Plagioclase archives of depleted melts in the oceanic crust 1

6 Mid-ocean ridge and ocean island basalts provide vital but incomplete insights into the chemical 7 structure of Earth’s mantle. For example, high-anorthite plagioclase carried by these basalts is 8 generally too primitive and incompatible-element depleted to have crystallized from them. 9 Moreover, erupted basalts rarely preserve the strong isotopic and incompatible-element 10 depletions found in some melt inclusions and mantle residua represented by abyssal peridotites. 11 By integrating experimental observations with published analyses of natural crystals and glasses, 12 we demonstrate that high-anorthite plagioclase is in equilibrium with melts generated by high13 degree melting of depleted mantle sources. Although such melts seldom erupt, their imprints on 14 crystal and melt inclusion records nonetheless suggest that high-anorthite plagioclase grows from 15 endmember but essentially unexotic magmas. The widespread occurrence of high-anorthite 16 plagioclase in both oceanic basalts and the oceanic crust hence indicates that depleted melts are 17 pervasive in the upper mantle and lower crust despite rarely reaching the surface. Plagioclase 18 archives therefore imply that depleted melts play much a greater role in lower crustal accretion 19 than typically recognized and that the upper mantle may also be more depleted than previously 20 thought. 21


INTRODUCTION 23
Mid-ocean-ridge and ocean-island basalts (MORB and OIB respectively; oceanic basalts 24 collectively) offer important windows into the chemical structure of Earth's mantle (Hofmann, 25 1997;Stracke, 2021). Over billions of years, lithospheric recycling at subduction zones has 26 created chemically, isotopically and lithologically enriched mantle domains that are ultimately 27 reflected in the compositions of erupted basalts (Chase, 1981). Melt extraction over geological 28 time has also created depleted domains that are well documented in abyssal peridotites but absent 29 from erupted records, reflecting the poor preservation of depleted melts during magmatic 30 evolution (Byerly and Lassiter, 2014;Warren, 2016;Neave et al., 2019). Melt inclusions in 31 primitive crystals, which are relatively resistant mixing-induced overprinting, thus provide vital 32 constraints on the chemical and isotopic variability of primitive melts and their mantle sources 33 (Sobolev and Shimizu, 1993;Maclennan, 2008bMaclennan, , 2008aStracke et al., 2019). However, whether 34 melt inclusions faithfully reflect the diversity of deep melt compositions remains to be seen. 35 Fortunately, crystals also record information about the melts from which they grow, and as 36 volumetrically significant components of magmas and cumulates may reflect the relative 37 abundances of chemically distinct melts at depth more closely than melt inclusions. 38 High-anorthite plagioclase (XAn > 0.8, where XAn = molar Ca/(Ca+Na+K)) is often a 39 major constituent of basalts from ocean islands and slow-to intermediate-spreading mid-ocean 40 ridges (Lange et al., 2013), as well as cumulates from ophiolites and the lower oceanic crust 41 (Browning, 1982;Lissenberg et al., 2013). However, such plagioclase crystals are rarely in 42 major-element equilibrium with erupted oceanic basalts (cf. Natland et al., 1983). Moreover, 43 they are often out of isotopic and incompatible-element equilibrium with their host liquids, 44 implying origins from different mantle melt distributions (Halldórsson et al., 2008; primitive melts from enriched sources (Neave et al., 2019). To place these isobaric observations 69 into the polybaric context of crustal magmatism, we performed new crystallization experiments 70 on the same analogues of the high-Ca# Háleyjabunga and low-Ca# Stapafell lavas from the 71 Reykjanes peninsula in Iceland at 100 and 600 MPa (Supplementary Material). 72 Plagioclase-liquid equilibria at 100 and 600 MPa are summarized in Fig. 1 alongside 73 published equilibria at 300 MPa from Neave et al. (2019). The depleted Háleyjabunga analogue 74 saturates in plagioclase at higher melt MgO contents (and temperatures) than the enriched 75 Stapafell analogue. While isobaric differences in plagioclase saturation conditions between the 76 two starting compositions reflect mantle-derived variability in melt Ca# and Al#, polybaric 77 differences reflect variability in the relative stabilities of plagioclase and clinopyroxene, with 78 plagioclase generally saturating at lower temperatures (and melt MgO contents) when 79 clinopyroxene stability is enhanced at higher pressures ( Fig. 1). Equilibrium plagioclase XAn also 80 correlates negatively with clinopyroxene stability and therefore pressure. Overall, melt 81 composition, which correlates with intensive conditions as well as source composition, exerts the 82 main control over XAn, and high-XAn plagioclase is only produced from the high-Ca# 83 Háleyjabunga analogue (up to XAn = 0.88 and 0.85 in the products of 100 and 300 MPa 84 experiments, respectively). Importantly, this demonstrates that high-XAn plagioclase is produced 85 from known, if highly depleted, oceanic basalt compositions under realistic intensive conditions 86 (cf. Grove et al., 1992;Kohut and Nielsen, 2003). Thus, even if the Háleyjabunga lava is at the 87 limit of erupted compositions (Fig. 3), our findings nonetheless suggest high-XAn plagioclase 88 crystals reflect the crystallization of endmember but otherwise unexotic melts derived from 89 depleted mantle sources. 90

PREDICTING PLAGIOCLASE-LIQUID EQUILIBRIA 91
By predicting equilibrium plagioclase XAn as a function of melt composition it is possible 92 to evaluate plagioclase-liquid equilibria in more systems than could ever be investigated 93 experimentally. While thermodynamic models allow phase relations to be robustly inter-and 94 extrapolated across wide parameter spaces (Ghiorso and Sack, 1995;Holland et al., 2018), 95 empirical models can be more precise when applied within their calibration ranges (e.g., Namur 96 et al., 2012). Moreover, it is possible to avoid making potentially erroneous assumptions about 97 crystallization conditions by calibrating an empirical model that predicts equilibrium XAn from 98 melt compositions alone; intensive conditions are implicit in melt compositions. 99 Performing multiple linear regression through calibration data (n = 98) from experimental 100 studies on basalts from mid-ocean ridges, an oceanic plateau and an ocean island yields the 101 following relationship between plagioclase XAn and melt composition ( Test data (n = 36) from experimental studies on basalts from mid-ocean ridges and a continental 105 hotspot with XAn ~ 0.6-0.9 are reproduced well by equation 1 (r 2 = 0.92; standard error = 0.02), 106 albeit with a slight offset to lower XAn, possibly because of Na loss from some furnace 107 experiments ( Fig. 2b; sources in the Supplementary Material). 108

ORIGINS OF HIGH-ANORTHITE PLAGIOCLASE 109
High-XAn plagioclase has been described in lavas from many mid-ocean ridge segments 110 and ocean islands (e.g., Lange et al., 2013). Here we apply our model to published Icelandic and 111 MORB glass compositions, though our findings are likely to be applicable in dry basaltic settings 112 where fewer compositions have been published. Equilibrium plagioclase XAn predicted from 113 Icelandic (n = 190) and MORB (n = 1687) glass compositions collated by Shorttle and Maclennan (2011) and Gale et al. (2013), respectively, are shown in Fig. 3. Predicted XAn 115 contents were also filtered for plagioclase saturation using a stability criterion from Gale et al. 116 (2014)  XAn. Crucially, some glasses from both datasets return stable high-XAn compositions (n = 33 and 121 22, respectively). Although these glasses are at the limit of natural variability in the case of 122 MORB, their occurrence nonetheless substantiates rare observations of natural high-XAn crystals 123 (XAn up to 0.89;Natland et al., 1983). Icelandic glasses return higher predicted maximum XAn 124 contents than MORB glasses (up to XAn = 0.89 and 0.85, respectively), likely reflecting 125 differences in tectonic setting, source composition and mantle temperature. 126 As well as being associated with high values of melt Ca# at any given melt MgO content 127 (Figs. 3A and 3B), high-XAn plagioclase is typically associated with low melt K2O contents 128 (often <0.1 wt.%; Figs. 3C and 3D), recapitulating the incompatible-element-depleted character 129 of erupted high-XAn crystals (Neave et al., 2014;Nielsen et al., 2020). Such high-Ca#, low-K2O 130 melts are typically generated by shallow melting of depleted sources that have experienced high 131 degrees of prior fractional melting (e.g., Grove et al., 1992;Shorttle and Maclennan, 2011). 132 High-XAn plagioclase is also associated with low melt FeO* contents (total Fe as FeO) at any 133 given melt MgO content (Figs 3E and 3F). This is particularly clear for Iceland, where low-FeO* 134 primitive melts (FeO* < 10 wt.%) are predicted to be equilibrium with high-XAn plagioclase but 135 high-FeO* primitive melts (FeO* > 10 wt.%) are not expected to be in equilibrium with 136 plagioclase at all (Fig. 3E). The depleted melts from which high-XAn plagioclase crystallizes are thus from dominantly peridotitic sources and have largely escaped contamination by melts from 138 enriched lithologies during ascent (Shorttle and Maclennan, 2011), though rare K2O-rich melts in 139 equilibrium with high-XAn plagioclase represent exceptions that may have interacted with 140 depleted harzburgites ( Fig. 3D; Nielsen et al., 2020). We hence argue that high-XAn plagioclase 141 crystals are the solid products of depleted melts feasibly derived from depleted residua recorded 142 by some abyssal peridotites (Byerly and Lassiter, 2014;Warren, 2016) that rarely erupt at the 143 surface despite sometimes being found in melt inclusions (Sobolev and Shimizu, 1993;144 Maclennan, 2008b;Stracke et al., 2019). 145

WIDESPREAD DEPLETED MELTS AT DEPTH 146
High-XAn plagioclase occurs throughout the oceanic realm (Fig. 4). In Iceland, it is 147 especially well documented in the Eastern Volcanic Zone (Fig. 4A), where isotopically and 148 incompatible-element-depleted high-XAn plagioclase may constitute >30 vol.% of basaltic lavas 149 that are otherwise relatively enriched and evolved (Halldórsson et al., 2008). High-XAn 150 plagioclase has also been reported from depleted picrites in the Northern Volcanic Zone of 151 Iceland that formed in response to deglaciation-driven decompression (Maclennan et al., 2003). 152 As well as being found throughout Iceland, high-XAn plagioclase is also well documented in 153 Galápagos, Réunion and Kerguelen (Fig. 4A), suggesting that incompatible-element-depleted 154 melts with high Ca# are more prevalent beneath ocean islands than implied from enriched OIBs. with high-XAn absent from seafloor lavas but present at depth, presumably as a consquence of 171 filtering by the axial melt lens. The association of high-XAn plagioclase with primitive olivine 172 and clinopyroxene in diverse settings indicates that is forms before mixing or reactive porous 173 flow fully overprint signatures from depleted mantle sources (Maclennan, 2008a;Lissenberg et 174 al., 2013).

PHASE EQUILIBRIA EXPERIMENTS Starting materials
The Háleyjabunga and Stapafell lavas are amongst the most geochemically extreme primitive basalts from Iceland in terms of their major element, trace element and isotopic compositions, which makes them excellent model systems for exploring the consequences of mantle-derived chemical variability (Gurenko and Chaussidon, 1995;Maclennan, 2008). The lavas are thought originate from lithologically distinct mantle sources, with the incompatibleelement-depleted Háleyjabunga lava being generated by high-degree melting of an initially fertile peridotite and the Stapafell lava being largely generated by modest-degree melting of a recycled and modally enriched (i.e., clinopyroxene-rich) peridotite (Shorttle and Maclennan, 2011;Neave et al., 2018). Of key relevance here is that the Háleyjabunga lava is relatively Caand Al-rich while the Stapafell lava is relatively Fe-and Na-rich, which results in fundamentally different phase equilibria between the two compositions (Neave et al., 2019b).
The synthesis of starting materials is described in detail by Neave et al. (2019b) and summarized below. Natural glass compositions from Condomines et al. (1983), Gurenko and Chaussidon (1995) and Peate et al. (2009) were corrected to the same initial melt MgO content of ~10.5 wt.%. Starting materials with these corrected compositions were then synthesized from reagent-grade oxide and carbonate powders that were fused twice in Pt crucibles at 1600 °C at the Institut für Mineralogie of the Leibniz Universität Hannover, Germany. Platinum crucibles were quenched in H2O after each fusion to ensure the production of homogenous glasses.

Experimental methods
Crystallization experiments were performed in an internally heated pressure vessel (IHPV) at the Institut für Mineralogie of the Leibniz Universität Hannover, Germany, using methods described in detail by Husen et al. (2016) and Neave et al. (2019b) and summarized below. Approximately 50 mg of each powdered starting material was loaded into Au80Pd20 capsules that had first been treated to contain 0.25-0.30 wt.% Fe to minimize Fe exchange with capsule materials (e.g., Gaetani and Grove, 1998). Capsules were suspended from a Pt wire in the hot zone of the IHPV. Experiments were performed at either 100 MPa or 600 MPa in an Ar pressure medium, and at temperatures that ranged from 1140 to 1240 °C. Pressure was monitored with a strain gauge manometer and did not vary by more than 5 MPa during the course of the experiments. Temperature was monitored with four unsheathed S-type thermocouples spaced across the hot zone and was typically within 5 °C of the target temperature. Experimental temperatures were approached by heating the furnace from room temperature to 10 °C below the target temperature at a rate of 50 °C/min; final heating was performed at a rate of 10 °C/min to avoid overshooting. Experimental durations varied from 48 hours for near-and super-liquidus experiments to 120 hours for lower-temperature experiments. Experimental products were quenched by fusing the Pt wires on which capsules were suspended, thereby dropping them into a cold zone at the base of the vessel.

Analytical methods
The major element content of experimental products was determined by electron probe microanalysis (EPMA) with a Cameca SX100 instrument at the Institut für Mineralogie of the Leibniz Universität Hannover, Germany. Silicon, Ti, Al, Cr, Fe, Mn, Mg, Ca, Na, K and P were measured in glasses with a beam size of 10 µm, an accelerating voltage of 15 kV and a current of 10 nA. Silicon, Ti, Al, Cr, Fe, Mn, Mg, Ca, Na and K were measured in minerals with a beam size of 1 µm, an accelerating voltage of 15 kV and a current of 15 nA. Gold, Pd and Fe were measured in capsules with a beam size of 1 µm, an accelerating voltage of 15 kV and a current of 40 nA. Elements were counted on peak for 20 s, with the exceptions of Si and Na that were counted on peak for 10 s to minimize drift and Na migration. Background counting times were half on-peak counting times. The following standards were used for calibration: wollastonite (Si and Ca), TiO2 (Ti), Al2O3 (Al), Cr2O3 (Cr), Fe2O3 (for Fe in silicates and Cr-spinel), Fe metal (for Fe in capsules), Mn3O4 (Mn), MgO (Mg), albite (Na), orthoclase (K), apatite (P), Au metal (Au) and Pd metal (Pd).
To ensure internal consistency across multiple sessions, analyses were normalized as follows: glass analyses were normalized to VG-2 basalt glass (NMNH 111240-52; using the preferred MgO content); clinopyroxene, low-Ca pyroxene and plagioclase analyses were normalized to Kakanui augite (NMNH 122142; using preferred values); olivine analyses were normalized to San Carlos olivine (NMNH 111312-44); and chromite analyses were normalized to Tiebaghi Mine chromite (NMNH 117075) (Jarosewich et al., 1980). Accuracy and precision were monitored by measuring the following standards that were also normalized for each analytical session: A-99 basaltic glass (NMNH 113498), Ney County Cr-augite (NMNH 164905) and Lake County plagioclase (NMNH 115900) (Jarosewich et al., 1980(Jarosewich et al., , 1987. Major (>1 wt.%) and minor (<1 wt.%) elements were determined with accuracies better than 2% and 10%, and 1σ precisions better than 2% and 15% respectively. Analyses of standards are provided alongside analyses of experimental products in the Supplementary Data. Compositions of experimentally produced glasses and plagioclase crystals are summarized in Supplementary Fig. 1.
Glass H2O contents were determined in experimental products with low crystal contents by Fourier-transform infrared (FTIR) spectroscopy with a Bruker IFS88 instrument at the Institut für Mineralogie of the Leibniz Universität Hannover, Germany, following the methods described by Husen et al. (2016). In short, H2O contents were determined in ~100-µm thick wafers using the peak attributed to the OH stretch vibration (3550 cm −1 ) using a molar absorption coefficient

Experimental oxygen fugacities
All experiments were conducted under nominally dry conditions (no H2O was added to dried starting materials), which resulted in melt H2O contents of 0.56-0.91 wt.% following the reduction of Fe2O3 in the starting glasses to FeO and the inward diffusion of trace H2 from the Ar pressure medium at high temperatures. Colorimetric analyses of experimental products with low crystal contents performed with the approach of Schuessler et al. (2008) returned Fe 3+ /ΣFe values of 0.13-0.23, which correspond to oxygen fugacity(fO2) conditions expressed with respect to the fayalite-magnetite-quartz (FMQ) buffer of FMQ+0.2 to FMQ+1.3 (Kress and Carmichael, 1991). Capsule compositions record broadly similar conditions of FMQ+0.0 to FMQ+1.2 (Barr and Grove, 2010). Estimated fO2 conditions are provided in the Supplementary Data.

PREDICTING PLAGIOCLASE-LIQUID EQUILIBRIA Rationale and data sources
Empirical models for predicting plagioclase-liquid equilibria and equilibrium plagioclase anorthite contents (XAn, where XAn = molar Ca/(Ca+Na+K)) are typically calibrated across large ranges of melt composition (Namur et al., 2012 and references therein). While such global calibrations facilitate internally consistent modelling across diverse situations they can also result in lower accuracy and precision than can be achieved by calibrating and applying models under more restricted conditions. Moreover, for technical reasons, the majority of published phase equilibria experiments have been performed at 0.1 MPa (i.e., 1 atm), meaning that plagioclaseliquid equilibria models are generally better constrained at pressures lower than those at which the majority of crustal magmatism takes place (Namur et al., 2012). Here we present a new empirical model optimized for predicting plagioclase-liquid equilibria in oceanic basalts evolving under crustal pressure and temperature conditions. Calibration data (n = 98) were sourced from relatively recent studies on H2O-poor (typically <1 wt.%) oceanic tholeiites at a range of pressures (0.1-700 MPa) that report high quality EPMA data. Specifically, data were sourced from experiments on ocean island basalts (OIBs) from Iceland (Neave et al., 2019a(Neave et al., , 2019b this study), mid-ocean-ridge basalts (MORBs) and plagioclase-saturated MORB liquids (Kohut and Nielsen, 2003;Voigt et al., 2017), and oceanic plateau basalts from Shatsky Rise (Husen et al., 2016). Only data from experimental runs containing <50 wt.% glass were used in the calibration to ensure that plagioclase and liquid pairs had approached equilibrium as closely as reasonably possible. The distribution of plagioclase XAn in the calibration dataset is summarized in Supplementary Fig. 2A which is analogous to Fig. 2 in the main text.
Test data (n = 36) to independently verify regression quality were sourced from studies on H2O-poor oceanic and continental tholeiites at range of pressures (0.1-1000 MPa). Specifically, data were sourced from experiments on MORBs (Grove et al., 1992;Yang et al., 1996) and continental tholeiites from Snake River Plain (Whitaker et al., 2007(Whitaker et al., , 2008. The distribution of plagioclase XAn in the test dataset is summarized in Supplementary Fig. 2B which is analogous to Fig. 2B in the main text.

Regression strategy
Least-squares multiple linear regression was then performed using the lm() function in R (R Development Core Team, 2016). The regression equation was selected by trial and error (e.g., Putirka, 2008). Namely, melt compositional parameters were variably combined and both overall  1) where Ca# = molar Ca/(Ca+Na) and Al# = molar Al/(Al+Si). All regression coefficients are highly significant (p < 0.001), and the regression is robust (r 2 = 0.88; standard error = 0.025). Adding further compositional parameters such as melt MgO or K2O contents did not improve the quality of the fit. The standard error of our new model (0.025) is considerably better than the standard errors of the models reviewed by Namur et al. (2012), which range from 0.044 to 0.090, though our model is only calibrated for oceanic basalts that are relatively poor in H2O; the models of Namur et al. (2012) have comparable standard errors of ~0.030. Relationships between experimental XAn, predicted XAn, melt composition and other intrinsic conditions (pressure, temperature and oxygen fugacity) are summarized in Supplementary Fig. 3. While the strong dependence of predicted XAn on Ca#melt is clear in Supplementary Fig.  3A, it is also important to note that experiments are well reproduced across a wide range of intrinsic conditions relevant to the evolution of oceanic basalts ( Supplementary Figs 3D-3F). The possible underestimation of XAn in the coolest experiments is of little significance for our study that focusses on the significance of high-XAn plagioclase crystals.
A simple linear regression through the test data reveals a similarly strong relationship between experimental and predicted XAn as observed for the calibration data (r 2 = 0.92; standard error = 0.020). Moreover, relationships between experimental XAn, predicted XAn, melt composition and other intrinsic conditions in the test dataset are comparable to those in the calibration dataset ( Supplementary Fig. 4). Importantly, Eq 1 reliably captures the high-XAn compositions reported from some experiments.

Verifying plagioclase stability
Although Eq 1 can reliably predict the plagioclase XAn in equilibrium with oceanic basalts, it does not account for plagioclase stability. That is, it will return metastable equilibrium plagioclase XAn values for melt compositions that are plagioclase undersaturated. Predicted values of plagioclase XAn were filtered for plagioclase stability using the following criterion from Gale et al. (2014): Kd An × Anliq + Kd Ab + Abliq = 1. (2) Values of Anliq and Abliq were determined from glass compositions while Kd An and Kd Ab were predicted from regressions analogous in form to Eq 1 (r 2 = 0.82 and 0.42, respectively). Natural glasses were then determined to be saturated in plagioclase if the value of Eq 2 was within the 95% confidence interval of values estimated from plagioclase-saturated experiments. That is, plagioclase was considered stable when Eq 2 returned 1.00±0.03, which allows some tolerance for both analytical and fitting uncertainties. As for predicting plagioclase-liquid equilibria we favoured this empirical approach over using thermodynamic models such as the MELTS algorithm (Ghiorso and Sack, 1995) to avoid making potentially erroneous assumptions about crystallization conditions.

PLAGIOGLASE IN OCEANIC BASALTS AND THE OCEANIC CRUST Data sources
High-XAn plagioclase crystals occur throughout the oceanic realm. They are often major constituents of basalts from ocean islands and slow-to intermediate-spreading mid-ocean ridges (Lange et al., 2013), as well as cumulates from ophiolites and the lower oceanic crust (Browning, 1982;Lissenberg et al., 2013). Here we collated data from diverse studies on oceanic samples that report high-XAn. We explicitly did not incorporate data from arc settings where elevated XAn contents likely result from elevated melt H2O contents initially suppressing plagioclase crystallization (Sisson and Grove, 1993).
Mineral data are rarely reported in consistent ways between different studies. For example, some studies only report macrocryst (i.e., phenocryst) compositions while others also consider microcryst and groundmass compositions; some separate core and rim analyses while others provide no textural information. In order to maximize our collated data of natural oceanic plagioclase compositions we therefore collated all available data, texturally constrained or not. Sources of natural plagioclase compositions are summarized in Supplementary Table 1.

Setting
Location Sources n OIB Iceland, Eastern Volcanic Zone