Partial Melting of Lower Oceanic Crust Gabbro: Constraints From Poikilitic Clinopyroxene Primocrysts

. Here, we examine crystal-scale records of partial melting in lower crustal gabbroic cumulates from the slow-spreading Atlantic oceanic ridge (Kane Megamullion; collected with Jason ROV) and the fast-spreading East Paciﬁc Rise (Hess Deep; IODP expedition 345). Clinopyroxene oikocrysts in these gabbros preserve marked intra-crystal geochemical variations that point to crystallization-dissolution episodes in the gabbro eutectic assemblage. Kane Megamullion and Hess Deep clinopyroxene core1 primocrysts and their plagioclase inclusions indicate crystallization from high temperature basalt ( > 1,160 and > 1,200 ◦ C, respectively), close to clinopyroxene saturation temperature ( < 50% and < 25% crystallization). Step-like compatible Cr (and co-varying Al) and incompatible Ti, Zr, Y and rare earth elements (REE) decrease from anhedral core1 to overgrown core2, while Mg# and Sr/Sr ∗ ratios increase. We show that partial resorption textures and geochemical zoning result from partial melting of REE-poor lower oceanic crust gabbroic cumulate (protolith) following intrusion by hot primitive mantle-derived melt

Thi s v e r sio n is b ei n g m a d e a v ail a bl e in a c c o r d a n c e wit h p u blis h e r p olici e s. S e e h t t p://o r c a . cf. a c. u k/ p olici e s. h t ml fo r u s a g e p olici e s. Co py ri g h t a n d m o r al ri g h t s fo r p u blic a tio n s m a d e a v ail a bl e in ORCA a r e r e t ai n e d by t h e c o py ri g h t h ol d e r s .
Successive magma batches underplate, ascend, stall and erupt along spreading ridges, building the oceanic crust. It is therefore important to understand the processes and conditions under which magma differentiates at mid ocean ridges. Although fractional crystallization is considered to be the dominant mechanism for magma differentiation, open-system igneous complexes also experience Melting-Assimilation-Storage-Hybridization (MASH, Hildreth and Moorbath, 1988) processes. Here, we examine crystal-scale records of partial melting in lower crustal gabbroic cumulates from the slow-spreading Atlantic oceanic ridge (Kane Megamullion; collected with Jason ROV) and the fast-spreading East Pacific Rise (Hess Deep; IODP expedition 345). Clinopyroxene oikocrysts in these gabbros preserve marked intra-crystal geochemical variations that point to crystallization-dissolution episodes in the gabbro eutectic assemblage. Kane Megamullion and Hess Deep clinopyroxene core1 primocrysts and their plagioclase inclusions indicate crystallization from high temperature basalt (>1,160 and >1,200 • C, respectively), close to clinopyroxene saturation temperature (<50% and <25% crystallization).
Step-like compatible Cr (and co-varying Al) and incompatible Ti, Zr, Y and rare earth elements (REE) decrease from anhedral core1 to overgrown core2, while Mg# and Sr/Sr * ratios increase. We show that partial resorption textures and geochemical zoning result from partial melting of REE-poor lower oceanic crust gabbroic cumulate (protolith) following intrusion by hot primitive mantle-derived melt, and subsequent overgrowth crystallization (refertilization) from a hybrid melt. In addition, toward the outer rims of crystals, Ti, Zr, Y and the REE strongly increase and Al, Cr, Mg#, Eu/Eu * , and Sr/Sr * decrease, suggesting crystallization either from late-stage percolating relatively differentiated melt or from in situ trapped melt. Intrusion of primitive hot reactive melt and percolation of interstitial differentiated melt are two distinct MASH processes in the lower oceanic crust. They are potentially fundamental mechanisms for generating the wide compositional variation observed in mid-ocean ridge basalts. We furthermore propose that such processes operate at both slow-and fast-spreading ocean ridges.
In an analog 2D study on the Rum layered intrusion (Scotland), Leuthold et al. (2014a) showed evidence for small scale partial melting of equigranular gabbro by invading hot picrite and formation of troctolitic restite, subsequently refertilized to poikilitic gabbro. In the present study, we have selected poikilitic gabbro samples from the slow-spreading Mid-Atlantic ridge, at Kane Megamullion, and from the fastspreading Galapagos triple junction, at Hess Deep (Figure 1). We combine microtextural observations and mineral chemical measurements on clinopyroxene oikocrysts (oikocrysts are large crystals with several inclusions, named chadacrysts) to investigate high temperature partial melting of lower oceanic crust gabbro primocrysts (primocrysts are early formed crystals; i.e., at high temperature) by intrusion of primitive hot mantle-derived melts, and subsequent crystallization from hybrid melt. This early-stage Melting-Assimilation-Storage-Hybridization (MASH; Hildreth and Moorbath, 1988) process is distinct from the latestage strong zoning and incompatible element fractionation commonly observed in clinopyroxene rims, resulting from reactive porous flow of interstitial differentiated melt with plagioclase ±olivine ±clinopyroxene (Coogan et al., 2000;Gao et al., 2007;Lissenberg et al., 2013;Lissenberg and MacLeod, 2016) or post-crystallization diffusive fractionation (Coogan and O'Hara, 2015).
The focus of this paper is threefold: (1) To identify gabbro partial melting and hybridization with mantle-derived melt by studying clinopyroxene oikocrysts cores; (2) To estimate the effect of gabbro assimilation on MORB chemistry; (3) To test whether lower oceanic gabbro partial melting is a thermally realistic process.

ANALYTICAL METHODS
Mineral element maps (Figure 2) were acquired using a fivespectrometer JEOL JXA-8200 electron microprobe analyser (EMPA) at ETH Zürich, with a 15 kV accelerating voltage and a beam current of 100 nA. Major element compositions of minerals were determined using a 15 kV accelerating voltage with a beam current of 20 nA. Natural and synthetic silicates and oxides were used as standards. Straight-line clinopyroxene core-rim profiles were not possible because of the occurrence of several plagioclase chadacrysts. Chemical maps and isolated measurements were used to reconstruct complete profiles.
Clinopyroxene and plagioclase trace element contents were analyzed in situ using a Thermo Element XR mass spectrometer at ETH Zürich, connected to a 193 nm Resonetics ArF Excimer laser. The laser was operated in a Laurin Technic S155 ablation cell with a spot size of 30 µm, a frequency of 5 Hz and a laser power density of 2 J·cm −2 . The EMPA data were used as internal standards for all analyses. We used NIST SRM610 for external standardization and the GSD-1G basalt glass as a secondary standard. The raw data were reduced off-line using the SILLS software (Guillong et al., 2008). For 30 µm spots, 1σ uncertainties for REE and Sr are 4-7% (14% for La) and the deviation from certified values on secondary standard (GSD-1G) is <3-5% (except ca. 6-9% for Gd and Er). Core-rim profiles were duplicated with a reduced element list and spot sizes of 50 µm. For 50 µm spots, 1σ uncertainties for REE, Sr and Zr are 2-5% and the deviation from certified values on secondary standard (GSD-1G) is <5% (except ca. 8% for Gd).

Mid-Atlantic Ridge Poikilitic Gabbro
The Kane Megamullion (Mid-Atlantic Ridge, 23 • 30 ′ N) is an oceanic core complex comprising a series of domes oriented parallel to the spreading direction . The average spreading rate is 14.4 mm/year to the west and 9 mm/year to the east (Williams, 2007). The plutonic suite was sampled by the Jason ROV during the R/V Knorr cruise 180-2 in the Fall of 2004. It is composed of troctolite (14.9% of all collected gabbro s.l. samples by number), olivine gabbro (34.1%), ferrogabbro (i.e., oxide gabbro) (26.9%) and metagabbros (24.1%) . We have studied a coarse-grained poikilitic gabbro (JAS117-63), crystallized under lower oceanic crustal conditions. This sample was previously described by Lissenberg and Dick (2008) who reported details of several poikilitic gabbros through the oceanic crustal section.

DISCUSSION
In the following discussion, we first use experimental petrology and MELTS (Ghiorso and Sack, 1995) calculations to determine the liquid line of descent of primitive mantle-derived melts and the associated cumulate composition and chemistry. We then use partition coefficients to calculate parental melts in equilibrium with core1 and core2 clinopyroxene. Using mineral textures and zoning, we discuss the core1 to core2 geochemical evolution with respect to equilibrium and fractional crystallization and to the MASH model of Hildreth and Moorbath (1988). We then explore the importance of lower oceanic crust partial melting undergoing intrusion by hot mantle-derived melt, using thermal modeling. Finally, we discuss the effects of gabbro partial melting and assimilation on the composition and chemistry of MOR lavas and the lower oceanic crust.

Mantle-Derived Melt Liquid Line of Descent
The mantle source partial melting conditions strongly determine the derivative melt composition (e.g., Klein and Langmuir, 1987;Hirose and Kushiro, 1993;Kelemen et al., 1997a;Hirschmann et al., 1998). Deep melting produces Mg-rich and Si-, Al-poor primitive melt and high degree partial melting generates Mg-, Cr-rich and Al-poor melt (Figure 8).
(see Table A (Righter et al., 2006) and assuming either equilibrium or fractional crystallization. Results are shown in Figure 8 and discussed hereafter. Experimental petrology (e.g., Tormey et al., 1987;Grove et al., 1992) and MELTS (Ghiorso and Sack, 1995) calculations on primitive mantle-derived melt at crustal pressures successively crystallize olivine on the liquidus, followed by plagioclase shortly followed by clinopyroxene, and finally low-Ca clinopyroxene, pigeonite and orthopyroxene. The cumulates formed in the models are Cr-spinel bearing dunite, troctolite, gabbro and gabbronorite, assuming crystallization with a strong component of fractionation. At 150 MPa, 0.2 wt.% H 2 O and an initial Fe 3+ /Fe tot ratio corresponding to NNO-1 (≈NNO-2 at 1050 • C) conditions, olivine saturates at ca. 1,225 • C, plagioclase at ca. 1,220 • C and clinopyroxene at ca. 1,205 • C. Plagioclase, olivine and mostly clinopyroxene saturation temperatures gradually increase with increasing pressure. Plagioclase stability strongly decreases at higher water content. Since plagioclase and clinopyroxene co-crystallize, the liquid fraction decreases significantly (i.e., up to ca. 2 vol.%/ • C, from ca. 90 vol.%). Low Ca-pyroxene (fractional and equilibrium crystallization) and/or orthopyroxene (equilibrium crystallization) saturate last, replacing olivine at temperatures <1,150 • C and <1,055 • C, respectively. Low-Ca clinopyroxene, pigeonite and orthopyroxene were not observed in the two studied samples and we did not observe any evidence that either existed at an earlier stage. However, orthopyroxene occurs as a minor cumulus phase in Hess Deep primitive layered gabbro (in the same drill site 345-U1415I-4R; sometimes occurring as isolated crystals in olivine-gabbro or in orthopyroxene-rich bands; Gillis et al., 2014a,b). Orthopyroxene also occurs as oikocrysts and as exsolution lamellae in clinopyroxene cores in Hess Deep high-level gabbroic cumulates (Natland and Dick, 1996).
Our MELTS (Ghiorso and Sack, 1995) calculations are comparable to the Mid Atlantic Ridge, East Pacific Rise and Galapagos rift MORB glass database of Gale et al. (2013) and PetDB (Lehnert et al., 2000), in the area of the studied samples. A selection of major elements is presented in Figure 8. The MELTS algorithm gives reasonable estimates of the melt, olivine and clinopyroxene Mg# (Leuthold et al., 2015), and the lower oceanic crust primitive gabbro bulk rock Mg# composition (Gillis et al., 2014a) overlaps with our gabbro cumulate MELTS calculations (Figure 8). The MELTS calculations and geochemical results appear identical at 150 MPa, at least until saturation of Fe-Ti oxides (ca. 1,100 • C). Upon gabbro fractionation, the liquid is progressively depleted in MgO, Mg#, Al 2 O 3 , Cr 2 O 3 and CaO. The melt SiO 2 concentration is nearly constant during crystallization of clinopyroxene and slightly decreases after the peritectic reaction of olivine to pigeonite or orthopyroxene. The melt SiO 2 strongly increases at constant Mg# with magnetite crystallization, which strongly depends on the magma redox condition. The selected starting composition is SiO 2 -poor (<49 wt%) in comparison to other high Mg# MORB, likely produced at lower pressure. The liquid line of descent of such a SiO 2 -rich MOR melt (SiO 2 >50 wt%, Mg#>60) and generated cumulates is comparable to the ones described here.

Origin of Clinopyroxene Core1 Primocrysts
The use of MELTS (Ghiorso and Sack, 1995) clinopyroxene chemistry to estimate the clinopyroxene core1 crystallization conditions is difficult, as CaO shows no zoning, Fe and Mg were partly reequilibrated and Cr in clinopyroxene is not considered in MELTS. MELTS predicts clinopyroxene strong TiO 2 enrichment upon crystallization, from 0.3 at 1,205 • C to 1.3 wt% at 1,100 • C, independent of equilibrium or fractional crystallization from parental Si-poor or Si-rich MOR melt. With low TiO 2 concentrations, Kane Megamullion clinopyroxene core1 KM (0.55 wt%) and Hess Deep clinopyroxene core1 HD (0.35 wt%) probably crystallized at high temperatures of ca. 1,160 • C and ca. 1,200 • C, respectively. MELTS also predicts a decrease in the plagioclase anorthite content upon fractional crystallization, from An 81 at 1,220 • C to An 43 at 1,000 • C. Plagioclase chadacrysts in Hess Deep (An 82−81 ) clinopyroxene core 1 crystallized at ca. 1,220 • C. The crystallization temperature estimates for clinopyroxene and plagioclase are in agreement. The clinopyroxene core1 KM and core1 HD crystallized after moderate (1,160 • C corresponds to ∼50% crystallization) to low (1,200 • C corresponds to ∼25% crystallization) crystal fractionation, respectively, from primitive mantle-derived melt.
Cr 2 O 3 , TiO 2 and the REE are relatively immobile elements whose concentration evolves quickly upon crystallization. Thus, they represent ideal elements to calculate clinopyroxene parental magma chemistry, using partition coefficients. Based on high MgO-basalt and basalt equilibrium and fractional crystallization experiments at one atmosphere and 200, 700, and 800 MPa (Grove and Bryan, 1983;Grove et al., 1992;Villiger et al., 2007;Bédard, 2014;Leuthold et al., 2015), the clinopyroxenemelt partition coefficient for Cr 2 O 3 decreases upon cooling from ca. 20 [at oxygen fugacity <NNO, typical for MOR lavas (Righter et al., 2006) Gale et al. (2013), and PetDB, (Lehnert et al., 2000)] sampled along Mid Atlantic Ridge, East Pacific Rise and Galapagos rift axes, close to Kane Megamullion and Hess Deep, are shown. Spinel lherzolite partial melting arrows show the chemical evolution of mantle-derived melt upon pressure (1 to 1.5 GPa) and partial melting (F = 0.5 to 20%) increases [pMELTS by Ghiorso et al. (2002); based on KLB-1 spinel lherzolite of Hirose and Kushiro (1993), see composition in Table A.3] [note: partial melting of olivine-poor lherzolite produces a melt with lower Mg#, as seen in experiments by Hirose and Kushiro (1993)]. The dunite (D), troctolite (T), gabbro (G), gabbronorite (GN) and ferrogabbro (FG) cumulates calculated using MELTS [fractional crystallization of primitive MOR glass (see Table A.3) at NNO-1, 0.2 wt.% H 2 O and 0.15 GPa] reproduces the lower oceanic crust gabbro primitive cumulate from Gillis et al. (2014a). The melt liquid lines of descent, assuming pure fractional crystallization (FC) and equilibrium crystallization (EC), show early MgO and Al 2 O 3 decreases and subsequent SiO 2 increase due to magnetite crystallization. At higher water content and/or oxygen fugacity, the melt becomes SiO 2 -rich at high Mg#. Gabbro partial melting and hybridization with hot mantle-derived melt results in little but distinct physical and chemical modifications of the percolating primitive MOR melts [but may be confused with other parameters (e.g., water content, oxygen fugacity, pressure, equilibrium vs. fractional crystallization)]. The clinopyroxene core1 parental magma chemistry was determined using the major element chemistry of glass sampled close to Kane Megamullion and Hess Deep and whose TiO 2 , Cr 2 O 3 , and REE (Ce/Sm and Sm on Figure 9) are equivalent to the calculated core1 parental magma (see text for discussion of the partition coefficients).
The parental magma REE composition for clinopyroxene core1 was calculated using the partition coefficient predictive model of Wood and Blundy (1997). The REE partitioning depends on the crystal chemistry, itself buffered by the change in temperature and melt Mg# from core1 to core2 to rim and outer  (Gale et al., 2013), and PetDB, (Lehnert et al., 2000) databases]. Mid-Atlantic Ridge basalt (36 • N, 9 • N) high-Mg olivine-hosted melt inclusions are shown for comparison (some of which contain too much clinopyroxene components) (Sobolev and Shimizu, 1993;Shimizu, 1998). The Kane Megamullion and Hess Deep clinopyroxene core1 are in chemical equilibrium with the basaltic glass having low Sm concentration and low Ce/Sm ratio. Such compositions can be reproduced by depleted spinel-lherzolite to harzburgite partial melting (liquid fraction F ≈ 0.05-0.15) (starting compositions by Jochum et al., 1989) (modal and non-modal and also equilibrium and Rayleigh fractional melting with aggregated melts give very similar results). The melt chemistry in equilibrium with the clinopyroxene core2 is Sm-depleted and has low Ce/Sm in comparison to core1. In combination with resorption textures (Figure 2) and complex zoning (Figures 3, 4), it seems likely that hybridization of lherzolite-derived melt (F = 10-20%) with gabbro partial melts (F = 10-30%) in variable proportions occurred. The MOR melt samples with such chemical characteristics may have been overprinted by gabbro assimilation during transport through and storage in the lower oceanic crust. Upon gabbro fractional crystallization (FC), the melt Sm concentration increases at nearly constant Ce/Sm ratio. See Table A.4 for geochemical modeling of partial melting. rim, and thus remains nearly unchanged upon differentiation. Our resulting estimates match previously reported compositions for the Mid Atlantic Ridge, East Pacific Rise and Galapagos rift Sm-poor (<4 µg/g Sm) and low Ce/Sm (<0.7) glasses (Figure 9). Close to Kane Megamullion, such glasses have 49.5-51.9 wt.% SiO 2 , Mg# of 0.67-0.51 (7-10 wt% MgO), Cr 2 O 3 of 0.04-0.05 wt% and TiO 2 of 1.0 to 1.9 wt% [see Gale et al. (2013), and the PetDB (Lehnert et al., 2000) databases]. Close to Hess Deep, such glasses have SiO 2 ranging from 48.1 to 50.8 wt%, high Mg# of 0.69-0.62 (MgO: 8.5-10 wt%), high Cr 2 O 3 (ca. 0.06 wt%) and low TiO 2 concentrations (0.9-1.2 wt%). Overall, Hess Deep clinopyroxene core1 HD and plagioclase chadacrysts apparently crystallized from more primitive melt, at higher temperature than Kane Megamullion primocrysts.

Clinopyroxene Core1 to Core2 Transition
The degree of partial mantle melting (e.g., Hirose and Kushiro, 1993) and mantle composition (e.g., Shimizu, 1998;Salters and Dick, 2002;Lambart et al., 2009) affect the basalt chemistry. Residual depleted lherzolite or harzburgite are REE-(especially LREE-), Zr-, Ti-depleted (Jochum et al., 1989) in comparison to lherzolite and may produce similar melt as core2 parental melt (Figure 9). However, higher degrees of peridotite partial melting, either resulting from higher liquid fraction (F) and/or from partial melting of residual depleted lherzolite/harzburgite, produces a melt with higher Cr 2 O 3 (Hirose and Kushiro, 1993) and lower Sr content, which thus fails to explain the core1 to core2 evolution. Melt compatible Cr 2 O 3 , Al 2 O 3 , Mg#, Sr/Sr * regularly decrease during fractionation of a gabbroic mineral assemblage, while incompatible TiO 2 , Zr, Y and REE increase (e.g., Grove et al., 1992;Leuthold et al., 2015). Simple equilibrium or fractional crystallization thus fails to explain the geochemical zoning observed, which is characterized by decreases in Cr, Al, Zr, Y, REE and increases in Mg# and Sr/Sr * (Figures 3, 4), thus pointing to incoherent behavior of compatible and incompatible elements.
Clinopyroxene oikocrysts from both the Kane Megamullion and Hess Deep show complex microtextures (Figure 2) associated with compositional zoning. The core1 to core2 contact is not related to the clinopyroxene crystalline structure but is wavy, with embayments developed especially along plagioclase chadacryst edges. For these reasons, we exclude the possibility that it results from sector zoning. Cooling and crystallization also fail to explain the clinopyroxene core1 resorption texture. Indeed, diffusion into an interstitial reactive melt fails to explain the sharp zonation pattern defined by the "immobile" elements (Cr, Ti, Al, REE) between core1 and core2. However, diffusion took place during protracted cooling of magma chambers, as shown by partly reequilibrated elements with higher diffusivities (e.g., Fe, Mg, Sr). Taken together, (1) the clinopyroxene resorption texture between core1 and core2; (2) the associated increase of Sr/Sr * ; and (3) the plagioclase dissolution textures and reverse-zoning (Figures 2B, 7A) all point to resorption of both clinopyroxene and plagioclase primocrysts and hence to gabbro dissolution. Olivine grains are in chemical equilibrium with the low Mg# clinopyroxene rim, assuming a D Fe−Mg of 0.3 for olivine (Ulmer, 1989) and 0.24 for clinopyroxene (Bédard, 2010). Fe-Mg diffusion in olivine is fast and its primary composition has been chemically equilibrated, erasing any zoning that may have existed. The overgrowth of anhedral clinopyroxene core1 primocryst by the core2 oikocryst is best explained by gabbro partial dissolution, preferentially along grain boundaries, and subsequent crystallization from a distinct magma (i.e., different composition and temperature). We discuss the origin and composition of this magma in the next section.

Melt Hybridization at the Origin of the Clinopyroxene Core2
Lower oceanic crust gabbroic cumulates typically have higher Mg# (average of 82.6 at Hess Deep, Gillis et al., 2014a) and lower TiO 2 (0.04-0.59 wt%) (Coogan, 2014;Gillis et al., 2014a) and REE (especially LREE) (Figure 6C; see also Kelemen et al., 1997b) than the most primitive MORB glass [Mg# up to 0.68, TiO 2 : 0.9-2.9 wt% (Gale et al., 2013); PetDB, (Lehnert et al., 2000)]. Thus, upon gabbro fractional crystallization, the melt Mg# would decrease and the incompatible TiO 2 and REE content would increase, at a slightly increasing Ce/Sm ratio (Figure 9). On the other hand, partial melting of gabbro cumulate produces a Ti-, Cr-and REE-poor melt with high Mg#. During increased or subsequent gabbro melting, the liquid progressively becomes more depleted in strongly incompatible elements (e.g., La, Ce) over moderately incompatible elements (e.g., Sm), thus resulting in a lower Ce/Sm ratio (Figures 6C, 9). In comparison to the core1 parental melt, the clinopyroxene core2 parental melt is Ti-, Cr-, REE-poor, with a low Ce/Sm ratio, and has a high Mg#, as predicted by gabbro partial melting.
We calculated the gabbro protolith Sr concentration to be 100 µg/g and Sr/Sr * to be 7, using MELTS (Ghiorso and Sack, 1995) gabbro modal abundances and measured trace element compositions in plagioclase and clinopyroxene cores. MOR primitive glasses have Sr/Sr * values close to 0.8-1.5 (Gale et al., 2013 andPetDB, Lehnert et al., 2000 databases). Upon hybridization of anatectic gabbro melt with MOR primitive magma, the hybrid melt Sr/Sr * increases. The plagioclase stability field also expands (Leuthold et al., 2015) and results in an initial rapid Sr depletion along the hybrid melt liquid line of descent. We can speculate on the Sr/Sr * peak observed at the Hess Deep core1 HD -core2 HD transition. The three analyses with Sr/Sr * up to 0.68 (see Figure 4D) are not related to a crack and far from any plagioclase inclusion. These high Sr/Sr * values might record plagioclase assimilation. The Kane Megamullion clinopyroxene experienced Sr diffusion over a few tenths to hundreds of micrometers upon cooling (Figure 3D), possibly erasing or attenuating sharp high Sr * /Sr peaks.
Geochemical modeling and basalt-gabbro isothermal reaction experiments show that modal partial melting of gabbroic cumulate produces a Cr-, Ti-, Zr-, REE-poor anatectic melt with high Mg# and Sr/Sr * (Leuthold et al., 2015;Leuthold and Ulmer, 2016). With a lower Ce/Sm ratio, the Kane Megamullion clinopyroxene core2 KM most likely crystallized from a hybrid melt of such a gabbro partial melt with low Cr-, Zr-and REEand high Mg# and newly injected mantle-derived primitive melt. We use Ce/Sm and Sm trace elements on Figure 9, to model the clinopyroxene core2 parental magma for both the Mid Atlantic Ridge and the East Pacific Rise samples (see Table A.4 for calculations). The REE models show that the parental melts to clinopyroxene core2 are best explained by 10% modal equilibrium melting of gabbro (minerals mode estimated from MELTS calculations). However, we prefer a model where the core2 crystallized from a hybrid melt generated by mixing of gabbro anatectic melt (10-30%) with variable proportions of (depleted) lherzolite partial melt (10-20%). Likewise, spinel inclusions occurring along plagioclase chadacrysts within core2 may saturate following mixing and hybridization of a newly injected olivine-saturated magma with a cogenetic, differentiated, more siliceous melt (Irvine, 1977), such as anatectic gabbro or troctolite (Bédard et al., 2000;O'Driscoll et al., 2009;Leuthold et al., 2015). The sharp REE decrease from core1 to core2, if the latter crystallized from a hybrid melt, indicates that most if not all of the REE-rich trapped interstitial melt was lost prior to recrystallization. This would be the case if the gabbro protolith was either an adcumulate [e.g., ultra-slow-spreading Southwest Indian Ocean ridge (Gao et al., 2007 and references therein), fastspreading ridge at Hess Deep (Natland and Dick, 1996)] or if it was a crystal mush and the interstitial melt was expelled prior to partial melting (e.g., Allibon et al., 2011). We propose that the clinopyroxene and plagioclase compositional data are best explained by partial melting of a gabbroic protolith (clinopyroxene core1 + plagioclase core and chadacrysts + olivine restite) that underwent intrusion/percolation by hot primitive mantle-derived melt. The products are a clinopyroxene-poor gabbro/troctolite residue and a hybrid melt saturated in clinopyroxene and plagioclase that subsequently crystallizes secondary clinopyroxene (core2 and rim) over rare relics and poikilitic gabbro calcic plagioclase (reversely zoned) rims (Figure 10). In addition, spinel may saturate in such hybrid basalt-silicic melts (Irvine, 1977;Bédard et al., 2000;O'Driscoll et al., 2009;Leuthold et al., 2015). This represents an alternative model for the petrogenesis of troctolite cumulate, classically interpreted as the product of low-pressure basalt fractional crystallization (see also Leuthold et al., 2014a;Coumans et al., 2016).
The partial melting-hybridization scenario for the origin of poikilitic gabbros is also consistent with the textural observations of clinopyroxene and plagioclase dissolution. During gabbro protolith partial melting, their stability increases in the basalt-gabbro hybrid melt (Leuthold et al., 2014a(Leuthold et al., , 2015. Large oikocrysts may form in a clinopyroxene-poor groundmass by a crystal coarsening process (Mills et al., 2011), where abundant small nuclei are dissolved and recrystallized as fewer large oikocrysts over relic crystals. In the Hess Deep sample, the random orientation of plagioclase chadacrysts in the clinopyroxene core1 HD and the preferred orientation of clinopyroxene rim HD plagioclase chadacrysts parallel to the main foliation (Figures 1B, 2D; see also Gillis et al., 2014b) suggests crystallization of clinopyroxene core1 HD predates cumulate compaction, while clinopyroxene rim HD postdates compaction. Leuthold et al. (2014a) interpreted similar textures in the Rum layered intrusion as the consequence of effective compactionrelated grain alignment because of the porosity increase following melt percolation and partial melting.
The striking similarities of the crystal textures and mineral chemical traverses in the Kane Megamullion and Hess Deep poikilitic gabbros points convincingly to similar processes acting at slow-and fast-spreading ridges. Records of gabbro hybridization with mantle-derived melt have also been documented in ophiolites (Bédard et al., 2000). Furthermore, similar crystal textures and geochemical trends were documented in the Rum Layered Suite, where olivinephyric and aphyric picritic sills (>1,200 • C) intruded gabbroic cumulate (1,180-1,160 • C), forming troctolitic restite and a clinopyroxene-saturated hybrid melt that percolated and refertilized the overlying gabbro, to produce poikilitic gabbro (Leuthold et al., 2014a(Leuthold et al., , 2015. In addition, Coumans et al. (2016) demonstrated that melt inclusions in reversely-zoned plagioclase were trapped following melting and assimilation of plagioclase-rich cumulate by hot primitive mantle-derived melt (Juan de Fuca near-ridge seamount). Hence, partial melting of cumulates by invading melts may be a widespread (global) process. The implications for cumulates and melts in the lower oceanic crust are discussed in further detail below.

The Clinopyroxene Oikocryst Rims
The geochemical evolution of poikilitic gabbro clinopyroxene rims has been discussed in several studies and will only be briefly discussed here. The clinopyroxene oikocrysts show strong enrichments in incompatible elements (TiO 2 , Zr and REE) and depletions in Cr 2 O 3 , Mg#, Eu/Eu * and Sr/Sr * toward their outer rims (Tables A.1, A.2). Such geochemical evolution is consistent with a fractional crystallization process. However, Lissenberg and MacLeod (2016) showed that fractional crystallization models fail to explain the observed outer rim enrichment, as well as the fractionation between incompatible elements. Previously published studies [e.g., Kane Fracture Zone in the Mid-Atlantic Ridge (Coogan et al., 2000;Lissenberg and MacLeod, 2016), Atlantis Bank in Southwest Indian Ridge (Gao et al., 2007;Lissenberg and MacLeod, 2016), Hess Deep in the Pacific Ocean (Lissenberg et al., 2013;Lissenberg and MacLeod, 2016)] explained the composition of the clinopyroxene outer rims by crystallization from pervasively infiltrated reactive differentiated melt. In that scenario, incompatible elements had been fractionated from one another by zone refining (e.g., McBirney, 1987) (chromatographic separation; a process by which a passing melt front depletes a solid framework in its incompatible elements through partial melting). Alternatively, Coogan and O'Hara (2015) modelled similar geochemical trends by in situ post-crystallization diffusive fractionation. Such latestage processes are likely superimposed on the earlier and highertemperature partial melting reactions described in this study.

The Troctolite-Medium-Grained Poikilitic Gabbro-Gabbro Suite
In the Kane Megamullion sample, the troctolite progressively grades to a medium-grained poikilitic gabbro and then to a gabbro ( Figure 1A). In all three layers, the presence of similar normally zoned An-rich plagioclase (Figure 7) points to crystallization from a high temperature melt. Also interstitial clinopyroxene and clinopyroxene rims within the three lithologies have a similar chemistry (Figure 5). Finer-grained olivine and plagioclase in the troctolite at the contact with the coarse-grained poikilitic gabbro might suggest higher nucleation rates next to a colder rock. The troctolite appears to have differentiated in situ to a troctolite, a gabbro with interstitial clinopyroxene, and then to a gabbro. This suite may Figure 11 | Thermo-mechanical modeling of a slow-spreading ridge (spreading of 2 cm/year) (results for a similar model applied to fast-spreading ridges are shown in Figure A.4). (A) Dykes and sills intruded within the last 100 ′ 000 years (100 ka). (B) Thermal profile after 100 ka. It is very similar to the initial gradient, based on the Sinton and Detrick (1992) model. (C) Residual melt after cooling and crystallization of mantle-derived melt injected throughout the crust. (D) Crustal melt at 100 ka, localized next to dykes and sills. The occurrence of crustal melt is discontinuous in time. (E) Considering extraction of interstitial melt from the lower to the upper crust (estimated here to be ca. 20%), the total melt fraction is generally less than ca. 20%.
have crystallized from hot mantle-derived melt, percolating and melting the coarse-grained poikilitic gabbro. The gabbro clinopyroxene core composition is identical to the poikilitic gabbro clinopyroxene core2 KM , and distinct from the troctolite interstitial clinopyroxene (Figure 5). We suggest these grains were transferred from the poikilitic gabbro into the percolating magma. Plagioclase rims in the coarse-grained poikilitic gabbro and troctolite have a similar chemical range (An 81 -An 68 ) (Figure 7, Table A.1). They crystallized either from a mixture of anatectic gabbro and mantle-derived melts in variable proportions, or from a cooling and fractionating magma, similar to the troctolite.

MELTS Models
Sufficient heat is needed for intrusive magma to partially melt a gabbroic cumulate, which requires specific conditions: (1) the gabbro cumulate must still be hot, (2) the temperature of the intruding melt must be higher than the solidus of the host gabbro cumulate, (3) the latent heat of fusion must not be much smaller than the latent heat of crystallization and (4) the mass ratio of assimilant to the invading hot melt must be low. We argue that each of these four conditions can be met: (1) Our MELTS (Ghiorso and Sack, 1995) calculations show gabbro (sensus stricto) fractionates from ca. 1,205 to ca. 1,155 • C and we assume that the cumulate did not have sufficient time to cool to distinctly lower temperature (see Figure 11, Figure A.4; Sinton and Detrick, 1992); (2) Based on our textural observations and geochemical model above, we argue that the reactive hot basalt was not saturated in clinopyroxene or plagioclase. The solidus temperature of the mantle below mid-ocean ridges is estimated to ca. 1,250-1,280 • C (e.g., Hirose and Kushiro, 1993;Hirschmann et al., 1998), which is higher than plagioclase and clinopyroxene saturation temperatures at crustal pressures (i.e., ca. 1,215 • C); (3) As gabbro cumulate crystallized along the same liquid line of descent as the intruding hot primitive basalt, the latent heat of fusion and crystallization are likely to be similar (e.g., Sleep and Warren, 2014). The basalt latent heat of crystallization (ca. 12 kJ/g) released during dunite and troctolite crystallization is relatively low, as those cumulates represent only ca. 15 vol.% of total cumulate. (4) The exact volume of intruding melt is difficult to estimate, as magma injection is irregular and may occur over a protracted period of time. Indeed, White et al. (2012) showed magma injection into dykes occurs in pulses over weeks. Also, in their study of a dolerite sill, Holness and Humphreys (2003) pointed out conduits where magma may flux over months. As a result, wide contact aureoles and partial melting of the conduit wall can occur (e.g., Huppert and Sparks, 1989;Bruce and Huppert, 1990). Thirdly, magma may be repeatedly injected and drained back into the source region [e.g., Rum layered intrusion (O'Driscoll et al., 2007); Kilauea Iki lava lake (USGS)]. The potential for the host gabbro to melt thus mostly depends on its initial temperature and the temperature and volume of the intruding melt.
We have run MELTS (Ghiorso and Sack, 1995) isenthalpic assimilation models (energy constrained) to test the extent of gabbro cumulate assimilation. We present one realistic case here: 25 g of a gabbro assemblage fractionated at 1,175 • C from mantle-derived melt at 150 MPa (see composition in Table A.3) and subsequently cooled down to 1,000 • C, would be entirely assimilated by 100 g of mantle-derived melt intruded at 1,220 • C. Upon complete assimilation, the liquid temperature would drop to 1,205 • C (85 g left) and 40 g of solid phase would have fractionated. The saturation temperature of clinopyroxene is increased to 1,210 • C, remains unchanged for plagioclase at 1,220 • C and is decreased to 1,220 • C for olivine. Isenthalpic models best represent continuous melting, crystallization and mixing processes, while isothermal models best represent natural situations where the assimilant melting and the invading melt crystallization precede mixing.
Thermal Modeling of the Crust MELTS (Ghiorso and Sack, 1995) isenthalpic assimilation models do not consider intrusion geometry (e.g., distance to the intrusion) or the time evolution of heat diffusion process. For this reason, we ran simulations at the crustal scale using the Karakas and Dufek (2015) thermo-mechanical model, where the evolution of crustal and mantle-derived melts is investigated during dyke and sill intrusion in the crust. We modified the two-dimensional thermal model for the conditions of slowand fast-spreading ridges and assumed an initial geometry constrained by Sinton and Detrick (1992). We simulated a computational domain of 10 km deep and 20 km wide (Figure 11, Figure A.4A) and assumed a crustal thickness of 6 km, with a 2 km upper crust and 4 km lower crust. We used two different initial temperature profiles (geotherms) that were based on the Sinton and Detrick models and the predicted isotherms of Dunn et al. (2000), for slow-and fast-spreading ridges respectively. We then simulated injection of 40 m thick dykes at different crustal levels, over a tectonic width of 4 km (Figure 11A, Figure A.4A), spreading the crust laterally. Dykes were intruded in a stochastic manner, corresponding to an average extension rate of 2 cm/year and 6 cm/year for slow-and fast-spreading ridges respectively. 80 m thick sills intruded in the lower crust, at dyke tips, thickening the crust. This effect is partly balanced by mantle upwelling at the base of the crust. The details of the thermal calculations and the methodology are explained in Karakas and Dufek (2015) and Karakas et al. (2017). To investigate the evolution of mantle-derived and crustal magmas during crystallization and melting, we used mantle-derived melt and crustal cumulate parameterized MELTS thermodynamic conditions. The mantle-derived melt liquid line of descent is identical to the one discussed above (see starting composition in Table A.3). The solidus temperature was set at 1,000 • C, with a fixed crystallization rate of 0.3%/ • C from 1,050 • C. As a cumulate, the gabbroic crust solidus temperature was set at 1,150 • C (i.e., above the pigeonite/orthopyroxene saturation temperature), assuming that interstitial melt was extracted at that temperature. Heat conduction drives crystallization of the dykes and sills and may generate partial melting of the host crust. The "residual melt" refers to the melt left behind after partial crystallization of mantle-derived melt (Figure 11C, Figure A.4C), while the "crustal melt" refers to the melt produced by partial melting of the crust (Figure 11D, Figure A.4D). On Figure 11E (Figure A.4E), the total melt (i.e., sum of residual melt and crustal melt) is reduced by 20% to account for interstitial melt extraction from the lower to the upper crust.
As spreading ridges are long-lived systems, the conductive geothermal gradient did not change significantly at the end of the simulations (i.e., after 100 ka) (Figure 11B, Figure A.4B).
The results of the thermal model show that partial melting of the crust occurs in the lower levels of the lower oceanic crust, adjacent to recently intruded dykes and sills. At slow-spreading ridges, crustal melting is low (< ca. 5% of the total melt volume) ( Figure 11D) and not continuous over time. At fast-spreading ridges, residual melt is abundant, forming a vertically extensive zone of crystal mush. Crustal melt represents ca. 20 vol.% of the total melt volume (Figure A.4D). According to our calculations, the degree of partial melting in contact with dykes and sills can reach ca. 50% locally (i.e., corresponding to a temperature increase to ca. 1,170 • C). Our thermal models confirm the above microtextural and geochemical observations of gabbro partial melting. In Figure 11E, Figure A.4E, the total melt is estimated to be generally less than 20% in slow-spreading ridges and less than 40% in fast-spreading ridges, after partial extraction of melt from the lower crust to the upper crustal level and seafloor.

Effect of Gabbro Partial Melting on Mantle-Derived Melt
The mineralogy and chemical composition of the cogenetic MOR melt and assimilated young oceanic gabbro are not significantly different. It is thus difficult to detect the assimilation of mafic cumulates in MORB. We discuss here the effects of gabbro assimilation on the mineral assemblage, the major element chemistry and the REE chemistry of the hybrid melt. Koepke et al. (2004) showed that partial melting of hydrated gabbro at 900-950 • C, with clinopyroxene reaction to orthopyroxene and pargasitic amphibole, produces felsic melt at the origin of the plagiogranite. Through isothermal assimilation of dry primitive gabbro, the hybrid melt liquid line of descent is shifted toward higher SiO 2 , Mg#, CaO and Al 2 O 3 values (Figure 8; Leuthold et al., 2015). The stability of clinopyroxene and plagioclase is increased. Due to the increase in SiO 2 , the olivine stability is decreased, replaced by orthopyroxene, which might occur as a cumulus phase even in primitive gabbro, as observed in the Hess Deep drilled cores (Gillis et al., 2014a,b). It also results in a Mg# increase in clinopyroxene (Bédard, 2010;Leuthold et al., 2015). High temperature cumulate also has low TiO 2 , FeO T , Na 2 O, K 2 O, which are consequently low in the hybrid melt.

Mineral Stability and Major Element Chemistry
We also used MELTS (Ghiorso and Sack, 1995) to quantify the effect of gabbro assimilation on the major element chemistry of mantle-derived melt (e.g., Leuthold et al., 2014a;Coumans et al., 2016). Models are shown on the Figure 8. The effect of 50% gabbro cumulate partial melting and subsequent mixing with mantle-derived melt differentiated to 1,200 • C (see compositions of anatectic melt and differentiated mantle-derived melt in Table A.3) is similar to a pressure decrease of ca. 100 MPa, a water increase of ca. 0.3 wt.%, a redox state change from NNO-2 to NNO-0.5 (i.e., more oxidized) or non-ideal fractional crystallization (i.e., with limited equilibrium crystallization).

REE Chemistry
MORB and lower crust clinopyroxene may show a wide range of REE, from REE-poor N-MORB (i.e., Ce/Sm) N <1.00) to REErich E-MORB (i.e., Ce/Sm) N >1.00) as a result of compositional heterogeneity in the mantle source and variation in partial melting processes (e.g., Kelemen et al., 1997a). Indeed, low-grade partial melting of a fertile lherzolite (not previously molten) or a refertilized peridotite would produce a REE-rich basalt with Ce/Sm N >1 (Figure 9). Upon further melting, the residue and melt REE concentration and Ce/Sm ratio progressively decrease. Partial melting of such a residual peridotite (i.e., depleted lherzolite, harzburgite) would produce melt with a low Ce/Sm N ratio (typically <1) (Figure 9). In the Hirose and Kushiro (1993) partial melting experiments of the relatively fertile KLB-1 spinellherzolite at 1.5 GPa, the solidus temperature is ca. 1,280 • C and the clinopyroxene is totally molten between 1,350 and 1,400 • C (i.e., leaving a harzburgitic residue), corresponding to a liquid fraction of 19-29%. pMELTS (Ghiorso et al., 2002) calculations using KLB-1 lherzolite composition at 1.5 GPa show that clinopyroxene is totally molten after 23% partial melting, at 1,455 • C.
The gabbro cumulates are REE-poor, with low Ce/Sm ratios (Kelemen et al., 1997b), in comparison to mantle-derived melt compositions ( Figure 6C). Upon crystal fractionation, differentiated MOR melts become REE-enriched. Conversely, gabbro cumulate partial melting produces a REE-poor melt with low Ce/Sm (Figure 9). In addition, upon increased gabbro melting, the clinopyroxene-hybrid melt REE partition coefficient will decrease (Leuthold and Ulmer, 2016), increasing the effect of REE depletion. When considering the REE concentrations of MOR melts, it is difficult to distinguish between assimilation of low REE gabbro and higher degrees of partial melting of lherzolite or depleted-lherzolite, particularly since primary melts are unlikely to exist.

CONCLUSIONS
We conclude that the oceanic lower crust poikilitic gabbros studied here record: (1) intrusion of hot mantle-derived primitive melt into the young, still-hot gabbroic lower oceanic crust, (2) cooling and crystallization of the intrusive magma and partial melting of the host gabbro cumulate to an equilibrium temperature, (3) mixing of the two melts, (4) possible extraction and percolation of the hybrid melt, (5) crystallization of secondary phases (e.g., clinopyroxene core2 and rim and plagioclase rim) and saturation of new phases (e.g., spinel) (i.e., refertilization) (see Graphical Abstract, Figure 10). Partial melting has affected primocrysts and modified the composition and textures (i.e., mineral modes, intra-crystal textures) of the lower oceanic gabbro cumulate, producing pyroxene-poor gabbro/troctolite residue and a hybrid melt [Cr-, Al-, REE-(especially LREE-) poor and with high Mg#] saturated in clinopyroxene and plagioclase. Importantly, the observations made here suggest that a similar MASH process operates in slowand fast-spreading ridges, as well as in layered intrusions (see Leuthold et al., 2014a) and arcs (e.g., Hildreth and Moorbath, 1988;Cooper et al., 2016). Thermal models confirm (locally high degree) partial melting of hot gabbro cumulate by recently intruded dykes and sills. However, the overall volume of crustal melt is low.
MORB magmas commonly have high Mg#, low REE concentrations and low Ce/Sm ratios. The Atlantic, East Pacific Rise and Galapagos major elements trends of natural glass follow liquid lines of descent from unmodified mantle-derived melts, extracted from variable depths. However, lower oceanic gabbro clinopyroxene oikocrysts and plagioclase microtextures and zoning point to hybridization of primitive magmas with oceanic crustal gabbro. The process of gabbro assimilation described here contributes to the large chemical variability of MORB at conditions close to the liquidus, in addition to mantle partial melting processes and crystallization at mantle and crustal depths. However, the lack of strong mineralogical and/or geochemical contrasts between co-genetic invading basaltic melt and gabbro cumulate makes it difficult to distinguish in erupted MOR lavas alone. Great care is therefore required to show modification during percolation through the lower oceanic crust filter, when only considering MORB major and trace elements. Discerning and quantifying possible reactions of primitive mantle-derived melt with lower oceanic crust is critical in the interpretation of oceanic basalt compositions, in order to avoid inadequate petrogenetic models derived from the incorrect assumption of equilibrium premises (e.g., degree of mantle partial melting deduced from the MORB geochemistry proxy). Further studies are necessary to evaluate how pervasive the MASH process is during the generation and differentiation of the oceanic crust.

AUTHOR CONTRIBUTIONS
JL has initiated the research project, analyzed samples and supervised M.Sc. student DK at every step. He has run the MELTS calculations, done the geochemical modeling and helped with the thermal modeling. He has written the manuscript and prepared the figures. CL has conducted detailed microscopic and geochemical study on the studied Kane Megamullion sample. He has helped writing the manuscript. BO has helped writing the manuscript. OK has written and run the thermal models. TF has provided the Hess Deep sample. He had acquired preliminary SEM images. DK has prepared the samples. She has helped acquiring preliminary SEM images and EMPA analyses and reducing LA-ICP-MS data. PU helped to initiate the project. We had fruitful discussions and he helped writing the manuscript. All authors have approved this submitted version. Figure A.1 | JAS117-63 sample. From left to right, layers of gabbro grading to a medium-grained poikilitic gabbro with interstitial clinopyroxene, to a fine-grained troctolite, in contact with a coarse-grained poikilitic gabbro.  Gillis et al., 2014b). Clinopyroxene oikocrysts in a foliated troctolitic groundmass.   Figure 11 for thermal modeling of a slow-spreading ridge). (A) Dykes and sills intruded within the last 100 ka. (B) Thermal profile after 100 ka. It is very similar to the initial gradient, based on Sinton and Detrick (1992) model. (C) Residual melt after cooling and crystallization of mantle-derived melt injected throughout the crust. (D) Crustal melt at 100 ka, localized next to dykes and sills. The occurrence of crustal melt is discontinuous in time. (E) Considering extraction of interstitial melt from the lower to the upper crust (estimated here to be ca. 20%), the total melt fraction is generally less than ca. 40%.   Table A.3 | Melt, gabbro and fertile spinel lherzolite compositions used in the geochemical models of Figure 9. Presentation 1 | Interactive microscopic view of the Mid-Atlantic ridge Kane Megamullion JAS117-63 gabbroic sample. In the slide show mode, use red buttons to change from plan polarized light (PPL) to cross polarized light (XPL), zoom in and out and rotate the sample.