The Neodymium Stable Isotope Composition of the Oceanic Crust: Reconciling the Mismatch Between Erupted Mid-Ocean Ridge Basalts and Lower Crustal Gabbros

The trace element and isotopic compositions of mid-ocean ridge basalts (MORB) provide an important cornerstone for all studies seeking to understand mantle evolution. Globally there is a significant over-enrichment in the incompatible trace element concentrations of MORB relative to levels which should be generated by fractional crystallization. Thermal and geochemical constraints suggest that MORB require generation in open system magma chambers. However, the petrology of lower oceanic crustal rocks suggests instead that these enrichments maybe formed through reactive porous flow (RPF). Stable isotope compositions are process dependent and therefore provide an excellent mechanism to compare these contrasting models. This study presents the first neodymium (Nd) stable isotope compositions of Indian MORB and well characterized gabbroic rocks from the lower oceanic crust sampled at the Southwest Indian Ridge (SWIR) (Hole 735B). Indian MORB is extremely homogenous with a mean δ146Nd of −0.025 ± 0.005‰ which is identical to the composition of Pacific MORB. Despite significant variability in the source composition of MORB globally (i.e., 143Nd/144Nd) their indistinguishable δ146Nd compositions suggests δ146Nd was homogenized through a consistent process (i.e., repeated melt addition in the open-system magma chambers across the global ridge network). In stark contrast, oceanic gabbros have δ146Nd ranging from −0.026 to −0.127‰, doubling the natural variability in Nd stable isotopes observed in terrestrial rocks. Clinopyroxene separates possess variable δ146Nd but are isotopically heavier than the gabbroic whole rocks at the same major element compositions. These large variations in δ146Nd cannot be generated solely by the fractionation or accumulation of magmatic minerals. Hole 735B preserves widespread evidence of RPF which could induce kinetic isotope fractionation during crystal growth. However, the maximum kinetic isotope fractionations that can be generated in clinopyroxene are only ca. 0.02‰, therefore several cycles of dissolution and reprecipitation of isotopic signatures at grain boundaries are required to explain the range of δ146Nd observed in the gabbros. The large disconnect between the average composition of the oceanic crust (δ146Nd = −0.067‰) and MORB, combined with limited evidence of melt extraction to the upper crust at Hole 735B, led to the conclusion that melts involved in RPF have not contributed in a substantial way to the Nd isotope composition of erupted MORB.

The trace element and isotopic compositions of mid-ocean ridge basalts (MORB) provide an important cornerstone for all studies seeking to understand mantle evolution. Globally there is a significant over-enrichment in the incompatible trace element concentrations of MORB relative to levels which should be generated by fractional crystallization. Thermal and geochemical constraints suggest that MORB require generation in open system magma chambers. However, the petrology of lower oceanic crustal rocks suggests instead that these enrichments maybe formed through reactive porous flow (RPF). Stable isotope compositions are process dependent and therefore provide an excellent mechanism to compare these contrasting models. This study presents the first neodymium (Nd) stable isotope compositions of Indian MORB and well characterized gabbroic rocks from the lower oceanic crust sampled at the Southwest Indian Ridge (SWIR) (Hole 735B). Indian MORB is extremely homogenous with a mean δ 146 Nd of −0.025 ± 0.005 which is identical to the composition of Pacific MORB. Despite significant variability in the source composition of MORB globally (i.e., 143 Nd/ 144 Nd) their indistinguishable δ 146 Nd compositions suggests δ 146 Nd was homogenized through a consistent process (i.e., repeated melt addition in the open-system magma chambers across the global ridge network). In stark contrast, oceanic gabbros have δ 146 Nd ranging from −0.026 to −0.127 , doubling the natural variability in Nd stable isotopes observed in terrestrial rocks. Clinopyroxene separates possess variable δ 146 Nd but are isotopically heavier than the gabbroic whole rocks at the same major element compositions. These large variations in δ 146 Nd cannot be generated solely by the fractionation or accumulation of magmatic minerals. Hole 735B preserves widespread evidence of RPF which could induce kinetic isotope fractionation during crystal growth. However, the maximum kinetic isotope fractionations that can be generated in clinopyroxene are only ca. 0.02 , therefore several cycles of dissolution and reprecipitation of isotopic signatures at grain boundaries are required to explain the range of δ 146 Nd observed in the gabbros. The large disconnect between the average INTRODUCTION The oceanic crust covers the majority of Earth's surface (∼60%) and therefore accurate knowledge of its composition is imperative for constraining the composition of the silicate Earth and the long-term geochemical fluxes in recycling systems. Oceanic crust is produced along the global ocean ridge system which stretches over 60,000 km and produces 75% of Earth's volcanism (Crisp, 1984). The composition of mid-ocean ridge basalts (MORB), therefore can be used to provide important insights into the composition of Earth's most extensive crustal reservoir while also placing constraints on the composition of the upper mantle (e.g., Langmuir and Hanson, 1980;Hofmann, 1997;Workman and Hart, 2005;Gale et al., 2013). The magmas parental to these basalts were traditionally considered to have evolved by simple fractional crystallization, producing an oceanic crust consisting of 1-1.5 km of basalt and dolerite underlain by 4-5 km of gabbro (e.g., Dick et al., 1984;Klein, 2003). Although the chemical and isotope compositions of MORB are relatively homogenous compared to other magma types (e.g., ocean island basalts), MORB glasses can display significant elemental variations (White, 1985;Warren et al., 2009;Jenner and O'Neill, 2012). This compositional scatter is the result of the interplay between the contrasting effects of: (i) source composition (Rehkämper and Hofmann, 1997;Nielsen et al., 2018); (ii) differences in the extent of melting reflecting variations in mantle potential temperature (Klein and Langmuir, 1987;McKenzie and Bickle, 1988) or spreading rate (Regelous et al., 2016); and (iii) fractional crystallization processes which can be influenced by both the pressure of crystallization and magma composition (Michael and Cornell, 1998;Herzberg, 2004). However, a significant over-enrichment in the incompatible trace elements concentrations (e.g., Th, La) of magmas is observed compared to that which is predicted by simple fractional crystallization alone (White and Bryan, 1977;Hekinian and Walker, 1987;O'Neill and Jenner, 2012). Two distinct petrogenetic models have been advocated as the cause of these enrichments: (1) differentiation in an open system replenished-tapped-crystallizing (RTX) magma chambers (O'Hara, 1968;O'Neill and Jenner, 2012;Coogan and O'Hara, 2015) and (2) melt rock interaction resulting in the chromatographic separation of elements in the lower crust (Kamenetsky et al., 1998;Lissenberg and Dick, 2008;Lissenberg et al., 2013;Lissenberg and MacLeod, 2016). There is compelling geochemical and geophysical evidence that magma chambers at mid-ocean ridges operate as steady state open systems (Detrick et al., 1987;Johnson and Dick, 1992;Sinton and Detrick, 1992;White, 1993), supportive of an RTX model. Plutonic rocks from the oceanic crust are considered the crystallization products of MORB differentiation (Coogan, 2014), however, whether MORB do in fact represent the erupted complement of the preserved oceanic gabbros remains debatable. At high temperature stable isotope fractionation can provide important insights in the nature and conditions of the processes involved in magma genesis (e.g., Dauphas et al., 2014;Millet et al., 2016;McCoy-West et al., 2018). This study presents high-precision double spike Nd stable isotope data for oceanic gabbros from Hole 735B and MORB from the Indian Ocean to test the various models of MORB genesis.

MATERIALS AND METHODS
Analyses were performed in the Arthur Holmes Geochemistry Labs at the Durham University. Basaltic glass, clinopyroxene or whole rock powders were weighed out to obtain 200 ng of natural Nd and then spiked with a 145 Nd− 150 Nd double spike (McCoy-West et al., 2017, 2020, and then dissolved using a conventional HF-HNO 3 hotplate digestion. Neodymium was separated using well-established chromatographic techniques. Briefly, the rare earth elements (REE) were separated from the sample matrix using BioRad AG50W-x8 cation exchange resin. A second elongated column filled with Eichrom Ln-Spec resin was then used to separate Nd from direct Sm isobaric interferences using dilute HCl. For a more detailed description of the methodology refer to McCoy-West et al. (2017) and McCoy-West et al. (2020).
Neodymium isotope measurements were performed using a Thermo-Fisher Triton Plus thermal ionisation mass spectrometer (TIMS). Neodymium was measured as a metal ion in static collection mode using a double Re filament assembly. Each analysis usually comprised 400 cycles of data acquisition (8 s per integration). Stable Nd isotope ratios are expressed using conventional delta notation, where δ 146 Nd which is the per mil deviation in the measured 146 Nd/ 144 Nd relative to the widely measured reference standard JNdi-1: Double spike deconvolution was undertaken using the Wolfram Mathematica program and is based on the algebraic resolution method (Millet and Dauphas, 2014 values agree within analytical uncertainty of previously published values for conventionally processed samples (see Figure 1 in McCoy-West et al., 2017) providing confidence of the accuracy of the δ 146 Nd obtained from the same measurement. As a secondary check of data quality, it is also now possible to independently deconvolve the δ 148/144 Nd of the samples to confirm the mass dependence of analyses (Figure 1). The data presented here have been normalized to JNdi-1 = 0, to correct for small offsets in the measured composition of JNdi-1 during different analytical periods due to differences in faraday collector efficiency (McCoy-West et al., 2020). Propagated uncertainties were calculated using the 95% standard error on the average of JNdi-1 for the correction period. Applying a secondary normalization such as this is common practice for radiogenic TIMS data (e.g., Bennett et al., 2007;McCoy-West et al., 2013) and stable isotope data generated by MC-ICP-MS (e.g., Goldberg et al., 2013;Millet and Dauphas, 2014;Neely et al., 2018;McCoy-West et al., 2019) in order to deal with small amounts of non-exponential mass bias that is not accounted for by internal corrections methods. The long-term reproducibility of the δ 146 Nd measurements is considered to better than ± 0.015 (McCoy-West et al., 2017, 2020.

Samples
This study focuses on samples from Ocean Drilling Program Hole 735B in the Atlantis II Transform Fault area located at 32 • S 57 • E on the Southwest Indian Ridge (SWIR) to the southeast of Madagascar (Figure 2). This region contains 12 Ma gabbroic ocean crust (John et al., 2004;Baines et al., 2009) that is exposed FIGURE 1 | Plot showing the mass dependent covariation between δ 146 Nd and δ 148 Nd for samples measured in this study. Shaded field represents an uncertainty in δ 148 Nd of ± 0.038 , which is based on the long-term uncertainty in δ 146 Nd (±0.015 ; McCoy-West et al., 2020) but has been increased proportionally based on the relative mass difference between the isotopes and the less precise counting statistics due to the smaller ion beam on 148 Nd. on a 5-km transverse ridge (Atlantis Bank) which formed as the result of detachment faulting from 13 to 10 Ma . The spreading rate in this region is ∼1.6 cm/year which is extremely slow (Baines et al., 2007). Hole 735B is located ∼95 km south of the present-day ridge axis and drilled 1508 m into the gabbroic rocks unroofed at the Atlantis Bank core complex, recovering a plutonic section dominated by series of generally coarse-grained olivine gabbros cross-cut by more evolved rocks (gabbronorites, oxide gabbros) and subordinate troctolite Coogan et al., 2001;Natland and Dick, 2001). More felsic leucocratic igneous rocks are rare, making up ≤1% of the Hole 735B core and are generally present as highly altered, small crosscutting veins (Dick et al., 2000). Petrologic evidence and fluid inclusion studies suggest the ocean crust at Hole 735B is 4 ± 1 km thick, with 60-70% of the total gabbro column sampled (Vanko and Stakes, 1991;Muller et al., 1997).
To supplement the gabbroic rocks analyses of MORB glass from the Indian Ocean where undertaken. These samples are widely distributed coming from all the major ridge segments in the Indian Ocean (SWIR, Southeast Indian Ridge and Central Indian Ridge; Figure 2) and have been previously been analyzed for Sr-Nd-Hf-Pb and Os radiogenic isotopes (Salters, 1996;Schiano et al., 1997;Escrig et al., 2004).

735B Gabbro Whole Rock
The gabbro's from Hole 735B possess highly variable δ 146 Nd with values ranging from −0.026 to −0.126 (n = 7; δ 146 Nd = 100 ppm; Figure 3 and Table 2), which extend to the lightest Nd stable isotope compositions so far observed. This large variation in δ 146 Nd is not correlated with the Nd concentration (1.4-4.2 ppm) or ε 143 Nd (ca. +9.5; excluding MS20-10 ε 143 Nd = + 6.8) with the samples possessing relatively similar radiogenic isotope compositions (Figure 4). Whole rock geochemistry is correlated with δ 146 Nd ( Figure 5): strong (r 2 > 0.6) positive correlations are observed with SiO 2 , Al 2 O 3 , Na 2 O, and CaO; with negative correlations observed against MgO, FeO T , and MnO. However, occasionally some samples appear as outliers, specifically sample BN-1(F) has a very light δ 146 Nd of −0.126 which is distinct from the strong trend formed by the rest of the gabbros ( Figure 5C). Despite these strong correlations between δ 146 Nd and major element contents, no correlation is observed with indicators of magmatic differentiation (e.g., Mg#; Figure 6A), with the gabbros possessing an extremely narrow range of Mg# from 0.73 to 0.79.

735B Gabbro Clinopyroxene
Clinopyroxene separates from Hole 735B possess between 2.8 and 13.3 ppm Nd (Figure 4A), excluding one separate from MS 41-18 that has a usually high Nd concentration of 80.9 ppm ( Table 2). Comparable to the whole rocks the clinopyroxene separates also possess extremely variable δ 146 Nd with values ranging from −0.016 to −0.090 (n = 7; δ 146 Nd = 84 ppm; Table 2) with a constant radiogenic isotope composition (ε 143 Nd = +9.0 to +9.6; Figure 4B). For the one gabbro MS20-10, where both the whole rock and clinopyroxene composition are available the clinopyroxene is resolvably heavier ( 146 Nd clinopyroxene−wholerock = + 0.047 ). Qualitatively this finding holds when comparing the other samples using their whole rock Al 2 O 3 and Na 2 O contents with the majority of clinopyroxene separates forming a secondary trend at heavier δ 146 Nd than the whole rock samples at similar major element compositions (Figures 5C,D).

Radiogenic Nd Isotope Variability
One of the fundamental strengths of radiogenic isotopes is they remain unfractionated by magmatic process (e.g., fractional crystallization; partial melting), but rather provide a timeintegrated tracer of parent-daughter fractionation (e.g., Sm from Nd). The heterogenous isotopic compositions of MORB globally has been used to map distinct mantle domains and trace long-term recycling of isotopically enriched components in the depleted MORB mantle (Zindler and Hart, 1986;Salters and Stracke, 2004;Hofmann, 2007). Indian MORB has less radiogenic Nd and more radiogenic Sr at the same Pb isotope composition than Atlantic, and Pacific MORB (see Figure 2 in Stracke et al., 2005). Despite this distinction significant variations in ε 143 Nd Frontiers in Earth Science | www.frontiersin.org 4 February 2020 | Volume 8 | Article 25  (Jacobsen and Wasserburg, 1984). δ 146 Nd NORM values have been corrected to JNdi-1 = 0 and errors have been propagated. To represent population uncertainty, both the two-standard deviation (2 SD) and 95% standard errors [95% SE = t × SD/(n) 1/2 ), where t = inverse survival function of the Student's t-test at the 95% significance level and (n−1) degrees of freedom] are presented for averages. #Represents a separately spiked aliquot of the same digestion processed through the entire chemical separation procedure. Global MORB average is based on this data herein and presented in McCoy-West et al. (2017).
are observed across the Indian Ocean between spreading ridges (Hamelin et al., 1986;Robinson et al., 2001;Coogan et al., 2004;Meyzen et al., 2005), and even within individual segments of the SWIR (Figure 2). The previous characterization of Hole 735B (Kempton et al., 1991;Hart et al., 1999;Holm, 2002) means it provides an ideal test case to examine this isotopic heterogeneity with depth (Figure 7). Previous analyses of gabbro from Hole 735B have ε 143 Nd values spanning a significant range from +7.7 to +10.6, which is nearly the entire range observed in Indian MORB. The average composition of Hole 735B drill core is calculated as +9.29 ± 0.14 (95% SE) and is within uncertainty of basalts collected adjacent to the overlying Atlantis II fracture zone (ε 143 Nd = +9.13 ± 0.19), suggesting that the major sources of gabbroic rocks and overlying MORB are broadly similar. The clinopyroxene and whole rock samples analyzed herein are generally indistinguishable from the Hole 735B average, except sample MS20-10 (ε 143 Nd = +6.8) which is the least radiogenic so far measured but remains well within the range of heterogeneity observed along the SWIR (Figure 7). The preservation of extreme isotopic heterogeneity in oceanic mantle between constituent phases and across small length is increasingly being recognized as a common phenomenon (Burton et al., 2012;Lambart et al., 2019). This is not an unexpected results given the depleted mantle represents a mixture of materials at different stages in their evolution, and is therefore unlikely to be a isotopically homogeneous (e.g., Salters and Stracke, 2004).

Stable Nd Variations in Hole 735B
Stable isotope ratios are time-independent and solely fractionated by the processes operating. At equilibrium, the magnitude of isotopic fractionation is proportional to the relative mass difference between the isotopes and inversely proportional to temperature (1/T 2 ) (Bigeleisen and Mayer, 1947;Urey, 1947). Due to the well characterized nature of Hole 735B it provides a unique opportunity to study stable isotope fractionations in the lower oceanic crust. The oceanic gabbros have δ 146 Nd values extending to extremely light compositions (−0.026 to −0.127 ; Figure 3). In Hole 753B at long wavelengths (>200 m) there is generally a good covariation between the chemical compositions of the major silicate phases, consistent with the in situ crystallization of four distinct olivine gabbro bodies (Dick et al., 2002). Three major mineralogical discontinuities have been identified in the lower portion of Hole 735B (picked on the basis of plagioclase anorthite content: see Figure 16 in Dick et al., 2002). These bodies are also clearly delineated by whole rock major element compositions, with the gabbroic rocks in each body generally become slightly more evolved up section (lower Mg#; Figure 8A). Two of the lightest gabbro samples in Hole 735B occur above 650 mbsf ( Figure 8B), however, when also considering the composition of the clinopyroxene separates no clear relationship between δ 146 Nd and depth is observed. Unfortunately, the current dataset remains too limited to constrain the variability of δ 146 Nd with depth, either throughout the entire drill core or relative to the individual position of the samples within the discrete gabbro bodies ( Figure 8B).

Resolving the Causes of Nd Stable Isotope Variation in Hole 735B
The lower oceanic crust is an extremely complex magmatic environment. Indeed, the crustal section at Hole 735B preserves 952 discrete intervals on basis of clear textural and mineralogic differences (Dick et al., 2002). The complexity of slow spreading ridges means that a range different process maybe contributing to the natural variations observed in δ 146 Nd. Ultimately, the whole rock compositions of the preserved gabbros integrate the effects of differences in mineral proportions, mineral compositions and any secondary modification through alteration.
Possible Low Temperature Fractionation?
Significant fractionation of both boron and iron isotopes have been shown to be caused by alteration in the oceanic crust (Spivack and Edmond, 1987;Rouxel et al., 2003). Boron is highly enriched in seawater and fluids and Fe is a major element in most mineral phases which is also sensitive to redox changes. In contrast, Nd a light (L)REE is generally moderately to highly incompatible in most silicate phases, possess a single valence  (Kempton et al., 1991;Hart et al., 1999;Holm, 2002;herein). The light blue represents the average ±2 standard deviations: (ε 143 Nd = 9.29 ± 1.21; 143 Nd/ 144 Nd = 0.513114 ± 62; n = 72) with the dark blue band representing the 95% standard error of the mean (ε 143 Nd = ± 0.14; 143 Nd/ 144 Nd = ± 8).
state (3+), and is extremely fluid immobile (Bau, 1991). These characteristics make Nd generally very robust to secondary modification. Oxygen isotope evidence shows that abundant seawater penetration appears to be limited to the upper part of the crust preserved at Hole 735B (≥500 mbsf; Kempton et al., 1991;Hart et al., 1999;Gao et al., 2006). Given that the samples analyzed here are below this depth and the extremely low concentration of Nd in seawater (ca. 3-4 ppt), it is reasonable to conclude that seawater interaction had no effect on the δ 146 Nd values measured here. Bach et al. (2001) undertook a detailed study of the effects of alteration in Hole 735B with an overall decrease in the intensity of hydrothermal alteration observed with increasing depth. Alteration becomes rarer below 600 mbsf with the lowermost part of the section (500-1500 mbsf) showing the effects of a complex low-temperature (ca. 250 • C) alteration history probably related to the tectonic uplift of the basement.
This low-temperature alteration is localized to fractured regions where intense alteration of the host rocks can be observed adjacent to secondary mineral (smectite, chlorite, calcite, zeolite, sulfide, and Fe-oxyhydroxides) filled veins (Bach et al., 2001). Bienvenu et al. (1990) showed that during intense alteration of basalt the REE can become mobilized during the breakdown of secondary minerals. Here, we have focused on the freshest possible samples to eliminate alteration as a factor in interpreting geochemical signals. The majority of the gabbros possess a restricted range of Nd concentration (1.3-2.0 ppm; Table 2) but span the entire range δ 146 Nd (Figure 4A), with no correlation observed that could be attributed to Nd loss. Neodymium stable isotope compositions have also been compared to other common indicators of alteration intensity such as loss on ignition ( Figure 9A) or enrichment of fluid mobile elements (e.g., Rb; Figure 9B) and remain uncorrelated with these parameters. Further evidence of the inability of alteration to fractionate δ 146 Nd comes from analyses of chondritic meteorites, which contain a range of exotic phases including REE enriched sulfides that are highly susceptible to alteration, that had experienced variable terrestrial alteration but showed no systematic change in δ 146 Nd values with increasing weathering intensity (see Figure 6 in McCoy-West et al., 2017). Therefore, we are confident the variations in δ 146 Nd observed here are generated by magmatic processes during formation of the oceanic crust.

Fractionation or Accumulation of Magmatic Crystals
The gabbros at Hole 735B are dominated by three major phases: (1) plagioclase is present in every rock type examined, is the most abundant phase (50-65% in representative gabbros; Dick et al., 1999), and typically occurs in near cotectic proportions with clinopyroxene and olivine (Bloomer et al., 1991); (2) clinopyroxene is next most prevalent phase, with the majority of gabbros containing from 15 to 45% , its abundance varies inversely with the amount of plagioclase (Dick et al., 2002;Niu et al., 2002); and (3) olivine is a subordinate component and generally only constitutes 5-10% of the olivine gabbros . Clinopyroxene is the major host of Nd in the gabbroic rocks of Hole 735B. Due to distortions of the crystal lattice partition coefficients are dependent on the temperature, pressure and composition of a mineral during crystallization (Wood and Blundy, 1997). Clinopyroxene is characterized by three crystallographically unique polyhedra, with the LREE mainly occupying the eightfold coordinated M2 site (Ross et al., 2013;Bédard, 2014). Higher Al 2 O 3 ( IV Al) contents significantly increase the compatibility of REE in clinopyroxene, due to the high Al 2 O 3 contents ( IV Al up to 0.4) in typical Hole735B gabbros D Nd values of 0.4-0.6 are likely (Hill et al., 2000). Due to the high modal abundance of plagioclase it could also host a significant amount of Nd. Plagioclase is a framework of linked (Si-Al)-O tetrahedra, with large interstices filled with Ca and Na ions. The REE occur in sixfold coordination in plagioclase (Cherniak, 2003), were they can replace Na following a coupled substitution due to their similar atomic radii (Nd = 0.983 Å; Na = 1.02 Å; Shannon, 1976 , 1985) and therefore olivine is not considered a likely host of Nd and will not be discussed further. Geochemical evidence shows that the gabbroic rocks preserved at Hole 735B are not simple quenched melts and instead they represent crystal cumulates. Their whole rock compositions have Mg# that are too high, TiO 2 contents that are too low and too wide a range in CaO/Al 2 O 3 to be in equilibrium with melts . Gabbro whole rock compositions (e.g., CaO, Sc) are instead largely a function of the relative modal proportions of clinopyroxene and plagioclase (see Figure 6 in Niu et al., 2002). Niu et al. (2002) demonstrated that it is possible to calculate the proportion of clinopyroxene using the CaO/Al 2 O 3 (R C/A ; Cpx# = [1.3835 × R C/A -0.5163]/[1.2571 × R C/A + 0.4587]) as it is insensitive to dilution by more mafic components (i.e., olivine). A strong correlation is observed between δ 146 Nd in Hole 735B gabbros and the proportion of clinopyroxene (Figure 6B), with increasing amounts of clinopyroxene correlated with lighter isotopic compositions. Stable isotope compositions are process dependent and therefore the accumulation or fractionation of a significant amount of crystal cargo with a unique isotopic composition can cause resolvable changes in the isotopic composition of a cogenetic rock suite (e.g., Sossi et al., 2012;Millet et al., 2016;McCoy-West et al., 2018). Stable isotope theory suggests that at equilibrium with a fixed valance state heavy isotopes will prefer the site with the stiffest bonds (i.e., lower coordination number; Schauble, 2004). The coordination of the REE in silicate glass is highly composition dependent with sixfold to ninefold coordination observed (e.g., Ponader and Brown, 1989). Existing experimental work suggests that Nd (and the LREE) are predominantly sevenfold coordinated in silicate and other oxide glasses, although sixfold coordination Frontiers in Earth Science | www.frontiersin.org 7 February 2020 | Volume 8 | Article 25 A strong negative correlation is observed between δ 146 Nd and cpx#. Regression line and uncertainty envelope (shaded area) is calculated through the data excluding sample BN-1(F) using Isoplot (Ludwig, 2008) assuming a 2.5% uncertainty on the cpx# and the long-term uncertainty on δ 146 Nd (±0.015). The clinopyroxene separates are variably offset, generally to heavier values, relative to the whole rock relationship consistent with the operation of a disequilibrium process.
can also be present (Mann and DeShazer, 1970;Robinson, 1971;Weber and Brawer, 1982;Rao et al., 1983). Given that Nd sits in the cubical coordinated M2 site in clinopyroxene ( VIII Nd) an isotopic fractionation would be predicted with clinopyroxene preferentially taking lighter Nd isotopes. This prediction is at odds with the observed subparallel trends of δ 146 Nd versus major element contents which suggest isotopically heavier clinopyroxene ( Figure 5) and the measured clinopyroxene-whole pair in gabbro MS20-10 (Figures 6, 8). Two solutions could account for this discrepancy either the theoretical prediction does not hold in this instance or alternatively a disequilibrium reaction is occurring.  (Kempton et al., 1991;Hart et al., 1999;Holm, 2002) are not shown above 100 m. Box and whisker plots are shown for basalts immediately adjacent to the Atlantis II fracture zone (Robinson et al., 2001;Coogan et al., 2004) and Southwest Indian Ridge MORB (Robinson et al., 2001;Coogan et al., 2004;Meyzen et al., 2005). The blue shaded field is the average composition of Hole 735B as shown in Figure 4.
To explain the trend seen in Figure 6B would require that the gabbros with highest clinopyroxene contents represent magmas that have previously fractionated a significant amount of isotopically heavy clinopyroxene and then migrated and crystallized with a significantly lighter whole rock composition. To explore this process, we undertook Rayleigh fractionation modeling to show the effect of crystallizing isotopically heavy clinopyroxene (Figure 10): based on the measured offset in sample MS 20-10 ( 146 Nd clinopyroxene−wholerock = +0.047 ). Even with an extreme amount of fractionation of solely clinopyroxene (up to 40%), which is rather unrealistic and not consistent with major element compositions of the gabbros, the lightest melt produced only has a δ 146 Nd of −0.048 which is significantly less than the range preserved in Hole 735B (Figure 10). Plagioclase probably also has a role to play in the evolution of the melts in Hole 735B. Given the lack of direct  Niu et al. (2002) and Bach et al. (2001) with comparative data taken from Dick et al. (1999), for clarity only the three major rock types are plotted. Dotted lines at 528, 941, and 1321 m represent major mineralogical discontinues identified in Dick et al. (2002) thought to delineate separate gabbro bodies. (B) Graph of δ 146 Nd in gabbroic rocks and clinopyroxene separates relative to sample depth. The yellow shaded area shows the average for Indian MORB (-0.025 ± 0.005 ; 95% SE) with the blue field the average composition of Hole 735B (-0.067 ± 0.033 ).
measurements of this phase, we can merely make an educated assumption based on stable isotope principles. Based on Nd coordination in plagioclase ( VI Nd) and silicate melts ( VI Nd or VII Nd) it would be predicted that plagioclase should have δ 146 Nd identical or slightly heavier than the melt. If the two phases occur in subequal proportions (as shown in Figure 6B) due to the difference in incompatibly, plagioclase will account for only about 1/5th of the total Nd of the rock. Therefore, even if Nd occurs in sevenfold coordination in melts and assuming a large 146 Nd plagioclase−wholerock of +0.1 , δ 146 Nd will change more rapidly but does not differ significantly from the pure clinopyroxene model already discussed (Figure 10). Ultimately, accumulation or fractionation of a mixture of clinopyroxene and plagioclase in subequal proportions is not capable of explaining the wide range of δ 146 Nd observed in Hole 735B gabbros instead disequilibrium processes are probably required.

Evidence for Reactive Porous Flow
The complicated histories preserved in plutonic rocks, collected from the three main ocean basins, have been used to propose that RPF is a nearly ubiquitous feature of the lower oceanic crust (e.g., Lissenberg and MacLeod, 2016). Complex zoning profiles, frozen mineral reaction textures, and mineral compositions and proportions that do not imitate models of simple fractional crystallization suggest that additional processes must have been involved in the formation of oceanic gabbros (Meyer et al., 1989;Coogan et al., 2000;Dick et al., 2002;Gao et al., 2007;Lissenberg and Dick, 2008;Lissenberg et al., 2013). RPF has been widely recognized through the over-enrichment of very to moderately incompatible trace elements in clinopyroxene (e.g., Ce/Y; see Figure 5 in Lissenberg and MacLeod, 2016), and poor correlations between mineral compositions and fractionation indices (e.g., elevated TiO 2 contents in clinopyroxene; see Figure 5 in Lissenberg and Dick, 2008).
Frontiers in Earth Science | www.frontiersin.org 9 February 2020 | Volume 8 | Article 25  Table 1 of Niu et al. (2002).ˆBeam Intensity on 148 Nd was too low (<0.1V) for δ 148 Nd to be accurately determined. The weighted δ 146 Nd of the oceanic crust was calculated by weighting the δ 146 Nd of the gabbro samples by their share of total Nd. *The average εNd composition of Hole 735B is calculated using published data (n = 72; Kempton et al., 1991;Hart et al., 1999;Holm, 2002;herein). Lissenberg and Dick (2008) presented a frozen example of RPF from the Kane Megamullion (mid-Atlantic Ridge), where ca. 5 cm wide channels of coarse-grained gabbro crosscuts a medium grained troctolite host rock. Percolation of clinopyroxene-saturated melt through the troctolite (Eq. 2), led to the dissolution of olivine, plagioclase (partial) and Cr-spinel in the troctolite and crystallization of high Mg# clinopyroxene and lower-anorthite plagioclase.
Olivine + plagioclase1 + Cr − spinel + melt1 Strong petrographic evidence is preserved that RPF has occurred widely in gabbros collected from the Atlantis Bank including: (1) textural evidence for the reaction of clinopyroxene with melt is preserved (i.e., complexly intergrown and ragged clinopyroxene grain boundaries); (2) troctolite-to-olivine gabbro reactions have been observed in the Hole 735B gabbros (see Figure 8 in Lissenberg and MacLeod, 2016); (3) significant core to rim enrichments of incompatible trace elements are preserved in clinopyroxene (Meyer et al., 1989;Kvassnes, 2004;Gao et al., 2007;Lissenberg and MacLeod, 2016); and (4) the composition of major silicate phases is poorly correlated with the evolution of the main gabbro bodies, instead a wide range in minor element concentrations (both compatible and incompatible) are seen in olivine and pyroxene at a given Mg#, consistent with widespread permeable flow of late-stage melt through these intrusions (Dick et al., 2002). Clinopyroxene with TiO 2 contents >0.3 wt% cannot be generated by fractional crystallization of a primitive MORB magma and instead have been interpreted to reflect RPF (Lissenberg and Dick, 2008;Lissenberg and MacLeod, 2016). A negative correlation is observed between δ 146 Nd in clinopyroxene and the TiO 2 of the clinopyroxene for samples from Hole 735B (Figure 11), consistent with increasing amounts of RPF producing isotopically light clinopyroxene. Furthermore, when the clinopyroxene separates are plotted on the Cpx# versus δ 146 Nd graph (Figure 6B), they are generally heavier, but variably offset from the well-defined whole rock trend consistent with a disequilibrium process where each individual sample is affected uniquely.
Given the multi-faceted evidence for RPF in Hole 735B next we explore the effects of this process on δ 146 Nd values. The preservation of highly fractionated isotopic signatures in minerals is a widespread phenomenon and can arise because of diffusion during crystal growth or through entrainment of crystals that are not in equilibrium with a melt (e.g., Lundstrom et al., 2005;Weyer and Ionov, 2007;Sio et al., 2013;McCoy-West et al., 2018). When kinetic isotope fractionation occurs light isotopes will preferentially diffuse faster than heavier ones (e.g., Richter et al., 2009), although diffusion in a medium is also dependent on temperature, pressure, Frontiers in Earth Science | www.frontiersin.org 10 February 2020 | Volume 8 | Article 25 chemical composition and the spatial anisotropy of the medium (Watkins et al., 2017). In silicate melts, the ratio of the diffusivities of two isotopes (D rel ) of the same element have been shown to depend on their masses Richter et al. (2003): where D H and D L are the diffusivities of the heavy and light isotopes, respectively, β is a dimensionless empirical parameter allowing comparison between different elements, and m H and m L are their masses (in this instance 145.91313 and 143.9101, respectively). Given there is currently no theoretical basis for predicting the value of the β exponent it must be determined experimentally (Watson and Müller, 2009). Unfortunately experiments for Nd do not exist, however, the experimental determined values of β in silicate melts for FIGURE 11 | Relationship between the TiO 2 content of clinopyroxene and δ 146 Nd in Hole 735B. Clinopyroxene major element data is only available for two of the whole rock gabbro samples . TiO 2 contents >0.3 wt% are considered to result from reactive porous flow. A strong negative correlation is observed between δ 146 Nd and the TiO 2 of the clinopyroxene, consistent with increased amounts of reactive porous flow producing isotopically lighter clinopyroxene.
heavier elements possess an extremely restricted range of 0.03-0.07 (Ca, Mg, Fe;Watkins et al., 2017). When modeling diffusion during crystal growth two assumptions are made: (1) the surface of the mineral is in equilibrium with the Frontiers in Earth Science | www.frontiersin.org 11 February 2020 | Volume 8 | Article 25 The magnitude of isotopic fractionation is proportional to the rate of crystal growth (R) relative to the diffusivity in the medium (D). By making geological reasonable assumptions it is possible to estimate the likely kinetic isotope fractionation in the major Nd hosts of the gabbros (clinopyroxene and plagioclase).
fluid at the solid-fluid interface and (2) that diffusion in the solid is negligible (which is valid given Nd diffusion in clinopyroxene and plagioclase is extremely slow; Van Orman et al., 2001;Cherniak, 2003). Therefore, any kinetic effect isotope fractionation occurs due to diffusion in the melt. Albarede and Bottinga (1972) introduced the concept of diffusive boundary layers showing that in fast growing crystals the uptake of impurities can diverge significantly from equilibrium, with the key parameter being the ratio of the growth rate to the diffusivity of the impurity (R/D; Figure 12). The kinetic uptake of REE impurities is more efficient when grain sizes are larger, or when a high melt flux moves through a region (Van Orman et al., 2002) making gabbros in the lower oceanic crust an ideal location for this process to occur. Watson and Müller (2009) developed a simple expression that calculates the maximum kinetic isotope fractionation in a growing crystal that would occur prior to the system reaching steady-state: where δ is the deviation from equilibrium in parts per thousand. R is the growth rate of the crystal (in cms −1 ), K eq is the equilibrium partition coefficient between the crystal and the growth medium, and x BL is thickness of the boundary layer. The magnitude of isotope fractionation is proportional to both the thickness of the diffusive boundary layer and the relative diffusivity (D rel ) of the two isotopes of interest which is dependent on the β factor for the system (Figure 12A). By assuming the diffusive boundary layer is 100 µm thick, the maximum observed in silicate systems (Kerr, 1995), and using the minimum experimentally determined β factor (i.e., β Nd = β Fe = 0.03; Richter et al., 2009) we can explore the size of fractionations in δ 146 Nd which can be ascribed to RPF ( Figure 12B). This modeling shows that for a given crystal growth rate, larger isotopic fractionations are predicted when Nd is more incompatible in a phase, and minimal fractionation is predicted at R/D < 1. Assuming an R/D of 10, which is at the upper end of values for phenocryst growth (Orlando et al., 2008;Maaløe, 2011), kinetic isotope fractionations of ca. 0.021 and 0.036 are predicted in clinopyroxene and plagioclase, respectively. A strong negative correlation is observed between δ 146 Nd and the modal proportion of clinopyroxene in Hole 735B gabbros ( Figure 6B). For kinetic process to explain this correlation would require that the extremely isotopically light gabbros represent quenched melts that have lost a proportion of their heavy isotopes during RPF in a different region of the lower crust (melt migration is observed at least on the short length scales in Hole 735B). Natland and Dick (2001) suggested that these gabbroic cumulates formed in a dense crystal mush dominated by fractures, channelized flow and intergranular porous flow which is the ideal environment to generate kinetic isotope fractionations. The mechanism responsible for generating variations in δ 146 Nd is clearly not straightforward, with the sample with the lightest δ 146 Nd clearly not adhering to this correlation [BN-1(F) = −0.127 ]. However, this sample was collected at 521 mbsf, within a region that contains several distinct lithostragraphic units, and within 10 m of a major boundary between distinct two gabbro bodies (Figure 8). This region probably experienced more concentrated melt flow aiding in the dissolution and reprecipitation of isotopic signatures at grain boundaries. Given the small magnitude of isotopic fractionation predicted in Figure 12, for RPF alone to account for the δ 146 Nd variations observed in Hole 735B gabbros one of the following is required: (1) crystal growth rates during this process are higher than previous recognized; (2) the β value for Nd is higher than the minimum experimentally determined value used for the predictions; or (3) several cycles of RPF are required to generated the extreme isotope compositions of some the gabbros. Further experimental work is required to better understand this process. Given the unrivaled complexity of the oceanic curst at slow spreading ridges it is fitting that no single process can conclusively explain the variability observed instead a combination of several processes is probably required.

Reconciling the Disconnect Between the Composition of the Oceanic Crust and Overlying MORB
The Homogenous δ 146 Nd of MORB The large variations observed in δ 146 Nd in the gabbroic rocks of Hole 735B are distinctly different than erupted Indian MORB which possesses a very restricted range of δ 146 Nd from −0.015 to −0.029 (Figure 3). The average composition of Indian MORB is δ 146 Nd = −0.025 ± 0.010 (Figure 8), which is identical within uncertainty to the composition of Pacific MORB (Figure 3) and the global MORB average of −0.024 ± 0.013 ( Table 1). In contrast, available mantle peridotite data display a significantly wider range of δ 146 Nd with values ranging from −0.045 to + 0.029 (McCoy-West et al., 2017). These peridotites have significantly lower Nd concentrations (0.10-0.81 ppm; McCoy-West et al., 2020) than erupted MORB and comprise samples from a variety of tectonic settings, therefore, several processes may be responsible for their variable δ 146 Nd including, but not limited to, partial melting, melt metasomatism, and mineralogical variations. In the MORB dataset no correlations are observed between δ 146 Nd and Nd radiogenic isotope compositions (Figure 4B), showing that mantle source heterogeneity and the process of formation are effectively decoupled. Given that δ 146 Nd is process dependent this means any initial variations in the δ 146 Nd of the source material need not be transferred to the final MORB, with the indistinguishable δ 146 Nd of global MORB suggesting that these magmas have been homogenized through the same global process. Partial melting concentrates Nd into the melt phase (modeling suggests that with only 10% melting >80% of the original Nd budget will be in the melt fraction; McCoy-West et al., 2015) helping effectively eliminate small scale heterogeneities from the source region. Further processing in the extensive magma chamber network observed at mid-ocean ridge systems globally, would be an ideal place to form such a homogenous reservoir. These magma chambers are required to be open systems to explain the trace element over-enrichment of MORB (O'Hara, 1977;Jenner and O'Neill, 2012), and therefore are constantly replenished with new melt providing an ideal mechanism to homogenize the δ 146 Nd of the magmas. Amongst the novel stable isotopes systems Nd appears unique with small heterogeneities greater than analytical precision observed for many other isotopic systems including Fe, Mo, Ba Bezard et al., 2016;Nielsen et al., 2018). It will be interesting if the homogeneity of the MORB reservoir is confirmed when new data becomes available.
Can the Melts in the Oceanic Crust Generate MORB in the SWIR?
The discussion above shows that to produce the extremely light values observed in Hole 735B gabbros requires some level of RPF. Advocates of RPF also suggest that this mechanism is responsible for the geochemical variability observed in erupted MORB glasses (e.g., Lissenberg and Dick, 2008;Lissenberg and MacLeod, 2016). Meaning that MORB magmas and gabbros in the oceanic crust should directly complement each other. There are several lines of evidence that suggest this is not the case for the samples from the SWIR: (1) there is a significant disconnect between the major element composition of MORB magmas and the lower oceanic crust with MORB glasses being significantly more evolved than the bulk composition of Hole 735B, whereas they should be the complimentary melts of the preserved oceanic crust (i.e., Mg#; Figure 6A). Furthermore, the average composition of the crust at Hole 735B is not sufficiently primitive in terms of Mg# or highly compatible trace element concentrations (e.g., Ni) for the bulk crust to be in equilibrium with the mantle (Coogan et al., 2001); (2) Hole 735B is largely comprised of cumulate gabbros (Dick et al., 2000;Natland and Dick, 2001). Despite significant Sr and Eu anomalies (from plagioclase accumulation) being observed in individual samples (1-4 m length scale), the average composition of the complete upper 500 m section of the Hole 735B displays no significant Sr or Eu anomalies (≤1%), this implies that there has been local separation of melt and solids, but no large scale removal of melts into the upper crust (Hart et al., 1999); and (3) stable Nd isotope compositions also suggest a significant disconnect between the lower oceanic crust and MORB compositions. The calculated average composition of the gabbro whole rocks is δ 146 Nd = −0.067 ± 0.033 (95% SE), with the weighted average based on the proportion of Nd in each sample being almost an identical −0.064 ( Table 2). This large isotopic discrepancy makes it very unlikely that significant melt has escaped the lower oceanic crust, and for the SWIR at least oceanic gabbros that have experienced RPF have had little impact on the compositions of erupted MORB.

CONCLUSIONS
Stable isotope compositions are process dependent and therefore provide an excellent mechanism to compare contrasting petrogenetic models. Here, we present the first neodymium stable isotope compositions of Indian Ocean MORB glass and gabbro whole rock and clinopyroxene separates from ODP Hole 735B on the SWIR which show: ( (2) In Hole 735B the δ 146 Nd of gabbroic rocks is highly variable ranging from −0.026 to −0.127 . Clinopyroxene separates also possess variable δ 146 Nd (δ 146 Nd = 74 ppm) but are variably offset to heavier values than the gabbro whole rocks at the same major element composition, with a 146 Nd clinopyroxene−wholerock of +0.047 observed in one paired sample.
(3) Rayleigh fractionation modeling shows fractionation of isotopically heavy clinopyroxene ± plagioclase has a limited effect on δ 146 Nd even after significant fractional crystallization (F = 40%). Therefore, the large variations in δ 146 Nd observed in the oceanic gabbros cannot be generated solely by the fractionation or accumulation of magmatic minerals. (4) Strong evidence exists to support RPF in Hole 735B, a process which will induce kinetic isotope fractionation during crystal growth. However, diffusion modeling based on geologically reasonable parameters, shows that minimal kinetic isotope fractionation ca. 0.02 will be induced in clinopyroxene. Thus, to produce the extremely light δ 146 Nd values observed in the gabbros would require multiple cycles of dissolution and reprecipitation of isotopic signatures at grain boundaries. (5) In Hole 735B there is a large offset, significantly greater than analytical uncertainty, between the average composition of the lower oceanic crust (δ 146 Nd = −0.067 ) and erupted Indian MORB (δ 146 Nd = −0.025 ). Combined with the geochemical evidence (e.g., Sr and Eu anomalies) for no large-scale removal of melts into the upper crust we conclude that melts involved in RPF have no significant impact on the Nd stable isotope composition of MORB erupted at the SWIR. Whether this conclusion is a global phenomenon remains to be tested.

DATA AVAILABILITY STATEMENT
All datasets generated for this study are included in the article/supplementary material.

AUTHOR CONTRIBUTIONS
AM-W and KB conceived the study. AM-W and M-AM developed the double spike. AM-W undertook the chemistry and mass spectrometry, and wrote the manuscript with input from all the authors.

FUNDING
This project was funded by a National Environmental Research Council (United Kingdom) Grant (NE/N003926/1) to KB. While at Monash AM-W was supported by ARC grant FL160100168 to Peter Cawood.