The field experiment was initiated in 2010 under a summer rice (Oryza sativa L.)–winter wheat (Triticum aestivum L.) rotation at the Agro–Ecological Station of the Chinese Ecosystem Research Network in the Taihu Lake region of China. The experimental soil belongs to the category of Anthrosols and originates from lacustrine sediments. The field experiment adopted a randomized complete block design with three replicates of four treatments as follows: no straw retention (NS), wheat straw retention in the summer season (WS), rice straw retention in the winter season (RS), and wheat and rice straw return in the summer and winter seasons, respectively (WRS). All four treatment groups received the same amount of mineral N, P, and K fertilizers ( Supplementary Table S1 ). Crop residues in the straw retention treatments, including rice and wheat residues, were chopped into pieces measuring 5 to 10 cm in length and incorporated into the soil in a rotary manner approximately one week before the next crop season.

During the rice season in July 2020, three replicate soil (0–15 cm) and standing water samples were randomly collected from each plot and stored in sterile plastic bags and bottles, respectively. The water samples were filtered through 0.45 μm membranes and processed for chemical analysis. The soil samples were homogenized by passing them through a 2 mm sieve, divided into three parts, and then processed for chemical analysis (stored at 4°C), soil incubation experiments (stored at 4°C), and molecular assays (stored at −80°C).

The ammonium nitrogen and nitrite nitrogen in the filtered standing water samples were measured using an ultraviolet spectrophotometer (UV-1280, Shimadzu, Japan) according to Committee of Analytical Method of Water and Wastewater (2002). The concentration of water organic nitrogen was analyzed using a Multi N/C 2100S analyzer (Analytikjena GmbH, Germany) (Ghani et al., 2003). Soil pH was measured using a portable meter (Mettler Toledo, Switzerland) at a soil:water ratio of 1:2.5. Soil electrical conductivity (EC) was detected based on a 1:2.5 soil-to-water ratio using a conductivity salinity meter (Y SI-30, Yellow Springs, USA). Soil organic carbon (SOC) and dissolved organic C (DOC) were measured using TOC-VCPH equipment (Shimadzu, Japan) based on Lu (2000). Soil NH₄ ⁺-N, NO₃ ⁻-N, and nitrite nitrogen (NO₂ ⁻-N) were desorbed with 2 mol L⁻¹ KCl (1:5 soil:solution) and analyzed with an Auto-Analyzer (Skalar, The Netherlands). Soil total nitrogen was measured using the dry combustion method (Lu, 2000). Soil organic nitrogen (SON) was calculated as the difference between soil total nitrogen and the combined soil NH₄ ⁺, NO₃ ⁻, and NO₂ ⁻. Soil available potassium (AK) and phosphorus (AP) were desorbed with 1 mol L⁻¹ NaHCO₃ and measured by atomic absorption spectrophotometry and the molybdenum-blue method, respectively. The physicochemical properties of soil and standing water are presented in Table 1 .

The denitrification enzyme activity (DEA) was measured in accordance with the acetylene (C₂H₂) inhibition method modified from Barton et al. (2000). Briefly, three replicates (10 g) of each fresh soil sample were added to 120 mL glass flasks. Each was amended with 20 mL of solution containing glucose (300 μg C⁻¹ soil) and potassium nitrate (50 μg N g⁻¹ soil). Each flask was sealed with rubber septa and an aluminum crimp cap, and the head space was evacuated and purged with helium gas for 1 min thereafter. Then, 10% of the headspace was replaced with C₂H₂ (12 mL) through injection. Finally, the replicates were incubated in the dark on a shaker (180 rpm, 25 °C) for 30, 60, 90, and 120 min. Approximately 5 mL of gas samples was transferred into pre-evacuated glass vials and determined using a gas chromatograph (GC-8A, Shimadzu, Japan).

Genomic DNA was extracted from the triplicate subsamples (0.33 g fresh soil) with Power Soil DNA Isolation Kits (QIAGEN, USA). The quantity and quality of the extracted DNA were measured with a NanoDrop spectrophotometer (Thermo Fisher Scientific, Wilmington, USA). The extracted DNA was divided into two parts, with one fraction for real-time PCR assays and the other fraction for high-throughput sequencing.

The denitrification-associated nosZ gene was amplified using the primer pairs nosZF (5′-CGCRACGGCAASAAGGTSMSSGT-3′) and nosZ (5′-CAKRTGCAKSGCRTGGCAGAA-3′). The reaction mixtures (10 μL) consisted of 5 μL of 2× SYBR green mix II (TaKaRa Biotechnology Co. Ltd., Dalian, China), 0.2 μL of 50× Rox Reference Dye (TaKaRa Biotechnology Co. Ltd., Dalian, China), 0.2 μL (10 μM) of forward and reverse primers, 5 ng of DNA template, and deionized water. The amplification systems consisted of 5 μL of SYBR Master Mix, 0.2 μL (10 μM) of forward and reverse primers, 1 μL of DNA template, and 3.6 μL of double-distilled water (ddH₂O). qPCR assay was performed at 95°C for 5 min (denaturation), followed by 40 cycles at 95°C for 30 s, 60°C for 30 s, and 72°C for 60 s (Henry et al., 2006). The qPCR amplification efficiencies and R² value of the triplicates were 91.4%–95.6% and > 0.99, respectively.

High-throughput sequencing of the nosZ gene was performed using the same primers (nosZF/nosZR) as those for qPCR. The conditions of PCR amplification were as follows: denaturation at 95°C for 3 min; followed by 35 cycles at 94°C for 30 s, 52°C for 45 s, and 72°C for 20 s; and a final elongation at 72°C for 5 min (Delorme et al., 2003). The PCR products were then purified, quantified, and sent for paired-end sequencing (2 × 300 bp) on an Illumina MiSeq platform (Illumina, San Diego, CA, USA).

Raw sequence data were assembled using FLASH2 (Magoč and Salzberg, 2011) and quality filtered using VSEARCH v2.15 (Rognes et al., 2016) and unoise3 in usearch (Edgar, 2016). In total, 769136 high-quality sequences were obtained from 16 samples. The high-quality sequences were clustered into operational taxonomic units (OTUs) based on a 97% nucleotide similarity cutoff (Edgar, 2010), and 2691 OTUs were finally generated. The taxonomic annotations of the OTUs were in accordance with the RDP Functional Gene Repository (FunGene, http://fungene.cme.msu.edu/) using a confidence threshold of 80%. The sequence data have been deposited in the NCBI Sequence Read Archive under the accession number PRJNA668401.

All data were tested for homogeneity of variance with Levene’s tests before analysis. One-way analysis of variance (ANOVA), permutation analysis of variance (PERMANOVA) tests, and the Kruskal-Wallis H-test were performed with SPSS 22.0, R 3.2.1 (http://www.rproject.org/), and STAMP, respectively, to test the significant differences in specific variables among treatment groups. Pearson correlation analysis was used to reveal the relationship between the activity and abundance of denitrifying bacteria using SPSS (version 20.0). Principal coordinate analysis (PCoA, vegan in R) was conducted to visualize assemblage conditions of the denitrifying bacterial community. Random forest analysis (RANDOMFOREST package in R) was used to evaluate the importance of environmental predictors for denitrifying bacteria. The co-occurrence networks were constructed based on the Molecular Ecological Network Analyses Pipeline (MENA) and visualized with Gephi 0.9.2 to detect the interactions among denitrifying bacteria. Structural equation modeling (SEM) was performed to investigate the direct and indirect effects of parameters on DEAs using Amos 21.

The physiochemical characteristics in the paddy field differed among distinct straw retention treatment groups ( Table 1 ). Inorganic N, especially NH₄ ⁺, comprised the largest proportion of N in the standing water. Compared to those with NS treatment, concentrations of NH₄ ⁺ and NO₃ ⁻ in the standing water significantly decreased, whereas those of DON and DOC increased after straw retention (P < 0.05). The conditions in the soil were much different. First, organic N (i.e., SON) took the dominant portion of soil N, followed by NH₄ ⁺, NO₃ ⁻, and NO₂ ^-. More importantly, compared to those with NS treatment, soil pH and soil NO₃ ⁻ and NO₂ ⁻ contents decreased, whereas those of NH₄ ⁺, SON, SOC, and DOC increased, after the 7-year implementation of straw retention (P < 0.05). Furthermore, whether in the standing water or soil, treatment groups with rice straw amendment (RS or WRS) showed more analogous trends. These results revealed that both in the standing water and soil of the paddy field, the NO₃ ⁻ value was reduced, and the organic C and N values were enhanced by the practice of straw retention, and particularly by rice straw amendment.

DEA was used to characterize the potential soil denitrification activity. In the current experimental field, soil DEA was observed to vary from 14.48 to 20.55 mg N₂O-N kg⁻¹ h⁻¹ ( Figure 1 ). Although the DEA value increased as an effect of straw retention, only treatments with rice straw amendment (RS and WRS) resulted in significant differences, of which the increments were estimated to be 41.93% and 45.80%, respectively. These results indicate that the employment of rice straw retention might amplify denitrifying N loss in the rice–wheat rotation system.

Analogous with DEA results, the abundance of the denitrifier-nosZ gene, which ranged from 1.52 to 2.27 × 10⁷ copies g⁻¹ dry soil, was enhanced with straw retention treatments. However, relative to that with NS, only the WRS treatment group showed significantly higher levels of nosZ gene abundance, which also did not correlate with the soil DEA (P > 0.05). The investigation into different phylogroups, but not the copy numbers of denitrifiers, thus appears to be particularly important to account for soil DEA changes as affected by crop residues.

The characteristics of nosZ-denitrifying communities showed clear variations in response to different straw retention patterns. In the current study, 2691 clustered OTUs were phylogenetically grouped as 69.9% bacteria and 30.1% archaea ( Figure 2A ), which were predominated by Planctomycetes (19.0%–22.3%), Euryarchaeota (15.1%–22.3%), and Proteobacteria (10.8%–12.7%) at the phylum level ( Figure 2B ). The ANOVA and STAMP results demonstrated that under treatments with rice straw retention (RS and WRS treatments), whether at the family or genera level, nosZ denitrifiers affiliated with the phyla Euryarchaeota (Haloferacaceae, Archaeoglobaceae, Halonotius) and Bacteroidetes (Cyclobacteriaceae, Lunatimonas) increased in relative abundance (P < 0.05), whereas those affiliated with the phylum Planctomycetes (Phycisphaeraceae, Planctomicrobium, Telmatocola) decreased (P < 0.05) ( Figures 3A, B ; Supplementary Table S3 ). PCoA and PERMANOVA results further identified the community distribution characteristics of nosZ denitrifiers, revealing that RS and WRS groups were significantly differentiated relative to the NS and WS groups ( Figure 3C ).

These findings suggested that treatments with and without rice straw resulted in pairwise similarity concerning both the community composition and structure of nosZ denitrifiers. Correspondingly, R mode and non-rice straw (non-R) mode co-occurrence networks were separately constructed to identify the key phylogroups and interactions among the nosZ denitrifiers ( Figure 4 ). Generally, non-R and R mode networks consisted of 202 and 262 nodes and 440 and 606 edges, respectively, the latter possessing a higher ratio of positive associations (88.0% > 82.1%), as well as higher average degree and clustering coefficient values, as shown from the topological features ( Figure 4A ; Table 2 ). Correspondingly, the Zi-Pi plot showed four (two module hubs and two connectors) and eight (three module hubs and five connectors) hub nodes (i.e., core OTUs), which are representative of keystone taxa of denitrifiers, within the non-R and R networks, respectively ( Figure 4B ; Supplementary Table S3 ). The keystone taxa include OTUs prominently from the phylum Planctomycetes (OTU2, OTU195) in the W network, whereas OTUs were prominently from the phyla Euryarchaeota (OTU585, OTU93, OTU414) and Bacteroidetes (OTU100, OTU90) in the R+S network. These findings revealed that the nosZ denitrifiers in the R mode were more positively interconnected and recruited more and specific keystone taxa, as compared with those in non-R mode.

Random forest analyses and Pearson correlation results demonstrated that the nosZ denitrifiers and associated keystone taxa were predominantly adjusted by soil NH₄ ⁺, SOC, pH, DOC/NO₃ ⁻, and DOC (in order of significance) ( Figure 4C ). On this basis, the direct and indirect effects of environmental factors on soil DEA, under the R and non-R modes, respectively, were assessed with SEM ( Figure 5 ; Table 3 ). Generally, more than 90% of soil DEA variance was explained by these two models. Soil DOC/NO₃ ⁻ (r_non-R = 0.85; r_R = 0.03), NH₄ ⁺ (r_non-R = −0.32; r_R = −0.61), and pH (r_non-R = −0.04; r_R = −0.11) were all key factors in both models, and these directly and indirectly regulated the soil DEA by modulating biotic factors including nosZ bacterial diversity (r_non-R = 0.05; r_R = 0.08) and keystone taxa (r_non-R = 0; r_R = 0.83). Interestingly, soil DEA was stimulated by a higher level of soil DOC/NO₃ ⁻, particularly in the non-R model, whereas it was the nosZ keystone taxa (r = 0.83, P < 0.001) that generated the largest positive effect on soil DEA exclusively in the R model. This underlines the fact that nosZ keystone taxa might play an extremely important role in stimulating soil DEA under rice straw retention conditions.

The availability of N is a key factor affecting crop development and the effect of the environment in an agricultural ecosystem (Ju et al., 2009; Zhang et al., 2015). Both in standing water and soil in the present study, the NO₃ ⁻ value was reduced with straw retention treatments (P < 0.05) compared to that with NS treatment. This is beneficial for N loss load reductions, particularly for areas at a high risk of runoff, and was consistent with the results of most studies focused on flooded paddy fields (Chen et al., 2023; Bhattacharyya et al., 2012; Leon and Kohyama, 2017). Distinctively, the NH₄ ⁺ content in the soil significantly increased after long-term straw retention, particularly for treatments with rice straw amendment (RS and WRS treatments). This was primarily attributed to the characteristics of inorganic N release and microbial immobilization from crop residue decomposition (Kahlon et al., 2013; Xu et al., 2024). More importantly, the available N level, which could be enhanced by stimulation with organic N mineralization following crop residue input, was amplified more with residues having a relatively lower C/N compared to that with residues with a higher C/N (Chen et al., 2021b; Tatti et al., 2017), respectively, corresponding to the rice straw (C/N = 62/1) and wheat straw (C/N = 110/1) in the current study. Nevertheless, the low levels of NO₃ ⁻ with straw retention treatments might imply higher activities of NO₃ ⁻ microbial reduction processes (such as denitrification, anammox, or dissimilatory nitrate reduction to ammonium) rather than its accumulation (Liu et al., 2020; Wang et al., 2021, 2024; Zhang et al., 2021c). Moreover, it is understandable that organic C and N, either in soil or standing water, were enhanced after long-term straw retention (P < 0.05), given the periodic supply of exogenous organic material in conjunction with the amplified microbial activity, which would support the accumulation of organic nutrients (Malhi et al., 2011; Wang et al., 2018; Xu et al., 2024).

The responses of the nitrous oxide reductase gene nosZ to fertilization in soil ecosystems have attracted considerable attention (Duan et al., 2018; Li et al., 2022; Philippot et al., 2011; Qin et al., 2021; Tao et al., 2022). In paddy fields, compositions of the dominant nosZ-denitrifying community were found to vary widely against divergent backgrounds (Cucu et al., 2017; Wang et al., 2020). In this experimental field, the nosZ denitrifiers were predominated by Planctomycetes (19.0%–22.3%), Euryarchaeota (15.1%–22.3%), and Proteobacteria (10.8%–12.7%). Most phylogroups of the latter two phyla derived from the nosZ I clade type, which can perform the complete denitrification process (Frostegård et al., 2022; Lin et al., 2023; Zhang et al., 2021a). In contrast, Planctomycetes-affiliated organisms were recently found to belong to the nosZ II clade type based on metagenomics-based analysis (Zhuang et al., 2020). The higher positive association ratio (88%) in the R mode network highlights stronger cooperative interactions among nosZ denitrifiers under rice straw retention, suggesting a more stable and synergistic community structure that facilitates nitrogen cycling (Tao et al., 2022; Zheng et al., 2019). The increased clustering coefficient further indicates the presence of tightly connected microbial clusters, potentially acting as hotspots for efficient nitrogen transformations. These findings underline the diverse ecological roles of nosZ denitrifiers in response to long-term straw retention and the importance of microbial interactions in driving soil nitrogen processes (Hallin et al., 2018; Wei et al., 2015).

Treatments with (R mode) and without (non-R mode) rice straw retention resulted in significant differences, in terms of either the community composition, structure, or keystone taxa of nosZ denitrifiers. Generally, the R mode resulted in the recruitment of more groups from Bacteroidetes and Euryarchaeota phyla as dominant bacteria or keystone taxa. For example, the genera Lunatimonas (phyla Bacteroidetes) and Halonotius (phyla Euryarchaeota), which originate from sea and lake sediments, are typical NO₃ ⁻-reducing salinophilic bacteria and archaea (Durán-Viseras et al., 2023; Song et al., 2019; Vavourakis et al., 2016). This indicates that the recruitment of these keystone taxa under rice straw retention likely contributes to the observed improvement in denitrification efficiency, aligning with findings that enriched organic carbon conditions stimulate the activity of keystone denitrifiers (Bano et al., 2024; Chen et al., 2025). In contrast, the predominant taxa Planctomicrobium and Telmatocola, which belong to the Planctomycetes phylum in non-R mode, include parthenogenetic anaerobic denitrifying bacteria (Cao et al., 2021) and serve as the dominant bacteria in the anammox process (Chen et al., 2021a; Zheng et al., 2019). This further implies that the specific community of nosZ denitrifiers in the R mode might thereby increase the competitiveness of the denitrification process with respect to nitrate–nitrogen allotropic reduction by increasing the abundance of uncoupled Lunatimonas and Halonotius colonies and their positive interaction ratios, which is partially supported by findings from the co-occurrence networks.

In the present paddy field, treatments with rice straw retention (RS and WRS treatments) rather than wheat straw amendment only (WS treatment) significantly increased the soil DEA (by ~ 41.93–45.80%). This corroborates 3-year observation data showing that the application of rice straw results in significant higher seasonal N₂O emissions than wheat straw only, under the equivalent input of inorganic fertilizers (Wang et al., 2019). However, neither differences in nosZ gene abundance among distinct straw retention modes nor correlation coefficients for DEA and nosZ gene abundance reach significance. This represents a prevailing controversy based on many studies (Attard et al., 2011; Kou et al., 2019; Kong et al., 2023; Yin et al., 2015), and perhaps in the future, the relationship between gene expression abundance and reactivity can be determined at the mRNA level.

Without considering the abundance of the nosZ gene, the associated microbial community characteristics, as well as the biotic environmental factors, were found to regulate the soil DEA in the experimental paddy field. Specifically, soil DOC and nosZ denitrifier diversity produced a positive effect on the soil DEA, and similar phenomena have been found in many ecosystems (Gao et al., 2019; Jiang et al., 2020; Surey et al., 2020; Yeerken et al., 2024). DOC provides electron donors for denitrification processes and improves microbial competition for soil NO₃ ⁻-N (Bai et al., 2015; Tao et al., 2022). In contrast, the soil pH, which ranged from 6.63 to 7.10, was partially deviated from the optimum neutral to slightly alkaline range (pH 7–8) (Liu et al., 2010) and had direct negative effects on soil DEA in this study. An interesting finding from this long-term field experiment is that it was rice rather than wheat straw amendment that significantly stimulated soil DEA through the contributions from the keystone taxa of denitrifiers—specifically, a few nitrite-reducing salinophilic bacteria and archaea. This implied that long-term rice straw amendment offers favorable conditions of electron or nitrite competition for these denitrifiers (Wang et al., 2024; Zhang et al., 2021b). The significant role of a few core phylogroups of denitrifiers has been increasingly observed by researchers (Qin et al., 2021; Tao et al., 2022). However, more evidence in terms of microbial cultivation is warranted to directly validate the underlying mechanisms.