How and when plume zonation appeared during the 132 Myr evolution of the Tristan Hotspot

Abstract

Increasingly, spatial geochemical zonation, present as geographically distinct, subparallel trends, is observed along hotspot tracks, such as Hawaii and the Galapagos. The origin of this zonation is currently unclear. Recently zonation was found along the last ∼70 Myr of the Tristan-Gough hotspot track. Here we present new Sr-Nd-Pb-Hf isotope data from the older parts of this hotspot track (Walvis Ridge and Rio Grande Rise) and re-evaluate published data from the Etendeka and Parana flood basalts erupted at the initiation of the hotspot track. We show that only the enriched Gough, but not the less-enriched Tristan, component is present in the earlier (70-132 Ma) history of the hotspot. Here we present a model that can explain the temporal evolution and origin of plume zonation for both the Tristan-Gough and Hawaiian hotspots, two end member types of zoned plumes, through processes taking place in the plume sources at the base of the lower mantle.

ARTICLE

Received 10 Oct 2014 | Accepted 10 Jun 2015 | Published 27 Jul 2015

Increasingly, spatial geochemical zonation, present as geographically distinct, subparallel trends, is observed along hotspot tracks, such as Hawaii and the Galapagos. The origin of this zonation is currently unclear. Recently zonation was found along the last B70 Myr of the

Tristan-Gough hotspot track. Here we present new SrNdPbHf isotope data from the older parts of this hotspot track (Walvis Ridge and Rio Grande Rise) and re-evaluate published data from the Etendeka and Parana ood basalts erupted at the initiation of the hotspot track. We show that only the enriched Gough, but not the less-enriched Tristan, component is present in the earlier (70132 Ma) history of the hotspot. Here we present a model that can explain the temporal evolution and origin of plume zonation for both the Tristan-Gough and Hawaiian hotspots, two end member types of zoned plumes, through processes taking place in the plume sources at the base of the lower mantle.

DOI: 10.1038/ncomms8799 OPEN

How and when plume zonation appeared during the 132 Myr evolution of the Tristan Hotspot

Kaj Hoernle1,2, Joana Rohde1, Folkmar Hauff1, Dieter Garbe-Schnberg2, Stephan Homrighausen1, Reinhard Werner1 & Jason P. Morgan3

1 GEOMAR Helmholtz Centre for Ocean Research Kiel, Wischhofstrasse 1-3, 24148 Kiel, Germany. 2 CAU Kiel University, Institute of Geosciences, Ludewig-Meyn-Strasse 10, D-24118 Kiel, Germany. 3 Royal Holloway, University of London, Department of Earth Sciences, Egham Hill, Egham TW20 0EX, UK. Correspondence and requests for materials should be addressed to K.H. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 6:7799 | DOI: 10.1038/ncomms8799 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8799

Although it has been proposed that some plumes have had constant compositions through time, such as the Louisville1 and Kerguelen2,3, spatial geochemical zonation

has been found in a variety of hotspot tracks in the Pacic Ocean. These include the Galapagos47, Hawaiian8,9, Samoan10, Marquesan10,11 and Society12 hotspot tracks, which are believed to be located at the border of the large, lower mantle, low shear-wave velocity province (LLSVP) beneath the southern Pacic9,10,12,13. The longevity of plume zonation is variable for the different Pacic hotspots. It can be traced for 1520 Myr along the Galapagos hotspot track6,7, for B4 Myr along the

Society hotspot track12,14, for B5.5 Myr along the Marquesas11,15 and for up to 2 Myr along the Samoan Islands10,14,16. The coexistence of the geochemically distinct Loa and Kea trends of the Hawaiian mantle plume has been traced for the last5 Myr along the island chain9. Little evidence exists for the presence of the Loa component in the Hawaiian seamounts, located west of the Hawaiian Islands and only the Kea component has been recognized so far in the Emperor Seamounts and older accreted Hawaiian complexes in Kamchatka17. To address questions concerning the longevity of plume zonation and how it originates, we investigate the entire history of the Tristan-Gough hotspot here and compare it with what is known about the history of the Hawaiian hotspot.

The Tristan-Gough hotspot track (Fig. 1) represents the classic evolution of a hotspot18 with active volcanic islands at its young end and ood basalt provinces at its older end. It has been termed one of the seven hotspots most likely to be derived from the lowermost mantle from a primary plume19 with its base currently located at the margin of the African LLSVP20,21. The emplacement of a Tristan-Gough plume head at the base of the Gondwana lithosphere caused massive volcanism forming the Parana (eastern South America) and Etendeka (western Africa) ood basalts at B132 Ma22 and may have contributed to the breakup of Africa from South America and the

formation of the South Atlantic Ocean basin23. As the Atlantic opened,the ridge-centred plume tail formed the Walvis Ridge on the African Plate and the Rio Grande Rise (possibly underlain by a thinned continental block) on the South American Plate. As the South Atlantic mid-ocean ridge drifted westwards away from the plume tail B5060 Ma, the hotspot became intraplate forming the Guyot Province, consisting of separate volcanic tracks leading to the active volcanic islands of Tristan da Cunha and Gough24.

Recently published SrNdHfPb isotope data of samples from the Tristan-Gough hotspot track reveal that bilateral chemical zonation can be traced from the Tristan da Cunha and Gough island groups along the Tristan and Gough subtracks for the last B70 Myr to the southwestern (SW) end of the Walvis Ridge13 (Fig. 1). The enriched Gough domain exhibits higher 207Pb/204Pb for a given 206Pb/204Pb and extends to lower 143Nd/144Nd and

176Hf/177Hf ratios and generally higher 87Sr/86Sr ratios compared with the more depleted (less enriched) Tristan domain. The domains form distinct elds on the uranogenic Pb isotope diagram (Fig. 2a). Published geochemical data are, however, insufcient to establish chemical zonation beyond 70 Ma.

Here we present new SrNdHfPb double spike (DS) isotope data from additional samples from the Guyot Province (obtained during R/V Akademik Kurchatov, AII-93 Atlantis II, ANT-XXIII/5 Polarstern, VM29 Vema, 51 S.A. Agulhas and SO233 Sonne cruises) and Deep Sea Drilling Programme (DSDP) Sites 525A, 527 and 528 at the SW end of the Walvis Ridge to test whether this part of the hotspot track is indeed zoned. More importantly we present new data from dredge (VM29 Vema, CIRCE Argo, AII93 Atlantis II, CH19 Jean Charcot, WALDAA002 Jean Charcot, RC11 Robert D. Conrad and SO233 Sonne cruises) and DSDP Leg 72 drill samples from the older part of the hotspot track (central and northeast Walvis Ridge and Rio Grande Rise) to determine whether the zonation can be traced further back in time, that is, going northeast along the hotspot track. Finally, we

g

e

d

15S

e

R

i

g

i

s

d

a

l

v Etendeka CFBs

i

W

R

c

Paran CFBs

n

ti

k

132 Ma

108114 Ma

Site 525A

132 Ma

la

c

At

a

Rio Grande Rise

t

r

d-

k

a

n

20S

c

Mi

ri

s

t

t

r

a

T

g

h

o

u

Guyot Province

87107 Ma

G

25S

Site 527

Site 528

Site 516F

8087 Ma

5867 Ma

7172 Ma

30S

49 Ma

36 Ma

33 Ma

Walvis Ridge

Rio Grande Rise

Guyot Province (Gough track)

Tristan da Cunha Islands

Gough Island DSDP site 525A

DSDP sites 527 and 528

Guyot Province (Tristan track)

46 Ma

47 Ma

Tristan da Cunha Islands

Gough Island

27 Ma

37 Ma

27 Ma

40S

40W 35W 30W 25W 15W 10W 5W 5E 10E

0

35S

Figure 1 | Bathymetric map of the South Atlantic Ocean and its margins showing the Tristan-Gough hotspot track. The map shows the Tristan-Gough hotspot track including the B132 Myr (ref. 22) old Etendeka and Parana continental ood basalt provinces, the Rio Grand Rise and Walvis Ridge (both within the age range B60115 Ma), the Guyot Province at the southwestern end of the Walvis Ridge containing the Tristan da Cunha and Gough Island groups (B060 Ma)24. Ages for late-stage volcanism are not shown. Age range in italics (87107 Ma) is estimated using a spatial age progression equation24. Sample locations are denoted by symbols: large symbols this study and small symbols as reported in ref. 13. Source of base map is http://www.geomapapp.org

Web End =http://

http://www.geomapapp.org

Web End =www.geomapapp.org . All sites are from the Deep Sea Drilling Project. The scale bar in the lower right hand corner indicates a distance of B500 km.

2 NATURE COMMUNICATIONS | 6:7799 | DOI: 10.1038/ncomms8799 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8799 ARTICLE

15.68

15.44

15.64

Tristan track Tristan Islands 527 and 528 Walvis Ridge

Gough track

Rio Grande Rise

Gough Island DSDP 525A

15.60

80%

90%

207 Pb/204 Pb

15.56

Average ingrowth corr.

15.52

15.48

17.2 17.6 18.0 18.4 18.8 19.2 19.6

Atlantic N-MORB

Atlantic N-MORB

80%

90%

90%

39.6

39.2

80% 90%

208 Pb/204 Pb

38.8

38.4

38.0

80%

37.6

206Pb/204Pb

Figure 2 | The 470 Ma Tristan-Gough hotspot track lavas have Gough-type isotopic compositions. On the (a) 206Pb/204Pb versus 207Pb/204Pb (uranogenic Pb) isotope diagram, older hotspot lavas (470 Ma) fall solely within the Gough eld of ref. 13. Even though they are completely separated on the uranogenic Pb isotope diagram, the Tristan and Gough elds completely overlap on the (b) 206Pb/204Pb versus 208Pb/204Pb (thorogenic Pb) isotope diagram, indicating that the Tristan compositions cannot simply be explained by mixing of Gough compositions with Atlantic MORB. Modelling shows that addition of 70 to 490% Atlantic N-MORB to Gough can explain the shift in Tristan to N-MORB in (a), but would require the

Tristan eld to be shifted towards Atlantic N-MORB in (b), for example, extending beyond the 490% mixing line towards N-MORB as in (a). Pb concentrations used for the mixing calculations are 0.57 p.p.m. for the MORB end members (based on average of global N-MORB62) and 3.2 p.p.m. for the Gough end members (average of Gough samples with MgO41 wt.%)

(Supplementary Dataset 1). Changing the assumed concentrations for Gough and MORB end members will simply shift the mixing percentages, but mixing of Gough and Atlantic N-MORB will still require Tristan to be shifted towards N-MORB compared with Gough on the thorogenic Pb isotope diagram. To compare isotope data of samples ranging in age between B0115 Ma, we use the measured compositions, assuming that the parent-daughter ratios were not signicantly fractionated during melting or subsequent differentiation. The average ingrowth corr. and associated 1s variation is the average correction needed for radiogenic ingrowth for the plotted samples (Supplementary Dataset 1). Since the age correction moves the data sub parallel to the boundary between the Gough and Tristan elds in (a), radiogenic ingrowth does not cause overlap between Gough and Tristan samples and therefore does not affect the classication of the samples. Analytical errors of data (Supplementary Dataset 1) in these and all subsequent isotope diagrams are smaller than symbol size. Numbers in the legend refer to DSDP Sites (Fig. 1). Atlantic MORB from PetDB (http://www.earthchem.org/petdb

Web End =http://www.earthchem.org/petdb) and additional Tristan-Gough data from GEOROC (http://georoc.mpch-mainz.gwdg.de/georoc/

Web End =http://georoc.mpch-mainz.gwdg.de/georoc/).

evaluate published data from the Etendeka and Parana ood basalts to assess the composition of the plume during its earliest history. These results are used to establish a model for the entire temporal (last 132 Myr) geochemical evolution of the Tristan-Gough hotspot. Finally, we compare the evolution of the Tristan

and Hawaiian plumes and evaluate the origin of geochemical zonation in both plumes.

ResultsIsotopic composition of Tristan-Gough hotspot track lavas. The new SrNdHfPb isotope data are presented in (Table 1). We report details about the samples, the major and trace element data and isotope data with errors in Methods and Supplementary Dataset 1. The SrNdHfPb isotope data of the new samples from the Guyot Province and DSDP Sites 525A, 527 and 528 at the SW end of the Walvis Ridge conrm the spatial geochemical zonation of the r70 Ma hotspot track proposed by (ref.13;

Fig. 2a). The data from the older (470 Ma) hotspot track, central and northeast portion of the Walvis Ridge and the Rio Grande Rise24, plot exclusively within the Gough eld. They also completely cover the range of the r70 Ma Gough track, showing that there is no systematic temporal change in the Gough source through time. The Rio Grande Rise samples and some samples from the Walvis Ridge are isotopically as enriched as DSDP Site 525A: the enriched mantle one (EMI) end member in the South Atlantic25,26.

Sr and Nd isotopes of Etendeka and Parana ood basalts. Evaluating the composition of the mantle plume head linked to the 132 Ma Parana-Etendeka ood basalt event is difcult because of potential interaction of melts with the continental lithosphere (both crust and mantle), which overall has a geo-chemically enriched composition in comparison with oceanic lithosphere. Lavas and dykes associated with the 132 Ma volcanic event range from pricritic and tholeiitic basalts through highly silica-undersaturated nephelinites to highly evolved rhyolites through phonolites. Intrusive equivalents range from gabbros to granites to syenites. These magmatic rocks show an extremely large range in measured 87Sr/86Sr of B0.7030.743 (with one sample having 0.924) and 143Nd/144Nd of B0.51170.5129 (Figs 3 and 4a). On plots of MgO versus 87Sr/86Sr and 143Nd/

144Nd isotope ratios (Fig. 3), the most magnesium-rich samples have the lowest 87Sr/86Sr and highest 143Nd/144Nd isotope ratios and show the least variation. With decreasing MgO, the range in

87Sr/86Sr and 143Nd/144Nd increases systematically with 87Sr/86Sr extending to higher and 143Nd/144Nd to lower ratios. If only samples with MgO Z11 wt.% (Sr4600 p.p.m.; Supplementary

Fig. 1) are considered, for example, the range in 87Sr/86Sr(0.70440.7091) and 143Nd/144Nd (0.512400.51288) is considerably reduced. The much larger range present in the more evolved compositions most likely reects derivation through continental lithospheric melting or through assimilation of such melts by differentiated mantle melts. Extensive feldspar fractionation in many highly evolved melts reduces the Sr concentration to very low values, whereas Rb continues to be incompatible in the melts and thus its concentration increases, resulting in very high Rb/Sr ratios. Radiogenic ingrowth in the evolved rocks with very high Rb/Sr ratios contributes to the elevated 87Sr/86Sr ratios. Although the 143Nd/144Nd of the mac (MgO Z11 wt.%) ood basalts is similar to the oceanic Tristan-Gough track rocks (Fig. 4a), the 87Sr/86Sr in some mac samples extends to higher ratios, which can be explained by small amounts of assimilation of crust with very radiogenic 87Sr/86Sr (for example, some crustal rocks have extreme 87Sr/86Sr up to 1.18) and/or through melting of enriched portions of the lithospheric mantle with radiogenic Sr and Pb isotope ratios27.

Pb isotopes of Etendeka and Parana ood basalts. In contrast to

87Sr/86Sr and 143Nd/144Nd where both upper and lower crustal interaction generally cause an increase in 87Sr/86Sr and decrease

NATURE COMMUNICATIONS | 6:7799 | DOI: 10.1038/ncomms8799 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8799

Table 1 | SrNdHfPb double-spike isotope data for Tristan-Gough hotspot track samples.

Sample Age Lat. Long. 87Sr/86Sr 143Nd/

144Nd

206Pb/

204Pb

207Pb/

204Pb

208Pb/

204Pb

176Hf/

177Hf

Tristan Track

AK16956 27 36.45 7.81 0.512654 18.9762 15.6093 39.3651 0.282835

Gough track

SO233 DR31 66 32.86 2.50 0.705358 0.512483 18.2685 15.5852 38.9360

AII93118 47 32.97 0.02 0.512631 18.1120 15.5381 38.4782 0.282799

PS69/4201DR 211 47 32.81 2.55 0.512423 18.3429 15.5607 39.1694 0.282682

PS69/4231DR 254 46 34.92 0.55 0.706357 0.512476 18.2090 15.5770 38.9231 0.282704

PS69/4241DR 261 46.5 34.92 0.55 0.704158 0.512694 18.4471 15.5623 38.9972 0.282841

V2991 47 32.63 1.12 0.704664 0.512694 18.3477 15.5671 38.6352 0.282822

AG5165 8 40.17 8.55 0.705545 0.512591 18.4323 15.6010 38.9192 0.282783

AG5171 8 40.17 8.55 0.704999 0.512565 18.3014 15.5863 38.8299

AG5172 8 40.17 8.55 0.704982 0.512551 18.3086 15.5892 38.8399 0.282763

Walvis Ridge Gough composition

CH19 DR32 114 19.37 9.33 0.705919 0.512464 17.9161 15.5279 38.3254 0.282724

CH19 DR41 113 19.85 9.02 0.705709 0.512342 17.5458 15.4796 38.1510 0.282668

CH19 DR42 112 19.85 9.02 0.705607 0.512387 17.5932 15.4933 38.1276 0.282683

WALDA002CH19DR403 113 19.85 9.02 0.705574 0.512397 17.5942 15.4930 38.1328 0.282675

SO233 DR842 114 20.35 7.64 0.706235 0.512308 17.4079 15.4574 38.0939

SO233 DR661 107 22.72 7.58 0.704928 0.512612 18.2968 15.5608 38.8369

SO233 DR711 101 22.67 5.13 0.704192 0.512472 18.7701 15.6325 38.9967

SO233 DR621 96 25.85 6.62 0.705231 0.512519 18.1463 15.5903 38.6825

SO233 DR341A 87 26.29 3.47 0.704735 0.512585 17.9160 15.5376 38.4723

V29111 85 26.15 5.58 0.704724 0.512503 17.9811 15.5529 38.4227 0.282774

CIR 139D2 85 26.45 5.89 0.704845 0.512496 17.8002 15.5259 38.3248 0.282725

CIR 139D3 85 26.45 5.89 0.704853 0.512509 17.8275 15.5310 38.3558 0.282723

Walvis III DR0435 79 27.26 4.39 0.704678 0.512649 18.0587 15.5635 38.4937 0.282841

DSDP Leg 74 525A 53 2W 91102 71 29.07 2.99 0.704909 0.512516 17.7478 15.5081 38.2496 0.282756

DSDP Leg 74 525A 56 2W 128134 71 29.07 2.99 0.704771 0.512490 17.6759 15.4972 38.1620 0.282748

DSDP Leg 74 525A 57 5W 141148 72 29.07 2.99 0.704841 0.512489 17.6544 15.4945 38.1603 0.282747

DSDP Leg 74 525A 63 2W 5978 72 29.07 2.99 0.705061 0.512431 17.7753 15.4910 38.2252 0.282702

Walvis Ridge Tristan composition

DSDP Leg 74 527 41 1W 4763 67 28.04 1.76 0.704144 0.512726 18.2048 15.5267 38.6715 0.282987

DSDP Leg 74 527 41 5W 95105 67 28.04 1.76 0.703608 0.512901 18.4908 15.5527 38.5336 0.283060

DSDP Leg 74 527 43 4W 5973 67 28.04 1.76 0.704547 0.512611 18.6188 15.5347 39.3297 0.282867

DSDP Leg 74 527 44 4W 5972 67 28.04 1.76 0.704427 0.512617 18.5224 15.5281 39.1353 0.282866

DSDP Leg 74 528 42 1W 2945 58 28.53 2.32 0.704744 0.512563 18.0458 15.5046 38.8273 0.282818

DSDP Leg 74 528 42 5W 3146 58 28.53 2.32 0.703996 0.512771 18.2698 15.5165 38.7332 0.283027

DSDP Leg 74 528 43 2W 8098 67 28.53 2.32 0.704092 0.512681 18.2553 15.5161 38.7148 0.282920

DSDP Leg 74 528 45 2W 109119 67 28.53 2.32 0.704285 0.512648 18.0917 15.5001 38.7408 0.282916

DSDP Leg 74 528 47 3W 6680 67 28.53 2.32 0.704456 0.512634 18.2089 15.5113 38.8174 0.282896

Rio Grande rise

RC1611RD 1 84 30.43 36.02 0.705578 0.512353 17.7807 15.4787 38.1734 0.282646

RC1611RD 2 84 30.43 36.02 0.705572 0.512344 17.7866 15.4813 38.1861 0.282647

RC1612RD 1 84 30.43 36.02 0.705687 0.512342 17.7822 15.4839 38.1913 0.282616

RC1612RD 3 84 30.43 36.02 0.705952 0.512273 17.7127 15.4736 38.1437 0.282588

DSDP Leg 72 516F 128 2W 6384 85 30.28 35.29 0.704988 0.512543 17.6396 15.4681 38.1425 0.282843

Full data, including errors are provided in Supplementary Dataset 1. Average 2s within run errors refer to the least signicant digits and are 5 for 87Sr/86Sr, 3 for 143Nd/144Nd, 5 for 176Hf/177Hf,

11, 11 and 33 for 206Pb/204Pb, 207Pb/204Pb and 208Pb/204Pb, respectively. External 2s errors based on standards measured along with the samples are 13 for 87Sr/86Sr, 7 for 143Nd/144Nd,

8 for 176Hf/177Hf, 24, 24 and 61 for 206Pb/204Pb, 207Pb/204Pb and 208Pb/204Pb, respectively. 87Sr/86Sr are the least radiogenic values obtained on dual analysis of 2 N HCl leached rock chips at 70 C for 1 h and 6 N HCl leached powders at 150 for 3 days. Ages in bold are from ref. 24. Ages in italics are either from dated samples from the same site or estimated based on a spatial linear age

equation24.

143Nd/144Nd isotope ratios27, Pb isotope ratios do not show good correlations with MgO or Pb concentration. The reason in part for the lack of correlation is that the lower crust generally has unradiogenic Pb isotope ratios, as indicated for example by the composition of granulites from the Namaqua-Natal Belt in South Africa28 and diorites derived from lower crustal melting during the Damara Orogen in Namibia29, whereas the upper crust extends to more radiogenic Pb isotope ratios, for example, Khan granodiorites and Kuiseb schists in Namibia30,31. On the

uranogenic Pb isotope diagram (Fig. 4b), where the Tristan and Gough domains form distinct elds, the Parana-Etendeka data form an array that completely covers the Gough eld but extends to more radiogenic (probably reecting upper crustal contamination) and less radiogenic (probably reecting lower crustal contamination) compositions. Some of this variation, for example, the lower trend towards more radiogenic Pb isotope ratios (Fig. 4b), could also result from contamination within the subcontinental lithospheric mantle, which can have more

4 NATURE COMMUNICATIONS | 6:7799 | DOI: 10.1038/ncomms8799 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8799 ARTICLE

0.7430

0.7380

Average ingrowth corr. Parana - Etendeka

0.7330

Gough track Tristan track Parana Etendeka

C

o

n

0.7280

ti

n

e

n

t

a

87 Sr/86 Sr

143 Nd/144 Nd

0.7230

li

l th

os

ph

ere

con

tam

inat

ion

0.7180

0.7130

0.7080

0.5128

0.5126

0.5124

min

ati

on

ta

sph

ere

con

lit

ho

ta

l

ti

n

e

n

o

n

C

0.5122

0.5120

0.5118

0.5116

0 2 4 6 8 10 12 14 16

MgO wt%

Figure 3 | Evolved Parana and Etendeka continental ood volcanism shows much greater isotopic variation than the oceanic Tristan-Gough hotspot track lavas. On plots of MgO vs. (a) 87Sr/86Sr and (b) 143Nd/

144Nd, the oceanic lavas have relatively constant composition and unradiogenic 87Sr/86Sr but radiogenic 143Nd/144Nd regardless of MgO content (degree of differentiation). The continental volcanism, however, shows a greater range in 87Sr/86Sr and 143Nd/144Nd extending to systematically more radiogenic 87Sr/86Sr and less radiogenic 143Nd/144Nd with increasing degree of differentiation (decreasing MgO). The increasing and substantially greater range in 87Sr/86Sr and 143Nd/144Nd for the continental, compared with oceanic, volcanic rocks primarily reects increasing amounts of assimilation during fractional crystallization of continental lithosphere by some magmas. Radiogenic ingrowth in some of the most evolved silica-saturated samples with very high Rb/Sr ratios also contributes to the extremely radiogenic 87Sr/86Sr. Average correction for radiogenic ingrowth for Parana and Etendeka is based on 498 (87Sr/86Sr) and 280 (143Nd/144Nd) analyses for which parent/daughter ratios are available. See Supplementary Dataset 5 and additional data from GEOROC (http://georoc.mpch-mainz.gwdg.de/georoc/

Web End =http://georoc.mpch-mainz.gwdg.de/georoc/).

radiogenic Sr and Pb but similar Nd isotope ratios27 to the Gough volcanic rocks. Surprisingly, no samples plot within the Tristan eld. If only ood basalts that do not show clear signs of contamination in 87Sr/86Sr (dened as 87Sr/86Sro0.7067 because0.7067 is the highest value in the oceanic hotspot track) are considered on the uranogenic Pb isotope diagram, they only show a slightly greater range than the Gough eld, which may reect slight Pb-isotopic contamination from the continental crust. The similarity in Pb-isotopic composition between the ood basalts with low 87Sr/86Sr and the oceanic Gough domain is consistent with the initial plume material having primarily had a Gough-type composition. In summary, Parana-Etendeka samples that show no obvious signs of being contaminated within the continental lithosphere plot within the Gough domain without any evidence for the presence of Tristan-type compositions in their source.

DiscussionNow we will evaluate possible models for explaining the origin of the Tristan and Gough domains and geochemical evolution of the Tristan-Gough hotspot. On the 87Sr/86Sr versus 143Nd/144Nd,

143Nd/144Nd versus 176Hf/177Hf, and the 206Pb/204Pb versus

207Pb/204Pb, 87Sr/86Sr, 143Nd/144Nd and 176Hf/177Hf isotope correlation diagrams (see ref.13, Figs 2a and 4 and Supplementary Fig. 2), the Tristan domain falls between the Gough and Atlantic MORB elds, suggesting interaction between a Gough-type plume and depleted upper mantle to generate the Tristan domain13. If this were the case, however, the Tristan domain would also have to be shifted towards the Atlantic MORB eld on the thorogenic Pb isotope diagram (Fig. 2b), since the denominator of the plotted ratios is the same element. Instead the Tristan domain almost completely overlaps the Gough domain and both are distinct from Atlantic MORB, in stark contrast to the nearly complete separation of the domains on the uranogenic Pb isotope diagram on which the Tristan domain overlaps the Atlantic MORB eld (Fig. 2a). Therefore, despite being generally more geochemically depleted in composition than the Gough domain, the Tristan domain cannot simply be derived through mixing of a Gough source and upper MORB-source mantle. Consequentially, we attribute both compositional types to the TristanGough plume, implying a deep source for both.

To evaluate the origin of plume zonation, a fundamental question concerns the distribution of both compositional types in the TristanGough plume. Is the plume chemically zoned with distinct Tristan and Gough domains13 or are these components uniformly distributed throughout the plume but sampled by different extents of melting, for example, under lithosphere with different thicknesses3234? In particular, was the Tristan composition also present in the early plume (B70132 Ma), but simply not sampled? Picritic and tholeiitic ood basalts, generated during the initial plume-head stage of volcanism, are typically associated with high temperatures and thus high degrees of melting (commonly 2030%)35 even in continental settings35,36. Therefore, the Etendeka and Parana ood basalts should have preferentially sampled the more depleted (or less enriched) Tristan component, if it were present. Even if the Tristan component was somehow more fertile (despite being more depleted isotopically), the lower-degree alkalic melts (alkali basalts, basanites and nephelinites) associated with the ood basalt volcanism should have sampled the more fertile component, if both components had been present. Since so far only Gough compositions have been identied in the ood basalts, despite the large range in rock types reecting a large range of melting conditions, the plume head is likely to have primarily had a Gough-type composition.

Although temperatures may have been lower during the formation of the early hotspot track from the plume stem than during the plume-head stage, the early-plume stem was located beneath the mid-Atlantic ridge, allowing upwelling to shallower depths and greater extents of pressure-release melting than possible beneath continental lithosphere. This situation should also have favoured melting of the more depleted Tristan component if it had been present in the plume. Certainly compared with the later, intraplate history of the hotspot (o60 Ma), degrees of melting should have been higher during melting of the early plume stem. Ratios of more to less incompatible trace elements (for example, La/Sm, which is relatively insensitive to differentiation except in highly evolved rocks) are inversely proportional to the melt fraction. Therefore, if lithospheric thickness controls the extent of melting, the La/Sm ratio should have been lower when the plume was ridge-centred and should have increased as the plume became progressively more intraplate after B60 Ma. As expected, the La/Sm ratio

NATURE COMMUNICATIONS | 6:7799 | DOI: 10.1038/ncomms8799 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8799

0.5130

0.703 0.708 0.713 0.718 0.723 0.728 0.733 0.738 0.743

Gough track Tristan track

Parana Etendeka

> 11 wt% MgO

> 11 wt% MgO

Average ingrowth correction ParanaEtendeka

Upper continental crust contamination

Lower continental crust contamination

0.5128

0.5126

143 Nd/144 Nd

0.5124

0.5122

0.5120

0.5118

0.5116

87Sr/86Sr

15.80

16.5 17.0 17.5 18.0 18.5 19.0 19.5 20.0

87Sr/86Sr < 0.7067

Gough track Tristan track

Parana Etendeka

15.75

Upper crustal / lithospheric mantle contam.

Average ingrowth correction Parana-Etendeka

15.70

15.65

87Sr/86Sr < 0.7067

207 Pb/204 Pb

15.60

15.55

15.50

Lower crustal contam.

15.45

15.40

206Pb/204Pb

Figure 4 | Larger range in isotopic composition of the continental ood volcanism most likely reects continental lithospheric contamination. The range in isotopic composition of continental ood volcanism and Tristan-Gough oceanic hotspot track volcanism are shown on (a) 87Sr/86Sr versus 143Nd/

144Nd and (b) 206Pb/204Pb versus 207Pb/204Pb isotope correlation diagrams. In (a) samples with MgO411 wt.% (enclosed within dashed line) show a much more restricted range in isotopic composition, yet 87Sr/86Sr ratio in some mac samples is still higher than in the oceanic part of the hotspot track, either reecting contamination by crustal material or lithospheric mantle, both of which can have extremely radiogenic Sr27,28. In (b), if only Etendeka and Parana ood basalts with 87Sr/86Sro0.7067 (highest value in the oceanic hotspot track) are considered on the uranogenic Pb isotope diagram (marked with a cross and enclosed within the eld dened by the dashed line), they only show a slightly greater range than the Gough eld, suggesting that 87Sr/

86Sr ratio can be used to effectively lter for continental lithospheric contamination. Arrows denote directions for upper and lower crustal and/or lithospheric mantle contamination. In (a) the arrow labelled Upper continental crust contamination extends into the eld for Damara S-type granites and points to the eld for Damara metasediments27. The arrow labelled Lower continental crust contamination points toward the Kaokoland gneisses (Pre-Damara basement)27. In (b) the Lower crustal contam. (contamination) arrow overlaps with and points to lower crustal granulites from the Namaqua-Natal Belt in South Africa28. The Upper crustal/lithospheric mantle contamination arrows point to the following rock groups in Namibia: (1) upper arrow Khan granodiorites (samples G12 and G13 (ref. 30) and 02/99 and 03/99 (ref. 31)) and (2) lower arrowKuiseb schists (sample Kh27 (ref. 30)) and lithospheric mantle, estimated to have a present-day composition of 206Pb/204PbB19.8 and 207Pb/204Pb B15.7 (based on sample VB32) beneath the

Spitzkoppe region in Namibia27. See Supplementary Dataset 5 and additional data from GEOROC (http://georoc.mpch-mainz.gwdg.de/georoc/

Web End =http://georoc.mpch-mainz.gwdg.de/georoc/). Average radiogenic ingrowth correction and 1s variation for Parana and Etendeka as dened in the Fig. 3 caption.

shows a very crude overall increase with decreasing age along the hotspot track (Supplementary Fig. 3), consistent with higher degrees of melting while the plume was ridge-centred. As noted previously, higher degrees of melting should have favoured melting of the more depleted Tristan component if it had been present in the plume during its earlier (B70120 Ma) history.

Finally, parental magma types range from tholeiite to alkali basalt on the Walvis Ridge and Rio Grande Rise. Therefore, if both Tristan and Gough components were present in the early plume stem (B70120 Ma), it is likely that both would have been sampled regardless of which was more fertile37,38. Since

none of the samples older than 70 Ma have clear Tristan-type compositions, we do not believe that the Tristan component, if present at all, was abundant in the early plume.

Over the last B70 Myr, the hotspot track consisted of distinct Tristan and Gough compositional geographic domains with the northwestern portion of the hotspot track (going to Tristan da Cunha Island) consisting of Tristan-type compositions and the southeastern part of the track (going to Gough Island) consisting of Gough-type compositions13 (Fig. 1). The rst known occurrence of Tristan-type lavas occurred near the SW end of the Walvis ridge when the hotspot was still ridge centred. After

6 NATURE COMMUNICATIONS | 6:7799 | DOI: 10.1038/ncomms8799 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8799 ARTICLE

B60 Ma, the plume became intraplate and the lithosphere overlying the hotspot generally became thicker with decreasing age13. Since the temporal variations in lithospheric thickness over the last 70 Myr far outweigh the spatial differences in the lithospheric thickness between the Tristan and Gough subtracks at any given time, we would expect a temporal rather than a spatial change in isotopic composition. We, however, do not see a shift in isotopic composition from one type of component to the other with age, but rather see a spatial separation throughout the last 70 Myr of the hotspots history. Although minor compared with the changes in lithospheric thickness with age, the Tristan track was located slightly closer to the Mid-Atlantic ridge during its formation and thus on slightly younger and thinner crust than the Gough track, which suggests that it may have formed through higher degrees of melting. If this were the case, we would expect more to less incompatible element ratios to be systematically lower in the Tristan lavas than in the Gough lavas. The La/Sm ratios of the lavas from the two subtracks, however, largely overlap at any given time or position along the hotspot track (Supplementary Fig. 3), suggesting that potential differences in lithospheric thickness between the two domains did not lead to a systematic difference in degree of melting. This is not surprising since the fractionated and overlapping heavy rare earth element patterns in both the Tristan and Gough lavas13 suggest that garnet was a residual phase during melting for both compositional types, and thus melting was likely to be well below the depths affected by the thickness of young lithosphere. In summary, since the isotopic composition of the erupted lavas does not appear to be related to differences in melting conditions and extent, we conclude that the plume primarily contained a Gough-type compositional range during its early history with a Tristan vertical compositional zone or stripe39,40 appearing on its northwestern side B70 Ma (ref. 13). Thereafter the plume remained bilaterally zoned. Below we discuss models to explain how a plume with relatively uniform composition turned into a zoned plume half way through its lifetime.

Although previous models have favoured the base of the lower mantle to be the source of plume zonation6,810,12,13, we will rst consider mid-mantle origins for the compositional variation. Plumes, for example, could entrain a sheath of depleted MORB-like material when they pass through the upperlower mantle transition zone41,42. Thus, we must pose the question as to whether the Tristan component could have been picked up in the transition zone. In material with a strongly temperature-dependent viscosity like the earths mantle, however, laboratory and numerical experiments suggest much smaller amounts of entrainment than for thermal plumes in an isoviscous uid40. Seismic imaging of examples of apparent plume broadening beneath 660 km suggests that there may be a higher viscosity transition zone43. This type of radial mantle viscosity structure, however, was also found to be associated with limited entrainment40. Furthermore, the high 208Pb/204Pb of the Tristan component distinguishes it from depleted upper mantle and it is hard to explain why an entrained sheath of material around the entire plume would only be sampled by melting on the plumes NW side, in particular when the plume was intraplate. Perhaps stalled slabs could be distributed irregularly in the transition zone, so that the plume may only have entrained such material on its NW side44. Recycled ocean crust is commonly believed to have a high time-integrated U/Pb ratio generating radiogenic 206Pb/204Pb at relatively unradiogenic

208Pb/204Pb over time and thus plotting beneath an extension of the MORB eld26. The Gough and Tristan components, however, have relatively low 206Pb/204Pb and high 208Pb/204Pb and plot above the MORB eld on the thorogenic Pb isotope

diagram. In conclusion, mid-mantle entrainment models do not appear to be able to adequately explain the zonation of the Tristan-Gough plume.

We propose the following model to explain the observed temporal and spatial changes in geochemistry of the Tristan-Gough hotspot track (Fig. 5). Both analogue40 and numerical39,40,45 experiments suggest that thermal plumes and starting plumes in the mantle will primarily consist of material from their basal boundary layer. According to plate motion reconstructions, the original eruption site of the ParanaEtendeka ood basalts lies vertically above and well within (B1,000 km) the western edge of the African LLSVP, dened by the 1% slow shear-wave velocity contour20,21. The TristanGough plume head is therefore likely to have been derived from the African LLSVP and, as dynamic plumes are near-vertical ow structures, its base will have sampled this region, consistent with this part of the LLSVP having a Gough-type source composition. In contrast, in these reconstructions the present Tristan-Gough hotspot is located south-southwest of the reconstructed ood basalt source region so that it now almost directly overlies the boundary of the African LLSVP. If these reconstructions are correct, the change in location could reect migration of the plume stem towards the margin of the LLSVP after the initial plume head event at an average relative speed of B1,000 km per 100 Myr or B1 cm per year. This scenario implies that both plume stem locations and the boundary of the LLVSP can slowly drift due to slow deep mantle ow46 moving at speeds of B110 mm per year, o10% of typical speeds of surface plate motions. Alternatively, the base of the plume may have remained xed in location throughout its early history. In this case, the plume may have exhausted the Gough material between the plume base and the margin of the LLSVP at B70 Ma, sucking the

LLSVP boundary into the base of the plume. A similar model has been proposed to explain the appearance of the Loa stripe in the Hawaiian plume, in which the Hawaiian plume draws in mantle from the Pacic LLSVP (Fig. 5)47. The appearance of Tristan-type compositions at B70 Ma in the Tristan-Gough hotspot track suggest that the boundary of the Atlantic LLSVP was drawn into the base of the plume at a northeastern (NE)-SW orientation, resulting in a bilaterally zoned plume assuming laminar ow in the plume conduit6,39,40,45,47,48. For largely NE-SW absolute plate motions, surface geochemical zonation is predicted by numerical experiments to form in a NE-SW direction, reecting the inow pattern at the base of the plume stem47. Since plume material is thought to rise within the stem at speeds 41 m per year (refs 46,49), it would have taken o3 Myr for Tristan-type mantle to appear in the hotspot melts after this component entered the plume stem. Over the last B70 Myr, the distance between the

Tristan and Gough subtracks has progressively increased, which could reect progressive bifurcation of a plume that roughly split along its compositional boundary24. In addition, the Tristan subtrack has become a progressively more prominent volcanic feature (in respect to erupted volume) than the Gough subtrack, suggesting that the intake of Gough-type material in the Tristan-Gough plume has systematically diminished at the expense of the Tristan-type material with decreasing age. If this trend continues, Gough material could also be exhausted from the southeastern side of the plume source in the future, such that the plume becomes entirely Tristan-like in composition. This model demonstrates how plume zonation can develop and also disappear, accompanied by an overall change in the composition of the plume from before to after its zoned history.

In contrast to this postulated evolution for the Tristan-Gough hotspot track, the Hawaiian hotspot track is not associated with a known large igneous province or ood basalt event at its initiation, and we only have the geologic record of a long-lived plume tail. The Loa component, presumably derived from the

NATURE COMMUNICATIONS | 6:7799 | DOI: 10.1038/ncomms8799 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8799

Pacic LLSVP9,10, is absent in the Emperor Seamounts and older accreted Hawaiian complexes in Kamchatka, with only the Kea component being present in the earlier history of the hotspot17. Thus far, the Loa component has primarily been found in the Hawaiian Islands8,9,10,17, but sufcient data are not available to evaluate its presence in the Hawaiian Seamounts. If the Hawaiian plume was located near to but outside of the Pacic LLSVP, it could have sucked in LLSVP material with time47. This is a potential explanation for the signicant increase in Hawaiian plume output (crustal volume ux) observed in the bathymetry and residual gravity anomalies beginning at Z15 Ma with a

further marked increase at B7 Ma (ref. 50), although passage of fracture zones bounding lithosphere of different ages/thicknesses over the hotspot no doubt also contributed to changes in volcanic ux51. In conclusion, while plumes appear to form near the boundaries of the LLSVPs20,21, plume stems may form or migrate to either side of LLSVP boundaries or, alternatively, remain stationary and pull in a migrating boundary of a LLSVP. Both situations would lead to a compositionally zoned plume. Tapping of LLSVP material, if it is hotter and/or more fertile than surrounding ambient deep mantle, would lead to an increase in productivity of the plume, whereas tapping of more depleted and/ or colder ambient mantle at the margin of a LLSVP would lead to a decrease in plume magma productivity. This scenario lends further support to the paradigm that plumes are long-lived upwelling structures within the convecting mantle, ow features that can persist despite a change in their source material. If so, there is the potential to further use geochemical evolution along other long-lived plume tracks to map the space-time variation of mantle composition near the core-mantle boundary.

Methods

Overview. Sixty dredge and drill core samples were analysed for major and trace elements and forty-one of them for SrNdHfPb DS isotope ratios in the framework of this study. A full table with all analytical data and sample information is provided in Supplementary Dataset 1.

Sampling. The samples were obtained from several repositories and were originally recovered during the US American cruises RC11 and RC16 (R/V Robert D. Conrad) in 1967 and 1972, the CIRCE cruise (R/V Argo) in 1968, cruise VM 29 (R/V Vema) in 1972, cruise AII93 (R/V Atlantis II) in 1975, during the French cruise (WALDA002) CH19 (R/V Jean Charcot) in 1971, during a Russian cruise with the R/V Akademik Kurchatov in 1975, during cruise 51 of the South African ship R/V S.A.Agulhas, during DSDP Legs 72 (Site 516F), 74 (Sites 525A, 527 and 528) with the R/V Glomar Challenger in 1980 and during German expeditions with the R/V Polarstern (ANT-XXIII5 (PS69)) in 2006 and the R/V Sonne (SO233) in 2014. Forty samples are from sites located along the Walvis Ridge, including dredge samples from the northern ank at its northeastern-most end close to the Namibian coast and from the southern ank of the southwestern part, from DSDP Sites 525A, 527 and 528, which form a transect across the southwestern end of the Walvis Ridge. Nine samples were obtained from the Rio Grande Rise on which isotope analyses were carried out on ve. These include two dredge sites at the border of a northwest-southeast trending canyon that cuts the Western Rio Grande Rise and from DSDP Site 516F on the main platform of the Rio Grande Rise. We

W E S N

Tristan-Gough Hotspot

~132 Ma Etendeka & Paran CFBs "Gough composition"

~ 2,100 km

Initial/starting plume head

African LLSVP

CMB

CMB

~ 12070 MaRio Grande Rise & Walvis Ridge "Gough"

Afr. Pl.

S. Am. plate

Ridge-centred plume tail

Removal

African LLSVP

CMB

CMB

< 70 MaSW Walvis & Guyot Province "Tristan" + "Gough"

Zoned plume

African LLSVP

Hawaiian Hotspot

10047 Ma Emperor Smnts. "Kea composition"

Pacific LLSVP

47 15 Ma Hawaiian Smts. "Kea"

Pacific Plate

Migration

Pacific LLSVP

50 Ma Hawaiian Islands "Loa" + "Kea"

Pacific LLSVP

Gondwana Pacific Plate

Figure 5 | Models illustrating the geochemical evolution of the Tristan-Gough and Hawaiian hotspots. (a) For the Tristan-Gough hotspot, the plume head is derived from the inner margin of the African low shear-wave velocity province (LLSVP) with enriched Gough composition. (b) The early plume stem continues to tap only LLSVP, but the margin of the LLSVP is continually drawn closer to the plume stem and/or the plume stem migrates towards the LLSVP boundary. (c) At B70 Ma, the LLSVP material on the northeastern side of the plume conduit is exhausted or the base of the plume may have migrated to the LLSVP boundary, so that depleted Tristan material from outside the LLSVP is also drawn into the plume conduit resulting in a zoned plume. (d) There is no geological evidence for the initiation of the Hawaiian plume. The oldest volcanic rocks associated with the hotspot are from B100 Myr old seamounts accreted to the forearc in Kamchatka17. The Hawaiian plume originally only taps the ambient depleted lower mantle (Kea component) between B10047 Ma. (e) At some time between 45 and 5 Ma (most likely between Z15 and 7 Ma when there was a peak in volcanic ux from the Hawaiian plume), enriched Loa material is drawn into the Hawaiian plume from the Pacic LLSVP (which has a distinct composition from the African LLSVP) and the plume becomes zoned. (f) After Z155 Ma the Hawaiian Plume has remained zoned. S.

Am., South American; Afr. Pl., African plate; CMB, core-mantle boundary. Proles are not to scale. No distinction has been made between the thickness of oceanic and continental lithosphere.

African Plate Pacific

Plate

CMB

CMB

8 NATURE COMMUNICATIONS | 6:7799 | DOI: 10.1038/ncomms8799 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8799 ARTICLE

also analysed ten additional primarily evolved dredge samples from the Gough subtrack and one from the Tristan subtrack within the Guyot Province.

Sample preparation. For the geochemical analyses only the least altered inner parts of the selected rock samples were used to prepare rock chips and powders. After initial jaw-crushing, the rock chips were sieved (fractions: o0.25, 0.250.5,0.51, 12, 24 and 44 mm) and thereafter repeatedly cleaned in an ultrasonic bath with deionized water for B2030 min until a clear solution was obtained. The freshest chips were then hand-picked under a binocular microscope. About510 g of 1- to 4-mm-sized whole-rock chips were taken to prepare bulk rock powders (in an agate mortar and agate ball-mill) for major and trace element and SrHf isotope analyses. In addition, 500 mg of whole-rock chips (0.51 mm fraction) was picked under a binocular microscope for SrNdPb isotope analyses.

Major element analyses. Major elements were determined by X-Ray Fluorescence Analysis (XRF) at GEOMAR Helmholtz Centre for Ocean Research Kiel (using a Philips XUnique PW 1480 X-ray orescence spectrometer), at the Institute of Mineralogy and Petrography at the University of Hamburg (using a Magix Pro PW 2540 XRF) and by inductively coupled plasma emission spectroscopy in the Acme Analytical Laboratories in Vancouver. Eleven international rock standards were measured along with the samples (JA2, JA3, JB2, JB3, JG2, JG3, JGB1, JR1, AGV2, BIR1, BHVO2 and BCR2). Results and information about data quality are presented in Supplementary Dataset 2.

Trace element analyses. Trace element analyses were carried out on an AGILENT 7500cs inductively coupled plasma mass spectrometer at the Institute of Geosciences at the Christian-Albrechts-University of Kiel after the methods of ref. 52 and the Acme Analytical Laboratories in Vancouver by inductively coupled plasma mass spectrometry (ICP-MS) subsequent to a lithium metaborate/ tetraborate fusion and nitric acid dissolution as well as an aqua regia digestion. BIR1, BHVO2, BCR2 and AGV2 were analysed as geochemical reference materials. Results and information about data quality are given in Supplementary Data set 3 (University Kiel) and Supplementary Data set 4 (Acme).

Isotope analyses. SrNdPbHf isotope analyses were carried out at GEOMAR Helmholtz Centre for Ocean Research Kiel by thermal ionization mass spectrometry (SrNdPb) and multi-collector ICP-MS (Hf) in static multi-collector mode on both types of instruments. Between 100 and 250 mg of sample chips were leached in 2 N HCl at 70 C for 1 h and subsequently triple rinsed in 18 MO water before digestion. For each sample additional Sr isotope analyses were generated on 100 mg of leached powders (6 N HCl at 150 C for 3 days) to remove possible seawater alteration effects more thoroughly. Ion chromatography followed established standard procedures53,54. Pb and most Sr isotope analyses were performed on a Finnigan MAT 262 RPQ2 while Nd and the remaining Sr isotope analyses were determined on a Thermo Fisher TRITON thermal ionization mass spectrometer. Nd and Sr ratios were normalized within run to 146Nd/

144Nd 0.7219 and 86Sr/88Sr 0.1194, respectively. Total chemistry blanks were

below 150 p.g. for both elements and thus negligible. Sample data are reported relative to 87Sr/86Sr 0.7102500.000013 (n 42; 2s external reproducibility) for

NBS987 on the MAT262 RPQ2 and 87Sr/86Sr 0.7102500.000013 (n 5; 2s

external reproducibility) on the TRITON. Nd sample data are reported relative to

143Nd/144Nd 0.5118500.000007 (n 25; 2s external reproducibility) for La

Jolla and our in-house SPEX Nd monitor 143Nd/144Nd 0.5117150.000005

(n 13; 2s external reproducibility). Pb mass bias correction followed the DS

technique55. DS-corrected values for NBS981 are 206Pb/204Pb 16.94160.0024,

207Pb/204Pb 15.49920.0024 and 208Pb/204Pb 36.72460.0061 (n 18; 2s

external reproducibility) and compare well with published double- and triple-spike data5660. Total chemistry blanks were 1040 p.g. for Pb and thus negligible. Hf chemistry followed the two-column procedure61 using unleached powders, and analyses were carried out on a VG Axiom MC-ICPMS and on a Nu plasma MCICPMS. Within-run mass bias correction used for 179Hf/177Hf was 0.7325. Total chemistry blanks were between 60 and 150 p.g. Hf. Our in-house SPEX Hf ICP standard solution (Lot no.9) was calibrated to JMC 475 (176Hf/177Hf 0.282163

(ref. 61)) and gave an average standard bracketing normalized ratio of 176Hf/ 177Hf 0.2821730.000008 (n 132; 2s external reproducibility; VG Axiom MC

ICPMS) and 0.2821700.000004 (n 48; 2s external reproducibility; Nu plasma

MC-ICPMS). Furthermore, USGS reference material BHVO2 gave 176Hf/177Hf

0.283107, which compares well with the mean of compiled values at GEOREM (176Hf/177Hf 0.2831090.000012; http://georem.mpch-mainz.gwdg.de/

Web End =http://georem.mpch-mainz.gwdg.de/ ).

Evaluation of isotope replicate analyses. Replicate analyses for SrNdPb isotopes were carried out on a second sample digestion for ve samples. Whole-rock chips were leached with warm 2 N HCl and powders with hot 6 N HCl as described above. While the reproducibility of 143Nd/144Nd lies within (three samples) or very close (two samples) to the external 2s errors of the standards, offsets outside the external 2s array of the standards are observed for radiogenic Sr and Pb isotope ratios on leached chips. In detail DSDP Leg 74 528 42 1 W 2945 lies slightly outside the external errors while a larger offset is observed for PS69/4241DR

261. The somewhat limited reproducibility for Sr on leached whole-rock chips is ascribed to variable degrees of seawater alteration even in visually homogeneous-sample chips and the inability of the leaching reagent to fully penetrate the sample and thus remove all Sr introduced through secondary processes. Notably, strong leaching of powders does not always yield the least radiogenic 87Sr/86Sr when compared with the corresponding data of the leached chips. Therefore, the least radiogenic 87Sr/86Sr of each sample is plotted and displayed in Table 1 of the manuscript, since it is presumably closest to the pristine magmatic value. Pb isotope ratios are reproduced within the external 2s errors for NBS981 for four out of ve samples in 207Pb/204Pb and three samples in 208Pb/204Pb, while 206Pb/204Pb is reproduced for only one sample under these specications. Interaction with seawater during low temperature alteration and seaoor weathering can lead to a heterogeneous enrichment of uranium that causes variable degrees of 206Pb ingrowth over time scales relevant in this paper. In this respect it is worth noting that 207Pb/204Pb is less affected by secondary U enrichment as 235U is 137.88 times less abundant than 238U presently. The two samples for which 208Pb/204Pb did not reproduce within the external 2s error of NBS981 appear to have undergone a more complex alteration history that includes Pb removal at high temperatures in addition to U enrichment at low temperatures. The slight variations caused by alteration, however, do not affect the overall scientic interpretations and conclusions derived from the Sr and Pb isotope data. Replicate Hf analyses on sample DSDP Leg 74 525A 57 5 W 141148 agreed within the external 2s errors of the standard.

References

1. Vanderkluysen, L. et al. Louisville Seamount Chain: Petrogenetic processes and geochemical evolution of the mantle source. Geochem. Geophys. Geosyst. 15, 23802400 (2014).

2. Frey, F. A. & Weis, D. Temporal evolution of the Kerguelen plume: geochemical evidence from 38 to 82 ma lavas forming the Ninetyeast ridge. Contrib. Mineral. Petrol. 121, 1228 (1995).

3. Class, C., Goldstein, S. L. & Galer, S. J. G. Discussion of Temporal evolution of the Kerguelen plume: geochemical evidence from B38 to 82 Ma lavas forming the Ninetyeast Ridge by F.A. Frey and D. Weis. Contrib. Mineral. Petrol. 124, 98103 (1996).

4. Geldmacher, J. et al. Hafnium isotopic variations in volcanic rocks from the Caribbean Large Igneous Province and Galpagos hot spot tracks. Geochem. Geophys. Geosyst. doi:http://dx.doi.org/10.1029/2002GC000477

Web End =10.1029/2002GC000477 4 (2003).

5. Harpp, K. S., Wanless, V. D., Otto, R. H., Hoernle, K. A. J. & Werner, R. The Cocos and Carnegie Aseismic Ridges: a trace element record of long-term plume-spreading center interaction. J. Petrol. 46, 109133 (2005).

6. Hoernle, K. et al. Existence of complex spatial zonation in the Galapagos plume for at least 14m.y. Geology 28, 435438 (2000).

7. Werner, R., Hoernle, K., Barckhausen, U. & Hauff, F. Geodynamic evolution of the Galapagos hot spot system (Central East Pacic) over the past 20m.y.: constraints from morphology, geochemistry, and magnetic anomalies. Geochem. Geophys. Geosyst 4, 1108 (2003).

8. Abouchami, W. et al. Lead isotopes reveal bilateral asymmetry and vertical continuity in the Hawaiian mantle plume. Nature 434, 851856 (2005).

9. Weis, D., Garcia, M. O., Rhodes, J. M., Jellinek, M. & Scoates, J. S. Role of the deep mantle in generating the compositional asymmetry of the Hawaiian mantle plume. Nat. Geosci. 4, 831838 (2011).

10. Huang, S., Hall, P. S. & Jackson, M. G. Geochemical zoning of volcanic chains associated with Pacic hotspots. Nat. Geosci. 4, 874878 (2011).

11. Chauvel, C. et al. The size of plume heterogeneities constrained by Marquesas isotopic stripes. Geochem. Geophys. Geosyst. 13, Q07005 (2012).

12. Payne, J. A., Jackson, M. G. & Hall, P. S. Parallel volcano trends and geochemical asymmetry of the Society Islands hotspot track. Geology 41, 1922 (2013).

13. Rohde, J. et al. 70 Ma chemical zonation of the Tristan-Gough hotspot track. Geology 41, 335338 (2013).

14. Clouard, V. & Bonneville, A. in Plates, Plumes & Paradigms. (eds Foulger, G. R., Natland, J. H., Presnall, D. C. & Anderson, D. L.) (2005).

15. Guillou, H. et al. Volcanic successions in Marquesas eruptive centers: a departure from the Hawaiian model. J. Volcanol. Geotherm. Res. 276, 173188 (2014).

16. Jackson, M. G. et al. Helium and lead isotopes reveal the geochemical geometry of the Samoan plume. Nature 514, 355358 (2014).

17. Portnyagin, M., Savelyev, D., Hoernle, K., Hauff, F. & Garbe-Schnberg, D. Mid-Cretaceous Hawaiian tholeiites preserved in Kamchatka. Geology 36, 903906 (2008).

18. Richards, M. A., Duncan, R. A. & Courtillot, V. E. Flood basalts and hot-spot tracks: plume heads and tails. Science 246, 103107 (1989).

19. Courtillot, V., Davaille, A., Besse, J. & Stock, J. Three distinct types of hotspots in the Earths mantle. Earth Planet. Sci. Lett. 205, 295308 (2003).

20. Burke, K., Steinberger, B., Torsvik, T. H. & Smethurst, M. A. Plume generation zones at the margins of large low shear velocity provinces on the core-mantle boundary. Earth Planet. Sci. Lett. 265, 4960 (2008).

NATURE COMMUNICATIONS | 6:7799 | DOI: 10.1038/ncomms8799 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 9

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8799

21. Torsvik, T. H., Smethurst, M. A., Burke, K. & Steinberger, B. Large igneous provinces generated from the margins of the large low-velocity provinces in the deep mantle. Geophys. J. Int. 167, 14471460 (2006).

22. Renne, P. R., Glen, J. M., Milner, S. C. & Duncan, A. R. Age of Etendeka ood volcanism and associated intrusions in southwestern Africa. Geology 24, 659662 (1996).

23. Segev, A. in From continental extension to collision: Africa-Europe interaction, the Dead Sea and analogue natural laboratories. EGU Stephan Mueller Special Publication Series. (eds SAPL, Cloetingh & Ben-Avraham, Z.) (Copernicus Publications, 2002).

24. Rohde, J. K., van den Bogaard, P., Hoernle, K., Hauff, F. & Werner, R. Evidence for an age progression along the Tristan-Gough volcanic track from new 40Ar/ 39Ar ages on phenocryst phases. Tectonophysics 604, 6071 (2013).

25. Richardson, S. H., Erlank, A. J., Duncan, A. R. & Reid, D. L. Correlated Nd, Sr and Pb isotope variation in Walvis Ridge basalts and implications for the evolution of their mantle source. Earth Planet. Sci. Lett. 59, 327342 (1982).

26. Zindler, A. & Hart, S. Chemical Geodynamics. Ann. Rev. Earth Planet. Sci. 14, 493571 (1986).

27. Thompson, R. N. et al. Origin of CFB magmatism: multi-tiered intracrustal Picrite-Rhyolite magmatic plumbing at Spitzkoppe, Western Namibia, during Early Cretaceous Etendeka magmatism. J. Petrol. 48, 11191154 (2007).

28. Class, C. & le Roex, A. South Atlantic DUPAL anomalydynamic and compositional evidence against a recent shallow origin. Earth Planet. Sci. Lett. 305, 92102 (2011).

29. Jung, S., Hoernes, S. & Mezger, K. Synorogenic melting of mac lower crust: constraints from geochronology, petrology and Sr, Nd, Pb and O isotope geochemistry of quartz diorites (Damara orogen, Namibia). Contrib. Mineral. Petrol. 143, 551566 (2002).

30. McDermott, F. & Hawkesworth, C. J. Intracrustal recycling and upper-crustal evolution: A case study from the Pan-African Damara mobile belt, central Namibia. Chem. Geol. 83, 263280 (1990).

31. Jung, S., Mezger, K. & Hoernes, S. Petrology of basement-dominated terranes:II. Contrasting isotopic (Sr, Nd, Pb and O) signatures of basement-derived granites and constraints on the source region of granite (Damara orogen, Namibia). Chem. Geol. 199, 128 (2003).32. Ballmer, M. D., Ito, G., van Hunen, J. & Tackley, P. J. Spatial and temporal variability in Hawaiian hotspot volcanism induced by small-scale convection. Nat. Geosci. 4, 457460 (2011).

33. Prytulak, J. & Elliott, T. TiO2 enrichment in ocean island basalts. Earth Planet. Sci. Lett. 263, 388403 (2007).

34. Stracke, A. & Bourdon, B. The importance of melt extraction for tracing mantle heterogeneity. Geochim. Cosmochim. Acta. 73, 218238 (2009).

35. Fitton, J. G. & Godard, M. in Origin and evolution of magmas on the Ontong Java Plateau. (eds Fitton, J. G., Mahoney, J. J., Wallace, P. J. & Saunders, A. D.) (Geol. Soc. Spec. Publ.229, 2004).

36. Herzberg, C. & Gazel, E. Petrological evidence for secular cooling in mantle plumes. Nature 458, 619622 (2009).

37. Niu, Y., Wilson, M., Humphreys, E. R. & OHara, M. J. The origin of intra-plate Ocean Island Basalts (OIB): the lid effect and its geodynamic implications.J. Petrol. 52, 14431468 (2011).38. Phipps Morgan, J. Isotope topology of individual hotspot basalt arrays: mixing curves or melt extraction trajectories? Geochem. Geophys. Geosyst. 1, 1003 (2000).

39. Farnetani, C. G. & Hofmann, A. W. Dynamics and internal structure of a lower mantle plume conduit. Earth Planet. Sci. Lett. 282, 314322 (2009).

40. Lohmann, F. C., Hort, M. & Morgan, J. P. Flood basalts and ocean island basalts: a deep source or shallow entrainment? Earth Planet. Sci. Lett. 284, 553563 (2009).

41. Fitton, J. G., Saunders, A. D., Norry, M. J., Hardarson, B. S. & Taylor, R. N. Thermal and chemical structure of the Iceland plume. Earth Planet. Sci. Lett. 153, 197208 (1997).

42. Kempton, P. D. et al. The Iceland plume in space and time: a Sr-Nd-Pb-Hf study of the North Atlantic rifted margin. Earth Planet. Sci. Lett. 177, 255271 (2000).

43. Nolet, G., Karato, S.-I. & Montelli, R. Plume uxes from seismic tomography. Earth. Planet. Sci. Lett. 248, 685699 (2006).

44. Sigloch, K., McQuarrie, N. & Nolet, G. Two-stage subduction history under North America inferred from multiple-frequency tomography. Nat. Geosci. 1, 458462 (2008).

45. Farnetani, C. G. & Hofmann, A. W. Dynamics and internal structure of the Hawaiian plume. Earth Planet. Sci. Lett. 295, 231240 (2010).

46. Morgan, W. J. Convection plumes in the lower mantle. Nature 230, 4243 (1971).47. Farnetani, C. G., Hofmann, A. W. & Class, C. How double volcanic chains sample geochemical anomalies from the lowermost mantle. Earth Planet. Sci. Lett. 359-360, 240247 (2012).

48. Kerr, R. C. & Mriaux, C. Structure and dynamics of sheared mantle plumes. Geochem. Geophys. Geosyst. 5, Q12009 (2004).

49. Schubert, G., Turcotte, D. L. & Olson, P. Mantle Convection in the Earth and Planets (Cambridge Univ. Press, 2001).

50. Vidal, V. & Bonneville, A. Variations of the Hawaiian hot spot activity revealed by variations in the magma production rate. J. Geophys. Res. 109, B03104 (2004).

51. Morgan, J. P., Morgan, W. J. & Price, E. Hotspot melting generates both hotspot volcanism and a hotspot swell? J. Geophys. Res. 100, 80458062 (1995).

52. Garbe-Schnberg, C.-D. Simultaneous determination of thirty-seven trace elements in twenty-eight international rock standards by ICP-MS. Geostand. Geoanalyt. Res. 17, 8197 (1993).

53. Hoernle, K. et al. Arc-parallel ow in the mantle wedge beneath Costa Rica and Nicaragua. Nature 451, 10941098 (2008).

54. Hoernle, K. & Tilton, G. R. Sr-Nd-Pb isotope data for Fuerteventura (Canary Islands) basal complex and subaerial volcanics: applications to magma genesis and evolution. Schweiz. Mineral. Petrogr. Mitt. 71, 318 (1991).

55. Hoernle, K. et al. On- and off-axis chemical heterogeneities along the South Atlantic Mid-Ocean-Ridge (5-11S): Shallow or deep recycling of ocean crust and/or intraplate volcanism? Earth Planet. Sci. Lett. 306, 8697 (2011).

56. Baker, J., Peate, D., Waight, T. & Meyzen, C. Pb isotopic analysis of standards and samples using a 207Pb-204Pb double spike and thallium to correct for mass bias with a double-focusing MC-ICP-MS. Chem. Geol. 211, 275303 (2004).

57. Baker, J. A., Peate, D. W., Waight, T. E. & Thirlwall, M. F. Reply to the: Comment on Pb isotopic analysis of standards and samples using a 207Pb-204Pb double spike and thallium to correct for mass bias with a double focusing MC-ICP-MS by Baker et al. Chem. Geol. 217, 175179 (2005).

58. Galer, S. J. G. & Abouchami, W. Practical application of lead triple spiking for correction of instrumental mass discrimination. Mineral. Mag. 62 A, 491492 (1998).

59. Thirlwall, M. F. Inter-laboratory and other errors in Pb isotope analyses investigated using a 207Pb-204Pb double spike. Chem. Geol. 163, 299322 (2000).

60. Thirlwall, M. F. Multicollector ICP-MS analysis of Pb isotopes using a

207Pb-204Pb double spike demonstrates up to 400 ppm/amu systematic errors in Tl-normalization. Chem. Geol. 184, 255279 (2002).61. Blichert-Toft, J., Chauvel, C. & Albarde, F. Separation of Hf and Lu for high-precision isotope analysis of rock samples by magnetic sector-multiple collector ICP-MS. Contrib. Mineral. Petrol. 127, 248260 (1997).

62. Gale, A., Dalton, C. A., Langmuir, C. H., Su, Y. & Schilling, J.-G. The mean composition of ocean ridge basalts. Geochem. Geophys. Geosyst. 14, 489518 (2013).

Acknowledgements

We are very grateful to the crews and funding sources of all the cruises and the Deep Sea Drilling Programme (DSDP) legs involved in obtaining samples from the Tristan-Gough hotspot track analyzed in this study. We thank S. Hauff and K. Junge for analytical support, J. Geldmacher for constructive comments, and the German Science Foundation (DFG) Priority Programme SAMPLE (grant HO1833/17-1, 2) and the German Ministry of Research and Education (BMBF; SO233-WALVIS II) for providing funding for this project. This research used samples provided by the Integrated Ocean Drilling Programme.

Author contributions

K.H., F.H. and R.W. conceived the project; K.H. was the primary interpreter of the data and writer of the paper with contributions from J.R., F.H., R.W. and J.P.M.; J.R., S.H. and F.H. generated the isotope and D.G.-S. the trace element data.

Additional information

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing nancial interests: The authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions/

Web End =reprintsandpermissions/

How to cite this article: Hoernle, K. et al. How and when plume zonation appeared during the 132 Myr evolution of the Tristan Hotspot. Nat. Commun. 6:7799doi: 10.1038/ncomms8799 (2015).

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

Web End =http://creativecommons.org/licenses/by/4.0/

10 NATURE COMMUNICATIONS | 6:7799 | DOI: 10.1038/ncomms8799 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.