Seismogenic plate-boundary faults at accretionary margins (e.g., the Nankai margin, southwest Japan) may occur where the uppermost part of subducting oceanic crust, composed of basaltic rocks, is in contact with the overriding plate of a lithified accretionary prism. The plate-boundary faults in ancient accretionary complexes typically record high-velocity slip under fluid-rich conditions. Although previous studies have emphasized the mechanical significance of fluids in terms of dynamic slip-weakening, the source of fluid in seismogenic subduction zones remains poorly constrained. In this work, we focus on the hydrous smectite in the uppermost oceanic crust, an alteration product of intact basalt before arrival at the trench axis. A comparison between (1) new mineralogical data on basalt drillcore recovered by Integrated Ocean Drilling Program (IODP) Expedition 322 at site C0012, a reference site for subduction input to the Nankai Trough, and (2) mineralogical data on basalt within ancient oceanic crust embedded in a fossil accretionary complex of the Shimanto Belt, southwest Japan, suggests that progressive smectite–chlorite conversion would liberate bound fluids at a rate of 0.34 to 0.65 × 10−14 s−1 along the plate interface. This rate of fluid production appears to be more than an order of magnitude greater than that from other possible sources, including from overlying sediments via smectite–illite conversion and the expulsion of pore fluids, and may facilitate seismic slip along plate-boundary faults.
 To identify the source of fluids within the seismogenic zone, we investigate the uppermost parts of oceanic crust. Before entering the subduction zone, such crust is covered by hemipelagic and terrigenous trench deposits. Accretionary processes at the subduction zone result in the scraping of most of these sediments. At depth within the seismogenic zone, the plate boundary fault is considered to lie near or within the oceanic crust [Ruff, 1989; Kimura and Ludden, 1995; Park et al., 2002]. Accordingly, to identify the primary fluid source along a seismogenic plate interface, it is necessary to examine the dehydration behavior of the uppermost oceanic crust, in addition to underthrusted or accreted sediments, yet this topic has not been considered in previous studies in contrast to the well-documented dehydration process at deeper setting where metamorphic reactions proceed (more than 300°C) [e.g., Peacock and Wang, 1999; Kuwatani et al., 2011]. The aim of the present study is to clarify the effects of diagenesis and low-grade metamorphism on hydrous clay minerals in the uppermost part of the subducting oceanic crust, in the context of their dehydration potential within the seismogenic zone.
2. Alteration of the Uppermost Oceanic Crust Prior to Reaching the Trench
 To constrain the reference physical and chemical states of the uppermost oceanic crust prior to entering the subduction zone, IODP Expedition 322 recovered a basaltic core (∼40 m thick) at site C0012 off the Kii Peninsula, central Japan (Figure 1a) [Saito et al., 2009]. Although the C0012 site locates at the crest of the bathymetric high (Kashinosaki Knoll), this topographic feature is supposed to be tectonic in origin and not due to the intra-plate volcanic activity [Ike et al., 2008]. Petrological/geochemical signatures of the core samples also suggest that the basaltic basement at the drilling site is representative of the tholeiitic basalt in the Shikoku Basin [Saito et al., 2010]. In this study, we analyzed eight samples from this core (collected at depth intervals of ∼4 m from the upper surface of the basaltic basement) (Figure 1b). The samples are composed of plagioclase and clinopyroxene phenocrysts in a clay matrix (Figure 2a). In places, there occur clay-mineral pseudomorphs after olivine (Figure 2b). Given that clay minerals are the main hydrous phase in the samples, they were identified and analyzed as follows. First, we measured the X-ray diffraction (XRD) patterns of air-dried bulk and random mounts of the original samples (Figure 3a). All of the XRD patterns show similar profiles, dominated by reflections corresponding to smectite, plagioclase, and clinopyroxene. The strong, sharp smectite (001) peak has a d-value of 15.5 Å, indicating a two-layer hydrated form of the mineral. Figure 3b shows XRD patterns for oriented mounts of the samples in Mg-saturated and ethylene-glycolated states. The Δ2θ value of ∼16° between the (002) and (005) basal reflections suggests that smectite occurs as an almost pure and discrete phase, without interstratification with chlorite (i.e., smectite content >90%) [Reynolds and Reynolds, 1997].
 The chemical composition of smectite was determined by energy dispersive X-ray spectrometer (EDX) in transmission electron microscopy (TEM). The average structural formula is (Ca0.24K0.074Na0.02)(Al0.18Fe1.18Mg1.61Ti0.02)(Si3.22Al0.78)O10(OH)2 on the basis of 22 negative charges (Table S1 in the auxiliary material). The smectite is trioctahedral and has high contents of Fe and Mg, indicating saponite. Based on this structural formula, the saponite density in a dry state and in a two-layer hydration state is estimated to be 2.92 and 2.24 (g/cm3), respectively. The saponite has a theoretical cation exchange capacity (CEC) of 133 (meq/100g), compared with an average value for bulk samples of ∼27 (meq/100g) (Table S2 in the auxiliary material). From these CEC values, the proportion of saponite in the bulk sample is estimated to be ∼20 wt%. Given the typical density of basalt (3.0 g/cm3), this value (∼20 wt%) is equivalent to ∼27 vol%. Such an abundance of saponite may have resulted from low-temperature interaction between intact basalt and seawater [Alt, 2004] during the long journey from the ridge to the trench (for more than ∼20 Ma based on the depositional age of the sediment just above the basaltic basement) [Saito et al., 2009]. The relatively homogeneous mineralogy throughout the analyzed core samples suggests that seawater circulation infiltrated to a depth of at least 40 m (greater than the drilling depth).
3. Diagenesis of the Oceanic Crust Within the Subduction Zone
 The mineral assemblage of ocean-plate basalt is modified by prograde diagenesis during subduction. An advanced state of such a reaction sequence is recorded by a basalt within an on-land, ancient accretionary complex, the Shimanto belt (Mugi mélange) of southwest Japan (Figure 1b). The basalt passed through the ∼150°C paleo-isotherm and would have been incorporated in the accretionary prism at this temperature via duplex underplating [Matsumura et al., 2003; Ikesawa et al., 2005]. A representative basaltic slab (∼20 m thick), composed of hyaloclastite, massive basalt, pillow basalt, and pillow breccia, is exposed along one of the subduction thrusts. The mineralogy of the basalt is dominated by mixed-layer saponite–chlorite (S–C) (∼30 wt%) followed by albite, calcite, quartz, and minor opaque minerals such as magnetite, hematite, and titanite [Kameda et al., 2011]. S-C occurs as a matrix clay mineral surrounding euhedral albite crystals. XRD analyses of the clay phase reveal that 30%–35% of the layers in the S–C are expandable saponite. This persistence of saponite layers in S–C at ∼150°C is an important factor when considering the source of fluid in the seismogenic zone.
 Kinetic simulations performed by Saffer et al. , focusing on the smectite–illite transition in underthrust sediments along the Muroto and Ashizuri transects across the Nankai Trough, indicate that the proportion of smectite in mixed-layer smectite–illite (S–I) is less than ∼20% at the décollement where it intersects the 150°C isotherm. Although the authors estimated a larger proportion of detrital smectite (45 wt%) in the sediments compared with the proportion of S–C in the present samples, a higher proportion of smectite appears to survive in mixed-layer S–C packets than in S-I. Consequently, the amount of bound water stored in the uppermost oceanic crust is similar to that in sediments across the plate interface. If the interlayer water is stably bounded in S–C, further subduction would release the fluids during progressive chloritization. However, in contrast to the S–I transition, the higher content of structural water in chlorite than in the starting saponite may result in the self-consumption of released water, which is likely to be dependent on the reaction mechanism. According to Bettison-Varga and Mackinnon , there are two potential reaction pathways for the S-C transition, one of which produces a volume increase, and the other a volume decrease. The former reaction requires high influxes of Mg and Fe; however, this is unlikely to occur because the possible sources of such elements (olivine) are largely absent (replaced by S–C) in the analyzed on-land basalts; hence, the latter process is more likely. In this case, the following reaction may occur:
where R in chlorite structural formula is cation occupying octahedral site. To construct the equation (1), we employ the averaged composition of saponite determined by TEM-EDS (Table S1 in the auxiliary material), Si content of chlorite in a metabasic rock (chlorite zone) of the Sanbagawa belt [Okamoto and Toriumi, 2005], and consider that saponite contains 4.5 moles of water per structural formula in a two-layer hydrated state, equivalent to those in aluminous smectite (montmorillonite) [Ransom and Helgeson, 1994]. Equation (1) indicates that 12% of the released water is self-consumed by the chloritization reaction.
 Based on the inferred reaction of saponite-to-chlorite conversion in S-C (equation (1)), we next estimate the rate of fluid production associated with this dehydration reaction within the seismogenic zone. Assuming a negligible increase in porosity, the rate of fluid production Γ(s−1) is expressed as follows [Bethke, 1986; Bekins et al., 1994]:
where H is the volume content of bound water per saponite layer, C is the volume fraction of S–C in the bulk sample, and Φ is porosity. Here we simply assume that H = 0.4 (corresponding to a hydrated two-layer form [Saffer et al., 2008]), 5% volume expansion of water by expulsion [Saffer et al., 2008], and C = 0.27. We adopt 0.1% of porosity, that was reported as an average value of the altered basalt incorporated in the Okitsu mélange of the Shimanto accretionary complex [Kato et al., 2004].
 The first term on the right-hand side in equation (2) is the rate of conversion from saponite to chlorite in S–C; however, there is no appropriate kinetic equation with which to describe this term. Consequently, a somewhat simple situation is considered where the saponite content in S–C shows a linear decrease with the increasing temperature that accompanies downward movement along the plate interface within the seismogenic zone. The initial saponite content in S-C is 35% at the up-dip limit of the seismogenic zone (∼150°C) [Kameda et al., 2011]. Because hydrated saponite layers in S–C (corrensite and randomly interstratified S–C) are likely to disappear above ∼260°C [Hillier, 1993], this temperature condition is defined as an end-point of the dehydration reaction. Accordingly, if the geometry of the accretionary wedge is known (i.e., taper angle, geothermal gradient, and rate of relative plate convergence), the reaction rate can be obtained as a function of these variables.
 Here we discuss fluid production along the Muroto and Ashizuri transects across the Nankai Trough. For this calculation, we use the parameters adopted by Saffer et al. : taper angles of 4–5° and 8–10°, and geothermal gradients of 72 and 48°C/km for the Muroto and Ashizuri transects, respectively, that are derived from reported heat flow values (180 and 120 mW/m2) and a normal heat conductivity of the accretionary wedge (2.5 W/m/K) [Hyndman et al., 1995]. A relative convergence rate of 68 km/Myr (= plate convergence rate+ prism outbuilding rate) is used for both transects. Equation (1) yields fluid production rates of 0.34 to 0.43 × 10−14 (s−1) and 0.46 to 0.65 × 10−14 (s−1) for the Muroto and Ashizuri transects, respectively. In Figure 4a, this rate is plotted on a diagram given by Saffer et al. , showing fluid production from underthrust sediments due to S–I conversion and from a compaction-induced source. Because we assume the reaction occurs at a steady rate, the rate of fluid production is invariable in a landward direction. A comparison of these estimates suggests that around the upper limit of the seismogenic zone, fluid production from basaltic rocks is comparable to that from S–I conversion and sediment compaction, whereas at deeper levels, the former is much greater than the latter.
Saffer et al.  further argued that if sediment is underplated to a certain depth and is incorporated into the overriding accretionary prism, fluid production is strongly suppressed from this point due to a reduction in the subduction velocity of the sediment. Figure 4b shows this scenario when underplating occurs at horizontal distances of 10 and 30 km from the trench axis. In this case, the fluid production rate from the subducting oceanic crust is more than an order of magnitude greater than that from S–I. This situation is achieved if the deeper part (i.e., next to be underplated) of the basaltic crust, which is exposed once the top of the basement has been peeled off, is also altered to produce a similar quantity of hydrous saponite. This scenario is applicable in the case of the analyzed basaltic slab in the Shimanto belt, because the thickness of this slab (∼20 m) is less than the extent of the alteration zone observed at the C0012 site.
 Our mineralogical analyses exhibit that S–C in the uppermost part of an oceanic plate is potentially a significant carrier of fluids into seismogenic subduction zones. However, whether the fluids released during S–C conversion are trapped in the plate boundary and contribute to seismic faulting depends on the competition between the rates of dehydration and diffusion. Quantitative argument on the behavior of such expelled fluids is now in progress, but low permeability of the onland altered basalt (as low as ∼10−19 m2 [Kato et al., 2004]) probably prevents diffusion of dehydrated fluids, so that fluid release associated with S-C conversion result in facilitating seismic slip along subduction faults, as recorded in ancient subduction zones.
 This research used samples and data provided by IODP. We thank Kei Matsuura (The University of Tokyo) for his assistance with the ICP measurement of the samples. We also thank editor M. Wysession and two anonymous reviewers for their critical comments and suggestions. This work was supported by Grants-in-Aid from JSPS for Research Activity Start-up (22840018) and for Scientific Research on Innovative Areas from MEXT (21107005).
 The Editor thanks the two anonymous reviewers for their assistance in evaluating this paper.