Alkaline basalts with geochemical features similar to those of intraplate ocean islands have been emplaced along the main trace of the Tepic-Zacoalco rift (TZR), a unique tectonic structure of the western Trans-Mexican Volcanic Belt in which extension is superimposed to a convergent margin. New geochemical and petrologic data on mafic volcanic rocks along the rift indicate the existence of a highly heterogeneous pre-subduction mantle wedge that has been slightly overprinted by slab-derived chemical agents. Most mafic volcanic rocks display geochemical and isotopic compositions that are indistinguishable from those of the Pacific islands Socorro and Isabel, and confirm the existence of an ancient, recycled, high-μ component (HIMU; μ = 238U/204Pb) in their mantle source. Olivines separated from samples carrying the HIMU signature have NiO and CaO contents similar to olivines from mid-ocean ridge basalt (MORB), indicating that the source of enrichment must be entirely hosted in peridotite. In contrast, more evolved rocks within the TZR have stronger subduction signatures and water contents, and display a distinctive isotopic array that points to slab-derived contributions. Olivines from these rocks are slightly less forsteritic but also extend to higher NiO and lower CaO contents than those from more mafic magmas, suggesting provenance from a secondary pyroxenite source. The overall geochemical evidence thus indicates that the pre-subduction background mantle wedge in western Mexico must be identical, and just as diverse, as that below the Pacific basin. Extension-driven mantle upwelling in a continental setting can only melt a dry peridotitic mantle to its lowest extents, and therefore preferentially sample its most enriched and easily fusible components. Yet the addition of even a small amount of slab-derived silica promotes a secondary petrologic transformation to pyroxene-rich lithologies that upon melting create magmas with compositions that are more akin to a volcanic arc setting.

The geochemical information on oceanic and continental basalts indicates that the Earth’s mantle is compositionally heterogeneous at all scales. Once considered to be constituted exclusively of a monotonous peridotite, our current understanding portrays a mantle that may be just as diverse as the continental crust (Hofmann, 2003). Extraction of the continents during the early evolution of the Earth left a lasting mark in the depletion of incompatible elements from the upper mantle, but the establishment of plate tectonics, and subduction in particular, has been continuously bringing enriched crustal materials back to the Earth’s interior, creating lithologic and chemical heterogeneities that are gradually disaggregated into the mantle by convective stirring and chemical diffusion (Hofmann and Hart, 1978; Katz and Rudge, 2011). Much of what we know about mantle heterogeneities comes from the study of oceanic hotspot volcanoes such as the Hawaiian Islands, not only because they are disconnected from global mantle convection, but also because they are conveniently isolated from the geochemical complexities found at subduction zones or underneath ancient continents. Nonetheless, isotopically enriched basalts similar to ocean-island basalts (OIBs) have been sporadically sampled at spreading centers that are far from any hotspot (Donnelly et al., 2004), and they have also been recognized at some convergent margins that are clearly isolated from the influence of deep-seated plumes (Hickey-Vargas et al., 1986; Langmuir et al., 2006; Gómez-Tuena et al., 2007b; Hoernle et al., 2008). Consequently, it has long remained unclear whether these unusual rock compositions are related to peculiar mantle compositions at the local scale, or if these mantle heterogeneities are a pervasive feature of the Earth’s upper mantle that can be preferentially sampled by more subtle tectonic perturbations.

Alkaline basalts with similar geochemical features to OIB have been identified all across the Trans-Mexican Volcanic Belt (Allan, 1986; Wallace and Carmichael, 1999; Verma, 2000a, 2000b; Gómez-Tuena et al., 2003, 2007b). Albeit volumetrically minor, these rocks have received disproportionate attention in the literature, mainly because of their unusual association in time and space with calc-alkaline volcanism that is more typical of a convergent margin. Perhaps not surprisingly, the origin of these rocks has been the subject of a myriad of interpretations that embrace a wide range of petrotectonic conditions. Some authors have proposed that these alkaline rocks may be associated with a mantle plume (Moore et al., 1994; Márquez et al., 1999), while others have related them to continental rifting (Verma, 2002), to the infiltration of deeper and hotter asthenospheric mantle due to slab rollback (Ferrari et al., 2001), or following a slab detachment event (Ferrari, 2004). Still other authors consider them as a direct consequence of convergent geodynamics (Luhr, 1997; Gómez-Tuena et al., 2011, 2013). As controversial as they may be, most of these interpretations are built upon the premise that the existence of an unusual rock type requires an exceptional explanation, but it has been increasingly recognized that assigning a tectonic connotation to a particular rock suite on the basis of composition alone can be misleading unless the petrologic character of the source and the conditions of melt generation are resolved.

In this contribution we provide comprehensive whole-rock geochemical data, as well as detailed chemical analysis of olivine, clinopyroxene, and plagioclase phenocrysts of intraplate (OIB-like) mafic rocks from the western sector of the Trans-Mexican Volcanic Belt, a well-studied area of the Mexican arc that has been experiencing tectonic extension and prolific volcanism since the early Pliocene (Demant, 1981; Luhr et al., 1985; Allan, 1986; Ferrari and Rosas-Elguera, 2000). Using these data, we identify the sources and processes of melt generation in order to elucidate the composition and petrologic nature of their mantle source and the role played by subduction in its compositional transformation.

The tectonic framework of the western Trans-Mexican Volcanic Belt (TMVB) comprises the subduction of the Rivera and Cocos plates beneath North America along the Middle American Trench (Fig. 1; Demant, 1978). Seismic hypocenters and recent tomographies indicate that subduction of the young Rivera plate is very steep (>50°), whereas the subduction of the Cocos plate becomes flat to the east (Pardo and Suárez, 1993, 1995; Pérez-Campos et al., 2008; Yang et al., 2009). The boundary between Rivera and Cocos plates extends below the southern Colima rift and gradually diverges to the north with the opening of a slab window (Yang et al., 2009). The active volcanic front in this portion of the arc is marked by a NW-SE–trending belt of monogenetic volcanoes (Bandy et al., 2001; Ownby et al., 2008; Gómez-Tuena et al., 2011) under which the Rivera plate resides at 140–150 km depth (Yang et al., 2009; Fig. 1A). Farther to the north the arc merges into the NW-SE–trending Tepic-Zacoalco rift (TZR; Demant, 1981), which has been interpreted as the result of an eastward “jump” of a segment of the East Pacific Rise (Luhr et al., 1985; Allan, 1986), or more recently as the extensional tectonic expression of rollback of the Rivera plate starting at ca. 10 Ma (Ferrari and Rosas-Elguera, 2000); the rollback accelerated during the Pliocene (DeMets and Traylen, 2000; Gómez-Tuena et al., 2013). Interestingly, seismic tomography has shown that the Rivera plate is currently deeper than 250–300 km or may be even absent under the TZR (Yang et al., 2009; Fig. 1A).

Extensional tectonics and volcanism have been closely related in western Mexico since the late Miocene (Ferrari and Rosas-Elguera, 2000), and both are the result of a complex evolution of the convergent margin that includes fragmentation of the Farallon plate and the opening of the Gulf of Baja California (DeMets and Traylen, 2000). The oldest rocks in the area belong to the Jalisco block—a Cretaceous batholith (ca. 100–75 Ma; Schaaf et al., 1995) that experienced significant uplift before the Pliocene that may continue to the present times (Righter et al., 2010; Ramírez-Herrera et al., 2011). Rocks of the Jalisco block mainly crop out to the south of the TZR, but some minor exposures have also been identified within the TZR (Gastil et al., 1978; Frey et al., 2007) and in geothermal wells (Ferrari and Rosas-Elguera, 2000). The subsequent magmatic episode belongs to the Oligocene–early Miocene Sierra Madre Occidental (SMO), one of the largest silicic igneous provinces on Earth (Bryan and Ernst, 2008). These rocks overlie Cretaceous granites and rhyolites within the TZR (Ferrari and Rosas-Elguera, 2000; Frey et al., 2007) but have never been identified in the Jalisco block, probably because they were eroded away during uplift, or because they were never emplaced above it. After an ∼10-m.y.-long hiatus of magmatic activity, volcanism resumed with the emplacement of voluminous tholeiitic and calc-alkaline basalts (11–8 Ma) that have been related to a mantle plume (Moore et al., 1994), a slab-detachment event (Ferrari, 2004), or a lithospheric foundering episode (Mori et al., 2009). Volcanism changed during the Pliocene (5.5–3 Ma), becoming bimodal within the TZR, and potassic lamprophyric to calc-alkaline within the Jalisco block (Lange and Carmichael, 1991; Frey et al., 2007; Ownby et al., 2008), probably as a result of a rapid rollback of the Rivera plate (DeMets and Traylen, 2000; Gómez-Tuena et al., 2013). After a period of stagnated convergence between 2.6 and 1 Ma, subduction resumed at an increased velocity (∼3 cm/yr) at ca. 1 Ma (DeMets and Traylen, 2000). Intraplate and rhyolitic volcanism diminished within the TZR during this period (Frey et al., 2007) and was replaced by the construction of five voluminous andesitic stratovolcanoes: Tequila, Ceboruco, Sangangüey, Tepetiltic, and San Juan (Fig. 1).

In this contribution we concentrate on the geochemical and petrological characteristics of Plio-Quaternary intraplate (or OIB-like) monogenetic volcanoes and plateau lavas emplaced across the TZR (Fig. 1). Some whole-rock data were already presented in recent publications (Gómez-Tuena et al., 2011, 2013), but Tables 1 and 2 provide complementary geochemical results on 19 additional samples, whereas Tables 5, 6, and 7 present new microanalytical data on olivine, plagioclase, and clinopyroxene crystals separated from selected samples.

Based on its similar petrography and geochemical characteristics, we define three rock suites according to their geographical location (Fig. 1): (1) the Sangangüey Basalt (SB) suite, a group of basaltic samples surrounding the Sangangüey volcanic field, which is located toward the western limit of the TZR; (2) the Amatlán de Cañas (AC) suite, which represents a group of samples collected in the southern edge of the TZR, close to its boundary with the Jalisco block; and (3) the Santa Rosa (SR) suite, which groups rock samples from the eastern edge of TZR and around the Tequila volcanic field. For comparison, some plots include previously published geochemical data from Isabel and Socorro Pacific islands (Bohrson and Reid, 1995, 1997; Housh et al., 2010) and from the compositionally similar Usmajac (Us) cinder cone (Gómez-Tuena et al., 2011) located within the Colima rift, ∼40 km to the north of Colima volcano (Fig. 1B). The stratigraphic and structural features of all these areas have been the focus of several previous studies that will be briefly summarized here.

Intraplate basalts in the TZR comprise cinder cones and basaltic plateaus emplaced mainly through preexistent NW-SE regional structures. Samples of the SB suite come from fissure-fed lava flows that filled the NW-SE Rio Grande de Santiago valley, and from cinder cones and associated lava flows erupted through NW-SE–trending extensional fractures at the flanks of the <200 ka andesitic Sangangüey stratovolcano (Nelson and Livieres, 1986). Plateau basalts and shield volcanoes of the AC suite (ca. 3.4–0.65 Ma) were emplaced directly over rocks of the Jalisco block, within a small half graben of early Pliocene age (Ferrari and Rosas-Elguera, 2000; Righter and Rosas-Elguera, 2001; Gómez-Tuena et al., 2011). Rocks belonging to the SR suite are plateau lavas and monogenetic volcanoes emplaced within the so-called Plan de Barrancas–Santa Rosa graben, a 30-km-wide, WSW-trending extensional structure flanked to the north by the Rio Grande de Santiago (Ferrari and Rosas-Elguera, 2000; Righter and Rosas-Elguera, 2001). Plateau lavas younger than 1 Ma were emplaced to the north of Tequila volcano from NW-SE–trending fissures located within the Rio Grande de Santiago valley, whereas cinder cones and associated flows younger than ca. 0.5 Ma were emplaced directly on top of the Santa Rosa basalt plateau (Lewis-Kenedi et al., 2005; Frey et al., 2007). Intraplate basalts of the SR suite form a bimodal association with rhyolitic volcanism between 1 and 0.2 Ma and includes the precursor activity of Tequila stratovolcano (Lewis-Kenedi et al., 2005; Frey et al., 2007).

Major elements of whole rocks were determined by X-ray fluorescence spectrometry using a Siemens SRS-3000 instrument at the Laboratorio Universitario de Geoquímica Isotópica (LUGIS) of the Universidad Nacional Autónoma de México (UNAM), following procedures described elsewhere (Lozano and Bernal, 2005). Trace-element data were obtained by inductively coupled plasma mass spectrometry (ICP-MS) using a Thermo Series XII instrument at Centro de Geociencias (CGEO), UNAM, following the sample preparation and measurement procedures described by Mori et al. (2007). The long-term reproducibility of the trace-element data at CGEO is given by the average concentrations and standard deviations of multiple digestions of the U.S. Geological Survey rock standards AGV-2, BHVO-2, BCR-2, and the Geological Survey of Japan JB-2, and has been reported in several previous publications (Mori et al., 2007, 2009; Gómez-Tuena et al., 2011).

Sample preparations for Sr, Nd, and Pb isotopic analyses were performed at the clean laboratory facilities of the CGEO following previously established chemical procedures (Gómez-Tuena et al., 2011, 2013). Strontium, Nd, and Pb isotope ratios were measured using a Thermo Neptune Plus multicollector–inductively coupled plasma mass spectrometer (MC-ICP MS) at CGEO, following measurement procedures described by Gómez-Tuena et al. (2013). The measured 87Sr/86Sr ratios were exponentially corrected for mass fractionation by normalizing to 86Sr/88Sr = 0.1194 and subsequently adjusted to a National Institute of Standards and Technology (NIST) Standard Reference Materials (SRM) 987 standard ratio of 87Sr/86Sr = 0.710230. During the course of this study, the mean value for NIST SRM 987 was 0.710264 ± 0.000012 (2σ, n = 22). The measured 143Nd/144Nd ratios were exponentially corrected for mass fractionation to 146Nd/144Nd = 0.72190, and further adjusted to a JNdi standard value of 143Nd/144Nd = 0.512115 (Tanaka et al., 2000), repeatedly measured during the same analytical sessions. Over the course of this study, the mean 143Nd/144Nd ratios of the JNdi standard were 0.512090 ± 0.000009 (2σ, n = 32). Using these corrections, the 143Nd/144Nd ratios for the La Jolla standard as an external monitor were 0.511859 ± 0.000007 (2σ, n = 9). The measured Pb isotopic ratios were exponentially corrected for mass fractionation using a NIST SRM 997 Tl solution with a reference 205Tl/203Tl value of 2.3871, which was added to the samples to match a Pb/Tl ratio of ∼4. The Tl fractionation-corrected values in the samples were further adjusted to the NIST SRM 981 standard values of 206Pb/204Pb = 16.9356, 207Pb/204Pb = 15.4891, and 208Pb/204Pb = 36.7006 (Todt et al., 1996). Over the course of this study, the Tl fractionation-corrected Pb isotopic compositions of the NIST SRM 981 standard were 206Pb/204Pb = 16.9305, 207Pb/204Pb = 15.4839, 208Pb/204Pb = 36.6739 (2σ of 57, 60, 56 ppm, respectively, n = 27). Reproducibility and accuracy of the isotopic measurements were verified by the analysis of U.S. Geological Survey (USGS) standards BHVO-2 and AGV-2, which were prepared and measured using the same procedures as the samples. The Sr, Nd, and Pb isotopic compositions obtained on these standards are within analytical uncertainty to the preferred values published in the GEOREM database (http://georem.mpch-mainz.gwdg.de/) and are reported in Table 2.

Olivine, plagioclase, and clinopyroxene phenocrysts were separated from the bulk rock using conventional magnetic and gravimetric techniques, mounted on epoxy resin, and polished to exposure for microanalysis. Compositions of olivine and plagioclase mineral phases were analyzed by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) at CGEO with a Resonetics Resolution M50 workstation, operating a 193 nm Argon fluoride (ArF) excimer laser (LPX220, Lambda Physik) coupled to a Thermo Series XII quadrupole mass spectrometer (see Solari et al., 2010 for further technical details). Analyses were performed by drilling with a spot size of ∼60 µm and a repetition rate of 5 Hz, employing ∼122 mJ output energy that corresponds to an on-target fluence of ∼7 J/cm2. The generated aerosol was transported to the ICP-MS using He as carrier gas (∼700–600 mL/min), which is mixed downstream with ∼700 mL/min of Ar as make-up gas. Analyses were obtained with 20 s of gas background acquisition followed by 30 s of ablation time.

Laser ablation–inductively coupled plasma–mass spectrometry peak intensities are often converted into elemental concentrations employing an internal standard, which is an element of a known concentration in the sample that has been determined by an independent analytical method such as an electron microprobe (EMP). Yet the internal standard method is costly and time consuming because it requires two analytical sessions at different facilities, and it also introduces additional analytical uncertainties due to interlaboratory calibration biases. Furthermore, and even if the analyzed target can be accurately identified in both instruments, spatial resolutions are often very different: EMP analyses are limited to a few microns of the sample surface, whereas laser spots are usually much larger and excavate tens of microns into the material. Such a spatial mismatch may not be important for homogeneous metallic alloys or minerals with a simple stoichiometry, but it may become problematic for the complex solid solutions of rock-forming minerals such as olivine, plagioclase, pyroxene, and garnet. The best results will obviously be obtained during a single analytical session on the same instrument, but trace-element concentrations are difficult to resolve by EMP, whereas LA-ICP-MS analyses are still limited by the lack of appropriate standards.

In order to minimize analytical uncertainties and interlaboratory biases, in this contribution we used an internal-standard independent calibration strategy that normalizes the concentrations of all major and minor elements as oxides to 100 wt%, after an external calibration against well-characterized natural olivine and plagioclase mineral standards that were repeatedly measured throughout the analytical sessions. Data reduction was performed offline with tailored spread sheets that used calculations thoroughly described elsewhere (Longerich et al., 1996; Halicz and Günther, 2004; Guillong et al., 2005; Liu et al., 2008). A matrix-matched external standardization corrects for potentially adverse spectral interferences, and significantly reduces the effects of different laser absorption behaviors that are commonly observed when using standards of a different matrix (Evans and Giglio, 1993; Kroslakova and Günther, 2007). The major drawback of this method is that natural mineral standards are not yet widely available, and those that exist are only well characterized for their most abundant constituents and perhaps a few trace elements.

During the course of this study, external standardization of unknown olivines was performed by measuring the well-calibrated Smithsonian USNM 111312/444 San Carlos olivine (Jarosewich et al., 1980) every 10 unknowns, and using the average elemental concentrations reported in Sobolev et al. (2005), normalized to 100%. In order to assess reproducibility and to ensure comparability of our results, a large olivine crystal fragment (sco-sms, Table 3) coming from the same locality as the Smithsonian San Carlos Olivine was also repeatedly measured during each analytical session. Our results on the sco-sms are equivalent within error to those reported by Straub et al. (2008, 2011), and also confirm their observation that this olivine crystal is similar but not identical to USNM 111312/444. Nonetheless, our results indicate that the sco-sms olivine is sufficiently homogeneous to be used as an in-house standard in future analytical sessions (Table 3).

Similarly, the calibration of unknown plagioclase crystals reported in this work utilized a natural labradorite microbeam standard (AS1285-AB) that is commercially available from Structure Probe Incorporated (SPI; http://www.2spi.com/). Standard measurements were performed every 10 unknowns, and the calibration used the SPI recommended published values normalized to 100%. For the case of plagioclase, an assessment of analytical reproducibility and accuracy is given by the standard deviation of multiple measurements of an in-house labradorite gemstone standard named La Gema, repeatedly measured during the course of this study. The composition of La Gema standard was independently characterized for homogeneity by EMP, and the results agree within error to those obtained by LA-ICP-MS (Table 4). Our results show that La Gema is homogeneous within the analytical uncertainty of the method, and can be used as an in-house standard in future analytical work.

Compositions of clinopyroxene phenocrysts were analyzed using a JEOL JXA-8900R electron microprobe at the Laboratorio Universitario de Petrología (LUP) of the UNAM. A 20 keV accelerating potential was used in the analyses, with a beam current of 20 nA and a spot size of ∼1 µm. Calibration was verified using the SPI #02757-AB diopside standard.

Petrographic Descriptions

The studied mafic rocks generally display porphyritic textures with a phenocryst assemblage made of olivine and plagioclase, with a few samples also containing scarce clinopyroxene phenocrysts (Fig. 2). Olivine phenocrysts up to 2 mm of maximum size frequently have rounded or embayed shapes. Plagioclase phenocrysts are fine-grained to very large (up to 10 mm), euhedral to subhedral crystals, and display carlsbad and albite twinning. A few large phenocrysts display disequilibrium sieve textures or corroded shapes, complex zoning, wedge-shaped twinning, or wavy extinction. In some cases, plagioclase phenocrysts form glomeroporphyritic aggregates (up to 5 mm) of fine- to coarse-grained phenocrysts with microphenocrysts of olivine. Clinopyroxene phenocrysts have medium-grained dimensions (∼2 mm), and typically display rounded shapes or wavy extinction. The groundmass consists of plagioclase microlites and titanomagnetite along with glass, olivine, and clinopyroxene. Textures of these rocks range from hypocrystalline in the more crystalline varieties, which generally come from lava flows, to hyalopilitic in the sparsely crystalline samples from cinder cones. A few samples display incipient chlorite alteration of the groundmass, and some olivines have been transformed to iddingsite, especially along rims and fractures. Some petrographic features may be distinguished among suites: basalts from the SB and the SR suites have phenocryst assemblages made of olivine, plagioclase, and scarce clinopyroxene phenocrysts, whereas rocks from Amatlán de Cañas have olivine and plagioclase but lack pyroxene. Furthermore, SB and SR usually display large phenocrysts of plagioclases and olivine, whereas the AC suite displays small- to medium-size phenocrysts and do not contain glomeroporphyritic aggregates.

Major and Trace Elements

Most of the studied rocks are Na-alkaline mafic volcanics that display a compositional variation spanning from trachybasalts (hawaiites) to basaltic-trachyandesites (mugearites), although samples from AC mostly straddle the alkaline and subalkaline fields (Fig. 3A). Most primitive rocks have higher TiO2, K2O, Na2O, and P2O5 but lower CaO contents than mid-ocean ridge basalts (MORB, Figs. 3B–3D), with compositions that are very similar to those found in basalts from the Mexican Pacific islands of Isabel and Socorro, which are also typical of ocean-island basalts (OIBs). Major-element contents are not significantly variable among the suites and mostly overlap with each other, following a near-tholeiitic differentiation trend (Fig. 3E). It is nonetheless noteworthy that samples from SB have the lowest silica contents at highly variable Mg#, whereas rocks from AC and SR range to higher SiO2 without exhibiting a significant drop in Mg# (Fig. 3F). Calcalkaline andesites and dacites erupted from large Quaternary stratovolcanoes in the area, such as Tequila and Sangangüey, also share these high Mg# (Gómez-Tuena et al., 2013), a characteristic that differs strongly from the differentiation tendency observed in Socorro Island, for instance, where peralkaline trachytes and rhyolites commonly have Mg# <10 (Bohrson and Reid, 1995, 1997).

Incompatible trace elements such as Rb and Th display a positive correlation with silica, following a tendency that overlaps data from Socorro Island (Figs. 4A and 4B). Lead contents in most mafic rocks are similar to Socorro and Isabel, and similarly increase with silica, albeit with a steeper slope than in the islands (Fig. 4C). Despite these similarities with neighboring OIBs, Sr contents remain almost constant in the studied rocks, while Sr sharply decreases in more evolved rocks from Socorro Island (Fig. 4D), whereas other incompatible elements such as Nb and Dy tend to decrease with increasing silica, and therefore follow a divergent trend to what is commonly observed in the oceanic islands (Figs. 4E and 4F).

Trace-element patterns allow a better discrimination among the studied rock suites (Fig. 5). Rocks from the SB suite are enriched in high field strength elements (HFSE) over most of the large ion lithophile elements (LILE), with trace-element concentrations that straddle typical OIB values, and a pattern that is almost indistinguishable from that of most primitive rocks from Socorro Island (Fig. 5A). Small positive spikes in Ba and Sr appear to be a characteristic feature of the most mafic volcanics, regardless of the geologic context in which they were generated. In contrast, samples from SR and AC, as well as the Usmajac cone, display subtle but clear positive anomalies in most LILE, Pb, and Sr with respect to the HFSE, and therefore tend to show characteristics that are more typical of magmatism from a convergent margin (Fig. 5B).

Isotope Geochemistry

The Sr-Nd isotopic compositions of the studied volcanic rocks mostly overlap with each other, and are bracketed between the composition of the East Pacific Rise–mid-ocean ridge basalts (EPR-MORB) and an upper crustal component similar to the Pacific sediments or the local upper continental crust (Jalisco block and Sierra Madre Occidental) (Fig. 6A). Most depleted rocks from SB entirely overlap the compositions of the Mexican Pacific islands and have Nd isotopic compositions that are only slightly less radiogenic than the Rivera plate MORB. In contrast, rocks from the SR and AC suites generally extend to more enriched Sr-Nd isotopic compositions, and display more radiogenic Sr isotopes at equivalent 143Nd/144Nd ratios than Socorro and Isabel (Fig. 6B). Interestingly, the Sr-Nd isotopic compositions of andesitic volcanoes in the area mostly plot along a subparallel mixing array and show little overlap with the rest of the suites (Gómez-Tuena et al., 2013).

Pb isotope compositions of the SB suite exhibit a distinctive mixing trend bracketed between the Pacific MORB and an enriched high-μ component (HIMU; μ = 238U/204Pb) that has also been recognized in the Mexican Pacific islands of Socorro and Isabel (Bohrson and Reid, 1995; Housh et al., 2010). In contrast, the isotopic compositions of AC and SR suites form an independent array bracketed between the MORB and the Pacific plate sediments and/or the local continental crust, with values that also overlap the compositions of andesitic stratovolcanoes in the area (Figs. 6C and 6D).

Rocks from the SB display a highly heterogeneous isotopic character that is mostly independent from silica contents or subduction proxies such as Pb/Nb ratios (Figs. 6E and 6F). In contrast, the isotopic compositions of the AC and SR suites are almost as diverse as those of the SB suite but show discernible correlations with both subduction (Pb/Nb) and fractionation (SiO2) indexes.

Olivine Compositions

Olivines from the studied rock suites (Table 5) are euhedral to subhedral phenocrysts (Figs. 2J and 2K) with sizes that range from 0.5 to 1 mm. Analyzed grains from SB and AC suites and the Usmajac cone show little compositional variations between their cores and rims, with just a few crystals displaying normal or reverse zoning. Olivines from the SR suite are compositionally more complex, with crystals with lower forsterite (Fo) contents often showing reverse zoning. Analyzed olivines from each sample have variable Fo contents that are often out of equilibrium with their whole-rock compositions (Fig. 7A). Olivine phenocrysts from SB suite tend to have higher Fo content than their host rocks, whereas olivines from the SR and AC suite are either in equilibrium with their hosts or extend to lower Fo contents.

Figures 7B and 7C show NiO and CaO contents in olivine phenocrysts from the studied area compared to olivines from the central TMVB (Straub et al., 2008, 2011), MORB, and the Hawaiian volcano of Koolau, which is taken as representative of a thick-lithosphere intraplate environment (Sobolev et al., 2007). In contrast to what has been documented in central Mexico, olivines from the SB suite entirely overlap the MORB olivine compositions, showing NiO contents that do not exceed 0.34 wt% at a near equilibrium mantle value of Fo90. Interestingly, more evolved olivines from the SR and AC suites share a similar range in NiO, but each sample follows a separate and subparallel array that overlaps the compositions observed in Koolau and the central TMVB, at much lower CaO abundances.

Plagioclase Compositions

Plagioclase phenocrysts were analyzed on selected samples from the SR and SB suites, and for the purpose of comparison, from an andesite that belongs to Sangangüey volcano (sample TPZ-10-27). Plagioclase phenocrysts from the mafic rocks (Table 6) are euhedral to subhedral (Figs. 2D–2I) with sizes that range from ∼1 to 2 mm to megacrysts that can reach up to 4 mm in size. These large plagioclase crystals often display carlsbad and albite twinning, but just a few of them develop clear optical disequilibrium textures. In contrast, plagioclase phenocrysts from andesitic stratovolcanoes usually display more complex zoning and sieve textures that are clear indications of disequilibrium growth.

Plagioclase phenocrysts from the SR suite have relatively homogenous intermediate compositions (Fig. 8A), ranging mostly between ∼An55 and An70, except for three crystals in TZ-09-14 that have <An50. In contrast, plagioclases from the SB suite are more variable and can reach up to An80 in SAN-10-01A, or go as low as An30 in sample RTZ-11-22. Overall, plagioclase phenocrysts from both suites are unzoned, except for a couple of crystals that show a slight reverse zoning mostly in the SR suite. This is in clear contrast to plagioclase phenocrysts separated from the Sangangüey andesite, which not only are highly variable (∼An50–90) but often display clear normal zoning (Figs. 8B and 8C).

Figure 9 shows that the measured An contents of each sample mostly overlap the experimentally determined equilibrium crystallization conditions of their respective whole-rock compositions (Lange et al., 2009), indicating that most plagioclases should be phenocrystic. Plagioclase compositions from the Sangangüey andesite are also similar to those reported previously in equivalent rocks from Tequila and Sangangüey (Crabtree and Lange, 2011).

Olivine compositions were used for the determination of magmatic temperatures employing the olivine-liquid equilibrium thermometer (Putirka et al., 2007). For the samples where fresh olivines were not available, liquidus temperatures were estimated using the clinopyroxene-liquid geothermometer (Putirka, 2008) (samples TZ-09-14 and SAN-10-09, Table 7), or from the whole-rock compositions using the MELTS algorithm (Ghiorso and Sack, 1995) (sample RTZ-11-35). The calculated temperatures were in turn used to estimate preeruptive water contents of the studied magmas by utilizing the plagioclase-liquid hygrometer (Lange et al., 2009). For the purpose of comparison, all models assume that olivines and plagioclases having the maximum forsterite and anorthite contents crystallized at a near Moho depth of 1.3 GPa, which corresponds to an average crustal thickness of ∼40 km in the region of the Tepic-Zacoalco rift (Molina-Garza and Urrutia-Fucugauchi, 1993). While this is an obvious simplification because mineral growth most likely occurred over a range of temperatures and pressures, the assumption allows us to obtain a useful comparison for the parameters of interest on each sample. Using this approach, maximum temperatures will be obtained using equilibrium olivines with the highest Fo content, whereas the best estimate of the initial amount of dissolved water in the melt will be recorded by plagioclases with the maximum An content (Table 8).

Figure 7A (Mg#liq versus Fo mole% for each sample) shows that olivine phenocrysts are not always in equilibrium with their host-rock compositions. This indicates that olivine crystallized at different thermal conditions from a variety of liquids, so that the temperature that can be calculated using the whole-rock composition only represents an average of a much wider range of conditions that are, nonetheless, uniquely recorded in the minerals (Thomson and Maclennan, 2013). Melt inclusions have been traditionally used as a direct estimate of the magmatic variability, but inclusions were rarely found in the studied samples, and those recognized were too small and thus prone to postentrapment modifications. For these reasons we followed an alternative approach for reconstructing the composition of equilibrium liquids. In our method, the olivine with the maximum Fo content (Fomax) from each sample was mathematically added to the whole-rock composition until equilibrium conditions were satisfied (KD(Fe–Mg) = 0.3; Roeder and Emslie, 1970). Magmatic temperatures were then calculated using Fomax and the reconstructed equilibrium liquid (Table 8). Figure 10 shows that the olivine addition method provides similar results as those obtained by the PRIMELT2 parameterization (Herzberg and Asimow, 2008), which calculates the composition of derivative liquids by successively adding equilibrium olivine to the whole-rock composition. But unlike PRIMELT2, which aims to find the composition of a primitive magma in equilibrium with a mantle olivine (∼Fo92), our method only calculates a magmatic composition in equilibrium with the maximum observed Fo for each sample (Fomax), because there is no a priori reason to assume that the studied magmas were ever in equilibrium with a typical peridotite source. It should be noted that (1) compositional reconstructions such as these are only entirely accurate if olivine is the only, or at least the most abundant, crystallizing phase (Herzberg and Asimow, 2008), and (2) the calculated temperatures will be slightly overestimated, if pyroxene and/or plagioclase were separated from the melt together with olivine. We therefore emphasize that calculated magmatic temperatures reported in this work should always be considered as maxima.

Temperature and water content estimates are mutually dependent and thus expected to correlate inversely (Table 8; Fig. 11A). Water abundances in the studied rocks are equivalent to those found in mafic magmas from intraplate or backarc settings (Stolper and Newman, 1994; Kelley et al., 2006; Langmuir et al., 2006), and are virtually identical to those determined in olivine-hosted melt inclusions from analogous basalts in central Mexico (Cervantes and Wallace, 2003) and from the compositionally similar Usmajac cinder cone (Vigouroux et al., 2008; sample CU-02-02). The estimated water abundance in the sole studied andesite is also similar to previously reported values on Sangangüey and Tequila (Crabtree and Lange, 2011) and to the values determined in melt inclusions from other Mexican andesitic stratovolcanoes (Roberge et al., 2009).

Interestingly, and as previously observed in the Mariana Trough (Stolper and Newman, 1994), Mexico (Cervantes and Wallace, 2003), and in most backarc settings (Kelley et al., 2006), the relative enrichment in water also correlates inversely with high field strength element (HFSE) contents and ratios, and positively with other subduction proxies such as LILE/HFSE and Pb/Nb ratios (Figs. 11B–11D). Since trace-element contents and ratios come from entirely independent measurements, we conclude that the temperatures and preeruptive water contents obtained in this work are sufficiently robust, and perhaps even more reliable than direct melt-inclusion measurements, which are usually affected by postentrapment modifications and many other analytical complications.

Crystal Fractionation and Crustal Contamination

Several lines of evidence indicate that the overall compositional diversity cannot be the result of a simple mechanism of crystal fractionation, not even considering secondary modifications due to crustal interactions. Rocks are almost always porphyritic and therefore some degree of mineral fractionation must have occurred, of course, but the variable isotopic compositions of the SB suite at almost constant silica and Pb/Nb ratios (Figs. 6E and 6F) are difficult to explain by crystallization or contamination alone. Rocks from the AC and SR suites have slightly more silica, stronger subduction signatures, and slightly different isotopic compositions that may indicate fractionation with some degree of crustal assimilation, but these rocks also display much higher Mg#, Sr, and H2O contents, and lower HFSE–rare earth element (REE) abundances than nearly identical magmas emplaced in the oceanic realm for which crystal fractionation has been well documented (Bohrson and Reid, 1995, 1997). It could be argued that an entirely different mineral assemblage involving Fe-Ti oxides, amphibole, or garnet dominates fractionation at the higher pressure conditions of a thick continental crust (Davidson et al., 2007; Alonso-Perez et al., 2009; Zellmer et al., 2012), but this only can happen if the primary melts are intrinsically more oxidized and hydrous than what has been documented for the most primitive rocks of the SB suite. The presence of dissolved water may indeed be responsible for suppressing plagioclase stability and creating derivative liquids with higher Sr contents than in the islands (Fig. 4D), but the fourfold decrease in Nb at increasing water contents and H2O/Nb ratios (Fig. 11) appears inconsistent with any known fractionation assemblage since water and Nb are similarly incompatible.

In addition, olivine phenocrysts from each sample display subparallel trends with different Fo at equivalent NiO contents that cannot be formed by a single liquid line of descent, and instead point to crystallization from different kinds of magmas. Similarly, plagioclase phenocrysts with high An contents are often found in more felsic liquids, not in hotter and more primitive magmas, a characteristic that is difficult to explain by crystal fractionation from a single cooling liquid but that is consistent with the existence of magmas with variable amounts of dissolved water (Lange et al., 2009).

In summary, the whole-rock compositions as well as their mineral chemistry indicate that the studied rock suites are not derivative liquids from a single parental melt. Each magma experienced its own fractionation pathway and had little interaction with other magmas or with the continental crust during transit to the surface. We thus conclude that the compositional variability derives primarily from an inherently heterogeneous mantle wedge that has been subsequently modified by hydrous chemical agents that likely derive from the subducted slab.

Insights into the Pristine Mantle Source

Most primitive rocks from the SB suite have almost identical chemical and Sr-Nd-Pb isotopic characteristics to those found in basalts from the ocean islands Socorro and Isabel. Isabel is just ∼30 km off the Mexican Pacific coast, sits on the North American plate, and perhaps on top of an attenuated continental lithosphere, yet its petrologic and geochemical character reflects melting of a mantle source that has not been modified by the continental crust or the subducted slab (Housh et al., 2010). Rocks of the SB suite are, however, more similar to basalts from Socorro Island (Bohrson and Reid, 1995), which is constructed on the Pacific plate, much farther away from any continental or subduction influence, and therefore likely to reflect the uncontaminated characteristics of the upper mantle in the region. Isotopically, these mafic rocks are more enriched than a typical MORB, but they nonetheless trend along a simple mixing line that extends to the compositions of Rivera and Cocos plates (Fig. 6A–6D). The compositional analogies of the studied rock suites with their oceanic counterparts strongly support the notion that the pre-subduction mantle wedge below western Mexico must be compositionally identical, and just as heterogeneous, as that below the Pacific basin (Gómez-Tuena et al., 2011, 2013), but how this mantle evolved, its petrologic nature, and the processes that allowed melting in its nearly pristine conditions in a continental convergent margin, are all questions that deserve further attention.

Rocks from the SB suite and the Pacific islands are invariably more enriched in trace elements than MORB, the origin of which cannot be solely the result of low extents of melting of the depleted upper mantle. TiO2 contents of alkaline basalts like these are too high to be formed by melting of a typical peridotite (Prytulak and Elliott, 2007), whereas their high 206Pb/204Pb ratios clearly reflect a long time-integrated isotopic growth of a source with much higher μ (238U/204Pb) than the MORB mantle. A high-μ mantle signature has traditionally been interpreted to be the result of recycling of deeply subducted oceanic basalts that reside in the convective mantle for billions of years (Hofmann and White, 1982; White, 1985; Zindler and Hart, 1986), albeit some more recent studies have suggested that it could have a metasomatic origin as amphibole- and/or clinopyroxene-rich cumulate veins located in the lithospheric mantle (Pilet et al., 2008, 2011; Niu et al., 2011), or as solidified low-degree melts derived from the subducted oceanic crust (Donnelly et al., 2004; Langmuir et al., 2006).

Mantle xenoliths collected all across Mexico demonstrate that the continental lithospheric mantle can be extremely heterogeneous at the local and regional scale (Luhr and Aranda-Gómez, 1997). Exotic mineral assemblages including kaersutite and phlogopite, as well as some extremely refractory compositions, indicate a long and complex history of melt extraction that is also often coupled with periods of metasomatic transformations. In effect, a highly heterogeneous lithospheric mantle is all but expected given that continental Mexico is formed from an ensemble of exotic terranes and superimposed arcs and rifts that date as far back as the Mesoproterozoic (Campa and Coney, 1983; Sedlock et al., 1993; Ortega-Gutiérrez et al., 1994). Although these geologic processes add inherent complexities to the Mexican lithospheric roots, several lines of evidence deem the lithospheric mantle an unlikely source for the origin of intraplate magmatism in the region.

Recovered mantle enclaves are mostly shallow spinel-to-plagioclase–bearing peridotites (Luhr and Aranda-Gómez, 1997) that are usually hosted in alkaline intraplate-like lavas with steep REE patterns that suggest melt extraction from the garnet stability field (Gómez-Tuena et al., 2003; Gómez-Tuena et al., 2007b). This chemical characteristic clearly indicates that the host magmas must come from a deeper source than the xenoliths they carry, regardless if they are emplaced in the oceanic or continental realm. In addition, while mantle xenoliths can be locally heterogeneous, their host lavas are remarkably alike at the regional scale, and not restricted to a specific geographic location, volcanic field, or tectonic setting. Alkaline rocks with high-μ signatures have been discovered not only all across the Trans-Mexican Volcanic Belt (Gómez-Tuena et al., 2007b) but also in the Mexican Basin and Range (Luhr et al., 1989; Pier et al., 1989; Luhr, 1997), along the eastern coast of the Gulf of Mexico (Gómez-Tuena et al., 2003; Orozco-Esquivel et al., 2007), in the extended margins of the Gulf of California (Ferrari et al., 2013), and even in abandoned spreading centers off the western coast of Baja California (Tian et al., 2011). Rocks with almost the exact same Pb isotopic signatures have also been found in the rear arc of Central America (Hoernle et al., 2008) and in the Galapagos Islands (Harpp and White, 2001). It thus seems clear that the highly heterogeneous nature of the lithospheric mantle underneath continental Mexico would hardly generate magmas with similar chemical characteristics thousands of kilometers apart, and in such a variety of geologic settings.

High-Nb basalts in some backarc basins lack significant isotopic ingrowths indicating that the process of trace-element enrichment can be a recent phenomenon, probably associated with the ongoing tectonic convergence (Langmuir et al., 2006). It is also conceivable that subduction of an isotopically enriched oceanic crust can transfer radiogenic high-μ Pb to the mantle wedge through the action of fluids or melts, as has been suggested for Central America (Hoernle et al., 2008). However, processes like these seem unlikely for the Mexican case because rocks carrying the high-μ signature have much lower water contents and lack any other geochemical proxy that could signal significant slab additions, whereas the ones that do (i.e., higher Pb/Nb(Zr) ratios) plot along a different trend in the Pb isotope mixing array (Figs. 11 and 12). Thus, if slab-derived metasomatism is responsible for the refertilization of the upper mantle in the region, the phenomenon not only must have occurred in the remote past in order to satisfy the isotopic constraints, but its effects would also be dispersed over a much wider region than what is currently being tapped by the Mexican convergent margin.

On Peridotites and Pyroxenites

Whichever mechanism is responsible for enrichment of the upper mantle in the region, most models converge in that the petrologic constituents of a source such as the high-μ must differ from the olivine-rich assemblage of a typical mantle peridotite, either in the form of recycled high-grade metamorphic assemblages, or as secondary melt/rock reaction products (Hofmann and White, 1982; White, 1985; Pilet et al., 2008, 2011; Niu et al., 2011). Trace elements and radiogenic isotopes of whole rocks are insufficient tools to distinguish if the enriched component resides as a discrete petrologic entity, such as pyroxenite or eclogite, or if these contaminants have been entirely hybridized into peridotite by long-term convective stirring and solid-state diffusion. Sobolev et al. (2005, 2007) and later Straub et al. (2008, 2011) overcame this limitation by showing that NiO and CaO contents in olivine phenocrysts can effectively identify the existence of non-peridotitic lithologies in the source of intraplate and arc magmas, because these elements partition differently in pyroxene and olivine, the most important mineral constituents of the upper mantle. Melts from normal peridotites will have high CaO and low NiO contents, and crystallize high CaO and low NiO olivines, due to the buffering effect of olivine in their mantle residuals. In contrast, melts from non-peridotitic lithologies, such as secondary reactive pyroxenites, will show the opposite effect if their source is entirely devoid of olivine.

Olivines from the SB suite have NiO and CaO contents that are virtually identical to olivines from MORB, strongly suggesting that crystallization occurred from a typical peridotite melt. This is a rather unexpected result given that their enriched trace-element and isotopic compositions are indicative of long-term recycling of subducted lithologies (Hofmann and White, 1982; Zindler and Hart, 1986). Nonetheless, recent studies in the Canary Islands (Gurenko et al., 2009) and the Cook-Austral Islands (Herzberg, 2006) have shown that primitive lavas carrying the high-μ signature also contain olivines that are indistinguishable from MORB compositions, indicating that the source of enrichment must also be primarily hosted in peridotite.

It thus appears that the ancient and recycled components carrying the high-μ signal in the Mexican magmas must be almost completely absorbed into peridotite and perhaps amount to just a few percent in volume of the mantle in the region. These components may be compositionally heterogeneous themselves, of course, but they are likely constituted by more fusible materials than mantle olivine. Consequently, they will be disproportionately represented at the initial stages of melting that characterize thick-crusted continental rifts and small volcano seamounts or ocean islands. Small heterogeneities like these will be easily diluted and become invisible if trapped into a geodynamic setting that promotes larger extents of mantle melting, such as below ocean ridges. Likewise, they will be virtually imperceptible if they become overprinted by secondary metasomatic transformations that are inherent of subduction zones.

Straub et al. (2008, 2011) recently showed that high-NiO olivines are not only produced in intraplate settings where secondary pyroxenites may be of volumetric significance (Sobolev et al., 2007), but that they may also be common in subduction zones where silica addition from the slab promotes the creation of metasomatic pyroxenites upon reaction with a peridotitic mantle. These authors have shown that remelting of these secondary pyroxenites can create a variety of primary liquids, from basalts to dacites, which upon reprocessing will gradually resemble more typical arc magmas. Thus, while high-NiO olivines in some intraplate settings reflect recycling of ancient and deeply transformed subduction lithologies, high-NiO olivines in arc magmas are the product of recycling and hybridization induced by the ongoing convergent margin.

Interestingly, olivines from the SR and AC suites have lower Fo contents but similar NiO than those from the SB suite, clearly indicating derivation from a different kind of liquid. While these olivines do not extend to the very high NiO contents observed in central Mexico, they do overlap the compositional trend of their less primitive counterparts, thus suggesting a similar derivative origin. Furthermore, their much lower CaO contents compared to the SB suite, MORB, and even Koolau, are strong indications that precisely these hydrous melts, carrying perceptible subduction signatures, are the ones that most likely derive from melting of a metasomatically produced pyroxenite source.

Subduction Effects on Mantle Transformation and Melting

We have shown above that the pristine mantle heterogeneity in the region appears to be ancient, asthenospheric, of regional distribution, not associated to a particular ocean basin, continent or tectonic boundary, and apparently hosted in peridotite. We have also suggested that highly enriched mantle components only become visible when melting occurs at its lowest extents, and without any influence by subduction. Therefore the upper mantle must remain close to partial fusion temperatures, so that very small decompression disturbances can be conducive to melting of its most fusible portions. Ocean ridges are obviously able to process much larger mantle volumes, and therefore function as homogenizing blenders that effectively erase the geochemical imprint of these small heterogeneities in their erupting magmas. Subduction zones are also able to process large mantle volumes, but as opposed to ocean ridges, the extent of melting in these settings is generally considered to be driven by the flux of water originating from the subducting oceanic crust. The working paradigm in convergent margins is that water effectively reduces the mantle solidus (Kushiro et al., 1968), and thus allows melting of peridotite under pressure-temperature conditions that are otherwise below the solidus. Since water contents in arcs and backarc basins are almost invariably correlated to geochemical proxies of melt fraction and slab contributions (Figs. 11 and 12), most authors conclude that petrogenesis in convergent margins occurs mainly by “flux melting” of the upper mantle (Stolper and Newman, 1994; Kelley et al., 2006; Langmuir et al., 2006).

The flux melting model has strong support from an experimental and theoretical basis, but it relies heavily upon the assumption that hydrous slab contributions are usually small, and thus unable to exert a significant petrologic transformation of the mantle wedge below arcs. Water and other volatiles clearly play a key role in the melting behavior of peridotite at subduction zones (Grove et al., 2012), but it has been increasingly recognized that water is not the only component being transferred to the mantle wedge at convergent margins. Melts from recycled sediments, eroded crust, and subducted basalts can bring significant amounts of silica to the mantle, and can act as much more effective metasomatic agents than solute-poor hydrous fluids (Gómez-Tuena et al., 2007a, 2011; Cooper et al., 2012). If the contribution of recycled silica is significant, reactive pyroxenites will form instead, the source of arc magmas will no longer be peridotitic, and the melting products will gradually become andesitic and not basaltic (Kelemen, 1998; Straub et al., 2011; Gómez-Tuena et al., 2013; Straub et al., 2013b).

As in any other convergent margin, the geochemical proxies for melt fraction like Zr/Nb ratios of the studied rock suites correlate with melt temperature, Pb/Nb(Zr) ratios, Pb isotopes, and water and silica contents (Figs. 11 and 12). Since, as shown above, these correlations cannot be the result of crystal fractionation or crustal contamination, they must reflect the effects of hydrous slab-derived contributions into the mantle wedge. At first sight, the elemental systematics are entirely consistent with a flux melting model, in which the addition of volatiles induces higher extents of melting and the formation of wet magmas with higher LILE/HFSE ratios. Silica contents of the AC and SR rocks are below 55 wt%, within the range of experimental hydrous melts of peridotite (Grove et al., 2012), and therefore it seems conceivable that they could be formed by flux melting of peridotite alone. Nonetheless, and as shown above, these rocks also display olivines that trend to much higher NiO and lower CaO contents than typical melts of peridotite, thus suggesting derivation from a different kind of primitive liquid. Are the olivines recording a unique petrologic transformation in the source of arc magmas that is not captured by other chemical proxies? And if so, then what do they tell us about the melting regime in the subduction zone?

It has been recently suggested that the observed trace-element variations in olivines only reflect differences in their temperature-dependent partitioning, and thus can be entirely explained by the melting regime and the so-called lid effect (Niu et al., 2011). Magmas formed in a thick crustal setting such as Hawaii equilibrate deeper and at higher temperatures than magmas formed at ocean ridges (Niu et al., 2011), and therefore their lavas will display high NiO and low CaO contents and consequently crystallize high-NiO and low-CaO olivines upon cooling. The larger extents of melting expected at ocean ridges will thus hinder the formation of high-NiO and low-CaO olivines due to their lower equilibration temperatures and pressures, which also leads to the possible exhaustion of pyroxene from their source (Niu et al., 2011).

Because crustal thickness does not change significantly along the TZR, the lid effect is not directly applicable to western Mexico. Yet the observed chemical variations in the Mexican rock suites are exactly the opposite of what can be deduced from the melting relationships suggested by Niu et al. (2011). In Mexico, rocks presumably formed at higher extents of melting are the ones that crystallized olivines with higher NiO and lower CaO. If water is the principal driving mechanism of melting in arcs, then olivines crystallized from wet melts should be expected to display higher CaO and lower NiO, because water expands the stability field of olivine at the expense of minerals like pyroxene and plagioclase (Grove et al., 2012). Since this is not observed, the olivine evidence from Mexico indicates that subduction must exert a significant lithologic modification in the source. In other words, the source of arc magmas is probably not a typical peridotite with higher water contents that melts to higher extents as commonly argued (Stolper and Newman, 1994; Kelley et al., 2006; Langmuir et al., 2006) but an entirely different pyroxene-rich lithology that is created within the wedge by silica-rich fluids or melts derived from the subducted crust. If true, then the formation of these secondary pyroxenites can bring some additional insights into the melting and tectonic regime operating in western Mexico, and arcs in general.

Intraplate Magmas and Slab Plumes

As mentioned before, dry intraplate magmas erupt side by side with other more typical arc volcanoes forming a compositional continuum that can be ascribed to both decompression melting and secondary pyroxenitization, presumably associated with addition of silica and other hydrous components from the subducted slab. The question then is to decipher the conditions in which these diverse magmas can be formed within the context of tectonic extension superimposed on a convergent margin.

Figure 13 shows possible petrotectonic models relevant for the western Mexican subduction zone. If a dry peridotite and a pyroxenite are allowed to ascend from depth along the same adiabat (Fig. 13A), the pyroxenite will start to melt sooner (and deeper) and thus will melt to higher extents than the peridotite due to its relatively lower solidus temperature. Lower extents of melting will be recorded in melts from the dry peridotite, whereas higher extents of melting will be observed by pyroxenite melts ascending from greater depths. Nonetheless, and similar to what was described by Niu et al. (2011) for the oceanic realm, high melt fractions will not show higher NiO and lower CaO olivines because pyroxene should be the main phase contributing to the melt, and thus be amongst the first minerals to be exhausted from the source along the melting column. Therefore, unless the secondary pyroxenite occupies a very large volume, as in the Hawaiian plume (Sobolev et al., 2005), it seems unlikely that a significant amount of pyroxene will remain in solid form during ascent through a longer decompression pathway.

It has been suggested that intraplate magmas in Mexico could mark the infiltration of deeper and hotter asthenospheric mantle through a slab window formed after a slab-detachment event (Ferrari, 2004). Geophysical models have also shown that a toroidal mantle flow may be introducing sub-slab mantle into the Mexican wedge as a consequence of slab rollback (Soto et al., 2009). It thus seems conceivable that conductive heating from a hotter upwelling mantle could induce melting of shallower and previously damped mantle portions of the wedge, while at the same time producing deeper mantle melts without any subduction influence (Ferrari, 2004). Nonetheless, even if sub-slab mantle may indeed be toroidally flowing into the wedge (Soto et al., 2009), Figure 13B shows that a wet peridotite that is conductively heated from below will start to melt at lower temperatures than a dry mantle, and will gradually melt more as the temperature increases. Initial fusion products will be relatively cooler, water-rich, and NiO-poor, while further displaying trace-element characteristics indicative of lower extents of melting. Dryer magmas will obviously be hotter and higher in NiO, but they could only be created at much higher extents of melting either because they will be ascending from greater depths, or because the wet source was largely diluted by melting at higher temperatures. In effect, in this model, the trace-element variations (Figs. 11 and 12), as well as the trace-element systematics of olivine phenocrysts (Fig. 7), will be opposite to what is observed in western Mexico.

Figure 13C provides our proposed solution to this petrologic dilemma. A nominally dry peridotite will remain solid in the mantle wedge unless perturbed by some kind of external influence. If it is allowed to ascend dry in an extensional tectonic regime like the one operating in western Mexico, it will probably generate very small melt fractions: the more fusible components will preferentially contribute to the melt, olivine will be the main residual phase, and the liquid compositions will likely be alkaline and similar to those observed in the SB suite and the Mexican oceanic islands. Alternatively, if the same peridotite can be metasomatically transformed into pyroxenite, it will inevitably start to melt because of the effective reduction in the solidus temperature. Melt compositions will in turn describe the well-documented positive correlations among trace-element proxies of slab addition and melt fractions (Figs. 11 and 12), but as opposed to flux melting of peridotite that produces basalt, the resulting magmas will gradually become andesitic, and they will crystallize high NiO and low CaO olivines because the original olivine in the source was effectively transformed into secondary pyroxene by the addition of slab silica. We then need a mechanism to create reactive pyroxenites at virtually the same depths where dry mantle upwelling is creating alkaline mantle melts.

Gómez-Tuena et al. (2011, 2013) recently proposed that the chemical diversity of arc volcanic rocks in western Mexico can be attributed to compositionally different kinds of fluids and/or melts, presumably extracted from the subducted assemblages at different metamorphic conditions. Nonetheless, and as opposed to what would be expected during extraction of metamorphic fluids at increasing slab depths (Kessel et al., 2005; Plank et al., 2009; Cooper et al., 2012), the compositions of arc volcanic rocks in western Mexico do not describe a direct relationship with the predicted thermal structure of a typical subduction zone: rocks at the volcanic front in Mexico record higher slab-fluid temperatures than rocks at the rear arc, even if the slab below the latter area rests at >300 km deep. In order to solve this puzzling discrepancy, Gómez-Tuena et al. (2013) proposed that stratovolcanoes in Mexico can be related to low-pressure melts of hybrid mechanical mixtures of eroded crustal lithologies, sediments, fragmented amphibolite-facies MORB, and highly hydrated mantle that are able to detach from the subducted slab due to differential buoyancy, forming a well-connected network of conduits and diapirs (Gerya and Yuen, 2003; Gerya et al., 2004; Behn et al., 2011; Gerya, 2011). Once incorporated into the hot core of the mantle wedge, these diapirs will gradually heat up by diffusion and experience a plethora of dehydration and metamorphic reactions that almost inevitably should lead to melting because of the lower solidus temperature of most of their constituents. Although constraining the sources and pathways of all these reactions is an extremely complex task (Tumiati et al., 2013), we speculate that the chemical continuum observed among volcanic rocks in western Mexico, from intraplate basalts to arc andesites (Figs. 11 and 12), must be intimately related to the low-pressure interaction of ascending slab-derived diapiric mélanges with the surrounding mantle (Fig. 13C).

Experimental evidence and supporting numerical models (Castro et al., 2010, 2013), as well geochemical evidence from andesitic volcanoes in western Mexico (Gómez-Tuena et al., 2013), support the notion that wholesale melting of subduction mélanges can create primitive andesitic magmas directly from the mantle at convergent margins. The transitional nature of rocks studied here indicates that the processes of mantle hybridization can also be more subtle, and perhaps even involve more intricate reactions among components. As proposed by Straub et al. (2011, 2013a), it may be possible that rocks from the AC and SB suites are registering the initial stages of reaction between diapir-derived siliceous melts and the peridotitic mantle in the form of silica-deficient pyroxenites. It also is conceivable that these rocks represent the wholesale melting products of highly diluted but discrete diapirs that contain only a very small proportion of recycled slab materials. Nonetheless, and whichever model we adopt to explain their petrogenesis, it seems clear that the petrologic and geochemical diversity in western Mexico cannot be simply the product of variable amounts of slab-derived water affecting an otherwise homogenous mantle peridotite. At least for the Mexican case, subduction seems to create profound transformations in the chemical and petrologic constituents of the upper mantle, and can therefore produce a plethora of magmatic compositions, from basalts to andesites and even dacites, that readily depart from the rather homogeneous basaltic crust that dominate the ocean basins.

Intraplate magmas emplaced in the Mexican arc are indeed unusual eruptive products for a convergent margin. Recording mantle melting under nearly anhydrous conditions and with negligible contributions from the continental crust, these kinds of magmas provide a unique opportunity to understand the compositional underpinnings upon which a continent rests and grows. However unusual these magmas may be, the geochemical evidence indicates that there is really nothing particularly exotic in the composition of their mantle source because magmas like these have been generated in a wide variety of tectonic settings over a gigantic area that includes the Pacific basin, continental Mexico, and may even extend to Central America. The mantle wedge below western Mexico could be experiencing replenishment by infiltration from below or the side as some models suggest (Ferrari, 2004; Soto et al., 2009), but there is no a priori reason to believe that these new additions must be radically different from the mantle that already resides in the mantle wedge or from the mantle that can be advected from behind the arc by subduction-induced convection. The only necessary condition that needs to be met for recognizing its pristine heterogeneous nature is allowing it to melt in a small amount and under nearly anhydrous conditions. In other words, what seems to be changing in western Mexico is not the composition of the pristine background mantle but the mechanism by which it is processed and transformed by the subduction zone.

Isotopic and trace-element constraints from the most pristine intraplate rocks unequivocally point to recycling of ancient subducted lithologies, but mineral chemistry indicates that these enriched contaminants must be relatively small, widely disaggregated, and almost entirely hosted in peridotite. Despite their pervasiveness, small heterogeneities like these only become visible when mantle melting occurs at its lowest extents, and are easily erased at larger melt fractions or when overprinted by even the smallest amount of slab contributions. Slab-derived hydrous fluxing has been traditionally held responsible for enhancing melting conditions at convergent margins, but evidence from Mexico indicates that the subduction flux can also impose a profound lithologic transformation of the mantle wedge, so that it can no longer be simply treated as a wet peridotite.

Trace-element compositions in olivines signal the importance of metasomatic transformations of the mantle source in arcs, presumably as a result of olivine consumption by the infiltration of silica-rich melts into the mantle wedge. Metasomatic silica addition must be intimately related to the convergent margin, but the petrologic and tectonic context of western Mexico precludes a direct derivation from the slab surface. The participation of hybrid subduction mélanges that detach from the downgoing slab and melt in the hot core of the mantle wedge can provide a viable alternative explanation for magma genesis in convergent margins.

This work is part of B.A. Díaz-Bravo’s Ph.D. dissertation, financed by a Consejo Nacional de Ciencia y Tecnología graduate student scholarship. Our most sincere thanks to Carlos Linares, Rosaisela Leija, Manuel Albarrán, Ernestina Salazar, Juan Vázquez, Patricia Girón, and Rufino Lozano for their help during sample preparation and analyses. We thank Susanne Straub, Laura Mori, Fernando Ortega, Alma Vázquez, Dante Morán, Luca Ferrari, and Vlad Manea for stimulating discussions. Susanne Straub kindly provided a fragment of the sco-sms olivine for analysis. The manuscript benefited greatly from the thoughtful comments of Brian Marshall and an anonymous reviewer. Editorial handling by Carol Frost is also highly appreciated. This research was funded by Programa de Apoyo a Proyectos de Investigación e Innovación Tecnológica (PAPIIT) Universidad Nacional Autónoma de México projects IN107810 and IN103907 to A. Gómez-Tuena.