Platinum-group element geochemistry of the Parana´ ﬂood basalts – Modelling metallogenesis in rifting continental plume environments

The 135 Ma Parana´-Etendeka Large Igneous Province (PELIP) is one of the largest areas of continental ﬂood basalt (CFB) volcanism in the world and is widely agreed to be a product of intracontinental melts related to thermal anomalies from the Tristan mantle plume. The province rifted during the break-up of Gondwana, as the plume transitioned into an oceanic geo-dynamic environment. This study reports analyses of plume-derived basalts from the Brazilian side of the PELIP (the Serra Geral Group) to investigate major, trace and platinum-group element (PGE) abundances in an evolving plume-rift metallo-genic setting, with the aim of contextualising metallogenic controls alongside existing magmatic interpretations of the region. The chalcophile geochemistry of these basalts deﬁnes three distinct metallogenic groupings that ﬁt with three modern multi-element magma classiﬁcations for Serra Geral lavas. In this scheme, Type 4 lavas have a distinctive PGE-poor signature, Type 1 (Central-Northern) lavas are enriched in Pd, Au and Cu, and Type 1 (Southern) lavas are enriched in Ru and Rh. Our trace element melt modelling indicates that the compositional variations result from changes in the melting regime between the gar-net and spinel stability ﬁelds, in response to the thinning and ‘unlidding’ of the rifting continent above. This process imposes progressively shallower melting depths and higher degrees of partial melting. Accordingly, Type 4 magmas formed from small degree melts, reducing the likelihood of sulphide exhaustion/chalcophile acquisition at source. Type 1 (Central-Northern) magmas incorporated components of the sub-continental lithospheric mantle (SCLM) in higher-degree partial melts; the SCLM was heterogeneously enriched via metasomatism prior to plume melting, and this produced enrichment in volatile metals (Pd, Cu, and Au) in these magmas. In contrast, the Ru-Rh enrichment in Type 1 (Southern) lavas is attributed to increased spinel-group mineral and sulphide incorporation from the mantle into higher degree partial melts close to the continental rift zone. Our models conﬁrm the importance of contributions from SCLM melts in precious metal mineral systems within CFB provinces, and reinforce the role of heterogeneous metasomatic enrichment underneath cratons in boosting intracontinental prospectivity with respect to ore deposits. (cid:1) 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/ licenses/by/4.0/).


INTRODUCTION 1.Mantle plumes and precious metals
Many of the world's important magmatic sulphide deposits are found within continental flood basalt (CFB) provinces (e.g.Lightfoot et al., 1993;Naldrett, 1997).Whereas the mineralisation processes at work in each CFB are primarily controlled by mid-to upper-crustal differentiation mechanisms (e.g.Thompson and Gibson, 1991;Ellam, 1992;Maier and Groves, 2011), the initial geochemical fingerprint of CFB magmas can have important implications for metallogenic fertility and understanding how metals are mobilised from the mantle (e.g.Hawkesworth et al., 1988;Wooden et al., 1993;Wilson et al., 1996;Bierlein et al., 2009).
Localities with prominent PGE orthomagmatic sulphide deposits share a similar geodynamic setting, with plumederived (i.e.intraplate) magmas interacting with thick continental cratons and their margins.Plumes represent buoyant material from mantle discontinuities that rise to the base of the crust (e.g.Morgan, 1971;Morgan, 1972;Shannon and Agee, 1998;Jellinek and Manga, 2004;Burke et al., 2008;Deschamps et al., 2011;Li and Zhong, 2017), inducing partial melting and magma generation in the asthenosphere and lowermost lithosphere (Jerram and Widdowson, 2005 and references therein).Decompression models for CFB melting predict ~5% of the lithosphere (particularly the sub-continental lithospheric mantle or SCLM) is involved in melt generation, and this contribution is dominant in the early stages of rifting (e.g.McKenzie and White, 1989;White and McKenzie, 1995;Turner et al., 1996;Howarth and Harris, 2017).An observed spatial association between PGE-rich ore deposits and Archaean lithosphere may indicate that the 'inheritance' of precious metals plays an important role in dictating aspects of the geochemistry of plume-derived magmas and mineralisation (e.g.Downes, 2001;Hawkesworth and Scherste ´n, 2007;Zhang et al., 2008;Maier and Groves, 2011;Maier et al., 2012;Griffin et al., 2013;Barnes et al., 2015).
Here, we examine the nature of the interactions between plumes and the SCLM and their role as a control on PGE (and other metal) content of CFB parental magmas.Work on the North Atlantic Igneous Province (NAIP) demonstrates that plume environments that transition from continental to oceanic settings, as well as changes in the thickness and nature of lithosphere above the region of partial melting, directly affect the PGE geochemistry of ascending plume-derived magmas (e.g.Andersen et al., 2002;Hughes et al., 2015).Continental lavas from Greenland and the British Isles exhibit distinctly higher Pt/Pd ratios than oceanic lavas from Iceland, attributed to a shifting ''metal basket" (i.e. the concentration and variety of precious and/or base metals) and variable SCLM incorporation in partial melts (Hughes et al., 2015).In this study, we investigate lavas associated with the Tristan plume and the Parana ´-Etendeka Large Igneous Province (PELIP) in the South Atlantic in order to test how (if) the PGE, Au and base metal geochemistry of the CFB province changed through time and space.The PELIP shares a similar transitional continent-to-ocean geodynamic setting to the NAIP.If the inherent controls of trace element and isotopic variation in regional magmas reflect distinct geodynamic controls, this should be evident in PGE and other chalcophile elements.New geochemical data for onshore South American PELIP flood basalts are compared and used in quantitative geochemical melt modelling, to assess how PGE geochemistry changes through the evolution of the CFB province, with a focus on the link between variable SCLM properties and resulting partial melt compositions.

Formation of the Parana ´-Etendeka Large Igneous Province (PELIP)
Currently positioned in the centre of the southern Atlantic Ocean, the Tristan plume has migrated through a cratonic, to continental margin, to mid-oceanic setting throughout the Cretaceous and Cenozoic (Fodor and Vetter, 1984).It is active today under the islands of Tristan da Cunha and Gough, but has a >130 million year history of volcanism that spans two continents and the entire width of the South Atlantic (Stewart et al., 1996).The plume head stage is thought to be responsible for the eruption of the PELIP, one of the largest CFB provinces in the world (Courtillot et al., 2003), in addition to the Rio Grande Rise and Walvis Ridge topographic features on the ocean floor (e.g.Camboa and Rabinowitz, 1984;O'Connor and Duncan, 1990;Ussami et al., 2013) (Fig. 1a).The Tristan plume is one of the world's ''primary hotspots" according to Courtillot et al. (2003), satisfying the criteria of creating linear chains of age-progressive volcanics, exhibiting large flood basalt buoyancy flux, high 3 He/ 4 He isotope ratios and significant low shear wave velocity anomalies in the mantle below.
The ~1.7 million km 3 PELIP is spread asymmetrically between (primarily) Parana ´, Brazil and Etendeka, Namibia (which were conjoined in the Cretaceous; Fig. 1a), with fifteen times higher volume of lavas found in South America than Africa (Fodor et al., 1989).Intraplate magmas are typically chemically variable and are derived from relatively undepleted mantle (e.g.Zindler and Hart, 1986;Stracke et al., 2005;Hawkesworth and Scherste ´n, 2007).The Tristan plume arrived at the Gondwanan lithosphere under the modern day central Parana ´region in Brazil at ca. 135 Ma, inducing partial melting in the asthenosphere (Stewart et al., 1996).The duration of eruption of the onshore PELIP has been debated, ranging from <1 million years (suggesting a high eruption rate of around 1.5 km 3 / year; Renne et al., 1996) to 10 million years (suggesting a much lower eruption rate of under 0.1 km 3 /year; Stewart et al., 1996).Recent studies suggest a conservative 0.8 km 3 /year for the majority of the flood eruption (Beccaluva et al., 2020), comparable to other CFB provinces, such as those that produced the Deccan and Karoo traps (Natali et al., 2017 and references therein).The PELIP eruption rate is broadly similar to the NAIP (a mean of 0.6 km 3 /year; Eldholm and Grue, 1994).
lithosphere (e.g.McKenzie and White, 1989;Turner et al., 1996;de Wit et al., 2008).North-westerly plate movement migrated plume activity to the Etendeka-Angola margin, synchronous with the eruption of the south-eastern lavas in Parana ´ (Beccaluva et al., 2020) (Fig. 1b).Following significant extension and basin formation as the South American and African continents drifted apart (Martins-Ferreira et al., 2020 and references therein), the asthenosphere above the Tristan plume continued to undergo partial melting and eruption of lava on the seafloor throughout the Cretaceous.Sustained volcanism formed ridges on either side of the mid-oceanic rift, the Rio Grande Rise (west) and the Walvis Ridge (east) that connect the PELIP to the modern active hotspot islands of Tristan da Cunha and Gough (e.g.O'Connor and Duncan, 1990;O'Connor and Jokat, 2015;Homrighausen et al., 2019).Fig. 1b summarises the different chemostratigraphic stages of PELIP magmatism with respect to congruent rifting.

Geochemical classification of the Parana ´flood basalts
The South American portion of the PELIP erupted over the Botucatu Sandstones and Proterozoic basement of the Parana ´Basin.The basin is oriented NE-SW with its morphology controlled by three major underlying structuresthe Ponta Grossa and Rio Grande arcs, and the Torres Trough (e.g.Cordani and Vandoros, 1967;Peate et al., 1992;Besser et al., 2018).On average, the basin-filling lavas in the region are >1 km thick, with thicknesses of up to 1.7 km towards the north (Gordon, 1947;Leinz, 1949;Leinz et al., 1966;Peate et al., 1990).The Parana ´lavas, referred to more formally as the Serra Geral Group in Brazil (e.g.Licht, 2018;Rossetti et al., 2018), are primarily basaltic to basaltic-andesitic and compositionally grouped into synchronous High-Ti (HT) in north-central Parana ánd Low-Ti (LT) in the south-east.The HT and LT lavas often inter-finger in central Parana ´due to their synchronous eruptive sequence (e.g.Licht, 2018).Volcanism in the southeast laterally evolved into bimodal mafic-felsic volcanism, producing significant amounts of silicic lavas (e.g.Peate et al., 1992;Polo et al., 2018).
Given that the Serra Geral lavas are thought to have erupted simultaneously throughout the province via different melting sources rather than by differentiation processes from a single source (Peate, 1997), the underlying stratigraphy of the Parana ´Basin is complex (Turner et al., 1999;Rossetti et al., 2018).Recent studies suggest that the synchronous HT and LT lavas were erupted from separate sub-surface magma plumbing systems spawned from plume-derived melting, and followed separate magmatic differentiation routes and localised melting processes (Natali et al., 2016;Licht, 2018), with rare evidence for minor interaction between HT and LT magmas in plumbing systems (De Min et al., 2017).LT magmas reflect higher degrees of crustal contamination in contrast to limited (or absent) crustal contamination for the HT magmas (e.g.Piccirillo et al., 1988;Peate et al., 1992;Peate and Hawkesworth, 1996;Turner et al., 1996;Rocha-Ju ´nior et al., 2013;Natali et al., 2017;Marques et al., 2018).
Whereas the mafic Serra Geral lavas are similar in both hand specimen and thin section, the magma-types exhibit small differences in modal mineralogy and larger differences bulk geochemical compositions.Consequently, pioneering studies on the geochemistry of the Serra Geral Group further classified the HT, LT and silicic magma-types based on major and trace element, and isotopic variations (e.g.Bellieni et al., 1984;Fodor, 1987;Peate et al., 1992;Peate, 1997).With time, increasingly more sophisticated distinctions across multi-element parameters have been sought (e.g. the sixteen Si-Ti-Zr-P divisions introduced by Licht, 2018).A generalised northern and southern stratigraphy for Serra Geral is given in Fig. 1c, and Table 1 compares the Peate et al (1992) and Licht (2018) classifications, with an indication of Si, Zr, Ti and P concentration cut-offs for the latter.In this paper, we address the intersection between geochemical characteristics of both naming schemes, with particular emphasis on PGE concentrations.
Paranapanema and Pitanga each contribute to 20% of the total Serra Geral volume, and Urubici contributes ~5%.Urubici lavas are the most sparsely observed at surface, although it is seen in the highlands of Sa ˜o Joaquim and as sills interbedded with Vale do Sol in the south-east of the CFB province (Besser et al., 2018).Paranapanema overlies Pitanga throughout most of the flood sequence; however it underlies Gramado in the very south of the basin (Bellieni et al., 1984).

Low-Ti Group
The LT basalts to basaltic-andesites (Gramado, Esmeralda and Ribeira) comprise around 50% of the total eruptive volume of the Serra Geral Group and are found in the south of the CFB province (Peate, 1997).Over a third of the total lava volume belongs to the Gramado magmatype, and 5-10% to the Esmeralda magma-type (Peate et al., 1992).As per Peate (1997), both magma-types see distinctive enrichment of heavy rare earth elements (HREE) and large-ion lithophile (LIL) elements relative to HFS elements and LREE, and a negative Ti anomaly when normalised against mantle compositions.Esmeralda has highly variable eNd (À4 to 4), while maintaining consistent initial Sr isotope values (( 87 Sr/ 86 Sr) i = 0.7046-0.7086)(Peate and Hawkesworth, 1996.Gramado lavas have the lowest TiO 2 (as low as 0.7 wt%) and eNd (À9 to À4) in the Serra Geral sequence, with higher and much more varied ( 87 Sr/ 86 Sr) i ratios (0.7075-0.7167) relative to Esmeralda, which sets it apart from all other mafic lavas in the Serra Geral Group (Peate and Hawkesworth, 1996).The Gramado magma-type is considered to be formed from the most crustally-contaminated Tristan plume-derived magmas.In the southern Parana ´Basin, Gramado is normally found on the flanks of the inland Serra Geral escarpment and at the coast near Torres, whilst Esmeralda is found on the flatter plateau tops.This reflects their general relative positioning within the lava pile as per Fig. 1c (Peate, 1997).Recent studies have further subdivided Gramado into the Torres and Vale do Sol Formations in the modern framework (Rossetti et al., 2018 and references therein).The Gramado and Esmeralda magma-types comprise the Type 1 (Southern) classification from Licht (2018) (<2.85 wt% TiO 2; Table 1).
Rarely found exposed at the surface (i.e.sampled primarily via drilling), the LT Ribeira lavas have low TiO 2 concentrations, but slightly higher Ti/Y than Gramado and Esmeralda (Ti/Y ratio has occasionally been used in place of TiO 2 as a more specialised LT/HT group discrim-inator - Peate et al., 1992;Peate and Hawkesworth, 1996).In contrast, the LT Ribeira ( 87 Sr/ 86 Sr) i signature is more in line with HT magma-types (0.7055-0.7060 compared to 0.7046-0.7167for LT; Peate et al., 1992), but lacks the accompanying HFS and LREE enrichment expected of the HT group.Despite having more in common geochemically with the LT group, the transitional Ribeira lavas are found interbedded with HT lavas in the northern zone (Peate et al., 1992), essentially an intermediate lava overlapping the geochemical properties of both HT and LT magma-types (Fig. 1c).

Silicic Group
Mafic lavas in central and southern Parana ´are often associated with the silicic lavas of the Chapeco ´and Palmas magma-types, predominantly within the bimodality of south-eastern LT group (Fig. 1c) (e.g.Bellieni et al., 1984;Peate et al., 1990;Peate et al., 1992;Garland et al., 1995;Peate, 1997).Chapeco ´lavas are higher in TiO 2 than Palmas, and their relative geographic location mimics that of the mafic HT/LT split, placing Chapeco ´further northwest than its counterpart.The rhyolites, dacites and related alkali complexes of Parana ´have been dated to co-exist with the pre-rift stages of flood basalt volcanism, which further reinforces the complexity of the multi-source melting beneath the province -the silicic lavas are likely derived from mid-to upper-crustal differentiation processes from the source of flood basalt volcanism (e.g.Milner et al., 1995).The Serra Geral silicic rocks have been interpreted to be genetically related to their mafic equivalents via fractional crystallisation -Chapeco ´represents partial melt extraction in HT basalts in the lower crust whilst Palmas represents the differentiation of LT basalts in the upper crust (Garland et al., 1995).Licht (2018) provided distinctions for eight high-Si magma-types with varying Zr, TiO 2 and P 2 O 5 concentrations -Palmas fits within Type 9 and Chapeco ´within Type 14, and they contribute to around 20% of samples classified in the region (Table 1).However, Table 1 Summary of geochemical classification schemes used in this study (Peate et al., 1992;Licht, 2018).The classification criteria for Peate et al. (1992) is referred to in Fig. 2, whilst the geochemical concentrations used for the sixteen types in Licht (2018) are given in this table along with a rough proportion of the total Serra Geral volume.Gra -Gramado, Esm -Esmeralda, Par -Paranapanema.as silicic rocks are very rarely associated with PGE mineralisation, the focus of our paper lies on the mafic magmatypes whilst using these more evolved compositions as reference material for geochemical trends.

Sample acquisition
We present a new suite of major, trace and PGE analyses for 85 basalt and 7 rhyolite samples from the Serra Geral Group.Samples were collected during two field expeditions, one in the north (focusing on HT lavas of Parana Śtate) and one in the south (focusing on LT and Silicic groups of the Santa Catarina and Rio Grande do Sul States).Fig. 1c shows the general sample localities within the context of the Parana ´Basin, with reference to underlying lava group.A complete sample database with field descriptions is provided in Supplementary A1.

Laboratory techniques
Weathered material and amygdales were removed from all samples prior to preparation, as these would interfere with primary geochemical concentrations.Rock samples were crushed to 1-2 mm chips using a jaw crusher before being milled in a chrome-steel TEMA mill to produce fine powders, which were used in all subsequent bulk geochemical analyses.
Major element oxides were measured using X-ray fluorescence (XRF) (after Kystol and Larsen, 1999;Tegner et al., 2009) at Camborne School of Mines, University of Exeter.Dried and ignited samples were fused in a furnace in platinum crucibles with lithium borate flux.Ammonium iodide solution was added to the molten sample to act as a wetting agent when poured into moulds.Cooled samples were analysed at 50-111 mA and 27-60 kV using a Bruker S4 Pioneer XRF spectrometer.Concentrations were corrected to anhydrous values.The AGV-1, BHVO-2, BIR and DNC-1 standards were used in analyses and reported alongside new data.
Trace elements were measured using inductively coupled plasma mass spectrometry (ICP-MS) (after McDonald and Viljoen, 2006), also at Camborne School of Mines.Powdered samples were dissolved using 4-Acid digestion technique (nitric, perchloric, hydrofluoric and hydrochloric acids), and elemental concentrations were detected by an Agilent 7700 Series mass spectrometer.The BCR-2 and Bir-1a standards were used in analyses and reported alongside new data.
Six PGE (Os, Ir, Ru, Rh, Pt, Pd) plus Au were measured using Ni-S fire assay followed by tellurium co-precipitation and ICP-MS analysis at Cardiff University (using methods developed by Huber et al., 2001;McDonald and Viljoen, 2006).A fifteen gram aliquot of each powdered sample was mixed with 12 g of borax flux, 6 g NaCO 3 , 0.9 g solid sulfur, 1.08 g Ni and 1 g silica, before being melted in a furnace at 1000 °C for 1.5 hours.The sulphide bead separated from the cooled melt was removed, dissolved in hydrochloric acid, co-precipitated with Te, filtered and diluted before element detection using a mass spectrometer.It should be noted Os was not measured for the northern sample set due to the lack of volatilisation-free methods at the time of analyses; samples analysed at a later date were measured for Os given the advancement of methodological accuracy.The TBD1 and WPR1 standards were used in analyses and reported alongside new data.Detection limits (i.e.lowest recorded blank values) were 0.0052 ppb for Os, 0.0026 ppb for Ir, 0.0603 ppb for Ru, 0.0152 ppb for Rh, 0.1458 ppb for Pt, 0.0662 ppb for Pd and 0.0070 ppb for Au.
Full raw major, trace and PGE-Au data, plus standard measurements for each method (with published values) and detection limits for PGE and Au are provided in Supplementary B. An amalgamated data sheet with duplicate measurements is available in Supplementary C. We acknowledge that chrome-steel mills can introduce small amounts of metal contamination to samples and although this is expected to be minimal and non-intrusive to our results (Evans et al., 2013), we have supplied quartz sand procedural blank measurements for PGE analyses alongside Supplementary B, processed identically to our basalts.
We have used a combination of two classification schemes alongside sampling locations in order to utilise the wealth of literature already available describing the characteristics of lavas belonging to each magma-type.In this way, we can contextualise subtleties in the dataset according to PGE and Au and therefore metallogenic processes in the new precious metal data set.Accordingly, lava samples have been classified into their classic magma-types (except Ribeira, which was not sampled) based on bulk concentrations of MgO and TiO 2 , plus Ti/Y and Ti/Zr ratios (established by Peate et al., 1992) and MgO vs. Zr, with literature data used for comparison (Fig. 2a-c; Supplementary A2).Classifications from Licht (2018) are also given for each sample (denoted by symbology), based on Si, Ti, Zr and P concentrations.The inherent issue with classifying samples based on limited bivariate relationships is evident -samples SG11-12, SG56 and SG78-79 do not conform to the Peate et al. (1992) classification for example, and thus their designation was assessed subjectively based on adjacent literature data clouds for the purpose of this study; Licht (2018) classifications are wider, but more robust.
The sample set is also classified using a total alkali vs. silica (TAS) plot in Fig. 2d.The mafic samples are predominantly basalts to basaltic andesites, with some Gramado samples classifying as andesites.Palmas samples plot within the dacitic field, while the two Chapeco ´samples plot with trachytes.All Parana ´lavas analysed fall within the subalkaline to transitional alkali-tholeiitic fields.

Major element oxides
Serra Geral basalts tend to have MgO concentrations consistently between 2 and 7 wt%; silicic lavas plot between 1 and 2 wt% MgO.Concentrations of TiO 2 highlight the established literature magma-types ->3 wt% for Urubici/ Pitanga or Type 4, 1.5-2.5 wt% for Paranapanema or Type 1 (Central-Northern), and < 1.5 wt% for Gramado/Esmeralda or Type 1 (Southern), plus the silicic lavas (Fig. 3a).All basic lavas exhibit Fe 2 O 3 between 10 and 16 wt% (Fig. 3b), with HT samples generally higher than LT.In Fig. 3c, LT lavas exhibit higher Al 2 O 3 concentrations than HT lavas (the opposite to Fe 2 O 3 ), with a slightly positive relationship with MgO; MnO concentrations follow this same slight positive relationship but between 0.15 and 0.35 wt% (Fig. 3d).A distinctly positive correlation between MgO and CaO (Fig. 3e), and negative correlations between MgO and Na 2 O (Fig. 3f), and MgO and K 2 O (Fig. 3g) do not show clear HT-LT distribution patterns.Finally, P 2 O 5 exhibits a bifurcated negative trend with MgO, with a shallow slope for LT and silicic lavas, and a steeper slope for HT lavas excluding Paranapanema, which follow LT trends (Fig. 3h).Peate (1997).Dashed lines for approximate Type 1 and 4 division (2.85 wt% TiO 2 ) from Licht (2018).Arrows refer to approximate development trajectory.In the key, dashes indicate that no data was available for that particular magma-type either in this study (large coloured markers) or the literature (small greyscale markers).Literature database with references is provided in Supplementary A2. S -Southern, C-N -Central-Northern.

Base metals
Trends in base metal enrichment are limited for the Serra Geral data set, with distinctively low overall Ni (<140 ppm; Fig. 4a) and Cr (<160 ppm; Fig. 4d) contents compared to other CFB provinces (e.g. the NAIP where onshore lavas generally exhibit Ni concentrations of 100-1,000 ppm and Cr concentrations of 100-2,000 ppm; Hughes et al., 2015).The Paranapanema samples display positive correlations between MgO and Ni, Co and Cr concentrations (Fig. 4a, c, d) that have a similar slope but higher concentrations than Gramado and Esmeralda samples; Urubici and Pitanga samples lack strong trends in all instances.Enrichment of Ni, Cr and Co (Fig. 4c) follow very similar trends to Fe 2 O 3 relative to MgO (Fig. 3b), with HT lavas generally containing slightly higher concentrations.Overall, Cu concentrations are better clustered into locality classifications than Ni, Cr or Co, with Paranapanema lavas exhibiting the highest mean Cu content at 254 ppm and Esmeralda exhibiting the highest singlesample concentrations (Fig. 4b).Unlike the Ni and Cr, concentrations of Cu and Co (Fig. 4b and c) are comparable to NAIP onshore lavas from Hughes et al. (2015), at around 10-500 ppm for Cu and 10-70 ppm for Co.

Rare earth elements and lithophiles
Chondrite-normalised (McDonough and Sun, 1995) REE plots for all Serra Geral magma-types have fractionated REE patterns, with LREE > HREE in all samples (Fig. 5a-f).Steeper gradients are observed in the HT lavas (Fig. 5a-c), with distinctive light REE enrichment especially in Urubici (Fig. 5a) and Pitanga (Fig. 5b); these two magma-types also exhibit parallel arrays, and all HT lavas have a narrower spread of concentrations.Paranapanema lavas (Fig. 5c) feature similar enrichment patterns to the LT group, not their designated HT group.The LT REE arrays (Fig. 5d-e) have more variable LREE concentrations than HT lavas, particularly in the more incompatible REE such as La, Ce, Pr, Nd and Sm.Esmeralda is the most depleted magma-type in all element concentrations (Fig. 5d).Silicic and LT lavas have prominent Eu troughs, with the Silicic group generally showing more enriched concentrations of all elements, particularly Chapeco ´(Fig.5f).
The three HT magma-types exhibit similar normalised incompatible element patterns to one another and are enriched in Ba, Nd, Hf and Zr, and depleted in Sr and P compared to the LT samples (Fig. 7a-c).Esmeralda (Fig. 7d) and Gramado (Fig. 7e) samples exhibit a wide range of multi-element normalised signatures, primarily lower than all other magma-types, with Ba, Nb, Sr, P and Ti negative anomalies and depletions in the most incompatible elements including Hf, Zr, Sm and Tb in particular.Chapeco ´and Palmas have similar normalised incompatible element enrichment trends, with pronounced Nb, Sr and Ti negative anomalies, and high Rb, Th, U, K, Hf and Zr concentrations (Fig. 7f).Chapeco ´features slightly higher concentrations overall and a positive Ba anomaly, whilst Palmas features a negative one.

PGE and Au
Chondrite-normalised PGE and Au plots (McDonough and Sun, 1995) demonstrate a relative enrichment of Pdgroup PGE (PPGE; Rh, Pt, Pd) and Au compared to Irgroup PGE (IPGE; Os, Ir, Ru) in all Serra Geral magmatypes (except Palmas, which was not analysed) (Fig. 8a-f).However, the degree of enrichment varies significantly across the sample set.This is most obvious between Serra Geral HT and LT groups, with HT exhibiting a steep slope from IPGE to PPGE (Fig. 8a-c) and the LT showing normalised IPGE concentrations up to an order of magnitude higher than HT samples, producing a flatter gradient into PPGE (Fig. 8d-e).Most importantly, Serra Geral LT samples, or Type 1 (Southern) samples, frequently display an unusual, distinctive 'hump' between Ru, Rh and Pt concentrations and significant Pt, Pd and Au depletions compared to the gradual slope seen in the HT plot (with particularly high Pd and Au).Paranapanema lavas, or Type 1 Central-Northern, have the highest Pd concentrations as shown by up to 0.1 Â chondritic values (compared to 0.0 01-0.01Â chondrite for other lavas) in Fig. 8c.
The highest concentrations of Ir, Ru and Rh (Fig. 9b-d) are found in LT rocks (around 1.2, 13.5 and 5.75 ppb, respectively), while Pt is most enriched in LT and Paranapanema (up to 30-40 ppb; Fig. 9e).Paranapanema lavas are distinctly Pd-rich in most instances (ranging 5-35 ppb compared to 1-15 ppb in all other lavas; Fig. 9f).Although Os concentrations are unavailable for HT lavas, LT lavas contain concentrations of a similar level to Ir in the same rocks (up to 1.2 ppb; Fig. 9a).Paranapanema lavas have the lowest Cu/Pd of the PELIP lavas (Fig. 10a), similar to the majority of NAIP lavas.Fig. 10b The IPGE/ PPGE enrichment split between LT and HT is such that LT lavas tent to have the lowest Pd/Ir ratios (Fig. 10b), in a similar range to most NAIP lavas from Hughes et al. (2015).Urubici and Pitanga magma-types evidently contain less PPGE compared to the LT samples in Fig. 10b.Fig. 10c and Fig. 10d compare PGE ratios to trace element signatures, to help characterise the chalcophile trends alongside magmatic differentiation trends from earlier figures.´.Grey shaded area -Urubici magma-type for reference.S -Southern, C-N -Central-Northern.

Magmatic trends
Our results show important geochemical trends across both the Peate et al. (1992) and Licht (2018) classification schemes that help contextualise precious metal trends, summarised and interpreted herein.Low MgO content (2-7 wt %; Fig. 3a-h) and Ni concentrations (<125 ppm; Fig. 4a) across all Serra Geral lavas implies that they are not primary mantle melts and have undergone significant differentiation, likely via fractional crystallisation (Xiao et al., 2004).Mineral fractionation vectors in Fig. 3a-h strongly suggest that olivine, clinopyroxene and plagioclase (plus minor amounts of spinel-group minerals) were progressively removed from the parent magmas leading to  (Fodor, 1987;Peate et al., 1990).As observed in thin section, the dominant mineral in all mafic samples is orthopyroxene, with clinopyroxene, plagioclase and spinel-group minerals, and a notable absence of olivine (typically 0-5% of assemblages), likely due to the earlier removal.Most major elements exhibit a single fractionation trend for each magma-type with a possible slight change in slope and groupings below ~4 wt% MgO, indicating that the fractionating mineral(s) were reasonably consistent for mafic lavas.For TiO 2 and P 2 O 5 (Fig. 3a and h), the fractionation trends  (Spandler and O'Neill, 2010).Considering the Serra Geral mafic lavas are particularly olivine-deficient in thin section and reasonably evolved with respect to SiO 2 and MgO concentrations (45-60 wt% and 2-7 wt%, respectively; Fig. 2d and 3a-h), it is likely that a significant amount of compatible base metals have been removed from the parent magmas prior to eruption in all magma-types.In a similar manner, Cr is depleted in lavas throughout the region, which we attribute to the removal of both olivine and chrome spinel.Copper appears to be the only base metal/chalcophile retained in the lavas in any significant amount (Fig. 4b), and thus is neither strongly compatible nor incompatible in the fractionating silicate and oxide mineral phases.If sulphide fractionation drove chalcophile depletion, Cu would be uniformly removed from melts given its strong affinity for sulphide liquids in silicate melts (D sulphideÀmelt Cu = 1,470; Mungall and Brenan, 2014).Since this is not the case, we propose that large degrees of olivine and chromite fractionation experienced by such MgO-depleted magmas would effectively remove Ni, Co and Cr while leaving Cu in melts.
Trace element concentrations and ratios provide evidence for markedly different geochemical processing between magma-types within the Serra Geral Group.Some Type 4 lavas feature as much as an order-of-magnitude enrichment in LREE compared to Type 1 lavas (particularly Southern), indicating differing degrees of partial melting or assimilation of continental crust (Fig. 5a-f).Combined with the trends of HT lavas towards OIB signa- tures in Fig. 6b and high-Ti melts akin to Karoo CFBs in Fig. 6c, we suggest Type 4 represents more strongly plume-related magmas in agreement with Ra ¨mo ¨et al. ( 2016), who attribute HT lava 87 Sr/ 86 Sr and eNd signatures to an enriched mantle (EM1) component.Given that the HT/Type 4 lavas align with the plume focus ca.135 Ma (Beccaluva et al., 2020), their signatures in our study signify the strongest asthenospheric signal.
More pronounced negative Eu anomalies in the LT group than HT lavas (Fig. 5) indicate that a larger amount of plagioclase fractionated from the parent magma, signalling a higher degree of crystal fractionation (FC) in this group compared to HT.The LT lavas, particularly Gramado, exhibit decreasing Nb/La with increasing SiO 2 (Fig. 6a), indicating the role of assimilation and fractional crystallisation (AFC) in the parent magmas.The depletion in trace element signatures and strongly negative Ta-Nb anomalies (Fig. 7d-e) indicate that contamination by continental crust material during magma ascent may have been important in their geochemical development, as widely documented by isotopic studies of the LT group (e.g.Peate et al., 1992;Rocha-Ju ´nior et al., 2013;Natali et al., 2017;Marques et al., 2018).The Ta-Nb anomalies in HT lavas (Fig. 7a-c) are significantly less pronounced, indicating a lower degree of crustal contamination in these magmas.Chapeco ´and Palmas typically trend with HT and LT, respectively, in Fig. 5f, 6a-d and 7f, suggesting they share a common source and perhaps evolved towards silicic compositions from their associated mafic magma-types.
Trace element observations provide an initial indication, prior to quantitative modelling, of melting conditions for the Serra Geral magmas.The steeply negative medium to HREE-LREE trends for Type 4 multi-element patterns in Fig. 5a-b, along with Type 4 samples trending towards high (Sm/Yb) n values in Fig. 6d suggest low degrees of partial melting and HREE retention in residual garnet.This could imply high pressure melting conditions commonly associated with plume melts beneath thick continental cratons, where the majority of the melting occurs in the garnetbearing asthenosphere (e.g.Arndt et al., 1998;Jourdan et al., 2007).In comparison, Type 1 lavas are strongly associated with trace element trends produced by the melting of spinel peridotite (Fig. 6d) (e.g.White and McKenzie, 1995), which corresponds with their shallow medium to HREE slopes in Fig. 5c-e.Paranapanema lavas trend with Type 1 (Southern) towards spinel peridotite melts (Fig. 6d), exhibit medium to heavy REE slopes between those Type 1 (Southern) and Type 4, but are grouped with Type 4 OIB sources in other discriminant diagrams (e.g.Fig. 6c) and as such must be regarded separately in terms of source interpretation, as in Licht (2018).
The spinel to garnet transition occurs where pressures and temperatures exceed 3 GPa and 1,570 °C (Milholland and Presnall, 1998;Walter et al., 2002).This translates to depths of 50-70 km in asthenospheric regions and 120-180 km in colder sub-cratonic settings (Kent and Fitton, 2000;Ziberna et al., 2013).In Cr-rich mantle sources or depleted rocks such as harzburgite, the transition can occur at significantly higher pressures (up to 10 GPa) and typically increases the depths at which spinel and garnet are both stable simultaneously (Klemme, 2004;Ziberna et al., 2013).The different spinel/garnet melting trends exhibited by the Serra Geral lavas reflect either 1) different degrees of melting of a garnet-bearing mantle source (with garnet exhaustion in the most extreme cases) or 2) separate melting sources and depths (between 50 and 100 km; Gibson et al., 2006) as a function of the rapidly evolving geodynamic setting.Both scenarios would require a change in melting depth between magma-types, to facilitate differential degrees of partial melting (Beccaluva et al., 2020), and polybaric melting processes are favoured by Ra ¨mo ¨et al. (2016) in their assessment of Pitanga and Paranapanema geochemical differences.We model the trace element geochemistry of Serra Geral lavas, including chalcophile elements, in Section 4.2.3, after firstly interpreting PGE enrichments in relation to partial melting information.

PGE enrichment
Trends in PGE enrichment fall consistently into the three major mafic subdivisions established by Licht (2018) in Figs. 8 and 9 -Pitanga and Urubici in Type 4, Paranapanema in Type 1 (Central-Northern), and Gramado and Esmeralda in Type 1 (Southern).As such, it is most appropriate to primarily use these newer classifications in lieu of the classic groups when discussing metallogenesis going forward.
Type 1 (Southern) lavas are more enriched in IPGE than all other lavas (Fig. 8a-f).This is especially true for Ru concentrations, which are an order of magnitude higher than typical intracontinental lavas (Barnes et al., 2015).Furthermore, Rh, the transitional PGE that shares properties with both IPGE and PPGE (Barnes et al., 1985;Holwell and McDonald, 2010 and references therein), is also enriched in Type 1 (Southern) lavas, forming a 'humped' shape on their plots in Fig. 8d-e.Contrasting this trend to the notable Pd and Au enrichment (and steep slope in the multi-element diagram) in the Type 1 (Central-Northern) lavas (Fig. 8c) and general depletion in PGE in Type 4 lavas (Fig. 8a-b), it is clear that PGE signatures are heterogeneous in the region.The discrepancy with depleted Ni and Co, and relatively enriched Cu in Fig. 4a and b supports preferential fractionation of Ni and Co via olivine, in the absence of significant Cu depletion via sulphide removal.As noted in Section 4.1, we suggest that sulphide fractionation is unlikely to control base metal and PGE concentrations in the PELIP magmas.Thus, we must identify different geochemical sources outside sulphide depletion to explain the separate Ir-Ru-Rh and Pd-Au-Cu enrichment associations in Type 1 lava varieties that make sense when integrated with their major and trace element interpretations.

Palladium in Type 1 (Central-Northern) lavas
The Pd-Au-Cu enrichment in Type 1 (Central-Northern) samples is clearly illustrated by their steep multi-element slope in Fig. 8c and high Pd/Ir ratios in Fig. 10b, which place the lavas as 1-2 orders of magnitude more Pd-rich than all other lavas in this study.These lavas often contain up to 30 ppm more Pd than any other group (Fig. 9f), particularly Type 4, and contain slightly higher mean Cu and Au concentrations than others (Fig. 4b and 10f).Given that Pd, Cu and Au are all mobile incompatible chalcophiles commonly associated with each other in fluid-rich environments (e.g.Holwell et al., 2019;Choi et al., 2020 and references therein), we propose that Pd, Cu and Au-bearing material is being preferentially incorporated into Type 1 (Central-Northern) melts.This supports Type 1 (Central-Northern) as a separate compositional group by Licht (2018) despite earlier classification alongside Urubici and Pitanga by Peate et al. (1992); the fundamental geochemical controls on Si, Ti, Zr and P in Serra Geral magmas established from this study evidently have an impact on PGE concentrations.Enrichment in Pd-Cu-Au is unique to Type 1 (Central-Northern) lavas in the PELIP and thus a different partial melting model is required to generate the compositional variances from other samples.If, as suggested by Rocha-Ju ´nior et al. ( 2013), the significant stratigraphic and spatial distance between the eruption of individual magma-types represents physically distinct sources in the mantle, we would expect each of the parental magmas to feature their own set of geodynamic controls on geochemistry.
The three metals enriched in Type 1 (Central-Northern) lavas -Pd, Au and Cu -are notable for being more incompatible in silicate melting environments than the other five PGE and base metals (e.g.Mitchell and Keays, 1981;Borisov et al., 1994;Holzheid et al., 2000;Righter et al., 2008;Tassara et al., 2017).Furthermore, Pd does not form micronuggets like the other PGE, tending to exist in base metal sulphides alongside Cu and Au (Holzheid et al., 2000).Metasomatic activity can easily redistribute such metals via fluids and melts at the upper mantle and base of the lithosphere in sub-cratonic settings over long periods of time (e.g.Zhang et al., 2008;Maier and Groves, 2011).Given that the Parana ´Basin exists on Proterozoic Gondwanan basement, the SCLM beneath modern day Brazil will have been subjected to numerous fertilisation events (Rocha-Ju ´nior et al., 2013) via subduction around the cratonic margin (Tassara et al., 2017).Subducting oceanic crust can release mobile elements in dehydrating fluids or small-degree melts as the plate descends, and these become concentrated in a mantle wedge and SCLM (e.g.Woodland et al., 2002;Griffin et al., 2013;Tassara et al., 2017;Rielli et al., 2018;Wade et al., 2019), similar to the origins of Cu-porphyry environments.Metasomatic sulphide minerals, and by extension chalcophile elements, can then be scavenged by melts generated in an intracontinental setting (Powell and O'Reilly, 2007).Generally, Pd will follow Cu during partial melting of the mantle and can be found in similar sulphide phases at upper mantle P/T conditions (e.g.Barnes et al., 1997;Mungall et al., 2005;Lorand et al., 2013 and references therein).If Cu-and Pd-bearing sulphides are melted and incorporated into magmas interacting with the SCLM, the two metals can share a coupled enrichment in resulting volcanic and intrusive products (e.g.Bockrath et al., 2004;Alard et al., 2011;Delpech et al., 2012;Lorand et al., 2013;Lorand and Luguet, 2016;McDonald et al., 2017;Tassara et al. 2017).
An alternative theory is that whilst the SCLM still received enrichment via metasomatic processes, subduction did not supply the metals.In Fig. 11a-b, relative PGE concentrations are compared to classic sub-lithospheric rocks from the literature, pyroxenite xenoliths from Cameroon (Abeng et al., 2012).Here, Type 1 (Central-Northern) samples plot far from classic intraplate xenoliths and instead are much more alike the pyroxenites.Thus, the enrichment of PPGE in the parent magma could have resulted from pyroxenite (i.e.recycled oceanic crust components) in the melt column and/or SCLM.Given the long history of the Gondwanan SCLM likely including many metasomatic and partial melting events, and the tendency of mantle plumes to sample both pyroxenite and peridotite from the deep mantle, this is plausible, yet would require Osisotope data to accurately confirm.Regardless of mechanism (subduction or mantle component-driven), the Type 1 (Central-Northern) parent magmas require enrichment in PPGE, Cu and Au prior to ascent via a metasomatic event to produce such PGE signatures.Ru/Rh, demonstrating the differences between the Serra Geral lava PGE ratios, with comparison data for intraplate xenoliths (Maier et al., 2012), Iceland (Momme et al., 2003) and Cameroon pyroxenites (Abeng et al., 2012).
In a study comparing intracontinental melts from depleted asthenospheric and metasomatised SCLM sources in the North China Craton, Wang et al. (2020) determined that the latter were 3-4 times more enriched in precious and base metals, likely via efficient extraction of SCLM sulphides during melting.Furthermore, spinel lherzolite xenoliths in intracontinental lavas from Ethiopia (Lorand et al., 2003), showing evidence of metasomatic deformation and melt percolation, acquired a characteristic enrichment in S-Cu-Pd-Au.Given the unique geochemical characteristics of Type 1 (Central-Northern) lavas for the PELIP, our metallogenic model must provide a mechanism for differentiating the magma-type from the others, particularly the Type 1 (Southern) group given that they melt at a similar depth according to garnet-spinel information in Fig. 6d.We propose that given the documented metasomatic association of Pd and Cu (e.g.Hughes et al., 2014;Tassara et al., 2017;Holwell et al., 2019;Choi et al., 2020) and potentially Au (although this is less constrained; Maier et al., 2012), and the consistently higher concentrations in shallowerderived continental melts from the PELIP (e.g.potentially involving more SCLM than asthenosphere), the Type 1 (Central-Northern) lavas obtained elevated concentrations of Pd, Cu and Au by the passive melting of metasomatised SCLM by the Tristan plume.Indeed, studies of Brazilian mantle xenoliths indicate a protracted metasomatic history beneath the craton (e.g.Rivalenti et al., 2000;Carlson et al., 2007), supporting this hypothesis.

Ruthenium and rhodium in Type 1 (Southern) lavas
The plateau between Ru, Rh and Pt in Type 1 (Southern) samples (Fig. 8d-e) likely reflects the source minerals included in parental melts.This plausibly indicates that a Ru-Rh-bearing mineral(s) is being incorporated into Type 1 (Southern) magmas from a mantle source in a separate system from other magmas.The melting depth and source for Type 1 magmas are different to Type 4 (Ra ¨mo ¨et al., 2016;Licht, 2018) (Fig. 6d), implying distinct melting processes in the northern and southern basalts (Beccaluva et al., 2020) that could explain the groups' differing metallogenic characters.Overall, the PGE behave in a similar manner chemically -in the presence of a sulphide melt portion, all six will be overwhelmingly concentrated within it rather than the silicate melt.However, the individual partition coefficients of the PGE between silicate minerals and melt vary significantly when not in the presence of sulphides (e.g.Peach et al., 1990;Peach et al., 1994;Righter et al., 2004;Park et al., 2017 and references therein), and the melting of a sulphide-deficient mantle will facilitate the heterogeneous incorporation of PGE into melts.It follows that the enrichment and depletion of PGE in Serra Geral lavas must be controlled by silicate or oxide differentiation processes in the magma.If the melts were depleted in sulphide, the majority of their PGE basket would also be removed and they would have higher Cu/Pd than what is displayed in Fig. 10a, more common in NAIP magmas.To add Ru and Rh to Type 1 (Southern) in the quantities observed in Fig. 8d-e and 9c-d would require melting of a source rich in these PGE alone, or at least in higher proportions than the other PGE.
Type 1 (Southern) and (Central-Northern) lavas follow a liquid line of descent commonly associated with the melting of spinel peridotites at shallower depths and lower pressures than Type 4 magmas (Fig. 6d).Experimental studies found Ru and Rh to be compatible in spinel-group minerals (Capobianco and Drake, 1990;Barnes and Picard, 1993;Pitcher et al., 2009), with spinel-silicate melt partition coefficients of 400-900 (Righter et al., 2004).This synchronises well with our trace element ratios (Fig. 6) -by melting a source with normative spinel-group mineral content, the generated magma would contain relatively high concentrations of the two PGE.In the mantle, Os, Ir, Ru and Rh are also highly compatible in sulphides, platinum-group minerals (PGM) and alloys, different mineral phases from the base metal sulphides, tellurides and arsenides that Pt and Pd are commonly found in (e.g.Mitchell and Keays, 1981;Alard et al., 2000;Helmy and Bragagni, 2017;Helmy and Fonseca, 2017).As such, the variety of sulphide compositions in the mantle source can be a crucial factor in fractionating the PGE from each other and isolating an IPGE and Rh rich mineral phase (i.e.Ni-sulphides, alloys or even spinel) for preferential melt incorporation (e.g.Keays et al., 1982;Barnes and Picard, 1993;Rehka ¨mper et al., 1997;Maier et al., 2003).A combination of spinelgroup minerals and IPGE-sulphides at source may provide the necessary PGE signatures to Type 1 (Southern) melts.However, Type 1 (Central-Northern) sample trace element signatures also indicate spinel peridotite as a melting source in Fig. 6d, cluster with Type 1 (Southern) in Fig. 6b, and sit distinct from the other magma-types in the Licht (2018) study.Type 1 (Central-Northern) lavas do not have the characteristic Ru-Rh enrichment in Fig. 8c and 9c-d.This suggests that although all Type 1 lavas likely formed by the partial melting of spinel peridotite mantle, the melting processes for the Southern and Central-Northern varieties must have been different to incorporate a different suite of PGE, or the mantle sources being melted exhibited different metasomatic characters.

Modelling melt compositions
Based on the interpretations of all Type 1 lava PGE signatures, we forward-modelled melt compositions appropriate for melting of primarily spinel peridotite in a sublithospheric setting, using mass balance calculations from Shaw (1970) and experimental melting mode values from Johnson (1998) (Fig. 12a-b).Mineral-melt partition coefficient sources are given in Fig. 12 and Table 2. Starting with bulk concentrations of Co, Ni, Cu, PGE and Au from i) metasomatised arc, ii) normal arc, iii) cratonic, and iv) off-craton peridotite xenoliths collated by Barnes et al. (2015), and the mineral mode (X) and melt mode (p) values displayed in Table 2, we successfully generated chondritenormalised multi-element PGE plots representative of the Serra Geral Type 1 magmas (summarised in Fig. 12; see Supplementary E for further details).Metasomatised and normal arc xenoliths represent re-enriched and depleted upper mantle sources in our Type 1 (Central-Northern) models, respectively.On-craton xenoliths are generally from kimberlites within the thickest area of a craton block whilst off-craton are from kimberlites in thinner mobile belts surrounding cratons (e.g.Wildman et al., 2017); these represent a progressively thinning intracontinental setting in our Type 1 (Southern) models.
An important caveat for our models is that strongly chalcophile elements (such as the PGE) will preferentially stay in sulphide minerals until the phase is melted entirely, and as such, sulphide melting strongly controls the concentration of PGE in the resulting magmas.In order to attain PGE concentrations contained in any of our samples (between 0.0001 and 0.1 Â chondritic; Fig. 8a-f), the sulphide phase, if present in the source (~1 wt% in our model), must be exhausted.Similarly, for IPGE and Rh compatible in spinel, this mineral phase must be entirely melted to release the metals into magmas.Given the spinel-melt partition coefficients for IPGE are often 3-4 orders of magnitude smaller than those for sulphides (Table 2), the effect on resulting concentrations pre-exhaustion is much less pronounced.Further, it must be noted that chalcophile sulphide-melt partition coefficients can be highly variable from experiment to experiment, so we have maintained those found by Mungall and Brenan (2014) for consistency whilst acknowledging the challenges of modelling sulphide melts.PGE, Cu, Co and Ni concentrations progressive melting are given in bivariate plots in Supplementary E for each scenario, in support of Fig. 12.
To generate Type 1 (Central-Northern) magmas, our models (Fig. 12a) show that elevated bulk Cu, Pt, Pd and Au concentrations in the source strongly influence the resulting melt enrichment.Metasomatised continental arc xenoliths (as in Wilson et al., 1996) typically feature such enrichment of Cu, Pt, Pd and Au.Beneath ancient cratons, the SCLM can be pre-conditioned by repeated subduction events throughout history (e.g.Mitchell and Keays, 1981;Powell and O'Reilly, 2007;Zhang et al., 2008;Hughes et al., 2014;Hughes et al., 2015;Holwell et al., 2019).The model predicts that at melting degrees of around 20%, metasomatic sulphides in an arc-like mantle will be exhausted at the conditions from Table 2, releasing chalcophile elements into the melt ('metasomatised arc' in Fig. 12a).Identical melting conditions with a less Pt, Pd, Au and Cu-enriched source would produce chondritenormalised plots with shallower gradients ('normal arc' in Fig. 12a, also from Wilson et al., 1996), indicating that bulk source composition is the key factor in melt enrichment.Further, a mantle source without sulphide in the mineral mode cannot sufficiently produce melts with elevated Pt, Pd or Au concentrations, as the elements in question do not partition strongly into the remaining silicate or oxide mineral phases.It should be noted that PGM and alloys are also known to host small amounts of the SCLM's PGE budget (e.g.Hughes et al., 2015;Gonza ´lez-Jime ´nez et al., 2020), but our model does not account for these phases.If Brazilian SCLM does host Pd-bearing PGM this would only work to boost our model further and reinforce the idea of metasomatically-enriched Pd-Au-Cu SCLM components.
Cratonic (Maier et al., 2012) and off-cratonic (Schmidt et al., 2003) peridotite xenoliths have a more even distribution in PGE concentrations, but in most cases are enriched in Ir, Ru and Rh, and depleted in Pt, Pd and Au with respect to arc sources (Table 2).Our models predict that once sulphides are exhausted, the resulting on/off craton melts produce nearly flat chondrite-normalised PGEenrichment patterns (Fig. 12b), but not the Ru-Rh 'hump' from Fig. 8d-e.However, given that evidence shows that Type 1 lavas were derived from melts nearer the locus of lithospheric rifting (Beccaluva et al., 2020), we assume a higher-degree (i.e 25%) melt fraction in the off-craton setting (i.e.thinner lithosphere, shallower/higher degree partial melts) as a best fit the PGE patterns in our Gramado and Esmeralda samples.This allows more spinel to be melted which increases Ir, Ru and Rh concentrations to match Type 1 (Southern) patterns (Fig. 8d-e).In the absence of sulphide, our model can produce similar PGE enrichment curves, implying that these lavas, unlike Type 1 (Central-Northern), do not rely on metasomatic sulphides in the melting system.Elevated source bulk concentrations in the relevant metals are the key driver in resulting melt composition in all models, provided the host phase can be sufficiently melted.
Our modelling demonstrates that Type 1 (Central-Northern) magma compositions are best explained by a mantle source region with metasomatic sulphides which become exhausted during partial melting, while Type 1 (Southern) magmas can be generated by higher degrees of partial melting to exhaust both sulphides and spinel-group minerals (if sulphides were present at all).Locally heterogeneous SCLM beneath Brazil (Fodor et al., 2002;Carlson et al., 2007), as exists in other CFB provinces worldwide (Karoo in Jourdan et al., 2007;Emeishan in Song et al., 2008;the North Atlantic in Hughes et al., 2015;and South Australia Wade et al., 2019), can generate diverse PGE signatures in melting signatures.

Melt components and the role of the SCLM in CFB metallogenesis
A unifying metallogenic model for Serra Geral must incorporate aspects of geodynamic, mineralogical and melting models to explain major and trace element signatures in the region.The variable enrichment and depletion of PGE and Au in Parana ´CFB magma-types immediately characterises Type 4 (the highest-Ti magma-type of the HT group) as relatively deficient in all seven precious metals.In contrast, the Type 1 mafic magma-types contain the modelled enrichments in two different sets of PGE/chalcophiles -  Johnson (1998) and Kinzler (1997), with addition of sulphide.b Dvalues for olivine from Brenan et al. (2003) and Spandler and O'Neill (2010).c D-values for orthopyroxene from Beattie et al., (1991) and Liu et al. (2014).d D-values for clinopyroxene from Hill et al. (2000) and Righter et al. (2004).e D-values for spinel from Liu et al. (2014), Park et al. (2017) and Righter et al. (2004).f D-values for sulphide from Mungall and Brenan (2014).g Arc xenolith concentrations from Wilson et al. (1996).h Craton xenolith concentrations from Maier et al. (2012).I Off-craton xenolith concentrations from Schmidt et al. (2003).Italicised values were unavailable in the literature for silicates, but assigned neutral values based on olivine values, given their lesser significance to melt composition compared to spinel and sulphide.

Setting
Mode (X) Pd, Au and Cu in Type 1 (Central-Northern), and Ru and Rh in Type 1 (Southern).The concentrations can be explained by variable element compatibility and mineral assemblages expected in the melting region (Fig. 12a-b), which in turn implies changing conditions beneath the Tristan plume.Between the generation of Type 4 and Type 1, the mantle source has changed to distinctly spinel lherzolite (Fig. 6d) (in agreement with Ra ¨mo ¨et al., 2016).This indicates a progressive shift to shallower melting depths, which may have allowed higher degrees of partial melting (i.e.20-25%, as modelled in Fig. 12a-b), permitting access to different zones and/or greater amounts of the SCLM as a melting source.We propose a multi-component metallogenic model (Fig. 13) to describe the incorporation of variable PGE assemblages throughout the geochemical magma-types in the Parana ´CFB, in line with the region's geodynamic development in the Cretaceous (135-128 Ma; Fig. 1a-b) and modelling from Fig. 12.
Firstly, we must establish the evolving geodynamic setting in Cretaceous Brazil as the mechanism behind the differing PGE enrichment.Prior to extension, a thick continental lithospheric 'lid' suppressed the degree of melting beneath the Gondwanan continent and confined CFB melt generation to deeper in the upper mantle (i.e. at the plume head), firmly within the upper mantle garnet lherzolite stability field (as in Arndt et al., 1993;White and McKenzie, 1995;Turner et al., 1996;Kerr, 1997).Following lithospheric thinning as a consequence of thermal weakening by the underlying mantle plume and the onset of continental rifting from the south-east ca.134 Ma, the underlying mantle decompressed and enabled partial melting at shallower depths i.e. within the spinel lherzolite stability field (e.g.Ellam, 1992;Stein and Stein, 1992).This thinning would be most prevalent in south-eastern Parana ´, near the future Atlantic coast, forming geographicallybound differences in lithospheric thickness and melting depth (Turner et al., 1996), decreasing from the north-west (HT lavas) to the south-east (LT lavas) (Beccaluva et al., 2020).This echoes the geodynamic setting of the Palaeogene proto-Icelandic rift-plume system in the North Atlantic -magmas from under the intact Greenlandic lithosphere and magmas at/near the developing Atlantic rift zone record different depths of melting and markedly different melt geochemistry as a result (e.g.Saunders et al., 1997;Kent and Fitton, 2000;Hughes et al., 2015;Lindsay et al., 2020).The transition from thick North Atlantic Craton to rifted lithosphere resulted in high-then low-Ti continental magma generation (Momme et al., 2006), reflected in their apparent depth and degree of melting, and garnet/spinel source signatures.In the case of Parana ´, we propose that partial melts that are generated at the garnet-spinel transition as a function of lithosphere extension can directly control the variable metal baskets in the Serra Geral magma-types as modelled (in line with Kerr, 1995;Klemme and O'Neill, 2000;Gibson et al., 2006).Although Serra Geral lavas generally follow the classic stratigraphy of Gramado to Esmeralda and Palmas (with minor Urubici intercalated) in the south and Pitanga to Paranapanema and Chapeco ´in the north (Fig. 1c), it must be reiterated that the lavas are synchronous across the Parana ´Basin.Evidence for synchronicity includes mixing between Esmeralda and Pitanga (De Min et al., 2017), Urubici occurrence between successive Gramado layers (Turner et al., 1996) and Paranapanema underlying the 'older' Gramado in places (Turner et al., 1999).The ongoing development and diversification of the magmas does not necessarily equate to their complex temporal and spatial surface distribution sequence.
The smaller degree of crustal contamination and more enriched mantle melt signatures place Type 4 lavas as close to a plume component as possible in the region, such that Urubici and Pitanga lavas are enriched in LREE and incompatible elements (Fig. 5a-b and 7a-b).Through time, as the lithosphere thinned, the melting region shifted to shallower depths in response to the lid effect, producing Type 1 (Central-Northern) magmas from a HT source (albeit with a drop in bulk TiO 2 concentration), evidenced by the transition from garnet to garnet-spinel to spinel lherzolite (Fig. 6d) (Ra ¨mo ¨et al., 2016).These melts successfully incorporated metasomatised (i.e.Pt, Pd, Au and Cu bearing) SCLM sulphides given the presumed increased degree of melting possible at reduced depths, modelled in Fig. 12a.Silicic Chapeco ´magmas that were formed by small degree partial melt extraction from the HT magmas would not inherit the Pd, Au or Cu concentrations contained within Type 1 (Central-Northern) because their subsequent upper crustal development would have fractionated such elements into early crystals and removed them from the felsic system.It is unclear whether the differences in melting depth were a) progressive from a single evolving and shallowing melting site or b) spatially distinct despite the general 'northern' HT source.Given the synchronous nature of Serra Geral lavas, we opt for the latter view.
Further to the east, Type 1 (Southern) magmas were sourced from beneath thinner lithosphere still, near the Atlantic rift and further from the plume focus -this pro-duced higher degree partial melts within the spinel lherzolite mantle (Fig. 6d) with much more dilute TiO 2 concentrations (Fig. 3a).These higher-degree melts were able to exhaust sulphide and spinel, to produce the IPGE and Rh dominant signatures in Fig. 8d-e and 12b.Beccaluva et al. (2020) suggest that where HT magmas were less viscous and were able to rise to the surface with minor assimilation of crustal material, LT magmas stalled in subsurface chambers prior to eruption.All LT Type 1 (Southern) lavas consequently exhibit geochemical evidence for contamination, such as high 87 Sr/ 86 Sr, low eNd and crustal U-Th-Pb isotopic signatures (Peate, 1997).The transition from Gramado to Esmeralda within Type 1 (Southern) did not visibly affect PGE concentrations (Fig. 9a-f).Palmas lavas formed from protracted upper crustal development of LT lavas, in this case with more extensive AFC than the mafic melts (Garland et al., 1995;Milner et al., 1995), which may explain the loss of PGE from parental Type 1 magmas.
Overall, the shifting melting depth imposed by the Tristan plume in response to the breakup of Gondwana drives regional metallogenesis.It is highly likely that the SCLM beneath Brazil is heterogeneous in terms of precious metal content hosted in metasomatic sulphide phases in the mantle.Whilst Type 4 magmas did not receive 'spiked' PGE signatures from increased degrees of melting, the shallowersourced Type 1 melts exhausted sulphides and, for the Southern magmas, spinel-group minerals to attain elevated precious metal contents.The scale of the PELIP, localised nature of metasomatism, and synchronicity of geochemically diverse regional lavas reinforces the idea that a variety of concurrent magmatic systems worked in unison to produce the complex stratigraphy of Serra Geral.

CONCLUSIONS
The bulk major and trace element, PGE and Au data and quantitative modelling presented in this chapter support a progressively thinning lithosphere in response to thermal uplift from the upwelling Tristan plume and congruent crustal extension.This in turn directly affects the depth and source of Parana ´intracontinental partial melts and their resulting metal baskets.
(1) Serra Geral Type 4 lavas (i.e.Urubici and Pitanga) do not contain any significant enrichment in precious metals, primarily due to their origin as deeper asthenospheric melts with plume-derived trace element enrichments but no obvious precious metal contribution from the SCLM.(2) Serra Geral Type 1 (Central-Northern) lavas (i.e. Paranapanema) contain elevated Pd concentrations, coupled with higher mean Au and Cu than the other magma-types, which we attribute to a shallower melting environment, but still primarily within the asthenosphere, promoting the incorporation of metasomatically-enriched Proterozoic SCLM.(3) Serra Geral Type 1 (Southern) lavas (i.e.Gramado and Esmeralda) display unique Ru-Rh enrichment that can be linked to shallow SCLM melting at the developing rift zone, and spinel and/or sulphide incorporation from the melting source.(4) The heterogeneous PGE enrichment of cratonic SCLM by metasomatism in intracontinental settings, as exhibited beneath the North Atlantic Igneous Province, the North China Craton and Ethiopia, promotes variable metallogenic characters throughout CFB provinces.This highlights the crucial role of plume-SCLM interaction as a potential precursor to the generation of PGE ore deposits in the upper crust from intracontinental magmas.

Fig. 1
Fig. 1. a) Schematic map showing the eruption of the Parana ´-Etendeka flood basalts over the conjoined South American and African continents (135 Ma) and their current formation with the Rio Grande Rise (RGR), Walvis Ridge (WR) and Tristan-Gough islands (TG) in between.b) Representative summary timeline of eruptive products of the PELIP with incipient rifting.c) Sample map for this study (Parana ŕegion) with High-Ti and Low-Ti provinces and CFB boundaries.Markers denote sample localities -see Supplementary X for further sample details.Representative stratigraphy is given for the north and south with magma-types classified as: north -Pitanga (PN), Paranapanema (PP) and Ribeira (RI); south -Gramado (GR), Esmeralda (ES), Urubici (UR) and Silicic (S), based on(Turner et al., 1999).

Fig. 2 .
Fig. 2. Classification diagrams based on Peate et al. (1992): a) MgO vs. Zr; b) MgO vs. TiO 2 ; c) Ti/Y vs. Ti/Zr.d) Total Alkali vs. Silica (TAS) petrological/geochemical discriminant diagram for the Serra Geral sample set.Colours refer to Peate classifications and symbols for Licht classifications as per the legend in 2a.Dashed lines for approximate HT/LT division are fromPeate (1997).Dashed lines for approximate Type 1 and 4 division (2.85 wt% TiO 2 ) fromLicht (2018).Arrows refer to approximate development trajectory.In the key, dashes indicate that no data was available for that particular magma-type either in this study (large coloured markers) or the literature (small greyscale markers).Literature database with references is provided in Supplementary A2. S -Southern, C-N -Central-Northern.

Fig. 12 .
Fig. 12. Melt modelling for Serra Geral magma PGE and Au concentrations based on the parameters from Table 2. a) Arc xenolith starting compositions, representing Type 1 (Central-Northern) magma generation.Metasomatised mantle sources can produce the characteristic Pt-Pd-Au enrichment in the corresponding Paranapanema samples at ~20% partial melting.b) Cratonic and off-cratonic xenolith starting compositions, representing Type 1 (Southern) magma generation.Magma produced at 20% partial melting (i.e.sulphides exhausted) releases PGE to roughly the chondrite-normalised concentrations in Gramado and Esmeralda samples, but the Ru-Rh 'hump' is not achieved until spinel is also exhausted (25% partial melting).

Fig. 13 .
Fig. 13.Metallogenic model for the Serra Geral flood basalts and their parent magmas based on geochemical interpretations, literature information and modelling (Fig. 12).Thinner lithosphere above the Tristan plume, either situated nearer to the future Atlantic rift (ie.southeast Parana ´and Etendeka) or from progressive craton extension (i.e.north-west Parana ´) promotes higher degrees of partial melting at shallower depths.Type 4 magmas diversify mainly by partial melt extraction and fractional crystallisation (FC) from a plume-proximal source with minimal crustal contamination.Type 1 (Central-Northern) magmas, generated when the HT source begins melting in shallow SCLM regions (F = 0.2), incorporate Pd, Au and Cu from metasomatic sulphides.Type 1 (Southern) lavas diversify by assimilation and fraction crystallisation (AFC), and mafic varieties contain elevated Ru and Rh content via higher degree melts (F = 0.25) of SCLM spinel and sulphides.Based on work by Beccaluva et al. (2020); Ra ¨mo ¨et al. (2016); Turner et al. (1999).S -Southern, C-N -Central-Northern.

Table 2
Parameters used in melting models.a Spinel peridotite melt modes from