Deformation mechanisms in mafic amphibolites and granulites: record from the Semail metamorphic sole during subduction infancy

. This study sheds light on the deformation mechanisms of subducted maﬁc rocks metamorphosed at amphibolite and granulite facies conditions and on their impor-tance for strain accommodation and localization at the top of the slab during subduction infancy. These rocks, namely metamorphic soles, are oceanic slivers stripped from the downgoing slab and accreted below the upper plate mantle wedge during the ﬁrst million years of intraoceanic subduction, when the subduction interface is still warm. Their formation and intense deformation (i.e., shear strain ≥ 5) attest to a systematic and transient coupling between the plates over a restricted time span of ∼ 1 Myr and speciﬁc rheological conditions. Combining microstructural analyses with mineral chemistry constrains grain-scale deformation mechanisms and the rheology of amphibole and amphibolites along the plate interface during early subduction dynamics, as well as the interplay between brittle and ductile deformation, water activity, mineral change, grain size reduction and phase mixing. Results indicate that increasing pressure and temperature conditions and slab dehydration (from amphibolite to granulite facies) lead to the nucleation of mechanically strong phases (garnet, clinopyroxene and amphibole) and rock hardening. Peak conditions (850 ◦ C and 1 GPa) coincide with a pervasive stage of brittle deformation which enables strain localization in the top of the maﬁc slab, and therefore possibly the unit detachment from the slab. In contrast, during early exhumation and cooling (from ∼ 850 down to ∼ 700 ◦ C and 0.7 GPa), the garnet–clinopyroxene-bearing amphibolite experiences extensive retrogression (and ﬂuid ingression) and signiﬁcant strain weakening essentially accommodated in the dissolution–precipitation creep regime including heterogeneous nucleation of ﬁne-grained materials and the activation of grain boundary sliding processes. This deformation mechanism is closely assisted with con-tinuous ﬂuid-driven fracturing throughout the exhumed amphibolite, which contributes to ﬂuid channelization within the amphibolites. These mechanical transitions, coeval with detachment and early exhumation of the high-temperature (HT) metamorphic soles, therefore controlled the viscosity contrast and mechanical coupling across the plate interface during subduction infancy, between the top of the slab and the overlying peridotites. Our ﬁndings may thus apply to other geodynamic environments where similar temperatures, lithologies, ﬂuid circulation and mechanical coupling between maﬁc rocks and peridotites prevail, such as in mature warm subduction zones (e.g., Nankai, Cascadia), in lower continental crust shear zones and oceanic detachments.

Abstract. This study sheds light on the deformation mechanisms of subducted mafic rocks metamorphosed at amphibolite and granulite facies conditions and on their importance for strain accommodation and localization at the top of the slab during subduction infancy. These rocks, namely metamorphic soles, are oceanic slivers stripped from the downgoing slab and accreted below the upper plate mantle wedge during the first million years of intraoceanic subduction, when the subduction interface is still warm. Their formation and intense deformation (i.e., shear strain ≥ 5) attest to a systematic and transient coupling between the plates over a restricted time span of ∼ 1 Myr and specific rheological conditions. Combining microstructural analyses with mineral chemistry constrains grain-scale deformation mechanisms and the rheology of amphibole and amphibolites along the plate interface during early subduction dynamics, as well as the interplay between brittle and ductile deformation, water activity, mineral change, grain size reduction and phase mixing.
Results indicate that increasing pressure and temperature conditions and slab dehydration (from amphibolite to granulite facies) lead to the nucleation of mechanically strong phases (garnet, clinopyroxene and amphibole) and rock hardening. Peak conditions (850 • C and 1 GPa) coincide with a pervasive stage of brittle deformation which enables strain localization in the top of the mafic slab, and therefore possibly the unit detachment from the slab. In contrast, during early exhumation and cooling (from ∼ 850 down to ∼ 700 • C and 0.7 GPa), the garnet-clinopyroxene-bearing amphibolite experiences extensive retrogression (and fluid ingression) and significant strain weakening essentially ac-commodated in the dissolution-precipitation creep regime including heterogeneous nucleation of fine-grained materials and the activation of grain boundary sliding processes. This deformation mechanism is closely assisted with continuous fluid-driven fracturing throughout the exhumed amphibolite, which contributes to fluid channelization within the amphibolites. These mechanical transitions, coeval with detachment and early exhumation of the high-temperature (HT) metamorphic soles, therefore controlled the viscosity contrast and mechanical coupling across the plate interface during subduction infancy, between the top of the slab and the overlying peridotites. Our findings may thus apply to other geodynamic environments where similar temperatures, lithologies, fluid circulation and mechanical coupling between mafic rocks and peridotites prevail, such as in mature warm subduction zones (e.g., Nankai, Cascadia), in lower continental crust shear zones and oceanic detachments.

Introduction
Metamorphic soles underlying ophiolites are 10-100 m thick metamorphosed oceanic slivers stripped from the slab during subduction infancy (Hacker, 1990;Dewey and Casey, 2013;Agard et al., 2016;Rioux et al., 2016;Guilmette et al., 2018). Their accretion to the upper plate mantle wedge (i.e., the future ophiolite; Fig. 1a) is indicative of the rheology of the nascent, still warm interface, which favors strong mechanical coupling between the two plates Soret et al., 2016Soret et al., , 2017. Internal changes in lithologies and metamorphic conditions reveal short-lived, stepwise ac-Published by Copernicus Publications on behalf of the European Geosciences Union.
How deformation is accommodated within the amphibolebearing sole rocks and how mafic slivers get detached from the top of the subducting slab and accreted to the upper plate at T ≥ 700 • C is so far unknown. To investigate the deformation mechanisms accompanying the progressive change in P and T conditions and mineralogy (i.e., essentially amphibole, with varying amounts of plagioclase, clinopyroxene and garnet), we provide detailed microstructural and mineral chemistry data from the well-constrained Semail ophiolite sole. In contrast to plagioclase and clinopyroxene (e.g., Bascou et al., 2002;Rybacki and Dresen, 2004;Hier-Majumder et al., 2005;Dimanov et al., 2007), the physical properties and mechanical behavior of amphibole and amphibolebearing polymineralic aggregates, commonly found over a large range of P -T and water activity conditions, have been far less studied and remain puzzling (see Getsinger and Hirth, 2014). Understanding their deformation mechanisms is therefore expected to shed light on the plate interface rheology during subduction infancy (Soret et al., , 2017Prigent et al., 2018b) or beneath warm subduction zones (where similar temperatures, lithologies and mechanical coupling are expected at ∼ 60 km depth; Abers et al., 2017), as well as on the rheological behavior of amphibole-bearing rocks and strain localization in mid-crustal environments (e.g., continental crust: Imon et al., 2004;Tatham et al., 2008;Getsinger et al., 2013;Giuntoli et al., 2018; oceanic detachment faults or transform faults: Boschi et al., 2006;Escartín et al., 2003).
There are different interpretations for the strong CPO of amphibole and its SPO. The presence of a CPO usually hints at dislocation creep as the dominant deformation mechanism (e.g., Wenk and Christie, 1991). Several studies report intracrystalline deformation in amphibole at high temperature (T ≥ 650-700 • C), compatible with dislocation glide along the most favorable slip systems {hk0} [001] (Berger and Stünitz, 1996;Diaz Aspiroz et al., 2007;Cao et al., 2010;Gomez Barreiro et al., 2010). However, other studies have suggested that amphibole is one of the strongest silicates, with a limited ability to deform by dislocation creep (e.g., Brodie and Rutter, 1985). Amphibole would rather behave as a rigid particle, rotating and fracturing (Hacker and Christie, 1990;Ildefonse et al., 1990;Nyman et al., 1992;Shelley, 1994;Berger and Stünitz, 1996;Imon et al., 2004;Ko and Jung, 2015), and/or dissolving and reprecipitating during deformation (Brodie, 1981;Hacker and Christie, 1990;Berger and Stünitz, 1996;Kruse et al., 1999;Imon et al., 2004;Marti et al., 2017Marti et al., , 2018Giuntoli et al., 2018). While variations in P -T conditions, water activity and grain size are known to control the activation of the different deformation mechanisms (e.g., Brodie and Rutter, 1985), their respective roles have not been clearly assessed in the specific case of amphibole.

The Semail metamorphic sole
The Semail metamorphic sole is a relatively thin unit (commonly ≤ 100 m thick) found at the base of the ≥ 10 km thick Semail ophiolite (Oman and UAE; Fig. 1a-c). This unit results from the juxtaposition of several slices of oceanic crust (Fig. 1c-d;Soret et al., 2017, and references therein) that were buried, strained and metamorphosed against the hot overlying mantle wedge, from granulite-facies (for the uppermost slice) to upper-greenschist-facies conditions (for the lowermost one). In practice, it is convenient to distinguish in the field a high-temperature (HT) sole, found directly below the mylonitic basal peridotite, from a low-temperature (LT) sole onto which the HT sole is thrust. This structural organization is found all along the Semail ophiolite (Fig. 1b-c;Gnos, 1998), from Wadi Tayin (Oman) in the south to Sumeini (Oman) and Asimah-Khubakhib (UAE) in the north. These characteristics are common to most metamorphic soles worldwide, regardless of the nature of the overlying ophiolite (see Agard et al., 2016).  Table 1 for details; (d) close-up view of the tectonic configuration along the nascent plate interface, during the formation and exhumation of the HTa, HTb and LT soles until reaching the steady state (sketch modified after Agard et al., 2016, andSoret et al., 2017); s: sediment, b: basalt, g: gabbro, m: mantle peridotite; (e) amphibole Ti vs. Si content in amphibolite samples from Sumeini (after Soret et al., 2017) outlining the discontinuous decrease in T conditions during the exhumation of the HT metamorphic sole; (f) compilations of P -T estimates and associated P -T paths (after Soret et al., 2017) from the Grt-Cpx amphibolite (HTa) to the Pl-rich amphibolite (HTb) and greenschist facies (LT) metamorphic sole units.
The HT sole is composed of two major subunits (Fig. 1c), being either made of garnet-clinopyroxene amphibolites (HTa, above) or plagioclase-rich amphibolites (HTb, below). The contrasting Ti content of amphibole, which correlates positively with crystallization temperature (Ernst and Liu, 1998) and shows a stepwise decrease from ≥ 0.2 a.p.f.u. (atoms per formula unit) in HTa to ≤ 0.15 a.p.f.u. in HTb ( Fig. 1e; Soret et al., 2017), indicates a ∼ 100 • C gap in peak temperature between the two units. Mapping of the Ti content is used hereafter to track different amphibole generations (Sect. 5.3).
The entire metamorphic sole shows a strong planar fabric with evidence of both pure and simple shear deformation ( Fig. 2a-c). In the LT sole, the lattice-preferred orientation of quartz in the interbedded quartzite is consistent with shear senses observed in the basal banded peridotite (Boudier et al., 1988;Gray et Gregory, 2000;Ambrose et al., 2018 EBSD data were collected using a CamScan X500FE CrystalProbe scanning electron microscope (SEM) at Géosciences Montpellier (France). Measurements were performed on polished thin sections at a 25 mm working distance with an accelerating voltage of 15 kV. For each sample (see map step size in Table 1), amphibole, plagioclase, clinopyroxene and garnet (together with prehnite, epidote and Ti-rich phases such as rutile, titanite and ilmenite) were indexed wherever present. Resulting maps were first filtered using the Channel 5 software suite (Schmidt and Olesen, 1989) to increase the quality of the maps. Isolated pixels were removed and non-indexed pixels with a minimum of six identical neighbors were filled with the same orientation. Maps were then processed using MTEX, a MATLAB toolbox for textural analysis (Hielscher and Schaeben, 2008;Bachmann et al., 2010;Mainprice et al., 2014).
In MTEX, misorientation angles (defined as the lowest rotation angle between two pixels about a common axis that brings two lattices into parallelism; e.g., Lloyd et al., 1997;Wheeler et al., 2001) were used to identify grain boundaries, applying a 10 • threshold between adjacent pixels. Grains with a surface smaller than 10 pixels were discarded. Twins in plagioclase were distinguished from grain boundaries by filtering out the 178 to 180 • misorientations during grain boundary identification. The grain size for each mineral was calculated using the equivalent circle diameter. Pole figures were calculated using the crystallographic orientation of each grain (i.e., one point per grain). All plots are represented in lower hemisphere projections in the specimen reference frame, with contours as multiples of a uniform distribution. Since thin sections were not perfectly cut parallelly to the lineation (hardly visible on the hand specimen), the latter was    defined on the pole figures by the amphibole c-axis maxima (point "X" in Figs. 8, 11 and 12). Mineral crystallographicpreferred orientation (CPO) strength was characterized using the J index (from 1 in the case of random orientation distribution to infinity in the case of an ideal single crystal; Bunge, 1982), calculated from the orientation distribution function (ODF; Mainprice et al., 2014). The shape-preferred orientation (SPO) was calculated from the average orientation of the long axis of the best-fit ellipse of each grain. The aspect ratio (AR) of grains corresponds to the length ratio between the long and short axes of the best-fit ellipse. Kernel average misorientation (KAM) analysis was used to map intragranular misorientations in clinopyroxene, plagioclase and amphibole. The KAM is defined as the average misorientation around a measurement point with respect to the 16 nearestneighbor points in a particular grain.
Mineral abbreviations used throughout the text and figures are given after Whitney and Evans (2010).

Sample overview
Samples from the Sumeini HTa unit are presented structurally from top to bottom, starting from the highest-grade samples collected near the contact with the overlying peridotite to garnet-free samples away from the contact. Mineral occurrences and proportions are given in Table 1. Within 2.5 m from the peridotite, the metamorphic sole is composed of a garnet-clinopyroxene-bearing amphibolite. Peak metamorphic assemblage and microstructures are well preserved in sample SE13-67 ( Fig. 2d; Soret et al., 2017). The garnet-clinopyroxene amphibolite SE13-67 exhibits a mylonitic foliation, locally transected at a relatively low angle (15-25 • ) by C -type shear bands indicating a dextral shear sense in the thin-section frame. The foliation comprises fine-grained aggregates of brown amphibole and clinopyroxene and locally comprises garnet and clinopyroxene porphyroclasts ( Fig. 2d). Three disjunctive domains are distinguished in thin sections: -Area 1 (A1) is dominantly composed of a clinopyroxene aggregate aligned in the foliation and is cut across at low angle by veins of epidote ± prehnite ± apatite (Figs. 2d, 3a).
Plagioclase is either rare or absent in the matrix, and most crystals have not been successfully indexed due to pervasive late sericitization by a minute assemblage of albite + epidote + muscovite. Epidote (≤ 2 vol. %), apatite (≤ 1 vol. %) and prehnite (≤ 5 %), commonly found in veins subparallel to the foliation (Fig. 2d), testify to a late stage of Ca-rich fluid circulation at low-temperature conditions (≤ 700 • C), which is documented across the entire HT metamorphic sole ( Fig. 2b; Soret et al., 2017).
The degree of high-temperature (700-800 • C) retrogression increases steadily towards the base of the HTa unit and is maximum at the contact with the HTb unit (Soret et al., 2017). Rocks located further away from the sole-peridotite contact therefore mostly record a lower temperature associated with post-temperature-peak deformation. About 8 m beneath the peridotite (sample SE13-69; Fig. 4a), the metamorphic sole exhibits a coarser-grained, more conspicuous foliation characterized by higher proportions of brown amphibole (58 %) and plagioclase (14 %), partly replacing clinopyroxene (12 %) and garnet (3 %). About 25 m below the contact (sample SE13-76; Fig. 4b), garnet and clinopyroxene no longer coexist in the metamorphic sole. Amphibole (23 %) has a brown-to-greenish-brown color and is closely associated with plagioclase (64 %), both with coarser grain size.
Amphibole grains are more scattered in the plagioclase-rich matrix and do not commonly form aggregates. The amphibolite also evidences precipitation of yellowish epidote (12 %) aligned along C -type shear bands together with fine-grained amphibole (Fig. 4b).
The underlying HTb unit is made of coarse-grained (typically 1 mm in diameter) plagioclase and greenish-brown amphibole aggregates, roughly in equal proportions, with rare ≤ 50 cm thick intercalations of biotite-plagioclase-rich amphibolite (Soret et al., 2017). Plagioclase-rich and darker amphibole-rich layers alternate throughout this unit (Fig. 2c).
Large amphibole grains display irregular boundaries in all samples, especially where bounded by -altered -plagioclase or prehnite (Figs. 5c,d,. They also show intragranular microfractures which are parallel to cleavage planes or perpendicular to the [001] direction ( Fig. 6a-d). These microfractures locally dissect the large grains, forming smaller grains with straight and sharp intraphase boundaries ( Fig. 6a-d). Later and wider prehnite-filled fractures are also found transecting the amphibolite (Fig. 5b).
The mean aspect ratio of amphibole (1.9 ± 0.6) does not vary with the distance from the peridotite nor with the mineral assemblage (Fig. S2a). Amphibole shape-preferred orientation (SPO), subparallel to the shear plane, is very strong in all samples and its intensity correlates positively with grain size and aspect ratio (Fig. S2a).
In the HTa unit, only rare (< 5 %) amphibole grains show microstructural evidence of intracrystalline plasticity as undulatory or patchy extinction ( Fig. 6a-b) and subgrain boundaries. However, in the sample best preserved from retrogression (SE13-67) the kernel average misorientation (KAM) maps outline the presence of intracrystalline misorientations in most amphibole grains. Most of these misorientations are associated with significant rotations (up to 6 • ; Fig. 6e-i). Grains with the high-microfracture density show the highest misorientation angles or new grain boundaries (red arrows; Fig. 6a, e). In the more retrogressed sam-  ples (SE13-69, SE13-76), amphibole grains also show a high density of microfractures (Fig. 6c, d), which are associated with high intragranular misorientations (Fig. 6h, i). In both samples, some of these intracrystalline misorientations are spatially correlated with undulatory extinction and subgrain formation (Fig. 6c, h), especially in the largest and/or most elongated amphibole grains (i.e., which can locally show intragranular fractures; Fig. 6d-i).
Where associated with amphibole (A2) and/or garnet (A3), the median grain size (51-57 µm) is similar to that of amphibole (Fig. S2d). Yet, clinopyroxene forms in places σ -type porphyroclasts (A2 in SE13-67; Figs. 2d, 3b-c, 5f) with high aspect ratios (≥ 2). They are usually aligned in the foliation plane (Figs. 2d, 5f, S2f), while the SPO for small clinopyroxene grains are more random (Fig. S2f). The core of clinopyroxene porphyroclasts is characterized by undulose extinction and trails of secondary fluid inclu-Solid Earth, 10, 1733-1755, 2019 www.solid-earth.net/10/1733/2019/  sions representing healed fluid-filled microfractures (dashed white arrows; Fig. 5g, h). Minute amphibole is found along the cleavages and the microfractures (white arrows; Fig. 5h). These trails are locally cut across by the wider prehnite-filled fractures. Clinopyroxene mantle also shows elongated fine-grained wings (consistent with dextral sense of shear in the thinsection frame; Fig. 2d) composed of small grains of variable size but with a median value similar to that of the new grains scattered in the matrix and in monomineralic layers (A1). Some of these small grains display undulose extinction (Fig. 5h). However, they rarely contain microfractures. Interstitial small amphibole grains (green arrows; Fig. 5g-h), randomly oriented, are typically found in the porphyroclast mantle and wings at the triple junction of the new clinopyroxene grains (Fig. 5g-h).

Plagioclase
Plagioclase is rare, even in the best-preserved metamorphic sole samples. The analysis of the crystallographic orientation in plagioclase grains is complicated by retrogression; therefore, the results shown below are restricted to the bestpreserved grains and may not be representative of all grains formed at high P and T .
In sample SE13-67, amphibole grains show an homogeneous Ti-high composition (Fig. 6j-k), confirming that these grains formed at or near peak conditions. Interstitial amphibole grains have a Ti-medium composition, indicating a nucleation during the onset of cooling. The opaque mineral and veins found along amphibole grain boundaries and microfractures correspond to ilmenite (green arrow; Fig. 6a, b, j, k) which also formed at slightly lower P -T conditions (Soret et al., 2017).
In the samples marked by a higher degree of retrogression at amphibolite-facies conditions (SE13-69, SE13-76), amphibole grains evidence a much more complex and patchy zoning pattern including smaller Ti-rich areas than in sample SE13-67. These Ti-rich areas are usually truncated by Ti-medium to Ti-low areas, especially along microfractures (red arrows; Fig. 6l-m). In addition to the precipitation of secondary plagioclase and ilmenite at amphibole lobate grain boundaries and/or along the microfractures (Fig. 6a-c), these zoning patterns indicate dissolution-precipitation during cooling (i.e., 800-700 • C). This also further supports the observation that Ti-medium and Ti-low amphibole grains are not in equilibrium with clinopyroxene and garnet but rather with plagioclase.
Clinopyroxene porphyroclasts have a jadeite-rich core similar in composition to that of inclusions in garnet and a thin jadeite-poor rim (Fig. 6a-c). The jadeite-rich core is truncated by microfractures which are both filled by Tirich to Ti-medium amphibole (red arrows; Fig. 6a-c). At the edges of the largest microfractures dissecting the porphyroclast core, compositions are jadeite poor.
Small clinopyroxene grains in mono-and polymineralic layers display a zoning pattern similar to that of the clinopyroxene porphyroclasts, with jadeite-rich cores rimmed and locally truncated by jadeite-poor domains (Fig. 6a-c). As for clinopyroxene porphyroclasts, some of the small grains exhibit amphibole-bearing microfractures in the jadeite-rich core only.
Amphibole J index ranges from 2.86 in the sample SE13-76 to 5.27 in the sample SE13-67 ( Fig. 10a-b). The CPO strength is slightly positively correlated with the modal proportion of amphibole but no systematic correlation is found with the distance to the peridotite ( Fig. 10a-b). In sample SE13-67, the CPO strength is significantly weakened around garnet and clinopyroxene porphyroclasts which show strong foliation deflection. In the more retrogressed sample SE13-69, where garnet and clinopyroxene coexist but do not deflect the foliation plane, amphibole has a strong CPO fabric (J index = 3.67; Fig. 10a-b) as in the garnet-free layers of SE13-67 (Area 2a). The weakest CPO (J index = 2.87; Fig. 10a-b) is observed away from the contact in the garnet-free sample SE13-76, which has the smallest amphibole modal proportion. Moreover, as highlighted with the SPO, amphibole grains with smaller grain size (< 80 µm) and/or aspect ratio (< 1.5) display a similar CPO pattern but significantly weaker in intensity compared to larger or more elongate grains (Fig. 10c).
The misorientation angle distribution of amphibole in all samples highly differs from the theoretical random distribution (i.e., the "Uniform" green curve; Fig. 8b). This is consistent with the strong intensity of amphibole CPO. Small misorientation angles (from 2 to 10 • ) are largely dominant. They correspond to rotations around the [001] axis and to a much lesser extent around the [010] axis (Fig. 8c) concentration is slightly higher in sample SE13-76 compared to the other samples.

Clinopyroxene
Clinopyroxene has a moderate to weak CPO. As in amphibole, the [001] maximum is subparallel to the lineation but more dispersed (Figs. 9d-f, 11a, S5b). The majority of CPO patterns has girdle concentrations of both (100) and (010) poles, except in sample SE13-69 where poles to (010) have a maximum normal to the foliation (Fig. 11a). In SE13-67, the strongest CPO (or J index) is found within monomineralic layers of SE13-67 (J index = 2.61 in A1; Fig. 10a).
In polymineralic aggregates, clinopyroxene J index ranges from 1.21 to 1.73 (Fig. 10b). In contrast with amphibole, both the SPO and CPO strengths of clinopyroxene appear to be also largely insensitive to grain size and aspect ratio (Fig. 10c).
The misorientation angle distribution of clinopyroxene in all samples does not significantly differ from the theoretical random distribution (i.e., the "Uniform" green curve; Fig. 11b), except in the monomineralic layer A1 in sample SE13-67. Small misorientation angles (from 2 to 10 • ) largely prevail. They correspond to rotations essentially around the [010] and the [001] axes (Fig. 11c) mechanical twinning axis, as previously reported for clinopyroxene deformed at high-temperature conditions (Frets et al., 2012).

Plagioclase
Plagioclase has a weak CPO (Figs. 12a, S5c-d). Its strength is similar to that of clinopyroxene with a J index varying between 1.51 and 2.06 (Fig. 10a). Plagioclase grains display a concentration of (001) poles normal to the foliation plane and [100] axes subparallel to the lineation (Fig. 12a). Poles to (010) show no preferred alignment. The CPO strength and patterns do not vary with the distance from the contact nor with the mineral assemblage. The misorientation angle distribution of plagioclase in all samples slightly differs from the theoretical random distribution (i.e., the "Uniform" green curve; Fig. 12b). Small misorientation angles (from 2 to 10 • ) are largely dominant. They correspond to rotations around the twinning axis [100] and to a much lesser extent around the [001] axis (Fig. 12c). However, the [001] axis becomes the main axis of intracrys-  Table 1); (c) the J index of amphibole and clinopyroxene as a function of their aspect ratio and grain size for the samples from the Sumeini cross section. talline misorientations in sample (SE13-76), where plagioclase forms monomineralic aggregates (Fig. 12a).

Discussion
The HTa metamorphic sole unit is characterized by an intense gradient of syn-kinematic retrogression increasing toward its base. This gradient allows us to constrain the combined deformation mechanisms in amphibolites at peak conditions (∼ 850 • C and 1 GPa; in sample SE13-67) and during cooler and more hydrated conditions (∼ 800 and ∼ 750 • C and 0.7 GPa, in samples SE13-69 and SE13-76, respectively).
Amphibolites recording peak conditions (e.g., SE13-67) show large contrasts in mineral strength and crystallographic fabric. Clinopyroxene forms σ -type porphyroclasts and monomineralic fine-grained aggregates wrapped around by monomineralic aggregates of amphibole or polymineralic aggregates of amphibole + clinopyroxene ± plagioclase (Fig. 2d). Where garnet is present, clinopyroxene forms finegrained aggregates partially mixed within an amphibole-rich matrix. Clinopyroxene appears therefore to be mechanically stronger than amphibole and the polymineralic matrix but weaker than garnet. Yet, such a strength contrast disappears in the more retrogressed garnet-clinopyroxene amphibolite (SE13-69) where mineral reactions are more pervasive.
Moreover, clinopyroxene aggregates show ubiquitous plastic deformation (as indicated by the presence of undulose extinction; Fig. 5g-h) but display a weak CPO. By contrast, amphibole aggregates in the amphibolites of the HT metamorphic sole have a very strong CPO (J index ∼ 5; Fig. 10) despite showing no or only little evidence of intracrystalline plastic deformation. Extensive phase mixing is observed throughout the rock, especially in more retrogressed samples. Finally, amphibole, clinopyroxene and garnet underwent intense grain size reduction by microfracturing, closely associated with fluid-driven mineral reactions.
In the light of our observations (see also Table 2) we discuss below the possible deformation mechanisms and their relative contributions to the mechanical behavior of the amphibolites of the soles.
If the strong amphibole CPO and the apparent intracrystalline misorientations (Fig. 8) result from plastic deformation, the misorientation axis should be perpendicular to the direction of crystal slip (e.g., Lloyd et al., 1997). In our study, however, intracrystalline misorientations are essentially accommodated around the [001] axis and only to a lesser extent around the [010] axis (Fig. 8b). Most of the misorientations are therefore incompatible with the activation of the amphibole easy slip system (100)[001] that would account for the observed CPO geometries and intracrystalline misorientations (Fig. 8a). Conversely, the [010] rotation axis (Fig. 8b), lying in the foliation plane perpendicularly to the general shear sense, is compatible with the activation of the  (Bascou et al., 2002;Getsinger and Hirth, 2014;Ingrin et al., 1991). This is consistent with the intracrystalline misorientations accommodated around the [010] axis (Fig. 11c), especially dominant in the fine-grained monomineralic aggregates. However, the CPO strength is low to moderate with some dispersion of the [001] axis even in the fine-grained monomineralic aggregates (Figs. 11a, S5b).

Brittle deformation
The high-microfracture densities in coarse amphibole grains (Figs. 5a, 6a-b) and the general fine-grained size in the bestpreserved sample (SE13-67) attest to grain size reduction through extensive brittle deformation starting at or after peak conditions. Healed microfracturing in secondary amphibole www.solid-earth.net/10/1733/2019/ Solid Earth, 10, 1733-1755, 2019 crystals grown during retrograde P -T conditions (in SE13-69 and SE13-76) indicates that brittle deformation also persisted during cooling. This is further supported by the ubiquitous presence of wider epidote-and prehnite-filled fractures crosscutting the entire HT amphibolite (Fig. 2d). The brittle behavior of amphibole within a rock deforming ductilely at high temperature has already been reported (Nyman et al., 1992;Berger and Stunitz, 1996;Imon et al., 2002;Diaz Aspiroz et al., 2007;Gomez Barreiro et al., 2010;Ko and Jung, 2015;Marti et al., 2017;Giuntoli et al., 2018) but ascribed to significantly lower deformation temperatures (≤ 700 • C). Preferential microfracturing along planes including the long ([001]) crystallographic axis could explain the similar aspect ratios of fine and coarse amphibole grains (essentially ≤ 2-3, as suggested by Gomez Barreiro et al. (2010). Microfracturing along the [001] crystallographic axis together with small rigid rotations (1-3 • ; Fig. 6e-h) may further explain some of the apparent intracrystalline misorientations in amphibole around the [001] rotation axis (Figs. 6j, 8c) Intense grain size reduction through brittle deformation in fact affected the whole rock at or near peak conditions as attested by the presence of fractured and rotated garnet and jadeite-rich clinopyroxene porphyroclasts in sample SE13-67. The lack of microfractures in the new small clinopyroxene suggests that fracturing preferentially occurred along cleavage planes of the porphyroclasts, and the precipitation of minute amphibole along these cleavage planes has probably facilitated the failure. Moreover, the intense brittle deformation followed by small rigid rotation easily explained the general weak of the CPO of clinopyroxene in the amphibolite.
The strong correlation between the aspect ratio and the CPO and SPO intensities of primary and secondary amphibole in all samples (Fig. 10) also advocates for a component of rigid grain rotation in the strain accommodation from peak to retrograde conditions (as previously reported in other settings; Ildefonse et al., 1990;Shelley, 1994;Berger and Stünitz, 1996;Díaz Azpiroz et al., 2007;Tatham et al., 2008;Elyaszadeh et al., 2018). Contrary to clinopyroxene, the high aspect ratio of amphibole allowed the preferential alignment of the long axes parallel to the stretching direction. Yet, the strong SPO and CPO parallel to the lineation in secondary amphibole could also be explained by oriented dissolutionprecipitation during cooling conditions (e.g., Shelley, 1994;Berger and Stünitz, 1996).

Dissolution-precipitation creep and grain boundary sliding
Lobate boundaries and patchy zoning of amphibole defined by Ti-high cores truncated by the precipitation of Ti-medium to Ti-low areas reveal that brittle deformation was closely associated with and/or promoted dissolution and precipitation ( Fig. 6k-m). This is also indicated by the nucleation of Ti-medium amphibole and jadeite-poor clinopyroxene within microfractures and cleavages of diopside-rich clinopyroxene (Figs. 5g, 7). Dissolution-precipitation creep was likely driven by the metastability of the peak assemblage during cooling and hydration and was further enhanced by the permeability created by the microfracturing. Such a positive feedback between viscous and/or brittle deformation and (hydrous) mineral reactions was recently described in shear experiments on amphibolites conducted at 600-800 • C and 0.5-1.5 GPa (Marti et al., 2017(Marti et al., , 2018 formation and dissolution-precipitation creep advocates for transient changes in pore-fluid pressure, either during dehydration (and/or melting) at peak conditions (e.g., Brodie and Rutter, 1985) or during the onset of retrogression (i.e., fluid ingression from adjacent slab dehydration; e.g., in a colder subduction context: Locatelli et al., 2018). The presence of randomly oriented minute amphibole at the grain boundaries of mantled clinopyroxene and the general intense phase mixing in the polymineralic matrix testify to the activation of grain boundary sliding (GBS) as a dominant mechanism accompanying dissolution-precipitation creep syn to post grain size reduction, even in the monomineralic layers (A1; SE13-67). The absence, in the retrogressed HTa amphibolite (sample SE13-69; Fig. 3a), of microstructural evidence (e.g., pressure shadows) for any strength contrast between phases known to have different strength further supports the role of GBS.

Mechanical evolution of the plate interface during subduction infancy
During initial subduction progress, the leading edge of the slab encounters an increasingly warmer, anhydrous mantle wedge acting as a buttress ( Fig. 14a; Agard et al., 2016). Accretion of metamorphic soles to the basal peridotites of the future ophiolite reflects this increased mechanical coupling between the plates across a restricted P -T time window (i.e., when viscosities of the slab crust and overlying mantle are similar; Agard et al., 2016). Our microstructural data allow us to document the evolution of the deformation at the top of the slab (stages A1-A2-B; Figs. 13, 14): -Approximatively from amphibolite to granulite facies and up to 850 ± 50 • C and 1.0 GPa, the slab crust progressively dehydrates and minerals such as amphibole (Ti-medium and Ti-poor) and plagioclase are replaced by mechanically stronger garnet and clinopyroxene in the HTa unit (and Ti-rich amphibole; stage A1; Fig. 13a). The leading edge of the slab thus likely hardens (as shown by the evolution of the strength curve in Fig. 14). When it reaches the strength of the mantle on top, HTa gets stripped and becomes the first slice accreted to the upper plate peridotite (stage A; Fig. 14).
Our observations indicate that a strong amphibole fabric dominates rock anisotropy at the top of the slab. This fabric is controlled by syn-kinematic-oriented growth and rigid grain rotation, as shown by the strong correlation between grain size, aspect ratio and fabric intensity. The onset of brittle deformation in the garnetclinopyroxene amphibolite leads to drastic grain size reduction (GSR; stage A1; Fig. 13a). This brittle deformation, possibly controlled by dehydration embrittlement (Davies, 1999), dehydration-driven stress transfer (Ferrand et al., 2017) and/or fluid ingression from adjacent slab dehydration acts as a precursor for strain localization in the garnet-clinopyroxene amphibolite within www.solid-earth.net/10/1733/2019/ Solid Earth, 10, 1733-1755, 2019 the downgoing slab (and therefore detachment of HTa; Fig. 14). Similarly, fluids released at the plate interface during slab burial are thought to trigger early strain softening of the base of the mantle wedge through GSR promoted by fluid/mantle interaction and dissolutionprecipitation processes ( Fig. 14b; Prigent et al., 2018b;Soret et al., 2016). Overall, these combined processes controlled the onset and location of strong mechanical coupling across the plate interface, between the top of the slab and the base of the nascent mantle wedge (stage A; Fig. 14).
-From 850 down to 750 • C and 1 to 0.7 GPa (stage A2 in Fig. 13a), early exhumation and cooling of the garnet-clinopyroxene-bearing HTa is associated with strong shearing and pervasive retrogression across the plate interface (Soret et al., 2017;Prigent et al., 2018b). Deformation in HTa is extensively accommodated by dissolution and precipitation assisted by grain bound-ary sliding (including mineral reactions, heterogeneous nucleation and phase mixing), leading to a second stage of significant strain softening of the exhuming HTa unit. Dissolution-precipitation creep and GBS are closely related to the brittle deformation events observed from peak conditions and operating until late exhumation (see Sect. 6.3) which facilitate fluid ingression and thereby enhanced mineral reactions and rock reequilibration within HTa. Syn-exhumation mechanical softening occurs also in the basal mantle wedge in the diffusion creep regime as it becomes colder and more metasomatized ( Fig. 14; Agard et al., 2016;Prigent et al., 2018a, b;Soret et al., 2016). Similar weakening on both sides of the plate interface maintains a coupling between HTa and the peridotites and explains their coeval early exhumation (from ∼ 1 to 0.7 GPa). sive fluid-rock interaction transforms the base of HTa into a two-phase matrix composed of syn-kinematic Timedium to Ti-low amphibole and plagioclase, with only minor amounts of biotite and epidote. Meanwhile, the top of the subducting slab (i.e., the future plagioclaserich HTb unit) consists of a similar mineral assemblage (Fig. 14). Our observations show that GBS is progressively inhibited while dissolution-precipitation creep remains the dominant deformation mechanism, thus behaving potentially like the HTb unit. In the absence of significant GBS, the modal increase in (wet) plagioclase has a major softening effect. Mechanical coupling between HTb and HTa and/or HTb and the weakening at the base of the mantle wedge is responsible for the detachment and accretion of HTb (Fig. 14).
These deformation stages between 850 and 750 • C result in the extreme thinning and distribution of similar HT metamorphic soles below the ophiolite (shear strain ≥ 5) and in the thinning of the base of the ophiolite itself (see discussion in Dewey and Casey, 2013).

Mechanical implications for warm subduction zones and oceanic/continental crustal-scale shear zones
We envision at least three major implications of this record of progressive deformation in metamorphic soles: -Since pressure is not expected to significantly influence deformation mechanisms in the ductile regime, we argue that deep recoupling mechanisms in mature warm subduction zones (at T ≥ 800 • C; e.g., Nankai, Cascadia; Wada and Wang, 2009;Abers et al., 2017) could be controlled, as for HTa metamorphic soles, by the extent of fluid-rock interactions and the degree of stability of high-pressure amphibole.
-Dissolution-precipitation creep and associated amphibole-forming reactions coupled with brittle deformation may also control strain localization at high-temperature conditions, and therefore the development of long-lived oceanic detachments and core complexes in slow-spreading environments (Boschi et al., 2006;Escartin et al., 2003;Gülcher et al., 2019), as well as continental crustal-scale shear zones in collisional settings under cooling conditions and exhumation (Getsinger et al., 2013;Giuntoli et al., 2018;Tatham et al., 2008).
-Finally, while many different interpretations have been invoked for the deformation mechanisms of amphibole and amphibolites (e.g., GBS, diffusion creep, dislocation creep; see Sect. 2), such diversity could reflect changes in P -T conditions, water activity and mineral stability. This study outlines that ongoing metamorphic processes and amphibole stability should be carefully taken into account when interpreting deformation mechanisms in amphibole, especially in shear zones experiencing hydration and cooling during exhumation, as a well-developed CPO in amphibole does not systematically testify to dislocation creep and does not exclude a component of fracturing and GBS in sheared amphibolites.

Conclusions
Based on microstructural analysis and mineral chemistry, this study highlights the mineral-scale mechanisms controlling the progressive deformation of sheared amphibolites from the Oman metamorphic sole during subduction infancy and unravels how strain is localized and accommodated in (hydrated) mafic rocks under amphibolite and granulite facies conditions. Metamorphic reactions and pore-fluid pressures driven by changes in P -T conditions and/or water activity exert a key control on the rheology of mafic rocks and the transition between brittle and viscous deformations at hightemperature conditions (as noted experimentally; Marti et al., 2017Marti et al., , 2018. At 850 ± 50 • C and 1 GPa, garnet-clinopyroxene amphibolites accommodate large shear strain essentially through fracturing, grain size reduction and grain boundary sliding, confirming that amphibole and amphibolite have only little ability to deform in the dislocation creep regime. The www.solid-earth.net/10/1733/2019/ Solid Earth, 10, 1733-1755, 2019 exact driving mechanism for the fracturing remains uncertain. Possible explanations include dehydration embrittlement, dehydration-driven stress transfer and/or fluid ingression. Lower temperatures (800-700 • C) and higher hydrated conditions on the return path lead to higher rock disequilibrium and therefore pervasive hydrous reactions. Strain in the exhumed amphibolite is dominantly accommodated in the dissolution-precipitation creep regime assisted and enhanced by a component of brittle deformation. Grain size reduction through fracturing and heterogeneous nucleation enables GBS and significant phase mixing, thereby producing a large mechanical weakening. However, the fabric intensity of amphibole is not significantly affected by the GBS activation (contrary to clinopyroxene and plagioclase) due to its specific high aspect ratio.
These mechanical evolutions, marked by strain hardening during increasing P -T conditions and dehydration and later large strain softening during cooling and hydration, are, respectively, coeval with the detachment and early exhumation of the different amphibolitic units of the HT metamorphic sole. They likely control the extent of the viscosity contrast and, therefore, the mechanical coupling between the top of the slab and the peridotites across the plate interface during subduction infancy Prigent et al., 2018b).
As amphibole is commonly found with plagioclase, clinopyroxene and/or garnet in high-temperature shear zones, these largely pressure-independent findings may be applied to other geodynamic environments where similar temperatures, lithologies, fluid circulation and mechanical coupling between mafic rocks and peridotites prevail, such as in mature warm subduction zones (e.g., Nankai, Cascadia) in lower continental crust shear zones and oceanic detachments.
Author contributions. MS, PA and BD conceived the initial idea of the study. MS, PA, BD and CP participated in the fieldwork. MS conducted EBSD and EMP (electron microprobe) analyses. MS and BI performed EBSD data processing using MTEX. MS prepared the manuscript with contributions from all co-authors.