Role of strain localization and melt flow on exhumation of deeply subducted continental crust

A section of anatectic felsic rocks from a high-pressure (>13 kbar) continental crust (Variscan Bohemian Massif) preserves unique evidence for coupled melt flow and heterogeneous deformation during continental subduction. The section reveals layers of migmatitic granofels interlayered with anatectic banded orthogneiss and other rock types within a single deformation fabric related to the prograde metamorphism. Granofels layers represent high strain zones and have traces of localized porous melt flow that infiltrated the host banded orthogneiss and crystallized granitic melt in the grain interstices. This process is inferred from: (1) gradational contacts between orthogneiss and granofels layers; (2) grain size decrease and crystallographic preferred orientation of major phases, compatible with oriented growth of crystals from interstitial melt during granular flow, accommodated by melt-assisted grain boundary diffusion creep mechanisms; and (3) pressuretemperature equilibria modeling showing that the melts were not generated in situ. We further argue that this porous melt flow, focused along the deformation layering, significantly decreases the strength of the crustal section of the subducting continental lithosphere. As a result, detachment folds develop that decouple the shallower parts of the layered anatectic sequence from the underlying and continuously subducting continental plate, which triggers exhumation of this anatectic sequence. LITHOSPHERE; v. 10; no. 2; p. 217–238; GSA Data Repository Item 2017403 | Published online 10 January 2018 https:// doi .org /10 .1130 /L666 .1


INTRODUCTION
Weakening of rocks by interstitial melt plays a significant role in the dynamics of orogenic systems (e.g., Vanderhaeghe 2009;Jamieson et al., 2010) and can result in large displacements along melt-bearing shear zones bounding regions of rapid uplift (Hollister and Crawford, 1986), extrusion of lower crustal rocks along orogenic sutures (Konopásek and Schulmann, 2005;Möller et al., 2015), or channel flow of partially molten lower crust during intracrustal indentation (Beaumont et al., 2001;Vanderhaeghe et al., 2003;Culshaw et al., 2006;Jamieson et al., 2007;Schulmann et al., 2008a;Maierová et al., 2012). Labrousse et al. (2015) suggested that melting of subducted upper continental crust at ultrahigh-pressure (UHP) conditions triggers its subsequent exhumation. Labrousse et al. (2015) showed that the partial melting of subducting crust in the subduction channel starts at depths exceeding 100 km, depending on the water content. Increasing amount of buoyant partially molten crust subsequently detaches from the subducting plate and rises upward in the form of a diapir-like structure. In contrast, based on a different numerical modeling study, Butler et al. (2015) argued that melting in subducted crust has only minor, if any effect on exhumation, at least for large UHP terranes, for example the Western Gneiss Region in Norway. However, numerical model results might be biased by the approach chosen to introduce the melt-enhanced weakening (Jamieson and Beaumont, 2011;Labrousse et al., 2015;Butler et al., 2015;Maierová et al., 2016). Furthermore, low resolution of these models (~1 km grid spacing) does not allow simulation of mesoscale to microscale deformation mechanisms coupled with porous melt flow and melt segregation (see Wallner and Schmeling, 2016).
One way to improve our understanding of melt-enhanced weakening coupled with melt transfer at high-pressure (HP) to UHP conditions, is to directly study such former anatectic regions in the field. Grain-scale to mesoscale melt topology studies mostly focus on anatectic high-temperature-medium-pressure regions that commonly show protracted episodes of melting (e.g., Sawyer, 2001;Brown and Solar, 1999;Hasalová et al., 2008aHasalová et al., , 2008bKisters et al., 2009;Bonamici and Duebendorfer, 2010;Sawyer, 2014;Kisters, 2012, 2016). In contrast, petrostructural studies of migmatitic UHP terranes have aimed to identify the incipient melting reactions close to the UHP conditions (e.g., Gordon et al., 2013;Chen et al., 2013Ganzhorn et al., 2014;Zhang et al., 2015;Stepanov et al., 2016), and rather neglected the grain-scale melt topology and deformation mechanisms with respect to the kinematic framework of their host. The mechanisms of coupled grain-scale deformation and melt flow in HP felsic crust are so far rather conceptual and inspired by results from material science (e.g., Fusseis et al., 2009), deformation experiments in mantle rocks (e.g., Kohlstedt and Holtzman, 2009), and granular compaction theory (McKenzie, 1984;Veveakis et al., 2015;. It is anticipated that melt transfer in partially molten rocks at high pressures is driven by either the pressure gradients 2 www.gsapubs.org | Volume 10 | Number 1 | LITHOSPHERE during grain boundary sliding at grain scale (e.g., Rosenberg and Handy, 2000;Fusseis et al., 2009;Holtzman et al., 2003;Weinberg and Regenauer-Lieb, 2010;Peč et al., 2015), or by compaction and/or decompaction of the matrix in porous viscoelastic matrix. The mesoscale structures coupled with melt transfer were explained either as ductile compaction instabilities, producing axial planar leucosomes , or porous channels (vug waves; Morgan and Holtzman, 2005; porous waves; Connolly and Podladchikov, 2007;Yarushina et al., 2015).
We have focused on the coupled role of melt transfer and rock deformation in the HP (~9-13 kbar) anatectic metagranitoids of the Eger Crystalline Complex (ECC), in western Czech Republic. The ECC represents a continental crust that has been subducted to UHP conditions underneath a continental plate, which is now exposed at the western margin of the Bohemian massif (e.g., Konopásek and Schulmann, 2005;Kotková and Janák, 2015;Haifler and Kotková, 2016). The ECC consists of granulites and orthogneisses that underwent different degrees of migmatization and deformation, resulting in a wide range of rock types, i.e., banded orthogneiss, migmatite, mylonite, and granofels.
In this study we aim to understand the origin of these different textural types and their relationship. We investigate different possibilities and suggest that the granofels represents banded orthogneiss that was transformed by pervasive grain-scale melt migration along the deformation fabric, as described for example in Závada et al. (2007) or Hasalová et al. (2008a). We demonstrate how such reactive porous flow channels gradually transform the banded orthogneiss into homogeneous granofels through a transitional rock where only relict banding is preserved. We propose that the granofels zones represent principal zones of weakness in the subduction channel and that melt infiltrating the deformed orthogneiss can promote significant weakening of the anatectic multilayer. We also investigate the origin of detachment folds in migmatites that develop above the weakened horizons. We then discuss the role of these folds in the mechanical evolution and exhumation of the HP to UHP rocks.

GEOLOGICAL SETTING
The Variscan orogen exposed in the Bohemian massif was recently interpreted in Schulmann et al. (2009Schulmann et al. ( , 2014 as a deeply eroded Andean-type subduction-collisional orogenic system. This model invokes subduction of the Saxothuringian plate beneath the upper plate represented by the Teplá-Barrandian and Moldanubian crust (Fig. 1A) and large-scale relamination of the subducted and buoyant lower plate felsic crust under the mafic lower crust of the upper plate. The subducted felsic crust likely plunged to upper mantle depths, as indicated by diamondand coesite-bearing felsic gneisses and granulites (Kotková et al., 2011;Haifler and Kotková, 2016;Nasdala and Massonne, 2000;Stöckhert et al., 2001) associated with peridotites and eclogites (Kotková and Janák, 2015). These rocks extruded together to the shallower crustal levels of the orogeny along the former suture interface.
The high pressure (HP) anatectic rocks, including the diamond-bearing granulites, form the ECC, which crops out along the Ohře (Eger) river between the towns of Kadaň and Stráž nad Ohří (Fig. 1). The ECC consists of banded orthogneisses, mylonites, migmatites, granofelses, and granulites, as well as of scarce metasedimentary inclusions. Pressure-temperature (P-T) conditions for the banded orthogneiss were estimated at 9 kbar at 700 °C (Závada et al., 2007), while granofels and granulite reveal much higher pressures of ~15 kbar at 760 °C, and 15-16 kbar at 775-840 °C, respectively (Konopásek et al., 2014;Kotková et al., 1996). U-Pb zircon dating revealed protolith ages clustered around ca. 475 Ma for both granofels and granulite (Zulauf et al., 2002;Konopásek et al., 2014), suggesting that they are derived from granitoid protolith of the same age. These ages are similar to other orthogneisses throughout the Saxothuringian domain (e.g., Tichomirowa et al., 2001;Košler et al., 2004). The high-P metamorphism was dated as 342 ± 1 Ma for the banded orthogneiss (U-Pb monazite ages; Zulauf et al., 2002) and as 339 ± 1.5 Ma and 342 ± 10 Ma for the granulites (U-Pb zircon ages; Kotková et al., 1996;Konopásek et al., 2014). Subsequent very fast cooling rates of 50 +25/-17 °C m.y. -1 were deduced from 40 Ar-39 Ar muscovite and biotite cooling ages of 341 ± 4 Ma in the banded orthogneiss (Zulauf et al., 2002). This rapid exhumation is attributed to a significant shift in the orogen dynamics due to the lower slab breakoff and rebound that squeezed the relaminated felsic crust of the lower plate sideways and upward along the margins and the arc region of the orogen . Such fast exhumation rates are necessary for preservation of the peak assemblages in UHP collisional systems (Hollister, 1982;Korchinski et al., 2014).

METHODOLOGY
Selected samples on the outcrops were collected conventionally or cut out from the outcrop using a portable gasoline-powered diamond saw. Extracted slabs were then dimensioned for polished thin sections. To construct the phase and melt topology maps of the entire thin sections, the cathodoluminescence (CL) image mosaics (~90-150 images) were captured using the CITL Mk5-2 CL microscope (http://www .cathodoluminescence .com/) at 700 mA with a capture time of 1.5 s at the Czech Geological Survey in Prague. The images were then stitched together using the software Kolor Autopano GIGA (www.kolor.com). Resulting image mosaics were subjected to image segmentation in ArcMap 10.2.1 (www.esri.com) using the Maximum likelihood classification tool. The results of image segmentation were then transformed to phase maps in the form of shapefiles (Data Repository [DR] File, Figs. DR1, DR2, DR3 1 ) that were then manually corrected and statistically analyzed in MATLAB™ using the PolyLX MATLAB™ toolbox (Lexa et al., 2005). Further details on the image segmentation and analysis procedure are provided in the DR File. In addition, we used an electron probe micro analyzer (EPMA) CL device as a complementary tool to study the melt topology alongside the compositional maps (Figs. DR1a,DR3) in the Laboratory of Electron Microscopy and Microanalysis at the Institute of Petrology and Structural Geology (IPSG, Faculty of Science, Charles University, Prague, Czech Republic). An integrated CL spectrometry system xCLent IV was attached to the field-emission gun (FEG) EPMA JXA-8530F (with 15 kV and 40 nA).
The crystallographic preferred orientation (CPO) data of K-feldspar, plagioclase, and quartz in two samples (a mylonite, sample EC12-8G and a granofels, sample EC12-8C2) were acquired using the electron backscattered diffraction (EBSD) in the Laboratory of Electron Microscopy and Microanalysis at the IPSG using the scanning electron microscope Tescan Vega equipped by EBSD detector NordlysNano (manufactured by Oxford Instruments). Automatic mapping of the CPO providing the EBSD maps was acquired at accelerating voltage 20 kV, 39 mm working distance, and ~5 nA beam current and 30 µm step size in automatic mode using the Aztec software (www.aztecsoftware .com/). The CPO data were processed and plotted using the MTEX MATLAB™ toolbox (Hielscher and Schaeben, 2008). To obtain a statistically relevant data set of crystallographic orientations, 11 maps in sample EC12-8G and 7 maps in sample EC12-8C2 were acquired in selected domains that are internally homogeneous (e.g., along the monomineralic layer in the mylonite sample). In addition, three maps (two in sample EC12-8G and one in sample EC12-8C2) of representative domains were acquired at a step size of 7 µm to reliably reconstruct the topology of grain-phase boundaries. The grains reconstructed from EBSD 1 GSA Data Repository Item 2017403, additional information about the methodology and extra datasets associated with the manuscript. The first section in the DR file, "Image analysis of melt topology," explains in detail the image analysis procedure (segmentation) of the cathodoluminescence image mosaics. The second section, "Thermodynamic modeling of PT equilibria," includes the details of phase equilibria modeling for the melt content and pressure and temperature estimations. "Additional datasets" shows additional micro structures in hand line-drawings alongside the cathodo luminescence images, wholerock and mineral compositional data, and geochemical diagrams comparing the composition of the different anatectic rock types and the leucosomes with respect to the protolith orthogneisses, is available at http://www.geosociety.org /datarepository /2017, or on request from editing@geosociety.org. data in MTEX (Bachmann et al., 2011) satisfy a condition of high-angle grain boundaries with 10° misorientation angle between neighboring grains (White and White, 1981). The strength of the preferred orientation, the texture J-index, as defined by Bunge (1982), was calculated in MTEX from the orientation distribution function ODF by Fourier method with de la Vallée Poussin kernel and a half width of 15°. Grain size and axial ratio statistics of the major phases were acquired by manual tracing of the CL images in ArcMap 10.2.1 and statistically processed in PolyLX MATLAB™ toolbox, because a large number of grains in the EBSD maps were truncated on the map boundaries. Alignment of melt pockets was statistically evaluated in rose diagrams as length-weighted manually traced two node line segments.
For chemical analyses of major and trace elements, the samples were crushed with a jaw crusher and pulverized in an agate mill in the laboratories of the Czech Geological Survey in Prague. Pulverized and homogenized samples were analyzed in the Bureau Veritas Mineral Laboratories (Vancouver, Canada). The total whole-rock characterization (code LF202; see http:// acmelab.com /pdfs /FeeSchedule -2016.pdf) was done using the lithium borate fusion inductively coupled plasma (ICP)-emission spectrometer (major and minor elements, code LF302) the lithium borate fusion ICP-mass spectrometer (MS) (trace elements, code LF100), and Aqua Regia ICP-ES and ICP-MS (trace elements, code AQ200) methods.
The compositional variation between the selected rock types, evaluated in terms of element transfer in migrating melt, is expressed using the relative abundance of specific element (i) and F v volume factor in the diagrams of Potdevin (e.g., Potdevin and Marquer, 1987;López-Moro, 2012). While the F v is given by the volume ratio between the transformed rock and the initial one, the difference in relative abundance of specific element (i) is expressed by: where Δm i is the relative gain or loss of mass, C i 0 and C i a are the initial and final concentrations, and ρ 0 and ρ a are the densities of these rocks.
The microanalyses of the rock-forming minerals were done at the Laboratory of Electron Microscopy and Microanalysis at the IPSG. They were acquired using the Jeol FEG-EPMA JXA-8530F (accelerating voltage 15 kV, beam current 20 nA). Garnet was analyzed with a focused beam and feldspars and micas were analyzed with a beam defocused to a diameter of 5 µm.

LITHOLOGIES AND DEFORMATION IN THE EGER CRYSTALLINE COMPLEX
We describe here the major types of anatectic rocks studied (Fig. 2), define the terminology used, and recount the succession of deformation fabrics (Fig. 1C). We then describe the structural pattern on the western and eastern part of the studied profile (Figs. 3,4), and detail the mesoscale features related to anatexis and melt migration on a key outcrop (Figs. 5,6), that has an excellent record of the earliest fabrics, in the town of Kadaň (Fig. 1B).

Rock Types
Based on the textural appearance and mineral assemblage we have distinguished five main rock types, described here (see Fig. 2). They are ordered in a sequence characterized by the gradual disappearance of monomineralic layering.
Mylonite is a highly deformed version of the muscovite-biotite-garnet-bearing banded orthogneiss and is compositionally identical to orthogneisses, but displays thinner bands (Fig. 2B).
Migmatite is defined here as a heterogeneous rock that dominantly consists of two parts; neosome, which represents the newly formed or reconstituted part of the migmatite, and a paleosome, which represents the residual part of the migmatite (Figs. 2C, 2D). Leucosome is the lighter colored part of the neosome, consisting dominantly of quartz and feldspars, and represents crystallized product of the melt. The migmatites retain their layered structure, can consist entirely of the neosome, and display 5-20 vol% of leucosomes.
Granulite is a felsic granulite ( Fig. 2H) that differs from granofels by presence of kyanite and a lack of muscovite and leucosomes. It has typical mineral assemblage of K-feldspar (40 vol%), plagioclase (15 vol%), quartz (40 vol%), and garnet, biotite, kyanite, and rutile (<5 vol%), and displays fine-grained equigranular texture. Granulites are typically surrounded by the granofelses. The ECC shows alternating layers of the individual rock types that are parallel to the S 1 compositional layering, most conspicuous in the banded orthogneiss ( Fig. 2A).

Deformation Fabrics
The ECC shows alternating layers/bands of the individual rock types that are parallel to the S 1 compositional layering, most conspicuous in the banded orthogneiss ( Fig. 2A). The ECC shows three major structural domains ( Fig. 1) that differ in the proportion of individual rock types and deformation styles that fold the S 1 fabrics. Domain A, in the western part of the studied profile, is dominated by banded orthogneiss with rare interlayers of the granofels. Domain B, in the east, displays an equal proportion of granofels and orthogneiss together with migmatites. Domain C consists exclusively of granofels and granulite. In this study, we focus on domains A and B (Figs. 1B, 1C).
There are three major deformation fabrics developed throughout the ECC. The S 1 layering is defined by S-type fabrics of banded orthogneisses and mylonites and parallel lithological boundaries of the alternating rock types (Figs. 3A-3C). In few subdomains that are unaffected by later phases of folding (Fig. 3B), the S 1 fabric is flat or dipping to the south at shallow angles. The S 1 layering is folded by upright open to closed F 2 folds associated with S 2 axial planar cleavage that dips south at steep angles (Figs. 1C, 3A, 3D, and 3E). In domain A, the subvertical S 2 foliation is affected by another generation of asymmetric open to closed chevron upright folds F 3 with northwest-southeast-trending fold axial planes S 3 and subvertical hinges (Figs. 1C and 3E). Resulting fold interference pattern creates curtain folds deforming curved and subhorizontal L 2 lineations and F 2 fold hinges (Fig. 3E).
In domain A, S 1 layering is preserved only locally in isolated zones of banded ortho gneiss (Figs. 3A, 3B). These zones are folded by upright open to closed folds, intercalated with narrow channels of granitic melt along both the S 1 layering and S 2 cleavage (Figs. 3A, 3D). The granofels contains leucosome patches and forms isolated layers (maximum 20 cm thick) parallel to the S 1 layering of the banded ortho gneisses (Fig. 3B). In domain B, S 1 layering as well as the lithological contacts between individual rock types are  (Table 3).

(H) Granulite shows fine-grained equigranular texture dominated by felsic minerals.
Strain and melt flow in deeply subducted continental crust | RESEARCH   The hinge zones of these folds display intense L-type fabrics resulting from the transposition of S 1 by S 2 and formation of L 2 intersection subhorizontal lineation. Some hinges of the D 2 folds in migmatitic orthogneisses or mylonites show patches of neosomes aligned parallel with the fold axial cleavage (Fig. 4D). Toward the west, the S 2 cleavage progressively rotates to steep north-northeast-south-southwest-trending orientation (Fig. 1C). Variations in plunge of both the b-axes of the small crenulation folds and the L 2 intersection lineations indicate strongly noncylindrical shape of these folds ( Fig. 4F and stereoplot in Fig. 5).

Macroscopic Relationships of Melt, Rock Types, and Structures
In this study we aim to understand the origin of granofels layers and their role in the exhumation of the entire ECC. Therefore, we have studied in detail the macroscale to microscale deformation pattern and crystallized melt distribution in the granofels layers, as well as in the surrounding banded orthogneiss in the representative outcrop in Kadaň (Fig. 5). This outcrop shows alternating layers (S 1 ) of banded orthogneisses, migmatites, fine-grained mylonites, and granofelses. The entire rock package was folded by north-verging open to closed Z-shaped rounded to chevron folds (Figs. 5 and 6). These asymmetrical folds have wavelengths from several decimeters to meters ( Fig. 5) with upright short limbs and long limbs, gently dipping to the south. Cusp-like interfaces of granofels locally protruding into the hinge zones of the larger folds suggest that this rock was less competent compared to the banded orthogneiss (Ramsay and Huber, 1987). The folds in the banded orthogneiss display asymmetric parasitic folds (wavelength of 5-10 cm) with axial planar patches of migmatites that consist entirely of neosomes (Figs. 4D and 6).
The estimated volume of former melt (as rare leucosomes, Fig. 4A) in banded orthogneiss ranges between 1 and 5 vol% (Fig. 4A). In the migmatites, the melt stored in neosomes (including the leucosomes) is more abundant (<25 vol%; Fig. 4B) than in the orthogneisses, www.gsapubs.org | Volume 10 | Number 1 | LITHOSPHERE and the neosomes form lenses either parallel to S 1 (Figs. 4B and 6) or mark the S 2 axial planar foliation, where they are connected to leucosomes parallel to the S 1 layering (Fig. 4D). The most characteristic features of granofels layers are their diffuse S 1 parallel transitions with the surrounding orthogneisses or mylonites, where the compositional layering typical of the latter rock gradually disappears into the granofels over a distance of ~5-10 cm (Figs. 4E, 7D, and 7E). In thick granofels layers (>2 m), ghosts of crenulated layered fabric can be found (Fig. 4E). These ghosts display faint thin layering in structural continuity with the fabrics of the orthogneisses in the surroundings of the granofelses. Boundaries of banded ortho gneisses and migmatites with the granofelses are locally traced by interconnected network of leucosomes (Fig. 7A). Locally, the hinge zones of the folded granofels layers display a well-developed network of S 1 parallel and axial plane and S 2 parallel leucosomes with estimated proportion of both sets at 15-25 vol% (Fig. 7E). While the S 1 parallel leucosomes typically form isolated lenses, as much as 10 cm long and 1 cm thick, the S 2 axial planar leucosomes are rather continuous and locally intrude the surrounding banded orthogneisses or migmatites (Figs. 7D,7E). Contacts between these two discordant sets of leucosomes are gradational, which suggests that they were interconnected and contained melt at the same time (Vanderhaeghe 1999(Vanderhaeghe , 2001Barraud et al., 2004;Vernon and Paterson, 2001;Sawyer, 2008;Kisters et al., 2009;Hall and Kisters, 2012).

MICROSTRUCTURAL OBSERVATIONS: TRANSITION BETWEEN BANDED ORTHOGNEISS AND GRANOFELS
We cut five rock slabs (A-E) from key domains on the Kadaň outcrop and made thin sections of the different rock and leucosome leucosome in granofels or migmatite patches in banded orthogneisses www.gsapubs.org | Volume 10 | Number 1 | LITHOSPHERE types. Slab A (Fig. 7A) captures a contact between a banded orthogneiss (grading continuously into migmatite) and a granofels, traced by a set of layer-parallel leucosomes. Slab B (Fig.  7B) depicts, from right to left, the disruption of the typical banded orthogneiss by increased anatexis. Slab C (Fig. 7C) represents thinly banded mylonite enclosing two granofels layers. Slab D (Fig. 7D) was taken from a hinge zone of a fold with S 1 crenulations and shows a contact between banded orthogneiss and a granofels, both with leucosomes. Slab E reveals another hinge zone of a fold with a 20-cm-thick granofels layer sandwiched in a fine-grained migmatite (Fig. 7E).
In the following subsections, we first detail the microstructural characteristics of the different rock types, then focus on the melt topology interpretation and melt content quantification from the microstructures for the individual rock types (see the DR File for the details of this method).
Locations of the thin sections used for the microstructural study are displayed in Figure 7.
Migmatites are characterized by continuous disintegration of the alternating layers of K-feldspar, plagioclase, quartz, and micas (Figs. 8D, 8E, 9B, and 9C) due to progressively increasing amount of interstitial grains of unlike phases. With the increasing disintegration, the monomineralic banding becomes disrupted and only ghost relicts of original monomineralic aggregates are preserved (Fig. 8E). Feldspar and quartz aggregates show highly irregular corroded shapes and grain boundaries that are traced by a polyphase fine-grained mixture of albite and/or oligoclase, quartz, and K-feldspar grains (Fig. 9B,  9C). Garnet grains are large, atoll shaped, and contain pools formed by irregularly intergrown grains of K-feldspar, plagioclase, and quartz (Fig. 9C). Muscovite is dispersed throughout the matrix. Increasing degrees of migmatization captured in the sequence of samples EC12-A, EC12-3A2, and EC12-3C2 (Figs. 8A, 8D, 8E; Table 3) are associated with decreasing content of micas from ~20 vol% to 15 vol%. The interstitial grains form irregular pools with tortuous appendices that sometimes line boundaries of several grains in the direction perpendicular to the residual layering (Figs. 9B, 9C). The pools are filled by a mixture of fine-grained anhedral K-feldspar, albite, and quartz. Albite grains in the pools locally form thick overgrowths on the residual oligoclase grains. Some of the pools formed by K-feldspar extend into pockets protruding into residual plagioclase (Figs. 8E, 9B). Other pools contain isolated inclusions or clusters of irregularly shaped, new grains of albite and quartz (Fig. 9C).
Granofels is characterized by fine-grained granoblastic aggregate of feldspars, quartz, and micas (Figs. 8F, 8G, and 9D; Tables 2, 3). Locally, K-feldspar grains form aggregates parallel to the S 1 fabric (Fig. 8F, 8G). Similarly, feldspars and quartz show a weak shape preferred orientation (SPO) parallel with the S 1 layering. K-feldspar and quartz grains are lined by irregular aggregates of quartz, plagioclase, and K-feldspar. Granofels sandwiched between mylonites (EC12-8C2 in Fig. 7C) commonly contains K-feldspar grains that have transgranular cracks oriented at high angles to the faint layering; some are filled by interstitial feldspar, mica, or circular grains of quartz (Fig. 9D). Muscovite crystals have straight boundaries, are dispersed homogeneously throughout the matrix, and are mostly oriented parallel to the S 1 fabric. Garnet grains (<5 mm) are strongly corroded and embayed by a mixture of anhedral feldspars and quartz. Kyanite (<0.2 mm) may be present in matrix as relicts enveloped by plagioclase.

Crystallized Melt Volume and Topology
Many have summarized criteria for recognition of former melt at grain scale in metamorphic rocks (e.g., Závada et al., 2007;Holness and Sawyer, 2008;Hasalová et al., 2008bHasalová et al., , 2011. Using these criteria we interpret the described interstitial films, and pools in the banded orthogneiss, migmatite, and granofels, as former crystallized melt. Estimated melt proportions from the phase maps range between 5 and 16 vol%, are highest in the migmatites and granofels (Table 2), and roughly correlate with the mesoscale estimates based on the proportion of leucosomes on the sampled slabs (Fig. 7). Leucosomes are regarded as macroscopically visible crystallized melt that accumulated in structural mechanical anisotropies (S 1 and S 2 fold cleavage). Leucosomes in the migmatites and granofelses consist of ~0.5 mm large anhedral grains of feldspars and quartz and are considered compositionally equivalent to the microscopic pools in the migmatites. These leucosomes do not contain any pools comparable to those in the migmatites. We have calculated P-T conditions for the banded orthogneiss (pseudosection modeling using Thermocalc, www .thermocalc .com/; data presented in the DR File; Fig. DR4) that indicate melting conditions of ~720-800 °C and 13 kbar. Using the upper temperature limit of 800 °C (at 13 kbar), maximum    www.gsapubs.org | Volume 10 | Number 1 | LITHOSPHERE melt content produced in banded orthogneiss is <~6 vol% (DR File).

SPO of Melt Pockets
The interstitial grains and/or films in the banded orthogneisses and the migmatite either do not show any SPO or very weak SPO parallel to layering (Figs. 8A, 8D, 8E). The SPO of the melt pockets is typically developed in the crenulated and axial parts of the folds (Fig.  8B). In these parts, the orthogneisses show melt pockets forming two maxima aligned at high angles to the layering in both limbs, one of them close to the fold axial cleavage S 2 (Fig. 8B). The granofels sample EC12-K6 from the crenulated fold hinge zone reveals an asymmetric SPO pattern with the major maximum again in the direction of axial cleavage S 2 (Fig. 8G). Similarly, in the mylonite sample EC12-8G and adjacent granofels sample EC12-8C2 (Figs. 8C, 8F), the SPO pattern of melt pockets shows a major maximum parallel with the axial planar cleavage S 2 .

Crystallographic Fabric (CPO) of the Mylonite and Adjacent Granofels
In order to define the dominant deformation mechanisms and evaluate the mechanical coupling of grains that form the solid framework in melt-bearing rocks, the crystallographic orientation of the major phases (K-feldspar, plagioclase, quartz) was measured in mylonite (sample EC12-8G) and adjacent granofels (sample EC12-8C2) (Fig. 10).
In the mylonite sample, the crystallographic preferred orientation (CPO) of K-feldspar measured in a K-feldspar band with subordinate interstitial plagioclase (Fig. 10A) reveals a relatively strong CPO fabric marked by J-index ~ 1.66 and a distinct submaxima of poles to (010) at the periphery of the diagram, centered 30°-40° and 20° inward, in the first and third quadrant, respectively. Two submaxima of [001] directions mark the periphery of the stereoplot in the first and fourth quadrant; another small maximum is subparallel to L 2 . The [100] directions display an incomplete girdle that crosses the center of the diagram at a high angle to the S 1 foliation. This diffuse girdle is marked by a strong submaximum between the second and third quadrant at 50° inward from the periphery. The [101] directions form a strong maximum near the center of the diagram, roughly parallel to both the foliation and lineation L 2 . Another submaximum of this direction is located in the second quadrant.
Interstitial K-feldspar grains (Fig. 10B) measured in the quartz-dominated band of the mylonite (sample EC12-8G) display higher intensity fabrics (J-index ~ 1.71) than for the Plagioclase measured in the K-feldspardominated band (Fig. 10A) reveals a high-intensity fabric (J-index ~ 2.33) marked by maxima of (010) poles in the third quadrant at ~60° from the periphery of the diagram and a smaller submaximum on the periphery of the fourth quadrant rotated 20° clockwise with respect to the foliation plane. The [001] directions reveal discrete submaxima in the first and between third and fourth quadrant. Directions [101] show single maximum located between the second and third quadrant, rotated ~50° inward from the periphery, and [100] directions show a maximum on the periphery of the diagram, rotated ~25° anticlockwise with respect to the foliation plane pole.
In the granofels sample (sample EC12-8C2) (Fig. 10C), the K-feldspar displays lower intensity fabrics (J-index ~ 1.45) than the band forming K-feldspar (Fig. 10A) and interstitial K-feldspar in quartz aggregates in the mylonite sample EC12-8G (Fig. 10B). The K-feldspar fabric pattern in the granofels is comparable to that of interstitial feldspar in quartz of the mylonite EC12-8G (Fig. 10B): well-defined maximum of [100] directions that is perpendicular to the foliation and stretches to an incomplete girdle at high angle to the foliation, and a broad point maximum of [001] that is nearly perpendicular to the lineation L 2 , within the foliation plane. Directions [101] form an elongated maximum between the third and fourth quadrant, while poles to (010) planes form a single maximum near the periphery between the second and third quadrant, centered 20° inward, that stretches to an incomplete girdle perpendicular to the foliation. Plagioclase reveals two distinct submaxima of [001] directions in the foliation plane that are perpendicular and parallel, respectively to the L 2 . Poles to (010) planes show a single maximum in the foliation plane perpendicular to the L 2 . While the directions [101] show single maximum between the second and third quadrant, [100] directions are marked by several submaxima in the center of the diagram, in the second quadrant and between the first and fourth quadrants, rotated 20° inward. Fabric intensity of the plagioclase in the granofels is high (J-index ~ 2.65) and maxima of the [101] are identical to those for stereoplots of interstitial plagioclase in mylonitic K-feldspar bands (Figs. 10A, 10C). Quartz in granofels (Fig. 10C) shows lower fabric intensity (J-index ~ 1.19) than in the mylonitic quartz bands (Fig. 10B) and displays a single maximum of c-axes at the periphery of the diagram in fourth quadrant, accompanied by a broad and broken girdle of {10-10} and {11-20} axes elongated within a plane rotated anticlockwise at 45°, with respect to the foliation plane.

Whole-Rock Geochemistry
To test the relative mobility of melt and/or fluids between the different rock types, we have compared their element budgets using bulkrock geochemistry (Lonka et al., 1998;Hasalová et al., 2008c;Oliot et al., 2014) (Fig. 11;  Fig. DR5). In addition, we have evaluated possible differences in composition of S 1 and S 2 parallel leucosomes (Fig. 11B). Orthogneisses and migmatites (four samples) and granofelses (four samples) are essentially compositionally similar (Fig. 11A), particularly in the major elements (Fig. 11C). Granofels is slightly depleted in FeO, MgO, TiO 2 , and Al 2 O 3 as well as in high field strength elements (HFSEs) (Fig. 11D). The leucosomes display depletion with respect to the banded orthogneiss (sample K1, Fig. 7D) in a wide range of incompatible elements (primarily the rare earth elements, REEs, and Y, Th, U, and Zr), as well as Ti ( Fig. 11B; DR File, Fig. DR5d). The migmatite, marked by leucosomes accumulated along the S 2 cleavage (EC86, Fig. 4D), as compared to banded orthogneiss K1, shows a similar range of depleted elements (REEs, Ca, Mg, Ti, Fe 3+ , Cs); however, this depletion is less significant than that for the leucosomes in the granofelses. In contrast, this migmatite is enriched in U, Ni, Rb, as well as As, Pb, and Mn ( Fig. 11B; Fig. DR5c). Leucosome samples (2E and S 1 + S 2 leucosomes) and the migmatite in the S 2 cleavage (EC86) reveal depletion in TiO 2 and MgO, although the content of all the other major oxides is similar (DR File, Figs. DR5b, 5d). Note: Modal analyses are in volume percent. Corresponding phase maps are shown in Figure 8A-G. Dash indicates no data. Note: Modal analyses are in volume percent. See the Methodology section in text for technical specifications of the device used for acquisition of the compositional maps. EC37 and EC12 in the sample name prefix correspond to the sampled localities 1 and 2 indicated in Figure 1 and shown in Figures 3 and 7, respectively. Mineral abbreviations: ksp-potassium feldspar; pl-plagioclase; q-quartz; g-garnet; bi-biotite; mu-muscovite; kykyanite; ap-apatite.
www.gsapubs.org | Volume 10 | Number 1 | LITHOSPHERE Figure 10. Representative raw electron backscattered diffraction (EBSD) maps with removed bad data and nonindexed points (white pixels) shown alongside the corresponding interpolated phase and grain boundary maps and equal area, upper hemisphere stereographic projections of crystallographic preferred orientation (CPO) for specific microstructural domains in the mylonite (EC12-8G) and the granofels (EC12-8C2) samples. Mineral abbreviations: pl-plagioclase; q-quartz; ksp-potassium feldspar. Max.maximum; Min.-minimum. Grain boundaries, defined by 10° misorientation angle, are shown by black contours in the raw and interpolated EBSD maps. Subgrain boundaries, defined by 2° misorientation angle, are shown by white contours in interpolated maps. Note that the CPOs in stereographic diagrams display data collected in 11 and 7 maps, respectively, for the EC12-8G and EC12-8C2 samples (see Fig.  8 Table DR2). Note that the granofelses and orthogneisses in A and B are represented by four samples in each group. Leucosomes show a range of compositions measured for five samples; one sample is represented by intergrowing leucosomes shown in Figure 7 (sample 2E); a pair of samples was drilled from S 1 parallel and another pair from S 2 parallel leucosomes, respectively, in both slabs D and E (see Figs. 7D, 7E). Isocon diagram (C) of Grant (1986) and diagrams of Potdevin (D) (Potdevin and Marquer, 1987; López-Moro, 2012) illustrate the loss-gain relationships for a range of elements for banded orthogneiss K1 and adjacent granofels K2 (see Fig. 7 for their location). Relative gain or loss of mass Δm i is normalized by the initial mass m i 0 in the volume-composition diagrams. Shaded area corresponds to an artificial fluctuation range, for which the elements are considered relatively immobile. While the isocon diagram (Grant, 1986) displays the comparison of only major elements, the composition-volume diagram (diagrams of Potdevin;Gresens, 1967;Potdevin and Marquer, 1987) shows the selected 35 major and minor elements, including the rare earth elements. Note that the detailed results of composition-volume calculations after Gresens (1967) for other samples shown in Figures 4D and 7A are presented in the DR File (Fig. DR5). HFSE-high field strength elements; LILE-large ion lithophile elements.

for locations of these maps). Mylonite sample EC12-8G is represented by two subdomains representing a K-feldspar-dominated band with interstitial plagioclase (A) and quartz band with interstitial K-feldspar (B). (C) Granofels is shown only in
www.gsapubs.org | Volume 10 | Number 1 | LITHOSPHERE

DISCUSSION
We focus here on the genesis of the granofels layers and we propose their origin through a feedback process linking strain localization and melt flow. Subsequently, we discuss possible deformation mechanisms related to complex melt-solid grains interactions during subduction of continental crust. We discuss the significance of asymmetrical folds and axial leucosomes and their potential role for the exhumation of the entire HP to UHP sequence.

Relationship Between Deformation and Melt Flow
The thickness of the granofels layers and their abundance progressively increase from domain A through B to C, where the granofels alternates with felsic granulites only (Fig. 1). This trend of increasing abundance of granofels at the expense of the layered orthogneiss is further discussed in terms of deformation mechanisms, the mode of grain-scale melt percolation, and pressure gradient.

Are Granofelses Cores of High Strain Zones?
The main question addressed in this paper is related to the origin of the granofels layers with regard to surrounding orthogneiss. Theoretically, there are four possibilities to explain the nature of granofels layers: (1) granofels represents rock of different protolith part of a compositionally heterogeneous crust; (2) granofels is a granodioritic dike intruding the deformed sequence parallel to the main anisotropy; (3) granofels is a diatexite representing a cumulate after melt extraction along the stress gradients; or (4) granofels is a core of a zone of increased strain accommodated by granular flow and porous melt migration that reflects strain partitioning in metagranitoid crust.
Numerous field observations show that the equigranular fine-grained granofels gradually passes to the mylonites and banded orthogneisses, creating characteristic diffuse transitions (Figs. 4E, 7C-7E). Such a diffuse sequence of rock types is commonly interpreted as a result of progressive deformation described in highgrade metagranitoids (Fitz Gerald and Stünitz, 1993;Kruse and Stünitz, 1999;Hasalová et al., 2008a), or alternatively due to an increase in melt fraction (e.g., Toé et al., 2013). Previous studies showed that the deformation gradient in metagranitoids at high temperatures is characterized by the development of augen-gneisses and banded gneisses at higher strains (Handy, 1990;Schulmann and Mlčoch, 1996). Further increase of deformation intensity results in complete disintegration of the layers associated with resorption of older grains and precipitation of new grains from the percolating fluids and/or melts in the core of the shear zones (Lonka et al., 1998;Hasalová et al., 2008aHasalová et al., , 2008bOliot et al., 2014). Typical microstructural features of this sequence are the decrease of grain size and progressive disintegration of the compositional layering by crystallization of interstitial unlike phases in host banded aggregates.
The microstructural trend described here is compatible with studied banded orthogneiss, mylonite, and fine-grained granofels sequence (Figs. 8 and 10; Table 1). Therefore, we argue that the granofels represents a zone of highest deformation in heterogeneously deformed metagranitoids, where the banding is disrupted by a combination of melt influx, straining and mechanical mixing (hypothesis 4). In contrast, the hypothesis (1) of a different protolith of studied rock types can be ruled out based on similarity in bulk rock chemistry (Fig. 11) and by similar Cambrian zircon ages (Zulauf et al., 2002;Konopásek et al., 2014) of both granofelses and orthogneisses that suggest similar protolith and origin of these rocks. Similarly, the hypothesis (2) of granodioritic vein intrusions is unlikely due to the lack of euhedral magmatic phenocrysts and related magmatic fabrics, diffuse boundaries with surrounding banded rock types (Fig. 4E), and relatively strong CPO of feldspars that is similar in both the granofels and the orthogneiss (Fig. 9). Hypothesis 3, explaining the granofelses as diatexites with cumulate composition due to melt extraction, cannot be completely ruled out. The latter is compatible with the refractory composition of the granofelses and leucosomes that are depleted in incompatible elements, suggesting their crystallization from residual melts. However, movement of the crystal melt mush en masse is not compatible with the development of CPO, abundant diffuse boundaries with surrounding banded ortho gneisses and mylonites, and lack of rotated ghosts or stringers of the host rocks in the granofelses.
There are two possible explanations for the relatively strong CPO fabric intensity in the granofelses: (1) inheritance of deformation fabrics from the orthogneisses that originated during earlier solid-state deformation and were later transformed to granofelses by pervasive grain-scale melt flow, and (2) oriented growth of grains from interstitial melt during granular flow.
The partial inheritance of fabrics is possible, as the fabric patterns of all the phases in both rock types are to some extent similar and granofelses locally contain strings of corroded K-feldspar grains. However, the most likely explanation for the origin of relatively strong fabric of interstitial feldspars in both the mylonite and the granofels is oriented growth of grains by crystallization from interstitial fluid and/or melt (Menegon et al., 2008;Boneh et al., 2017). This is supported by similarity of CPO fabrics and fabric intensities displayed by interstitial feldspars in host mylonite bands, i.e., plagioclase in K-feldspar and K-feldspar in quartz bands (Figs. 10A, 10B), to those displayed by plagioclase and K-feldspar in the granofels, respectively (Fig. 10C). Another argument for the oriented growth rather than dislocation creep is the lack of subgrains or internal distortion of the small interstitial feldspar grains, which is compatible with their crystallization from interstitial melt (Fig. 12, inset). At the same time, however, CPO fabrics in the mylonitic bands formed dominantly by K-feldspar and quartz (Figs. 10A, 10B), respectively, are compatible with activity of dislocation creep. The CPO pattern in quartz for the mylonite sample EC12-8G can be interpreted in terms of basal <a> slip system. As some of the poles to (0001) occur in the center of the pole figure (Fig. 10B), they suggest a development of a wide single or crossed-girdle c-axis pattern and consequently confirm the correctness of our speculated kinematic section. The dominant slip system in band forming K-feldspar in the mylonite is not clear.

Are Granofelses the Loci of Melt Flow?
We described microstructural features such as narrow films, pools, corroded grain boundaries, and albitic rims (Fig. 9) that indicate the presence of former melt (e.g., Sawyer, 2001;Holness and Sawyer, 2008;Hasalová et al., 2008a). We are aware of the oversimplification of our melt topology quantification approach as the crystallizing melt usually forms overgrowths on preexisting grains and cannot be always identified. The amount of crystallized melt is variable, being highest in migmatites and granofelses (>16 vol%; Table 2) compared to 5-10 vol% in the orthogneisses. However, it has been shown that the amount of crystallized new magmatic phases is not necessarily proportional to the amount of melt that passed through the rock, as the shear zones can be regarded as open systems (Hasalová et al., 2008b(Hasalová et al., , 2008c(Hasalová et al., , 2011Goncalves et al., 2012). Alternatively, large interstitial oligoclase grains may have crystallized in larger pools, as the composition of minerals crystallized from melt is also related to the available pore space (Holness and Sawyer, 2008). Therefore, the quantified melt contents in this study should be regarded as minimum estimates.
In this context, the identified melt in the granofels may represent only a fraction of the total residual melt, as the fine-grained granoblastic texture reflects crystallization of interstitial melt and resorption of old grains during earlier stages of shear zone history. Our P-T calculations show that at peak conditions (800 °C at 13 kbar) <~6 vol% melt can be produced. Larger melt fraction can be produced only if hydrous fluids were introduced (Weinberg and Hasalová, 2015). We therefore infer that most of the interstitial melt in these rocks originated in external sources and probably migrated from deeper parts of the subducted crust.
If we assume flux of simple granitic melt through the banded orthogneiss, we would expect the granofels to display composition enriched in incompatible elements, specifically the large ion lithophile elements (LILEs), with respect to the surrounding orthogneisses. Instead, comparison of the banded orthogneiss with the granofels shows little differences in major and trace elements. Granofels is slightly depleted in FeO, MgO, TiO 2 , and Al 2 O 3 , reflecting less ferromagnesian minerals like biotite, as well as in incompatible REEs, Cs and Th, and U (HFSEs; Fig. 11). Such a small difference suggests a relatively similar protolith. However, depletion in the incompatible elements can be considered as a result of partial melting of granitic sources, responsible for dissolution of accessory, REE-bearing minerals such as zircon, monazite, xenotime, and apatite in melt and its escape from the system (Janoušek et al., 2004;Lexa et al., 2011). We note that similar chemical changes can result from hydrous fluid percolation (Marquer et al., 1994;Goncalves et al., 2012). However, the described P-T conditions of 700-800 °C and >13 kbar do not allow free hydrous fluid percolation and impose stability of silicate melt. Because we cannot constrain total melt transferred in the granofelsic shear zones, or pinpoint precisely the amount of melt that crystallized in respective studied anatectic domains (namely in the granofelses), it is difficult to speculate on the mode of melt assisted deformation. Melt is usually considered as effectively mobile when its fraction reaches the rheological critical melt percentage (RCMP) of ~25%-30% or similar solid to liquid transition of ~40%-60%, giving rise to diatexites (Arzi, 1978;Rutter and Neumann, 1995;Vigneresse et al., 1996;Vigneresse and Tikoff, 1999;Rosenberg and Handy, 2005). The RCMP could have been easily reached locally and transiently in structurally defined channels represented by S 1 layering and S 2 fold axial cleavage (Fig. 3D, 3E, 4B, and 4D). Nevertheless, destruction of the deformation layering may occur at lower melt fractions if the deformation mechanism is dominated by grain boundary sliding (granular flow) (Sawyer, 1994(Sawyer, , 1996White et al., 2005). Therefore, complete fabric disintegration and granofels formation can be also explained by continuous reactive porous grain-scale melt flow through dynamic porosity during the granular flow of the grains below the melt connectivity threshold marked by 7-10 vol% (Rosenberg and Handy, 2005). Such transition from banded mylonites to nebulitic migmatites (or granofelses) was convincingly demonstrated by Hasalová et al. (2008aHasalová et al. ( , 2008bHasalová et al. ( , 2008c for a middle crustal, melt-assisted shear zone. In summary, the granofelses can be interpreted as diatexites that resulted from reactive porous melt flow at grain scale that percolated through highly strained domains in the layered orthogneisses and mylonites (Fig. 12). This scenario assumes that the orthogneisses and mylonites reflect strain gradients in the subducted slab that developed before the partial melting and that such strain gradients were precursory for later pervasive melt flux through the layered sequence.
The melt flow in the high strain zones is accompanied by oriented growth by crystallization of new interstitial phases in host aggregates producing the CPO, higher amount of former melt, depletion of HFSEs and REEs, and high total amount of transferred melt (Fig. 12). Such percolating melt was not in chemical and textural equilibrium with the host rock, and therefore dissolved residual grains and precipitated new interstitial phases (Figs. 9 and 12; Table 2). Such a reactive porous flow is associated with chemical modification of the deformed host rock, as shown in Figure 11.

Mode of Porous Grain-Scale Flow in Granofels
Our microstructural and CPO study  shows that the banded and mylonitic ortho gneisses were deformed by dislocation creep combined with melt-assisted grain boundary sliding. In contrast, the granofels underwent melt-enhanced granular flow by unspecified grain-size sensitive deformation mechanisms, and its CPO likely reflects primarily the oriented growth of interstitial grains from melt (Fig. 10). This transition is usually associated with increase of dynamic porosity by dilation of grain boundaries and enhancement of grainscale fluid and/or melt flow (e.g., Závada et al., 2007;Schulmann et al., 2008b;Fusseis et al., 2009;Okudaira et al., 2015).
The increasing abundance of granofels at the expense of the banded orthogneisses along the studied profile (domains A→C, Fig. 1) can be attributed to porous melt flux from higher melt pressures in deeper levels of the subducting continental slab undergoing partial melting. At high pore pressures, the porous melt flow can www.gsapubs.org | Volume 10 | Number 1 | LITHOSPHERE be also driven by larger scale pressure gradients, giving rise to the porous waves or vug waves (Morgan and Holtzman, 2005;Connolly and Podladchikov, 2007), or by compaction-induced segregation . Porous waves are simply described as plume-or lense-shaped batches of melt that migrate through intergranular network of viscoelastic anatectic rocks, where RCMP is transiently reached in the cores of such instabilities.

Geodynamic Consequences of the Localized Porous Melt Flow
We discuss here two contrasting roles of the granofelses during continental subduction: (1) they record the partitioning or localization of deformation in high strain zones that focused porous melt flow and weakening of the crust; and (2) they played a key role in the exhumation of high-P rocks through development of detachment folds decoupled from continuously subducting continental crust.

Porous Melt Flow and Strain Partitioning During Continental Subduction
The studied sequence of anatectic rocks is interpreted to result from continental subduction of felsic crust of the Saxothuringian lower plate beneath the Teplá-Barrandian upper plate (Konopásek and Schulmann, 2005;Schulmann et al., 2014). These reconstructions are compatible with the P-T conditions estimated for the banded orthogneiss (700°-750° at 9-13 kbar), granofels (780° at 15 kbar), and felsic granulite (750-800 °C at 14-16 kbar; Závada et al., 2007;Konopásek et al., 2014;Kotková et al., 1996) that were subducted into mantle depths and rapidly exhumed. When these data are plotted in the P-T diagram (Fig. 13A) they outline a typical subduction-type metamorphic gradient with the lowest peak P-T conditions typically developed in the banded orthogneisses and higher P-T conditions in the granofels and granulite. The microstructural evolution from banded orthogneisses to granofels and the prograde P-T trend presented here are consistent with the model of Saxothuringian continental subduction of Chopin et al. (2012); however, in contrast to the latter study, we propose that the granofelses represent zones of active deformation coupled with melt percolation enabling the metamorphic reequilibration. The P-T signature in Figure 13A corresponds to the regional distribution of orthogneiss, granofels, and granuliterich domains presented in a schematic crustalscale section (Fig. 13B). The cross section can be subdivided into two regions characterized as domain A and B in the west (increasing proportion of granofels with depth) and domain C (granofels-granulite region) in the east. We interpret domain C as a source region for the melts percolating upward along zones of localized deformation producing the granofels layers. This model is fundamentally similar to that explaining Paleozoic eclogitization of subducted Precambrian granulites in the Norwegian Western Gneiss Region (Austrheim, 1987) and the Cenozoic eclogitization of Permian granulites in Italian Alps (Lardeaux and Spalla, 1991). Austrheim (1987) and Lardeaux and Spalla (1991) proposed that the fluids percolated through subducted continental crust along shear zones, leading to progressive eclogitization of deformed and recrystallized anhydrous granulites. The eclogitization is heterogeneous and leads to development of high-P assemblages in localized shear zones surrounding low-strain domains characterized by lower metamorphic grade. In analogy to these studies, we argue that the granofels layers represent Carboniferous high strain zones that equilibrated at higher P-T conditions by percolating melt, compared to the Cambrian-Ordovician orthogneisses, which correspond to lowstrain domains, thereby preserving lower P-T conditions.

Exhumation of High-P Rocks through Development of Detachment Folds Decoupled from Subducting Crust
The asymmetrical folding of the layered sequence of anatectic rocks can explain melt percolation triggered exhumation of the lowdensity felsic crust during continental subduction (Labrousse et al., 2011(Labrousse et al., , 2015. The style of folding and deformation is also similar to that explaining exhumation of eclogites in Scandinavian Caledonides that were exhumed due to progressive weakening of continental crust along eclogite facies (hydrated) shear zones (Jolivet et al., 2005). This style, schematically presented by Labrousse et al. (2011Labrousse et al. ( , 2015, is similar to the detachment folds developed above a weak layer (e.g., Blay et al., 1977;Solar and Brown, 2001;Barraud et al., 2004). This folding can be observed on the outcrop scale in domain A (Fig. 4), where the décollement planes probably coincide with boundaries between rheologically weak migmatites or granofelses impregnated by residual interstitial melt and the orthogneisses. This interpretation is supported by melt topology identified in folded granofels layers marked by shape-preferred melt films oriented preferentially parallel to the axial planar cleavage of asymmetrical folds (Figs. 8B, 8G). Strain incompatibilities during the folding around the arrested tips of deformation zones (Fig. 14) imply stress gradients that redistribute the melts on the grain scale to mesoscale (Kruger and Kisters, 2016). At first, dilatancy can account for thickening of the S 1 parallel leucosomes and influx of weak and meltcontaining granofels into the hinge domains of incipient folds (Fig. 7E, 1 in Fig. 14). In the next stage, as the rate of amplification decreases and the system passes to post-buckle flattening (Price and Cosgrove, 1990), the melt pockets and leucosomes align parallel to the axial planar fold cleavage, similar to the melt pockets in the matrix (2 in Fig. 14), in form of compaction-induced melt bands (Figs. 4C, 4D, 8B, and 8G;. The melt-coated planes of this planar cleavage are regarded as new transfer zones that can promote retroward transport of the more deeply subducted and folded domains atop the shallower sequences of the same slab (Fig. 13B).
The detachment folding seems to be a key mechanism responsible for exhumation of high-P partially molten sequences along the subduction channel. The exhumation process in the  (Figs. 4C, 4D, 7D, and 7E). Incipient folding at the arrested tips of shear zones (or layer-bound porous waves) at first creates dilatancy that collects residual melt from the granofelses (and orthogneisses) and thickens the S 1 leucosomes (arrow 1 in inset). The second stage of detachment folding is associated with coalescence of melt pockets (arrow 2 in inset) and leucosomes aligned parallel to S 2 fold axial planes. Melt-lubricated fold planes in last stages of the folding represent new zones of weakness that can act as thrusts and accommodate backflow of the more deeply subducted portions of the crust. Gray layers show the granofelses, orange layers are banded orthogneisses, and migrating melt is shown in red.
www.gsapubs.org | Volume 10 | Number 1 | LITHOSPHERE ECC is probably also driven by buoyancy, as modeled by Labrousse et al. (2015), but likely takes place already at much shallower depths of ~50 km. While the model of Labrousse et al. (2015) shows geometries of exhumed partially molten crustal batches at the scale of several kilometers, the detachment folding can explain backflow of partially molten rocks along the subduction interface at scales ranging from meters to hundreds of meters.

CONCLUSIONS
A model of porous melt flow along the metamorphic layering in high-P felsic crust was established based on detailed field-based structural analysis, microscale melt topology mapping, crystallographic fabric acquisition, thermodynamic equilibria modeling, and whole-rock composition correlations. We have defined zones of porous melt flow that coincide with zones of concentrated deformation and are approximated by the extent of granoblastic granofels layers alternating with migmatized banded orthogneisses and ortho gneiss mylonites. Increasing abundance and thickness of these granofels layers are associated with increasing melt pore pressure during granular flow at progressively increasing depths of the subducted slab. Porous melt flow in these deformation zones is reflected by depleted content in incompatible elements in the granofels in contrast to the host-rock orthogneisses, and ubiquitous evidence for grain-scale textures indicative of mineral resorption and precipitation of new grains from interstitial melt. In our model, the granofels layers represent the pathways of throughgoing porous melt flow and are regarded as shear zones that equilibrated at progressively increasing prograde metamorphic conditions of the subduction channel. Moreover, weakening of subducting felsic crust is promoted by redistribution of melt along the shear zones traced by the granofels layers. This weakening facilitated detachment folding at the arrested tips of the shear zones, where melt is driven into the hinge zones and is injected along the axial cleavage of amplifying folds. The detachment folding is regarded as a key mechanism responsible for initiation of exhumation of deeply subducted and partially molten continental crust.