Carbonate formation temperatures were calculated based on the O isotopic compositions of carbonates and of the solutions from which these precipitated. For the carbonates formed within the subduction channel, we used the calibrations from O'Neil et al. (1969) and Friedman and O'Neil (1977). The latter was as well used for the single calcite sample that was recovered from Asùt Tesoru, whereas the calibration curve from Grossman and Ku (1986) was used for authigenic aragonite formed within the Yinazao mud volcano.

Host rocks of the carbonate ± silicate ± sulfate veins are ultramafic, metavolcanic, and metasedimentary clasts. Metavolcanic and metasedimentary clasts originate exclusively from the deeper-sourced mud volcanoes Fantangisña and Asùt Tesoru, whereas ultramafic hosts were recovered at Asùt Tesoru and the shallow-sourced Yinazao (Table 1).

Studied peridotites include two samples of dunite and one sample of harzburgite. One of the dunites is about 30 % serpentinized, whereas the other peridotites are fully serpentinized and characterized by mesh textures and distinct generations of crosscutting serpentine ± brucite ± magnetite veins (Fig. S1 in the Supplement). The harzburgite contains hydroandradite intergrown with serpentine in bastite texture after pyroxene (Fig. S1b) and scarce late veins that contain carbonate phases.

Seven variably metamorphosed volcanic rocks (Fig. 2a; Table 1) contain relict primary minerals including lath-shaped plagioclase and titaniferous clinopyroxene (Fig. 2a–c). The Ti-rich pyroxenes mostly plot in the field of alkali basalt, as classified by Leterrier et al. (1982; Fig. 2e). The primary mineral assemblages were largely altered to prehnite–pumpellyite and blueschist facies mineral assemblages, including chlorite, prehnite, pumpellyite, glaucophane, lawsonite, titanite, and vesuvianite (representative analyses available in Table S1 in the Supplement). These samples are characterized by complex networks of carbonate–silicate veins.

The metasedimentary quartzite is brecciated into smaller fragments in a fine-grained matrix that is rich in calcite and hematite (Fig. 2d). Fine calcite veins crosscut the quartz fragments; the breccia is crosscut by a centimeter-wide vein of aragonite, calcite, and barite.

The serpentinized peridotite clasts represent forearc mantle material. In contrast, the likely origin of the metavolcanic and metasedimentary rocks is the upper oceanic crust and overlying deep marine sediments of the subducting plate. The presence of remnant Ti-rich pyroxene implies an ocean island provenance for the metavolcanic rocks (Fig. 2; Leterrier et al., 1982), in agreement with petrographic observations and geochemical data from IODP Expedition 366 and demonstrating the recycling of Pacific Plate seamount material via serpentinite mud volcanism (Fryer et al., 2018b, 2019). Secondary phases within the metavolcanic clasts illustrate a high-pressure metamorphic overprint in accordance with estimated peak pressure conditions at depth beneath Fantangisña (slab depth ∼14 km) and Asùt Tesoru (∼18 km; Oakley et al., 2007, 2008; Hulme et al., 2010). Glaucophane and lawsonite, present in samples from Asùt Tesoru, indicate beginning blueschist metamorphic conditions as observed by earlier studies of similar blueschist materials from the serpentinite mud volcanoes (e.g., Maekawa et al., 1995; Fryer et al., 1999).

Quartzite breccias have not been previously described from the serpentinite mud volcanoes. We suggest that this sample represents lithified sedimentary siliceous ooze from the subducted Pacific Plate. Siliceous oozes cover large areas of the western Pacific seafloor (cf. ODP Site 801 and Site 1149; Shipboard Scientific Party, 2000), as abyssal depths in many places are below the carbonate compensation depth (∼4.5 km), leading to dissolution of the otherwise widespread calcareous plankton (e.g., Pälike et al., 2012). With progressive metamorphism during subduction, the silica-rich sediments were likely transformed into quartzite (cf. Kastner et al., 1981). The brecciation indicates the exposure of the quartzite to stresses at temperatures below that of the brittle–plastic transition zone (< 300–350 ∘C; e.g., Stipp and Kunze, 2008), which is in agreement with the estimated temperature of 150 ∘C for the subduction channel below Fantangisña (Hulme et al., 2010).

For all veins, geochemical data and textural relationships of the vein assemblages demonstrate that the precipitates formed (i) within the subduction zone and (ii) after deformation of the clasts and their eventual detachment from the forearc mantle and subducting slab, respectively. (i) The carbonate REE patterns differ substantially from those of carbonates formed during seawater circulation within mafic portions of the ocean crust where the formation of Ca carbonate veins is a common process (e.g., Alt and Teagle, 1999). Rausch (2012) studied carbonate veins within the upper oceanic crust from the west Pacific (Site 801 and Site 1149, ODP Leg 185) in host lithologies comparable to those subducted at the Izu–Bonin–Mariana convergent margin (Fig. 4). These carbonates exhibit higher REE concentrations than the carbonates in this study and enrichments in HREEs compared to our deep carbonates, which are characterized by significant depletions in HREEs and commonly with negative Ce anomalies (Fig. 4). Only one sample (U1498B-8R-1, 10–15 cm) shows a slight HREE enrichment similar to Pacific crust carbonates, yet total REE concentrations are considerably lower. Hence, we rule out that the deep carbonates, namely those carbonates hosted in veins by metavolcanic and metasedimentary clasts, formed during alteration of the ocean crust before subduction. In contrast, vein carbonates in ultramafic clasts do exhibit REE concentrations and patterns similar to those of carbonates from Site 801 and Site 1149 (Fig. 4). But these veins are hosted by peridotite clasts from the forearc mantle, and hence they must have formed within the subduction zone. Further credence to the formation within the forearc is lent by the Sr isotopic signatures of the carbonates: their 87Sr/86Sr compositions are similar to the most unradiogenic values of the slab fluids (Fig. 5). These unradiogenic signatures represent the pore waters least influenced by mixing with seawater, and we interpret them to reflect the endmember compositions. The similar Sr isotopic composition of the endmember fluids and the deep carbonates imply their likely precipitation from such slab-derived fluids at depth. (ii) We further base this interpretation on textural evidence including the radial growth (e.g., Fig. 3c) and euhedral crystal shapes (e.g., Fig. 3d) of vein precipitates associated with the deep carbonates that indicate growth in stress-free environments, probably subsequent to deformation/brecciation. Their host rocks show evidence of high-pressure/low-temperature metamorphism along a cold geotherm as brittle deformation and recrystallization to pumpellyite or glaucophane. In contrast, no such indications of metamorphism were identified for the deep carbonates or associated silicates and/or barite.

We conclude that the vein assemblages including the deep carbonates have precipitated from fluids subsequent to the high-strain conditions that affected their metavolcanic and metasedimentary hosts. The veins and cavities that now host the deep carbonates can either have formed before their host rocks were detached from the slab, for instance by high pore fluid pressures resulting from ongoing fluid release (cf. Hacker et al., 2003; Taetz et al., 2018), or during breakup of the clasts within the subduction channel. During and after vein assemblage formation, relatively deviatoric stress-free environments must have prevailed to form and preserve the observed undeformed mineral textures. Nevertheless, pressure–temperature conditions during vein precipitate formation were likely similar to the peak metamorphic conditions under which their host rocks recrystallized, as indicated by the presence of aragonite and its intergrowth with lawsonite and phengite (cf. Table 1). Aragonite itself is restricted to high-pressure systems (e.g., Johannes and Puhan, 1971). Lawsonite is characteristic of metamorphic overprint of mafic rocks under blueschist facies conditions (e.g., Black, 1977) at high pressures and temperatures from 150 to 400 ∘C (Essene et al., 1965; Diessel et al., 1978). Intimate intergrowth of lawsonite with aragonite (Fig. 2a) hence clearly implies that these veins formed at great depth. Furthermore, the uniform REE + Y patterns as well as Sr isotopic compositions of those deep veins suggest that they have precipitated from compositionally alike fluids with a Sr isotope signature of the invoked pore water endmember.

Temperatures of formation are difficult to constrain from metamorphic assemblages in subduction zones, but carbonate formation temperatures can be calculated from their δ18O values if the O isotopic composition of the precipitating solution is known. Pore waters of the most deeply rooted mud volcano (Asùt Tesoru) have δ18Ofluid values of +1.7 ‰ to +2 ‰ (Debret et al., 2019); however, it is unlikely that these fluids retained their O isotopic composition during their ascent from the decollement. Temperature estimates based on these isotopic compositions would range from ∼60 to 95 ∘C for deep carbonates (Fig. 7; O'Neil et al., 1969; Friedman and O'Neil, 1977). These values are much lower than the temperatures of 150 to 250 ∘C estimated for the depth of the slab interface beneath the serpentinite mud volcanoes (Hulme et al., 2010), and it seems unlikely that the δ18O values of the pore waters of the mud volcanoes are representative of the slab-derived fluids from which the carbonates precipitated. The precipitation of the deep carbonates presumably occurred at (considerably) higher temperatures. Alt and Shanks III (2006) calculated isotopic compositions of fluids released at depth that are in equilibrium with the weighted mean of subducting sediment (their Table 4). These authors estimated the δ18O value of such fluid at 12 ‰ at 200 ∘C and 16 ‰ at 300 ∘C. Adopting their approach, we calculated δ18Ofluid in equilibrium with subducting sediment based on fractionation coefficients for calcite–water, quartz–water, albite–water, illite–water, and chlorite–water from Zheng (1993a, b, 1999). For the inferred slab temperatures of 150 ∘C below Fantangisña and of 250 ∘C below Asùt Tesoru we obtained δ18Ofluid compositions of 9.3 ‰ and 14.9 ‰, respectively. If the deep carbonates formed in equilibrium with these fluids, estimated precipitation temperatures are shifted to 130 to 175 ∘C and to 220 to 310 ∘C, respectively (Fig. 7). These estimates overlap with those for carbonates hosted in metavolcanic and metasedimentary clasts from the Conical seamount that have similar O isotopic compositions (δ18O = 16.7 ‰–24.5 ‰; Fig. 6) and are interpreted to have formed from slab-derived fluids at temperatures of up to 250 ∘C (Alt and Shanks III, 2006). Fluids expelled from sediments at shallow depths in subduction zones may generally have high δ18O values. For instance, δ18O of ∼13 ‰ is estimated for slab-derived fluids from 25 to 45 km depth in the Catalina Schist subduction complex (Bebout and Barton, 1989).

However, incipient serpentinization in the subduction channel is likely to remove 18O from the fluids into serpentine (e.g., Alt and Shanks III, 2006; Debret et al., 2019). Consequently, in the subduction channel, which likely contains eroded variably serpentinized materials from the basal plane of the mantle wedge, the actual δ18Ofluid may be somewhat lower. The temperatures estimated on the basis of calculated δ18Ofluid values should therefore be considered maximum estimates. But besides that, the estimated formation temperatures of 130 to 310 ∘C are well in line with the presence of pectolite and prehnite associated with the deep carbonates in some of the veins. These phases require formation temperatures that well fit the above-proposed temperatures: pectolite may form at temperatures as high as 300 to 350 ∘C (e.g., Craw and Landis, 1980; Esteban et al., 2003), and prehnite may be stable in water–rock systems up to temperatures of ∼300 ∘C (e.g., Bird and Spieler, 2004). Pectolite and prehnite are often associated with rodingitization (e.g., Austrheim and Prestvik, 2008), a process in which crustal rocks interacting with serpentinization fluids undergo Ca metasomatism including the formation of Ca–Al silicates, such as hydrogarnet, prehnite, and vesuvianite (see, e.g., Bach and Klein, 2009, for details).

We suggest that the deep carbonates formed from dehydration fluids in the subduction channel close to where they were released from the downgoing slab. This idea is supported by the trace element composition of the deep carbonates, specifically the generally high concentrations of REEs and negative Ce anomalies. The low REE mobility and oxidizing conditions indicated by these REE compositions do not support the idea that the fluids had been modified significantly by serpentinization reactions taking place where peridotitic clasts dominate the subduction channel mélange.

Furthermore, prevailing serpentinization within the subduction channel would produce reducing conditions due to the release of H2 to the aqueous solutions (e.g., McCollom and Bach, 2009). But the pronounced negative Ce anomalies indicate that carbonate precipitation took place under fairly oxidizing conditions where Ce was tetravalent (e.g., German et al., 1991). Cerium is the only REE that can exist in the trivalent and tetravalent oxidation state; in aqueous solutions, tetravalent Ce is less soluble than trivalent Ce, resulting in its precipitation and the removal of Ce from the fluid and the creation of the observed negative Ce anomalies. Cerium anomalies may hence be used as a redox proxy (e.g., German and Elderfield, 1990; Tostevin et al., 2016). We calculated the predominant redox states of Ce and Eu in the aH2(aq) vs. temperature plane to identify the low reduction potential under which the carbonates have formed (Fig. 8; see figure caption for details on the calculations). Within the estimated temperature range of ∼130 to 310 ∘C, tetravalent Ce will be stable under fairly oxidizing conditions. Since not all deep carbonates exhibit Ce anomalies, the conditions presumably shifted towards slightly more reducing conditions, i.e., approaching the field of trivalent Ce stability (Fig. 8). Yet, the redox conditions under which tetravalent Ce is expected to predominate are clearly within the stability region of SO42- and CO2 (Fig. 8). This result is consistent with the abundance of carbonate veins and the presence of barite in the quartzite sample. Our estimate puts the upper boundary of H2 activity in the fluids well below levels imposed by serpentinization reactions, which would be > log -2 for temperatures between 200 and 300 ∘C (McCollom and Bach, 2009). Under such reducing conditions, CO2 is no longer the predominant C species (Fig. 8) but is instead reduced to CH4 by H2 produced during serpentinization. Calcium carbonate formation, for instance in ultramafic/serpentinite clasts, is hence less likely to occur in such conditions. These findings lend further credence to the hypothesis that the carbonate precipitation occurred in an environment that was not dominated by serpentinization reactions. It is hence likely that the deep carbonates formed within the subduction channel, where serpentinization of the mantle wedge has only a minor influence on the overall redox conditions. Instead, dehydration processes induced by prograde reactions in the subducting slab are thought to drive more oxic conditions. These processes include the release of pore waters due to sediment compaction, dehydration of opal-CT, decomposition of clay, and decarbonation (e.g., Mottl et al., 2004; Fryer et al., 2018b). On the basis of the textural relationships implying strain-free environments, we can speculate that the deep carbonate precipitation occurred after the metavolcanics/metasediments were detached from the subducting lithosphere (Fig. 9). Our isotope-based temperature estimates are in line with this interpretation as they well match inferred slab temperatures of ∼150 ∘C beneath Fantangisña and ∼250 ∘C beneath Asùt Tesoru.

Within the mantle wedge, serpentinization is expected to become a dominant process and the redox conditions will become reducing, as indicated by the presence of magnetite and awaruite in the serpentinite clasts (Kahl et al., 2015). These reducing conditions prevail in the mantle wedge and mud volcano conduits: Kawagucci et al. (2018) suggested that the fluids venting at the South Chamorro mud volcano had as much as 40 mM H2 before most of the H2 was consumed by microbial sulfate reduction. We hence suggest that there is a redox gradient across the subduction channel set up by the release of oxidized CO2 and SO42--bearing fluids from the slab that become progressively more reducing by ongoing serpentinization reactions of wedge peridotite in the upper part of the subduction channel where peridotite dominates volumetrically over metasediment and/or oceanic crust (Fig. 9).

The veined metavolcanic clasts were exposed to these reducing conditions imposed by serpentinization reactions, but the carbonate phases preserved a record of earlier oxidizing conditions in the subduction channel. However, evidence for late-stage reactions between the metavolcanic clasts and serpentinization fluids exists in the form of vein phases such as nesquehonite, Mg(HCO3)(OH) ⚫ 2(H2O), present in few clasts (cf. Table 1). This phase is stable only at temperatures of < 100 ∘C (Davies and Bubela, 1973; Suzuki and Ito, 1974; Ballirano et al., 2010) and forms exclusively in alkaline environments (Wilson et al., 2009). Thaumasite, Ca3Si(CO3)(SO4)(OH)6 ⚫ 12(H2O), another low-temperature alteration phase in mafic rocks, forms preferentially in the presence of carbonate and sulfate at low temperatures and under alkaline conditions (e.g., Karpoff et al., 1992; Schmidt et al., 2008). Within the marine realm, the formation of thaumasite occurs at temperatures of < 100 ∘C (e.g., Noack, 1983; Karpoff et al., 1992). Nesquehonite and thaumasite are not directly intergrown with the deep carbonates but rather occur in veins that in some cases crosscut the deep carbonates. They likely formed during low-temperature alteration of the clasts below or within the mud volcanoes.

The precipitation of shallow carbonates within the mud volcanoes took place under conditions drastically different from the deep carbonate formation, as indicated by their contrasting O isotope compositions (Fig. 6, Table 3). Radiating clusters of aragonite at Yinazao have exceptionally high δ18O values (∼36.6 ‰), which leads to unrealistically low calculated formation temperatures (following Grossman and Ku, 1986) if the lowest δ18Ofluid value (-1.5 ‰; Debret et al., 2019) is assumed. Temperatures calculated for the highest values of δ18Ofluid at Yinazao (+0.6 ‰) would be just below 1 ∘C. Temperatures of 3.0 and 15.6 ∘C are computed for the calcite crystal (δ18O = 33 ‰) recovered from Asùt Tesoru (using Friedmann and O'Neil, 1977) and the range of δ18Ofluid values of pore water (-0.2 ‰ to +2.5 ‰; Debret et al., 2019). These estimates imply a formation of the shallow carbonates at low temperatures, which is in accordance with results of earlier work (Gharib, 2006; Tran et al., 2014).

The very low REE + Y contents of authigenic aragonite and calcite and of carbonate hosted in veins of the serpentinite clasts and one metavolcanic clast may point to a fluid source distinct from the slab-derived fluids. Low total REE + Y concentrations would be plausible if seawater was mixed in the shallow muds of the serpentinite mud volcanoes (REE concentrations in NW Pacific bottom water are < 60 pM; e.g., Piepgras and Jacobsen, 1992). The co-occurrence of iowaite with the shallow carbonates implies oxic conditions, further strengthening our conclusion of the presence of seawater (cf. Heling and Schwarz, 1992; Bach et al., 2004). Seawater would as well provide Cl- during the formation of iowaite. Yet, this interpretation is inconsistent with the similar 87Sr/86Sr isotopic compositions of the deep and shallow carbonates, indicating that entrainment of seawater in the shallow parts of the system was minimal (no apparent seawater fraction for two out of three authigenic carbonate samples; cf. Fig. 6). The lower REE + Y contents of the shallow carbonates may hence be more plausibly explained by (i) a loss of the REEs en route to the surface (cf. Kahl et al., 2015; see also next section) or (ii) lower precipitation temperatures that will favor crystal-chemical discrimination against REEs.

If the pore waters from the mud volcanoes indeed represent cooled slab-derived fluids, then the oxygen isotopic composition must have been considerably changed from high initial δ18O values to overall seawater-like compositions near the seafloor. This shift in δ18O may be caused by isotopic exchange in the mud between water and serpentine, which has higher δ18O values than mantle wedge peridotites (e.g., Alt and Shanks III, 2006; Debret et al., 2019) and may hence explain the postulated decrease in δ18O values in the interacting fluids. The magnitude of the shift would indicate very low water-to-rock ratios.

Our data provide important insights into understanding the transfer of carbonic fluids from the subducting slab and reactions of these fluids with rocks in the subduction channel. The data show that C is mobilized from the downgoing slab at subduction depths as shallow as ∼14 km and lower blueschist facies temperatures within the Mariana subduction system (∼150 ∘C). The mobilization of C at these low metamorphic grades is in accordance with experimental evidence (Caciagli and Manning, 2003) that shows that calcite solubility increases with pressure and temperature at pressures ≥0.3 GPa. Some of the C mobilized subsequently rises through the forearc with upwelling fluids and is either incorporated into carbonate precipitates within the serpentinite mud volcanoes or reduced to methane and transported back to the seafloor by serpentinite mud volcano activity (Mottl et al., 2004; Hulme et al., 2010; Fryer et al., 2018b). Associated with the redox gradient within the subduction channel from oxidizing conditions at the slab interface to more reducing conditions at the interface with the mantle wedge, a change in the C oxidation state from carbonate to methane is likely. In the model we propose (Fig. 9), metavolcanic and metasediment clasts react with slab-derived carbonic fluids and form carbonates in fractures and voids. DIC contents are likely lowered by (1) removal in CaCO3 veins and (2) partial reduction to CH4 by H2 produced during serpentinization. Both mechanisms would help explain why deep carbonate veins were not observed in serpentinite clasts.

The abundance of carbonate veins in clasts from mud volcanoes with slab depths of ≥14 km shows that a large portion of C may be trapped by carbonate mineralization within the subduction channel. The REE + Y characteristics of these carbonates indicate that the redox conditions within the subduction channel are oxic and within the stability field of DIC (and not methane). Carbon isotope compositions (δ13C = -0.6 ‰ to +3.3 ‰) also show that CO2 reduction, which would deplete 12C from the DIC pool, has not occurred. Carbonate veins in clasts of metasediment were also recovered from Conical (∼19 km depth to slab; Shipboard Science Party, 1990); these carbonate minerals have C isotopic compositions (δ13C = +0.07 ‰ to +2.5 ‰; Alt and Shanks III, 2006) similar to the carbonates studied here. The presence of these carbonates shows that the mobilization of C continues to greater depths. The bulk of these carbonates presumably is not entrained into the serpentinite mud volcano conduits and may remain in the forearc mantle or be subducted deeper.

Based on the pore water compositions of the Mariana serpentinite mud volcanoes, it has previously been suggested that the decarbonation of the subducted slab initiates at depths of < 30 km (e.g., Mottl et al., 2004; Hulme et al., 2010; Fryer et al., 2018b). Our examination of the deep carbonate precipitates formed within the subduction channel provides information on geochemical conditions at depths of ∼14 to 18 km and potential C sources. Samples from these shallow slab depths have not previously been described, and hence shallow decarbonation processes have not been assessed. Geochemical modeling, experiments, and field evidence have focused on potential C mobilization at greater depths and temperatures of > 300 ∘C (e.g., Molina and Poli, 2000; Kerrick and Connolly, 2001; Newton and Manning, 2002; Dasgupta et al., 2004; Gorman et al., 2006; Frezzotti et al., 2011; Ague and Nicolescu, 2014; Sverjensky et al., 2014; Piccoli et al., 2016; Schwarzenbach et al., 2018; Sieber et al., 2018). Budgets for the deep carbon cycle are hence plagued by very large uncertainties (e.g., Dasgupta and Hierschmann, 2010; Füri et al., 2010; Kelemen and Manning, 2015).

Reservoirs of C as well as its mobilization and fluxes in shallow regions of subduction zones are important components of Earth's carbon cycle. Our data are in line with the suggestion that simple devolatilization reactions do not account for the mobilization of C within these shallow portions of the subduction system (e.g., Kerrick and Connolly, 2001). Instead, we suggest that fluid-induced dissolution of carbonate (e.g., Poli et al., 2009) is the more likely process that leads to the release of C under shallow forearc conditions. Kelemen and Manning (2015) estimated the global C flux into the leading edge of the mantle wedge to > 1 Mt a-1. This budget includes C loss due to venting of carbonic fluids as it occurs in the Mariana forearc but does not cover the solid storage of C within the shallow portions of subduction systems. Such solid storage may sequester large amounts of the C released from the slab (e.g., > 90 % in the Costa Rican forearc; Barry et al., 2019). Our study highlights that additional work is needed to better understand the shallow processes that affect the cycling of C.