LETTERdoi:10.1038/nature12758

Foundering of lower island-arc crust as anexplanation for the origin of the continental MohoOliver Jagoutz1 & Mark D. Behn2

A long-standing theory for the genesis of continental crust is that itis formed in subduction zones1. However, the observed seismicproperties of lower crust and upper mantle in oceanic island arcs2,3

differ significantly from those in the continental crust4. Accordingly,significant modifications of lower arc crust must occur, if conti-nental crust is indeed formed from island arcs. Here we investigatehow the seismic characteristics of arc crust are transformed into thoseof the continental crust by calculating the density and seismic struc-ture of two exposed sections of island arc (Kohistan and Talkeetna).The Kohistan crustal section is negatively buoyant with respect tothe underlying depleted upper mantle at depths exceeding 40 kilo-metres and is characterized by a steady increase in seismic velocitysimilar to that observed in active arcs. In contrast, the lower Talkeetnacrust is density sorted, preserving only relicts (about ten to a hundredmetres thick) of rock with density exceeding that of the underlyingmantle. Specifically, the foundering of the lower Talkeetna crustresulted in the replacement of dense mafic and ultramafic cumulatesby residual upper mantle, producing a sharp seismic discontinuity atdepths of around 38 to 42 kilometres, characteristic of the conti-nental Mohorovicic discontinuity (the Moho). Dynamic calculationsindicate that foundering is an episodic process that occurs in mostarcs with a periodicity of half a million to five million years. More-over, because foundering will continue after arc magmatism ceases, thisprocess ultimately results in the formation of the continental Moho.

Continental crust is characterized by a lower-velocity upper crust(seismic P-wave velocity VP < 5–6 km s21) and a higher-velocity lower

crust (VP < 7–7.5 km s21), separated from the underlying mantle(VP < 8–8.5 km s21) by the sharp Moho4. Globally, with the exceptionof active orogenic belts or rifts, the continental Moho occurs at a rela-tively constant depth of 41 6 6 km (ref. 4). In contrast, the seismicstructure of the lower crust in many active arcs is defined by a transi-tional increase from lower crustal velocities of VP < 7 km s21 to sub-Moho velocities of VP < 7.6–7.7 km s21, significantly slower than thesub-Moho velocities observed in continental regions. The sharply definedMoho seen in continental crust is generally absent in arcs5–7; instead aweak discontinuity (increase in VP from about 6.8 to 7.2 km s21) isobserved that has been interpreted to indicate either a contact betweenmafic lower crust and unusually hot upper mantle or an intra-crustalcontact between mafic and ultramafic cumulates6. Accordingly, ifcontinental crust is formed in arcs, significant reworking of arc lowercrust must occur to transform the transitional lower-crust–mantleinterface in arcs into the sharply defined crust–mantle discontinuityof continental regions.

It is widely accepted that mafic/ultramafic rocks in arc lower crustcan become denser than the underlying upper mantle and could founderback into the upper mantle8–10. This process can explain the andesiticchemical composition of continental crust ultimately derived from basalticmantle melts10,11. Previous studies have proposed that the maximumobserved thickness of the continental crust (about 70–80 km) is con-trolled by the depth interval at which a density inversion occurs12; how-ever, the relationship between lower crustal foundering and the locationand nature of the Moho has not been established. Specifically, crustal

1Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, Massachusetts 02139-4307, USA. 2Department of Geology andGeophysics, Woods Hole Oceanographic Institution, Woods Hole, Massachusetts 02543, USA.

6 7 81.6

1.7

1.8

1.9

VP/V

S

1.6

1.7

1.8

1.9

VP/V

S

VP (km s–1)

6 7 8

VP (km s–1)

50 70 90SiO2 (wt%)

Pre

ssure

(G

Pa)

0 400 800 1,200

2.0

1.6

1.2

0.8

0.4

0

Temperature (°C)

80 mW

m –2

60 m

W m

–2

40 m

W m

–2

0.8 GPa, 750 °C 0.8 GPa, 900 °C a b c

Talkeetna

Kohistan

α-q

uartz

β-q

uartz

β-quartzα-quartz

Figure 1 | Seismic and petrological constraints on the thermal regime inarcs. a, b, Seismic velocities of representative lower-crustal rocks fromcontinents and arcs in the a–quartz (a) and b–quartz (b) stability fields(n 5 428). Boxes indicate the seismic properties observed in the lower crust ofactive arcs (see Methods for references). c, Pressure versus temperaturediagram showing the location of the a-quartz to b-quartz transition and

metamorphic pressure and temperature recorded in the Kohistan andTalkeetna sections18,19,31. Yellow and brown stars indicate the pressure andtemperature conditions used to calculate panels a and b, respectively.The observed VP/VS and VP values constrain the spatially averagedtemperatures to lie within the a-quartz field17.

0 5 D E C E M B E R 2 0 1 3 | V O L 5 0 4 | N A T U R E | 1 3 1

Macmillan Publishers Limited. All rights reserved©2013

rocks can become density unstable with respect to the upper mantleover a significant depth interval from 20 to 60 km or more, dependingon compositions and temperature conditions in the arc lower crust(see Methods for detailed discussion). Detailed knowledge of the com-position and temperature regime in the arc lower crust is thereforeessential to assess how such a foundering process could influence theseismic properties of the crust–mantle interface.

To constrain the depth at which foundering occurs in arcs we cal-culate the density and seismic properties of rocks from Kohistan andTalkeetna, the two best-exposed oceanic arc sections (see Methods forgeological setting). Previous studies have suggested that the Talkeetnaarc crust is generally less dense than the underlying upper mantle perido-tites (‘density stable’)13,14, whereas in Kohistan lower-crustal rocks denserthan the upper-mantle peridotites (‘density unstable’) are preserved15.Here we present thermodynamic modelling of observed crustal com-positions at pressures and temperatures appropriate for the formationof the arc sections (see Methods for details of the modelling). We usethese results to reconstruct the detailed density and seismic structureof the two arc sections during their formation to (1) determine the depthat which the density inversion in arcs occurs, and (2) explore the effectof foundering on the seismic properties of the arc lower crust.

To calculate the seismic/density structure of the Kohistan and Talkeetnaarcs, we first estimated the temperature in the lower crust of active arcs.Although the thermal structure of an active arc is transient owing to theinteraction between a conductive geothermal gradient and perturba-tions from frequent melt infiltration events (see ref. 16 for example), wecan use VP/VS (where VS is shear-wave velocity) estimates from the arclower crust in combination with geothermometry on metamorphicmineral assemblages to infer the spatially averaged thermal conditionsduring the construction of the arc crust. Estimates show that VP/VS

in the lower crust of active arcs is variable, but is generally 1.70–1.80with a corresponding VP of 6.5–7.5 km s21 (Fig. 1a, b)17. This low VP/VS

indicates that quartz-bearing lithologies are present in the arc lowercrust and the quartz must be mostly the low-temperature alpha-quartzpseudomorph (Fig. 1a, b).

These observations constrain the spatially averaged temperature in arclower crust to less than 800–850 uC at approximately 25–40 km depth,consistent with a conductive geothermal gradient of about 60–70 mW m22

(Fig. 1c). Similar metamorphic temperatures are preserved in Kohistan(700–800 uC at about 40–50 km; ref. 18), whereas higher temperaturesare recorded in Talkeetna (about 900–1,000 uC at depths of around 40 km;ref. 19) (Fig. 1c). On the basis of these results we calculated density andseismic properties along appropriate geotherms for Kohistan (60 mW m22)and Talkeetna (80 mW m22) (Fig. 1c). However, as discussed in theMethods and shown in Extended Data Fig. 1, the effect of temperatureon key metamorphic reactions controlling density and seismic structureis modest and does not influence the main conclusions of this study.

In both the Kohistan and Talkeetna sections an abrupt increase inVP is observed between the dominant felsic/intermediate plutonic rocksof the upper crust (6.3–6.4 km s21) and underlying mafic arc crust (6.9–7.1 km s21) (Figs 2 and 3). However, the lower crust in the two arc sec-tions differs significantly. In Kohistan, VP in the lower crust increaseslinearly between 35 km depth and 50 km depth with two minor dis-continuities (Figs 2 and 3). The first is an increase in VP at about 40 kmdepth between gabbroic rock (about 7.0 km s21) and mafic garnetgranulite (about 7.5 km s21). The second is an increase in VP betweenthe garnet granulite and the underlying ultramafic rocks (about 8.0 kms21) at approximately 50 km depth. This contact between garnet gran-ulite and ultramafic rocks has traditionally been interpreted to reflectthe seismic Moho of the Kohistan arc20,21. The discontinuity at around40 km coincides with a density inversion where the lowermost 10 kmof crust is significantly denser (Dr 5 rcrust 2 rmantle 5 40–280 kg m23)than the underlying mantle (Figs 2 and 3). This density inversioncorresponds to a pressure of about 1.0–1.2 GPa and is related to theappearance of garnet as a stable phase in mafic lithologies (the ‘garnet-in’ reaction; see Methods). Rocks above this discontinuity are generally

density-stable compared to a depleted upper mantle, whereas the rocksbelow are generally density-unstable and could founder back into theupper mantle.

In Talkeetna, the Moho is a sharp contact between the basal gab-bronorite and the underlying depleted mantle22 occurring at pressures(about 1 6 0.14 GPa; ref. 19) comparable to those of the observed densityinversion in Kohistan, and corresponding to a maximum crustal thick-ness of around 40 km depth (Fig. 3). Petrological considerations indi-cate that significant volumes of mafic/ultramafic cumulates are missingfrom the base of the Talkeetna arc22. Density-unstable garnet granulites(VP < 7.5–7.7 km s21), similar to those preserved in Kohistan, are onlypresent as relicts in a thin layer (less than about 100 m thick) situatedbetween the basal gabbronorite and the upper mantle (Fig. 2)22. Withthe exception of these garnet granulites, the Talkeetna arc crust is gene-rally density-stable (Dr 5 < 2160 kg m23) and a single large increasein VP is calculated at around 40 km depth, between gabbroic rocks(about 7.0 km s21) and the underlying depleted harzburgite (about7.9–8.0 km s21).

The calculated seismic properties of the density-stable Talkeetna lowercrust match those of the continental lower crust, whereas the density-unstable Kohistan lower crust has seismic characteristics comparableto the sub-Moho structure in active arcs (Fig. 3 and Extended Data Fig. 2).An important difference between the two sections is that the Talkeetnacrust is density sorted, whereas the lower Kohistan arc is not (Fig. 3).Density sorting of the Kohistan arc lower crust, in which unstable cumu-lates are replaced by harzburgitic sub-arc mantle, would result in a lowercrust with seismic properties comparable to those of Talkeetna and thecontinental lower crust (Figs 2 and 3). From these observations wepropose that density sorting of arc lower crust is a crucial mechanism

a b

Izu-

Bonin

Moho depth

Den

sity

un

sta

ble

7 8 3,000 3,500

Density (kg m–3)

SiO2 (wt%)

40 50 60 70 80

Den

sity s

tab

le

VP (km s–1)

50

40

30

20

10

0

Dep

th (km

)

Average continental crust

(ref. 4)

Den

sity o

f th

e u

pp

er

man

tle

Kohistan

(running average)

Figure 2 | Detailed VP and density depth-structure of the exposed Kohistanarc section compared to the average continental crust and the Izu-Bonin arccrust. a, The seismic characteristics of the reconstructed Kohistan arc aresimilar to those of the Izu-Bonin arc crust (pink line, with pink shadingindicating the variability)32 and differ significantly from the average seismiccharacteristics of continental lower crust (purple)4. Specifically, a sharp Mohothat defines the crust–mantle interface in continents is absent in arcs.b, The depth of the continental Moho coincides with the depth at which crustalrocks from Kohistan become density-unstable with respect to the depletedupper mantle (green line). Black lines indicate the running average.

RESEARCH LETTER

1 3 2 | N A T U R E | V O L 5 0 4 | 0 5 D E C E M B E R 2 0 1 3

Macmillan Publishers Limited. All rights reserved©2013

for the transformation of an arc-type Moho to a continental-typeMoho.

Density sorting of the lower crust can occur during collision, tectonicunderplating and/or normal arc buildup23. To constrain the maximumthickness that an unstable layer can achieve before foundering occurs,we calculated the timescale for the initiation of a Rayleigh–Taylor-typeinstability at the base of the arc crust as a function of the density andtemperature of the underlying mantle and the density and thickness of

the unstable layer13,24–26 and compared it to the timescale for crustalgrowth for different magma supply rates10 (see Methods for details). Fora given temperature, we assume that the unstable layer will grow untilthe timescale for instability initiation is less than the time required toform the layer. For temperatures below about 700–800 uC, instabilitytimes exceed reasonable geological timescales, because the high visco-sity of the ‘cold’ crust and underlying mantle inhibits instability growth(Fig. 4). In contrast, for temperatures over 800 uC, instability growthbecomes more efficient and the unstable layer grows to a thickness ofonly a few kilometres before foundering into the underlying mantle (ona timescale of 0.5–5 million years). The difference in the observed ther-mal regime between Kohistan and Talkeetna (Fig. 1c) is consistent withthe preservation of a thick layer of density-unstable material at the baseof the Kohistan arc section, whereas in the warmer Talkeetna arc foun-dering is predicted to be more efficient, resulting in the preservation ofa significantly thinner unstable layer (Fig. 4).

Our results show that foundering can explain both the location andprimary seismic characteristics of the continental Moho. In active arcs,an unstable layer removed by foundering will be rebuilt within a fewmillion years and so the chance of seismically imaging a newly density-sorted lower crust with a sharp Moho at about 40 km is low. After mag-matism ceases at an arc, Moho temperatures will remain high until thegeotherms conductively relax16. As long as the Moho temperature remainsabove about 700 uC, foundering will continue, but the foundering layerwill not be rebuilt. Instead, it will be replaced by upper-mantle rocks,resulting in the formation of a density-sorted continental lower-crust/upper-mantle interface with a sharp Moho discontinuity. We specu-late that unusually thick Archaean continental crust with a preserved5–10-km-thick transitional zone between crust and mantle representslower crust that has not been density sorted27. More detailed seismicstudies of stable continental regions are needed to test the abundanceof such preserved relicts.

METHODS SUMMARYCalculation of density and seismic velocity. We used Perple_X (ref. 28) to calculatesubsolidus thermodynamic phase equilibria for a range of whole-rock composi-tions from the Kohistan11 and Talkeetna22,29 arcs assuming 1 wt% H2O. Seismicvelocities and densities of the stable mineral assemblage and mode were calculated

0.5

Ko

his

tan

Talk

eetn

a

500 600 700 800 900 1,000 1,1000

5

10

15

20

25

1

5

0.5

1

5

Mantle temperature (°C)

Cum

ula

te layer

thic

kness (km

)

215 km3 km–1 Myr–1

290 km3 km–1 Myr–1

Δρ =

50 kg

m–3

Δρ =

200 k

g m

–13

Figure 4 | Modelled thickness of the density-unstable layer at the base of arccrust. The thickness was calculated by equating the timescale required foran instability to form14 with the timescale required to grow a cumulate layerbased on estimated magma fluxes10 (blue and red curves, respectively). Theboxed numbers are times required to grow the layer in millions of years. Layergrowth assumes that 70% of the original melt mass is partitioned into thecumulate layer10. Solid and dashed curves are based on different densitycontrasts between the layer and underlying mantle. The vertical bands indicatethe approximate Moho temperatures for the Talkeetna (orange) and Kohistan(green) arcs, and horizontal fields indicate the preserved thickness of thedensity-unstable layer in the two arcs.

VP (km s–1) VP (km s–1)

6 7 8

Density (kg m–3) Density (kg m–3)

2,600 3,200 3,800

Dep

th (km

)

54

48

42

36

30

24

18

12

6

0

)

Fels

ic p

luto

nic

s

Volcanic

rocks

Diorite/

gabbro

Granite-granulite/

granite-gabbro

Hornblendite/

garnetite

Volcanic

rocks

Felsic

plutonics

Diorite/

gabbro

Granite-granulite/

pyroxenite

Kohistan Talkeetna

???

6 7 82,600 3,200 3,800

Harzburgite/

dunite

Gabbro

norite

Quartz diorite

and tonalite

Density

unstable

Gap

Model A

of ref. 14

(gap filled

by felsic

plutonics)

Pyroxenite

Figure 3 | Schematic illustrations of the lithological, seismic and densityproperties of the Kohistan and Talkeetna arc sections. Shown are simplified,schematic crustal columns (after refs 19, 22 and 33) and the calculated averageseismic VP velocities and densities (black lines) of the main crustal buildingblocks of the two arcs. The thickness of the different units was approximated

using the calculated densities and existing barometric pressure estimates19,22,33.The pink and purple lines indicate the seismic velocities of the Izu-Bonin andcontinental arc crust, respectively (as in Fig. 2). The VP and VS estimates ofKohistan are after Fig. 2, and those of Talkeetna are recalculated after ref. 14.

LETTER RESEARCH

0 5 D E C E M B E R 2 0 1 3 | V O L 5 0 4 | N A T U R E | 1 3 3

Macmillan Publishers Limited. All rights reserved©2013

using a compilation of geophysical mineral properties30. We implemented anupdated version of the compilation (B. Hacker and G. Abers, personal communi-cation, 2010) into Perple_X. The variable intrusion pressures of the rocks studiedare from refs 11, 19 and 31 for Kohistan and Talkeetna, respectively; correspondingdepths were calculated by integrating the calculated density profiles for pressure.Temperatures were calculated along a 60 mW m22 geotherm for the Kohistan arc31

and a 80 mW m22 geotherm for Talkeetna19. We used the following solid solutionmodels: Atg, Chl (HP), Ctd (HP), Cpx (HP), Ep (HP), GlTrPg, Gt (HP), Pheng(HP), O (HP), Opx (HP), Pl (h), San, Sp, and T. To investigate the influence ofvariable oxygen fugacity (fO2 ) on the seismic velocity structure of an arc, we calcu-lated seismic properties and densities at Fe31/Fetotal values of 0.15, 0.25 and 0.35.All results were plotted with Fe31/Fetot 5 0.25 but the results discussed here are notdependent on fO2 .Calculation of instability timescales. Our calculation of the thickness of the unstablelayer follows the approach of ref. 13. The timescale required to form an instabilityscales inversely with the thickness of the dense layer and the density contrastbetween the layer and underlying mantle (that is, thicker layers and greater densitycontrasts lead to shorter instability times)24,25. For temperature-dependent viscositythe instability time decreases exponentially with increasing mantle temperature.For a given temperature, we assume that the unstable layer will grow until the time-scale for instability initiation is less than the time required to form the layer.

Online Content Any additional Methods, Extended Data display items and SourceData are available in the online version of the paper; references unique to thesesections appear only in the online paper.

Received 19 April; accepted 3 October 2013.

1. Taylor, S. R. & McLennan, S. M. The Continental Crust: Its Composition and Evolution(Blackwell, 1985).

2. Holbrook, W. S., Lizarralde, D., McGeary, S., Bangs, N. & Diebold, J. Structure andcomposition of the Aleutian island arc and implications for continental crustalgrowth. Geology 27, 31–34 (1999).

3. Calvert, A. in Arc–Continent Collision 87–119 (Springer, 2011).4. Christensen, N. I. & Mooney, W. D. Seismic velocity structure and composition of

the continental crust; a global view. J. Geophys. Res. B 100, 9761–9788 (1995).5. Takahashi, N., Kodaira, S., Tatsumi, Y., Kaneda, Y. & Suyehiro, K. Structure and

growth of the Izu-Bonin-Mariana arc crust: 1. Seismic constraint on crust andmantle structure of the Mariana arc-back-arc system. J. Geophys. Res. 113,B01104 (2008).

6. Tatsumi, Y. et al. Structure and growth of the Izu-Bonin-Mariana arc crust: 2. Roleof crust-mantle transformation and the transparent Moho in arc crust evolution.J Geophys. Res. 113, B02203 (2008).

7. Shillington,D. J., Van Avendonk,H. J. A.,Holbrook,W.S.,Kelemen,P.B.&Hornbach,M. J. Composition and structure of the central Aleutian island arc from arc-parallelwide-angle seismic data. Geochem. Geophys. Geosyst. 5, Q10006 (2004).

8. Kay, R. W. & Mahlburg Kay, S. Delamination and delamination magmatism.Tectonophysics 219, 177–189 (1993).

9. Arndt, N. T. & Goldstein, S. L. An open boundary between lower continental crustand mantle: its role in crust formation and crustal recycling. Tectonophysics 161,201–212 (1989).

10. Jagoutz,O.&Schmidt,M.W.Thenature andcomposition of the crustal delaminatein arcs. Earth Planet. Sci. Lett. 371–372, 177–190 (2013).

11. Jagoutz, O. & Schmidt, M. W. The formation and bulk composition of modernjuvenile continental crust: the Kohistan arc. Chem. Geol. 298–299, 79–96 (2012).

12. Sobolev, S. V.& Babeyko, A. Y. Phase transformations in the lower continental crustand its seismic structure. Geophys. Monogr. Ser. 51, 311–320 (1989).

13. Behn, M. D., Hirth, G. & Kelemen, P. B. Trench-parallel anisotropy produced byfoundering of arc lower crust. Science 317, 108–111 (2007).

14. Behn,M.D.&Kelemen, P.B.Stability of arc lowercrust: insights fromtheTalkeetnaarc section, south central Alaska, and the seismic structure of modern arcs.J. Geophys. Res. 111, B11207 (2006).

15. Jagoutz, O., Muentener, O., Schmidt, M. W. & Burg, J. P. The respective roles offlux- and decompression melting and their relevant liquid lines of descent forContinental Crust formation: evidence from the Kohistan arc. Earth Planet. Sci. Lett.303, 25–36 (2011).

16. Kelemen, P., Rilling, J. L., Parmentier, E. M., Mehl, L. & Hacker, B. R. in InsideSubduction Factory Vol. 138, 293–311 (ed. Eiler, J. M.) (Geophysical Monograph,American Geophysical Union, 2003).

17. Shillington, D. J., Van Avendonk, H. J. A., Behn, M. D., Kelemen, P. B. & Jagoutz, O.Constraints on the composition of the Aleutian arc lower crust from VP/VS.Geophys. Res. Lett. 40, 2579–2584 (2013).

18. Ringuette, L., Martignole, J. & Windley, B. F. Magmatic crystallization, isobariccooling, and decompression of the garnet-bearing assemblages of the JijalSequence (Kohistan Terrane, western Himalayas). Geology 27, 139–142 (1999).

19. Hacker, B. R. et al. Reconstruction of the Talkeetna intraoceanic arc of Alaskathrough thermobarometry. J. Geophys. Res. 113, B03204 (2008).

20. Miller, D. J. & Christensen, N. I. Seismic signature and geochemistry of an islandarc; a multidisciplinary study of the Kohistan accreted terrane, northern Pakistan.J. Geophys. Res. B 99, 11623–11642 (1994).

21. Kono, Y., Ishikawa, M., Harigane, Y., Michibayashi, K. & Arima, M. P- and S-wavevelocities of the lowermost crustal rocks from the Kohistan arc: implications forseismic Moho discontinuity attributed to abundant garnet. Tectonophysics 467,44–54 (2009).

22. Kelemen, P., Hanghoj, K. & Greene, A. in The Crust Vol. 3 Treatise on Geochemistry(ed. Rudnick, R. L.) 593–659 (Elsevier-Pergamon, 2003).

23. Hacker, B. R., Kelemen, P. B. & Behn, M. D. Differentiation of the continental crustby relamination. Earth Planet. Sci. Lett. 307, 501–516 (2011).

24. Jull, M. & Kelemen, P. B. On the conditions for lower crustal convective instability.J. Geophys. Res. B 106, 6423–6446 (2001).

25. Conrad, C. P. & Molnar, P. The growth of Rayleigh-Taylor-type instabilities in thelithosphere for various rheological and density structures. Geophys. J. Int. 129,95–112 (1997).

26. Conrad, C. P., Behn, M. D. & Silver, P. G. Global mantle flow and the development ofseismic anisotropy: differences between the oceanic and continental uppermantle. J. Geophys. Res. 112, B07317 (2007).

27. Guggisberg, B., Kaminski, W. & Prodehl, C. Crustal structure of the Fennoscandianshield: a traveltime interpretation of the long-range FENNOLORA seismicrefraction profile. Tectonophysics 195, 105–137 (1991).

28. Connolly, J. A.D.Computationofphaseequilibriaby linearprogramming:a tool forgeodynamic modeling and its application to subduction zone decarbonation.Earth Planet. Sci. Lett. 236, 524–541 (2005).

29. Greene, A. R., DeBari, S. M., Kelemen, P., Blusztajn, J. S. & Clift Peter, D. A detailedgeochemical study of island arc crust: the Talkeetna arc section, south–centralAlaska. J. Petrol. 47, 1051–1093 (2006).

30. Hacker, B.R.&Abers,G.A. Subduction factory3: anExcel worksheetandmacro forcalculating the densities, seismic wave speeds, and H2O contents of minerals androcks at pressure and temperature. Geochem. Geophys. Geosyst. 5, Q01005(2004).

31. Jagoutz, O. et al. TTG-type plutonic rocks formed in a modern arc batholith byhydrous fractionation in the lower arc crust. Contrib. Mineral. Petrol. 166,1099–1118 (2013).

32. Kodaira, S. et al. New seismological constraints on growth of continental crust inthe Izu-Bonin intra-oceanic arc. Geology 35, 1031–1034 (2007).

33. Jagoutz, O. & Schmidt, M. W. The formation and bulk composition of modernjuvenile continental crust: the Kohistan arc. Chem. Geol. 298–299, 79–96 (2012).

Acknowledgements Theworkwas supported byNSF grantnumbersEAR 0910644 (toO.J.) and EAR 1316333 (to M.D.B.). We thank N. Arndt for comments that helped toimprove the manuscript. J. Connolly’s help in recalibrating the elastic propertycalculation of Perple_X is appreciated, as are discussions with P. Kelemen andB. Hacker.

Author Contributions O.J. designed the project. Both authors conducted thecalculations, contributed to the interpretation of the results and wrote the manuscript.

Author Information Reprints and permissions information is available atwww.nature.com/reprints. The authors declare no competing financial interests.Readers are welcome to comment on the online version of the paper. Correspondenceand requests for materials should be addressed to O.J. (jagoutz@mit.edu).

RESEARCH LETTER

1 3 4 | N A T U R E | V O L 5 0 4 | 0 5 D E C E M B E R 2 0 1 3

Macmillan Publishers Limited. All rights reserved©2013

METHODSThe density of arc lower crust rocks. Crustal rocks that are denser than the under-lying upper mantle peridotite are considered density unstable. Density-unstable rockscan either form as relatively Fe-rich ultramafic cumulates derived from mantle-derived melts, which are as dense or slightly denser than the underlying mantleat magmatic temperature but can become significantly denser upon cooling9,41.Additionally, dense garnet-bearing cumulates can form from hydrous basaltic–andesitic liquids at high pressures (above about 1 GPa)42. Three important pressure-dependent metamorphic reactions result in the formation of dense minerals (suchas spinel and garnet), which strongly control the density of Fe-rich and Al-rich(such as gabbroic) compositions in the arc lower crust.

The three main densification reactions likely to be important in the lower arccrust are: (1) The breakdown of plagioclase next to olivine, which occurs at pres-sures of about 0.6–0.7 GPa (ref. 43): olivine 1 plagioclase R pyroxene 1 spinel (I)(2) The formation of metamorphic garnet due to the breakdown of plagioclase at0.8–1 GPa (ref. 43): plagioclase 1 orthopyroxene R garnet 1 quartz (II) (3) Thebreakdown of plagioclase at about 1.2–1.6 GPa: albite R jadeite 1 quartz (III).

The importance of each reaction for densification depends on the bulk com-position of the system. Reaction (I) is important for olivine and plagioclase-richrocks (troctolite and olivine-gabbro)15, which probably form in thin arcs wheremagma fractionation occurs at shallower crustal levels and the olivine 1 plagioclasestability field is increased42. Reaction (II) is important for rocks with high Fe/Mgratios and high Al-content and low Si-content, such as cumulates formed fromhydrous arc magmas at increased pressures15. Reaction (III) will only be relevantfor strongly over-thickened arc crust.

Additionally, the depth range corresponding to pressures of about 0.8–1 GPa atwhich reaction (II) occurs varies significantly depending on the density structureof the arc crust. In juvenile arcs, where most of the arc crust is composed of rockswith approximately basaltic compositions with densities of around 2,900–3,100 kgm23, pressures of 0.8–1 GPa correspond to depths of about 26–34 km. In contrast,in mature arcs that have a significant thickness of granitic upper crust (such as theIzu-Bonin arc) with densities as low as 2,600–2,800 kg m23, pressures of 0.8–1 GPacan correspond to depths of up to 32–40 km. Accordingly, the depth range in whichdelamination—owing to the formation of magmatic/metamorphic garnet and/orpyroxene and spinel—occurs is 20–70 km, depending in detail on the compositionof the rocks in the arc crust.Geological setting. The Kohistan arc, exposed in northeast Pakistan, was a long-lived Jurassic/Cretaceous to Tertiary island arc that formed in the equatorial partof the Neotethyan ocean separating India and Eurasia before the India–Asia col-lision. The Kohistan arc exposes a complete arc section ranging from unmeta-morphosed sediments in the north to upper-mantle rocks in the south. Pressureand temperature estimates for the lowermost mafic arc crust indicate pressures inexcess of about 1.5 GPa for the crust–mantle transition.

The Talkeetna arc, exposed in south central Alaska, is a Triassic island arc thatwas active from about 200–175 million years ago. It exposes rocks ranging fromunmetamorphosed sediments and associated volcanics in the north of the arc toupper-mantle rock in the south of the arc. Owing to large-offset strike–slip fault-ing, the middle crust is partly missing19. The lowermost mafic arc crust recordsmaximum pressures of 1.0–1.1 GPa, indicating a slightly shallower crust–mantletransition in the Talkeetna compared to the Kohistan.Calculation of density and seismic velocity. We used Perple_X (ref. 28) to calcu-late subsolidus thermodynamic phase equilibria for a wide range of whole-rockcompositions from the Kohistan11 and Talkeetna22,29 arcs assuming 1 wt% H2O.Seismic velocities (VP, VS) and densities of the stable mineral assemblage andmode were calculated using a compilation of geophysical mineral properties30.We implemented an updated version of the compilation (B. Hacker and G. Abers,personal communication, 2010) into Perple_X. The variable intrusion depths ofthe rocks studied are from refs 11, 19, 31 and 44 for Kohistan and Talkeetna,respectively. Temperatures were constrained along a 60 mW m22 geotherm con-strained for the Kohistan arc31,44 and a 80 mW m22 geotherm for the Talkeetnaarc19. We used the following solid solution models in our calculation: Atg, Chl(HP), Ctd (HP), Cpx (HP), Ep (HP), GlTrPg, Gt (HP), Pheng (HP), O (HP), Opx(HP), Pl (h), San, Sp and T.

To investigate the influence of variable fO2 on the seismic velocity structure of anarc, we calculated seismic properties and densities at Fe31/Fetotal values of 0.15,0.25 and 0.35. All results were plotted with Fe31/Ftotal 5 0.25 (ref. 45) but theresults discussed here are not dependent on fO2 .The effect of temperature on the density structure. The thermal regime in thelower crust is poorly constrained and probably highly variable through time owingto the intrusion of hot basaltic liquids. To evaluate the effect of variable tempera-ture we calculated the density and seismic structure of the Kohistan and Talkeetnacrust at 40, 60 and 80 mW m22 geotherms (Extended Data Fig. 1). Because mag-matic and metamorphic phase boundaries involving significant volume changes(and corresponding density changes) are dominantly pressure-dependent, and onlyto a limited extent temperature-dependent, the density structure is only marginallyinfluenced by the thermal structure. The most important reaction at higher tempera-ture is the breakdown of hydrous phases (for example, amphibole), which generallybreak down to a denser phase (for example, pyroxene). However, this transforma-tion has only a limited effect on density and seismic properties (Extended Data Fig. 1).Calculation of instability timescales. Our calculation of the thickness of the unstablelayer follows the approach of ref. 13. The timescale required for an instability toform scales inversely with the thickness of the dense layer, and the density contrastbetween the layer and underlying mantle (that is, thicker layers and greater densitycontrasts lead to shorter instability times)24,25. In addition, for temperature-depen-dent viscosity the instability time decreases exponentially with increasing mantletemperature. For a given temperature, we assume that the unstable layer will growuntil the time required for instability initiation is less than the time required to formthe layer.References for Fig. 1. The VP/VS and VP estimates for different arcs in Fig. 1 aretaken from refs 46–49.

34. Nakanishi, A. et al.Crustal evolution of the southwestern Kuril Arc,Hokkaido Japan,deduced from seismic velocity and geochemical structure. Tectonophysics 472,105–123 (2009).

35. Kopp, H. et al. Deep structure of the central Lesser Antilles Island Arc: relevance forthe formation of continental crust. Earth Planet. Sci. Lett. 304, 121–134 (2011).

36. Iwasaki, T. et al. Crustal and upper mantle structure in the Ryukyu Island Arcdeduced from deep seismic sounding. Geophys. J. Int. 102, 631–651 (1990).

37. Iwasaki, T. et al. Precise P and S wave velocity structures in the Kitakami massif,Northern Honshu, Japan, from a seismic refraction experiment. J. Geophys. Res.99, 22187–22204 (1994).

38. Kodaira, S. et al. Seismological evidence for variable growth of crust along the Izuintraoceanic arc. J. Geophys. Res. 112, B05104 (2007).

39. Takahashi, N. et al. Crustal structure and evolution of the Mariana intra-oceanicisland arc. Geology 35, 203–206 (2007).

40. Calvert, A. J., Klemperer, S. L., Takahashi, N. & Kerr, B. C. Three-dimensional crustalstructureof the Mariana island arc fromseismic tomography. J.Geophys. Res. 113,B01406 (2008).

41. Muntener, O. & Ulmer, P. Experimentally derived high-pressure cumulates fromhydrous arc magmas and consequences for the seismic velocity structure of lowerarc crust. Geophys. Res. Lett. 33, L21308 (2006).

42. Muntener, O., Kelemen, P. B. & Grove, T. L. The role of H2O during crystallization ofprimitivearc magmasunder uppermost mantle conditionsandgenesis of igneouspyroxenites; anexperimental study. Contrib. Mineral. Petrol. 141, 643–658 (2001).

43. Kushiro, I. & Yoder, H. S. Jr. Anorthite-forsterite and anorthite-enstatite reactionsand their bearing on the basal-eclogite transformation. J. Petrol. 7, 337–362(1966).

44. Jagoutz, O. The fine scale seismic structure of an exposed island arc section basedon field and petrological constraints. AGU Fall Meet. Abstr. 2628 (2011).

45. Cottrell, E. & Kelley, K. A. The oxidation state of Fe in MORB glasses and the oxygenfugacity of the upper mantle. Earth Planet. Sci. Lett. 305, 270–282 (2011).

46. Zhang, H. et al. High-resolution subducting-slab structure beneath northernHonshu, Japan, revealed by double-difference tomography. Geology 32, 361–364(2004).

47. Syracuse, E. M. et al. Seismic tomography and earthquake locations in theNicaraguan and Costa Rican upper mantle. Geochem. Geophys. Geosyst. 9,http://dx.doi.org/10.1029/2008GC001963 (2008).

48. Eberhart-Phillips, D. et al. Imaging the transition from Aleutian subduction toYakutat collision in central Alaska, with local earthquakes and active source data.J. Geophys. Res. 111, http://dx.doi.org/ 101029/2005JB004240 (2006).

49. Wang, Z. & Zhao, D. Vp and Vs tomography of Kyushu, Japan: New insight into arcmagmatism and forearc seismotectonics. Phys. Earth Planet. Inter. 157, 269–285(2006).

LETTER RESEARCH

Macmillan Publishers Limited. All rights reserved©2013

Extended Data Figure 1 | Seismic velocity and density along differentgeotherms for the Kohistan arc. Plotted are the mean and range in VP anddensity as calculated along the 40, 60 and 80 mW m22 geotherms.

RESEARCH LETTER

Macmillan Publishers Limited. All rights reserved©2013

Extended Data Figure 2 | Seismic velocities of the lower arc and continentalcrust. Histogram showing distribution of average seismic velocities directlyabove and below the Moho in continents (red, after ref. 4) and from active arcs(refs 6, 8, 32, 34–40). Also shown are the range of VP for density-stable anddensity-unstable rocks from the Kohistan and Talkeetna arcs, as dashed fieldscalculated from this study. In the arcs, sub-Moho rocks have on average aVP that is 0.5 km s21 slower than do sub-Moho rocks in continents. Theobserved low velocities in the arcs agree with the velocities calculated fordensity-unstable crustal rocks from Kohistan.

LETTER RESEARCH

Macmillan Publishers Limited. All rights reserved©2013