This study was performed in a typical swamp meadow ecosystem at the Drought and Nitrogen Deposition Interaction Experiment Platform in Xiangdong village (33°37′16″N, 102°52′21″E, 3,400 m above sea level), Zoige county, Sichuan Province (northeastern Qinghai-Tibetan Plateau). The mean temperature in the region ranges from −0.7 to 1.1°C, with the coldest temperature in January at −10.5°C and the hottest in July at 11°C, respectively. The mean annual precipitation is 650–750 mm, which mainly occurs from June to September. The soil type is peat and swamp soil. The dominant plant species are Poa poophagorum, Elymus nutans, Carex atrofusca, and Potentilla anserina. Soil samples from the study site were collected before the experiment and analyzed. The soil properties of the top 0–20 cm layer were: pH 6.04, total organic C 66.8 mg g⁻¹, total N 5.09 mg g⁻¹, and total P 0.90 mg g⁻¹.

In June 2019, a 90 × 90 m plot was enclosed with a 1.6-m-high fence, which prevented herbivores from entering. An interactive experiment between drought and N enrichment was established using a randomized complete block design with two levels of drought (CK and drought) and six levels of N enrichment (0, 2, 4, 8, 16, and 32 g N m⁻² yr.⁻¹). Four replicate blocks were established and each block had 12 treatments randomly assigned to 4 × 4 m plots. Each plot was 2 m away from neighboring plots. We fixed rainout shelters on the plots to simulate drought in September 2019 and each rainout shelter had a roof made of curved bands of transparent acrylic that intercepted 50% of the rainfall while having a minimal impact on other environmental factors (Zhang et al., 2019a). Iron sheets were installed around the plots to a depth of 40 cm belowground to prevent lateral movement of water between the plots (Zhang et al., 2019b). Six levels of N enrichment consisted of the range from the plant N limitation to saturation according to a previous study conducted near the experimental site (Song et al., 2017). At the end of the 2020 growing season, the effect of the drought treatment on soil water content was tested and was significant, so we initiated the N addition treatment in May 2021. Coated slow-release urea was spread onto the soil surface of the plots by hand at the beginning of the growing season each year.

Before the experimental treatments were initiated, soil respiration collars (PVC pipe, 20 cm inner diameter) were installed in the ground in each plot. Two soil respiration collars were set up per plot; one shallow (5 cm) for measuring Rs, and one deep (40 cm) for measuring Rh. The trench method was successfully used in previous studies (Hanson et al., 2000; Keeler et al., 2009; Sayer and Tanner, 2010).

Soil respiration (Rs) and heterotrophic respiration (Rh) were measured with a portable soil carbon flux automatic measurement system (PS-9000, LICA, Beijing, China) once every 2 weeks during the growing season. The above-ground parts of newly growing plants were cut off at the surface in both collars in advance. Soil temperature (ST) and soil water content (SWC) were simultaneously measured with a soil temperature and humidity probe on a portable soil carbon flux automatic measurement system in the 10 cm soil layer during the carbon flux measurement.

At the end of August 2022, 48 soil samples were collected by drilling 0–20 cm soil layers in 48 plots, removing the stones, roots, and other impurities, and dividing them into two subsamples. Subsample 1 (10 g) was wrapped in foil, quickly placed in liquid nitrogen, transported back to the laboratory, and stored at −80°C for soil DNA extraction. Subsample 2 (200 g) was transported back to the laboratory and stored at −4°C to determine the soil physicochemical properties.

Soil-dissolved carbon (DOC) was extracted by adding 50 mL of 0.5 M potassium sulfate to subsamples of 12.5 g homogenized soil and agitating the sample on an orbital shaker at 120 rpm for 1 h. The filtrate was analyzed using a TOC analyzer (multi N/C 3100, Analytik Jena, Germany). Soil microbial biomass carbon (MBC) and microbial biomass nitrogen (MBN) were estimated using a chloroform fumigation extraction method (Brookes et al., 1985). Soil NH₄⁺ and NO₃⁻ concentrations were determined by extraction with 2 M KCl solution followed by colorimetric analysis on a FIAstar 5,000 Analyzer (FIAstar 5,000 Analyzer, Foss Tecator, Hillerød, Denmark). Soil pH was determined in a 1:2.5 soil: water solution (w/v).

Soil DNA was extracted from a 0.3 g soil sample using the DNeasy® PowerSoil® Kit (Qiagen, Hilden, Germany) following the manufacturer’s instructions. The universal primers used for prokaryotic 16S rDNA amplification were 515F (5′-GTG CCA GCM GCC GCG GTA A-3′) (Caporaso et al., 2011) and 806R (5′-GGA CTA CHV GGG TWT CTA AT-3′) (Walters et al., 2016). The universal primers used for fungal ITS rDNA amplification were ITS3-F (5’-GCATCGATGAAGAACGCAGC-3′) and ITS4-R (5’-TCCTCCGCTTATTGATATGC-3′) (White et al., 1990). The PCR was carried out in 50 μL PCR reaction volumes, containing 15 μL of NEB Next Ultra II Q5 Mix (NEB, New Ipswich, MA, United States), 3 μL of the forward primer (10 μM), 3 μL of the reverse primer (10 μM), 1 μL of template DNA, and 8 μL of nuclease-free water. The thermal-cycling conditions were 95°C initial denaturation for 30 s, followed by 30 cycles of denaturation at 95°C for 5 min, annealing at 56°C for 30 s; extension at 72°C for 40 s, and a final extension at 72°C for 10 min. The PCR products were purified using the GeneJET Gel Extraction Kit (Thermo Fisher Scientific Inc., Waltham, MA, USA). Illumina NovaSeq high-throughput sequencing was performed with a paired-end (2 × 250 bp) sequencing strategy (Magigene Co., Ltd., Guangdong, China).

The paired-end raw sequences were spliced using USEARCH (v11), and the low-quality sequences and primers were removed following the UPARSE pipeline (Edgar, 2013). The UNOISE3 denoising algorithm was used to generate zOTU representative sequences, and the zOTUs with sequence numbers <9 were removed (Edgar, 2016). The zOTU table was generated by mapping the zOTU representative sequence with the merged sequences via the otutab script. The taxonomic information annotations of the prokaryotes and fungi were prepared based on the SILVA138 and UNITE8.2 databases in QIIME2 (Quast et al., 2012; Nilsson et al., 2018; Bolyen et al., 2019). Finally, 24,131 and 4,004 zOTUs were obtained for prokaryotes and fungi, respectively. The prokaryotic and fungal sequence numbers in each sample were rarefied to 71,712 and 65,914, respectively, for subsequent analysis.

The rrn copy number of each OTU was searched using the rrnDB database and estimated according to its closest relatives with a known rrn copy (Stoddard et al., 2015). Then, we calculated the abundance-weighted average rrn copy number for each soil sample (Wu et al., 2017). We calculated the product of the estimated rrn copy number and the relative abundance of each OTU and summed these values of all OTUs for each sample.

Estimates of microbial activities, including hydrolases (α-glucosidase [AG], β-glucosidase [BG], and β-D-cellobiosidase [CB]) and oxidases (peroxidase [PEO] and polyphenol oxidase [PPO]) were based on the maximum potential activity of extracellular enzymes (Guan, 1986; Wallenstein et al., 2009).

The soil suspensions to measure the hydrolases were prepared by homogenizing a 1.0 g soil sample in 100 mL of 50 mmol L⁻¹ sodium acetate buffer. Then, a mixture of soil homogenate, methylumbelliferyl (MUB), and a MUB-linked substrate was placed in a black polystyrene 96-well microplate and incubated in the dark for 4 h at 25°C. The hydrolytic enzyme activities were expressed as nmol g⁻¹ h⁻¹. The soil suspensions for oxidases were prepared by homogenizing a 1.0 g soil sample in 10 mL of 1% pyrogallol solution. The mixture was placed in an incubator after shock and cultured at 30°C for 2 h. The oxidative enzyme activities were expressed as mg g⁻¹ h⁻¹. The LCI was calculated as the ratio of lnPPO to the sum of lnPPO and lnBG (Duan et al., 2023). The units of the BG and PPO activities were converted to nmol g⁻¹ MBC h⁻¹ before calculating the LCI.

All statistical analyses were performed in R version 4.1.3 (R Development Core Team, 2022). Repeated-measurement analysis of variance was employed to examine the effects of drought, added N, and date on seasonal Rs and Rh using linear mixed-effect models and the R package nlme. The drought and nitrogen-added treatments were set as fixed effects, block was set as a random effect, and a corAR1type time-autocorrelated covariance matrix was used to avoid violating the assumption of sphericity for repeated measurement of Rs and Rh data. To examine the effects of drought and added N on soil properties, seasonal mean Rs and Rh, rrn copy number, and soil extracellular enzyme activities, we used linear mixed-effect models with the lme4 and lmerTest packages, setting drought and the nitrogen addition treatments as fixed effects and block as the random effect (Bates et al., 2015; Kuznetsova et al., 2017; Pinheiro et al., 2022). Multiple comparisons followed by linear mixed-effect models were performed with the R package lsmeans (Lenth, 2016). The standardized regression coefficient and marginal R² were utilized to assess the effect size of fixed factors on Rh (Nakagawa and Schielzeth, 2013). The marginal R² was calculated using the partR2 package (Stoffel et al., 2021). Pearson’s correlation coefficients between factors were examined and visualized using the R package ggcor.

Principal coordinates analysis (PCoA) and permutational multivariate analysis of variance (PERMANOVA) were performed using the vegan package (Oksanen et al., 2018) to reveal the effects of drought and added nitrogen on soil prokaryotic and fungal community composition. All community composition distances were calculated based on Bray-Curtis dissimilarities. Actinobacteriota, Acidobacteriota, and Chloroflexi were classified as the K-selected (oligotrophic-associated) bacterial phyla, and Proteobacteria, Bacteroidota, and Firmicutes were the r-selected (copiotrophic-associated) bacterial phyla (Phung et al., 2004; Fierer et al., 2007; Nemergut et al., 2010; Francioli et al., 2016). Basidiomycota was classified as a K-selected fungal phylum, and Ascomycota and Mortierellomycota were as r-selected fungal phyla (Yao F. et al., 2017; Wu et al., 2021). The bacterial or fungal phyla ratios of K- to r-strategists were calculated based on the relative abundance. The responses of the relative abundance of the bacterial and fungal lineages (from phylum to class) to drought were determined using the linear discriminant analysis effect size (LEfSe) method (Segata et al., 2011). The LEfSe was performed using the online Huttenhower Galaxy server (huttenhower.sph.harvard.edu/galaxy) with a setting LDA score > 4.0.

A structural equation model (SEM) was used with the R package piecewiseSEM (Lefcheck, 2016) to examine the causal pathways by which drought and adding N affected Rh. Based on our knowledge of the effects of drought and added N on Rh, we developed an a priori model to allow a hypothesized causal interpretation of the linkages between SWC, DOC, the LCI, prokaryotic community composition, fungal community composition, the abundance-weighted rrn copy number, the ratio of K- to r-selected bacteria (B_K:r), the ratio of K- to r-selected fungi (F_K:r), soil hydrolase activities, soil oxidases activities, and Rh. Prokaryotic and fungal community composition was represented by PC1 from the Bray-Curtis distance-based principal coordinate analysis. Soil hydrolase activities were calculated as the sum of the activities of AG, BG, and CB, and soil oxidase activities were calculated as the sum of PEO and PPO activities.

After the drought treatments, SWC decreased significantly from 33.02 to 20.89% (p < 0.001), while ST increased significantly from 13.65°C to 14.53°C (p < 0.001; Supplementary Figure S1 and Supplementary Tables S1, S2). DOC increased significantly under drought and added N (p < 0.001; Supplementary Figure S1 and Supplementary Tables S1, S2). MBC did not change statistically under the drought and added N treatments (Supplementary Figure S1 and Supplementary Table S1). MBN decreased significantly under drought (p = 0.0024), and increased under added N (p < 0.001; Supplementary Figure S1 and Supplementary Tables S1, S2). Soil inorganic N concentration, NH₄⁺, and NO₃⁻, increased under the added N treatment (p = 0.0003, p < 0.001; Supplementary Figure S1 and Supplementary Tables S1, S2). Soil pH decreased significantly from 5.68 to 5.65 by 0.03 under the added N treatment (p = 0.00081; Supplementary Figure S1 and Supplementary Tables S1, S2).

The temporal dynamics of soil respiration were consistent with the heterotrophic components, and all maximum rates occurred in July (Figures 1A,B; Table 1). Drought significantly decreased the mean growing season Rh and Rs values by 40.07 and 52.24%, respectively (p < 0.001, p < 0.001; Figures 1C,D; Supplementary Tables S1, S2). While adding N had no effect on the mean growing season Rh and Rs values (Figures 1C,D; Supplementary Tables S1, S2). The mean Rh/Rs ratio increased significantly over the growing season by 22.04% under drought (p < 0.001; Figure 2B; Supplementary Tables S1, S2), but did not change significantly under the N addition gradient (Figure 2A; Supplementary Table S1). Notably, the effects of adding N on Rs and Rh were not significant for the seasonal dynamics and the mean value (Figures 1A,B; Supplementary Table S1). Drought significantly increased the Rh to Rs ratio from 52.91 ± 1.17 to 64.57 ± 1.70 (p < 0.001), whereas adding N did not affect the Rh to Rs ratio (Figure 2A; Supplementary Tables S1, S2).

As shown in Figure 3A, the dominant prokaryotic phylum was Proteobacteria, followed by Acidobacteria, Verrucomicrobiota, Actinobacteriota, and Bacteroidota. As shown in Figure 3B, the dominant fungal phylum was Ascomycota, followed by Basidiomycota and Mortierellomycota.

Drought significantly changed prokaryotic (p = 0.0001, PERMANOVA) and fungal (p = 0.0001, PERMANOVA) community composition, while adding N had no effect (Supplementary Table S3). Drought increased B_K:r from 0.90 ± 0.042 to 1.08 ± 0.052 (p = 0.0051), and F_K:r from 0.35 ± 0.049 to 0.67 ± 0.126 (p = 0.014), respectively (Figures 4A,B; Supplementary Tables S1, S2). The rrn value decreased significantly from 1.94 ± 0.019 to 1.88 ± 0.017 under drought (p = 0.035; Figures 4A–C; Supplementary Tables S1, S2). However, adding N did not significantly change prokaryotic or fungal community composition or rrn (Figures 4A–C; Supplementary Table S1).

As revealed by the LEfSe analysis of bacteria at the phylum level, the relative abundance of Acidobacteriota and Proteobacteria decreased under drought, while the relative abundance of Actinobacteriota increased. The relative abundance of Thermoleophilia increased under drought, while the relative abundance of Blastocatellia, and Gammaproteobacteria decreased (Figure 4D). The LEfSe analysis of fungi at the phylum level revealed that drought significantly reduced the relative abundance of Mortierellomycota. Drought increased the relative abundance of Dothideomycetes but decreased that of Eurotiomycetes, Sordariomycetes, and Mortierellomycetes (Figure 4E).

Drought significantly increased the activities of AG and BG by 39.42 and 90.53%, respectively (p = 0.0072, p < 0.001, p < 0.014; Figures 5A,B; Supplemenatry Tables S1, S2). However, the activities of PEO and PPO decreased by 20 and 14.29%, respectively under drought (p < 0.001, p = 0.0026; Figures 5D,E; Supplemenatry Tables S1, S2). Adding N did not significantly change any of the enzyme activities (Figure 5; Supplemenatry Table S1). Drought significantly decreased the LCI from 0.53 to 0.52 (p < 0.001; Figure 5F, Supplemenatry Table S2).

Pearson’s correlations showed that Rh and Rs were significantly positively correlated with SWC, MBN, rrn, PEO, PPO, and the LCI (Supplementary Figure S3), and significantly negatively correlated with DOC, 16SPC1, ITSPC1, and BG (Supplemenatry Figure S3). Rh was significantly positively correlated with MBC, while Rs were significantly negatively correlated with AG and B_K:r (Supplementary Figure S3).

We analyzed the relationships between potential drivers (i.e., soil properties, microbial properties, and enzyme activities) and Rh to explore the controls of Rh. As a result, Rh was significantly associated with the factors (Figure 6A). Briefly, for soil properties, ST and DOC significantly attenuated Rh (Figures 6C,D). In contrast, Rh was significantly facilitated by the increases in SWC, MBC, and MBN (Figures 6B,E,F). While Rh was not correlated with NH₄⁺-N, NO₃⁻-N, or pH (Figure 6A). For microbial properties, the prokaryotic community composition, fungal community composition, and B_K:r were negatively correlated with Rh (Figures 6G,H,I), while rrn was positively correlated with Rh (Figure 6J). For enzyme activities, Rh decreased with AG and BG but increased with PEO, PPO, and LCI (Figures 6K–O).

The SEM explained 84% of the variation in Rh (Figure 7). Standardized total effects from the SEM showed that drought had a significant negative effect on Rh while adding N did not (Figures 7, 8). Except for drought, oxidases were the most significant factor affecting Rh (Figures 7, 8), in which oxidases exerted a positive effect. The standardized direct effect sizes of drought, hydrolases, and oxidases on Rh were − 0.5834, 0.0679, and 0.4279, respectively (Figure 8), but only the drought and oxidase pathways were significant (Figure 7). Drought, SWC, LCI, rrn, ITSPC1, B_K:r, F_K:r, and hydrolases were important factors that exerted indirect effects on Rh (Figures 7, 8). LCI, F_K:r, and hydrolases contributed to increased oxidases, the second most important factor that affected Rh, while B_K:r contributed to decreasing oxidases.

Previous studies have shown that drought and adding N change the microbial community composition (Allison et al., 2007; Treseder, 2008; Gao et al., 2021). However, our results show that only drought altered the microbial community composition (Figures 3C,D), which partially supported our hypothesis 1. The results of a study in semi-arid grasslands showed that reducing precipitation increases oligotrophs and decreased copiotrophs, which was consistent with our B_K:r results (Li et al., 2022). A lower proportion of r-selection and average potential growth rate under drought has been suggested by a lower abundance-weighted average rrn copy number (Roller et al., 2016). Consistent with a previous study (Bu et al., 2018), the abundance of Gammaproteobacteria, which has often been associated with copiotrophic bacteria (Kurm et al., 2017), was suppressed by drought (Figure 4C). Here, the increased B_K:r, F_K:r, and decreased rrn support that drought favors oligotrophic taxa. Water affects microbial dynamics as a transport medium (Tecon and Or, 2017). Our drought treatments significantly reduced SWC (Supplementary Tables S1, S2) and, therefore, decreased diffusion pathways of dissolved nutrients (Tecon and Or, 2017). The breakdown of hydrological links is thought to be the main reason why drought affects soil community composition (Carson et al., 2010). Pearson’s correlation analysis revealed that SWC was significantly correlated with B_K:r, F_K:r, and rrn (Supplementary Figure S3). Thus, the drought-induced decrease in SWC drove the microbes to shift to an oligotroph-dominated community.

The results in extracellular enzyme activities did not accord with our hypothesis 2. In this study, drought increased hydrolase activities, including AG and BG. Previous studies have reported that the potential activities of AG and BG increase under dry conditions, indicating a decrease in hydrolase turnover (Alster et al., 2013; Ochoa-Hueso et al., 2020). Furthermore, partly consistent with our third hypothesis, only drought decreased Rh in this study (Figure 1C), which agrees with some previous research on terrestrial ecosystems (Zhou L. Y. et al., 2016; Zhou X. et al., 2016; Veach and Zeglin, 2020; Zheng et al., 2021). Our SEM showed that drought reduced oxidases via the LCI, B_K:r, F_K:r, and hydrolases and subsequently, Rh (Figure 7). Drought affected fungal community composition by decreasing SWC and the LCI, then increasing F_K:r and decreasing oxidase activities. The B_K:r and F_K:r values increased, and the rrn value decreased in this study, indicating that drought shifted the microbial community from r-strategist-dominated to K-strategist-dominated (Figures 4A–C) (Roller et al., 2016; Duan et al., 2023). K-strategists are more associated with oxidases than r-strategists and effectively utilize recalcitrant C, including lignin (Chen et al., 2022; Morrissey et al., 2023). Our SEM showed that the increase in B_K:r and the decrease in rrn promoted hydrolase activity. However, hydrolases had no significant effect on Rh (Figure 7). Soil hydrolases and oxidases are related to labile and recalcitrant C, respectively (Sinsabaugh et al., 2008; Burns et al., 2013). Here, drought reduced the LCI (Figure 6F), indicating that the C substrate was more vulnerable to decomposition (Moorhead et al., 2013). According to the optimization of the cost/benefit ratio (Allison et al., 2011), more labile soil C substrate should increase the activities of hydrolase, and decrease that of oxidase. Microbial community structure may be less important for the turnover of more labile C because a broad phylogeny of taxa is capable of metabolizing simpler compounds (Berg and McClaugherty, 2014). In that case, the rate-limiting variables for Rh may depend more on oxidase than hydrolase activity of the microbial community even if the microbial life history strategy shifts. The r-strategist-dominated soils generally have a higher respiration rate than K-strategist-dominated soils, which may decrease Rh by reducing oxidase activities (Bailey et al., 2002; Six et al., 2006; Fierer et al., 2007; Waring et al., 2013; Malik et al., 2016). Oxidase genes were identified in γ-Proteobacteria, indicating it could be a potential oxidase producer (Tian et al., 2014). In this study, the relative abundance of γ-Proteobacteria decreased under drought (Figure 4D), causing the reduction of oxidase activities. According to the LEfSe, Mortierellomycota contributed to the change in the F_K:r value at the phylum level (Figure 4E). Moreover, members of Mortierellomycota are sensitive to reduced precipitation (Han et al., 2024). A previous study found that the relative abundance of Mortierellomycota is lower in dry ecosystems (Tedersoo et al., 2014). The phyla Mortierellomycota mostly includes saprotrophs in the soil (James et al., 2006; Tedersoo et al., 2018). Saprophytic fungi perform the initial steps in the decomposition of cellulose, lignin, and other complex macromolecules (Gessner et al., 2010; Berg and McClaugherty, 2014). Mortierellomycota is involved in the decomposition of recalcitrant C (Větrovský and Baldrian, 2013; Fang et al., 2018; Shi et al., 2020), this may be the reason why F_K:r increases oxidase activities. From the correlation analysis and structural equation model (Figures 7, 8), the bacterial community composition shift exerted a greater effect on oxidase activities than fungal community composition shift, which ultimately led to the decline of oxidase activity, and thus a decrease in Rh.

The estimate of Rh in this study may have some limitations. First, the Rh value may have been overestimated, because the trenched subplots allowed root ingrowth underneath the collar (>0.6 m depth) into the subplots (Sayer and Tanner, 2010). Second, trenched subplots may exhibit a different microbial community composition from non-trenched subplots, which would change the Rh value (Chen et al., 2016). Third, long-term collar deployment leads to bias in soil respiration measurements, which contributes to higher soil bulk density and lower microbial biomass; inside long-term collars can underestimate Rh (Ma et al., 2023). Finally, the conclusions were drawn from Rh during the growing season. It is necessary to consider the seasonal pattern of Rh in response to drought and added N, including the growing and nongrowing seasons. In addition, we also assayed the potential activities of soil extracellular enzymes. An assay of potential enzyme activities usually provides an unlimited and relatively simple soluble substrate and is usually performed at a constant temperature, which is inconsistent with reality and may result in a misestimate of enzyme activity (Wallenstein and Weintraub, 2008).

In this study, adding N had no significant effect on Rh. Adding N has been reported to reduce Rh by decreasing microbial biomass (Treseder, 2008; Liu and Greaver, 2010). Here, MBC did not change significantly after adding N (Supplementary Table S1). Previous global meta-analysis revealed that MBC decreases with increasing experimental duration, indicating that the negative effects of adding N on microbes become more pronounced over time (Zhang et al., 2018). We predict that adding N will decrease Rh because soil microbes suffer progressive inhibition and continue to decrease in the long term. However, drought increases the availability of nitrogen, which can harm phenols, and reductions in phenols can increase Rh. Therefore, experiments on the effects of long-term drought and added nitrogen on Rh are full of uncertainties and should be continuously conducted.