereby potentially favoring soil N retention under long-term high warming. In contrast, persistently wetter conditions in the swamp meadow may have promoted dissolved N transport and increased the potential for N losses through hydrological transport and gaseous pathways¹². This interpretation is also consistent with the observed increase in isotope-derived f_gas index and the changes in genes related to denitrification potential under warming in the swamp meadow (Supplementary Fig. 15). However, these indicators should be interpreted as indirect evidence of altered N loss pathways rather than direct measurements of gaseous N losses.

Fig. 5. Conceptual diagram of divergent topsoil N responses to long-term warming in alpine and swamp meadows.

Red arrows represent measured N-cycling processes, and grey arrows represent unmeasured potential N loss pathways. Plus and minus symbols indicate positive and negative responses under 12-yr high warming in 2012 warming cohorts, respectively.

Limitations and future work

Our long-term warming experiment, soil–plant N-cycling measurements, and meta-analysis provide evidence that topsoil N stocks respond divergently to warming in alpine and swamp meadows. However, several limitations should be acknowledged. First, the relationships among warming-associated soil moisture changes, topsoil N stocks, and N transformation processes remain correlative. Because soil moisture was not independently manipulated, and because the alpine and swamp meadows differ in their vegetation and hydrological setting, our analyses cannot fully separate the independent effect of moisture from those of warming and other inherent ecosystem-level differences. Future studies combining warming with drainage or irrigation treatments would help resolve these effects^34,38. Second, several important N fluxes were not directly quantified, and key N transformation processes were measured only once. Therefore, the observed N-cycling responses should be interpreted cautiously in relation to changes in topsoil N stocks and should be viewed as point-in-time process-level evidence rather than as complete explanations of long-term topsoil N stock changes. Given the strong seasonality and freeze–thaw dynamics of permafrost ecosystems, repeated seasonal measurements are needed to better resolve temporal variation in soil N cycling^39,40. Third, all treatments were sampled on the same calendar date, but warming may have altered cumulative thermal exposure (Supplementary Table 10), plant phenology, and microbial activity. Therefore, identical sampling dates may not represent equivalent thermal or phenological stages among treatments. Future studies using growing-degree-days-based or phenology-adjusted sampling would help address this limitation⁴¹. Finally, our measurements focused on the upper 0–10 cm soil layer. Because warming may also affect N redistribution among deeper soil layers⁴², plant compartments, and hydrological flow paths, broader vertical and ecosystem-scale observations are needed to better constrain the fate of N under long-term warming.

Our findings have important implications for understanding and predicting N cycling feedbacks under climate change across the Tibetan Plateau. This region is expected to undergo continued warming and changes in precipitation regimes, which may alter the distribution and hydrological conditions of alpine ecosystems⁴³. Our results show that warming-induced changes in topsoil N stocks differed between alpine and swamp meadows, with relatively greater N retention observed in alpine meadows and stronger topsoil N declines observed in swamp meadows. These contrasting responses were associated with ecosystem-specific moisture changes under warming. These findings highlight the importance of incorporating ecosystem-specific moisture dynamics under warming into models of C–N feedbacks in permafrost regions.

Methods

Study area

The study was conducted in the Fenghuo Mountains, within the core area of the Hoh Xil Nature Reserve, on the central Qinghai–Tibetan Plateau, China (34°43.716′N, 92°53.567′E), at an elevation of approximately 4850 m above sea level. The region has a mean annual temperature of −5.3 °C and a mean annual precipitation of 269.7 mm, with approximately 80% of precipitation occurring during the growing season from May to September. Alpine meadows and swamp meadows are the two dominant meadow types in this region, together covering approximately 80% of the land area. Alpine meadows are dominated by Carex parvula, Carex atrofusca subsp. minor, Carex scabriostris, Kobresia humilis, Kobresia capillifolia, Kobresia myosuroides, and Kobresia graminifolia, whereas swamp meadows are mainly composed of Kobresia tibetica, Carex moorcroftii, Caltha palustris, and Ranunculus japonicus²⁵. The region is underlain by permafrost, with an active layer thickness ranging from 80 to 150 cm⁴⁴. The soil is characterized by a loamy clay texture and is classified as Gelisol according to the U.S. Soil Taxonomy and as Cryosol according to the UN-FAO World Reference Base of Soils. Detailed information about study site is provided in the Supplement Materials (Supplementary Table 1).

Warming experiment

The warming experiment included two warming magnitudes and two separate warming cohorts in both the alpine and swamp meadows (Supplementary Fig. 1). The 12-yr warming cohort was initiated in 2012, and the 9-yr warming cohort was initiated in 2015. Because these cohorts were initiated in different years and established in separate experimental plots, they do not represent repeated measurements of the same experimental units over time. Therefore, we treated them as separate warming cohorts rather than as a continuous warming-duration gradient. In each meadow type, three 6 × 6 m blocks with similar vegetation were randomly selected within a flat 100 × 100 m fenced area. Each block contained both the 2012 and 2015 warming cohorts. Within each block, control, low-warming, and high-warming were established for the 2012 cohort. The 2015 cohort followed the same design in the alpine meadow, whereas the high-warming in the swamp meadow was excluded because accidental damage to the OTCs resulted in insufficient replication. Experimental warming was imposed using open-top chambers (OTCs) made of translucent synthetic glass³¹. Two OTC sizes were used to represent projected low (≥ 2 °C) and high (≥ 4 °C) warming scenarios for high-altitude regions by the end of the century⁴⁵: a shorter chamber, 40 cm high and covering 1.2 m², representing low warming, and a taller chamber, 80 cm high and covering 2 m², representing high warming. The OTCs were installed at the beginning of each warming experiment and remained in place throughout the experiment.

Two separate groups of shorter and taller OTCs were randomly installed for continuous monitoring of soil temperature and soil volume moisture content to avoid potential sensor damage caused by repeated destructive soil sampling. Soil temperature and volumetric soil moisture content at 5 cm depth were recorded at 30-min intervals in warming and control using automated thermometers and thermistor sensors (EM50, Decagon, USA) beginning in 2012. To avoid disturbance from soil sampling, no samples were collected within the OTCs until 2018.

The warming levels are hereafter referred to as nominal low- and high-warming levels, corresponding to the intended +2 and +4 °C warming levels, respectively. Because of sensor malfunction and missing records in some later years, valid soil temperature observations were available for the alpine meadow from 2013 to 2020 and for the swamp meadow from 2013 to 2016 (Supplementary Fig. 2). Faulty sensor records and periods with missing or clearly erroneous values were excluded from the analysis. However, because continuous microclimate records were not available throughout the entire experimental periods, the realized warming magnitudes could not be directly verified for all 9-yr or 12-yr treatments. Therefore, these warming levels should be interpreted as nominal treatments rather than continuously verified temperature increases throughout the full 9-yr or 12-yr experimental periods. Based on the available valid observations, warming treatments resulted in higher growing-season soil temperatures at 5 cm depth compared with the corresponding controls during the monitoring periods for both alpine and swamp meadows (Supplementary Fig. 2). Soil moisture responses differed between meadow types: growing-season soil moisture was reduced on average by approximately 6% under nominal low warming and 16% under nominal high warming in the alpine meadow during 2013–2020, whereas it increased by approximately 18 and 23%, respectively, in the swamp meadow during 2013–2018 (Supplementary Table 2).

Soil and plant sampling and analysis

In July 2024, one composite soil sample per subplot, either OTC or control, was collected from the 0–10 cm soil layer by combining three soil cores taken within each subplot using an 8-cm-diameter hollow steel sampler (500 cm³ per core). Immediately after collection, all samples were stored at approximately 4 °C in a portable refrigerator for 2 days in the field and then promptly transported to the laboratory for subsequent analyses. All analyses were completed within 2 weeks of sampling. Fresh soil samples were sieved through a 2-mm mesh to remove roots and debris. The sieved soil was divided into two subsamples. One subsample was stored at −20 °C for metagenomic analyses, whereas the other was stored at 4 °C for subsequent analyses. A portion of the fresh soil was air-dried and finely ground with a ball mill for determination of soil N concentration using a CN elemental analyzer (Vario Macro, Elementar, Langenselbold, Germany). Additional fresh soil aliquots were oven-dried at 105 °C to determine gravimetric soil water content (Supplementary Table 3). Fresh soils were used to measure ammonium N (NH₄⁺-N), nitrate N (NO₃⁻-N), and total free amino acids (TFAA) concentrations.

Soil NH₄⁺-N and NO₃⁻-N concentrations were extracted using 2 M KCl at a soil-to-solution ratio of 1:10 (w/v), shaken for 1 h, filtered, and analyzed using a continuous flow analyzer (SAN+, Skalar Analytical BV, Breda, The Netherlands). Soil TFAA were quantified using the o-phthaldialdehyde and 3-mercaptoethanol fluorescence method^46,47 and measured with a Cary Eclipse fluorescence spectrophotometer (Agilent, Santa Clara, CA). Soil bulk density (BD) was determined following the core method described by ref. ⁴⁸. Samples were dried at 105 °C to constant mass and weighed. Coarse rock fragments (> 2 mm) were then separated and weighed, and soil BD was calculated using sample masses before and after fragment separation.

In July 2024, aboveground plant biomass and belowground root biomass were sampled separately in each subplot. For aboveground biomass, one 20 × 20 cm quadrat was randomly placed within each subplot, and all aboveground plant material within the quadrat was harvested. For belowground root biomass, three soil cores were randomly collected from the 0–10 cm soil layer within the same quadrat using a 3.8-cm-diameter root corer. Roots from the three cores were pooled to obtain one composite root sample per subplot. In the laboratory, roots were carefully washed and oven-dried at 65 °C to constant mass. Aboveground biomass was converted to an areal basis according to the quadrat area, whereas belowground root biomass was converted to an areal basis according to the total basal area of the three root cores (Supplementary Note. 1). In the laboratory, green leaves and fine roots were carefully cleaned and oven-dried at 65 °C to constant mass, and ground to a fine powder in a ball mill. Previous observations at the same experimental site showed no substantial warming-induced changes in plant species richness or community composition in either the alpine or swamp meadow^25,49. Plant total N concentrations were determined using a CN elemental analyzer (Vario Macro, Elementar, Langenselbold, Germany). Plant N stocks were calculated as the product of dry biomass per area (kg m⁻²) and tissue N concentration (g N kg⁻¹). Topsoil N stocks were calculated as the product of bulk density, soil N concentration, and soil layer depth.

Soil nitrogen transformation processes

We quantified three key soil N transformation processes—gross protein depolymerization (GPDR), gross mineralization (GM), and gross nitrification (GN)—in 2024 using ¹⁵N isotope dilution techniques^21,50,51. This approach also allowed estimation of microbial immobilization of inorganic ¹⁵N-labelled substrates, that is, ammonium and nitrate. Specifically, a ¹⁵N-labelled mixture of 20 algal amino acids, (¹⁵NH₄)₂SO₄, and K¹⁵NO₃, all at 98 atom% ¹⁵N and purchased from Cambridge Isotope Laboratories and Sigma Aldrich, was applied to trace GPDR, GM, and GN, respectively. The amount of ¹⁵N tracer added comprised approximately 20% of the corresponding native target pool (Supplementary Table 4), minimizing potential fertilization effects while ensuring sufficient sensitivity for tracer determination.

Soil samples were incubated at 15 °C in the laboratory to determine potential gross N transformation rates under standardized conditions. This temperature falls within the growing-season temperature range at our study sites and has also been used in previous studies conducted at nearby sites¹². For GPDR, two aliquots per sample were incubated and extracted with 20 mL of cold (4 °C) 1 M KCl at 15 and 60 min after tracer addition⁵⁰. For GM and GN, soil samples were extracted at 0.5 and 24 h after tracer addition using the same KCl extraction procedure²¹. All suspensions were shaken for 1 h, filtered through ash-free cellulose filters, and stored at −20 °C until isotope analysis. The ¹⁵N abundances of TFAA-N, NH₄⁺-N, and NO₃⁻-N were determined using chemical conversion method, in which each N form was converted to N₂O prior to isotopic analysis^50,52,53.

Metagenome sequencing and microbial functional genes analysis

Soil genomic DNA was extracted and fragmented using a Covaris M220 instrument (Gene Company Limited, China). Libraries were prepared with the NEXTFLEX Rapid DNA-Seq Kit (Bioo Scientific, Austin, TX, USA) and sequenced as paired-end 150 bp reads on an Illumina NovaSeq 6000 platform with the NovaSeq 6000 S4 Reagent Kit v1.5 (300 cycles) at Majorbio Bio-Pharm Technology Co., Ltd. (Shanghai, China). After quality control and adapter removal with fastp v0.21.0, we obtained 995.8 Gb of clean reads, with 12.3–17.5 Gb per sample and an average of 14.5 Gb. De novo assembly was performed with MEGAHIT v1.2.9, ORFs were predicted using Prodigal v2.6.3, and redundant sequences were clustered with CD-HIT v4.8.1 at 95% nucleotide identity⁵⁴. Gene abundances were estimated by read mapping, and functions were annotated against the KEGG database⁵⁵.

Beyond the well-studied genes involved in mineralization (ureC), nitrification (amoA), and denitrification (nirK, nirS, and nosZ), we also targeted microbial genes involved in organic N depolymerization, such as chiA⁵⁶, and inorganic N assimilation, such as gdh, to explore microbial functional potential associated with the observed changes in N cycling (Supplementary Table 5). Chitinase ORFs were identified using six chitinase hidden Markov models from the KEGG database (K01183, K03791, K13381, K17523, K17524, and K17525), of which only the first three, K01183, K03791, and K13381, were detected in our dataset.

Data analysis

We first applied linear mixed-effects models separately for each meadow type to evaluate the effects of warming treatments on topsoil N stocks, plant N stocks, combined soil + plant N stocks within the 0–10 cm soil–plant system, plant C:N ratios, and soil gross N cycling rates. The 9-yr and 12-yr warming treatments were established as separate experimental cohorts in different years rather than as repeated observations of the same subplots. We therefore treated warming cohort as a categorical treatment and did not interpret differences between the 9-yr and 12-yr cohorts as a continuous temporal trajectory. In these models, warming treatment was specified as a fixed factor, with five levels in the alpine meadow: control, 9-yr low warming, 9-yr high warming, 12-yr low warming, and 12-yr high warming; and four levels in the swamp meadow: control, 9-yr low warming, 12-yr low warming and 12-yr high warming. Block was included as random factor. Post-hoc pairwise comparisons were performed using Tukey’s HSD adjustment for multiple comparisons. Model assumptions were checked using residual-versus-fitted plots, quantile–quantile plots, and the Shapiro–Wilk test for residual normality. These diagnostics indicated no major deviations from model assumptions. We further calculated log response ratios (LnRRs) to quantify warming effects on soil microbial functional genes and enzyme activities. For each block, log response ratios (lnRRs) were calculated as LnRR = ln (X_warming/X_control), where warmed plots were compared with their corresponding control plots within the same block.

To evaluate whether warming effects on topsoil N stock and GPDR were associated with warming-related variation in soil moisture, we fitted additional linear mixed-effects model separately for each meadow. These models included warming treatment and soil moisture as fixed effects, with block included as a random factor. We compared the estimated warming-treatment effects between warming-only models and models including soil moisture as a covariate (Supplementary Table 6). These analyses were used to assess whether soil moisture was associated with part of the variation in warming responses. We further used bootstrap approaches to estimate effect sizes and 95% confidence intervals (Supplementary Fig. 3). In addition, we calculated the LnRRSM, LnRRNstock, and gross protein depolymerization rates (LnRRGPDR) by comparing each warmed plot with its corresponding control plot within the same meadow type and block. Linear mixed-effects models were then used to evaluate the relationships of LnRRSM with LnRRNstock and LnRRGPDR within each meadow type. In these models, LnRRSM was included as a fixed effect, while block was treated as a random factor to account for the dependence among multiple warming treatments that shared the same block control. All analyses were performed in R v4.3.1⁵⁷.

To assess the generality of our experimental findings, we conducted a global meta-analysis synthesizing studies from permafrost-affected ecosystems with warming durations exceeding six years (Supplementary Note 2 and Supplementary References). Relevant published studies were identified through searches of the Web of Science and Google Scholar databases using the keywords “(Warming) AND (Soil N OR Soil nitrogen) AND (Soil moisture) AND (Tundra OR Arctic OR Subarctic OR Permafrost OR Tibetan OR Alpine)” (Supplementary Fig. 4). Based on these criteria, we compiled a total of 65 observations (including 31 observations for soil N stocks and 34 observations for soil N concentrations) from 30 studies (Supplementary Table 7). We extracted data on soil N and soil moisture responses to warming and calculated effect sizes using the standardized mean difference, Hedges’d, following ref. ⁵⁸. Sampling variance was calculated accordingly, and missing standard deviations (SDs) were conservatively estimated as 10% of the mean where SD was not reported⁵⁹; this applied to only four observations, accounting for 6.0% of the dataset. Bayesian meta-analysis was implemented in the brms package, which is robust for relatively small sample sizes because it estimates both posterior distributions and parameter uncertainty⁶⁰. Study identity was treated as a random effect to account for non-independence of multiple observations from the same study.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Y.P., T.W., and R.C. designed the research. Y.P., T.D., J.L., and M.L. performed the investigations. Y.P. curated and analyzed the data. G.W. and R.C. contributed to funding acquisition and project administration. Y.P. wrote the original draft with inputs from T.W., W.W., Y.Y., and R.C. All authors contributed to reviewing and editing the manuscript.

Peer review

Peer review information

Nature Communications thanks Tonghua Wu and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the National Natural Science Foundation of China [42430507, 42271123, 42477271], Sichuan Science and Technology Program [2025ZYD0009], the Youth Innovation Promotion Association CAS [Y2022096] and Innovation Cross Team-Key Laboratory project of the Chinese Academy of Sciences.

Data availability

Data used in this study are available from the Figshare⁶¹ data repository (https://doi.org/10.6084/m9.figshare.33152726). The sequencing data generated in this study have been deposited in the National Genomics Data Center (https://ngdc.cncb.ac.cn/gsa/s/y92UKchF).

Code availability

The code used for data analysis and figure generation is available in Figshare⁶¹ at https://doi.org/10.6084/m9.figshare.33152726.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76894-6.