Tracing the oxygen isotope composition of the upper Earth's atmosphere using cosmic spherules

Molten I-type cosmic spherules formed by heating, oxidation and melting of extraterrestrial Fe,Ni metal alloys. The entire oxygen in these spherules sources from the atmosphere. Therefore, I-type cosmic spherules are suitable tracers for the isotopic composition of the upper atmosphere at altitudes between 80 and 115 km. Here we present data on I-type cosmic spherules collected in Antarctica. Their composition is compared with the composition of tropospheric O2. Our data suggest that the Earth's atmospheric O2 is isotopically homogenous up to the thermosphere. This makes fossil I-type micrometeorites ideal proxies for ancient atmospheric CO2 levels.

F ree molecular oxygen (O 2 ) is released by photosynthesis into the atmosphere and is essential for all breathing animals. With exception of data for the last 800,000 years from air inclusions in polar ice, little direct information is available about concentration and isotope composition of ancient atmospheric O 2 . This is due to the limited interaction between air molecular oxygen and the lithosphere.
Among the rare rocky materials that contain atmospheric oxygen [1][2][3] there are particular types of micrometeorites (microscopic extraterrestrial dust particles) called cosmic spherules 4 . Roughly 10 tons of small extraterrestrial particles are deposited onto the Earth's surface per day 5 . The particles collide with the Earth's atmosphere at velocities of 11-70 km s À 1 (ref. 6) and are visible as shooting stars when they are decelerated and at altitudes up to B80-115 km 7,8 . A portion of these extraterrestrial particles totally melts during the atmospheric entry and is termed cosmic spherules. Cosmic spherules that are composed of Fe,Ni oxides are termed 'I-type cosmic spherules' (in the following, we use the short version 'I-type spherules' [9][10][11] . These I-type spherules formed by oxidation of extraterrestrial Fe,Ni metal alloys, which are ubiquitous components of meteorites. Because oxygen in I-type spherules originates entirely from the atmosphere, they are excellent probes for the isotopic composition of upper atmospheric oxygen. The isotopic composition of atmospheric oxygen, in turn, is a proxy for the global primary production (GPP) and atmospheric CO 2 levels 1,2,12-14 . It is not clear, however, if the atmospheric oxygen is isotopically homogenous up to the meso-and thermosphere, where cosmic spherules interact with air.
The stable isotope composition of tropospheric O 2 (99.8% 16 O, 0.04% 17 O, 0.2% 18 O) is controlled by the steady state between photosynthesis and respiration (mass-dependent Dole effect 15 ), evapotranspiration and mass-independent fractionation in the stratosphere 14,16 . For the composition of the modern troposphere values of 23.4rd 18 Or24.2% and À 0.566rD' 17 Or À 0.430% have been reported in the literature 14,[17][18][19][20] ( are due to evaporation and do not reflect the isotope composition of the upper atmosphere; a conclusion that was supported by further measurements 24,[26][27][28][29] . From the oxygen and iron isotope composition, Engrand et al. 24 modeled evaporative mass losses for I-type spherules of 54-85%. Because evaporative fractionation is strictly mass-dependent, I-type spherules still provide unique information about the mass-independent anomaly in D' 17 O of their upper mesospheric oxygen source. The reconstruction of variations in atmospheric D' 17 O from fossil cosmic spherules 4,30-33 would be a new paleo-CO 2 proxy. The only published D' 17 O data on I-type spherules by Clayton et al. 4 and Engrand et al. 24 , however, have intrinsic uncertainties that are too large (0.1 to 41%) to provide reliable information on the composition of the upper atmosphere.
We present new high-precision oxygen isotope data of tropospheric O 2 and compare these data with new high-precision oxygen and iron isotope data from Antarctic I-type spherules. These data are combined with results of oxidation and evaporation experiments to test if the Earth atmosphere is isotopically homogenous and if isotope ratios of fossil cosmic spherules are suitable paleo-CO 2 proxies.

Results
Oxygen isotope composition of tropospheric air. The mean composition of air oxygen from our study (series B; Supplementary  Table 2).

Discussion
The interaction between cosmic Fe,Ni metal and the Earth atmosphere during deceleration is considered to proceed in two consecutive steps. The first step is the atmospheric heating and oxidation of the infalling Fe,Ni metal alloy (fractionation in oxygen isotopes only). The second step is the melting and evaporation of the Fe,Ni oxides (fractionation in both, oxygen and iron isotopes).
Information about the oxygen isotope fractionation that is associated with the oxidation step is obtained from experiments (this study; see Methods) and from iron meteorite fusion crust data 4,35,36  For the experiments and iron meteorite fusion crusts, atmospheric oxygen is the oxidant. Above the ozone layer, however, a considerable fraction of molecular oxygen is steadily dissociated into atomic oxygen (for example, 38 ). Atomic oxygen is a hazard for low Earth orbit space flights due to its highly corrosive nature. Because I-type spherules are oxidized at high altitudes, atomic oxygen may have contributed to the oxidation. However, Clayton et al. 4 and Genge 10 stated that no discrimination between atomic and molecular oxygen is likely during oxidation upon atmospheric entry because the collision energy between infalling meteoroids and air particles is higher than the O 2 bond strength. This implies that I-type spherules sample the bulk upper atmosphere oxygen (O and O 2 ).
Our experiments and the fusion crust literature data 4 show that a oxidation o1 (for 18 O/ 16 O; see Equation 5), but also reveal considerable variation. For I-type spherules, we assume that 0.9428ra oxidation r1. The lower limit is given by pure Graham's law 39 fractionation with atomic oxygen being the moving species.
The second process affecting the isotopic composition of I-type spherules is evaporation 25 . Because iron isotopes are not affected by oxidation 35 but only by evaporation, d 56 Fe can be used as monitor for the degree of evaporation f (refs 24,25,40; Equation 1). The d 56 Fe of the infalling metal is assumed to be 0 ± 1% relative to the IRMM-014 standard material 41 .
Wang et al. 40 The difference in d 18 O between the spherules before evaporation and air oxygen gives the degree of fractionation during the oxidation step. We obtained values between À 22 and À 12% during oxidation of the infalling Fe,Ni alloys (Fig. 3). The results of the calculation are listed in Supplementary  Table 3 and illustrated in Fig. 4. The gas that oxidized the studied Antarctic I-type spherules had À 0.510rD' 17 Or À 0.420% (mean À 0.460 ± 0.020%; this study; 14,17 which is in excellent agreement with measured D' 17 O ¼ À 0.469% for the modern troposphere (Fig. 4).
Our data from oxygen and iron isotope analyses of Antarctic I-type spherules are consistent with an oxygen source with a D' 17 O similar to that of modern tropo-and stratospheric molecular oxygen within ± 0.02%. No oxygen reservoir with a markedly different D' 17 O participated in the oxidation of I-type spherules, suggesting that the Earth atmosphere is isotopically homogenous up to the mesosphere in B70-80 km (Fig. 5).
Our results imply that the oxygen isotope composition (D' 17 O) of the bulk atmosphere can be reconstructed from combined oxygen and iron isotope data of I-type cosmic spherules. This has an important implication for the reconstruction of past atmospheric CO 2 levels. Blunier et al. 42 showed that the D' 17 O of atmospheric molecular oxygen, indeed, varies with CO 2 partial pressures. This is predicted from experiments 12 and mass balance modeling 14 . The oxygen and iron isotope composition of unaltered fossil I-type cosmic spherules 4,30-33 will thus provide information on the D' 17 O of the ancient atmosphere and past CO 2 levels. The resolution of the calculated D' 17 O O2 isB0.07% (single I-type cosmic spherule; see Fig. 4), which translates (at modern GPP) to an uncertainty in the CO 2 mixing ratio ofB200 p.p.m. 14 . Lower than modern GPP levels would lead to an even higher resolution of calculated CO 2 levels. The reconstruction of CO 2 levels based on 17 O in I-type spherules is therefore considerably more precise than CO 2 reconstruction from 17 O of sulfate 1 and reaches far more back into Earth history than 17 O from air inclusions in ice cores 42 and fossil mammal bioapatite 2,13 . However, to use D' 17 O of air oxygen as paleo-CO 2 -barometer the GPP at that time needs to be known 14 . This may limit the usability of the new proxy. The apparent disadvantage, however, can be turned into a fortune. If the atmospheric CO 2 concentration is known from other, independent proxies 43 , D' 17 O of atmospheric molecular oxygen in combination with mass balance modeling 14 turns into a proxy for the GPP. There is little doubt that I-type cosmic spherules were deposited during the entire geological history of the Earth. The question is whether sufficiently large and unaltered fossil I-type spherules can actually be recovered from sediments. The recent find of unaltered 2.7 Ga old I-type cosmic spherules 33 is very promising in this respect.

Methods
Sampling and experiments. We studied a total of 21 aliquots of four samples for oxygen isotopes and one aliquot of the four samples each for iron isotopes. The samples are part of the Transantarctic Mountain collection 44 . The sample sizes range between 400 and 550 mm with masses between 160 and 370 mg. The samples were inspected for weathering products by optical microscopy. No weathering products (for example, brownish ferrihydride or goethite) were observed. The spherule densities were determined prior to crushing. Their diameters and masses were measured and from this their densities calculated (Supplementary Table 4). The I-type spherule densities vary between 4.3 and 4.8 g cm À 3 (Supplementary Table 4).
The density data confirm that the I-type spherules are composed of Fe,Ni oxides with little or no remaining Fe,Ni metal. The studied samples fall within the density range (5.0 ± 0.5 g cm À 3 ) observed by Feng et al. 45 ). Wüstite ([Fe,Ni]O 0.94 ) has a density of 5.7 and magnetite ([Fe,Ni] 2 FeO 4 ) of 5.2 g cm À 3 . In contrast, iron metal has a density of 7.9 g cm À 3 . The apparent lower density of the spherules compared to wüstite and magnetite is explained by B20 vol.% pore space.
The samples are all spherical due to melting during their atmospheric entry. The bulk elemental composition and the mineralogy of the studied spherules were not determined. Electron microprobe analysis of I-type spherules from the same collection yielded 91 ± 5 wt.% FeO, 2.8 ± 0.5 wt.% NiO, and MgO, Al 2 O 3 , and o0.5 wt.% SiO 2 (ref. 44). This composition is similar to results of Engrand et al. 24 and Herzog et al. 28 who report values of 92-93 wt.% FeO and 4-5 wt.% NiO.
For isotope analysis, spherules were wrapped in Al foil and gently crushed in a steel mortar. We obtained 21 aliquots (4-8 per spherule) with masses of 20-50 mg. As magnetite is the dominant phase in I-type spherules 24,46 , we used terrestrial magnetite along with NBS-28 quartz for tests. For NBS-28, we adopted a d 18 O ¼ 9.65% and D' 17 O ¼ À 0.054% (D' 17 O from 47 , with revision from 34 ).
I-type spherules form by oxidation of Fe,Ni alloys at high temperatures during their atmospheric entry. We conducted three metal oxidation experiments at the University of Göttingen to study the oxygen isotope fractionation associated with high temperature oxidation of metal in air. A powdered iron base alloy with 7.5 wt.% Ni and 0.6 wt.% Co was used as an analogue material for I-type spherules. Between 0.8 and 1.1 mg powder was placed on a ceramic plate in the hot zone of a Gero HTRV vertical gas-mixing furnace. The furnace was flushed with 300 ml min À 1 air. Oxidation occurred between 1,510 and 1,590°C for 30 min.
Air samples were taken at the Göttingen University North Campus outside the Geoscience Building (51°33 0 23 00 N 9°56 0 46 00 E). The air was taken from the balcony on the 4th floor using a 5 ml syringe yieldingB1 ml standard temperature and pressure O 2 gas.
Oxygen isotope analyses. Variations in stable oxygen isotope ratios of a sample are expressed in form of the d notation relative to the ratios in VSMOW2 water (Equation 3) with i standing for masses 17 and 18: The fractionation between two reservoirs (A, B) is expressed in form of the fractionation factor a (Equation 5). The reservoirs could be two phases in equilibrium or products (B) and educts (A) of a kinetic process.
The i in Equation 5 stands for isotopes with masses 17 and 18. For mass-dependent processes, the relation between a 17/16 and a 18/16 is linked through the triple oxygen isotope fractionation exponent y O (Equation 6).
For oxygen y O varies between 0.5000 and 0.5305 (refs 37,48,49). Only for very small a values, y values may fall outside the 0.5-0.5305 range 50 . Such effects are neglected here. As a rule, low y values are associated with kinetic effects, whereas higher y values are associated with equilibrium fractionation processes. The triple oxygen isotope ratios of I-type spherules and the high-T oxidation experiment run products were analysed at the University of Göttingen on O 2 extracted by infrared laser fluorination 51 , following the protocol described in Pack et al. 2 . In brief, sample O 2 was liberated by laser fluorination (F 2 ) and analysed in continuous flow mode in a Thermo MAT253 gas source mass spectrometer. NBS-28 quartz was used for normalisation relative to VSMOW2 scale (d 17 O ¼ 5.04%, d 18 O ¼ 9.65%, D' 17 O ¼ À 0.054%; using the revised calibration of San Carlos olivine from 34 ). The total masses of the spheres were 160-370 mg and thus suitable for measurement of multiple aliquots of a single spherule. For the oxidation experiments, B0.6-1 mg aliquots were analysed.
Oxygen from air was extracted using the same line that was described by Pack et al. 34 for their water (including VSMOW2 and SLAP2) and silicate analyses (San Carlos olivine) (Fig. 6).
For each extraction, 5 ml aliquots of air standard temperature and pressure (STP) were injected through a liquid nitrogen cooled glass U-trap (for removal of moisture and CO 2 ; 'trap 7' , Fig. 6). The dry, CO 2 free mixture of Ar, N 2 and O 2 was transferred to 'trap 2' that was filled with 5 Å molecular sieve pellets. In an early protocol (S01-S05; Supplementary Table 1), Ar was separated from O 2 at À 100°C, which resulted in very long trapping times. In an improved protocol (B01-BP2; Supplementary Table 1), separation of Ar was performed using the cryo unit of the Hewlett-Packard 5890 gas chromatograph at À 80°C. After Ar had passed through the gas chromatograph (monitored using a Pfeiffer Prisma  14,[18][19][20]54 ). These data agree well with the mesospheric I-type spherule proxy data from this study (solid red squares with 1s-error bars).
quadrupole mass spectrometer at the end of the He capillary; 'He out' in Fig. 6), temperature was raised to À 30°C for the elution of O 2 and separation of N 2 . The improved protocol is similar to the protocol described by Young et al. 14 . The purified O 2 was analysed forB60-90 min in dual inlet mode.
To test accuracy and precision of laser F 2 in combination with continuous flow mass spectrometry of small samples, a set of experiments were performed on magnetite (as analogues of I-type spherules) and NBS-28 quartz (Fig. 7).
Our tests on NBS-28 quartz and magnetite (Fig. 7) show that precise analyses of d 18 The symbols A and B in Equation 8 can either stand for two phases that are in equilibrium or for educts (B) and products (A) of a reaction with associated kinetic fractionation. The high-T approximation for equilibrium iron isotope fractionation is y Fe ¼ 0.6784 (ref. 37). As in case of oxygen, variations in y Fe provide insights into the fractionation process. The iron isotope compositions were measured at the University of Bonn using a Thermo Scientific Neptune MC-ICP-MS instrument and glassware for sample introduction. Samples were measured 2-3 times, non-consecutively, during long analytical sessions of around 8 hours. Analyses were carried out in high-resolution mode with sufficient transmission to allow routine analyses of an 1 p.p.m. iron sample solutions. Each sample measurement was bracketed by two analyses of an IRMM-014 iron solution that was made up to closely match the iron concentration of the sample. The external reproducibility of the data was typically ± 0.07%, whereas the internal reproducibility was generally about a factor of two better. A detailed description on sample preparation and mass spectrometry can be found in Hezel et al. 53 and Hezel et al. 36 .
Monte Carlo simulation. The errors in d 18 O and D' 17 O of the upper atmospheric oxygen were estimated using a Monte Carlo approach. The composition was computed 500 times. Each input parameter was varied independently for each run (Supplementary Table 5). We used a normal distribution for the variation within the respective error interval. The computation was performed using the Mathematica software.
Data availability. The authors declare that the data supporting the findings of this study are available within the paper and its supplementary information.