Only at slow-spreading ridges are centimeter- to decimeter-sized veins of SiO₂-enriched rocks known. These are often called “oceanic plagiogranites” (see Koepke et al., 2007, for definition) or “felsic veins”, cutting the lower gabbroic crust in many places. These rocks generally account for less than 1 vol % of the crust (see Koepke et al., 2007, for details). The best example of such felsic veins from slow-spreading ridge systems is the gabbroic oceanic core complex “Atlantis Bank” at the Southwest Indian Ridge (SWIR, Fig. 1), where ODP (Ocean Drilling Program) and IODP (International Ocean Discovery program) drilled cores at three different sites (see the review of Dick et al., 2019). The longest drill core was drilled at Site 735B between 1987 and 1998, encountering ∼ 1.5 km of gabbroic rocks, which are cut by hundreds of igneous felsic veins, consisting of diorite, tonalite, and trondhjemite (Dick et al., 2000). Two models have been suggested for the generation of felsic veins in gabbroic complexes from slow-spreading ridges: (1) crystal fractionation in MORB magmas, resulting in highly differentiated melt (e.g., Berndt et al., 2005; Niu et al., 2002; Chen et al., 2019; Zhang et al., 2025), and (2) hydrous remelting of gabbroic rocks in the deep oceanic crust (e.g., Flagler and Spray, 1991; Brophy and Pu, 2012; Koepke et al., 2004, 2007; Zhang et al., 2025). Both models have been verified by experimental studies (e.g., Berndt et al., 2005; Koepke et al., 2004; Wolff et al., 2013).

With regard to model 2, the corresponding experiments on hydrous partial melting of gabbro were conducted using natural starting material and significantly different grain sizes. While Koepke et al. (2004) used very fine-grained material to achieve global equilibrium conditions, Wolff et al. (2013) used microrocks of the same gabbro to shed light on the kinetics of the partial-melting process. Although far from global equilibrium, the observed phase relations and compositional trends in the microrock experiment results are practically identical to those in runs using a very fine-grained powder. This provides a detailed insight into the reaction kinetics (Wolff et al., 2013). Interestingly, textural features and compositional characteristics obtained in the partial-melting experiments on microrocks are in good agreement with corresponding features of interstitial assemblages in some natural oceanic gabbros from slow-spreading ridges. This suggests that some natural late-stage parageneses may be better explained as residual phases left behind after a hydrous partial-melting event rather than by the crystallization of a late, evolved percolating melt.

Two longer cores were drilled at the top of the Atlantis Bank: Hole 735B (ODP Legs 118 and 176), which was drilled between 1987 and 1998 (see Dick et al., 2000, for details), and Hole U1473A, which was drilled by IODP Expedition 360 in 2016 (for details, see MacLeod et al., 2017; Dick et al., 2019; Fig. 1). Comparing the two drilling processes demonstrates continuity in both holes, emphasizing the complex interplay of magmatic accretion and steady-state detachment faulting over a period of ∼ 128 kyr (MacLeod et al., 2017). The results show that the lower crust of the Atlantis Bank was built up through repeated cycles of intrusion by several bodies of olivine gabbro, each hundreds of meters in size.

Several dozen publications on the gabbro rocks drilled in both holes provide a foundation of knowledge on how the deep oceanic crust under slow-spreading ridges is formed. The gabbros from both drill cores are composed of a variety of gabbroic lithologies, from primitive olivine gabbros to evolved oxide-rich gabbros, cut by hundreds of millimeter- to centimeter-thick felsic veins (Dick et al., 2000; MacLeod et al., 2017). Examples of findings relating to gabbros drilled in Hole 735B can be found in the work of Natland et al. (1991) and Dick et al. (2000, 2002). Several publications focusing on gabbros from Hole U1473A over the last 7 years demonstrate the geoscience community's general interest in the Atlantis Bank gabbros. Petrological studies focusing on magma formation and evolution show that “simple” fractional crystallization is not the main process behind the formation of crustal magmas. Instead, the importance of other processes is emphasized: assimilation/fractional crystallization as the key mechanism in lower crust formation (Dhar et al., 2022), the role of compaction in melt extraction and accumulation (Ferrando et al., 2021), the pervasive reactive transport of ferrobasaltic melts formed during early crystallization of Fe–Ti oxides as monitored by amphibole geochemistry (Ferrando et al., 2022), melt rock reaction by the intergranular reactive porous flow of a melt migrating through the mush framework (Sanfilippo et al., 2020), processes controlling halogen (F, Cl, Br, I) abundances (Kendrick, 2019), the importance of melt migration by intrusion processes or porous flow at all scales (Boulanger et al., 2021), the role of multi-stage melt migration due to compaction and syn-magmatic deformation during crustal accretion (Zhang et al., 2020).

Several papers have been published focusing on the felsic veins drilled in the Atlantis Bank at the SWIR. By investigating seven felsic veins cutting through gabbro in Hole U1473A, Nguyen et al. (2018) determined that the formation occurred through fractional crystallization. Based on a structural geological survey of 14 felsic veins, Ma et al. (2020) discussed two generations of felsic veins in the core drilled in Hole U1473A. For type-1 veins, they proposed late fractional crystallization as the formation model. For type-2 veins, which are associated with dikes, the authors considered it to be the case that anatectic melts were formed by hydrous partial melting of gabbros, triggered by fluids derived from seawater. While the focus of this study is on structural geology, no valid geochemical or petrological evidence is presented for the mode of formation of the veins studied, except for some microstructures and An-rich rims in the plagioclases of the host gabbros, suggesting a genesis by hydrous partial melting of gabbro. Other authors investigating these veins focused on fractional crystallization as the relevant process, resulting from oxide fractionation at a late stage of tholeiitic magma evolution (Niu et al., 2002; Chen et al., 2019). For the felsic veins in Hole 735B, Chen et al. (2019) concluded that the formation of the felsic melts was primarily due to extreme fractional crystallization, as suggested by the bulk TiO₂ vs. MgO trend. However, this trend is highly heterogeneous and is based on only a few data points, which appear to have been affected by contamination from cumulate phases in the host gabbros. The most recent paper investigating the formation mode of felsic veins drilled in Hole U1473A is by Zhang et al. (2025), with a focus on the geochemistry of zircons and geochemical modeling. These authors provide compelling evidence that both extreme fractionation of MORB and hydrous partial melting of gabbros played significant roles in the formation of felsic melts at the Atlantis Bank.

Koepke et al. (2005a, b) focused on microstructures in gabbros from slow-spreading ridges and suggested that the felsic veins could be the result of partial melting triggered by high-temperature hydrothermal fluids that were transported along grain boundaries deep into the crust via high-temperature shear zones. In Koepke et al. (2005b), one example from the Atlantis Bank at the SWIR drilled in Hole 735B is presented. This model is also supported by the relatively low TiO₂ concentration in the veins, which is a consequence of the depleted nature of typical cumulate gabbros in the lower oceanic crust (see Koepke et al., 2007, for details). Finally, Natland et al. (1991) proposed liquid immiscibility between an oxide-rich and a silicate-rich melt as the possible origin of the felsic veins from Hole 735B. Pietranik et al. (2017) examined the composition of zircons in felsic veins and oxide gabbros from Hole 735B. Their findings suggest that there are different ways in which felsic melt can form. The zircons generally exhibit variable compositions, which is consistent with their crystallization from melts formed by crystal fractionation of mid-ocean ridge basalt (MORB) and hydrous partial remelting of deep cumulate gabbros.

This study aims to provide evidence that hydrous partial melting of gabbros is one of the processes responsible for the formation of felsic veins in gabbros from Hole 735B. Until now, the microstructures of residual phases produced by hydrous partial-melting reactions have only been documented in a few locations in the recent oceanic crust generated at slow-spreading ridges (Koepke et al., 2005a, b). Of the Atlantis Bank at the SWIR, where the majority of studies on felsic veins cutting gabbros have been published, only one rock example has been presented by Koepke et al. (2005b) to date. This paper addresses the issue by presenting numerous additional examples of these characteristic microstructures in drilled gabbros from the Atlantis Bank at the SWIR.

Unlike the many studies on felsic veins from Hole 735B mentioned above, this paper focuses not on the petrology and geochemistry of the felsic veins themselves but on the microstructures preserved in the gabbros, which imply that hydrous partial melting has occurred. Key features of this process include zones of An-enriched plagioclase on grain boundaries, which form parageneses with interstitial brown amphibole and orthopyroxene. These can be interpreted as the residual phase assemblage left behind following a hydrous partial-melting event. The relevant microstructures have been found in about half of the gabbroic rocks from the ODP drill core 735B investigated in this study, emphasizing the importance of hydrous partial melting of gabbro as a mode of felsic melt formation. As these microstructures are often overlooked during petrographic work, this paper provides a catalogue of examples demonstrating how they can be identified using back-scattered electron (BSE) imaging alongside microanalytical tools (Figs. S1–S14 in the Supplement).

In total, 29 gabbroic samples from drill core 735B have been investigated. Of these, 14 samples exhibited microstructures and compositions at grain boundaries that indicated the occurrence of hydrous partial-melting processes. For these 14 samples, electron microprobe data are presented (Table 1). Additionally, analytical profiles for each sample across grain boundaries between plagioclase grains affected by hydrous partial melting are presented, as well as back-scattered electron images (see the Supplement). A cathodoluminescence (CL) light image for one sample is also presented (Fig. 2).

Mineral phases were analyzed using an electron microprobe. Around 5000 analyses, primarily via profiles through grain boundaries affected by hydrous partial melting, have been presented. Of these, 79 averages were evaluated based on approximately 800 individual analyses. The data are shown in Table 1.

A Cameca SX100 electron microprobe, equipped with five spectrometers and operating with the Cameca software “Peak Sight”, was used. All data were obtained using a static (fixed) beam, Kα emission from all elements, and a 15 kV acceleration potential. The “PAP” matrix correction (Pouchou and Pichoir, 1991) was applied using natural and synthetic standards. Most element concentrations were obtained with a beam current of 15 nA and a counting time of 10 s for the peak and background. Higher beam currents (40 nA) and increased counting times (60–120 s) were applied for F and Cl in amphibole and for Mg, Ti, and K in plagioclase, leading to more accurate trace element data and lower limits of detection. The limits of detection (in wt %) are as follows: TiO₂ in mafic silicates – 0.03; TiO₂ in plagioclase – 0.01; Al₂O₃ in mafic silicates – 0.02; Cr₂O₃ and NiO in silicates – 0.04; MnO in mafic silicates – 0.03; MgO in plagioclase – 0.005; CaO in olivine – 0.02; Na₂O and K₂O in mafic silicates – 0.02; K₂O in plagioclase – 0.005; Cl in brown amphibole – 0.005; F in brown amphiboles amphibole – 0.1. As the F content of brown amphibole is mostly below the detection limit and because the data are highly scattered, the F content of amphiboles has not been considered and has been excluded from the data table.

The gabbroic rocks investigated in Hole 735B are oxide-bearing olivine gabbros. They exhibit a granular texture ranging from fine to coarse, and foliation is sometimes visible due to crystal–plastic deformation. This phenomenon is evident in numerous gabbros from this site (e.g., Dick et al., 2000). The rocks consist of primary plagioclase, clinopyroxene, and olivine and often contain prismatic orthopyroxene. There is frequently a notable secondary mineral growth consisting of brown amphibole, orthopyroxene, and Fe–Ti oxides, which rim the primary minerals and form granular interstitial assemblages.

The focus of this paper is these interstitial assemblages. They are usually explained by typical late-stage crystallization, precipitated from the final melt drops enriched in SiO₂ and incompatible elements such as P and Zr. This enables the crystallization of apatite and zircon, respectively (see Koepke et al., 2018, for details). However, if such late-stage paragenesis also contains zones of plagioclase that are strongly enriched in An content, it is much more probable that the minerals form a residual assemblage after hydrous partial melting of the gabbro (Koepke et al., 2005a, b; Wolff et al., 2013). This is demonstrated in Fig. 2, which shows such an assemblage. Microscope images taken using plan-polarized or cross-polarized light do not reveal the zones of An-enriched plagioclases. It is not possible to distinguish between the two modes of forming interstitial parageneses in such an image (Fig. 2a). However, a solution is provided by BSE images, in which the An-enriched plagioclases can be easily identified. These normally nucleate at the grain boundary between grains of primary plagioclase. Due to their higher density and, consequently, greater electron backscattering effect, these zones appear to be brighter in BSE images, enabling them to be distinguished from the darker host plagioclase (Fig. 2b). BSE images of these parageneses in the samples investigated, which are indicative of proceeded hydrous partial melting of gabbro, are presented in the Supplement.

Another way to identify zones containing An-enriched plagioclase is to use CL light. Exposing the sample surface to electrons results in photon emissions that produce luminescence in the mineral phases, which has a characteristic color corresponding to the luminescence wavelength. This is a complex function of trace element substitutions and defect structures in the irradiated minerals. As the An-rich plagioclase produced by partial melting exhibits a notably distinct trace element composition (see Sect. 3.2), the reactive plagioclase at grain boundaries exhibits a notably different luminescence color (yellow-green color in Fig. 2c), contrasting with the primary plagioclase with lower An content (blue in Fig. 2c).

The mineral analyses obtained using (electron probe microanalysis) EPMA are presented in Table 1. The primary minerals are highly homogeneous, with no significant zoning. The Mg# (molar MgO/(MgO + FeO) ⋅ 100) of olivine and the An content of primary plagioclase are relatively low, ranging from 56 to 82 and from 49 to 67 mol %, respectively. This is because the gabbros from the Atlantis Bank at the SWIR are relatively evolved (see Coogan, 2014; Dick et al., 2000, 2002), as can be seen in the Mg# in olivine versus the An content in plagioclase diagram (Fig. 3). This diagram shows the magmatic evolution of the gabbros investigated, which follows the same trend as the whole 735B drill core (Dick et al., 2002).

As mentioned above, the focus of this paper is on the secondary interstitial minerals generated by the hydrous partial melting of gabbros. This is indicated by zones of An-enriched plagioclase at the boundaries between primary plagioclase grains, which coexist with interstitial brown amphibole and orthopyroxene. To verify the An enrichment in the plagioclase, BSE imaging and EPMA profiling through such zones at plagioclase–plagioclase grain boundaries have been performed. This is demonstrated by an example, showing a typical interstitial paragenesis consisting of brown amphibole, orthopyroxene, and An-enriched plagioclase, which are embedded within the primary granular cumulate assemblage of olivine, clinopyroxene, and plagioclase lower in An (Fig. 4). The plagioclase compositions in such zones exhibit distinctive characteristics. Aside from the substantial enrichment in An content, they are significantly depleted in trace elements, as illustrated in Fig. 4 for the elements Mg, Ti, and K. The corresponding analyses with EPMA have been performed by applying significantly higher beam currents and counting times, resulting in more accurate trace element data and lower limits of detection (for details see Sect. 2.2). A further characteristic feature of such zones is the asymmetrical, heterogeneous nature of the compositional profiles (Fig. 4), which is the reason why it is not possible to present simple average values of the An-enriched plagioclase. Representative analyses of plagioclase with the strongest An enrichment in such zones are presented in Table 1 instead.

Mg# of orthopyroxene varies between 66 and 78, which is significantly higher than values for the corresponding primary olivines of the same rock, varying between 56 and 76 (Table 1). Brown amphibole forming rim and interstitial parageneses with orthopyroxene and An-enriched plagioclase are of pargasitic/hastingsitic composition, with high concentrations of Na₂O (from 2.1 to 2.9 wt %) and Al₂O₃ (from 10.7 to 13.7 wt %, Table 1). The TiO₂ contents are relatively high, leading to formation temperatures by applying the Ti-in-amphibole geothermometer of Ernst and Liu (1998) from 717 to 1017 °C (Table 1), mostly indicating a magmatic formation overlapping with the magmatic–metamorphic regime. Volatile contents are generally low, with F contents mostly below the EPMA detection limit and a Cl content varying from below the detection limit to 0.01 wt %.

Since gabbroic rocks from the Atlantis Bank are well known for their intense late-stage crystallization during magmatic processes, resulting in widespread occurrence of amphibole-bearing oxide gabbros (e.g., Dick et al., 2000, 2002; MacLeod et al., 2017), clear evidence is needed to confirm the formation of a similar interstitial assemblage as residual paragenesis following hydrous partial melting of gabbro. This will be discussed in more detail below, presenting all of the characteristic features that are indicative of a residual phase assemblage generated by the hydrous partial melting of gabbro.

The most striking feature is the presence of zones enriched in An at the grain boundaries between the primary plagioclases, which form parageneses with brown amphibole and orthopyroxene (Fig. 4). The reason for the An-enrichment in plagioclase is the transition from a dry system during cumulate formation to a system with high water activities during the partial-melting event. This causes plagioclase to crystallize with a high An content (e.g., Sisson and Grove, 1993; Berndt et al., 2005). This feature has also been reproduced in experimental studies of partial melting of gabbro (Koepke et al., 2004; Wolff et al., 2013). In an experimental study using “microrocks” as starting material, Wolff et al. (2013) managed to replicate the interstitial nature of the residual paragenesis following a partial-melting process. The authors highlighted the discontinuous nature of the partial-melting reaction, as evidenced by the interstitial growth of amphibole and orthopyroxene associated with primary olivine and clinopyroxene, as well as the growth of An-enriched plagioclase zones on grain boundaries between primary plagioclase grains. As presented in Fig. 5, the difference in An content between the An-enriched zones and the primary plagioclases in the investigated gabbros varies considerably, from 12 to 35 mol %. The zoning of the plagioclase, as indicated by the presence of An-rich rims, contrasts sharply with the typical zoning observed in many oceanic gabbros, including those from the Atlantis Bank. An example of normal zoning in plagioclases is presented in Fig. 6 for an olivine gabbro from the 735B drill core. As expected, an EPMA profile through a rim of such a normal-zoned plagioclase shows that it is impoverished in An content (Fig. 6). This is a consequence of the composition of typical late-stage melts, which are relatively evolved compared to the primitive nature of MORB that fed the gabbro's main stage crystallization (e.g., Koepke et al., 2018). The presence of zones of An-enriched plagioclase on grain boundaries in oceanic gabbros is essential for a hydrous partial-melting formation model. The corresponding microstructures found in the 735B gabbros investigated, including EPMA traverses through zones of An-enriched plagioclase, are presented in the Supplement.

In summary, the three observations listed above strongly suggest that the observed microstructures formed through hydrous partial melting of gabbro rather than normal fractional crystallization at a later stage. Another compelling argument is that relating to the low bulk rock concentration of TiO₂ in the felsic veins, which is characteristic of the depleted nature of the cumulate gabbros starting material. Figure 7 shows the corresponding TiO₂ vs. SiO₂ diagram, including all published bulk data of felsic veins from the Atlantis Bank drill cores. The vast majority of the data points lie below the line representing the minimum TiO₂ values obtained from experiments on MORB differentiation in tholeiitic systems. This implies that at least some of the felsic veins in the Atlantis Bank drill core are formed by hydrous partial melting of gabbro. For this diagram to be applicable, the felsic veins must represent frozen melt and not cumulate rocks, as the surrounding gabbros do. Many of the felsic veins exhibit granophyric textures, consisting of intergrowths of quartz and albite-rich plagioclase (see MacLeod et al., 2017). This texture is believed to result from eutectic crystallization (e.g., Johannes and Holtz, 1996), which provides compelling evidence that the felsic veins formed from a melt. In summary, the results of this study independently support Zhang et al.'s (2025) conclusion that hydrous partial melting occurred in the gabbros of the Atlantis Bank, producing anatectic melts.

In light of our finding that the drilled gabbros from Hole 735B underwent hydrous partial melting, the question arises as to whether the water-rich fluids were magmatic or hydrothermal. In general, only a very small amount of water is needed to produce water-saturated melts at low pressures. For example, at 1 kbar (equivalent to a depth of 3 km), around 2.5 wt % of water is needed to create a water-saturated melt (Berndt et al., 2002). While this corresponds to 100 % melt, the amount of melt is significantly lower during anatectic processes at ridges (see the experiments of Koepke et al., 2004). Thus, at this pressure, only a few droplets of water are enough to trigger partial melting. It is interesting to consider the water/rock ratio, which, in this case, is ∼ 0.025 (to generate 100 % melt). In real conditions of hydrous partial melting under mid-ocean ridges, the water-to-rock ratio can be even lower. This implies that only a small amount of water at temperatures just above the gabbro solidus is necessary to initiate the partial-melting process in the gabbro. A key question is whether these water-rich fluids are of magmatic or hydrothermal origin. The usual approaches to answering this question are (1) applying stable isotopes to the felsic veins and (2) analyzing chlorine and fluorine in the magmatic phases in the related rocks, especially in amphibole and apatite. We will focus on these two aspects below.

Relatively high 87Sr/86Sr ratios and depleted δ18O values in bulk rocks from the lower oceanic crust typically indicate that seawater is the most likely source of a water-rich fluid operating at a very high temperature in the deep oceanic crust (e.g., Bosch et al., 2004). However, if the water/rock ratios are very low (e.g., <0.025, as discussed above for the case of initial hydrous partial melting of gabbro), the application of stable isotopes is hampered since it does not lead to unequivocal results. Stable isotopes only lead to clear results if the water/rock ratios are relatively high. The same applies to mantle-like zircon δ18O values, as analyzed by Zhang et al. (2025) for felsic veins from IODP Hole U1473. These values were interpreted as reflecting limited seawater involvement. However, as for bulk rocks, mantle-like zircon δ18O values do not rule out an origin from seawater-derived hydrous partial melting if the water/rock ratio is very low.

Ferrando et al. (2022) examined brown amphiboles found in gabbros drilled from Hole U1473 to determine the nature of the fluid present. According to these authors, the low Cl content of brown amphiboles indicates that they crystallized from melts containing a magmatic hydrous component. This suggests that the microstructures of hydrous partial melting observed in the gabbros in this study could have been formed by magmatic fluids percolating through the gabbro at temperatures above its solidus. On the other hand, Zhang et al. (2025) highlighted the crucial role of detachment faulting in the lower ocean crust of the Atlantis Bank core complex in facilitating seawater-induced hydrous partial melting at temperatures exceeding 750 °C. In summary, the discovered microstructures in the gabbro drilled in Hole 735B clearly indicate that hydrous partial melting occurred in the gabbro. However, the origin of the water that triggered this process remains unclear.

The residual microstructures observed in the Hole-735B gabbros, which are indicative of hydrous partial melting, imply that felsic melts have been generated. However, interstitial phases representing frozen melt associated with the residual microstructures have never been observed, e.g., granular felsic leucosomes associated with quartz. Therefore, they were not frozen in situ, as is typical of high-grade metamorphic migmatite terrains in the deep continental crust. This raises the question: what happened to these melts? In the following section, two options are discussed.

An important fact is that the gabbros from the Atlantis Bank core complex generally crystallized in a tectonically active regime, as evidenced by crystal–plastic deformation (e.g., Dick et al., 2019; MacLeod et al., 2017). Therefore, it is highly probable that the felsic melts, which were generated in a ductile regime, migrated into zones with less pronounced deformation, where they finally solidified. These locations for crystallization include fractures or cracks within gabbroic rocks in a cooler, eventually brittle regime. This results in the formation of widespread felsic veins within the gabbro massif. It is also highly probable that the anatectic melts mix with late-stage vagabonding melts formed by extended crystal fractionation, producing hybrid melts that exhibit characteristics of both modes of felsic melt generation. Such mixing could be indicated by the compositions of magmatic amphiboles that crystallized within the felsic veins, which show moderate F and Cl contents. This is the case for some amphiboles within felsic veins from Hole U1473A in the Ma et al. (2020) data set.

The following outlines our model for the formation of interstitial, late-stage parageneses with zones of plagioclase strongly enriched in An content by partial melting of oceanic gabbro at slow-spreading ridges initiated by hydrothermal fluids; this is illustrated in Fig. 8. This model corresponds to the ideas of Zhang et al. (2025), who proposed seawater-involved hydrous partial melting for their group-1 felsic veins cutting gabbros in Hole U1473A. The process involves three distinct stages:

The first step is the main-stage crystallization of olivine, plagioclase, and clinopyroxene in the axial magma beneath the mid-ocean ridge. After accumulation, the cumulate cools into the subsolidus regime but remains above the wet solidus temperature of gabbro (Fig. 8a).