Volatile Cycling of H2O, CO2, F, and Cl in the HIMU Mantle ... · 1986; Hart, 1988; Hofmann, 1997; White, 2010]. Although the cause of the chemical heterogeneity in the mantle has

RESEARCH ARTICLE10.1002/2014GC005473

Volatile cycling of H2O, CO2, F, and Cl in the HIMU mantle: Anew window provided by melt inclusions from oceanic hotspot lavas at Mangaia, Cook IslandsRita A. Cabral1, Matthew G. Jackson2, Kenneth T. Koga3, Estelle F. Rose-Koga3, Erik H. Hauri4,Martin J. Whitehouse5, Allison A. Price2, James M. D. Day6, Nobumichi Shimizu7,and Katherine A. Kelley8

1Department of Earth and Environment, Boston University, Boston, Massachusetts, USA, 2Department of Earth Science,University of California Santa Barbara, Santa Barbara, California, USA, 3Universit�e Blaise Pascal - CNRS - IRD, LaboratoireMagmas et Volcans, OPGC, Clermont-Ferrand, France, 4Department of Terrestrial Magnetism, Carnegie Institution of Wash-ington, Washington, DC, USA, 5Department of Geosciences, Swedish Museum of Natural History, Stockholm, Sweden,6Geosciences Research Division, Scripps Institution of Oceanography, La Jolla, California, USA, 7Department of Geologyand Geophysics, Woods Hole Oceanographic Institution, Woods Hole, Massachusetts, USA, 8Graduate School of Oceanog-raphy, University of Rhode Island, Narragansett, Rhode Island, USA

Abstract Mangaia hosts the most radiogenic Pb-isotopic compositions observed in ocean island basaltsand represents the HIMU (high m5 238U/204Pb) mantle end-member, thought to result from recycled oce-anic crust. Complete geochemical characterization of the HIMU mantle end-member has been inhibiteddue to a lack of deep submarine glass samples from HIMU localities. We homogenized olivine-hosted meltinclusions separated from Mangaia lavas and the resulting glassy inclusions made possible the first volatileabundances to be obtained from the HIMU mantle end-member. We also report major and trace elementabundances and Pb-isotopic ratios on the inclusions, which have HIMU isotopic fingerprints. We evaluatethe samples for processes that could modify the volatile and trace element abundances postmantle melt-ing, including diffusive Fe and H2O loss, degassing, and assimilation. H2O/Ce ratios vary from 119 to 245 inthe most pristine Mangaia inclusions; excluding an inclusion that shows evidence for assimilation, the pri-mary magmatic H2O/Ce ratios vary up to �200, and are consistent with significant dehydration of oceaniccrust during subduction and long-term storage in the mantle. CO2 concentrations range up to 2346 ppmCO2 in the inclusions. Relatively high CO2 in the inclusions, combined with previous observations of carbon-ate blebs in other Mangaia melt inclusions, highlight the importance of CO2 for the generation of the HIMUmantle. F/Nd ratios in the inclusions (30 6 9; 2r standard deviation) are higher than the canonical ratioobserved in oceanic lavas, and Cl/K ratios (0.079 6 0.028) fall in the range of pristine mantle (0.02–0.08).

1. Introduction

Isotopic measurements performed on ocean island basalts (OIBs) have demonstrated that the Earth’s mantleis compositionally heterogeneous [e.g., Gast et al., 1964; Allegre, 1982; Zindler et al., 1982; Zindler and Hart,1986; Hart, 1988; Hofmann, 1997; White, 2010]. Although the cause of the chemical heterogeneity in themantle has been a point of discussion in the scientific community for half a century [Gast et al., 1964], it hasbeen suggested to arise from the subduction of oceanic and continental crust material into the mantle[Cohen and O’Nions, 1982; Hofmann and White, 1982; White and Hofmann, 1982]. Different types of sub-ducted material (e.g., oceanic crust, continental crust, and sediment) have been proposed to give rise tospecific mantle end-member isotopic compositions—HIMU (high m5 238U/204Pb), EM1 (enriched mantle I),and EM2 (enriched mantle II)—which are then melted and erupted as OIB melts at hot spots. EM2 is gener-ally considered to be derived from continental material [White and Hofmann, 1982; Jackson et al., 2007],but the origin of EM1 remains more controversial [e.g., Weaver, 1991; Eiler et al., 1995; Eisele et al., 2002;Honda and Woodhead, 2005; Salters and Sachi-Kocher, 2010; Hart, 2011]. The HIMU mantle end-member isthought to be derived from ancient subducted oceanic crust [e.g., Chase, 1981; Hofmann and White, 1982;Zindler et al., 1982; Nakamura and Tatsumoto, 1988; Dupuy et al., 1989; Graham et al., 1992; Hauri and Hart,1993; H�emond et al., 1994; Roy-Barman and Allegre, 1995; Woodhead, 1996; Hanyu and Kaneoka, 1997;

Key Points:� Volatile element concentrations have

never been reported on HIMU end-member lavas� Lavas from Mangaia represent melts

of the HIMU mantle end-member� We provide volatile and trace

element concentrations on Mangaiamelt inclusions

Supporting Information:� Readme� Tables S1–S7� Supporting information text� Filtering

Correspondence to:R. A. Cabral,[email protected]

Citation:Cabral, R. A., M. G. Jackson, K. T. Koga,E. F. Rose-Koga, E. H. Hauri,M. J. Whitehouse, A. A. Price,J. M. D. Day, N. Shimizu, andK. A. Kelley (2014), Volatile cycling ofH2O, CO2, F, and Cl in the HIMUmantle: A new window provided bymelt inclusions from oceanic hot spotlavas at Mangaia, Cook Islands,Geochem. Geophys. Geosyst., 15,4445–4467, doi:10.1002/2014GC005473.

Received 25 JUN 2014

Accepted 20 OCT 2014

Accepted article online 25 OCT 2014

Published online 28 NOV 2014

CABRAL ET AL. VC 2014. American Geophysical Union. All Rights Reserved. 4445

Geochemistry, Geophysics, Geosystems

PUBLICATIONS

http://dx.doi.org/10.1002/2014GC005473

http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1525-2027/

http://publications.agu.org/

Kogiso et al., 1997; Salters and White, 1998; Schiano et al., 2001; Lassiter et al., 2003; Stracke et al., 2003;Stroncik and Haase, 2004; Nishio et al., 2005; Chan et al., 2009; Parai et al., 2009; John et al., 2010; Hanyu et al.,2011; Kawabata et al., 2011; Salters et al., 2011; Krienitz et al., 2012; Cabral et al., 2013]. Because of the fluidmobility of Pb during subduction zone processing, Pb loss from the slab results in high U/Pb ratios in thesubducted slab reservoir, which evolves highly radiogenic Pb-isotope compositions over time [Hofmannand White, 1982; Zindler and Hart, 1986; Hauri and Hart, 1993; Kelley et al., 2005; Hanyu et al., 2011, 2014].Alternatively, the high U/Pb inferred for the HIMU mantle source may be a result of U incorporation intooceanic crust during submarine alteration [Hart and Staudigel, 1982; Hofmann and White, 1982; Chauvelet al., 1992; Kelley et al., 2003], and this elevated U/Pb ratio preserved during subduction would evolve theradiogenic Pb-isotopic signature associated with the HIMU mantle domain [Chase, 1981; Hofmann andWhite, 1982]. A recycled origin for the HIMU mantle end-member has been further supported by sulfur iso-topic measurements from olivine-hosted melt inclusions from Mangaia, which confirm that the materialerupted at Mangaia was once at Earth’s surface >2.45 billion years ago [Cabral et al., 2013].

During its life on the seafloor, oceanic lithosphere, including crust, experiences low temperature alterationand becomes H2O-rich and CO2-rich (and gains up to several weight percent H2O and CO2 in the top por-tion of the crust during submarine alteration) [e.g., Staudigel et al., 1996; Alt and Teagle, 1999; Bach et al.,2001; Wallmann, 2001; Dixon et al., 2002; Gillis and Coogan, 2011]. If the HIMU mantle end-member is trulycomposed of subducted oceanic crust that is subsequently melted and erupted, this mantle end-membercan present important insights into how volatiles are returned to the mantle and recycled back to the sur-face at hot spots. This is important, as the distribution of volatiles in the mantle (particularly H2O) is a signifi-cant variable in determining the rheology of the mantle, the depth and extent of mantle melting, as well asthe composition of melts and how they evolve [e.g., Hirth and Kohlstedt, 1996; Gaetani and Grove, 1998; Asi-mow and Langmuir, 2003; Hirschmann, 2006; Portnyagin et al., 2007]. The results of Cabral et al. [2013] alsodemand that the HIMU mantle sampled by Mangaia lavas experiences long-term storage in the mantle,which has important implications for mantle dynamics and the mechanisms of long-term mantle storagefor this end-member. This discovery highlights the need for careful geochemical characterization of HIMUlavas from Mangaia to better understand the origins of this mantle end-member. Despite its importance forunderstanding mantle recycling processes, from subduction zones to hot spots, volatile measurements arenot available from the extreme HIMU mantle end-member sampled by Mangaia lavas.

Due to the lack of deep sea dredging expeditions in the Cook Islands, submarine volcanic glasses are notavailable for volatile abundance characterization of HIMU mantle end-member lavas from Mangaia. Volatileshave not been measured on HIMU mantle end-member lavas, which extend to 206Pb/204Pb values of 21.9 inhigh-precision measurements made on basalts from Mangaia Island [e.g., Hauri and Hart, 1993; Woodhead,1996; Kogiso et al., 1997; Hanyu et al., 2011]. In an effort to obtain volatile measurements of this end-member,we have homogenized crystalline melt inclusions hosted in olivine phenocrysts extracted from subaeriallyerupted whole rocks collected from Mangaia. These analyses represent the first coupled Pb isotope and major,volatile, and trace element compositions measured on glasses from the HIMU mantle end-member.

We find that Mangaia melt inclusions host H2O and CO2 concentrations that are relatively high (1.73 wt %and 2346 ppm, respectively; corrected for olivine fractionation). We use the new data to constrain volatileabundances in the HIMU mantle end-member as well as volatile cycling in the mantle, from subductionzones to hot spots.

2. Sample Descriptions, Locations, and Methods

Sample descriptions and locations, as well as a description of the geology of Mangaia, are provided in thesupporting information. A complete description of melt inclusion preparation (including the homogeniza-tion procedure) and methods for analysis of Pb-isotopes and major, trace, and volatile concentrations arealso provided in the supporting information.

3. Results

The melt inclusions in this study are alkalic (Figure 1). The forsterite values of the olivine hosting the inclu-sions in this study range from 75.9 to 86.1, with an average of 81.6 (supporting information Table S1). The

Geochemistry, Geophysics, Geosystems 10.1002/2014GC005473


largest range in forsterite con-tent is exhibited by olivinesfrom hand sample MGA-B-47(Fo 5 80.3–86.1, supportinginformation Tables S1 and S2).The melt inclusion data pre-sented in Figure 2 have beencorrected to be in equilibriumwith the host olivine by addi-tion or subtraction of equilib-rium olivine following themethods outlined in Jacksonet al. [2012], and discussion ofthe major, trace, and volatileelement data below refers toabundances following this oli-vine fractionation correction,unless stated otherwise (sup-porting information Tables S1and S3–S5). The MgO contentsof the melt inclusions do notspan the entire range encapsu-lated by the whole rocks (3.10–20.47 wt %); instead they arerestricted to a range of 4.51–10.80 wt % MgO (supportinginformation Table S5). The meltinclusions tend to have lowerMgO than the host wholerocks, and this likely owes tothe fact that the whole rocksexamined in this study arephenocryst-rich and thereforemay reflect cumulate composi-

tions. The SiO2 concentrations generally cluster in the range defined by whole rocks from Mangaia, butsome inclusions (mostly from MGA-B-47) exhibit somewhat higher SiO2 at a given MgO than the wholerocks. Two inclusions in particular (both from hand sample MG1006) have lower SiO2 than previously identi-fied in Mangaia lavas. The whole rock data suggest the crystallization of a Ti-rich phase in lavas where MgOreaches <5 wt %. The melt inclusion MgO concentrations are generally above 5 wt % (with the exception ofsample MGA-B-47-RAC3 with 4.51 wt % MgO, corrected for olivine fractionation) and, with the possibleexception of MGA-B-47-RAC3, do not reach the point of extreme magmatic evolution where TiO2 concentra-tions are affected by fractionation of the Ti-rich phase. In sample MGA-B-47, there is one inclusion with TiO2

significantly higher than the whole rock and melt inclusion analyses. The melt inclusions generally overlapwith whole rock data in a plot of Al2O3 versus MgO (Figure 2). At a given MgO, most of the inclusions plotwithin the range of whole rocks for K2O and Na2O, but a large subset of the inclusions is shifted to higheralkali element concentrations than observed in the whole rocks. Melt inclusions generally have CaO concen-trations and CaO/Al2O3 ratios that are similar to the whole rock data at a given MgO, except for inclusionsfrom sample MGA-B-47, which have higher CaO, and a single inclusion from MG1001, which has lower CaO.There is a clear depletion in FeO for melt inclusions from the MGA-B-47 whole rock, which has beenobserved in melt inclusions previously and predicted experimentally (discussed below) [Danyushevsky et al.,2000; Gaetani and Watson, 2000]. Melt inclusions from the other samples lie on the FeO versus MgO trendformed by whole rocks, except for the two melt inclusions from the MG1006 whole rock, which show an ele-vation in FeO compared to the rest of the melt inclusion and whole rock data. These two inclusions areanomalous in many respects, and have the lowest SiO2, among the lowest CaO and CaO/Al2O3, and amongthe highest alkali abundances in the melt inclusion suite.

Figure 1. Na2O 1 K2O versus SiO2 plotted with the alkali-tholeiite basalt line from MacDon-ald and Katsura [1965] and applicable rock types. Mangaia melt inclusions are plotted asgray symbols and were corrected for olivine addition or subtraction (following the methodsoutlined in Jackson et al. [2012]) to be in equilibrium with their host olivine (see supportinginformation Table S5 for olivine-corrected concentrations). The whole rocks hosting therespective melt inclusions are plotted as the same symbol as the melt inclusions, but inblack. Previously published whole rock data are shown as white circles and were obtainedfrom Georoc [Sarbas, 2008]. One sample (M17) from Woodhead [1996] contains anomalouslylow Na2O. We argue that this is a result of significant alteration, which is supported by thehigh loss on ignition (LOI) of 9.33% for this sample; we have removed this sample from thefigures. Whole rock data for MGA-B-25 is not available. Samples from this study plot in thealkalic field.



Previously published whole rock trace element data were obtained from the Georoc database [Sarbas,2008] and, with few exceptions (see caption of Figure 3), were filtered for trace element analyses made byICP-MS (inductively coupled plasma mass spectrometry) [Hauri and Hart, 1993; Woodhead, 1996; Hanyu

Figure 2. Major element variability in Mangaia melt inclusions compared to their respective whole rocks and previously published wholerocks from the island. Samples from this study use the same symbol convention as seen in Figure 1. Error bars in the corner of each plotrepresent the average 2r measurement error of the melt inclusions. Data shown are for melt inclusions that have been corrected to be inequilibrium with the host olivine (see supporting information Table S5 for olivine-corrected concentrations).



Figure 3. Trace element spidergrams of Mangaia melt inclusions compared to Mangaia whole rocks. Mangaia melt inclusion data (cor-rected for olivine fractionation to be in equilibrium with the host olivine, see supporting information Table S5) are plotted as black lines,except for inclusion MG1001-5, which is plotted as a gray dashed line. All data are normalized to the chondrite-based primitive mantle val-ues of McDonough and Sun [1995]. (a) Previously published whole rock trace element data were acquired from Georoc [Hauri and Hart,1993; Woodhead, 1996; Sarbas, 2008; Hanyu et al., 2011] and were filtered to include only samples with a full suite of trace element data(Ba, U, Nb, La, Ce, Pr, Nd, Sr, Zr, Hf, Sm, Y, and Yb) obtained by ICP-MS analysis (except for Zr from Hauri and Hart [1997], which was meas-ured by XRF). A complete set of trace element data by ICP-MS did not exist from Kogiso et al. [1997], so we exclude the data from this refer-ence from the figure. Only whole rock samples that have MgO compositions >5 wt % are plotted to avoid the effects of extreme crystalfractionation (the majority of melt inclusions in this study have more than 5 wt % MgO). Additionally, whole rock samples with greaterthan 4% LOI are also excluded from the figure to avoid effects of alteration. The estimated primary melt concentration for Mangaia, calcu-lated from inclusion MG1001-3b following olivine fractionation corrected to be in equilibrium with mantle olivine (with forsterite 90 com-position), is the following (in ppm): Ba 5 256.0, U 5 0.93, Nb 5 42.3, K 5 6511, La 5 38.0, Ce 5 68.4, Pr 5 9.3, Nd 5 33.5, Sr 5 515, Zr 5 156,Hf 5 3.1, Sm 5 7.2, Ti 5 16,750, Y 5 28.9, Yb 5 2.3. Using the melt model outlined in section 3.2, �3% melting of the HIMU mantle source(shown as a black dashed line) will generate this primary melt composition if the trace element concentrations in the source are the fol-lowing (in ppm): Ba 5 7.7, U 5 0.028, Nb 5 1.27, K 5 195, La 5 1.15, Ce 5 2.20, Pr 5 0.35, Nd 5 1.51, Sr 5 18.1, Zr 5 12.4, Hf 5 0.29,Sm 5 0.51, Ti 5 2520, Y 5 10.4, Yb 5 1.4. (b) Melt inclusion MG1001-5—which has an anomalous spidergram pattern—plotted as a graydashed line. It is compared to the average of Cape Verde calcio-carbonatites from Fuerteventura Island [Hoernle et al., 2002], which is plot-ted as a solid black line. Inclusion MG1001-5 has several geochemical similarities with thee calico-carbonatite average, including negativeanomalies of U, K, Zr, Hf, and positive anomalies of Nb and Sr. However, the inclusion does not have the negative Ti anomaly found in thecarbonatite.



et al., 2011]. The whole rock trace element data and the new melt inclusion trace element data are shownas spidergrams in Figure 3. The majority of the melt inclusion trace element data agree well with the previ-ously published whole rock trace element data. One exception is inclusion MG1001-5 (dashed gray line inFigure 3), which has a spidergram pattern significantly different than the other melt inclusions and previ-ously published whole rocks. This sample was subjected to two replicate analyses by SIMS at Arizona StateUniversity, and an additional analysis by LA-ICP-MS at URI (supporting information Table S3), and the repli-cate analyses gave the same spidergram pattern. A key feature of this melt inclusion is that the spidergramhas strongly negative Zr and Hf anomalies, which mimic more subtle negative Zr and Hf anomalies in HIMUlavas from Mangaia [Kogiso et al., 1997; Hanyu et al., 2011] and other OIBs from the Cook-Austral Islands[Chauvel et al., 1997]. Notably, this inclusion does not have anomalous major or volatile element abundan-ces. The trace element pattern is qualitatively similar to the average calcio-carbonatite (including tracephases, like apatite, that are associated with calcio-carbonatite) from Fuerteventura (Cape Verde hot spot)presented by Hoernle et al. [2002], including negative anomalies U, K, Zr, Hf, and positive anomalies of Nband Sr. However, the inclusion does not have the negative Ti anomaly found in the calcio-carbonatites, andwe cannot offer an explanation for the origin of the anomalous trace element pattern in this inclusion.

The Mangaia melt inclusions are characterized by high H2O and CO2 contents (Figure 4a), up to 1.73 wt %and 2346 ppm (supporting information Table S5 and Figure 4b). Inclusions from MGA-B-47 form a separategroup with lower H2O than the melt inclusions from other hand samples; melt inclusions from this samplealso appear to have lower abundances of FeO (Figure 2) and H2O. The lower H2O (and H2O/Ce, see below)in inclusions from this sample suggest water loss from the inclusion through the host olivine (see section4.1). An inclusion from sample MG1006 exhibits high CO2, but like MGA-B-47 it has low H2O abundances(1750 ppm and 0.57 wt %, respectively; supporting information Table S5). However, unlike MGA-B-47 inclu-sions, this inclusion does not exhibit evidence for lower FeO abundance. Excluding this inclusion and allinclusions from MGA-B-47, the majority of the Mangaia melt inclusions lie along a broad array centered at�1.5 wt % H2O (varying from 1.14 to 1.73 wt %, supporting information Table S5). This array, defined by 27inclusions from hand sample MG1001 and three inclusions from hand sample MGA-B-25 only, spans CO2

concentrations ranging from 284 to 2346 ppm (supporting information Table S5). Excluding melt inclusionsamples from MGA-B-47 (discussed below), the melt inclusions record entrapment pressures ranging from250 bar to approximately 1.75 kbar (see Figure 4a). This translates to entrapment depths of up to 6.4 km(1.75 kbar) for melt inclusions with the highest concentrations of CO2 (assuming a rock density of igneouscrust of 2800 kg/m3) [Workman et al., 2006].

Figure 4c provides ratios of volatile (H2O and CO2) to nonvolatile (Ce and Nb) trace elements that have simi-lar mineral-melt partition coefficients during mantle melting and magmatic differentiation processes.Excluding MGA-B-47 and MG1006 inclusions, H2O/Ce varies from 119 to 245; the highest H2O/Ce value (245;sample MG1001A12) is associated with the highest (least degassed) CO2/Nb ratio (49.5; sampleMG1001A12), but this inclusion exhibits evidence for assimilation (see below).

Koleszar et al. [2009] suggested that the MORB mantle has a CO2/Nb ratio of 300 [Koleszar et al., 2009], butCO2/Nb in the various mantle reservoirs may not be constant, and it has also been suggested that the OIBmantle may have higher CO2/Nb than the MORB mantle [Saal et al., 2002; Cartigny et al., 2008; Koleszaret al., 2009]. The Mangaia melt inclusions reach a maximum CO2/Nb value of only 49.5, suggesting signifi-cant degassing (see section 4.4), and CO2 and CO2/Nb ratios in the inclusions must still be viewed as mini-mum values.

Fluorine concentrations range from 1073 to 1950 ppm (supporting information Table S5) in the Mangaiamelt inclusions. When compared to Nd, they are offset to higher F/Nd (�30) than the F/Nd ratio of 21observed in other MORB and OIB samples (Figure 5) [e.g., Workman et al., 2006]. F/Nd may not always reflectmantle source characteristics if assimilation and alteration processes are involved. For example, Lassiteret al. [2002] studied a HIMU lava from Raivavae (sample RVV318, 206Pb/204Pb 5 20.94) that showed strongevidence for brine and oceanic crust assimilation—and melt inclusions from this sample have high F/Ndratios (60–183). Koleszar et al. [2009] also identified high F/Nd in five melt inclusions that range from approx-imately 35 to 100 and argue that this departure from the canonical ratio of fresh OIBs and MORBs of 21 mayrelate to open-system behavior of F. We note that the Mangaia melt inclusion suite has elevated F/Nd ratios,but these inclusions do not exhibit signatures associated with significant assimilation prior to entrapment(see discussion, Figure 6).



Cl/K2O ratios for the Mangaia melt inclusions are quite low (Figure 6). The average Cl/K2O of the melt inclu-sions is 0.067 (including melt inclusions from MGA-B-47), which corresponds to a Cl/K ratio of 0.079 6 0.028and overlaps with the range for pristine oceanic lavas (Cl/K 5 0.02–0.08) found by Michael and Cornell[1998] and Stroncik and Haase [2004].

Figure 7 compares Cl/Nb and F/Nd ratios in the Mangaia inclusions. Cl/Nb ratios, like Cl/K ratios, can beused to monitor for assimilation of altered oceanic crust or brines (see discussion section 3.5). While the F/

Figure 4.



Nd in the Mangaia inclusions is shifted to higher values than observed in the EM1 and EM2 hot spots (Pit-cairn, Samoa, and Societies), even higher F/Nd ratios are found in melts from the Azores, Galapagos, andAustrals hot spots.

Sulfides are abundant in Mangaia melt inclusions. Data for S concentrations in the silicate portions of the meltinclusions are provided in supporting information Table S4. Since the concentration of S in the silicate portionof the inclusions is buffered by the presence of sulfides, we cannot use the measured S concentrations tomake inferences about sulfur in the mantle source and we do not discuss sulfur further in this paper. Somemeasurements of Pb isotopes, S isotopes, and major elements of the sulfides from these inclusions can befound in Cabral et al. [2013]. Sulfides from samples MGA-B-25 and MGA-B-47 were found to have anomaloussulfur isotopic compositions that indicate mass independent fractionation [Cabral et al., 2013].

Excluding melt inclusion measurements where the ion probe primary beam overlapped with phase boun-daries (see supporting information), new measurements (supporting information Table S5) of Pb-isotopicratios in the Mangaia inclusions plot in a tighter cluster near the extreme HIMU mantle end-memberdefined by Mangaia whole-rock compositions (Figure 8). The limited isotopic variability of the new SIMSmeasurements presented here for homogenized and nonhomogenized melt inclusions is similar to thatobserved in the Paul et al. [2011] LA-ICP-MS study. We do not observe the significant variability in Pb-isotopic measurements in melt inclusions identified in previous SIMS studies of Mangaia inclusions [Saalet al., 1998, 2005; Yurimoto et al., 2004]. Melt inclusions from MGA-B-25 (n 5 4 different inclusions analyzed)show the least Pb-isotopic variability, while MGA-B-47 exhibits the most Pb-isotopic variability (n 5 12 differ-ent inclusions analyzed). MG1001 exhibits intermediate isotopic variability (n 5 23 inclusions analyzed).Only one melt inclusion was analyzed from sample MG1002, and no melt inclusions from MG1006 wereanalyzed. The sample with the greatest range in olivine forsterite contents, MGA-B-47, also has the mostheterogeneous Pb-isotopic compositions.

The new Pb-isotopic data in these melt inclusions exhibit approximately twice as much variability asobserved in high-precision Pb-isotopic measurements of leached whole rocks from the island. One meltinclusion plots away from the other melt inclusions in Pb-isotopic space (MGA-B-47-RAC3), and expands themelt inclusion data field significantly; without this single inclusion, the melt inclusion data field is onlyslightly larger (but shifted to somewhat higher 207Pb/206Pb and 208Pb/206Pb ratios) than the data fielddefined by high-precision Pb-isotopic measurements of leached whole rocks from Mangaia.

4. Discussion

In order to constrain volatile element abundances of primary magmas and to ultimately make inferencesabout the composition of the HIMU mantle sampled by Mangaia lavas, it is first necessary to identify proc-esses that operate on magmas during ascent and eruption that might modify the volatile budgets of themagmas. In the case of melt inclusions, such processes can include degassing before entrapment of themelt inclusion, diffusive H2O loss from the melt inclusion through the host olivine, and assimilation of

Figure 4. Volatile contents of Mangaia melt inclusions. All Mangaia melt inclusions plotted in this figure are corrected to be in equilibriumwith the host olivine (see supporting information Table S5). (a) CO2 and H2O contents of the Mangaia melt inclusions with vapor saturationcurves generated by SolEx [Witham et al., 2012] and the open-system degassing trend for alkali basalt from Dixon et al. [1997]. Major ele-ment data from one of the most primitive and undegassed samples in the suite (MG1001-5, corrected for olivine fractionation) was usedfor the input values to SolEx. Melt inclusions from MGA-B-47 have experienced diffusive water loss and water data for these inclusions arenot included in subsequent figures; an inclusion from MG1006 has similarly low water, which is excluded in subsequent figures. As inWorkman et al. [2006], we interpret the nearly vertical trend (marked by steeply sloping arrow in the top plot) defined by inclusions fromsample MG1001 (n 5 27 inclusions) and MGA-B-25 (n 5 3 inclusions) to indicate CO2 degassing with negligible water loss prior to meltinclusion entrapment (except for inclusions from MGA-B-47, which suffered diffusive H-loss post entrapment). (b) The Mangaia melt inclu-sion data in comparison to other OIB and MORB data sets, which consist of data obtained from submarine glasses and melt inclusions.Data for the high CO2 inclusions from the Galapagos (Fernandina and Santiago; Koleszar et al., 2009) are ones for which CO2 from thevapor bubble was calculated back into the melt composition, and this may explain the unusually high CO2 (extends up to 5821 ppm) insome of the Fernandina inclusions (A. Saal, personal communication, 2014); when a similar calculation is applied to Mangaia melt inclu-sions, the concentration of CO2 increases by< 4% on average and< 0.02% for H2O on average (see corrected CO2 and H2O values in sup-porting information Table S4). Additionally, in this figure, samples were removed from plotting if the relevant data source [Clague et al.,1995; Dixon et al., 1997; Dixon and Clague, 2001; Dixon et al., 2002; Hauri, 2002; Saal et al., 2002; Workman et al., 2006; Cartigny et al., 2008,and references therein; Koleszar et al., 2009; Kendrick et al., 2014; Metrich et al., 2014] indicated that the sample was severely degassed dueto shallow eruption, experienced extensive assimilation, or was otherwise compromised; all samples removed, and the reason for theirremoval, are outlined in the supporting information. (c) CO2/Nb and H2O/Ce ratios for Mangaia melt inclusions.



altered oceanic crust andbrines before entrapment.When evaluating the geo-chemistry of olivine-hostedmelt inclusions from Mangaia,it is critical to evaluate all ofthese processes to determinewhether primary magmaticcharacteristics can be inferredfrom this new suite of meltinclusions. Identification of pri-mary melt volatile characteris-tics permits inferences aboutthe mantle source compositionof the HIMU end-member and,thus, volatile cycling throughthe mantle resulting fromsubduction-injection of oce-anic crust.

4.1. Volatile ElementConcentrations: SourceVersus Process4.1.1. Diffusive Loss of Feand HMelt inclusions from wholerock sample MGA-B-47 showmarked depletion in FeO com-pared to melt inclusions fromthe other three whole rock

samples, which we interpret to be representative of Fe loss from the inclusions via diffusion through thehost olivine (Figure 2). Fe loss from olivine-hosted melt inclusions has been documented in other melt inclu-sion suites and is thought to occur following long-term storage and slow cooling in the host magma, wherethe residence time is sufficient to permit Fe diffusion through the olivine [e.g., Danyushevsky et al., 1992,2000; Sobolev and Danyushevsky, 1994; Gaetani and Watson, 2000]. The diffusivity of H in olivine is greaterthan Fe, and if sufficient time elapsed for significant Fe loss to occur, then the time scale for diffusive H-lossfrom the inclusions was also surpassed and H loss would have taken place provided the external melt hadlower H2O than the melt inclusions. The lower H2O in MGA-B-47 inclusions is not likely due to magmadegassing, as inclusions from MGA-B-47 have elevated CO2 (149–1335 ppm, supporting information TableS5), so the inclusions were trapped at great depths before significant degassing of H2O. Therefore, we inter-pret the low H2O concentrations in MGA-B-47 inclusions (0.16–0.70 wt %, supporting information Table S5)to be a result of diffusive water loss from the inclusions from this hand sample, and the paired Fe loss andH2O loss in inclusions are complementary and indicative of longer time scales of storage in a magma cham-ber following olivine entrapment of the inclusions. Thus, in the discussion below regarding water abundan-ces in the HIMU mantle, we exclude the inclusions from hand sample MGA-B-47.

Melt inclusions from the other hand samples exhibit no evidence for Fe loss, and have FeO abundances simi-lar to previously published whole rocks from Mangaia. The other inclusions also have H2O (and H2O/Ce) con-centrations that are higher than the H2O in MGA-B-47. Diffusive loss from the inclusions via the olivine, ordiffusive incorporation of water into the melt inclusion, is unlikely to have occurred in inclusions from MGA-B-25 and MG1001. The inclusions from these two lavas form overlapping arrays in plots of CO2 versus H2O andCO2/Nb versus H2O/Ce (Figure 4). It would seem unlikely that diffusive water loss or incorporation in inclusionsfrom two different lavas would generate the same arrays by coincidence, though we note that the arrays arebiased by a single inclusion with the highest CO2/Nb versus H2O/Ce. Diffusive loss of H from the inclusionsmay occur following eruption during slow cooling of lavas, but it is difficult to rule out such a mechanism

Figure 5. Fluorine versus neodymium of Mangaia melt inclusions in comparison to otherOIBs and MORBs. Mangaia melt inclusions are corrected to be in equilibrium with their hostolivine (supporting information Table S5) and are shown as gray symbols following the for-mat of previous figures. Other OIB and MORB data are shown for comparison and consist ofdata obtained from submarine glasses and melt inclusions [Lassiter et al., 2002; Workmanet al., 2006; Koleszar et al., 2009; Kendrick et al., 2014; Metrich et al., 2014]. Data were filteredfor degassing and assimilation as in Figure 4, with the exception of the data from the Aus-trals: Austral samples are divided into three categories, following Lassiter et al. [2002]: Type 1inclusions are not thought to have experienced assimilation, Type 2 inclusions (which plotoutside the figure) are thought to indicate the assimilation of Cl-rich brines, and Type 3inclusions (which also plot outside the figure) are thought to be secondary inclusions. Man-gaia melt inclusions cluster around an average value of F/Nd of approximately 30, which ishigher than the ‘‘canonical’’ F/Nd ratio for OIB and MORB of 21 [e.g., Workman et al., 2006].MGA-B-47 data are shown because, unlike H, F does not show evidence of loss from theinclusion.



lowering the H2O in the melt inclusions from the primary magmatic values. For example, there is no clear rela-tionship between H2O and Ce in the inclusions. However, owing to the fact that small melt inclusions aremore susceptible to diffusive loss of H2O than larger inclusions [Chen et al., 2011], a positive relationshipbetween H2O and melt inclusion diameter will be present if diffusive loss of H2O has occurred. We observe nosuch relationship in inclusions from MGA-B-25 and MG1001 (Figure 9), which argues against diffusive loss ofH2O from the inclusions; however, the inclusions from sample MGA-B-47 that have suffered Fe loss and H2Oloss do show a weak relationship between melt H2O and melt inclusion diameter, which is consistent with dif-fusive loss of H from inclusions in this sample. However, while diffusive loss from MGA-B-25 and MG1001inclusions seems unlikely (given the lack of relationship between H2O and inclusion size), it cannot be ruledout, and the H2O/Ce ratios in the melt inclusions should be treated as minima. For inclusions from samplesMGA-B-25 and MG1001, we also conclude that diffusive exchange through the olivine did not modify theabundance of the other incompatible trace elements, like F, Cl, and S, because their diffusivities are expectedto be significantly slower than those of Fe and H [Cherniak, 2010; Bucholz et al., 2013].

After excluding inclusions from MGA-B-47, H2O/Ce ratios exhibit some variability in Mangaia inclusions.While H2O concentrations and H2O/Ce ratios are relatively high in the Mangaia melt inclusions (excluding

Figure 6. Mixing models demonstrating the effect of assimilating different crustal and brine materials. Mangaia melt inclusion data are plotted as gray symbols following previous figureformatting. Loihi melt inclusion data are outlined in a light gray field and a dark gray field outlines Loihi pillow glasses [Kent et al., 1999a, and references therein]. The model lines andend-member compositions (altered basalt, seawater, and two different brine compositions) for the assimilation model are from Kent et al. [1999a]. The Mangaia primary magma in themodel uses olivine fractionation corrected data from melt inclusion MG1001-14 to define the primary basalt composition because it has the most primitive MgO content (10.8 wt %, cor-rected for olivine fractionation, see supporting information Table S5) in the inclusion suite. Like most of the inclusions in this study, MG1001-14 has a relatively low Cl/K2O ratio (0.064)typical of mantle-derived lavas that have not experienced significant assimilation of hydrothermally altered material. Notably, the composition of one melt inclusion (MG1001A12, whichalso has the highest H2O/K2O and H2O/Ce in the melt inclusion suite) may be explained by the assimilation of altered basalt and/or seawater, and this sample is indicated in all plots ofthis figure. Data from MGA-B-47 and MG1006 inclusions are not shown owing to diffusive water loss.



inclusions from sample MGA-B-47), some H2O loss may occurby diffusive H2O loss frominclusions [e.g., Chen et al.,2011; Gaetani et al., 2012] andslight loss of H2O can occur viadegassing prior to inclusionentrapment (see below), andthis may explain the lowerH2O/Ce ratios (down to 119)observed in some of the Man-gaia inclusions.

4.1.2. Degassing of H2Oand CO2

Models examining solubility ofH2O and CO2 in mantle meltshave shown that, during open-system degassing, the vaporphase is dominated by CO2

until very low pressures (�100bars) [Dixon, 1997]. As in Dixon[1997] and Workman et al.[2006], we interpret the nearlyvertical degassing trend (Fig-ure 4a) and entrapment pres-sures >100 bar to indicatenegligible H2O loss from themelt inclusions from hand sam-ples MG1001 and MGA-B-25.

Owing to the low solubility ofCO2 in mantle melts, in onlyrare cases has CO2 beenobserved to be undersatu-rated, thereby providing infor-mation about primary meltCO2 abundances [Saal et al.,2002]. In the case of Mangaia,and all other OIB and MORBlocations examined to date,extensive degassing of CO2 isevident. Therefore, primarymelt CO2 abundances cannotbe calculated with the presentdata set, as estimates deter-mined for the primary meltCO2 concentration using CO2/Nb ratios have large uncertain-ties owing to variability in theCO2/Nb ratio [Saal et al., 2002;Cartigny et al., 2008; see sec-tion 4.4), particularly if carbo-natite is present duringmelting [Dasgupta et al., 2009].

Figure 7. Cl/Nb versus F/Nd for Mangaia melt inclusions compared to global OIBs andMORBs. Mangaia melt inclusion data are plotted as gray symbols following previous figureformatting. The dashed box represents the canonical F/Nd value from Workman et al. [2006]and the range of Cl/Nb suggested for the mantle [Lassiter et al., 2002; Saal et al., 2002; Work-man et al., 2006; Koleszar et al., 2009; Shaw et al., 2010; Kendrick et al., 2014; Metrich et al.,2014, www.earthchem.org/petdb]. Data from other locations are provided for comparisonand consist of data collected from submarine glasses and melt inclusions [Lassiter et al.,2002; Workman et al., 2006; Koleszar et al., 2009; Kendrick et al., 2014; Metrich et al., 2014].See the caption of Figure 5 for description of Austral sample groups. Data are filtered in thesame way as in the caption of Figure 4.

Figure 8. 208Pb/206Pb versus 207Pb/206Pb measurements on Mangaia melt inclusions fromthis study. Data from this study obtained on homogenized inclusions are shown as solidblack circles (supporting information Table S5). Data from sulfide inclusions are plotted aswhite diamonds; the average of two analyses is plotted for one of the sulfide analyses (sup-porting information Table S5). FOZO values are estimated and are from Hart et al. [1992].The field for previously published whole rock basalt data from Mangaia shows only the dataobtained from leached whole rocks that were measured by MC-ICP-MS [Hanyu et al., 2011]or by a double-spike TIMS method [Woodhead, 1996]. The error ellipse (in the bottom rightportion of the figure) represents the 2r reproducibility (but not the absolute value) of thesecondary standard measurements during the session (these measurements are shown indi-vidually in supporting information Figure S2).



http://www.earthchem.org/petdb

Assimilation during magmaascent: Assimilation of oceaniccrust and/or brines duringascent and eruption of magmacan modify the Cl and H2Oabundances of the magma[Michael and Schilling, 1989;Jambon et al., 1995; Michaeland Cornell, 1998; Kent et al.,1999a, 1999b; Lassiter et al.,2002; Stroncik and Haase, 2004;Kendrick et al., 2013a]. Kentet al. [1999a] explored thisissue through the analysis ofmatrix glass and olivine-hostedglass inclusions from Loihi sea-mount, Hawaii. Following Kentet al. [1999a], we examined thevariability of Cl/K2O and H2O/K2O in the melt inclusions fromMangaia (Figure 6). The major-ity of melt inclusions plot in asmall cloud near the mostprimitive (highest MgO) Man-

gaia melt inclusion composition (see Figure 6 caption for description) and generally exhibit limited variabili-ty, especially in comparison with the variability seen in the Loihi samples, which are known to haveexperienced assimilation. However, we note that assimilation of altered basalt, as shown in Figure 6, canincrease the H2O/K2O (and H2O/Ce) without significantly increasing the Cl/K2O, and thus Cl/K2O alone can-not be used to evaluate assimilation of altered oceanic basalt. However, in plots showing B/K2O, H2O/K2O,B/Cl, and H2O/Cl, one inclusion (MG100A12) is systematically shifted in a direction that suggests assimilationof altered oceanic crust and seawater (Figure 6). This inclusion also has the highest H2O/Ce (245) in thesuite, and the high value may owe to assimilation of H2O. However, the other inclusions do not appear tobe shifted consistently in the direction of assimilation of crustal materials, brines, or seawater. These inclu-sions, which exhibit no clear evidence for assimilation, have H2O/Ce values that range up to �200. However,we cannot exclude the possibility that minor assimilation has occurred in other inclusions in the Mangaia suite,which may result in the incorporation of non-magmatic H2O. If this is the case, then the H2O/Ce of the Man-gaia HIMU melts will be overestimated. Further work analyzing HIMU samples at other localities, particularlyquenched submarine glasses, will be important for constraining the H2O content of this mantle end-member.

4.2. H2O in the HIMU Source and Constraints on Cycling of Subducted Oceanic CrustThe crustal recycling origin hypothesis for the HIMU mantle end-member requires that material (oceaniccrust) once at the Earth’s surface was processed through a subduction zone, experienced long-term storagein the mantle, and was reerupted at an OIB locality, such as Mangaia. The input of water-rich altered oceaniccrust to subduction zones may represent a significant contribution of H2O to the mantle [e.g., Hacker, 2008],and data from the Mangaia melt inclusions can provide constraints on the water that remains in the oceaniccrust during subduction.

In order to calculate representative H2O abundances for the HIMU mantle source, it is important to considerthe melting process itself. Owing to the incompatibility of H2O, mantle melting concentrates water in the meltrelative to the mantle source and subsequent fractional crystallization further increases the concentration ofH2O in the melt. Comparing H2O concentrations to the concentrations of other similarly incompatible ele-ments (e.g., Ce and K) provides a means to ‘‘see’’ through the melting process, as H2O/Ce ratios are not signifi-cantly modified during mantle melting to generate OIB and MORB [e.g., Michael, 1995; Dixon et al., 2002].

When H2O is plotted against K2O and Ce, lavas that erupt at different hot spot localities fall on differenttrends (Figure 10). Melt inclusions from Mangaia have relatively high H2O concentrations compared to other

Figure 9. Melt inclusion H2O contents compared to melt inclusion long axis diameter. Man-gaia melt inclusion data are plotted as gray symbols following previous figure formatting.The data have been corrected to be in equilibrium with the host olivine. The lack of correla-tion for melt inclusions from samples MGA-B-25 and MG1001 suggest no significant diffu-sive H2O loss from the inclusion. However, the negative correlation observed for meltinclusions from hand sample MGA-B-47 does suggest diffusive H2O loss, and is consistentwith the previous observations from this study.



submarine glasses and meltinclusions in the global OIBand MORB data set, but haveintermediate Ce and K2O con-centrations; lavas erupted atEM1 and EM2 localities alsotend to have elevated H2O, buthave higher Ce and K2O at agiven H2O abundance. Lavasfrom the enriched mantle hotspots, which host elevated87Sr/86Sr and lower 143Nd/144Ndratios [Zindler and Hart, 1986;Hofmann, 1997], exhibit Ceconcentrations that are nearly 2times higher at a given H2Oabundance than the HIMUmelts.

Plots of H2O/Ce versus Sr andPb isotopes also show fields fora number of OIB and MORBlocalities (Figure 11). Mangaialavas and melt inclusionsanchor the low 207Pb/206Pband low 208Pb/206Pb region ofthe global data set, and in situSIMS measurements of Pb iso-topes on the Mangaia meltinclusions from this study haverelatively homogeneous com-positions that are similar towhole rocks from the island[Nakamura and Tatsumoto,1988; Dupuy et al., 1989; Hauri

and Hart, 1993; Woodhead,1996; Kogiso et al., 1997; Tat-

sumi et al., 2000; Schiano et al.,2001; Hanyu et al., 2011]. ESClavas with the most radiogenicPb-isotopic compositions(206Pb/204Pb, 207Pb/204Pb, and208Pb/204Pb up to 19.91, 15.64,and 39.66, respectively, whichtranslates to 208Pb/206Pb 5 1.99and 207Pb/206Pb 5 0.785) have

H2O/Ce ratios (up to 244) that are similar to the highest H2O/Ce ratios (�200) in the subset of Mangaia meltinclusions that show no evidence for assimilation. We note that other localities have even higher H2O/Ce,including Pico Island in the Azores (H2O/Ce 5 340) [Metrich et al., 2014] and N. Atlantic MORB (350) [Dixon

et al., 2002], which indicates that the HIMU mantle reservoir does not have the highest H2O/Ce.

We note that Dixon et al. [2002] argued for a lower H2O/Ce ratio for the HIMU mantle of 100, but this is a fac-tor of two lower than the highest H2O/Ce values from the ESC lavas with the most radiogenic Pb-isotopiccompositions [Dixon et al., 2002] and the Mangaia inclusions (�200). In contrast to the HIMU mantle, H2O/

Figure 10. H2O versus Ce and K2O compared to global OIB and MORB data. Mangaia meltinclusion data are plotted as gray symbols following previous figure formatting, and havebeen corrected to be in equilibrium with the host olivine (supporting information Table S5).The H2O contents found in Mangaia melt inclusions are among the highest for OIBs andMORBs. Data arrays from other OIB and MORB localities are plotted for comparison and con-sist of data collected from submarine glasses and melt inclusions [Schilling et al., 1985; Clagueet al., 1995; Douglass et al., 1995; Dixon et al., 1997; Langmuir et al., 1997; Pan and Batiza, 1998;Dixon and Clague, 2001; Dixon et al., 2002; Hauri, 2002; Kingsley, 2002; Saal et al., 2002; Simonset al., 2002; Yang et al., 2003; le Roux et al., 2002; Workman et al., 2004, 2006; Cartigny et al.,2008, and references therein; Koleszar et al., 2009; Kelley et al., 2013; Kendrick et al., 2014;Metrich et al., 2014]. Easter microplate (EMP) and Easter seamount chain (ESC) are abbreviatedfor clarity. Kilauea glass and Hawaii melt inclusion samples [Clague et al., 1995; Hauri, 2002] donot have trace element data and therefore do not plot in the H2O versus Ce plot. The volatilesdatabase was filtered in the same way as described in the caption of Figure 4.



Ce values for EM-type (high 87Sr/86Sr) lavas from Societies, Pitcairn, and Samoa extend to the low H2O/Ceratios (H2O/Ce< 100) [Dixon et al., 2002; Workman et al., 2006; Kendrick et al., 2014]. Thus, the EM mantle sig-nature anchors the high 87Sr/86Sr, low H2O/Ce portion of the global mantle array. Sr-isotopes were notmeasured on the Mangaia inclusions, but if the Sr-isotopic compositions of the inclusions are similar towhole rocks from the island (just as Pb-isotopic compositions of the inclusion suite examined here are simi-lar to the Pb-isotopic compositions of the whole rocks from Mangaia), then Mangaia, together with the ESClavas with the most radiogenic Pb-isotopic ratios, anchors the low 87Sr/86Sr and high H2O/Ce portion of theglobal mantle array.



Given a H2O/Ce ratio of the HIMU mantle of �200, it is possible to calculate the H2O concentration of thismantle end-member if a reliable estimate for the mantle source Ce can be obtained. There are somebounds that can be placed on the Ce concentration of the HIMU mantle that can be used to constrain theH2O abundance in the HIMU mantle source. It is important to note that the Ce concentrations for the HIMUmantle source are not known, and there are many uncertainties associated with estimating the abundancesof incompatible elements in the mantle source of lavas. The following discussion uses estimates from the lit-erature, and assumes that the HIMU mantle is a mixture of altered oceanic crust [Hoffman and White, 1982]and a depleted plume component, which is similar to earlier models for this mantle source [Hauri and Hart,1993; Stracke et al., 2003]. It is reasonable to assume that the Ce concentration will not be more depletedthan the depleted plume component, which is estimated to have a Ce concentration of 1.3 ppm [Jacksonand Jellinek, 2013], but not more enriched than altered oceanic crust, which is estimated to have a Ce con-centration of 6 ppm (as provided by Dixon et al. [2002]). The proportion of the two end-members (alteredoceanic crust and depleted plume component) contributing to the HIMU mantle source is estimated fromprior work suggesting a two-component mixture for generating the HIMU Mangaia mantle source [e.g.,Hauri and Hart, 1993]. If we assume the Mangaia HIMU mantle has a �20% contribution of recycled oceaniccrust, which is consistent with coupled Os-isotopic and Pb-isotopic constraints on Mangaia lavas [Hauri andHart, 1993], and if the remaining 80% of the Mangaia source is composed of the depleted plume compo-nent, then the Mangaia HIMU source has a Ce concentration of 2.2 ppm. Isotopic constraints suggest thathigher proportions of recycled oceanic crust in the HIMU source are unlikely, particularly if the Mangaiamantle source is >2.5 Ga [Hauri and Hart, 1993; Day et al., 2009, 2010]. Dixon et al. [2002] assumed a Ce con-centration of 6 ppm for the HIMU source, which assumes that HIMU is pure recycled altered oceanic crustwithout a depleted mantle contribution; other estimates for the Ce concentration of altered oceanic crustand MORB are available in the literature [Kelley et al., 2003; Gale et al., 2013], but we use the estimate fromDixon et al. [2002] for direct comparison with their model results exploring the H2O abundances in the man-tle. We emphasize that the bulk major element composition of HIMU lavas, and the compositions of the oli-vines, are not well matched by melting a source comprised of purely mafic material [Jackson and Dasgupta,2008; Herzberg et al., 2014], and we do not advocate a model for the melting of a purely mafic lithology inthe Mangaia mantle; instead, Herzberg et al. [2014] argue that the HIMU mantle consists of fertile peridotiteenriched by mixing completely with a minor component of pyroxenite. Similarly, the contribution of 20%mafic material to the HIMU source is assumed to be mixed with the depleted-mantle peridotite such thatthe final mixture is an enriched peridotite [Herzberg, 2011]. Using the estimated H2O/Ce ratio for Mangaiaprimary melts (�200), and a Ce concentration of 2.2 ppm, the H2O concentration in the HIMU mantle sourceis estimated to be 440 ppm.

The primary uncertainty in the estimate of H2O concentration of the HIMU source derives from uncertaintyin the source Ce concentration, which in our calculation scales with the amount of recycled oceanic crust inthe HIMU mantle. We have adopted a recycled oceanic crust mass fraction of 20% in our calculation basedon isotopic constraints: a higher proportion of recycled oceanic crust will yield a higher estimated Ce con-centration and a higher HIMU source H2O concentration, and a lower proportion of oceanic crust yields alower HIMU source H2O concentration.

Using a simple melt model, a HIMU mantle source water content of 440 ppm can easily generate the H2Oabundances estimated for Mangaia primary melts. Following olivine fractionation correction so that themelt is in equilibrium with mantle olivine (Fo90), the inclusion with the highest H2O abundance (MG1001-3b

Figure 11. H2O/Ce versus 87Sr/86Sr, 207Pb/206Pb, and 208Pb/206Pb isotopes for Mangaia melt inclusions and global OIBs and MORBs. The Pb-isotopic data for the Mangaia inclusions (supporting information Table S5) are relatively homogeneous and similar to Mangaia whole rockvalues. Therefore, while Sr-isotopes were not measured in the inclusions, they are also assumed to be homogeneous and similar to the Man-gaia whole rock values. (a) We applied an average 87Sr/86Sr to all of the melt inclusions using data from Hanyu et al. [2011] for age corrected,leached whole rocks. Data from other OIB and MORB sources are provided for comparison and consist of submarine glass samples and meltinclusions [Staudigel et al., 1984; Hannan and Schilling, 1989; Devey et al., 1990; Dosso et al., 1991; Fontignie and Schilling, 1991; Woodhead andDevey, 1993; Douglass et al., 1995; Dixon et al., 1997; Kingsley and Schilling, 1998; Douglass et al., 1999; Dosso et al., 1999; Frey et al., 2000; Dixonand Clague, 2001; Dixon et al., 2002; Kingsley, 2002; Simons et al., 2002; Yang et al., 2003; Honda and Woodhead, 2005; Workman et al., 2006;Cartigny et al., 2008, and references therein; Hanyu et al., 2011; Kelley et al., 2013; Kendrick et al., 2013a, 2014]. Not all data from Figure 10 areshown in this figure because isotopic compositions are not available. The box labeled ‘‘HIMU’’ is the estimated range of the HIMU sourcevalue, and excludes the Mangaia inclusion with the highest H2O/Ce (245; inclusion MG1001A12) owing to possible assimilation of seawateror altered oceanic crust. The volatiles database was filtered in the same way as described in the caption of Figure 4.



has 1.73 wt % H2O, when corrected to be in equilibrium with the host olivine; supporting information TableS5) has a water concentration of 1.37 wt % (and an H2O/Ce ratio of �200). Using mineral-melt partitioncoefficients and a mantle source mineralogy with clinopyroxene (12% modal abundance), orthopyroxene(25%), olivine (55%), and garnet (8%) modes provided in Kelemen et al. [2004] and summarized in Jacksonet al. [2010], and assuming that H2O and Ce have the same mineral-melt partition coefficients during mantlemelting, we use a modal aggregated fractional melting model to explore the conditions under which amantle source with 440 ppm H2O can generate melts with 1.37 wt % H2O.

If the HIMU source has 440 ppm H2O, then the water abundance of the HIMU primary melt (1.37 wt %) isachieved when the melt fraction is 3.0%. This degree of melting is within the range previously suggestedfor melt generation in the Cook-Australs [Herzberg and Gazel, 2009].

The melt model can also be used to provide an estimate of the refractory lithophile element concentrationsin the Mangaia mantle source. When the melt fraction is �3%, the melt model yields a Ce concentration of68.5 ppm in the melt (assuming 2.2 ppm Ce in the HIMU source). This value is identical to the Ce concentra-tion (68.5 ppm) for the inclusion with the highest H2O (MG1001-3b) following olivine fractionation correc-tion so that the melt is in equilibrium with mantle olivine (Fo90), and this trace element composition isadopted as the Mangaia primary melt composition (see Figure 3 caption). Because the melt model does areasonable job of matching the Ce concentration in Mangaia primary melts, we use the same melt model(with �3% melting), together with the estimated Mangaia primary melt trace element composition, toinvert for the source composition of the other trace elements in the primary melt shown in Figure 3 (andpresented in the caption). Using the specified melt model, this source will generate the trace element abun-dances in the Mangaia primary melt composition when the melt fraction is �3%. This source composition ispresented in the caption of Figure 3.

4.3. Dehydration of the Subducted Slab to Generate the HIMU SourceAltered oceanic crust is estimated to have a H2O/Ce ratio of 2500–5000 [Dixon et al., 2002] and a water con-centration of 2–3 wt %. If the HIMU mantle hosts a significant component of recycled oceanic crust, then animportant question remains: How did the H2O/Ce ratio become so low in the HIMU mantle? Subductedmaterial experiences a two-stage dehydration process: (1) stage one is defined by subduction zone process-ing [e.g., Hacker, 2008; van Keken et al., 2011], where H2O is lost from the slab, and (2) stage two is definedby diffusion of water out of the subducted slab during long-term storage in the mantle [e.g., Workmanet al., 2006]. Following Dixon et al. [2002], we can look at the net effect of these two processes by comparingpresubduction H2O/Ce ratios of mature oceanic crust to the H2O/Ce ratios in HIMU melt inclusions eruptedat Mangaia. Compared to the presubduction mature oceanic crust H2O/Ce of 2500–5000, we find that theHIMU mantle source today, which is suggested to be a mixture of recycled material and depleted plumecomponent mantle [e.g., Hauri and Hart, 1993], has an H2O/Ce ratio of �200 (which is a factor of two higherthan the H2O/Ce ratio suggested for the HIMU source [�100] by Dixon et al. [2002]). A reduction of the H2O/Ce ratio from 2500–5000 to 200 represents a factor of 12.5–25 reduction in original H2O content in the sub-ducted slab compared to the loss of Ce.

It is useful to deconvolve the relative effects of stage one of dehydration (subduction zone processing) fromthe effects of stage two of dehydration (long-term mantle storage, diffusive water loss). van Keken et al.[2011] reported estimates of water content entering subduction zones and surviving to reach >230 kmdepth based on petrological models and global subduction zone rock fluxes. Using H2O values of igneouscrust from before subduction (9.3 3 108 Tg/Myr) [Hacker, 2008] and after subduction (1.83108 Tg/Myr) [vanKeken, 2011], igneous crust is estimated to dehydrate 81% from the subduction process alone. This con-strains long-term storage dehydration of the slab by diffusive water loss to account for the remaining reduc-tion of water in the HIMU mantle reservoir, which can efficiently reduce H2O contents in subducted slabson geologic time scales (see below).

Diffusion rates for trace elements (like Ce) are very low and heterogeneities in the solid mantle will not equi-librate with these elements on length scales of more than 10 m over the age of the Earth [e.g., Hofmannand Hart, 1978; Van Orman et al., 2001]. In contrast, hydrogen diffuses quickly and the H2O/Ce ratio can bereduced by diffusive loss of H2O into the ambient mantle. Using the H2O concentration values of alteredoceanic crust from Dixon et al. [2002] (2–3 wt %) and the amount of dehydration calculated above (81%),the recycled slab after the subduction zone would have 0.38–0.57 wt % (3800–5700 ppm) H2O. If the



Mangaia mantle, which is estimated to have 440 ppm water, is composed of recycled slab (20% by mass)and a depleted component (80%) that has a water content similar to ambient mantle (116 ppm) [Saltersand Stracke, 2004], then the slab component in the Mangaia mantle has a water content of �1740 ppm.This is 2.2–3.3 times lower than the estimated water content of a slab subducted to postarc depths, and weargue that diffusive loss from the slab during long-term storage in the mantle can reduce the water contentin the slab from postsubduction values.

We model the effect of diffusion in a manner similar to Workman et al. [2006]. A diffusion calculation [seeZhang, 2008, equation (3-48b)] was performed assuming the placement of postsubduction mature oceaniccrust in an infinite reservoir of depleted mantle. We estimate depleted mantle to have 116 ppm H2O [Saltersand Stracke, 2004], adopt a thickness of 7 km for the slab, and assume a storage time of 2.5 Ga. The range ofwater contents remaining in the slab after subduction (0.38–0.57 wt % H2O) are used as inputs for the diffu-sion calculation. The diffusion coefficient of H in olivine was found to be 1.4 3 1028 m2/s at 1600�C for the[100] (fast) axis by Mackwell and Kohlstedt [1990]. In this simple diffusion calculation, following subduction,the slab experiences approximately 90% diffusive loss of water during 2.5 Ga of residence in the mantle: thewater content of the slab is reduced from 3800 to 5700 ppm to approximately 340 to 450 ppm. Thus, diffu-sive loss of H2O from the slab following subduction zone dehydration would make the slab much drier(340–450 ppm H2O) than our model for the slab (1740 ppm H2O) in the HIMU mantle. However, if the slabis thicker, or slabs pile up on each other in the mantle (effectively thickening the diffusive distance), thenthe diffusive loss of H2O would be reduced. Additionally, we note that pyroxene-rich and garnet-rich litholo-gies have a higher water storage capacity than olivine-rich lithologies [Hirschmann et al., 2009; O’Leary et al.,2010; Ardia et al., 2012], and higher modal pyroxene and garnet abundances in the HIMU mantle sourcemay allow the preservation of higher water abundances during long-term storage in the mantle. This is notan unreasonable model, as HIMU sources have long been suggested to host a significant mafic component[Hofmann and White, 1982; Hauri and Hart, 1993; Day et al., 2009] or to be a fertile peridotite that has beenenriched by mixing completely with a component of pyroxenite (thereby increasing the pyroxene abun-dance in the enriched peridotite resulting from the mixture) [Herzberg et al., 2014].

4.4. CO2 in the Generation of HIMU LavasIt is apparent that the CO2 concentrations in Mangaia melt inclusions are high compared to CO2 abundan-ces measured in many other OIB settings (Figure 4b). In fact, carbonate blebs were previously noted in crys-talline melt inclusions hosted in olivines [Saal et al., 1998; Yurimoto et al., 2004], and Saal et al. [1998] foundthat melt inclusions from two of the whole rocks (MGA-B-47 and MGA-B-25) examined in this study host car-bonatite blebs (unfortunately, we did not examine these particular melt inclusions). We infer that someinclusions in the present study may have also hosted carbonate blebs prior to homogenization, but we didnot observe carbonatite blebs in the inclusions posthomogenization. Saal et al. [1998] suggested that carbo-natite plays a key role in the genesis of HIMU lavas, and Hauri et al. [1993] found that carbonatite plays animportant role in modifying mantle xenoliths from the island of Tubuai (located in the Cook-Australs). Theassociation of carbonate with lavas derived from the HIMU mantle source could be an important clue forultimately understanding the origin of this mantle end-member [Jackson and Dasgupta, 2008].

Partial melt compositions of carbonated mafic lithologies are shown experimentally to give rise to meltsthat are similar to the inferred compositions of HIMU primary melts (matching geochemical characteristicssuch as SiO2, TiO2, Al2O3, CaO, Na2O, and CaO/Al2O3) [Dasgupta et al., 2004; Gerbode and Dasgupta, 2010;Mallik and Dasgupta, 2012]. It has also been suggested that there is interaction of this melt product with sur-rounding peridotite, which leads to a closer match for FeO and is consistent with a recent study of theHIMU mantle source lithology by Herzberg et al. [2014] [Dasgupta et al., 2004; Gerbode and Dasgupta, 2010;Mallik and Dasgupta, 2012]. Experimental evidence shows that it is difficult to generate strongly SiO2-under-saturated and high CaO/Al2O3 melt compositions, like those observed in HIMU lavas, without high concen-trations of CO2 in the mantle source [Kushiro, 1975; Eggler, 1978; Dasgupta et al., 2004, 2007; Gerbode andDasgupta, 2010; Mallik and Dasgupta, 2012], which strongly suggests that the HIMU mantle end-member isCO2-rich.

Unfortunately, it is not possible to reliably estimate the CO2 abundance in the primary melts of the HIMUmantle owing to large uncertainties in the CO2/Nb ratio that has been used to estimate mantle source CO2

abundances [Saal et al., 2002]. The CO2/Nb ratio of the depleted mantle is estimated to be 300 [Koleszar



et al., 2009], which is slightly higher than the value of 239 that was originally proposed [Saal et al., 2002],and this is likely a lower limit for the CO2/Nb ratio of the OIB mantle. Cartigny et al. [2008] observed a rela-tionship between CO2/Nb and (La/Sm)N in MORB. If we assume that this relationship applies to the Mangaiamelts, which have an average (La/Sm)N of 2.9 (supporting information Table S5), then the CO2/Nb of theMangaia mantle source is �1000; this value, when paired with the estimated Nb concentration of the HIMUmantle source (see caption of Figure 3), yields unreasonably high CO2 in Mangaia primary melts. If carbona-tite is present during melting, CO2 and Nb will have dramatically different partition coefficients, and CO2/Nbratios in the primary melt will exceed 104 [Dasgupta et al., 2009]. This serves to underscore the great uncer-tainty in the CO2/Nb ratio of primary melts, particularly when carbonatite is present.

While it is not possible to reliably estimate the CO2 in abundance in the HIMU mantle source or in HIMU pri-mary melts, the HIMU mantle is thought to have elevated CO2 [Hauri et al., 1993; Saal et al., 1998; Jacksonand Dasgupta, 2008; Castillo, 2014]. The origin of this elevated CO2 may be traced to the recycled crustal ori-gin of the HIMU mantle, where CO2-rich and H2O-rich altered oceanic crust [e.g., Staudigel et al., 1996; Altand Teagle, 1999; Bach et al., 2001; Wallmann, 2001; Dixon et al., 2002; Gillis and Coogan, 2011] is subductedinto the mantle and recycled into the origin of HIMU lavas. The carbonated nature of the HIMU mantlesource is consistent with experimental work attempting to model the major element composition of HIMUlavas [Dasgupta et al., 2004; Gerbode and Dasgupta, 2010; Mallik and Dasgupta, 2012] and has importantimplications for the deep storage of recycled carbon in the mantle.

4.5. F/Nd and Cl/Nb Signature of HIMU MeltsThe F and Cl concentrations in melt inclusions are useful for constraining the undegassed, preeruptive com-positions because their degassing pressure is typically less than 100 bars [Spilliaert et al., 2006]. Additionally,F and Cl are not thought to diffuse in or out of the host olivine [Bucholz et al., 2013]. Normalizing F and Cl toelements with similar bulk partition coefficients, namely Nd and Nb, respectively [Dalou, 2011], the F/Ndand Cl/Nb ratios measured in melt inclusions have been used to characterize those of the source [e.g., Rose-Koga et al., 2012]. Known as a canonical ratio, F/Nd is largely insensitive to mantle melting, and natural sam-ples (melt inclusions and glasses from MORB and OIB) display a relatively constant value of 21 over a rangeof Nd concentrations that vary over two orders of magnitude [Workman et al., 2006]. In cases where a meltinclusion has undergone assimilation of crustal materials or brines, F/Nd as high as 60–183 have beenobserved [Lassiter et al., 2002]. However, such inclusions are often accompanied by extremely elevated Cl/Nb ratios (from 1271 to 7166; see section 2) [Lassiter et al., 2002]. Typically, Cl/Nb ranges from 5 to 35 amongthe data of MORB and OIB glasses and melt inclusions [Lassiter et al., 2002; Saal et al., 2002; Workman et al.,2006; Koleszar et al., 2009; Shaw et al., 2010; Kendrick et al., 2014; Metrich et al., 2014; www.earthchem.org/petdb]. Notably, once altered oceanic crust assimilation has been identified and excluded, any deviationfrom the relatively small range of these geochemical ratios indicates the presence of material that is notconsidered to be typical mantle.

Mangaia melt inclusions display Cl/Nb values similar to MORB melt inclusions and samples from other OIBs,such as the Azores and Samoa (Figure 7). The Cl/Nb ratio of the source of Mangaia is likely similar to that ofthe MORB source. On the contrary, Mangaia melt inclusions show a higher F/Nd (30 6 9; 2r standard devia-tion) than the ‘‘MORB-OIB’’ canonical value of 21, and the Mangaia inclusions range up to a F/Nd ratio of 41.The higher F/Nd values are mainly related to higher F concentrations in the Mangaia melt inclusions (about30% higher), while Nd concentrations are comparable to other OIB data (e.g., 45 6 5 ppm; the Azores, Samoa,and this study). The Nd concentration in the HIMU source is estimated to be 1.38 ppm (see Figure 3), which,when combined with the average Mangaia melt inclusion F/Nd ratio (30), yields F abundances in the HIMUsource of 41 ppm. By comparison, the F concentrations of DMM and chondrite-based primitive mantle areestimated to be 11 and 25 ppm, respectively [McDonough and Sun, 1995; Salters and Stracke, 2004].

The recycled HIMU component with high H2O and CO2 is also inferred to have high F. While F and Cl arelikely elevated in the subducting slab, most of the subduction input is expected to be stripped off duringprograde dehydration, as reflected in elevated F and Cl concentrations in many arc magmas and forearc ser-pentines [e.g., Straub and Layne, 2003; John et al., 2011; Rose-Koga et al., 2012; Kendrick et al., 2013b; Debretet al., 2014]. The data from Mangaia melt inclusions suggest that the recycled component in the HIMUsource is characterized by the fractionation of F from Cl, as Mangaia samples have Cl/Nb ratios similar toambient mantle, while F/Nd and H2O/Ce ratios remain higher than ambient mantle. However, the precise





mechanism responsible for fractionation of F and H2O from Cl in subduction zones is not well known. Thehydroxyl site in silicate minerals—which has a strong affinity for F [e.g., Smith, 1981; Smith et al., 1981; Wuand Koga, 2013]—provides an ideal mineralogical site to promote recycling of H2O and F, but not Cl. Thus,the residual mineralogy of the subducted slab after it has gone through prograde dehydration/melting,including phlogopite, humite, apatite, and/or amphiboles (together with their high pressure polymorphs),will play an important role in the fractionation of H2O and F from Cl [Rose-Koga et al., 2014].

5. Conclusions

Pb-isotopic studies of lavas from Mangaia have established this locality to be the surface expression of theHIMU mantle end-member, which has long been considered to contain subducted oceanic crust in themantle source [Hofmann and White, 1982; Zindler and Hart, 1986; Hauri and Hart, 1993; Kelley et al., 2005;Hanyu et al., 2011]. Sulfur isotopic work on melt inclusions from the island has established that the materialerupted at Mangaia was once at the Earth’s surface> 2.45 Ga, confirming the recycled origin hypothesis forthe HIMU mantle [Cabral et al., 2013].

Previous work on HIMU melt inclusions has been restricted to the study of crystalline inclusions from Mangaia[Saal et al., 1998; Yurimoto et al., 2004; Paul et al., 2011], which cannot be used to measure volatile abundan-ces. Additionally, because of the lack of dredging expeditions to this location, the scientific community hasbeen without glasses from Mangaia on which to perform volatile measurements. Here, we have homogenizedcrystalline melt inclusions hosted in olivine phenocrysts from the island. We have performed in situ volatile,trace, and major element analyses in the melt inclusions, in addition to in situ Pb-isotopic analyses. We findMangaia melt inclusions host elevated H2O (up to 1.73 wt %) and CO2 (up to 2346 ppm) contents.

We suggest that the H2O abundance in the HIMU mantle source is �440 ppm, which can generate the H2Oabundance observed in the melt inclusion with the highest H2O concentration when the melt fraction is�3%. The inclusions have relatively high concentrations of CO2 (up to 2346 ppm), which is consistent withprevious observations of carbonatite in Mangaia melt inclusions. The high levels of CO2 in HIMU lavas, andhence the HIMU mantle source, are consistent with the experimental studies that require a carbonated pro-tolith to generate the unique major element composition of the melts.

We also find the F/Nd ratios of the melt inclusions (30 6 9; 2r standard deviation) to be higher than thecanonical value (21), while Cl/Nb ratios in Mangaia inclusions are likely similar to the MORB mantle. We sug-gest that this may owe to the influence of hydroxyl sites in silicate minerals, which may promote recyclingof H2O and F, but not Cl.

ReferencesAllegre, C. J. (1982), Chemical geodynamics, Tectonophysics, 81, 109–132.Alt, J. C., and D. A. H. Teagle (1999), The uptake of carbon during alteration of ocean crust, Geochim. Cosmochim. Acta, 63, 1527–1535.Ardia, P., M. M. Hirschmann, A. C. Withers, and T. J. Tenner (2012), H2O storage capacity of olivine at 5–8GPa and consequences for dehy-

dration partial melting of the upper mantle, Earth Planet. Sci. Lett., 345–348, 104–116, doi:10.1016/j.epsl.2012.05.038.Asimow, P. D., and C. H. Langmuir (2003), The importance of water to oceanic mantle melting regimes, Nature, 421, 815–820.Bach, W., J. C. Alt, Y. Niu, S. E. Humphris, J. Erzinger, and H. J. B. Dick (2001), The geochemical consequences of late-stage low-grade altera-

tion of lower ocean crust at the SW Indian Ridge: Results from ODP Hole 735B (Leg 176), Geochim. Cosmochim. Acta, 65, 3267–3287.Bucholz, C. E., G. A. Gaetani, M. D. Behn, and N. Shimizu (2013), Post-entrapment modification of volatiles and oxygen fugacity in olivine-

hosted melt inclusions, Earth Planet. Sci. Lett., 374, 145–155, doi:10.1016/j.epsl.2013.05.033.Cabral, R. A., M. G. Jackson, E. F. Rose-Koga, K. T. Koga, M. J. Whitehouse, M. A. Antonelli, J. Farquhar, J. M. D. Day, and E. H. Hauri (2013),

Anomalous sulphur isotopes in plume lavas reveal deep mantle storage of Archaean crust, Nature, 496, 490–493, doi:10.1038/nature12020.

Cartigny, P., F. Pineau, C. Aubaud, and M. Javoy (2008), Towards a consistent mantle carbon flux estimate: Insights from volatile systematics(H2O/Ce, dD, CO2/Nb) in the North Atlantic mantle (14� N and 34� N), Earth Planet. Sci. Lett., 265, 672–685, doi:10.1016/j.epsl.2007.11.011.

Castillo, P. R. (2014), Recycling of marine carbonate as a possible solution to the Pb paradox, Mineral. Mag., 358.Chan, L.-H., J. C. Lassiter, E. H. Hauri, S. R. Hart, and J. Blusztajn (2009), Lithium isotope systematics of lavas from the Cook-Austral Islands:

Constraints on the origin of HIMU mantle, Earth Planet. Sci. Lett., 277, 433–442, doi:10.1016/j.epsl.2008.11.009.Chase, C. G. (1981), Oceanic island Pb: Two-stage histories and mantle evolution, Earth Planet. Sci. Lett., 52, 277–284.Chauvel, C., A. Hofmann, and P. Vidal (1992), HIMU-EM: The French Polynesian connection, Earth Planet. Sci. Lett., 110, 99–119.Chauvel, C., W. McDonough, G. Guille, R. Maury, and R. Duncan (1997), Contrasting old and young volcanism in Rurutu Island, Austral chain,

Chem. Geol., 139, 125–143.Chen, Y., A. Provost, P. Schiano, and N. Cluzel (2011), The rate of water loss from olivine-hosted melt inclusions, Contrib. Mineral. Petrol.,

162(3), 625–636, doi:10.1007/s00410-011-0616-5.

AcknowledgmentsM.J. acknowledges NSF grants EAR-1145202, EAR-1348082, EAR-1347377,and OCE-1153894 that supported thiswork. E.F.R.-K. thanks the EuropeanSynthesys FP7 ‘‘Capacities’’ SpecificProgram for financing part of theanalytical cost of this research.K.T.K. acknowledges French ANR grantANR-09-BLAN-038 (project SlabFlux)that supported this work. The Nordsimfacility is funded and operated as ajoint Nordic research infrastructureunder an agreement with NOS-N. Thisis Laboratory of Excellence ClerVolc(ANR-10-LABX-006) contribution 126and Nordsim laboratory contribution376. We thank J.E. Dixon, P.J. le Roux,A.E. Saal, and P. Cartigny for advice incompiling the OIB and MORB databasethat was used for creating many of thefigures. We also thank P.S. Hall for hisassistance in clarifying the diffusionmodel. We acknowledge constructivereviews from Bill White, John Lassiter,and an anonymous reviewer.



http://dx.doi.org/10.1016/j.epsl.2012.05.038


http://dx.doi.org/10.1038/nature12020





http://dx.doi.org/10.1007/s00410-011-0616-5

Cherniak, D. J. (2010), Diffusion in accessory minerals: Zircon, titanite, apatite, monazite and xenotime, Rev. Mineral. Geochem., 72(1), 827–869, doi:10.2138/rmg.2010.72.18.

Clague, D. A., J. G. Moore, J. E. Dixon, and W. B. Friesen (1995), Petrology of submarine lavas from Kilauea’s Puna Ridge, Hawaii, J. Petrol., 36,299–349.

Cohen, R. S., and R. K. O’Nions (1982), Identification of recycled continental material in the mantle from Sr, Nd and Pb isotope investiga-tions, Earth Planet. Sci. Lett., 61, 73–84.

Dalou, C. (2011), Fluorine and chlorine fractionation in the sub-arc mantle: An experimental investigation, thesis, Univ. Blaise Pascal,Clermont-Ferrand, France.

Danyushevsky, L. V., A. V. Sobolev, and N. N. Kononkova (1992), Methods of studying melt inclusions in minerals during investigations onwater-bearing primitive mantle melts (Tonga Trench boninites), Geochem. Int., 29, 48–62.

Danyushevsky, L. V., F. N. Della-Pasqua, and S. Sokolov (2000), Re-equilibration of melt inclusions trapped by magnesian olivine phenoc-rysts from subduction-related magmas: Petrological implications, Contrib. Mineral. Petrol., 138, 68–83.

Dasgupta, R., M. M. Hirschmann, and A. C. Withers (2004), Deep global cycling of carbon constrained by the solidus of anhydrous, carbo-nated eclogite under upper mantle conditions, Earth Planet. Sci. Lett., 227, 73–85, doi:10.1016/j.epsl.2004.08.004.

Dasgupta, R., M. M. Hirschmann, and N. D. Smith (2007), Partial melting experiments of peridotite 1 CO2 at 3 GPa and genesis of AlkalicOcean Island Basalts, J. Petrol., 48(11), 2093–2124, doi:10.1093/petrology/egm053.

Dasgupta, R., M. M. Hirschmann, W. F. McDonough, M. Spiegelman, and A. C. Withers (2009), Trace element partitioning between garnetlherzolite and carbonatite at 6.6 and 8.6 GPa with applications to the geochemistry of the mantle and of mantle-derived melts, Chem.Geol., 262, 57–77, doi:10.1016/j.chemgeo.2009.02.004.

Day, J. M. D., D. G. Pearson, C. G. Macpherson, D. Lowry, and J.-C. Carracedo (2009), Pyroxenite-rich mantle formed by recycled oceaniclithosphere: Oxygen-osmium isotope evidence from Canary Island lavas, Geology, 37(6), 555–558, doi:10.1130/G25613A.1.

Day, J. M. D., D. G. Pearson, C. G. Macpherson, D. Lowry, and J. C. Carracedo (2010), Evidence for distinct proportions of subducted oceaniccrust and lithosphere in HIMU-type mantle beneath El Hierro and La Palma, Canary Islands, Geochim. Cosmochim. Acta, 74(22), 6565–6589, doi:10.1016/j.gca.2010.08.021.

Debret, B., K. T. Koga, C. Nicollet, M. Andreani, and S. Schwartz (2014), F, Cl and S input via serpentinite in subduction zones: Implicationsfor the nature of the fluid released at depth, Terra Nova, 26(2), 96–101, doi:10.1111/ter.12074.

Devey, C. W., F. Albarede, J.-L. Chemin�ee, A. Michard, R. M€uhe, and P. Stoffers (1990), Active submarine volcanism on the society hotspotswell (west Pacific): A geochemical study, J. Geophys. Res., 95, 5049–5066.

Dixon, J., D. A. Clague, P. Wallace, and R. Poreda (1997), Volatiles in alkalic basalts form the North Arch Volcanic Field, Hawaii: Extensivedegassing of deep submarine-erupted alkalic series lavas, J. Petrol., 38(7), 911–939.

Dixon, J. E. (1997), Degassing of alkalic basalts, Am. Mineral., 82, 368–378.Dixon, J. E., and D. A. Clague (2001), Volatiles in basaltic glasses from Loihi Seamount, Hawaii: Evidence for a relatively dry plume compo-

nent, J. Petrol., 42, 627–654.Dixon, J. E., L. Leist, C. Langmuir, and J.-G. Schilling (2002), Recycled dehydrated lithosphere observed in plume-influenced mid-ocean-

ridge basalt, Nature, 420, 385–389.Dosso, L., B. B. Hanan, H. Bougault, J.-G. Schilling, and J.-L Joron (1991), Sr-Nd-Pb geochemical morphology between 10� and 17�N on the

Mid-Atlantic Ridge: A new MORB isotope signature, Earth Planet. Sci. Lett., 106, 29–43.Dosso, L., H. Bougault, C. Langmuir, C. Bollinger, O. Bonnier, and J. Etoubleau (1999), The age and distribution of mantle heterogeneity

along the Mid-Atlantic Ridge (31–41�N), Earth Planet. Sci. Lett., 170, 269–286.Douglass, J., J.-G. Schilling, and R. H. Kingsley (1995), Influence of the discovery and Shona mantle plumes on the southern Mid-Atlantic

Ridge: Rare earth evidence, Geophys. Res. Lett., 22, 2893–2896.Douglass, J., J.-G. Schilling, and D. Fontignie (1999), Plume-ridge interactions of the discovery and Shona mantle plumes with the southern

Mid-Atlantic Ridge (40�–55�S), J. Geophys. Res., 104, 2941–2962.Dupuy, C., H. G. Barsczus, J. Dostal, P. Vidal, and J.-M. Liotard (1989), Subducted and recycled lithosphere as the mantle source of ocean

island basalts from southern Polynesia, central Pacific, Chem. Geol., 77, 1–18.Eggler, D. (1978), The effect of CO2 upon partial melting of peridotite in the system Na2O-CaO-Al2O3-MgO-SiO2-CO2 to 35 Kb, with an anal-

ysis of melting in peridotite-H2O-CO2 system, Am. J. Sci., 278, 305–343.Eiler, J. M., K. A. Farley, J. W. Valley, E. M. Stolper, E. H. Hauri, and H. Craig (1995), Oxygen isotope evidence against bulk recycled sediment

in the mantle sources of Pitcairn Island lavas, Nature, 377, 138–141.Eisele, J., M. Sharma, S. J. G. Galer, J. Blichert-Toft, C. W. Devey, and A. W. Hofmann (2002), The role of sediment recycling in EM-1 inferred

from Os, Pb, Hf, Nd, Sr isotope and trace element systematics of the Pitcairn hotspot, Earth Planet. Sci. Lett., 196, 197–212.Fontignie, D., and J.-G. Schilling (1991), 87Sr/86Sr and REE variations along the Easter Microplate boundaries (south Pacific): Application of

multivariate statistical analyses to ridge segmentation, Chem. Geol., 89, 209–241.Frey, F. A., D. Clague, J. J. Mahoney, and J. M. Sinton (2000), Volcanism at the edge of the Hawaiian plume: Petrogenesis of submarine

alkalic lavas from the North Arch volcanic field, J. Petrol., 41, 667–691.Gaetani, G. A., and T. L. Grove (1998), The influence of water on melting of mantle peridotite, Contrib. Mineral. Petrol., 131, 323–346.Gaetani, G. A., and E. B. Watson (2000), Open system behavior of olivine-hosted melt inclusions, Earth Planet. Sci. Lett., 183, 27–41.Gaetani, G. A., J. A. O’Leary, N. Shimizu, C. E. Bucholz, and M. Newville (2012), Rapid reequilibration of H2O and oxygen fugacity in olivine-

hosted melt inclusions, Geology, 40(10), 915–918, doi:10.1130/G32992.1.Gale, A., M. Laubier, S. Escrig, and C. H. Langmuir (2013), Constraints on melting processes and plume-ridge interaction from comprehen-

sive study of the FAMOUS and North Famous segments, Mid-Atlantic Ridge, Earth Planet. Sci. Lett., 365, 209–220, doi:10.1016/j.epsl.2013.01.022.

Gast, P. W., G. R. Tilton, and C. Hedge (1964), Isotopic composition of lead and strontium from Ascension and Gough Islands, Science, 145,1181–1185.

Gerbode, C., and R. Dasgupta (2010), Carbonate-fluxed melting of MORB-like pyroxenite at 2.9 GPa and genesis of HIMU Ocean IslandBasalts, J. Petrol., 51(10), 2067–2088, doi:10.1093/petrology/egq049.

Gillis, K. M., and L. A. Coogan (2011), Secular variation in carbon uptake into the ocean crust, Earth Planet. Sci. Lett., 302, 385–392, doi:10.1016/j.epsl.2010.12.030.

Graham, D., S. Humphris, and W. Jenkins (1992), Helium isotope geochemistry of some volcanic rocks from Saint Helena, Earth Planet. Sci.Lett., 110, 121–131.

Hacker, B. R. (2008), H2O subduction beyond arcs, Geochem. Geophys. Geosyst., 9, Q03001, doi:10.1029/2007GC001707.



http://dx.doi.org/10.2138/rmg.2010.72.18


http://dx.doi.org/10.1093/petrology/egm053

http://dx.doi.org/10.1016/j.chemgeo.2009.02.004

http://dx.doi.org/10.1130/G25613A.1

http://dx.doi.org/10.1016/j.gca.2010.08.021

http://dx.doi.org/10.1111/ter.12074

http://dx.doi.org/10.1130/G32992.1



http://dx.doi.org/10.1093/petrology/egq049


http://dx.doi.org/10.1029/2007GC001707

Hanyu, T., and I. Kaneoka (1997), The uniform and low 3He/4He ratios of HIMU basalts as evidence for their origin as recycled materials,Nature, 390, 273–276.

Hanyu, T., et al. (2011), Geochemical characteristics and origin of the HIMU reservoir: A possible mantle plume source in the lower mantle,Geochem. Geophys. Geosyst., 12, Q0AC09, doi:10.1029/2010GC003252.

Hanyu, T., et al. (2014), Isotope evolution in the HIMU reservoir beneath St. Helena: Implications for the mantle recycling of U and Th, Geo-chim. Cosmochim. Acta, 143, 232–252, doi:10.1016/j.gca.2014.03.016.

Hart, S. (2011), The Mantle Zoo: New species, endangered species, extinct species, Mineral. Mag., 3, 983.Hart, S., and H. Staudigel (1982), The control of alkalies and uranium in seawater by ocean crust alteration, Earth Planet. Sci. Lett., 58, 202–

212.Hart, S. R. (1988), Heterogeneous mantle domains: Signatures, genesis and mixing chronologies, Earth Planet. Sci. Lett., 90, 273–296.Hart, S. R., E. H. Hauri, L. A. Oschmann, and J. A. Whitehead (1992), Mantle plumes and entrainment: Isotopic evidence, Science, 256, 517–

520.Hauri, E. (2002), SIMS analysis of volatiles in silicate glasses. 2: Isotopes and abundances in Hawaiian melt inclusions, Chem. Geol., 183, 115–

141.Hauri, E. H., and S. R. Hart (1993), Re-Os isotope systematics of HIMU and EMII oceanic island basalts from the south Pacific Ocean, Earth

Planet. Sci. Lett., 114, 353–371.Hauri, E. H., and S. R. Hart (1997), Rhenium abundances and systematics in oceanic basalts, Chem. Geol., 139, 185–205.Hauri, E., N. Shimizu, J. Dieu, and S. Hart (1993), Evidence for hotspot-related carbonatite metasomatism in the oceanic upper mantle,

Nature, 365, 221–227.H�emond, C., C. Devey, and C. Chauvel (1994), Source compositions and melting processes in the Society and Austral plumes (South Pacific

Ocean): Element and isotope (Sr, Nd, Pb, Th) geochemistry, Chem. Geol., 115, 7–45.Herzberg, C. (2011), Identification of source lithology in the Hawaiian and Canary Islands: Implications for origins, J. Petrol., 52(1), 113–146,

doi:10.1093/petrology/egq075.Herzberg, C., and E. Gazel (2009), Petrological evidence for secular cooling in mantle plumes, Nature, 458, 619–622, doi:10.1038/

nature07857.Herzberg, C., R. A. Cabral, M. G. Jackson, C. Vidito, J. M. D. Day, and E. H. Hauri (2014), Phantom Archean crust in Mangaia hotspot lavas and

the meaning of heterogeneous mantle, Earth Planet. Sci. Lett., 396, 97–106.Hirschmann, M. M. (2006), Water, melting, and the deep Earth H2O cycle, Annu. Rev. Earth Planet. Sci., 34, 629–653, doi:10.1146/10.1146/

annurev.earth.34.031405.125211.Hirschmann, M. M., T. Tenner, C. Aubaud, and A. C. Withers (2009), Dehydration melting of nominally anhydrous mantle: The primacy of

partitioning, Phys. Earth Planet. Inter., 176, 54–68, doi:10.1016/j.pepi.2009.04.001.Hirth, G., and D. L. Kohlstedt (1996), Water in the oceanic upper mantle: Implications for rheology, melt extraction and the evolution of the

lithosphere, Earth Planet. Sci. Lett., 144, 93–108.Hoernle, K., G. Tilton, M. J. Le Bas, S. Duggen, and D. Garbe-Sch€onberg (2002), Geochemistry of oceanic carbonatites compared with conti-

nental carbonatites: Mantle recycling of oceanic crustal carbonate, Contrib. Mineral. Petrol., 142, 520–542.Hofmann, A. W. (1997), Mantle geochemistry: The message from oceanic volcanism, Nature, 385, 219–229.Hofmann, A. W., and S. R. Hart (1978), An assessment of local and regional isotopic equilibrium in the mantle, Earth Planet. Sci. Lett., 38,

44–62.Hofmann, A. W., and W. M. White (1982), Mantle plumes from ancient oceanic crust, Earth Planet. Sci. Lett., 57, 421–436.Honda, M., and J. D. Woodhead (2005), A primordial solar-neon enriched component in the source of EM-I-type ocean island basalts from

the Pitcairn Seamounts, Polynesia, Earth Planet. Sci. Lett., 236, 597–612, doi:10.1016/j.epsl.2005.05.038.Jackson, M. G., and R. Dasgupta (2008), Compositions of HIMU, EM1, and EM2 from global trends between radiogenic isotopes and major

elements in ocean island basalts, Earth Planet. Sci. Lett., 276, 175–186, doi:10.1016/j.epsl.2008.09.023.Jackson, M. G., and A. M. Jellinek (2013), Major and trace element composition of the high 3He/4He mantle: Implications for the composi-

tion of a nonchonditic Earth, Geochem. Geophys. Geosyst., 14, 2954–2976, doi:10.1002/ggge.20188.Jackson, M. G., S. R. Hart, A. A. P. Koppers, H. Staudigel, J. Konter, J. Blusztajn, M. Kurz, and J. A. Russell (2007), The return of subducted con-

tinental crust in Samoan lavas, Nature, 448, 684–687, doi:10.1038/nature06048.Jackson, M. G., R. W. Carlson, M. D. Kurz, P. D. Kempton, D. Francis, and J. Blusztajn (2010), Evidence for the survival of the oldest terrestrial

mantle reservoir, Nature, 466, 853–856, doi:10.1038/nature09287.Jackson, M. G., D. Weis, and S. Huang (2012), Major element variations in Hawaiian shield lavas: Source features and perspectives from

global ocean island basalt (OIB) systematics, Geochem. Geophys. Geosyst., 13, Q09009, doi:10.1029/2012GC004268.Jambon, A., B. D�eruelle, G. Dreibus, and F. Pineau (1995), Chlorine and bromine abundance in MORB: The contrasting behaviour of the

Mid-Atlantic Ridge and East Pacific Rise and implications for chlorine geodynamic cycle, Chem. Geol., 126, 101–117.John, T., G. D. Layne, K. M. Haase, and J. D. Barnes (2010), Chlorine isotope evidence for crustal recycling into the Earth’s mantle, Earth

Planet. Sci. Lett., 298(1–2), 175–182, doi:10.1016/j.epsl.2010.07.039.John, T., M. Scambelluri, M. Frische, J. D. Barnes, and W. Bach (2011), Dehydration of subducting serpentinite: Implications for halogen

mobility in subduction zones and the deep halogen cycle, Earth Planet. Sci. Lett., 308(1–2), 65–76, doi:10.1016/j.epsl.2011.05.038.Kawabata, H., T. Hanyu, Q. Chang, J.-I. Kimura, A. R. L. Nichols, and Y. Tatsumi (2011), The Petrology and geochemistry of St. Helena alkali

basalts: Evaluation of the oceanic crust-recycling model for HIMU OIB, J. Petrol., 52(4), 791–838, doi:10.1093/petrology/egr003.Kelemen, P. B., G. M. Yogodzinski, and D. W. Scholl (2003), Along-strike variation in lavas of the Aleutian island arc: Implications for the gen-

esis of high Mg# andesite and the continental crust. Inside the Subduction Factory, Geophys. Monogr. Ser, 138, 223–276.Kelley, K. A., T. Plank, J. Ludden, and H. Staudigel (2003), Composition of altered oceanic crust at ODP Sites 801 and 1149, Geochem. Geo-

phys. Geosyst., 4(6), 8910, doi:10.1029/2002GC000435.Kelley, K. A., T. Plank, L. Farr, J. Ludden, and H. Staudigel (2005), Subduction cycling of U, Th, and Pb, Earth Planet. Sci. Lett., 234, 369–383,

doi:10.1016/j.epsl.2005.03.005.Kelley, K. A., R. Kingsley, and J.-G. Schilling (2013), Composition of plume-influenced mid-ocean ridge lavas and glasses from the Mid-

Atlantic Ridge, East Pacific Rise, Gal�apagos Spreading Center, and Gulf of Aden, Geochem. Geophys. Geosyst., 14, 223–242, doi:10.1029/2012GC004415.

Kendrick, M. A., R. Arculus, P. Burnard, and M. Honda (2013a), Quantifying brine assimilation by submarine magmas: Examples from theGal�apagos Spreading Centre and Lau Basin, Geochim. Cosmochim. Acta, 123, 150–165, doi:10.1016/j.gca.2013.09.012.

Kendrick, M. A., M. Honda, T. Pettke, M. Scambelluri, D. Phillips, and A. Giuliani (2013b), Subduction zone fluxes of halogens and noblegases in seafloor and forearc serpentinites, Earth Planet. Sci. Lett., 365, 86–96, doi:10.1016/j.epsl.2013.01.006.



http://dx.doi.org/10.1029/2010GC003252


http://dx.doi.org/10.1093/petrology/egq075



http://dx.doi.org/10.1146/10.1146/annurev.earth.34.031405.125211

http://dx.doi.org/10.1146/10.1146/annurev.earth.34.031405.125211

http://dx.doi.org/10.1016/j.pepi.2009.04.001



http://dx.doi.org/10.1002/ggge.20188



http://dx.doi.org/10.1029/2012GC004268



http://dx.doi.org/10.1093/petrology/egr003

http://dx.doi.org/10.1029/2002GC000435


http://dx.doi.org/10.1029/2012GC004415

http://dx.doi.org/10.1029/2012GC004415



Kendrick, M. A., M. G. Jackson, A. J. R. Kent, E. H. Hauri, P. J. Wallace, and J. Woodhead (2014), Contrasting behaviours of CO2, S, H2O andhalogens (F, Cl, Br, and I) in enriched-mantle melts from Pitcairn and Society seamounts, Chem. Geol., 370, 69–81, doi:10.1016/j.chemgeo.2014.01.019.

Kent, A. J., M. D. Norman, I. D. Hutcheon, and E. M. Stolper (1999a), Assimilation of seawater-derived components in an oceanic volcano:Evidence from matrix glasses and glass inclusions from Loihi seamount, Hawaii, Chem. Geol., 156, 299–319.

Kent, A. J., D. A. Clague, M. Honda, E. M. Stolper, I. D. Hutcheon, and M. D. Norman (1999b), Widespread assimilation of a seawater-derivedcomponent at Loihi Seamount, Hawaii, Geochim. Cosmochim. Acta, 63, 2749–2761.

Kingsley, R. H. (2002), The geochemistry of basalts from the Easter Microplate boundaries and the Salas y Gomez seamount chain: A com-prehensive study of mantle plume-spreading center interaction, dissertation, Univ. of Rhode Island, Narragansett, R. I.

Kingsley, R. H., and J.-G. Schilling (1998), Plume-ridge interaction in the Easter-Salas y Gomez seamount chain-Easter Microplate system: Pbisotope evidence, J. Geophys. Res., 103, 24,159–24,177.

Kogiso, T., Y. Tatsumi, G. Shimoda, and H. G. Barsczus (1997), High l (HIMU) ocean island basalts in southern Polynesia: New evidence forwhole mantle scale recycling of subducted oceanic crust, J. Geophys. Res., 102, 8085–8103.

Koleszar, A. M., A. E. Saal, E. H. Hauri, A. N. Nagle, Y. Liang, and M. D. Kurz (2009), The volatile contents of the Galapagos plume: Evidencefor H2O and F open system behavior in melt inclusion, Earth Planet. Sci. Lett., 287, 442–452, doi:10.1016/j.epsl.2009.08.029.

Krienitz, M.-S., C.-D. Garbe-Sch€onberg, R. L. Romer, A. Meixner, K. M. Haase, and N. A. Stroncik (2012), Lithium isotope variations in OceanIsland Basalts—Implications for the development of mantle heterogeneity, J. Petrol., 53(11), 2333–2347, doi:10.1093/petrology/egs052.

Kushiro, I. (1975), On the nature of silicate melt and its significance in magma genesis: Regularities in the shift of the liquidus boundariesinvolving olivine, pyroxene, and silica minerals, Am. J. Sci., 275, 411–431.

Langmuir, C., et al. (1997), Hydrothermal vents near a mantle hot spot: The Lucky Strike vent field at 37�N on the Mid-Atlantic Ridge, EarthPlanet. Sci. Lett., 148, 69–91.

Lassiter, J. C., E. H. Hauri, I. K. Nikogosian, and H. G. Barsczus (2002), Chlorine–potassium variations in melt inclusions from Raivavae andRapa, Austral Islands: Constraints on chlorine recycling in the mantle and evidence for brine-induced melting of oceanic crust, EarthPlanet. Sci. Lett., 202, 525–540.

Lassiter, J. C., J. Blichert-Toft, E. H. Hauri, and H. G. Barsczus (2003), Isotope and trace element variations in lavas from Raivavae and Rapa,Cook-Austral Islands: Constraints on the nature of HIMU- and EM-mantle and the origin of mid-plate volcanism in French Polynesia,Chem. Geol., 202, 115–138, doi:10.1016/j.chemgeo.2003.08.002.

le Roux, P. J., A. P. le Roex, and J.-G. Schilling (2002), Crystallization processes beneath the southern Mid-Atlantic Ridge (40–55�S), evidencefor high-pressure initiation of crystallization, Contrib. Mineral. Petrol., 142, 582–602.

MacDonald, G. A., and T. Katsura (1965), Eruption of Lassen Peak, Cascade Range, California, in 1915: Example of mixed magmas, Geol. Soc.Am. Bull., 76, 475–482.

Mackwell, S. J., and D. L. Kohlstedt (1990), Diffusion of hydrogen in olivine: Implications for water in the mantle, J. Geophys. Res., 95, 5079–5088.

Mallik, A., and R. Dasgupta (2012), Reaction between MORB-eclogite derived melts and fertile peridotite and generation of ocean islandbasalts, Earth Planet. Sci. Lett., 329–330, 97–108, doi:10.1016/j.epsl.2012.02.007.

McDonough, W. F., and S. S. Sun (1995), The composition of the Earth, Chem. Geol., 120, 223–253.Metrich, N., V. Zanon, L. Creon, A. Hildenbrand, M. Moreira, and F. O. Marques (2014), Is the ‘‘Azores Hotspot’’ a Wetspot? Insights from the

Geochemistry of Fluid and Melt Inclusions in Olivine of Pico Basalts, J. Petrol, 55(2), 377–393, doi:10.1093/petrology/egt071.Michael, P. (1995), Regionally distinctive sources of depleted MORB: Evidence from trace elements and H2O, Earth Planet. Sci. Lett., 131,

301–320.Michael, P. J., and W. C. Cornell (1998), Influence of spreading rate and magma supply on crystallization and assimilation beneath mid-

ocean ridges: Evidence from chlorine and major element chemistry of mid-ocean ridge basalts, J. Geophys. Res., 103, 18,325–18,356.Michael, P. J., and J.-G. Schilling (1989), Chlorine in mid-ocean ridge magmas: Evidence for assimilation of seawater-influenced compo-

nents, Geochim. Cosmochim. Acta, 53, 3131–3143.Nakamura, Y., and M. Tatsumoto (1988), Pb, Nd, and Sr isotopic evidence for a multicomponent source for rocks of Cook-Austral Islands

and heterogeneities of mantle plumes, Geochim. Cosmochim. Acta, 52, 2909–2924.Nishio, Y., S. Nakai, T. Kogiso, and H. Barsczus (2005), Lithium, strontium, and neodymium isotopic compositions of oceanic island basalts

in the Polynesian region: Constraints on a Polynesian HIMU origin, Geochem. J., 39, 91–103.O’Leary, J. A., G. A. Gaetani, and E. H. Hauri (2010), The effect of tetrahedral Al31 on the partitioning of water between clinopyroxene and

silicate melt, Earth Planet. Sci. Lett., 297(1–2), 111–120, doi:10.1016/j.epsl.2010.06.011.Pan, Y., and R. Batiza (1998), Major element chemistry of volcanic glasses from the Easter Seamount Chain: Constraints on melting condi-

tions in the plume channel, J. Geophys. Res., 103, 5287–5304.Parai, R., S. Mukhopadhyay, and J. C. Lassiter (2009), New constraints on the HIMU mantle from neon and helium isotopic compositions of

basalts from the Cook-Austral Islands, Earth Planet. Sci. Lett., 277(1–2), 253–261, doi:10.1016/j.epsl.2008.10.014.Paul, B., J. D. Woodhead, J. Hergt, L. Danyushevsky, T. Kunihiro, and E. Nakamura (2011), Melt inclusion Pb-isotope analysis by LA-MC-

ICPMS: Assessment of analytical performance and application to OIB genesis, Chem. Geol., 289, 210–223, doi:10.1016/j.chemgeo.2011.08.005.

Portnyagin, M., K. Hoernle, P. Plechov, N. Mironov, and S. Khubunaya (2007), Constraints on mantle melting and composition and nature ofslab components in volcanic arcs from volatiles (H2O, S, Cl, F) and trace elements in melt inclusions from the Kamchatka Arc, EarthPlanet. Sci. Lett., 255, 53–69, doi:10.1016/j.epsl.2006.12.005.

Rose-Koga, E. F., K. T. Koga, P. Schiano, M. Le Voyer, N. Shimizu, M. J. Whitehouse, and R. Clocchiatti (2012), Mantle source heterogeneityfor South Tyrrhenian magmas revealed by Pb isotopes and halogen contents of olivine-hosted melt inclusions, Chem. Geol., 334, 266–279, doi:10.1016/j.chemgeo.2012.10.033.

Rose-Koga, E. F., K. T. Koga, M. Hamada, T. H�elouis, M. J. Whitehouse, and N. Shimizu (2014), Volatile (F and Cl) concentrations in Iwateolivine-hosted melt inclusions indicating low-temperature subduction, Earth Planets Space, 66, 81.

Roy-Barman, M., and C. J. Allegre (1995), 187Os/186Os in oceanic island basalts: Tracing oceanic crust recycling in the mantle, Earth Planet.Sci. Lett., 129, 145–161.

Saal, A. E., S. R. Hart, N. Shimizu, E. H. Hauri, and G. D. Layne (1998), Pb isotopic variability in melt inclusions from Oceanic Island Basalts,Polynesia, Science, 282, 1481–1484, doi:10.1126/science.282.5393.1481.

Saal, A. E., E. H. Hauri, C. H. Langmuir, and M. R. Perfit (2002), Vapour undersaturation in primitive mid-ocean-ridge basalt and the volatilecontent of Earth’s upper mantle, Nature, 419, 451–455.






http://dx.doi.org/10.1093/petrology/egs052



http://dx.doi.org/10.1093/petrology/egt071







http://dx.doi.org/10.1126/science.282.5393.1481

Saal, A. E., S. R. Hart, N. Shimizu, E. H. Hauri, G. D. Layne, and J. M. Eiler (2005), Pb isotopic variability in melt inclusions from the EMI-EMII-HIMU mantle end-members and the role of the oceanic lithosphere, Earth Planet. Sci. Lett., 240, 605–620, doi:10.1016/j.epsl.2005.10.002.

Salters, V., and W., White (1998), Hf isotope constraints on mantle evolution, Chem. Geol., 145, 447–460.Salters, V. J. M., and A. Sachi-Kocher (2010), An ancient metasomatic source for the Walvis Ridge basalts, Chem. Geol., 273, 151–167, doi:

10.1016/j.chemgeo.2010.02.010.Salters, V. J. M., and A. Stracke (2004), Composition of the depleted mantle, Geochem. Geophys. Geosyst., 5, Q05004, doi:10.1029/

2003GC000597.Salters, V. J. M., S. Mallick, S. R. Hart, C. E. Langmuir, and A. Stracke (2011), Domains of depleted mantle: New evidence from hafnium and

neodymium isotopes, Geochem. Geophys. Geosyst., 12, Q08001, doi:10.1029/2011GC003617.Sarbas, B. (2008), The GEOROC database as part of a growing geoinformatics network, translated by S. R. Brady, A. K. Sinha, and L. C. Gun-

dersen, Geoinformatics. Data to Knowledge, Proceedings, U. S. Geol. Surv. Sci. Invest. Rep., pp. 42–43.Schiano, P., K. W. Burton, B. Dupr�e, J.-L. Birck, G. Guille, and C. J. Allegre (2001), Correlated Os-Pb-Nd-Sr isotopes in the Austral-Cook chain

basalts: The nature of mantle components in plume sources, Earth Planet. Sci. Lett., 186, 527–537.Schilling, J.-G., H. Sigurdsson, A. N. Davis, and R. N. Hey (1985), Easter microplate evolution, Nature, 317, 325–331.Shaw, A. M., M. D. Behn, S. E. Humphris, R. A. Sohn, and P. M. Gregg (2010), Deep pooling of low degree melts and volatile fluxes at the

85�E segment of the Gakkel Ridge: Evidence from olivine-hosted melt inclusions and glasses, Earth Planet. Sci. Lett., 289(3–4), 311–322,doi:10.1016/j.epsl.2009.11.018.

Simons, K., J. Dixon, J.-G. Schilling, R. Kingsley, and R. Poreda (2002), Volatiles in basaltic glasses from the Easter-Salas y Gomez SeamountChain and Easter Microplate: Implications for geochemical cycling of volatile elements, Geochem. Geophys. Geosyst., 3, 1–29, doi:10.1029/2001GC000173.

Smith, J. V. (1981), Halogen and phosphorus storage in the Earth, Nature, 289, 762–765.Smith, J. V., J. S. Delaney, R. L. Hervig, and J. B. Dawson (1981), Storage of F and Cl in the upper mantle: Geochemical implications, Lithos,

14, 133–147.Sobolev, A. V., and L. V. Danyushevsky (1994), Petrology and geochemistry of boninites from the north termination of the Tonga Trench:

Constraints on the generation conditions of primary high-carboninite magmas, J. Petrol., 35, 1183–1211.Spilliaert, N., N. M�etrich, and P. Allard (2006), S-Cl-F degassing pattern of water-rich alkali basalt: Modelling and relationship with eruption

styles on Mount Etna volcano, Earth Planet. Sci. Lett., 248, 772–786, doi:10.1016/j.epsl.2006.06.031.Staudigel, H., A. Zindler, S. Hart, and T. Leslie (1984), The isotope systematics of a juvenile intraplate volcano: Pb, Nd, and Sr isotope ratios

of basalts from Loihi Seamount, Hawaii, Earth Planet. Sci. Lett., 69, 13–29.Staudigel, H., T. Plank, B. White, and H. Schmincke (1996), Geochemical fluxes during seafloor alteration of the basaltic upper oceanic crust:

DSDP Sites 417 and 418: Subduction: Top to Bottom, Geophys. Monogr., 96, 19–38.Stracke, A., M. Bizimis, and V. J. M. Salters (2003), Recycling oceanic crust: Quantitative constraints, Geochem. Geophys. Geosyst., 4(3), 8003,

doi:10.1029/2001GC000223.Straub, S. M., and G. D. Layne (2003), The systematics of chlorine, fluorine, and water in Izu arc front volcanic rocks: Implications for volatile

recycling in subduction zones, Geochim. Cosmochim. Acta, 67(21), 4179–4203, doi:10.1016/S0016–7037(03)00307-7.Stroncik, N. A., and K. M. Haase (2004), Chlorine in oceanic intraplate basalts: Constraints on mantle sources and recycling processes, Geol-

ogy, 32(11), 945, doi:10.1130/G21027.1.Tatsumi, Y., K. Oguri, G. Shimoda, T. Kogiso, and H. G. Barsczus (2000), Contrasting behavior of noble-metal elements during magmatic dif-

ferentiation in basalts from the Cook Islands, Polynesia, Geology, 28, 131–134.van Keken, P. E., B. R. Hacker, E. M. Syracuse, and G. A. Abers (2011), Subduction factory. 4: Depth-dependent flux of H2O from subducting

slabs worldwide, J. Geophys. Res., 116, B01401, doi:10.1029/2010JB007922.Van Orman, J. A., T. L. Grove, and N. Shimizu (2001), Rare earth element diffusion in diopside: Influence of temperature, pressure, and ionic

radius, and an elastic model for diffusion in silicates, Contrib. Mineral. Petrol., 141, 687–703, doi:10.1007/s004100100269.Wallmann, K. (2001), The geological water cycle and the evolution of marine d18O values, Geochim. Cosmochim. Acta, 65, 2469–2485.Weaver, B. L. (1991), The origin of ocean island basalt end-member compositions: Trace element and isotopic constraints, Earth Planet. Sci.

Lett., 104, 381–397.White, W. M. (2010), Oceanic Island Basalts and mantle plumes: The geochemical perspective, Annu. Rev. Earth Planet. Sci., 38, 133–160, doi:

10.1146/annurev-earth-040809-152450.White, W. M., and A. W. Hofmann (1982), Sr and Nd isotope geochemistry of oceanic basalts and mantle evolution, Nature, 296, 821–825.Witham, F., J. Blundy, S. C. Kohn, P. Lesne, J. Dixon, S. V. Churakov, and R. Botcharnikov (2012), SolEx: A model for mixed COHSCl-volatile

solubilities and exsolved gas compositions in basalt, Comput. Geosci., 45, 87–97, doi:10.1016/j.cageo.2011.09.021.Woodhead, J. D. (1996), Extreme HIMU in an oceanic setting: The geochemistry of Mangaia Island (Polynesia), and temporal evolution of

the Cook-Austral hotspot, J. Volcanol. Geotherm. Res., 72, 1–19.Woodhead, J. D., and C. W. Devey (1993), Geochemistry of the Pitcairn seamounts. I: Source character and temporal trends, Earth Planet.

Sci. Lett., 116, 81–99.Workman, R. K., S. R. Hart, M. Jackson, M. Regelous, K. A. Farley, J. Blusztajn, M. Kurz, and H. Staudigel (2004), Recycled metasomatized litho-

sphere as the origin of the Enriched Mantle II (EM2) end-member: Evidence from the Samoan Volcanic Chain, Geochem. Geophys. Geo-syst., 5, Q04008, doi:10.1029/2003GC000623.

Workman, R. K., E. Hauri, S. R. Hart, J. Wang, and J. Blusztajn (2006), Volatile and trace elements in basaltic glasses from Samoa: Implicationsfor water distribution in the mantle, Earth Planet. Sci. Lett., 241, 932–951, doi:10.1016/j.epsl.2005.10.028.

Wu, J., and K. T. Koga (2013), Fluorine partitioning between hydrous minerals and aqueous fluid at 1GPa and 770–947�C: A new constrainton slab flux, Geochim. Cosmochim. Acta, 119, 77–92, doi:10.1016/j.gca.2013.05.025.

Yang, H.-J., F. A. Frey, and D. A. Clague (2003), Constraints on the source components of lavas forming the Hawaiian North Arch and Hono-lulu Volcanics, J. Petrol., 44, 603–627.

Yurimoto, H., T. Kogiso, K. Abe, H. G. Barsczus, A. Utsunomiya, and S. Maruyama (2004), Lead isotopic compositions in olivine-hosted meltinclusions from HIMU basalts and possible link to sulfide components, Phys. Earth Planet. Inter., 146, 231–242, doi:10.1016/j.pepi.2003.08.013.

Zhang, Y. (2008), Geochemical Kinetics, pp. 173–324, Princeton Univ. Press, Princeton, N. J.Zindler, A., and S. Hart (1986), Chemical geodynamics, Annu. Rev. Earth Planet. Sci., 14, 493–571.Zindler, A., E. Jagoutz, and S. Goldstein (1982), Nd, Sr and Pb isotopic systematics in a three-component mantle: A new perspective, Nature,

298, 519–523.





http://dx.doi.org/10.1029/2003GC000597

http://dx.doi.org/10.1029/2003GC000597

http://dx.doi.org/10.1029/2011GC003617


http://dx.doi.org/10.1029/2001GC000173


http://dx.doi.org/10.1029/2001GC000223

http://dx.doi.org/10.1016/S0016-7037(03)00307-7

http://dx.doi.org/10.1130/G21027.1

http://dx.doi.org/10.1029/2010JB007922

http://dx.doi.org/10.1007/s004100100269

http://dx.doi.org/10.1146/annurev-earth-40809-152450

http://dx.doi.org/10.1016/j.cageo.2011.09.021

http://dx.doi.org/10.1029/2003GC000623





Supporting Information for “Volatile cycling of H2O, CO2, F, and Cl in the HIMU mantle: A new window provided by melt inclusions from oceanic hotspot lavas at Mangaia, Cook Islands”

Rita A. Cabral (1), Matthew G. Jackson (2), Kenneth T. Koga (3), Estelle F. Rose-Koga (3), Erik H. Hauri (4), Martin J. Whitehouse (5), Allison A. Price (2), James M.D. Day (6), Nobumichi

Shimizu (7), Katherine A. Kelley (8)

(1) Boston University (2) University of California, Santa Barbara (3) Université Blaise Pascal (4) Carnegie Institution of Washington (5) Swedish Museum of Natural History (6) Scripps

Institution of Oceanography (7) Woods Hole Oceanographic Institution (8) University of Rhode Island

Geochemistry, Geophysics, Geosystems, 2014

Introduction

The supplement contains sample descriptions for the hand samples collected from Mangaia. We describe melt inclusion sample preparation and homogenization techniques. We also include the analytical methods for measurements of Pb isotopes and major, trace, and volatile elements in the melt inclusions as well as a summary of the methods for whole rock major and trace element compositions. Additionally, for the Mangaia melt inclusions, we include data tables for each suite of elements as well as olivine fractionation corrected compositions. The method for filtering the global MORB and OIB volatiles database is outlined.

“Supporting Information.pdf” Detailed sample descriptions and analytical methods used to measure Pb-isotopic ratios as well as major, trace, and volatile element compositions in Mangaia melt inclusions. Additionally, whole rock major and trace element measurement methods are discussed.

“Filtering.pdf” Outlines the filtering methods used for generating the global MORB and OIB volatiles database used in many of the figures.

“Table S1.xlsx” Measured major element compositions of melt inclusions hosted in Mangaia olivines.

“Table S2.xlsx” Major element compositions of olivines hosting Mangaia melt inclusions.

“Table S3.xlsx” Measured trace element abundances in Mangaia melt inclusions.

“Table S4.xlsx” Measured volatile element and P abundances of Mangaia melt inclusions.

“Table S5.xlsx” Coupled Pb-isotopic ratios and major, trace, and volatile element compositions of melt inclusions. Major, trace, and volatile element compositions are corrected to be in equilibrium with their host olivine.

“Table S6.xlsx” GPS coordinates of all hand samples.

“Table S7.xlsx” Measured lead isotope compositions of the secondary standard.

Supporting Information

1. Samples and methods

1.1 Sample locations and descriptions

Mangaia is the most southerly island located within the Cook Island chain (Fig. S1) and

is thought to have formed by long-lived hotspot volcanism that is anchored in the east by

Macdonald Seamount, an active submarine volcano [e.g., Stoffers et al., 1989; Chauvel et al.,

1997; Bonneville et al., 2002]. Mangaia resides on oceanic crust that is ~75 million years old

[Saal et al., 2005] and has K-Ar ages up to 19.3 ± 0.6 Ma (we note that our samples have not

been age dated) [Dalrymple et al., 1975; Turner and Jarrard, 1982]. The island also appears to

have been uplifted [Dickinson, 1998]. Most of the island is covered by dense, tropical vegetation

and heavily eroded terrain. The rim of the island consists of uplifted coral reefs (termed

‘makatea’) that transition to low lying swamp and farmland before reaching the core of the

island, which represents a topographic high. The majority of the volcanics are located at the

center of Mangaia in deeply-eroded stream valleys that incise the volcano [Wood and Hay,

1970]. The island has a basaltic (commonly ankaramitic), volcanic core that is cross-cut by dykes

[Wood and Hay, 1970]. Due to poor outcrop exposure on the island, the stages of volcanism and

time period over which the volcano was active are poorly constrained.

Olivine and pyroxene phenocrysts in Mangaia lavas frequently host melt inclusions, and

all melt inclusions examined in this study and previous studies are highly crystalline [Saal et al.,

1998, 2005; Yurimoto et al., 2004; Paul et al., 2011]. Saal et al. [1998] and Yurimoto et al.

[2004] provided detailed descriptions of the phases commonly present in Mangaia inclusions,

including: Ti-augite, Cr-spinel, ilmenite, sphene, phlogopite, carbonate, sulfide phases,

magnetite, apatite, amphibole (hornblende [Yurimoto et al., 2004] and kaersutite [Saal et al.,

1998]), and Si-rich glass.

The five volcanic samples examined in this study were collected during two separate

field expeditions to Mangaia, and are described here and in Hauri and Hart [1993]. MGA-B-25

and MGA-B-47 were collected in 1990, and whole rock geochemical data for these two samples

were published previously [Hauri and Hart, 1993, 1997]. Saal et al. [1998; 2005] first measured

Pb-isotopic compositions in olivine and clinopyroxene-hosted melt inclusions isolated from these

hand samples. MG1001, MG1002, and MG1006 were collected in 2010, and whole rock major

and trace element data are available elsewhere [Herzberg et al., 2014]. These five samples are

alkali basalts hosting abundant clinopyroxene and olivine phenocrysts, and were found close to

the center of the island in outcrops created by roads and drainage channels (Fig. S1, Table S6).

Owing to ~20 million years of weathering, some samples have developed weathering rinds on

their exteriors. We were careful to select the least altered material for study, and only inclusions

hosted in fresh olivines were chosen for analysis.

Sample MGA-B-25: This sample was obtained from the fresh core of a boulder in the interior of

the island, exposed in a road cut along a stream draining the north-east side of Mangaia. The

sample is olivine and clinopyroxene-phyric (20% olivine, 20% augite), with phenocrysts up to 2

cm in diameter in a fine-grained groundmass.

Sample MGA-B-47: This sample was obtained as a single cobble exposed in a road cut along a

stream north of the village of Tamarua, on the south side of the island. The sample is olivine and

clinopyroxene-phyric (50% olivine, 10% augite) with phenocrysts up to 2 cm in diameter in a

fine-grained groundmass.

Sample MG1001: Small outcrop exposed by a road cut near a stream on the south side of the

island. The outcrop was a stratified flow such that bands of variable phenocryst abundances and

sizes were encountered along a vertical cross-section of the flow. The hand sample is olivine and

pyroxene-phyric with fresh olivine phenocrysts commonly 1 cm in diameter and clinopyroxene

phenocrysts up to 2 cm in diameter. The hand sample is composed of 40% olivine, 10%

clinopyroxene, and 50% fine-grained groundmass.

Sample MG1002: This sample was found in an outcrop exposed in a stream channel on the north

side of the island. It is a cobble that was contained within the wall of the stream channel and

lodged in lithified soil at 5 m depth. Owing to the depth of burial, height, location on the island,

and lithified state of the soil, it is likely to be proximal to its source. Also, the sample could not

have been transported far by natural processes, as it was recovered close to (< 1 km) the center of

the island. The sample is olivine and clinopyroxene-phyric, with fresh olivine phenocrysts up to

1 cm in diameter and clinopyroxene phenocrysts up to 2 cm in diameter. The hand sample is

composed of 20% olivine, 30% clinopyroxene, and 50% fine-grained groundmass.

Sample MG1006: Boulder was found in road-cut in a valley at the end of a road near Ivirua

village. The sample may not be part of an outcrop, as it was lodged in soil or lithified sediment,

but buried deeply enough that possible transport by humans is highly unlikely. The sample

hosted large pyroxene phenocrysts up to 1 cm. The hand sample is composed of 20% olivine,

10% cpx, and 70% fine-grained groundmass.

1.2 Selection, preparation, and homogenization of the melt inclusions

The Mangaia hand samples were reduced to centimeter-size pieces using a rock saw or

hammer before being crushed in a jaw crusher. The crushed samples were individually sieved

and the freshest olivines hosting melt inclusions were hand-picked under a binocular microscope.

Melt inclusions that were compromised due to cracks running through the olivine grain and that

pass near or through the melt inclusion were avoided. The olivine with the largest inclusions

were mounted in CrystalbondTM and then doubly-polished (without exposing the inclusion) by

first using silicon carbide paper, followed by polishing with diamond suspension, and finishing

with a polish of 0.3 µm alumina paste. The samples were cleaned by sonication in milliQ water

and acetone to remove residual CrystalbondTM and other impurities.

The inclusions in each of the hand samples were crystalline, as are all reported melt

inclusions from Mangaia [Saal et al., 1998, 2005; Yurimoto et al., 2004; Paul et al., 2011]. Some

common features prior to melt inclusion homogenization include post-entrapment daughter

crystals, a rim of olivine that crystallized onto the inclusion walls, a vapor bubble, a sulfide

phase, and spinel crystals that existed prior to entrapment and that did not dissolve during

homogenization. Using a homogenization technique (see below), we were able to reverse the

extensive crystallization observed in Mangaia inclusions. Post-homogenization, the inclusions

are free of crystals, glassy, and light brown in color. This process allows us to obtain major,

trace, and volatile element compositions for the bulk melt inclusions, which was previously

impossible given their highly crystalline nature.

The homogenization process used in this study follows the technique outlined in Le

Voyer et al. [2008]. The melt inclusions were heated in a Vernadsky-type heating stage under a

microscope at the Laboratoire Magmas et Volcans at the Université Blaise Pascale in Clermont-

Ferrand, France, which allowed us to visually confirm homogenization and eliminate those that

burst during heating and those that crack and heal during the process, as inclusions are not

exposed prior to homogenization. Homogenization was performed at 1 atm, and the oxygen

fugacity was held between 10-10 and 10-9 atm with purified He. Homogenization times are short:

The olivines hosting the inclusions were raised incrementally from room temperature to the

melting point of the target inclusion over a period of less than 20 minutes. The temperature was

monitored by a thermocouple attached to the furnace and was calibrated by a piece of gold

placed on the surface of the olivine near the melt inclusion (~100 µm piece, Au melting point =

962 °C). Using the same Vernadsky heating stage that we employed, Chen et al. [2011]

demonstrated that this homogenization procedure does not result in any measurable water loss

from the inclusions over the (relatively short) times and temperatures that our homogenization

experiments were conducted. Similarly, the other volatiles are not degassed or lost by diffusion

during the homogenization process. Melt inclusions that burst or were breached during the

homogenization process were discarded.

Once the inclusions had been heated and quenched, the host olivines were mounted on

glass slides and carefully ground on silicon carbide paper to expose the homogenized inclusion

of interest. The inclusions were then polished with diamond suspension, and the final polishing

step employed 0.3 µm alumina paste. After polishing, the inclusions were placed in vials filled

with acetone and ultrasonicated for five minutes (to remove Crystal BondTM), followed by

ultrasonication in milliQ water for five minutes.

Each exposed inclusion, hosted in olivine, was mounted in indium and pressed in a vise

using a protective Al guard to prevent contamination from the surface of the vise jaws. The

mount was subjected to a final overnight press and then left in a vacuum furnace at 100°C

overnight. The mount hosting the exposed melt inclusions was polished once more with 0.3 µm

alumina paste and cleaned with MilliQ water. The sample mount was then gold coated prior to

volatile, trace, and Pb-isotopic analyses by ion probe. Following these analyses, the gold coat

was removed by polishing with 0.3 µm alumina paste and cleaning with MilliQ water before

being carbon coated for major element analyses by electron microprobe.

Any samples that showed evidence of bursting during the homogenization process are not

presented here.

1.3 Analytical techniques

1.3.1 Major element measurements in melt inclusions and host olivine

Major element compositions for the melt inclusions were measured at the Laboratoire

Magmas et Volcans, Clermont-Ferrand, France on a Cameca SX 100 electron microprobe

analyzer using standard procedures reported elsewhere [Le Voyer et al., 2008]. In brief, a

defocused electron beam with a 15 kV accelerating voltage and an 8 nA current was used for the

melt inclusion measurements. Phosphorus measurements made by ion probe (where available)

are reported in Table S1 for a subset of the inclusions.

Host olivine data were collected at Rutgers University using a 300 nA beam with a 20 kV

accelerating voltage and are reported in Herzberg et al. [2014]. In cases where data from Rutgers

University were not available for a particular sample, the olivine was measured at the

Laboratoire Magmas et Volcans, Clermont-Ferrand, France (using the same beam conditions

used for melt inclusion measurement, but with a focused beam). This was necessary for two

samples: MGA-B-25-1 and MGA-B-47-RAC16. Host olivine data not in Herzberg et al. [2014]

are presented in Table S2.

1.3.2 Major elements in the host whole rocks

Whole rock geochemical data for the hand samples from which olivine-hosted inclusions

were separated are reported in Herzberg et al. [2014], Hauri and Hart [1993], and Hauri and

Hart [1997].

1.3.3 Trace element measurements in melt inclusions

Trace element analyses were performed at the Secondary Ion Mass Spectrometry (SIMS)

Labs at Arizona State University (ASU) using a Cameca IMS 6f ion microprobe. Data on the

melt inclusions were collected over the course of two, one-week-long sessions (the first session

in October 2012 and the second in June 2013) that utilized identical machine configuration and

analysis protocols, discussed below (and following the methods outlined in Wanless et al. [in

press]). One melt inclusion, MG1001-5, gave a significantly different primitive mantle

normalized trace element pattern (or spidergram) from the rest of the suite. This anomalous

spidergram was re-analyzed during a third ASU session and also during an analytical campaign

using a laser ablation inductively coupled plasma mass spectrometer (LA-ICP-MS) at the

University of Rhode Island (both sessions were in January 2014).

During the three ASU sessions, we used a 16O- primary beam accelerated to ~12.5 keV at

~10 nA. A 9 kV secondary voltage was placed on the sample to accelerate the secondary ions.

Only secondary ions with 75 ± 20 eV excess kinetic energy were admitted to the mass

spectrometer. This “energy filtering” approach effectively eliminated molecular ions from the

mass spectrum except for dimers (such as BaO+). This approach has been used successfully for

several decades [Shimizu et al., 1978; Zinner and Crozaz, 1986; Wanless et al., in press]. Each

analysis began by cycling the magnet through the masses of interest six times prior to peak

calibration. Samples were then subjected to a 200 second pre-sputter before beginning

measurement, which consisted of up to 20 individual cycles (for a total analysis time of ~25

minutes). During each analysis cycle, intensities for the following isotopes were collected: 11B (2

s), 30Si (1 s), 47Ti (2 s), 88Sr (2 s), 89Y (2 s), 90Zr (2 s), 93Nb (2 s), 138Ba (2 s), 139La (3 s), 140Ce (3

s), 141Pr (3 s), 144Nd (3 s), 147Sm (3 s), 174Yb (5 s), 180Hf (5 s), 238U (5 s). The beam size (nA) and

duration of each measurement were adjusted (decreased and increased, respectively) to

accommodate the measurement of small diameter melt inclusions.

We used the basaltic reference glasses KL2-G, BCR-2G, BHVO-2G, and ALV519-4-1

[Melson et al., 2002; Jochum and Nohl, 2008 [accessed 3/2011]; Gale et al., 2013] to generate

the working curves for calculating the final concentrations of the elements in our samples.

ALV519-4-1 was also used as a secondary standard to monitor drift over the course of each

analytical session. Typical in-run precision on the samples (2σ standard error of the mean) is <

6% for all trace elements, except for U (typically < 9%). Uncertainties introduced by the working

curves are < 8% for all elements, except for Y and Nb (19%) and Yb (27%). The new trace

element data are reported in Table S3, together with analyses of 519-4-1 run as a secondary

standard.

The LA-ICP-MS session for inclusions MG1001-5 and MG1001-1 at the University of

Rhode Island was done using the New Wave UP213 Nd-YAG laser and Thermo X-Series II

quadrupole ICP-MS with techniques following those outlined by Kelley et al. [2003] and Lytle et

al. [2012]. Analyses were done using a 30 µm spot and 5 Hz repeat rate, and the data were

normalized to 43Ca as the internal standard. Calibration curves were linear (r2 > 0.99 for all

elements reported), and trace element abundances for reference glass 519-4-1 are accurate to

within 10% difference from concentrations reported by Gale et al., [2013]. These replicate

measurements are also included in Table S3, and the trace element patterns and abundances

exhibit excellent agreement with the ion probe measurements (< 10% difference, on average).

All ion probe analyses inadvertently made on grain boundaries (i.e., olivine-melt

inclusion or spinel-melt inclusion grain boundary) were discarded for both the volatiles and trace

elements (below).

1.3.4 Volatile elements in melt inclusions

Volatile element analyses were completed in one, four-day session in November, 2013

using the Cameca IMS 6f ion microprobe at the Department of Terrestrial Magnetism SIMS lab,

Carnegie Institution of Washington, Washington D.C. (Table S4). Methods have been detailed

elsewhere [Hauri et al., 2002], but are summarized briefly here. A focused Cs+ primary beam of

14 nA was used to sputter the inclusions, and an electron gun compensated for charge build-up.

The position for analysis was determined through ion imaging of OH, F, and S. An ion image of

12C was made prior to each analysis in order to avoid any C contribution due to surface

contaminants residing in vapor bubbles, cracks, or edges of the melt inclusions. The beam was

then rastered (75 µm raster for the pre-sputter) over the sample for 300 seconds to clean the

sample surface of contaminants before measurement. During each analysis, the following

isotopes were analyzed to determine the abundances of the respective volatile elements and P:

12C (10 s), 17OH (5 s), 19F (5 s), 30Si (5 s), 31P (5 s), 32S (5 s), and 35Cl (5 s). Each volatiles

measurement lasted approximately 10 minutes, and was comprised of five individual cycles

lasting 40 seconds each (after the pre-sputter). ALV519-4-1 was used as the secondary standard

to monitor for drift during the session. After correcting the data for drift and standard offset, the

2σ error of the mean is as follows: < 6% for H2O; < 5% for CO2, F, and Cl; < 3% for P and S. A

glass from the EMP, D52-5 [Simons et al., 2002], was analyzed during the analytical session as

an additional secondary standard.

1.3.5 Pb-isotope measurements by multi-collector SIMs

In situ Pb-isotopic measurements were carried out at the NORDSIM facility at the

Swedish Museum of Natural History’s Laboratory for Isotope Geology in Stockholm, Sweden

using the Cameca 1280 ion microprobe and methods described elsewhere [Whitehouse et al.,

2005; Rose-Koga et al., 2012]. All measurements were made during a continuous, four-day

analytical session in June of 2012.

Samples experienced a 180 second pre-sputter with a 25 × 25 µm raster to remove the Au

coating and clean the surface of the melt inclusion from contaminants prior to analysis. A 200

µm mass aperture was used in conjunction with a -13 kV 16O2- primary ion beam to create a

semi-elliptical crater. A 10 kV secondary ion beam was centered by scanning the transfer

deflectors in a 4000 µm field aperture. The beam was also maximized to a peak energy

distribution (45 eV window) by scanning the sample voltage. The magnetic field was fixed using

an NMR field sensor during the analytical session to ensure high precision. The entrance slit was

120 µm wide and the exit slit was 250 µm wide. Three secondary electron multipliers (EMs)

were used to simultaneously count ions (206Pb+, 207Pb+ and 208Pb+) in multicollection mode, and

the mass resolution (M/∆M) was 4845, sufficient to resolve Pb from molecular interferences in

glasses. Each EM was located to allow concurrent detection of 206Pb, 207Pb and 208Pb. Each

measurement was comprised of 40-120 cycles, with each cycle representing 20 seconds of

integration, so measurements typically lasted 15-45 minutes following the pre-sputter. A 60 ns

deadtime correction was applied, but results in a negligible correction. The background levels of

Pb on the EMs were typically < 0.02 cps (counts per second), which was insignificant at the level

of Pb signal measured (typically >200 cps on mass 208, except for MG1002-1 [145 cps]; see

Table S5). Isotopic ratios for each inclusion were determined by taking a global average of each

block average, where each block consisted of five cycles. Statistical outliers for each block were

identified as the values outside of 2σ bounds of an average. The standard error of the mean for

each measurement was determined by taking the standard deviation of the block averages and

dividing by the square root of the number of blocks.

Under these conditions, melt inclusions were analyzed together with frequent analyses of

the GOR132-G and KL2-G standards. The GOR132-G glass standard (20 ppm Pb, Jochum et al.,

2006) was used as the primary Pb-isotopic reference, which allows for the correction of

instrumental mass fractionation and detector efficiencies. The Pb-isotopic values for GOR132-G

that we use are 207Pb/206Pb = 0.81644 and 208Pb/206Pb = 2.0103 [Jochum et al., 2006]. The basalt

glass KL2-G (2.07 ppm Pb; Jochum et al., 2006) was used as a secondary standard after

measurement of GOR132-G and we recovered its accepted 207Pb/206Pb value (0.8214; Jochum et

al., 2006) within error (0.8195 ± 0.0055, 2σ standard deviation of the mean of 10 measurements,

see Fig. S2 and Table S7). We also recovered its accepted 208Pb/206Pb value (2.0239; Jochum et

al., 2006) within error (2.0219 ± 0.0083, 2σ standard deviation of the mean of 10 measurements,

see Fig. S2 and Table S7). The in-run relative standard errors on KL2-G (2σ) for 207Pb/206Pb

ranged from 0.26% to 0.64% and for 208Pb/206Pb ranged from 0.20% to 0.52%.

To evaluate whether pre-sputtering actually eliminated surface contamination, we

analyzed an olivine grain using the same method and machine settings that were used on

standards and glassy melt inclusions, including the 180 second pre-sputter. Olivine should

contain no lead, so any lead measured must be derived from surface contamination or the gold

coat. We found that even after the 180 second pre-sputter, there was a small residual Pb signal

(1-3 cps on mass 208) that lasted for approximately the first 20 measurement cycles (where 1

cycle = 20 seconds). However, following those first 20 cycles, the lead intensity decreases to < 1

cps on mass 208. This residual Pb may owe to surface contamination or is hosted in the gold coat

that is sputtered on the perimeter of the beam. We note that the signal intensity of the Pb signal

that we measured on the olivine the first 20 cycles of measurement (~1-3 cps, immediately

following the pre-sputter) is dwarfed by the Pb signal that we measure on the melt inclusions

(145 to 1031 cps). In summary, the 180 second pre-sputter did not eliminate all of the surface

contamination. Therefore, the first 20 cycles of measurement following the pre-sputter were

removed from all standard and melt inclusion analyses and only the truncated data are shown in

the figures and tables. This processing step did not cause large differences in the resulting Pb-

isotopic ratios. It was a precautionary step to ensure absolute confidence in the data.

Additionally, we excluded Pb-isotopic measurements for which the primary beam

inadvertently overlapped with phase boundaries (e.g., spinel-glass boundaries, olivine-glass

boundaries), as grain boundaries may host contaminant material not easily removed by polishing

and cleaning with MilliQ water.

Figure S1. Map of Mangaia and sample locations. Map of Mangaia with sample locations,

roads, and stream channels marked. The map of Mangaia was adapted with modification from

Woodhead [1996]. GPS coordinates of samples can be found in Table S6.

Figure S2. Pb-isotopic values for glass standard KL2-G measured over the course of the

four-day analytical session during which all melt inclusion measurements were made. New

analyses of KL2-G are shown by white circles. Their averages with 2σ in run error are shown as

dashed and dotted black lines, respectively. The Pb-isotopic composition reported by Jochum et

al. [2006] for KL2-G is shown by the solid gray line.

References Cited:

Bonneville, A., R., Le Suavé, L., Audin, V., Clouard, L., Dosso, P.Y., Gillot, P., Janney, K.,

Jordahl, and K., Maamaatuaiahutapu (2002), Arago Seamount: The missing hotspot found in the

Austral Islands, Geology, 30, 1023-1026.

Chauvel, C., W., McDonough, G., Guille, R., Maury, and R., Duncan (1997), Contrasting old

and young volcanism in Rurutu Island, Austral chain, Chem. Geol., 139, 125-143.

Chen, Y., A., Provost, P., Schiano, and N., Cluzel (2011), The rate of water loss from olivine-

hosted melt inclusions, Contrib. to Mineral. Petr., 162 (3), 625-636, doi: 10.1007/s00410-011-

0616-5.

Dalrymple, G.B., R.D., Jarrard, and D.A., Clague (1975), K-Ar ages of some volcanic rocks

from the Cook and Austral Islands, Geol. Soc. Am. Bull., 86, 1463-1467.

Dickinson, W. (1998), Geomorphology and geodynamics of the Cook-Austral island-seamount

chain in the South Pacific Ocean: Implications for hotspots and plumes, Int. Geol. Rev., 40,

1039-1075.

Gale, A., M., Laubier, S., Escrig, and C.H., Langmuir (2013), Constraints on melting processes

and plume-ridge interaction from comprehensive study of the FAMOUS and North Famous

segments, Mid-Atlantic Ridge, Earth Planet. Sci. Lett., 365, 209-220, doi:

10.1016/j.epsl.2013.01.022.

Hauri, E.H., and S.R., Hart (1993), Re-Os isotope systematics of HIMU and EMII oceanic island

basalts from the south Pacific Ocean, Earth Planet. Sci. Lett., 114, 353-371.

Hauri, E.H., and S.R., Hart (1997), Rhenium abundances and systematics in oceanic basalts,

Chem. Geol., 139, 185-205.

Hauri, E., J., Wang, J.E., Dixon, P.L., King, C., Mandeville, and S., Newman (2002), SIMS

analysis of volatiles in silicate glasses 1. Calibration, matrix effects and comparisons with FTIR,

Chem. Geol., 183, 99-114.

Herzberg, C., R.A., Cabral, M.G., Jackson, C., Vidito, J.M.D., Day, and E.H., Hauri (2014),

Phantom Archean crust in Mangaia hotspot lavas and the meaning of heterogeneous mantle,

Earth Planet. Sci. Lett, 396, 97-106.

Jochum, K.P., B., Stoll, K., Herwig, M., Willbold, A.W., Hofmann, M., Amini, S., Aarburg, W.,

Abouchami, E., Hellebrand, B., Mocek, I., Raczek, A., Stracke, O., Alard, C., Bouman, S.,

Becker, M., Dücking, H., Brätz, R., Klemd, D., de Bruin, D., Canil, D., Cornell, C.-J., de Hoog,

C., Dalpé, L., Danyushevsky, A., Eisenhauer, Y., Gao, J.E., Snow, N., Groschopf, D., Günther,

C., Latkoczy, M., Guillong, E.H., Hauri, H.E., Höfer, Y., Lahaye, K., Horz, D.E., Jacob, S.A.,

Kasemann, A.J.R., Kent, T., Ludwig, T., Zack, P.R.D., Mason, A., Meixner, M., Rosner, K.,

Misawa, B.P., Nash, J., Pfänder, W.R., Premo, W.D., Sun., M., Tiepolo, R., Vannucci, T.,

Vennemann, D., Wayne, and J.D., Woodhead (2006), MPI-DING reference glasses for in situ

microanalysis: New reference values for element concentrations and isotope ratios, Geochem.

Geophy. Geosy., 7 (2), doi: 10.1029/2005GC001060.

Jochum, K.P., and U., Nohl (2008), Reference materials in geochemistry and environmental

research and the GeoReM database, Chem. Geol., 253, 50-53, doi:

10.1016/j.chemgeo.2008.04.002.

Kelley, K.A., T., Plank, J., Ludden, and H., Staudigel (2003), Composition of altered oceanic

crust at ODP Sites 801 and 1149, Geochem. Geophy. Geosy., 4 (6), doi:

10.1029/2002GC000435.

Langmuir, C.H., J.F., Bender, A.E., Bence, and G.N. Hanson, (1977), Petrogenesis of basalts

from the FAMOUS area: Mid-Atlantic Ridge, Earth Planet. Sci. Lett., 36, 133-156.

Le Voyer, M., E.F., Rose-Koga, M., Laubier, and P., Schiano (2008), Petrogenesis of arc lavas

from the Rucu Pichincha and Pan de Azucar volcanoes (Ecuadorian arc): Major, trace element,

and boron isotope evidences from olivine-hosted melt inclusions, Geochem. Geophy. Geosy., 9

(12), doi: 10.1029/2008GC002173.

Lytle, M.L., K.A., Kelley, E.H., Hauri, J.B., Gill, D., Papia, and R.J., Arculus (2012), Tracing

mantle sources and Samoan influence in the northwestern Lau back-arc basin: Geochem.

Geophy. Geosy., 13 (10), doi: 10.1029/2012GC004233.

Melson, W.G., T., O'Hearn, and E., Jarosewich (2002), A data brief on the Smithsonian Abyssal

Volcanic Glass Data File, Geochem. Geophy. Geosy., 3 (4), 1-11, doi: 10.1029/2001GC000249.

Paul, B., J.D., Woodhead, J., Hergt, L., Danyushevsky, T., Kunihiro, and E., Nakamura (2011),

Melt inclusion Pb-isotope analysis by LA–MC–ICPMS: Assessment of analytical performance

and application to OIB genesis, Chem. Geol., 289, 210-223, doi:

10.1016/j.chemgeo.2011.08.005.

Rose-Koga, E.F., K.T., Koga, P., Schiano, M., Le Voyer, N., Shimizu, M.J., Whitehouse, and R.,

Clocchiatti (2012), Mantle source heterogeneity for South Tyrrhenian magmas revealed by Pb

isotopes and halogen contents of olivine-hosted melt inclusions, Chem. Geol., 334, 266-279, doi:

10.1016/j.chemgeo.2012.10.033.

Saal, A.E., S.R., Hart, N., Shimizu, E.H., Hauri, and G.D., Layne (1998), Pb Isotopic Variability

in Melt Inclusions from Oceanic Island Basalts, Polynesia, Science, 282, 1481-1484, doi:

10.1126/science.282.5393.1481.

Saal, A.E., S.R., Hart, N., Shimizu, E.H., Hauri, G.D., Layne, and J.M., Eiler (2005), Pb isotopic

variability in melt inclusions from the EMI–EMII–HIMU mantle end-members and the role of

the oceanic lithosphere, Earth Planet. Sci. Lett., 240, 605-620, doi: 10.1016/j.epsl.2005.10.002.

Shimizu, N., M.P., Semet, and C.J., Allègre (1978), Geochemical applications of quantitative

ion-microprobe analysis, Geochim. Cosmochim. Acta, 42, 1324-1334.

Simons, K., J., Dixon, J.-G., Schilling, R., Kingsley, and R., Poreda (2002), Volatiles in basaltic

glasses from the Easter-Salas y Gomez Seamount Chain and Easter Microplate: Implications for

geochemical cycling of volatile elements, Geochem. Geophy. Geosy., 3 (7), 1-29, doi:

10.1029/2001GC000173.

Stoffers, P., R., Botz, J.-L., Cheminée, C.W., Devey, V., Froger, G.P., Glasby, M., Hartmann, R.,

Hékinian, F., Kögler, D., Laschek, P., Larqué, W., Michaelis, R.K., Mühe, D., Puteanus, and

H.H. Richnow (1989), Geology of Macdonald Seamount region, Austral Islands: Recent hotspot

volcanism in the south Pacific, Marine Geophysical Researches, 11 (2), 101-112.

Turner, D.L., and R.D., Jarrard (1982), K-Ar dating of the Cook-Austral island chain: A test of

the hot-spot hypothesis, J. Volcanol. Geoth. Res., 12, 187-220.

Wanless, V.D., M.D., Behn, A.M., Shaw, and T. Plank, Variations in melting dynamics and

mantle compositions along the Eastern Volcanic Zone of the Gakkel Ridge: Insights from

olivine-hosted melt inclusions, Contrib. Mineral. Petr., doi: 10.1007/s00410-014-1005. In press.

Whitehouse, M.J., B.S., Kamber, C.M., Fedo, and A., Lepland (2005), Integrated Pb- and S-

isotope investigation of sulphide minerals from the early Archaean of southwest Greenland,

Chem. Geol., 222, 112-131, doi: 10.1016/j.chemgeo.2005.06.004.

Wood, B.L., and R.F., Hay (1970), Geology of the Cook Islands, New Zeal. J. of Geol. Geop.,

82, 4-99.

Woodhead, J.D. (1996), Extreme HIMU in an oceanic setting: The geochemistry of Mangaia

Island (Polynesia), and temporal evolution of the Cook—Austral hotspot: J. Volcanol. Geoth.

Res., 72, 1-19.

Yurimoto, H., T., Kogiso, K., Abe, H.G., Barsczus, A., Utsunomiya, and S., Maruyama (2004),

Lead isotopic compositions in olivine-hosted melt inclusions from HIMU basalts and possible

link to sulfide components, Phys. Earth Planet. In., 146, 231-242, doi:

10.1016/j.pepi.2003.08.013.

Zinner, E., and G., Crozaz (1986), A method for the quantitative measurement of rare earth

elements in the ion microprobe, Int. J. Mass Spectrom., 69, 17-38.

Database Filtering Guidelines

Location Name Reference Filter Hawaii Melt Inclusions

Volatile and major elements: Hauri, E. (2002), SIMS analysis of volatiles in silicate glasses, 2: isotopes and abundances in Hawaiian melt inclusions, Chem. Geol., 183, 115-141.

Excluded KK31-12-17, KOO-17A-2, -6, -7, -16 due to potential brine/halite contamination following Fig. 11a in Hauri [2002].

Hawaii North Arch

Volatile elements: Dixon, J., D.A., Clague, P., Wallace, and R., Poreda (1997), Volatiles in alkalic basalts form the North Arch Volcanic Field, Hawaii: extensive degassing of deep submarine-erupted alkalic series lavas, J. Petrol., 38 (7), 911-939. Major elements: Clague, D.A., R.T., Holcomb, J.M., Sinton, R.S., Detrick, and M.E., Torresan (1990), Pliocene and Pleistocene alkalic flood basalts on the seafloor north of the Hawaiian Islands, Earth Planet. Sci. Lett., 98, 175-191. Trace elements: Yang, H.-J., F.A., Frey, and D.A., Clague (2003), Constraints on the source components of lavas forming the Hawaiian North Arch and Honolulu Volcanics, J. Petrol., 44, 603-627. Isotopes: Frey, F.A., D., Clague, J.J., Mahoney, and J.M., Sinton (2000), Volcanism at the edge of the Hawaiian plume: Petrogenesis of submarine alkalic lavas from the North Arch volcanic field, J. Petrol., 41, 667-691.

No samples excluded.

Discovery and Shona

Volatile elements: Dixon, J.E., L., Leist, C., Langmuir, and J.-G., Schilling (2002), Recycled dehydrated lithosphere observed in plume-influenced mid-ocean-ridge basalt, Nature, 420, 385-389. Major elements: le Roux, P.J., A.P., le Roex, and J.-G., Schilling (2002), Crystallization processes beneath the southern Mid-Atlantic Ridge (40–55°S), evidence for high-pressure initiation of crystallization, Contrib. Mineral. Petr., 142, 582-602. Trace elements: Douglass, J., J.-G., Schilling, and R.H., Kingsley (1995), Influence of the Discovery and Shona mantle plumes on the southern Mid-Atlantic Ridge: Rare earth evidence, Geophys. Res. Lett., 22, 2893-2896. Isotopes: Douglass, J., J.-G., Schilling, and D., Fontignie (1999), Plume-ridge interactions of the Discovery and Shona mantle plumes with the southern Mid-Atlantic Ridge (40°-55°S), J. Geophys. Res., 104, 2941-2962.

Dixon et al. [2002] noted potential water assimilation for sample 19D-1 in the caption of Fig 1a. Therefore, this sample was excluded.

North Atlantic MORB

Volatile and trace elements (Asimow in prep never published): Dixon, J.E., L., Leist, C., Langmuir, and J.-G., Schilling (2002), Recycled dehydrated lithosphere observed in plume-influenced mid-ocean-ridge basalt, Nature, 420, 385-389. Major elements: Langmuir, C., S., Humphris, D., Fornari, C., Van Dover, K., Von Damm, M.K., Tivey, D., Colodner, J.-L., Charlou, D., Desonie,

Excluded samples D9-10, D46a, D44-1, RC136, D17-1, D17-5, D19-1A, D21-5, RC63, D22-5, D22-6 due to either having > 8% degassing of H2O (as calculated in Table S2 of Dixon et al. [2002]), or having been dredged at depths < 1000 m (also documented in Table S2 of Dixon et al. [2002]).

C., Wilson, Y., Fouquet, G., Klinkhammer, and H., Bougault (1997), Hydrothermal vents near a mantle hot spot: the Lucky Strike vent field at 37°N on the Mid-Atlantic Ridge, Earth Planet. Sc. Lett., 148, 69-91. Isotopes: Dosso, L., H., Bougault, C., Langmuir, C., Bollinger, O., Bonnier, and J., Etoubleau (1999), The age and distribution of mantle heterogeneity along the Mid-Atlantic Ridge (31–41°N), Earth Planet. Sc. Lett., 170, 269-286.

Samoa

Trace and volatile elements, and some Sr isotopes: Workman, R.K., E., Hauri, S.R., Hart, J., Wang, and J., Blusztajn (2006), Volatile and trace elements in basaltic glasses from Samoa: Implications for water distribution in the mantle, Earth Planet. Sc. Lett., 241, 932-951, doi: 10.1016/j.epsl.2005.10.028. Major elements and isotopes: Workman, R.K., S.R., Hart, M., Jackson, M., Regelous, K.A., Farley, J., Blusztajn, M., Kurz, and H., Staudigel (2004), Recycled metasomatized lithosphere as the origin of the Enriched Mantle II (EM2) end-member: Evidence from the Samoan Volcanic Chain, Geochem. Geophy. Geosy., 5 (4), doi: 10.1029/2003GC000623.

Following Workman et al. [2006], samples from dredges 63, 68, 70, and 73 were excluded because they were dredged at depths of ≤ 1000 m dredge depth.

Siqueiros

Saal, A.E., E.H., Hauri, C.H., Langmuir, and M.R., Perfit (2002), Vapour undersaturation in primitive mid-ocean-ridge basalt and the volatile content of Earth's upper mantle, Nature, 419, 451-455.

Excluded samples Siq6-2-2, Siq6-3, Siq6-8, and A-3 due to having Cl/K > 0.02.

Easter micro plate and seamount chain

Volatiles elements (H2O & CO2): Simons, K., J., Dixon, J.-G., Schilling, R., Kingsley, and R., Poreda (2002), Volatiles in basaltic glasses from the Easter-Salas y Gomez Seamount Chain and Easter Microplate: Implications for geochemical cycling of volatile elements, Geochem. Geophy. Geosy., 3, 7, 1-29, doi: 10.1029/2001GC000173. Major elements: Schilling, J.-G., H., Sigurdsson, A.N., Davis, and R.N., Hey (1985), Easter microplate evolution, Nature, 317, 325-331. Pan, Y., and R., Batiza (1998), Major element chemistry of volcanic glasses from the Easter Seamount Chain: Constraints on melting conditions in the plume channel, J. Geophys. Res., 103, 5287-5304. Trace elements: Kingsley, R.H. (2002), The geochemistry of basalts from the Easter Microplate boundaries and the Salas y Gomez seamount chain: A comprehensive study of mantle plume-spreading center interaction, University of Rhode Island, Narragansett, RI. Dissertation. Kelley, K.A., R., Kingsley, and J.-G., Schilling (2013), Composition of plume-influenced mid-ocean ridge lavas and glasses from the Mid-Atlantic Ridge, East Pacific Rise, Galápagos Spreading Center, and Gulf of Aden, Geochem. Geophy. Geosy., 14, 223-242, doi: doi:10.1029/2012GC004415. Isotopes: Hanan, B.B., and J.-G., Schilling (1989), Easter Microplate

Excluded samples EN113-4D-1, -5D-1C, -8D-1, -13D-1, 26D-1, -26D-10, -30D-1, -35D-1, -44D-5, GL07-52-5, 53-5, 60-1 and 58-2 due to brine assimilation following Simons et al. [2002] sections 4.4.2, 4.4.3, and 4.4.4. Samples that are indicated by Simons et al. [2002] as having suffered assimilation have Cl/Nb ratios > 54. Therefore, in addition to samples indicated as having experienced assimilation, samples in Simons et al. [2002] that have Cl/Nb > 54 have been removed. Excluded samples GL07-30-1, 32-4, 36-1, 36-3, 39-1, 42-1, 42-3, 43-1, 43-2, 46-2, 47-4, and 48-5 due to extensive degassing (following Simons et al. [2002] section 4.3.2).

Evolution: Pb Isotope Evidence, J. Geophys. Res., 94, 7432-7448. Fontignie, D., and J.-G., Schilling (1991), 87Sr/86Sr and REE variations along the Easter Microplate boundaries (south Pacific): Application of multivariate statistical analyses to ridge segmentation, Chem. Geol., 89, 209-241. Kingsley, R.H., and J.-G., Schilling (1998), Plume-ridge interaction in the Easter-Salas y Gomez seamount chain-Easter Microplate system: Pb isotope evidence, J. Geophys. Res., 103, 24159-24177. Kingsley, R.H. (2002), The geochemistry of basalts from the Easter Microplate boundaries and the Salas y Gomez seamount chain: A comprehensive study of mantle plume-spreading center interaction, University of Rhode Island, Narragansett, RI. Dissertation. Kelley, K.A., R., Kingsley, and J.-G., Schilling (2013), Composition of plume-influenced mid-ocean ridge lavas and glasses from the Mid-Atlantic Ridge, East Pacific Rise, Galápagos Spreading Center, and Gulf of Aden, Geochem. Geophy. Geosy., 14, 223-242, doi: doi:10.1029/2012GC004415.

Fernandina and Santiago

Koleszar, A.M., A.E., Saal, E.H., Hauri, A.N., Nagle, Y., Liang, and M.D., Kurz (2009), The volatile contents of the Galapagos plume; evidence for H2O and F open system behavior in melt inclusion, Earth Planet. Sc. Lett., 287, 442-452, doi: 10.1016/j.epsl.2009.08.029.

Excluded samples STG06-29-02B, -06, -11B, -13, -23A, -23B, -25A, -25B, -27, and -32 due to significant degassing, as noted by Koleszar et al. [2009] (see section 3.2 of their paper).

Excluded samples AHA-D25C-2-57, STG06-29-16A, -16B, -16C, -17, and -47 because of severe degassing noted by Koleszar et al. [2009] (see Fig. 3B in their paper).

Pitcairn and Societies

Major, trace, and volatile elements: Kendrick, M.A., M.G., Jackson, A.J.R., Kent, E.H., Hauri, P.J., Wallace, and J., Woodhead (2014), Contrasting behaviours of CO2, S, H2O and halogens (F, Cl, Br, and I) in enriched-mantle melts from Pitcairn and Society seamounts, Chem. Geol., 370, 69-81, doi: 10.1016/j.chemgeo.2014.01.019. Pitcairn Isotopes: Honda, M., and J.D., Woodhead (2005), A primordial solar-neon enriched component in the source of EM-I-type ocean island basalts from the Pitcairn Seamounts, Polynesia: Earth Planet. Sc. Lett., 236, 597-612, doi: 10.1016/j.epsl.2005.05.038. Woodhead, J.D., and C.W., Devey (1993), Geochemistry of the Pitcairn seamounts, I: source character and temporal trends, Earth Planet. Sc. Lett., 116, 81-99. Societies Isotopes: Devey, C.W., F., Albarede, J.-L., Cheminée,A., Michard, R., Mühe, and P., Stoffers (1990), Active submarine volcanism on the society hotspot swell (west Pacific): A geochemical study, J. Geophys. Res., 95, 5049-5066.

Excluded samples 45DS-1, 46DS-2, 46DS-7, and 33GTV2 due to shallow dredge depths of < 1000 m.

Popping Rock

Major, trace, and volatile elements: (average of literature values for trace elements*) Cartigny, P., F., Pineau, C., Aubaud, and M., Javoy (2008), Towards a consistent mantle carbon flux estimate: Insights from volatile systematics (H2O/Ce, δD, CO2/Nb) in the North Atlantic mantle (14° N and 34° N), Earth Planet. Sc. Lett., 265, 672-685, doi: 10.1016/j.epsl.2007.11.011. Isotopes: (using an average of three D43 measurements) Dosso, L., B.B., Hanan, H., Bougault, J.-G., Schilling, and J.-L., Joron (1991), Sr-Nd-Pb geochemical morphology between 10° and 17° N on the Mid-Atlantic Ridge: A new MORB isotope signature, Earth Planet. Sc. Lett., 106, 29-43. *And references therein (used to build Table 1 in Cartigny et al., 2008).


Loihi

Volatile, major, and trace elements: Dixon, J.E., and D.A., Clague (2001), Volatiles in basaltic glasses from Loihi Seamount, Hawaii: Evidence for a relatively dry plume component, J. Petrol., 42, 627-654. Isotopes: Staudigel, H., A., Zindler, S., Hart, and T., Leslie (1984), The isotope systematics of a juvenile intraplate volcano: Pb, Nd, and Sr isotope ratios of basalts from Loihi Seamount, Hawaii, Earth Planet. Sc. Lett., 69, 13-29.

Excluded samples KK18-6, KK27-14, and KK31-12 because of degassing (see Fig. 3 of Dixon and Clague [2001]).

Kilauea

Clague, D.A., J.G., Moore, J.E., Dixon, and W.B., Friesen (1995), Petrology of submarine lavas from Kilauea's Puna Ridge, Hawaii, J. Petrol., 36, 299-349.

Excluded samples 1684, 1717, D45a, D45b, and 1685 due to shallow dredge depths of < 1000 m.

Azores

Metrich, N., V., Zanon, L., Creon, A., Hildenbrand, M., Moreira, and F.O., Marques (2014), Is the “Azores Hotspot” a Wetspot? Insights from the Geochemistry of Fluid and Melt Inclusions in Olivine of Pico Basalts, J. Petrol, 55, 2, 377-393, doi: 10.1093/petrology/egt071.


Australs

Lassiter, J.C., E.H., Hauri, I.K., Nikogosian, and H.G., Barsczus (2002), Chlorine–potassium variations in melt inclusions from Raivavae and Rapa, Austral Islands: constraints on chlorine recycling in the mantle and evidence for brine-induced melting of oceanic crust, Earth Planet. Sc. Lett., 202, 525-540.

No samples excluded. However, samples that have experienced assimilation (i.e., Type 2 and Type 3 inclusions, as defined by Lassiter et al. [2002]) are noted in all figures.

Documents

Volatile Cycling of H2O, CO2, F, and Cl in the HIMU Mantle ... · 1986; Hart, 1988; Hofmann, 1997; White, 2010]. Although the cause of the chemical heterogeneity in the mantle has