Tracing the subducting Pacific slab to the mantle transition zone with hydrogen isotopes

Hydrogen isotopes have been widely used as powerful tracers to understand the origin of terrestrial water and the water circulation between the surface and the deep interior of the Earth. However, further quantitative understanding is hindered due to a lack of observations about the changes in D/H ratios of a slab during subduction. Here, we report hydrogen isotope data of olivine-hosted melt inclusions from active volcanoes with variable depths (90‒550 km) to the subducting Pacific slab. The results show that the D/H ratio of the slab fluid at the volcanic front is lower than that of the slab fluid just behind the volcanic front. This demonstrates that fluids with different D/H ratios were released from the crust and the underlying peridotite portions of the slab around the volcanic front. The results also show that the D/H ratios of slab fluids do not change significantly with slab depths from 300 to 550 km, which demonstrates that slab dehydration did not occur significantly beyond the arc. Our estimated δD‰ value for the slab materials that accumulated in the mantle transition zone is > − 90‰, a value which is significantly higher than previous estimates.

www.nature.com/scientificreports/ hydrogen isotopic data on samples of such small sizes. Furthermore, D/H data of the melt inclusions should be used carefully, because the ratios do not necessarily represent those of the original magma due to degassing and post-entrapment modification of the melt inclusion compositions by shallow-level processes 9,11,16,17 .
In this study, to trace the evolution of the D/H ratios of the subducting Pacific slab from the volcanic front to the mantle transition zone (MTZ), the hydrogen isotopic compositions of olivine-hosted melt inclusions were determined for basaltic scoria samples from six active volcanoes, including Iwate (Z = ~ 90 km), Akita-Komagatake (Z = ~ 100 km), Me-Akan (Z = ~ 110 km), Oshima-Oshima (Z = ~ 180 km), Rishiri (Z = ~ 300 km), and Fukue (Z = ~ 550 km) (Fig. 1). The Iwate, Akita-Komagatake, and Oshima-Oshima volcanoes belong to the NE Japan arc, and the Me-Akan and Rishiri volcanoes belong to the Kuril arc. Fukue is an intraplate-type volcano, beneath which the subducted Pacific Plate is stagnated in the MTZ (Fig. 1). The MTZ here is considerably hydrous, as shown by electrical conductivity observations 18 . The influx of fluids derived from the stagnant Pacific slab generates primary magmas from the melting of the asthenospheric mantle at the Fukue volcano 19 . Therefore, the samples from the five arc volcanoes and the Fukue volcano provide information about the variations in the D/H ratios of fluids released from the cold subducting Pacific slab at depths from ~ 90 to ~ 550 km. Notably, the slab depth of ~ 300 km for Rishiri is one of the deepest among arc volcanoes globally, and Fukue is one of the rare active volcanoes that originated from the magmatism rooted in the MTZ.
Numerous scoria samples were collected from various tephra fall deposits at different eruption stages for each of the six volcanoes, and the deposits containing scoria with the highest-quality olivine-hosted melt inclusions were carefully selected. The melt inclusion-bearing olivine crystals were then separated from the scoria samples collected from the selected deposits. To carefully examine the effect of the post-entrapment modification of melt inclusions, volatile content analyses were conducted on at least 25

Results and discussion
The descriptions of the scoria samples used in this study are provided in Supplementary Methods. The samples exhibit basaltic compositions (Supplementary Table S1) and contain olivine as the primary mafic phenocryst phase. The samples from the Iwate, Akita-Komagatake, Me-Akan, Oshima-Oshima, and Rishiri volcanoes show trace element concentration patterns that are characteristic of arc basalt. In contrast, the trace element concentration pattern for the samples from the Fukue volcano has an affinity to intra-plate basalt ( Supplementary Fig. S1). The Pb isotopic compositions of all the samples lie within the triangular field defined by three subduction zone components: sediment, altered oceanic crust, and depleted mantle ( Supplementary Fig. S2).
Volatile (H 2 O, CO 2 , F, S, and Cl) and P 2 O 5 contents and hydrogen isotopic compositions were analysed on melt inclusions using the ion microprobe (Cameca IMS-1280HR, Ametek Cameca) at the Kochi Institute for Core Sample Research, JAMSTEC. As most measured melt inclusions were glassy without daughter minerals ( Supplementary Fig. S3), we did not homogenise the inclusions at high temperatures. The details of the analytical procedures, including sample preparation, analytical conditions, and standard data, are provided in Refs. 20,21 and "Methods". The volatile and major elemental compositions and the δD‰ values of the melt inclusions are listed in Supplementary Table S2. www.nature.com/scientificreports/ It is well established that the D/H ratios of olivine-hosted melt inclusions do not necessarily represent those of mantle-derived magmas 9,11,16,17 . Mantle-derived magma is not saturated with volatiles at deep levels because volatiles are more soluble in melt at higher pressures. During the ascent of the magma in the crust, the magma becomes saturated with volatiles and vesiculation occurs. Degassing of the magma can cause fractionation of the D/H ratio 22 because the D/H ratio of an H 2 O-rich vapor phase is higher than that of the melt phase. In addition, during residence of the magma in a crustal magma chamber, the D/H ratio of the magma may be modified by crustal assimilation. Because of these, the D/H ratios of the melt inclusions entrapped in olivine phenocrysts after magma degassing and/or crustal assimilation can be significantly different from those of the original magma. As the magma continues to further ascend, the H 2 O content of the melt progressively decreases. The D/H ratios of olivine-hosted melt inclusions that were entrapped before the magma degassing can be modified by the reequilibration of the olivine with the surrounding melt because of faster diffusion of H relative to D through the host olivine 16,17 . For these reasons, to extract the information about the mantle processes from melt inclusion data, it is necessary to carefully evaluate the effects of shallow-level processes on the D/H ratios.
The volatile content data are shown in CO 2 -H 2 O diagrams for all the melt inclusions from the individual volcanoes in Supplementary Figs. S4-S9. At all the volcanoes, the melt inclusions have variable H 2 O contents, and those with relatively higher H 2 O contents have highly variable CO 2 contents at a given H 2 O content. Some melt inclusions have shrinkage bubbles, and, therefore, the original CO 2 contents would have been even higher 17 . CO 2 is the first compound to be lost by degassing during magma ascent from the upper mantle because it is the least soluble species in silicate melt among the major volatile elements 23 . Therefore, the marked variation in CO 2 content at a given H 2 O content suggests that the melt with high CO 2 and high H 2 O contents experienced minimal loss of H 2 O by degassing before the entrapment 24 . In contrast, the low CO 2 and low H 2 O contents of some melt inclusions may be explained by the entrapment of melts that were degassed during magma ascent (i.e. open-system and closed-system degassing trends in Supplementary Fig. S6a). The high CO 2 melt inclusions trapped at high pressures might have experienced re-equilibration with lower H 2 O melt through the host olivine at lower pressures, as explained above. δD‰ values of the melt inclusions are expected to correlate negatively with the H 2 O contents when a diffusive loss of H 2 O occurs significantly from the melt inclusions 16,17 . At all the volcanoes, some melt inclusions with lower H 2 O contents tend to have higher δD‰ values, and they might have been affected by diffusive re-equilibration. However, the δD‰ values and the H 2 O content of CO 2 -rich melt inclusions do not exhibit a significant negative correlation, suggesting that CO 2 -rich melt inclusions did not experience significant diffusive re-equilibration. This is consistent with the observation that the δD‰ values of the high CO 2 melt inclusions are essentially constant irrespective of the CO 2 contents (Supplementary Figs. S4-S9). Based on these evaluations, the data of the melt inclusions with the three highest CO 2 contents were selected as the representative data sets for each volcano.
Of the remaining melt inclusions, some inclusions with high H 2 O and lower CO 2 contents may also preserve the original D/H ratios, because they experienced minimal loss of H 2 O by degassing from the original magma with high CO 2 and high H 2 O contents (see the degassing trends in Supplementary Fig. S6a). However, these inclusions were entrapped at lower pressures than those with high CO 2 contents, and there would be a possibility that the magmas experienced additional processes such as crustal assimilation before the entrapment at the shallow levels. Therefore, the data of the inclusions with low CO 2 contents were not used as the representative data sets for each volcano.
The average H 2 O contents of the three representative inclusions are 3.5 wt.% for Iwate, 2.7 wt.% for Akita-Komagatake, 3.8 wt.% for Me-Akan, 4.1 wt.% for Oshima-Oshima, 3.8 wt.% for Rishiri, and 2.0 wt.% for Fukue. The Cl/F and H 2 O/F ratios of the melt inclusions are shown as a function of the slab depth in Fig. 2a,b, respectively. These ratios are significantly higher for all the volcanoes than those of the depleted mantle or the majority of mid-ocean ridge basalts (MORB) 25,26 . This observation reconfirms that the fluids derived from the Pacific slab were involved in the magma genesis for the volcanoes. In particular, the involvement of the slab-derived fluids at the Fukue volcano supports the 'Big mantle wedge model' 27,28 , according to which deep dehydration from the stagnant slab in the MTZ plays a primary role in the overlying volcanism. Notably, the slab fluids for the Fukue volcano were not derived from the Philippine Sea Plate because the plate is too hot to efficiently carry water to the depths of the MTZ 19 . The Cl/F and H 2 O/F ratios at Fukue (Z = ~ 550 km) are lower than those of the frontal-arc Iwate and Me-Akan volcanoes. This can be attributed to preferential partitioning of Cl and H 2 O into slab fluids relative to F over the slab source materials 29 .
The δD‰ values of the representative melt inclusions are plotted against the slab depths in Fig. 2c. The value of − 80 ± 10‰ 7 was commonly used as the reference δD‰ value for the depleted mantle, and it was later revised to − 60 ± 5‰ 30 . However, recent work suggests that the δD‰ values of the depleted mantle could be heterogeneous worldwide, and this value for some depleted MORB, such as those from the North Atlantic, is − 90 ± 10‰ 12 . The present study used the traditional δD‰ value of − 80 ± 10‰ for the Pacific depleted mantle. The δD‰ values of the samples from all the volcanoes are higher than those of the depleted mantle, which can be interpreted to result from the addition of D-enriched slab fluids to the source depleted mantle 9 . The water dissolved in a subduction-related primary magma is derived from the H 2 O in the slab-derived fluid and the H 2 O contained in the depleted mantle peridotite. We estimated that the contributions of the slab fluids to the H 2 O budget in the primary magmas from the studied volcanoes range from 92 to 98% (Supplementary Methods), and hence, the δD‰ values of the representative melt inclusions (Fig. 2c) can be regarded as those of the slab-derived fluids at the individual volcanoes.
One of the crucial observations is that the δD‰ values of Iwate are lower than those of Akita-Komagatake, which is located just behind the frontal arc of the Iwate volcano. This observation cannot be explained by the single reservoir model proposed in Ref. 9  Recently, the role of the slab peridotite layer as the H 2 O source for subduction zone magmas has been receiving increased attention 11,12,31 . The crust portion of the subducting slab, consisting of altered basaltic rocks [altered oceanic crust (AOC) and gabbro (GAB)] and sediment (SED), and the overlying mantle wedge base peridotite (MwP) are considerably dehydrated before reaching the depth beneath the volcanic front (Fig. 3). Contrastingly, the underlying peridotite portion of the slab (SlbP), which is effectively hydrated during bending at the trench, can survive as a water reservoir beneath the arc 12,31 (Fig. 3). This scenario can explain the observed spatial variation in the δD‰ data. When the slab fluid for Iwate was supplied primarily from the crust portion of the slab (green star in Fig. 3), the D/H was likely low because this portion had already been considerably dehydrated at shallower levels. Conversely, the main source of the fluid beneath the Akita-Komagatake volcano is considered to have been the underlying peridotite portion of the slab (yellow star in Fig. 3). In this case, the fluid would be relatively D-enriched because this portion was not significantly dehydrated before reaching ~ 100 km depth. Therefore, considering that the initial D/H ratios of the crust and peridotite portions of the slab were similar at the time of the subduction, the fluid released at ~ 2.5 GPa from the crust portion would have lower D/H ratios than that released at ~ 3.5 GPa from the underlying peridotite portion 11,12 . The low δD‰ values of the Me-Akan samples (− 60 to − 70‰) may also reflect that the slab fluid responsible for the magmatism was sourced primarily from the crust portion of the slab, considering that the volcano is located at the volcanic front of the Kuril arc.
The δD‰ values of the melt inclusions from the Mariana arc basalts are high, ranging from − 55 to − 12‰ 9 , compared to those of the frontal-arc Iwate and Me-Akan volcanoes. This can be attributed to the fact that the  31,34 , and water is supplied primarily by the breakdown of serpentine in the peridotite layer 12 . No significant decrease in the δD‰ values of slab fluids deeper than ~ 300 km suggests that significant dehydration of the peridotite portion of the slab did not occur during the subduction to the MTZ. This result is consistent with the observation that only a few instances of magmatism, including those of Rishiri and Fukue, are caused by a fluid release from the Pacific slab with Z > ~ 200 km. The host basalts of these samples (i.e. Rishiri and Fukue) have significantly enriched Pb isotopic features compared with the depleted mantle ( Supplementary Fig. S2). This suggests that the fluids that were dehydrated from the peridotite portion of the slab interacted with the overlying crust portion (i.e. 'slab rehydration' 12 ) before the release to the wedge mantle.
Because δD‰ values of ~ − 60‰ at Fukue are representative of the slab fluid beneath the Fukue volcano (Supplementary Method), the δD‰ values of the residual slab materials that were stored in the MTZ can be estimated to be ~ − 90‰, assuming the D/H fractionation factor (1000 lnα) between the slab and fluid of − 30‰ (Ref. 9 and the references therein). If we consider the possible pressure dependence on the fractionation factor 35 , the δD‰ of the residual slab materials could be higher than ~ − 90‰. Our new estimate of the δD‰ values for the recycled slab materials is much higher than − 234‰ estimated by Ref. 9 and is also slightly higher than − 100‰ to − 120‰ obtained by Ref. 11 . At Koolau volcano (Hawaii) where recycled slab materials are considered to have contributed significantly to the source mantle 36 , a δD‰ value of − 120‰ was estimated 16 . If the δD‰ value of the Earth's primordial H 2 O is − 60 to − 80‰ 13,37 , then the lower δD of − 120‰ compared to both the primordial δD and the δD‰ value of > − 90‰ for slab materials stored in the MTZ may suggest that the recycled materials in the Koolau basalts experienced further D/H fractionation during the transport from the MTZ to the bottom of the mantle as a cold plume and or during the residence at the core-mantle boundary until the ascent as a hot plume.

Methods
Analytical methods. Whole-rock major and trace element analyses as well as Sr, Nd, and Pb isotopic analyses on the scoria samples were carried out at the Faculty of Science, Hokkaido University, Japan. Major element compositional analyses of melt inclusions and the host olivine crystals were also conducted at Hokkaido University. The details of the analytical methods were provided in Supplementary Methods. The CO 2 , H 2 O, F, Cl, S, and P 2 O 5 content analyses of the melt inclusions were conducted using an ion microprobe CAMECA IMS-1280HR, AMETEK CAMECA, at the Kochi Institute for Core Sample Research, Figure 3. Schematic of the subduction-zone processes in the NE Japan arc. The figure is modified from Fig. 11A of Ref. 42 illustrating the chemical geodynamics in a cold subduction system. The inset, modified from Fig. 2a of Ref. 31 , shows the result of calculations for the NE Japan arc using the Arc Basalt Simulator version 5. Namely, the results show variations in the total amount of H 2 O released from the slab (the cyan area) and in the amounts of H 2 O stored in the individual layers of the subducting slab with depth. MwP denotes the mantle wedge base peridotite layer just above the subducting slab; SED and AOC refer to the sediment layer and the altered oceanic crust layer, respectively; GAB_U and GAB_L denote the upper and lower gabbro layers, respectively; and SlbP refers to the slab peridotite layer. The green, yellow, and red stars indicate the possible main H 2 O sources of the slab fluids for Iwate, Akita-Komagatake, and Oshima-Oshima, respectively. The red circles denote the melt segregation depths obtained by Refs. 43,44 (the segregation depth at Oshima-Oshima is assumed to be similar to that at Sannome-gata 43 ). Fluids released from the slab surface may be dragged to a slightly deeper depth because they can be trapped in a triple junction of the olivine grains in the overlying mantle wedge base peridotite layer (MwP) 31 . This figure was created using Canvas 14. www.nature.com/scientificreports/ JAMSTEC, following the methods described in Ref. 20 . Melt inclusion-bearing olivine crystals separated from the scoria samples were individually mounted in an acrylic resin, and they were polished until melt inclusions were exposed. The polished crystals were then mounted in an indium-filled Al-disc of 1 inch to minimise volatile background. To dry the sample completely, the cleaned Al-disc was kept in an oven at 80 °C and 10 -7 torr for more than 48 h prior to Au coating. The Au-coated Al-disc was stored in the airlock chamber of the spectrometer at < 10 -8 torr overnight prior to the analyses. We used a 20 keV Cs + primary beam of 500 pA defocused to a diameter of 10-15 µm. A − 10 keV electron gun of ~ 100 μm in diameter was used to avoid charging the sample surface. We set the field aperture size to 5 × 5 µm of the field of view of the secondary ion image to eliminate secondary ion signals from surface contamination. The instrument was operated with a mass resolving power of ~ 6000, which allows us to distinguish completely the interferences of 34 S 1 H on 35 Cl, 17 O on 16 O 1 H, 29 Si 1 H on 30 Si, and 31 P 1 H on 32 S. We detected negative secondary ions of 12 C, 16 OH, 19 F, 30 Si, 31 P, 32 S, and 35 Cl by an axial electron multiplier using a magnetic peak switching method. The total measurement time for each analysis, including 20 s for pre-sputtering, 120 s for auto-centring of secondary ions, and 10 cycles for measurements, was ~ 7 min. Fourteen in-house basaltic glass standards with broad ranges of volatile contents were used 20 . The concentrations of volatiles and P 2 O 5 in melt inclusions were determined by calibration curves of the 30 Si-normalised intensities of volatile elements and 31 P of a set of basaltic standards. Analytical reproducibilities of CO 2 , H 2 O, F, P 2 O 5 , S, and Cl concentrations were 3.6, 0.8, 2.0, 0.9, 0.8, and 1.3%, respectively, which were obtained by repeated analyses (n = 72) on different glass fragments (n = 38) of a homogeneous MORB basalt (EPR-G3) over seven analytical sessions. The maximum uncertainties were less than 5% for H 2 O, 15% for CO 2 , and 10% for Cl, F, S, and P 2 O 5 .
The hydrogen isotope analyses were conducted on the same melt inclusions as those used for the volatile content analyses. The same microprobe as above was used, and a method described in Ref. 21 was followed. We used a 20 keV Cs + ion beam of ~ 5 nA defocused to ~ 15 µm in diameter. A − 10 keV electron gun of ~ 100 µm in diameter was used. We set the field aperture size to 7 × 7 µm, which is smaller than the beam size, to minimise hydrogen contamination from the edge of the primary beam. We measured negative secondary ions of 16 OH − and 16 OD − in multi-detection mode with a Faraday cup (FC) of 10 +11 or 10 +12 Ω resistance and axial electron multiplier, respectively. To sufficiently separate the 16 OD − from 18 O − , 17 OH − , and 16 OHH − , we set the mass resolving power (M/∆M) of the axial detector to approximately 10,000. The mass resolving power of the FC detector was set to ~ 5000 with a slit of 250 µm in width, allowing sufficient separation of the 16 OH − signal from the 17 O − signal. The total measurement time for each analysis, including 20 s pre-sputtering, ~ 120 s for auto-centring of 16 OH − to the field and contrast apertures, and 4 s × 50 cycles for measurements, was ~ 6 min. FC background stability was ~ 700 cps (2sd), which did not affect the D/H ratios as long as OH was > 0.4 Mcps or H 2 O was > 0.2 wt.%.
Three in-house basaltic glass standards were used for the calibration of the analyses. The water contents and δD‰ values are 0.54 wt.% and − 107 ± 20‰ (2sd), 0.94 wt.% and − 109 ± 3‰, and 3.6 wt.% and 37 ± 5‰, respectively. The δD‰ values were determined by TCEA/IRMS (high-temperature conversion elemental analyser/ isotope ratio mass spectrometer). The standard bracketing method was adopted to determine the D/H ratios of the melt inclusions. The working standard of Hawaiian submarine basaltic glass (H 2 O content of 1.4 wt.% and δD‰ of − 95 ± 5‰) was embedded in the sample mount, and it was measured five times for every ten unknown samples. We calculated the instrumental mass fractionation factor and analytical error by averaging 10 data points of the standard glass. If the 2se of the melt inclusion was greater than the 2sd, we adopted the former as the analytical error; otherwise, the latter was adopted as the analytical error. Water contents of the melt inclusions were typically higher than that of the working standard in this study, and hence, analytical errors of most melt inclusions were 2sd of the working standard averages of each bracket.

Data availability
The datasets generated and/or analysed during the current study are available from the corresponding author upon reasonable request.