Volume 21, Issue 10 e2020GC009081
Research Article
Free Access

Isotopic Evidence for Multiple Recycled Sulfur Reservoirs in the Mangaia Mantle Plume

J. W. Dottin III

Corresponding Author

J. W. Dottin III

Department of Geology, University of Maryland, College Park, MD, USA

Correspondence to:

J. W. Dottin III,

[email protected]

Search for more papers by this author
J. Labidi

J. Labidi

Institut de Physique du Globe de Paris, CNRS, Université de Paris, Paris, France

Search for more papers by this author
M. G. Jackson

M. G. Jackson

Department of Earth Science, University of California, Santa Barbara, CA, USA

Search for more papers by this author
J. Woodhead

J. Woodhead

School of Earth Sciences, University of Melbourne, Parkville, Victoria, Australia

Search for more papers by this author
J. Farquhar

J. Farquhar

Department of Geology, University of Maryland, College Park, MD, USA

Earth System Science Interdisciplinary Center, College Park, MD, USA

Search for more papers by this author
First published: 23 September 2020
Citations: 9


Mangaia, an ocean island in the Cook-Austral volcanic chain, is the type locality for the HIMU mantle reservoir and has also been shown to exhibit evidence for recycled sulfur with anomalous δ34S and Δ33S that has been attributed an Archean origin. Here we report bulk S-isotope data from sulfide inclusions in olivine and pyroxene phenocrysts from one of the previously analyzed and four additional Mangaia basalts to further test for the prevalence of anomalous S in the HIMU mantle source feeding Mangaia. We document compositions that range from −5.13‰ to +0.21‰ (±0.3 2σ), +0.006‰ to +0.049‰ (±0.016 2σ), −0.81‰ to +0.69‰ (±0.3 2σ) for δ34S, Δ33S, and Δ36S, respectively. These data extend the range of measured compositions and suggest S-isotope heterogeneity in the HIMU mantle source at Mangaia. We show that S-isotope compositions of bulk sulfide in olivine is not in isotopic equilibrium with bulk sulfide in pyroxene from the same samples and that samples from a confined area (M4, M10, M12, and M13) in the northern central part of the island show a distinct covariation for δ34S and Δ33S. This isotopic variation (forming an array) suggests mixing of sulfur from two sources that were captured at different stages of crystallization by phenocrysts in the Mangaia HIMU sulfur endmember.

Key Points

  • We report bulk S-isotope compositions of sulfide in melt inclusions of olivines and pyroxenes from Mangaia
  • Isotopic disequilibrium in a single sample records a process of trapping various compositions during magma mixing
  • Data collected identify a new, potentially young, recycled component with negative δ34S and slightly positive Δ33S in HIMU source

Plain Language Summary

The sulfur isotope composition of the mantle is heterogenous from crust formation and recycling continental and oceanic crust. Basalts, from an intraplate volcanic hotspot volcano in the Pacific Ocean, Mangaia, show isotopic evidence for hosting Archean-related sulfur that is sourced from a recycled oceanic crustal component. The prevalence of this Archean-related sulfur in the plume source that feeds Mangaia is not well constrained. In this work, we present sulfur isotope data on sulfur from minerals that crystallized deep in the magma chamber and identify a new, potentially younger, recycled sulfur source that demonstrates the diversity of materials mixing in the magma chamber prior to eruption.

1 Introduction

Covariations in the radiogenic isotope signatures of strontium, neodymium, and lead seen in various Mid-Ocean Ridge Basalts (MORB) and Ocean Island Basalts (OIB) have been used to identify distinct compositional reservoirs that exist in the mantle as a result of melt removal and recycling of oceanic and continental crust. These reservoirs have been termed Depleted MORB Mantle (DMM), Enriched Mantle I and II (EM I, EM II), and HIMU (high μ = 238U/204Pb). Depleted MORB mantle is characterized as having relatively low 206,207,208Pb/204Pb and 87Sr/86Sr but high 143Nd/144Nd and argued to have formed from the process of melt removal during crust formation (Salters & Stracke, 2004; Workman & Hart, 2005). HIMU is characterized by high 206,208Pb/204Pb (and variable 207Pb/204Pb), low 87Sr/86Sr, and intermediate 143Nd/144Nd. The HIMU reservoir is thought to have formed from the recycling of oceanic crust that experienced Pb loss from sulfides during subduction followed by the ingrowth of radiogenic Pb due to a high 238U/204Pb (“high μ”) time integrated ratio (e.g., Chauvel et al., 1992; Kelley et al., 2005) with a wide range of proposed formation ages of 550 Ma to 3.0 Ga (Hauri & Hart, 1993; Mazza et al., 2019; Nebel et al., 2013; Thirlwall, 1997).

Mangaia, located in the Cook-Austral volcanic chain, is the type locality for the HIMU mantle reservoir that, together with Bermuda, preserve the highest reported radiogenic 206Pb/204Pb isotope compositions of any OIB locality suggesting an old and recycled mantle source (Hauri & Hart, 1993; Mazza et al., 2019; Woodhead, 1996). Jackson and Dasgupta (2008) and Cabral et al. (2014) argue that HIMU basalts are associated with a carbonated mafic source on the basis of major element characteristics (low SiO2 and high CaO/Al2O3) and elevated volatile contents of CO2. Weiss et al. (2016) suggest that the HIMU mantle source of Mangaia was metasomatized by carbonatitic fluids and imply a connection to recycling processes. The sulfur isotope composition of sulfides in sulfide inclusions trapped in magmatic olivine from Mangaia basalts has been measured in situ using secondary ion mass spectrometry (SIMS) and also using conventional bulk isotope methods (Cabral et al., 2013). The δ34S values reported from Mangaia (~−17‰ to −3‰) are depleted in heavy isotopes compared to bulk sulfur isotope measurements of MORB (Mid Ocean Ridge Basalt) glasses (average δ34S of Pacific-Antarctic basalts = −0.89 ± 0.11 (1σ), Labidi et al., 2014) and additionally show small but significant negative Δ33S in two samples by Secondary Ion Mass Spectrometry (SIMS), down to −0.24 ± 0.15‰ (2σ) (weighted average = −0.25 ± 0.07‰, 2σ) and −0.35 ± 0.23‰ (weighted average = −0.34 ± 0.08‰) in MGA-B-25 and MGA-B-47, respectively. Taken together, the data collected via SIMS and conventional bulk S-isotope measurements reveal compositions that are consistent with a subducted component in the source of Mangaia melts that hosts Archean sulfur that experienced mass-independent fractionation (Cabral et al., 2013). What remains unanswered is whether anomalous Archean S is common in the HIMU mantle reservoir sampled by Mangaia or whether other Mangaia basalts contain what might be considered more typical mass-dependent sulfur.

Here, we present bulk S-isotope compositions (δ34S, Δ33S, and Δ36S) of sulfide inclusions within magmatic olivine and clinopyroxene phenocrysts from five Mangaia samples. These olivine and clinopyroxene phenocrysts trap sulfide inclusions during their growth at depth in the magma chamber and, like melt inclusions (e.g., Sobolev & Shimizu, 1993), they may shed light on melt mixing of diverse sources. Our data provide new constraints on the various recycled components that dominates HIMU mantle melts feeding lavas at Mangaia.

2 Samples and Methods

We present new multiple sulfur isotope data on five basaltic samples from Mangaia: MGA-B-25, M4, M10, M12, and M13. Samples are from a road cut along the northeast side of the island (MGA-B-25) and from Vau Roa stream Valley (M4, M10, M12, and M13). Samples previously analyzed (Cabral et al., 2013) using SIMS and discussed in this manuscript are from a road cut adjacent to a stream north of Tamarua village on the south side of the island (MGA-B-47) and a small outcrop revealed by a road cut near a stream on the south side of the island (MG1001B). The work by Cabral et al. (2013) also presents a bulk sulfide measurement from an olivine separate of MGA-B-47. Further, note that MGA-B-25 was analyzed in Cabral et al. (2013) but only using SIMS methods. We analyzed newly prepared aliquots of minerals from MGA-B-25, while M4, M10, M12, and M13 (n = 4) are samples that have not previously been analyzed for sulfur isotopes. With these new measurements we are able to (1) test for S-isotope heterogeneity at different locations on the island (possibly related to separate flows) and (2) evaluate the prevalence of negative Δ33S in MGA-B-25 and Mangaia basalts generally.

2.1 Acid Digestion

All samples are olivine and clinopyroxene phenocryst separates that were hand-picked and cleaned by sonication in preparation for S-isotope analyses. Crushed and weighed Ol and Px mineral separates were placed into Teflon reaction vessels for hydrofluoric acid digestion using methods developed by Labidi et al. (2012) for the liberation of sulfide from silicates. Note that these methods have demonstrated the ability to digest and consistently receive 100% ± 5% of S as sulfide powders as well as sulfides dissolved in silicate melts (Labidi et al., 2012). Sulfides were extracted in this study, as the method used here only extracts sulfur of reduced valence, leaving behind sulfate if present in the samples. The digestion apparatus was first purged for ~15 min with N2. After purging, the samples were acidified and digested in a heated solution (70–80°C) of 3.2 M CrCl2, 12 M HCl, and 29 M HF in the amounts of 10, 5, and 5 ml, respectively. During this digestion, sulfide was released as H2S, bubbled (~3–5 bubbles every ~1–2 s) through a water trap to remove acid vapors, before being bubbled and trapped in weakly acidic AgNO3 where S is precipitated as Ag2S. The Ag2S was centrifuged and rinsed three times with Milli-Q water. After rinsing, the Ag2S was dried at ~70°C and subsequently weighed to determine sulfur concentration.

2.2 Fluorination

The weighed Ag2S was wrapped in foil and placed into nickel reaction tubes three to five times excess F2. Reactions between Ag2S and F2 occurred at ~250°C for at least 12 hr. The product SF6 was separated from residual F2 by freezing into a liquid-nitrogen-cooled trap. Residual F2 was passivated through a reaction with heated (~110°C) KBr salt. After all fluorine was passivated, the liquid nitrogen trap was replaced by an ethanol slush (~−108°C to −115°C) to retain cocondensed HF, and the SF6 was sublimated into a separate liquid-nitrogen-cooled trap, from which it was injected into a gas chromatograph (GC) with a 1/8-inch diameter, 6-foot long Haysep-Q GC column and a 5A GC column using a helium flow rate of 20 ml/min. SF6 was monitored while passing through the GC and captured in liquid nitrogen cooled metal coils. Once captured, the SF6 was thawed to determine procedural yields as pressure and, lastly, stored in individual glass manifolds.

2.3 Mass Spectrometry

Isotopic variations of the four stable isotopes of sulfur (32S [95.2%], 33S [0.75%], 34S [4.25%], and 36S [0.02%]) were determined using gas source mass spectrometry of purified SF6. Measurements were made by monitoring SF5+ ion beams at m/z of 127, 128, 129, and 131 of the purified SF6. Each sample underwent three measurements that consists of eight 26-s cycles measuring the sample and the reference gas. Data are reported in per mil using the following notation:

Analyses are normalized to analyses of Canyon Diablo Troilite on a scale that place IAEA-S1 at a composition of δ34S = −0.401, Δ33S = 0.116, and Δ36S = −0.796 (Antonelli et al., 2014). For normalization, raw delta values from sample analysis are converted to ratios and normalized to the isotope ratio associated with measurements of CDT (Antonelli et al., 2014). Δ33S and Δ36S are calculated from the CDT normalized delta values. Uncertainties on measurements are estimated from the long-term reproducibility on measurements (including wet chemistry, fluorination, and mass spectrometry) of standard materials and are 0.3, 0.016, and 0.3 (2σ) for δ34S, Δ33S, and Δ36S, respectively. As a test of our methods and uncertainty, we have analyzed MORB glasses that were previously analyzed in Labidi et al., (2014), namely, PAC2 DR 32-1G, PAC2 DR 06-6G, and PAC1 CU 03-3G, and one MORB glass that is not reported, PAC2 DR 20-04 (see supporting information Table S1). When a small difference in the calibration between the two labs (CDT here vs. a V-CDT calibration described in Labidi et al., 2014) are taken into account, the data are reproducible and within uncertainty of measurements between both labs. The values of CDT normalized to IAEA S-1 in the two studies are δ34S = −0.401 ± 0.15‰, Δ33S = +0.116 ± 0.008‰, and Δ36S = −0.796 ± 0.080‰ for UMD and δ34S(V-CDT) = −0.14 ± 0.52, Δ33S = +0.106 ± 0.008, and Δ36S = −0.782 ± 0.140 for Labidi et al. (2014) (all 2SD).

3 Results

Bulk sulfur isotope compositions of sulfide minerals hosted in olivine and pyroxene mineral separates from Mangaia basalts are reported in Table 1 and illustrated in Figures 1 and 2. We extracted 0.54 mg to 1.57 mg of Ag2S from ~300 to 400 mg of sample resulting in sulfur concentrations that range from 96 to 333 ppm. S concentration in Ol and Px from the same sample does not match and shows no systematic S concentration in olivine versus pyroxene separates (Table 1).

Table 1. S-isotope Compositions of Bulk Sulfide in Ol and Px Separates From Mangaia Basalts
Sample Rock type S (ppm) δ34S Δ33S Δ36S 206Pb/204Pb 207Pb/204Pb 208Pb/204Pb Latitude Longitude
MGA-B-25 Px Ankaramite 290 −5.1 0.028 0.0 21.678 15.807 40.512 −21.9133 −157.9144
MGA-B-25 Ol Ankaramite 333 −4.0 0.016 −0.3 21.678 15.807 40.512 −21.9133 −157.9144
M4 Px Ankaramite 185 −1.2 0.026 −0.1 21.668 15.828 40.565 −21.93 −157.93
M10 PX Ankaramite 112 −0.8 0.009 −0.2 21.718 15.837 40.583 −21.93 −157.93
M10 OL Ankaramite 161 −0.3 0.008 −0.2 21.718 15.837 40.583 −21.93 −157.93
M12 PX Ankaramite 127 −1.6 0.030 −0.2 21.708 15.831 40.561 −21.93 −157.93
M12 OL Ankaramite 322 −2.3 0.035 0.7 21.708 15.831 40.561 −21.93 −157.93
M13 PX Ankaramite 238 0.2 0.006 0.0 21.666 15.817 40.539 −21.93 −157.93
M13 OL Ankaramite 96 −2.1 0.049 0.5 21.666 15.817 40.539 −21.93 −157.93
MGA-B-47a Pictrite −3.28 −0.13 21.784 15.813 40.734 −21.9379 −157.9234
  • Note. Data compilation of S-isotope and radiogenic Pb isotope compositions presented.
  • a S-isotope data from Cabral et al. (2013). Radiogenic Pb for M4-M13 from Woodhead (1996). Radiogenic Pb data for MGA-B-25 and MGA-B-47 from Hauri and Hart (1993). Estimated coordinates are sourced from the GeoRoc database. Rock characterization sourced from Woodhead (1996) and Saal et al. (1998).
Details are in the caption following the image
δ34S versus Δ33S for bulk sulfide data from sulfide inclusions in olivine and pyroxene phenocrysts. We also plot a mixing triangle between (a) the convective mantle (δ34S = 0, Δ33S = 0), (b) the proposed recycled endmember from Cabral et al. (2013) (δ34S = −22, Δ33S = −0.6), and (c) our new proposed second recycled endmember (δ34S = −7.5, Δ33S = +0.2). The dashed mixing lines within the triangle show that the S-isotope composition of bulk sulfide measurements represent a mixture of ~80% the recycled (c) component, providing a potential explanation for the mismatch between the SIMS analysis and the bulk measurement.
Details are in the caption following the image
Δ33S versus Δ36S for bulk sulfide data from sulfide inclusions in olivine and pyroxene phenocrysts. With the exception of two data points from sulfide in olivine (M12 Ol and M13 Ol), the data overlap with data from MORB (Labidi et al., 2013; Labidi et al., 2014; Labidi & Cartigny, 2016).

The δ34S compositions measured range from −5.1‰ to +0.2‰ (±0.3 2σ). Sulfide minerals hosted in olivine are not in isotopic equilibrium with sulfide minerals from pyroxene in the same sample showing −4.0‰ and −5.1‰ for MGA-B-25 Ol and MGA-B-25 Px, respectively, −0.3‰ and −0.9‰ for M10 Ol and M10 Px, respectively, −2.3‰ and −1.6‰ for M12 Ol and M12 Px, respectively and −2.1‰ and 0.2‰ M13 Ol and M13 Px, respectively (Figure S3). This reflects remarkable intrasample (olivine vs. pyroxene in the same lava) and intersample (olivines and pyroxenes from different samples) δ34S variability in Mangaia lavas. However, there is no relationship between the magnitude of the δ34S value and type of mineral separate. There is also no systematic fractionation among the mineral separates from the same sample.

The Δ33S and Δ36S range from +0.002‰ to +0.049‰ and −0.81‰ to +0.69‰, respectively. Δ33S values of sulfide in Ol and Px from the samples measured are similar with exception of one sample (M13) showing Δ33S = +0.049‰ and +0.006‰ for Ol and Px, respectively. Similarly, Δ36S values are similar for sulfides in pyroxene and olivine, with the exception of M12 showing Δ36S = +0.7‰ and −0.3‰ for Ol and Px, respectively. A subset of samples (M4, M10, M12, and M13) yield a rough negative correlation between δ34S and Δ33S (Figure 1). With the exception of two data points (M13 Ol and M12 Ol), the data plot at slightly higher Δ33S and lower Δ36S compared to previously published data from MORB (Figure 2).

Details are in the caption following the image
Whole rock Pb isotopes of Mangaia lavas, Pb isotopes in olivine-hosted melt inclusions from Mangaia, and Δ33S of sulfide in melt inclusions within olivine and pyroxene phenocrysts. (a) 208Pb/206Pb versus 207Pb/206Pb for olivine-hosted melt inclusions from Mangaia. Melt inclusion data are published in Cabral et al. (2014), Paul et al. (2011) and Saal et al. (1998), and whole rock data are shown for reference Hauri and Hart (1993) and Woodhead (1996). Data from melt inclusions demonstrate Pb isotope heterogeneity in a single sample that suggests mixing of multiple melt sources. (b) Δ33S versus 206Pb/204Pb. Pb isotope data are from whole rock measurements and S-isotope data represent bulk sulfide digestions of sulfur in melt inclusions within olivine and pyroxene phenocrysts. *The S measurement of bulk sulfide in olivine from MGA-B-47 was presented in Cabral et al. (2013). The S-isotope data and Pb isotope data show potential for a relationship that suggest mixing of multiple S sources and would link MGA-B-47 to an older mantle component.

4 Discussion

An unusual observation of HIMU mantle sampled at Mangaia made by Cabral et al. (2013) is the negative δ34S and the presence of S-MIF observed in individual sulfides by SIMS and by SF6 techniques similar to those used here. The signatures observed by Cabral et al. (2013) yielded a field extending from δ34S of −3.3 (±2‰) to −17.3 ‰ (±0.6‰) and with near-zero to negative Δ33S. The observation of samples with negative Δ33S that were measured both by SIMS and with one measurement by SF6 (−0.12 ± 0.04‰, 2σ) carried out at UMD (on sample MGA-B-47) is ascribed to the presence of subducted and recycled Archean components in the Mangaia source.

The data collected in this study on sulfide inclusions in mineral separates of Ol and Px from Mangaia also show a range of δ34S but do not extend to values as negative as those presented in Cabral et al. (2013) (−5.2‰ to +0.2‰). Our data also lack the negative Δ33S values seen in Cabral et al. (2013) and extend from to +0.006 to +0.049‰ (±0.016‰). Our data include analyses of sulfur from phenocrysts of one sample studied by Cabral et al. (2013) (MGA-B-25) for which they report a negative Δ33S by SIMS, because this sample had not been analyzed using the SF6 method by Cabral et al. (2013). Cabral et al. (2013) reported a 33S depletion (−0.12 ± 0.04‰, 2σ) in an olivine separate from MGA-B-47. This was based on a measurement in our laboratory using the same fluorination and mass spectrometry as that is described in the methods section of this manuscript but with a wet chemistry protocol involving hydrochloric acid and Cr (II) to digest sulfide minerals. MGA-B-47 was not reanalyzed here due to lack of available olivine material with inclusions. The protocol we use here includes HF, in addition to HCl and Cr (II), after Labidi et al. (2012). The HF technique is more efficacious at digesting silicate and thus can extract sulfides trapped within phenocrysts which may be missed by the HCl + Cr (II) protocol used in Cabral et al. (2013). While the sulfide concentrations we observe (96 to 333 ppm) are higher than those reported in Cabral et al. (2013) for the SF6 analysis of olivine from MGA-B-47 (~30 ppm), we cannot say whether this is due to some sulfide loss during sample processing in the laboratory or whether this reflects a lower sulfide content for the material from MGA-B-47 compared to the samples studied here. Regardless, both data sets (from Cabral et al., 2013, and this study) reveal significant variability for both δ34S and Δ33S that reflect some aspect of the history of these materials. We discuss below the arguments that support an origin associated with mixing of sulfur from different sources at the time of phenocryst formation.

4.1 Phenocrysts and Links to Sulfide Concentration

A unique aspect of this work is that our samples are not whole rock samples or glasses but rather phenocryst separates containing sulfide inclusions (Cabral et al., 2013). Sulfur is incompatible and thus not incorporated in olivine and pyroxenes structures (Gaetani & Grove, 1997; Peach et al., 1990). Digestion of sulfide from inclusions such as those documented by Cabral et al. (2013, 2014) is also the most straightforward way to explain the amount of sulfur we see from digestion of phenocrysts. We observe between 100 and 300 ppm S in the bulk mineral separates (Table 1). Considering a silicate melt inclusion from Mangaia typically hosts ~1,000–2,000 ppm S (Cabral et al., 2014), olivines would be required to be composed of 5–20 wt% melt inclusions to account for our concentration data. This is significantly more than the <0.1 vol% observed for phenocrysts (Cabral et al., 2014). Our S concentrations, however, are consistent with the presence of exsolved sulfide inclusions, which consist of approximately 32 wt% S and thus require <0.02 vol% as sulfide inclusions in phenocrysts.

Trapping of sulfide liquids by growing phenocrysts is supported by observations in previous work: Near-perfect spherical-shaped sulfide inclusions were observed in olivine (often coexisting with silicate melt inclusions) by Cabral et al. (2013, 2014). The sulfide droplets were shown to display an association of Cu, Ni, and Fe sulfides that exsolved upon further cooling (Cabral et al., 2013), which is typical of magmatic sulfides observed in MORBs (Peach et al., 1990). For sulfide to be trapped as inclusions in phenocrysts, the associated melt must be sulfide saturated and contains coexisting sulfide liquid droplets. Such liquids could have exsolved from the silicate melt as a result of changes in melt composition associated with phenocryst formation or other processes. Liquid sulfide droplets would subsequently be trapped by growing olivine and pyroxene phenocrysts. Note that the amount of sulfide inclusions in any of the phenocrysts must reflect a variety of factors, including the level of sulfide saturation and the behavior of sulfide liquids and phenocrysts during the time leading up to and including entrapment. Similar to melt inclusions not yielding straightforward information for characterizing the liquid/mineral ratios of parental melts, the amount of sulfide inclusions may not directly constrain the sulfide concentration of the parent melt(s).

Melt inclusions in olivine phenocrysts from Mangaia record pressures of entrapment between 250 bar to 1.75 kbar that translate into approximately 0.9 to 6.4 km depth (Cabral et al., 2014) and magmatic melt temperatures of >1200°C. Again, at these conditions sulfides are above the liquidus (Craig & Kullerud, 1968). At these pressures, available thermodynamic models argue against sulfur degassing as a significant, or even likely, process (Burgisser et al., 2015; Gaillard et al., 2011). Once trapped, the phenocrysts encapsulate and isolate sulfide sulfur from further processing such as degassing upon eruption and posteruption weathering and provide snapshots of the diversity of magma chamber compositions prior to mixing and eruption. This process is analogous to the entrapment of melt inclusions which reveal heterogeneities inherited from mantle sources, before late mixing and eruption (e.g., Sobolev & Shimizu, 1993).

4.2 Covariation of δ34S and Δ33S in Mangaia Samples M4, M10, M12, and M13

We observe covariation of δ34S and Δ33S for sulfide from bulk Ol and Px in samples M4, M10, M12, and M13. It is conceivable that the covariation is the result of a high temperature process that yields mass-dependent fractionation. One may suggest that the Mangaia melts started with near-zero δ34S and Δ33S values, similar to MORB values (Labidi et al., 2014)and subsequently evolved toward low δ34S and high Δ33S. However, the change of Δ33S with δ34S is not explained by the typical 0.515 exponent for high-temperature isotopic fractionations (Cao & Liu, 2011; Hulston & Thode, 1965). An exponent of 0.499 ± 0.005 is needed to fit the δ33S-δ34S data (Figure S2) from this study (specifically that of M4, M10, M12, and M13). Although observations of exponents outside of 0.515 have been proposed for lower temperature processes (e.g., Deines, 2003; Eldridge & Farquhar, 2018), we know of none that translate to the conditions at high temperature.

While the armoring of sulfides by phenocrysts likely makes them immune to effects of degassing, we examine degassing further since it is a common process in basalts and, as such, is a typical explanation for shifts to negative (or positive) δ34S (Beaudry et al., 2018; de Moor et al., 2013; Mandeville et al., 2009). Given the mass fractionation scenario proposed by Fiege et al. (2015) at QFM (quartz-fayalite-magnetite) redox conditions, a loss of 85% of the sulfur as S (IV) during equilibrium (i.e., closed system) degassing would be required to achieve a δ34S shift of ~−3‰. This is insufficient to account for the range of observed δ34S in our data set and also that observed by Cabral et al. (2013). There is also no clear evidence that degassing can explain the change of Δ33S with δ34S from 0.515 by significant amounts. Although analyses of degassed OIB show shifts for δ34S, they show no significant Δ33S variations, even when large δ34S fractionations are produced (Beaudry et al., 2018; Dottin et al., 2020).

Partitioning of sulfur to S (VI) (sulfate) is also known to shift the δ34S of residual sulfide to negative values. A case has been made for the presence of sulfate in OIBs (Jugo et al., 2010; Labidi et al., 2015). The equilibrium fractionation between sulfide and sulfate for δ34S is ~3‰ at 1200°C (Mandeville et al., 2009, and references within). The possibility that this accounts for a 34S depletion of the magnitude we observe would require a significant partitioning of sulfur to sulfate, and a sulfate/Σsulfur ratio greater than values seen in OIB (e.g., 0.2 observed in Samoan lavas in Labidi et al., 2015). In the extreme case where 50% of total sulfur in the melt is sulfate, sulfides would be shifted from the bulk sulfur composition by ~1.5‰. This fractionation is not large enough to account for the 34S depletion observed in the Mangaia sulfide inclusions. Furthermore, there is no evidence that, at high temperature, sulfide and sulfate would coexist with distinct Δ33S values resulting from fractionation with a mass-dependent exponent different from 0.515.

Altogether, we conclude that there is no reasonable process identified to explain the δ34S and Δ33S covariation among sample M4, M10, M12, and M13. We argue that, instead, the covariation is rather likely caused by mixing processes in the mantle sources prior to eruption. Most of our data are accounted for by two-component mixing, between the ambient mantle and a component with low δ34S and positive Δ33S (Figure 1a, endmembers A and C). Olivine and pyroxene separates from a given sample are in S-isotope disequilibrium: They do not record identical δ34S-Δ33S compositions. Instead, the data from the various mineral separates plot along the two-component mixing array δ34S-Δ33S space (Figure 1). This indicates a contribution of two isotopically distinct magma sources that are heterogeneously delivered to the magma mixture and sampled by various phenocrysts in a given sample. Note that MGA-B-25 (measured via IRMS) does not fall along the two-component mixing relationship between endmembers A and C. We suggest that this reflects the contribution of a third component: the recycled endmember identified in Cabral et al. (2013). We show in Figure 1a that this sample can be accounted for by a mixture between the ambient mantle, the endmember identified in Cabral et al. (2013) (Figure 1, endmember B), and the endmember identified in this study (Figure 1, endmember C).

The heterogeneous contributions of these three components recorded in phenocrysts of our samples are in agreement with previous arguments for magma mixing prior to eruption at Mangaia. This is supported by (1) highly variable forsterite content of olivine from individual Mangaia lavas and (2) Pb isotopic variability in individual olivine-hosted melt inclusions in Mangaia lavas (Saal et al., 1998) (but see Paul et al., 2011, and Cabral et al., 2014, who identify somewhat less Pb isotopic variability in Mangaia melt inclusions; Figure 3). Our suggestion is also in agreement with petrographic observations of resorption features on Ol and Px phenocrysts from M4, M10, M12, and M13 (Woodhead, 1996), which are consistent with magma mixing.

4.3 Nature and Age of the Recycled Components

The HIMU mantle reservoir is generally thought to reflect mantle mixed with subducted oceanic crust that experienced sulfide loss, causing depletion in Pb (Chauvel et al., 1992; Kelley et al., 2005; Lassiter et al., 2003); however, some proposals also call for a mantellic origin (Thirlwall, 1997), formation by recycled carbonate (Castillo, 2015; Castillo, 2016; Castillo et al., 2018), recycled serpentinites (Kendrick et al., 2017), carbonatitic metasomatism of the subcontinental lithospheric mantle that in turn delaminated and sank to the core-mantle boundary (Weiss et al., 2016), or an origin in metasomatic veins hosted in ancient subducted mantle lithosphere (Pilet et al., 2008). A wide range of ages have been proposed for the U/Pb fractionation that generated HIMU (~550 Ma to 3.0 Ga; Hauri and Hart, 1993; Mazza et al., 2019; Nebel et al., 2013; Thirlwall, 1997). Below, we offer a hypothetical scenario that allows the S-isotope data to be reconciled in light of the most common model for HIMU source generation: altered oceanic crust (AOC) + peridotite + a carbonated contribution (e.g., Jackson & Dasgupta, 2008)

4.3.1 Subduction Processes

When including an AOC component, we must consider the potential effect of sulfide loss during slab subduction on δ34S. If a parcel of subducted oceanic crust started with ~1,000 ppm S (Alt & Shanks, 2011; Rouxel, Ono et al., 2008) with an average δ34S of +3‰ (Alt, 1995), and assuming Pb is hosted in sulfides (Kelley et al., 2003), Pb loss by sulfide breakdown in the subduction zone results in 400–20 ppm S (i.e., 60% to 98% sulfide loss) in the HIMU mantle source. A process resulting in the preferential loss of 34S could explain the unusually low δ34S observed at Mangaia. Assuming 60% sulfide loss occurred during slab devolatilization, such would require an average fractionation of ~14‰ between the fluid (preferentially partitioning 34S) and the residual slab (preferentially retaining 32S) to explain our data, and a ~34 ‰ fluid-slab residue fractionation to explain the lowest δ34S in Cabral et al. (2013). This calculation represents an upper limit. Instrumental bias during analytical sessions results in a fractionation on δ34S as large as −3.8 ± 0.7 (1SD) (Cabral et al., 2013) when calibrating pyrrhotite measurements. If 98% of sulfide was lost, an average fractionation of ~9‰ between the fluid and the residual slab to explain our data, and a ~21‰ fluid-slab residue fractionation to explain the lowest δ34S in Cabral et al. (2013). These fractionations are large for temperatures associated with slab dehydration and are not observed in studies of metamorphism (Li et al., 2020). We note additionally that in diamond sulfides, subducted components typically do not show evidence for S isotopic fractionation associated with subduction of MIF-Sulfur, displaying δ34S ≥ 0 and Δ33S ≥ 0 (Farquhar et al., 2002; Smit et al., 2019; Thomassot et al., 2009) and at localities where δ34S is fractionated (Beaudry et al., 2018; Hofmann et al., 2009; Labidi et al., 2013, 2015; Roerdink et al., 2016; Thomassot et al., 2015), the composition is attributed to that of the protolith and not subduction processes. In our study we adopt this interpretation and conclude that while sulfide breakdown in subduction zones causes U/Pb fractionation in HIMU, it may not account for variable δ34S observed globally, between Mangaia (this study; Cabral et al., 2013) and other HIMU locations (e.g., Beaudry et al., 2018).

4.3.2 Mixing AOC, Peridotitic Mantle, and a Carbonated Component

The variable δ34S values of HIMU may be inherited from the composition of the protoliths, prior to subduction. Modern AOC has, on average, a δ34S that is positive due to the widespread occurrence of massive hydrothermal sulfides that generally carry positive values (Alt, 1995). While such material is consistent with the positive δ34S estimated for HIMU at other localities (e.g., Labidi et al., 2014; Labidi & Cartigny, 2016), it contrasts with the low δ34S values in HIMU Mangaia samples observed here. The bulk δ34S composition of the first kilometer of oceanic crust is estimated at −6 to −7 ‰ (Alt & Shanks, 2011; Rouxel, Shanks, et al., 2008), with positive Δ33S up to +0.162‰ ± 0.014 (Ono et al., 2012). This first kilometer is where the deep biosphere is able to process seawater sulfate and precipitate biogenic sulfides. If this portion of the slab is the contributor of the AOC component in HIMU, such can explain the low δ34S values at Mangaia with the caveat that a unique process is required to contribute a specific parcel of a slab (specially the upper portion) to the Mangaia mantle source.

An additional component we must consider in that of the carbonated component in HIMU lavas (Jackson & Dasgupta, 2008, and references therein). Cabral et al. (2014) argue for a carbonated source at HIMU Mangaia on the basis of inferred (predegassing) high primary CO2 in melt inclusions that are required to explain their elevated CaO/Al2O3 and low SiO2, as well as the presence of carbonatite observed in Mangaia melt inclusions by Saal et al. (1998) and Yurimoto et al. (2004). Castillo et al. (2018) suggest a component of recycled carbonate in the source of HIMU, and such may be the source of carbonation for the HIMU mantle source. If the carbonated component originates from marine carbonates, or carbonated shales, as suggested by Castillo et al. (2018), they could have influenced the isotopic composition of sulfides. Archean and Proterozoic carbonates are not typically sulfide rich (Veizer et al., 1989; Veizer Clayton, & Hinton, 1992; Veizer, Plumb, et al., 1992) and frequently possess positive δ34S; thus these lithologies do not provide an obvious way to connect the carbonated component to the full negative δ34S at Mangaia; however, the possibility of such a connection remains.

We can establish a simple mass balance of the carbonated component and AOC component in the HIMU mantle source. In our scenario, the carbonated component has a δ34S of ≈−7.5‰ (Figure 1) and <10,000 ppm S estimated from carbonatite sulfur contents as a proxy for the carbonated component (Gold, 1963), the AOC component also has a δ34S of ≈−7.5‰ (Figure 1) but <400 ppm S (Labidi et al., 2014), and the peridotitic mantle has a δ34S of = 0‰ and 200 ppm S (Lorand & Luguet, 2016). It is clear that regardless of the AOC-peridotite mixing ratio, the S-isotope budget would be dominated by relatively small contributions of the carbonated component, irrespective of the δ34S compositions of AOC and peridotite. The AOC contribution was suggested to be as high as 50% in the source of Mangaia (Hauri & Hart, 1993). We suggest that in addition to 50% recycled AOC in the HIMU mantle source, <10% of the carbonated material is necessary to satisfy our mixing model (Figure 1), in line with numerous suggestions for a C-rich component in the Mangaia mantle source (e.g., Cabral et al., 2014; Jackson & Dasgupta, 2008; Saal et al., 1998; Yurimoto et al., 2004).

4.3.3 Unusual δ34S in HIMU OIB

The contribution of negative δ34S components at Mangaia is unusual. It contrasts with recycled components observed at Samoa (EM II) and Discovery (EM I) hotspots, where positive δ34S values are observed (Dottin et al., 2020; Labidi et al., 20132015). Another HIMU hotspot, the Canary Islands, also shows a mantle source with positive δ34S values (Beaudry et al., 2018), and MORBs that incorporate variable amounts of HIMU components (Sobolev et al., 2007) appear to be influenced by a source with δ34S = +3 ± 2‰ (Labidi et al., 2014; Labidi & Cartigny, 2016). Importantly, the unusual 34S-depleted signature present in the Mangaia samples (this study and Cabral et al., 2013) as well as the EM1 Pitcairn lavas in Delavault et al. (2016) are associated with variable Δ33S that may be linked to the ages of the contributing source components.

It has been suggested that recycling of negative δ34S components may have occurred in the Archean (Cabral et al., 2013; Delavault et al., 2016; Farquhar & Jackson, 2016) and the Proterozoic (Canfield, 2004) as a way to offset the positive δ34S components observed in sediments. Negative δ34S values are not typically observed in Archean rocks, although they have been observed among positive δ34S values in some formations, such as barite deposits (Roerdink et al., 2016), gabbros (Thomassot et al., 2015), Archean conglomerates (Hofmann et al., 2009), and metamosphosed Archean crust (Kitayama et al., 2012). This hypothesis posits that an underrepresented reservoir of 34S-depleted material was lost to recycling and stored in the deep mantle, in the source of hotspots like Mangaia. EM-2 and EM-1 locations, where sedimentary recycling is suspected to occur, show the contribution of a crustal component with positive δ34S and near-zero Δ33S (Labidi et al., 20132015) and negative δ34S associated with variable Δ33S (Delavault et al., 2016), respectively. Thus evidence from mantle samples suggests recycling of sulfur with both negative and positive δ34S. Our measurements, along with those of Cabral et al. (2014), indicated that at Mangaia the negative δ34S dominates.

4.3.4 Information on Relative Age From Δ36S Data

Additional constraints on the relative age of sulfur contributing slightly positive Δ33S may be provided by the relationship between Δ33S and Δ36S. Archean rocks typically show covariation of Δ33S and Δ36S with a slope of ~−1 (Farquhar et al., 2007). This means that if a Δ33S 0.05‰, like that observed here, results from the contribution of an Archean protolith in the Mangaia mantle source, the corresponding Δ36S is predicted to be −0.05‰, which is not resolved with our uncertainty on Δ36S (±0.3‰). Post-Archean rocks show covariation of Δ33S and Δ36S with a slope of ~−7, which would result in a very small Δ36S shift to −0.4‰, which would be marginally resolvable given our estimated uncertainty. Our data generally plot below (more negative Δ36S) and to the right (more positive Δ33S) of typical MORB (Figure 2). The shift for Δ36S of the data cluster is more consistent with a biological fractionation than a mass-independent (Archean) fractionation, but an Archean component with little to no mass-independent anomaly could satisfy our observations. However, we argue the most parsimonious explanation for the data here is that the material we measured is post-Archean, in addition to the contribution of Archean S (as identified in Cabral et al., 2013) in the MGA-B-25 data presented here. We note that the data do, however, have large uncertainties, and our hypothesis would need to be confirmed with further analyses.

4.4 Recycling Two Reservoir Ages

With the samples presented here, the Pb isotope data of whole rock basalts from Woodhead (1996) and Hauri and Hart (1993) all show extreme 206Pb/204Pb, indicating the endmember HIMU character of the Mangaia mantle source, consistent with modern geochemical studies of Mangaia lavas (Hanyu et al., 2011, and references therein). Further constraints on the nature of the recycled components at Mangaia may be provided in relationships between 206Pb/204Pb and sulfur isotopes: Whole rock Pb isotope and bulk sulfide S-isotope measurements reveal heterogeneity in Mangaia lavas. There is some Pb isotopic heterogeneity in the whole rocks, but the Pb isotope compositions of individual melt inclusions in Mangaia lavas reveal substantially more heterogeneity than the whole rock data (Saal et al., 1998; Yurimoto et al., 2004) though recent reports of Pb-isotopic heterogeneity in Mangaia inclusions reveal somewhat less heterogeneity (Cabral et al., 2014; Paul et al., 2011) (Figure 3a). Thus, compared to the bulk olivine and clinopyroxene analyses presented here, the greater S isotopic heterogeneity in individual melt inclusions (Cabral et al., 2013) may not be surprising in this light. Additionally, the remarkable δ34S disequilibrium between olivine and pyroxene in a single sample (this study) supports a scenario in which the melt inclusions are sampling isotopically heterogenous compositions, as supported by the Pb-isotopic heterogeneity in Mangaia melt inclusions.

Our S-isotope data are from mineral separates extracted from the same whole rocks presented in Woodhead (1996) and Hauri and Hart (1993). Interestingly, the S-isotope compositions presented in this study and Cabral et al. (2013) from mineral separates show the potential for a relationship between whole rock Pb isotopes and bulk S isotopes from mineral separates (Figure 3b), where whole rock 206Pb/204Pb increases with decreasing Δ33S in mineral separates. We note however that the whole rock Pb isotope data use an older TIMS method that did not employ double or triple spike to track in-run mass fractionation (Hauri & Hart, 1993; Woodhead, 1996). Thus, these older TIMS data are likely to have larger uncertainties than were originally reported by these authors (Hauri & Hart, 1993; Woodhead, 1996). Nonetheless, the variability in Pb isotope compositions of melt inclusions may be indicative of mixing of multiple compositions in the HIMU mantle source beneath Mangaia, suggesting the relationship we observe between bulk S-isotope compositions from mineral separates and whole rock Pb isotope compositions, also reflects variability in the HIMU mantle source. If real, the data would reflect mixing between components of different ages, where higher 206Pb/204Pb and the negative Δ33S in Mangaia is associated with an older mantle component.

5 Conclusions

The data presented here on sulfide inclusions within olivine and pyroxene phenocrysts from Mangaia reveal a heterogeneous S-isotope composition. Compositions range from −5.13‰ to +0.21‰ (±0.3), +0.006‰ to +0.049‰ (±0.016), −0.81‰ to +0.69‰ (±0.3) for δ34S, Δ33S, and Δ36S, respectively. The Δ33S values are different from those presented in Cabral et al. (2013) who generally show negative Δ33S and indicate that some, possibly the preponderance, of HIMU sulfur sampled at Mangaia comes from a different reservoir, likely of younger post-Archean age. Both studies, however, report 34S-depleted sulfur, and this appears to be a characteristic of Mangaia HIMU that we attribute to a crustal origin, following observations by studies like Rouxel, Shanks, et al. (2008) and Alt and Shanks (2011). Thus, there appears to be variability in the material derived from the same plume plumbing system at Mangaia.

This work focuses on S-isotope compositions of bulk sulfide in melt inclusions in Ol and Px from the same sample, and this sulfur is not in isotopic equilibrium but shows covariation in δ34S and Δ33S which we link to the capturing of S-isotope compositions of various melts as they are added to the magma chamber. The data for Mangaia presented here, and provided by Cabral et al. (2013), support mixing of melts with at least three distinct sulfur isotope endmembers: (1) a convective mantle component with δ34S = 0 and Δ33S = 0; (2) a component with negative δ34S and slightly positive Δ33S, and (3) a component with negative δ34S and negative Δ33S.


We acknowledge funding for this project from National Science Foundation EAR-1348082 (to J. F. and M. J.). We also thank Editor Marie Edmonds for handling our manuscript and for helpful reviews from Emilie Thomassot and an anonymous reviewer.

    Data Availability Statement

    Data set for this manuscript exists as open access in the Digital Repository at the University of Maryland (DRUM; https://doi.org/10.13016/zs1g-knnj).