Using a metatexite from the Spanish Betic Cordillera as an example, we show that in situ and otherwise impossible to retrieve compositional information on natural anatectic melts can be reliably gained from experimentally rehomogenized melt inclusions in peritectic garnets. Experiments were conducted on single garnet crystals in a piston cylinder apparatus until the complete homogenization of crystal-bearing melt inclusions at the conditions inferred for the anatexis. The compositions of quenched glasses, representative of the early anatectic melts, are leucogranitic and peraluminous, and differ from those of leucosomes in the host rock. The H2O contents in the glasses suggest that melts formed at low temperature (∼700 °C) may not be as hydrous and mobile as thought. Providing for the first time the precise melt composition (including the volatile components) in the specific anatectic rock under study, our approach improves our understanding of crustal melting and generation of S-type granites.


Partial melting (anatexis) of the metasedimentary crust and extraction of that melt produce S-type granites in the upper crust and granitic leucosomes in migmatites, promoting the chemical differentiation of the continental crust (Sawyer et al., 2011). The presence of melt also causes rock weakening, with dramatic effects on crustal geodynamics and mountain building (Brown et al., 2011). Water plays a significant role during anatexis, determining melting conditions and reactions, as well as the amount of granitic melt produced (Sawyer et al., 2011). It also affects the viscosity of anatectic melts, important for their segregation and ascent rates (Scaillet et al., 1996).

Despite the fact that S-type granites and leucosomes provide important compositional and geodynamic constraints to crustal melting (Clemens and Stevens, 2012; Collins and Richards, 2008; Sawyer, 2008), their nature as primary anatectic melts is called into question by several lines of evidence (Stevens et al., 2007; Sawyer, 2008). Experimental petrology and thermodynamic modeling have been widely and successfully used for inferring the composition of melts and the fluid regime during anatexis (e.g., Montel and Vielzeuf, 1997; Stevens et al., 1997; White et al., 2011). This information is then used in geochemical modeling of migmatitic crust (Sawyer, 2008) and in the modeling of continental crust generation and differentiation (Hacker et al., 2011). However, the comparison between experiments and thermodynamic calculations shows some differences, primarily regarding the melt composition (White et al., 2011). In addition, these unnatural approaches assume melt compositions at equilibrium with the solid residual assemblage, but kinetics may play an important role during crustal anatexis (e.g., Acosta-Vigil et al., 2010). On the natural side, classic melt inclusions (MI), trapped in minerals of felsic igneous rocks during magma crystallization, provide the composition of evolved magmas (Webster and Thomas, 2006), and not of the primary melts as they are produced during anatexis. Therefore, characterizing the precise composition (including the nature and concentration of volatiles) of natural primary anatectic melts remains a major challenge, particularly at the onset of anatexis.

In the past decade, the study of primary glassy MI in peritectic minerals of anatectic enclaves in lavas from southeastern Spain (Acosta-Vigil et al., 2010, and references therein) has provided important information on crustal melting. However, partially melted enclaves in extrusive rocks are exceptionally rare in nature, and cannot be considered common anatectic scenarios. Cesare et al. (2009) showed that peritectic minerals in more conventional migmatitic granulites can trap MI that formed by incongruent melting reactions. However, the geochemical and petrological “message” enclosed in these minute strongboxes has remained unintelligible (Clemens, 2009), mainly because of the analytical difficulty in retrieving the original composition of the trapped melt. The reassessment of regional anatectic terranes worldwide (see Cesare et al., 2011; Ferrero et al., 2012) showed that the presence of MI hosted in peritectic minerals is the rule rather than an exception. This has opened the possibility for conducting routine studies on the geochemical characterization of anatectic melts in natural samples from different geodynamic settings, provided that an appropriate method for retrieving the information within these MI is found.

We took advantage of a new experimental approach for remelting crystallized MI, and present this method along with an unprecedented analytical and geochemical study of primary MI hosted in garnets from a metasedimentary migmatite at Ronda (Betic Cordillera, southern Spain). We show that MI from migmatites provide information on melt compositions and physical properties, fluid regime, and melting mechanisms. We also compare, for the first time in the literature, the compositions of these MI with those of leucosomes from the same rock.


The sample investigated is a stromatic metatexite migmatite (see the GSA Data Repository1 and Fig. DR1 therein). During melting at T (temperature) ∼700 °C and P (pressure) ∼5 kbar, small amounts (<1%) of peritectic garnet grew and trapped primary MI (Figs. 1A and 1B; see the Data Repository). The former presence of melt in the rock is also recorded by mineral pseudomorphs after melt films (Holness et al., 2011) and by leucosomes containing feldspars with euhedral shapes (Figs. 1C and 1D). MI in the migmatite were characterized microstructurally (Cesare et al., 2011; Ferrero et al., 2012); they occur in clusters of tens, preferentially located at the cores of garnets (Fig. 1B), and show a negative crystal shape with an average diameter of ∼5 μm. Most MI are partially crystallized, containing biotite, muscovite, quartz, and glass (Fig. 2), and coexist in the same cluster with nanogranites (i.e., fully crystallized MI; Cesare et al., 2009) and with very rare preserved glassy inclusions.


Previous attempts to re-melt crystallized MI in migmatitic garnets with routine techniques (i.e., the high-T heating stage; Bodnar and Student, 2006) produced extensive decrepitation and interaction between the melt and host, which made the samples unsuitable for a geochemical study (Cesare et al., 2009). We remelted MI using a single-stage piston cylinder apparatus, loading single garnet crystals and powdered silica into Au capsules (for supplementary methods, see the Data Repository). Experiments were run at 5 kbar and temperatures of 700, 750, and 800 °C for 24 h, both under dry conditions and with excess H2O to evaluate the occurrence of H2O diffusion out of the MI during experiments. At 700 °C, some MI are completely rehomogenized, without decrepitation or production of peritectic phases (Fig. 3). When heated at 750 °C and 800 °C, the number of remelted inclusions increases, but they are often characterized by ≤4-μm-long decrepitation cracks and some bubbles containing vapor CO2 with a density of 0.14 g/cm3 (Fig. DR2; Table DR1).

All glasses from both dry and wet experiments, performed at conditions (700 °C, 5 kbar) closest to those inferred for the onset of anatexis, show homogeneous leucogranitic peraluminous compositions with aluminum saturation index [ASI = molar Al2O3/(CaO + Na2O + K2O)] between 1.04 and 1.35 (Table DR2) and H2O contents measured by Raman spectroscopy in the range 3.1–7.6 wt% (Fig. DR3; Table DR2). No significant differences in H2O concentrations were observed for MI remelted under dry and wet conditions.

The progressive increase of the experimental temperature produced an increase in the internal pressure of the volatile-rich MI (Bodnar and Student, 2006), causing MI decrepitation (Fig. DR2) and then depressurization. Because the solubility of CO2 in silicate melts is much lower than that of H2O, and given that the solubility of CO2 in rhyolitic melts decreases with decreasing pressure (Tamic et al., 2001), decrepitation resulted in vesiculation of CO2 bubbles at experimental temperatures of 750 and 800 °C (Fig. DR2). The absence of CO2 bubbles in MI remelted at 700 °C (Fig. 3) indicates that CO2 did not exsolve from the melt of rehomogenized MI, and that our experimental approach, unlike homogenization at ambient pressure (see Bodnar and Student, 2006), maintains the original fluid concentration in the MI.

Using the major element and H2O contents measured in rehomogenized MI and a temperature of 700 °C, we calculated viscosities between 104.1 and 107.1 Pa·s for these low-T anatectic melts (Table DR2).


Partial melting of the metasedimentary protolith produced hydrous peraluminous leucogranitic melts now preserved as MI in peritectic garnet. However, the amount of biotite and muscovite in nanogranites cannot account for all the H2O measured in the glass. Micro-Raman mapping of nanogranites below the garnet surface has documented the presence of micropores and nanopores filled with liquid H2O (Fig. 4), in agreement with the presence of a primary porosity observed during the microstructural studies (Cesare et al., 2011; Ferrero et al., 2012). Therefore, during crystallization of these hydrous melts to produce nanogranites, H2O was exsolved in microbubbles and nanobubbles, but part of the H2O was also consumed by micas, favoring the crystallization of muscovite over K-feldspar (Ferrero et al., 2012). Leucogranitic leucosomes in the migmatite show higher SiO2, TiO2, CaO, and #Mg than the MI, and oxide (wt%) totals close to 100% (Table DR2). Normative compositions of the rehomogenized MI plot close to the 5 kbar haplogranite eutectic at aH2O < 0.5 and overlap those of rare preserved glassy MI and those of some experimental glasses from the literature (Fig. 5). However, these experimental melts produced at variable P-T-aH2O conditions and using different starting materials show a very large spread in composition (Fig. 5). This reinforces the importance of analyzing MI in migmatites in order to make accessible the melt composition for the specific rock under investigation.

Two peraluminous leucosomes plot close to the quartz-orthoclase cotectic far away from the eutectic (Fig. 5), suggesting that they could approach unmodified (i.e., primary) anatectic melts that were produced at higher temperature compared to the MI (i.e., at, or closer to, the peak metamorphic conditions; see Fig. 1). A third leucosome, far from the cotectic line, could have been affected by differentiation processes or by the presence of residual quartz. Alternatively, all these leucosomes may be genetically unrelated to MI and derive from migration and redistribution of melt from other volumes of the migmatitic sequence (Brown et al., 2011).

Similar H2O concentrations measured in rehomogenized MI from dry and wet experiments (Fig. DR3) and the lack of evidence of decrepitation in the analyzed MI (Fig. 3) indicate that the variable H2O concentrations can be considered representative of the melt phase that was trapped in the MI. A large proportion (∼70%) of the rehomogenized MI show primary H2O contents between 3 and 6 wt% (Fig. DR3), indicating an overall H2O-undersaturated environment during melting at ∼700 °C and ∼5 kbar. However, considering (1) the occurrence of CO2 in MI, and of graphite in the rock; (2) that the devolatilization of hydrous phases produces graphite-saturated C-O-H fluids with maximum XH2O ∼ 0.85 at ∼5 kbar and ∼700 °C (Connolly and Cesare, 1993); and (3) that the H2O saturation for granitic melts in equilibrium with C-O-H fluids (with XH2O ∼0.85) at 5 kbar is ∼8 wt% (Tamic et al., 2001), the highest concentrations of H2O measured in MI (∼7–8 wt%; Fig. DR3) suggest that at the very beginning partial melting took place at fluid-saturated conditions. We therefore infer that melting in the metatexite occurred initially and locally at the fluid-saturated solidus owing to the presence of H2O-rich intergranular fluids, and then likely progressed by mica breakdown melting.

Most of the primary H2O contents (3–6 wt%) measured in the MI differ from those of experimentally determined liquidus curves for eutectic or minimum compositions in the subaluminous granite system (Holtz et al., 2001). At the P-T conditions inferred for the onset of melting in the metatexite, these models would predict H2O contents of ∼6–7 wt% in the melts. Although there are uncertainties in both approaches that could lead to a partial overlap of melt H2O contents, we speculate that the discrepancy is due to (1) the compositions of the natural melts that do not correspond exactly to eutectics or minima, and (2) the presence of excess Al2O3 in the peraluminous systems and of other minor components (e.g., FeO) that may modify the solidus and liquidus phase relationships along with the H2O solubility curves.

The calculated viscosity of the melt produced at the onset of melting in the metatexite is approximately two or three orders of magnitude greater than the viscosity generally assumed for granitic melts produced at ∼700 °C (104.3–104.5 Pa·s; Scaillet et al., 1996; Holtz et al., 2001). This would imply much slower rates and much longer time scales for their segregation and ascent.


Based on what the natural samples contain, our study confirms the conclusions of previous experimental and theoretical studies that partial melting of the metasedimentary crust produces peraluminous leucograntic melts. However, our results also document that MI in peritectic minerals represent a unique tool to obtain in situ quantitative information on crustal anatexis and formation of S-type granites, making accessible the precise melt composition for any anatectic terrane. Moreover, the consistency of the compositional data collected refutes concerns (see Clemens, 2009) about the use of MI to recover melt chemistry. Following our new approach, the partially crystallized and nanogranite inclusions in peritectic garnets of migmatites from orogenic belts worldwide can be successfully rehomogenized and analyzed. Because the source is the primary control on granite magma chemistry (Clemens and Stevens, 2012), the characterization in situ (i.e., in the source region) of the first melt droplets produced at the onset of melting is of paramount importance in crustal petrology and will allow us to constrain the geochemical modeling of S-type granitic magmatism and continental crust differentiation using directly measured, natural primary melt compositions. In principle, the composition of the melt at different stages during crustal anatexis could even be identified by analyzing MI in different microstructural locations (e.g., in different mineral hosts) or by studying MI in a protolith affected by variable degrees of melting in a prograde anatectic sequence.

The study of MI will also constrain more precisely the time scales of melt extraction in anatectic terranes. Although these are the first values of viscosity ever produced in migmatites and need support from the study of other MI in comparable samples from other settings, our data suggest that higher values for the viscosity of granitic melts produced at the onset of crustal anatexis should be considered in models of the segregation and ascent of these melts.

We thank R. Angel, J.A.D. Connolly, M. Faccenda, L. Hollister, F. Ridolfi, C. Rosenberg, and D. Vielzeuf for discussion, and G. Stevens and two anonymous reviewers for constructive comments. This work was funded by the Italian Ministry of Education, University and Research (grant PRIN 2007278a22) and by the University of Padua, Progetto di Ateneo CPDA107188/10, to Cesare; by a Ramón y Cajal research contract and grants CGL2007-62992, CTM2005-08071-C03-01, CSD2006-0041 to Acosta-Vigil; and by U.S. National Science Foundation grant EAR-1019770 to Bodnar.

1GSA Data Repository item 2013028, supplementary geological and petrographic information, supplementary methods, Tables DR1–DR3, and Figures DR1–DR4, is available online at www.geosociety.org/pubs/ft2013.htm, or on request from editing@geosociety.org or Documents Secretary, GSA, P.O. Box 9140, Boulder, CO 80301, USA.