An experimental cabbage (B. oleracea var. capitata) field (Supplementary Figure 1) was established in 2002 at the National Institute of Vegetable and Tea Science, NARO (36° 01’ N, 140° 06’ E; 21 m above sea level), with 18 plots (25 m²) of six treatments (Supplementary Figure 1A), arranged in a randomized complete block design (n = 3): NF, no fertilizer; PF, conventional fertilization combining chemical fertilizers and cow manure compost; CF, chemical fertilizers alone; MC-1, cow manure compost at 250 kg N ha^–1; MC-2, cow manure compost at 500 kg N ha^–1; and MC-3, cow manure compost at 750 kg N ha^–1. The soil chemical characteristics and fertilizer N used for each treatment are shown in Supplementary Table 1.

Cabbage was cultivated in winter and summer in 2009 and 2010. N₂O flux in the field experiment was measured using the static closed chamber method (Yan et al., 2001). The chamber was set on the ground to enclose a growing cabbage plant to measure N₂O flux from the plant–soil system comprising both the soil and crop surfaces. Precipitation, as well as soil temperature, water content, pH, ammonia, and nitrate were measured (Supplementary Figure 2). To measure direct N₂O flux from an unharvested leaf (Supplementary Figure 1B) in situ, unharvested leaves in the field were quickly transferred to a 17-L plastic container and closed. The amount of N₂O gas accumulated in the container was measured.

Unharvested outer cabbage leaves and soil were collected from the PF-treated plots. To imitate harvest-season field conditions, a microcosm was prepared (n = 6): two unharvested leaves were placed on 5 kg of soil in a 17 L plastic container. One leaf was placed on the other so that the upper one avoided adhesion to soil particles. The microcosms were incubated at 25°C. N₂O flux from the microcosms was periodically measured during incubation. When the microcosms showed the highest N₂O emission, the upper leaf was taken and cut into several pieces, and the N₂O flux of the subsamples was measured to identify the most active point on the leaf surface, that is, the N₂O-emitting hot spot.

N₂O flux was measured using a Shimadzu GC-2014 gas chromatograph equipped with a thermal conductivity detector, or a Shimadzu GC-14B equipped with an electron capture detector (Shimadzu Co., Kyoto, Japan). Dissolved N₂O and oxygen concentrations in liquid cultures were measured using microsensors for N₂O and O₂, respectively (Unisense, Aarhus, Denmark). The end product of denitrification was measured using the acetylene block method, and gas chromatography–mass spectrometry analysis using ¹⁵NO₃^–. Nitrate and ammonium concentrations were analyzed using the Cu–Cd reduction method and indophenol-blue method, respectively. Total organic carbon and nitrogen were measured using a TOC–V/TN analyzer (Shimadzu Co., Kyoto, Japan).

Two media were used to screen for denitrifiers: R2A medium (Difco, BD, Franklin Lakes, NJ, United States) with 1 g L^–1 potassium nitrate added (R2A-N) and a customized medium comprising cabbage extract (CE). To prepare the CE medium, cabbage leaves taken from the field were boiled in four volumes of water for 30 s to inactivate cellular enzymes and then homogenized with a blender. Plant debris in the crude extract was then removed by filtration through cotton gauze and filter paper, and the filtrate was autoclaved and stored at 4°C before use. The culture (2.5 mL) was incubated under static conditions in a 20 mL glass serum vial (Nichiden-Rika Glass, Kobe, Japan) at 25°C in the dark (n = 3). During the incubation, the cultures were not sealed, allowing air to pass naturally through the gas phase within the vial.

Bacterial cells were extracted from the N₂O-emitting hotspots using a Nycodenz density gradient method (Ikeda et al., 2009). The cell suspension was serially diluted with CE or R2A-N liquid media, and denitrification activity was measured by determining N₂O production. Denitrifiers were screened using the dilution-plate method. The isolated denitrifiers were classified via 16S rRNA gene sequencing analysis, using the bacterial universal primer set 27f and 1492r, as described elsewhere (Tago et al., 2015). N₂O reductase activity of bacterial culture was measured via the acetylene block method (Tiedje, 1994).

Total DNA was extracted from isolated denitrifiers belonging to Agrobacterium spp., using the Genomic-Tip 100/G kit (Qiagen, Valencia, CA, United States). Genome sequencing was performed using PacBio RSII single-molecule real-time (SMRT) sequencing technology (PacBio, Menlo Park, CA, United States) and an Illumina MiSeq platform (Illumina, Inc., San Diego, CA, United States). Read processing and hybrid assembly were performed as previously reported (Guo et al., 2020). Briefly, single-end reads were trimmed from the MiSeq paired-end reads using sickle v.1.33,² with default settings. The Unicycler pipeline (v.0.4.8) (Wick et al., 2017) was used for combined assembly of the PacBio and the MiSeq reads, after which genome polishing was performed using the built-in tool Pilon (v.1.24) (Walker et al., 2014). Automatic annotation and identification of rRNA and tRNA genes and CDSs were performed using the DDBJ Fast Annotation and Submission Tool (DFAST) pipeline (Tanizawa et al., 2018), based on Prokka software (Seemann, 2014). The CDSs identified as “hypothetical proteins” by DFAST were manually identified and annotated against the NCBI non-redundant database using the BLAST.

A representative denitrifier, Agrobacterium sp. strain 6Ca8, was cultured in CE medium at 25°C under static conditions (denitrifying conditions) (n = 3). As a negative control, the culture was aerated by stirring at 60 rpm to restrict cellular denitrification activity. Under the aerated conditions (the negative control), only a small amount of dissolved N₂O was detected in the culture (Supplementary Figure 3). Cells of 6Ca8 at the early- and mid-log phases (OD₆₀₀ of approximately 0.4 and 0.9, respectively) were collected, and the total RNA was extracted and purified using an RNeasy kit (Qiagen, Valencia, CA, United States). The extracted nucleic acid was treated with TURBO DNase (TURBO DNA-free kit, Thermo Fisher Scientific, Waltham, MA, United States), followed by the Ribo-Zero rRNA Removal kit (Bacteria; Illumina, Inc., San Diego, CA, United States), according to the manufacturer’s protocols. The mRNA was purified using an RNA Clean & Concentrator-5 kit (Zymo Research, Irvine, CA, United States), according to the manufacturer’s protocols. mRNA quality was determined using a Bioanalyzer and a Prokaryote Total RNA Pico Chip (Agilent Technologies, Foster City, CA, United States).

For library preparation, the NEBNext Ultra RNA Library Prep Kit for Illumina and NEB Multiplex Oligos for Illumina (New England Biolabs, Ipswich, MA, United States) were used for fragmentation, adapter ligation, cDNA synthesis, and PCR amplification, according to the manufacturer’s protocols. The library was loaded on an agarose gel containing Synergel (Diversified Biotech Inc., Dedham, MA, United States; Synergel:agarose, 5:3), and the appropriate fragment size (200–300 bp) was purified using a gel extraction kit (Qiagen, Valencia, CA, United States). The final products were sequenced on an Illumina MiSeq platform (Illumina, Inc., San Diego, CA, United States). To ensure high sequence quality, the remaining sequencing adaptors and reads with a Phred quality score > 15 (or > 20, for leading and tailing sequences) and reads shorter than 80 bp were removed using Trimmomatic v.0.30, with Illumina TruSeq3 adapter sequences used for clipping (Bolger et al., 2014). The remaining paired reads were analyzed using FastQC³ for quality control, and Bowtie2 (v. 2.2.2) (Langmead and Salzberg, 2012) for mapping onto the 6Ca8 genome (DDBJ/EMBL/GenBank accession: AP026433–AP026435). After converting the output BAM files to BED files using the bamtobed function in BEDTools (v. 2.14.3) (Quinlan and Hall, 2010), gene expression levels were calculated as TPM using an in-house script (Sato et al., 2019). The data from early-log phase cells were further analyzed, because the difference in the expression of denitrification-related genes between the denitrifying and control (aerobic) conditions was larger for early-log than mid-log phase cells (Supplementary Table 2).

Cells cultured in CE medium (n = 3), at the initial and late-log phases, were filtered to remove cells and contaminants, then subjected to NMR analysis. The filtrate (140 μL) was added to 560 μL of 125 mM potassium phosphate buffer (pH or pD 7.2) in deuterium oxide (D₂O, 99.9%, Cambridge Isotope Laboratories, Andover, MA, United States) containing 1.25 mM of 2,2-dimethyl-2-silapentane-5-sulfonate sodium salt (DSS, Sigma–Aldrich, St. Louis, MO, United States). NMR spectra were recorded on a Bruker AVANCE500 spectrometer (Bruker BioSpin GmbH, Rheinstetten, Germany) equipped with a dual carbon/proton CPDUL cryoprobe, that fits 5-mm-diameter NMR tubes, according to the previously described procedure (Li et al., 2021) with a slight modification. Data acquisition was done in acquisition mode with a spectral width of 20 ppm, in digital quadrature detection, with a proton 90° pulse value of 20 μs, offset frequency of 4.7 ppm, 4 s relaxation delay, 65,536 data points, and 128 scans. The metabolites were identified and quantified relatively using the Chenomx NMR Suite (Chenomx, Edmonton, Alberta, Canada).

The identified carbon sources consumed during cell growth in CE were further subjected to a culturing experiment to verify that they were electron donors for denitrification. Strain 6Ca8 was incubated in minimal medium with 5 mM carbon source and 10 mM potassium nitrate (n = 3). Cell growth and nitrate consumption for 20 h were measured. N₂O production rate was measured using the cells at late-log phase. Concentrations of sugars and succinic acid were measured using the F-kit (sucrose/D-glucose/D-fructose and succinate; R-Biopharm, Darmstadt, Germany) according to the manufacturer’s protocol. Pyroglutamic acid content was determined using a Bruker AVANCE III HD 500 NMR spectrometer (Bruker BioSpin GmbH, Rheinstetten, Germany) equipped with a cryogenic probe. NMR measurements were performed using the Bruker standard pulse program “zgpr” with the above-mentioned acquisition parameters with a proton 90° pulse value of 15 μs. The obtained spectra were preprocessed and quantified on TopSpin software v.3.6.2 (Bruker BioSpin GmbH, Rheinstetten, Germany) based on the DSS internal standard.

The distribution and abundance of the isolated denitrifiers in senescent cabbage phyllosphere was characterized via amplicon sequencing and qPCR analyses. Frozen subsamples of fresh and senescent leaves were homogenized by bead beating using a Multi beads shocker MB-200 (Yasui Kikai, Osaka, Japan) at 2,000 rpm, and DNA was extracted using the Power Plant Pro DNA isolation kit (Qiagen, Valencia, CA, United States). To amplify the V4 region of the bacterial 16S rRNA gene, we modified the Earth Microbiome Project protocol to use 16S rRNA primers 515f (5’-GTGYCAGCMGCCGCGGTAA-3’) and 806r (5’-GGACTACNVGGGTWTCTAAT-3’), with the addition of a peptide nucleic acids (PNAs)-matching plastid (5’-GGCTCAACCCTGGACAG-3’) and mitochondrial DNA (5’-GGCAAGTGTTCTTCGGA-3’) (Lundberg et al., 2013). The qPCR analysis was completed using the StepOnePlus™ Real-Time PCR system (ThermoFisher Scientific, Waltham, MA, United States).

For amplicon sequencing, the PCR products were purified using AMPure XP beads (Agencourt Bioscience, Beverley, MA, United States). A second PCR and sequencing was then carried out, according to the protocols supplied with the Illumina MiSeq platform. Using the QIIME 2 pipeline (v. 2021.4) (Caporaso et al., 2010) and the dada2 plugin (Callahan et al., 2017), the paired-end fastq files were processed via primer trimming, quality filtering, merging of paired ends, chimera removal, singleton removal, and construction of a feature table of ASVs. Because the quality scores of the sequences are lower at the ends of the reverse reads, the reverse reads were truncated to 263 bp using the “–p-trunc-len-r” option implemented in the dada2 plugin. Taxonomic identification of ASVs in the feature table was conducted using the SILVA database (release 138) (Quast et al., 2013), using the Qiime2 feature-classifier plug-in (Bokulich et al., 2018). ASVs classified as chloroplasts or mitochondria were removed. The sequence reads of each sample were rarefied to 5,684 reads per sample, and percentage relative abundance and alpha diversity were calculated using the QIIME2 pipeline.

Single-regression analyses were conducted using BellCurve for Excel (SSRI Inc., Tokyo, Japan) to analyze the relationships between N₂O flux and field environmental factors. based on statistical analysis (significant level of 5%). The significance of the difference in the gene expression between the denitrification condition and control (aerobic condition) in RNA-seq analysis was tested with a t-test.

We have shown that residues of soybean, potato, and cabbage crops contribute to N₂O emissions during harvest season in these fields (Itakura et al., 2012; Akiyama et al., 2020). To understand whether the crop residue itself serves as a local hotspot of N₂O emission, cabbage (B. oleracea var. capitata) was cultivated in winter and summer under six patterns of fertilizer application (Supplementary Table 1). During winter cultivation, the N₂O flux occurred immediately after fertilization (0.077 mg N₂O-N m^–2 h^–1) in the treatments with conventional fertilization (PF). N₂O fluxes after fertilization was likely due to the accumulation of ammonium and nitrate in the soil (Figures 1B,C). During summer cultivation, the N₂O flux reached a maximum of 0.064 mg N₂O–N m^–2 h^–1 after fertilization in the treatments with PF. Another remarkably large flux occurred just after the harvest under summer cultivation in the treatments of PF (0.605 mg N₂O–N m^–2 h^–1), chemical fertilization (CF) (0.143 mg N₂O–N m^–2 h^–1), and cow manure compost fertilization with 500 kg N ha^–1 (MC-2) (0.075 mg N₂O–N m^–2 h^–1) and 750 kg N ha^–1 (MC-3) (0.126 mg N₂O–N m^–2 h^–1). This N₂O peak could not be explained by soil ammonia and nitrate because their concentrations were low (Figures 1B,C). In contrast, nitrate accumulated in the unharvested outer leaves (Supplementary Table 1), and the leaf nitrate contents were highly positively correlated with cumulative N₂O emissions during the summer cultivation harvest season (Table 1). N₂O flux declined after cabbage residues were entirely removed from the field or when they were incorporated into the soil (Figure 1A, white arrow). These results suggest that leaf nitrate provides a direct link between unharvested aboveground residues and N₂O emissions.

To confirm direct N₂O emissions from the phyllosphere of the residue, we measured N₂O flux from the crop–soil system and unharvested leaves alone (Supplementary Figure 1B) in the PF treatment. Unharvested leaves produced N₂O, with some leaves showing higher flux than the crop–soil system (Table 2), indicating that the unharvested leaves could be a source of N₂O emissions.

To experimentally confirm the results of the field experiment, a microcosm was prepared with six replicates, using the soil and the unharvested leaves collected from the PF treatment. In five of the six replicates, N₂O was emitted as the leaves became senescent and decomposed (Table 3), and the amount of emission varied depending on the position of the leaf. pH, which affects microbial activity, was higher in the senescent leaves (maximum pH 8.57) than in fresh leaves (average pH 5.47). Furthermore, bacterial populations, measured by qPCR of the bacterial 16S rRNA gene, were considerably larger in senescent leaves (> 10⁸ copies g^–1 fresh wt) than in fresh leaves (at 10⁵ copies g^–1 fresh wt). Nitrate remained in the senescent leaves. Nitrate was moderately and negatively correlated with N₂O flux (correlation coefficient, r = –0.552; Supplementary Figure 4), suggesting that greater nitrate consumption was associated with higher N₂O production. Although the results represented only a snapshot of the nitrate and N₂O levels in senescent leaves, they suggest that nitrate may have been utilized for N₂O production.

Heading crops and leafy vegetables tend to accumulate nitrate in their body because of nitrogen fertilizer applications (Turan and Sevimli, 2005; Czech et al., 2012). Correlations between nitrate concentration and N₂O emissions have been reported for some crop residues (Muhammad et al., 2011). After harvest, plant growth ceases and senescence proceeds; during the senescence, plant tissues disorganize, and degradation of chloroplast, protein, nucleic, and lipid is activated (Havé et al., 2017). Our results suggest that unharvested cabbage leaves contain nitrate, and that the leaf senescence could trigger the direct N₂O emissions from its phyllosphere. During the senescence, nutrient release from the plant tissue, and microbes colonizing the leaf surface could then quickly access these nutrients in order to increase their populations. Under these circumstances, the senescent leaves could become a hotspot for N₂O emissions, independently of soil.

N₂O is produced via two microbial processes: nitrification and denitrification. Amplicon sequencing of bacterial community structure in the phyllosphere of senescent leaves revealed no detectable sequence reads corresponding to nitrifying bacteria (Supplementary Figure 5). As nitrate is the primary electron acceptor for denitrification, we postulated that N₂O is directly emitted from the phyllosphere of senescent cabbage leaf, and denitrification is the main process responsible for its production. This suggestion can be elucidated by isolation and identification of denitrifiers functioning in senescent leaves and analysis of the metabolism of the leaf contents by the isolates.

Although denitrification capability is widespread among plant-associated bacteria, it remains uncertain whether denitrification occurs in the phyllosphere. Our results suggest that the phyllosphere of senescent leaves provides favorable conditions for N₂O production via denitrification. We therefore isolated denitrifiers from the senescent cabbage leaves showing high N₂O flux, using two liquid media: a conventional R2A medium supplied with nitrate (R2A-N medium) and our unique CE medium made from cabbage leaf extract. The denitrifiers were classified into five genera (Supplementary Table 3): Achromobacter, Agrobacterium, Alcaligenes, Brucella, and Stenotrophomonas. Most of the denitrifiers isolated from the CE medium were classified into Agrobacterium spp., and they were further used to clarify the process of denitrification in senescent leaves. As Agrobacterium is one of the most common genera in the phyllosphere (Vorholt, 2012), it provides a suitable model for understanding the processes of denitrification and N₂O emissions in senescent leaves.

To evaluate N₂O emissions by denitrifiers in the phyllosphere of senescent leaves, we cultured the representative isolate, Agrobacterium sp. strain 6Ca8, in CE medium under static conditions imitating the leaf environment. During cell growth, nitrate was consumed concurrently with the release of N₂O (Figures 2A–C), indicating that this strain produces N₂O using the nitrate from the cabbage extracts as an electron acceptor. Dissolved oxygen in the CE medium was consumed within 4.5 h of incubation (Figure 2D), indicating that the cells grew under conditions that were initially aerobic and subsequently anaerobic. This finding has also been reported for plant-associated denitrifiers such as Agrobacterium fabrum C58 (Baek and Shapleigh, 2005) and Ensifer meliloti (Torres et al., 2014). A transition from aerobic to anaerobic conditions may therefore be required for these plant-associated microbes to activate denitrification. Since the phyllosphere is subjected to stressful conditions, such as desiccation and rainfall, which alter water and oxygen contents, plant-associated microbes might need to change their respiration mode depending on the prevailing oxygen level.

To reveal the intracellular molecular mechanisms of denitrification by senescent phyllosphere-derived denitrifiers, we sequenced the complete genome of Agrobacterium sp. strain 6Ca8 (Supplementary Figure 6A), followed by RNA-seq analysis. This revealed a series of denitrifying genes, nap, nir, and nor, encoding nitrate, nitrite, and nitric oxide reductases, respectively, in its genome (Supplementary Figure 6B). RNA transcripts of 6Ca8 cells in CE medium at the early-log phase (at 8 h; Figure 2A) were subjected to RNA-seq analysis. Expression of the nitrate reductase genes napEFDABC was 2.6–4.4-fold higher under denitrifying conditions than under aerobic conditions (Table 4). The expression of the nitrite reductase genes nirVK and nitric oxide reductase genes norDQBCFE increased more than 25.36-fold under denitrifying conditions, which were among the top 20 transcripts with the highest fold-change in expression from aerobic to denitrifying conditions (Supplementary Table 4). The expression of other genes related to electron biogenesis (coding cytochromes, quinols, and pseudoazurin) and of those related to Nir and Nor regulation (fnrN, sinR, nnrR, nnrS, and nnrU) (Zumft, 1997; Baek et al., 2008) increased by up to 60-fold under denitrifying conditions (Table 4). The expression of genes involved in heme cofactor biosynthesis required for Nor (Hino et al., 2010) was higher under denitrifying conditions. This indicates that the expression of genes involved in denitrification was upregulated in 6Ca8 cells.

Electron transport in the respiratory chain shifted from aerobic respiration to denitrification (Table 4). Under denitrifying conditions, the expression of genes encoding NADH hydrogenases (Complex I), representative enzymes for aerobic respiration, decreased, while that of genes encoding microaerobic respiratory enzymes of Complex IV, the cbb₃- and bd-type cytochrome oxidases (Buschmann et al., 2010; Safarian et al., 2016), increased. Nevertheless, the expression of Complex I genes remained high and was comparable to that of denitrification genes, as indicated by the TPM values (Table 4). These results indicate that denitrification and aerobic respiration occurred simultaneously under conditions mimicking the phyllosphere environment.

In the central metabolic pathways, namely, the glycolysis and pentose phosphate pathways and the tricarboxylic acid (TCA) cycle, the expression of genes required to transform the primary precursors essential for cellular biosynthesis, including amino acids, nucleotides, and lipids, and for ATP production was altered (Figure 3; Supplementary Table 5). These results suggest that denitrifiers alter gene expression to produce energy and to initiate cellular biosynthesis, by changing the respiration mode to use either oxygen, nitrate, or both.

Simultaneous aerobic and anaerobic respiration has been reported in facultative anaerobic bacteria such as Escherichia coli (Basan et al., 2015). Microorganisms select the optimal ATP-producing respiration processes depending on the oxygen concentration to optimize their growth rate while minimizing proteomic costs (Basan et al., 2015). If nitrate can be utilized more easily than oxygen as an electron acceptor, denitrification can occur even in the presence of oxygen. Therefore, the co-expression of the genes for denitrification and aerobic respiration observed here is consistent with previous findings. In agreement with this, Chen and Strous (2013) reported that aerobic respiration and denitrification can occur simultaneously if the oxygen concentration around the bacterial cells is kept very low via oxygen consumption by actively respiring microorganisms, even when the oxygen concentration in the gas phase is relatively high. Parkin (Parkin, 1987) estimated that a water or microbial film at least 160-μm thick could achieve anaerobic conditions at the leaf surface. Such conditions can occur in the phyllosphere of senescent cabbage leaves in the fields: the leaves are rich in organic carbon and nitrate as electron donors and acceptors, respectively, thereby promoting the consumption of ambient oxygen and consequently, inducing denitrification. Using optimal pathways to generate the primary precursors for cellular biosynthesis and ATP production enables microbes to minimize metabolic costs and conserve energy (Noor et al., 2010).

The ecological significance of carbon sources in terms of electron supply for denitrification remains poorly understood. We analyzed carbon sources in the CE medium to identify the electron donors used for denitrification. The late-log phase 6Ca8 culture (at 24 h) (Figure 2A) in CE medium was subjected to NMR analysis. The CE medium contained low-molecular weight compounds, such as amino acids, simple organic acids, and sugars (Supplementary Table 6). Glucose, fructose, and sucrose, and components of the TCA cycle (i.e., malate and succinate) were consumed during incubation. Gluconate, methylguanidine, pyroglutamate, and aspartate were also consumed, while pyruvate, formate, acetate, and mannose were accumulated.

The 15 compounds consumed in the culture were further evaluated for availability. When fructose, glucose, pyroglutamate, succinate, and sucrose were used as sole sources of carbons, cell growth, N₂O production, and nitrate consumption was observed (Table 5). These five substrates were consumed simultaneously (Supplementary Table 7), indicating that they acted as electron donors for denitrification in the phyllosphere of senescent cabbage leaves. We did not include malate or aspartate as electron donors because nitrate was not significantly consumed (malate, P = 0.46; aspartate, P = 0.69).

As substrates for denitrification, glucose and succinate are known to have high-energy yields, as reflected in the following formulae:

5C₆H₁₂O₆ + 24NO₃^– + 24H⁺ → 12N₂ + 30CO₂ + 42H₂O

(ΔG°′ = –2,670 kJ mol^–1) (Strohm et al., 2007)

C₄H₆O₄ + 2.8NO₃^– + 2.8H⁺ → 1.4N₂ + 4CO₂ + 4.4H₂O

(ΔG°′ = –1507.59 kJ mol^–1) (Maier and Pepper, 2015)

Pyroglutamate is widely used as a precursor for the synthesis of other amino acids. Although there are no reports relating pyroglutamate to denitrification, glutamate, a pyroglutamate derivative, is a common substrate for denitrification. The Gibbs free energy of the formation (Gf°) of glutamate is comparable to that of succinate (Gf°: glucose, -917.22 kJ mol^–1; succinate, -690.23 kJ mol^–1; and glutamate, -699.6 kJ mol^–1) (Madigan et al., 2014). In cabbage crops, the accumulation of succinate, glutamate, and malate is linked to nitrate uptake (Turan and Sevimli, 2005). Thus, cabbage leaves have the potential to become preferred sites for denitrification and N₂O emission, by accumulating both electron donors and acceptors. We assumed that the other two electron donors, fructose and sucrose, are utilized similarly to glucose, as they are glucose derivatives. This assumption is supported by our RNA-seq analysis showing significant expression of the genes involved in the conversion of sucrose to glucose via fructose (aglA and xylA) under denitrifying conditions (Supplementary Table 5).

Our study is the first to identify the electron donors enabling denitrification in crops. Although we analyzed low-molecular weight carbon substrates, these can also have high molecular weights. Further studies are needed to reveal other electron donors—the hidden drivers for denitrification.

We analyzed the genomes of the isolated denitrifiers: Agrobacterium spp. strains 6Ca8, 5Ca39, and 5Ca50 (Supplementary Figure 6A). In the genome of strains 6Ca8 and 5Ca39, nap, nir, and nor formed a genetic cluster (Supplementary Figure 6B). This cluster has been found in the representative denitrifier, A. fabrum C58, and is commonly distributed in other denitrifying strains of Agrobacterium spp. and related genera, according to the Agrogenom database (Lassalle et al., 2017).⁴ In contrast, the genome of strain 5Ca50 encoded the N₂O reductase genes nosRZDFYLX, and organization of the denitrification genes differed from that of 6Ca8 and 5Ca39 genomes: in 5Ca50, nir and nor are encoded on a large chromosome and nap and nos on a small chromosome. Presence or absence of nos in the genome of 5Ca50 and 6Ca8 was supported by culturing experiments, which confirmed that 5Ca50 has N₂O reductase activity, and that N₂O is the end product of denitrification by strain 6Ca8.

To elucidate the distributions of 6Ca8, 5Ca39, and 5Ca50 in the phyllosphere of senescent cabbage leaves, we analyzed the bacterial community structure of the senescent leaves and the relative abundances of the these three isolated denitrifiers, based on 16S rRNA gene amplicon sequencing (Supplementary Figure 5 and Table 3). Sixteen bacterial orders were identified as dominant in the cabbage leaf phyllosphere. Alteromonadales, Burkholderiales, and Caedibacterales dominated in the senescent leaves, while Rhodobacterales and Exiguobacterales dominated in the fresh leaves (Supplementary Figure 5). ASVs identical to strains 6Ca8 and 5Ca50 were detected in up to 5.30 and 3.37%, respectively, of the senescent leaves, while were rare (at up to 0.641%) in fresh leaves (Table 3). Considering that the bacterial population increased explosively in the senescent leaves (based on the 16S rRNA gene copies; Table 3), these two strains might develop their populations due to plant senescence. No ASV of strain 5Ca39 was detected in any of the samples, suggesting that this strain is a minor denitrifier.

Accumulating evidence indicates that denitrification is a modular process performed by denitrifiers with partial or complete nitrate- or nitrite-respiration pathways (Lycus et al., 2017). Our genome analysis indicates that 6Ca8 and 5Ca39 produce N₂O, while 5Ca50 has the potential to reduce N₂O. Although we did not detect direct evidence of interactions between strains 6Ca8 and 5Ca50, our findings suggest that the phyllosphere of senescent cabbage leaves could be a site for both N₂O production and reduction, and that their activity and interaction may influence the amount of N₂O released into the atmosphere.