Arc-like magmas generated by mélange-peridotite interaction in the mantle wedge

The mechanisms of transfer of crustal material from the subducting slab to the overlying mantle wedge are still debated. Mélange rocks, formed by mixing of sediments, oceanic crust, and ultramafics along the slab-mantle interface, are predicted to ascend as diapirs from the slab-top and transfer their compositional signatures to the source region of arc magmas. However, the compositions of melts that result from the interaction of mélanges with a peridotite wedge remain unknown. Here we present experimental evidence that melting of peridotite hybridized by mélanges produces melts that carry the major and trace element abundances observed in natural arc magmas. We propose that differences in nature and relative contributions of mélanges hybridizing the mantle produce a range of primary arc magmas, from tholeiitic to calc-alkaline. Thus, assimilation of mélanges into the wedge may play a key role in transferring subduction signatures from the slab to the source of arc magmas. Mélange rocks are predicted to form at the slab-mantle interface in most subduction zones, but their role in arc magmatism is still debated. Here, the authors show that melting of peridotite hybridized by mélange rocks produces melts that carry the major and trace element abundances of natural arc magmas.

S ubduction zones are widely studied because they are a major locus of volcanic and seismic hazards. In particular, the compositions of arc magmas have been used to understand the magmatic processes operating in the deep Earth. During subduction, hydrated oceanic crust and sediments are subducted and recycled back into the Earth's interior. Although the fate of subducted sediments is uncertain, their signature is imprinted in the chemistry of most arc magmas around the world 1 . Sediments are globally enriched in many trace elements (e.g., K, Rb, Th, rare earth elements) relative to peridotite mantle 2 , thus small volumes of sediments can drastically shift the trace element budget of the mantle wedge. Arc magmas are also characteristically enriched in fluid-mobile large-ion lithophile elements (LILE) such as Ba and Sr, and depleted in high field strength elements (HFSE) such as Nb, relative to mid-ocean ridge basalt (MORB) 3 . The LILE enrichment has usually been attributed to mantle wedge metasomatism by slab-derived fluids 4 produced during dehydration of the subducting slab. The HFSE depleted character, on the other hand, has been attributed to different processes such as a 'pre-subduction' mantle depletion 5,6 , selective retention of HFSE by accessory phases (e.g., rutile, sphene, and perovskite) stabilized in the mantle wedge and/or in the slab 7,8 , and preferred partitioning of HFSE into orthopyroxene during melt-rock reaction 9 . Although extensive geochemical studies have suggested that arc magma chemistry reflects variable contributions from a depleted MORB mantle (DMM), altered oceanic crust (AOC) and sediments 10,11 , experimental studies have faced challenges to simultaneously reproduce both the major and trace element characteristics of tholeiites and calc-alkaline melts, the most common types of arc magmas. In addition, the processes by which typical trace element signatures are produced and transferred to arc magmas remain a matter of debate. In particular, it has been recently argued that the trace element and isotope variability of global arc magmas could not be reconciled with the classic model of arc magma genesis, which invokes hybridization of the mantle wedge by discrete pulses of melted sediments and aqueous fluids from dehydrating AOC. Instead, the trace element and isotope data of global arcs can only be reconciled if physical mixing of sediments + fluids + mantle takes place early on in the subduction process before any melting occurs 12 . This prerequisite redefines the order of events in subduction zones and supports an important role for mélange rocks in arc magmatism. However, the trace and major element chemistry of melts that would result from the interaction of natural mélange rocks with a peridotitic mantle in subduction zones has never been investigated experimentally and remains unknown. Such information is critical to determine whether mélange rocks are viable contributors to arc magmatism worldwide.
Mélange rocks are observed in field studies worldwide 13 and are believed to form by deformation-assisted mechanical mixing, metasomatic interactions and diffusion at different P-T conditions along the slab-mantle interface during subduction [13][14][15][16] . Mélanges are hybrid rocks composed of cm to kmsized blocks of altered oceanic crust, metasediments, and serpentinized peridotite embedded in mafic to ultramafic matrices 14,17,18 . These matrix rocks include nearmonomineralic chlorite schists, talc schists, and jadeitites with variable amounts of Ca-amphibole, omphacite, phengite, epidote, and accessory minerals (e.g., titanite, rutile, zircon, apatite, monazite, and sulfides), among others. Although the volumes of mélange rocks at depth are poorly constrained, several km-thick low-seismic velocity regions observed at the slab-top in subduction zones worldwide indicate the persistence of hydrated rocksmélange zonesat the slab-mantle interface 14,15,19 . This km-scale estimate of mélange rocks from seismic observations is corroborated by numerous field studies of exhumed high-pressure terranes reporting thicknesses ranging from several hundreds of meters up to several kilometers 14,17,[20][21][22] . Mélange rocks display significant spatial heterogeneity, but detailed field observations indicate that chemical potential gradients between juxtaposed lithologies (e.g., metasediments, eclogite, and serpentinized peridotites) may be reduced to homogenous matrices through diffusion and fluid advection processes as mélanges mature 14,23 . For the purpose of this first study, we will assume that mélange matrices are broadly representative of the bulk composition of the mélange and provide a relevant first-order approximation of mélange compositional variability as they form at the expense of, and reflect chemical contributions from their protoliths 15,21 . Although more compositions will be studied in the future, the mélange matrix samples used here reflect two contrasting members in the range of mélange materials that we use to explore possible melt compositions produced by mélangeperidotite interaction.
Laboratory 24 and numerical simulations of subduction process [25][26][27][28] have shown that hydration and partial melting may induce gravitational instabilities at the slab-mantle interface, which can develop into diapiric structures composed of partially molten materials. Although these diapirs have not been unambiguously imaged in active subduction zones, we note that alongarc geophysical studies are rare, that the current resolution of seismic techniques may not be appropriate to image mixed mélange-peridotite lithologies, and that magnetotelluric approach, sensitive to interconnected free fluids, would not easily detect the presence of mélanges, where most of the water may be crystallographically bounded. With their intrinsic buoyancy, mélange diapirs have been predicted to form at the slab-top, migrate to the overlying mantle 25,26 , and transfer the compositional signatures of slab-derived rocks to the source region of arc magmas 23,29 . In particular, physical mixing and homogenization of viscous mélange diapirs and sub-solidus mantle peridotites is predicted in the hot zones of the mantle wedge 30 . Recent findings on ophiolitic zircon grains also support the idea that material can be transported in the wedge via cold plumes 31 . However, as stated previously, the major and trace element compositions of melts that would be produced by melting of a mélange-hybridized mantle wedge remains unexplored.
Here we present the first experimental study on the generation of arc-like magmas by melting of mélange-hybridized mantle sources. We perform piston-cylinder experiments at 1.5 GPa and 1150-1350°C and simulate a scenario where mélange materials rise as a bulk 26,32 into the hot corner of the wedge and homogenize with the peridotite mantle ( Fig. 1). Using powder mixtures of DMM-like natural peridotite (LZ-1, Supplementary Fig. 1; 85-95 vol. %) and natural mélange rocks from a high-pressure terrane (SY400B, SY325; 5-15 vol. %), we show that experimentally produced glasses display the major and trace element characteristics typical of arcs magmas (e.g., high Ba contents, high Sr/Y ratios, and negative Nb anomaly). Our study provides evidence that the compositional signatures of sediments and fluids, initially imparted to mélange rocks during their formation at the slab-mantle interface, can be delivered to the source region of arc magmas by mixing of mélange materials with mantle wedge peridotites, and variably enhanced during melting of mélangehybridized peridotite source. We show that depending on the types and relative contributions of mélange materials that hybridize the mantle wedge, the compositions of the melts vary from tholeiitic to calc-alkaline. We further discuss how lithological heterogeneities observed in supra-subduction ophiolites and arc xenoliths could represent direct evidence for peridotitemélange interactions.

Results
Experimental techniques. We performed piston-cylinder experiments to investigate the composition of melts produced by partial melting of a natural DMM-like peridotite hybridized by small proportions of natural mélange matrix. We used two starting mixes that consisted of homogenized 'peridotite + sediment-dominated mélange matrix' (PER-SED mix) and homogenized 'peridotite + serpentinite-dominated mélange matrix' (PER-SERP mix). Both mélange matrices are fine-grained chlorite schists from Syros (Greece) with estimated water contents between 2-3 wt. %. These two types of natural mélange matrices span a range of compositions that reflect the first-order variability of global mélange rocks in terms of mineralogy (Supplementary Data 1), immobile element chemistry (Fig. 2), and trace element chemistry ( Supplementary Fig. 6). As mélange rocks should be volumetrically small compared to peridotite in the mantle wedge, we only added limited volumes (5-15%) of natural mélange matrix to a natural lherzolite powder (85-95%). We note that mélange rocks would not necessarily represent 5-15 vol.% of the sub-arc region at all times because of the 3-D nature of mélange diapirs. Certain regions of the wedge could be hybridized by different amount of mélange materials at different times. Although this experimental design is more challenging because it produces small melt pools, it simulates a more realistic scenario. Experimental melts were collected using glassy carbon spheres placed at the top of Au-Pd capsules. The natural peridotite (LZ-1; from Lherz, France) displays modal proportions and major and trace element compositions similar to DMM ( Supplementary  Fig. 1). The PER-SED and PER-SERP starting materials were partially melted at 1.5 GPa and 1280-1350°C, conditions applicable to arc magmatism 33,34 . In addition, near-solidus (1230°C ) and solidus (1150°C) experiments were performed to better constrain the solid phase assemblage at the beginning of and before melting, respectively. The quenched, dendrite-free glasses were analyzed for major elements using electron microprobe (EPMA) at the Massachusetts Institute of Technology. In addition, chemical maps for major elements were acquired on all experiments ( Fig. 3 and Supplementary Fig. 2). Trace element compositions of glass pools were analyzed using a Cameca 3 F secondary ion mass spectrometer (SIMS) at the North East National Ion Microprobe Facility (Woods Hole Oceanographic Institution). Backscattered electron (BSE) images and energy dispersive spectroscopy (EDS) maps were acquired on all experiments using a Hitachi tabletop SEM-EDS TM-3000. The major and trace element compositions of starting mixes and experimental melts are summarized in Supplementary Data 1 and 2, respectively. We assessed approach to equilibrium by performing a time-series of experiments at 1.5 GPa and 1280°C, with run durations ranging from 3 h to 96 h. The capsules were preconditioned to minimize Fe loss, although we still observed a decrease in FeO T (total iron) with increasing run duration. We observed that melt compositions performed between 72 and 96 h were indistinguishable in terms of SiO 2 , Al 2 O 3 , MgO, Na 2 O, CaO, K 2 O, MnO, and TiO 2 , within 1 s.d. value (Supplementary Fig. 4). Thus, a 72-h run duration was chosen to closely approach equilibrium in those experiments. Mass balance calculations yielded a sum of squared residuals <0.39 (FeO excluded), attesting for a close system for all other major oxides. Phase proportions for each experiment were calculated from mass balance calculations and are reported in Supplementary Data 3. Additional information is provided in the Supplementary Information. The alkali contents of melts produced from PER-SED experiments are higher than those from PER-SERP experiments at similar temperatures, due to the higher alkali contents of PER-SED starting material (Supplementary Data 1 and Supplementary  Fig. 6). The FeO T contents of peridotite-mélange melts are lower than global arc data due to some limited Fe loss to the capsule ( Supplementary Fig. 7). We now compare the major element compositions of experimental melts ( Fig. 4 and Supplementary  Fig. 7) with fractionation-corrected global arc data 35 (normalized to MgO = 6 wt.%), primitive arc melts compilations 33,34 , and previous experimental studies (Supplementary Data 4). Experimental hydrous peridotite melt compositions 36 match well the major element compositions of global arcs, although alkali contents are expectedly lower than in most arc magmas. Experimental melts from mantle hybridized by slab melts 37 are lower in CaO, and higher in TiO 2 , Na 2 O, K 2 O, and SiO 2 compared to arc datasets. Experimental melts from olivine hybridized by sediment melt 38 are lower in CaO, and higher in Na 2 O and K 2 O compared to arc datasets. Experimental mélange-type 1 and type 2 melts 27,39 are both lower in CaO and MnO and higher in K 2 O and SiO 2 compared to arc datasets. Interestingly, the major element compositions of experimental mélange-type 2 melts 40 , which are partial melts from the sediment-dominated mélange material used in this study (SY400B), plot in the continuity of PER-SED experiments but with higher elemental abundances. Experimental melts from mantle hybridized by sediment melts 41 are higher in K 2 O compared to arc datasets. Conversely, partial melts of peridotite hybridized by mélange materials produced in this study plot within or near the compositional field defined by arc datasets for SiO 2 , MgO, Na 2 O, K 2 O, TiO 2 , P 2 O 5 , and CaO. In terms of alkali contents, lower degree melts (10-19%) of PER-SED experiments are slightly higher than global arcs but plot within the field of global arcs at higher degree of melting (25-31%). Overall, partial melts of peridotite hybridized by mélange materials are similar to partial melts of hydrous peridotites and match well the alkali and major element compositions of typical arcs magmas.
Experimental melts from PER-SED experiments range from the boundary between tholeiitic and calc-alkaline fields to high-K calc-alkaline field (Fig. 5). On the other hand, experimental melts from PER-SERP experiments plot tightly within the tholeiitic field. Experimental mélange-type 2 melts 40 show a strong enrichment in K 2 O and plot in the ultrapotassic shoshonitic field. Our results, along with the experimental data of Cruz-Uribe et al. 40 , highlight a continuum in alkali enrichment from tholeiitic melts produced by melting of mantle hybridized by serpentinitedominated mélange, to calc-alkaline melts produced by melting of mantle hybridized by sediment-dominated mélange materials, to ultrapotassic shoshonitic melts from melting of pure sedimentdominated mélange materials.
Trace element composition of the melts. The trace element compositions of hybrid peridotite-mélange melts are presented in N-MORB-normalized spider diagrams ( Fig. 6) along with global arc data 35 , with emphasis on the dominant primitive arc magma types 33 (i.e., calc-alkaline and tholeiite), and published experimental studies that provided both major and trace element contents of experimental melts (Supplementary Data 4). Primitive calc-alkaline arc magmas are geochemically characterized by up to two orders of magnitude higher trace element concentrations compared to primitive arc tholeiites. Pure sediment melts 7 and melts from olivine hybridized by sediment melts 38 have higher trace element concentrations than global arc magmas and display elemental fractionations that are different from global arcs (e.g., Ba/Th, Sr/Nd). Other previous studies 37,40,42 display trace element abundances that plot in the highest range for natural arc magmas, but with major element compositions that are missing CaO or reflect ultra-potassic melts (high K 2 O). Here we show that, compared to N-MORB, partial melts of hybrid peridotitemélange materials display enrichment in LILE (e.g., Ba, Th, Sr, K), high LREE/HREE (e.g., Ce/Yb), high LILE/HFSE (e.g., high Th/Nb, Ba/Nb, and K/Ti), and plot tightly within the trace element fractionation range defined by global arc data (Fig. 7). Experimental melts from PER-SED experiments record elevated trace element concentrations and show fractionations that are characteristic of primitive calc-alkaline magmas. Sr/Nd ratios still fall within the range of global arcs (Fig. 7), although within the lower range of values. Experimental melts from PER-SERP experiments display trace element concentrations that are an order of magnitude lower than melts from PER-SED experiments, and show a slight enrichment in Sr relative to Ce and Nd. In PER-SED experiments, Zr-Hf are slightly enriched compared to Sm and Ti, whereas in PER-SERP experiments, Zr-Hf are not fractionated from Sm and Ti. Trace element concentrations in the melts generally decrease with increasing temperature, consistent with dilution at higher melting extents in the absence of accessory phases that would retain trace elements in the residue. Overall, melts produced from melting of a peridotite source hybridized by mélange rocks (this study) carry trace element signatures typical of natural arc magmas. In particular, peridotite hybridized by serpentinite-dominated and sediment-dominated mélanges produced melts that carry the trace element characteristics of arc tholeiites and calc-alkaline magmas, respectively.

Discussion
Geodynamic models of rising mélange diapirs have predicted an uneven distribution of mélange rocks in the mantle wedge that involves both complete and incomplete mixing of mélange rocks and peridotites 30 . Our experiments investigate a scenario where the peridotite mantle wedge and limited volumes of mélange rocks are fully mixed and form a new hybrid source that partially melts (Fig. 1). As the extent and volumetric significance of mélange rocks at the slab-mantle interface are still debated, a growing number of studies support their ubiquitous occurrence and importance at the slab-mantle interface. Petrologic modeling 43 , numerical instability analysis of subduction zones 44,45 , and metamorphic P-T-t histories of exhumed high-pressure mélange terranes [46][47][48] strongly support the possibility of exhumation of high-pressure rocks through diapirism within the mantle wedge. Further experiments will model how the path of mélange materials would be affected by the thermal structure of individual subduction zones but are beyond the scope of the current study.
For the purpose of this study, we consider that the two endmember mélange matrices from Syros (Fig. 2) offer compositions that represent a reasonable first-order approximation of global mélange variability. Our choice of using natural chlorite schist   39 , while the experimental mélange-type 2 melts are from Cruz-Uribe et al. 40 Our experiments are plotted as averages with error bars representing 1 s.d. All the data, including the literature, are plotted on volatile-free basis matrix from Syros (Greece) was guided by the fact that the Syros mélange record the mechanical and metasomatic interactions at P-T conditions appropriate for slab-mantle interface at depths of about 50-60 km in subduction zones 16,21,23 . In addition, the chlorite ± talc-dominated assemblage in global mélange matrices (including Syros mélange) is relatively insensitive to pressure 49,50 , making them a reasonable proxy to the type of mélange extending down to sub-arc depths 14,15,51 . Importantly, our natural starting mélange materials record minimal late-stage modification and overprinting during their exhumation, making their mineralogy, elemental, and volatile concentrations 21 closely approximate the in-situ compositions of mélange rocks at the slab-mantle interface. Thus, the present study offers a reasonable approximation of subduction dynamics where mélange rocks formed at 1.6-2.2 GPa, detach from the slab and homogenizes with peridotite in the hot zones of the mantle wedge at 1.5 GPa (~45 km depth). Results from our experiments support the idea that primary melts in arcs are not only limited to MgO − rich (up to 15.9 wt.%) basalt but may also resemble trachyandesite and basaltic trachyandesite with MgO contents of around 7 wt.% (Supplementary Data 2), covering the MgO range of primitive arc magmas 33 . All of our experiments display CaO, K 2 O, Na 2 O, TiO 2 , and P 2 O 5 that more accurately reproduce the chemistry of global arc magmas compared to previous studies that simulated hybridization of the wedge by discrete slab melts or discrete sediment melts. The fact that the hybrid source is largely peridotite-like (85-95%) explains why realistic, arc-like major element compositions can be produced in our experiments. Indeed, the large dominance of mantle-equilibrated arc magmas from different subduction zones should reflect the fundamental control of mantle peridotites in controlling the major element compositions of primary arc melts 34,52 .
The presence of mall mélange components within the mantle wedge significantly affects the trace element budget of melts generated by melting of a mélange-hybridized mantle source. Although hydrous melting of peridotite would typically produce melts that display a MORB-like trace element pattern 53,54 , the trace element compositions of peridotite-mélange melts show striking similarity with global arc magmas, with enriched LILE such as Ba, Th, and K, and depleted HFSE such as Nb and Ti. Previous experimental studies on mantle hybridization by slab melts 37 and sediment melts 42 also produce melts enriched in LILE and depleted in HFSE (Fig. 6d); however their major element compositions mostly reflect (ultra-) potassic shoshonitic melts (high K 2 O) that occur lesser widely in subduction zones worldwide. Traditionally, melts with high Sr/Y signature have been interpreted as slab melts due to the presence of garnet in the melting residue 55 while the high Th/Nb signature was interpreted to record contribution from sediments melts, as Th can be mobilized more efficiently in sediment melts 56 . In addition, high Ba contents have traditionally been ascribed to addition of fluids 57 . The peridotite-mélange melts plot tightly within the range defined by global arcs for ratios that have traditionally required discrete sedimentary, slab melt, and/or AOC fluid addition to the arc magma source 57 . In particular, the peridotitemélange melts carry arc-like Sr/Y, Th/Nb, Ba/Th, K/Ti ratios among others (Fig. 7).
In nature, there exists a large compositional variability in primitive arc magmas, ranging from arc tholeiites to calc-alkaline and shoshonites. However, such compositional variability and their spatial distributions (or the lack thereof) have not been satisfactorily explained. Primitive arc tholeiites are usually thought to be produced by decompression style melting (similar to MORB), whereas the classically invoked model for the formation of primitive calc-alkaline magmas envisages their production by flux melting of the mantle induced by the addition hydrous slab components. These slab components are responsible for the up to two orders of magnitude higher trace element concentrations of primitive calc-alkaline magmas relative to N-MORB. For instance, the elevated Th-Zr-TiO 2 concentrations of primitive calc-alkaline magmas reflects higher slab contributions in their sources 33 . We have shown that melts produced from melting of a mantle hybridized by sediment-dominated mélanges (PER-SED) strongly resembled primitive calc-alkaline magmas whereas melts produced from melting of a mantle hybridized by serpentinite-dominated mélanges (PER-SERP) strongly resembled primitive arc tholeiites, both in terms of major (e.g., K 2 O, TiO 2 ) and trace element abundances (e.g., Ba, Th, Zr) and in terms of fractionation characteristics (Fig. 6).
It is critical to determine whether those abundances and fractionations are simply inherited from the starting material or if they are enhanced during melting of the mélange-hybridized peridotite. We make several important observations regarding elemental abundances and fractionations in the melt compared to the starting materials. The bulk starting compositions of PER-SED 95-5 and PER-SERP 85-15 experiments (the two types of experiments that are dominated by ultramafic componenteither peridotite or serpentine) fall either outside of the global arc range or within the lower range of values observed in arcs (Fig. 6a, c). It is thus clear that melting plays an important role in producing elemental abundances that are similar to values observed in global arc magmas.
The bulk composition of PER-SED 85-15 experiments (more strongly influenced by a sediment-dominated mélange) is already similar to global arcs for most elements (Fig. 6b), and less surprisingly, melting produces melts that are also similar to arcs. Yet, regardless of abundances, characteristic element ratios acquire a slightly enhanced "arc-like" signature for most elemental ratios  during melting of mélange-hybridized peridotite. Specifically, Ba/ Th, Sr/Y, Zr/Hf, Zr/Sm, and K/Ti slightly increased in melts compared to the starting materials; Ba/Nb, Sr/Nd, and Sm/Nd stay relatively unchanged whereas Th/Nb and Th/Zr slightly decreased compared to the starting materials (Fig. 7). Experimental melts produced from PER-SED experiments have higher Ba than melts produced from PER-SERP experiments because the sediment-dominated mélange matrix initially had a higher Ba content than the serpentine-dominated mélange matrix (Supplementary Figs. 6 and 8). Still, melts that are produced during melting of PER-SED and PER-SERP starting materials have slightly higher Ba/Th, Sr/Y, Zr/Hf, Zr/Sm, and K/Ti and slightly lower Th/Nb and Th/Zr ratios (compared to starting materials), and thus are not only inherited from the starting materials.
In Supplementary Fig. 9, we show that primitive arc magmas mainly record Nb/Ce N < 1 (normalized to N-MORB 58 ), but their Zr/Sm N can be below or above 1 and is unrelated to the magma type. In addition to Nb depletion and low Nb/Ce ratios, depletion in Zr and Hf is seen for example in shoshonites from Sulawesi and Fiji, and in calc-alkaline basalts from Solomon and Bismarck 33 . However, Zr, Hf, and Zr/Hf are actually variable in natural primitive arc magmas. Elevated Zr-Hf and Zr/Sm N ( >1) observed in low-degree melts from PER-SED experiments are features that are observed in natural arc magmas such as calcalkaline andesites from Japan and New Zealand, calc-alkaline basalts from Mexico, and depleted andesites from Izu-Bonin. Meanwhile, low-degree melts from PER-SERP experiments have Zr/Sm N < 1 that overlap with some HFSE-depleted arc magmas such as tholeiitic basalts and andesites from Japan, Cascades and Tonga arcs. We note that in PER-SED experiments, elevated Zr/ Sm (and Hf/Sm) does not reflect inheritance from the mélange matrix ( Supplementary Fig. 6). Instead, the variability in Zr-Hf contents and Zr/Hf in natural mélange matrices most likely reflect some Zr-Hf mobility in the absence/destabilization of zircon 15 . Overall, the trace element characteristics of our experimental melts plot well within the range of primitive arc magmas (Fig. 7). Thus, these experiments do not only reproduce elemental abundances (major and trace) but also elemental fractionations observed in global arc magmas. In addition, we show that although the trace element compositions of peridotitemélange melts are partly inherited from the mélanges themselves (i.e., some characteristic subduction signatures may be already imprinted at the slab interface), those arc-like abundances and fractionation signatures can be readily produced and variably enhanced during melting of a mélange-hybridized mantle source (i.e., additional fractionation should occur in the mantle wedge).
Using chemical maps and high-resolution BSE images, we did not observe accessory phases, unlike what had been found in pure mélange melt residues 40 . Our results indicate that elements that have similar incompatibilities during pure peridotite melting can still be slightly fractionated during mélange-hybridized peridotite Experimental pure sediment melts are from Skora and Blundy 7 . Experimental melts of a sediment-dominated mélange material (mélange-type 2 melts) are from Cruz-Uribe et al. 40 The literature data are plotted as averages with error bars representing 1 s.d melting. Also, we did not observe HFSE-or REE-compatible accessory phases that could retain these elements in the residue. Niobium depletion was in part inherited from the starting bulk compositions (Supplementary Fig. 6) but we hypothesize that it was enhanced by the preferential partitioning of Nb into orthopyroxene 9 . In particular, the presence of an opx-rich reaction zone in all 72-h experiments could have contributed to Nb depletion in the melts. The opx-rich band is likely due to reaction of hydrous melts with the peridotite assemblage, as has been observed in previous studies 59 . Natural pyroxenites, including orthopyroxenites, have been ubiquitously found in exhumed mantle sections. Previous experimental 27,59 and field-based studies [60][61][62] have pointed out that orthopyroxenites should form as reaction products of hydrous melts and mantle minerals. The ubiquitous occurrence of orthopyroxenites exposed in suprasubduction zone ophiolites such as in the Josephine 62 , Coast range ophiolites 63 , and UHP Maowu Ultramafic Complex 64 , and sampled in arc-related xenoliths 65 , may also potentially record the hybridization of mantle wedge by mélange materials 31 . Thus, the incorporation of mélange diapirs into the mantle wedge may also have implications for the formation of mineralogical and lithological heterogeneities in the mantle. This study has important implications for the understanding of subduction zone magmatism. During subduction, mélange diapirs may propagate, and dynamically mix with the overlying mantle. Our study shows that depending on the nature and relative contributions of the hybridizing mélange materials in the source of arc magmas, a large variety of primary magmas with characteristic arc-like signatures can be produced. As LILEenriched shoshonitic melts are expected to form from melting of pure sediment-dominated mélange materials 40 , our study shows that both primitive arc tholeiites and primitive calc-alkaline magmas, which are the two most abundant magma types in subduction zones worldwide, can be produced by melting of mantle hybridized by serpentinite-dominated and sedimentdominated mélange materials, respectively. The rarer occurrence of ultrapotassic shoshonites as compared to tholeiites and calc-alkaline magmas likely reflects the volumetric significance of peridotites in the wedge and the dilution effect due to mixing of mélange materials with mantle wedge peridotites. The absence of systematic along-and across-arc spatial distributions of primitive tholeiitic and calc-alkaline arc magmas is consistent with the complexity involved in mélange-diapir ascent paths, and their eventual distributions and mixing with peridotite in the mantle wedge.
In summary, this experimental study provides unique constraints for the role of mélange materials in arc magmatism, as invoked in previous studies. We have shown that melting of a mélange-hybridized peridotite represents a mechanism to generate melts with major element, trace element and trace element ratios characteristic of tholeiitic and calc-alkaline arc magmas. In these experiments, the compositions of starting materials, P-T conditions, and melting degrees were designed to be as realistic as possible compared to natural processes in the mantle wedge. Where mélanges can form and ascend into the wedge, variations in their compositions, thicknesses, and relative contributions in the arc magma source will likely result in the formation of compositionally diverse primary arc melts and can result in the formation of lithological heterogeneities in the mantle. Mélange transfer from the subducting slab to the mantle wedge may be one of several mechanisms by which arc magmas are produced, but we emphasize that both major and trace element of experimental melts need to be reported to better assess how closely we can simulate arc processes. Although further experiments will help decipher the type and amount of mélange materials that could be involved in individual subduction zones, we show that hybridization of peridotite by buoyant mélange rocks is a viable process to transfer crustal signatures from the slab surface to arc magmas.

Methods
Starting material preparation. Alteration-free, natural peridotite (LZ-1; typelocality in Lherz, France) was grinded to a fine powder using agate ball mill. The bulk composition of LZ-1 is similar to DMM 66 in major and trace element compositions ( Supplementary Fig. 1) and is here considered to be representative of peridotite mantle wedge. Following grinding, the LZ-1 powder was loaded into a nickel bucket and preconditioned in a 1-atm vertical gas-mixing furnace at 1100°C with fO 2 maintained at the FMQ buffer (Fayalite-Magnetite-Quartz buffer) for 96 h. This fO 2 was adjusted by changing the partial pressures of CO and CO 2 gases  Fig. 7 Trace element ratios of experimental melts compared to natural arc magmas. Trace element fractionations of experimental peridotite-mélange melts (a-c) compared to the bulk starting compositions (yellow, green, and red lines) and global arc ratios defined by Turner and Langmuir database 35 (white rectangles) in the furnace, and is within the range of estimated fO 2 for sub-arc mantle 67,68 . Two chlorite schist matrices from Syros (Greece) were selected to represent two end-member compositions of global mélange rocks: the sediment-dominated mélange matrix (SY400B) and the serpentinite-dominated mélange matrix (SY325). Both natural mélange matrices contain water contents of~2-3 wt. %. We acknowledge that there exists a wide range in chemical and mineralogical compositions of exhumed mélange rocks worldwide and that there is no single rock material that can represent such wide variability. In order to account for this and capture its first-order variability, we selected two mélange matrix rocks from Syros (Greece) based on mineralogical assemblages (Supplementary Data 1), immobile element chemistry (Fig. 1), and trace element chemistry ( Supplementary Fig. 5). In Supplementary Data 1, the mineralogical assemblages of SY400B and SY325 are consistent with being derived from a sediment-like and ultramafic/serpentinite-like protoliths, respectively. Using immobile element systematics, Fig. 1 shows a mixing trend between serpentinites and sediment/upper crustal rocks to account for the range in global mélange variability where mélange material SY400B plotted close to GLOSS composition while SY325 plotted close of DMM composition. In Supplementary Fig. 5, the trace element composition of SY400B closely resemble the GLOSS composition while SY325 broadly resemble the DMM-like peridotite, with exception for some highly fluid-mobile elements (e.g., U, K). SY400B and SY325 from Syros record minimal late-stage modification and overprinting during their exhumation, making their mineralogy, elemental and volatile concentrations 21 closely approximate the in-situ compositions of mélange rocks at the slab-mantle interface. Taken together, the mineralogy, immobile element (Cr vs Cr/Al) and trace element chemistry strongly support for the representability of mélange materials SY400B and SY325 to cover for the first-order variability in global mélange composition. Since Syros mélange is one of the most studied and wellconstrained exhumed high-pressure mélange terranes in terms of its metamorphic P-T-t condition 69,70 and mélange formation 21,71,72 , we have more control on the conditions at which our starting materials have been subjected to and the processes that led to their formation. These natural mélange materials were grinded to fine powders using agate ball mill. Experimental setup. Partial melting experiments were performed in 0.5′′ endloaded solid medium piston cylinder device 73 at the Woods Hole Oceanographic Institution (USA). The starting mixes were packed in Au 80 Pd 20 capsules conditioned (Fe-saturated) to minimize Fe loss 36 . The Au 80 Pd 20 capsules were conditioned by packing MORB-like basalt powder (AII92 29-1) in the capsules and firing them at 1250°C in a 1-atm vertical gas-mixing furnace with fO 2 maintained at 1 log unit below FMQ buffer for 48 h. Ideally, we would have used actual starting materials to condition the capsules, but due to limited quantities of starting materials we considered that a MORB-like basalt would provide enough Fe to saturate the capsules. The silicate glass was removed from the Au 80 Pd 20 capsules using warm HF-HNO 3 bath.
When loading the starting material into the conditioned Au 80 Pd 20 capsules, approximately 35-45 mg of the starting mix was first packed in the capsule and then topped with 3.5-4 mg of vitreous carbon spheres (80-200 µm in diameter) to act as melt entrapments. The capsule was triple-crimped and welded shut. All the experiments were performed in a CaF 2 pressure assembly with pre-dried crushable MgO spacers, straight-walled graphite furnace and alumina sleeves. The sealed capsule was strategically positioned in the assembly such that the top portion of the capsule is the side that contains the vitreous carbons spheres to facilitate easy migration of melt during the experiment. Silica powder was placed in between the sealed capsule and alumina sleeve to fill up the space and maintain the capsule's position. Lubricated Pb foils were used to contain the friable CaF 2 assembly and to provide lubrication between the assembly and the bore of the pressure vessel.
The experiments were performed at 1280-1350°C and 1.5 GPa, relevant to arc magma generation 74,75 . Run duration was set at 72 h after verifying approach to equilibrium from a 3 h to 96-h time-series (see paragraph below). Pressure was applied using the cold piston-in technique 76 where the experiments were first raised to the desired pressure before heating them at desired temperature at the rate of 60°C/min. The friction correction was determined from the Ca-Tschermakite breakdown reaction to the assemblage anorthite, gehlenite, and corundum 77 at 12-14 kbar and 1300°C and is within the pressure uncertainty ( ± 50 MPa). Thus, no correction was applied on the pressure in this study. Temperature was monitored and controlled in the experiments using W 97 Re 3 /W 75 Re 25 thermocouple with no correction for the effect of pressure on thermocouple electromotive force. Temperatures are estimated to be accurate to ±10°C and pressures to ±500 bars, and the thermal gradient over the capsule was <5°C. The experiments were quenched by terminating power supply and the run products were recovered. The capsules were longitudinally cut in half before mounting in epoxy. All the mounted capsules were polished successively on 240-to 1000-grit SiC paper before the final polishing on nylon/velvet microcloth with polycrystalline diamond suspensions (3-0.25 µm) and 0.02 µm colloidal silica. Vacuum re-impregnation of capsules with epoxy was performed to reduce plucking-out of the vitreous spheres during polishing.
Approach to equilibrium. Approach to equilibrium was evaluated by performing a time-series of experiments using PER-SED (95-5) starting material at 1.  (Supplementary Fig. 4). It has been shown experimentally that hydrous melting of peridotite produces melts with lower FeO* (~6 wt. %) contents than anhydrous equivalents (~8 wt. %) 36 but we also observed a decrease in FeO T with increasing run duration, which suggests Fe loss. We speculate that this Fe loss/depletion reflects one or a combination of the following causes: (1) Fe diffusion to the Au 80 Pd 20 capsule due to incomplete Fe saturation during conditioning; (2) formation of orthopyroxenite reaction zone, which could have further contributed to Fe depletion. Other observation that indicates a close approach to equilibrium in our experiments is the homogenous distribution of minerals in the matrix away from the reaction zone, and homogeneous major element compositions within single capsule.
Another way of assessing equilibrium between the melt and minerals, and check whether the experiment behaved as a closed system, is based on the quality of mass balance calculations performed for all the major elements. Using the MS-Excel optimization tool Solver, we obtained low values for the sum of squared residuals (<0.39), for all the major elements, except for Fe, attesting for a close approach to equilibrium for all other major oxides in our experiments, and confirming a small amount of Fe loss in the capsule walls. Phase proportions for each experiment estimated from the mass balance calculations were verified visually in every experiment.
Electron microprobe analysis. Major element compositions of the quenched melts and coexisting silicate minerals from all experimental run products were analyzed using JEOL JXA-8200 Superprobe electron probe micro-analyzer at Massachusetts Institute of Technology. Analyses were performed using a 15 kV accelerating potential and a 10 nA beam current. The beam diameter varied depending on the target point. For quenched melt pools, beam diameters varied between 3 μm to 10 μm (mostly 5 μm) depending on the size of the melt pools. For silicate minerals, a focused beam (1 µm) was used. Data reduction was done using CITZAF package 78 . Counting times for most elements were 40 s on peak, and 20 s on background. In order to prevent alkali diffusion in glass, Na was analyzed first for 10 s on peak and 5 s on background. All phases (melt and coexisting minerals) were quantified using silicate and oxide standards. The compositional maps for different major elements were performed at similar instrumental setup using a focused beam. Major element compositions of melts and minerals are reported in Supplementary Data 2 and 5, respectively.
Secondary ion mass spectrometry. Concentrations of selected trace elements in melt pools (usually <30 µm diameter) were obtained using a Cameca IMS 3f ion microprobe at the Northeast National Ion Microprobe Facility (NENIMF) at the Woods Hole Oceanographic Institution (WHOI). Analyses were done using 16 O − primary ion beam with 8.4 keV voltage, 500 pA to 1 nA current and~10 µm beam diameter. No raster was used in the beam. Positive secondary ions are accelerated to a nominal energy of 4.5 keV. The energy window of the mass spectrometer was set to 30 eV. 30 Si was set as the reference isotope and ATHO-G, T1-G, StHs6/80-G glasses were used as standards 79 . Analyses were carried out in low mass resolution (m/δm = 330) with an energy offset of −85 V. Secondary ions were counted by an electron multiplier. A 1800 µm diameter field aperture size was used for most of the measurements. We did not use the field aperture to block any of the ion image from the sample since the spot was already very small. Each measurement consists of four minutes of pre-sputtering, then five cycles with an integration of 10 s/cycle for 30 Si and 10 s/cycle for elements 88 Sr, 89 Y, 90 Zr, 93 Nb, 138 Ba, and 30 s/cycle for 140 Ce, 143 Nd, 147 Sm, 174 Yb, 180 Hf, 232 Th, and 238 U. Th concentrations are reported if 1SE error is above detection limit. 1SE error for U is below detection limit for all measurements so U is not reported. In total, each analysis spot requires a total analysis time of approximately 60 min. Reduced trace element concentrations of melts obtained by SIMS are reported in Supplementary Data 2. Internal errors from analyses (2 SE) and error from calibration curves (2SE) have been propagated and are incorporated in the total 2SE error reported in the figures and Datasets.
X-ray fluorescence technique (XRF). Whole-rock elemental concentrations of LZ-1, SY400B, and SY325 were analyzed using X-ray fluorescence technique for major elements and inductively coupled mass spectrometer technique for trace elements at GeoAnalytical Laboratory at Washington State University. As stated before, whole-rock compositions (major and trace elements) of LZ-1, SY325, Major element variability of residual phases. Major element compositions of residual minerals are homogeneous through the capsule in individual experiments, and vary between experiments due to differences in temperature and starting compositions ( Supplementary Fig. 10). They are within the range of values observed in peridotites worldwide, although Fe loss probably artificially increased Mg# of minerals (93-96 in olivine; 91-95 in clinopyroxene; 92-95 in orthopyroxene). Temperature (1280-1350°C) has variable effect on mineral compositions. For example, with increasing temperature, olivines display a slight decrease in Al 2 O 3 , a constant CaO, and a slight increase in MgO. D MgO ol/melt decreases with increasing temperature. Orthopyroxenes display a slight decrease in TiO 2 and Al 2 O 3 with increasing temperature, whereas SiO 2 and MgO increase, and CaO is constant. As predicted experimentally, D Al2O3 opx/melt decreases with increasing temperature and D Na2O opx/melt increases with increasing temperature 80