Extreme enrichment in atmospheric 15N15N.

Molecular nitrogen (N2) comprises three-quarters of Earth's atmosphere and significant portions of other planetary atmospheres. We report a 19 per mil (‰) excess of 15N15N in air relative to a random distribution of nitrogen isotopes, an enrichment that is 10 times larger than what isotopic equilibration in the atmosphere allows. Biological experiments show that the main sources and sinks of N2 yield much smaller proportions of 15N15N in N2. Electrical discharge experiments, however, establish 15N15N excesses of up to +23‰. We argue that 15N15N accumulates in the atmosphere because of gas-phase chemistry in the thermosphere (>100 km altitude) on time scales comparable to those of biological cycling. The atmospheric 15N15N excess therefore reflects a planetary-scale balance of biogeochemical and atmospheric nitrogen chemistry, one that may also exist on other planets.

INTRODUCTION

Nitrogen is a major component of many planetary atmospheres. On Earth, it probably first appeared early, degassing from the mantle because of its low solubility in oxidized silicate melts (, ). Its present-day budget is dominated by biological cycling. Nitrogen fixation (natural + industrial) is the major sink of N2, whereas denitrification (via nitrate or nitrite reduction and ammonia oxidation) is its major source. Global rates are typically estimated to be between 400 and 500 TgN year−1 (–), with a total atmospheric inventory of 3.92 × 109 TgN. Abiotic N2 cycling mechanisms are much slower and include N2 fixation by lightning (~5 TgN year−1) and geologic outgassing [0.1 TgN year−1 ()]. This vast difference in magnitude between biotic and abiotic fluxes suggests that N2—particularly its isotopic composition—records biological nitrogen cycling at the planetary scale. A planetary-scale perspective on nitrogen cycling would offer new opportunities to evaluate Earth’s enigmatic nitrogen budget.

We describe a new isotopic tracer that unlocks this signal by exploiting natural variations in 15N15N, the N2 molecule containing two rare isotopes. The tracer characterizes the relative natural abundances of 14N14N, 14N15N, and 15N15N as a Δ30 value, which quantifies the excess in 15N15N relative to a random distribution of 15N and 14N atoms in N2 molecules. Mathematically, the Δ30 tracer is defined by Δ30 ≡ 30R/(15R)2 − 1, where 30R = 15N15N/14N14N and 15R = 15N/14N. Other isotopic notations used throughout include δ15N, δ29N2, and δ30N2, defined similarly: δ15N = 15R/15Rair − 1, δ29N2 = 29R/29Rair − 1, and δ30N2 = 30R/30Rair − 1, where 29R = 14N15N/14N14N. All δ and Δ30 values are reported in per mil (‰).

At isotopic equilibrium, a small 15N15N excess is expected because of its higher thermodynamic stability compared to 14N14N and 14N15N molecules. Thermodynamic control of Δ30 values would cause them to vary monotonically between 1.9 and 0.07‰ for temperatures between 200 and 1000 K, respectively (fig. S1) (). When out of isotopic equilibrium, however, they may express a wider range of values that reflect biogeochemical processing (–). Crucially, the Δ30 value is mainly sensitive to the chemistry that makes and breaks N–N bonds, allowing it to evolve independently from the 15N/14N ratio in N2 on time scales governed by the rates at which its bonds are broken and remade: The Δ30 value traces N–N bond-forming chemistry by quantifying the number of naturally occurring 15N-atom pairs. We analyzed the Δ30 composition of N2 from a variety of natural and laboratory samples to an estimated accuracy of ±0.3‰, free of isobaric interferences, using an ultrahigh-resolution isotope ratio mass spectrometer (). The measurements were calibrated against N2 gas that was driven to isotopic equilibrium by reordering on a strontium nitride catalyst at 800°C.

RESULTS

Air sampled from the University of California, Los Angeles (UCLA) and Rice University shows a 19‰ excess in 15N15N relative to N2 containing a random distribution of 14N and 15N isotopes [that is, Δ30 = 19.1 ± 0.1‰ (95% confidence interval); Fig. 1]. Stratospheric air sampled at 32 km in 2004 (, ) and dissolved air sampled from the surface ocean in 2017 are analytically indistinguishable from surface air (table S1; P > 0.4). The atmosphere’s high Δ30 value is 10 times larger than that for thermodynamic equilibrium in its coldest regions (that is, Δ30 ~ 2‰ at 180 K), indicating that the atmospheric Δ30 value is governed by kinetic processes. The stratospheric sample suggests that the Δ30 value is homogeneous up to at least 32 km.

Fig. 1

Isotopic composition of N2 from natural samples and laboratory experiments.

The covariation of all three isotopic variants of N2 is shown by plotting δ30N2 versus δ29N2. Mass-dependent fractionation curves for air and high-temperature equilibrated N2 are also shown. Error bars are smaller than the data points.

To determine the effects of biological cycling on the Δ30 value of N2, we analyzed (i) N2 produced by cultures of denitrifying bacteria and (ii) 15N15N isotopic fractionation factors for nitrogen-fixing cyanobacteria. Axenic cultures of Pseudomonas stutzeri and Paracoccus denitrificans produced N2 with Δ30 values ranging from 0.4 to 1.4‰, with a decreasing trend as the reaction progressed. The denitrification Δ30 values for high nitrate conversion are lower than those for isotopic equilibrium at the culturing temperatures (Δ30,equil = 1.0‰ at 303 K), indicating that kinetic and/or combinatorial isotope effects are expressed in biological denitrification (table S2). Isotopic fractionation resulted in either no change or a slight decrease in Δ30 values for nitrogen fixation in batch cultures of Anabaena variabilis expressing the molybdenum-containing variant of nitrogenase (Fig. 1 and table S1). This observation implies that the 15N15N/14N14N isotopic fractionation factor for N2 fixation is roughly equal to the square of that for 15N/14N [that is, 30α ≈ (15α)2], resulting in little change to Δ30 values in the residual N2 reservoir during N2 fixation.

Neither bacterial denitrification nor N2 fixation can explain the atmospheric Δ30 value. Moreover, the low Δ30 values in N2 coming from bacterial denitrification must be offset by another process characterized by extremely high Δ30 values (>>19‰). Therefore, we evaluated whether other processes relevant to the atmospheric N2 budget can elevate atmospheric Δ30 values to such a degree. Fungal denitrification produces N–N bonds as N2O and expresses different isotope effects from those found in bacteria (, ), but it contributes on the order of 1% to the global N2 budget (). Its contribution to Δ30 values is likely negligible on the basis of mass balance. Anaerobic ammonia oxidation (anammox) may have an important impact on atmospheric Δ30 values; thus, we measured N2 samples of opportunity from an anammox reactor at the University of Utah, which had been flushed with air-N2 to render it anaerobic (). These samples showed Δ30 values lower than atmospheric values (table S2), indicating that the Δ30 value of anammox-N2 is less than that of the atmosphere. A large difference in 15N/14N ratios and kinetic isotope effects for ammonium and nitrite in nature () suggests that end-member Δ30 signatures for the anammox process are near or less than zero because of combinatorial effects and mass-dependent fractionation (). Experiments with enriched cultures are needed to determine the range of Δ30 values characteristic of anammox-N2 and its contributions to the global budget, but the evidence presented here suggests that it cannot explain the high atmospheric Δ30 value.

Mixing of N2 reservoirs that have similar Δ30 values but different 15N/14N ratios can yield elevated Δ30 values. However, the difference in the reservoir 15N/14N ratios must be at least 300‰ different, larger than any thus far observed on Earth, to cause Δ30 = 19‰ upon mixing (see Materials and Methods). Other physical effects such as gravitational and thermal fractionation are mass-dependent, yielding changes in Δ30 value that are much smaller in magnitude than that required to explain the atmospheric Δ30 value (–).

The slow biological recycling time of N2 in the atmosphere [~10 million years (My) based on the N2 inventory and biological fluxes] suggests that atmospheric chemistry could affect its isotopic composition. Extreme δ15N enrichments (>500‰ versus Earth’s N2) have been observed in the atmospheres of Mars (, ) and Titan (), possibly arising from unusual isotope effects in photochemistry near the N2 dissociation threshold (80 to 100 nm; 12.4 to 15.5 eV) (–). These and other unconventional isotopic fractionation mechanisms may also be operating in the middle and upper atmosphere of Earth, influencing both the 15N/14N ratios and the proportions of 15N15N in atmospheric N2.

We tested this possibility by conducting high-voltage radio-frequency discharge experiments with an Oudin coil to simulate the chemistry of N2 in the upper atmosphere. Pure N2 gas and O2/N2 mixtures were introduced into a glass vacuum chamber that contained electrical feedthroughs made of tungsten. Experiments lasting 1 hour yielded N2 with Δ30 values ranging from –3 to +23‰, with a marked dependence on initial pressure and O2/N2 ratio (Fig. 2 and table S3). The Δ30 values of electrolyzed N2 show a step-like increase when O2 is added, and they decrease with increasing pressure above ~3 mbar. Partial isotope equilibration on tungsten surfaces may be important below this pressure (). The pressure dependence of postelectrolysis Δ30 values above 3 mbar is independent of the initial isotopic composition, suggesting that a laboratory steady state is approached under these conditions and that N2 bonds are being broken and remade. Together, these experiments demonstrate that gas-phase chemistry of N2 and its ions can potentially explain the extreme Δ30 values observed in the atmosphere.

Fig. 2

Results of laboratory electrolysis experiments demonstrating clumped-isotope reordering.

Initial isotopologue compositions were either N2 that had been equilibrated at 800°C (circles) or pure tank N2 (triangles). Surface chemistry effects likely became dominant below ~3 mbar. Error bars are smaller than the size of the data points.

The marked increase in Δ30 values with the addition of O2 in the experiments points to an oxygen-containing species in the key 15N15N-concentrating step(s). The decrease in steady-state Δ30 values with increasing pressure suggests that this step becomes less important at higher pressures. At these higher pressures, the system is affected more by a step that disfavors 15N–15N bonds. This balance between two opposing reactions would qualitatively explain the laboratory steady state. Our kinetic model of the experiments () suggests that the most important N–N bond-breaking mechanisms are electron impact dissociation (N2 + e– → N+ + N + 2e–) and the ion-molecule reaction N2+ + O → NO+ + N. Nitrogen-nitrogen bond-making occurs fastest via N + NO → N2 + O. These reactions compete primarily with recombination on chamber surfaces and gas-phase recombination (that is, N + N + M → N2 + M, where M is a third body or wall). On the basis of the dramatic increase in Δ30 values when O2 is introduced, we hypothesize that the N2+ + O and/or N + NO reactions are key to concentrating 15N in 15N15N molecules in the laboratory experiments: Both reactions require an oxygen source and they contribute a larger proportion to the total N–N bond recycling rate as pressure decreases. The other reactions involving oxic species comprise minor channels, typically several orders of magnitude slower (fig. S4). We note that any trace oxygen source in the N2-only experiments (for example, from a surface; it need not be O2) could be responsible for the pressure dependence for Δ30 values observed there. However, a Δ30 pressure dependence for the N2-only chemistry cannot be ruled out.

In the atmosphere, N2+, NO, N, and O are also key species, particularly in the thermosphere (100 to 500 km altitude). We estimated the time scale of photochemical N2 cycling in the atmosphere using species concentrations derived from the Whole Atmosphere Community Climate Model, eXtended version (WACCM-X) model, from the surface to ~500 km (), and temperature-dependent reaction rates. We find that in model year 2001, 302 TgN is recombined via N + NO, whereas 0.01 TgN is recombined by the N + NO2 and N+ + NO channels combined (Fig. 3). The most important chemical channel of N2 destruction, N2+ + O, destroys about 74 TgN year−1. Photolytic, electrolytic, and nonthermal N2 bond rupture cannot be explicitly calculated from the WACCM-X species concentrations; hence, we assume that the sum of all destruction mechanisms balances chemical recombination to maintain a steady state for N2 concentrations. In such a steady state, the thermosphere would recycle a mass of N2 equal to the atmospheric inventory (3.92 × 109 TgN) in 13 My. Locally nonthermal conditions in the thermosphere and variations in incident solar radiation can alter this estimate (see Supplementary Text). Nevertheless, the time scale of atmospheric N2 recycling is likely comparable to that for biological N2 cycling.

Fig. 3

Calculated global, annual mean outputs from the WACCM-X model (year 2001).

(A) Temperatures, (B) species concentrations, and (C) gas-phase thermal reaction rates relevant to N–N bond rupture and formation are shown. Nonthermal effects are important in the upper atmosphere but are not included in these calculations. Photolysis reactions have been omitted from this plot.

Furthermore, mass exchange between the upper and lower atmosphere is sufficient to mix high-Δ30 thermospheric N2 into the troposphere on these time scales. The anthropogenic CO2 rise has recently been detected in the lower thermosphere by satellites (, ), suggesting that gases emitted at the surface can enter the thermosphere within several hundred years and vice versa. The atmosphere’s Δ30 value therefore carries signatures of both gas-phase and biological processing of N2.

DISCUSSION

We interpret the atmospheric Δ30 value in N2 (hereafter Δ30,atm) as the balance between two planetary-scale cycles that drive Δ30 toward characteristic end-member values. Atmospheric chemistry, primarily in the thermosphere, drives Δ30,atm toward its end-member value, Δ30,thermosphere, at a rate represented by Fthermosphere (TgN year−1). Biological N2 cycling drives Δ30,atm toward its end-member value Δ30,bio at the rate Fbio. Our data suggest that N2 fixation does not alter atmospheric Δ30 values significantly (that is, it does not discriminate strongly and fractionates mass-dependently); thus, the biological signature of Δ30,bio is primarily that associated with denitrification. The steady-state global Δ30 mass balance is thus approximately described by Fbio(Δ30,atm − Δ30,bio) + Fthermosphere(Δ30,atm − Δ30,thermosphere) ≈ 0 (see Supplementary Text). Using Δ30,bio = 0.4‰ and the WACCM-X–derived Fthermosphere = 302 TgN year−1, we find that Δ30,thermosphere = 40 to 50‰ would be consistent with current bottom-up estimates of the global denitrification flux of N2 (Fbio ~ 400 to 500 TgN year−1; Fig. 4A). A larger Δ30,thermosphere end-member value would imply a larger global biological N2 cycling rate and vice versa. Refining Δ30,bio, Δ30,thermosphere, and Fthermosphere may thus allow one to constrain the average global rate of denitrification integrated over the past ~10 My. In principle, this idea can also be applied to extraterrestrial atmospheres to determine rates of biogeochemical nitrogen cycling once the controls on Δ30,thermosphere and Fthermosphere (for example, those tied to solar flux and the presence of oxic species) are understood. On smaller scales, the Δ30 tracer should be an in situ measure of denitrification that does not rely on tracer injections and their associated uncertainties (). It may be particularly useful for studying nitrogen cycling in oceanic oxygen minimum zones and subsurface terrestrial environments.

Fig. 4

Applications of the Δ30 tracer.

Using atmospheric Δ30 values to constrain the global denitrification rate (A) [the dashed line in (A) represents the current atmospheric Δ30 value] and the nitrogen sources of geologic N2 outgassing (B). Error bars are smaller than the size of the data points.

More broadly, Δ30 measurements are of utility as an unambiguous measure of air content in geochemical mixtures. For example, gases that are mixtures of atmospheric N2 and geologically outgassed N2 can be partitioned into contributions from atmospheric N2, mantle nitrogen, and crustal nitrogen using Δ30 values, δ15N values, and isotopic mass balance in N2 (Fig. 4B). For high-temperature geologic nitrogen sources, Δ30 signatures are expected to be low because of combinatorial effects or isotope exchange equilibrium (for example, Δ30 = 0.07‰ at 1000 K), whereas δ15N values will reflect the source material [for example, +7‰ versus air for sedimentary sources and −5‰ for mantle sources (, ), although mantle δ15N values may be lower ()]. The residence time of gases in hydrothermal systems (weeks or less) is short relative to time scales of isotopic equilibration in those environments (that is, T < 400°C without engineered catalysts; see heating experiments in Materials and Methods); air is expected to retain its high Δ30 value under such conditions. The Δ30 value of N2 from fumaroles should therefore allow one to quantify the proportion of air in the sample, whereas the δ15N value can be used to partition the remaining nitrogen between mantle and recycled-sediment sources.

Fumarole gases collected at the summit crater of the Momotombo volcano (), in the Nicaraguan volcanic front, show low Δ30 values of 1.5 and 3.9‰. Gases drawn from fumaroles in Valles Caldera (New Mexico, USA) and Yellowstone Caldera (Wyoming, USA) show higher Δ30 values (10.8 and 11.6‰, respectively), indicating a higher proportion of background air contribution. We find that 82 and 91% of the volcanic nitrogen in the Momotombo samples come from subducted sediment sources. The caldera samples, representing mantle hot spots or continental rifts with little influence from subducted sediments, contain primarily mantle-derived N2: 11 and 3% of the nonatmospheric N2 come from sedimentary N2 at Valles Caldera and Yellowstone Caldera, respectively. Although noble gas–based source apportionment yields similar subducted sediment fractions for the Momotombo samples (86 and 94%, respectively), they disagree for the Yellowstone Caldera sample (3% sedimentary from Δ30 values versus 19% from N2/He ratios; table S5). Signatures of mantle and crustal outgassing are variable, and postsampling gas loss will affect gas species ratios much more so than isotopic ratios; hence, the combined Δ30-δ15N approach is likely more robust for these questions. Applying the Δ30 tracer to volcanic N2 outgassing may constrain the 15N isotopic composition of the primordial mantle, which has implications for planetary accretion, differentiation, and the development of the early atmosphere ().

MATERIALS AND METHODS

Analytical methods

Natural and laboratory-generated samples of N2 were analyzed on an ultrahigh-resolution gas source mass spectrometer that separates 14N14N, 14N15N, and 15N15N from their main isobaric interferences (). Spike-dilution experiments suggest that residual isobaric effects on N2 are minor, comparable to other analytical uncertainties. Measurements were calibrated to a thermodynamically based reference frame by equilibrating N2 gases at 800°C with strontium nitride (fig. S1) (). Methodological accuracy was verified using the pinhole-diffusion and N2 re-equilibration experiments described below.

N2 gas was analyzed at mass/charge ratio (m/z) = 28, 29, and 30 (14N14N+, 14N15N+, and 15N15N+, respectively) on the Nu Instruments Panorama mass spectrometer (UCLA). 14N14N+ and 14N15N+ ions were collected on Faraday cups with intensities ranging from 15 to 480 pA for 14N14N+, with all natural samples but the volcanic samples run between 50 and 480 pA. Volcanic samples were run between 18 and 50 pA. A secondary electron multiplier was used for 15N15N+, with corresponding count rates ranging from 1 × 103 to 4 × 104 counts per second (cps), or 0.06 to 2.5 fA. Abundances of 14N14N, 14N15N, and 15N15N in atmospheric N2 are approximately 99.3%, 0.7%, and 14 parts per million, respectively. Typical mass resolving power (MRP) was near 50,000, calculated as the ratio of the ion mass and the mass difference between 95 and 5% of the peak height; for example, MRP = 30.00022 atomic mass unit (amu)/0.0006 amu = 50,000. Nearly baseline resolution is achieved between 14N16O+ and 15N15N+, despite a mass difference of only 0.002 amu (fig. S2). Replicate Δ30 measurements of air with 14N16O+ ion currents ranging from 104 to 105 cps, afforded by having previously reduced the source surfaces with methane, were indistinguishable. The effect of the isobaric interference between 12C18O+ and 15N15N+ (mass difference, 0.001 amu) was evaluated using serial 12C18O spike dilutions and found to be comparable to the uncertainty in the external reproducibility of the measurement (fig. S2).

Samples of N2 were purified using gas chromatography (GC) with methods described previously for methane (). Briefly, a two-column setup at 60°C was used to separate N2 from other gases, first through a stainless steel molecular sieve 5A column [3 m × 1/8″ outside diameter (OD)] followed by a stainless steel HaySep D column (2 m × 1/8″ OD). This setup was interfaced with an all–stainless steel, oil-free high-vacuum line (base pressure ~10−6 mbar) through which samples were introduced and recollected for analysis. Some air samples were purified on a separate, single-column GC system (30 to 60 cm × 1/8″ OD molecular sieve 5A, 80/100 mesh) over a range of temperatures, interfaced with a glass vacuum line (), with indistinguishable results.

Calculations, instrument calibration, and testing

The bulk 15N/14N ratio, in terms of 14N14N, 14N15N, and 15N15N abundances, is expressed asTherefore, to obtain both a δ15N ≡ (15Rsample/15Rair − 1) value and a Δ30 value, one must solve for 15R of the sample using 29R and 30R. Our method is one often used in clumped-isotope geochemistry, namely, to measure a δ15N standard (air; 15Rair = 0.003676) and a Δ30 standard (N2 isotopically equilibrated at 800°C; see fig. S1) against the same laboratory working standard gas and iterate 29R and 30R of that gas until its isotopic composition is found. Subsequent measurements are related to the δ15N and Δ30 standards through this working gas.

We note in passing that Eq. 1 also shows how mixing between N2 reservoirs will alter 15R nonlinearly. Moreover, because Δ30 ≡ 30R/(15R)2 − 1, Δ30 values vary nonlinearly with mixing as well. A 50:50 mixture between components that both have Δ30 values of zero, but with a 20‰ difference in δ15N, will have Δ30 = +0.1‰. If the difference in δ15N of the two components is 40‰, then the gas will be characterized by Δ30 = +0.4‰.

In order to reorder N2 toward high-temperature equilibrium, samples were expanded into either breakseals or valved quartz sample tubes containing strontium nitride (Sr3N2, a black powder; ESPI Metals and Materion) and heated in a horizontal tube furnace to 800°C for between 4 and 96 hours. Time-series experiments conducted at 400° and 800°C (Δ30 = 0.21 and 0.07‰ at equilibrium, respectively) showed a ninefold change in reordering rate (fig. S2), but we also found that this rate depended on the age of the solid catalyst; it decomposes slowly over time due to reaction with water to make ammonia. The condition of the catalyst and the relative proportions of catalyst to N2 gas allowed some variation of δ15N values of the equilibrated gas (table S1). We also reordered N2 from a 21-mbar electrolysis experiment (Δ30 = −2‰) to Δ30 = 0.0 ± 0.1‰ on strontium nitride at 800°C. This experiment, combined with the time series described above, indicates that heating N2 over strontium nitride yields N2 with a nearly random distribution of isotopes.

In general, the catalyst is crucial to the isotopic reordering of N2. An experiment using titanium nitride instead of strontium nitride as the N2-reordering catalyst showed significantly slower reordering (Δ30 = 19 → 7‰ in 7 days at 900° to 1000°C) with no change in δ15N. The experiment suggests that the N2 isotopes are not equilibrating on the quartz.

To verify instrumental accuracy with pure N2 samples, we performed pinhole-diffusion experiments using the same apparatus as that described previously for O2 (). Briefly, 1 mbar of N2 in a 5-liter bulb was allowed to diffuse through a 75 ± 7.5 μm critical orifice, which was comparable to the mean free path of N2 at that pressure (~95 μm). The amount of undiffused gas remaining was determined manometrically, and both the diffused gas and the residue gas were analyzed to determine the difference in their δ29N2 and δ30N2 values. Measured values were in agreement with those predicted for Rayleigh fractionation of gases separated according to Graham’s law of effusion within typical analytical uncertainty ranges (that is, ±0.2‰ in δ30N2) when a correction for back diffusion/viscous flow was included (table S6). The correction in which the effective fractionation factor αeffective is related to Graham’s law fractionation factor αeffusion by (αeffective − 1) = (αeffusion − 1)(1 − Pds/Pus) has been described previously; the Pds/Pus factor here is similar in magnitude to that found previously on the same apparatus for O2 diffusion (). The consistency between theory and measurements for these experiments suggests that isotopic reordering on the tungsten filament in the mass spectrometer does not exceed analytical uncertainties over a range of ~40‰ in δ30N2.

Other potential impurities in natural samples included O2 and methane. We performed analyses of air-N2 that included up to 14% O2 and found that Δ30 values increase 0.5‰ at 3 to 4% O2 and 1.8‰ at 14% O2, accompanied by 0.2 and 0.5‰ decreases in δ15N values, respectively. We estimate typical O2 content to be <<1% after GC purification. In the volcanic N2 samples, small amounts of methane (<10%) were identifiable and collected in the tail of the N2 peak. To examine the effect that this methane impurity may have on the N2 isotopologue ratios, we made a 10% methane/N2 mixture of similar sample size and purified it. We did not observe any difference in Δ30 value from the methane-free tank gas.

Finally, we tested for potential analytical artifacts due to variations in N2 sample size. Air-N2 samples ranging from 5 to 160 μmol in size were prepared and analyzed over a range of ion currents. Bulk isotope composition exhibited larger variability at small sample sizes, about ±0.2‰ in δ15N versus ±0.05‰ for larger samples. The Δ30 values were indistinguishable up to ~250 pA on m/z = 28 (smaller samples had larger analytical uncertainty because of counting statistics). Above 250-pA ion currents, the Δ30 values for air-N2 decreased by as much as ~0.4‰ when the ion current was 400 pA on m/z = 28. The origin of this effect is enigmatic: No dependence on ion current intensity was observed for heated gases, and ion-counting rates (104 cps) were lower than typical rates that one expects would require dead-time corrections. A possible explanation for this effect is isotope exchange between NO and N2 in the ion source, similar to the exchange reaction observed for N2O (). Whatever the source of this ion-intensity effect, it was only relevant for N2 dissolved in seawater (the largest samples), and the effects are small. We therefore report Δ30 values for those samples before and after correcting for this ion-intensity effect based on the air-N2 measured at similar ion currents (table S1). On the basis of the comprehensive analytical tests above, we estimate an overall accuracy of ±0.3‰ in Δ30 values.

Natural samples

Sampling procedures for UCLA air and the stratospheric sample have been described previously (, ). The stratospheric sample was #3-A01-R(1141) from a scientific balloon flight in 2004 (Fort Sumner, New Mexico; 32.3 km) that has also been characterized for a variety of other trace gases and isotope ratios (, , , ). The stratospheric air samples were purified at UCLA. Seawater samples were collected from the San Pedro Ocean Time-series site (33°33′N, 118°24′W), a coastal site near Los Angeles, on 12 April 2017. They were siphoned from 10-liter Niskin flasks into prepoisoned, pre-evacuated (<10−3 mbar) bottles according to established methods (). The bottles were allowed to degas and equilibrate on an orbital shaker (110 rpm) for 48 hours before the water was removed, and the headspace gas was collected onto silica gel for purification. Dissolved N2 was separated from the other gases at Rice University on a molecular sieve 5A column (3.05 m × 1/8″ OD, 80/100 mesh) at −80°C and recollected on a silica gel U-trap using a method described previously (). We also isolated atmospheric N2 using this method, and the results were indistinguishable from gases collected and purified at UCLA.

Biological culturing conditions and experiments

Denitrifying bacteria

P. stutzeri and P. denitrificans were cryogenically stored (−80°C) in tryptic soy broth (TSB; Caisson Labs) and sterile glycerol 1:1 (v/v). Stock cultures were reestablished in 5 ml of TSB amended with sodium nitrate (NaNO3, 10 mM; Sigma-Aldrich) under aerobic conditions at a constant temperature with continuous agitation (18 hours). Denitrifier cultures were grown at 30°C. Individual colonies were obtained from reestablished stock cultures by the streak-plate technique on tryptic soy agar (Caisson Labs) amended with NaNO3 (10 mM). Streak-plate cultures were sealed with parafilm and incubated (aerobic, 30°C). The plates were stored at 4°C for up to 2 weeks before establishment in liquid medium for experiments.

The two species were established with one colony from stored stock culture plates in 5 ml of TSB amended with NaNO3 (10 mM) in a 20-ml culture tube (Thermo Fisher Scientific). Cultures were grown aerobically with agitation (30°C for 18 hours) to late exponential phase (optical density at 600 nm = 2.0). Culture turbidity was determined with a Spectronic 20 spectrophotometer (Bausch and Lomb). Two 160-ml sterile serum bottles containing 50 ml of carbon minimal medium (CMM; 10 mM NaNO3 and 10 mM sodium succinate; Sigma-Aldrich) () were each inoculated with 200 μl of the aerobic culture. The bottles were stoppered (Geomicrobial Technologies Inc.) and crimp-sealed, and the headspace was sparged with ultrahigh purity (UHP) helium for 20 min. Cultures were incubated (30°C for 18 hours) with agitation. The cells were transferred to 50-ml conical Falcon tubes (Corning) and centrifuged (3000g for 30 min). The cell pellet was washed in CMM lacking carbon and nitrogen and centrifuged (3000g for 30 min). The supernatant was decanted, and the cells were dispersed in CMM lacking a carbon or nitrogen source (optical density at 600 nm = 0.2). The cells were aliquoted (9.6 ml) into sterile flasks outfitted with a vacuum “sidearm” valve (160 ml), and a carbon source was added (30 mM sodium succinate). The flask was stoppered (Geomicrobial Technologies Inc.) and crimp-sealed, and the sidearm closure was sealed. An anaerobic environment was created by sparging the cells (UHP He, 60 min). Sparging was accomplished by inserting one sterile stainless steel needle (#20, Thomas Scientific) carrying He gas through the stopper into the medium while a second sterile stainless steel needle was inserted through the stopper and into the headspace to allow gas to exit. Following sparging, the flasks were allowed to reach atmospheric pressure and then the reactions were initiated by injecting the nitrogen source (30 mM USGS34 potassium nitrate) with a gas-tight syringe. Individual reactions were stopped by injecting 10 M sodium hydroxide (400 μl).

Nitrogen-fixing cyanobacteria

A. variabilis [American Type Culture Collection (ATCC) 29413] was cultivated in BG-11 medium () from lyophilized samples as per ATCC instructions. Cultures were maintained by streaking onto BG-11 agar under a 12-hour light/12-hour dark regime at 25°C.

Cyanobacterial cultures were started from streak plates of A. variabilis by adding culture to 10 ml of BG-11 medium and incubating aerobically (22°C, 12-hour light/12-hour dark regime). Cultures were transferred to larger vessels twice in 1:10 dilutions to reach a final culture volume of 250 ml. Transfers were conducted only when cultures reached a chlorophyll-a density of greater than 1 μg/ml (). The cells were centrifuged (4000g for 15 min) and washed 1× in 50 ml of BG-11 medium without nitrogen (BG-11o). The cells were resuspended in 250 ml of BG-11o medium, which had been allowed to equilibrate with laboratory air for 1 day with gentle stirring. Contamination of outside agents was prevented by fitting a sterile cotton plug to the flask opening and covering this with sterile tin foil. Laboratory equilibration of the medium served to ensure that atmospheric N2 was the only source of nitrogen in the medium. The culture was then transferred to a sterile “Emerson” flask with a sterile gravity syphon. The Emerson flask (200 ml) was filled beyond the neck and then sealed shut to remove any headspace. Before stoppering the flask, the seal was sterilized with 70% ethanol and rinsed with Milli-Q water. Nitrogen fixation reactions were considered to have begun once the Emerson bottle was stoppered. Reactions were conducted in an environmental chamber (22°C, 12-hour light/12-hour dark regime), and the flasks were turned over once a day to provide gentle agitation. After between 2 and 7 days, the dissolved N2 was transferred by equilibration following established protocols (). After allowing the culture to settle to the bottom of the bottle, 150 ml of culture medium was transferred to an evacuated Emerson bottle (300 ml) with a sterile syphon. The transferred medium was allowed to equilibrate for at least 6 hours before isolation.

Isolation and trapping of N2 for analysis

Quantitative distillation of N2 from samples was achieved under high vacuum (4.5 μmHg). A metering valve and two stainless steel liquid N2 cryogenic traps (OD = 6.35 mm) preceded the sample bottle. The first cryogenic trap condensed and separated water from gases, whereas the second trap was filled with coarsely crushed silica to trap all remaining gases (15 min). Isolated N2 was cryogenically transferred to a 30-cm Pyrex tube (OD = 6.35 mm) containing 50 mm of crushed silica gel for 15 min. To accomplish this, the second cryogenic trap was isolated from the first cryogenic trap and the vacuum line was evacuated. The system was then isolated from the vacuum system and the isolated gases were desorbed from the silica by heating with warm tap water. Pyrex sample tubes were sealed with an acetylene torch while still under cryogen conditions. All silica gel traps were conditioned under high vacuum at 200°C (24 to 48 hours) before sampling.

The specific method for isolation of N2 from sidearm flasks for denitrification experiments differed slightly from the previous method. Sidearm bottles were attached to the vacuum line with ultra-torr Cajon fittings. The sidearm bottle was opened to the vacuum line while the metering valve remained closed. The metering valve was then slowly opened to maintain a vacuum (50 to 70 μmHg) while the sample passed through the cryogenic traps, until high vacuum was again achieved (35 to 40 min). The cryogenically isolated gases were then desorbed from the silica as previously described.

Analytical correction for air contamination

During the P. denitrificans experiments, a procedural blank was performed, which revealed a small residual gas pressure equal to 1 to 5% of the experimental N2 gas yield, depending on reaction progress. This blank (0.8 μmol) was presumed to be entirely N2 with atmospheric isotopic composition. The reported isotopic composition for both the P. denitrificans and the P. stutzeri experiments includes a correction for this blank. We cannot rule out a difference in blank amount between the two experiments. However, the more fully reacted experiments had larger N2 gas yields and therefore a smaller contamination from this blank; thus, their Δ30 values are more likely to reflect those intrinsic to bacterial denitrification.

Samples of opportunity from anammox bioreactors

Water from a semi-batch–fed anammox bioreactor at the University of Utah () was sampled into 60-ml septa vials (bubble-free, until they overflowed, ensuring no gas headspace) and poisoned with 600 μl of saturated ZnCl2 solution. Samples were shipped back to Michigan State University for isolation of the dissolved gases.

Electrolysis experiments and modeling

Laboratory electrolysis

Electrolysis experiments were performed in a valved-off Bayart-Alpert ion gauge with one lead (the ion collector) attached to the Oudin coil and the others grounded (fig. S3). The chamber was preconditioned initially by applying a voltage to the electrical lead under vacuum for ≥1 hour to degas the filament and then applying a voltage under N2 atmosphere for ≥1 hour to replace the adsorbed species with nitrogen. In the experiments, voltage was applied to the electrical lead until continuous visible emission was observed in N2 or an O2/N2 mixture. The same analog setting (that is, rotating dial position at the base of the coil) was used for all samples. The reaction was allowed to proceed for 1 hour.

The emission field changed color and expanded as pressure was varied (fig. S3). The colors and patterns were similar for both N2 only and O2/N2 mixtures. These changes suggest that local electron and ion concentrations and electron energies also varied with pressure. Consequently, we expect that ion production rates, energy-dependent product branching ratios, wall reaction rates, and the rates of reactions dependent on these quantities likely varied with total pressure as well. Because of the potential variability in reaction rates expected in the experiments, our kinetic model was aimed at determining the most important N–N bond-altering mechanisms in the gas phase rather than explaining the laboratory results quantitatively.

Model of electrolysis experiments

The model contained 41 reactions for the O2-N2 system, including charge transfer reactions (table S4). It was run at 298 K for 1 hour of simulated time using the program Kintecus (). Electron temperatures were held constant at 10,000 K. Only positive ion chemistry was included except for electron impact and recombination processes. Electronic states were not specified; ground electronic states and thermal vibrational states for all reactions and products were assumed unless otherwise noted. Surface chemistry was not explicitly included. Electron impact ionization was only included for N2 and O2, with prescribed (but adjustable) branching ratios between ionization and fragmentation channels (typically 3:1). A more explicit description would render the number of variables unmanageable for this conceptual model. In a similar vein, electron densities were prescribed and held constant within each simulation as a way to specify the total rate of electron impact processes.

Gross rates of N–N bond-altering processes were tracked by including “dummy” products in those reactions that did not affect other variables such as temperature or pressure. Rates of electron impact ionization and dissociation were varied over a large range, but rates of at least 1013 to 1014 cm3 s−1 were required for the fastest N–N bond-altering processes to reorder an amount of N2 equal to the experiment’s inventory between 1 and 10 mbar (2.5 × 1016 to 2.5 × 1017 cm−3). For the N2-only experiments, only electron impact and the termolecular N + N + M → N2 + M reaction would be relevant to the destruction and formation of N2 unless some other trace oxygenated constituent were present (for example, NO). We note that tungsten surfaces can catalyze isotope exchange between N2 molecules (); at low pressures, therefore, these reactions likely influenced Δ30 values resulting from electrolysis, driving them down toward isotopic equilibrium values in competition with gas-phase isotopic fractionation.

Sample results for three different electron number densities are shown in fig. S4. In all cases, electron impact fragmentation of N2 is the only important N2 bond-breaking mechanism by an order of magnitude except at low pressures. At low pressures, N2+ + O becomes comparable in rate. The importance of this channel decreases at higher pressures. Of the bond-making reactions, N + NO is the most important by at least a factor of 100 in all cases.

Determining the atmospheric N2 recycling time scale

Globally integrated rates of N2 bond-altering processes in moles per year were calculated from monthly-averaged outputs of the WACCM-X model () for model year 2001 (1.9° latitude × 2.5° longitude grid). The data were outputs of present-day control runs of the Community Earth System Model version 1.0 (data set f.e10.FWX.f19_f19.control.001) available from the Earth System Grid (www.earthsystemgrid.org). Globally integrated species concentrations and reaction rates were calculated using methods similar to those previously described (). Briefly, reaction rates were calculated at each grid point using monthly-averaged species mixing ratios, temperatures, and pressures. Because N2 is not an explicit field in this version of the model, its concentrations were calculated by subtracting the number densities of O and O2 from the total number density. Reaction rates were then integrated over grid volumes covering all 81 atmospheric levels using the latitude, longitude, and mean geopotential height of each grid point as the vertical coordinate. Accuracy checks on this integration scheme included reproducing the total atmospheric N2 inventory (1.40 × 1020 mol N2 modeled versus 1.4 × 1020 mol N2 expected) and total atmospheric volume (2.74 × 1014 km3 modeled versus 2.76 × 1014 km3 using a 500-km exobase). Weighting for residence time in different regions of the atmosphere was not included because the time scale for whole-atmospheric mixing [<<106 years based on cross-homopause and intrathermosphere diffusion coefficients of >10−5 km2 s−1 ()] should be much faster than the time scale found for atmospheric N2 bond recycling (107 years). The time scale for atmospheric N2 recycling in years is defined as the total N2 inventory divided by the globally integrated rate of N–N recombination assuming a steady state between N2 destruction and recombination.

Mass balance of volcanic N2

Mixing fractions of air N2, recycled-sediment N2, and mantle N2 in volcanic N2 samples (that is, fair, fRS, and fmantle, respectively) are calculated from N2/He ratios and δ15N values using the end-member compositions shown in table S5 and the mass balance equations ()

Alternatively, they are calculated using Eqs. 4 to 6 and δ15N and Δ30 values for each end-member (converted to 29R and 30R using Eq. 1)

In both methods, only fair, fRS, and fmantle are unknown, and they can be computed using a set of three equations.