Slab-derived devolatilization fluids oxidized by subducted metasedimentary rocks

Metamorphic devolatilization of subducted slabs generates aqueous fluids that ascend into the mantle wedge, driving the partial melting that produces arc magmas. These magmas have oxygen fugacities some 10–1,000 times higher than magmas generated at mid-ocean ridges. Whether this oxidized magmatic character is imparted by slab fluids or is acquired during ascent and interaction with the surrounding mantle or crust is debated. Here we study the petrology of metasedimentary rocks from two Tertiary Aegean subduction complexes in combination with reactive transport modelling to investigate the oxidative potential of the sedimentary rocks that cover slabs. We find that the metasedimentary rocks preserve evidence for fluid-mediated redox reactions and could be highly oxidized. Furthermore, the modelling demonstrates that layers of these oxidized rocks less than about 200 m thick have the capacity to oxidize the ascending slab dehydration flux via redox reactions that remove H2, CH4 and/or H2S from the fluids. These fluids can then oxidize the overlying mantle wedge at rates comparable to arc magma generation rates, primarily via reactions involving sulfur species. Oxidized metasedimentary rocks need not generate large amounts of fluid themselves but could instead oxidize slab dehydration fluids ascending through them. Proposed Phanerozoic increases in arc magma oxygen fugacity may reflect the recycling of oxidative weathering products following Neoproterozoic–Palaeozoic marine and atmospheric oxygenation. Metasedimentary rocks atop the downgoing slab oxidize ascending slab-derived dehydration fluids by removing reduced species, according to petrological analysis of subduction complex metasedimentary rocks and reactive transport modelling.

M agmatic arcs above subduction zones produce most of the world's explosive volcanism and host giant ore deposits of copper, molybdenum, gold and other valuable metals. Arc magmas are considerably more oxidized than mid-ocean ridge basalts 1-5 and generate volcanic eruptions that can inject sulfur gases (mainly SO 2 ) into the stratosphere, producing sulfate aerosols that trigger transient tropospheric cooling and stratospheric heating 6 . The origins of the oxidized signature in arcs, as well as in the underlying lithospheric mantle [1][2][3][4][5]7 , are vigorously debated 4,[8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23] . One family of hypotheses holds that the fluids released by subducting slabs are inherently oxidized relative to the pristine igneous rocks generated at mid-ocean ridges. The oxidation may take place during seafloor hydrothermal alteration of mafic crust and/or serpentinization of ultramafic rocks at mid-ocean ridges before subduction. During subduction zone devolatilization, these rocks release fluids with a high oxidation potential to the mantle wedge and give rise to oxidized arc magmas via flux melting 4,[8][9][10][11][12][13][14][15] . In contrast, another set of hypotheses posits that the oxidized signature is acquired in the mantle or crust overlying the subduction zone 16,17 . Some proposed oxidative pathways include the loss of reducing components (for example, H 2 ) from ascending melts (or fluids) to the surrounding mantle 16,18 or fractional crystallization of Fe 2+ -rich phases such as garnet in deep lithospheric magma chambers 19 .
The veneer (~400 m thick 24 ) of sediments that covers subducted slabs worldwide provides another potential oxidative pathway but has received much less attention than other slab lithologies. Oxidized Fe 3+ -bearing sedimentary detritus containing goethite (FeO(OH)) and haematite (Fe 2 O 3 ) from weathered continental sources can be transported to marine depositional settings thousands of kilometres from the shore (for example, the Bengal Fan 25 ). Aeolian fluxes of oxidized minerals to deep ocean basins can occur on similar scales (for example, Indian and Pacific Oceans 26 ). Furthermore, highly oxidized oceanic (meta)sediments (such as Fe-and Mn-rich cherts) are widely known from exhumed subduction complexes including those of California 27 , Japan 27 and New Zealand 28 in the Circum-Pacific region, the Alps 29 and other localities globally 30 . Moreover, the substantial oxidation potential of sediment entering subduction zones, such as the Mariana subduction zone, has been clearly documented in extensive ocean floor drill cores 8,31 . What is urgently needed now is a field-based evaluation of the redox states of subducted metasedimentary rocks and the extent to which they can regulate the f O 2 of devolatilization fluids released from downgoing slabs.
To address this important gap in knowledge, we investigated metasedimentary rocks from two forearc subduction complexes in the Aegean region, Greece (Methods and Extended Data Fig. 1). The samples are from three islands (Andros, Naxos and Tinos) that form part of the Cycladic Blueschist Unit (CBU) and from Crete.
Metamorphic devolatilization of subducted slabs generates aqueous fluids that ascend into the mantle wedge, driving the partial melting that produces arc magmas. These magmas have oxygen fugacities some 10-1,000 times higher than magmas generated at mid-ocean ridges. Whether this oxidized magmatic character is imparted by slab fluids or is acquired during ascent and interaction with the surrounding mantle or crust is debated. Here we study the petrology of metasedimentary rocks from two Tertiary Aegean subduction complexes in combination with reactive transport modelling to investigate the oxidative potential of the sedimentary rocks that cover slabs. We find that the metasedimentary rocks preserve evidence for fluid-mediated redox reactions and could be highly oxidized. Furthermore, the modelling demonstrates that layers of these oxidized rocks less than about 200 m thick have the capacity to oxidize the ascending slab dehydration flux via redox reactions that remove H 2 , CH 4 and/or H 2 S from the fluids. These fluids can then oxidize the overlying mantle wedge at rates comparable to arc magma generation rates, primarily via reactions involving sulfur species. Oxidized metasedimentary rocks need not generate large amounts of fluid themselves but could instead oxidize slab dehydration fluids ascending through them. Proposed Phanerozoic increases in arc magma oxygen fugacity may reflect the recycling of oxidative weathering products following Neoproterozoic-Palaeozoic marine and atmospheric oxygenation.
Both complexes reflect Tertiary subduction of the African plate beneath Eurasia, which continues today in the Hellenic subduction zone. The subduction complexes of the Aegean are among the most extensively studied and best-exposed on Earth.

oxidized rock types and textures
The oxidized rock types we investigated are likely to be less familiar than the classic blueschists and eclogites of subduction complexes. Because of the high Fe 3+ contents of the rocks, Fe 2+ -rich minerals including almandine-rich garnet porphyroblasts are uncommon or absent. Consequently, many of the rocks have an unremarkable appearance in outcrop. For this reason, we posit that oxidized metasediments have largely escaped attention in petrological studies and, thus, are probably much more common than is recognized at present. Moreover, they are not restricted to the islands we study; for example, oxidized metasedimen-tary (and metaigneous) rocks are known from the CBU on Evia 32 and Sifnos 11,33,34 .
A wide variety of oxidized rock types are exposed, including metabauxite, which is well known from Crete and Naxos (>90 localities on Naxos alone 35 ). The metabauxite protoliths were deep lateritic weathering horizons developed on carbonate sequences. Haematite and rutile are widespread, and relic soil pisolites were preserved in places (Fig. 1a).
We also examined two rocks from Tinos that are intercalated with oxidized metasedimentary layers. One is an epidote-and Na amphibole-rich metabasaltic blueschist. The other is an Na-rich 'albitite' schist with complex mineralogy that includes Na amphibole, jadeite-aegirine (Na-Fe 3+ -Al) clinopyroxene, magnetite and haematite (Fig. 2b). This highly sodic rock is reminiscent of jadeitite and may have similar origins 36 .
Multiple samples preserve textural evidence for reduction during metamorphism. For example, garnets whose cores contain only haematite inclusions can be found surrounded by matrix magnetite, indicating the reduction of haematite to magnetite some time after core growth (Fig. 2a). Garnet molar Fe 3+ decreases ~70% from cores to rims (see Data availability). In another example, haematite-rich domains are cut by a network of Na amphibole-bearing veinlets in which haematite has been converted to coarse magnetite; these veinlets are inferred to be fluid infiltration pathways (Fig. 2b).    Figure 2c shows a garnet core that contains Fe 3+ -rich Na-Ca amphibole and clinopyroxene, whereas the matrix contains glaucophane with Fe 2+ > Fe 3+ . In Fig. 2d, haematite is found in garnet interiors, but the matrix hosts Fe 3+ -poor ilmenite. A redox profile is shown in phyllite from Crete, in which the rock distal to a cross-cutting quartz vein is rich in haematite, whereas the altered selvage rock proximal to the vein lacks haematite and is strongly depleted in iron (Fig. 2e).

f O 2 estimates
We used various combinations of eight independent oxybarometers to estimate metamorphic f O 2 (Methods), including the simple haematite-magnetite and haematite-ilmenite-rutile equilibria that are independent of the activity of H 2 O. When possible, equilibria were applied to assemblages within garnet (preserved as inclusions) and in the rock matrix to evaluate f O 2 changes during metamorphism. Figure 3 shows that metasedimentary lithologies can preserve a highly oxidized signature during subduction. The estimates of f O 2 relative to the fayalite-magnetite-quartz buffer (in log 10 units; ΔFMQ) cover an extraordinary range exceeding seven orders of magnitude; all are at or above the typical ΔFMQ values of arc magmas. Rocks that lack magnetite can have f O 2 values that extend far above the haematite-magnetite buffer, up to ΔFMQ ~9. These include Fe-and Mn-rich haematite-bearing schist, as well as haematite + rutile-bearing metabauxite, quartzite and metapelitic phyllite, and epidote-rich schist (samples 1-5). The extreme f O 2 of the manganiferous metasediments is consistent with comparable localities elsewhere 29 . Another substantial fraction of the rock suite, which includes marble, quartzite, metabasalt intercalated with oxidizing metasediments and metapelitic schist, clusters between ΔFMQ ≈ 2 and the haematite-magnetite buffer (samples [6][7][8][9][10][11][12]. Inclusion assemblages within garnet may record f O 2 values that are ~1-4 log 10 units higher than matrix assemblages. This is consistent with textural evidence and indicates synmetamorphic reduction following initial garnet growth (Fig. 2a,c,d), as has also been documented in metabasalts 11 . Furthermore, reduction need not cause a drop in f O 2 . For example, in Fig. 2b, the magnetite-rich veinlets formed at the expense of the intervening haematite-rich domains, but magnetite and haematite coexist. Thus, although the conversion of haematite to magnetite was proceeding, both phases were present so the f O 2 was constrained to be near the haematite-magnetite buffer as the bulk-rock Fe 2+ /Fe 3+ increased. We infer that reactive fluids ascending from deeper in the slab caused the reduction documented in Figs. 2 and 3 and were oxidized as a result.

Fluid fluxes and metasedimentary rock oxidizing capacity
The time-integrated fluid flux (q TI ; m 3 fluid m −2 rock ) must be used to evaluate changes in redox state due to the infiltration of externally derived reactive fluids (Methods). Consider a rock column with 1 m 2 cross-sectional area extending vertically through a slab. Devolatilization fluids are progressively released and flow up and out of this column into the mantle wedge as the slab (and column) descends. Thus, q TI , as measured at the top of the column, increases with depth. For the subarc depth interval 80-150 km, we took q TI = 220 m 3 fluid m −2 rock (refs. 21,24 Table 1 for ΔFMQ values. local devolatilization of metasediment is over 1,500 times smaller and will therefore be dominated by the external flux ascending from deeper in the slab (Supplementary Information). Some reduction begins in the forearc (Figs. 2 and 3), but most is expected in the subarc where >80% of the fluid release in the 0-150 km depth interval occurs 21,24 . In addition, ~40% of the forearc fraction is derived from metasediments 21,24 ; if they were inherently oxidized, they would release oxidized fluids during dehydration. We used reaction-transport theory to assess whether oxidizing metasediments have the capacity to oxidize these fluids [37][38][39] (Methods). For one-dimensional reactive transport dominated by advection (flow), the reacted and unreacted rocks are separated by a reaction (redox) 'front' that propagates in the flow direction (Fig. 4a). The amount of bulk-rock Fe 2 O 3 reduced, the fluid composition and a redox reaction are required for the calculations (Fe is the most abundant redox-sensitive element). We investigated reductions of 2.5, 5, 10 and 20 wt% bulk-rock Fe 2 O 3 , the likely range for the studied lithologies (Methods), by O-H, C-O-H and S-O-H fluids at representative subarc conditions of 700 °C and 3 GPa. We note that for a given fluid flux, reaction fronts for different chemical or isotopic tracers will travel different distances as a function of their partitioning behaviour 37 .
The reduction of one mole of Fe 2 O 3 by molecular H 2 in aqueous (O-H) fluids can be described by FeO and Fe 2 O 3 are considered to be generically present in oxides or silicates. The fluid composition can be determined if f O 2 is known. Dehydration fluids released from relatively reducing subarc mafic crust and serpentinite are probably in the ΔFMQ range 1 to −2 (refs. 20,21,23 ); we took −1 as representative. The ΔFMQ of the haematite-magnetite buffer is representative of the oxidized metasediments and thus the fluids released from the top of the slab. With these bounding ΔFMQ values, we could quantify the capacity of the metasediments to oxidize the dehydration fluids passing through the slab cover into the mantle wedge at subarc conditions. The calculations were not particularly sensitive to the metasedimentary f O 2 value as long as it was around or above that of haematite-magnetite. We note that some slabs may release more oxidized fluids 9,13,23 ; these would be little modified by flow through the oxidized metasediments. Figure 4b shows how thick metasedimentary layers would need to be to oxidize the slab dehydration flux. For O-H fluids, they are remarkably thin, ranging between ~20 cm (20 wt% Fe 2 O 3 reduced) and ~2 m (2.5 wt% Fe 2 O 3 reduced). This is because the amount of H 2 in the ascending dehydration fluids is small, and the redox buffer capacity of the metasediments is large 39,40 .
In C-O-H fluids, one mole of methane (CH 4 ) will reduce four moles of Fe 2 O 3 We calculated the input CH 4 mole fraction ( X CH 4 ) assuming graphite saturation, which yielded the maximum possible X CH 4 and is thus the most conservative value. Thicker sequences of metasediment are required to oxidize this flux relative to the O-H case (Fig. 4b). This is because the CH 4 concentrations are higher than those of H 2 , and the CH 4 :Fe 2 O 3 ratio is 1:4. Nonetheless, the thicknesses are still only ~10-30 m. As X CH 4 in the input fluid is still relatively small, the evolved CO 2 is also small and will, in general, not precipitate carbonate phases unless they were already stable in the rock. S 2− species will be dominant in S-O-H fluids at low f O 2 (refs. 21,23 ). The oxidation of H 2 S to produce SO x can be represented using SO 2 (ref. 23 ), the most abundant S species in volcanic gases 6 Fluid and/or minerals can host the product Fe and S. For such reactions to go strongly to the right, f O 2 must be above that for f H 2 S = f SO 2 (isofugacity; Methods). As shown in Fig. 3, this would be the case for the highest f O 2 rocks at forearc conditions; the f O 2 for isofugacity drops sharply with increasing pressure (P) and temperature (T) and would be below haematite-magnetite at T > ~650 °C for typical subduction geotherms. Thus, SO x will be important in fluids equilibrated with oxidized subarc metasediments.
For O-H and C-O-H fluids we used molecular H 2 , CH 4 and CO 2 as their thermodynamic and mixing properties are reasonably well established (Methods). For S-O-H fluids, calculations based on aqueous species 21 including H 2 S aq (the DEW model 41 ) tend to give higher total S concentrations than those based on molecular H 2 S. We calculated the input mole fraction X H 2 S at ΔFMQ −1 using the molecular approach for typical mid-ocean ridge basalt 42 at pyrite saturation to represent fluids exiting the top of the metaigneous portion of the slabs. The H 2 S:Fe 2 O 3 ratio will vary depending on the valence of S in SO x ; the maximum ratio is 1:4 (for sulfate). Taking H 2 S:Fe 2 O 3 = 0.25, the metasediment thickness needed to oxidize the

Redox front
Redox front slab flux is greater than that for the O-H or C-O-H cases but is <60 m (Fig. 4b). Using S concentrations from aqueous species calculations 21  Graphitic carbon is not common, occurring in isolated metasedimentary horizons intercalated with more oxidized rocks. The fluids in graphitic rocks need not be very reducing. Assuming the mean regional fluid X CO 2 for the CBU to be 0.007 ± 0.001 (2σ) 43 together with the reaction C + O 2 = CO 2 yielded ΔFMQ ~0.3 ± 0.1 at graphite saturation. As most CBU rocks lack graphitic carbon, this is a minimum estimate. Regardless, as noted above, fluids near FMQ equilibrated with graphitic carbon would have little ability to reduce oxidized metasediments.

The metasedimentary oxidative filter
Our results show that oxidized metasedimentary rocks have the capacity to oxidize the dehydration flux of fluids ascending from slabs at subarc depths. In general, metasedimentary rocks will be at the top of a slab and, thus, will be the last rock type encountered by the fluids before they enter the mantle wedge. Consequently, oxidized metasediments will act as an 'oxidative filter' that imposes a high-f O 2 fingerprint on the slab fluids that ultimately drive flux melting and arc magmatism (Fig. 5). This model can reconcile evidence for the release of relatively reducing (for example, H 2 S-bearing) fluids from subducted metabasalts and serpentinites in the subarc 20,21,44,45 with the presence of an oxidized (for example, sulfate-bearing) slab fluid component 46 in arc lavas 10,14 . In addition, any high-f O 2 fluids generated in underlying hydrothermally altered metabasalt or serpentinite 9,11,13 would pass through the filter with their oxidizing character preserved. Moreover, the filter does not preclude oxidation processes operating in the overlying mantle wedge or lithosphere. Such metasediments could also undergo dehydration or partial melting 45 themselves, contributing to the oxidized flux.
The O-H and C-O-H fluid models require reduction of <10% of an average subducted sedimentary sequence (400 m thick 24 ) to oxidize the slab flux. The S-O-H models require a higher, but still reasonable, proportion of <~15-50%. Consequently, the oxidizing potential of the metasedimentary sequence could be realized even if the rocks experienced thinning by offscraping in an accretionary prism or by compaction, if flow was channelized or if the sequence was not composed entirely of oxidized metasediments. On the other hand, thrust faulting or folding in the subduction channel would lead to greater thicknesses. A further implication is that considerable amounts of surface-derived oxic components in slabs could be subducted past the subarc deep into the mantle, consistent with geochemical modelling 47 and an oxygen mass balance of the Marianas subduction zone 8 .
A rough assessment suggests that fluids ascending from metasediments could oxidize the mantle at a rate of ~4 km 3 yr −1 , comparable to the global arc magma generation rate of ~2.5 km 3 yr −1 (ref. 48  The f O 2 of arc magmas ranges over two to three orders of magnitude 2-5 . At least some of this variability could be related to the oxidative capacity of subducted metasedimentary sequences. Some sequences of oceanic affinity, such as the Palaeozoic Tianshan high-pressure/ ultrahigh-pressure metamorphic belt 49 , are relatively poor in oxic components 21 . By contrast, oxidative weathering-derived detritus would be expected to be important in depositional basins more proximal to continents. The Aegean setting represents a hybrid case that contains both oxidized oceanic (for example, Mn-rich) and weathering-related sedimentary source components. Whether flow is pervasive or channelized to some degree 12,21,50-52 will increase the variability of the redox signal delivered to arcs. Postulated increases in the f O 2 of Phanerozoic island arcs relative to Precambrian equivalents 53 may be related to the global emergence of oxidative weathering driven by Neoproterozoic-Palaeozoic marine and atmospheric oxygenation [54][55][56] , and thus reflect the ultimate recycling of weathering products in subduction zones.

online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41561-022-00904-7.   If the metaigneous devolatilization fluids were already oxidized, they would pass through the metasediments with their oxidizing character intact. In the inset diagram, an accumulation of pillow basalts is shown in green.

Methods
Mineral abbreviations follow Whitney and Evans 57 , except Haem is used for haematite. Amphibole nomenclature follows Hawthorne et al. 58 .
Metamorphism. The CBU underwent high-pressure/low-temperature metamorphism during the Eocene. Metamorphic P-T conditions reached 500-550 °C and 1.5-2.0 GPa (refs. [59][60][61] 67 for all others. Nonideal mixing among these was treated using the molecular models of refs. 68,69 . Thermodynamic data for SO 2 were also included 70 ; the critical pressure was adjusted slightly from 7.87 MPa to 9.87 MPa to better fit available volume relations at elevated P-T conditions 71 using the equation of state of ref. 67 . SO 2 was assumed to mix ideally; this assumption had no impact on the f H2S − f SO2 isofugacity calculations. Equilibria involving the above C-O-H-S species were calculated using Theriak-Domino 72 version 11.03.2020 (Supplementary Table 2). The 'fluid' standard state was adopted, which specifies unit activity of the pure substance at the P-T conditions of interest. O 2 was not considered as a fluid constituent because its concentrations are so low that there is effectively no free O 2 in the fluid 39 .
We estimated fluid S concentrations in two ways. First, we considered the average mid-ocean-ridge basalt composition of ref. 42 at ΔFMQ −1 with enough added S to stabilize pyrite at subarc conditions of 700 °C and 3 GPa (1.895 × 10 −3 molar S/Si ratio; Supplementary Information). Such fluids need not be generated in the metabasalt; they could also be derived from underlying reduced serpentinite that achieved redox equilibrium with metabasalt. Using the fluid standard-state model above and pseudosection calculations following ref. 43 , this yields a total S mole fraction of 5.6 × 10 −4 (almost entirely H 2 S).
Second, we considered the aqueous species treatment used in the DEW model 41 computed in ref. 21 . The standard state is: unit activity for a hypothetical 1 molal solution referenced to infinite dilution at the P-T conditions of interest. This treatment gives an average S mole fraction of 1.6 × 10 −3 for fluids released from metaigneous rocks when integrated over subarc depths of 75-150 km (ref. 21 ). The two approaches should give comparable results as the fluids are supercritical 23 , but the DEW result is larger by a factor of ~3. We attribute this to the fact that DEW considers a much wider range of aqueous species than the molecular model, including Cl complexes, thus facilitating a better and more complete accounting of all sulfur species in fluids. We emphasize that a factor of ~3 difference is still quite good agreement for calculations of this nature, and that our conclusions regarding oxidation by metasediments are unaffected by the choice of model. Figure 4b shows the results of both approaches.
For the f O2 estimates, nonideal mixing was considered for garnet 73,74 , epidote 65 , clinopyroxene 75 , haematite-ilmenite and amphibole; for the last two of these we used the models in the AX 62 program (T.J.B. Holland, University of Cambridge f O2 calculations. Mineral compositions (Data availability) for f O2 estimation were obtained using the JEOL JXA-8530F field-emission gun electron-probe microanalyser (EPMA) at Yale University. Analyses used natural and synthetic standards, off-peak background corrections, a 15 kV accelerating voltage and a 10 nA beam current. The beam diameter ranged from focused to 5 μm, depending on the grain size and mineral type (5 μm was used for all hydrous phases). Rhombohedral oxides in three of the four lowest-f O2 samples (ΔFMQ ∼2-3; jagti90A, jagti106B, jagti154F) contain exsolution lamellae that were reintegrated with the host grain using multiple (up to 12) EPMA analysis spots per sample 76 .
We utilized the following equilibria The last of these was used to compute the f O2 for H 2 S-SO 2 isofugacity ( f H2S = f SO2 ) assuming unit activity of H 2 O (a H2O = 1). Decreasing a H2O decreases the f O2 for H 2 S-SO 2 isofugacity, but low a H2O fluids are unlikely 43 . Donohue and Essene 77 defined equilibrium Ep1 and used it to estimate f O2 for several rocks, including high-pressure calcsilicate from the Bergen Arcs, Norway. They obtained high ΔFMQ in the range of 3.5-4 for the calcsilicate. This rock is from a continental subduction zone and was subjected to Neoproterozoic granulite facies metamorphism and potential metasomatism before the Palaeozoic high-pressure event. Thus, it is from a very different setting than that studied herein.
Most f O2 estimates were made for the Cycladic P-T conditions of 500 °C and 1.5 GPa. Using 550 °C and 2 GPa yielded very similar results. For the low-grade sample from Crete, we used 400 °C and 1.0 GPa. Because we report f O2 estimates in terms of ΔFMQ, the results are not strongly dependent on the P-T conditions used in the calculations (see uncertainty analysis).
The various oxybarometers involve different calculation assumptions but yielded comparable results for a given sample (Fig. 3 and Supplementary Table 1). Large variations in the activity of H 2 O would be expected to produce large variations in f O2 estimates for H 2 O-bearing and H 2 O-absent equilibria in a given sample, but this was not observed. Magnetite is nearly pure, so we take a Mag = 1; the activity would have been less than that if magnetite originally contained impurities such as Ti 4+ that were lost subsequent to high-pressure/low-temperature metamorphism. However, there is no evidence for any such losses and using a Mag < 1 simply increases the haematite-magnetite f O2 estimates.

Uncertainty analysis.
We evaluated the effect of P-T uncertainties on f O2 estimates for individual reactions using a Monte Carlo analysis of the haematite-magnetite buffer with 2σ uncertainties of ±50 °C and ±0.4 GPa. This yielded a standard deviation on ΔFMQ of ±0.22, which is comparable to that of Gerrits et al. 11  smaller ΔFMQ uncertainty of ±0.12. To evaluate the effects of uncertainties on mineral analyses, thermodynamic data and the extent and timing of equilibration, we calculated the standard deviation on ΔFMQ with respect to the mean for four samples with multiple f O2 estimates made using different reactions (jagti75A, jagti90A, jagti106B and jagti154F; 17 total estimates). This yielded a standard deviation of ±0.21. Summing this and the haematite-magnetite uncertainty in quadrature ( √ 0.22 2 + 0.21 2 ) yielded ±0.30, our preferred value for the standard deviation of an individual ΔFMQ estimate. This is far smaller than the observed range of ~9 log 10 units (Fig. 3). The uncertainties for the Mn-rich sample from Andros (jagan1A-1) are deemed larger due to uncertainties in the thermodynamic properties of clinozoisite-epidote-piemontite solid solutions. Nonetheless, the extremely high f O2 of such rocks is clear (ΔFMQ ~9) and is comparable to estimates made on similar rocks elsewhere 29 .
Time-integrated fluid flux calculations and the fluid:rock ratio. The fluid:rock ratio (FRR) is a measure of the amount of fluid infiltration needed to drive a given reaction in a rock, but in general it will underestimate the fluid flux required to propagate a reaction front. Consider a volumetric FRR of 1 m 3 fluid m −3 rock . This seems like a modest number, but it only considers a rock volume in isolation and does not account for the spatial extent of flow. Imagine a 1-km-long vertical column of rock with a 1 m 2 cross-section through which this fluid flows vertically. To react the entire column, 1 m 3 of fluid is required for every 1 m 3 of rock (the FRR). Thus, 1,000 m 3 of fluid is required to react 1,000 m 3 of rock, far greater than the FRR implies. The FRR must be multiplied by the length scale of flow to obtain the time-integrated fluid flux 37 (q TI ), yielding 1, 000 m 3 fluid m −2 rock for our example. The one-dimensional conservation of mass expression describing fluid advection (flow) with chemical reaction for a chemical species, s, in the fluid is 79 in which C s is the concentration of s (mol m −3 fluid ), vx is the average flow velocity in the x direction (m s −1 ), ϕ is the porosity, R s,l is the rate of production or consumption of s by reaction l (mol m −3 fluid s −1 ) and t is time. Assuming local fluidrock chemical equilibrium, one simple overall redox reaction (l = 1) and constant porosity, this expression can be integrated and recast to give q TI at the fluid inlet 39 Here, L is the length of a unit column of rock that has been reacted. A reaction front, which moves in the direction of flow, separates the reacted and unreacted regions (Fig. 4a). The V f term is the molar volume of the fluid, M s is the moles of s produced or consumed per unit volume rock and ΔX s is the difference between the mole fraction of s in the fluid upstream and downstream of the front. The term in parentheses on the right-hand side of the equation is the volumetric FRR (m 3 fluid m −3 rock ). The Lϕ term is negligible if the porosity is small 37-39 ; we set ϕ = 0.001 (ref. 80 ). Even the comparatively large value of ϕ = 0.01 contributes only 1 m 3 m −2 to q TI for L = 100 m front propagation. We take V f =1.495 × 10 −5 m 3 mol −1 , the value for H 2 O at 700 °C and 3.0 GPa (ref. 66 ).
If the metasedimentary sequence is too thin then the buffer capacity of the rocks will be exceeded; some of the incoming fluid will be oxidized, but not all of it. Thus, to be conservative and constrain the maximum thicknesses required, we modelled the largest likely redox state changes. These involved the most reduced reactant valences and the most oxidized product valences: H 2 -H + , C 4− -C 4+ and S 2− -S 6+ . Although there may be variations in speciation on the product side of the overall reactions, our main concern was the nature and concentration of the reducing species that enter the rock. For the O-H, C-O-H and S-O-H fluids we considered, these species were dominantly H 2 , CH 4 and H 2 S, respectively 21,23,39,40 . Aqueous Fe 2+ and Fe 3+ species were not considered as they are likely to be less important for long-distance redox transport 23 .
The molecular fluid compositions were calculated for representative subarc conditions of 700 °C and 3.0 GPa (Supplementary Table 2). As discussed in the main text, input fluids were speciated at ΔFMQ −1, and output fluids at the haematite-magnetite buffer (ΔFMQ 2.5). For example, for the O-H fluid, the input has a mole fraction H 2 ( XH 2 ) of 6.20 × 10 −5 , whereas the output has 1.06 × 10 −6 (>98% of the H 2 is oxidized). This yields a ΔXH 2 value of 6.094 × 10 −5 , which is very close to the input value because the output fluid has little H 2 . For the C-O-H fluid, ΔX CH4 is 1.85 × 10 −5 , and virtually all of the input CH 4 is oxidized to produce CO 2 . At 700 °C and 3.0 GPa, haematite-magnetite is at a higher f O2 than f H2S − f SO2 isofugacity and, thus, SO x will be in greater abundance than H 2 S in the output fluid (Fig. 3). For the fluid standard-state S-O-H fluid, the X H2S in the input fluid is ~7 times greater than the output, yielding ΔX H2S = 4.78 × 10 −4 . This means that ~85% of the input H 2 S is oxidized (to produce SO 2 and S 2 in our calculation) at the haematite-magnetite buffer. The proportion is >95% at ΔFMQ 3.0. Minor H 2 is also present in C-O-H and S-O-H fluids; its oxidation is treated as described above.
The M s value is the amount of rock-hosted Fe 3+ that can be reduced. Sample jagti68B had the lowest iron content in the sample suite (2.9 wt% as Fe 2 O 3 , total 43 ). On average, the metabauxites of Naxos contain ~20 wt% (ref. 35 ). Consequently, we took the generous range of 2.5-20 wt% to represent the amount of Fe 2 O 3 available for reduction. The total iron content could be higher; this range simply represents the mass fraction that is reduced. For a representative rock density of 3,200 kg m −3 , 1 wt% Fe 2 O 3 corresponds to 2.004 × 10 2 moles Fe2O3 m −3 rock . Reaction (1) shows that 1 mole of H 2 will reduce 1 mole of Fe 2 O 3 . As a result, the MH 2 required to reduce 1 wt% of Fe 2 O 3 is 2.004 × 10 2 moles H2 m −3 rock . The MH 2 values for other weight per cent values will scale proportionately. As noted in the text, the M CH4 and M H2S values are 0.25 MH 2 (for example, a CH 4 :Fe 2 O 3 ratio of 1:4).
To estimate the q TI due to devolatilization fluids exiting the top of the slab, we took 38,500 km as the effective trench length 21,24 , a convergence rate of 6.2 cm yr −1 (refs. 21,24 ), a 45° slab dip angle and a fluid density of 1,150 kg m −3 . This density represents the range for H 2 O from 1,090 kg m −3 at 500 °C and 1.5 GPa to 1,210 kg m −3 at 700 °C and 3.0 GPa (ref. 66 ). With these values, we calculated q TI in the range 210 m 3 m −2 (ref. 24 ) to 230 m 3 m −2 (ref. 21 ) for the 80-150 km depth interval; we used 220 m 3 m −2 herein. Note that this value included metasedimentary devolatilization; however, 94% of the flux is generated by underlying altered oceanic crust and serpentinite 21,24 . Subtracting the metasedimentary contribution yielded ~200-220 m 3 m −2 , which is within the uncertainties of the calculation.
The thickness of metasediment required for H 2 oxidation is remarkably small, but CH 4 and H 2 S oxidation will involve greater thicknesses. Considerable amounts of sediment cover slabs worldwide 81 ; we took the average thickness of the metasedimentary sequence to be 400 m (ref. 24 ). This is probably a minimum average, because vertical fluid flow up through such a sequence dipping at 45° along the slab would actually traverse a distance of ~570 m. Steeper dips would lead to even greater thicknesses. This would provide appreciably more oxidizing power than the 400 m we consider; thus, our conclusions are conservative.
The largest subarc time-integrated fluid flux estimates of which we are aware are ~300 m 3 m −2 (ref. 82 ). These would increase the length scales of flow needed for metasedimentary fluid oxidation, but all examples in Fig. 4b would remain <275 m. For example, for representative 5 wt% Fe 2 O 3 reduction via an S-O-H fluid (DEW model), the redox front L is still only ~130 m, much smaller than typical metasediment thicknesses.
If diffusion and mechanical dispersion operated in addition to advection and/or if there were kinetic departures from local fluid-rock equilibrium, redox fronts would be smeared out to some degree but the L is still valid for q TI estimation [37][38][39] . Diffusive mass transfer will occur adjacent to conduits such as veins (Fig. 2e). These features are too small to resolve individually with our methods but are incorporated in a general way by the continuum approach of equation (4). Fluid channelization at larger scales in subduction complexes is well documented 50 , for example in metaigneous rocks below the metasedimentary cover 21,51 , in veins 21,83 and along lithologic contacts 52,83,84 , but reaction progress and oxygen isotope evidence indicate that fluxes within the metasedimentary units of the CBU were largely pervasive 43 .
It is difficult for oxidized fluids to change the redox state of rocks that are already highly oxidized 39,40 (that is, very large fluid fluxes would be needed). Thus, it is not uncommon to find highly oxidized intercalated rock layers with differing f O2 values 32 . However, if such sequences were infiltrated by fluids with substantially lower f O2 , reduction would occur as described herein and in refs. 39,40 .

Rock descriptions.
White mica refers to undifferentiated K-rich and/or Na-rich micas. The samples are listed in the approximate order of decreasing f O2 . Extended Data Fig. 1 shows the sample locations and general geologic relations for Tinos 43,85,86 . 1. Sample jagan1A. Mn-and Fe-rich quartzitic schist, Andros (Fig. 1b).
Composed mainly of quartz, Mn-bearing epidote and garnet, white mica, chlorite and haematite. The epidote can contain an appreciable piemontite (Ca 2 Mn 3+ Al 2 Si 3 O 12 (OH)) component. The most Mn-rich compositions occur in the cores of grains that form aggregates several millimetres in diameter (Fig. 1b). We speculate that these were originally small seafloor Mn nodules.  (Fig. 1e). This is representative of a very common rock type on Tinos, composed predominantly of phengite + epidote + quartz + chlorite + haematite + rutile or sphene ± albite ± carbonates ± Na-bearing amphibole/clinopyroxene. The high-pressure/low-temperature origin of such rocks is clear, as phengite is very Si rich and can attain 3.52 Si per formula unit (see Data availability). Veins and adjacent selvages in such rocks can contain coarse haematite and rutile (Fig. 1f). Location: 37° 32.693′ N, 25° 13.634′ E. 6. Sample jagti134N. Magnetite + haematite + garnet quartzite, Tinos (Fig. 2a).
Rutile and ilmenite coexist in garnet cores, transitioning to sphene in garnet rims and the matrix. The f O2 estimate is for the core assemblage of rutile + ilmenite. Location: 37° 37.375′ N, 25° 02.850′ E.

Data availability
The electron-probe microanalyses of minerals can be downloaded from https:// doi.org/10.5281/zenodo.5809204. The rock samples and petrographic thin sections are in the collections of the Yale Peabody Museum of Natural History, Division of Mineralogy and Meteoritics.