Artificial ponds as hotspots of nitrogen removal in agricultural watershed

Small waters, like ponds, are the most abundant freshwater environments, and are increasingly recognized for their function in ecosystem service delivery. In agricultural watershed, artificial ponds play an essential role in reducing nitrogen pollution. However, until now artificial ponds remain the least investigated part of water environments. The importance of microbial activities has seldom been discussed, which makes the microbial pathways and processes rates in nitrogen removal poorly understood. To illustrate the role of artificial ponds in microbial nitrogen removal in agricultural watersheds, 21 pond sediments and 11 soils are collected in an agricultural watershed of China. Results show that surface sediments in ponds carry significantly higher dissolved inorganic nitrogen (9.1–21.9 mg/kg) and total organic matter (64.8–113.0 g/kg) compared to the surrounding agricultural soils. High rates of microbial nitrogen removal in ponds (12.4–25.5 nmol N g−1 h−1) are observed, which are 2–9 times higher than those in dryland soils. In pond sediments, denitrification dominates (> 90% N-loss) the microbial nitrogen removal process with only a minor contribution of anaerobic ammonium oxidation. A high potential of N2O production (up to 9.4 nmol N g−1 h−1) occurs in ponds along with the rapid nitrogen removal. For denitrifier genes, nir gene are always more abundant than nosZ gene. Additionally, the nirS gene is more abundant under flooded conditions, while nirK gene prefers higher dissolved oxygen and NO3− in drylands. These findings highlight the ecosystem function of ponds in agricultural watersheds, and provide new ideas on pollution control and global nitrogen cycling.


3
Vol:. (1234567890) 2005). The number of ponds is enormous globally, with an estimated number of 277 million ponds less than 1 ha in size (Downing et al. 2008). These small water bodies account for more than 90% of the global 304 million standing waterbodies, or 30% of global standing water by surface area (Downing et al. 2006). Ponds provide a number of vital ecosystem services including hydrological regulation, conservation of biodiversity and pollution mitigation (Brazier et al. 2021;Chen et al. 2019;Mushet et al. 2020). In nutrient cycling, the burial rates for organic carbon in ponds were 20-30 times higher than that reported for other habitats on a global scale (Downing et al. 2008). Moreover, small ponds (< 1000 m 2 ) make up 8.6% of the global lakes and ponds area, but contribute 15.1% of CO 2 and 40.6% of CH 4 emissions (Holgerson & Raymond 2016).
Many regions are known for large numbers of small waterbodies, including the vernal pool systems in North America, Mediterranean temporary ponds in southern Europe and North Africa, ponds of arid and desert areas in South Africa and Australia (Boix et al. 2016;Dahl 2014;Hobbie 1980;Wissinger et al. 2016). The agricultural activities in Asian countries particularly benefit from pond construction (Chen et al. 2017), especially for China with approximately 42% of the population living in rural areas . Due to the monsoon climate, the distribution of water resources in China is extremely imbalanced spatially and temporally (Wang & Zhang 2011), and the construction of ponds is a common mean to reserve water resources for agricultural irrigation and rural life (Chen et al. 2017). The history of ponds construction dates back to 3,000 years ago (Yin et al. 1993), and the number of ponds (< 10,000 m 2 ) has reached 4 million in China to date (Lü et al. 2021).
The construction of ponds in agricultural watershed altered the landscape as well as the ecological processes. For example, artificial ponds significantly mitigated the nutrients output by intercepting and degrading the diffuse pollutants from agricultural fields and rural life sources (Capps et al. 2014). It had been reported that the traditional multi-pond systems retained about 98% of total nitrogen and phosphorus output carried by rainfall runoff in an agricultural watershed in China (Yan et al. 1998). In the United States of America, 64% of nitrate (NO 3 − ) and 36% of total nitrogen output was abated through farm ponds in an agricultural watershed (Brunet et al. 2021). On a global scale, approximate 25% nitrogen removal of watersheds occurred in ponds and such small waterbodies (Harrison et al. 2009).
To dates, the removal of nitrogen by artificial ponds was mostly attributed to their interception of pollutants, plant uptake and N burial in sediment (Lü et al. 2019;Verhoeven et al. 2006;Xue et al. 2020;Youn Chi & Pandit 2012). However, microbial nitrogen transformation was seldom discussed in pond ecosystems, although it is essential for nitrogen removal. In nitrogen cycling, denitrification is generally considered as the dominant process in the freshwater ecosystems (Hernandez & Mitsch 2007;Seitzinger 1988;Wang et al. 2019a). Denitrification processes reduce NO 3 − to NO 2 − , NO, N 2 O, and eventually to N 2 , in a stepwise manner, under anaerobic conditions (Seitzinger et al. 2006). N 2 O and N 2 are usually the two main end products in microbial nitrogen removal. Nitrite reduction catalyzed by nitrite reductase (NIR, encoded by the nirS or nirK genes) followed by NO reduction by nitric oxide reductase (NOR, encoded by nor genes) is the main source of N 2 O production , and N 2 O can be reduced to N 2 via N 2 O reductase (encoded by the nosZ gene) (Hallin et al. 2018). The release of N 2 O which is an intermediate of denitrification is regulated by functional genes of both nitrite reductase and N 2 O reductase.
Anammox (anaerobic ammonium oxidation) is an alternative nitrogen removal pathway to denitrification, in which NH 4 + is oxidized by NO 2 − by autotrophic anammox bacteria yielding N 2 as end product (Jetten et al. 2003;Wang et al. 2012). Unlike denitrification, the anammox process is not restricted by the organic matter content and directly oxidized NH 4 + to N 2 without emission of N 2 O Kartal et al. 2011;Strous et al. 2006). Identification of the reaction rates, relative contribution, and microbial communities of the two pathways is crucial to understand the end products, environmental drivers, and processes dynamics in the nitrogen removal of artificial ponds.
In this study, a typical mountainous agricultural watershed in southwest China was selected. The land of the agricultural watershed was mainly composed of drylands, paddy fields and artificial ponds. We collected sediment samples from the artificial ponds of different uses and soil samples from dryland and paddy fields. By investigating the nutrient contents, microbial activities, and microbial community composition, we aim to show the roles of artificial pond (i) as nutrient reserve, (ii) in nitrogen removal of the agricultural watershed, and (iii) relative roles of denitrification and anammox.

Study area and sample collection
The study area, a mountainous agricultural watershed characterized by abundant and scattered artificial ponds, is located in Liuyin Town, Chongqing, southwest China (29° 56′ 56"-29° 57′ 43" N, 106° 37′ 12"-106° 38′ 13" E) (Fig. 1). The study area was approximately 13.60 km 2 , and the altitude was from 280 to 496 m. It was roughly estimated that 300 artificial ponds (< 10,000 m 2 , ~ 0.8 km 2 in total) existed in this area for irrigation and aquaculture. In this study, sediment samples were collected from seven ponds, including four irrigation ponds (P1, P2, P5 and P6) and three aquaculture ponds (P3, P4 and P7) (Table S1) in July 2021. At each pond three parallel sediment columns (30 cm depth) and the overlying water were collected. The sediment columns were segmented into three Sects. (0-10 cm, 10-20 cm, and 20-30 cm). Three parallel surface soil samples (0-10 cm) in dryland or paddy fields around the ponds were collected. The crop types in dryland include corn, cowpea fields and orchard gardens. In total, 21 pond sediments and 11 agricultural soils were collected in the study region.
Field samples were stored in sterile plastic bags and transported to the laboratory in a cooler box (4 °C) for subsequent analysis. One subsample was immediately incubated to determine microbial nitrogen removal activity, a second subsample was used for physicochemical analyses, and a small portion was stored at -80 °C for molecular analysis. All samples were analyzed in triplicate, and the values were averaged to represent site conditions. Physicochemical analysis Sediment/soil ammonium (NH 4 + ), nitrite (NO 2 − ), and nitrate (NO 3 − ) were extracted from 5 g of fresh sediment/soil with 25 ml of 2 M KCl (1:5 wt./vol). The supernatant was filtered through a 0.22 μm membrane filter and the compounds were determined via a spectrophotometric detection assay (Wu et al. 2016). Moisture content was measured by oven-drying at 105 °C until a constant weight was achieved. The pH was determined in 1:2.5 sediment/water (wt./vol) suspensions after shaking and centrifugation, with a Mettler Toledo pH analyzer (S220, Switzerland). Total organic matter (TOM) was measured as loss on ignition at 550 °C (LOI 550) using a Muffle furnace. The TN and TP were determined with the potassium persulfate oxidationultraviolet spectrometry method (Apha 1998), using a UV spectrophotometer (UVmini-1240, Japan).

Measurements of potential denitrification and anammox rates
The denitrification and anammox rates of sediment/soil samples were measured using the slurry incubation and isotope pairing technique (Risgaard-Petersen et al. 2004). Fresh sediments/soils were mixed with water in the ratio of 1:7 (sediment: water) and flushed with ultrahigh purity He for 30 min to make anaerobic sediment slurries. These slurries were darkly pre-incubated at in-situ temperature for 36-48 h to remove background NO x − (NO 3 − and NO 2 − ) and DO. After pre-incubation, the slurries were transferred into 12.5 mL tubes (Exetainers, Labco, UK) via injectors, these tubes were divided into two groups. The first group were used to analyze F n (fraction of 15 NO 3 − in NO x − pool), and the second group of tubes were injected with 15 NO 3 − (99.6 atom%) solution to 100 μM final concentration. The tubes were incubated in the incubator at in-situ temperature and were stopped by adding 200 μL of 50% ZnCl 2 at 0 and 2 h from the beginning of incubation. 29 N 2 and 30 N 2 produced in the tubes were determined with a membrane inlet mass spectrometry (MIMS, HPR40, Hiden, UK), and the rates of denitrification and anammox were calculated (Thamdrup & Dalsgaard 2002). The calculation equations are as follows: where R D (nmol N g −1 h −1 ) represents the total rate of 15 NO 3 − -based denitrification, D 29 is the 29 N 2 production rate from denitrification, P 30 (nmol N g −1 h −1 ) is the total 30 N 2 production rate; F n represents the fraction of 15 N in total NO 3 − . The potential rates of anammox were estimated by the following equation Xiao et al. 2018): where R A (A 29 ) and P 29 (nmol N g −1 h −1 ) represent the potential rate of 15 NO 3 − -based anammox (or 29 N 2 production rate from anammox) and total 29 N 2 production rate, respectively.
Determination of potential N 2 O production rate N 2 O production rates were measured with headspace equilibrium gas chromatography using the samples prepared as describes in 2.3 of Methods and Materials section (Hou et al. 2015). Specifically, after inactivation and settling, the 12.5 mL tubes were injected with 5 mL of ultrahigh-purity He gas to replace the water phase and create headspace. Then, the tubes were shaken violently for 1 h to make gas-liquid equilibrium. The concentration of N 2 O in the headspace gas was measured with a gas chromatograph (GC-2014C, Shimadzu, Japan) which was equipped with electron capture detector (ECD). N 2 O concentrations in the headspace after equilibrium were calculated according to equation: where C G (nmol L −1 ) is the concentration of N 2 O in the headspace after equilibrium, C g (ppb: nmol mol −1 ) is the volumetric concentration of N 2 O in the headspace, R (0.082057 L atm mol −1 K −1 ) is the ideal gas constant, T (K) is the temperature of the water samples after equilibrium, and P (hPa) is the pressure of laboratory.
The dissolved N 2 O concentrations were determined according to equation: (1) where C L (nmol L −1 ) is the concentration of dissolved N 2 O in the incubated water, R (0.082057 L atm mol −1 K −1 ) is the ideal gas constant, α is the ratio of gas to liquid after replacement, and K 0 (mol L −1 atm −1 ) is the equilibrium constant which is calculated from Weiss formula (Weiss & Price 1980), as follow: The accumulation of N 2 O per unit of incubation time is the rate of N 2 O production which was calculated according to equation: where D N 2 O (nmol N g −1 h −1 ) is the rate of N 2 O production, T 0 and T 2 represent 0 and 2 incubation time respectively, C L0 and C L2 represent the concentration of N 2 O at T 0 and T 2 respectively. DNA extraction, sequencing, and data processing Total genomic DNA was extracted from sediment/soil samples (approximately 0.5 g) using the FastDNA SPIN Kit for Soil (MP Biomedicals, USA) following the manufacturer's instruction. The concentration of extracted DNA was measured with a NanoDrop Lite (Thermo Fisher Scientific, Wilmington, DE, USA), and the DNA quality examined by 1% (wt./vol) agarose gel electrophoresis.
The nirS gene (performing NO 2 − reduction) was used to study the denitrification community, and was amplified by PCR using primers cd3aF and R3cd (Yergeau et al. 2007). PCR amplification of Anammox-specific 16S rRNA genes was based on primers A438f and A684r (Han & Gu 2013). More details about the conditions of PCR amplification are presented in Table S2. Prior to high-throughput sequencing, the PCR products were purified using the MiniBEST Agarose Gel DNA Extraction Kit (TaKaRa Bio, Japan). Subsequently, purified amplicons were pooled in equimolar and paired-end (PE) sequenced (2 × 300) on an Illumina MiSeq PE300 platform. The raw sequences were merged and quality filtered in Quantitative Insights in Microbial Ecology (QIIME) (Caporaso et al. 2010) and Mothur (Schloss et al. 2009). OTUs with identity thresholds (93% for nirS and 97% for anammox 16S rRNA) were defined using Usearch (v. 7.0 http:// drive5. com/ uparse/). Rare OTUs with less than 0.01% of the total sequences were removed. To avoid biases arising from sequencing depth and to make samples comparable, sequences were rarified to a uniform sequencing depth based on the sample with the lowest sequences. The diversity indices (Shannon, Simpson, Ace and Chao1) and rarefaction curves were calculated also in Mothur referring to previous studies Ye et al. 2021). The raw sequences of nirS and Anammox-specific 16S rRNA genes used in this study were deposited in the Sequence Read Archive (SRA, https:// submit. ncbi. nlm. nih. gov/ subs/ sra/) of NCBI under the accession numbers PRJNA780407 and PRJNA780073.

Quantitative PCR analysis
The abundance of bacterial 16S rRNA genes, -nir gene (nirS and nirK), nosZ gene (nosZ I and nosZ II) and anammox-specific 16S rRNA genes in sediments/soil samples were quantified by a LightCycler ® R480 II Real-Time PCR (Roche, Switzerland). Each sample was analyzed in triplicate. The standard curves used for calculation were achieved with plasmid DNA with known concentrations and copy numbers. qPCR results with high amplification efficiency (90-110%) and correlation coefficient values of the standard curve (r 2 > 0.97) were integrated into the analysis. The specificity of PCR amplifications was defined by melting curve analysis and gel electrophoresis. The primers, reaction systems, and programs are shown in Table S2.

Statistical analysis
To test significant differences between samples, oneway analysis of variance (ANOVA) based on Tukey's post hoc analysis was conducted with the 'multcomp' package in v. R 4.1.0. The plots were conducted with Prism 8 software (version 8.0.2). Principal coordinates analysis (PCoA) ordination and non-metric multidimensional scaling (NMDS) was used to reveal differences in denitrification and anammox bacteria community structures based on Bray-Curtis dissimilarities. Significant differences among the different samples were tested by an analysis of similarities test (ANOSIM). To identify the environmental factors likely to affect the composition of denitrification and anammox bacteria communities, Canonical Correlation Analysis (CCA) was used in Canoco 5 software. Partial least squares pathway modelling (PLS-PM) was conducted to infer the effects of sediment/soil conditions (pH and moisture), nutrients (TN, TP, and TOM) and DIN (NH 4 + , NO 3 − , and NO 2 − ) on denitrifier abundance (nirS, nirK, nosZ I and nosZ II), denitrifier community (composed with genera of nirS-type denitrifier which contain all OTUs) and potential nitrogen removal in R with the 'plspm' package (Sanchez and Trinchera 2012). 1000 bootstraps were performed on model pathways to evaluate pathway coefficients and coefficients of determination (R 2 ).
The highest N 2 O production rates were observed in paddy soils (9.4 ± 3.0 nmol N g −1 h −1 , ANOVA, P = 0.09) (Fig. 3e). The ratios of N 2 O/ (N 2 O + N 2 ), which reflected the potential of N 2 O release during nitrogen removal, showed high values up to 54.4% and 47.6% in dryland and paddy soils, respectively. In the pond sediments, samples taken at different depths showed little variances (ANOVA, P = 0.22) (Fig. 3f).

Diversity and composition of denitrification and anammox bacteria
The diversity of the key players in nitrogen removal was determined by amplicon sequencing. For nirS, Fig. 3 Potential rates of sediment (sediments are surface to bottom from left to right) and soil samples in this study. a Potential denitrification rate, b potential anammox rate, c the contribution of anammox to nitrogen removal (ra%), d potential N 2 production rate, e potential N 2 O production rate, f the ratio of (N 2 O/ N 2 O + N 2 ). Different lowercase letters above the boxes indicate significant differences between different groups based on one-way ANOVA with Tukey's test (P < 0.05). Boxwhisker element depicts the first quartile, median, and third quartile in the form of a colored box; minimum and maximum scores are depicted as whiskers after filtering the 330,501 raw sequences, 165,568 high-quality sequences were obtained which resulted in 1443 OTUs. Shannon, Simpson, Chao1 and ACE indexes were calculated to determine the alpha diversity (Table S3 and Fig S2a). Paddy soils showed the highest Shannon diversity (4.81-5.46) and Chao richness (471.12-760.17) than those in pond sediments (2.62-4.38 and 296-587, respectively) and dryland soils (3.18-4.56 and 265-476, respectively).
The most abundant 50 OTUs, covering 62.5% of the nirS sequences, were all affiliated to Proteobacteria. At the genus level, significant differences were discovered in pond sediments, paddy, and dryland soils. Most sequences identified in pond sediments were affiliated to Azoarcus (59.45 ± 5.14%), and followed by Steroidobacter and Dechloromonas. Rhodanobacter was the dominant genus in dryland soils (58.1 ± 16.2%), and Steroidobacter had the highest relative abundance in paddy soils (44.1 ± 2.1%) ( Fig. 5a and S3a). The PCoA showed that the first two axes explained 50.4% of nirS community (Fig. 5c).
The nirS community showed a significant separation between samples of sediment, dryland soil and paddy soil (ANOSIM, P = 0.01), and low variance was discovered in different layers of pond sediments.
According to the CCA analysis, pH, NH 4 + and DIN were the main factors significantly related to nirS community structure (P < 0.01, Fig S4a).
The dominant OTUs (30 OTUs, covering 94.2% of the sequences) were all affiliated to Planctomycetes. Only 9 OTUs were affiliated to anammox genus Ca. Brocadia, and other sequences were divided into two unknown clusters (cluster 1 and 2) (Fig. S3b). Cluster 1 was close to the uncultured bacterium from freshwater wetland and lake sediments, and cluster 2 was more similar with samples from South China Sea sediment and paddy soil. In pond sediments, less sequences were affiliated with anammox bacteria Fig. 4 Abundances of nitrogen removal related genes in this study including a nirS (reduce NO 2 − to NO), b nirK (reduce NO 2 − to NO), c nosZ I (reduce N 2 O to N 2 ), d nosZ II (reduce N 2 O to N 2 ), e the ratio of nir/nosZ, f bacteria 16S rRNA, g anammox 16S rRNA. Different lowercase letters above the boxes indicate significant differences between different groups based on one-way ANOVA with Tukey's test (P < 0.05). Boxwhisker element depicts the first quartile, median, and third quartile in the form of a colored box; minimum and maximum scores are depicted as whiskers (12.4 ± 1.9%), while sequences in dryland and paddy soils were dominated by genus Ca. Brocadia (82.7% and 67.2%, respectively) (Fig. 5b). The distribution of the bacteria for all samples were further analyzed by NMDS (Fig. 5d, explained by 80.81% of the variances). The community was significantly separated into two groups (pond sediments versus paddy/dryland soils, ANOSIM, P = 0.002). The community of sediment in different depths showed little variation. The CCA analysis indicated that TOM and DIN significantly affected the bacterial community structure (P < 0.01, Fig S4b).

Drivers of nitrogen removal
PLS-PM analysis was used to access the direct and indirect effect of a single factor or the combination of interacting factors on potential nitrogen removal rates (Fig. 6). Moisture content and pH showed significantly positive effects on nutrients (TN, TP and TOM), DIN and denitrifier community (coefficients = 0.665, 0.741 and 0.546 respectively; P < 0.001). DIN had a negative effect on denitrifiers gene abundance (coefficients = − 0.552, P < 0.05). By contrast, DIN showed positive effects on denitrifier community composition (coefficients = 0.427, P < 0.001) and potential nitrogen removal rates (coefficients = 0.616, P < 0.05). Denitrifiers gene abundances also had a significantly positive effect on potential nitrogen removal rates (coefficients = 0.483, P < 0.05).

Extensive storage of DIN and TOM in ponds
Ponds are often overlooked freshwater ecosystems due to their relatively small area to the total freshwater surface of the earth (Oertli 2018). However, the ponds appear to be a nutrient pool (DIN and TOM) in the studied agricultural watershed. The dissolved inorganic nitrogen (DIN) in pond sediments (up to 21.9 mg/kg) was significantly higher than those of surronding agricultural soils (6.0-9.3 mg/kg in surface soil), and so was the TOM which was 105.0 g/kg in surface sediments versus 56.5-76.5 g/kg in agricultural soils. Taking into consideration the respective areas of ponds and agricultural fields, the ponds may contain 18.5% of the total DIN and 14.9% of the TOM, while only covering 7.1% of the total area in the targeted research region.
Addtionally, the DIN in pond sediments was at a high level in comparison with other types of waterbodies, such as freshwater rivers in America and Germany (4.75-7.80 mg/kg) (Arce et al. 2018;Kim Fig. 6 Directed graph of the Partial Least Squares Path Model (PLS-PM). Each box represents an observed variables or latent variables. Path coefficients are calculated after 1000 bootstraps. Coefficients differ significantly from 0 are indicated by *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001. The model is assessed using the Goodness of Fit statistic, a measure of the overall prediction performance (The GoF index is 0.58). Path coefficients are reflected in the width of the arrow with red indicating a positive effect and blue a negative effect Zimmer-Faust et al. 2017), eutrophic lake(0.01-8.54 mg/kg) (Cao et al. 2009) and Miyun and Jiulonghu Reservoir in China (1.5-11.55 mg/kg) Shen et al. 2017). The TOM content of ponds was also higher than those of freshwater, like Congo and Cross River (6.2-85.0 g/kg) (Mata et al. 2020;Pisani et al. 2013), eutrophic lakes in Korea and New Zealand (32-79 g/kg) (Gu et al. 2017;Hickey & Gibbs 2009) as well as the Three Gorges and the Sulejów Reservoir (13.8-61.8 g/kg) (Han et al. 2020;Trojanowska & Izydorczyk 2010). That is because the ponds were mostly located close to human settlements, and thus were apt to receive pollution discharged from daily life and farming (Aguilar et al. 2012). The high content of TOM and high ratio of NH 4 + in DIN (93.3%-98.3%) also suggested that the nutrients in ponds probably originated from human activities (Li et al. 2013), and the anoxic condition in pond sediment inhibited the nitrification process (Cerco & Cole 1993).

High rate of nitrogen removal in ponds
The nitrogen removal that occurred in pond sediment is a considerable rapid process when comparing with other agricultural units. The rates of nitrogen removal (denitrification plus anammox) in pond surface sediments (12.4-25.5 nmol N g −1 h −1 ) were 2-9 times higher than those in dryland soils (3.0-15.7 nmol N g −1 h −1 ), and were comparable with those of paddy field soils (16.1-24.5 nmol N g −1 h −1 ). The nitrogen removal rates were also higher than those in estuaries influenced by extensive human activity (3.1-7.5 nmol N g −1 h −1 ) , high-elevation rivers (0.05-1.14 nmol N g −1 h −1 ) (Zhang et al. 2021b), constructed, coastal and riparian wetlands (0.30-4.12 nmol N g −1 h −1 ) (Coban et al. 2015;Gao et al. 2016;Wang et al. 2019b), and comparable with agricultural ditches (0.39-64.40 nmol N g −1 h −1 ) (Deng et al. 2020;Speir et al. 2020). The high rates of nitrogen removal in ponds were largely caused by the nutrient enrichment which was proven to have a positive effect on denitrification, mineralization processes (Doroski et al. 2019) and the overall geochemical dynamics (Cheng & Basu 2017). It was further confirmed using PLS-PM and Pearson correlation analysis that the DIN and TOM both showed positive correlations with the nitrogen removal rates (Fig. 6 and S1).
In this study, denitrification dominated (> 90%) the nitrogen removal process in ponds, which was consistent with data from several natural habitats (tidal rivers, marine, reservoirs) (McCarthy et al. 2015;Tall et al. 2011;Zhou et al. 2019). In previous studies, the high proportion of denitrification is often associated with high level of organic matter or carbon content (Ballantine and Schneider 2009;Bruesewitz et al. 2011;Small et al. 2016), and it was ascribed to the physiology of many denitrifiers consuming organic matter in their metabolism (Baker et al. 2000;Jones 1995). It was also identified in this study that lower denitrification rates were detected in deeper sediments and dryland soils with lower organic matter content. Pearson correlation analysis further confirmed the significant correlation (r = 0.437, P < 0.05) between TOM and denitrification rates ( Fig S1). The DIN (mostly NH 4 + ) also had a positive effect (coefficients = 0.616, P < 0.05) on denitrification rates (Fig. 6).
In addition, a significant negative correlation between NO 3 − and denitrification rates was observed, which was different from previous studies (Bruland et al. 2006;Hunt et al. 2004). It was probably because the denitrification rates in dryland soils, which had higher NO 3 − concentrations, were significantly lower than those of pond sediments and paddy soils. Alternatively, higher TOM in the pond sediments promoted denitrification and consumed NO 3 − , while nitrification was inhibited under anaerobic conditions and more NH 4 + deposited in the sediment, the opposite occurred in the soils. The NO 3 − and TOM act as electron acceptor and donor respectively, to regulate the denitrification process. TOM is more decisive compared to NO 3 − in freshwater ecosystems (Burgin et al. 2010;Hill & Cardaci 2004). Hence, the higher denitrification rates in the sediments were more likely to be regulated by the high content of TOM in sediment. In general, the human nutrients input (DIN and TOM) not only turns the ponds into a nutrient pool, but also shapes the pond into a hotspot of microbial nitrogen removal.
By contrast, the anammox process only contributed 4.3 ± 3.0% of nitrogen removal in ponds, which was consistent with the levels reported in other natural ecosystems (river estuaries, inland rivers) (Dale et al. 2009;Kessler et al. 2018;Zhou et al. 2014). However, in some habitats anammox bacteria were highly abundant and responsible for up to 41% of the total N loss such as estuaries in Costa Rica , paddy fields in China (Shen et al. 2014;Zhu et al. 2011) and constructed wetlands (Zhu et al. 2011). Anammox is more active in environments where NO 3 − is readily available (Engström et al. 2005;Trimmer et al. 2003). However, anammox as an autotrophic process is usually more competitive than denitrification in low organic carbon environments (Plummer et al. 2015). Thus, in surface sediments and paddy soils, which had higher organic matter, anammox was responsible for only 5% of the nitrogen removal. By contrast, higher anammox contributions (up to 13.4% of total nitrogen removal) were detected in dryland soils with higher content of NO 3 − and lower content of TOM. On the basis of the nitrogen removal rates and sediments/soils density data (1.41-1.89 g cm −3 ), an estimated total loss by the combination of denitrification and anammox processes was 47.8 and 25.0 t N yr −1 in dryland and paddy soils, while the pond surface sediments in the study area contributed 14.0 t N yr −1 . Thus, 16.2% of the nitrogen was removed from the ponds with only 7.12% of the total area in the targeted research region. The pond is a neglected hotspot for nitrogen removal in agricultural watershed. Furthermore, numerous studies have been conducted to show the strong nitrogen removal capacity in paddy fields as in this study (Li et al. 2013;Rahman et al. 2018). Often neglected but widespread wetland-like agricultural drainage ditches play an important role in the removal of agricultural nonpoint source nitrogen pollution through denitrification (Davis et al. 2015;Roley et al. 2012). These two types of habitats act as intermediates controlling the transport of nitrogen pollution to ponds or downstream and remove some of the nitrogen pollution by themselves. Therefore, a combination of ponds, ditches and paddy fields may be able to better manage and reduce the export of nitrogen pollution out of the watershed.
Strong potentials of N 2 O production appeared along with the nitrogen removal in ponds. The N 2 O production rates from pond sediments (6.5 ± 2.2 nmol N g −1 h −1 ) were higher than many habitats like East China Sea (0-0.09 nmol N g −1 h −1 ), subtropical estuary in Southeast China and Douro River estuary (0.03-3.40 nmol N g −1 h −1 ), riparian zones in Han River basin (0-0.04 nmol N g −1 h −1 ) and constructed wetlands (0-1.60 nmol N g −1 h −1 ) (David et al. 2013;Li et al. 2021b;Lin et al. 2017;Liu et al. 2016;Teixeira et al. 2010). The N 2 O/ (N 2 O + N 2 ) percentages in pond sediments (34.2 ± 13.2%) were comparable or higher than those of a nitrogen-enriched subtropical estuaries and stormwater ponds (3.38-51.00%) (Blaszczak et al. 2018;Li et al. 2021b;Su et al. 2021). Despite the high rates of denitrification in ponds, the terminal product of nitrogen removal in pond sediment was more likely to be N 2 compared to dryland soil (N 2 O/ (N 2 O + N 2 ) up to 86.8%). Previous study demonstrated that the ratio of N 2 O/ (N 2 O + N 2 ) was negatively correlated with pH (Samad et al. 2016), matching the result of Pearson correlation analysis in our study (r = − 0.558, P < 0.001). Under lower pH (< 7.0) conditions the nosZ genes (particularly nosZ II genes) and activity of the N 2 O reductase may be restrained, which would increase N 2 O emission and the N 2 O/(N 2 O + N 2 ) ratio in dryland soils (Čuhel & Šimek 2011;Jones et al. 2014;Pan et al. 2012). Thus, more than half of the N 2 O was emitted as an intermediate product of denitrification processes in dryland soils.

Contributors to microbial nitrogen removal
The high abundance of nirS, nirK, nosZ I and nosZ II genes also supported that denitrifiers were the main contributors to nitrogen removal in ponds, in which these denitrifier gene abundances were 2-5 orders of magnitude higher than those of anammox bacteria. Despite the frequent exchange of water/soil between ponds and surrounding agricultural fields, the denitrification community in pond sediments, which was mainly composed by genus Azoarcus (59.5 ± 21.0%), was significantly different from that in the surrounding agricultural soils. By contrast, the denitrification community was more similar with those reported in other freshwater ecosystems (i.e. lake and river sediments) Kim et al. 2011;Zhang et al. 2021a). Hence, it could be concluded that environmental properties like pH, DIN and TOM rather than geographic distance were stronger determinants for the community composition of denitrifier in ponds.
The nirS and nirK genes were detected in high copy numbers in pond surface sediments ((7.2 ± 4.0) × 10 8 and (1.2 ± 0.6) × 10 8 copies g −1 , nirS and nirK respectively), being higher than those reported for marine (Lee & Francis 2017;Lindemann et al. 2016;Zheng et al. 2021) and freshwater sediments (Jin et al. 2020;Wang et al. 2019a;Zhu et al. 2018). This suggested extensive denitrification processes in ponds. It was noted that nirS genes were more abundant than nirK genes in pond sediments while nirK genes dominated over nirS gene in dryland soils. This is consistent with the properties of nirK type denitrifiers which prefer high DO and NO 3 − (Desnues et al. 2007;Knapp et al. 2009). It was reported that a higher frequency of co-occurrence of nosZ with nirS than with nirK was reported before, suggesting that nirS type denitrifiers are more likely to perform complete denitrification (Clark et al. 2012;Hallin et al. 2018). Taken this into account, our results would point to lower production of N 2 O in the ponds compared to dryland soils and this supports the rate measurements.
In this study, the abundance of nosZ II gene (10 7 -10 9 copies g −1 ) was greater than that of the nosZ I gene (10 5 -10 7 copies g −1 ), and similar results were found in other terrestrial ecosystems with the ratio of 1.5-10 (Jones et al. 2013;Juhanson et al. 2017;Pascazio et al. 2018;Su et al. 2021;Tsiknia et al. 2015). It was reported that nosZ II played a greater role in reducing N 2 O emission in terrestrial habitats (Jones et al. 2014;Xu et al. 2020). However, high nosZ II gene abundance together with high N 2 O production potential (N 2 O/ (N 2 O + N 2 )) were observed in dryland soils in this study. It was probably because the activity of N 2 O reductase was restrained by the low pH in dryland soils (Čuhel & Šimek 2011;Jones et al. 2014;Pan et al. 2012).
The nir/nosZ ratio had been widely used as indicator of N 2 O production potential observed with positive correlation (Domeignoz-Horta et al. 2015;Saarenheimo et al. 2015;Zhao et al. 2020). Nevertheless, we observed a negative correlation between the nir/nosZ and N 2 O/ (N 2 O + N 2 ) ratio in the pond sediments (r = − 0.452, P < 0.05), which coincided with previous studies (Linton et al. 2020;Mafa-Attoye et al. 2020). These results implied that the nir/nosZ ratio only is not a good indicator for N 2 O production in natural habitats.

Conclusions
In China and other Asian nations, building domestic ponds is an intriguing spontaneous familybased engineering endeavor (Ichinose et al. 2007). Artificial ponds provide a variety of services for agricultural and dwelling households. Beyond the traditional functions, our study revealed that the presence of artificial ponds plays a key role in nutrients reserve and nitrogen removal. The pond sediments had higher DIN (9.1-21.9 mg/ kg) and TOM (64.8-113.0 g/kg) comparing with those in soils of dryland and paddy fields, thus play potential role of nutrient reserve in the agricultural watershed. Besides, the pond sediments also showed higher rates of nitrogen removal (12.4-25.5 nmol N g −1 h −1 ), 2-9 times higher than those in dryland soils. It was estimated that 16.2% of the nitrogen was removal from ponds with only 7.1% of area in targeted watershed. It was worth pointing out that the artificial ponds received little additional nutrient input from outside the watershed, which was different from dryland and paddy field with massive of fertilizers input. Hence, the artificial ponds could be a sink of nitrogen pollution in such agricultural watersheds.
In a sense of ecological understanding, the spatial structure of these artificial ponds at a landscape scale fostered the ecological flows and interactions between habitats for whole ecosystem integrity. The artificial ponds together with the farmland and settlement place formed an integrated water-land system. The function of nitrogen removal of artificial ponds inspires us to consider reducing pollution output by constructing ponds along with the farmland and forming an integrated system encompassing both waters and lands. The present study proved the overlooked function in nitrogen removal of artificial ponds and also call attentions to these small and scattered water bodies for the significant roles in local pollution control and global nitrogen cycling.