Introduction

Over the past 150 years, there has been a significant influx of reactive nitrogen (N) into the biosphere, consequently accelerating microbial N transformations1,2. Denitrification is an important sink for reactive N, by which nitrate (NO3−) is reduced to gaseous forms of N, including nitrous oxide (N2O), a major greenhouse gas and contributor to ozone depletion2. N2O can persist in the atmosphere for more than 100 years3, and without effective mitigation, anthropogenic emissions of N2O are expected to rise due to the accelerating use of synthetic N fertilizers2. An improved understanding of the environmental factors that control N2O production and consumption is a prerequisite for the development of mitigation strategies4.

Denitrification involves multiple reduction steps catalyzed by various enzymes, including NO3− reductases (NAR), nitrite (NO2−) reductases (NIR), nitric oxide (NO) reductases (NOR), and N2O reductases (NOS). In complex environments like soil, denitrification is mediated by a diverse network of facultative anaerobes that either have a complete or partial set of denitrification enzymes5,6,7. This modular nature of the denitrification pathway affects soil denitrification phenotypes, i.e., the kinetics of the four steps of denitrification and the transient accumulation of the intermediates in response to anoxia7,8, which determines the soil’s propensity to emit N2O to the atmosphere. Understanding these phenomena is crucial for designing biotechnologies for mitigating N2O emissions from farmland.

A soil’s propensity to emit N2O is intricately linked to a series of environmental factors, one of which is oxygen (O2) that acts as a superordinate repressor of denitrification via FNR-like transcriptional regulators, two-component systems such as NarX-NarL, and small RNAs9,10,11,12. Detailed pure culture studies have revealed that denitrifying bacteria avoid entrapment in anoxia (without energy to synthesize denitrification enzymes) in various ways, such as the ‘early’ transcription of NAR, and sometimes NOS, at the cusp of anoxia13,14,15, and the preservation of intact denitrification enzymes throughout oxic episodes16. These extant enzymes are ready to be used during subsequent anoxic episodes, representing an ‘anoxic legacy’ that enhances denitrification rates. However, denitrification reductases appear to vary in their persistence during oxic respiration, influencing the transient accumulation of intermediates during subsequent anoxic episodes. For example, in vivo studies of various bacterial strains provided evidence that while NOS remains intact when cells engage in aerobic respiration14,17,18, the cd1-containing NirS enzyme is irreversibly damaged16. This implies that repeated exposure to anoxic episodes will lead to an accumulation of intact NOS, but not NIR, resulting in a gradual reduction of the transient N2O accumulation. That being said, much remains to be clarified regarding the mechanisms through which O2 causes a transient inhibition or permanent damage to denitrification reductases, and if the Cu-containing NirK5 is as sensitive as NirS to O2. It is also uncertain to which extent various denitrifying organisms can concurrently engage in aerobic respiration and denitrification by directing electrons to both oxidases and reductases, as suggested by Chen and Strous19 and references therein. This phenomenon would possibly depend on the organism’s repertoire of high-affinity terminal oxidases20.

Taking into consideration the large respiratory flexibility in bacteria21 and the fact that protein synthesis is estimated to account for ~75% of the energy cost in growing bacteria22, it is conceivable that O2 fluctuation regimes exert a selective pressure, favoring certain denitrifying organisms. Frequent short anoxic spells would favor organisms that minimize their energy expenditure by not synthesizing a complete denitrification proteome. This could be due to the lack of some of the genes or the late synthesis of some of the denitrification enzymes in hypoxia. If this is the case for N2O reductase, it would result in increased N2O emissions. The expression of complete denitrification would be energetically advantageous if anoxic spells last longer.

The principal objective of our study was to determine how a soil´s pre-existing oxygenation history affects denitrification progression and N2O emissions when exposed to anoxic conditions. Using soil microcosms that received regular additions of NO3− and organic carbon (dried clover powder), we imposed three O2 legacy pre-treatments: Ox, oxic conditions (no imposed anoxia); SA, alternations between short anoxic periods and longer periods of oxic conditions; and LA, alternations between long anoxic periods and shorter oxic periods. Denitrification phenotypes for each O2 legacy treatment were investigated during a final anoxic incubation using time-resolved, high-sensitivity rate measurements of N-oxide gas kinetics. We analyzed the abundance of denitrification genes and transcripts using metagenomic and metatranscriptomic techniques. We hypothesized that: (1) repeated anoxic treatments would lead to higher denitrification rates and increased abundance of denitrification genes and transcripts compared to the oxic pre-treatment; and 2) short anoxic spells would select for partial denitrifiers lacking NosZ through gene deletion or regulatory control, subsequently resulting in increased N2O emissions compared to long anoxic spells.

Results

Oxygen legacy establishments

The O2 pre-treatment regimes, consisting of 11 consecutive cycles (Fig. 1), instilled three different O2 legacies in the soil. Gas kinetics in the three treatments (Ox, LA, and SA) were monitored during this time to ensure that the implemented O2 legacy regimes were indeed causing functional changes. For LA, the initial rate of denitrification (calculated as the rate of electron flow to denitrification during the first 7 to 10 h) displayed a relatively steady linear increase in the initial rates, from 1.5 µmol e− vial−1 h−1 for Cycle 1 to 3.2 µmol e− vial−1 h−1 for Cycle 11 (Fig. 2a). The N2O-index (IN2O, a proxy for a soil’s propensity to emit N2O23), showed a nonlinear but gradual decline throughout the 11 cycles (Fig. 2b), consistent with the drastic decrease in the time needed to fully reduce N2O to N2 from approximately 30 h during Cycle 1 to 6 h during Cycle 11 (Supplementary Fig. 1a, b). The N2 production varied between cycles (Supplementary Fig. 1b). During Cycle 1, N2 production plateaued at ~7 µmol N vial−1, demonstrating the consumption of innate and added soil NO3−, whereas for the remaining cycles, N2 plateaus were between 2.5 and 3.7 µmol N vial−1. The NO concentration ([nM]liquid) also decreased with each successive cycle, from a peak concentration of 760 nM in Cycle 1 to 63.3 nM in Cycle 11 (Supplementary Fig. 1c, d). The initial VNOS (N2O → N2) increased from Cycle 1 (0.04 µmol N vial−1 h−1) to Cycle 11 (0.63 µmol N vial−1 h−1) (Supplementary Fig. 1e). Cycle 1 and the final anoxic incubation showed very high NO accumulation compared to the other cycles in the O2 establishment phase (in most cases 4–7 times higher; Supplementary Fig. 1c, d), which most likely slowed the initial VNOS (Supplementary Fig. 1e), leading to a lag in N2 production (Supplementary Fig. 1b). Mass balance calculations for LA showed that 42–76% of the added NO3− was recovered as N2 in Cycles 2–11 (except for Cycle 8 where, for unknown reasons, only 25% of the added N was used for denitrification) (Supplementary Fig. 1f). The remaining NO3− was plausibly reduced by dissimilatory nitrate reduction to ammonium (DNRA). We do not know how much nitrate was produced during the O2 legacy establishment phase (via nitrification during the oxic periods), thus the amount of added NO3− that was not accounted for in N2–N represents the minimum proportion of NO3− that was reduced by DNRA. The percentage for Cycle 1 could not be estimated since the recovered N2 exceeded the provided NO3−.

Fig. 1: Establishment of the oxygen legacies.
figure 1

Three oxygen legacy treatments were instilled over 11 repeated cycles, all of which were initiated simultaneously across all treatments. Treatments included: Ox, oxic conditions (no imposed anoxia); SA, alternations between short anoxic periods and longer periods of oxic conditions; and LA, alternations between long anoxic periods and shorter oxic periods. The treatments underwent the following regimes: Ox microcosms were maintained oxic throughout the entire instillment period. At the start of each cycle, they were amended with 8.5 mg crushed clover supplemented with 0.55 µmol KNO3. SA and LA microcosms were made anoxic at the start of each cycle, immediately after the addition of 8.5 mg crushed clover supplemented with 5.5 µmol KNO3. The SA microcosms were opened to oxic conditions once N2O accumulation reached between 100-200 ppm (before detectable N2 accumulation), while the LA microcosms were opened when the N2 concentration approached a stable level, indicating the completion of denitrification. Moisture levels were monitored and kept constant throughout the incubation with the addition of water. The total length of time for each cycle is presented, and the time periods during which the vials were kept under anoxic and oxic conditions during each cycle are indicated. Data for Cycle 9 was lost; thus, only the duration of the entire cycle is known. Source data are provided as a Source Data file.

Full size image
Fig. 2: Initial denitrification rate and N2O index of the long anoxic (LA) treatment during the oxygen legacy establishment phase and final anoxic incubation.
figure 2

a Initial denitrification rate (VeD[T0]) estimated by linear regression of electron flow to denitrification during the first 7–10 h of incubation using the sum of electron flow to all intermediates. b N2O index (IN2O), a proxy for a soil’s propensity to emit N2O. Eight of 11 cycles are presented; for cycles 6, 7, and 9, there were issues with the robot incubation system resulting in poor-quality data. Values for the final anoxic incubation are also presented. The cycles in the O2 legacy establishment phase are represented by circle symbols, whereas the final anoxic incubation is represented by square symbols. Source data are provided as a Source Data file.

Full size image

The SA vials were monitored until N2O had accumulated to 2 µmol N vial−1 upon which they were re-exposed to O2. Thus, the denitrifier community in this treatment was not allowed to complete denitrification. The time it took for the SA vials to reach this level of N2O decreased from 8 h in Cycle 1 to only 2 h by Cycle 11 (Fig. 1). Rates of N2O reduction and N2 production were not calculated due to the small number of measuring points.

In the Ox treatment, the initial rate of aerobic respiration was monitored at the start of each successive cycle, in which the rate of electron flow to aerobic respiration increased from an average of 11.8 µmol e− vial−1 h−1 in Cycle 1 to 20.3 µmol e− vial−1 h−1 by Cycle 8 (Supplementary Fig. 2a). The Ox treatment was also monitored for NO and N2O to check if any significant denitrification occurred due to anoxic microsites within the soil matrix. The results show that practically no denitrification took place: NO was barely detectable (0.5–1 nmol NO vial−1), and N2O concentration remained close to the atmospheric background (Supplementary Fig. 2b), indicating that the majority of the organisms in Ox had ample provision of O2 throughout the establishment phase.

The variation in lengths of the oxic/anoxic periods in the different cycles (Fig. 1) mirrors the situation in the soil where hypoxic/anoxic spells in microenvironments of the soil matrix may last from minutes to days/weeks and possibly from seconds to years24. Also noteworthy is that the varying duration of the oxic phase through Cycles 1–11 was apparently inconsequential. In the LA treatment, the increase in initial denitrification rates as well as the decline in IN2O (Fig. 2) were unaffected by the length of the oxic phases, which varied from 30 to 72 h (Fig. 1).

Final anoxic incubation

The three O2 legacy treatments showed distinct denitrification phenotypes during the final anoxic incubation (Fig. 3). The initial NO3− reduction rate (VNAR), calculated as the coefficient of the linear regression equation of NO3− consumption during the first two hours of incubation, was similar across treatments with an average of 4.0 µmol N vial−1 h−1 (Table 1), whereas NO2− reduction was greatest in Ox, as indicated by the two-times significantly greater NO2− accumulation in LA and SA (Fig. 3). This was consistent with the significantly greater initial NO2− reduction rate VNIR (NO2− → NO) in Ox (0.89 µmol N h−1) compared to LA and SA (average of 0.47 µmol N vial−1 h−1) (Table 1). The Ox treatment also had a significantly greater initial VNOR (NO → N2O) and VNOS (N2O → N2) compared to LA and SA (Table 1) (calculations for VNIR, VNOR, and VNOS are detailed in Lim et al.25). Consistent with the initial VNIR, VNOR, and VNOS, Ox completed denitrification the earliest (~25 h; Fig. 3a) with the greatest maximum denitrification rate of 0.85 µmol N vial−1 h−1 and lowest accumulation of NO (18.1 µmol N vial−1) and N2O (49.9 µmol N vial−1) (Table 1). The LA treatment was the second fastest to complete denitrification of NO3− (~75 h; Fig. 3c) with a maximum denitrification rate of 0.55 µmol N vial−1 h−1, NO accumulation of 31.1 µmol N vial−1, and N2O accumulation of 134 µmol N vial−1 (Table 1). The SA treatment took the longest to complete denitrification (>100 h; Fig. 3b) with a maximum denitrification rate of 0.46 µmol N vial−1 h−1 and the greatest accumulation of intermediates totaling 146 µmol N vial−1 for NO and 592 µmol N vial−1 for N2O (Table 1). The IN2O was 0.31, 0.60, and 0.91 for Ox, LA, and SA, respectively (Table 1). Mass balance calculations showed that 77, 82, and 89% of the added NO3− was recovered as N-gases (thus used for denitrification) in the LA, SA, and Ox treatment, respectively.

Fig. 3: Denitrification gas kinetics in the oxygen legacy treatments during the final anoxic incubation.
figure 3

Gaseous emissions of nitric oxide (NO; μmol vial−1), nitrous oxide (N2O, µmol N vial−1), and N2 (µmol N vial−1). Nitrate (NO3−) and nitrite (NO2−) concentrations (µmol N vial−1) during the first 2 h of the incubation are included in sub-panels. a Oxic treatment (Ox). b Short anoxic treatment (SA). c Long anoxic treatment (LA). Data are presented as mean values ± the standard error (n = 3 biological replicates derived from individual soil microcosms, where each microcosm served as an individual experimental unit). Source data are provided as a Source Data file.

Full size image
Table 1 Indicators of denitrification progression in the oxygen legacy treatments during the final anoxic incubation
Full size table

PLFA analysis

The repeated clover amendments induced microbial growth, as evidenced by the 77–91% increase in total and bacterial PLFA abundance for all clover-amended soils compared to the original soil (Fig. 4a, b). The fungi grew more than bacteria, as indicated by the doubling of the fungal: bacterial PLFA ratio (treatment average of 0.077 vs. 0.038 in the original soil) (Fig. 4c). A PCA analysis of the total microbial community structure separated the clover-amended soils from the original soils along PC1, which explained 47.6% of the variation. Ox separated from SA and LA along PC2, which explained 20.5% of the variation (Fig. 4d).

Fig. 4: Soil microbial biomass and community structure of the oxygen legacy treatments based on phospholipid fatty acid (PLFA) analysis.
figure 4

a Total microbial biomass. b Bacterial biomass. c Fungi:bacteria biomass ratio. d PCA analysis of the total microbial community PLFA structure. Samples for the original soil were collected pre-experiment after soil processing, whereas samples for the Ox, SA, and LA treatments were collected during the final anoxic incubation at T = 2.0 h. Data are presented as mean values ± the standard error (n = 3 biological replicates derived from individual soil microcosms, where each microcosm served as an individual experimental unit). The distribution of individual replicate values is presented. Source data are provided as a Source Data file.

Full size image

Metagenomic and metatranscriptomic community analysis

Supplementary Table 1 provides data on the total sequenced reads after read trimming for both the metagenomic and metatranscriptomic data. To check if treatments had any effect on the composition of metagenomes and metatranscriptomes, we conducted PCoA analyses of the MASH-based distances. While the metagenomes were not significantly different (Supplementary Fig. 3a, PERMANOVA P = 0.055), the metatranscriptomes demonstrated significant spatial clustering (Supplementary Fig. 3b, PERMANOVA P = 0.001). Sequence α-diversity based on the metagenomic data was estimated by calculating their Nd sequence diversity using Nonpareil26. The order of Nd sequence diversity as a metric for α-diversity from highest to lowest was: Original soil (25.1 ± 0.53) > Ox (24.7 ± 0.55) > LA (24.1 ± 0.33) > SA (23.9 ± 0.59) (Supplementary Fig. 4).

Denitrification, DNRA, and ROS-scavenging enzyme gene and transcript read abundances in the metagenomes and metatranscriptomes

The most abundant denitrification gene in the original soil was narG, resulting in ~2.5 times more gene reads per total million reads (RPM) than napA, nirK, and qnor (Fig. 5a; Supplementary Table 2). The abundances of nirS, cnor, and nosZ clades I and II were all rather low, with values between 5 and 10 RPM. The narG abundance was greater in the O2 legacy treatments than in the original soil, plausibly reflecting the growth of organisms with narG induced by the repeated addition of clover in these treatments; however, this increase in narG was greater in SA and LA than in Ox (Supplementary Table 2). The abundance of the other denitrification genes was generally higher in the legacy treatments than in the original soil, albeit not statistically significant, except for napA and nosZ I, which increased by 30% in Ox (P = 0.011) and LA (P = 0.017), respectively (Supplementary Table 2).

Fig. 5: Abundance and clustering of denitrification gene and transcript reads from the metagenomes and metatranscriptomes of the oxygen legacy treatments during the final anoxic incubation.
figure 5

a Abundance of denitrification gene reads. Reads encode NO3− reductases (napA and narG), NO2− reductases (nirK and nirS), NO reductases (cnor and qnor), and N2O reductases (nosZ clades I and II). b Uniform Manifold Approximation and Projection (UMAP) for dimension reduction analysis of metagenomic profiles based on denitrification gene read abundances. c Abundance of denitrification transcript reads. d UMAP for dimension reduction analysis of metatranscriptomic profiles based on denitrification transcript read abundances. Reads were normalized to reads per total million reads prior to analysis. Metagenomic data: the original soil was sampled pre-experiment after soil processing, whereas the Ox, SA, and LA treatments were sampled at T = 2.0 h during the final anoxic incubation. Metatranscriptomic data: the Ox, SA, and LA treatments were sampled at T = 0.25 h (circles), 1.0 h (triangles), and 2.0 h (squares) during the final anoxic incubation. A detailed list of denitrification gene and transcript read abundances is found in Supplementary Tables 2 and 3. Source data are provided as a Source Data file.

Full size image

Using the denitrification gene read abundances from the metagenomic analysis, the UMAP dimension reduction method demonstrated distinct local clustering of treatments, where the UMAP1 axis differentiated the denitrification profile of the original soil from that of the O2 legacy treatments (Fig. 5b). These trends were consistent with the global clustering of denitrification metagenomes, as demonstrated by a PCA plot generated from the same denitrification gene read abundances (Supplementary Fig. 5a), and likely reflected the clover and NO3− additions.

The metatranscriptomic analysis revealed that the O2 legacy treatments impacted the transcriptional activity of some of the denitrification genes during the first 2 h of the final anoxic incubation (Fig. 5c; Supplementary Table 3). The transcript abundance of narG, nirK, qnor, and nosZ increased over time in Ox, while only napA increased over time in LA, and no transcripts increased over time in SA. Overall, Ox and SA exhibited marginal differences in denitrification gene transcription except for qnor and nosZ II which were significantly more abundant in Ox than in SA, and for narG which was greater in SA than in Ox. LA had the most notable treatment effect on denitrification gene transcription, resulting in significantly greater transcript abundances of napA, narG, nirS, cnor, and nosZ I compared to Ox and SA, and significantly lower nosZ II transcript abundance compared to Ox.

Using the denitrification transcript read abundances, the UMAP dimension reduction method demonstrated a distinct local clustering of treatments, where the UMAP1 axis differentiated the denitrification profile of LA from that of Ox and SA (Fig. 5d). These trends were consistent with the global clustering of denitrification metatranscriptomes, as demonstrated by a PCA plot generated from the same denitrification transcript read abundances (Supplementary Fig. 5b).

The gene read abundance of nrfA, which is involved in DNRA, was consistently low across the original soil and Ox, SA, and LA (average of 16.4–25.8 RPM), while its transcript abundance was affected by treatment: greatest in LA (average of 63.3 RPM), much lower in Ox (average of 10.4 RPM), and almost undetectable in SA (average of 1.05 RPM) (Supplementary Fig. 6). We also investigated the abundance of genes and transcripts of selected ROS-scavenging enzymes, including sodN, sod2, sod1, katE, and katG (Supplementary Fig. 7). Notably, the legacy treatments had no distinguishable effect on the gene or transcript abundance of these genes. Gene abundance was similar across the legacy treatments except for sod1, which was greater in Ox than in SA. For transcript abundance, SA and Ox had a lower abundance of sod1 compared to LA, and Ox had a lower and greater abundance of sodN and katE, respectively, compared to LA and SA.

The relative proportion of nosZ clade gene and transcript reads in the metagenomes and metatranscriptomes

In the original soil, the relative proportion of nosZ clade II gene reads in the metagenomes was twice as large as clade I read (61 vs 29%); however, in Ox, LA, and SA, the proportion decreased to 53–55%, while clade I exhibited a corresponding increase of ~10%. Gene reads not identified as belonging to one of the clades accounted for <10% of all soils (Fig. 6a; Supplementary Table 4). Throughout the first 2 h of the anoxic incubation, the proportion of nosZ transcripts was significantly different between all treatments (HSD, P ≤ 0.05), with nosZ II accounting for 80–90% in Ox, 60–70% in SA, and 40–50% in LA (Fig. 6b; Supplementary Table 4).

Fig. 6: The relative proportion (%) of nosZ clades I and II gene and transcript read abundance from the metagenomes and metatranscriptomes of the oxygen legacy treatments during the final anoxic incubation.
figure 6

a Metagenomic data. b Metatranscriptomic data. Reads were annotated as either nosZ clade I or clade II. Reads that were assigned to both clades were annotated as ambiguous. The metagenomic and metatranscriptomic data were collected as described in Fig. 5. A detailed list of clade proportions is found in Supplementary Table 4 of the Supplementary Information. Source data are provided as a Source Data file.

Full size image

Taxonomic annotation of metagenomes and metatranscriptomes

The most abundant bacterial phyla detected in the metagenomes are shown in Fig. 7a. The overall order from highest to lowest read abundance in all treatments was Actinobacteria > Proteobacteria > Acidobacteria > Chloroflexi > Planctomycetes > Verrucomicrobia > Bacteroidetes > Gemmatimonadetes > Firmicutes, in exception for Ox which had a greater abundance of Proteobacteria (2.3 × 105 reads) than Actinobacteria (1.8 × 105 reads) (Fig. 7a).

Fig. 7: Taxonomic annotation of metagenomes and denitrification gene and transcript reads from the oxygen legacy treatments during the final anoxic incubation.
figure 7

a Taxonomic annotation of reads from the metagenomes. b Taxonomic annotation of denitrification gene and transcript reads from the metagenomes and metatranscriptomes. Reads belonging to the NO3− reductase narG and napA, NO2− reductase nirK, NO reductases qnor, and N2O reductases nosZ clades I and II are presented. Reads were annotated at the phylum level except for reads belonging to Proteobacteria, which were annotated at the class level. The unclassified reads reflect members of the community that are not well represented in the reference databases. The metagenomic and metatranscriptomic data were collected as described in Fig. 5. A detailed list of taxonomic annotations is in Supplementary Table 5 and Supplementary Data 1, which also includes read abundances for other denitrification genes not presented here.

Full size image

The annotated reads of denitrification genes in the metagenomes are shown in Fig. 7b. The reads of napA, nirK, and qnor were dominated by Actinobacteria and Proteobacteria, primarily Alpha-, Beta-, and Gammaproteobacteria (Fig. 7b). Conversely, the metagenomic reads of nirK in the original soil were dominated by Actinobacteria, Acidobacteria, Verrucomicrobia, and Bacteroidetes, while qnor reads were primarily populated by Bacteroidetes. Across treatments, Proteobacteria and Bacteroidetes were the most abundant phyla in the metagenomic reads of nosZ I and nosZ II, respectively.

The metatranscriptomic reads for napA were low in Ox and SA throughout the first 2 h of the final incubation and were dominated by Gammaproteobacteria, while LA had greater napA transcript read abundance, increasing with time, and also dominated by Gammaproteobacteria (Fig. 7b). The metatranscriptomic reads for nirK were dominated by Alpha-, Beta-, and Gammaproteobacteria, yet their abundance profiles varied across treatments: Ox exhibited an increase in abundance over time, SA exhibited an increase in abundance at 2.0 h, and LA had a greater abundance of reads annotated as Gammaproteobacteria at 0.25 h but a greater abundance of reads annotated as Alphaproteobacteria at 2.0 h (Fig. 7b). The annotated metatranscriptomic reads of nosZ I was low in both Ox and SA compared to LA, with LA exhibiting high abundances of Betaproteobacteria throughout the incubation. Across treatments, the metatranscriptomic reads of nosZ II were dominated by Bacteroidetes, particularly in Ox, which exhibited an increase of reads from this phylum over time. A detailed list of taxonomic annotations is in Supplementary Table 5 and Supplementary Data 1.

Discussion

Habitats favoring denitrification are those with fluctuating O2 concentrations. In these settings, carbon sources may be limited but still available, and NO3− is generated by nitrification during oxygenated periods or supplied through fertilization. The prevailing understanding is that the ability to denitrify provides a fitness advantage by securing survival and growth when confronted with anoxic conditions27. It, therefore, lies close at hand to infer that microbial communities adapted to regular shifts between O2-rich and deprived conditions will develop an ‘efficient’ denitrification process with minimal transient production of intermediates. The present study does not support this assumption. Instead, and contrary to our hypothesis, it was the Ox treatment that showed the most efficient denitrification process during the final anoxic incubation, seen through its faster denitrification rate and lower accumulation of NO2−, NO, and N2O compared to both SA and LA (Fig. 3, Table 1). Our experiment is based on one agricultural soil, cautioning against drawing extensive conclusions. Nonetheless, the results underscore that our current understanding of the fitness value of denitrification is incomplete. Importantly, our findings are supported by a study of complex denitrifying communities in seawater28, which showed that samples collected from the oxic layer consumed N2O under anoxic conditions at a faster rate than samples collected from anoxic depths at the same station.

The rapid denitrification in Ox compared to LA and SA could theoretically be due to a greater biomass of denitrifiers, but this was apparently not the case. The total microbial biomass reached similar levels across all three legacy treatments (Fig. 4), and the denitrification gene abundances were also similar (Fig. 5a). This may appear counterintuitive since the growth yield (per mol C) is greater in aerobic than in anaerobic respiration9. However, the soils in SA were kept primarily under oxic conditions throughout the O2 legacy establishment phase, while soils in LA were kept oxic for over 50% of this time (Fig. 1). This indicates that most growth occurred via aerobic respiration in all three legacy treatments during the establishment phase.

The metatranscriptomic analysis also failed to provide a clearcut explanation for the ‘efficient’ denitrification in the Ox treatment, as the denitrification transcripts were generally more abundant in LA than in the other treatments (Fig. 5c, Supplementary Table 3). A notable difference between the treatments was that Ox had the greatest level of nosZ II transcription (Figs. 5c and 6b) and the lowest IN2O (Table 1). This may be attributed to a selection process during the establishment phase that favored bacteria that thrived on clover, used O2 as an electron acceptor, and had a general propensity towards the nosZ II gene type. Studies have demonstrated that a pre-existing history of oxic conditions increased nosZ II transcription at the onset of anoxia29,30, and others have suggested that nosZ II has an enhanced affinity for N2O compared to nosZ I31. Collectively, these characteristics may explain the rapid N2O consumption in Ox during the final anoxic incubation.

The nosZ clade II gene is found in a wide range of taxonomically diverse organisms, many of which have truncated denitrification pathways or are classified as ‘non-denitrifier’ N2O reducers32. Conversely, the nosZ clade I gene is more commonly found in canonical denitrifiers, many of which belong to the Proteobacteria and carry other denitrification genes such as nirS6,32,33. In our study, the differences between the nosZ clades align with the slightly greater diversity and increased activity of Bacteroidetes in Ox and the increased transcription of nirS and activity of Proteobacteria in LA (Figs. 5 and 7, and Supplementary Fig. 4). This implies that LA favored active canonical denitrifiers, while Ox favored active truncated denitrifiers that performed denitrification in a modular fashion with each reductase providing selectable benefits independent of the others6,7. These findings highlight the uncertainty surrounding any selective advantage of being a complete denitrifier in a complex denitrifying community and suggest that the ‘the sharing of work’ (partnering) between organisms may result in a more efficient biogeochemical process, emphasizing the need for more research into this understudied field of microbial ecology.

A comparison of LA and SA during the final anoxic incubation shows, in agreement with our hypothesis, that antecedent soil conditions of long anoxic pulses led to a lower accumulation of NO and N2O (Table 1) and an earlier completion of denitrification (Fig. 3) compared to antecedent conditions of short anoxic pulses. These functional differences in denitrification progression were not due to differences in the abundance of denitrification genes (Fig. 5a) or in initial VNIR, VNOR, and VNOS (Table 1). Transcription of the denitrification genes was, however, generally higher in LA than in SA during the first 2 h of the final anoxic incubation (Fig. 5c, Supplementary Table 3), indicating that antecedent soil conditions characterized by brief anoxic pulses favored organisms employing distinct transcriptional control mechanisms. This could be due to low transcription rates of all cells or some sort of bet-hedging, where only a fraction of the cells in SA transcribed the denitrification genes as a strategy to conserve energy if O2 were to return34,35. If so, the initial denitrification rates in SA being equal to those in LA (Table 1) may be explained by the activity of denitrification reductases that were produced during earlier anoxic events. In contrast, LA favored organisms that synthesized a full denitrification proteome at the onset of anoxia. The metabolic cost was rewarded in terms of a more rapid denitrification progression and a lower transient production of intermediates. This points to the succession of a fluctuation-adapted denitrifier community in LA that was strongly dependent on the length of the anoxic-oxic cycles, similar to what was suggested by Pett-Ridge and Firestone36.

During the O2 legacy establishment phase, LA exhibited a progressively more efficient denitrification phenotype through repeated exposure to denitrifying conditions, as evidenced by an increasing denitrification rate and the concurrent decrease in N2O index and maximum NO concentration with each successive anoxic–oxic cycle (Fig. 2; Supplementary Fig. 1c, d). One reason could be that frequent switches between anoxic and oxic conditions enhanced denitrification at high O2 concentrations as a result of ‘aerobic denitrification’37. While we have reservations about this term, we acknowledge the potential for detectable denitrification in the presence of O2 in two scenarios: (1) denitrification that occurs in anoxic microsites of a seemingly oxic system38; or (2) the co-respiration of O2 and N-oxides within the same cell when electron flow to terminal oxidases is restricted by low O2 concentrations19, provided there is no O2-induced damage to the denitrification reductases.

The trend of progressively increasing denitrification rates and decreasing accumulation of denitrification intermediates in LA during the O2 legacy establishment phase was unexpectedly interrupted in the final anoxic incubation (Fig. 2). One tentative explanation for this ‘phenotype reversion’ is potential O2-induced damage to denitrification reductases during the legacy establishment phase, combined with the increased NO3− concentration in the final anoxic incubation. The final incubation showed a significantly lower initial VNIR and greater accumulation of NO2− in LA compared to Ox (Table 1; Fig. 3c), despite LA having greater transcription of NIR genes (Fig. 5c, Supplementary Table 3). These findings may reflect an impairment of NIR, particularly NirS, which is known to be irreversibly damaged by O216. Subsequently, the accumulated NO2− may have undergone abiotic reactions with soil components25, leading to NO accumulation in LA (Supplementary Fig. 1c). NO accumulation would also occur if NOR was hampered by oxidative stress. In return, elevated NO concentrations may have affected NOS, resulting in increased N2O accumulation during the final incubation39,40. The absence of this effect in LA during the O2 legacy establishment phase could be due to undamaged reductases effectively managing the repeated additions of 2 mM NO3−; however, in the final anoxic incubation, 4 mM NO3− may have exceeded their capacity, subsequently slowing relative N-oxide reduction rates. Although speculative, this combined effect may also serve as an additional explanation for the lower accumulation of intermediates in Ox compared to LA and SA (Fig. 3), as Ox likely did not produce reductases during the legacy establishment phase, thus avoiding oxidative damage and utilizing only ‘freshly’ synthesized reductases in the final anoxic incubation.

A recent study pointed to the role of ROS in the inhibition of N2O reduction in soils and sediments during the transition from anoxic to oxic conditions41. In our study, we found no clear O2 legacy treatment effect on the abundance of ROS-scavenging genes and transcripts across treatments when measured during the final anoxic incubation (Supplementary Fig. 7). This does not preclude the possibility that ROS-scavenging enzymes mitigated the toxicity of ROS produced during the O2 legacy establishment phase, as these enzymes were not monitored during this period.

The specific type of active denitrifiers favored by SA remains unclear. The transcriptional dynamics during the final anoxic incubation were similar to that of Ox, as evidenced by the close clustering of their denitrification metatranscriptomes (Fig. 5d). Yet, SA was genotypically and phenotypically more similar to LA, subsequently sharing close clustering of denitrification metagenomes (Fig. 5b) and exhibiting slowed denitrification kinetics during the final anoxic incubation (Fig. 3). Future approaches should investigate oxidative damage to denitrification reductases using detailed proteomic analyses, as described for another group of enzymes42, thereby providing a more holistic understanding of functional changes in response to fluctuating O2 conditions.

Our study supports the hypothesis that antecedent conditions of long anoxic pulses resulted in a faster denitrification phenotype at the onset of anoxia compared to a history of short anoxic pulses. Surprisingly, a history of constant oxic soil conditions gave rise to the fastest denitrification phenotype at the onset of anoxia and favored a denitrifier community dominated by active nosZ clade II-bearing partial or non-denitrifiers, suggesting efficient partnering of the reduction steps among organisms. Overall, our study underscores the necessity for further investigations into the interactions among organisms involved in denitrification and highlights that knowing the O2 legacy of a complex environment is crucial for accurately predicting N2O emissions arising from denitrification.

Methods

Soil collection and soil characteristics

Loamy soil was collected from the top 15 cm of a grassy ley field in Ås Norway (59°39'46''N, 10°45'40''E), pooled, sieved (2 mm), and stored at 4 °C. The soil had 3.0% total carbon content, 0.26% total N content, and a pH of 5.2 in a 1:2 0.01 M CaCl2 slurry. A portion of the original sieved soil was frozen at −80 °C for nucleic acid analysis, while the remaining soil was divided into 120 mL glass serum vials at 8.5 g dry weight (dw) soil vial−1 and stored at 4 °C for 2 weeks during diagnostic testing. Diagnostic tests were run to determine the soil moisture content and clover and NO3−-amendment rates that would enhance the activity of the soil microbial community and ensure detectable transcription while avoiding toxic levels of gaseous N-oxides and hypoxic/anoxic soil microsites. These tests are described in detail in the Supplementary Information. Each soil microcosm was adjusted to 40% of the water holding capacity (0.32 mL water g−1 dw soil) and amended with 8.5 mg dried clover powder containing either 5.5 µmol or 0.55 µmol NO3− (treatment specifications detailed below).

Oxygen legacy establishment phase

Over a four-week period, three distinct O2 legacy treatments were imposed on the soil microcosms: soils that were kept ‘oxic’ (Ox) or pulsed with short periods (short anoxic, SA) or long periods (long anoxic, LA) of anoxia. Each treatment had 15 vials for replication purposes and to permit destructive sampling of various measurements during the final anoxic phase. A subset of vials was placed in a robotized incubation system set at 15 °C for continuous denitrification gas kinetics. For these measurements, SA and LA were performed with five replicates (n = 5; except Cycle 11 n = 3), whereas Ox was performed with two replicates (n = 2) due to space limitations in the robotic incubation system and the expectation that no or very little denitrification would occur in this treatment. The remaining vials were treated equally and incubated ‘off-line’ (outside of the incubation robot).

The O2 legacies were instilled over 11 repeated cycles, all of which were initiated simultaneously across all three treatments (Fig. 1). The treatments underwent the following regimes: Ox microcosms were maintained oxic throughout the entire instillment period. At the start of each cycle, they were amended with 8.5 mg crushed clover supplemented with 0.55 µmol KNO3. SA and LA microcosms were made anoxic at the start of each cycle, immediately after the addition of 8.5 mg crushed clover supplemented with 5.5 µmol KNO3. The SA microcosms were opened to oxic conditions once N2O accumulation reached between 100 and 200 ppm (before detectable N2 accumulation), while the LA microcosms were opened when the N2 concentration approached a stable level, indicating the completion of denitrification. Figure 1 presents the length of time for each cycle during the O2 legacy establishment phase, including the time that each treatment was incubated under oxic and anoxic conditions. To ensure even distribution of the clover, it was mixed into the soil by rolling the vials for 30 seconds. The Ox treatment received 1/10th of the NO3− than the SA and LA treatments so that the system was not overwhelmed with toxic levels of NO2−. The SA and LA vials were made anoxic by crimp sealing the vials with butyl septa and using an automated manifold for six cycles of gas evacuation and helium (He) filling23. The Ox vials were also crimp sealed with butyl septa but were not He-washed. Unfortunately, gas data was lost for Cycle 9, so only the duration of the entire cycle is known. Throughout the O2 legacy establishment phase, vials were weighed, and water was added to maintain the desired soil moisture content. During periods of O2 exposure, vials were sealed with parafilm punctured with small holes to secure oxic conditions but with minimum loss of moisture.

Final anoxic incubation

NO3− and NO2− were measured after Cycle 11 to test for their accumulation during the O2 legacy establishment phase. No significant differences between the treatments were observed (Fig. 3). To secure a long enough time span of anaerobic respiration to obtain high-resolution denitrification kinetics, the amount of NO3− was doubled in the final incubation, since preliminary experiments showed that the 5.5 µmol NO3− addition used for LA and SA in the O2 establishment phase was the minimum amount needed. Thus, 11 µmol KNO3− on 8.5 mg crushed clover was uniformly added to each treatment. This addition corresponded to 4 mM NO3− in the soil moisture, which is well within the range that microbes are expected to encounter in soil43. The Ox, SA, and LA treatments were made anoxic by six rounds of He-washing, as described above. The vials were then split into two sets. The first set was placed in the robot incubation system set at 15 °C for continuous gas kinetic measurements using three replicates per treatment (n = 3). The second set of vials (12 vials per treatment) was treated equally but incubated ‘off-line’ and destructively sampled throughout the incubation for analyses of NO3−, NO2−, PLFA, metagenomics, and metatranscriptomics. Each treatment, time, and replicate combination served as an individual destructive sample that was discarded after sampling. Measurements for NO3− and NO2− were sampled at T = 0 h, 0.25 h, 1 h, and 2 h after He-washing, metatranscriptomics at 0.25 h, 1 h, and 2 h after He-washing, and PLFA and metagenomics at 2 h after He-washing. These measurements were performed using three replicates per each treatment-time combination (n = 3), except for the metatranscriptomic analysis, which was performed using two replicates per each treatment-time combination (n = 2) apart from LA and SA at 0.25 h, where only one replicate per treatment was used. Measurements for multiple parameters at the same time point were taken from the same destructive vial used for each treatment, time, and replicate combination. Soil from these time points was immediately analyzed for NO3− and NO2− or was snap-frozen in liquid N2 and stored at −80 °C until nucleic acid extraction.

Gas kinetics

Denitrification kinetics were monitored with frequent headspace measurements of O2, NO, N2O, and N2 using a robot incubation system and OpenLAB CDS 2.3 software for gas chromatograph (GC) data acquisition (Agilent), as described in detail by Molstad et al.44,45. Briefly, the microcosm vials were maintained in a thermostatic water bath. Gas samples were taken at intervals from the headspace by an autosampler coupled to a peristaltic pump. Gas samples were pumped into a gas chromatograph (Varian; 7890 A GC, Agilent) for analysis of O2, N2O, and N2, and into a chemiluminescence NOx analyzer (M200E, Teledyne) for analysis of NO. For each sampling, an equal volume of He was pumped back into the vial to maintain ~1 atm pressure in the vials. Measurements for N2 were mathematically compensated for the dilution that took place when sampling the headspace.

Nitrate and nitrite concentrations

NO3− and NO2− measurements were conducted as described by Lim et al.25. Briefly, 0.5 g of fresh-weight soil was collected, vortexed in 0.5 mL MilliQ water, and then centrifuged to remove the soil particles. The supernatant (10 μL) was immediately injected into a purging device with reducing agents: either acetate + NaI (room temperature) for converting NO2− to NO, or HCl + VCl3 (95 °C) for converting NO3− + NO2− to NO. The NO was transported by a stream of N2 to a chemiluminescence detector system. The integrated NO peaks were used to estimate NO2− and NO3− + NO2−, in which NO3− concentrations were determined by subtracting the NO2− from the NO3− + NO2− values.

Nucleic acid extraction

DNA for metagenomic sequencing was extracted according to Lim et al.46, which is a modified version of the method by Griffiths et al.47. RNA for metatranscriptomic sequencing was extracted using the RNA PowerSoil Total RNA Isolation Kit (Mo Bio catalog No. 12866-25). This is described in detail in the Supplementary Information.

DNA and RNA sequencing

DNA samples from the original soil and three O2 legacy treatments (collected at T = 2.0 h) were sent for metagenomic sequencing at the JP Sulzberger Columbia Genome Center in New York City, USA. DNA was sequenced using the Illumina HiSeq next-generation sequencing platform with 2 × 100 bp paired end reads. RNA samples (collected at T = 0.25, 1.0, and 2.0 h) were sequenced from the O2 legacy treatments. Raw RNA was sent for metatranscriptomic sequencing at the Roy J. Carver Biotechnology Center at the University of Illinois. RiboZero technology was used to remove ribosomal RNA from the samples to enrich messenger RNA. The remaining RNA was converted to cDNA and subsequently sequenced using the Illumina HiSeq next-generation sequencing platform to generate single-end 160 bp reads. DNA and RNA met the purity and quality standards of the respective sequencing centers. Initial quality assurance/quality control (QA/QC) was carried out by the respective sequencing centers, which included trimming sequencing adapters and bar codes from sequence reads. Additional QA/QC was performed using the CLC Genomics Workbench 4.0, which involved trimming low-quality reads and short reads ( < 80 bp for DNA; <70 bp for RNA). After read trimming, the number of reads ranged from 14 to 18.2 million in the metagenomic samples and between 10.2 and 12.1 million in the metatranscriptomic samples.

Metagenomic and metatranscriptomic analyses

Read quality was verified using FastQC48. Compositional dissimilarities between treatment metagenomes and metatranscriptomes were assessed using MASH-based distance estimations49. The Nonpareil (Nd) sequence diversity was calculated using the quality-controlled read sets in the Nonpareil 3 software (version 3.3) with default parameters and the R software provided with the program26. For functional annotation of denitrification reductase genes and transcripts, reads were aligned using DIAMOND50 with an e-value cut-off of 1 × 10−3 and compared against a custom data set described in Nadeau et al.51 and Roco et al.52 using the methods described in Frostegård et al.53. Outputs of annotated reads were normalized to reads per million of total reads to account for differing sequencing depths. Reads assigned to nosZ were classified into clades I and II using the full-length sequences presented in Juhanson et al.54 and Conthe et al.55. Downstream filtering and final assignment were performed using the custom code described in Nadeau et al.51. This approach is described in detail in the Supplementary Information. Uniform Manifold Approximation and Projection for Dimension Reduction analysis (UMAP; arXiv:1802.03426 [stat.ML]) was carried out on numerical data frames of denitrification genes (napA, narG, cnoR, qnoR, nirK, nirS, nosZ I, and nosZ II) using the UMAP package in R (parameters: n_neighbors=4, random_state=123, n_epochs=200, metric =“euclidean”). Reads derived from the denitrification reductases were extracted from read sets53 and uploaded to KBase56, where their taxonomic assignment at all phylogenetic levels was performed using default values in KAIJU57. The gene and transcript read abundance of enzymes involved in dissimilatory nitrate reduction to ammonium (DNRA; nrfA), and of select reactive O2 species (ROS)-scavenging enzymes (sodN, sod1, sod2, katE, and katG) were also investigated using the Functional Mapping and Analysis Pipeline (FMAP) tool58.

PLFA analysis

The PLFA analysis was done as described in Frostegård et al.59 using GC conditions described in Jia et al.60. Individual FAMEs were identified using a pre-established database and quantified in relation to the internal standard using TurboChrom (v 6.1.1; Perkin-Elmer)60. Conversion of PLFA data to bacterial numbers and fungal/bacterial ratios was done according to Frostegård and Bååth61.

Statistical analyses

Graphs were generated using Microsoft Excel or ggplot262 and tidyverse63 in RStudio (version 4.3.3). Statistical analyses were performed using the base statistics software in RStudio or vegan64. In brief, one-way analysis of variance (ANOVA) tests were performed to determine the effect of O2 legacy treatment on the indicators of denitrification progression (Table 1), PLFA profiles, and gene read abundances. Global ANOVA tests were performed to determine the main O2 legacy treatment effect, main time effect, and interaction effect between treatment and time on NO3− and NO2− concentrations and transcript read abundances. When differences occurred (α = 0.05), mean pairwise comparisons were performed using Tukey’s test (HSD) for the post hoc analysis of the ANOVA models. Normality was tested using homoscedasticity plots and the Shapiro-Wilks test of the residuals. The estimated effect size of the test statistics was determined using Eta-squared (95% CI) and the effectsize65 package in RStudio. The MASH-based distance estimations were analyzed using a permutational multivariate analysis of variance (PERMANOVA, permutations = 999), where multilevel pairwise comparisons were performed using the pairwise Adonis wrapper function from Vegan. Supplementary Table 6 presents the F-values, P-values, effect size, and degrees of freedom generated from the statistical analyses used in this study.

Reporting summary

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