Introduction

Evidence for the early evolution of the ancient oceanic crust is presumed to be preserved in deep continental roots (Aulbach et al. 2016; Jacob 2004). However, these are not directly accessible and their study is limited to xenoliths exhumed by kimberlite magmas. Although eclogite xenoliths are extremely rare, being < 1% of mantle xenoliths in kimberlites (Schulze et al. 2000) they may be the only relics of the ancient crust (Beard et al. 1996; Jacob and Foley 1999; Schulze and Helmstaedt 1988). The roots of most cratons contain scattered eclogite bodies, sometimes associated with diamonds, possibly concentrated around 120–150 km depth (Garber et al. 2018; Jacob 2004; Pearson 1999). Determining the origin of mantle eclogites is key to understand the formation and thickening of stable continental domains (Aulbach 2012; Jacob 2004; Pearson 1999). Prior to or during their entrapment, most mantle xenoliths underwent kimberlitic/carbonatitic metasomatism that changed their primary composition (Howarth et al. 2014; Huang et al. 2012; Misra et al. 2004). This must be taken into account before any protolith can be reliably identified.

Several hypotheses have been proposed for the origin of mantle eclogites and their preservation in the cratonic root: (1) mantle-derived high-pressure cumulates (Caporuscio and Smyth 1990; Harte and Gurney 1975; Hatton 1978; Hatton and Gurney 1977; Lappin 1978; Lappin and Dawson 1975; MacGregor and Carter 1970; O’Hara and Yoder 1967; Smyth and Caporuscio 1984; Smyth et al. 1989); (2) metamorphosed fragments of a subducted oceanic crust (Agashev et al. 2018; Aulbach et al. 2016; Barth et al. 2002; Beard et al. 1996; Helmstaedt and Doig 1975; Jacob 2004; Jacob and Foley 1999; Jacob et al. 1994; MacGregor and Manton 1986; Radu et al. 2017; Riches et al. 2016a; Schulze and Helmstaedt 1988; Schulze et al. 2000; Shervais et al. 1988; Shu et al. 2016; Viljoen et al. 2005; Withers et al. 1998); and (3) subduction-derived melt residues (Barth et al. 2001; Ireland et al. 1994; Rollinson 1997; Shchipansky 2012; Snyder et al. 1993).

The high-pressure cumulate model fails to account for δ18O values outside the ‘normal’ mantle range (+ 5.5 ± 0.4‰ Mattey et al. 1994), the presence of a parental basaltic melt deeper than 200 km, the absence of ubiquitous olivine (Jacob 2004), or the excess silica and aluminium to produce kyanite for the entire range of bulk Mg# (Ater et al. 1984). Moreover, this model is most commonly based on grosspydite xenoliths, misclassified as eclogites (Harte and Gurney 1975; Lappin 1978; MacGregor and Carter 1970; O’Hara and Yoder 1967; Smyth et al. 1989).

Although seemingly more robust, the two crustal models (2, 3) have been in turn contested (Gréau et al. 2011; Griffin and O’Reilly 2007; Huang et al. 2014), and redox conditions have been proposed as alternative mechanisms that could have generated the recorded positive Eu, while Sr anomalies and δ18O values higher than mantle (> 5.5‰) have been attributed to mantle metasomatism. However, these alternative explanations do not account for eclogites with such a wide range in δ18O values, and little consensus has been reached regarding the petrogenesis of mantle eclogites (Aulbach and Jacob 2016; Griffin and O’Reilly 2007; Huang et al. 2016; Radu et al. 2017).

Geological background

The Kaapvaal Craton broadly covers the centre of southern Africa, and is delimited by the Limpopo belt (~ 2.7 Ga) to the north, the Namaqua-Natal Province mobile belt (~ 1.1–1.9 Ga) to the south and east and by the overthrusted Kheis belt (~ 2.0 Ga) to the west (van Reenen et al. 1992). It generally stabilized between 3.7 and 2.6 Ga (de Wit et al. 1992), followed by a later disruption of the north-central part of the craton by the major intracrustal intrusion of the Bushveld Igneous Complex (James et al. 2003) at ~ 2.06 Ga. Seismic investigation shows a thin continental crust, with an intermediate to felsic composition at the base and a sharp, almost flat Moho (Schmitz and Bowring 2003), attributed to intensive re-melting of the lower crust, possibly related to the Ventersdorp extensional tectonics and magmatism (2.7 Ga). This is further supported by coeval granulite xenoliths, which have been interpreted as evidence for ultrahigh-temperature metamorphism (in excess of ~ 1000 °C in the lowermost crust) that may have affected the western and central part of the craton, including the Kimberley area (Schmitz and Bowring 2003). Partial melting and deep crustal ductile flow may have led to differentiation at the base of the crust and high-pressure–temperature (HP-HT) crystallization (James et al. 2003).

The Roberts Victor Mine (28°29′ S–25°33′ E) is located ca. 95 km northeast of Kimberley, in the centre of the Kaapvaal Craton (Fig. 1). It consists of two small Group-2 kimberlite pipes intruded at ca. 128 Ma (Smith et al. 1985), containing phenocrysts of olivine, diopside, spinel, perovskite and apatite, in a phlogopite-dominated groundmass, characterized by fine-grained textures; with elevated contents of SiO2, K2O, Pb, Rb, Ba and LREE-enriched composition (Field et al. 2008). The xenolith suite comprises Beaufort Group sandstone and Karoo basalt from the overlying strata, but is mostly dominated by mantle eclogites (> 95%), as well as xenocrysts and severely altered peridotites, often diamond-bearing (Hatton 1978). Eclogite xenoliths interpreted to have retained the age of their formation have been previously dated at 1.7 Ga (Jagoutz et al. 1984), 1.4–2.2 Ga (Jacob et al. 2005), 0.7–1.1 Ga and 1.2–1.5 Ga (Huang et al. 2012) and 0.9–1.9 Ga (Gonzaga et al. 2010).

Fig. 1
figure 1

Map of the Kalahari craton and separate terrains including the Kaapvaal and Zimbabwe cratons and the Limpopo Mobile Belt. K Kimberly cluster, P Premier/Cullinan cluster, M Murowa/Sese kimberlite, V Venetia kimberlite cluster, RV Roberts Victor mine, F Finsch, J Jagersfontein mine, S Star, L Letseng, M Monastery, O Orapa, Lth Letlhakane. After Van der Meer et al. (2013) and Katayama et al. (2009)

Full size image

The Jagersfontein Mine (29°46′ S–25°25′ E) is located ca. 170 km southeast of Kimberly, close to the craton margin. It is one of the large Group-1 kimberlite pipes, dated at ~ 85 Ma (Smith et al. 1985), containing olivine, monticellite, calcite, phlogopite, ilmenite, spinel, perovskite and apatite in a micaceous groundmass (Field et al. 2008), with a relatively coarse-grained texture. The xenolith suite includes peridotites, eclogites, pyroxenites and megacrysts. Two main metasomatic imprints (hydrous and carbonate) have been described for the eclogite suite (Pyle and Haggerty 1998). Both peridotitic and eclogitic inclusions have been found contained within diamonds (Field et al. 2008). Some xenoliths exhibiting omphacite exsolution in garnet have been interpreted as being sourced from the upper Transition Zone limit (Haggerty and Sautter 1990). The only age constraints for eclogite formation are inferred from sulphide inclusions in diamonds (1.1–1.7 Ga), and the strongly radiogenic 187Os/188Os has been interpreted as possible indicator for an ancient, remobilized eclogitic source (Aulbach et al. 2009).

Petrography

The primary mineralogy of the eclogite xenoliths analyzed in this paper consists of a high-grade mineral assemblage of garnet (gt) and omphacitic clinopyroxene (cpx). Most are bimineralic, with rutile (ru) as the main accessory phase. Less frequently, they may contain coesite (typically found inverted to palisade quartz), diamond, graphite, kyanite (ky) and corundum (cor) as primary minor phases. They have variable grain size, grain boundaries, as well as modal abundance and colour. A few samples show exsolution textures such as topotactic (crystallographically controlled) lamellae or lenses of garnet and/or kyanite in clinopyroxene, or needle-like rutile in both omphacite and garnet. Samples showing interactions with kimberlitic/carbonaceous fluids typically contain secondary phlogopite, plagioclase feldspar and/or sanidine. Textures and mineral assemblages for each sample are summarized in Appendix A in the electronic supplementary material.

Bimineralic eclogites

Bimineralic eclogites range in size from a few centimetres (Jagersfontein) up to > 20 cm in diameter (Roberts Victor) and consist of omphacite and garnet, and are characterized by two main textural types: described as metasomatized and non-metasomatized by MacGregor and Carter (1970) and Huang et al. (2012). The former is the most common (Type I Gréau et al. 2011; MacGregor and Carter 1970), and is defined by subhedral to rounded, or “cauliflower”-shaped garnet grains, with serrated grain boundaries (Fig. 2a, b), underlined by a finely crystallized, seemingly opaque, secondary rim. Garnet can be found as grain clusters, ‘necklaces' and less as isolated grains, marked by irregular intra-crystalline fine fractures associated with scattered glass and melt inclusions that give an overall ‘dusty' appearance. Garnet ranges in size and modal abundance, with colour from dark to pale orange but showing no textural zoning. Omphacite is present as inequigranular, sub- to anhedral grains, often with destabilized symplectic rims and with poorly preserved cores (Fig. 2a, b). The xenomorphous crystals are highly altered along irregular fractures by carbonate-rich fluids and form an interstitial matrix around garnet. The colour of omphacite varies from dark to light green (in plane polarized light), with irregular grain boundaries doubled by symplectic plagioclase, diopside and fine-grained phlogopite. Rutile is interstitial and texturally bound to the fine-grained secondary minerals.

Fig. 2
figure 2

Scanned thin sections of representative Type I (a, b), Type II (c–f) (MacGregor and Carter 1970) bimineralic, kyanite- (e) and corundum-bearing (f) eclogites from Roberts Victor (a, c, e, f) and Jagersfontein (b, d) mines

Full size image

Non-metasomatized eclogites (defined as Type II Gréau et al. 2011; MacGregor and Carter 1970) are less common than Type I, and are characterized by sub- to anhedral garnet, which is less fractured and more systematically coarse grained, and bordered by a fine opaque rim (Fig. 2c, d). The colour of the garnet varies from dark to pale orange. A few samples have fine, elongated anhedral garnet bordering clinopyroxene. Omphacite occurs as a sub- to anhedral, inequigranular matrix. Its colour ranges from dark to pale green, with well-preserved translucent cores, often surrounded by symplectic (diopside-plagioclase) rims of variable thickness. Often, rutile is found as needle-like exsolution spread throughout both garnet and omphacite.

High-Al eclogites

High-Al eclogites contain mineral phases such as kyanite and corundum. Kyanite-bearing xenoliths are characterized by sub- to anhedral, inequigranular garnets with straight to slightly indented grain boundaries (Fig. 2e), ranging from light to dark brown in colour. Omphacite is broken down into a white “spongy” clinopyroxene matrix, sometimes with preserved translucent green cores and xenoblastic grain boundaries. Kyanite is found as small (30–40 µm), subhedral to tabular or rounded, isolated or clustered grains. Microcrystalline, opaque rims surround grain boundaries.

Omphacite is the dominant mineral (~ 60 vol%) in corundum-bearing eclogites, ranging from small (~ 0.3 cm) rounded grains, to large (~ 2 cm) subhedral light green crystals (Fig. 2f). Garnet is found as subhedral and rounded inequigranular grains, with straight grain boundaries and of typically light brown colour. The modal abundance of corundum varies from ~ 2–17 vol% and is found as pale to intense pink, subhedral to anhedral crystals. Often corundum grains are surrounded by garnet or kyanite-garnet coronae. Rutile is scarce, mostly interstitial.

Three types of exsolution texture are present (Fig. 3): fine oriented lamellae, coronae and granular clusters. The topotaxial lamellae (~ 50–300 µm) consist of garnet or kyanite bordered by garnet. Some samples exhibit crosscutting garnet lamellae in-between the host omphacite cleavage planes (Fig. 3a). There is continuity between the lamellae and corona textures, as lamellae can be found blending into the coarse coronae. The corona around corundum grains can be monophasic—consisting exclusively of garnet, or polyphasic—consisting of kyanite and garnet (Fig. 3b). The granular clusters consist of rounded, fine-grained (~ 0.1 mm) garnet grains.

Fig. 3
figure 3

BSE images of representative exsolution textures in corundum-bearing eclogites from Roberts Victor. Large omphacite grains exhibit topotaxial garnet, garnet and kyanite exsolution lamellae (a). Some samples contain crosscutting garnet lamellae in-between the cleavage planes. Corundum grains are often surrounded by monophasic (garnet) or polyphasic (kyanite and garnet) coronae (b)

Full size image

Analytical techniques

Mineral (major element) compositions were acquired using a JEOL JXA-8320 electron microprobe at the Geology Department, Rhodes University, South Africa, equipped with four WDS, one FCS and two L-Type spectrometers and 12 diffraction crystals and at the Laboratoire Magmas et Volcans, Cézeaux, France, using a Cameca SX-100 electron microprobe, equipped with four WDS spectrometers and 12 diffraction crystals. Operating conditions were 15 keV and a beam current of 10 nA. The counting time was of 20 s for all elements, except K and Na (60 s), and 10 s for background. All data were processed using full ZAF Cameca PAP corrections.

Trace element compositions were acquired by laser ablation inductively coupled plasma spectrometry at the Laboratoire Magmas et Volcans, Cézeaux, France, using a Resonetics Resolution M-50 laser powered by an ultra-short pulse ATL AtlexExcimer laser system, beam diameter of 100 µm, 8 Hz repetition rate and 6 mJ laser pulse energy (detailed description in Müller et al. 2009). Reproducibility and accuracy of the analyses were estimated through repeated analyses of BCR-2g standard at the beginning and at the end of every run. Data reduction was carried out with the software package GLITTER® (Macquarie Research Ltd, 2001) (van Achterbergh et al. 2001). For each analysis, the time resolved signal for each element was monitored to discard perturbations related to inclusions, fractures or mixing.

Whole-rock compositions have been calculated based on individual mineral analyses and estimated modal abundance. Mineral proportions were determined on thin section scans and compositional maps (~ 10–20% uncertainty due to variable grain size and distribution), based on colour contrast between the different phases with the use of the Adobe®Photoshop® image-editing software. The errors in surface to volumetric conversions are negligible because the minerals concerned have similar densities and there is no preferred orientation of grains or layering in the rock.

The O isotope analyses were performed at the Department of Geological Sciences, University of Cape Town (UCT), South Africa, by the laser fluorination method described by Harris and Vogeli (2010). Mineral separates showed no optical evidence for alteration, fractures or inclusions. Duplicate and triplicate grain batches were analyzed for 15 out of a total of 36 samples and were reproducible within 0.67‰, and one of the samples with particularly low mineral δ18O values has been analyzed up to 4 times the garnet, with a maximum difference of 0.17‰. Between 2.5 and 3.0 mg of each mineral (~ 4 grains in average) were reacted in the presence of ~ 10 kPa of BrF5. The O2 produced was collected onto molecular sieve at liquid nitrogen temperatures. Oxygen isotope ratios were measured off-line using a Finnigan DeltaXP mass spectrometer, in dual inlet mode, on O2 gas. All values are reported relative to the SMOW scale, as δ = [(18O/16O) sample/(18O/16O) standard − 1] × 1000. The long-term average difference between duplicates of the in-house MONGT standard is 0.12‰ (n = 283), which corresponds to a 1σ error of 0.077.

Mineral chemistry

Worldwide, bimineralic eclogites that were metasomatized by fluids preceding the kimberlite magma (Type I) are characterized by alkali, LREE, Ba, Sr, and HFSE enrichment and are defined by ≥ 0.8 wt% K2O in clinopyroxene and/or ≥ 0.9 wt% Na2O in garnet (Gréau et al. 2011; Huang et al. 2012). Conversely, non-metasomatized eclogites (Type II) have alkali contents that are lower (Huang et al. 2012; McCandless and Gurney 1989) and are poor in LREE and LILE. Similarly, eclogite xenoliths have been divided into group A (Mg-rich), group B and group C (with increasing jadeite and grossular contents) based on the Ca–Mg ratio in garnet and Na content in clinopyroxene (Taylor and Neal 1989). The former had been interpreted as high-pressure cumulates and the latter as relics of a subducted oceanic crust. Selected mineral composition and structural formulae from the samples analyzed in this paper are reported in the electronic supplementary material.

Major elements

Eclogite xenoliths from both Roberts Victor and Jagersfontein are eclogites sensu stricto, with Na/(Na + Ca) in clinopyroxene > 0.20 (Clark and Papike 1968), and have a wide range in omphacite K2O contents (~ 0.02–0.34 wt%) and garnet Na2O contents (~ 0.01–0.16 wt%). They are roughly equally distributed among the Types IA, IB and IIA, IIB (as previously defined by Huang et al. 2012), respectively (Fig. 4), with variable FeO (6.23–21.62 wt%) and MgO (4.59–20.63 wt%) garnet content. There are three main compositional clusters for garnet endmembers: Py35–40Alm20−40Grs20−35; Py40–45Alm15−30Grs20−40 and Py45–75Alm20−45Grs5−15 and the samples range from groups A to C after Taylor and Neal (1989) (Fig. 5a) also shown by the wide spread in clinopyroxene composition with negatively correlated Na2O (1.11–8.81 wt%) and MgO (4.82–17.54 wt%) composition. Similarly, the omphacite compositions cover a broad range in Al2O3 and Na2O (~ 7–64% jadeite component), with a positive correlation (Fig. 5b).

Fig. 4
figure 4

Na2O in garnet vs. K2O in clinopyroxene for eclogite xenoliths from Roberts Victor and Jagersfontein mines. Bimineralic eclogites showing metasomatic textural traits (i.e., indented grain boundaries, numerous melt inclusions and secondary minerals) are marked with dark blue solid square symbols and eclogites with pristine textures are marked with black solid square symbols. Si–Al-rich eclogites are marked with blue (ky-bearing) and violet (cor-bearing) solid circle symbols. The two Type I and Type II fields are delimited as defined by McCandless and Gurney (1989) and Huang et al. (2012)

Full size image
Fig. 5
figure 5

Ternary grossular (Grs)–pyrope (Py)–almandine (Alm) plot (molar proportions) in garnet (a) and Na2O vs. Al2O3 plot in clinopyroxene (b) in eclogite xenoliths from the Kaapvaal craton comprising samples from Roberts Victor and Jagersfontein mines. Bimineralic eclogites are marked with black square solid symbols, ky-eclogites with blue solid circle symbols and cor-eclogites with purple solid circle symbols. A, B, C groups are delimited as defined by Taylor and Neal (1989)

Full size image

Kyanite-bearing samples from this study also cover a wide compositional range and belong to both the metasomatized—Type I and pristine—Type II categories (Radu et al. 2017), as well as in the transitional group. Garnet is typically MgO poor (~ 4.6–11.4 wt%). All metasomatized kyanite eclogites are characterized by low garnet FeO (~ 10.8–14.4 wt%) and jadeite-rich clinopyroxene (Jd50–57) and are classified as Type IB and IK (after Huang et al. 2012). Our non-metasomatized samples belong to the Type IIB group, except for RV333, which is a Type IIA. All kyanite-bearing eclogites are classified as Group B and Group C (Fig. 5) after Taylor and Neal (1989) and, with the exception of the two pyrope-rich samples, they are typically characterized by grossular-rich garnet and jadeite-rich clinopyroxene.

Corundum is restricted to the Type II—pristine eclogites (Fig. 4) that have very low clinopyroxene K2O content (0.01–0.02 wt%). The large, primary, garnet grains have higher MgO (14.9–16.5 wt%), specific to Type IIA, with respect to garnet exsolutions (4.59–11.8 wt% MgO), classified as Type IIB. Clinopyroxenes with exsolution textures are typically more magnesian (~ 5.97–9.98) and less sodic and aluminous (~ 4.07–5.51 wt% Na2O; 14.0–19.9 wt% Al2O3), with lower Ca-Tschermack (Ca-Ts) (~ 0.11–0.37) than non-exsolved, primary grains (~ 4.82–9.07 wt% MgO; 3.07–7.87 wt% Na2O; 16.5–20.4 wt% Al2O3; 0.16–0.51 CaTs). The Na and Mg content in clinopyroxene and the typically grossular-rich garnets with distinct trends towards Ca-enrichment for the exsolutions (Fig. 5), place corundum eclogites in groups B and C after Taylor and Neal (1989), proportional to their degree of exsolution.

Temperatures have been calculated with the garnet-clinopyroxene Fe–Mg geothermometer proposed by Krogh (1988) for 1 GPa, 3 GPa and 6 GPa, and the tendency lines were intercepted with the local conductive model geotherm (Pollack and Chapman 1977) that corresponds to a 39 mW m−2 surface heat flow (Griffin et al. 2003). Metasomatized, Type I eclogites, cover a wide range in pressure–temperature conditions, of ~ 800–1300 °C and ~ 3.2–6.5 GPa. Nevertheless, these temperatures remain unreliable, due to possible chemical disequilibrium induced by the percolating fluids. The non-metasomatized, bimineralic eclogites show a bi-modal distribution, where Type IIA corresponds to lower PT conditions (~ 650–1000 °C; ~ 2.0–4.5 GPa) than Type IIB (~ 1000–1300 °C; ~ 4.5–6.5 GPa). This indicates a compositionally stratified distribution of eclogites in the SCLM. PT equilibrium conditions yielded by ky- and cor- bearing eclogites are overall higher (~ 960–1150 °C; 4.1–5.6 GPa and ~ 950–1300 °C; 4.0–6.5 GPa, respectively) than the PT estimates for bimineralic eclogites specific for the lowermost part of the cratonic keel.

Trace elements

The previously defined Type I eclogites from this study are characterized by a narrow LREE and LILE concentration range, with variable degrees of LREE/HREE fractionation. Clinopyroxene has convex upward REE patterns, with (La/Nd)N of 0.16–0.56 and (Nd/Yb)N of 5.33–76.2 (subscript N denominates values normalized to the primitive mantle McDonough and Sun 1995). The Sr clinopyroxene concentration ranges between 2.5 and 19.9, roughly overlapping the ~ 5.0–15.1 estimated values for eclogitic clinopyroxene (Jacob et al. 2003). Garnet REE spectra have strongly fractionated LREE ((La/Nd)N of 0.005–0.26) and generally flat MREE-HREE ((Nd/Yb)N of 0.03–0.89). We distinguish two compositional groups, characterized by HREE-rich garnet (between 8 and 11 times the primitive mantle) and HREE-poor garnet (between 1.5 and 3 times).

Our Type II eclogites comprise two compositional groups and associated LREE and LILE distribution patterns, for both garnet and clinopyroxene (Fig. 6a, b). The first group predominantly consists of Type IIA eclogites and is characterized by clinopyroxene with LREE concentrations superior to the primitive mantle values (between 1 and 30 times primitive mantle). The clinopyroxene REE patterns have flat LREE distribution ((La/Nd)N of 0.20–1.38) and strongly fractionated MREE-HREE (Nd/Yb)N of 3.43–167.1 (Fig. 6a). The primitive mantle-normalized clinopyroxene Sr concentration ranges between 1.42 and 39.0. All extended trace element spectra of clinopyroxene are strongly fractionated (Nb/La)N ratio (0.001–0.50), and show positive Sr anomalies (1.71–30.0 Sr*, Sr* = SrN/sqrt(SmN*NdN)) (Fig. 6c). Similarly, garnet has high and slightly fractionated LREE profiles ((La/Nd)N of 0.01–0.05), and flat MREE-HREE distribution ((Nd/Yb)N of 0.08–1.68). The extended trace elements spectra have (Nb/La)N of 0.68–25.6 and weak positive Zr inflections are observed in garnet (Fig. 6b, d).

Fig. 6
figure 6

REE (a, b) and extended trace elements patterns (c, d) in clinopyroxene and garnet, normalized to the primitive mantle (McDonough and Sun 1995), in bimineralic, Type II (non-metasomatized) eclogites from the Kaapvaal craton comprising samples from Roberts Victor and Jagersfontein mines. Type IIA (Mg rich) are marked with blue solid square symbols and Type IIB (Mg poor) with orange solid square symbols

Full size image

The second compositional group includes both Type IIA and IIB eclogites (Fig. 6) and is defined by sub-primitive mantle LREE concentrations in clinopyroxene (nine times lower than primitive mantle values). REE patterns typically have fractionated LREE ((La/Sm)N of 0.03–0.01), less pronounced for HREE ((Sm/Yb)N of 4.04–13.2). Sr content in clinopyroxene (0.51–1.22 SrN) is inferior to the LREE-rich group and typical eclogitic values. Although less pronounced, all LREE-poor clinopyroxene have positive Sr anomalies (Sr* 1.40–7.95). Garnet composition is LREE depleted, with strongly fractioned (La/Sm)N of 0.001–0.03 and flat or slightly enriched HREE distribution ((Sm/Yb)N of 0.20–2.83). The extended trace element patterns show negative inflections in Zr and Hf.

Kyanite-bearing eclogites have convex upward clinopyroxene REE patterns (Fig. 7), with fractioned LREE (0.03–9.39 (La/Nd)N) and HREE (19.6–114 (Nd/Lu)N). Sr primitive mantle-normalized concentrations are typically between ~ 5 and 22 times higher than primitive mantle values. Garnet REE spectra have fractionated LREE (0.003–0.06 (La/Sm)N) and flat HREE with 1.08–1.97 (Sm/Yb)N. All samples have positive Sr anomalies in clinopyroxene (4.78–26.1 Sr*). Similarly, both clinopyroxene and garnet have positive Eu anomalies (Eu*cpx 1.05–1.55, Eugt 1.06–3.97, Eu* = EuN/sqrt(SmN*GdN)), except for sample RV168 where they are preserved only in the garnet fraction. The extended incompatible element patterns show additional positive inflections for Pb and Hf in clinopyroxene and equivalent negative inflections in garnet.

Fig. 7
figure 7

REE (a, b) and extended trace elements patterns (c, d) in clinopyroxene and garnet, normalized to the primitive mantle (McDonough and Sun 1995), in kyanite (blue solid circle symbols) and corundum eclogites (violet solid circle symbols) from Roberts Victor and Jagersfontein mines

Full size image

Minerals in corundum-bearing eclogites have overall incompatible elements concentrations lower than the primitive mantle (Fig. 7). Sample RV179 has fractionated LREE and MREE ((La/Sm)N = 2.97) and HREE-enriched ((Sm/Lu)N = 0.79) clinopyroxene patterns; and convex downward garnet REE spectra, with fractionated LREE and MREE ((La/Sm)N = 0.16) and flat HREE ((Sm/Lu)N = 0.44). Sample RV344 has convex upward clinopyroxene patterns ((La/Sm)N = 0.03; (Sm/Lu)N = 10.4) and garnet REE patterns with strongly fractionated LREE ((La/Sm)N = 0.003) and MREE ((Sm/Dy)N = 0.47) and flat HREE ((Sm/Lu)N = 1.02), higher than the primitive mantle. Both samples have positive Eu anomalies in clinopyroxene and garnet (Eu*cpx 1.11–4.91, Eu*gt 1.13–4.16), and sample RV179 shows positive Sr anomaly in clinopyroxene (Sr* = 12.7).

Oxygen isotopes

The δ18O values of garnet, clinopyroxene, kyanite and corundum from this study (Table 1) range from higher to significantly lower than the mantle (+ 5.5 ± 0.4‰ Mattey et al. 1994) (Fig. 8). Whole-rock δ18O values estimated from mineral δ18O values and modal proportions (see “Analytical techniques” for method) vary from 1.4‰ to 7.5‰.

Table 1 δ18O values (the ‰ deviation from 18O/16O in the standard) of garnet, clinopyroxene, kyanite, corundum and recalculated bulk rock based on mineral abundance in eclogite xenoliths from Roberts Victor and Jagersfontein mines, South Africa. PT estimates were determined with Krogh (1988) geothermometer, see “Major elements” for method
Full size table
Fig. 8
figure 8

Plot of a δ18O values of garnet (gt), corundum (cor) and kyanite (ky) vs. clinopyroxene (cpx) from this study; b Δgt−cpx vs. temperature (°C) calculated from Fe–Mg exchange between garnet and clinopyroxene after Krogh (1988). Metasomatized (Type I) bimineralic eclogites (Roberts Victor mine) are marked with dark blue solid square symbol. Non-metasomatized (Type II) bimineralic eclogites are marked with black solid square symbols. Ky-cor-bearing samples are marked with light blue (kyanite) and violet (corundum) solid circle (gt), diamond (ky) and triangle (cor) symbols. Red solid square marks ‘normal’ mantle values 5.5 ± 0.4‰ (Mattey et al. 1994). Analytical error bars are inferior to the symbol size

Full size image

Three metasomatized (Type I) samples have been analyzed for comparison. The δ18O values of garnet range from 5.4 to 6.1‰. Among the non-metasomatized eclogites, we distinguish low-δ18O eclogites, which have garnet values ranging from 1.1 to 4.8‰ in agreement with the coexisting clinopyroxene (1.4–4.9‰), and high-δ18O eclogites, with mineral δ18O values that range from the expected ranges for peridotitic mantle (5.2–6.0‰ in garnet and 5.2–6.0‰ in clinopyroxene), to higher values (6.0–7.5‰ in garnet and 5.5–7.5‰ in clinopyroxene).

The low-δ18O group includes bimineralic samples, as well as corundum-bearing eclogites. The former have garnet values between 2.3 and 3.9‰ and between 4.5 and 4.8‰ (Fig. 9). The lowermost values typically correspond to the low-magnesium (Type IIB) eclogites characterized by high-pressure–temperature equilibrium conditions, whereas the higher δ18O values correspond exclusively to the magnesium-rich, Type IIA eclogites, characterized by lower PT conditions. Similarly, the high-δ18O group with mantle-like or higher δ18O values corresponds strictly to Type IIA eclogites (Fig. 9). The corundum-bearing eclogites have δ18O values lower than the ‘normal’ mantle, with whole rock estimates of 1.4 and 4.8‰. Moreover, sample RV344 has the lowest δ18O values recorded so far in mantle eclogites, with a garnet average of 1.1 (n = 4, 1σ = 0.22). By contrast, kyanite-bearing (~ 5.6–6.7‰) eclogites have greater than equal to ‘normal’ mantle δ18O values.

Fig. 9
figure 9

Garnet δ18O vs. a MgO in garnet and b temperature (°C) calculated from Fe–Mg exchange between garnet and clinopyroxene after Krogh (1988) in bimineralic Type II (non-metasomatized) and corundum-bearing eclogites from this study. Pristine eclogites with Mg-rich garnets (Type IIA) have generally higher δ18O values and correspond to lower PT conditions. Non-metasomatized eclogites with Mg-poor garnets (Type IIB) have lower δ18O values and correspond to higher PT equilibrium conditions. Type IIA is marked with blue solid square symbols, Type IIB with orange solid square symbols and corundum-bearing eclogites are marked with violet solid circle symbols

Full size image

Discussion

Origin and evolution of mantle eclogites

Extensively metasomatized eclogites can be distinguished from non-metasomatized eclogites based on their destabilized textures (e.g., serrated grain boundaries, abundant melt inclusions), high alkali and LREE compositions (Aulbach et al. 2016; Aulbach and Viljoen 2015; Gréau et al. 2011; Huang et al. 2012). Although some authors have attempted to recalculate the composition of pristine eclogites from metasomatized samples, (Aulbach and Jacob 2016; Aulbach and Viljoen 2015), it is the non-metasomatized eclogites that most likely represent the closest composition to their protolith (Gréau et al. 2011; Radu et al. 2017).

The Archaean Earth would have had higher mantle heatflow expressed as potential temperature TP (McKenzie and Bickle 1988), leading to a higher degree of adiabatic melting compared to modern mid-oceanic ridge (MOR) settings (Herzberg 2011). In this situation, the primary magmas would have had higher Mg and lower Si and Na contents compared to present-day MOR basalts or boninites (Aulbach and Jacob 2016; Herzberg and O’Hara 1998; Jacob and Foley 1999). The average potential protolith made of Archaean oceanic crust is, therefore, assumed to be compositionally equivalent to picrite (MgO ~ 22.0 wt%) (Aulbach and Viljoen 2015; Jacob and Foley 1999; Schmickler et al. 2004). For comparison, a basaltic composition corresponding to modern MOR is assumed, as well as gabbro for low-pressure plagioclase-rich cumulate, with granophyre as the most evolved term of an evolving mafic magma (Fig. 11).

Bimineralic eclogites

Metasomatized eclogites

The Type I (Na2O in garnet > 0.9 wt%; K2O in omphacite > 0.8 wt%), metasomatized eclogite, from our study has textural aspects consistent with fluid interaction: numerous inclusions, indented crystal boundaries and subrounded to rounded grains. The reconstructed whole rock trace elements composition of metasomatized eclogites duplicates the two previously observed compositional groups characterized by HREE-rich (8–11 times the primitive mantle) and HREE-poor (1.5–3 times the primitive mantle) garnets (Fig. 10). The former shows typically flat REE profiles with slightly fractionated LREE, (La/Sm)N of 0.10–0.56 and (Sm/Lu)N of 0.31–0.83, similar to a tholeiitic basalt composition (Fig. 10a, c). The latter group shows comparable LREE distribution, (La/Sm)N of 0.05–0.55 and more fractionated HREE, (Sm/Lu)N of 1.53–3.73, with significantly lower MREE-HREE concentrations (Fig. 10b, d). There is a distinct correlation between the amount of HREE (Sm–Lu) and the FeO and CaO contents in garnet. Eclogites containing garnets with low-Ca and high-Fe contents typically have high-bulk ΣHREE, whereas samples with Ca-rich, Fe-poor garnets have low-bulk ΣHREE composition. This could be due to the substitution of HREE for Ca, more markedly in the garnet than in the clinopyroxene structure, controlled by similar ionic radii (Harte and Kirkley 1997) or to enrichment by silicate melts (Aulbach et al. 2016).

Fig. 10
figure 10

Reconstructed whole-rock extended trace (a, b) and REE patterns (c, d) in bimineralic Type I (metasomatized) eclogites from the Kaapvaal craton compared to tholeiitic basalt composition (GEOROC compilation). Primitive mantle values are from McDonough and Sun (1995)

Full size image

Non-metasomatized eclogites

The Type II eclogite xenoliths from this study are in apparent textural equilibrium, with translucent subhedral grains, showing no evidence of extensive metasomatism and are characterized by garnet with Na2O < 0.9 wt% and K2O in omphacite < 0.8 wt%. The reconstructed whole-rock compositions of non-metasomatized eclogites correlate with the previously described compositional types IIA and IIB.

Type IIA eclogite bulk compositions are rich in MgO (11.15–19.71 wt%) and vary broadly from low to high CaO (5.85–17.52 wt%), Al2O3 (6.23–22.18 wt%) and Na2O (0.17–4.27 wt%) contents (Fig. 11). The decreasing Mg# could be following olivine (± pyroxene) fractionation in an oceanic crustal protolith, translated also by the increasing FeO content (Aulbach and Viljoen 2015). Furthermore, Type IIA eclogites systematically show an inverse correlation between Na2O and Al2O3 content with respect to MgO (Fig. 11c, d), which can be equally accounted for by progressive fractional crystallization of olivine ± pyroxene in an evolving “basaltic” liquid (Aulbach et al. 2016). The high CaO and, less marked, Al2O3 contents at relatively constant MgO are interpreted as evidence for a small fraction of plagioclase (± diopsidic clinopyroxene) accumulation, as main mineral phases preferentially incorporating Ca with respect to Mg (Beard et al. 1996). The REE and extended incompatible trace element patterns of reconstructed Type IIA whole-rock compositions (Fig. 12a, b) are predominantly flat and superior to primitive mantle values (0.8–20 times higher). These compositions are comparable to average oceanic picrite content with the exception of Nb, Ta and La. Moreover, Type IIA eclogites show positive correlation between ΣREE and Y concentrations, with respect to TiO2 content. It is inferred that REE and HFSE enrichment is caused by olivine fractionation (Aulbach et al. 2016). All Type IIA eclogites register positive Eu* (1.06–2.71) and Sr* (1.11–12.21) anomalies, inversely correlated with TiO2 content (Fig. 11e, f). This is in agreement with the previous inference, arguing for the presence of plagioclase in a protolith issued from an evolved picritic liquid for Type IIA eclogites (Beard et al. 1996; Jacob and Foley 1999).

Fig. 11
figure 11

Relationships between major elements, Eu* (Eu* = EuN/sqrt(SmN*GdN)) and Sr* (Sr* = SrN/sqrt(SmN*NdN)) in reconstructed whole-rock compositions of bimineralic Type II (non-metasomatized) eclogites, kyanite- and corundum-bearing eclogites from the Kaapvaal craton, and magma series endmembers (from GEOROC compilation). The ultramafic pole is represented by Icelandic picrite and the mafic pole by olivine-bearing tholeiite, whereas the felsic endmember is represented by granophyre and a plagioclase cumulate composition is represented by gabbro. The subscript “N” shows primitive mantle-normalized concentrations (McDonough and Sun 1995)

Full size image
Fig. 12
figure 12

Reconstructed whole-rock REE (a, c) and extended trace elements patterns (b, d) in bimineralic Type II (non-metasomatized) eclogites from the Roberts Victor and Jagersfontein mines compared to tholeiitic basalt and picrite compositions (GEOROC compilation). Primitive mantle values are from McDonough and Sun (1995)

Full size image

The reconstructed bulk rock compositions of Type IIB eclogites have low MgO (6.93–10.12 wt%) and medium to high CaO (10.66–15.54 wt%), Al2O3 (14.26–20.10 wt%) and Na2O (2.58–3.96 wt%) contents. Covering a lower and narrower range of values, MgO content shows a similar positive correlation with Mg# and inverse correlation with Al2O3 and Na2O as for Type IIA eclogites (Fig. 11a–d). It is inferred the progressive enrichment in Al2O3 and Na2O along with MgO depletion is the result of olivine (± pyroxene) fractional crystallization in oceanic crust (Aulbach et al. 2016; Aulbach and Viljoen 2015), precursory to the eclogite formation. Concurrently, iron oxide crystallization and plagioclase accumulation could lead to progressive FeO decrease and CaO increase, respectively. The REE and extended incompatible elements patterns of reconstructed Type IIB whole-rock compositions (Fig. 12c, d) have strongly fractionated LREE ((La/Sm)N ~ 0.001) and flat to slightly enriched MREE-HREE ((Sm/Lu)N ~ 0.07–0.19) predominantly higher than primitive mantle values (~ 5–10 times). The high-HREE concentrations, with flat to enriched distribution, are similar to average picrite or tholeiitic basalt compositions. Strongly fractionated LREE may be the result of partial melt loss during burial. All samples have negative Zr and Hf anomalies, which could indicate a metasomatic HFSE depletion of the protolith source (Aulbach et al. 2011). Most Type IIB samples have positive Eu* (0.90–1.61), with variable Sr* (0.18–3.78) from negative to positive anomalies, inversely correlated with TiO2 content and have a positive correlation between ΣREE and Y concentrations with respect to TiO2 content, interpreted as the result of olivine fractionation (Aulbach and Viljoen 2015). These composition trends support the interpretation that Type IIB eclogites are derived from a picritic to basaltic protolith which underwent differentiation, (Aulbach et al. 2016; Beard et al. 1996; Jacob 2004; Jacob and Foley 1999; Taylor and Neal 1989), followed by small degrees of partial melting during eclogitization and burial (Schulze et al. 2000; Shervais et al. 1988; Viljoen et al. 2005).

High-Al eclogites

The kyanite- and corundum-bearing eclogites have low reconstructed bulk MgO content (5.37–11.47 wt%), and have medium to high CaO (13.03–17.54 wt%) and Al2O3 (22.62–33.34 wt%) contents. The FeO content (1.84–7.43 wt%) progressively decreases from kyanite- to corundum-bearing eclogites. Conversely, Na2O (1.78–4.56 wt%) and Mg# progressively increase from kyanite-(0.34–0.69) to corundum-bearing eclogites (0.71–0.75) (Fig. 11a-d) without a strong variation of MgO content, interpreted as evidence for a protolith dominated by plagioclase accumulation possibly concurrent with iron oxide crystallization (Agashev et al. 2018; Schulze et al. 2000; Shervais et al. 1988; Shu et al. 2016).

Reconstructed bulk trace element compositions of kyanite-bearing eclogites (Fig. 13a, b) have REE patterns with poorly fractionated, enriched LREE ((La/Sm)N 0.61–1.18) and slightly fractionated to flat HREE ((Sm/Lu)N 2.83–3.94). These incompatible element concentrations are similar to a gabbroic composition and are marked by positive Eu* (1.31–1.46) and Sr* (1.63–4.61) anomalies (Jacob 2004). This is further supported by decreasing TiO2 contents with ΣREE and Y, interpreted as evidence for plagioclase accumulation, equally translated into increasing Eu* and Sr*, from negative to positive anomalies, inversely correlated with TiO2 content (Aulbach et al. 2016). Reconstructed whole-rock composition of corundum-bearing RV179 eclogite (Fig. 13c, d) has flat LREE and HREE, ((La/Nd)N=1.03) and ((Dy/Lu)N=0.88), and concave MREE patterns; incompatible element concentrations inferior to the primitive mantle (~ 9 times less) and pronounced positive anomalies in Eu and Sr (Eu*=4.32; Sr*=5.38). This is consistent with a pyroxene-plagioclase cumulate protolith. Conversely, corundum-bearing sample RV344 has reconstructed whole-rock trace element compositions with fractionated LREE ((Ce/Sm)N=0.04) and MREE ((Sm/Dy)N=0.39) and almost flat HREE ((Dy/Lu)N=0.88), having MREE-HREE concentrations superior to the primitive mantle (~ 3 times primitive mantle), and less pronounced Eu-positive anomaly (Eu*=1.12) and negative Sr* anomaly (~ 0.09). Based on the recalculated bulk rock composition, sample RV344 is interpreted as more likely derived from a gabbroic protolith which has undergone partial melting (Shu et al. 2016).

Fig. 13
figure 13

Reconstructed whole-rock REE and extended trace elements patterns in kyanite-(a, b) and corundum-bearing eclogites (c, d) from the Kaapvaal craton, compared to the composition of gabbro and pyroxenite, respectively (GEOROC compilation)

Full size image

Pressure–temperature conditions (based on equilibrium Fe–Mg exchange estimated temperatures) for kyanite-bearing eclogites were obtained in the range of ~ 4.1–5.6 GPa and ~ 960–1150 °C and at ~ 4.0–6.5 GPa and ~ 950–1300 °C for corundum-bearing eclogites. This is coherently sustained by textural and chemical arguments: kyanite-bearing samples exhibit cloudy clinopyroxene, formed through inner structure breakdown by fast decompression, as well as by the high-jadeite and vacancy-bearing CaEsk content in the preserved omphacite cores attesting to their origin from high pressure (Schroeder-Frerkes et al. 2016). Nevertheless, the connection with the bimineralic eclogites and their geometric relation is not yet clear.

Oxygen isotope constraints on eclogite petrogenesis

Oxygen isotopes are robust geochemical tracers because oxygen is ~ 50% of the rock; and solid-state diffusion rates are slow, even in the mantle (e.g., Peck et al. 2003 and references therein). Crustal material can, therefore, be assumed to preserve its δ18O values during the process of recycling into the mantle (Jacob et al. 1994; Neal et al. 1990). Moreover, no significant change in δ18O value would be expected as a result of interaction with potential kimberlite-derived fluids, and mantle metasomatism should change δ18O values towards a mantle-like range. Numerous studies of eclogites have used oxygen isotopes to constrain their protolith, leading to an important database of δ18O variations in mantle eclogites. However, many of these studies on South African eclogites are focused on fully (Garlick et al. 1971; Gréau et al. 2011; MacGregor and Manton 1986; Schulze et al. 2000) or partly metasomatized (Riches et al. 2016a) eclogites, and only few non-metasomatized eclogites have been analyzed (Huang et al. 2016; Jacob et al. 2005; Viljoen et al. 2005).

The strong correlation (Fig. 8a) between garnet and omphacite δ18O values (r2 = 0.98) is interpreted as evidence that the omphacite δ18O has not changed with respect to that of the co-existing garnet since their formation. The difference between clinopyroxene and garnet δ18O values (Δcpx−gt) ranges from − 0.22 to + 0.31‰ (n = 35), except for samples: JG29 (− 0.64 to − 0.74‰), RV203 (− 0.45‰), RV209 (+ 0.44‰), RV220 (+ 0.42‰) and RV320 (+ 0.45‰); the Δgt−ky and Δgt−cor range from − 0.37 to − 0.30‰ (n = 3) and from + 0.05 and + 0.31‰ (n = 4), respectively. A slight increase in the Δcpx−gt is observed with increasing temperature (Fig. 8b), contrary to what has been previously observed for Lace eclogites (Aulbach et al. 2017b). Constant differences in δ18O values in the Δgt−ky and Δgt−cor present in each sample is additional evidence for oxygen isotope equilibrium at mantle temperatures. On the basis of the above, it is concluded that the bulk δ18O values of the eclogite xenoliths estimated from mineral data in this work approximate the δ18O values of the protolith.

Recalculated bulk rock δ18O values for the Type IIA (magnesium rich) eclogites vary between 3.0 and 7.5‰. Values lower than average mantle are most easily explained by high-temperature hydrothermal alteration of the original ocean crust, whereas higher than 5.5‰ δ18O values can be explained by low-temperature alteration (Bindeman 2008). High and low-δ18O values are present in the upper and lower part of the oceanic crust, respectively (Beard et al. 1996; Eiler 2001; Gregory and Taylor 1981; Jacob et al. 1994; Schulze and Helmstaedt 1988; Shervais et al. 1988). Similarly, we attribute the recalculated bulk rock δ18O values from 2.3 to 3.9‰, for our Type IIB (low magnesium) eclogites to interaction of oceanic crust with seawater at high temperatures (~ 350 °C Bowers and Taylor Jr. 1985) (Eiler 2001; Garlick et al. 1971; MacGregor and Manton 1986; Riches et al. 2016a; Shervais et al. 1988; Vogel and Garlick 1970).

Taking into account either of the two main models for TTG formation: partial melting of oceanic crust during subduction or anatexis of thickened lower crust in the roots of oceanic plateaus (Zhang et al. 2009), a limited contribution of crustal material is typically expected (Condie 2013). This is consistent with the δ18O values previously recorded in TTGs and adakites, which range from 5.4 ± 0.3‰ (King et al. 2000) to above the mantle values 7.1–8.3‰ (Bindeman et al. 2005; Faure and Harris 1991; Whalen et al. 2002). If mantle eclogites were melt residues from Archaean crust formation, they would have δ18O values that are only slightly lower (< 0.4‰, Korolev et al. 2018), as a function of the degree of melt produced and extracted and of the modal abundance of omphacite and garnet (i.e., increasing melt fraction leads to a progressively 18O-enriched residue), with modelled values in the range of 5.2–5.8‰ (Bindeman et al. 2005). Therefore, the TTG residue model fails to account for the variability in δ18O values of our mantle eclogites (Table 1) and, in particular the δ18O values that are lower than the peridotitic mantle.

Estimated bulk rock δ18O values are between 5.6 and 6.7‰ for kyanite-bearing eclogites and between 1.4 and 4.8‰ for corundum-bearing eclogites. Compared to bimineralic eclogites, kyanite-bearing eclogites are interpreted to be derived from oceanic crust whose upper part underwent isotopic exchange during low-temperature alteration (Eiler 2001; Garlick et al. 1971; Schulze et al. 2013), and corundum-bearing eclogites are interpreted to be derived from the lower section of an oceanic crust that was subjected to high-temperature alteration (McCulloch et al. 1981; Shervais et al. 1988; Vogel and Garlick 1970). The garnet δ18O values of 1.1‰ (n = 4, 1σ = 0.22) in one of the corundum eclogites is significantly lower than any other δ18O values previously measured in Phanerozoic ophiolites. There are several possible explanations for these low-δ18O values. (1) They could be preserved from an oceanic crust derived from an Archaean mantle having lower δ18O values than its more modern counterparts; this is supported by a recent study on Archaean komatiites from Barberton where, based on δ18O values of fresh olivine crystals, it was inferred that mantle-derived magmas in the Archaean had significantly lower δ18O values than the mantle today [3.5 ± 0.6‰ Byerly et al. (2017)]. (2) Low-δ18O values recorded in this study could be explained if the seawater involved in the alteration had lower δ18O values than the seawater that infiltrated Phanerozoic ophiolites (Gregory and Taylor 1981). (3) Alteration could have taken place at higher temperatures than in the case of Phanerozoic ophiolites, leading to a smaller rock-water oxygen fractionation. This will be discussed further below.

Influence of mantle metasomatism on δ18O

Some authors have put forward a metasomatic origin for δ18O in eclogite xenoliths caused by interaction with carbonaceous, hydrous melts or fluids (Gréau et al. 2011; Huang et al. 2012). This model, is based on intra-sample variability in garnet δ18O value, correlating with tracers of metasomatism (i.e., LREE, HFSE, Sr, etc.), and decoupling between the oxygen and magnesium isotope composition (Huang et al. 2016).

There are several issues concerning the metasomatic interpretation of oxygen isotope values in mantle eclogites that we will address point by point. First, no variability between mineral replicates and no correlation between typical metasomatic indicators (e.g., LREE, Hf, Zr) and δ18O values are observed in the extended set of non-metasomatized eclogites from this study, where both lower and higher δ18O values with respect to ‘normal’ mantle are present. Second, according to a recent study of both oxygen and magnesium isotopes in mantle eclogites (Huang et al. 2016), the variable δ26Mg (− 1.09 to − 0.17‰) with constant δ18O values (2.34–2.91‰) are interpreted as evidence for a metasomatism arguing that they could not be inherited from an altered oceanic crustal protolith. However, studies on both continental (Li et al. 2011; Wang et al. 2014) and cratonic eclogites (Wang et al. 2012) show that a wide range of δ26Mg values (− 1.38 to + 0.17‰, Wang et al. 2015) can be inherited from an oceanic crust which underwent hydrothermal alteration prior to subduction.

Moreover, a link between low-δ18O values (< 5‰) and mantle metasomatism is inherently unlikely for reasons of mass balance. Oxygen is the most abundant element in mantle rock forming minerals (about 45% by weight). In this case, even a small (i.e., < 1‰) change in the oxygen isotopic ratio of mantle eclogites, with respect to the ‘normal’ mantle values [+ 5.5 ± 0.4‰ Mattey et al. (1994)] would require substantial amounts of fluids with unrealistically low-δ18O values (Ionov et al. 1994; Riches et al. 2016a). Mass balance calculations for a volumetrically significant reactive agent that could change δ18O values in eclogites without equilibrating with the surrounding mantle shows that a 1‰ change requires a maintained continuous flow from the source and across the upper mantle, of a fluid extremely different in δ18O values, at a ≥ 2:5 fluid–rock ratio (Riches et al. 2016a). A metasomatic agent with extremely low-δ18O values, overprinting the lower part of the cratonic root, without re-equilibrating with the surrounding peridotitic mantle seems unlikely and any mantle metasomatic agent is predicted to either shift the δ18O values towards mantle-like values or have a negligible contribution (Aulbach et al. 2017b; Czas 2018). We, therefore, infer that oxygen isotope signature measured in eclogite xenoliths does not have a mantle source.

Origin of oxygen isotope variations in the protolith

Oxygen isotope ratios in mafic rocks may vary with respect to ‘normal’ mantle values as a result of fluid–rock interaction at variable temperature and crustal contamination. The δ18O values of the eclogitic garnets from Roberts Victor vary from 1.1‰ to 7.4‰. This partly overlaps previously documented δ18O values in eclogite suites worldwide (Beard et al. 1996; Caporuscio 1990; Garlick et al. 1971; Gréau et al. 2011; Huang et al. 2014, 2016; Jagoutz et al. 1984; MacGregor and Manton 1986; Neal et al. 1990; Ongley et al. 1987; Riches et al. 2016b; Schulze et al. 2000; Shervais et al. 1988; Sommer et al. 2017) (Fig. 14) and includes values that are lower than those previously measured. Bimineralic eclogites have “typical” δ18O values according to the majority of the eclogites previously documented (Fig. 15a–c). Nevertheless, most of the oxygen isotope database on mantle eclogites from the Kaapvaal craton seems to be on metasomatized eclogites, based on their given textures or high alkali content. Conversely, pristine eclogites have been significantly less documented due to their low abundance. Unlike Type I eclogites that typically have mantle or higher than mantle δ18O values, pristine Type II eclogites have lower values (with ~ 77% of δ18O in garnet < 5‰).

Fig. 14
figure 14

Oxygen isotope compositions of eclogite garnets from Jagersfontein and Roberts Victor mines analyzed in this study (black bars) and literature data from worldwide eclogite suites (pink bars): Orapa (Aulbach et al. 2017a); Udachnaya (Jacob et al. 1994; Shatsky et al. 2016); Mir (Beard et al. 1996); Lace (Aulbach et al. 2017b); Roberts Victor (Garlick et al. 1971; Jagoutz et al. 1984; Ongley et al. 1987; MacGregor and Manton 1986; Caporuscio 1990; Schulze et al. 2000; Gréau et al. 2011; Huang et al. 2014, 2016; Riches et al. 2016a, b); Kaalvallei (Viljoen et al. 2005); Bellsbank (Neal et al. 1990; Shervais et al. 1988); Bobbejaan (Caporuscio 1990); Koidu (Barth et al. 2001, 2002)

Full size image
Fig. 15
figure 15

Histograms showing oxygen isotope compositions of garnets from Roberts Victor mine from this study, compared to literature data from: Caporuscio (1990); Garlick et al. (1971); Gréau et al. (2011); Huang et al. (2014, 2016); Jagoutz et al. (1984); MacGregor and Manton (1986); Ongley et al. (1987); Riches et al. (2016a, b); Schulze et al. (2000), in a Type I (metasomatized) eclogites; b in Type II (non-metasomatized) eclogites; c in Type II (black), Type I eclogites (dark blue), kyanite- (light blue) and corundum-bearing (violet) eclogite xenoliths; d in bimineralic Type IIA (high Mg) and Type IIB (low Mg) eclogites

Full size image

As previously discussed, the non-metasomatized high Mg, Type IIA, eclogites characterized by lower equilibrium temperatures have a wide range in δ18O values, from lower to higher than the ‘normal’ mantle, whereas Type IIB eclogites, with a low-Mg content and high equilibrium temperatures always have lower δ18O values (Fig. 15d). Kyanite-bearing eclogites have δ18O values within or above the ‘normal’ mantle range, similar to Type IIA eclogites, whereas the corundum-bearing samples have lower δ18O values than peridotitic mantle. Sample RV344 has average garnet δ18O values of 1.1‰ (n = 4, 1σ = 0.2), ~ 1‰ lower than any previously recorded value in mantle eclogites (2.20‰ MacGregor and Manton 1986).

The variation in δ18O values cannot be accounted for by fractionation under high-temperature mantle conditions (Bindeman 2008 and references therein; Shervais et al. 1988) and are, therefore, not consistent with a mantle origin for eclogite xenoliths that does not involve crustal recycling. It is inferred the bulk eclogite δ18O values we estimated most likely reflect the range of δ18O values of a subducted oceanic crust profile. Taking into account that low-δ18O values are typically restricted to the lower segment of the oceanic crust and high-δ18O values to the upper section (Gregory and Taylor 1981), it is unclear if the two eclogite types (i.e. IIA; IIB) are derived from different parts of the same oceanic crust (i.e., ~ 150 km; ~ 210 km) or are derived from oceanic crust subducted at different times. Korolev et al. (2018) quantified the effects of typical processes associated with subduction (i.e., dehydration, partial melting, exogenous fluid flux, and progressive garnet overgrowth) on δ18O values to be less than 1‰ individually or ~ 2.5‰ a possible total contribution; and of local mantle metasomatism as typically no more than 1–1.2‰. Nevertheless, the effects of metasomatism are strongly dependant on the difference between the initial rock composition and the δ18O values of the metasomatic agent. It is, therefore, possible that the initial composition of the protolith has been slightly affected during subduction and concurrent metamorphism, nevertheless this excludes a mantle origin for the δ18O values measured.

The δ18O values recorded in our samples extend at least 2‰ lower than δ18O values previously obtained from Cretaceous ophiolites (Alt and Bach 2006; Gregory and Taylor 1981). To produce δ18O values as low as ~ 1‰ in the Phanerozoic (assuming a δ18O value of seawater = 0‰), seawater–rock interactions would require either higher water to rock ratio or higher temperatures than envisaged for the Cretaceous ophiolites, which seem improbable (Kastings et al. 2006 and references therein). Most of the oxygen budget on Earth is contained in silicate minerals in the crust and mantle (Walker and Lohmann 1989) and the oxygen isotope record of the hydrosphere is controlled by the exchange of oxygen with silicate rocks (Perry and Ahmad 1983). Although in a recent study on Archaean komatiites, the δ18O values of olivine phenocrysts from Barberton (3.5 ± 0.6‰) are interpreted as evidence for lower δ18O values in the bulk Archaean mantle (Byerly et al. 2017), this seems unlikely to be the general case on the grounds of mass balance, as there is no evidence for a δ18O-enriched, unaltered Archaean oceanic crust. This is further supported by δ18O values in Archaean eclogite xenoliths (e.g., Aulbach et al. 2017a, b; Jacob 2004; Korolev et al. 2018 and references therein) and eclogitic inclusions in diamonds (Deines et al. 1991) that overlap with the ‘normal’ modern mantle range.

Conversely, in the context of a mantle with higher heatflow the associated oceanic crust is expected to be thicker and more buoyant (Bickle 1986; Smithies et al. 2003). The equivalent of modern high-temperature hydrothermal systems found at axial mid-ocean ridges, might, therefore, be restricted to shallower depths (Kastings et al. 2006). In such a system with less effective fluid circulation, low-temperature oxygen isotope exchange is expected to dominate, generating lower seawater δ18O values (Kastings et al. 2006; Walker and Lohmann 1989). At the same time, enhanced continental weathering due to higher pCO2 levels and the presence of extensive greenstone belts (areas of high low-temperature hydrothermal alteration De Wit et al. 1982) are believed to have had a significant contribution in the low-δ18O values of Archaean seawater (Jaffres et al. 2007; Walker and Lohmann 1989). This is supported by oxygen isotope variations recorded in (bio)chemical marine sedimentary rocks, such as chert and calcite, indicating seawater had lower δ18O values in the past, estimated at ~ − 10‰ in the Archaean, compared to present values (~ 0.00‰ SMOW) (Burdett et al. 1990; Hren et al. 2009; Jaffres et al. 2007; Veizer and Prokoph 2015; Walker and Lohmann 1989). Although Archaean seawater temperature may have been higher, leading to lower δ18O values of altered oceanic crust (Jaffres et al. 2007 and references therein) it cannot be the only explanation for the δ18O values recorded in some of our samples or in some Archaean sedimentary rocks (~ 8‰ lower than modern carbonates, Kastings et al. 2006; Knauth and Lowe 1978, 2003; Veizer et al. 1999). In conclusion, the low-δ18O values measured in the Kaapvaal eclogites from this study are most likely the result of the interaction of an oceanic crustal protolith with Archaean seawater that had lower δ18O value than present as suggested by Kastings et al. (2006) and references therein.

Conclusions

  1. 1.

    The correlation between garnet and clinopyroxene δ18O values (r2 = 0.98) and the constant Δcpx−gt (− 0.22 to + 0.31‰, n = 35), Δgt−ky (− 0.37 to − 0.30‰, n = 3) and Δgt−cor (+ 0.05 to + 0.31‰, n = 4) among the minerals present in each of our samples are evidence for oxygen isotope equilibrium at mantle temperatures. We, therefore, conclude that the whole rock δ18O values estimated in this work for eclogite xenoliths, were most likely inherited from their protolith.

  2. 2.

    Bimineralic eclogites from the Kaapvaal craton, showing little or no signs of kimberlite metasomatism, can be defined as high-magnesium (Type IIA) and low-magnesium (Type IIB) eclogites. Based on reconstructed whole-rock major and trace element compositions and oxygen isotope ratios, both compositional types are interpreted to be derived from an Archaean oceanic crust formed from an evolved picritic liquid. The wide range in δ18O values (1.1–7.6‰) present in samples with a narrow major and trace element compositional range is interpreted as being caused by hydrothermal alteration of the ocean crust at variable temperatures. Type IIB eclogites were sampled from greater depths (~ 160–240 km) than Type IIA (~ 70–160 km), and suffered a small degree of melt loss before or during eclogitization. It is unclear whether both types are derived from the same oceanic crust, or from ocean crust recycled at different times.

  3. 3.

    Our non-metasomatized eclogites predominantly have estimated bulk δ18O values that are lower than asthenospheric mantle and one corundum-bearing eclogite from Roberts Victor has bulk δ18O values of 1.4‰ which is the lowest so far measured in mantle eclogites. A mantle metasomatic agent with extremely low-δ18O values overprinting the lower part of the cratonic root before the kimberlite eruption, without re-equilibrating with the surrounding peridotitic mantle is considered unlikely. It is more likely that oxygen isotope ratios in non-metasomatized eclogites are inherited from the ancient protolith.

  4. 4.

    Kyanite-bearing eclogites have positive Sr and Eu anomalies and, based on reconstructed whole-rock compositions, are interpreted to be derived from a gabbroic protolith, whereas corundum-bearing eclogites were derived from a pyroxene-dominated crustal cumulate.

  5. 5.

    Assuming our data represent a profile through oceanic crust that has its lowest δ18O values ~ 2‰ lower than profiles reconstructed from Cretaceous ophiolites, this is most probably the result of interaction with Archaean seawater having lower δ18O values, perhaps combined with a higher thermal regime for seawater–rock interaction.