Edinburgh Research Explorer Northeast- or southwest-dipping subduction in the Cretaceous Caribbean gateway?

of a dismembered island arc that located along the to the inter-American region Caribbean) in the mid-late Cretaceous. the timing of Caribbean lithospheric movement into the inter-American region is contro- versial, with one hypothesis advocating that it happened in the Hauterivian-Albian (132.9 – 100.5 Ma), and a second hypothesis proposing the Turonian-Campanian (93.9 – 72.1 Ma). In order to investigate this problem, is- land arc rocks are studied on St. John, U.S. Virgin Islands, which are Barremian (127 Ma) to Santonian (83.6 Ma) in age. Immobile trace element and Nd – Hf radiogenic isotope ratios demonstrate that the arc rocks are derived from the partial melting of an Atlantic MORB-like mantle source region that has been variably contaminated with slab-derived ﬂ uids composed of continental detritus and slow sediment clay components. We argue that the lack of a mantle plume geochemical signature in the rocks supports the idea that the movement of Caribbean lithosphere into the inter-American region occurred in the late Cretaceous (post-Santonian) due to a subduction polarity reversal caused by collision of the Caribbean oceanic plateau with the


Introduction
The Caribbean plate is situated between the much larger North and South American plates and its eastern and western margins are separated from oceanic crust in the Atlantic and Pacific by active subduction zones. The Caribbean is predominantly composed of an 8-20 km thick Cretaceous oceanic plateau (COP) with Cretaceous to Miocene marginal arc sequences (e.g., Edgar et al., 1971;Hauff et al., 2000;Leroy and Mauffret, 1996;Neill et al., 2010). The COP was initially formed in the Pacific region (on the Farallon plate), possibly at the Galapagos hotspot (Buchs et al., 2018;Thompson et al., 2003;Torró et al., 2017;Wright and Wyld, 2011). At the same time, a large island arc was located along the western entrance to the inter-American region (termed the Great Arc of the Caribbean) (Burke, 1988).
After the COP was emplaced onto Farallon oceanic crust it was transported to the northeast and both the plateau and Great Arc moved into the inter-American region and developed into a separate plate (the Caribbean plate) (e.g., Boschman et al., 2014;Burke, 1988;Pindell et al., 2011). As the COP and Great Arc were translated eastwards into the inter-American region, the Great Arc fragmented to form the modern islands of Cuba, Jamaica, Hispaniola, Puerto Rico and the Virgin Islands. Additional arc fragments are represented by the submarine Aves Ridge (west of the Lesser Antilles) and the southern islands of Aruba, Curaçao and Bonaire and parts of Trinidad and Tobago (Escuder Viruete et al., 2008;Kerr et al., 1999;Neill et al., 2011Neill et al., , 2014Pindell et al., 2011;Wright and Wyld, 2011).
The timing of Caribbean lithospheric movement into the inter-American region is controversial. One hypothesis proposes that the Caribbean plate began to be emplaced between the Americas in the Hauterivian-Albian (132.9-100.5 Ma) (Boschman et al., 2014(Boschman et al., , 2019Marchesi et al., 2016;Pindell et al., 2011;Pindell and Kennan, 2009;Torró et al., 2016Torró et al., , 2017. The proposed causal mechanism to allow this early emplacement is variable in the literature, but has been linked to an increase in Atlantic seafloor spreading rate that either (1) generated subduction on a pre-existing inter-American transform fault (Pindell et al., 2011) or (2) initiated a polarity reversal on the existing Great Arc (Pindell and Kennan, 2009). It has also been proposed that an arc-arc collision may have caused the reversal at this time (Boschman et al., 2019).
A second hypothesis advocates that in the Turonian-Campanian (93.9-72.1 Ma) the COP collided with the Great Arc of the Caribbean and initiated a subduction reversal whereby northeast-directed subduction gave way to southwest-directed subduction (Buchs et al., 2018;Burke, 1988;Hastie et al., 2013;Kerr et al., 1999Kerr et al., , 2003Smith et al., 1998;Thompson et al., 2003;Vallejo et al., 2006). Boschman et al. (2014) also states that the collision model has no kinematic problems and is a plausible scenario.
Studying temporal geochemical changes in the igneous rocks of the Great Arc of the Caribbean can help us determine when Caribbean lithosphere moved into the inter-American region and the timing of subduction polarity reversal. In this contribution, we determine the geochemistry of  Ma) island arc rocks on the U.S. Virgin Island of St. John in order to establish when a plume-like mantle signature replaced a depleted mid-ocean ridge basalt (MORB) mantle signature. Our results will help explain when northeast directed subduction tapping a MORB-like proto-Caribbean mantle changed to southwest directed subduction tapping mantle plumederived asthenosphere beneath the Caribbean oceanic plateau and so can help resolve when the Caribbean plate began to move into the inter-American region.

St. John, U.S. Virgin Islands
The Virgin Islands, the eastern extremity of the Great Arc ( Fig. 1a-c), are now located in the northeast corner of the Caribbean plate. Puerto Rico and the Virgin Islands represent a microplate with the Puerto Rico Trench being the current northeast tectonic boundary between the Caribbean plate to the south and the North American plate to the north ( Fig. 1a) (Mann et al., 2002;Rankin, 2002;Smith et al., 1998). The three U.S. Virgin Islands of St. Thomas, St. John and St. Croix lie to the southwest of the British Virgin Islands, but all are located where the extinct Greater Antilles arc meets the modern-day active Lesser Antilles to the east (Fig. 1a).
The WIF consists of extrusive altered silicic rocks known as keratophyres and related volcaniclastic rocks (~80%), basaltic lavas and breccias (~20%) and minor cherts (Rankin, 2002). The keratophyres are classified by Rankin (2002) into multiple varieties based on the size and mineralogy of the phenocryst populations. The Careen Hill Intrusive Suite is largely composed of intrusive keratophyre (with minor trondhjemite and gabbro) (Rankin, 2002). The plutonic rocks intrude one another and the WIF. The conformably overlying LF is dominantly volcaniclastic, but basaltic, andesitic and keratophyre lavas are found sparsely throughout the formation (Rankin, 2002). The TF is composed of a variety of sedimentary rocks with rare basalts and andesites. Rocks of the Careen Hill Intrusive Suite, WIF, LF and TF have been affected by tropical and hydrothermal alteration and are cut by younger dolerite, gabbro, lamprophyre and tonalite dykes (Rankin, 2002).

Water Island Formation mafic-intermediate rocks
The basalt and basaltic andesite lavas are porphyritic with <5% subto euhedral micro-phenocrysts of plagioclase. Rare clinopyroxene and hornblende phenocrysts up to 3 and 1 mm, respectively, are minor phenocryst constituents. Phenocrysts sit in a commonly aphanitic groundmass of randomly orientated plagioclase needles, (<0.1 mm) with small clinopyroxenes and opaque minerals. Many samples exhibit a randomly orientated groundmass, however, locally plagioclase laths can be aligned and impart a weak foliation to the rock. The basalts have undergone variable alteration and low degree metamorphism with the growth of chlorite, prehnite, pumpellyite and calcite, and in some samples, clinopyroxene is locally replaced by epidote and amphibole (hornblende), and plagioclase to sericite. Amygdales are present as elongated cavities <3 mm-long filled with variable combinations of quartz and epidote with radiating elongate crystals, hornblende, devitrified glass, plagioclase and occasionally, prehnite. Polycrystalline quartz and sericite veins are common and likely formed as secondary features, post-eruption.

Water Island Formation silicic rocks (includes the Careen Hill Intrusive Suite)
For the remainder of this article, we refer to the evolved rocks in the WIF and CHIS as the WIF because the two rock suites are geochemically identical and the WIF term has been used extensively in previous literature. The WIF keratophyres have been grouped into multiple varieties based on the size and mineralogy of the phenocryst populations by Rankin (2002). A brief summary of the samples used in this study is given below, but for a more detailed classification of the WIF keratophyres the reader is referred to Rankin (2002). Like the basalts, the keratophyres are altered and are predominately porphyritic with phenocrysts of varying size. Typically, all keratophyres have aphanitic groundmasses composed of laths of randomly orientated or aligned albitised plagioclase and interstitial quartz. Opaque minerals make-up~10% of the groundmass and minor hornblende and biotite can be present. Secondary mineralisation includes metamorphic chlorite, muscovite and biotite. Some keratophyres also show rare orthoclase altered to chlorite and sericite alteration is common at the edges and within plagioclase phenocrysts.

Louisenhoj Formation
The basalts and andesites are largely porphyritic with vesicles in the andesites and amygdales in the basalts. Phenocrysts of stubby plagioclase, 3-5 mm, are abundant and contribute up to 50% of the samples. The plagioclase is predominately albitised. Clinopyroxene (10%) phenocrysts are also common in the samples of this study. The andesites also contain interstitial quartz and biotite within the groundmass and their amygdales are filled with chlorite, quartz and prehnite.

Tutu Formation
The basalts are porphyritic with abundant (up to 50%) phenocrysts of 3-4 mm long plagioclase. Clinopyroxenes are less common, but occur as larger phenocrysts up to 5 mm and are locally replaced by amphibole. The groundmass has a weak foliation because of the alignment of plagioclase and clinopyroxene together with minor amphibole (hornblende), chlorite and randomly distributed opaque minerals.

Methodology
Thirty-seven samples from all of the pre-Eocene volcanic formations on St. John have been analysed for major and trace elements at Cardiff University and a sub-set of twelve samples has been analysed for Nd-Hf radiogenic isotopes at Durham University ( Fig. A1 and Tables A1 and A2, Appendix).
Following removal of weathered surfaces and obvious vein material the samples were crushed in a jaw crusher and powdered using an agate Tema ball mill. Approximately 2 g of powdered sample was heated in a porcelain crucible to 900°C for two hours to determine the loss on ignition (LOI). For the chemical analysis, 0.1 g ±0.0005 of sample was mixed and fused with 0.4 g ± 0.0005 Li-metaborate flux.
This was dissolved in a~5% HNO 3 solution with a 1 ppm Rh spike and made up to 100 ml using 18 Ω distilled water. Major element concentrations were determined by using a JY Horiba Ultima 2 inductively coupled plasma optical emission spectrometer. 1 ml of the solution was further diluted with 1 ml of In and Tl spike and 8 ml of 2% HNO 3 and analysed using a Thermo X7 series inductively coupled plasma mass spectrometer to obtain trace element contents. The full major and trace element analytical procedure is described in McDonald and Viljoen (2006). Nd-Hf radiogenic isotopes were determined at the Arthur Holmes Laboratory of the Durham Geochemistry Centre, Durham University. The samples were dissolved using standard HF-HNO 3 digestion followed by elemental pre-concentration based on the method of Dowall et al. (2007). Sample solutions were run through 1 ml pipette tips containing dilute Sr-spec resin to remove the Sr-bearing fractions. The high field strength element (HFSE)-and Rare Earth Element (REE)-bearing fraction was run through 10 ml polypropylene columns containing Bio-Rad AG50W-X8 200-400 mesh cation-exchange resin. Neodymium was collected as part of a general REE fraction. The HFSEbearing fraction was then run through a 10 ml polypropylene column containing Bio-Rad AG1-X8 anion-exchange resin to separate Hf from Ti. All samples were taken up in 400-500 μl of 3% HNO 3 . Radiogenic isotopes were analysed on a Thermo Neptune multi-collector inductively coupled plasma mass spectrometer. Neodymium was run with J&M and Sm-doped J&M standards giving a mean 143 Nd/ 144 Nd of 0.511104 ± 0.000011 (2σ, n = 13, uncertainty = 22.4 ppm). Results were normalised to a preferred value of 0.511110. Hf was run with standard JMC475 giving a mean 176 Hf/ 177 Hf of 0.282146 ± 0.000009 (2σ, n = 9, uncertainty = 31.3 ppm). Results were normalised to a preferred value of 0.282160. For the age correction of the samples we use the decay constants (λ): Lu-Hf = 1.876 × 10 −11 and Sm-Nd = 6.54 × 10 −12 . For chondritic values we use 143 Nd/ 144 Nd = 0.512638, 147 Sm/ 144 Nd = 0.19670, 176 Hf/ 177 Hf = 0.282772 and 176 Lu/ 177 Hf = 0.03320.

Element mobility and classification
Previous studies (e.g., Escuder Viruete and Castillo Carrión, 2016; Lidiak and Anderson, 2015;Neill et al., 2011;Torró et al., 2017) have shown that igneous rocks in the Caribbean region have suffered intense hydrothermal alteration and tropical weathering. As such, the large ion lithophile elements (LILE: K, Cs, Ba, Na) are frequently mobilised and cannot be used for petrogenetic interpretations. We can test the mobility of the LILE in rocks of the Virgin Islands by plotting bivariate diagrams with an element of interest on the ordinate and Nb on the abscissa. Niobium is plotted because it is one of the most immobile trace elements (e.g., Cann, 1970;Hill et al., 2000;Kurtz et al., 2000). If an element is plotted against Nb, the data should form a linear trend if the samples are co-genetic and the elements are moderately to highly incompatible and immobile.
Light rare earth element (LREE) bivariate diagrams show a broad and dominant positive liquid line of descent that demonstrates that the elements are incompatible, immobile and that the data could be consistent with a co-genetic petrogenesis for the entire St. John sample suite (Fig. 2a-c). However, the middle and heavy REE diagrams show the samples progressively forming three broad trends, as indicated on Fig. 2d-f. The M-and HREE should be immobile, therefore, the three trends suggest that the samples, as a whole suite, are not co-genetic. The silicic WIF and samples of the LF form a high M-HREE trend at a given Nb content and the TF forms a dominant middle trend. The Zr-Nb diagram shows two broad trends with the silicic WIF and TF samples forming low and high Zr liquid lines of descent at a given Nb content respectively (Fig. 2g). In contrast, the LILE data show relatively large amounts of scatter on the Rb, Ba, U and Cs-Nb bivariate plots in Fig. 2i-l that is consistent with the elements being mobilised during sub-solidus alteration. The trends appear more consistent on the Sr and Th-Nb diagrams and the Sr data shows very low Sr contents for the WIF silicic rocks.
In conclusion, the LILEs have been mobilised in the samples of this study, but Th, REE and HFSE are immobile and show relatively coherent liquid lines of decent (similar to other Caribbean igneous rocks, e.g., Marchesi et al., 2016). There are three broad magmatic trends seen in the bivariate diagrams and possibly several more distinct trends within the individual formations. However, two of the three broad trends are mostly represented by rocks of the TF and silicic WIF respectively. As such, the igneous rocks of St. John are not co-genetic, but are composed of multiple magmatic suites Igneous rocks are routinely classified using the total alkali silica diagram, but because of the mobility of the LILEs in the St. John rocks we have opted to classify the samples using the immobile element Th-Co diagram of Hastie et al. (2007). The WIF mafic lavas mostly plot as tholeiitic and calcalkaline basalts and basaltic andesites while the WIF silicic rocks plot in the dacite/rhyolite field (Fig. 3). Similarly, the lavas of the LF predominantly classify as tholeiitic basalts to andesites and the rocks of TF are tholeiitic basalts (Fig. 3).

Multielement diagrams
Normal mid-ocean ridge basalt (N-MORB) normalised multielement plots in Fig. 4 show that all the rocks from St. John have large negative Nb-Ta anomalies. The WIF mafic rocks have predominantly MORBlike HREE contents, positive Sr anomalies and enriched Th and U (Fig. 4a). The silicic WIF also has MORB-like HREE, but the rocks have negative Sr, P and Ti anomalies (Fig. 4b). The LF rocks have relatively uniform HREE concentrations, but positive Sr anomalies and a wide range of incompatible element contents (Fig. 4c). The rocks of the TF have flat, MORB-like HFSE and HREE contents and a small positive Sr anomaly (Fig. 4d). John. Data from this paper and the grey fields represent data ranges from . Normalising data are from Sun and McDonough (1989 Nd/ 144 Nd(i) diagram (Fig. 5a) the three formations have distinctive radiogenic isotope compositions with LF rocks having the most depleted (radiogenic) values and the TF samples having the least depleted (nonradiogenic) composition. The three formations are geochemically distinct and are therefore derived from compositionally different mantle source regions.

Nature of the mantle wedge
The main aim of this contribution is to determine the nature of the mantle component in the Cretaceous magmatic rocks of the Virgin Islands. In island arc magma petrogenesis the HFSE can be mobilised in slab-derived partial melts but are not mobilised easily by aqueous fluids from the dehydration of the slab (Keppler, 1996;Kessel et al., 2005;Pearce and Peate, 1995). Therefore, if the slab component in the St. John rocks is represented by a slab-fluid, the HFSE can be used to determine the compositional nature of the mantle wedge. We can test this for the Virgin Islands magmatic rocks in this study by plotting a Zr/Yb-Nb/Yb diagram from Pearce and Peate (1995) (Fig. 5b). Basic to intermediate samples from St. John lie within a defined mantle array. If Zr had been mobilised in a slab partial melt and added to the mantle wedge, the Zr/Y ratios of the St. John samples would plot above the mantle array. Fig. 5b shows that ratios of the HFSEs have not been mobilised in a slab-derived component (i.e., they are conservative) and so can be used to study the relative enrichment of the mantle source that underwent partial melting to form the St. John magmas.
Mantle plume-derived magmas are usually generated from less incompatible element depleted source regions than N-MORB and so have higher Nb concentrations. Therefore, on a Nb/Y-Zr/Y diagram, mantle plume-derived rocks have higher Nb/Y ratios at a given Zr/Y than N-MORB. This diagram was initially developed to investigate Icelandic lavas (Fitton et al., 1997) where enriched mantle plume rocks and N-MORB fall above and below a lower reference line respectively, but it works well for other mantle plume-derived rocks (e.g., Hastie et al., 2008). Data from basic and intermediate rocks from St. John and rocks from the U.S. Virgin Islands from  are plotted in Fig. 5c. The samples lie below the lower reference line suggesting that the mantle component involved in the petrogenesis of the Cretaceous Virgin Island arc rocks had an N-MORB composition.
On the 176 Hf/ 177 Hf(i) -143 Nd/ 144 Nd(i) diagram the St. John samples have more enriched (unradiogenic) compositions relative to modern N-MORB and they plot in the Caribbean oceanic plateau field (Fig. 5a). Nevertheless, unlike the HFSE, Nd is moderately non-conservative and will be mobilised in a slab-derived aqueous fluid (e.g., Keppler, 1996;Kessel et al., 2005;Pearce and Peate, 1995). Therefore, the St. John 143 Nd/ 144 Nd(i) values represent mixing between the mantle wedge source region and a slab-related component. The original St. John mantle compositions, prior to contamination with the slab-derived fluid, would have had higher (more radiogenic) 143 Nd/ 144 Nd(i) values. Therefore, it is likely, based purely on Nd-Hf radiogenic data, that the composition of the mantle wedge involved in the petrogenesis of the St. John magmas had a N-MORB composition; thus, supporting the Nb/Y-Zr/ Y data.

Composition of the slab component
Many elements that are used to investigate the nature of slab components in island arc-derived lavas cannot be used in this study because they are mobilised during sub-solidus alteration processes (e.g., K and Ba: Elliott, 2003). However, Hastie et al. (2009Hastie et al. ( , 2013 developed the Th/La-(Ce/Ce*) Nd diagram that uses immobile trace elements to determine the geochemical components of the slab-derived flux in the petrogenesis of island arc magmatic rocks. Such fluxes could be composed of a variety of geochemical end members that include fluids from the basaltic crust, continental and volcanic detritus, and slow sedimentation-rate-clays (SSC). Such SSCs include hydrogenous Fe-Mn oxides, fish debris-rich clay and hydrothermal sediments.
The WIF silicic rocks on St. John have crystallised accessory allanite and apatite (Rankin, 2002). These Th and LREE-rich phases could potentially have an impact on the Th/La-(Ce/Ce*) Nd diagram; thus, only basic to intermediate samples from the WIF, LF and TF are plotted. Furthermore, sub-solidus alteration processes can preferentially mobilise REE 3+ and Y, relative to Ce 4+ . Therefore, the hydrothermal fluids responsible for sub-solidus alteration can impart negative and positive Ce anomalies depending on whether the fluid has been removed or added to the rock (Cotten et al., 1995;Patino et al., 2003). Basicintermediate rocks analysed in this study are plotted on a La-Ce bivariate diagram that shows a linear trend with R 2 = 0.99 (Fig. 6a). If any Ce anomalies in the St. John dataset are formed by variable metasomatism of REE 3+ fluids the La-Ce ratio would be expected be variable and not show a linear correlation coefficient close to unity. Hydrothermal alteration would also disturb Y concentrations and may lead to high loss of ignition values (LOI) if the rocks have not been extensively silicified. The St. John samples do not show systematic correlations between Y or LOI versus (Ce/Ce*) Nd (Fig. 6b and c); thus, suggesting that Ce anomalies in the St. John samples are unrelated to secondary alteration processes. The St. John samples analysed in this study are plotted on the Th/La-(Ce/Ce*) Nd diagram and are interpreted to be generated predominantly by mixing between a mantle-like component and continental detritus (Fig. 6d). The average Th/La values for the basic to intermediate samples of the WIF and LF are 0.16 (not including sample SJ52 that has anomalously high values) and 0.11 for rocks of the TF. This suggests that WIF and LF magmas contain a higher proportion of, or a more enriched, continental detritus component than the TF magmas. Furthermore, some of the samples are slightly displaced to lower (Ce/Ce*) Nd values, which implies a small contribution from a SSC component rich in fish debris and/ or hydrothermal sediments. The average (Ce/Ce*) Nd values for the basic to intermediate samples from the WIF and LF range from 0.93-0.97 (again, not including sample SJ52), but the TF rocks are relatively low with 0.89. Therefore, this implies that the TF magmas contain a higher proportion of, or a more (Ce/Ce*) Nd depleted, SSC component than the WIF and LF magmas.
These findings are generally supported by Virgin Island data in  where WIF, LF and TF mafic rocks form a broad mixing array between a mantle component and continental detritus (Fig. 6e). However, the  TF data have a higher average Th/La ratio of 0.15 and the LF have a low ratio of 0.12. All the samples are displaced to relatively low (Ce/Ce*) Nd values with WIF, LF and TF having 0.85-0.91. In particular, some of the LF rocks have very low (Ce/Ce*) Nd values down to 0.67. Consequently, data in this study and in  suggest that all the formations show evidence of variable contributions from both continental detritus and fish debris and/or hydrothermal sediments.
The St. John lavas do not have adakite compositions, which would be expected if the slab component was a partial melt derived from the magmatic portion of the subducting plate. A partial melt from any sedimentary veneer would mobilise the HFSEs, which is not seen in Fig. 5b. In addition, a 176 Hf/ 177 Hf(i) -Th/Hf diagram is shown in Fig. 7a, and if the slab component was a sedimentary melt, the St. John samples would be expected to be displaced to lower 176 Hf/ 177 Hf(i) and higher Th/Hf ratios due to the mobility of Th and Hf in a slab-derived sedimentary melt. In contrast, Nd does not require a slab melt because it will be mobilised during slab dehydration (Pearce and Peate, 1995) and the St. John 143 Nd/ 144 Nd(i) values will represent mixing between the mantle wedge and the slab fluid. Fig. 5a shows that the TF is relatively more enriched in 143 Nd/ 144 Nd(i) relative to WIF, and LF and WIF is slightly more enriched than LF. Thus, if the  database is combined with our new analyses, Th/La and (Ce/Ce*) Nd suggest that there is no simple case where one formation dominantly shows evidence for a stronger association with either a continental or SSC slabderived component. However, 143 Nd/ 144 Nd(i) values demonstrate that the TF has been contaminated with either a larger proportion of, or a more enriched sedimentary slab-derived component relative to WIF and LF samples.
As with previous studies, it is difficult to model this mixing because we do not know the nature of the sediments that were subducting beneath the Great Arc~127-85 Ma. However, if we assume a mixture between N-MORB (Sun and McDonough, 1989) and the Global Subducting Sediment (GLOSS II) end member of Plank (2014) we can model the generation of the St. John rocks using a 143 Nd/ 144 Nd(i) -Th/Hf diagram (Fig. 7b). For the radiogenic isotope data, the mixing curve is calculated by: where C M is the 143 Nd/ 144 Nd(i) value of the mixture, C A 143/144 and C B 143/144 are the 143 Nd/ 144 Nd(i) ratios in components A and B respectively, C A and C B represent the corresponding Nd concentrations of components A and B respectively, and X is the mass fraction of component A in the mixture (DePaolo, 1981). Also, the Th and Hf mixing concentrations are calculated by: where C M is the concentration of Th or Hf in the mixture, C A and C B are the abundances of Th or Hf in components A and B respectively, and X is the proportion of component A. The mixing curve shows that the compositions of the St. John island arc rocks can be replicated with a mixture of 95-85% N-MORB and 5-15% GLOSS II, with samples from the TF requiring a larger percentage of GLOSS II.

Petrogenesis of the WIF silicic rocks
The silicic rocks of the WIF could have been formed by either large degrees of fractional crystallisation of mafic parental magmas or the partial melting of mafic source rocks in the lower arc basement (Jolly et al., 2008;. The similarity of the 76 Hf/ 177 Hf (i) and 143 Nd/ 144 Nd(i) ratios strongly suggests that the WIF mafic rocks represent either the parental magmas or the source rocks for the WIF silicic rocks (Fig. 6a). A.R. Hastie, S. Cox and A.C. Kerr Lithos 386-387 (2021) 105998 Chondrite-normalised REE diagrams in Fig. 8a and b show that the rocks of the WIF have slight concave-up patterns that is frequently interpreted as due to amphibole-controlled fractionation by either crystallisation from a magma or as a residual phase in a metabasic source region. To quantify such patterns Davidson et al. (2013) developed the Dy/Dy* [chondrite normalised (cn) Dy/(La 4/13 × Yb 9/13 )] versus Dy/Yb cn diagram. A MORB field on this diagram extends from low Dy/Yb cn < 1 and Dy/Dy* > 1 to low Dy/Dy* < 1 and Dy/Yb cn > 1 (Fig. 8c). Primitive arc magmas plot in the MORB field and subsequent amphibole (± cpx) fractional crystallisation should lead to the generation of more evolved arc magmas with both Dy/Dy* and Dy/Yb cn < 1.0. Fig. 8c shows the WIF mafic rocks predominantly plotting in the MORB array. Some of the mafic samples plot with Dy/Dy* ratios <1 and these represent the small number of samples in Fig. 8b that are relatively enriched in LREE compared to HREE. However, the majority of the mafic WIF rocks are relatively depleted in LREE and have Dy/ Dy* ratios >1. The WIF silicic rocks plot on a moderately linear trend from the LREE-depleted mafic rocks to compositions with Dy/Dy* and Fig. 8. (a and b) Chondrite-normalised REE diagrams showing raw data from this study and grey fields representing analyses from . (c) Dy/Dy*-Dy/Yb diagram from Davidson et al. (2013) and (d-h) major element bivariate diagrams showing fractional crystallisation trends of a plagioclase (58%), clinopyroxene (10%), magnetite (12%), orthopyroxene (10%), amphibole (9.7%) and apatite (0.3%) mineral assemblage from a parental magma representing WIF sample SJ53. Tick marks on trends represent 10% increments. Virgin Island data in c-h are from this paper and . Normalising values from McDonough and Sun (1995).
A.R. Hastie, S. Cox and A.C. Kerr Lithos 386-387 (2021) 105998 Dy/Yb(cn) <1. This suggests that amphibole controlled fractional crystallisation of the WIF mafic rocks can form the silicic samples. It also shows that no garnet is involved as garnet fractionation would increase Dy/Yb(cn). Woodhead (1988) modelled the formation of intra-oceanic island arc lavas by fractional crystallisation processes using a mineral assemblage of plagioclase, clinopyroxene, magnetite and olivine in the proportions 60:25:10:5 respectively. Evidence of plagioclase and magnetite crystallisation in the silicic WIF is apparent with low Sr (Fig. 2m) and Fe contents (not shown). The intermediate Virgin Island rocks that could represent the parental magmas for the keratophyres would be too silicic for olivine saturation and would instead crystallise orthopyroxene. Finally, the negative P anomalies in the keratophyres (Fig. 4) suggests that apatite was also crystallising. Therefore, we have modified the crystallisation mode of Woodhead (1988) to include plagioclase, clinopyroxene, magnetite, orthopyroxene, amphibole and apatite in proportions of 58:10:12:10: 9.7:0.3. We have calculated the removal of this bulk mineral assemblage from a theoretical WIF intermediate magma by mass balance using the following equation: where C l is the concentration of the magma after a given proportion of crystallisation, C 0 is the original magma composition that is taken as that of the relatively evolved mafic sample SJ53, C min is the composition of the bulk mineral assemblage being removed from the magma and X is the mass fraction of the mineral assemblage being removed. Calculated major element liquid lines of decent in Fig. 8d-h show that it is possible to generate silicic WIF igneous rocks from an intermediate WIF parent through the removal of~50-70% of the crystallising assemblage We have tested the major element crystallisation results by modelling the evolution of trace element concentrations during fractional crystallisation of the same parental intermediate WIF magma using the equation: C l and, C 0 are as in Eq.
A.R. Hastie, S. Cox andA.C. Kerr Lithos 386-387 (2021) 105998 et al. (1984), McKenzie and O'Nions (1991), Bacon and Druitt (1988) and Bédard (2006) and F is the proportion of melt remaining. Fractional crystallisation trends in Fig. 9a-f show that the trace element abundances of the keratophyres can be explained by 50-70% crystallisation of a WIF intermediate parental magma. Also, the models show that Dy/Dy* and Dy/Yb(cn) <1 can be replicated and the whole compositional range of the silicic rocks can be formed by the evolution of the compositionally heterogeneous WIF mafic rocks. Finally, the low Both the mineral and melt modes are taken from Beard and Lofgren (1991). The non-modal batch equation is used from Shaw (1970) and D values are derived from Bédard (2006). (b) Dy/ Dy*-Dy/Yb diagram from Davidson et al. (2013) showing the metabasic partial melting trend intersects the silicic WIF data with~10-18% partial melting. (c-h) Selected trace element variation diagrams showing partial melt trends for the arc composition from Beard and Lofgren (1991). Partial melt fractions range from 1 to 18% and the internal increments represent 5, 10 and 15%. Virgin Island data are derived from this paper and . Normalising values from Sun and McDonough (1989).
proportion of crystallising apatite in our models does not have any substantial effect on the Dy/Yb and Dy/Dy* ratios that we use to justify amphibole removal. Nevertheless,  suggest that the lack of intermediate magmatism and the overwhelming volume of silicic (80%) compared to mafic material (20%) (Fig. 1d) implies that the keratophyres are most likely generated by crustal fusion. We have tested this crustal fusion scenario by modelling the partial melting of a theoretical garnet-free metabasic source region. The low Dy/Dy* and Dy/Yb(cn) ratios suggest that amphibole is required in the residue and so an anhydrous pyroxenite source region is unsuitable. The nonmodal partial melt equation from Shaw (1970) is used for the calculation: where C l is the concentration of the resultant melt, C 0 is the concentration of the source region before partial melting (taken as SJ53 again), F is the mass fraction of melt generated, D is the bulk partition coefficient prior to partial melting and P is the average of the partition coefficients weighted by the proportion contributed by each phase to the melt. The starting mineral and melt modes are from Beard and Lofgren (1991) because they supply data for the partial melting of low-pressure, garnet-free amphibolites from an arc setting. The mineral mode includes plagioclase, amphibole, magnetite, quartz, clinopyroxene, orthopyroxene and ilmenite. We slightly modify this modal proportion by adding apatite that is required to generate relatively low P concentrations in the keratophyres (Fig. 4). The starting mode comprises plagioclase (0.4973), amphibole (0.3261), quartz (0.0854), clinopyroxene (0.002) orthopyroxene (0.0136), magnetite (0.0531), ilmenite (0.02) and apatite (0.0025) and the melt mode comprises plagioclase (0.2641), amphibole (1.747), quartz (0.4576), clinopyroxene (−1.0755) orthopyroxene (−0.2983), magnetite (−0.0994), ilmenite (0.0021) and apatite (0.0025). The melt mode indicates that amphibole and quartz would be exhausted after~18% partial melting; thus we restrict the range of modelled melts in Fig. 10 to 1-18%. Our calculations show that the WIF silicic multielement patterns, Dy/Dy* and Dy/Yb(cn) ratios <1 and trace element abundances on bivariate diagrams can be replicated by~1-18% partial melting of a WIF mafic source region ( Fig. 10a-h).
Although we have shown that the composition of the WIF silicic rocks can be generated by partial melting of a relatively shallow metabasic source region with compositions similar to the WIF mafic rocks, geochemical modelling can also support a fractional crystallisation model for generating the keratophyres. Distinguishing between the two processes is extremely difficult; however, we favour the partial melt process due to the relatively large volumes of silicic material and lack of intermediate compositions.

Geochemical summary
The Barremian to Santonian (127-83.6 Ma) St. John rocks in this study represent pre-and post-COP island arc magmatism that has IAT and CA compositions. All the rocks, except for the silicic WIF samples, have immobile trace element and Nd-Hf radiogenic isotope ratios that show that the St. John samples are derived from the partial melting of an Atlantic MORB-like mantle source region that has been variably contaminated with slab-derived fluids composed of continental detritus and slow sediment clay components. In contrast to younger Campanian (83.6-72.1 Ma) igneous rocks in the Great Arc (e.g., Hastie et al., 2013), none of the rocks from St. John show evidence for a mantle plume-like geochemical signature.

Evolution of the Inter-American region in the cretaceous
It has been proposed that Caribbean lithosphere moved into the inter-American region in the Hauterivian-Albian (132.9-100.5 Ma) (Boschman et al., 2014;Marchesi et al., 2016;Pindell et al., 2011;Pindell and Kennan, 2009;Torró et al., 2016Torró et al., , 2017. If so, partial melting in the convergent margin with southwest-directed subduction would likely tap mantle plume-like material instead of depleted MORB mantle after~90 Ma. Hastie et al. (2013) showed that such a mantle plume signature is not seen in island arc igneous rocks throughout the Caribbean until after the late Campanian (<75 Ma) (Fig. 5c). Similarly, Barremian to Santonian (127-83.6 Ma) island arc magmatic rocks on St. John show no geochemical evidence for the involvement of a mantle plume and, thus, a COP component before the Santonian. This is surprising because our younger samples post-date the 94-88 Ma COP and would be expected to show an oceanic plateau signature with southwest-directed subduction. This indicates that, before~84 Ma, subduction in the inter-American region was most likely to the northeast and magmatism was generated by partial melting a mantle wedge composed of Atlantic-like N-MORB.
In keeping with our earlier conclusions (e.g., , we still maintain that one of the major problems with an Early Cretaceous reversal is that the Farallon plate was subducting to the east and beneath the Americas since the Permian (e.g., Mann et al., 2007;Spikings et al., 2015). Relatively localised southwest-directed subduction of the proto-Caribbean oceanic crust in the Early Cretaceous would lead to a serious space problem. Regardless of what mechanism is invoked in the overlying American and Proto-Caribbean lithosphere to initiate subduction reversal, the underlying Farallon plate would have continued to subduct to the northeast. To allow the proto-Caribbean lithosphere to be consumed in a southwest-directed subduction zone the Farallon plate must be locally prevented from subducting to the northeast. The Turonian-Campanian (93.9-72.1 Ma) hypothesis satisfies these constraints because the collision of the COP with the Great Arc would jam the subduction zone and prevent the Farallon plate from locally subducting in the inter-American region. This allows both subduction reversal and subduction back-step to isolate the COP and the Great Arc and results in the development of the Caribbean plate. In contrast, models that invoke an early Cretaceous reversal have to explain how the Farallon plate is prevented from locally subducting to the northeast. A further problem with early Cretaceous reversal involves the potential southwest-directed subducting plate at depth in the mantle. If oceanic lithosphere was subducting to the southwest from~130-100 Ma and the mantle plume that generated the Caribbean oceanic lithosphere is dominantly 88-90 Ma the southwest-directed slab at depth may prevent mantle plume material from being emplaced close to the active island arc. This may particularly be true if young Proto-Caribbean lithosphere subducted at a relatively shallow angle. Nevertheless, we note that the formation of southwestdirected subduction further to the west in an intra-oceanic island arc would help negate a space problem. Indeed, Boschman et al. (2019) advocate that a convergent margin with southwest-directed subduction could form at~125 Ma to the west of existing northeast-directed subduction between the Americas. The two subduction zones subsequently collide at~100 Ma to locally reverse northeast-directed subduction in the Proto-Caribbean region. However, forming a subduction zone is very difficult. It is easier for India to underthrust the Himalaya than it is to form a subduction zone to the south of Kerala. Slab pull will be pulling old and thick oceanic lithosphere below the Americas at 135-100 Ma in a northeast-direction. Why and how would very thick and very old oceanic lithosphere to the west of the Americas locally suddenly stop being pulled to the northeast, break and then reverse direction to subduct to the southwest?
We continue to argue that before the Turonian to Campanian (93.9-72.1 Ma) a northeast directed subduction zone existed along the western seaboard of the Americas generating the Great Arc of the Caribbean (Hastie et al., 2013). In the late Cretaceous (~88-94 Ma) the main volume of the COP erupted onto the Farallon plate. The plateau was then transported to the northeast and collided with the Great Arc of the Caribbean. The collision choked the northeast-directed subduction zone and generated a subduction polarity reversal at~80-75 Ma where southwest-directed subduction initiated on the eastward seaboard of the Great Arc. The movement of the COP between the American continents in the Turonian-Campanian (93.9-72.1 Ma) would have reduced or prevented the flow of oxygenated bottom water through the inter-American passage (Buchs et al., 2018;Kerr, 1998). This may have been the causal mechanism of, or at least contributed to, the formation of the Cenomanian-Turonian mass extinction event that is evidenced by a global oceanic anoxia event at this time (OAE-2) and also continued anoxic events up to the Santonian (OAE-3) (Buchs et al., 2018;Kerr, 2005).

Conclusions
Cretaceous igneous island arc rocks on St. John, U.S. Virgin Islands are Barremian (127 Ma) to Santonian (83.6 Ma) in age. The rocks are derived from partial melting an Atlantic MORB-like mantle source region that has been variably contaminated with slab fluids with dissolved continental detritus and slow-sediment clays. The most silicic rocks can be modelled with fractional crystallisation processes from more primitive parental magmas and also by partial melting of a low-pressure metamorphosed basaltic source region. However, the relatively large volumes of silicic material relative to mafic and intermediate compositions and lack of intermediate compositions favour a partial melt origin. The lack of mantle plume geochemical signature in the arc rocks strongly suggests that the translation of Caribbean lithosphere into the inter-American region occurred in the late Cretaceous (post-Santonian) due to a subduction polarity reversal caused by collision of the COP with the Great Arc.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.