Clinopyroxene precursors to amphibole sponge in arc crust

The formation of amphibole cumulates beneath arc volcanoes is a key control on magma geochemistry, and generates a hydrous lower crust. Despite being widely inferred from trace element geochemistry as a major lower crustal phase, amphibole is neither abundant nor common as a phenocryst phase in arc lavas and erupted pyroclasts, prompting some authors to refer to it as a ‘cryptic’ fractionating phase. This study provides evidence that amphibole develops by evolved melts overprinting earlier clinopyroxene—a near-ubiquitous mineral in arc magmas. Reaction-replacement of clinopyroxene ultimately forms granoblastic amphibole lithologies. Reaction-replacement amphiboles have more primitive trace element chemistry (for example, lower concentrations of incompatible Pb) than amphibole phenocrysts, but still have chemistries suitable for producing La/Yb and Dy/Yb ‘amphibole sponge’ signatures. Amphibole can fractionate cryptically as reactions between melt and mush in lower crustal ‘hot zones’ produce amphibole-rich assemblages, without significant nucleation and growth of amphibole phenocrysts.

T he composition of the lower crust of arcs plays an important role in magmatism and mantle-to-upper crust fluxes of metals and volatiles 1,2 . Nominally anhydrous, clinopyroxene-dominated lithologies in the lower crust are refractory and prone to foundering 3,4 . In comparison, amphibole-bearing assemblages are more prone to melting [5][6][7][8] and/or assimilation 9,10 . However, the fractionation of amphibole is often considered to be cryptic 1,11,12 given the lack of amphibole phenocrysts in many arc lavas and pyroclasts 13 , even where the presence of amphibole cumulates and autoliths is unequivocal 14 . Conversely, clinopyroxenites are less common as cumulates and autoliths than fractionation models suggest, particularly in mature arc crust 15 .
Savo, Solomon Islands, southwest Pacific ( Fig. 1), is a recently eruptive volcano formed by the on-going subduction of the Indo-Australian plate beneath the Solomon Islands Arc 16 . Savo Island is the upper third of the 1,400-m high edifice of the volcano, which is built onto an unknown basement 17 . Crustal thickness is B14 km (ref. 17). The mantle beneath Savo, and much of the Solomon Islands Arc, has potentially been metasomatized by slab melts from the earlier-subducted Pacific slab, and the currently subducting Indo-Australian slab [18][19][20][21][22] . Historical eruptions (16th and 19th centuries 17 ) have been dominated by mugearites to sodic trachytes (mildly alkaline equivalents of basaltic andesite to dacite) 18 . The interpreted fractionation sequence is typical of arc magmas with early oxides and olivine followed closely by clinopyroxene, with later amphibole 18 . Plagioclase feldspar is a ubiquitous phenocryst phase in the erupted volcanic suite, but with limited involvement in the major geochemical evolution of the suite, and is interpreted to crystallize late (that is, upon or just before extrusion of the magma) with little to no participation in fractionation, leading to elevated Sr/Y values (100-250 in the evolved rocks) 17 . Amphibole occurs as a phenocryst in the more evolved parts of the suite, but consistent La/Yb and Dy/Yb trends across the range of analysed SiO 2 indicate its fractionation from more primitive melts too-hence fractionation is cryptic early, and non-cryptic later in the suite.
This study examines the suite at Savo volcano, in particular the phenocrysts and entrained nodules of hornblendites, clinopyroxenites and mixed amphibole-clinopyroxene lithologies. Nodules are interpreted to represent a range of cumulates and reactionreplacement lithologies, with considerable amphibole formed by melt-clinopyroxene reactions. Consistent with experimental literature and empirical evidence from a range of amphibolebearing lithologies, the first appearance of amphibole in crystallizing (mafic) arc magma is likely to be at the expense of clinopyroxene (and/or olivine) by reaction-replacement. This reaction-replacement mechanism is effective at generating a 'cryptic' fractionation signature for amphibole as it develops in relatively immobile crystal mushes, and unlike true cumulates, there need not be crystal phases nucleating and growing in the liquid melt. Amphibole phenocrysts at Savo have greater enrichments of incompatible trace elements relative to the nodules, suggesting their separate and later crystallization from more evolved melts. The decoupling of phenocrysts and nodule chemistry illustrates that crystal fractionation may be cryptic, in that early formation and fractionation of minerals is not necessarily recorded in the crystal assemblage of the daughter magmas, and that phenocryst population is not representative of the crystallization history of a magma.

Results
Mineralogy and petrography. The erupted rocks of Savo contain abundant nodules (Fig. 2) of clinopyroxenites (±olivine), hornblendites and assemblages with varying proportions of clinopyroxene and amphibole (±olivine). Gabbroic assemblages are rare, with plagioclase occurring rarely as a highly calcic, anhedral phase. This is in line with the chemistry of the evolved rocks (Eu  anomaly poorly developed or absent, Sr behaves as an incompatible element), which suggests a limited role for plagioclase fractionation 18 , likely a result of elevated H 2 O in the magma 4,23 .
Magnetite is present in all samples as a minor phase, often as inclusions, suggesting it is an early-crystallizing mineral.
Clinopyroxenites are dominated (90% by volume) by clinopyroxenes that straddle the augite-diopside composition. Olivine occurs as isolated crystals within the nodules, and is typically Fo 70-80 . Clinopyroxenites contain variable amounts of amphibole, and can be considered an end member of a continuum through mixed amphibole-clinopyroxene to hornblendites ( Fig. 2a-e).
Nodules were analysed for whole rock geochemistry by X-ray fluorescence (major and trace elements) and solution inductively coupled plasma mass spectrometry (trace elements), and crystal chemistry by scanning electron microscope-energy dispersive X-ray microanalysis (SEM-EDXA; major elements) and laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS; trace elements). Details of analyses are described in Methods and Supplementary Data 1.

Discussion
The textures of amphibole in clinopyroxenites and clinopyroxene-amphibole cumulates through to granoblastic hornblendites support a reaction-replacement 24 origin, with replacement of clinopyroxene as blebs along cleavage planes 25 (Fig. 2a) at crystal margins (producing pseudo-interstitial and poikilitic textures; Fig. 2b) and at nodule rims (Fig. 2c). Ultimately reaction replacement produces hornblendite with a granoblastic texture, and varying proportions of subhedral, relict clinopyroxene (Fig. 2d,e). The formation of amphibole by meltclinopyroxene reaction is a widely reported and long-recognized phenomenon in cumulates 26,27 , mantle pyroxenites 28 and metasomatized peridotitic mantle 29,30 . Idiomorphic, framework adcumulates ( Fig. 2f) are precipitation cumulates, formed by the nucleation, growth and settling of amphibole from a melt 24 .
The trace element geochemistry (see Supplementary Data 1) of the samples can be used to determine a genetic sequence for these textural classes. Incompatible elements such as Zr and Pb increase in the melt with continued fractionation; later phases should have higher concentrations of these elements. Figure 3 shows that the mineral-textural sequence: clinopyroxene-amphibole blebcrystal margins-granoblastic amphibole is reflected in the trace element geochemistry. Clinopyroxene phenocrysts are generally enriched in incompatible trace elements (average values in mg kg Decoupling of phenocryst and cumulate populations indicates that cumulates form earlier than phenocrysts, and that populations of the latter may not be representative of the former. Despite the implied variation in parent melt, and the different formative processes of amphibole, all populations are capable of contributing to an 'amphibole sponge' La/Yb and Dy/Yb signature (Fig. 4). Normalized rare earth element (REE) profiles (see Fig. 5) show similar forms-albeit at different relative enrichments-for clinopyroxenes and amphiboles of different texture classes.
Volcanic suites with La/Yb, Dy/Yb trends indicative of amphibole fractionation (including Savo; Fig. 4) show no inflections marking an 'amphibole-in' point in the petrogenetic sequence, and the suites show amphibole fractionation patterns from the most mafic parts to the most evolved 1 . Dy/Dy* trends in arc magmas also suggest amphibole as a key fractionating phase, but do not preclude clinopyroxene involvement 31 . Experimental data on hydrous basaltic fractionation [32][33][34] , and the maximum thermal stability of amphibole (B1,100°C; ref. 35) both support the formation of clinopyroxene before amphibole, which, as in this study, develops at the expense of the earlier-formed clinopyroxene (and/or olivine). Progression from clinopyroxene to amphibole fractionation is not marked by inflections in La/Yb and Dy/Yb because: (a) REE ratios for clinopyroxene and hornblende are similar; (b) clinopyroxene fractionation alone has little impact on daughter melt SiO 2 (ref. 1) and hence, trends in SiO 2 space are suppressed; and (c) this study demonstrates that amphibole fractionation may be partial and progressive, which will smooth any abrupt changes in chemistry inferred from a single 'mineral-in' phase boundary. The continuous trend in the Savo La/Yb, Dy/Yb data from B50 wt% SiO 2 (whole rock), as well as those of Davidson et al. 1 , coupled with the relatively limited influence of clinopyroxene on residual melt SiO 2 , means that the melts participating in amphibole-forming melt-mush reactions are expected to be low SiO 2 (basalt or basaltic andesite), with relatively low abundances of incompatible trace elements, albeit higher than the melt before clinopyroxene crystallization (reflected in the crystal chemistry; Fig. 3).
Melts reacting with clinopyroxene need not be co-genetic; clinopyroxene may be formed as a cumulate from earlier magmas, and progressively replaced with amphibole by later melts ascending through the cumulate pile. The formation of an ARTICLE 'amphibole sponge' in the lower crust of arcs can therefore occur by repeated injection of melts into, and reactive transport 36 through hot zone-type systems 37 . The lower thermal stability of amphibole means that these cumulates may be periodically re-melted by the addition of high-temperature (41,110°C) primitive melts, providing a mechanism by which lower crustal sulphides 3 entrained within cumulate assemblages are reworked into ascending magmas and ultimately contribute to metallogenesis. Amphibole-forming reactive transport of a melt through a clinopyroxene mush will impart an amphibole fractionation signature 36 irrespective of amphibole appearance or absence as a phenocryst phase. The absence of amphibole phenocrysts in melts segregated from the mush may be explained by a range of nonexclusive mechanisms. The precursory clinopyroxene mush may already have surpassed the liquid-to-solid transition 38 and be effectively immobilized in the lower crust. Figure 6 shows mechanisms by which reaction-replacement may act to generate cryptic fractionation of amphibole. Path A-D shows a basaltic melt injected into a 25-km-deep crustal hot zone 37 . As basalt cools and crystallizes along path A-B, clinopyroxene and olivine are the dominant silicate phases. At B, amphibole appears at the expense of clinopyroxene 32 . At C, a dacitic liquid with 10 wt% H 2 O is extracted from the hot zone and ascends adiabatically. Along C-D, the liquid is super-liquidus, and any entrained crystals may be resorbed 37 ; this crustal interval will therefore be underrepresented or absent in the crystal record of the magma. At point D, the dacitic melt crosses the liquidus and begins to crystallize, but outside of the amphibole stability field. Hence, the magma has been subject to fractionation at 1.0 GPa (equivalent to the lower crust of a continental arc), with no phenocryst record of amphibole. Path E-F illustrates basalt injection, cooling and crystallization at an initial pressure of 0.3 GPa (comparable to the lower crust of an intra-oceanic arc, and the setting for Savo). Fractionation is olivine and clinopyroxene dominated, with amphibole appearing at the expense of clinopyroxene at F. Basaltic andesite liquids (similar to the most primitive melts bearing amphibole phenocrysts erupted at Savo) extracted from the melt-mush pile at F, ascend adiabatically along F-G. With reference to the basaltic andesite phase fields, the extracted liquid can quickly intersect the liquidus and crystallize amphibole during ascent until relatively low pressures. Thus, the magmas erupted from point G may carry non-resorbed cumulates from F, and will be amphibole-phyric, although phenocrysts may show decompression-related breakdown due to moving out of the stability field of amphibole in the shallow crust 39 . The high Na 2 O content of the Savo magmas 18 may increase the stability field of amphibole at lower pressures 23 .
The onset of amphibole formation in primitive arc magmas is at the expense of clinopyroxene precursors. Where intermediate and more evolved arc magmas are generated at hot zones, amphibole fractionation can be achieved by residual melts reacting with earlier-formed clinopyroxene mushes. Melt segregation in hot zones 38 provides mechanisms by which a largely crystal-free melt is separated from the early-formed clinopyroxene and reactionreplacement amphiboles. Phenocryst formation is decoupled from this process, and as a result, fractionation of amphibole in arc magmas may be cryptic.

Methods
Whole rock chemistry. Whole rock cumulate samples were analysed with a PANalytical Axios Advanced XRF spectrometer at the University of Leicester. Major element analyses determined on fused glass beads prepared from ignited powders using a 1:5 sample to (80% Li metaborate:20% Li tetraborate) flux ratio. Trace elements were determined from 32 mm diameter pressed powder pellets produced by mixing 7 g fine ground sample powder with 12-15 drops of a 7% polyvinyl alcohol solution (Moviol 8-88). A range of reference materials were used to calibrate both instruments. The precision (1s) of the major element data, across a range of compositions, was estimated to be o3% for SiO 2 , Al 2 O 3 , Fe 2 O 3 , MgO, CaO and Na 2 O, o7% for TiO 2 and MnO, and o10% for K 2 O and P 2 O 5 . For trace elements at concentrations above 10 p.p.m. the precision was o15% for Ba; o10% for Ce, Co, Cr, Cu, Ni, Sc; and o5% for Ga, Rb, Sr, V, Zn, Zr. For trace elements at concentrations below 10 p.p.m., the estimated for precision were o1 p.p.m. for Ga, Nb, Rb; o2 p.p.m. for Co, Cu, La, Nd, Ni, Y; and o4 p.p.m. for Ba, Cr, Th, Zr. Measured values for reference materials were within 1s of accepted values.
Additional trace element on whole rock cumulate samples were analysed by solution ICP MS at Durham University, using a Thermo Scientific X-Series 2 ICP-MS, and digestion methods outlined in ref. 40 Statistical errors and detection limits are included in Supplementary Data 1.
Microanalytical chemistry. Microanalytical major element data are from the British Geological Survey's FEI QUANTA 600 environmental SEM, coupled with an Oxford Instruments INCA Energy 450 EDXA system that utilizes an Oxford Instruments X-Max large area (50 mm 2 ) Peltier-cooled silicon-drift detector. Quantitative EDXA analyses were carried out using an accelerating voltage of 20 kV. Reproducibility for the major elements is ±2% of the quoted values, based on repeat spot analysis.
Microanalytical trace element data are by LA-ICP-MS, with a New Wave Research UP-213 laser ablation system with He carrier gas, attached to a Thermo Finnigan Element 2 ICP-MS at the University of Notre Dame's Midwest Isotope and Trace Element Research Analytical Center (MITERAC). The laser was operated at 5 Hz with a 5-mm spot size, with analysis intervals of B30 s background counts, B20 s laser warm-up, 60 s ablation and B45 s washout. Elemental abundances were determined using GLITTER software (XP version, Simon Jackson, Macquarie University) with Ca from EDXA-SEM used as the internal standard and NIST SRM 612 glass used as the external standard. Statistical errors and detection limits are included in Supplementary Data 1.