Introduction

Molecular nitrogen (N2) constitutes approximately 78% of the Earth’s present atmosphere and is an essential element for sustaining life. It plays a vital role in the N biogeochemical cycle because molecular nitrogen (N2) is fixed by microorganisms as ammonium (NH4+) and successively transformed into nitrates (NO3−) in the Earth superficial reservoirs1. Past variations in the atmospheric content of nitrogen, which is still under debate, may have played an essential role in regulating the early Earth’s climate2,3 as greenhouse gas. Uncertainties in understanding the secular changes in the Earth’s atmospheric nitrogen concentration are partially caused by the lack of knowledge about the deep geological cycle of nitrogen and particularly its abundance and chemical form in the terrestrial mantle4. Compared to other volatile elements, nitrogen is depleted by one order of magnitude in the mantle and in the whole Earth5. This is the so-called “missing nitrogen”. However, the causes of the nitrogen depletion, compared to the terrestrial volatile inventory, have not been yet identified. One hypothesis suggests losses from the early Earth’s atmosphere by meteorite impacts6. However, meteorites and comets have likely delivered nitrogen to the proto-Earth7,8, making difficult to explain the “missing nitrogen” solely by atmospheric dissipation. A second hypothesis proposes the existence of a nitrogen reservoir in the deep Earth. For example, nitrides such as cubic boron nitride, titanium nitride, and iron nitride have been discovered respectively as inclusions of ophiolite9,10 or of super-deep diamonds originating from the upper mantle and the lower mantle11,12. Previous studies suggested that there are possible avenues for nitrogen reservoir formation in the deep Earth. Magma ocean6,13,14,15,16,17 and subduction are an event and a process which formed nitrogen reservoir in the deep Earth4,18,19,20,21.

During the magma ocean, nitrogen was thought to dissolve into liquid iron because of its moderately siderophile behavior14,22,23. However, Dalou et al.6 reported that a high carbon to nitrogen ratio (C/N = 6.1 × 102 from Marty5) in the bulk silicate Earth (BSE) cannot be solely explained by the core formation because carbon is more siderophile than nitrogen under an oxygen fugacity (fO2) of IW-0.4 to IW-3.5 (IW: Fe–FeO buffer). If a major nitrogen reservoir was formed in the Earth’s core, then nitrogen would not be depleted relative to carbon in BSE, which conflicts with the nitrogen depletion compared to carbon in the BSE, as reported by Marty5. First principle molecular dynamic simulations clarified that nitrogen solely cannot explain both the density and the bulk sound velocity of the outer core24. However, there is a possibility that the Earth’s core plays an important role in reserving nitrogen. Li et al.13 conducted high-pressure experiments under Ni–NiO and Fe–FeO buffers, reporting that the deeper part of the upper mantle can become a nitrogen reservoir through the solidification of a magma ocean because the maximum nitrogen solubilities in forsterite and enstatite were determined as ca. 10 ppm and 100 ppm13, respectively. In addition, Yoshioka et al.15 reported a 21.5 ± 18.1 ppm nitrogen solubility in bridgmanite, in high-pressure experiments with a multi-anvil apparatus and under reduced conditions close to the Fe–FeO buffer. This is the only study of the nitrogen solubility of bridgmanite which, occupies approximately 80 vol% of the lower mantle25.

A nitrogen reservoir might have also been formed in the mantle by subducting slabs26, because the inward nitrogen flux of subduction is one order of magnitude higher than that of the outgassing flux27,28,29,30. The Earth is the only planet in the Solar system with active plate tectonics31. Such volatile cycle via subduction has remained a very important research topic in geochemistry. To date, various NH4-bearing silicate minerals have been synthesized under conditions corresponding to cold subducting slabs in the deep upper mantle4,18,20,21. Sediments and eroded continental crust in subducting slabs can be nitrogen-rich. Average concentrations of nitrogen in sediments and upper crusts are, respectively, 560 ppm and 150 ppm32. However, experimental conditions reported in earlier studies4,18,20,21 are limited to the deep upper mantle (< 13 GPa). Consequently, nitrogen behavior in the mantle transition zone (> 13 GPa) and in the lower mantle (> 24 GPa) are still unknown. Therefore, we conducted high-pressure experiments at lower mantle conditions to investigate nitrogen incorporation into deep mantle minerals such as stishovite and bridgmanite.

Results

Description of the run products

The experimental conditions are presented in Table 1 and Supplementary Fig. S1. Figure 1 illustrates the cell assembly used for the high-pressure experiments. Figure 2 and Supplementary Figs. S3–S8 present the backscattered electron (BSE) images of recovered cell assemblies and recovered samples. Each outer gold (Au) capsule included an inner platinum (Pt) capsule surrounded by a Fe–FeO buffer. The two inner Pt capsules of the cell assembly accommodated the Al-free and Al-bearing samples, respectively. The outer Au and the inner Pt capsule partially melted during the runs, because of iron and water coexistence (Fig. 2 and Supplementary Figs. S3–S8). However, no Fe–FeO leaked out of the Au capsules. After all the runs were completed, a subsequent EDS analysis revealed that pure metallic iron was still present and FeO contained no Mg. The Fe and FeO coexistence after the runs indicated that the experiments’ oxygen fugacity conditions were that of the lower mantle. Therefore, we inferred that hydrogen was generated in the outer Au capsule and permeated the inner capsule. The MgO-rich hydrous melt in the inner Pt capsule was expected to contain 15NH3 under a hydrogen-coexisting environment. Run products consisted of stishovite, bridgmanite, and a hydrous melt. The resulting single crystals of stishovite and bridgmanite were larger than 40 μm and 80 μm, respectively. These grain sizes were sufficiently large to be analyzed by nanoscale secondary ion mass spectrometry (NanoSIMS). Some bridgmanite samples contained iron with concentrations up to approximately 1.75 at% in Al-bearing samples (1,500 °C, OT2258; 1,400 °C, OT2259; 1,620 °C, OT2293). Iron contamination from the Fe–FeO buffer might result from gaps between the Pt double capsules or the iron diffusion into the capsule. The quenched hydrous melt was found to be MgO-rich for all recovered samples (e.g. Si/Mg ratio = 0–0.25 in OS3083).

Table 1 Experimental conditions and run products in recovered samples.
Full size table
Figure 1
figure 1

Cell assembly for high-pressure and high-temperature experiments.

Full size image
Figure 2
figure 2

(Left) Back scattered electron (BSE) image of recovered sample from 28 GPa and 1,400 °C (OT2259). (Right) BSE images of two samples in OT2259: (a) Al-free system and (b) Al-bearing system. Circles indicate analyzed points. St denotes stishovite, Brg bridgmanite Brg2 and Brg4 in the Al-bearing sample contained iron. Dotted circles indicate analysis points located in cracks or intergranular, which are not discussed in this paper. This figure was created by Ko Fukuyama using Microsoft Power Point and Excel (OFFICE 2016).

Full size image

Nitrogen in stishovite and bridgmanite and its solubility in stishovite under various temperatures

We compared the respective nitrogen contents of stishovite and bridgmanite recovered from the run products by looking at the 15N16O−/30Si− ratio. Because of the insufficient mass resolution of the NanoSIMS (mass resolving power M/ΔM ≒ 4,000), the 15N16O− and 29SiH2− ions could not be separated. Therefore, we exclusively calculated the 15N16O− ion counts by subtracting the potential 29SiH2− ion counts. These latter have been estimated from the 28SiH2− ion counts based on the 29Si−/28Si− isotopic ratio (5.06 × 10–2). The Supplementary Figure S9 shows the time-dependence of the representative ion counts. In stishovite, the 15N16O− ion counts were higher than 28SiH2− ion counts, indicating that the contribution of the 29SiH2− to 15N16O− was negligible. Contrastingly, in bridgmanite, the 15N16O− ion counts were much lower than those of 28SiH2−. The estimated 29SiH2− ion counts were 65.1% of the original 15N16O− ion counts at maximum (Supplementary Fig. S9b). Therefore, in bridgmanite, we were able to estimate only the maximum possible concentration.

Figure 3 and Supplementary Fig. S10 present nitrogen incorporation rates in stishovite and bridgmanite, respectively, in both the Al-free and Al-bearing systems. These results suggest that the nitrogen solubility of stishovite is markedly higher than that of bridgmanite, for both systems. Although nitrogen solubilities of different silicate minerals from 15N16O−/30Si− cannot be compared directly because of a matrix effect, it is noteworthy that the ion intensity ratio of stishovite was one to two orders of magnitude larger than that of bridgmanite, in all recovered samples (Fig. 3 and Supplementary Fig. S10). The nitrogen solubility in stishovite in the Al-free system can be higher than that in the Al-bearing system (see Supplementary Fig. S11). Due to the low counts of 15N16O ions, the nitrogen solubility in bridgmanite between the Al-free and Al-bearing systems was difficult to evaluate.

Figure 3
figure 3

15N16O− to 30Si− count ratio in the Al-free system; samples recovered from (a) 28 GPa, 1,400 °C (OT2259); (b) 28 GPa, 1,500 °C (OT2258); (c) 28 GPa, 1,620 °C (OT2293); (d) 28 GPa, 1,700 °C (OS3083). St denotes stishovite and Brg, bridgmanite; n is the number of analyzed points; error bars represent the standard deviation. This figure was created by Ko Fukuyama using Microsoft Power Point and Excel (OFFICE 2016).

Full size image

Figure 4 presents the nitrogen solubility in stishovite, estimated using nitrogen-implanted quartz glass as standard samples. The nitrogen solubility in stishovite varied from 90 ppm (μg/g) to 404 ppm (μg/g). Stishovite can incorporate higher concentrations of nitrogen than bridgmanite (21.5 ± 18.1 ppm)15. In addition, the nitrogen solubility in stishovite increases with temperature. This implies that nitrogen might be incorporated in the stishovite crystal structure as N3− into oxygen vacancies or as N2 into Schottky defects13 (lattice point defects formed by unoccupied cation and anion sites in a stoichiometric ratio). Although oxygen vacancies might be more numerous in crystal structures at higher temperatures, substitutions between N3− and O2− require charge compensation. In stishovite, nitrogen is not expected to exist as NH4+ because NH4-bearing silicate minerals release nitrogen at temperatures of 750–1,000 °C20,33,34 much lower than those of our experimental conditions. Nitrogen solubilities in mantle minerals were investigated from high-pressure and high-temperature experiments: e.g. forsterite (7.1 ± 1.6 ppm), Al-bearing enstatite (50–100 ppm), wadsleyite (188.8 ± 15.6 ppm), ringwoodite (110.4 ± 109.8 ppm), bridgmanite (21.5 ± 18.1 ppm), and Ca-perovskite (28.3 ± 23.6 ppm)13,15. Nitrogen solubility in stishovite can be higher than in these minerals.

Figure 4
figure 4

Nitrogen solubility in stishovite at different temperatures. All points refer to single measurement points of the sample; errors were obtained using the square root of 15N16O− ion counts. This figure was created by Ko Fukuyama using Microsoft Power Point (OFFICE 2016).

Full size image

Discussion

Metamorphism of SiO2-rich sediments and nitrogen-containing continental crust form stishovite in subducting slabs. Moreover, Lee et al.35 reported that subduction erosion plays an important role in nitrogen transport into the mantle based on the nitrogen isotope data. The subduction erosion invokes the removal of continental material at the frontal or basal areas of continental margins. The mineral proportion of clinopyroxene, which can incorporate a significant quantity of nitrogen4,13 in subducted sediments and eroded continental crust, decreases with increasing depth36. In addition, NH4-bearing silicate minerals are unstable at high temperatures20,33,34. Therefore, stishovite, which can incorporate nitrogen at higher temperatures up to 1,700 °C, is the major mineral in the subducted sediment and, presumably, in the subducted eroded continental crust. Ingalls et al.37 reported that continental crusts can subduct into the mid upper mantle in spite of its high buoyancy. Moreover, stishovite has the highest nitrogen solubility in major minerals formed by the phase transition in subducting slabs. We propose stishovite as the major nitrogen carrier down to the transition zone and the lower mantle, if stishovite can inherit nitrogen from clinopyroxene4,13 or some NH4-bearing high-pressure minerals18 which are stable up to 10 GPa and 700 °C18,36. This P–T condition corresponds to a depth of 300 km where stishovite forms from metamorphism of the subducted sediments.

Based on the mass of subducted material, we estimated the mass of nitrogen exclusively transported by stishovite. Current subduction rates derived from sediments and continental crusts are, respectively, 1.65 km3/year and 2.1 km3/year, consisting of > 1.7 km3/year from subduction erosion and 0.4 km3/year from crustal subduction during continental collision38. Stishovite accounts for about 40 vol% of the subducted sediment and about 30 vol% of the continental crust, at pressures higher than 10 GPa or depths greater than 300 km36. Stishovite can incorporate up to 90 μg/g nitrogen at 28 GPa and 1,400 °C (see Fig. 4), corresponding to the present cold slab geotherm39. From these parameters, we estimated that stishovite can transport a maximum nitrogen quantity of 5.0 × 108 kg/year into the mantle transition zone and the lower mantle. If stishovite can transport nitrogen into the lower mantle, then almost all the transported nitrogen can be retained there. This is because the current nitrogen outgassing from the lower mantle [OIB (Ocean island basalt): 1.15 × 105 kg/year] is three orders of magnitude lower than the nitrogen amount transported into the lower mantle by stishovite28. Thus, the maximum estimation of nitrogen transported into the lower mantle ranged from 0.46 to 0.53 PAN (PAN: Present Atmospheric Nitrogen) since 3.6–4.2 billion years ago, when plate tectonics started40,41,42,43 although these start times of plate tectonics are controversial. Furthermore, outgassing C/N ratios (mol) from subduction (C/N = 5.2 × 103) and OIB (C/N = 7.1 × 103) are approximately one order of magnitude higher than that from MORB (Mid-ocean ridge basalt, C/N = 6.8 × 102)28,44. Bridgmanite and metallic iron in the lower mantle incorporate nitrogen15,23 and can store a maximum of 25 PAN15; this nitrogen storage capacity is greater than that of either the upper mantle or the transition zone15. Subduction is expected to play an essential role in the co-evolution between the atmosphere and the deep mantle, especially in the lower mantle. While metallic iron in the reduced lower mantle is expected to play a crucial role in reserving much nitrogen23, silicate minerals are the only nitrogen reservoir in subduction zones under oxidized conditions. K-hollandite and garnet are formed by the transition of the subducted sediments and continental crust in the mantle. These minerals can also incorporate nitrogen, as Watenphul et al.18 and Li et al.13 reported based on the experiments carried under lower pressures corresponding to those of the upper mantle. Although nitrogen solubilities in K-hollandite and garnet have not been estimated under the high-pressure corresponding to the lower mantle, the maximum of the nitrogen mass transported into the lower mantle via subducting slabs might be greater than the estimates solely based on stishovite transport. Compared to carbon, the formation of nitrogen reservoir in the lower mantle can deplete nitrogen effectively and might explain the “missing” nitrogen.

Nitrogen concentration in the Archean atmosphere remains controversial. Some researchers have reported that the partial pressure of nitrogen (pN2) was half as low as that of the present atmosphere45,46,47. Others suggested that the partial pressure of nitrogen (pN2) in the Archean atmosphere might have been 1.4–3 times higher than that at present. The whole Earth was not frozen in the Archean Earth despite the fainter Sun48. The higher pN2 in the atmosphere might reconcile the faint young Sun paradox2: greater nitrogen contents in the early Earth’s atmosphere might have enhanced the greenhouse effect of CO22. Goldblatt et al.2 suggested that nitrogen was transported into the deep Earth via subduction zones lowering the higher initial concentration in the atmosphere down to the present level. These results are supported by high-pressure experiments of Mallik et al. reporting that the Archean atmosphere had a 1.4–1.6 times19. Barry and Hilton49 also reported that the early atmospheric N content was approximately 50% higher than it is today based on a model using N–He–Ne–Ar isotope data Our results are consistent with the previous studies2,19,49 reporting that the mass of nitrogen in the ancient atmosphere was greater than that at present if the fluxes of nitrogen outgassing from the ancient and the present OIB were similar. Nitrogen transport into the lower mantle by stishovite and the formation of a deep nitrogen reservoir might have affected the global nitrogen cycle and the history of the Earth’s atmosphere. If the lower mantle could capture the nitrogen transported by stishovite, then a deep “hidden” nitrogen reservoir has formed in the lower mantle since the plate tectonics started (Fig. 5). Stishovite might have played a key role in the co-evolution of the atmosphere and the deep mantle.

Figure 5
figure 5

Deep “hidden” nitrogen reservoir formed by subducting slabs. UM the upper mantle, LM lower mantle, OIB oceanic island basalt.

Full size image

Methods

High-pressure experiments using a multi-anvil apparatus

High-pressure and high-temperature experiments were conducted at the Geodynamics Research Center, Ehime University, Japan. We used a Kawai-type 2000 tons multi-anvil apparatus (Orange-2000) and a Kawai-type 3,000 tons multi-anvil apparatus (Orange-3000) installed at the Geodynamics Research Center, Ehime University, Japan. All experiments were conducted at 28 GPa and 1,400–1,700 °C, for a 2 h heating duration.

We used tungsten carbide anvils (Tungaloy F-grade and Fujilloy F08) with 4 mm truncated edge length (TEL). Two kinds of starting materials were prepared: a powdered mixture of MgO and SiO2 (quartz) for the ideal bridgmanite composition (MgSiO3), and a powdered mixture of Al2O3, MgO, Mg(OH)2, and SiO2 (quartz) for the Al-bearing hydrous bridgmanite composition (MgSi0.9Al0.1H0.1O3). From the hydrous starting material, we intended to grow the large crystals of stishovite and bridgmanite necessary for SIMS measurements. To distinguish sample-derived nitrogen from atmospheric or instrumental contamination, more than 99.6% 15N-substituted ammonium nitrate (15NH415NO3, SHOKO SCIENCE Corp.) was used as nitrogen source. Contamination with atmospheric 15N is negligible because the 15N natural abundance is more than two orders of magnitude lower than that of the atmospheric 14N. Starting materials and 15NH415NO3 were enclosed in a Pt capsule. The starting material to nitrogen source mass ratio was approximately 5:1 for each experiment. For run no. OS3083, the starting material was separated from the 15NH415NO3 by 30 μm thick Au foil. For the other runs, the starting material and 15NH415NO3 were mixed. The Pt capsule was surrounded by a Fe–FeO buffer, which contained water to reproduce the oxygen fugacity corresponding to the lower mantle50,51,52 and enclosed in an Au outer capsule. The two Au capsules of the cell assembly were insulated from the Re heater by a 25 μm thick magnesia sleeve. We used 150 mesh iron powders and iron oxide (FeO) powders with particle sizes smaller than 200 mesh for the Fe–FeO buffer (Fe:FeO = 2:1, wt%); 20–50 μl of water were added per each 0.5 g Fe–FeO. As previously described, Fig. 1 depicts the cell assembly used for this study. Thermal insulation was provided by a LaCrO3 sleeve. The Pt sample capsule was produced combining two Pt tubes, with 0.1 mm thick walls and outer diameters of 1.3 and 1.5 mm, by welding each end of the capsules53. An Au capsule was made from an Au tube with 0.1 mm wall thickness and 2.5 mm outer diameter. The temperature was measured by a W–Re (W3%Re–W25%Re) thermocouple inserted in the octahedron, attached to the Au capsules. Hydrogen fugacity in the inner and outer capsules was assumed to be equal because Pt has a high hydrogen permeability compared to Au. At high temperatures, 15NH415NO3 decomposes into 15N2O and H2O; under the fO2 condition, 15NH3 is expected to form in 15N–H–O fluid under the fO2 condition in the inner Pt capsule.

The chemical composition of the quenched minerals was analyzed using a FE-SEM (JSM-7000F; JEOL) under operating conditions of 15 kV and 87.4–130.4 μA. For phase evaluation, Raman spectra were obtained using a micro-Raman spectrometer with an Ar ion laser of 514.5 nm and 6 mW power. Both characterizations were performed at the University of Tokyo, Japan.

Nitrogen analysis of silicate minerals using NanoSIMS

A quantitative nitrogen analysis was conducted using a NanoSIMS 50 (Cameca) installed at the Atmosphere and Ocean Research Institute, The University of Tokyo (AORI), Japan. This NanoSIMS has a magnetic sector mass analyzer with seven parallel detection systems to simultaneously detect up to seven ion species as secondary ions. We employed a Cs+ primary ion beam with a 2 nA current, 5 μm diameter, and a 5 μm × 5 μm or 10 μm × 10 μm raster. We detected nitrogen as 15N16O−, as described by Li et al.11 To avoid charge-up, an e− gun was used at 500–800 nA. Count times for the detection of 30Si−, 30SiH−, 28SiH2−, and 15N16O−were, respectively, 2 s, 2 s, 10 s, and 10 s. The time of analysis varied between 600 and 1,500 s. The post-bombardment crater volume was measured using laser scanning microscopy (Olympus Corp.).

We prepared ion-implanted standard samples to estimate the nitrogen concentrations in stishovite at the National Institute for Materials Science (NIMS). We used 14N-implanted quartz glass as the nitrogen standard for stishovite. The irradiation of 14N+ ions into the quartz glass was performed at 132 keV, with doses of 2.44 × 1014 and 3.66 × 1015 ions/cm2 (Ion Implanter RD-200l; Nissin Electric Co. Ltd.). Supplementary Figure S2 presents the calibration line for the nitrogen analysis using NanoSIMS.