Oxidising agents in sub-arc mantle melts link slab devolatilisation and arc magmas

Subduction zone magmas are more oxidised on eruption than those at mid-ocean ridges. This is attributed either to oxidising components, derived from subducted lithosphere (slab) and added to the mantle wedge, or to oxidation processes occurring during magma ascent via differentiation. Here we provide direct evidence for contributions of oxidising slab agents to melts trapped in the sub-arc mantle. Measurements of sulfur (S) valence state in sub-arc mantle peridotites identify sulfate, both as crystalline anhydrite (CaSO4) and dissolved SO42− in spinel-hosted glass (formerly melt) inclusions. Copper-rich sulfide precipitates in the inclusions and increased Fe3+/∑Fe in spinel record a S6+–Fe2+ redox coupling during melt percolation through the sub-arc mantle. Sulfate-rich glass inclusions exhibit high U/Th, Pb/Ce, Sr/Nd and δ34S (+ 7 to + 11‰), indicating the involvement of dehydration products of serpentinised slab rocks in their parental melt sources. These observations provide a link between liberated slab components and oxidised arc magmas.


Supplementary Table 1 | Major element (wt%) and sulfur (ppm) abundances and valence state in 'bulk' (i.e. potentially containing anhydrite) low-temperature (LT) inclusions, melt pockets (MP) and vein melt inclusions (vein MI) in the Kamchatka sub-arc mantle peridotites in this study
The major element and S data for vein MI are from Bénard et al. 5 b.d., below detection   Major element analyses by SEM (shown in Supplementary Figs 3 & 4) Analyses in italics are those of sulfides also characterised by Raman spectrometry (shown in Supplementary Fig. 6) b.d., below detection    Fig. 1a) f fO 2 calculated using the olivine-spinel temperature and the oxybarometer of Ballhaus et al. 28    - Instrumental mass fractionation f (( 34 S/ 32 S) corrected /IMF/0.04416375-1)×1000 g Data from Fiege et al. 31 ; this standard has a δ 34 S true of 1.3  Fig. 1a) c These melts are saturated in a pure H 2 O phase at 0.2 GPa  Fig. 4b)

Supplementary Discussion
The samples investigated in this study are all sub-arc mantle-derived spinel harzburgite brought to the surface by recent volcanic activity at Avacha and Ritter volcanoes, located in Kamchatka Two main modes of occurrence of LT inclusions were observed: LT inclusions in 'trails' or rims in partially reacted spinel (see Fig. 2 in Ionov et al. 3 and Supplementary Fig. 7), and LT inclusions entirely peppering strongly reacted spinel (Figs 2a & 3a). The LT inclusions contain only glass or glass and amphibole, sometimes with accessory sulfides 3 Fig. 3a, b & Supplementary Fig. 7).
The net rounded shape displayed by all the MI in Ionov et al. 3 supports their formation at high pressure and temperature in the mantle, for instance in contrast with the 'sieve' mineral-melt texture developed during recent interactions between peridotite xenoliths and their carrier magmas at sub-surface conditions 4,6 . The texture of LT inclusions and MP is also not consistent with an origin related to the late-stage percolation of the carrier magma in the deep crust, as this has been rarely but clearly identified in some Kamchatka mantle xenoliths (from the same collection as those from Ionov et al. 3 ) with the formation of Fe-rich amphibole veins intruding brittle fractures 7 .
Based on a comparison of the compositions of amphibole in the carrier magma, the Fe-rich veins and in LT inclusions, it has been demonstrated that these are not related to each other 3 .
After heating experiments were conducted by Ionov et al. 3 to homogenise LT inclusions (i.e. dissolve the possibly occurring daughter amphibole), it was further established by these authors that the distinct melt compositional fields formed by LT inclusions and MP could not be related to the pervasive infiltration of the host andesite magma (see Fig. 6 in Ionov et al. 3 and Supplementary Fig. 1). Finally, the trace element abundances in both glasses of LT inclusions and MP are too depleted to be formed by percolation of the carrier magma in the xenoliths, and suggest instead a more primitive origin (see Fig. 8 in Ionov et al. 3 and Fig. 4a).
All XANES, Raman spectrometry and EPMA data reported in this study for the determination of S and Fe valence states, respectively in LT inclusions and MP and their host spinel, have been acquired on spinel grains from sample Av33 that have not been experimentally treated (i.e. unheated). Only Kamchatka inclusions found 'naturally homogeneous', i.e. free of daughter silicate minerals such as amphibole in LT inclusions, were investigated here.
Subsequently to the study by Ionov et al. 3 , Bénard et al. 2 have reported the major element compositions of olivine and orthopyroxene from sample Av33, for which equilibrium oxygen fugacity (fO 2 ) is calculated here (Fig. 1a). These compositions are given in Supplementary Table   9.
All XANES data on vein MI reported in this study for the determination of their S valence state have been acquired on spinel grains that have not been experimentally treated (i.e. unheated).
Instead, XANES data have been acquired on the 'naturally homogeneous' vein MI from samples Av24 and Av25 (Supplementary Table 1).
In this study, we also report additional spinel-hosted MI in a mantle-derived harzburgite xenolith from the West Bismarck Arc, which petrological and geochemical features are similar to LT inclusions and MP from Kamchatka. The new West Bismarck MI are rounded and disseminated in the spinel harzburgite sample 67-02D(7) originally described by Bénard et al. 2,4 and do not show any relationship with a melt channel source (e.g. a vein) at the sample scale.
Their texture and distribution in harzburgite suggest a formation during the re-crystallisation of pre-existing (i.e. of residual origin) spinel grains in the presence of a pervasively percolating melt, which is very similar to the case of LT inclusions 3 (Fig. 3b).
It has been shown that the petrographic features West Bismarck MI cannot be related to the late-stage percolation of the carrier magmas of the xenoliths at sub-surface conditions, which produces sieve textures 6 (see Figs A1 & A2 in Bénard et al. 4 ). As for Kamchatka inclusions, the major element compositional range of West Bismarck MI is very different to that of the picrite magmas carrying the xenoliths 2 ( Supplementary Fig. 1).
All Raman spectrometry and EPMA data reported in this study for the determination of S orthopyroxene and spinel from sample 67-02D (7), for which equilibrium fO 2 is calculated here ( Fig. 1a & Supplementary Table 10). These compositions are given in Supplementary Table 6 (spinel) and Supplementary Table 9 Fig. 1) with the surrounding sub-arc mantle lithosphere, which is indicated for instance by the heterogeneous texture and modified chemistry (generally (Cr, Fe 3+ , Fe 2+ )-rich with irregular 'halos' and rims) of their host mantle spinel (see Fig. 3 in Ionov et al. 3 and Fig. 3a & Supplementary Fig. 7). This indeed suggests the infiltration of an 'exotic' melt in a primary mantle spinel, which also implies that this liquid was potentially out of equilibrium with the residual minerals from the shallow sub-arc mantle lithosphere before being trapped ( Supplementary Fig. 1).
After the entrapment of the parental melts of LT inclusions, MP formed in the spinel  for a review).
All XANES spectra in In addition to S 6+ as SO 4 2dissolved in glass, the vein MI contain S 2evidenced by a peak at 2472.5 eV (Fig. 2g & Supplementary Fig. 5d). This is in the energy range where crystalline sulfide such as pyrrhotite and pyrite have their most prominent XANES signals 13,14 . The peak at 2472.5 eV also varies in intensity and is most prominent in XANES spectra from areas with high S concentrations (0.24≤S 6+ /∑S≤0.44), while it only appears as a hump in the background of the other XANES spectra (0.62≤S 6+ /∑S≤0.88, Fig. 2d, g & Supplementary Fig. 5c, d) in the glass as for MP ( Fig. 2c-g & Supplementary Fig. 5c, d).
A high photon flux during XANES measurements may cause beam damage affecting S valence states in glasses [13][14][15] . Indeed, S 4+ is visible in some spectra with a peak at 2478 eV ( Fig. 2e-g & Supplementary Fig. 5b, d). As the beam damage only affects glasses and has not been observed in solid S-bearing compounds [13][14][15] , it is confirmed that XANES spectra from areas with low S concentrations for all inclusions predominantly represent glass analyses ( Fig. 2b- Cr/(Cr+Al)), the effects of dissolution and re-precipitation reactions on spinel compositions are thus characterised by an increase in Cr# and decrease in Mg#, whereas an opposite trend is observed for halo formation (see Fig. 3 in Ionov et al. 3 ). This records two distinct processes, with the second event (i.e. halo formation) occurring at a post-entrapment stage.

From the typical variations in spinel composition in halos around LT inclusions and West
Bismarck MI (Supplementary Tables 5 & 6 19 . However, the situation for inclusions formed in the sub-arc mantle shortly before eruption 20 may be very distinct from those residing in magma chambers for longer periods of time. In particular, the temperature effect on re-equilibration during cooling is likely more limited; it could even be insignificant in some cases, such as for LT inclusions formed at 900-1000°C (as inferred from homogenisation experiments 3  This process ultimately conducted to the formation of irregular halos. In a scenario where LT inclusions and West Bismarck MI result of a dissolution and reprecipitation process of mantle spinel by SiO 2 -rich parental melts, both phases should tend to reach redox equilibrium when the inclusions form. It appears to be the case for most of the inclusions in this study, if one estimates their fO 2 using the S 6+ /∑S equilibrium curve for basaltandesite silicate melts for low-pressure (≤0.2 GPa) conditions [21][22][23] (Fig. 1a). This is confirmed by the calculated Fe 3+ /∑Fe in spinel, based on the Fe 3+ /∑Fe equilibrium in silicate melts 24 Table 3) may correspond to higher fO 2 conditions (by ≥0.5 log unit) than those calculated using oxybarometry 27,28 and the stable composition of spinel in the percolated harzburgite samples (Fig. 1a). This suggests that a certain extent of redox disequilibrium may have existed between the percolating inclusion parental melts and the harzburgites before interactions through dissolution and re-precipitation took place, but also that this disequilibrium was partly preserved after entrapment for some inclusions. The preservation of a certain extent of disequilibrium is likely related to the specific mechanisms of pervasive melt percolation in mantle peridotites, which commonly involve kinetic fractionation with only partial re-equilibration of melts and rocks 29 .
High fO 2 conditions (ranging 1-1.5 log units above FMQ) recorded by the percolated harzburgites in this study likely result of a redox process by the S 6+ -bearing parental melts of LT inclusions and West Bismarck MI following reaction (2), or a similar one (Fig. 1a). Such process is consistent with the unusually high oxidation states of the percolated harzburgite samples in this study 2,4 . The difference in atomic composition between the irregular halos and the flat portions of the EPMA profiles show that Fe 2+ -Mg 2+ and (Fe 3+ , Al 3+ )-Cr 3+ cation substitutions in reactions (1), (3) and (4)