Introduction

Nitrous oxide (N2O) is a potent climate gas, with ~273 times the global warming potential of carbon dioxide (CO2)1 and strong ozone-depleting properties2. The atmospheric concentration of N2O continues to rise through the use of nitrogen-based fertilisers, fossil fuel combustion, biomass burning and sewage discharge3 and has already increased by approximately 20% since 17504. Not surprisingly, given its atmospheric potency, research to date has focused on these N2O sources with N2O sinks being relatively understudied5,6,7. The few studies reporting on both N2O sources and sinks8,9,10,11 often simply document the sinks as concentrations below that expected for water (marine or freshwater) at equilibrium with the atmosphere and the true mechanism behind this N2O deficit remains largely unknown.

In terrestrial and aquatic environments, N2O can be produced from both microbial nitrification12 either via hydroxylamine oxidation (NH4+ → NH2OH → N2O), or hybrid formation (NO2− + NH2OH → N2O)13, and incomplete denitrification (NO3− → NO2− → NO→ N2O[\(\to\) N2])14. Where oxygen is limiting and/or completely absent, N2O can be further reduced to N2 in the last step of microbial denitrification (N2O→N2) that is typically mediated by facultative anaerobic bacteria14,15. As such, any undersaturation – indicating a sink for N2O – as observed in some waters has routinely been attributed to that last step in denitrification.

However, such N2O undersaturation has typically been reported in well-oxygenated, shallow freshwaters8,9,10,16,17,18,19,20,21 (down to 13% of air equilibration, typically ~70–100%) and surface-ocean-waters5,11,22,23,24,25,26,27,28 (down to 34%, typically ~90%) where canonical denitrification is unlikely to explain any undersaturation in N2O. While N2O consumption by denitrification has been reported in both anoxic and oxic-to-anoxic transitioning waters in the Eastern Tropical North Pacific6, the reasons for N2O undersaturation in general remain poorly understood, with many instances of N2O undersaturation remaining unaccounted for8,9,10,19,20,21 or simply being dismissed as analytical artifacts24,29. Further, as N2O sources generally increase at higher concentrations of ammonium and nitrate (i.e., fixed, bio-available N)8,25, any potential undersaturation in N2O could be masked by stronger production of N2O from nitrification and denitrification. This might explain why many accounts of N2O undersaturation have been reported in N limited environments5,9,19,21,22.

In recent years, evidence has been presented for an additional pathway to denitrification for N2O reduction, namely – N2O dependent N fixation – that has been reported for pure cultures of marine Trichodesmium and Crocosphaera5. N2O fixation has also been reported in the surface waters of the Eastern Tropical South Pacific5,23, where the measured N2O fixation activity could contribute some (0.2 – 60%) of the total N2O reduction5. As long ago as 1954, it was shown30 that 15N2O could be assimilated by soybean root nodules with activity comparable to 15N2 assimilation. These findings show that N2O fixation (e.g. N2O→NH4+) represents an alternative N2O reduction pathway to the terminal step in denitrification (N2O→N2) that may explain some of the undersaturation reported for N2O. Within the widespread accounts of N2O undersaturation found in well-oxygenated waters, only a few studies mentioned the possibility of N2O fixation5,22,23 and it is not widely acknowledged.

Primary production and N2 fixation are tightly coupled in N-limited ecosystems31. Some early studies (1952–1986) showed that N2O is a competitive inhibitor for N2 fixation and it could also be a substrate for the enzymatic nitrogenase complex32,33,34,35, indicating that N2O fixation (e.g. N2O→NH4+) may be related to N2 fixation (N2→NH4+). Further, with its N≡N bond N2 fixation has a high activation energy (~1 to 2 eV vs. 0.65 eV and 0.32 eV for respiration and photosynthesis, respectively)36,37 which makes fixing N2 in the cold energetically unfavourable. As a consequence, the abundance of diazotrophs has been shown to decrease as temperatures decline36. In contrast, the energy required to fix N2O (Eq. 1, ΔG defined for freshwater at 10 °C, see Supplementary Text 1) is lower than that for N2 (Eq. 2) and being able to fix N2O could confer an ecological advantage to some microbes either in the cold or when resources (light or reduced substrates) in general are limiting.

$$\begin{array}{ccc}{0.5{{{{{\rm{N}}}}}}}_{2}{{{{{\rm{O}}}}}}+{1.5{{{{{\rm{H}}}}}}}_{2}{{{{{\rm{O}}}}}}\to {1{{{{{\rm{NH}}}}}}}_{3}+{1{{{{{\rm{O}}}}}}}_{2} & \Delta G=+ 247{{{{{\rm{kJ}}}}}} & ({{{{{\rm{per}}}}}}\,{{{{{{\rm{NH}}}}}}}_{3})\end{array}$$
(1)
$$\begin{array}{ccc}0.5{{{{{{\rm{N}}}}}}}_{2}+1.5{{{{{{\rm{H}}}}}}}_{2}{{{{{\rm{O}}}}}}\to 1{{{{{{\rm{NH}}}}}}}_{3}+0.75{{{{{{\rm{O}}}}}}}_{2} & \Delta G=+ 291{{{{{\rm{kJ}}}}}} & ({{{{{\rm{per}}}}}}\,{{{{{{\rm{NH}}}}}}}_{3})\end{array}$$
(2)

While the ~18% energy saving for fixing N2O versus N2 is seemingly modest, it is comparable to the recognised 21% saving delivered by assimilating NO3− (Eq. 3) rather than fixing N2 (ref. 38) (see Supplementary Text 1).

$$\begin{array}{ccc}{{{{{{\rm{NO}}}}}}}_{3}^{-}+3{{{{{{\rm{H}}}}}}}^{+}+2{{{{{{\rm{e}}}}}}}^{-}\to 1{{{{{{\rm{NH}}}}}}}_{3}+1.5{{{{{{\rm{O}}}}}}}_{2} & \Delta G=+ 241{{{{{\rm{kJ}}}}}} & ({{{{{\rm{per}}}}}}\,{{{{{{\rm{NH}}}}}}}_{3})\end{array}$$
(3)

With both the last step in denitrification (N2O→N2) and N2O fixation (N2O→NH4+) providing sinks for N2O it is ecologically important to distinguish between these two parts of total N2O reduction. Further, any genuine direct N2O fixation (N2O→NH4+) needs to be distinguished from indirect N2O fixation i.e., that which could occur after the initial reduction of N2O to N2 (N2O→N2→NH4+). Despite the few studies5,23,30 documenting N2O fixation so far, to the best of our knowledge, there has been no characterisation of N2O fixation in relation to canonical N2 fixation through the dual use of 15N2O and 15N2 in natural communities.

In 2005, we set up 20 experimental ponds (each with 1 m3 water volume, 0.5 m depth) in East Stoke, Dorset, UK, to experimentally study the whole-ecosystem effects of climate warming39,40,41. Here, however, we exploited the fact that our experimental ponds are also N-limited41, being fed only by rain water, to characterise any potential N2O fixation in a controlled, experimental system. Despite being artificial, the ponds have well-established freshwater ecosystems39,40,41,42 with diverse cyanobacteria communities42, among which some Nostocales43 and Oscillatoriales44 are known to fix N2.

Here, we show that the ponds are undersaturated in both N2 and N2O and further hypothesise that the pond communities fix both gases to support primary production. Then, due to the different energy demands of N2 and N2O fixation, we hypothesise that the two processes will respond differently to temperature. We use incubations with pond biomass and 15N2 and 15N2O stable isotope techniques to quantify their fixation activity, distinguish direct from indirect N2O fixation and characterise the temperature dependence of each N-fixing process. Finally, with no known freshwater candidates for N2O fixation to date, we explore the recognised N2 fixing community in relation to N2O fixation. We ask whether: 1, is N2O fixation mediated by the total nitrogenase (nifH) community simply in relation to the relative availability of N2O to N2; or 2, is N2O fixation preferentially mediated by a subset of the nifH community?

Results

Contrasting seasonalities in undersaturation for N2 and N2O

Concentrations of dissolved N2O and N2 were both significantly below atmospheric equilibration (p < 0.001, Fig. 1a) and the ponds are sinks for both atmospheric N2O and N2. Overall, N2O was more under-saturated than N2 (p < 0.001, t = −17.5, d.f. = 240.6, two-sided, Fig. 1a), with a mean value of 79.1% ± 1.1% (mean ± s.e., as below) of air saturation compared to 98.5% ± 0.2% for N2. Furthermore, the seasonality in N2O saturation was far more pronounced than for N2 (Best fitting Generalised Additive Mixed Models, GAMMs, Supplementary Table 1), with a strong minimum for N2O in December and maximum saturation in summer (Fig. 1b). Conversely, N2 saturation peaked in winter and was lower in spring and summer (Fig. 1c).

Fig. 1: Seasonal and overall levels of N2O and N2 saturation in the ponds.
figure 1

a Box-whisker plots showing overall that the saturation of N2O was lower than that for N2 in the ponds (p < 0.001, two-sided, Supplementary Table 1). The dashed line denotes 100% atmospheric equilibrium for the gases. b, c show the saturation in N2O had a different seasonal pattern compared to N2 (Note the different scales on the y-axes). The solid lines in b and c represent the best fitting GAMM models (two-sided, Supplementary Table 1). d N2O saturation increased at higher temperatures while, in contrast, e N2 saturation declined. The lines in d and e are simple first-order linear regressions (two-sided). f Overall concentration of total inorganic nitrogen (TIN) in the ponds (n = 213 samples for 11 months in 20 ponds). g The ratio of TIN to soluble reactive phosphorous (SRP) in the ponds (n = 168 samples for 11 months for 20 ponds, SRP data omitted below the detection limit). The dashed line in g denotes the Redfield ratio of N to P of 16:1. Each box in a, f and g shows the 25th to 75th percentiles, horizontal lines the median, open circles denote outliers and whiskers extend to 1.5 times the interquartile range. n = 230 and n = 215 samples for N2O and N2 saturation, respectively, for 20 ponds, in 11 months from November 2019 to April 2022 (see Supplementary Table 2).

Full size image

The concentrations of dissolved inorganic nutrients (nitrite, NO2−; nitrate, NO3−; ammonium, NH4+, and soluble reactive phosphorus, SRP) were low in the ponds, with NO2−, NO3− and NH4+often at or below the limit of detection. The concentration of total inorganic nitrogen (TIN as the sum of NO2−, NO3− and NH4+) was 0.85 ± 0.03 µM across all sampling months (Fig. 1f). SRP concentrations were 0.14 ± 0.01 µM, on average, and, at 5 to 1, the median N to P ratio was markedly lower than Redfield45 (16 to 1), indicating primary production in the ponds to be N limited (Fig. 1g). As the ponds were N-limited, primary production must be sustained largely by N fixation (and any unknown atmospheric N deposition), which may have resulted in the undersaturation of N2 and N2O in the ponds.

Interestingly, N2O saturation increased with water temperature (p < 0.001, Fig. 1d), suggesting relatively higher net reduction of N2O in the cold (see Supplementary Fig. 1 for concentration data). Whereas N2 saturation showed the opposite pattern, with relatively more net N2 reduction at higher temperatures (p < 0.001, Fig. 1e) in spring and summer. Moreover, the saturation of dissolved O2 in the ponds (at the same depth where the samples for N2 and N2O were collected) was generally around air-equilibration (104.8% ± 1.8%, median 99.6%), with N2 saturation decreasing at higher O2 saturations, while N2O saturation increased with higher O2 saturation (Supplementary Fig. 2). Oxygen saturation was positively correlated with temperature (Supplementary Fig. 2c), probably due to higher temperatures in spring and summer promoting primary production. Therefore, maximum N2 undersaturation was probably related to higher primary production in spring and summer40. The negative and positive correlations between N2 or N2O and O2 respectively, indicated different controls for the reduction of N2 and N2O.

N2O and N2 fixation by biomass in the ponds

To rationalise the undersaturation in both N2O and N2 in our ponds, we measured fixation of either 15N2O or 15N2 (at a range of temperatures, see below) by biomass collected from the ponds (Supplementary Fig. 3). We found 15N assimilated into biomass from either 15N2 or 15N2O in the majority of our incubations (87%, 572 out of 658 incubations, Fig. 2a), with higher rates of 15N assimilation with 15N2 than for 15N2O with both floating and benthic biomass (p < 0.001, t = 6.5, d.f. = 369.7, two-sided, Fig. 2b). On average, 11.5 ± 0.9 and 5.3 ± 0.3 nmol g−1 d−1 (mean ± s.e.) of 15N were assimilated into biomass with either 15N2 or 15N2O, respectively (Fig. 2b). The rate of 15N2 assimilation was higher in the floating than the benthic biomass (p = 0.001, t = 3.3, d.f. = 179.4, two-sided), while 15N2O assimilation was consistent between the two biomass types (p = 0.24, t = −1.2, d.f. = 319.1, two-sided).

Fig. 2: Evidence for the direct assimilation of 15N into biomass from either 15N2 or 15N2O.
figure 2

15N assimilated into biomass from either 15N2 or 15N2O in different months. a Each box shows the 25th to 75th percentiles, horizontal lines the median, open circles denote outliers, and whiskers extend to 1.5 times the interquartile range. b Rates of 15N2 assimilation were higher than for 15N2O assimilation in both biomass types. The red dashed line marks the upper threshold for indirect 15N assimilation if 15N2O was first reduced to 15N2 prior to assimilation (see Table 1), red shaded area 95% C.I. Data plotted are means ± s.e. Yellow: difference between floating and benthic biomass with 15N2 (p = 0.001, two-sided), black: difference between 15N2 and 15N2O in floating or benthic biomass (p < 0.001, two-sided). To exclude extreme outliers the data plotted in a and b are 95% of the dataset (2.5–97.5% percentiles). Data in a and b were collected in 5 calendar months for 2 biomass types from 8 to 10 ponds per sampling date (n = 303 and n = 322 incubations for 15N2 and 15N2O treatments, respectively). c No significant production of 15N2 from 15N2O in the floating biomass, but d significant production of 15N2 in the benthic biomass (p = 0.04, two-sided). The horizontal line is the mean and shaded area 95% C.I. (n = 351 incubations for both controls and 15N2O treatments).

Full size image

To distinguish direct N2O fixation (N2O → NH4+, Eq. 1) from indirect fixation i.e., that after an initial reduction of N2O to N2 ([N2O → N2] N2 → NH4+, Eq. 2) through denitrification, we first checked for any production of 15N2 from 15N2O. Overall, the production of 15N2 in the 15N2O treatments was not significant in the floating biomass incubations (Fig. 2c), though 15N assimilation from 15N2O was significant (Fig. 2b). In contrast, 30% of the benthic incubations showed measurable 15N2 production (p = 0.04, two-sided, Fig. 2c and Supplementary Fig. 5) but with comparable 15N assimilation from 15N2O (Fig. 2b). Some denitrification is expected given the sediments recognised capacity to consume oxygen39, and our 12 h/12 h light/dark incubation-cycle generated oxygen minima overnight that likely facilitated the reduction of N2O to N2 via denitrification.

In addition, we also compared rates of assimilation against a theoretical upper threshold for indirect assimilation of 15N2O after reduction to 15N2 (Table 1 and Fig. 2b). For example, any 15N2 from the reduction of 15N2O would be assimilated in proportion to the 15N-labelling of the total N2 pool, which would be predominantly ambient 14N2 (Table 1). As a result, any indirect assimilation of 15N from 15N2O should have been ~14-fold lower than what we measured in the incubations where we added 15N2 directly e.g. 0.8 nmol N g−1 d−1 vs. 11.5 nmol N g−1 d−1 (Table 1). In contrast, we measured far higher rates of 5.3 nmol N g−1 d−1 with 15N2O, compared to the upper threshold of 0.8 nmol N g−1 d−1, on average (0.69 to 0.92 nmol N g−1 d−1, 95% C.I., Fig. 2b). Such disproportionately high activity suggests direct assimilation of 15N from 15N2O into biomass in our freshwater ponds.

Table 1 Rationalising N2O assimilation as direct N2O fixation
Full size table

Here we added 15N2O to our incubations at concentrations many times higher than atmospheric equilibration (9 µM vs. 0.01 μM) and our rates of 15N2O assimilation are likely upper-potentials. We also characterised the kinetic effect of N2O concentration on total N2O reduction from 9.2 nM (atmospheric equilibration) to 20,000 nM (Supplementary Fig. 6), which enabled us to estimate N2O reduction by biomass at in situ concentrations in the ponds. We then scaled these in situ N2O reduction estimates by the amount of benthic biomass in the ponds and compared them to our estimates of N2O flux into the ponds calculated using our measurements of N2O saturation (Fig. 1 and see Supplementary Text 2). Accordingly, we estimated in situ N2O reduction by the benthic biomass to be −0.75 µmol N2O m−2 d−1 (Supplementary Text 2) which is equivalent to 56% of the N2O flux into the ponds of −1.33 µmol N2O m−2 d−1, on average (range of −3.65 to 0.02 µmol N2O m−2 d−1, including low emissions to the atmosphere in summer). The remaining ~44% of the N2O flux is probably driven by microbes associated with the floating biomass (Fig. 2b) or free-living in the water column42 and we are confident that our laboratory biomass incubations can rationalise the undersaturation in N2O we measured in our ponds. In addition, N2 flux into the ponds was −3,934 µmol N2 m−2 d−1, on average (Supplementary Text 2).

Multiple fates for total 15N2O reduction

Apart from 15N2O being assimilated into biomass and the fraction reduced to 15N2 (above), some fixed 15N2O as 15NH4+ could potentially “leak” into the pond-water medium to, in turn, be nitrified to 15NO2− and 15NO3− (together 15NOx−) – all of which comprise total 15N2O reduction. We characterised total 15N2O reduction and the proportions of the different end-products and found significant 15N2O reduction in the majority (289 out of 372 incubations, 78%) of our incubations enriched with 15N2O. The mean rate of total 15N2O reduction was 364 ± 27 nmol N g−1 d−1, with the highest rate of total 15N2O reduction occurring in December for both floating (850 ± 178 nmol N g−1 d−1) and benthic biomass (784 ± 158 nmol N g−1 d−1).

To compare summer to winter, we pooled data from November, February and December, for winter, and August and September for summer. Total 15N2O reduction was highest in winter at 507 ± 49 nmol N g−1 d−1, compared to 237 ± 21 nmol N g−1 d−1 in summer, on average (p < 0.001, t = −5.1, d.f. = 213.4, two-sided) in both floating (p < 0.001, t = −5, d.f. = 68.6, two-sided) and benthic (p = 0.02, t = −2.3, d.f. = 148.5, two-sided) biomass (Fig. 3a). The patterns in total 15N2O reduction measured in the incubations agreed with the seasonal pattern of N2O saturation in the ponds (Fig. 1): overall, N2O was consumed in both seasons and the ponds were net sinks for N2O, with higher N2O reduction in winter, corresponding to greater undersaturation in N2O in winter.

Fig. 3: The fate of total 15N2O reduction in biomass.
figure 3

a Total 15N2O reduction was highest in winter compared to summer in both floating (p < 0.001, two-sided) and benthic biomass (p = 0.02, two-sided). Data plotted are means ± s.e. from 95% (2.5% to 97.5% percentiles) of the dataset for summer and winter where n = 180 and n = 160 incubations in summer and winter, respectively (two months for summer, three months for winter). b Production of dissolved inorganic nitrogen in 15N2O incubations. TIN: Total inorganic nitrogen. n = 6 incubations for biomass from 6 ponds. Each box shows the 25th to 75th percentiles, horizontal lines the median, and whiskers extend 1.5 times the interquartile range. c Distribution of the ratio of 15NOx− production to total 15N2O reduction, dashed line is the median. Data in c are from December, 2020, when the highest rate of total 15N2O reduction was measured (n = 47 incubation bottles, 5 temperatures with 2 biomass types). d Simplified diagram showing possible pathways for N2O reduction in relation to canonical N2 fixation.

Full size image

To test whether N2O initially fixed intracellularly as NH4+ (Eq. 1) could leak into the water i.e., to be available to the wider ecosystem, we performed additional incubations with samples for nutrient measurements (without formaldehyde, see methods). Although the concentration of NH4+ was often below the limit of detection for the colorimetric assay (~0.2 μM), the stronger signal for NH4+ with N2O (p = 0.03) indicated some N2O fixed as NH4+ could “leak”. The concentration of total inorganic nitrogen (TIN) was, on average, 0.32 μM higher in incubations enriched with N2O than the controls (p = 0.001, Fig. 3b). We also characterised the production of 15NOx− from 15N2O in December, 2020 (winter), when rates of total 15N2O reduction were highest. The rate of 15NOx− production was 280 ± 46 nmol g−1 d−1 (mean ± s.e.), which accounted for 11.7% (median) of total 15N2O reduction (Fig. 3c). Together, these results show that some N2O fixed as NH4+ can be lost to the water and further oxidised to NOx− through nitrification both of which could be assimilated into PON by the wider community (Fig. 3d).

The temperature dependence of N2O fixation

As seasonal changes in temperature drove contrasting patterns in N2O and N2 saturation, we characterised the effect of temperature on N2O and N2 reduction by incubating biomass from the ponds at temperatures from 6°C to 25°C. Assimilation of 15N from 15N2 increased at higher temperatures (p = 0.005, t = 2.8, d.f. = 301, Fig. 4a), with an estimated Q10 of 1.38. In contrast, assimilation of 15N from 15N2O was consistent across all temperatures with no discernible temperature sensitivity. The large variance in Fig. 4a may in part be due to simply normalising the 15N assimilation data to a unit of dry biomass in each incubation, whereas the communities responsible for N2 or N2O assimilation could be heterogeneous in the biomass samples and across different months of the year. In addition, rates of 15NOx− production from 15N2O were also consistent across incubation temperatures (Fig. 4b), which, again, suggested that N2O fixation is not sensitive to temperature (i.e., Fig. 4a, b).

Fig. 4: Temperature sensitivity of 15N assimilation from 15N2 and 15N2O and 15NOx− production from 15N2O.
figure 4

a Temperature sensitivities for 15N assimilation from 15N2 and 15N2O were different (p = 0.025, two-sided), increasing at higher temperatures for 15N2 (slope: 0.35, 95% CI: 0.12–0.5) but 15N2O. As the data were skewed, we used median regression models to minimise bias from outliers. The regression in a, uses the whole dataset but only 95% of the dataset (2.5% to 97.5% percentiles) are presented (n = 303 and n = 322 for 15N2 and 15N2O, respectively, for 5 months, 2 biomass types). b similarly, 15NOx− production from 15N2O was also invariant to temperature. Data in b are from December, 2020, when total 15N2O reduction was highest (n = 47 incubations, 5 temperatures, 2 biomass types), blue line is a first-order linear regression, mean ± s.e. As the temperature sensitivity of 15N assimilation and 15NOx− production was consistent between floating and benthic biomass, the data in a and b have been pooled for both biomass types.

Full size image

nifH communities in relation to N2O reduction

The fact that here N2O fixation appears less sensitive to temperature than N2 fixation supported our hypothesis that fixing N2O is less energy demanding than fixing N2. Here we aimed to address our question of whether N2O fixation is mediated by the whole N2 fixing community or a sub-set, using a long-term incubation with N2O-enriched biomass.

N2O reduction was most rapid during the first 3 days of the incubation and started to decline after the increase in total inorganic nitrogen (TIN, sum of NO3−, NO2− and NH4+) from 0.44 µM to 0.76 µM and (Fig. 5a and Supplementary Fig. 7a). We terminated the incubation after 25 days when oxygen production from photosynthesis started to decline (Fig. 5a and Supplementary Fig. 8b) and characterised the abundance and structure of the nifH community.

Fig. 5: Biomass nifH community in relation to N2O reduction.
figure 5

a Biomass incubated for 25 days reduced 76 nmol N2O-N g−1 (dry weight) on average (n = 40 incubations for biomass enriched with N2O) with activity peaking before inorganic nitrogen accumulated. Data plotted are means ± s.e. b Redundancy analysis (RDA) revealed positive correlations between the initial rates of N2O reduction and the relative abundance of 22 nifH OTUs, including 15 Cyanobacterial OTUs and 7 Proteobacterial OTUs (arrows in green). c Heat-map (white to dark red) of the relative abundance of the 22 nifH OTUs (in green), identified in b, in samples (columns, n = 10) in rank order of increasing rate of N2O reduction. n = 10 in b and c. N2O reduction is presented as a black arrow in b and as a grey ascending triangle in c.

Full size image

We first tested whether N2O reduction was related to the abundance of the whole nifH community (copy numbers of nifH per g wet biomass) but found no relationship (p = 0.21, F = 1.74, d.f.=18). We then tested whether N2O reduction was related to a sub-community by looking for any changes in diversity or composition of the nifH community over the 25-day incubation. Our primers amplified 894 well-represented OTUs ( > 20 reads in at least 3 samples) of which only 227 were identified as nifH OTUs (see Methods and Supplementary Figs. 9 and 10). However, neither the diversity (p = 0.81, t = 0.24, d.f. = 17.5, two-sided t-statistic tested on the means of Shannon index) nor the composition (p = 0.99, PERMANOVA, see also Supplementary Fig. 11 for unchanged nifH community composition at 3, 10 and 25 days) of the overall nifH community changed significantly during the 25-day incubation.

As an alternative, we used redundancy analysis (RDA) to ordinate the relative abundances of nifH OTUs and the initial rates (i.e., in 10 samples up to day 3) of N2O reduction to identify any likely N2O fixing candidates (Fig. 5b). Since N2O reduction may be mediated by a subset of the whole nifH community, we relaxed our definition of a well-represented OTU to include <20 reads in at least 3 samples (see Supplementary Fig. 9) which retained 72 out of 227 nifH OTUs. Of those 72 OTUs, the relative abundance of 22 were positively correlated with N2O reduction i.e., in the ordination their arrows pointed in a similar direction to the arrow for N2O reduction. The positive correlations for the 22 OTUs, including 15 Cyanobacteria and 7 Proteobacteria, were further explored by visualising their relative abundance in each biomass sample in rank order of increasing rate of N2O reduction (Fig. 5c). Among the 15 Cyanobacterial OTUs, Pegethrix-like OTU392 and OTU394 (100% identical protein sequence to Pegethrix) and the Fischerella-like (>99% identical) OTU396 appeared to not only be more common, but they were also more strongly correlated with the initial rates of N2O reduction. While OTU412 and OTU389 were also identical to Pegethrix they were either relatively rare or less-well correlated, respectively. Despite the two Methylomonas-like (>99 % identical) Proteobacterial OTU444 and OTU462 being less common than the Cyanobacterial candidates, exisiting in only four samples, their higher relative abundances coincided with higher rates of N2O reduction. Moreover, combinations of the five strongest (OTUs 392, 394, 396, 444, 462) N2O fixing candidates were not only present in the ten samples used to determine the initial rates of N2O reduction but all 40 samples enriched with N2O for our 25-day incubation (Supplementary Fig. 9).

Discussion

In ecosystems with limited fixed nitrogen (e.g. inorganic NO2−, NO3−, NH4+), primary production is tightly coupled to N-fixation – typically recognised to be N2 gas (Eq. 2). The undersaturation in N2O reported here means that the reduction of N2O was greater than its rate of delivery either from the atmosphere or biological sources in the ponds, which shows that these N-limited ponds were overall sinks for N2O, including direct N2O fixation (Eq. 1).

Others have argued for direct N2O fixation on the premise that if N2 production was not detected in incubations with N2O, then N2O fixation was direct5,30. Here, besides not detecting 15N2 production in 76% of our incubations (Fig. 2b), our disproportionate fixation of N from 15N2O relative to that measured with 15N2 provides more substantive evidence for direct N2O fixation (Table 1). Direct N2O fixation represents an alternative to the only widely recognised sink for N2O – namely denitrification6,29. In addition, our estimation of in situ N2O fixation helps to rationalise the undersaturation and resultant flux of N2O into our ponds (see Supplementary text 2). Further, as the scale of N2O undersaturation in our ponds (Fig. 1a) is in line with many other studies also reporting undersaturation in N2O in freshwaters (typically ~70–100%)10,16,17,18,21, this indicates that direct N2O fixation could explain the unaccounted for N2O undersaturation in many freshwaters8,9,10,16,17,18,19,20,21.

We can see similar seasonal trends to what we report here in previous accounts of N2O undersaturation. For example, boreal lakes, ponds and rivers show undersaturation in N2O which is strongest at coldest temperatures19. In Boreal peatlands, N2O was undersaturated mostly in spring, increasing to maximum oversaturation in summer, then decreasing to near equilibrium in autumn46. From the same study, soils acted as net N2O sources at higher temperatures, while most N2O sinks occurred below 13 °C46. In the surface waters of the Baltic Sea, N2O was most undersaturated in winter (December), but was oversaturated in summer and autumn11. However, these studies generally lacked a clear explanation for the occurrence and the temperature dependence of N2O undersaturation, whereas we now offer an explanation.

Our findings demonstrate different temperature dependencies for N2 and N2O fixation. This difference in N2 versus N2O is supported not only by the opposing seasonal patterns in N2 and N2O saturation in our ponds, but also by the experimentally determined different temperature sensitivities for the assimilation of N2 and N2O by biomass in our incubations. Moreover, the results from our incubations support the seasonal patterns in N2 and N2O saturation in the ponds – with the higher rates of N2O reduction in incubations in winter than in summer, matching the stronger N2O undersaturation in the ponds in winter, and the elevated temperature effect on N2 assimilation agrees with that for N2 saturation in our ponds. Here the apparent lack of temperature sensitivity of N2O fixation (Fig. 4) suggests that the N-fixing communities may be strongly adapted to substrate limitation (Supplementary Fig. 6), with dissolved N2O typically at ~10 nM in the ponds compared to ~490 µM for N2. This strong kinetic effect of substrate availability on N2O fixation has also been reported in incubations with surface seawater23 and pure culture of marine cyanobacteria Trichodesmium sp5. Similar stronger limitation of activity by substrate over temperature is also recognised in other autotrophs such as the methane oxidising methanotrophs47,48.

The contrasting temperature sensitivies of N2O and N2 fixation are probably associated with the energetic advantage of using N2O instead of N2 as a N-substrate for N-fixation49. Here we compiled data for studies measuring N2 fixation in both aquatic and terrestrial ecosystems (Supplementary Fig. 12 and the references cited therein) which clearly shows N2 fixation increases at higher temperatures. On the other hand, as dissociating the N bond in N2O only requires about half of the energy compared to N2 (refs. 49,50), N2O should be easier to fix at colder temperatures and a higher proportion of total N fixation could be dependent on N2O in the cold. For example, with our pond biomass the fraction of total N-fixation coupled to N2O at 6°C was 26% higher than that at 25°C (Fig. 4a). This energy saving offered by fixing N2O in the cold might explain why N2O undersaturation in our ponds was strongest during the colder months and may also explain the undersaturation reported in cold, boreal environments and Baltic Sea11,19,46.

In addition, in northern latitudes cold temperatures typically occur alongside lower light and the ~18% energy saving from fixing N2O (Eq. 1) compared to N2 (Eq. 2) could provide an over-wintering strategy for a subset of the photosynthetic community. Besides photosynthesis, some chemosynthetic microbes could also benefit. For example, some Methylomonas spp. are known to fix N2 and were identified here amongst our potential N2O fixing candiates. Some 32% of the energy yielded from oxidising CH4 to CO2 can be expended on assimilating CH4 into biomass (equation S6, Supplementary Text 1) with another 56% needed to fix N2 compared to 47% for N2O. Given that methane concentrations are lowest in our ponds in winter39 - and more widely methane production is known to be tightly coupled to primary production in spring and summer51 - fixing N2O could offer an advantage over N2 when resources are limited.

The concept that N2 fixation in general is routinely supressed by inorganic N (>~1 µM) has been revised52. For example, while there are numerous examples of N2-fixing activity being suppressed by inorganic nitrogen in pure cultures of Trichodesmium spp., others have reported its activity in euphotic ocean waters (whole water and Trichodesmium spp. isolates) with ~5 µM to 21 µM NO3− (refs. 52,53), which may indicate short-term tolerance to NO3−. In contrast, the N-fixing community in our ponds is exposed to chronic, year round inorganic nitrogen starvation (i.e., <1 µM). Our 25-day incubation showed N2O reduction declined after an increase in inorganic nitrogen (NO3− and NH4+) to ~0.8 µM (Fig. 5a), indicating a low-threshold concentration for inorganic nitrogen that apparently limits N2O fixation. Such a threshold also reflects the co-occurance of N2O undersaturation (Fig. 1a) with <1 µM inorganic nitrogen throughout the year in our ponds (Fig. 1f). In contrast, N-rich ecosystems generally act as N2O sources8,25, while N2O sinks - mediated through N2O fixation – are likely to be found in pristine, cold ecosystems19,21,22,26,27,28.

To date, it is not clear which microorganisms are responsible for N2O fixation in natural ecosystems. A few studies have reported N2O fixation in sea water5,23 and soybean root nodules30, but only one study, on pure cultures of the marine cyanobacteria Trichodesmium sp. and Crocosphaera sp., has related the nifH gene to N2O fixation5. Since being set up in 2005, our nitrogen-limited ponds have developed diverse diazotroph communities42, comparable to those in estuaries54, freshwater55 and seawater56. Here we set out to link cause to effect by attempting to enrich N2O fixing candidates but failed to detect any changes in the total nifH community after 25 days enrichment with N2O (Supplementary Fig. 11). This might have been due to the decline in rate of N2O reduction, as a result of the parallel accumulation in inorganic nitrogen, or longer incubations being needed to detect any change in the total nifH community that is likely slow-growing. In our ordination analysis, five OTUs emerged as the strongest potential candidates for N2O fixation i.e., their relative abundance is at least correlated with N2O reduction activity. Of these, one is identical to Fischerella, a common diazotroph in freshwater55 and seawater57 and also identified in seawater undersaturated in N2O23 and another two being identical to Pegethrix58, a newly identified genera of filamentous Cyanobacteria. In addition, proteobacterial methanotrophic diazotrophs have been related to N-fixing activity in N-limited freshwaters59, and relatives of our two Methylomonas-like candidates are known to grow on N2 as their sole nitrogen source60. Whether or not our candidate N2O-fixers are responsible for the widely reported undersaturation in N2O in natural waters needs characterising directly but our first attempt here at least suggests N2O fixation is likely mediated by a subset of diazotrophic communities.

In addition, our work shows that N2O fixation can occur in an abundance of N2 i.e., against the high natural N2 background. This indicates that N2O fixation could happen in natural ecosystems replete in N2 and provides further insight into the communities responsible for N2O fixation. For example, nifH communities could fix N2 and N2O randomly, with the ratio of N2O to N2 fixation being simply proportional to the relative availability of N2O to N2. However, the distinct seasonal patterns we measured for N2 and N2O undersaturation (Fig. 1), coupled to disproportionate rates of N2O fixation (Fig. 2) and higher proportion of N2O fixation at colder temperatures (Fig. 4a) - all indicate that a specialised subset of the nifH community (Fig. 5) likely favoured N2O over N2 at colder temperatures in support of our nifH ordination analysis.

To put our estimates of N fixation into an ecological context, we compared estimates of the N2 flux (Supplementary Text 2) with former estimates of gross primary production (GPP) in the ponds40. For example, the average net N2 flux into the ponds was 3934 µmol N2 m−2 d−1, which, assuming Redfield ratios of 106:16 for C:N, could sustain GPP of 52,126 µmol C m−2 d−1 and which is comparable to GPP measured previously of 51,488 to 70,792 µmol C m−2 d−1 (ref. 40). Moreover, the seasonal dynamics in N2 flux in our study also matched that of GPP reported previously40, with both peaking in the summer (Fig. 1b). In contrast, while the flux of N2O is comparatively minor (~0.03 %) in terms of supporting GPP, it is great enough to maintain a strong sink for N2O.

To date, denitrification in either anoxic or oxic-to-anoxic transitioning waters is still the only widely recognised sink for N2O6,7,29. Here, as an alternative to denitrification, direct N2O fixation can rationalise the undersaturation in N2O in our ponds and could also explain the various unaccounted for N2O sinks – of similar magnitude – reported in natural, pristine waters10,17,19,21,22,28,46. As N2O undersaturation is favoured in the cold, rising temperatures could erode this natural sink for such a potent climate-gas.

Methods

Nutrient analysis

Temperature and O2 were measured in each pond using HQD portable metre (Hach). Samples of water for nutrient analysis were filtered (0.45 μm PES, 25 mm, pre-washed with deionized water) into Falcon tubes, kept cool and frozen at −20 °C back in the laboratory. Samples were thawed overnight at 4 °C and analysed by standard wet-chemistries for NO2−, NO3−, NH4+ and SRP on an autoanalyzer (San++, SKALAR Analytical B.V.)61 against certified reference materials, traceable to NIST. The limits of detection were 0.05 μM and 0.1 μM for NO2− and NOx− (NO2− + NO3−), respectively, 0.2 μM for NH4+ and 0.05 μM for SRP. SRP and total inorganic nitrogen (TIN, NO3− + NO2− + NH4+) below detection limits were omitted from any calculations.

Dissolved N2 and N2O in the ponds

For dissolved N2 and N2O analyses, water samples were taken carefully at mid-water-depth (~20 cm from the surface) from each pond using a 60 mL syringe and tubing. Five gas-tight vials (12 mL Exetainer, Labco, two vials for N2O and three for N2) for each pond were allowed to overflow three times, preserved with ZnCl2 (50 μL of 50% w/v)62, closed and mixed by hand. Extra pond water samples for reference N2 saturation were collected and preserved along with samples of air.

In a temperature-controlled laboratory at 22 °C, references for N2 saturation were prepared by equilibrating the pond water with the laboratory air, and then water and air samples were collected as for the field samples. Helium headspaces (2 mL 99.999% purity) were created in all sample and reference vials, followed by 24 h equilibration on an orbital shaker (SSL1, Stuart) in the same laboratory and all vials weighed to determine the exact volume of headspace and water.

For N2O, 100 μL of sample headspace was injected by an autosampler into a gas chromatograph fitted with a µECD (Agilent Technology UK Ltd., South Queensferry, UK) along with air samples using conditions described previously63. Calibration was performed against known concentrations of N2O from a NOAA standard (traceable to the SI unit “amount of substance fraction”) at 359.73 ppb or 120 ppb and 1.04 ppm and 96 ppm from BOC, UK, cross-calibrated to the NOAA standard. The precision for N2O concentration was 2% (coefficient of variation, n = 10). The total concentration of N2O in each vial was calculated using solubility coefficents64 as described before63 and the degree of over- or under-saturation calculated by comparison to the expected concentration of N2O for pond water at equilibrium with the atmosphere (see Supplementary Fig. 1).

For N2 analysis, we used the published N2:Ar method65. 100 μL of headspace was injected by an autosampler into an elemental analyzer, (Flash EA 1112 series, Thermo Finnigan) to remove O2 by the hot-copper reduction, before passing to a continuous flow isotope ratio mass spectrometer (CF-IRMS, Delta V Plus, Thermo Finnigan). Throughout each run, air samples were analysed to correct for drift and the expected concentrations for N2 or Ar in the headspace (Chs) calculated using the solubility of N2 and Ar for air at both field (Kfield) and laboratory (Klab) temperature66:

$${{{\mbox{C}}}}_{{{\mbox{hs}}}}\times {{{\mbox{V}}}}_{{{\mbox{hs}}}}+{{{\mbox{C}}}}_{{{\mbox{hs}}}}\times {K}_{{{{{{\rm{lab}}}}}}}\times {{{\mbox{V}}}}_{{{\mbox{aq}}}}={{{\mbox{C}}}}_{{{\mbox{field}}}}\times {{{{\mbox{K}}}}_{{{\mbox{field}}}}\times {{\mbox{V}}}}_{{{\mbox{aq}}}}$$
(4)

Where Vhs and Vaq are the volumes of headspace and water in a vial and Chs and Cfield the concentration of either gas in the headspace or field air, respectively. The saturation of N2 in the samples was then derived by comparing the measured to expected ratio of N2 to Ar in the samples to that in the pond water ref. 67:

$${{{\mbox{N}}}}_{2}{{\mbox{Saturation}}}(\%)={\left(\frac{{{{\mbox{N}}}}_{2}/{{\mbox{Ar}}}\, {{\mbox{measured}}}}{{{{\mbox{N}}}}_{2}/{{\mbox{Ar}}}\, {{\mbox{expected}}}}\right)}_{{{\mbox{Sample}}}}\bigg/{\left(\frac{{{{\mbox{N}}}}_{2}/{{\mbox{Ar}}}\, {{\mbox{measured}}}}{{{{\mbox{N}}}}_{2}/{{\mbox{Ar}}}\, {{\mbox{expected}}}}\right)}_{{{\mbox{reference}}}}\times 100$$
(5)

Precision for the ratio of N2 to Ar for triplicate reference water and air standards was 0.1% and 0.05% (coefficient of variation), respectively. We also tested the effect of calculating N2 saturation with different references, with the ratio for deionized versus pond water being 99.7% and 99.82%, on average (n = 20 and n = 8, respectively).

Biomass incubations to characterise N2 and N2O fixation

Two types of biomass were collected from the ponds for the routine (24-h) incubations (Supplementary Fig. 3) and see below for the 25-day incubation. Floating assemblages on the ponds, comprising Oedogonium spp. and microorganisms attached to the filaments (Supplementary Fig. 3b) and green or yellow benthos assemblages (Supplementary Fig. 3c), sampled avoiding the sandy sediments beneath. Samples of biomass were collected in sterile Falcon tubes (50 mL) and transported back to the laboratory in a cool box and stored overnight at 15 °C before preparing the incubations. Despite the ponds being low in dissolved inorganic nitrogen, we standardised nutrient concentrations in the incubations by using an artificial pond water medium devoid of fixed N comprising: CaCl2 (0.5 mM), KCl (1 mM), MgSO4 (0.25 mM), KHCO3 (0.7 mM) and NaHCO3 (0.5 mM) in deionized water. P was added to 0.08 μM of NaPO4, based on measured SRP concentrations in the ponds.

Incubations of biomass with 15N2 and 15N2O tracers

Floating or benthic biomass was weighed (~3 g wet weight) into 12 mL gas-tight vials and, for the 15N2O treatment, filled with oxygen-saturated artificial pond water and closed. 100 μL of water was replaced by 15N2O stock solution (see below) with a gastight syringe. For the 15N2 treatment, each vial was filled with 10 mL of oxygen-saturated artificial pond water and 2 ml of 15N2 stock and closed (see below). All incubations were prepared without a headspace to ensure 15N-substrate concentrations were the same under different temperatures. Parallel controls were prepared in the same way without either 15N-gas.

T0 (Time zero) vials were killed with 200 μL of 50% (w/v) formaldehyde immediately and the remainder incubated in temperature-controlled orbital incubators (SI500, Stuart at 50 cycles min−1) at 6, 10, 15, 10 and 25 °C on a 12 h light/12 h dark cycle for 24 h. Time final (Tf) samples were killed as above, brought to 22 °C, helium headspaces created in all vials, and all allowed to equilibrate for 24 h, as above.

Stock solutions for 15N2 and 15N2O additions to the biomass incubations

To avoid the recognised equilibration problems with the 15N2 “bubble method”, especially during short-term incubations68, we first made an aqueous 15N2 stock with the artificial pond water. 200 mL of artificial media were injected into a 0.5 L gas sampling bag along with 40 mL 15N2 gas (98% atom % 15N, Sigma-Aldrich) and allowed to equilibrate for 24 h while gently rocking. 15N2O stock solutions were prepared by replacing 3 mL of water with 15N2O gas (98% atom % 15N, Cambridge Isotope Laboratories, Inc.) in a 50 mL sealed serum bottle.

The solubility of N2 is low and to maximise 15N2 labelling we added a relatively large amount (~2 ml) of 15N2 stock solution to the 12 mL vials (~16% v/v). To keep the dissolved nutrients and gases at background levels in the 15N2 treatments -and the same as in the controls and 15N2O treatment- we used artificial freshwater medium instead of deionized water for preparing 15N2 stocks. In contrast to N2, N2O is highly soluble, with only ~100 μL of 15N2O stock being needed in each treatment (~0.8% v/v) to reach the comparable concentration of 15N-N2 addition (~10 μM). 15N2O and 15N2 stocks were prepared fresh before each experiment and their respective dosages tested by spiking controls.

Characterising total 15N2O reduction

The concentration of 15N2O in the samples was measured on a CF-IRMS (Delta V Plus, Thermo-Finnigan) with an automated trace gas pre-concentrator (PreCon, Thermo-Finnigan)63. A sub-sample from the headspace of each sample was transferred to a 12 mL air-filled gas-tight vial. The high 15N-labelling of 15N2O (on average, 97.7% of 46N2O) in the 15N2O treatment meant only a small aliquot of sample (10 μL) was needed to keep the signal within the measurable range of 46N2O. Mass-to-charge ratios were measured for m/z 44, 45 and 46 and the concentration of N2O determined by calibration against known amounts (0.02–2 nmol) of natural abundance N2O (96 ppm N2O standard, BOC, UK)63. Note, whether the mass spectrometer is calibrated with high purity 15N2O or natural abundance N2O, the signal-to-mole ratio is constant (Supplementary Fig. 13). Here the concentration of total 15N2O is expressed as 15N2O = 45N2O + 2 × 46N2O and the reduction of 15N2O calculated by subtracting 15N2O concentrations in Tf samples from T0 samples.

Characterising any dissimilatory reduction of 15N2O to 15N2

Any production of 15N2 in the 15N2O treatments was measured by the CF-IRMS (Delta V Plus, Thermo Finnigan), after bypassing the copper reduction step to avoid reduction of 15N2O to 15N2 (ref. 69). The concentration of total N2 was calculated from the solubility of N2 (ref. 66) and the signal of total N2 mole masses i.e., m/z 28, 29 and 30 in the samples and air standards70. Drift in m/z 30 was corrected by inserting air standards for every 10 samples. Changes in the concentration of 15N2 (Δ15N2, nmol N d−1) were calculated by the excess 15N2 in 15N2O treatments compared to the controls, where Δ15N2 = Δ29N2 + 2 x Δ30N2. The limit of detection for Δ15N2 in the incubations is ~0.14 μM.

Characterising assimilation of 15N2O or 15N2 into biomass

After all the gas measurements, samples were centrifuged and the supernatants filtered (as above). The remaining biomass was dried, homogenised and sub-samples weighed into tin caps (6 × 4 mm, Elemental Microanalysis) for elemental analysis as described previously41. The level of 15N enrichment in biomass incubated with either 15N2 or 15N2O was then calculated by the difference in excess 15N atom % relative to the controls, where excess 15N atom % is the difference in 15N atom % between T0 and Tf in the 24-h incubations:

$${}^{15}{{{{{\rm{N}}}}}}\,{{{{{\rm{enrichment}}}}}}={({{{{{\rm{excess}}}}}}{}^{15}{{{{{\rm{N}}}}}}\, {{{{{\rm{atom}}}}}}\%)}_{{{{{{\rm{Treatment}}}}}}}-{({{{{{\rm{excess}}}}}}{}^{15}{{{{{\rm{N}}}}}}\, {{{{{\rm{atom}}}}}}\%)}_{{{{{{\rm{Control}}}}}}}$$
(6)

Rates of 15N assimilation (nmol 15N g−1 day−1) by biomass into particulate organic nitrogen (PON) were calculated as:

$${}^{15}{{{{{\rm{N}}}}}}\,{{{{{\rm{assimilation}}}}}}\,{{{{{\rm{rate}}}}}}={{{{{\rm{PON}}}}}}\times {}^{15}{{{{{\rm{N}}}}}}\,{{{{{\rm{enrichment}}}}}}/(\varDelta {{{{{\rm{t}}}}}}\times {{{{{\rm{dw}}}}}})$$
(7)

Where PON is particulate organic nitrogen in a sample of biomass, \(\Delta {{\mbox{t}}}\) is the incubation time (24 h), and dw the dry weight (g).

Characterising 15N2O fixation

As total 15N2O reduction includes both assimilatory 15N2O fixation and dissimilatory 15N2O reduction to 15N2, total 15N2O fixation can be calculated by subtracting 15N2 production from total 15N2O reduction:

$${{{{{\rm{Total}}}}}}{}^{15}{{{{{\rm{N}}}}}}_{2}{{{{{\rm{O}}}}}}\,{{{{{\rm{fixation}}}}}}={{{{{\rm{Total}}}}}}{}^{15}{{{{{\rm{N}}}}}}_{2}{{{{{\rm{O}}}}}}\,{{{{{\rm{reduction}}}}}}-{}^{15}{{{{{\rm{N}}}}}}_{2}\, {{{{{\rm{production}}}}}}$$
(8)

Where total 15N2O fixation includes 15N2O assimilated into biomass, as well as any fixed 15N2O present in the pond water medium as dissolved inorganic nitrogen (15DIN, e.g., 15NH4+, 15NO2− and 15NO3−):

$${{{{{\rm{Total}}}}}}{}^{15}{{{{{\rm{N}}}}}}_{2}{{{{{\rm{O}}}}}}\,{{{{{\rm{fixation}}}}}}={}^{15}{{{{{\rm{N}}}}}}_{2}{{{{{\rm{O}}}}}}\,{{{{{\rm{assimilation}}}}}}+{}^{15}{{{{{\rm{D}}}}}}{{{{{\rm{IN}}}}}}\,{{{{{\rm{production}}}}}}$$
(9)

We characterised any 15DIN production coupled 15N2O fixation by measuring 15NOx− (i.e., 15NO3− + 15NO2−) with sulfamic acid62, testing for any effect of formaldehyde on the 15NOx− assay (Supplementary Fig. 14). Due to the high 15N2O background, it was not possible to measure any 15NH4+ from 15N2O using the sensitive sodium-azide assay as it converts 15NO2− to 15N2O. Further, as the formaldehyde preservative interferes with the colorimetric NH4+ assay, we did additional incubations in October 2021 without formaldehyde, following the exact incubation procedure described above. Here, samples for DIN were immediately centrifuged and frozen at −20 °C while parallel samples for gases were treated as above. Concentrations of DIN in controls and 15N2O treatments were measured by the automated wet-chemistry autoanalyzer (see ‘Nutrient analysis’ in Methods), while changes in N2O concentrations were measured by GC/µECD63.

nifH communities in relation to N2O reduction over a 25-day incubation

To characterise any nifH communities potentially involved in N2O fixation, we incubated floating biomass from the ponds with excess N2O for as long as possible, looking either for changes in the nifH community or relationships between particular nifH families and N2O reduction. Floating biomass was collected from 10 ponds in May 2021 (as above) and once back in the laboratory kept in a temperature-controlled room at 15°C overnight. The next day, 7 g wet biomass was transferred into 70 mL serum bottles (n = 80, 8 serum bottles per pond), filled with water and sealed. N2O stock solution (600 µL as above) was injected into half of the serum bottles, while venting water through a needle, to create an initial N2O concentration of ~10 µM and the remaining serum bottles left unamended as controls. Temperature-controlled incubations were carried out on a 12 h:12 h light/dark cycle as above. A total of 20 serum bottles (10 with N2O and 10 controls) were sacrificed after 0, 3 and 10 days of incubation, respectively, while the last 20 serum bottles were incubated until the daily maximum in oxygen started to decline. Oxygen was measured with optical sensors (OXSP5, FireSting®, Pyro Science GmbH, Germany) at ~2-hourly intervals after lights and the incubations terminated on day 25 when daily maximum oxygen started to decline (Supplementary Fig. 8).

Sub-samples of water were transferred from each serum bottle into a 3 ml gas-tight vial (Exetainers, Labco) after 0, 3, 10 and 25 days of incubation, fixed with 50 µl formaldehyde, sealed and stored at room temperature. After creating a helium headspace N2O concentrations were measured by GC/µECD (as above). The remaining water was filtered and frozen at −20 °C for later quantification of NO3−, NO2− and NH4+, as above. Biomass was frozen at −20 °C until DNA extraction (June 2021) from ~0.5 g of wet biomass (DNeasy PowerSoil kit, Qiagen) as per the manufacturer’s instructions.

nifH gene abundance (qPCR) and library preparation

Gene abundance of nifH was determined using qPCR with IGK3/DVV (forward, 5’-GCIWTHTAYGGIAARGGIGGIATHGGIAA-3’; reverse, 5’- ATIGCRAAICCICCRCAIACIACRTC-3’)71 using a CFX384 Touch Real-Time PCR (Bio-Rad) in 10 µL reactions containing 5 µL SensiFAST SYBR No-ROX mastermix (Meridian Bioscience), 0.8 µL of each primer (10 µM), 0.8 µL DNA template and 2.6 µL molecular biology quality water (MBQW). The qPCR programme was 98 °C (3 min) then 40 cycles of 98 °C (15 s), 58 °C (60 s), 72 °C (60 s). Standard curves (105 to 108 copies per µL) were prepared from plasmid DNA containing nifH and product specificity confirmed by endpoint melt curve analysis.

A three step PCR was used to prepare the nifH library72. nifH was amplified using IGK3/DVV in 10 µL of MyTaq Red Mix (Bioline), 0.8 µL of each primer (10 µM), 0.8 µL of DNA template and 7.6 µL MBGW on a T100 Cycler (Bio-Rad) at (1) 94 °C (5 min); (2) 36 cycles of 94 °C (30 s), 57 °C (45 s), 72 °C (30 s); (3) 72 °C (10 min). These PCR products were then re-amplified with IGK3/DVV appended with overhang MiSeq adaptors in 25 µL containing 12.5 µL of MyTaq Red Mix (Bioline), 1 µL of each primer (10 µM), 1 µL of amplicons from the first step as template and 9.5 µL of MBGW. The PCR programme was: (1) 94 °C (4 min); (2) 12 cycles of 94 °C (30 s), 57 °C (45 s), 72 °C (30 s); (3) 72 °C (7 min). PCR products were cleaned using AMPure XP beads and multiplexing barcodes added by the Index PCR in 25 µL containing 12.5 µL of MyTaq Red Mix (Bioline), 0.5 µL of each primer (10 µM), 0.5 µL of DNA template and 11 µL of MBQW at (1) 95 °C (3 min); (2) 8 cycles of 98 °C (20 s), 57 °C (15 s), 72 °C (15 s); (3) 72 °C (5 min). Final amplicons were quantified (Qubit 2.0 Fluorometer (Invitrogen)) and normalised to 4 nM (SequalPrep Normalization Plate Kit, Invitrogen), combined and sequenced (Illumina MiSeq, 300 base paired-ends).

Sequence processing pipeline and phylogenetic analysis

Paired-end de-multiplexed files were imported into QIIME2 (v.2021.11) on the Apocrita HPC facility at Queen Mary University of London73 (Supplementary Fig. 9) and processed using DADA2 to trim primers, remove low-quality sequences and chimeras74. Sequences were clustered into species-level OTUs at 95% similarity72, singletons and sequences >356 bp or <333 bp and low-abundance OTUs (<20 reads and in 3 samples or less) were removed. Amino acid sequences were aligned to known nifH and non-nifH references and a phylogenetic tree constructed using COBALT75. The primers IGK3/DVV can amplify non-nifH homologues including the chlorophyll synthesis genes BChL and ChlL and these were identified after translating the OTU sequences using “Translate” in MEGA (version 10.2.2). Translation initiation site adjustment and frameshifts were detected using blastp76. Amino acid sequences were aligned to known nifH and non-nifH references and a phylogenetic tree constructed using COBALT75. The non-nifH OTUs were identified using distinct non-conservative short sequence motifs and visualised using the iTOL tool77 that appeared as two separate clusters on the phylogenetic tree (see Supplementary Fig. 10). Approximately 82% of the sequences were non-nifH homologues, which is common when using general nifH- primers71. Non-nifH sequences were removed and q-PCR estimates of nifH gene abundances were corrected for the proportion of non-nifH sequences in each sample.

Statistical analysis

Statistical analysis and plotting were performed in R78 using RStudio (Version 1.3.1093). We used generalised additive mixed effects models (GAMMs)79 to characterise the seasonal patterns in N2 and N2O saturation, fitting sampling month as a fixed effect and each replicate pond as random effects. We included an interaction term for sampling month by gas (N2 or N2O) to explore any distinct seasonality in N2 and N2O saturation. Models were ranked by the small sample-size corrected Akaike Information Criterion (AICc) using the ‘MuMIn’ package (Supplementary Table 1).

Rate data for total 15N2O reduction and 15N assimilation were skewed, potentially due to normalising to unit dry biomass which may not account for the true abundance of N2 and N2O fixers. Therefore, we present 95% of data (2.5% to 97.5% percentiles) for both datasets (Figs. 2a, c, 3a and 4a) and fitted quantile regression models (‘quantreg’80) rather than mean regression models to the full dataset for rate of 15N assimilation to minimise any bias from outliers (median regression lines, Fig. 4a). The difference between the temperature response of 15N2 and 15N2O was compared using the ‘emmeans’ package.

nifH Shannon diversity was calculated using the ‘estimate_richness’ function in the ‘phyloseq’ package81 and any changes in the nifH community calculated using the ‘adonis’ function from the ‘Vegan’ package82 (with Original UniFrac distance). Principal Coordinates Analysis (PCoA) was used to test the significance of either incubation day or excess N2O on nifH community composition by Permutational multivariate analysis of variance (PERMANOVA) and redundancy analysis (RDA) to ordinate N2O reduction and nifH relative abundance.

Reporting summary

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