Fossilized Melts in Mantle Wedge Peridotites

The shallow oxidized asthenosphere may contain a small fraction of potassic silicate melts that are enriched in incompatible trace elements and volatiles. Here, to determine the chemical composition of such melt, we analysed fossilized melt inclusions, preserved as multiphase solid inclusions, from an orogenic garnet peridotite in the Bohemian Massif. Garnet-poor (2 vol.%) peridotite preserves inclusions of carbonated potassic silicate melt within Zn-poor chromite (<400 ppm) in the clinopyroxene-free harzburgite assemblage that equilibrated within the hot mantle wedge (Stage 1, > 1180 °C at 3 GPa). The carbonated potassic silicate melt, which has a major element oxide chemical composition of K2O = 5.2 wt.%, CaO = 17 wt.%, MgO = 18 wt.%, CO2 = 22 wt.%, and SiO2 = 20 wt.%, contains extremely high concentrations of large ion lithophile elements, similar to kimberlite melts. Peridotites cooled down to ≅800 °C during Stage 2, resulted in the growth of garnet relatively poor in pyrope content, molar Mg/(Mg + Fe + Ca + Mn), (ca. 67 mol.%). This garnet displays a sinusoidal REE pattern that formed in equilibrium with carbonatitic fluid. Subsequently, subduction of the peridotite resulted in the formation of garnet with a slightly higher pyrope content (70 mol.%) during the Variscan subduction Stage 3 (950 °C, 2.9 GPa). These data suggest the following scenario for the generation of melt in the mantle wedge. Primarily, infiltration of sediment-derived potassic carbonatite melt into the deep mantle wedge resulted in the growth of phlogopite and carbonate/diamond. Formation of volatile-bearing minerals lowered the density and strength of the peridotite. Finally, phlogopite-bearing carbonated peridotite rose as diapirs in the mantle wedge to form carbonated potassic silicate melts at the base of the overriding lithosphere.

in incompatible trace elements such as light-REE (LREE), Pb, U, and Li. Here, we examine fossilized melt inclusions 11 to constrain the major and trace element compositions of potassium-rich carbonatitic melts in the mantle wedge.

Mantle wedge peridotites in the Bohemiam Massif
To study melts within the mantle wedge, we investigated an orogenic garnet peridotite from the Moldanubian Zone of the Bohemian Massif (BM) in the Czech Republic, which experienced corner flow in the mantle wedge and contains MSIs (i.e. fossilized melt) 7,12,13 . Orogenic garnet peridotite bodies within the Moldanubian Zone occur as lenticular bodies ranging in size from several centimetres to kilometres across, and are surrounded by felsic granulites and gneisses. A previous study has shown that Mg-Cr-rich garnet peridotites within granulite equilibrated in a low P/T regime (2.5-3.0 GPa, 1015-1335 °C) and were possibly derived from the asthenosphere, whereas those occurring within migmatitic orthogneiss equilibrated in a medium P/T regime (3.3-6.0 GPa, 875-1180 °C) within the lithospheric mantle wedge 14 . The Blanský les granulite 15 in the BM has a cylindrical dome structure and contains numerous serpentinized garnet peridotite bodies that occur mainly in a wide NW-SE striking serpentinite belt in the central part of the granulite massif. At the SE margin of the serpentinite belt, fresh spinel-garnet peridotite crops out within the Plešovice quarry. Figure 1 summarizes the thermal history of the Plešovice peridotite 7,12 , which may have experienced P-T conditions within the diamond stability field at the earliest stage (Stage 0, > 4 GPa). The peridotite was then uplifted to shallow depths (Stage 1) 7 and subsequently cooled into the chlorite stability field (Stage 2, < 750 °C). It was then subducted into the high P-T garnet lherzolite field (Stage 3, 950 °C at 2.9 GPa). At the final stage, the peridotite was partially recrystallized to spinel lherzolite assemblage (Stage 4, 750 °C at 1.5 GPa). Previous Sm-Nd geochronological studies have yielded ages of 375 to 329 Ma for the garnet peridotites in the Moldanubian Zone of the BM 14 .
The spinel-garnet peridotite at Plešovice is a depleted lherzolite consisting mainly of olivine (Ol) and orthopyroxene (Opx) (14-16 vol.%) with minor clinopyroxene (Cpx) (1 vol.%) and calcic amphibole (Amp) (1-2 vol.%) and accessory chromian spinel (Chr), phlogopite (Phl) and apatite (Ap) 7 . Mineral abbreviations follows ref. 16 . Two lithological types of the spinel-garnet peridotite, Type A and Type B, have been distinguished by variations in the abundance of garnet (Grt) (~2 and ~5 vol.%, respectively). Garnet occurs as sporadic large grains (>5 mm) in Type A peridotites, whereas they are ubiquitous and show a range of grain sizes (0.1-30 mm) in Type B peridotites (Fig. 2a,b). We document the trace element composition of the minerals in order to correlate metasomatic events within the existing framework of P-T constraints 7,12 .

Reconstruction of melt compositions using multiphase solid inclusions
Multiphase solid inclusions (MSIs) have been widely observed in high-pressure minerals within UHP metamorphic rocks and orogenic garnet peridotites, and are interpreted as the crystallization products of dense, solute-rich supercritical fluids or melts 11,24 . Many studies have been undertaken on MSIs in order to reconstruct the original chemical composition of supercritical fluids or melts formed in subduction zones 11,13,24 . The Plešovice garnet peridotite preserves fossilized melts as MSIs that consist of volatile-rich minerals 7,13 . Based on rare accessory minerals, including priderite and burbankite, parental melts to the MSIs are speculated to have been kimberliteor carbonatite-like 13 , although quantitative estimates of melt compositions are lacking. All observed Plešovice MSIs are within Chr 1 grains enclosed in Grt 2 /Grt 3 crystals. Large grains (>3 mm in length) of Chr 1 contain exsolution lamellae of diopside and enstatite 12 (Fig. 4a), suggesting that chromite contained pyroxene components during Stage 0 12 (i.e. Chr 0 ). However, we found no MSIs in these larger Chr 1 grains. Smaller Chr 0 grains recrystallized into exsolution-free chromite (Chr 1 ) + diopside (Cpx 1 ) + enstatite (Opx 1 ) (Fig. 4b), and most MSIs occur in these exsolution-free Chr 1 grains. Chr 1 has a relatively high Cr# [molar Cr/(Cr + Al) = 0.57-0.64] but is deficient in Zn (as low as ~350 ppm; Fig. 1b) and equilibrated during Stage 1 (Fig. 1a). MSIs show negative crystal shapes and consist mainly of hydrous aluminosilicate minerals, phosphates and carbonates (Fig. 4c-f). Single-phase phlogopite inclusions also occur within Chr 1  We examined 37 polished thin sections by electron microscopy and found 243 inclusions that are free of severe secondary alteration (Supplementary Table S3). These inclusions occur mainly within Type A peridotites (187), but some occur within Type B peridotites (56). The main carbonates in the MSIs are dolomite and calcite, with a dolomite/calcite volume ratio of 2.0 ± 0.6. Rarely, carbonates are magnesite, Ba-Ca-Mg carbonate, strontianite, and burbankite 13 . Other minerals identified in the MSIs include Nb-rich rutile, a Mn-Fe-Ti-W-Ta phase (possibly koragoite), Cr-priderite 13 , monazite, a U-Th oxide and various sulphide minerals including maucherite, orcelite, millerite, galena and pentandite (Supplementary Table S3). The occurrence of such a wide variety of minerals in the MSIs can be explained by use of the phase rule (see Supplementary Discussion -application of the phase rule to multiphase solid inclusions). Some MSIs contain chlorite, calcic amphibole, talc, serpentine, brucite, and diaspore. These MSIs are connected with the matrix via annealed fractures (Supplementary Fig. S2f). We did not use such altered MSIs for the reconstruction of melts.
Crystals of apatites and phlogopites are ubiquitous within both the MSIs and the matrix, and their chemical compositions correlate with textures, for both major elements (Fig. 4g,h; Supplementary Table S1) and trace elements (Fig. 5a,b; Supplementary Table S2). Apatite grains in MSIs are rich in X Cl = Cl/(F + Cl + OH) < 0.1} and Ce relative to U (Ce N /U N > 3), and are clearly distinguished from apatites in the surrounding kelyphite (X F < 0.1, X Cl > 0.5; Ce N /U N < 1) (Figs 4g, 5a). Apatites in the MSIs show high REE contents, with fractionated REE patterns (Ce N /Yb N ≈ 100) (Fig. 5a). Phlogopite grains in MSIs are strongly deficient in Na ( < 0.03 cations per 23 oxygens) and depleted in Pb and Sr, whereas phlogopite grains in both the matrix and proximal kelyphite are rich in Na (>0.1 cations per 23 oxygens), Pb and Sr (Figs 4h, 5b). The observed large compositional gaps between apatites and phlogopites in MSIs, compared with those in the matrix, suggest that the MSIs formed under different physical and chemical conditions to the matrix minerals.
The bulk chemical compositions of MSIs in Type A and Type B peridotites have been reconstructed using the mineral proportions (vol.% in Supplementary Table S3) Table S1). This chemical composition is similar to that of Group II kimberlites 25 and suggests that the carbonated potassic silicate melts (hereafter referred to as CPS) formed in equilibrium with a garnet-harzburgite residue at temperatures above the carbonated peridotite solidus (i.e. T > 1200 °C 25 ). Although the alkali contents in the CPS melts are much higher than those of melts in the peridotite-CO 2 system 26 , the alkali contents are much lower than those of melt inclusions within diamond 27 and sediment-derived carbonatites 28,29 . CPS melts with K 2 O contents of ~5 wt.%, similar to those measured in our samples, have been produced experimentally as partial melts in the peridotite-CO 2 system doped with 1.5 wt.% K 2 O 30 (Fig. 6). Therefore, a source for the CPS melt may have been mantle peridotite that was metasomatized by potassic melts. As shown in Fig. 1, the CPS melt could have been stable within the Plešovice peridotite under the conditions of Stage 1 (~1300 °C and 3 GPa), under which chromite is stable in garnet harzburgite 31 (Fig. 1). Upon cooling, the rock would have crossed reaction curve (1) at ~1220 °C 32 , at which point the melt would have changed to a calcio-carbonatite. We hypothesize that CPS melt crystallizes to form roughly equal proportions of carbonatite melt and phlogopite, resulting in production of a melt with the following composition: MgO = 12 wt.%, CaO = 35 wt.%, CO 2 = 47 wt.%, and SiO 2 < 1 wt.% (Supplementary Table S1).  Table S2) in combination with appropriate partitioning coefficients for these minerals and carbonatite melts 33,34 . For comparison, the trace element patterns of a sediment-derived potassic carbonatite 35  The trace element compositions of the CPS melts were calculated using the compositions of apatite and phlogopite (Supplementary Table S2) using the following relation: C i (melt) = D i (melt/crystal) × C i (crystal), where D i represents the partition coefficient of element i 33,34 (see Methods). The calculated trace element pattern for the CPS melt is similar to those of sediment-derived carbonatites 35 (Fig. 5c), showing significant enrichment in LILE such as Rb, Ba, Th, and U, and light REE. However, the quantity of HFSE in the CPS melt is much lower than in sediment-derived carbonatites 35 (Fig. 5c). Fractional crystallization of phlogopite transformed the CPS melt to a carbonatite, whose trace element pattern is represented by the blue line in Fig. 5d. This composition closely resembles that of average calcio-carbonatite lavas worldwide 36 (Fig. 5d). Subsequently, the calcio-carbonatite was likely transformed into a carbonatitic fluid through the reaction of carbonatite and enstatite to diopside, forsterite and CO 2 37 .

Equilibrium modelling of REE between minerals and melts
Here we aim to decipher the metasomatic signatures recorded in silicate minerals that were stable under the conditions of Stages 1 and 2, and we examined whether the CPS melt and carbonatite were in equilibrium with the residual silicate minerals. Firstly, orthopyroxene and olivine in Type A peridotite have high Mg# [Mg/ (Mg + Fe 2+ )] (Mg# Opx ≈ 92 in Table S1; Mg# Ol ≈ 91.5 7 ). Due to their relative insensitivity to changes in T and P, the Mg# of orthopyroxene and olivine serve as indicators of melt depletion 19 . At ~3 GPa, an olivine Mg# of 91.5 equates broadly to the point of complete clinopyroxene consumption at ~25% partial melting 19,38 . Modelling that considers batch melting has shown that orthopyroxene in a residual peridotite acquires a REE pattern with a high HREE/LREE ratio 19 . Therefore, the observed high LREE content in Opx 1 cannot be formed as a result of partial melting events, but it requires the metasomatic addition of LREE after partial melting. Here we examine the hypothesis that a CPS melt coexisted with Opx 1 during Stage 1. To test this hypothesis, we calculate REE patterns of a hypothetical melt that coexisted with Opx 1 , utilizing partitioning coefficients determined by an experiment at 1300 °C and 6 GPa (Run#: M355 39 ). As shown in Fig. 7a, the calculated REE pattern of melt coexisting with Opx 1 matches fairly well with that of the CPS melt. This suggests that the CPS melt was a metasomatic agent during Stage 1.
After Stage 1, the CPS melt crystallized phlogopite, was transformed to carbonatite, and then may have been converted into a carbonaceous fluid during cooling to Stage 2. It has been shown that carbonaceous fluids derived from carbonatite have similar trace element compositions to the source carbonatite 40 . Both are strongly enriched in incompatible trace elements and have the potential to significantly alter the trace element signatures of Stage 2 minerals. Grt 2 in Type A peridotite shows sinuous REE patterns (Fig. 7b) that may have been the product of metasomatism by LREE-rich metasomatic agents 41 . Here we attempt to reproduce the sinuous REE patterns in Grt 2 by modelling interaction between a pre-metasomatic garnet 40 and the calculated composition of the carbonatite melt. Firstly, we assume various proportions of melt (0.01, 0.1, 0.5, 2, and 5 wt.%) relative to the mass of pre-metasomatic garnet 40 . To simulate the reaction between carbonatite melt and garnet, we solved following two equations simultaneously, (1) C REE (garnet) + p × C REE (carbonatite) = constant (mass-balance equation), and (2) C REE (garnet)/C REE (carbonatite) = D REE (garnet/carbonatite) 39 , where p represents proportion of melt. This modelling shows that only 0.5-1.0 wt.% melt relative to garnet can produce the observed sinuous REE pattern observed in Grt 2 grains of Type A peridotites (Fig. 7b). As Type A peridotite contains 2 vol.% garnet, the required melt fraction is only about 0.02-0.04 vol.% (the densities of pyrope and carbonatite are assumed to be 3.7 and 2.2 gcm -3 , respectively).

Discussion
We have reconstructed the major and trace element composition of carbonated potassic silicate (CPS) melt inferred to have been present in the mantle wedge during Stage 1 (~1300 °C at 3 GPa). This melt could have been sourced from sedimentary rocks subducted to depths of > 250 km 28 . Infiltration of the sediment-derived carbonatite (i.e. K 2 O > 27 wt.%, CaO = 17 wt.%, CO 2 = 38 wt.%, SiO 2 = 0.3 wt.% in run am-5B 28 ) into the mantle produces a phlogopite-bearing peridotite. Due to the highly reduced conditions in the deep mantle wedge, the carbonatite may have recrystallized to diamond through the reduction of carbon dioxide during Stage 0 12,42 (Fig. 1). Subsequent convective upwelling of the volatile-rich peridotite induced melting beneath the overriding lithosphere by oxidation 42 , producing a CPS melt during Stage 1 (Fig. 1). The low abundances of HFSE in the CPS melts can be explained by the persistence of ilmenite in a residual peridotite to temperatures 100-120 °C above the solidus 30 .
It has been proposed that the P-T paths of mantle wedge peridotites are controlled by corner flow 43 , which can be envisaged as two types ( Supplementary Fig. S5). Models of slab-induced flow in the mantle wedge show that the flow lines are sub-horizontal and directed towards the slab in the uppermost part of the mantle wedge 44 , consistent with the initial P-T paths observed for the Nonsberg 43 and Higashi-Akaishi peridotites 45,46 . However, this model cannot account for initial upward flow, such as that inferred for the Plešovice and Zhimafang peridotites 8 . The volatile contents in the mantle wedge potentially control the style of corner flow 5 , as well as directly influencing the chemical compositions of mantle wedge melts. Mantle wedge peridotites originating in the spinel lherzolite facies record magmatic activity producing high-temperature silicate melts 6,45,46 , whereas those originating in the garnet lherzolite facies preserve volatile-rich melts as MSIs. In the Zhimafang peridotite, an alkali-rich hydrous silicate melt stable at ~5 GPa has been identified based on the study of MSIs (consisting of calcic amphibole, chlorite, phlogopite, and talc) enclosed within primary garnet 11 . Thus, we propose that volatile-rich partial melts can create upward flow in the mantle wedge.
When the CPS melt cooled below 1220 °C it started to crystallise phlogopite. Thus, phlogopite crystallization should have taken place near the base of the lithosphere or a short distance above the thermal boundary layer (~1300-1400 °C) 47 . Recent high-resolution seismic studies have demonstrated a substantial reduction in shear wave velocities (V S ) at depths of 60-150 km in the cratonic lithosphere 48,49 , in the oceanic lithosphere-asthenosphere boundary (LAB) 50,51 and in the mantle wedge 52 . Several hypotheses have been proposed to explain this reduction in V S : 1) the presence of partial melt 50,51 2) elastically accommodated grain boundary sliding 53 and 3) the presence of hydrous minerals produced through metasomatism 48,49 . Phlogopite is seismically slow and can cause a reduction in V S . For example, the formation of 5-7 vol.% phlogopite in the lithosphere can reduce V S by up to 6% 49 . This velocity reduction matches values observed at the oceanic LAB (a 7-9% reduction 50,54 ) and can explain modest velocity reductions observed in the cratonic lithosphere (a 2%-24% reduction 48,49 ).
Recently petrologists have proposed that physical mixing of subducted sediment and mantle wedge peridotites may result in the formation of volatile-rich hybrid rocks in low-temperature subduction zones 55 . Such hybrid rocks could create low-V S regions in the subduction mélange. In warmer subduction/collision zones, subducted sediment would produce potassic carbonatites 28 . Interaction between potassic carbonatites and mantle peridotites results in the creation of phlogopite-bearing carbonated peridotite 5 or a phlogopite-rich metasomatic layer (i.e. glimmerite 56 ). In both cases, blebs of hybrid rocks or phlogopite-bearing carbonated peridotite/glimmerite  18 . The calculated REE pattern of the melt agrees well with that of the reconstructed REE pattern for the MSIs. Panel (b) shows a result of REE modelling of a reaction between pre-metasomatic garnet 40 and a calculated carbonatite melt. The red-coloured curves are calculated patterns for varying volumes of carbonatite melt that reacted with garnet. Garnet-melt partition coefficients are after ref. 39 . Only 0.5-1.0% carbonatite melt relative to garnet is needed to generate the observed sinuous REE pattern in Type A peridotite.
Scientific RePORtS | (2019) 9:10116 | DOI:10.1038/s41598-018-28264-6 will rise buoyantly from the surface of the subducting slab into the sub-arc mantle, producing low seismic anomalies along their ascent pathways 55 . Finally, these diapirs would melt to produce potassic melts. Melting of phlogopite in the mantle wedge generates ultrapotassic magmas with high K 2 O and MgO (>3 wt.%) 57 , such as those observed in the Bohemian massif. Orogenic garnet peridotite in the Bohemian Massif (Austria) contains glimmerite veins 58 as well as CPS melts. Both are key features that give rise to orogenic ultrapotassic magmatism.
In this study, we constrained the composition of mantle wedge melts. High contents of LILE, LREE and carbon in the CPS melts were likely derived from subducted carbonate-rich sediments 28,56 . Although the presence of this type of potassium-and volatile-enriched melt in the asthenosphere has been speculated to exist 59,60 , and is sometimes referred to as "failed kimberlite" 25 , this study is the first to report a natural example from mantle wedge samples. Such CPS melts may be ubiquitous in the deep mantle wedge, but they are rarely present in sufficient quantity to reach the Earth's surface. Due to small melt fractions, the CPS melts are expected to have differentiated to carbonaceous fluid within the lithosphere, with carbon dioxide returned to the surface through arc volcanism 1 .

Methods
Major element analyses were performed using an EPMA (JEOL JXA-8900L at the University of Tokyo) and a FE-EPMA (JEOL JSM-8350F, at the University of Tokyo). Most silicate minerals, carbonates, phlogopite, and phosphates were analysed by EPMA. The analytical conditions were 15 kV, a surface beam current of 12 nA [1-3 µm beam diameter (b.d.)] for silicates and phosphates, and 3 nA (scanning mode with magnification of 20,000) for carbonate with a ZAF correction. The sulphides, priderite, burbankite, Ba-rich carbonates, and monazite were analysed by FE-EPMA. Analytical conditions were 15 kV and a surface beam current of 6 nA (3 μm b.d.) for priderite, 15 kV and 12 nA (1 um b.d.) for sulphides, and 25 kV and 20 nA (10 μm beam diameter) for Ba-rich carbonates, burbankite, and monazite. Oxides, silicate minerals, and natural and synthetic carbonates with known compositions were used as analytical standards.
Trace element analyses were performed on an iCAP Q inductively coupled plasma mass spectrometer (Thermo Fisher Scientific) attached to an LSX-213 G2 + Nd:YAG laser ablation system (Cetac Technologies) at the University of Tokyo. Each analysis consisted of a 25-sec baseline measurement followed by a ~35-sec measurement with laser ablation sampling. The data were obtained from 15-25 µm ablation pits with a laser repetition rate of 4-8 Hz. An external calibration was made against NIST 611 SRM, and an internal standard calibration was applied using the Ca (apatite and garnet), Mn (chromite and orthopyroxene) and K (phlogopite) concentrations determined by EPMA.
Trace element concentrations in melts were calculated using those in apatite and phlogopite in multiphase solid inclusions (MSIs). According to the phase diagram and crystal shapes, the first mineral to crystallize from parental melt is phlogopite followed by apatite. The phlogopite crystallization reaction is as follows: Melt → 0.474 residual melt-1 + 0.526 phlogopite (wt.% basis). The concentrations of Na, Rb, Cs, Ba, Ta, Nb, Hf, Zr and Ni in a residual melt-1 were calculated using those in phlogopite grains in conjunction with trace element partitioning coefficients between phlogopite and carbonatite 34 . The apatite crystallization reaction is as follows: Melt → 0.442 residual melt-2 + 0.526 phlogopite + 0.032 apatite. Some trace elements (Pb, Th, U, and REE) in the residual melt (melt-2) were estimated using those in the apatite grains in combination with trace element partitioning coefficients between apatite and carbonatite 33 .
Partitioning coefficients of Rb, Ba, Nb, and Zr between phlogopite and carbonatite are from ref. 34 , and those of Cs and Ta are from ref. 61 . The partitioning coefficients of Hf was assumed to be 0.19 62 , which agrees with the average value in the literature 63 . In order to reduce the error of partitioning coefficients between apatite and carbonatite 33 , we calculated the standard error by dividing σ by the square root of the number of data (n = 60) (i.e. σ/√n) 33 .
The calibration of Zn in chromite geothermometer of ref. 17 {T = h (Zn ppm)} adopted the temperature based on the Ni-in-garnet geothermometer of ref. 17 {T = f (Ni ppm)}, which has been shown to be inconsistent with the calibration proposed by ref. 64 {T = g (Ni ppm)}. Therefore, we corrected Zn-in-chromite thermometer as follows: T = g {f −1 {h (Zn ppm)}} to ensure that it is consistent with the calibration proposed by ref. 64 .