Molybdenum isotopes unmask slab dehydration and melting beneath the Mariana arc

How serpentinites in the forearc mantle and subducted lithosphere become involved in enriching the subarc mantle source of arc magmas is controversial. Here we report molybdenum isotopes for primitive submarine lavas and serpentinites from active volcanoes and serpentinite mud volcanoes in the Mariana arc. These data, in combination with radiogenic isotopes and elemental ratios, allow development of a model whereby shallow, partially serpentinized and subducted forearc mantle transfers fluid and melt from the subducted slab into the subarc mantle. These entrained forearc mantle fragments are further metasomatized by slab fluids/melts derived from the dehydration of serpentinites in the subducted lithospheric slab. Multistage breakdown of serpentinites in the subduction channel ultimately releases fluids/melts that trigger Mariana volcanic front volcanism. Serpentinites dragged down from the forearc mantle are likely exhausted at >200 km depth, after which slab-derived serpentinites are responsible for generating slab melts.

in the main text to make it clearer. In addition, samples of SW flank from Pagan are not included in Figure S1c. Line 222: less incompatible rather than less mobile. Line 224 to 226: This may be problematic. From the trend of the samples from Pagan, melt derived from slab sediment is likely characterized by low δ98/95Mo (-0.3 ‰) and Mo/Nb, Ba/Nb etc. However, the trend of samples from NW Rota-1 is contrasting, which is unlikely explained by different proportion of sediment melt contribution. Line 229: typo of "in". Line 243: It is not clear what is the Mo isotope composition of the melt derived from the overlying subducted oceanic crust and sediments. Figure 6: Please explain the model in detail in the caption.
Reviewer #2: Remarks to the Author: Heye Freymuth, review of the manuscript: Molybdenum isotopes unmask slab dehydration and melting at Mariana arc by Li et al.
The presented data build on previously published Mo isotope data for the Mariana arc and other arcs. There are several interesting aspects of the dataset that represent a step forward compared to previous publications: 1) Two individual sites are studied in detail, 2) some of the samples are from a rear-arc locality and 3) primitive samples with high MgO are included. I therefore believe that this is a valuable contribution that should be considered for publication in Nature Communications. I nevertheless found a number of issues that should be addressed before publication, in particular regarding the inference that the Mo isotope data can be used to trace serpentinized forearc mantle dragged down by the subducted slab.
The interpretation that serpentinized forearc mantle that is dragged down by the subducting slab and dehydrates to produce the geochemical fluid signature is based on the discussion lines 245-246: "…...this mechanism fails to explain the Mo isotope variations in the Pagan primitive lavas, in that neither an aqueous fluid nor a hydrous melt component shows a mixing trend with the depleted mantle component" and lines 246-255: " Mo isotope variations and their correlation with elemental proxies for aqueous fluid and hydrous melt require that the source material for Pagan volcano was first metasomatized by an aqueous fluid phase to elevate its δ 98/95 Mo and to decrease its Ce/Mo while keeping its Hf/Nd and Hf isotopes intact. The fluid modified source material was further fluxed by hydrous melt to decrease its δ 98/95 Mo, εHf, and Hf/Nd and increase its Ce/Mo (Figures 4c and 4d). The geochemistry of Pagan primitive lavas thus need an intermediate carrier that retains both fluid and melt from the subducted slab before being released to the magma source. Serpentinite dragged down from the fore-arc mantle is the most likely candidate for this carrier." First, Mo is fluid-mobile and highly enriched in arc magmas compared to the mantle. Thus, there is no need for mixing trends with the depleted mantle component. The fluid and slab melt components will dominate the amount of Mo in arc magmas whereas the mantle component is likely close to negligible. It should therefore be unlikely to see mixing towards the mantle. Second, there is no need for a temporal progression in the addition of the various components (as depicted by arrows in Fig. 4). The simplest interpretation of the Pagan data and trends between d98Mo and Hf/Nd would be that the slab component added to the NE flank is more aqueous while the slab component added to the SW flank is more dominated by hydrous melt. Both are then added to the mantle, hence the more mantle-like Hf/Nd at the NE flank (as the fluid adds little Hf and Nd) and the more hydrous melt-like Hf/Nd in the SW flank (influenced by some slab melts). With such a model there is no need for interim storage of fluids in wedge serpentinites. I don't want to argue against the model of serpentinite dragged down from the fore-arc. It may well be viable. But I'm not convinced that the Mo isotope data presented here are tracing this process.
It is unfortunate that observations made in the manuscript are entirely qualitative. Mass balance and/or quantitative mixing models would be very useful here in demonstrating whether the above scenarios are viable. See e.g. models in the recent study by Villalobos-Orchard et al. for the Izu section of the IBM arc.
All Mo isotope date reported here for samples with < 7 wt. % MgO were not used later on due to concerns about effects of fractional crystallisation. Yet, published data used as reference were not filtered in the same way and in fact, previously published data are almost entirely for samples with < 7 wt. % MgO. Those studies have argued against significant modification of Mo isotope ratios by fractional crystallisation, at least for basalts and basaltic andesites. In the Pagan data presented here, only two samples are significantly isotopically lighter than the rest and interestingly, a similar "trend" does not exist for NW Rota. While it is clearly best to focus on primitive samples (and this study reports some of the very few Mo isotope data for high MgO samples) it seems arbitrary to selectively ignore some of the samples with lower MgO.
There is no information on sample preparation in the method section. This is particularly important because the samples are from submarine eruptions and hence easily altered by seawater. Fig. 3A shows a much steeper trend in d98Mo vs. Ce/Mo for Pagan than for previously published arc sections, suggesting an isotopically heavier fluid. This is an interesting aspect of the data, in particular with regards to the mass balance of Mo in subduction zones and beyond and thus worth highlighting.
Other comments on lines: 46-47: It's odd to introduce the fluids as derived from altered oceanic basalt and then later discussing them as derived from serpentinites. That classic AOC model is not really up to date any more.
136-137: Only one Pagan sample has higher Ce/Mo than the depleted mantle in Fig. 3a.
146: Radiogenic Hf and Nd in the mantle doesn't need metasomatic input. 179-180: Why just sediment melt? The AOC may well melt, too.
188: "To reconcile this dilemma…" I think this dilemma needs to be explained in more detail and assessed quantitatively, in particular with respect to the subsequent sentence stating that the sediment contribution is "minimal".
The insights from the manuscript can be largely applicable to multiple and diverse disciplines of the earth and marine sciences. With that said, I think with moderate revisions the manuscript will have high and overall positive impact and is worth publishing in a journal such as Nature Communications.
--------------------------------Main comments: The manuscript contains both new Mo isotope data, as well as published isotope and elemental datasets on these same rock samples. Some of the authors are world-leading experts on arc geochemistry and not surprisingly their sample selection is excellent. The differences with the previous Mo isotope datasets from the same arc (*but different volcanoes) are a bit worrisome, but the explanation of the authors is quite convincing and not analytical in nature. Some of the conclusions and ideas in this manuscript are in direct clash with the views of Chen et al, Nature Comm. 2019 (you have this reference on your line 390), which is also involving Mo isotopes and review of the role of serpentinites (from an Raspas, which is exhumed and well preserved ophiolite complex). In a way this is great and good for your story, making the interpretation novel. In fact your story is basically like their figure 4, but turned upside down. Also worth noting is their AOC compositions as shown on a Mo isotopes vs Mo/Ce plot, where AOC seems to be off the trends shown by the ophiolite, which is in itself not supporting the altered crust/AOC as a "player". I particularly agree that there are some really important insights in the correlations seen and their careful consideration does lead to the conclusions that the hydrated in the forearc and previously depleted mantle peridotites are an end member in the arc magma sources. This has been suggested more than a decade ago via trace element arguments, with Pb-Nd-B isotopes (Tera and co-workers, Ishikawa-san & Nakamura-san; Ryan and co-workers, among others). Adding another tracer for support of the forearc mantle contribution to arc magmas is outstanding achievement! It is novel in that the data is extending the variations of arc volcanic rocks previously reported and via combination of FME/Nb ratios it is convincingly supporting a widely debated issue of the type of serpentine input into the arc magmatic source-lithospheric in origin (deep MORB mantle at bottom of gabbroic slabs) or forearc modified and down dragged with the slab to depths and with ultimately (ultra)-depleted mantle protoliths. I urge the authors to browse through the recently published paper (in Nature Comm.) that reports on the modelling of fluid penetration in deep slabs to form serpentinites and the d11B signatures of the altered oceanic crust (that is eventually subducted). This study (reference is below) is another independent evidence for lack of arc contributions from the hydrated lithospheric section of the slabs. This fact is leaving the forearc fluid-modified mantle as the only other viable alternative (and more reasonable to be honest as hydrous slabs will be hard to subduct due to low density!). McCaig, A.M., et al. No significant boron in the hydrated mantle of most subducting slabs. Nature Comm. 9, 4602 (2018). https://doi.org/10.1038/s41467-018-07064-6 The conclusions of (this) manuscript are also similar to the study of Kimura-san, where there is fantastic and very quantitative estimates of the nature of inputs from the slab and mantle under cross arc volcanic chains across the Izu arc (fluid X= 2-4% shallowly and meltX=1.5-4.5% for the deep melts). Their conclusions were derived from trace elements and also from Sr and Pb isotopes in combination with trace elements. All such data is apparently available for the mafic rocks form the Marianas (this study), so some links perhaps can be further established. I accent on the idea of the manuscript to be a bit more quantitative. I suggest that the latest version of arc basalt simulator (ABS) can used to show some parallels with the Kimura-san's 2010 study (doi:10.1029/2010GC003050) . Perhaps there may be some bridges that can prove useful and importantly some quantifications from the Marianas may be revealing a more global case.
Your figures 3-bottom two segments with Ba/Nb and Cs/Nb-here one may argue that there are two trends, but those trends are not necessarily the one you highlight. If we want to include the Izu arc in the discussion, it will be easier to have trend 1[steep trend] consisting of all NWRotaPagan (COB 2)the high Ba/Nb samples from Alamagan and Guguanall of the Izu arc data (with a mixing end member DM); and trend 2 (more vertical one) = including Agrigan and Uracas volcanoesNW Pagan and the highest 98/95 Mo samples from Alamagan and Guguan volcanoes. I urge the authors to take the forearc serpentinized mantle trace elements from, say Savov et al., 2007-JGR and see where on the trace element graphs these potential end-member compositions will/may plot. Site 801C is clearly not very useful to explain the mixing relationships and so adding something in the high Ba/Nb and Cs/Nb end of the plot and always with very high 98/95 Mo will be quite useful. What is in the upper right corner of your diagrams, anyway? Why is Izu arc and the NE flank of Pagan trending in this direction? Please also note that there may be two different trends on your figure 4 (Mo isotopes vs. Hf isotopes). What is sitting in the upper right corner there? Please add an end-member on the plot. Again the trends here do not go through 801C basalt composite. I suppose that some of the highly modified serpentinites will have elevated 98/95Mo (aqueous fluid arrows on the other plots show vertical enrichments of 98/95Mo) and so these may be good to show (or if anyone have looked at serpentinites for Mo isotopes-those will be good to see as they are central to your hypothesis). There is now published dataset for adakitic melts, effects of hornblende in the source of melts and Mo isotopes. The work is published in Geochim Cosmochim. Acta (https://doi.org/10.1016/j.gca.2021.01.020). It appears to show that hornblende crystallization may dominate the Mo isotope variations in arc magmas. Please consider this new dataset, especially when you discuss the deep sourced magmas. There may be useful information about the effect of mineralogy and mantle metasomatism. Also please check if there is anything in the literature on Mo isotopes in metasomatized/serpentinized rocks.
There are small details that I think may be helpful to clarity, especially for the non-expert (petrologists and arc geochemist) audience of Nature Comm. I list below some of those points and some of the text edits that will help: Minor details linked to the text. line 39-will help if you state "intraoceanic". The mentioning of end-membr convergent margin is not helpful. I rather add that it is "non accreting" and also add that it is quite sediment starved, making it appropriate, if not unique for the cross arc cycling studies. line 47-no CAPS for Boron line 48-The cited studies does not involve Hf isotopes. Perhaps the study of J.Pearce on the Hf isotopes of the Marianas will be good to add. line 50-Please note that the Nd and Hf are NOT a good tracers of fluids as the elements in question are highly fluid immobile. In the case of Nd-please note that there is increasing amout of evidence in the literature that the 143/144Nd of serpentinites may indeed vary vastly (research by Bizimis and coworkers) and the process behind it is not well understood. line 53-"inconsistent with the Mariana arc lavas"-perhaps add the range here. Line 54-these are indeed some moderately elevated 87/86Sr values, but this is in respect to MORBs. Otherwise, in respect to anything in the slab those are immensely low or better-unradiogenic. line 74-77-what are the errors for the Mo isotope ratios. If those are large, then the MORBs and sediments may nearly overlap. This is critical point why we need AOC as end-member and why you have shown exactly that on your plots (ODP Site 801 end member).
Line 84-You need to state how we know these depths. What methods were used to determine where the slabs are and at what dip they sink in the mantle. A reference will also be good.
line 108-East Knoll is with CAPS. line 112-you may want to state somewhere what is the slab DIP. One way to do this is maybe in the caption of your schematic summary diagram. In any case-for non specialists there is a need to explain that. Another option will be to add it as a method. line 125-but this is not too distant range I respect to the Izu VF (0.16 per mil) , isn't it? This is one of the reasons that the range of the isotope errors needs to be properly reported. Line 132-Cs/Nb are high in fluid, no doubt. This has been shown nicely in Savov et al, 2007-JGR manuscript. Line 140-I recommend that for the same of consistency you stick to your Ce/Mo (as in your line 126) and do not confuse things by introducing Mo/Ce. So here you may say " high Ce/Mo". line 168-elevated Sr isotope ratios in respect to MORB. And only so very little! line 171-172-Need to tell us if this fluid is realistic (see my recommendation for use of McCaig et al, 2019) . In any case-if this fluid exists [doubt that!] is released-it will react with the abundant fresh gabbro and diabase and will not manage to do anything. If you eliminate this option early-then you can use some lines for properly introducing the forearc down dragging processes (which are not vague) and link to the Mariana md volcanoes, which are physical evidence for the high Ba and Ca you need. Line 180-why is this. Maybe clarify your arguments. line 184-this increasingly make more and more sense to me. line 193 ( and several other places in the manuscript)-a just published GCA paper by Anders et al (2021) is reporting very interesting story from Izu arc and is telling us tat sediments are not playing a role and that the slabs are melting. Please add this paper insights (Sr, Pb, Nd, Hf isootpes are VERY abundant!) to your story. I think it is highly relevant to see what is the entire range of mantle melts, nicely shown vs. sediments-all nicely summarized in their plots. line 196-need reference here after t"sediments" line 229-"from in" line 232-hence the two trends on figure 4B (Hf isotopes) line 236-This is a good place to cite some forearc serpentinite peridotite major and trace element paper. line 239-it will be more thorough if here you also cite McCaig et al. (2019). Line 246-here refer to some of your figures line 254-this "dragging down" of the serpentinized mantle needs to be either explained in a bit more detail or some key references need to be given. As it is, for average reader, this is not clear enough. line 255-this is SUPER! I really think this is a great selling point of your paper and you may want to further accent on this fact, in addition to the mixing trends and arc magma genesis. People should start plotting this as mixing end member and not some composite samples of AOC or sediments, which may but may not be relevant. line 263-"Breakdown of serpentine …"-see McCaig et al. (2019). Also note than at high T this breakdown will lead to formation of chlorite-rich protoliths. Line 272-please give reference here for the depth.
Just a note to the authors, that serpentinites often have Ba/Th (10^3 and 10^4).

Overall response to reviewers：
The reviewer's main comments are answered as follows: (1) Key contribution.
Our key contribution in this study is that we can use the Mo isotopes to discriminate among different sources of subduction related serpentinites. Forearc mantle and slab lithosphere serpentinites play different roles in generating subduction zone fluids. On the one hand, forearc serpentinites downdragged by the slab act as an important intermediate carrier for slab fluid and melt transfer in the subduction channel. On the other hand, dehydration of serpentinites in the slab lithosphere is critical for triggering slab dehydration and melting at depth. We have rewritten part of the introduction and discussion sections. The following information has been added in the revision: Recent geochemical and geophysical observations proposed that dehydration of serpentinite may play a more critical role in explaining these observed elemental and isotopic paradoxes (Cai et al., 2018;Savov et al., 2005Savov et al., , 2007Tamura et al., 2014;Freymouth et al., 2015). Two kinds of serpentinite may have contributed fluids for the Mariana arc mantle. The first is serpentinite formed in the shallow forearc mantle wedge as a result of the incorporation of low-temperature fluid (80-350℃) from the slab during its initial burial (<19km; Freyer et al., 1996Freyer et al., , 2020Savov et al., 2005Savov et al., , 2007. Forearc mantle serpentinite that diapirs and extrudes at the seafloor are extremely enriched of fluid mobile elements (e.g., B, As, Cs, Sb, Li) and has very high δ 11 B (> 15‰), indicating large slab inventory depletion of B (>75%), Cs (>25%), As (>15%), Li (>15%), and Sb (>8%) during its initial stage of dehydration and significant residual slab δ 11 B (~-6±4‰) decreasing as a result of selective removal of heavy 11 B by the fluid (Benton et al., 2001;Savov et al., 2005Savov et al., , 2007Pabst et al., 2012). The slab would experience further fluid mobile elements depletion and δ 11 B decreasing during deeper subduction. Physically addition of forearc serpentinites to the sub-arc mantle or subduction channel would be required to explain the enrichment of fluid mobile element and heavy B isotope (e.g. +4.5 to +12.0‰ in the Izu Arc; Ishikawa and Nakamura, 1994;Straub and Layne, 2002) characteristics of the volcanic front basalt (Benton et al., 2001;Savov et al., 2005Savov et al., , 2007Tonarini et al., 2011;Ryan and Chauvel, 2014). (2) Quantitative calculation.
A major criticism from the referees is the qualitative character of our original explanation. In the revision, we did detailed quantitative calculations using the Mo-Nd-Hf elemental and isotopic compositions for slab components (sediments, altered MORB, fresh MORB, and MORB-like eclogite) and the ambient mantle, partition coefficients during slab dehydration/melting at different P-T condition  (3) The effects of fractional crystallization on Mo isotope and Mo/Ce ratios.
The basic law for fractional crystallization of amphibole affecting the Mo/Ce of the magma is that Mo is more incompatible than Ce for a given amphibole-bearing magma system. It is confirmed by the trace element partitioning during amphibole-bearing garnet lherzolite melting (Adam and Green, 2006) and magma differentiation of the Kos Plateau Tuff (Voegelin et al., 2014). Also, amphibole preferentially incorporates middle REEs over heavy REEs, as well as Nd over Hf during oceanic crust melting or magma fractional crystallization (e.g., Tiepolo et al., 2007). Therefore, amphibole fractional crystallization will induce increasing of Hf/Nd and decreasing in Ce/Mo and Dy/Yb of the residual magma. Villalobos-Orchard et al., 2020). Therefore, samples that experienced significant amphibole fractional crystallization should be excluded before discussing the slab dehydration and mantle melting process.
For other Izu and Mariana data cited in the text and diagrams, we didn't find the amphibole fractional crystallization effects based on proxies such as Hf/Nd and Dy/Yb. We think these samples didn't experience significant fractional crystallization of amphibole, thus their Ce/Mo and Mo isotopes mostly reflect the slab dehydration and mantle melting processes.
We analyzed Mo isotopes for serpentinite mud samples from the Asùt Tesoru mud volcano (collected during IODP Exp. 366; informally known as Big Blue Seamount) during the review process, and the data are presented in the Source Data document and now plotted in indicating the forearc mantle serpentinite could be potential source components for the IBM basalts (Fig. 3).
We should point out that the forearc serpentinites generated at shallow depth will be further metasomatized by the slab fluid when they are dragged down by the slab to a depth of > 80 km, where the slab starts to couple with the overriding mantle and being heated up to devolatilize. The mantle wedge serpentinite we discriminated is not exactly equivalent to serpentine mud volcanoes that erupt at the forearc corresponding to a very shallow plate depth (< 19 km).
This information have been added in the revision. While I find the new Mo isotope data of high quality and the proposed model very interesting, I also think the discussion should be more thorough, some of which need quantitative calculations.
General comments: 1. In the introduction section, the authors could more stress on what is newly constrained here or refined by Mo isotope results.
Please see reply to the 'Overall response to reviewers (1)'.
2. It is better to re-assess the effects of fractional crystallization on Mo isotope and Mo/Ce ratios and re-consider the evolved samples in this study.
Please see reply to the 'Overall response to reviewers (3)'.

Given the authors proposed/identified several slab components based Mo isotopes
and other geochemical indicators, it is useful to quantify each contribution in more detail for the samples from two volcanoes.
Please see reply to the 'Overall response to reviewers (2)'.
Overall, I think this will be a good and useful paper in the field of subduction science and Mo isotope research after revision.

More detailed comments:
Line 38 to 68: In the introduction part, the authors could more stress on which mechanism regarding this process is newly constrained here or refined by Mo isotope results.
We have rewritten part of the introduction section, please see reply to Overall response to reviewers (1).
Line 69 to 70: It is better to add refs (e.g., Bali et al., 2012, EPSL). Line 140 to 141: It is better to keep consistency to use Ce/Mo or Mo/Ce in the whole text.
We select to use Ce/Mo consistently in the text and diagrams.
Line 192 to 193: It will be helpful to quantify each contribution in terms of Sr-Pb-Mo isotopes using simple calculation of mixing models.
We think detailed quantify calculation based on Sr-Pb isotope is difficult as the slab surface would experience long-term, and dynamic interaction with the serpentinite fluid from the deep slab lithosphere (Chen et al., 2019;Freymuth et al., 2015;Klaver et al., 2020). These processes affect the Sr-Pb-Mo elements and isotopes a lot. In our calculation, we use Hf-Nd-Mo systematics to constrain the contributions from different source, as Hf-Nd is hardly to be affected by the serpentinite fluid. Please see reply to the 'Overall response to reviewers (2)'.
Line 209: The information in Figure S1c is quite interesting and important, which needs to be included in the main text to make it clearer. In addition, samples of SW flank from Pagan are not included in Figure S1c.
This diagram has been moved to the main text (Figure 4 now).
Line 222: less incompatible rather than less mobile.
After detailed consideration, we think it is not a partition effect. The reason is the slab is already depleted of Mo when it heated up to melt after early dehydration at shallow depth. It is supported by the high Ce/Mo of the eclogite that didn't experience melting  The presented data build on previously published Mo isotope data for the Mariana arc and other arcs. There are several interesting aspects of the dataset that represent a step forward compared to previous publications: 1) Two individual sites are studied in detail, 2) some of the samples are from a rear-arc locality and 3) primitive samples with high MgO are included. I therefore believe that this is a valuable contribution that should be considered for publication in Nature Communications. I nevertheless found a number of issues that should be addressed before publication, in particular regarding the inference that the Mo isotope data can be used to trace serpentinized forearc mantle dragged down by the subducted slab.

General comments:
The interpretation that serpentinized forearc mantle that is dragged down by the subducting slab and dehydrates to produce the geochemical fluid signature is based on the discussion lines 245-246: "…...this mechanism fails to explain the Mo isotope All Mo isotope date reported here for samples with < 7 wt. % MgO were not used later on due to concerns about effects of fractional crystallisation. Yet, published data used as reference were not filtered in the same way and in fact, previously published data are almost entirely for samples with < 7 wt. % MgO. Those studies have argued against significant modification of Mo isotope ratios by fractional crystallisation, at least for basalts and basaltic andesites. In the Pagan data presented here, only two samples are significantly isotopically lighter than the rest and interestingly, a similar "trend" does not exist for NW Rota. While it is clearly best to focus on primitive samples (and this study reports some of the very few Mo isotope data for high MgO samples) it seems arbitrary to selectively ignore some of the samples with lower

MgO.
Please see reply to the Overall response to reviewers (3).
We have re-evaluated the amphibole fractional crystallization effects on the arc basalt geochemistry. Amphibole fractional crystallization will induce increasing of Hf/Nd and decreasing of Ce/Mo and Dy/Yb of the residual magma (e.g., Tiepolo et al., 2007;Davidson et al., 2007;Voegelin et al., 2014;Wille et al., 2018). For the Izu and Mariana data cited in the text and diagrams. We can't find the amphibole fractional crystallization effects based on proxies such as Hf/Nd and Dy/Yb. We think these

The information has been suitably added in the text.
There is no information on sample preparation in the method section. This is particularly important because the samples are from submarine eruptions and hence easily altered by seawater.
Information on sample preparation has been added in the method section. All samples were pulverized in an agate ball mill after sawing and jaw crushing.

Other comments on lines:
46-47: It's odd to introduce the fluids as derived from altered oceanic basalt and then later discussing them as derived from serpentinites. That classic AOC model is not really up to date any more.
We have rewritten part of the introduction section. Please see reply to 'Overall response to reviewers (1)'.
In the revision, we first point out the problems that have been identified from radiogenic isotope studies and then introduced the importance of serpentinite (formed in forearc mantle and slab lithosphere) dehydration in completing these understandings.
136-137: Only one Pagan sample has higher Ce/Mo than the depleted mantle in According the constraints from Hf-Nd isotopes, the slab surface is composed by AMOC and sediments with a ratios of 9:1. We use slab surface melting in the whole text of the revision. It contains contributions from both of the AMOC and sediments.
188: "To reconcile this dilemma…" I think this dilemma needs to be explained in more detail and assessed quantitatively, in particular with respect to the subsequent sentence stating that the sediment contribution is "minimal".
Please see reply to 'Overall response to reviewers (2)'. We have added detailed calculations for the different slab components based on Mo-Nd-Hf systematics.
According to the Hf-Nd isotope and Hf/Nd ratio constrains, the slab melt was derived from a mixed source consisting of subducted AMOC and sediments with a mass ratio of 9:1 ( Figures 5C and S3). Then the slab melts should have much higher 87

Reviewer #3 (Remarks to the Author):
Review of MS NCOMMS-21-02686 "Molybdenum isotopes unmask slab dehydration and melting at Mariana arc" Reviewed by Ivan Savov (Univ. Leeds) This is a very well written manuscript, with excellent graphics and good data quality, including data of a novel tracer (Mo isotopes). The manuscript supports interesting and currently debated hypothesis for the mechanisms of elemental and isotope cycling from subducting plates to the surface volcanoes. The insights from the manuscript can be largely applicable to multiple and diverse disciplines of the earth and marine sciences. With that said, I think with moderate revisions the manuscript will have high and overall positive impact and is worth publishing in a journal such as Nature

Communications. Main comments:
The manuscript contains both new Mo isotope data, as well as published isotope and elemental datasets on these same rock samples. Some of the authors are worldleading experts on arc geochemistry and not surprisingly their sample selection is excellent. The differences with the previous Mo isotope datasets from the same arc Also worth noting is their AOC compositions as shown on a Mo isotopes vs Mo/Ce plot, where AOC seems to be off the trends shown by the ophiolite, which is in itself not supporting the altered crust/AOC as a "player". I particularly agree that there are some really important insights in the correlations seen and their careful consideration does lead to the conclusions that the hydrated in the forearc and previously depleted mantle peridotites are an end member in the arc magma sources.
We thank the comments and suggestions above and below. Our key contribution of this study is that we have discriminate the different role We have re-written part of the introduction and text to clarify our main finding.
This has been suggested more than a decade ago via trace element arguments, with Pb-Nd-B isotopes (Tera and co-workers, Ishikawa-san & Nakamura-san; Ryan and co-workers, among others). Adding another tracer for support of the forearc mantle contribution to arc magmas is outstanding achievement! It is novel in that the data is extending the variations of arc volcanic rocks previously reported and via combination of FME/Nb ratios it is convincingly supporting a widely debated issue of the type of serpentine input into the arc magmatic source-lithospheric in origin (deep MORB mantle at bottom of gabbroic slabs) or forearc modified and down dragged with the slab to depths and with ultimately (ultra)-depleted mantle protoliths. I urge the authors to browse through the recently published paper (in Nature Comm.) that reports on the modelling of fluid penetration in deep slabs to form serpentinites and the d11B signatures of the altered oceanic crust (that is eventually subducted). This study (reference is below) is another independent evidence for lack of arc contributions from the hydrated lithospheric section of the slabs. This fact is leaving the forearc fluid-modified mantle as the only other viable alternative (and more reasonable to be honest as hydrous slabs will be hard to subduct due to low density!). We thank the suggestion for making mass balance calculation using the ABS. We have tried to make the calculation using the latest version, ABS5 (Kimura et al., 2017).

McCaig
This software is designed to calculate the total subduction inputs. Given that we've identified different slab components, shallow slab fluids, deep slab melts, and slab lithosphere serpentinite fluids, we chose to perform manual calculations using available parameters, which allowed us to calculate Mo and Mo isotopes of slab The endmember in the upper right corner should be the mantle wedge serpentinite dragged down by the slab from a depth shallower than 80km. The Asùt Tesoru mud volcano is much closer to the Pagan volcano than other serpentinite mud volcanos with data reported. We analyzed Mo isotopes for some samples from the Asùt Tesoru mud volcano during the review process as part of a collaboration with co-author Ryan, and the data are now plotted in the Fig. 3. As you supposed they have high δ 98/95 Mo. We should point out that these serpentinites will be further metasomatized by the slab fluid when they are dragged down by the slab to a depth of > 80 km, where the slab starts to couple with the overriding mantle and being heated up to devolatilize.
The mantle wedge serpentinite we discriminated is not exactly equivalent to serpentine mud volcanoes that erupt at the forearc corresponding to a very shallow plate depth (< 19 km).
There is now published dataset for adakitic melts, effects of hornblende in the source of melts and Mo isotopes. The work is published in Geochim Cosmochim. Acta This reference is citied in the revision.
We added Mo and Mo isotope data for serpentinite samples. Please see reply to the 'Overall response to reviewers (2)' for details.
There are small details that I think may be helpful to clarity, especially for the non-expert (petrologists and arc geochemist) audience of Nature Comm. I list below some of those points and some of the text edits that will help.

Minor details linked to the text:
line 39-will help if you state "intraoceanic". The mentioning of end-membr convergent margin is not helpful. I rather add that it is "non accreting" and also add that it is quite sediment starved, making it appropriate, if not unique for the cross arc cycling studies.
The sentences have been re-written in the revision: The Izu-Bonin-Mariana (IBM) arc system stretches over 2,800 km from near Tokyo, Japan to south of Guam, USA, and is a typical intraoceanic island arc with negligible inputs from the sub-arc crust to the lavas. The IBM system is an endmember non-accreting convergent margin with a thin sedimentary cover on the downgoing slab, which means that input and output fluxes in the subduction zone may be more confidently assessed (Stern et al., 2004).

Rewording:
The aqueous fluid component has been presumed to derive from subducted altered mafic oceanic crust (AMOC) based on heavier boron isotopes and more radiogenic Nd isotopes, indicating a less sediment affected mantle source (Elliott et al., 1997;Ishikawa and Tera, 1999). The hydrous melt component is from subducted sediments, given its high Th/Nb and less radiogenic Nd isotopes (Elliott et al., 1997).
line 50-Please note that the Nd and Hf are NOT a good tracers of fluids as the elements in question are highly fluid immobile. In the case of Nd-please note that there is increasing amout of evidence in the literature that the 143/144Nd of serpentinites may indeed vary vastly (research by Bizimis and co-workers) and the process behind it is not well understood. Please see reply above. Here we introduce the problems of study based on other isotopes and trace element ratios.
line 53-"inconsistent with the Mariana arc lavas"-perhaps add the range here.
Line 54-these are indeed some moderately elevated 87/86Sr values, but this is in respect to MORBs. Otherwise, in respect to anything in the slab those are immensely low or better-unradiogenic.
We select to use the word "unradiogenic". Corrected.
line 112-you may want to state somewhere what is the slab DIP. One way to do this is maybe in the caption of your schematic summary diagram. In any case-for non specialists there is a need to explain that. Another option will be to add it as a method. Brucite in the slab serpentinite will breakdown at temperature 300-400 (Plümper et al., 2016;Peters et al., 2020). According the study of Erro-Tobbio meta-serpentinites (Ligurian Alps, Italy), the fluid can escape via channel networks from the slab mantle. The reaction is: Antigorite + Brucite= 2 Olivine + 3 H 2 O (Plümper et al., 2016;Peters et al., 2020). According the thermal structure of the Mariana arc, the slab surface has much higher temperature than the slab Moho depth  (Freymuth et al., 2015(Freymuth et al., , 2016(Freymuth et al., , 2019. These isotopes need the fluid (generated by slab serpentinite breakdown) composition to travel through and equilibrate with the overlying oceanic crust to modulate the isotopes to that of pristine MORB. On the one hand we think the mantle wedge serpentinite will be dragged down, on the other hand we think they will be further metasomatized by the slab fluids. The slab fluid may be related to breakdown of the plate serpentinite.
We did some rewording in the revision to clarify our discussion. The fluid-rock interaction is affected by the thermal structure of the slab, therefore, fluids may behave differently on the way down and up.
Brucite in the slab serpentinite will breakdown at temperature 300-400 (Plümper et al., 2016;Peters et al., 2020). According the study of Erro-Tobbio meta-serpentinites (Ligurian Alps, Italy), the fluid can escape via channel networks from the slab mantle. The reaction is: Antigorite + Brucite= 2 Olivine + 3 H 2 O (Plümper et al., 2016;Peters et al., 2020). According the thermal structure of the Mariana arc, the slab surface has much higher temperature than the slab Moho depth Please see reply to the main concern. The Ba/Th is also significantly affected by the degree of mantle depletion for the serpentinite. As the Th for samples from the South Chamorro can be 2-3 orders of magnitude lower than those from the Asùt Tesoru samples, their Cs/Th are reasonably 2-3 orders of magnitude higher than the later. We think the depletion of Th may be related to early melt extraction from the forearc mantle, either during subduction initiation or more ancient event (Li et al., 2019a;Parkinson et al., 1998). Different degrees of forearc mantle depletion may also affect the corresponding volcanic arc basalt chemistry, e.g., the high Ba/Nb and Cs/Nb signature of the Izu volcanic front samples (Figure 3).

Reviewers' Comments:
Reviewer #1: Remarks to the Author: In the revised version of this manuscript as well as in their detailed replies, the authors have adequately addressed my concerns. Now this work will be a valuable contribution that I recommend for publication in Nature Communications in its present form.
Reviewer #2: Remarks to the Author: The paper has improved during revisions and is well written and presented. Data for seamount samples were also added which are interesting.
But the way the seamount samples are presented could be improved. Despite there being only five datapoints, the data are shown as rectangular fields in Fig. 3 which are quite large and make it impossible to judge where the actual data are located in the plots. I suggest that these are replaced by the actual data instead.
My main point of criticism is that I believe the importance of the forearc serpentinites in generating the arc magma chemical inventory is overstated: -The main argument against a model that could explain the data without the involvement of forearc serpentinites is discussed in lines 273-279, namely that the temperature variation corresponding to varying slab depths beneath Pagan is too low to explain significant variation in slab melt + slab fluid mixtures. This is a rather surprising argument, as -if I understand it correctly -it assumes that fluids and melts generated beneath the arc move towards the surface strictly vertically. There is a lot of evidence that fluid and melt migration pathways are highly complex and some studies explicitly suggest lateral flow and that magmas might frequently tap into backarc sources to produce geochemical variation (see e.g. Ishizuka et al. 2015, EPSL).
-Very similar geochemical models have been presented that do not require forearc serpentinites as a source for Mo, see Villalobos-Orchard et al. 2020 whose models can be directly compared to those in Fig. 5. In conclusion, I believe that there is a possibility that forearc serprentinites are involved in arc magma formation and this is well worth pointing out, as done in this manuscript. Yet, the same data can be explained without such a model and the new data are by no means a smoking gun for the involvement of fore-arc serpentinites. It could be presented as the author's preferred model but unless a stronger case can be made I think it is not appropriate to present it as 'requirement'.
There seem to be some inconsistencies in the models. In particular, the need for the deep slab melt (beneath NW Rota) to have higher d98Mo than the shallower slab melt beneath Pagan is odd, as the slab looses isotopically heavy Mo at shallower levels. From the text, it seems that the 6 GPa melt is actually a mixture of a melt and a fluid, while at 4 GPa the fluid and melt components are treated separately. This needs some clarification. Can the geochemical models be added to the d98Mo vs. Ce/Mo plot (or a panel added to Fig.5)? All the parameters seem to be available. It would be interesting to see if the models fit the data in that plot.
Lines 169-172: The 'much steeper trend' is impossible to see in Figure 3a because trendlines in that figure should not be linear if one of the axes is logarithmic.
Lines 379-381: The argument that no mixing trends are seen with the depleted mantle (DM) is not good here (see also my previous review). Even the geochemical models in Fig. 5 include the DM, yet there are no mixing trends seen with it either.

Reviewer #2 (Remarks to the Author):
The paper has improved during revisions and is well written and presented. Data for seamount samples were also added which are interesting.
But the way the seamount samples are presented could be improved. Despite there being only five data points, the data are shown as rectangular fields in Fig. 3 which are quite large and make it impossible to judge where the actual data are located in the plots. I suggest that these are replaced by the actual data instead.
Thanks for the suggestion. The serpentinite data have been presented with the actual data points in the new Fig. 3. We have also adjusted the y-axis to show all the relevant data as possible.
My main point of criticism is that I believe the importance of the forearc serpentinites in generating the arc magma chemical inventory is overstated: -The main argument against a model that could explain the data without the involvement of forearc serpentinites is discussed in lines 273-279, namely that the temperature variation corresponding to varying slab depths beneath Pagan is too low to explain significant variation in slab melt + slab fluid mixtures. This is a rather surprising argument, as -if I understand it correctly -it assumes that fluids and melts generated beneath the arc move towards the surface strictly vertically. There is a lot of evidence that fluid and melt migration pathways are highly complex and some studies explicitly suggest lateral flow and that magmas might frequently tap into back arc sources to produce geochemical variation (see e.g. Ishizuka et al. 2015, EPSL).
-Very similar geochemical models have been presented that do not require forearc serpentinites as a source for Mo, see Villalobos-Orchard et al. 2020 whose models can be directly compared to those in Fig. 5.
In conclusion, I believe that there is a possibility that forearc serprentinites are involved in arc magma formation and this is well worth pointing out, as done in this manuscript. Yet, the same data can be explained without such a model and the new data are by no means a smoking gun for the involvement of fore-arc serpentinites. It could be presented as the author's preferred model but unless a stronger case can be made I think it is not appropriate to present it as 'requirement'.
We thank these constructive suggestions and accept the criticisms. We have added some words to indicate that this is our preferred model, which need to be checked by future research.  Although the role of mixing of slab components in the mantle wedge or mixing of magmas in the plumbing system cannot be fully ruled out (Ishizuka et al., 2015), this does not explain the Th/Nb versus Pb/Ce variation in the Pagan samples (Fig. S4).
Our δ 98/95 Mo versus Ce/Mo plot (Fig. 3a) also doesn't support a model of mixing magmas from the rear arc to the volcanic front. The phenomena that the fluid signature gradually decreased while the melt signature gradually increased along with the increasing of slab depth best fit the Pagan data of this study (Fig. S4).
Covariations between fluid proxies and melt proxies indicate the two slab components were possibly first captured by a source contributor before being released to the magma mantle source.
We have indicated that the forearc serpentinites added to the Pagan volcano source is not exactly the same as the Asùt Tesoru serpentinites. The fluid component for Pagan volcano bears similarities to the Asùt Tesoru serpentinites in that they have high δ 98/95 Mo, low Ce/Mo and high Mo/Nb. However, the high Cs/Nb and Ba/Nb signature of the volcanic lavas (Figure 3) are difficult to explain through serpentinite inputs alone, as Cs is only moderately (25% to 30 %) while Ba is only slightly (< 2%) mobile off the subducting slab at 10-40 km depths (Savov et al., 2005(Savov et al., , 2007. Further metasomatism of the subduction channel material by slab fluids released at depths > 40 km is necessary, under slab thermal conditions hot enough to mobilize Ba. Other more fluid sensitive isotope systematics, like boron, can be used to test our model (Benton et al., 2001;Savov et al., 2005Savov et al., , 2007Pabst et al., 2012). Detailed geophysical observations may also be helpful.
There seem to be some inconsistencies in the models. In particular, the need for the deep slab melt (beneath NW Rota) to have higher d98Mo than the shallower slab melt beneath Pagan is odd, as the slab looses isotopically heavy Mo at shallower levels.
From the text, it seems that the 6 GPa melt is actually a mixture of a melt and a fluid, while at 4 GPa the fluid and melt components are treated separately. This needs some clarification.
According the δ 98/95 Mo vs. Ce/Mo diagram (Fig. 3a), it is clear the hydrous melt for the NW Rota-1 is not a mixture of fluid and melt for the Pagan. As the slab would lose isotopically heavy Mo at shallower levels, the heavy Mo and high Ce/Mo signature of NW Rota-1 need the deeply subducted crust to be fluxed with high δ 98/95 Mo fluid.
This phenomena is first observed from this study according our knowledge. This fluid may be from the deep lithosphere that has been rinsed by early fluid (experienced early Mo loss) at shallow depth. We have improved a little bit of the calculations to better depict the slab dehydration and melting process based Ce/Mo and δ 98/95 Mo.  (Table S1). It indicates a deep lithosphere fluid for melting of the slab beneath the SW Pagan is not necessary (Fig. S5). The Sr-Pb isotope signatures of the slab may have been buffered toward the MORB values as a result of long-term early lithosphere fluid percolation at shallow depth.
3) The slab at 6 GPa is assumed to have experienced early melting at 4 GPa (900 ℃; F=10%). Deep lithosphere fluid at 6 GPa is assumed to equilibrate with a MORB-like source that has experienced 2% fluid percolation and Mo loss at shallow depth. Then melting of a mixture of 98% slab + 2% deep fluid at 6 GPa can generate the Ce/Mo and δ 98/95 Mo signature of the NW Rota-1.
These calculations suggest that the dehydration of the slab lithosphere serpentinite may be episodic. At slab depth < 170 km, the dehydration is controlled by mineral reaction of Antigorite + Brucite= 2 Olivine + 3 H 2 O at low temperature (> 300 ;Plümper et al., 2016;Peters et al., 2020). According the slab Moho temperature of the south Mariana arc, the reaction between antigorite and brucite can start at ~ 100 km slab depth (Fig. D1). At slab depth > 200km, the slab Moho can reach a temperature of > 550 . Then dehydration of the lithosphere is possibly controlled by breakdown of antigorite (Ulmer and Trommsdorff, 1995).
These information has been added in the text. Can the geochemical models be added to the d98Mo vs. Ce/Mo plot (or a panel added to Fig.5)? All the parameters seem to be available. It would be interesting to see if the models fit the data in that plot. The information has been added in the text.
Lines 169-172: The 'much steeper trend' is impossible to see in Figure 3a because trendlines in that figure should not be linear if one of the axes is logarithmic.
The new Fig. 3a is changed to be a linear-linear correlation. Now, the 'much steeper trend' is clear.
Lines 379-381: The argument that no mixing trends are seen with the depleted mantle (DM) is not good here (see also my previous review). Even the geochemical models in Fig. 5 include the DM, yet there are no mixing trends seen with it either.
Thanks for pointing out this. This sentence has been deleted from the text. We have accordingly changed the new Fig. 3a to be a linear-linear correlation.