Sulphide Sinking in Magma Conduits: Evidence from Mafic–Ultramafic Plugs on Rum and the Wider North Atlantic Igneous Province

Ni–Cu–PGE (platinum group element) sulphide mineralization is commonly found in magmatic conduit systems. In many cases the trigger for formation of an immiscible sulphide liquid involves assimilation of S-bearing crustal rocks. Conceptually, the ﬂuid dynamics of sulphide liquid droplets within such conduits is essentially a balance between gravitational sinking and upwards entrainment. Thus, crustal contamination signatures may be present in sulphides preserved both up-and down-ﬂow from the point of interaction with the contaminant. We examine a suite of ultramaﬁc volcanic plugs on the Isle of Rum, Scotland, to decipher controls on sulphide accumulation in near-surface magma conduits intruded into a variable sedimentary stratigraphy. The whole-rock compositions of the plugs broadly overlap with the compositions of ultramaﬁc units within the Rum Layered Complex, although subtle differences between each plug highlight their individuality. Interstitial base metal sulphide minerals occur in all ultramaﬁc plugs on Rum. Sulphide minerals have magmatic d 34 S (ranging from –1 (cid:2) 3 to þ 2 (cid:2) 1 ø ) and S/Se ratios (mean ¼ 2299), and demonstrate that the conduit magmas were already S-saturated. However, two plugs in NW Rum contain substantially coarser (sometimes net-textured) sulphides with unusually light d 34 S (–14 (cid:2) 7 to þ 0 (cid:2) 3 ø ) and elevated S/Se ratios (mean ¼ 4457), not represented by the immediate host-rocks. Based on the Hebrides Basin sedimentary stratigraphy, it is likely that the volcanic con duits would have intruded through a package of Jurassic mudrocks with characteristically light d 34 S (–33 (cid:2) 8 to –14 (cid:2) 7 ø ). We propose that a secondary crustal S contamination event took place at a level above that currently exposed, and that these sulphides sank back to their present position. Modelling suggests that upon the cessation of active magma transport, sulphide liquids could have sunk back through the conduit over a distance of several hundreds of metres, over a period of a few days. This sulphide ‘withdrawal’ process may be observed in other vertical or steeply inclined magma conduits globally; for example, in the macro-dykes of East Greenland. Sulphide liquid sinking within a non-active conduit or during magma ‘suck-back’ may help to explain crustal S-isotopic

H u g h e s , H a n n a h S. R., M c D o n al d, I ai n ORCID: h t t p s://o r ci d.o r g/ 0 0 0 0-0 0 0 1-9 0 6 6-7 2 4 4, Boyc e, Adri a n J., H ol w ell, D. A. a n d Ke rr, An d r e w C r ai g ORCID: h t t p s://o r ci d.o r g/ 0 0 0 0-0 0 0 1-5 5 6 9-4 7 3 0 2 0 1 6. S ul p hi d e si n ki n g in m a g m a c o n d ui t s: e vi d e n c e fro m m afic ul t r a m afic pl u g s o n R u m a n d t h e wi d e r N o r t h Atla n tic Ig n e o u s P r ovi n c e. Jou r n al of P e t r olo gy 5 7 (2) , p p . 3 8 3-4 1 6. 1 0. 1 0 9 3/ p e t r olo gy/ e g w 0 1 0 file P u blis h e r s p a g e : h t t p:// dx. d oi.o r g/ 1 0. 1 0 9 3/ p e t r olo gy/ e g w 0 1 0 < h t t p:// dx. d oi.o r g/ 1 0. 1 0 9 3/ p e t r olo gy/ e g w 0 1 0 > Pl e a s e n o t e: C h a n g e s m a d e a s a r e s ul t of p u blis hi n g p r o c e s s e s s u c h a s c o py-e di ti n g, fo r m a t ti n g a n d 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 .

INTRODUCTION
Upper crustal conduits (pipes, dykes and sills) within mafic-ultramafic magmatic systems provide favourable settings for magmatic sulphide genesis, promoted by crustal contamination, resulting in some of the world's largest orthomagmatic Ni-Cu and platinum group element (PGE) deposits. Contamination can trigger S-saturation of the magma, particularly if the country rocks contain significant amounts of S in the form of sulphides or sulphates; this can be recognized by S isotope signatures outside the typical mantle range of 0 6 2ø (e.g. Ohmoto & Rye, 1979;versus Li et al., 2002;Ripley et al., 2003;Keays & Lightfoot, 2010;Hayes et al., 2015;Smith et al., 2016). In addition, sulphide liquid that ponds within conduits can become enriched in chalcophile elements, possibly to economic levels, as a result of interaction with magma continuing to pass through the conduit (e.g. Kerr & Leitch, 2005;. Sulphides within conduits commonly occur as massive to semi-massive accumulations and are present as rounded globules or droplets of varying sizes; for example, at Norilsk-Talnakh, Russia (Czamanske et al., 1992), the Insizwa Complex, South Africa (Lightfoot et al., 1984), Voisey's Bay, Canada (Huminicki et al., 2008), a mafic dyke in Uruguay (Prichard et al., 2004) and in East Greenland macrodykes (Holwell et al., 2012). Droplets are generally interpreted to represent sulphide liquid that has been transported upwards, downwards or laterally along the magmatic plumbing system. The fluid dynamics involved in this transport, such as settling rate versus entrainment rate, has been the subject of many recent studies based on both empirical and experimental evidence (e.g. de Bremond d'Ars et al., 2001;Mungall & Su, 2005;Godel et al., 2006Godel et al., , 2013Barnes et al., 2008;Chung & Mungall, 2009;Arndt, 2011).
Sulphide-and chromitite-bearing ultramafic units hosting platinum group minerals (PGM) are documented within the British Paleogene Igneous Province (BPIP) on the islands of Skye, Mull and Rum (Hulbert et al., 1992;Butcher et al., 1999;Pirrie et al., 2000;Power et al., 2000;Prout et al., 2002) and highlight the potential for PGE mineralization in the Scottish portion of the North Atlantic Igneous Province (NAIP; Andersen et al., 2002). This is especially the case considering the potential for contamination offered by the thick crustal sequences of Mesozoic S-bearing rocks in the region (e.g. Hesselbo & Coe, 2000;Yallup et al., 2013;Hughes et al., 2015). Recent studies on Skye demonstrate the involvement of sediments, with light S isotope tenors, in triggering S-saturation in ascending BPIP magmas, particularly in vertical dyke swarms (Hughes et al., 2015). On Rum, ultramafic, vertical or steeply inclined volcanic plugs (pipes) intrude a variety of country rocks, including sediments at the base of the Mesozoic Hebrides Basin. Power et al. (2003) identified a crustal d 34 S signature for intercumulus (net-textured) sulphides in one volcanic plug in NW Rum, but could not establish whether the sulphide liquids that formed these had been entrained upwards or had sunk through the conduit.
In this study, we use the whole-rock S-isotope composition, and sulphide-specific (in situ) S-isotope and trace element compositions, of ultramafic plugs on Rum to constrain if crustal contamination took place. The size range and textural associations, trace element composition of sulphide minerals and S-isotopic composition are established for 12 plugs on the island to understand the provenance of sulphide liquids in these conduits. We use these data to decipher controls on sulphide accumulation and compare plug sulphide compositions with those from peridotitic cyclic units of the Eastern Layered Series, part of the Rum Layered Suite. Where crustal contamination is evident [e.g. in NW Rum, as first discovered by Power et al. (2003)], these data are combined with numerical modelling of settling and entrainment rates to investigate the roles of magma entrainment and gravitational settling and sinking of sulphide liquids within conduit systems and to assess the distances that sulphide liquids could sink and over what timescale. We provide new constraints on the development of orthomagmatic Ni-Cu-PGE mineralization in small BPIP intrusions and discuss these in relation to wider processes of sulphide migration within magmatic systems.

REGIONAL GEOLOGICAL SETTING
The North Atlantic Igneous Province (NAIP) formed after the impingement of the Icelandic mantle plume under thick continental lithosphere, during a period of continental flood basalt magmatism (e.g. Saunders et al., 1997). Magmatism initiated c. 62 Ma within what is now the UK, Greenland and Baffin Island, and ultimately led to the opening of the Atlantic Ocean (Saunders et al., 1997). The main products of this prolonged period of magmatism were tholeiitic and alkali basalts. The British Palaeogene Igneous Province (BPIP) is part of the earliest magmatic series of the NAIP, which includes Palaeogene rocks in the Hebrides and along the west coast of Scotland and Northern Ireland (Saunders et al., 1997;Fig. 1).
In Scotland, the BPIP includes magmatic rocks on the Isles of Mull, Skye and Arran, the Small Isles (Rum, Eigg, Muck, Canna and Sanday), and on the mainland at Ardnamurchan and Morvern (Emeleus & Bell, 2005). The Scottish BPIP extends across a number of Archaean and Proterozoic basement terranes (e.g. Emeleus & Bell, 2005), covered by Neoproterozoic (Dalradian and Moine) metasediments (Kinnaird et al., 2007), which themselves are overlain by Mesozoic sediments from the Hebrides Basin (Hesselbo & Coe, 2000). Volcanic plugs occur within or close to lava fields and central intrusive complexes, and are primarily exposed on the Isles of Mull and Rum (Emeleus & Bell, 2005). Emeleus & Bell (2005) provided a detailed review of radiometric and palynological age determinations of various intrusions and lavas across the BPIP.

Geology of the Isle of Rum
The Rum Central Complex straddles the Long Loch Fault (Fig. 1), which probably acted as a crustal lineament that facilitated magma ascent (e.g. Upton et al., 2002). The complex developed during at least three magmatic phases, forming in < 500 kyr (Emeleus & Bell, 2005;Troll et al., 2008).
In Phase 1, silicic magmatism occurred along an arcuate fault system (the Main Ring Fault) at 60Á33 6 0Á21 Ma  followed rapidly by Phase 2, which formed the mafic-ultramafic Rum Layered Suite (60Á53 6 0Á04 Ma; Hamilton et al., 1998). Ultramafic magmas were intruded into the complex as multiple intrusions and replenishment of the chamber with picritic magma is thought to have disrupted previously formed cumulate layers (e.g. Bé dard et al., 1988). The mafic-ultramafic Rum Layered Suite can be divided into the Western, Central and Eastern Layered Series, bounded by the Main Ring Fault (Fig. 1). Rare late-stage picrite dykes [described by Upton et al. (2002)] may represent feeders to the layered suites. However, the small melt fraction (6-7%) inferred to have formed these mildly alkalic dykes and two olivine phenocryst populations observed within single picrite dykes, such as M9 (e.g. Upton et al., 2002), preclude these feeders from representing a parental magma composition. During Phase 3, deep subaerial erosion and unroofing of the complex followed caldera collapse and intermittent burial by basalt flows from the Skye lava field (Chambers et al., 2005).

Ultramafic plugs of the Isle of Rum
In excess of 40 volcanic plugs occur on Rum, comprising gabbros and feldspathic peridotite intrusions (senso stricto olivine melagabbros) probably formed during Phase 2. Plugs predominantly intrude through Torridonian and Triassic sediments, as well as portions of the Rum Layered Suite (McClurg, 1982;Wadsworth, 1994;Holness, 1999). The sedimentary host-rocks of the volcanic plugs (Triassic,Fig. 2a;and Torridonian, typically appear bleached, fissile and metasomatized. Locality details of the plugs sampled during this study are given in Table 1 (also see Fig. 2a-c).
The plugs have rounded or elongate-to teardropshaped outlines (e.g. Fig. 2a). Their plan-view shapes probably result from varying inclinations of conduits intersecting the present-day topography. Plugs sampled during this investigation are distinct from the small 'apophyses' of peridotite within, or straddling, the Main Ring Fault that may predate or be contiguous with the Eastern Layered Series (e.g. Fig. 2b; Butcher et al., 1985;Wadsworth, 1994). None of the plugs have been radiometrically dated, and the timing of their intrusion is based upon cross-cutting field relationships alone. Most of the plugs lie within 1-2 km of the Long Loch Fault (Fig. 1) and range in size from $100 to 600 m diameter (McClurg, 1982;Volker & Upton, 1990). The plugs may have fed lava flows, but no associated lavas have been proven (Emeleus & Bell, 2005).
The structural setting and silicate mineralogy of volcanic plugs on Rum has been studied in detail by Volker & Upton (1990), Wadsworth (1994), Holness (1999) and Holness et al. (2012). Ultramafic plugs are thought to represent the conduits of olivine-rich melts or 'crystal mushes', fed from an underlying magma chamber along faults such as the Long Loch Fault. These crystal mushes were lubricated by interstitial basaltic magma, hence the contemporaneous series of gabbroic plugs on the island could represent similar conduits without the entrained olivine mush. Rare plugs in the south of Rum contain rounded troctolite, dunite and chromitite xenoliths, probably derived from part of the Rum Layered Suite (Volker & Upton, 1990).
Plagioclase (10-15 modal %) and clinopyroxene (up to 8%, sometimes poikilitic) are interstitial to cumulus olivine and euhedral and rounded Cr-spinel; accessory amphibole and biotite are also observed (Wadsworth, 1994). Layering (defined by olivine cumulates) is recorded in the ultramafic peridotite plugs and typically dips shallowly at the margins (striking parallel to the margins) with the central portions of plugs taking on a colloform texture. Changes in modal mineralogy can also define layering, now weathered to 'indentation layering'. At the West Sgaorishal plug (Fig. 2a, Plug 1) indentation layering is particularly well developed and is observed to be closely spaced at the margins (5-10 cm), dipping inwards towards the plug centre at 30-50 , with layer spacing increasing towards, and ultimately disappearing, at the centre. In this plug, olivine accounts for 55-60 vol. % of the rock, but olivine abundance increases to 88 vol. % at the centre. In detail, there is a marked asymmetry in the olivine content from west to east, such that olivine is more concentrated on the western edge (see Holness et al., 2012). In contrast, layering can be poorly developed, laterally discontinuous (e.g. Loch Sgaorishal plug; Fig. 2a, Plug 2) or completely absent in other plugs. Overall, layering developed after magma flow had ceased and/or during magma withdrawal back into the feeder system [compare 'withdrawal' of magmas in Hawaiian conduits, as suggested by Upton (2004)].
Olivine compositions in plugs are equivalent to those from feldspathic peridotites in the Eastern Layered Series (Fo 89-82 ) and Western Layered Series (Wadsworth, 1994;Emeleus, 1997, and references therein;Holness et al., 2012) and distinct from olivine compositions in the troctolite portions of the Rum Layered Suite (Fo 84-70 ; Emeleus, 1997, and references therein).
The sedimentary host-rocks of the volcanic plugs (Triassic,Fig 2a;Torridonian, typically appear bleached, fissile and metasomatized. Studies of in situ anatexis of country rocks (e.g. Holness, 1999;Holness et al., 2012) and thermal modelling of the West Sgaorishal plug (Plug 1; Fig. 2a) suggest that the plugs were active magma conduit for only a few months and cooled over a period of c. 30 years (Holness et al., 2012).

The Hebridean sedimentary succession and country rocks to the Rum plugs
The crustal sequence on the Isle of Rum includes Archaean basement rocks of the Lewisian Gneiss Complex, Mesoproterozoic Torridonian sediments, and Triassic sediments in NW Rum. A thick sequence of Jurassic and Cretaceous sediments probably also covered the island (at least in part), as suggested by extensive outcrops of these rocks preserved on the neighbouring Isle of Skye and 'wedges' of Jurassic mudrocks preserved in the Main Ring Fault of the Rum Central Complex (Emeleus, 1997; for stratigraphic log see Hughes et al., 2015).
The Mesoproterozoic Torridonian formations of NW Scotland include the Stoer, Sleat and Torridon Groups (Stewart, 2002;Kinnaird et al., 2007); however, the Stoer Group is thought to be absent on the Isles of Skye and Rum (Figs 1 and 2). Mudrocks are present within the Sleat (and, to a lesser extent, Torridon) Group; however, they are less extensive and thinner than the Jurassic mudrocks of the Hebrides Basin. Sulphur isotope studies of Torridonian sulphates and sulphides in Stoer Group mudrocks have identified the light isotopic shift of d 34 S associated with bacterial sulphate reduction in small freshwater pools (Parnell et al., 2010, Fig. 2. Geology of areas surrounding the peridotite and gabbro plugs sampled during this study. (a) NW Rum; (b) central Rum across the Long Loch Fault; (c) SE Rum near Dibidil. Sample locations are shown as dots. Plugs are numbered and correspond to Table 1. Petrology, 2016, Vol. 57, No. 2 Hughes et al. (2015) established a whole-rock d 34 S range of þ1Á4t oþ4Á7ø for Mesoproterozoic Torridonian sediments on Skye and Rum. They also found that these rocks contained low concentrations of sulphur (<0Á018 wt %). The stratigraphy of the Triassic to Cretaceous rocks of the Hebrides Basin is well correlated (e.g. Morton & Hudson, 1995;Hesselbo & Coe, 2000, and references therein) and the S concentration of this Mesozoic sequence is generally much greater than that of the older Torridonian sediments, Moinian metasediments, or crystalline basement (Hughes et al., 2015). Therefore the Mesozoic Hebrides Basin represents the most effective source of crustal S available for magmatic contamination. Sulphur isotopes for Mesozoic (specifically Jurassic) sediments have been reported by Raiswell et al. (1993) and Hudson et al. (2001) from elsewhere in the British Isles, but Hughes et al. (2015), together with results from Yallup et al. (2013), have provided the first detailed Hebridean S-isotopic framework.

Journal of
In summary, Jurassic mudrocks and ironstones have the highest S abundance in the Hebrides Basin, with d 34 S ranging from -33Á8 to -14Á7ø (Hughes et al., 2015). d 34 S generally becomes lighter from the Lower Jurassic (-18Á4 to -14Á7ø) to the Upper Jurassic (-33Á8t o-29Á2ø), although the ironstones in the Lower Jurassic (e.g. Raasay Ironstone Formation) generally have the highest bulk S content (up to c. 28 wt %, in comparison with some Upper Jurassic shales with 0Á1 wt %). In contrast, sediments of the Triassic and Cretaceous succession have volumetrically very few mudrocks and no ironstones, and are therefore less likely to act as major crustal S sources (Hughes et al., 2015). Triassic sediments from Rum (sandstones, conglomerates, breccia and cornstone of the Monadh Dubh Formation) and the wider Hebrides Basin contain calcretes-thus if S were to be present in these rocks, it would most probably be present as sulphate with a heavy d 34 S composition (e.g. Power et al., 2003).

Magmatic sulphide mineralization and S contamination in the NAIP
On the Isle of Rum, disseminated and net-textured coarse sulphides (up to 5% modal abundance; Power et al., 2003) occur in the West Sgaorishal peridotite plug (Plug 1; Fig. 2a) within a few metres of the margin. Sulphides are most abundant at the plug margin, and become increasingly rare and fine grained towards the centre (Power et al., 2003). Sulphide-rich zones have not been identified in the other plugs on Rum. PGM are associated with sulphides in Plug 1 and include Pt-and Pd-tellurides, bismuth-tellurides and electrum (Power et al., 2003). Previously reported d 34 S values of the coarse sulphides in Plug 1 range from -18Á3t o-9 Á2ø, and suggest that the magmas forming the plug experienced crustal contamination by Jurassic sedimentary rocks (Power et al., 2003). However, given the absence of crustal rocks with low d 34 S currently surrounding the plug, Power et al. (2003) suggested that either S contamination took place in a now-eroded parental magma chamber, or Jurassic sediments originally surrounded the plug locally (and have also since been eroded). Hughes et al. (2015) found that for steeply dipping intrusions, such as basaltic dykes, incorporation of bacterially reduced sulphur from wall-rock shales produced d 34 S signatures in magmatic sulphides ranging from -30Á7t o-2 Á3ø, depending on the degree of contamination and the d 34 S signature of the local crustal contaminant. In contrast, horizontal picritic sills on Skye do not record such crustal contamination and lack the anomalous d 34 S ( Hughes et al., 2015), indicating that the mechanism by which a conduit intruded the surrounding country rock (vertical or horizontal) and the ambient flow regime within it (i.e. turbulent or nonturbulent) were important factors in the contamination process; for example, a turbulent regime is more likely to result in brecciation of the conduit wall rocks, allowing for more efficient S incorporation into the magma (Robertson et al., 2015).
Magmatic Cu-Ni-PGE-Au sulphide mineralization in the wider NAIP is recorded in a number of settings. The largest deposit is the Au-and Pd-dominant Platinova Reef in the Skaergaard Intrusion, formed through prolonged fractional crystallization that eventually triggered S-saturation (e.g. Bird et al., 1991;Andersen et al., 1998;. In this instance, crustal contamination played no role in the formation of an immiscible sulphide liquid. Mineralization is also found as sulphide globules in small hyperbyssal intrusions, such as the Miki Fjord and Togeda Macrodykes of East Greenland (Holwell et al., 2012). Sulphur isotope analyses identified a strong crustal contribution attributed to pyritic Cretaceous black shales, and therefore in these instances crustal S contamination was the trigger for S-saturation (Holwell et al., 2012).
Overall, if crustal contamination is suspected to be the main cause for S-saturation, the isotopic composition of disseminated sulphide mineralization in magma conduits, such as sills, dykes and plugs, can be used as a vectoring tool towards areas of more massive sulphide mineralization (e.g. Lightfoot et al., 1984;Czamanske et al., 1992;Huminicki et al., 2008;Barnes et al., 2016). Hence understanding the source and location of crustal contamination, and the potential crustal horizon(s) that triggered S-saturation in relation to the prevailing fluid dynamic regime in the magmatic conduit, is a valuable method for targeting sulphide-hosted mineralization up-or downstream within the magmatic plumbing systems.

METHODS
Twenty-six whole-rock samples from 12 volcanic plugs from across the Isle of Rum were sampled for major and trace elements, S isotopes (bulk and in situ), and mineralogical investigations (sample details are given in Table 1; localities are shown in Fig. 2a-c; bulk-rock geochemical data are presented in the Supplementary Material, available for downloading at http://www.pet rology.oxfordjournals.org). Additionally, 15 samples of various Torridonian and Triassic sediments were collected from areas adjacent to the plugs, and elsewhere across the island, to assess the crustal S isotopic framework of the island (Table 1).

Bulk-rock and in situ S-isotope analysis
For samples with visible sulphide minerals >500 mm 2 , polished blocks were subjected to sulphide in situ laser combustion, following Wagner et al. (2002). Raw in situ d 34 S data were corrected by the factor d 34 S pyrite ¼ d 34 S SO2laser þ 0Á8ø, as laser combustion causes a small and predictable fractionation of sulphur isotope compositions for the SO 2 gas produced, compared with the actual d 34 S of the sulphide mineral (Wagner et al., 2002). At the Scottish Universities Environment Research Centre (SUERC), SO 2 gases released by combustion were analysed on a ThermoFisher Scientific MAT 253 dual inlet mass spectrometer (conventional samples) and VG Isotech SIRA II mass spectrometer (for laser samples).
Samples without sufficiently large sulphides underwent whole-rock chemical S extraction by the chromium reducible sulphur method (CRS) followed by conventional sulphur analysis: a detailed description and evaluation of this method has been given by Hughes et al. (2015). In summary, CRS is based on and adapted from the techniques of Zhabina & Volkov (1978), Canfield et al. (1986), Tuttle et al. (1986), Hall et al. (1988), Newton et al. (1995), Nielsen & Hanken (2002) and Labidi et al. (2012). Powdered rock samples were loaded into a flat-bottomed glass reaction vessel under a nitrogen atmosphere and connected to a gas bell jar containing 0Á1 M AgNO 3 solution. A 1M solution of chromous chloride (CrCl 2 ) was injected through a septum to cover the powdered sample and was heated to 150 C for 3 h to react as follows: This process leads to the complete breakdown of elemental and reduced sulphur species into H 2 S, even in the presence of Fe 3þ .H 2 S produced is streamed to the gas bell jar to react with AgNO 3 solution, precipitating Ag 2 S. At the end of the reaction, this black precipitate was filtered under vacuum, freeze dried and collected ready for conventional S analysis. During conventional analysis, SO 2 gas samples were run at SUERC using a ThermoFisher Scientific MAT 253 dual inlet mass spectrometer.
Bulk-rock S concentrations were determined using a LECO CS230 Carbon/Sulphur Determinator at the University of Leicester. Sample mass ranged from 0Á1 to 1Á0 g depending on relative bulk S content. Each sample was run in triplicate to monitor precision. Accuracy was monitored by regular analysis of the reference material BAS ECRM 877-1. The limit of minimum detection for this method is 0Á018 wt % S (3 Â standard deviation of the mean blank).

Mineralogy and mineral composition
Quantitative microanalysis was carried out on a Cambridge Instruments S360 scanning election microscope (SEM), using an Oxford Instruments INCA Energy energy-dispersive X-ray spectroscopy (EDX) analyser, with operating conditions set at 20 kV and specimen calibration current of $2 nA at a fixed working distance of 25 mm. Analytical drift checks were carried out every 2 h using a Co reference standard and a comprehensive suite of standards from MicroAnalysis Consultants Ltd were used to calibrate the EDX analyser.
Sulphide trace element concentrations were determined by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) using polished samples and a New Wave Research UP213 UV laser system attached to a Thermo X Series 2 ICP-MS system, following procedures outlined by Prichard et al. (2013) and Smith et al. (2014). Further details are available in the Supplementary Material.
As part of our study, we also analysed the sulphide compositions of a selection of peridotites from each of the major cyclic units in the Eastern Layered Series (ELS), to place plug compositions into context. The results are presented in Table C

Sulphide mineralogy and trace element compositions
Details of the mineralogy, including size, texture and mineral association of base metal sulphides in the Rum plugs are presented in Table 2 together with a record of any PGM present. SEM-based PGM searches were carried out only on samples from Plugs 1, 2 and 10, as previous work had identified a correspondence between sulphide and PGM occurrences (e.g. Power et al., 2003). As our study focuses on the S-isotopic composition of the plugs and trace element compositions of base metal sulphides within the plugs, we did not undertake a comprehensive search for PGM across all plug samples. Selected back-scattered electron (BSE) images of sulphide mineral textures are shown in Fig. 3.
With the exception of plugs 1 and 2, all the peridotite plugs contain small sulphide blebs (e.g. Fig. 3a),     (Table 2). Occasionally a rim of iron oxide is present around the sulphide blebs in more altered samples (e.g. Fig. 3b). Sulphides normally occur at the margins of cumulus olivine ( Fig. 3a and c), or within intercumulus plagioclase and clinopyroxene, and sulphides are locally in contact with granular Cr-spinel (e.g. Fig. 3d). In rare cases, sulphides occur within the fractured and serpentine-filled margins of cumulus olivine (e.g. Fig. 3e). In contrast, plug 1 at West Sgaorishal contains coarser sulphide minerals, up to 4 mm in length, which occur interstitially to pyroxene and plagioclase, or as net-textured assemblages ( Fig. 3f-h; Table 2). Sulphides have been observed to almost enclose cumulus olivine, as well as ilmenite and Cr-spinel (e.g. Fig. 3g and h), and include pyrrhotite and pentlandite (and pyrrhotite-pentlandite flames) as well as chalcopyrite. Minor galena is also observed; however, unlike the other plugs, no bornite or chalcocite was encountered in plug 1. Finer sulphide minerals are of the same species as the course sulphides occurring in the centre of plug 1. PGM (plugs 1 and 2 only) are found within sulphide minerals or at silicate-sulphide mineral contacts. These range in size from 0Á5t o7mm wide and include Pd-and Pt-tellurides, bismuth-tellurides, Pt-arsenides and Pt-Sbarsenides (Table 2). In plug 2 (Loch Sgaorishal) sulphides range from 50 to 500 mm and are interstitial to cumulus olivine, with rare rounded sulphides ($50 mm) occurring as inclusions within cumulus olivine, albeit in serpentinefilled fractures. Pentlandite in plug 2 contains small (<1 mm wide) Ir and Pt sulpharsenides of the irarsiteplatarsite series. In both plugs 1 and 2, small grains of electrum were observed in pyrrhotite (normally <5 mm) as well as one crystal of Pb-telluride (Table 2).
Representative trace element compositions, including S/Se ratios of sulphide minerals and PGE abundances in plug sulphides (grouped per region of Rum), are presented in Table 3 (full data are available in  Supplementary Material Tables C and D). Overall, the variation of PGE concentrations in sulphide minerals from plugs in various areas of Rum (NW, SE and central) overlaps significantly (Fig. 4a-d), such that there is no systematic difference in PGE tenor between plugs. Pentlandite is the principal carrier of Pd but there is no correlation between Pd and Rh for any of the plugs (Fig. 4a), nor are Rh concentrations higher in pentlandite or correlated with Ni content (Fig. 4b). By contrast, Rh and IPGE (Os, Ir, Ru) abundances in pentlandite are not distinct from other base metal sulphides or 'mixed sulphide' analyses ( Figs. 5b and d).
The chondrite-normalized PGE composition of coarse sulphides at the margins of plug 1 (Fig. 5a-c) is subtly different from the composition of fine sulphides at the centre (Fig. 5d). This difference is particularly noticeable for Os, Ir and Ru, with the concentration of all PGE in marginal sulphides being lower than in finer sulphides at the centre. Overall, for each type of base metal sulphide (chalcopyrite, pentlandite, pyrrhotite), the sulphide PGE patterns of plug 1 (particularly the margins) are fairly consistent and parallel to one another. However, chalcopyrite (Fig. 5a) generally has lower PGE abundances than pentlandite and pyrrhotite ( Fig. 5b and c, respectively), but the overall normalized PGE spectra are similar or parallel between Ni-Fe mineral species (e.g. Fig. 5b and c). Sulphides in plug 2 (Fig. 5e) also have two distinctive PGE patterns, with sulphides at the margin being comparatively homogeneous, and of a similar composition to those from the margin of plug 1. However, two sulphides analysed from the centre of plug 2 have differing PGE compositions-one that overlaps the sulphide compositions of the margins, and one entirely different sulphide composition with notably elevated IPGE. In the instance of plug 2, this may be due to a fundamental difference in sulphide mineralogy, as the sulphide analysed in sample RM35 has a notably lower Cu content (Table 3).
Overall, sulphide compositions from plugs 4-12 are heterogeneous (Fig. 6). This may in part be explained by their being analyses of 'mixed sulphide' species (owing to their small size) but this cannot account for all the observed variation, as even sulphides with comparable Fe, Cu and/or Ni contents fluctuate between plugs. We note that whole-rock geochemistry is similarly variable between these plugs (albeit subtly) (see Supplementary Material). Broadly, peridotite plug sulphide compositions resemble those from peridotite units in the Eastern Layered Series (Fig. 6).
S-isotope compositions and S/Se ratios of the Rum plugs Figure 7a shows the average d 34 S value for whole-rock samples per plug (data in Table 4, along with S wt %) and for the ELS, plotted against S/Se ratio per sulphide. S/Se was calculated from LA-ICP-MS analyses of Se and calibrated by prior quantitative SEM analyses of S from the same sulphide grain (or in cases of small mixed sulphides a mean stoichiometric S abundance was assumed) and therefore quantifies S/Se on a mineral-by-mineral basis per sample. Sulphide minerals large enough to be analysed by laser combustion (i.e. only in plug 1) on a sulphide mineral-by-mineral basis (Table 5) have been averaged to the mean d 34 S value serpentinized fracture at margin of cumulus olivine, with intercumulus clinopyroxene, plug 5. (f) Coarse sulphides as a poikilitic phase with clinopyroxene and plagioclase, plug 1 margin. (g) Poikilitic sulphides surrounding cumulus olivine and elongate ilmenite, with intercumulus plagioclase, plug 1 margin. (h) Similar to (g), demonstrating poikilitic, sometimes 'net-textured' sulphides surrounding granular cumulus olivine phenocrysts. In both (g) and (h) the contrast has been adjusted to display the sulphide internal texture and mineralogy. A summary of sulphide characteristics and PGM is presented in  Hughes et al. (2015) for various sediments on the isles of Rum and Skye are also shown in Fig. 7b.
The S/Se ratios of ELS samples are mostly outside the mantle sulphide range (2550-3750; Lorand et al., 2003) with only 8% of analyses falling within this range, 84% of analyses being below mantle values (see Supplementary Material Table D). The full range of ELS S/Se ratios is 489-5416 (Fig. 7a) and d 34 S ranges from -4Á4t oþ1Á2ø. Excluding plug 1 and one anomalous value from plug 8, the S/Se ratios of plug sulphides are up to 4000. With the exception of plugs 1 and 2, the peridotite plugs have d 34 S ranging from -0Á6t oþ2Á1ø,i n comparison with the ELS peridotites, which range from  -1Á0t o0 Á9ø. Thus, most of the peridotite plug sulphides have d 34 S and S/Se ratios comparable with, and within range of, sulphides from ELS peridotite units. In contrast, sulphides from plug 1 have extremely light d 34 S, ranging from -14Á7to-8Á0ø, and S/Se ratios ranging from 3000 to 12 000. The centre of Plug 1 shows least variation and deviation of S/Se from the ELS; however, d 34 S is -10Á9ø. Whereas the S/Se ratios of plug 2 are within range of the ELS, the d 34 S composition ranges from -6Á5to-2Á9ø ( Fig. 7a and Table 4).

Magmatic context of the Rum plugs in relation to the Rum Central Complex
The peridotite plugs included in this study have wholerock major and trace element and PGE signatures that are subtly distinctive from one another (Supplementary  Material Table B) but, broadly, the range of major element compositions overlaps with those recorded for the various portions of the Rum Layered Suite, including peridotite units from the Eastern, Western and Central Layered Series (see Emeleus, 1997;O'Driscoll et al.,2009, and references therein) and the M9 picritic dyke (Upton et al.,2002). The Layered Suites on Rum are the result of multiple injections of ultramafic magma, in an open magmatic system (e.g. Wager et al., 1960). Current consensus postulates that layering resulted from interaction between cumulates and later magmatic pulses derived from subsequent intruding picrites (e.g. Bé dard et al., 1988;Volker & Upton, 1990;Emeleus, 1997;Holness, 2005;O'Driscoll et al., 2007;Leuthold et al., 2014). The S/Se data for the ELS display a broad range from 489 to 5416 ( Fig. 7 and Supplementary Material Table D). Open circles represent pentlandite where it could visually be recognized and targeted by LA-ICP-MS. This serves to highlight PGE vs Ni abundance for known pentlandite, for comparison with fine sulphide grains where analysis by individual base metal sulphide was not possible, resulting in a 'mixed sulphide' analysis. Grey shaded area in (b)-(d) delineates the pentlandite field (based on a minimum of 17 wt % Ni, assuming a 50:50 mix between pentlandite and pyrrhotite). Note that where PGE abundances were less than the detection limit, results are calculated as half the detection limit and delineated by a small downward arrow.
Mantle S/Se ratios typically range from 2550 to 3750 (Lorand et al., 2003) and although most of the ELS data are within this range there are significant deviations. Loss of S via oxidation of sulphide minerals will lower the primary S/Se ratio. This is also consistent with the common observation of partially oxidized sulphide minerals in thin sections of the ELS; we suggest that this is the cause of ELS S/Se ratios <2000. In contrast, the causes of an increase in S/Se ratio (i.e. >3750) are two-fold: this can be the result of either hydrothermal mobilization of S, forming a series of hydrothermal sulphide minerals, or contamination of the magmas by  Table 3 for summary of data.) It should be noted that where PGE abundances were less than the detection limit, results are calculated as half the detection limit and delineated by a small downward arrow. Grey shaded areas show the range of compositions for chalcopyrite (cp), pentlandite (pn) and pyrrhotite (po) measured in the Eastern Layered Series (ELS; see Supplementary Data Table D) for comparison in plots (a)-(c). Dashed line shows the mean 'mixed sulphide' composition from the centre of plug 1. In plots (d) and (e) the grey shaded area delineates ELS, pn only.  Table 3 for details.) It should be noted that where PGE abundances were less than detection limit, results are calculated as half detection limit with small downward arrow. Grey shaded areas delineate range of compositions for pentlandite (pn) in the Eastern Layered Series (ELS) for comparison. crustal S (which is devoid of Se, and thus has a high S/ Se ratio). Although d 34 S is commonly used as a proxy for crustal S contamination, recent studies suggest that with enough magma throughput in the magmatic (conduit) system, the 'S signal' can be swamped by magmatic d 34 S (-2Á0t oþ0Á4ø for the Iceland plume; Torssander, 1989; and -2Á3 6 1Á5ø for the Isle of Skye picrites; Hughes et al., 2015), whereas the S/Se ratio remains a more reliable proxy for crustal contamination (Ihlenfeld & Keays, 2011).
We record a minimum in d 34 S locally in unit 1 (ELS) of -4Á4ø, but earlier studies have reported d 34 S as low as -14Á8ø in ELS unit 1 rocks with sulphide concentrations locally up to 3 modal % (Faithfull, 1985;Hulbert et al., 1992). These lighter d 34 S values occur strictly within unit 1 rocks in the vicinity of metre-scale rafts of Jurassic mudrocks (Faithfull, 1985). In this study, we did not obtain a d 34 S composition for the Jurassic mudrock rafts, but we see no reason to disagree with the inference made by Hulbert et al. (1992). In general, Hebridean Jurassic mudrocks, including those in proximity to dolerite and picrite sills on Skye, have a similarly light S-isotopic signature, although those adjacent to intrusions have been significantly degassed owing to contact metamorphism and devolatilization, thereby reducing their bulk S content (Yallup et al., 2013;Hughes et al., 2015). In higher cyclic units in the ELS, d 34 S falls within the typical magmatic range; however, sporadic increases in S/Se ratio (>4350) may indicate localized crustal S contamination. This probably took place during intrusion of magmas into the layered suite, and may not be recorded in the d 34 S either because of the contaminant having an S-isotopic composition similar to the magma or because high volumes of magma throughput have obliterated the signal. Ultimately, the complexities of sulphide composition, S/ Se ratio and S-isotopic composition throughout the ELS and Rum Layered Suite as a whole require further investigation, which is beyond the scope of this study.
The diversity of S/Se ratios and PGE compositions measured in ELS sulphides from our study overlaps with sulphide compositions from the plugs. Therefore  Table E). (b) Histogram of d 34 S for Proterozoic (Torridonian from Rum and Skye) and Mesozoic (specifically Jurassic mudrocks and ironstones from Skye and Ardnamurchan) sediments present in this area of the Hebrides; data from Hughes et al. (2015). we suggest that the plugs (conduits) record various crystal mush injection events broadly coincident with magma (or mush) injections into the Rum Layered Suite. As such, the magma conduits represented by the peridotite plugs were not formed from a single eruption event, and probably represent multiple eruptions.

Feeders and magma staging chamber for the plugs and Rum Layered Suite
The presence of dunite, chromitite, and troctolite xenoliths within peridotite plugs situated inside or close to the Main Ring Fault (Volker & Upton, 1990) suggests that ejection of crystal mushes that fed these plugs occurred whilst crystallization of the layered series was taking place. This demands a physical brecciation of wall-rock material, and does not provide evidence that the conduit crystal mushes were derived from the layered suites directly, but instead indicates that they erupted through them. One striking observation regarding the Rum plugs is the lack of troctolite plugs. In contrast, feldspathic peridotites (strictly olivine melagabbros) described in our study exist alongside gabbroic plugs that have no phenocryst assemblage or other evidence for their having been intruded as a crystal mush, unlike the peridotite plugs. It is possible that the gabbroic plugs represent the basaltic magmas ejected from a staging magma chamber, and there is no evidence to suggest that these gabbroic plugs correlate strictly with the troctolite portions of units in the Rum Layered Suite. Indeed, per cyclic unit in the Rum Central Complex, the peridotite base and troctolite tops are typically paired in the Eastern Layered Series (Emeleus & Troll, 2011, and references therein;O'Driscoll et al., 2014). Regardless of the mechanism that formed the overall layering of this suite, it is apparent that each of the troctolite layers is a fractionation feature associated and paired with each peridotite layer. Therefore because the peridotitic crystal mush that was ejected through the volcanic pipes is of a comparable composition to the peridotitic bases of the layered suite cyclic units, this suggests that the crystal mushes of both came from a common source, and not necessarily that the layered suites directly fed the plugs themselves. This is supported by the observation that small entrained sulphide minerals in both the plugs and layered suite peridotites are directly comparable with one another (according to both S/Se and d 34 S, except for plugs 1 and 2, which are discussed below). In other words, there must have been a separate deeper staging chamber in which a peridotitic crystal mush evolved, before being tapped by the magmatic plumbing system to both the layered suites and the plugs.

The origin of sulphides within the Rum plugs
With the exception of plugs 1 and 2, the sulphide minerals observed are of low abundance and are small, often rounded phases (<150 mm in diameter); sometimes these sulphides occur as embayments within cumulus olivine (e.g. Fig. 3c and e, and Table 2). All plugs have distinctive chondrite-normalized PGE and Au patterns and S/Se ratios, some of which closely resemble the compositions of sulphides in peridotite units of the ELS (e.g. Figs 6-8). The Fo content of cumulus olivine in the plugs suggests that previously crystallized olivine phenocrysts were entrained within a basaltic or picritic magma, such that the peridotite plugs were essentially crystal mush conduits (e.g. Wadsworth, 1994;Holness et al., 2012) sometimes with entrained chromitite, dunite or troctolite xenoliths (e.g. Volker & Upton, 1990). We propose a similar scenario for the entrainment of sulphide liquids.
The magmatic temperature of the peridotite plugs on Rum has previously been estimated at 1400 C, based on picritic magma temperatures (Holness et al., 2012). The solidus temperature of an Fe-bearing sulphide liquid is c. 1100 C ( Bowles et al., 2011;after Kullerud & Yoder, 1959;Arnold, 1971) suggesting that sulphides would be liquid at the estimated temperature of emplacement for the plugs. Given that the 'lubricating' magmas were energetic enough to entrain up to 50% olivine crystals (e.g. Holness et al., 2012) and sizeable xenoliths (e.g. Volker & Upton, 1990), it is highly likely that these were also capable of entraining immiscible sulphide droplets. A similar scenario of upward sulphide liquid entrainment has been suggested in the vicinity of Grasvally in the Northern Limb of the Bushveld Complex by Smith et al. (2016), suggesting that S-saturation occurred in a deeper master chamber (albeit via deep crustal contamination with a distinctive S-isotopic and S/Se composition). Therefore, we suggest that at least some sulphides (with mantle S/Se ratios) were formed prior to intrusion of the plugs, and were transported as sulphide liquid droplets into these conduits.
Sulphur content at sulphide saturation (SCSS), as defined by Li & Ripley (2005), is dependent upon the bulk composition of the magma (MgO, FeO T , SiO 2 , Na 2 O and K 2 O), pressure and temperature. Assuming the magmatic temperature of 1400 C from Holness et al. (2012) and using the bulk compositions of gabbroic plugs (e.g. plugs 3 and 9) and ultramafic plugs (e.g. plug 12 at Dibidil, and plug 1 at West Sgaorishal) reported in the Supplementary Material, we find that SCSS would be 2985 ppm (plug 12), 3510 ppm (plug 1) and 1830 ppm for the two gabbroic plugs (plugs 3 and 9) at 1 km depth [using the equation of Li & Ripley (2005)]. For the ultramafic plugs, this is apparently well in excess of the measured bulk S concentration (Table 4). However, we highlight that this bulk concentration potentially incorporates up to 50% cumulus olivine, and therefore the actual S concentration of the interstitial magma carrying the crystal mush will have been significantly higher if this is taken into account (i.e. up to double that measured). Further, a late picrite dyke (M9) on Rum is calculated to reach an SCSS at 1400 C and 1 km depth of 3000 ppm S. Upton et al. (2002) observed up to 60 vol. % olivine phenocrysts and we observe two sets of olivine phenocrysts in the M9 dyke, thus highlighting that the bulk composition of this dyke is not suitable for use in modelling the parental magma.
According to the equation of Li & Ripley (2005), pressure (i.e. depth) has little effect on SCSS in comparison with temperature. Therefore if we reduce the ambient temperature to 1200 C (i.e. still 100 C above the liquidus temperature of sulphide) we see a significant drop in SCSS, to $1600 ppm for the ultramafic plugs and $1000 ppm for the gabbroic plugs. Accommodating for the bulk geochemical analyses incorporating up to 50% cumulus olivine for the ultramafic plugs, and using this lower temperature, we suggest that SCSS is within the range of the S content now observed in plugs across Rum. Coupled with observations of sulphide mineral embayments in cumulus olivine (as noted above), and Li & Ripley's calculations of decreasing SCSS with fractional crystallization of a magma, it is plausible to assume that sulphide saturation could be achieved deeper in the magmatic system of Rum, prior to plug emplacement.
Several examples of sulphide liquid entrainment in magma conduits have been recorded worldwide; for example, the Norilsk-Talnakh magma conduit setting for orthomagmatic Ni-Cu-PGE mineralization, in which sulphides were initially exsolved in a deeper staging chamber before being flushed out to higher magmatic plumbing system levels (Brü gmann et al., 1993;Lightfoot & Keays, 2005). Sulphide entrainment is also suggested for the Platreef (Bushveld Complex) where sulphides were transported away from a deeper zone of sulphide liquid formation McDonald et al., 2009;Ihlenfeld & Keays, 2011;Sharman et al., 2013). In the Platreef, evidence for this sulphide transportation was based on Cu/Pd, S/Se ratios, S-isotope compositions and trapped sulphide . Plotted for comparison are the calculated mean Cu abundances of the sulphide liquid [calculated by normalizing the bulk Cu concentration to 100% sulphide, using the method outlined by Huminicki et al. (2005)] and weighted mean Pd abundances (per sample weighted according to the number of analyses of chalcopyrite, pentlandite and pyrrhotite) for sulphides from plug 1, centre and margins. melt inclusions (Holwell et al., 2011). Clearly, the process of entrainment of pre-formed sulphides, other silicate minerals and xenoliths is a common feature of many magmatic systems. However, in many of these cases only one part of the full system is exposed, with earlier processes at depth being inferred. The volcanic plugs and the Layered Suite of Rum provide a rare opportunity to compare layered intrusion and volcanic pipe compositions and mechanisms of sulphide transport at the same present-day erosional level.
The d 34 S composition of the peridotite plugs (þ0Á1 6 1Á0ø, excluding plugs 1 and 2) and peridotite cyclic units of the ELS (-0Á5 6 1Á4ø; Supplementary Material Table E) overlap with the typical Icelandic mantle plume range of -2Á0ø to þ0Á4ø (Torssander, 1989), mid-ocean ridge basalt (MORB) -0Á91 6 0Á5ø (Labidi et al., 2012) and the picritic Trotternish Sills on the Isle of Skye with average d 34 Sof-2Á3 6 1Á5ø (Hughes et al., 2015). Thus we suggest that, for most plugs and the ELS, S-saturation was reached prior to emplacement, in the deeper magmatic plumbing system where olivine crystallization and fractionation was taking place, and did not involve the addition of crustal sulphur with d 34 S different from magmatic compositions. We note that the d 34 S values observed in the ELS and plugs (excluding plugs 1 and 2), are slightly heavier (-0Á5t o þ0Á1 6 1ø) than the Icelandic plume d 34 S signature of -0Á8ø to -2Á3ø [Torssander (1989) and Hughes et al. (2015), respectively] and that this may be due to small changes in the magmatic source signature. With ejection of crystal mushes from a staging chamber, sulphide liquid was entrained and emplaced into the peridotite plug magma conduits and Rum Layered Suite alike. However, in the case of plugs 1 and 2 significant deviation of d 34 S and elevation of S/Se ratio indicates that localized crustal contamination also took place.

Potential crustal S-sources for the West and Loch Sgaorishal plugs (plugs 1 and 2)
The d 34 S of plug 1 (West Sgaorishal) ranges from -14Á7 to -8Á0ø with a S/Se ratio of sulphide minerals ranging from 3000 to 12 000, whereas the d 34 S of plug 2 (Loch Sgaorishal) varies from -6Á5t o-2 Á9ø (Fig. 7a). Further, the sulphide S/Se ratios of plug 1 are elevated above the range of most ELS sulphides, and often exceed 10 000 (a critical value for crustal sedimentary sulphide compositions; Ihlenfeld & Keays, 2011, and references therein). These S-isotope compositions are significantly lighter than typical mantle or Icelandic plume values, and, together with S/Se, demonstrate that crustal S must have been added to the plug 1 and 2 magmas. There is also a change in textural character of the sulphides in plug 1, which became more abundant and coarser grained. Hence the following questions arise: what was this sulphur source, and why is this light d 34 S recorded only in two plugs from NW Rum?
The d 34 S of sedimentary S (typically pyrite) can be highly variable owing to bacterially mediated sulphate reduction processes during diagenesis. The range of d 34 S can be of the order of tens of per mil (ø) with both negative and positive values. Palaeozoic and Mesozoic sedimentary sulphides generally have more negative values than those observed in Precambrian rocks (Canfield & Teske, 1996;Parnell et al., 2010).
Most plugs on Rum have intruded through Torridonian (Applecross Formation) sandstones at the current erosion level, but the NW tip of plug 1 also intrudes the unconformable base of a sequence of Triassic limestones and calcretes (Binns et al., 1974;Fyfe et al., 1993). Hughes et al. (2015) established that the d 34 S of these Triassic sediments is isotopically heavy (with very low S concentrations) and thus cannot account for the d 34 S of plugs 1 and 2 via assimilation. This leaves three possible explanations: (1) Torridonian lacustrine sediments with a light d 34 S (e.g. Parnell et al., 2010) are present below plugs 1 and 2; (2) a segment of Jurassic mudrock with suitably light d 34 S occurs in a downthrown faulted block of a major fault zone (e.g. Main Ring Fault) through which plugs 1 and 2 have intruded; or (3) the d 34 S signature of plugs 1 and 2 is derived from above the current erosion level and involves the Lower Jurassic Broadford Group that probably covered this area at the time of conduit activity.
Despite some Torridonian lacustrine sediments from the mainland Stoer Group having d 34 S as low as -30Á1 6 17Á3ø (Parnell et al., 2010), all the Sleat and Torridon Group Mesoproterozoic sediments analysed by Hughes et al. (2015) had values > 0Á0ø. Further, the Stoer Group is not recorded in the sedimentary pile exposed on either Skye or Rum, suggesting that scenario (1) is extremely unlikely.
Hebridean Jurassic mudrock and ironstone d 34 S values range from -10 to -35ø, and associated basaltic dykes intruded through this sequence record d 34 S values as light as -30Á7ø (Hughes et al., 2015). Blocks of Jurassic mudrock are recorded as being present within the Main Ring Fault on Rum at Allt nam B a and near Dibidil Figs 1 and 2c). In contrast, no such sedimentary blocks are reported from the Long Loch Fault, not least owing to this being a strikeslip fault with no vertical motion that would have downthrown these sedimentary units. It is possible, however, that the subsurface route of the plug magma conduits could have intersected a downthrown block of Jurassic mudrocks, assuming that these conduits were fed from somewhere near the centre of the Rum volcanic edifice. However, only plugs 1 and 2 from the NW quadrant of the island record this light d 34 S and the absence of such a light d 34 S in plugs 3-5 renders this scenario highly improbable; and necessarily very localized, if true.
By a process of elimination, to account for the light d 34 S, we are left with a scenario whereby crustal sulphur contamination demonstrably took place only in plugs 1 and 2. If contamination were taking place above the current erosion level, in Lower Jurassic sediments that very probably overlay this region, there remains a problem as to how these crustally contaminated sulphides are now present below the level of the potential contaminant. One possibility may be that the sulphides moved down the plug or conduit. Below we consider whether this is feasible given the upward injection of magma in a conduit, and whether the observed characteristics of the sulphides in plugs 1 and 2 support this model. What were the physical constraints on sulphides sinking and could this effect have been so localized that the other plugs (3-5) further 'downhill' on the NW quadrant of Rum, be too low in altitude to record this process?

Dynamic conduit modelling and sulphide sinking Hadamard-Rybczynski conduit modelling
The settling of sulphide liquid droplets through a magma conduit can be modelled either by using Stoke's Law or by the Hadamard-Rybczynski equation (e.g. Lightfoot et al., 1984) as established by the experiments of de Bremond d' Ars et al. (2001). The latter takes into account the non-Newtonian fluid properties in the conduit, as well as the differences in both density and viscosity of the silicate and sulphide liquids present in the system, and is expressed as where V s is settling velocity (m s -1 ), r is sulphide droplet radius (m), Dq is the difference in density between sulphide liquid and silicate magma (¼ 1400 kg m -3 ), mf and md are the viscosities of silicate magma (0Á1P as ; Paterson et al., 1998) and sulphide liquid (0Á01 Pa s) respectively, and g is the acceleration due to gravity (9Á8m s -2 ). Sulphide droplets will sink (with velocity V d ) through the conduit pipe once the settling velocity (V s ) exceeds the upward velocity of magma intruding into and through the pipe (V z ), such that Sulphide droplet settling occurs when V d < 0m s -1 . The analogue experiments of de Bremond d' Ars et al. (2001) indicated that sulphide droplets can be transported in suspension by flowing magmas, and that owing to the large surface tension of the droplets, they do not coalesce during active transport. In part, this could explain the disparity and size of sulphide minerals in peridotite plugs 4-12. These plugs generally do not record matrix banding or indentation layering, suggesting that prolonged quiescent periods were not prevalent during cooling, hence perturbing the ability of the conduit to form crystal layering (e.g. Wadsworth, 1994) and thus arresting sulphide droplet settling and coalescence. Further, the low abundance of sulphides in plugs 4-12 would also hamper coalescence. The size and distribution of fine sulphides in these plugs probably reflects the mode of transportation and entrainment that occurred during conduit activity.
For scenarios involving higher volumes of sulphide liquid, plots of V s and V d according to sulphide droplet radius are presented in Fig. 8a and b. Sulphides in the margins of plug 1 range in diameter from 0Á5t o4m m (Table 2); most intercumulus sulphides are 1-3 mm in diameter. V z was varied in equation (2) until sulphides of the minimum average radius observed in the margin of plug 1 (r $ 0Á5 mm) had V d < 0m s -1 . This required an upward magma velocity of 0Á01 m s -1 (Fig. 8a and b), such that sulphide droplets would sink through the conduit at a rate of 0Á001 m s -1 (V d ). By lowering V z further to 0Á001 m s -1 , sulphides with r ¼ 0Á2 mm would also begin to sink at a similar rate. Thus, the upward flow of magma through the conduit would essentially have to have ceased before sulphide droplets of the sizes observed in the margins of plug 1 would settle. The viscosity of the silicate magma was assumed to be 0Á1Pas (Paterson et al., 1998); however, if the conduit was instead filled with olivine phenocrysts entrained in basaltic magma then the overall effective viscosity of the 'silicate magma' would be higher. Holness et al. (2012) suggested a crystal load in the range of 10-40 vol. % for the peridotite plugs, based on the volume of phenocrysts entrained in picrite dykes on Rum. Even by increasing viscosity to 1 Pa s [i.e. an assumed 50% crystal mush entrained in a mafic magma, with viscosity ¼ 1 Pa s based on Paterson et al. (1998)]a t V z ¼ 0Á001 m s -1 , sulphides of r $ 0Á5 mm would still be capable of sinking through the conduit. In summary, it is possible that sulphide liquids forming through crustal contamination above the present-day level of erosion and exposure of plug 1 (and plug 2) would have sunk through the conduit once active upwards magma injection had ceased. Sulphide liquid sinking may have taken place either in a passively cooling (stationary) environment and/or sulphide liquid was 'sucked' downwards if magma withdrawal were taking place.
By assuming that the cooling rate of a stationary magma is controlled by Newton's Law of Cooling, it would take the margin of plug 1 c. 5 days to cool below 1100 C, where T a is the temperature of the country rock [assumed to be 600 C, based on Holness et al. (2012)], T o is the initial magma temperature of the conduit [assumed to be 1400 C; Holness et al. (2012)], T is the target temperature of 1100 C for the sulphide liquidus (e.g. Bowles et al., 2011) and j is a thermal diffusivity constant of 10 -6 m 2 s -1 . Given a downward velocity V d of 0Á001 m s -1 for sulphide droplets through the quiescent conduit, sulphides would be capable of settling through a distance of the order of 400 m, downwards from the initial level of S-saturation. As mentioned above, however, constraints on the SCSS of these plugs may suggest that their emplacement temperature was lower than 1400 C. Therefore if we use 1200 C as the emplacement temperature, we estimate that sulphides could still have settled more than 100 m down the conduit. Ultimately, this settling distance is likely to be subject to estimation errors; for example, as sulphide droplets coalesce during settling, their settling velocity will change. In addition, as the conduit cools towards 1100 C, the effective viscosity of both the sulphide liquid and silicate magma will increase, thus slowing its descent. The 'permeability' of the host silicate mush (i.e. amount of partial melt and degree of ponding of sulphide liquid above areas of low permeability; e.g. Holwell & McDonald, 2006;Hutchinson & McDonald, 2008) may not be uniform throughout a single volcanic plug. However, the interfacial tension at silicatesulphide liquid interfaces may preclude sinking of sulphide droplets below a critical size range (e.g. Mungall & Su, 2005). This depends on the proportions of silicate melt to solidified silicate minerals in the system, and would suggest that compaction-driven sulphide segregation is possible (Rose & Brenan, 2001).
The estimated thickness of Triassic sediments in the Hebrides Basin in the vicinity of Rum on the 'Skye High' of this Mesozoic sedimentary basin ranges from tens of metres on Ardnamurchan, to 100 m at Raasay off the Isle of Skye, and 300 m at Gruinard Bay on the Scottish mainland (Steel et al., 1975). Given that the current erosion level of plug 1 occurs at the base of these Triassic sediments (Figs 2a and 9a) and the estimated minimum potential settling distance of sulphide droplets is 100 m (assuming a lower emplacement temperature of 1200 C) to 400 m [using the Holness et al. (2012) emplacement temperature of 1400 C], this is within reach of S-rich lower Jurassic sediment horizon(s), assumed to have been present above the current erosion level (Fig. 9a). Assimilation of wall-rock sediments, particularly light d 34 S crustal sulphur from Jurassic units, would induce sulphide saturation at a high level in the conduit (above the current erosion level; e.g. Fig. 9b). Thus, we can reasonably envisage that once active magma-mush injection had ceased, silicate and sulphide liquids from above sank through the conduit, forming olivine cumulates with interstitial sulphide liquid that recorded light crustal S isotope compositions and high S/Se ratios inherited from hundreds of metres above (e.g. Fig. 9c).
Although plug 2 also records d 34 S in sulphides significantly less than the local magmatic compositions of around 0ø, it has lower abundances of sulphide than plug 1. Projecting the dip of the base of the Triassic sedimentary units near plug 1 (approximately 20 NW; Fig. 9a) and even accounting for the current topography and elevation of the area between plugs 1 and 2, the lower sulphide abundance in plug 2 may reflect the greater distance between the potential Lower Jurassic contamination zone and the present-day level of exposure (Fig. 9a). Further, the lack of a crustal d 34 S signature in plugs 3, 4 and 5 is consistent with the model, in that their current level of exposure is >400 m below the projected contamination horizon, so that any sulphides that did settle downwards are likely to have been removed by erosion (i.e. Fig. 9a). The lack of a crustal d 34 S signature in plugs from elsewhere on the island, particularly farther east, may also suggest that the Mesozoic sedimentary sequence of the Hebrides Basin was not present, as the Isle of Rum sits on the 'Skye High' within the basin such that only the feather-edge of the basin's sediments could have been deposited over part of the island (see Hughes et al., 2015).

Jurassic shale geochemical interaction with volcanic plugs
The sedimentary host rocks to the volcanic plugs (Triassic, Fig. 2a; Torridonian, Fig. 2a-c) typically appear bleached, fissile and metasomatized. Detailed studies of in situ anatexis of these rocks have provided insights into the longevity of the plugs as active magma conduits (e.g. Holness, 1999;Holness et al., 2012). Static melting occurred up to 15 m from the gabbro plugs and 6 m from the larger peridotite plugs, suggesting that the peridotite plugs were a relatively short-lived magma conduits (active for c. 1-2 years; Holness et al., 2012) whereas the gabbro plugs were longer lived features (c. 40 years; Holness, 1999). However, Holness (1999) concluded that anatectic melt migration did not occur, so that very limited contamination of the plug margins was likely to have taken place. However, Yallup et al. (2013) and Hughes et al. (2015) (and references therein) demonstrated that the volatile content (S, C, H 2 O, etc.) of mudrocks is particularly vulnerable to mobilization by magmas during contact metamorphism of wallrocks and/or their brecciation. In this way, we envisage a scenario whereby S is preferentially mobilized, imparting a strong crustal contamination signature onto the plug magmas, but without other non-volatile trace element contamination.

Comparison with Paleogene macrodykes in East Greenland and other similar intrusions
A similar set of geological circumstances exists in East Greenland, where Holwell et al. (2012) described a suite of mineralized macrodykes, radiating away from the Skaergaard intrusion and intruding Archaean basement and overlying pyrite-rich Cretaceous black shales (with d 34 S -23 to -30ø) and Paleogene flood basalts. Magmatic Cu-PGE-Au sulphide mineralization, with crustal d 34 S signatures (-10 to -26ø), is present along the margins of the subvertical Miki Fjord Macrodyke, which is observed to intrude all three lithological units. Further north, the Togeda Macrodyke, which is exposed at a level where it intrudes the basement gneiss, has similar sulphides that have a light S-isotope signature (Holwell et al., 2012). Holwell et al. (2012) concluded in both cases that Cretaceous shales were the trigger for sulphide saturation and that sulphide droplets were entrained within the macrodyke magmas during intrusion. The presence of sulphides in the Togeda Macrodyke, beneath the stratigraphic level of the contaminant, was Owing to the orientation of the Mesozoic Hebridean Basin Triassic and Jurassic sediments relative to the present-day topography, the current exposures of plugs 3, 4 and 5 are too great a distance below the Jurassic contamination horizon to record crustal d 34 S and voluminous sulphide liquids. However, plugs 1 and 2 still record light crustal d 34 S. (b) Active volcanic plug intrudes through Torridonian sediments, and Triassic and Jurassic sediments of the Hebrides Basin (since eroded) above. The plug entrains tiny droplets of high-tenor sulphide liquid from a staging chamber. Wall-rock metasomatism and partial melting (especially recorded in the siliceous Torridonian Applecross Formation sandstones) takes place. In the upper portions of the plug, S-rich shales and siltstones from the Jurassic sediments contaminate the magmas with isotopically light crustal S. (c) Active magma intrusion of volcanic plug ceases, and magma-entrained olivine mushes are left to settle through the conduit (forming cumulate layers, particularly at the conduit margins). Sulphide droplets entrained in the plug, including more voluminous sulphide liquids from crustal S contamination horizons (above) in Jurassic sedimentary units, settle through the conduit. Sulphide liquids amalgamate during settling, and form an interstitial or intercumulus phase around the cumulate olivine. Sulphide liquid is particularly abundant around the conduit margins (owing to wall-rock S contamination) and has a higher concentration of PGE, higher S/Se ratio and lighter d 34 S. explained by northward injection of magma, which could then incorporate S from sediments exposed deeper to the south (Holwell et al., 2012).
A similar situation to that described above for the Rum plugs is also possible for the Togeda Macrodyke. The Cretaceous sedimentary units are inferred to have been present some 500 m or so above the present erosion level. Given the much larger size of the droplets (up to 10 mm) in the macrodykes compared with the Rum plugs (mostly 1-3 mm), the likelihood of large droplets settling down through the dyke conduit by a vertical distance of up to 500 m is plausible. For example, assuming a sulphide droplet with radius (r)o f 0Á5 mm and a silicate liquid (not crystal mush) of viscosity 0Á1 Pa s, then owing to the silicate-sulphide liquid density difference, the sulphide droplets would be capable of sinking through a magma conduit with upward silicate magma velocities (V z )o fu pt o1 Á1ms -1 . However, in a dyke scenario, lateral movement of magmas along the dyke (owing to a pressure gradient from the point of intrusion along an opening fracture) must also be factored in. Hence both lateral and downward migration of sulphide droplets may be envisaged for planar features such as dykes (e.g. Todega Macrodyke), in contrast to pipe-shaped conduits.
The downward movement of sulphide liquids in a cooling magmatic system has also been suggested for ultramafic complexes in Madagascar (McDonald, 2008). In this model, sulphide liquids migrate laterally, as well as downwards, towards the centre of the intrusion during crystallization. During their migration, the sulphide liquids fractionated such that Pt-and Pd-bearing PGM exsolved from, and detached from, the sulphide liquid (becoming trapped in olivine mushes), leaving Cu-rich sulphide liquid to continue sinking towards the centre of the intrusion (McDonald, 2008).
Our model of sulphide sinking in a cooling magmatic system (where magma flow has recently ceased) differs markedly from sulphide sinking models suggested for other mineralized conduit settings; for example, at Norilsk-Talnakh where Arndt (2011) envisaged that sulphide liquids sank (or were re-injected) through an active silicate magma conduit, with a significant upward velocity of magma injected from beneath. Sulphide pools may be re-entrained by pulses or batches of new magma (e.g. Robertson et al., 2014), leading to complex scenarios of re-deposition within large, dynamic or long-lived conduit systems, with ultimate sulphide tenor upgrading (Barnes et al., 2016, and references therein).

N-factor modelling: silicate/sulphide liquid ratios
Another parameter that can be calculated for a sulphide-bearing conduit is the proportion of sulphide liquid to silicate magma within the system. In a closed system, this is typically denoted by the 'R-factor'; however, in the case of open-system conduits this is termed the 'N-factor' and was modelled for the sill-hosted Norilsk-Talnakh Ni-Cu-PGE orthomagmatic mineralization in Siberia (Brü gmann et al., 1993;Naldrett & Fedorenko, 1995;Naldrett et al., 1996). In an open system such as a magma conduit silicate magma batches are flushed through the conduit system. If this system is Journal of Petrology, 2016, Vol. 57, No. 2 S-saturated, sulphide liquids may become trapped within the conduit whilst silicate magmas continue to flush through. Owing to their strong chalcophile nature, PGE and Cu will partition from each fresh silicate magma batch into the resident sulphide liquid, thereby upgrading the PGE tenor of this sulphide in a mechanism analogous to zone refining (Brü gmann et al., 1993;Naldrett & Fedorenko, 1995;Naldrett et al., 1996). The proportion of silicate magma (as a time integrated volume) relative to the volume of the sulphide liquid will determine this degree of PGE upgrading.
The N-factor is typically used for conduits in which magmas move upwards past sulphide accumulations. However, in the situation we propose for Rum, sulphide droplets are formed at the top of the magma conduit and slowly settle through it (e.g. Brü gmann et al., 1993). The effect is the same; in both cases silicate magma and sulphide liquids must migrate past one another to achieve upgrading of PGE tenor. Thus the N-factor model is appropriate in this instance to assess the volumes of ambient sulphide liquid in the conduit.
The N-factor equation, based on Cox et al. (1979) is as follows: where Y i is the concentration of metal i in the sulphide, X i is the initial concentration of metal i in the silicate, and D is the partition coefficient for metal i between silicate and sulphide; N ¼ volume silicate magma volume sulphide liquid : As the value of N increases, the concentration of strongly chalcophile metals such as Pt and Pd rapidly increases. For plug 1, the concentration of trace elements such as Pd and Cu according to the volume ratio of silicate magma to sulphide liquid (N-factor) can be modelled using equation (4), assuming D Cu ¼ 1470 and D Pd ¼ 190 000 (Mungall & Brenan, 2014), with an initial silicate magma composition of 50 ppm Cu and 10 ppb Pd (comparable with picritic compositions and Cu and Pd abundances in other low-sulphide plugs such as plug 12). By finding the intersection between the model curves for Pd and Cu and the measured concentration of these metals in the plug, we can estimate that the ratio of silicate magma to sulphide liquid (N-factor) ranged from 1000 to 1500 (specifically using sample RM33 from plug 1) (Fig. 8c).
To cross-check if the N-factor calculated is reasonable, we have used the average abundance of Pd measured in sulphides (Table 3) calculated per sample for plug 1, and plotted this against the calculated Cu abundance for a homogenized sulphide liquid, based on assigning all bulk-rock Cu to chalcopyrite, and normalizing to 100% sulphides. As sulphide assemblages comprise pyrrhotite, pentlandite and chalcopyrite (with only accessory bornite, chalcocite and galena), we have followed the method of Huminicki et al. (2005) using the whole-rock measured abundances of S, Cu, Ni and Fe. All of the whole-rock Cu budget was assigned to chalcopyrite, and after subtracting the mass of S required, and the mass of Ni bound within olivine [assuming a Ni content of 2400 ppm in olivine and an average of 60% olivine in the plug, based on Holness et al. (2012)] all of the remaining Ni was assigned to pentlandite. The proportion of pyrrhotite was then estimated from the remaining S. Following the observations and methods of , such that chalcopyrite contains negligible PGE in solid solution, we then recalculate the measured whole-rock PGE content to 100% pyrrhotite þ pentlandite, as these are the principal PGEbearing base metal sulphides. By this method, the Cu and Pd compositions of sulphides in plug 1 can be modelled by equation (4) if N ¼ 1000-1500 ( Fig. 8c and d). This includes the centre of the plug (e.g. RM30). However, the outlier RM95 at the southern plug margin has a substantially higher concentration of Cu (almost 120 000 ppm), which may suggest that the margin of the plug (as represented by RM95) has acquired extra Cu from the surrounding wall-rock. The Triassic Sandstone Formation in this area contains rare malachite reported with carbonate calcretions .

The significance of sulphide sinking versus upward entrainment
Comparison with examples outside the BPIP shows that a process of sulphide sinking through cooling, recently inactive conduits may well be widespread and could account for the presence of high-volume sulphides with crustal S compositions at levels stratigraphically lower than that at which they originally formed. Thus, the study of mineralized magmatic sulphide systems also needs to consider the possibility of contamination from sources above the level of mineralization, which may no longer be exposed. In terms of exploration models, this is significant in that intrusions exposed below the stratigraphic level of any potential contaminant (or where possible contaminants have been eroded) still have the potential to host magmatic sulphides. If a settling sulphide liquid had the ability to pond in a depression in the magmatic plumbing system, then massive sulphide accumulations that result from crustal contamination could potentially be found hundreds of metres below the level of the contaminant that triggered sulphide saturation. Our work provides quantifiable and traceable evidence for the processes of sulphide sinking. As such, it offers support to the growing number of studies that cite sulphide sinking, slumping and/or re-injection within conduits as a mechanism for magmatic sulphide accumulation and upgrading (Benko et al., 2015;Saumur et al., 2015;Barnes et al., 2016, and references therein;). This contrasts with the more traditional view of upward entrainment of sulphides into higher level intrusions (e.g. Naldrett et al., 2009Naldrett et al., , 2011.