Introduction

The carbon isotopic signature of the present–day upper mantle that feeds mid–ocean ridge basalts (MORB) is a crucial value for understanding planetary–scale processes, such as (i) the origin of Earth’s carbon1,2, (ii) the efficiency of carbon recycling at subduction zones3,4 and (iii) the sources of volatiles emitted to the atmosphere by volcanoes5,6,7. However, direct measurements of MORB mantle δ13C values are lacking. Our most reliable estimate is based on carbon isotope measurements of CO2 gas trapped in vesicles from two ‘popping rock’ samples (2πD43 and OT03–09) collected at 14 °N and 34 °N in the Mid–Atlantic Ridge8,9,10. These rare samples have been interpreted as closed systems with vesicles retaining all CO2 degassed from the melt during magma ascent and both indicate an initial MORB mantle δ13C of –4‰ (all δ13C values given relative to Pee Dee Belemnite). A second estimate of the MORB mantle carbon isotope signature comes from studies of peridotitic diamonds. These samples are not from mid–ocean ridges but are collected from kimberlites and lamproites around the world. As such, they are more representative of the upper mantle beneath cratons and provide samples ~1 to 3 Ga old11, much older than today’s MORBs. Nevertheless, δ13C measurements on peridotitic diamonds are widely used to characterise the carbon isotopic signature of the present–day upper mantle indicating a mantle δ13C value of –5‰12,13,14. A third estimate of the MORB mantle carbon isotope signature comes from studies of volcanic gases emitted by volcanoes in the Afar Depression, a region where the Aden and Red Sea Ridges meet the East African Rift. Gas samples collected during the 1978 Asal Rift eruption15 and at Erta Ale volcanoe in the seventies16 show δ13C values around –6.5‰. These values however, although sometime cited as such, are not directly representative of the mantle δ13C because they represent surface degassing and the isotopic fractionation between gas and melt during magmatic degassing is significant17,18,19,20.

Here we present a novel, more direct approach to determining the isotopic signature of the MORB mantle by studying a series of CO2–undersaturated MOR melts that preserve direct isotopic equilibrium with their mantle source. We take advantage of the recent development of high precision δ13C SIMS analyses in basaltic glasses21 and present the first such measurements on a series of melt inclusions (a few δ13C analyses of melt inclusions have been reported by previous workers22 but with very high uncertainties around ±5‰). We investigate olivine–hosted melt inclusions from the glassy rims of a pillow basalt erupted at >2,650 m depth from the South West Indian Ridge (SWIR; 37 °51′48″S, 49 °20′12″E). These samples have previously been characterised for major, minor, trace and volatile elements23 and have been shown to record one of the few occurrences of CO2 undersaturated melts, magmas that never reached saturation with a CO2 gas phase23,24,25,26. All melt inclusions analysed were large (87–440 µm diameter), four out of ten contained shrinkage bubbles, which were always less than 1% of the MI volume, and none had evidence of decrepitation or visible cracks prior to polishing (Fig. 1 and Supplementary Fig. 1).

Fig. 1: SWIR CO2-undersaturated melt inclusions δ13C signature.
figure 1

CO2 content versus δ13C (A), CO2/Ba versus δ13C (B) and Rb/Nb versus δ13C (C) for SWIR melt inclusions. Inclusions containing a shrinkage bubble are marked with a white dot in their centre. Repeat analyses performed on the same inclusion are marked with a black cross. Lower right inset is a transmitted light photomicrograph of a typical olivine crystal with three large glassy melt inclusions. Error bars are one standard deviation.

Full size image

Results and discussion

Ten melt inclusions were analysed, one of which was large enough for three replicate analyses (Table 1). The error on repeat analyses in this inclusion (1σ = ± 0.8‰; n = 3) is comparable to the analytical error of each individual analysis (average 1σ = ± 1.3‰) indicating no intra-inclusion heterogeneity within analytical error. We found no correlation between the CO2 content and the δ13C signature of the melt inclusions, as expected for undersaturated melts that have not undergone CO2 degassing and therefore not experienced carbon isotope fractionation due to CO2 degassing (Fig. 1). Supplementary Fig. 2 illustrates the predicted relationship between δ13C and CO2 for both closed and open system styles of degassing, indicating that neither model aligns with our observations. Furthermore, degassing is incompatible with the uniform CO₂/Ba ratio observed across all our melt inclusions. Note that while ten melt inclusions were analysed for δ13C only seven were analysed for trace elements by LA-ICPMS (analyses from ref. 23). We found no significant correlation between the MI δ13C and any major, minor, trace or volatile element concentrations. Since the MI trace elements are generated by varying degrees of mantle partial melting23 this suggests that δ13C does not fractionate during partial melting, confirming a long–standing but rarely stated assumption. Finally, we found no correlation between the δ13C of the MIs and their oxidation state (Supplementary Fig. 3). The close correspondence between the analytical error on each analysis ( ± 1.5‰ on average) and the standard deviation across the entire population ( ± 2.2‰), together with the lack of correlation with other variables, serves as evidence in favour of treating our measurements as multiple analyses of a singular unknown. Taken as a group, our MIs δ13C analyses define a normal distribution (kurtosis= –0.6; skewness = –0.04) with a mean δ13C signature of –7.7‰, a median of –8.0‰, a weighted mean of –7.5‰ and a 95% confidence level of ±1.4‰. Inclusions containing a shrinkage bubble are evenly distributed in the population in term of both their δ13C value and CO2 contents, excluding them would only slightly shift the population mean to lower values (–8.0‰, median of –8.3‰, weighted mean of –7.8‰). Since these inclusions have the same CO2/Ba ratio as those without bubbles (Fig. 1), it is reasonable to assume that they have not lost any CO2 to their shrinking bubble.

Table 1 δ13C measurements in SWIR melt inclusions by SIMS (see method section)
Full size table

Taken together, our measurements hence indicate a SWIR mantle source δ13C at –7.5 ± 1.4‰. This is different from the –4‰ mantle source determined from the Mid Atlantic Ridge (MAR) popping rocks8,9. The statistical significance of this difference given our measurement population can be tested, giving a Z–score of 1.65, which translates into a 5% probability that the true value is equal to or greater than –4‰ (using the population mean and standard deviation). In other words, we have 95% confidence in the fact that the SWIR mantle source is lighter in carbon isotopes than the previously accepted modern MORB mantle (89% confidence when compared to the –5‰ value of peridotitic diamonds). We therefore argue that the modern–day upper mantle must be heterogeneous in terms of its carbon isotopic signature.

Why is the SWIR mantle lighter in δ13C and how much heterogeneity in mantle δ13C should we expect? One feature of the mantle beneath the South West Indian Ridge is that it has been recognised as having been modified by a subduction component27,28, which is not the case for the section of the Mid–Atlantic Ridge from which the popping rocks originated28. A reasonable hypothesis, therefore, would be that regions of the mantle which are most affected by subducting slab components would be expected to show greater variability in their δ13C signature. Yang et al.28, proposed a trace element systematic to identify the influence of subduction fluids on MORB mantle sources. Following their approach, most MORB show little influence of a subduction fluid-contaminated component in their source mantle, with Ce/Pb vs. Nb/U ratios close to the “canonical” values (of ~25 and 47 respectively, Fig. 2) so we would expect their source δ13C values to be close to those of the MAR popping rocks (at ~ –4‰). Our SWIR samples however, fall along the trend of “back-arc-like MORBs”28, MORBs whose source has been influenced by subduction fluids. We can therefore develop a two-component mixing model between two mantle sources, one highly metasomatized by subduction fluids and one without subduction influence. We use the global trend to identify the subduction fluid metasomatized endmember (Fig. 2). Assuming a fluid-immobile character of Nb, the global minimum value of Nb/U ( ≈ 2) is expected to indicate the most likely value of the subduction fluid metasomatized mantle endmember. The “normal” MORB average (with Ce/Pb and Nb/U ratios of ~25 and 47) is taken as the other endmember, and the mixing line constrained by the average SWIR DR 75 melt inclusions, is shown in Fig. 2 (see supplementary methods for modelling detailed). The corresponding concentrations of magma derived from the metasomatized mantle source are, Pb = 0.020 ppm, and Ce = 2.46. The δ13C values of the metasomatized mantle endmember can be derived from our mixing model using a δ13C value of –7.5 ± 1.4‰ and a CO2/Nb ratio of 1134 ± 41523 for SWIR and using a δ13C value of –4‰ and a CO2/Nb ratio of 557 ± 7925 for the “normal” MORB source mantle. This would give a δ13C value of –10‰ for the metasomatized endmember (Fig. 3).

Fig. 2: Trace element systematics of global and SWIR MORBs.
figure 2

Nb/U versus Ce/Pb diagram showing the distribution of MORB melt inclusions25,47,48,49,50, glasses28 and whole rock28. Vertical blue line marks the furthest extent of the “back-arc-like MORBs”28 trend to minimum value of Nb/U ( ≈ 2). Yellow line shows modelled mixing between the “normal” MORB and a subduction-fluid metasomatized mantle endmember passing through the average composition of this study and previously published DR75 melt inclusions23, indicated as black points with error bars (1 standard deviation).

Full size image
Fig. 3: The influence of subducted carbon on MORB mantle δ13C signature.
figure 3

Nb/U versus δ13C showing modelled mixing lines between the “normal” MORB mantle and three possible subduction-fluid metasomatized mantles endmembers. Error bars on the SWIR average (large circle symbol) show 95% confidence intervals based on 31 analyses for Nb/U from ref. 23 and 9 analyses for δ13C, this study. Error bars on individual analyses are one standard deviation analytical errors and their propagation for the Nb/U ratio. Horizontal dashes show the 25, 50 and 75% of mixing along each mixing lines. The SWIR melt mantle δ13C can be explained by ~80% mixing between the “normal” MORB mantle and a metasomatized mantles component with a δ13C = –10‰ signature.

Full size image

We expect that other ridge systems influenced by subduction zone fluids, such as the Chile Rise, the Gakkel Ridge and the southernmost part of the Mid–Atlantic ridge near the Bouvet Triple Junction to sample mantle with a δ13C signature distinct from –4‰. However, as the bulk δ13C of subducting slabs can vary widely, depending on the relative proportions of organic carbon (at ~ –25‰), altered oceanic crust (at ~ –5‰) and marine carbonate (at ~0‰) components29, a simple relationship between tracers of slab influence and mantle δ13C is not necessarily expected (Fig. 3). Therefore, the metasomatized mantle component we infer here (with a –10‰ δ13C signature) is unlikely to be globally widespread; instead, we envision the existence of several such components with specific carbon isotopic signatures, mixing with the normal MORB mantle (with Fig. 3 depicting the possible ranges). These metasomatized mantle components should also get episodically sampled by mantle plumes and result in a variety of mantle δ13C signatures at hotspot volcanoes. Indeed while studies at the Kilauea30, Réunion31 and Society32 hotspots all point to source δ13C values around –3 to –4‰, CO2 rich ( > 3000 ppm) vesicles found in submarine basalts from the Pitcairn hotspots clearly show a distinct source δ13C values around –6‰33 while minerals hosted fluid inclusions in mantle xenoliths from the Canary Hotspot (El Hierro) show initial δ13C values around 0‰34.

Decades of geochemical studies of MORBs have shown that the upper mantle is heterogeneous in isotope and trace element systematics (such as Hf and Nd35, Fe and Mg36,37,38 and Ca39,40 among many others41,42,43), our data show that this heterogeneity also applies to the upper mantle carbon isotopic signature. Precise estimate of carbon fluxes in and out of the mantle and of the size of the mantle carbon reservoir through geological time are difficult to obtain. Assuming an influx of carbon from subduction to the mantle in the order of ~1013 g of C/yr44, a outflux of carbon from the mantle of similar magnitude and a reservoir size of ~1023 g of C44 and further assuming that these fluxes are representative of ~3.5 billion years of subduction on earth one would predict that less than 0.1% of the carbon in the Earth’s current mantle is recycled from subduction. The fact that we identify a heterogeneity in the mantle carbon isotopic composition suggests either that the current flux and/or reservoir size estimates are incorrect, or that subducted carbon is not easily mixed with the rest of the mantle. The latter view is consistent with the subduction recycling model, in which distinct subduction lithologies are preserved and preferentially contribute to chemical heterogeneities of mantle derived magmas45. An alternative explanation would be that the earth’s mantle is heterogeneous in its primordial carbon isotopic composition.

Methods

We used olivine–hosted melt inclusions, doubly polished down to 0.25 μm with corundum mats prepared in ref. 23. These samples were mounted in indium such that the MI side exposed to the beam would be the one on which no previous SIMS nor EMPA analyses were obtained and were no previous metallic coating had been applied.

Carbon isotopes were measured using a CAMECA ims–1270 at CRPG (Nancy, France). The area for analysis is chosen away from any cracks or gaps on the surface of the sample. The samples were sputtered with a Cs primary beam with presets ranging from 1 nA to 3 nA and the extracted ions were accelerated by a 10 kV potential. Depending on the intensity of the signals on the standards and samples, the primary intensity was adjusted to maintain the signal of 13C in the electron multiplier detector (i.e. <500 000 cts/s). For the samples the primary intensity was held at 3 ± 0.5 nA while for the standards (given higher variability in their CO2 content) the primary intensity varied from 0.8 to 3.6 nA. As shown in Fig. S4B the instrument mass fractionation (as determined from the glass standards) is constant over this beam intensity range. A pre–sputtering of 120 s with a 15 µm grid was set to avoid any surface contamination. A 10 µm beam, centred on the gridded area, was then used for analysis. Contamination of the carbon isotopic signal from instrument background can be assessed using the natural MORB standard EDUL_DR52. This sample has a CO2 concentration of ~350 ppm, comparable to our lowest sample (~300 ppm) and four times lower than our highest ( ~ 1200 ppm). This sample has a δ13C signature characterised by step−heating extraction and isotope-ratio mass spectrometry analysis of –8.4 ± 0.3‰ (error is one standard deviation on three repeat analyses)46. Our analysis of this sample (without including it to determine the IMF hence treating it entirely as an unknown) yields a δ13C –8.1 ± 1.0‰ (error is one standard deviation on five repeat analyses). The fact that we find a δ13C value for a sample with CO2 content comparable to our lowest sample undistinguishable from the reference value, shows that for all our samples any potential background contamination would have been at a level lower than our uncertainty.

Secondary negative ions 12C and 13C were detected by an axial electron multiplier (EM) using a magnetic peak switching method. 18O was collected in a Faraday cup (in FC2). We used mass 11.8 for the background measurement of the EM and mass 17.8 for the background of the FC. The counting times were 4 s (EM background), 4 s (12 C), 20 s (13 C), 4 s (FC2 background) and 2 s (18 O). The wait times between the masses were 3, 1, 1, 1 and 1 s, respectively, and a dead time correction of 89 ns was applied. For each measurement, 30 cycles were collected, so each analysis took about 30 minutes. Using the analytical conditions in Table S1, the mass resolving power (MRP) was 5000, which is sufficient to discriminate mass 13C from mass 12C1H. Instrument mass fractionation was characterised using a series of eleven synthetic standards of MORB composition (CI_Ref_6, CI_Ref_9, CI_Ref_10, CI_Ref_11, CI_Ref_15, CI_Ref_19, CI_Ref_20, CI_Ref_22, CI_Ref_23, CI_Ref_25, and CI_Ref_27) and one natural MORB standards (EDUL_DR52) all described in details in ref. 21 (Fig. S4). With a primary beam of 1.5 nA on a ~ 5000 ppm CO2 glass (Cl_Ref_27), 270,000 counts per second (cps) of 12C and 2900 cps of 13C were measured, corresponding to a secondary ion intensity of ~40 cts/ppm C/nA for 12C and ~0.4 cts/ppm C/nA for 13C.