ARTICLESPUBLISHED ONLINE: 23 JANUARY 2017 | DOI: 10.1038/NGEO2877

Evolution of carbonated melt to alkali basalt in theSouth China SeaGuo-Liang Zhang1,2,3*, Li-Hui Chen4*, Matthew G. Jackson3 and Albrecht W. Hofmann5

CO2 is considered to play a key role in the melting of the deep upper mantle, and carbonated silicate melts have been widelypredicted by partial melting experiments to exist at mantle depths of greater than 80 km. However, such melts have not beenshown to exist in nature. Thus, the relationship between CO2 and the origin of silicate melts is highly speculative. Here wepresent geochemical analyses of rocks sampled from the South China Sea, at the Integrated Ocean Discovery Program SiteU1431. We identify natural carbonated silicate melts, which are enriched in light rare earth elements and depleted in Nb andTa, and show that they were continuously transformed to alkali basalts that are less enriched in light rare earth elements andenriched in Nb and Ta. This shows that carbonated silicate melts can survive in the shallow mantle and penetrate through thehot asthenosphere. Carbonated silicate melts were converted to alkali basaltic melts through reactions with the lithosphericmantle, during which precipitation of apatite accounts for reduction of light rare earth elements and genesis of positive Nb–Taanomalies. We propose that an extremely thin lithosphere (less than 20 km in the South China Sea) facilitates extrusion ofthe carbonated silicate melts, whereas a thickened lithosphere tends to modify carbonated silicate melt to alkali basalt.

Most of the Earth’s carbon is stored in the deep interior1,2,and CO2 can strongly affect the melting behavioursof the deep upper mantle3. Alkali basalts are generally

SiO2-undersaturated and cannot be produced by melting of ‘dry’mantle2,4. Many alkali basalts carry strong indications of a role forCO2; for example, phenocrysts with CO2-rich melt inclusions5,6,mantle xenoliths metasomatized by carbonatitic melts7, and alkalibasalt associations with carbonatites8–10. It is well known that CO2can reduce the SiO2 content in the melt and facilitate low-degreemantle-derivedmelt that is enriched in alkalis3; thus, CO2 could playa fundamental role in the origin of alkali basalts. However, naturalcarbonatites, including those from the two known oceanic locations(Cape Verde and Canary islands9,11), generally have CaO that is toohigh andMgO that is too low to be primary carbonatemelts12. Thus,direct field observation onmantle-derived primary carbonate meltsis lacking, and the role of CO2 in the genesis of alkali basalts ishighly speculative.

Near-solidus partial melting experiments on carbonated mantlerocks result in dolomitic melts, which transform rapidly tocarbonated silicatemelts with increasingmelting degree3. A numberof experiments have been conducted to test if melting of aspecified mantle lithology (for example, carbonated peridotite13,14and carbonated eclogite15,16), or interactions between carbonatedmelt and the asthenospheric mantle17, can generate primary alkalibasaltic melts. However, the experimentally derived carbonatedsilicate melts have not been shown to exist in nature. It alsoremains unresolved on how carbonated silicate melts with negativeNb–Ta12,18 evolve to alkali basalts with positive Nb–Ta19–21. Herewe show a suite of volcanic clasts that vary continuously fromcarbonated silicate melts to alkali basalts in an oceanic setting(South China Sea, Fig. 1), which have implications for how naturalcarbonated melts are transformed to alkali basalts.

Compositions of carbonated silicate melt-alkali basaltThe South China Sea basin, which was active between 32–16Ma(Myr ago) (ref. 22), is amarginal spreading basin located to the southof the continent shelf of China (Fig. 1). A seamount chain (Fig. 1)was formed along the fossil ridge shortly after cessation of the SouthChina Sea spreading23. The International OceanDiscovery Program(IODP) Expedition 349 (Site U1431) drilled into the mid-oceanridge basalt (MORB) crust near the fossil ridge. Before reachingthe MORB crust, a∼300m-thick volcanic breccia layer (Fig. 1) wasrecovered, which was formed by multiple volcanic episodes (from12.8Ma to 7.4Ma)24 shortly after cessation of spreading. The brecciacontains abundant volcanic clasts that range in diameter frommillimetres to approximately ten centimetres, and the geochemistryof the clasts is examined in this study.

Globules (formed by exsolution from magma at reducedpressures as magma ascends) are most abundant in the clastsat the bottom of breccia layer, which decrease in abundanceupwards (Supplementary Information). For example, typicalglobules in sample 21R-1-W-(9/11) (Supplementary Information)consist of four phases: P2O5-CaO-Ti-F-rich glass, silicate glassthat is MgO-FeOt-rich and CaO-depleted, silicate glass that isK2O-Na2O-rich and CaO-depleted, and carbonate (calcium calciteand MgO-rich calcite). In situ analyses of major and trace elementsconfirm that the carbonates in the globules are of magmatic origin(Supplementary Information). The compositional complementarityof these phases indicates that they were derived from meltimmiscibility (Supplementary Information). Electron microprobeanalyses on the glass of plagioclase-hosted melt inclusions in thissample indicate a P–Ca–Ti–F-rich and low-SiO2 primary melt (seediscussion in Supplementary Information), which is consistent withthe high P2O5 and CaO concentrations observed in experimentallyderived carbonated silicate melt12,25,26.

1Key Laboratory of Marine Geology and Environment, Institute of Oceanology, Chinese Academy of Sciences, 266071 Qingdao, China. 2Laboratory forMarine Geology, Qingdao National Laboratory for Marine Science and Technology, 266000 Qingdao, China. 3Department of Earth Sciences, University ofCalifornia Santa Barbara, Santa Barbara, California 93106-9630, USA. 4State Key Laboratory for Mineral Deposits Research, School of Earth Sciences andEngineering, Nanjing University, 210023 Nanjing, China. 5Max-Planck-Institute für Chemie, Postfach, 55020 Mainz, Germany.*e-mail: zhangguoliang@qdio.ac.cn; chenlh@nju.edu.cn

NATURE GEOSCIENCE | VOL 10 | MARCH 2017 | www.nature.com/naturegeoscience 229

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO2877

600

700

800

900

Volcaniclastic breccia

Pelagic brown sandstone

Seafloor basalt

Met

res

belo

w s

eaflo

or (m

bsf)

7.4 Ma

8.3 Ma

9.75 Ma

11 Ma

11.6 Ma

11.9 Ma12.8 Ma

20

18

16

8

14

10

12

Latit

ude

(° N

)

Longitude (° E)110 112 116 118114 120

China continental shelf

HainanSouth China Sea

Fossil spreading ridge

Manila Trench

Subduction zone

SCS deep-water basin

U1431

South ChinaIndia

Pacific Ocean

Philippine Sea

SCS

Indian Ocean

HN

80° E 100° E 120° E 140° E

20° N

10° N

0°

10° S

Figure 1 | Map of the South China Sea, sampling location and stratigraphy of recovered cores. Ages marked on the stratigraphy chart are based onoccurrences of calcareous nannofossils in the interlayer sediment24. Green circles in the stratigraphy diagram indicate sampling positions. SCS,South China Sea; HN, Hainan.

20 30 40 50 60 70 800

10

20

30Late stage (<8.3 Ma)Early stage (>8.3 Ma)Carbonatedperidotite meltsCarbonatedeclogite melts

CaO

(wt%

)

0

2

4

6

8

P 2O5 (

wt%

)

SiO2 (wt%)20 30 40 50 60 70 80

SiO2 (wt%)20 30 40 50 60 70 80

SiO2 (wt%)

OIB

OIB

0

50

100

150

200

La (p

pm)

OIB

Figure 2 | Plots showing relationships of CaO, P2O5 and La versus SiO2 for Site U1431 volcanic clasts. The symbol of the early stage sample marked witha cross indicates sample 21R-1-W-(9/11), which has been analysed by electron microprobe and LA-ICP-MS (Supplementary Information). Carbonatedperidotite melts experimentally derived at a pressure of 3 GPa (refs 13,26), and carbonated eclogite melts derived at pressures from 2.9–9 GPa(refs 15,27,28) are plotted for comparison. Data of alkali OIBs (alkali oceanic island basalts: including the Canary, Cape Verde, Society, Samoa, Caroline andAustral-Cook islands) are from http://georoc.mpch-mainz.gwdg.de/georoc/Entry.html.

Results of whole-rock geochemistry are shown in SupplementaryTable 1. The samples form an alkaline series on the plot of totalalkaline versus SiO2 (Supplementary Fig. 1). They show highlyvariable major and trace element compositions and trace elementratios (Supplementary Figs 2–4). As SiO2 increases they show co-variations in concentrations ofmajor element (for example, decreas-ing CaO, P2O5 and TiO2, and increasing K2O and Na2O, see Fig. 2and Supplementary Fig. 2), trace elements (for example, decreasingLa, and increasing Nb, see Fig. 2 and Supplementary Fig. 3), and

trace element ratios (for example, decreasing La/Nb, La/Yb, Ce/Pb,Sm/Hf and Y/Ho, see Supplementary Fig. 4).

The clast samples are grouped into early (789.8–704.2mbsf(>8.3Ma)) and late (651.8–617mbsf (<8.3Ma)) stages. The earlystage clasts have relatively low SiO2 (31–48wt%), and high P2O5(mostly 2.6–6.8 wt%), CaO (6.4–25.6 wt%) andMgO (2.1–6.4 wt%).The major element compositions of many of the early stagesamples are in the range of the experimentally derived carbonatedeclogite melts (Fig. 2 and Supplementary Fig. 2). Most of the

230

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

NATURE GEOSCIENCE | VOL 10 | MARCH 2017 | www.nature.com/naturegeoscience

NATURE GEOSCIENCE DOI: 10.1038/NGEO2877 ARTICLES

1

10

100

Late stage (<8.3 Ma)Early stage (>8.3 Ma)21R-1-W-(9/11)Typical alkali OIB

RbNb K Ce Pr P Sm

HfTi Tb Y Er Yb

Th Ta La Pb Sr NdZr

Eu Gd Dy Ho Tm Lu

500

Ba

Conc

entr

atio

n/pr

imiti

ve m

antle

Figure 3 | Trace element patterns for Site U1431 volcanic clasts. Note thatthe early stage volcanic clasts (>8.3 Ma) are dominated by carbonatedmelts with negative Nb–Ta–Zr–Hf, and the late stage volcanic clasts(<8.3 Ma) are dominated by OIB-type patterns. Data of primitive mantlefor normalization are from ref. 35. The typical alkali OIB for reference is theaverage of alkali basalts from the Society, Samoa, Caroline andAustral-Cook islands. The data sources are as in Fig. 2.

early stage samples show strong enrichments in light rare earthelements (LREEs) and negative anomalies of Nb–Ta–Zr–Hf (Fig. 3),which are consistent with the experimentally derived carbonatedmelts12. The early stage samples also have positive Y anomalies(Y/Y∗ values, calculated as YN/(DyN∗HoN)

0.5, that range from1.05–1.96) (Fig. 3). The early stage sample 21R-1-W-(9/11) (seeSupplementary Information) has the highest CaO (25.6 wt%) andP2O5 (6.75wt%), the lowest SiO2 (31.0 wt%) (Fig. 2), and thestrongest enrichment in LREEs and negative anomalies of high fieldstrength elements (HFSEs) (for example, Hf/Hf∗ values, calculatedas HfN/(NdN×SmN)

0.5, down to 0.24). These samples are distinctfrom the carbonatites with high CaO (>48wt%) and very lowSiO2 (<4wt%) and MgO (<2wt%) from the Cape Verde andCanary islands11.

Comparatively, the late stage clasts have higher SiO2(47–63wt%),lower P2O5 (0.8–2.0 wt%),CaO (0.8–11wt%) andMgO(1.3–5wt%),less enrichment in LREEs (LaN/SmN of 1.5–3.1), and positiveanomalies of Nb–Ta–Zr–Hf (Fig. 3). Thus, the late stage clastsare similar to typical alkali ocean island basalt (OIB) (Fig. 3).

However, these alkali basalts have relatively low MgO amongglobal OIBs (Supplementary Fig. 2d). They also have negative Yanomalies (Y/Y∗ values of 0.60–0.97) (Fig. 3). The most siliceoussample (SiO2 of 63wt%) has the lowest CaO (0.8 wt%) andthe strongest positive anomalies of Nb–Ta–Zr–Hf and negativeanomalies of Sr, Eu and Ti. Despite the geochemical differencesbetween the early and late stage samples, they form a continuoustransition from carbonated silicate melts to alkali basalts (Fig. 2 andSupplementary Figs 2 and 3).

Mantle sourcesThese samples show limited variation in Sr–Nd–Pb–Hf isotopes,which are more enriched than the normal (N)-MORB-like samplesat the same drill site (Fig. 4 and Supplementary Fig. 5). Despitethe limited range in isotopic compositions, the early stage (morecarbonate rich) samples are obviously more geochemically enrichedthan the late stage (more carbonate poor) samples (Fig. 4). Similarly,the early stage samples also have higher La/Yb and La/Sm ratiosthan the late stage samples (Supplementary Fig. 6), which mightsuggest a deep origin (and smaller degree of melting) for thecarbonated melts that ‘sample’ the enriched mantle components3.The correlations of 87Sr/86Sr with trace element ratios (for example,La/Yb, La/Sm) (Supplementary Fig. 6) and Sr–Nd–Pb–Hf isotopeswith SiO2 content (Supplementary Fig. 7) indicate that a depletedmantle component has been involved in the origin of these volcanicclasts. The isotopic systematics suggests that the depleted mantlecomponent involved is the asthenospheric or lithospheric mantle asrepresented by the N-MORB-like samples at Site U1431 (Fig. 4 andSupplementary Fig. 5).

Carbonated peridotite13 and carbonated eclogite15 are widelyconsidered as the mantle sources of carbonated melts. However,the systematically low MgO (<7wt%) and relatively high TiO2(Supplementary Fig. 2e) of the carbonated melts in this studyexcludes them as primary partial melts of carbonated peridotitewith high MgO (>15wt%)13,26 (Supplementary Fig. 2d). Meltingof carbonated eclogite produces carbonated melts with low MgO(<10wt%)15,27,28 (Supplementary Fig. 2d), which are similar to thecarbonated melts in this study. Subduction of altered oceanic crustwith secondary carbonate minerals is considered as an effective wayto introduce carbonated eclogite to the mantle29. The relatively high206Pb/204Pb ratios (Fig. 4), and the strong negative Pb anomalies(Fig. 3), inmany of the early stage samples support the origin of these

0.702 0.704 0.706−5

0

5

10

15

87Sr/86Sr

Nd

ε

EPR MO

RBs

IR MORBs

SCS Seamounts

SCS N-MORBs at site U1431

17 18 19

37

38

39

EPR MORBs

IR MORBs

SCS Seamounts

206Pb/204Pb

208 P

b/20

4 Pb

NHRL

CHUR

Figure 4 | Plots of εNd versus 87Sr/86Sr, and 208Pb/204Pb versus 206Pb/204Pb for Site U1431 volcanic clasts. Arrows indicate depletion trend towards theN-MORB samples at Site U1431 (unpublished data, Supplementary Table 2). Legends are as in Fig. 2. Sources of data for comparison: the East Pacific Rise(EPR) and Indian Ridge (IR) MORBs are from http://www.earthchem.org/petdb; the South China Sea seamount basalts are from refs 23,44. NHRL, NorthHemisphere reference line, based on ref. 45.

NATURE GEOSCIENCE | VOL 10 | MARCH 2017 | www.nature.com/naturegeoscience

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

231

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO2877

10 20 30 40 50 600

2

4

6

8a

Y/Ho

P 2O5 (

wt%

)

0 1 2 3 4 5 60

1

2

3

4

5

6

La/Nb

Sm/H

f

OIB

OIB

0.020.04

0.06

0.080.02

0.040.060.08

Apatite precipitation

Apatite precipitation

OceanicCarbonatiteOceanic

carbonatite

PM

0 2 4 6

La/Nb

Met

res

belo

w s

eaflo

or (m

bsf)

15 30 45 60 75

Y/Ho

0 2 4 6 8800

760

720

680

640

600

P2O5 (wt%)

0.7035 0.704087Sr/86Sr

24 21 18 15

Lithospherethickness (km)

Lithospherethickness (km)

Lithospherethickness (km)

Lithospherethickness (km)

OIB

OIB

OIB

7.4 Ma

8.3 Ma

11 Ma

9.75 Ma

24 21 18 1524 21 18 15 24 21 18 15

c d e f

b

Figure 5 | E�ects of apatite fractionation and lithosphere thickness. a,b, E�ects of apatite fractionation. c–f, Down-hole chemical changes. Legends are asin Fig. 2. Data sources: OIBs as in Fig. 2; the oceanic carbonatites are from ref. 9; primitive mantle (PM) are from ref. 36. Sample 25R-1-W-(116/120) withthe highest La/Nb ratio is selected for calculation of apatite (Ca5(PO4)3(F,OH)) fractionation. The partition coe�cients between apatite/melts are fromref. 30. Red dashed line, variation of lithosphere thickness underlying Site U1431 with time. Lithosphere thickness, based on the equation given in ref. 46 ofH=7.49×

√(t) by assuming cessation of the South China Sea spreading at 16 Ma, where H is the lithosphere thickness (km) and t is the lithosphere

age (Ma).

melts from recycled oceanic crust, which has a high time-integratedU/Pb ratio due to significant Pb loss in the subduction zone. Thehigh La/Yb ratios in the early stage samples can be explained bythe effects of residual garnet in the source, which is consistent withthe low solidus and deep melting of carbonated eclogite. Primarycarbonated eclogite melts have also been shown by partial meltingexperiments to have strongly depletedHFSEs and enriched LREEs27.Thus, carbonated eclogite is most likely the source of carbonatedsilicate melts in this study, and the evolution from carbonated meltsto alkali basalts discussed here is limited to melts derived fromcarbonated eclogite.

E�ects of apatite precipitationThe decrease of REE concentrations (for example, La) and increaseof HFSE concentrations (for example, Nb) with increasing meltevolution (that is, higher SiO2, Supplementary Fig. 3) suggestfractionation of minerals rich in REEs and depleted in HFSEs.Titanite and pyrochlore are REE-rich minerals that have beenfound in carbonatite9,10. However, increasing HFSE concentrations

with increasing SiO2 in the samples of this study exclude the effectsof precipitation of these minerals, which are rich in HFSEs. Thelowering of P2O5 (with increasing SiO2) as carbonated silicate meltsevolve towards alkali basalts suggests precipitation of a phosphatemineral (Fig. 2). Apatite is a candidate phosphate, as it has high REEconcentrations and low HFSE concentrations10,30,31. Precipitationof apatite is also supported by the lowering of CaO (togetherwith lowering P2O5) (Fig. 2) with increasing magmatic evolution.Apatite can occur in lithospheric mantle xenoliths metasomatizedby carbonated melts32 and in alkali volcanic rocks associated withcarbonatites10. Although the partitioning of REEs between apatiteand melt can vary significantly as a function of the compositionsof the host melt33–35, the strong enrichment of REEs and depletionof HFSEs and Pb in apatite are widely accepted30. The lowering ofthe La/Nb, Ce/Pb and Sm/Hf ratios (Fig. 5a,b and SupplementaryFig. 8) with decreasing P2O5 in the melts of this study point tothe effect of apatite precipitation. For example, ∼8wt% apatiteprecipitation from sample 25R-1-W-(116/120) generates P2O5concentration and La/Nb, Ce/Pb and Sm/Hf ratios that are in

232

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

NATURE GEOSCIENCE | VOL 10 | MARCH 2017 | www.nature.com/naturegeoscience

NATURE GEOSCIENCE DOI: 10.1038/NGEO2877 ARTICLESthe range of alkali OIBs, including the alkali basalts in this study(Fig. 5a,b and Supplementary Fig. 8).

Ratios of Y/Ho are generally considered unfractionated relativeto the primitive mantle (Y/Ho of 28.9)36 during melt generationbecause global MORBs and OIBs have relatively constant Y/Horatios (28 ± 4)37. However, many of the early stage samples inthis study have much higher Y/Ho ratios (up to 59, Fig. 5a) thanthat of the primitive mantle. Mantle melting with the presenceof fluorine (F) can strongly elevate Y/Ho ratios in the melt33,34.The carbonated melts in this study are highly enriched in F (forexample, up to 4wt% in plagioclase-hosted melt inclusion glass,Supplementary Information). Thus, the high Y/Ho ratios may beexplained by high F abundances during melting. Nonetheless, ascarbonated silicate melts evolve, the Y/Ho ratios decrease to therange observed in alkali OIBs (Fig. 5a and Supplementary Fig. 4e)through precipitation of apatite, because the partition coefficientbetween F-rich apatite and silicate melt is much higher for Y thanHo (for example, DY/DHo of 2.5; ref. 30); in fact, Y/Ho ratios are ashigh as 80 in natural F-rich apatite35. Similar to La/Nb, Ce/Pb andSm/Hf, the Y/Ho ratio of sample 25R-1-W-(116/120) is also loweredto the range of alkali OIBs after fractionation of apatite (Fig. 5a).

Melt evolution in the lithospheric mantleIn Sr–Nd–Pb–Hf isotopic space, the evolution of carbonated meltstowards alkali basalts shows the addition of a depleted component(Supplementary Figs 6 and 7). On the basis of a comparison withthe MORB-like samples at Site U1431 (Fig. 4), it is predictedthat a depleted component, the asthenospheric and/or lithosphericmantle, has played a fundamental role in melt evolution. Thecompositional variation over time, from the carbonated meltsdeeper in the drill core to the alkali basalts shallower in the core,most likely reflects a temporal evolution (Fig. 5c–f). Reaction ofcarbonated peridotite melts with the asthenospheric mantle is onemechanism thatmay explain the isotopic depletion from carbonatedmelts to alkali basalt13. However, following reactions with hotasthenosphere, the K2O and Na2O concentrations in carbonatedmelts are reduced3,17. This is opposite to the chemical variationsfrom the carbonated melts to alkali basalts in this study, whichshow higher K2O and Na2O concentrations than the carbonatedmelts (Supplementary Figs 2 and 3). Thus, the transition fromcarbonated melts to alkali basalts is unlikely to have occurred inthe asthenosphere.

Carbonated melts are also in disequilibrium with the cooledlithospheric mantle38,39, in which orthopyroxene (Opx) canbe dissolved into the melt with concomitant precipitation ofclinopyroxene (Cpx) and olivine (Ol)38. In this way, harzburgite andlherzolite are metasomatically converted to wehrlite. This processoccurs as the melt cools, which permits fractionation of apatite.Precipitation of apatite (Ca5(PO4)3(F,OH)) cannot account for thefull range of CaO in the samples in this study, which indicatesprecipitation of other phases via the following reactions:

2Mg2Si2O6(Opx)+CaMg(CO3)2(melt)

=CaMgSi2O6(Cpx)+2Mg2SiO4(Ol)+2CO2(melt) (1)

Mg2Si2O6(Opx)+Melt1=Mg2SiO4(Ol)+Melt2 (2)

Through reaction (1), harzburgite/lherzolite are converted to wehr-lite as Opx is dissolved and Cpx and Ol precipitate from the melt38.Thus, CaO can be lowered in the melt through precipitation ofCpx. Additionally, through reactions with the carbonated melt (thereaction (2)), Opx dissolution and concomitant precipitation of Olincreases the SiO2 in the silica-undersaturated melts reported inthis study. Increased SiO2 in carbonated melts has been supportedby experiment that shows reactions of carbonatitic melts with Opxduring kimberlite generation39. Thus, the increasing SiO2 from

20 km

>>20 km

SCS

litho

sphe

ricm

antle

Norm

al oceanic lithospheric mantle

Carbonatedsilicate melt

Carbonatedsilicate melt

Carbonated silicatemelt to alkali basalts Alkali basalts

Crus

t Crust

Evolving carbonated silicate melts

Melt-rock reactions

Figure 6 | Schematic evolution of carbonated silicate melts in thelithosphere. The diagram shows the control of lithosphere thickness onreactive evolution of mantle-derived carbonated silicate melts. Anextremely thin lithosphere tends to allow extrusion of carbonated melts,whereas a thickened lithosphere tends to modify carbonated silicate meltsto alkali basalts.

carbonated melts to alkali basalts in this study can be explainedby melt-rock reaction, except for the most siliceous sample (withSiO2 of 63%) which could have experienced extensive magmadifferentiation. Through such reactions, dissolution of Opx fromthe country rock introduced geochemically depleted componentsinto the melts during melt evolution (Supplementary Figs 6and 7), which can help explain the presence of geochemicallydepleted Sr–Nd–Pb–Hf isotopic compositions in the alkali basalts.Precipitation of Cpx and Ol from carbonated melts through suchreactions might also explain the decreasing Ni, Cr and Co with meltevolution (Supplementary Fig. 3).

E�ects of lithosphere thicknessCarbonate melts have been considered to be stable only in themantle depth of more than 80 km because of the ‘carbonate ledge’(a boundary that prevents carbonated melt from reaching the shal-low depth). However, recent work shows that carbonated meltsfavour hot geotherms40. The extrusion of carbonated melts near theridge of the South China Sea indicates that deep carbonatitic meltscan penetrate through the hot asthenosphere. Although carbon-ated melts have been proposed to rise rapidly through the mantledue to their low viscosity41, the results of this study suggest thatcarbonated silicate melts can move through the lithosphere viareactive infiltration, during which they tend to be modified to alkalibasaltic melts. This implies that the extent to which the carbonatedsilicate melts react with the lithospheric mantle may be closelyrelated to the thickness of the overlying lithosphere (Fig. 5c–f). Thetemporal transition in compositions of, for example, P2O5, La/Nb,Y/Ho and 87Sr/86Sr, from carbonated melts to alkali basalts havelikely been controlled by a thickened mantle lithosphere duringvolcanic activity over∼5Myr (Fig. 5c–f). Because these carbonatedmelts were formed near the South China Sea fossil ridge shortly(∼4Myr) after the spreading cessation, a sufficiently thin litho-sphere (∼20 km) is inferred such that chemical reactions betweencarbonatedmelts and the overlying lithosphere were limited (Fig. 6).Thick lithosphere would transform all carbonated silicate melts toalkali basaltic melts, which could be the case for the seamount alkali

NATURE GEOSCIENCE | VOL 10 | MARCH 2017 | www.nature.com/naturegeoscience

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

233

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO2877

basalts, formed on the thin lithosphere near the fossil ridge of theSouth China Sea, with isotopic compositions similar to the alkalibasalt samples in this study (Fig. 4). A thick oceanic lithosphere isexpected to lower the degree of melting at a hotspot42. However,in contrast to the SiO2-dominated hotspot volcanism, the low-degree carbonated melts in this study were generated under thinlithosphere. It is difficult at this stage to assess the significance ofthe melt/lithosphere reactions in the origin of global alkali basalts.However, this model potentially explains the small-scale alkali vol-canism that is unlikely related tomantle plume activities, such as thehighly vesicular (CO2-rich) young (4–8Ma) alkali basalts at the petitspot volcanic province43 located on the thick, 135Ma northwesternPacific Plate.

MethodsMethods, including statements of data availability and anyassociated accession codes and references, are available in theonline version of this paper.

Received 7 August 2016; accepted 9 December 2016;published online 23 January 2017

References1. Sleep, N. H. & Zahnle, K. Carbon dioxide cycling and implications for climate

on ancient Earth. J. Geophys. Res. 106, 1373–1399 (2001).2. Dasgupta, R., Hirschmann, M. M. & Stalker, K. Immiscible transition from

carbonate-rich to silicate-rich melts in the 3 GPa melting interval ofeclogite+CO2 and genesis of silica-undersaturated ocean island lavas. J. Petrol.47, 647–671 (2006).

3. Dasgupta, R., Hirschmann, M. M. & Smith, N. D. Water follows carbon: CO2

incites deep silicate melting and dehydration beneath mid-ocean ridges.Geology 35, 135–138 (2007).

4. Kogiso, T., Hirschmann, M. M. & Frost, D. J. High-pressure partial melting ofgarnet pyroxenite: possible mafic lithologies in the source of ocean islandbasalts. Earth Planet. Sci. Lett. 216, 603–617 (2003).

5. Andersen, T. & Neumann, E. R. Fluid inclusions in mantle xenoliths. Lithos 55,301–320 (2001).

6. Hudgins, T. R., Mukasa, S. B., Simon, A. C., Moore, G. & Barifaijo, E. Meltinclusion evidence for CO2-rich melts beneath the western branch of the EastAfrican Rift: implications for long-term storage of volatiles in the deeplithospheric mantle. Contrib. Mineral. Petrol. 169, 1–18 (2015).

7. Neumann, E.-R., Wulff-Pedersen, E., Pearson, N. J. & Spencer, E. A. Mantlexenoliths from Tenerife (Canary Islands): evidence for reactions betweenmantle peridotites and silicic carbonatitie melts inducing Ca metasomatism.J. Petrol. 43, 825–857 (2002).

8. Silva, L. C., Le Bas, M. J. & Robertson, A. H. F. An oceanic carbonatite volcanoon Santiago, Cape Verde Islands. Nature 294, 644–645 (1981).

9. Hoernle, K., Tilton, G., Le Bas, M. J. & Garbe-Schönberg, D. Geochemistry ofoceanic carbonatites compared with continental carbonatites: mantlerecycling of oceanic crustal carbonate. Contrib. Mineral. Petrol. 142,520–542 (2002).

10. Weidendorfer, D., Schmidt, M. W. & Mattsson, H. B. Fractional crystallizationof Si-undersaturated alkaline magmas leading to unmixing of carbonatites onBrava Island (Cape Verde) and a general model of carbonatite genesis inalkaline magma suites. Contrib. Mineral. Petrol. 171, 1–29 (2016).

11. Doucelance, R., Hammouda, T., Moreira, M. & Martins, J. C. Geochemicalconstraints on depth of origin of oceanic carbonatites: the Cape Verde case.Geochim. Cosmochim. Acta 74, 7261–7282 (2010).

12. Foley, S. F. et al . The composition of near-solidus melts of peridotite in thepresence of CO2 and H2O between 40 and 60 kbar. Lithos 112S,274–283 (2009).

13. Dasgupta, R., Hirschmann, M. M. & Smith, N. D. Partial melting experimentsof peridotite+ CO2 at 3 GPa and genesis of alkalic ocean island basalts.J. Petrol. 48, 2093–2124 (2007).

14. Pilet, S., Baker, M. B. & Stolper, E. M. Metasomatized lithosphere and theorigin of alkaline lavas. Science 16, 916–919 (2008).

15. Kiseeva, E. S. et al . An experimental study of carbonated eclogite at3.5–5.5 GPa—implications for silicate and carbonate metasomatism in thecratonic mantle. J. Petrol. 53, 727–759 (2012).

16. Herzberg, C. et al . Phantom Archean crust in Mangaia hotspot lavasand the meaning of heterogeneous mantle. Earth Planet. Sci. Lett. 396,97–106 (2014).

17. Mallik, A. & Dasgupta, R. Reactive infiltration of MORB-eclogite-derivedcarbonated silicate melt into fertile peridotite at 3 GPa and genesis of alkalimagmas. J. Petrol. 54, 2267–2300 (2013).

18. Dalou, C., Koga, K. T., Hammouda, T. & Poitrasson, F. Trace elementpartitioning between carbonatitic melts and mantle transition zone minerals:implications for the source of carbonatites. Geochim. Cosmochim. Acta 73,239–255 (2009).

19. Sun, S. S. & McDonough, W. F. Chemical and isotopic systematics of oceanicbasalts: implications for mantle composition and processes. Geol. Soc. Lond.Spec. Publ. 42, 313–345 (1989).

20. Willbold, M. & Stracke, A. Trace element composition of mantle end-members:implications for recycling of oceanic and upper and lower continental crust.Geochem. Geophys. Geosyst. 7, Q04004 (2006).

21. Kalfoun, F., Ionov, D. & Merlet, C. HFSE residence and Nb/Ta ratios inmetasomatised, rutile-bearing mantle peridotites. Earth Planet. Sci. Lett. 199,49–65 (2002).

22. Briais, A., Patriat, P. & Tapponnier, P. Updated interpretation of magneticanomalies and seafloor spreading stages in the South China Sea: implicationsfor the Tertiary tectonics of Southeast Asia. J. Geophys. Res. 98,6299–6328 (1993).

23. Yan, Q. S., Shi, X. F., Wang, K. S., Bu, W. R. & Xiao, L. Major element, traceelement, and Sr, Nd and Pb isotope studies of Cenozoic basalts from the SouthChina Sea. Sci. China D 51, 550–556 (2008).

24. Expedition 349 Scientists. Opening of the South China Sea and its Implicationsfor Southeast Asian Tectonics, Climates, and Deep Mantle Processes Since theLate Mesozoic International Ocean Discovery Program Expedition 349Preliminary Report (2014); http://dx.doi.org/10.14379/iodp.pr.349.2014

25. Baker, M. B. & Wyllie, P. J. High-pressure apatite solubility in carbonate-richliquids: implications for mantle metasomatism. Geochim. Cosmochim. Acta 56,3409–3422 (1992).

26. Hirose, K. Partial melt compositions of carbonated peridotite at 3 GPa and roleof CO2 in alkali-basalt magma generation. Geophys. Res. Lett. 24,2837–2840 (1997).

27. Hammouda, T., Moine, B. N., Devidal, J. L. & Vincent, C. Trace elementpartitioning during partial melting of carbonated eclogites. Phys. Earth Planet.Inter. 174, 60–69 (2009).

28. Gerbode, C. & Dasgupta, R. Carbonate-fluxed melting of MORB-likepyroxenite at 2.9 GPa and genesis of HIMU ocean island basalts. J. Petrol. 51,2067–2088 (2010).

29. Dasgupta, R., Hirschmann, M. M. &Withers, A. C. Deep global cycling ofcarbon constrained by the solidus of anhydrous, carbonated eclogite underupper mantle conditions. Earth Planet. Sci. Lett. 227, 73–85 (2004).

30. Chazot, G., Menzies, M. A. & Harte, B. Determination of partition coefficientsbetween apatite, clinopyroxene, amphibole, and melt in natural spinellherzolites from Yemen: implications for wet melting of the lithospheric mantle.Geochim. Cosmochim. Acta 60, 423–437 (1996).

31. Hammouda, T., Chantel, J. & Devidal, J.-L. Apatite solubility in carbonatiticliquids and trace element partitioning between apatite and carbonatite at highpressure. Geochim. Cosmochim. Acta 74, 7220–7235 (2010).

32. Hervig, R. L. & Smith, J. V. Dolomite-apatite inclusion in chrome-diopsidecrystal, Bellsbank kimberlite, South Africa. Am. Mineral. 66,346–349 (1981).

33. Bau, M. Controls on the fractionation of isovalent trace elements in magmaticand aqueous systems: evidence from Y/Ho, Zr/Hf, and lanthanide tetrad effect.Contrib. Mineral. Petrol. 123, 323–333 (1996).

34. Irber, W. The lanthanide tetrad effect and its correlation with K/Rb, Eu/Eu∗,Sr/Eu, Y/Ho, and Zr/Hf of evolving peraluminous granite suites. Geochim.Cosmochim. Acta 63, 489–508 (1999).

35. Liu, C. Q. & Zhang, H. The lanthanide tetrad effect in apatite from the AltayNo. 3 pegmatite, Xingjiang, China: an intrinsic feature of the pegmatitemagma. Chem. Geol. 214, 61–77 (2005).

36. McDonough, W. F. & Sun, S. S. The composition of the Earth. Chem. Geol. 120,223–253 (1995).

37. Arevalo, R. Jr & McDonough, W. F. Chemical variations and regional diversityobserved in MORB. Chem. Geol. 271, 70–85 (2010).

38. Dalton, J. A. & Wood, B. J. The compositions of primary carbonate melts andtheir evolution through wallrock reaction in the mantle. Earth Planet. Sci. Lett.119, 511–525 (1993).

39. Russell, J. K., Porritt, L. A., Lavallée, Y. & Dingwell, D. B. Kimberlite ascent byassimilation-fuelled buoyancy. Nature 481, 352–356 (2012).

40. Hammouda, T., Chantel, J., Manthilake, G., Guignard, J. & Crichton, W. Hotmantle geotherms stabilize calcic carbonatite magmas up to the surface.Geology 42, 911–914 (2014).

41. Kono, Y. et al . Ultralow viscosity of carbonate melts at high pressures. Nat.Commun. 5, 5091 (2014).

42. Ellam, R. M. Lithospheric thickness as a control on basalt geochemistry.Geology 20, 153–156 (1992).

234

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

NATURE GEOSCIENCE | VOL 10 | MARCH 2017 | www.nature.com/naturegeoscience

NATURE GEOSCIENCE DOI: 10.1038/NGEO2877 ARTICLES43. Hirano, N. et al . Volcanism in response to plate flexure. Science 313,

1426–1428 (2006).44. Tu, K. et al . Magmatism in the South China Basin 1. Isotopic and trace-element

evidence for an endogenous Dupal mantle component. Chem. Geol. 97,47–63 (1992).

45. Hart, S. R. A large-scale isotope anomaly in the Southern Hemisphere mantle.Nature 309, 753–757 (1984).

46. Yoshii, T., Kono, Y. & Ito, K. in Geophysics of the Pacific Ocean Basin and itsMargin (eds Sutton, G. H., Manghnani, M. H. & Moberly, R.) 423–430(Geophysical Monograph 19, 1976).

AcknowledgementsWe thank the dedicated effort of the ship’s crew and scientific staff of the DrillshipJOIDES Resolution. We thank X.-J. Wang for Sr–Nd–Hf isotope analyses, J.-X. Zhao andL.-F. Zhong for Pb isotope analyses, and A. Kylander-Clark for LA-ICP-MS analyses.G.-L.Z. and M.G.J. thank F. Spera for discussions on melt immiscibility. We thankC. Class for constructive suggestions. This work was financially supported by the

Strategic Priority Research Program of the Chinese Academy of Sciences(XDA11030103), the National Natural Science Foundation of China (41522602,41376065, 41372064), and AoShan Talent Program Supported by Qingdao NationalLaboratory for Marine Science and Technology (No. 2015ASTP).

Author contributionsG.-L.Z. designed this project and wrote the manuscript. M.G.J., L.-H.C. and A.W.H.contributed to the scientific discussion, including interpretation of the results andscientific implications.

Additional informationSupplementary information is available in the online version of the paper. Reprints andpermissions information is available online at www.nature.com/reprints.Correspondence and requests for materials should be addressed to G.-L.Z. or L.-H.C.

Competing financial interestsThe authors declare no competing financial interests.

NATURE GEOSCIENCE | VOL 10 | MARCH 2017 | www.nature.com/naturegeoscience

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

235

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO2877

MethodsWhole-rock major and minor elements.Whole-rock major and minorelements (Cr, Sr, V, Zn and Ba) were analysed on fused glass discs with anAxios sequential X-ray fluorescence spectrometer. Samples were fused at 1,050 ◦Cusing a lithium tetraborate flux (Li2B4O7) in a mixture consisting of 0.5 g ofsample and 5 g of lithium tetraborate. The loss on ignition (LOI) of samplepowders was determined at a temperature of 1,000 ◦C. The measurements weremonitored using standards GSR-3 and BHVO-2. The accuracy of the analyticalresults was controlled by measuring standard reference material BHVO-2(Supplementary Information).

Whole-rock trace elements. For whole-rock trace element analyses, 25mg ofsample powder was precisely weighed and transferred into Teflon beakers. Then1.5ml HF and 1.5ml HNO3 were added in turn and heated in closed Teflonbeakers at 50 ◦C for 24 h. Beakers were opened heated at 120 ◦C to evaporate thedissolution. Then 0.5ml HF and 2.5ml 1:1 HNO3 were added, and the beakerswere transferred into a digestion bomb. The bombs were heated in an oven at170 ◦C for 72 h. Then the beakers were opened and heated at 120 ◦C to evaporatethe dissolutions. The residues were added to 1:1 HNO3, then sealed and heated at120 ◦C for 30min. The solutions were diluted 1,000 times using 2% distilledsuper-pure HNO3 and analysed by ICP-MS (Thermal X Series II) using In, Rh andRe as inner standards. The standard solutions (American Lab Tech Company) werediluted to 1 µg l−1, 10 µg l−1, 50 µg l−1 and 100 µg l−1 to produce the calibrationcurve, with linear regression coefficients all above 0.9999. To evaluate the accuracy,BHVO-2 and GSR-3 were run as external standards (Supplementary Information).

Whole-rock Sr–Nd–Pb–Hf isotope. Sr, Nd, and Hf isotope analyses in this paperwere accomplished at the State Key Laboratory for Mineral Deposits Research inNanjing University. To remove weathering effects and any potential seawatercontamination, the sample powders (200mg for Sr–Nd and 100mg for Hf isotopeanalyses) were leached in hot 6N twice sub-boiled HCl for 30min. Samples werethen rinsed repeatedly (5–6 times) with milli-Q water, then ultrasonicated for30min to remove any residual traces of acid. For the pretreatment of Sr–Nd isotopeanalyses, after evaporation, all samples were dissolved for more than 36 h on ahotplate at approximately 130 ◦C with a mixture of HF-HNO3 acids in closedSavillex beakers. After evaporation to dryness, concentrated HNO3 was added in allsamples to volatilize the residual HF in an open environment, followed by additionsof concentrated HCl and drying (repeat twice). Finally, the samples were dissolvedin 4N HCl acid. Sr–Nd were separated from the rock matrix by chromatographicextraction using a cation exchange resin (200–400 mesh Bio-Rad 50WX8).

For Hf isotope analyses, after leaching, all samples were wetted with HClO4 andthen dissolved for a week on a hotplate at approximately 140 ◦C with HF acid inclosed Savillex beakers. After evaporation to dryness, 6N HCl acid was added to allsamples and dried (repeat twice). Finally, the samples were dissolved in 3N HClacid for 12 h at 80 ◦C and then were prepared for column chemistry. Hf wasseparated from the rock matrix by chromatographic extraction using an EichromLn-spec resin. Detailed chemical separation procedures are given by Yangand colleagues47.

The isotopic compositions of Sr were determined on a Finnigan MAT Triton TIthermal ionization mass spectrometer. The mass fractionation for Sr isotopic ratioswas based on 86Sr/88Sr= 0.1194. The measured value for the international standardNBS987 was 0.710251± 0.000014 (2σ , n=7) for 87Sr/86Sr. Nd and Hf isotopic datawere obtained using a Neptune Plus (Thermo Fisher Scientific) multicollectorinductively coupled plasma mass spectrometer (MC-ICP-MS). Mass biascorrections on Nd and Hf were made with the exponential law and using146Nd/144Nd= 0.7219 and 179Hf/177Hf= 0.7325. The international JNdi-1 standardgave 143Nd/144Nd= 0.512093± 0.000003 (2σ , n=8). The measured value for theinternational Hf standard JMC475 was 176Hf/177Hf= 0.282178± 0.000006(2σ , n=7).

For lead isotope analyses, 200mg of sample powder was leached in 2ml 4N HClat 50 ◦C for 20min ultrasonically to ensure sufficient contact with HCl, then rinsed

with milli-Q water three times and then with 2% HNO3. Samples were dissolved in2ml conc. HNO3 and 10ml conc. HF at 80 ◦C overnight with the cap not tightlyclosed. On the second day, the lid was closed tightly and the temperature increasedto 140 ◦C, before checking whether the samples were fully dissolved or not. If fullydissolved, samples were dried down at 80 ◦C, then 1ml conc. HNO3 was added todry down again with no lid, 10ml 1N HCl+0.25N HNO3 added and the cap lidsclosed tightly overnight at a temperature of 90 ◦C maintained with hot blocks. Acheck was made whether all fluorite had fully dissolved. If yes, lids were openedand the solution dried down at 80∼ 90 ◦C in a wall oven with hot blocks. When thesolution was almost dry, we added 1ml 7N HNO3 and dried down without lids,then added 1ml 7N HNO3 again. After adding 2ml 1N HNO3 to dissolve theresidue and heating at 80 ◦C with tight lids, the solution was ready for Pb columnchemistry the next day. The AG1-8 anion resin was used for Pb separation usingthe standard procedure. Pb isotopes were analysed using a Nu instrumentMC-ICP-MS at the University of Queensland. The blank for Pb of the procedurewas 0.13 ppb. Samples were ‘spiked’ with a Tl standard (203Tl-205Tl isotopes), whichwas used as an internal standard to correct for mass-dependent isotopicfractionation. The entire procedure was monitored using the standard SRM-981,and multiple analyses (n=10) of SRM-981 yielded 206Pb/204Pb of 16.9410(σ =0.0010), 207Pb/204Pb of 15.4944 (σ =0.0012) and 208Pb/204Pb of 36.7179(σ =0.0022). Analyses on BCR-2 (GeoReM preferred values of 18.76, 15.62 and38.73 for 206Pb/204Pb, 207Pb/204Pb and 208Pb/204Pb, respectively) resulted in 18.7573,15.6186 and 38.7258 for 206Pb/204Pb, 207Pb/204Pb and 208Pb/204Pb, respectively.

In situmajor element analyses. Quantitative analyses of in situmajor elements forglass phases and minerals were carried out at the State Key Laboratory for MineralDeposits Research, Nanjing University, using a JEOL JXA-8100M electronmicroprobe (EMP). The operating conditions were: accelerating voltage 15 kV andprobe current 2× 10−8 A. The diameter of the electron beam was 1 µm. Thecounting times at the peaks were 20 s for Si, Ti, Al, Fe, Mg, Ca and Na, 30 s for Mn,P and K, and 50 s for F. All data were corrected with standard ZAF correctionprocedures. Natural minerals were used as standards.

In situ trace element analyses. Laser-ablation inductively coupled plasma massspectrometry (LA-ICP-MS) analyses were conducted at the University ofCalifornia, Santa Barbara, following the methods described in Kylander-Clark andcolleagues48. Carbonate grains were ablated using a Photon Machines Excite193 nm excimer laser using a 40 µm spot, ablating at 10Hz for 30 s, following a 20 sbaseline, with an approximate fluence of 1 J cm−2. The aerosol+He carrier gas wasmixed with Ar, and sent to an Agilent 7700 quadrupole ICP-MS for elementalconcentration analysis. Data was standard- and drift-corrected using Iolite v2.5(ref. 49). NIST 612 was used as the primary reference material and 43Ca was used asthe internal standard (Supplementary Information).

Data availability. The data that support the findings of this study are availablefrom the corresponding author upon request. The following publicly available datasources were used in this study: Petrological Database (PetDB): http://www.earthchem.org/petdb. GEOROC Database: http://georoc.mpch-mainz.gwdg.de/georoc/Entry.html.

References47. Yang, Y. H., Zhang, H. F., Chu, Z. Y., Xie, L. W. &Wu, F. Y. Combined chemical

separation of Lu, Hf, Rb, Sr, Sm and Nd from a single rock digest and preciseand accurate isotope determinations of Lu-Hf, Rb-Sr and Sm-Nd isotopesystems using Multi-Collector ICP-MS and TIMS. Int. J. Mass Spectrom. 290,120–126 (2010).

48. Kylander-Clark, A. R., Hacker, B. R. & Cottle, J. M. Laser-ablation split-streamICP petrochronology. Chem. Geol. 345, 99–112 (2013).

49. Paton, C., Hellstrom, J., Paul, B., Woodhead, J. & Hergt, J. Iolite: freeware forthe visualisation and processing of mass spectrometric data. J. Anal. At.Spectrom. 26, 2508–2518 (2011).

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

NATURE GEOSCIENCE | www.nature.com/naturegeoscience