Tungsten-182 evidence for an ancient kimberlite source

Significance

Kimberlites are igneous rocks derived from deep mantle sources. Recent studies have suggested that certain kimberlites originated from a mantle source with relatively primitive chemical composition that was created by early Earth processes. We present W isotope data for a global suite of kimberlites with variable formation ages and find their mantle source(s) to be characterized by ¹⁸²W/¹⁸⁴W averaging ∼6 ppm lower than the upper mantle ratio. This result is consistent with derivation of some kimberlites from one or more early formed mantle reservoirs. The low ¹⁸²W/¹⁸⁴W of these kimberlites is indicative of an ancient mantle source modified by some form of core–mantle interaction, an early silicate fractionation event, an overabundance of late-accreted materials, or a combination of these.

Abstract

Globally distributed kimberlites with broadly chondritic initial ¹⁴³Nd-¹⁷⁶Hf isotopic systematics may be derived from a chemically homogenous, relatively primitive mantle source that remained isolated from the convecting mantle for much of the Earth’s history. To assess whether this putative reservoir may have preserved remnants of an early Earth process, we report ¹⁸²W/¹⁸⁴W and ¹⁴²Nd/¹⁴⁴Nd data for “primitive” kimberlites from 10 localities worldwide, ranging in age from 1,153 to 89 Ma. Most are characterized by homogeneous μ¹⁸²W and μ¹⁴²Nd values averaging −5.9 ± 3.6 ppm (2SD, n = 13) and +2.7 ± 2.9 ppm (2SD, n = 6), respectively. The remarkably uniform yet modestly negative μ¹⁸²W values, coupled with chondritic to slightly suprachondritic initial ¹⁴³Nd/¹⁴⁴Nd and ¹⁷⁶Hf/¹⁷⁷Hf ratios over a span of nearly 1,000 Mya, provides permissive evidence that these kimberlites were derived from one or more long-lived, early formed mantle reservoirs. Possible causes for negative μ¹⁸²W values among these kimberlites include the transfer of W with low μ¹⁸²W from the core to the mantle source reservoir(s), creation of the source reservoir(s) as a result of early silicate fractionation, or an overabundance of late-accreted materials in the source reservoir(s). By contrast, two younger kimberlites emplaced at 72 and 52 Ma and characterized by distinctly subchondritic initial ¹⁷⁶Hf/¹⁷⁷Hf and ¹⁴³Nd/¹⁴⁴Nd have μ¹⁸²W values consistent with the modern upper mantle. These isotopic compositions may reflect contamination of the ancient kimberlite source by recycled crustal components with μ¹⁸²W ≥ 0.

Kimberlites are igneous rocks generated by melting of source rocks in the upper mantle at depths of >150 km (1) and may derive from much greater depths, as indicated by the occasional occurrence of diamond xenocrysts indicative of high-pressure origins in the transition zone or lower mantle (2–4). Kimberlites are notable for their enrichments in incompatible trace elements and volatile species, especially H₂O and CO₂, and have peculiar high–field strength element fractionations that have also been attributed to deep mantle sources (5). They are found on every continent (6), and formation ages range from ∼2,850 Ma to recent (e.g., ref. 7).

Two recent studies examined the long-lived ¹⁴⁷Sm-¹⁴³Nd and ¹⁷⁶Lu-¹⁷⁶Hf isotopic systems in a global suite of fresh kimberlites, with formation ages ranging from ∼2,050 Ma to present day (8, 9). Most of the kimberlites that are older than ∼200 Ma are characterized by initial ¹⁴³Nd/¹⁴⁴Nd slightly above the range of chondritic evolution and corresponding initial ¹⁷⁶Hf/¹⁷⁷Hf ratios within the range of chondritic evolution (e.g., ref. 10) (SI Appendix, Fig. S1). The collective ¹⁴³Nd-¹⁷⁶Hf isotopic growth trajectories of these samples, which Woodhead et al. (8) termed “primitive” kimberlites, are much more restricted than and well resolved from those of midocean ridge basalts (MORB), which provide representative samples of the convective upper mantle. The Nd and Hf isotopic growth trajectories of the primitive kimberlites require derivation from one or more isolated, chemically homogeneous sources that maintained relatively constant and chondritic to slightly suprachondritic Sm/Nd and Lu/Hf ratios for at least the ∼2,000-Ma range of sample ages that define the trend and likely extending back to the earliest stages of Earth history (8, 9).

In contrast to the primitive kimberlites, the initial ¹⁴³Nd/¹⁴⁴Nd and ¹⁷⁶Hf/¹⁷⁷Hf ratios of many kimberlites younger than 200 Ma are distinctly subchondritic (8). These kimberlites were termed “anomalous” by Woodhead et al. (8) The authors suggested that the subchondritic compositions reflected variable contributions of ancient recycled components to the mantle source reservoir of the primitive kimberlites beginning at about that time, signaling a partial perturbation of the isolated source reservoir.

As samples of kimberlites extend back only to ∼2,850 Ma, the existing data for long-lived isotopic systems are not very sensitive to when the distinct compositional characteristics of the source reservoir of the primitive kimberlites were established. Consequently, to further investigate the nature of the kimberlite source(s), data are reported here for the short-lived ¹⁸²Hf-¹⁸²W and ¹⁴⁶Sm-¹⁴²Nd isotopic systems. Short-lived radiogenic isotope systems are ideal for investigating formational differentiation events that occurred within the first ∼600 Ma of Earth history. The ¹⁸²Hf-¹⁸²W system (t_1/2 = 9 Ma) system is comprised of moderately siderophile (W) and lithophile (Hf) elements. Both elements are strongly incompatible in silicate systems. Hence, the system can record evidence of both metal–silicate interactions and silicate–silicate fractionation processes that occurred within the first ∼70 Ma of Solar System history while ¹⁸²Hf was extant (e.g., ref. 11). The complementary ¹⁴⁶Sm-¹⁴²Nd system (t_1/2 = 103 Ma), consisting of two strongly incompatible lithophile trace elements, is useful for identifying silicate crystal–liquid fractionation processes that occurred within the first ∼600 Ma of Solar System history (e.g., refs. 12 and 13).

Despite the very short-lived nature of the Hf-W system, the modern Earth’s interior does not have a homogeneous μ¹⁸²W value (in which μ values are the part-per-million difference in ¹⁸²W/¹⁸⁴W and ¹⁴²Nd/¹⁴⁴Nd between sample and laboratory reference materials). Assuming that the μ¹⁸²W value of the bulk silicate Earth (BSE) is 0, comparable to laboratory reference materials, and that the bulk Earth has a chondritic μ¹⁸²W value of ∼ −190, mass balance dictates that the core has a μ¹⁸²W value of ∼ −220 (14). The difference in values between the presumed BSE and core indicates that core segregation occurred largely while ¹⁸²Hf was extant (e.g., refs. 15 and 16). Further, certain ancient and modern rocks are characterized by well-resolved differences in μ¹⁸²W values compared to the presumed BSE value (e.g., refs. 17 and 18). By contrast, while both positive and negative μ¹⁴²Nd values are common among Archean rocks, post-Archean rocks are characterized by minimal isotopic heterogeneity, presumably as a result of efficient mantle mixing (e.g., refs. 19–21).

If kimberlites tap a mantle source that formed early in Earth history and was isolated from interactions with the incompatible element–depleted convecting upper mantle sampled by MORB and with chemically evolved subducted crustal materials, the kimberlite source may have μ¹⁸²W and μ¹⁴²Nd values that differ from materials derived from the upper mantle and crust. Toward this end, we report μ¹⁸²W and μ¹⁴²Nd values for eleven and six, respectively, of the whole-rock samples examined for ¹⁴³Nd-¹⁷⁶Hf by prior studies (8, 22), using previously well-documented methods (Materials and Methods). The samples examined for ¹⁸²W/¹⁸⁴W include nine “primitive” and two “anomalous” kimberlites, as defined by ref. 8. All samples examined for ¹⁴²Nd/¹⁴⁴Nd are classified as primitive. The studied kimberlites are from three regions (Siberia, southern Africa, and western Canada), with eruption ages ranging from ∼1,153 to 52 Ma (Table 1). In addition to the whole-rock kimberlites, an ilmenite megacryst (PREM-1) from the South African Premier kimberlite pipe was examined. As megacrysts in kimberlites crystallize from precursor kimberlite melts at mantle depths (e.g., refs. 22 and 23), the ilmenite is examined here to gauge the effects of possible crustal contamination on the isotopic composition of the Premier whole rocks. Two whole-rock samples of the ca. 1,715 Ma Harney Peak Granite, South Dakota, were also analyzed for ¹⁸²W/¹⁸⁴W as part of the same measurement campaign. They serve as a reference to the isotopic composition of average continental crust during the Proterozoic and supplement data for other crustal materials of varying age.

Table 1.

Tungsten and Nd isotopic compositions of the studied kimberlites

Sample	Location	Age (Ma)	μ182W (N 6/3)*	2SE	μ183W (N 6/4)†	2SE	W (ppb)	μ142Nd	2SE
Primitive kimberlites
Aikhal A4	Siberia	376a	−7.7	4.1	−1.4	4.8	3,638	+5.4	1.9
INT-I2 (Internationalnaya)	Siberia	377b	−7.8	3.3	−2.9	3.8	2,829	+2.5	2.5
Mir-M3	Siberia	361c	−6.8	3.3	−4.8	3.7	870	+2.6	2.2
Mir-M3 (dup.)			−5.5	2.9	−0.8	3.6
Mir-M3 (average)			−6.2	3.1
UD-2 (Udachnaya)	Siberia	367d,e	−7.8	3.6	−2.4	4.1	1,999	+2.8	2.6
LQ-7 (Liqhobong)	Lesotho	91f	−6.6	3.0	−2.6	3.4	987
LQ-7 (dup.)			−1.7	4.3	−0.6	4.4
LQ-7 (average)			−4.2	3.7
Monastery	South Africa	89g	−5.7	2.9	−1.4	3.4	550
Premier-C	South Africa	1,153h	−5.3	3.6	−4.0	4.3	3,320	+1.8	2.2
Premier-C (dup.)			−7.7	3.1	−4.4	3.5
Premier-C (average)			−6.5	3.3
Premier-P	South Africa	1,153h	−6.2	4.5	−4.6	5.4	2,438	+1.3	1.9
Premier-P (dup.)			−4.1	3.4	−3.5	4.1
Premier-P (average)			−5.2	4.0
DRY-1 (Drybones)	Western Canada	441i	−4.2	3.4	−3.0	3.9	743
Primitive Average (n = 13, 2SD)			−5.9	3.6				+2.7	2.9
PREM-1 (ilmenite megacryst)			−5.4	3.9	−2.9	4.5	432
Anomalous kimberlites
HL-12–1 (Hardy Lake)	Western Canada	72j	+3.4	3.1	+1.4	3.6	319
GRZ-2 (Grizzly)	Western Canada	52j	−4.1	3.8	−3.4	4.2	681
GRZ-2 (dup.)			−0.6	5.1	+3.3	5.6
GRZ-2 (average)			−2.4	4.5
Harney Peak granites
81BH2-1	United States	1,715k	+2.3	5.1	−1.2	6.3	195
81BH2-1 (dup.)			+4.7	3.3	−3.6	4.0
81BH2-1 (average)			+3.5	4.3
81BH3-1B	United States	1,715k	0.0	3.4	−4.2	4.1	110
81BH3-1B (dup.)			−0.6	2.9	−2.9	3.2
81BH3-1B (average)			−0.3	3.2

μ¹⁸²W, μ¹⁸³W, and μ¹⁴²Nd are the deviations of ¹⁸²W/¹⁸⁴W, ¹⁸³W/¹⁸⁴W, and ¹⁴²Nd/¹⁴⁴Nd, respectively, from that of laboratory standards, in parts per million. dup.: duplicate analysis. All analyses labeled as duplicates represent separate sample dissolutions. Numbers in italics are the average of two analyses. Uncertainties are reported as either the 2SE run statistics of individual analyses or the average of duplicate analyses. Emplacement ages are from ^aref. 62, ^bref. 63, ^cref. 64, ^dref. 65, ^eref. 66, ^fref. 67, ^gref. 68, ^href. 69, ⁱref. 70, ^jref. 71, and ^kref. 72. Refer to SI Appendix for complete W (SI Appendix, Table S6) and Nd (SI Appendix, Table S7) isotopic data.

^*182W/¹⁸⁴W ratios normalized to ¹⁸⁶W/¹⁸³W.

^†183W/¹⁸⁴W ratios normalized to ¹⁸⁶W/¹⁸⁴W.

Prior studies of ¹⁸²W/¹⁸⁴W and ¹⁴²Nd/¹⁴⁴Nd in kimberlites have been limited. Scherstén et al. (14) reported ¹⁸²W/¹⁸⁴W data for three South African kimberlites. Although individual analyses were not resolved from the laboratory standard within measurement uncertainties, the average of the combined data yielded a μ¹⁸²W value of −7 ± 4 (2SD). Another study reported ¹⁸²W/¹⁸⁴W data for two ∼85-Ma kimberlites collected from Minas Gerais, Brazil, with an average μ¹⁸²W value of 11 ± 39 (2SD), and the analyses were all within uncertainties of the standard measurements (24). Most recently, Tappe et al. (25) reported ¹⁸²W/¹⁸⁴W data for 18 kimberlites from Africa, ranging in age from 1,835 Ma to modern, with an average μ¹⁸²W value of 0 ± 4 (2SD). That study found no resolvable differences in ¹⁸²W/¹⁸⁴W between kimberlites and standards (see comparison to data from other studies in SI Appendix) and concluded that the sources of these kimberlites did not sample ancient or core-equilibrated mantle domains.

The μ¹⁴²Nd values for a global suite of approximately 20 kimberlites were reported in Boyet and Carlson (13). That study reported no resolvable differences from the laboratory standard that, at that time, showed a 2SD external reproducibility of ±6 ppm.

Results

The abundances of W in the primitive kimberlites analyzed are generally high, ranging from 550 to 3,638 ppb, consistent with the incompatible element–enriched nature of kimberlites (Table 1). The two anomalous kimberlites have W concentrations of 319 and 681 ppb, which are at the lower end of concentrations recorded in the primitive kimberlites. No correlation between W concentration and kimberlite emplacement age is observed.

Thorium, U, and W are similarly strongly incompatible trace elements in silicate systems, so Th–U–W systematics are useful to evaluate secondary disturbance effects on W. For the primitive kimberlites, ratios of W/Th and W/U are broadly positively correlated (R² = 0.67) (SI Appendix, Fig. S2). Given that W and U can be highly mobile in aqueous fluids, whereas Th is not, this positive correlation is consistent with a magmatic origin for W. Further, the W/Th ratios of the primitive kimberlites range only from 0.05 to 0.19, all of which are within the “canonical range” of 0.04 to 0.30 for mantle-derived magmas (26, 27) (SI Appendix, Table S1 and Fig. S3). The canonical range defines the range of W/Th ratios that would likely be produced by igneous processes, assuming normal W and Th concentrations in the mantle source reservoir. The observation that all of the primitive kimberlites have concentration ratios within the canonical range implies that the W abundances and isotopic compositions of the primitive kimberlites were likely not strongly modified by fluid-related processes that commonly affect kimberlite magmas after their emplacement (6, 26). By contrast, the W/Th of two of the anomalous kimberlites from the Lac de Gras field (Slave Craton; western Canada), HL-12–1 and GRZ-2, are 0.02 and 0.03, respectively, falling below the canonical mantle range. This most likely suggests modest W loss from the rock or parental magma, leaving open the possibility of W isotopic exchange between the rock/melt and fluids.

During the course of the analytical campaign, the long-term external reproducibility for the University of Maryland Alfa Aesar laboratory W reference material was 3.3 ppm (n = 14) 2SD with a 2SE of 0.9 ppm (SI Appendix, Table S2). The primitive kimberlites are characterized by negative μ¹⁸²W values ranging from −1.7 to −7.8 (n = 13 including four duplicates), when corrected for mass spectrometric fractionation using ¹⁸⁶W/¹⁸³W (Table 1 and Fig. 1). The independently measured ¹⁸³W/¹⁸⁴W ratios of all samples analyzed are within ±5 ppm of the average ratio for the reference material and serve as an additional check on measurement quality. The μ¹⁸²W values of duplicate analyses of four primitive kimberlite samples differed by no more than 4.9 ppm and are in good agreement with the results for the reference material that the external reproducibility of individual analyses is ≤±3.3 ppm.

Fig. 1.

μ¹⁸²W values for primitive (closed symbols) and anomalous kimberlites (open symbols) from Siberia (circles), southern Africa (triangles), and Canada (squares). Light and dark gray areas represent the 2SD (±3.3 ppm) and 2SE (±0.9 ppm) long-term external reproducibility of the Alfa Aesar W laboratory reference standard, respectively. Sample error bars are 2SE of individual analyses. Light and dark blue areas represent the 2SD (±3.6 ppm) and 2SE (±1.0 ppm) of the mean of primitive kimberlites, respectively. Results for samples in which two separate analyses (including sample digestion) were performed are shown as the smaller symbols, with their averages and 2SE in the larger symbols. The black bar represents the variation in average μ¹⁸²W values of African kimberlites reported by Tappe et al. (25)

Collectively, the primitive kimberlites have an average μ¹⁸²W value of −5.9 ± 3.6 ppm (2SD). Seven out of the nine primitive kimberlite samples have μ¹⁸²W values and associated uncertainties that are resolved from the 2SE of the laboratory standard value (μ¹⁸²W = 0 ± 0.9), considered to be representative of the BSE. If the primitive kimberlites were derived from a single, isotopically homogeneous reservoir as suggested by refs. 8 and 9, all of the data for the nine primitive kimberlites can be combined to obtain a 2SE = ±1.0. These results indicate that the mantle source reservoir for most or all of the primitive kimberlites examined here was less radiogenic than the presumed BSE value of 0 but not as strongly depleted in ¹⁸²W as some modern ocean island basalts (OIB), which extend to negative μ¹⁸²W values of −22 (28). The ilmenite megacryst PREM-1 has a μ¹⁸²W value of −5.4 ± 3.9, which is identical within uncertainties of the associated whole-rock samples from Premier (−6.5 ± 3.3 and −5.2 ± 4.0).

The data differ from those reported by Tappe et al. (25), which found no anomalies among ¹⁸²W/¹⁸⁴W ratios in the southern African kimberlites they examined. The reason for the difference in the results of the two studies is not clear. The difference may result from limited overlap among sample locales. Even where sample locales are the same (Premier and Monastery), different samples may not yield the same isotopic compositions, as most kimberlites are composite bodies, with varying ages of the components. We also note that that some of the kimberlites examined by ref. 25 do not have W/Th ratios within or near the canonical range suggested by refs. 26 and 27. Of 25 kimberlite samples for which W/Th data are available in that study, seven samples do not have W/Th ratios within or near magmatic range between 0.04 and 0.30 (SI Appendix, Fig. S3). The samples with noncanonical W/Th ratios may have experienced fluid-related process, which can affect W isotopic composition by introducing extraneous W. Moreover, most samples examined by ref. 25 have subchondritic initial Hf isotope ratios and do not satisfy the primitive kimberlite criteria outlined by ref. 8.

The average μ¹⁸²W value of −2.4 for the “anomalous” Lac de Gras kimberlite GRZ-2 is consistent with the primitive kimberlites but also overlaps within uncertainty of the value obtained for the laboratory reference material. The μ¹⁸²W value of 3.4 for anomalous sample HL-12–1, also from Lac de Gras, is distinctly higher than any primitive kimberlite but also overlaps the laboratory reference material within uncertainty.

The two samples of Harney Peak Granite have average μ¹⁸²W values of −0.3 and +3.5 (n = 2 for both), in good agreement with estimates of the value presumed for the BSE, based on data for MORB (17, 18, 29), and post-Archean samples of continental crust (i.e., glacial diamictites) (30).

All six primitive kimberlite samples analyzed for μ¹⁴²Nd have values within the 2SD of the Shin Etsu Nd standard (JNdi) (±6.4 ppm, SI Appendix, Table S3). The average μ¹⁴²Nd value for the samples is +2.7 ± 2.9 (2SD; n = 6, Table 1 and Fig. 2), which is identical within uncertainties to the average kimberlite values reported by ref. 13 (average = 2.1 ± 8.4 [2SE], n = 19) and is not resolved from current estimates for the upper mantle (21).

Fig. 2.

Averaged μ¹⁴²Nd values for primitive kimberlites. Light and dark gray areas represent the 2SD (±6.4 ppm) and 2SE (±1.7 ppm) external reproducibility of the JNdi standard, respectively. Sample error bars are 2SE of individual analyses. Light and dark blue areas represent the 2SD (±2.9 ppm) and 2SE (±1.2 ppm) of the mean of primitive kimberlites, respectively. Symbols are the same as in Fig. 1.

Discussion

Anomalous Kimberlites.

In addition to plotting well below the Nd and Hf isotopic evolution trends of primitive kimberlites (SI Appendix, Fig. S1), the two anomalous kimberlites are also characterized by generally higher μ¹⁸²W values than the primitive kimberlites. This indicates either derivation from an isotopically distinct source or some form of contamination either in the mantle source or en route to the surface. The latter possibility is unlikely, given that the anomalous kimberlites are characterized by high Mg/Yb and Si/Al, implying theses samples have been minimally affected by crustal contamination (Materials and Methods). These observations provide permissive evidence for the interpretation that the anomalous kimberlites formed from the primitive kimberlite source, which was subsequently modified by the incorporation of a deeply subducted component with low ¹⁴³Nd/¹⁴⁴Nd and ¹⁷⁶Hf/¹⁷⁷Hf ratios, affecting the source during the past ∼200 Ma (8) (SI Appendix, Fig. S1).

Woodhead et al. (8) modeled potential contaminants as consisting of subducted slab assemblages (90% MORB and 10% terrigenous sediment) that formed at 3.0 to 3.5 Ga. Given the ancient nature of the contaminant required by the Hf and Nd isotopic data, these materials could plausibly have had either near-zero or positive μ¹⁸²W values, as rocks formed before ∼2.5 Ga generally are characterized by near-zero or positive μ¹⁸²W values (17, 26). In order to explore this general scenario further, we modeled mixing between possible contaminants and the putative primitive kimberlite source (refer to Discussion and modeling parameters in SI Appendix). The terrigenous sediment contaminant components at 3.5 and 3.0 Ga are assumed to have μ¹⁸²W values of either 12.8 (solid curves in Fig. 3), which is the average W isotopic composition of Eoarchean and Paleoarchean crustal rocks from ref. 26, or 0 (dashed curves). Assuming the anomalous kimberlites were derived from a primitive source that had been contaminated with similar recycled crustal materials, the μ¹⁸²W values of HL-12–1 and GRZ-2 are consistent with mixing if the terrigenous sediment component of the contaminant had μ¹⁸²W ≥ 0.

Fig. 3.

Plot of μ¹⁸²W versus initial εNd (part per 10,000 deviation in initial ¹⁴³Nd/¹⁴⁴Nd compared to chondritic radiogenic growth) for primitive kimberlites and anomalous kimberlites. Sample error bars of μ¹⁸²W are 2SE of individual analyses. Light and dark gray areas represent the 2SD (±3.3 ppm) and 2SE (±0.9 ppm) long-term external reproducibility of the University of Maryland Alfa Aesar W laboratory reference standard, respectively. Light and dark blue areas represent the 2SD (±3.6 ppm) and 2SE (±1.0 ppm) of the primitive kimberlite mean, respectively. Symbols are the same as in Fig. 1. Curves are shown for models of mixing between subducted slab assemblages (90% MORB and 10% terrigenous sediment) formed at 3.5 and 3.0 Ga and the composition of the mantle source reservoir for the primitive kimberlites. The μ¹⁸²W values of the terrigenous sediment component are assumed to be +12.8 (solid curves), consistent with the average W isotopic composition of Eoarchean and Paleoarchean rocks from ref. 26, or 0 (dashed curves). Modeling parameters are provided in SI Appendix, Table S4.

Primitive Kimberlites.

The primitive kimberlites examined here come from widely disparate geographic locations (Siberia, southern Africa, and western Canada) and with eruption ages ranging from 1,153 Ma to 89 Ma. The chondritic to slightly suprachondritic initial ¹⁴³Nd/¹⁴⁴Nd and ¹⁷⁶Hf/¹⁷⁷Hf ratios of the primitive kimberlite reservoir could have been created by the mixing of depleted mantle and enriched crustal materials. However, such a mixing scenario is unlikely to have generated a source with comparatively uniform isotopic evolution through at least 2,000 Mya, given likely changes in the compositions of subducted crustal materials and the physical conditions present in subduction zones through time (9).

The negative μ¹⁸²W values among the primitive kimberlites are also unlikely to have resulted from contamination of the parental magmas en route to the surface. Assimilation of lithospheric mantle material, which is ubiquitous in kimberlites (e.g., ref. 31), can be ruled out because the lithospheric mantle traversed and entrained by kimberlites is highly refractory (32) and, hence, highly depleted in W. This contrasts with the W-rich nature inferred for the kimberlitic parental melts. Further, no evidence currently indicates that lithospheric mantle is characterized by negative μ¹⁸²W values. A prominent role for crustal contamination in the primitive kimberlites can also be discounted, based on two observations. First, all of the primitive kimberlite samples have been filtered for substantial effects of crustal contamination (Materials and Methods). Second, except for some modern, volumetrically limited OIB, most crustal rocks examined to date, including those from the Slave Craton (33), the host to the Drybones Bay kimberlite, are characterized by near-zero or positive μ¹⁸²W values. Further, in the case of the Premier kimberlite, the isotopic composition of the ilmenite megacryst, recording the likely mantle isotopic signature, is identical within uncertainties to whole-rock samples. We conclude that the ¹⁸²W isotopic compositions of the whole-rock kimberlites most likely reflect the compositions of their mantle source(s).

Consistent with the interpretation espoused by refs. 8 and 9, one possible interpretation of the ¹⁸²W/¹⁸⁴W data is that the primitive kimberlites were derived from a single, deep, and isolated mantle reservoir collectively characterized by a uniform μ¹⁸²W value of −5.9 ± 1.0 and μ¹⁴²Nd = +2.7 ± 1.2, in which uncertainties cited are 2SE. The μ¹⁸²W value, but not the μ¹⁴²Nd value, is well resolved from current estimates of the BSE, indicating derivation from an ancient reservoir that remained isolated from the convecting upper mantle and subducted recycled materials for a period of at least 1,000 Mya. Alternatively, if the interpretation of a common, isotopically uniform mantle source for primitive kimberlites is either incorrect or if ¹⁸²W isotopic compositions were somehow decoupled from ¹⁴³Nd and ¹⁷⁶Hf, the ¹⁸²W results, combined with the results of ref. 25, could be interpreted as a continuum of μ¹⁸²W values ranging from ∼ −8 to 0. Regardless of the preferred interpretation, the cause of the negative anomalies in some kimberlites must be explained.

Three general categories of primary processes have been previously proposed to explain negative μ¹⁸²W values inferred for mantle reservoirs: 1) interactions between the core and the mantle (e.g., refs. 28 and 34), 2) silicate crystal–liquid fractionation while ¹⁸²Hf was extant (35), and 3) derivation from mantle source reservoirs containing excesses of late-accreted materials (36). All three processes are potentially capable of resulting in the ¹⁸²W/¹⁸⁴W compositions observed in the primitive kimberlites and are not mutually exclusive. Only the second of these processes would have some consequences for ¹⁴²Nd/¹⁴⁴Nd, because neither the core nor late-accreted materials have enough Nd of sufficiently distinct isotopic composition to cause measurable shifts in μ¹⁴²Nd. Conceptual cartoon descriptions of these three general categories of processes to produce negative ¹⁸²W anomalies are provided in Fig. 4 A–C.

Fig. 4.

Models for generating the ¹⁸²W composition of the primitive kimberlite mantle source reservoir. (A) Interaction between the mantle source reservoir and the core, or core-derived materials, (B) early silicate–liquid fractionation while ¹⁸²Hf was extant, and (C) overabundance of late accreted materials.

Modern OIB are characterized by μ¹⁸²W values that range from ∼0 to −24 (18, 28, 34, 37). The negative values in OIB have mainly been attributed to possible interactions between the mantle and core, coupled with complex mixing of core-modified mantle with other mantle components. Such interactions are plausible since the modern mantle is characterized by a W concentration of <10 ppb (38, 39), while the core has an elevated W concentration of ∼500 ppb, and a strongly negative μ¹⁸²W of ∼ −220 (14, 40). Given that some OIB have isotopic compositions that overlap with those of the primitive kimberlites, core–mantle interactions could potentially have delivered W characterized by negative μ¹⁸²W values to the mantle source(s) of the kimberlites in a similar manner to that envisioned for OIB. Yet compared to the primitive kimberlites, the anomalies in OIB are much more variable both globally and even within individual OIB systems (28). Application of the core–mantle mixing model developed for OIB to the primitive kimberlites is then problematic because it requires retention of relatively constant mixing ratios between one or more mantle reservoirs bearing the core isotopic signature and other mantle reservoirs over at least ∼1,000 Mya. While this process seems implausible, it cannot be ruled out at this time.

Early silicate–liquid fractionation for example due to crystallization of a magma ocean could also have led to the creation of one or more mantle reservoirs characterized by negative μ¹⁸²W values. Tungsten and Nd are more highly incompatible than Hf and Sm, respectively, under conditions of both high- and low-pressure melting/crystallization of mantle assemblages (41, 42). Silicate–liquid fractionation of Hf/W and Sm/Nd within the first ∼70 Ma of Solar System history, therefore, would have led to correlated ¹⁸²W and ¹⁴²Nd anomalies. Given the modest size of the average ¹⁸²W anomaly for primitive kimberlites, it is important to assess the magnitude of a corresponding ¹⁴²Nd anomaly that would result from the same process, as well as the corresponding isotopic evolution of ¹⁴³Nd/¹⁴⁴Nd and ¹⁷⁶Hf/¹⁷⁷Hf. Toward this end, isotopic evolution models were calculated for both high- and low-pressure conditions of magma ocean crystallization, assuming fractionation events at 32 and 65 Ma following Solar System formation, within an initially uniform mantle reservoir with Hf/W, Sm/Nd, and Lu/Hf ratios equivalent to estimates for BSE (SI Appendix). For example, if high-pressure fractional crystallization of a melted mantle domain occurred at 32 Ma following Solar System formation, a fractionated melt could evolve to the μ¹⁸²W average value of −5.9 ppm after <2% fractional crystallization. Such fractionation, in turn, would result in an eventual negative offset of <2 ppm of ¹⁴²Nd/¹⁴⁴Nd from BSE by the time ¹⁴⁶Sm was no longer extant. This magnitude of offset is not resolvable with the μ¹⁴²Nd data for the primitive kimberlites. This scenario also results in a negligible modification to the long-term evolution of ¹⁴³Nd/¹⁴⁴Nd and ¹⁷⁶Hf/¹⁷⁷Hf, though it would not lead to the slightly suprachondritic ¹⁴³Nd/¹⁴⁴Nd values observed in most of the primitive kimberlites.

If the fractionation event occurred during the later stages of the lifetime of ¹⁸²Hf (e.g., 65 Ma following Solar System formation), a change in the Hf/W of ∼40% is required to generate the observed ¹⁸²W anomaly. The amount of fractional crystallization required to achieve this change under high- and low-pressure conditions ranges from ∼30% for high-pressure to >99% for low-pressure crystallization. Such changes would be accompanied by relatively large changes to Sm/Nd and Lu/Hf and consequent negative offsets of ∼15 and 34 ppm, respectively, for μ¹⁴²Nd, as well as distinctly nonchondritic evolution toward lower ¹⁴³Nd/¹⁴⁴Nd and ¹⁷⁶Hf/¹⁷⁶Hf thereafter. Therefore, if the anomalous μ¹⁸²W isotopic composition of a kimberlite source was the result of an early silicate fractionation process, the process most likely occurred within the first 30 to 35 Ma of Solar System history.

This explanation for the primitive kimberlite data might be problematic in that it may be difficult to reconcile with the concept of late accretion. Late accretion is a commonly proposed process during which ∼0.5 to 2% of Earth’s mass was added to the mantle in the form of siderophile element–rich planetesimals with chondritic bulk compositions, subsequent to the end of core formation and likely the Moon-forming giant impact (e.g., refs. 43 and 44). The process has typically been invoked to provide the mantle with most of its inventory of highly siderophile elements (HSE), which occur in generally chondritic relative abundances in the BSE, and ∼10% of its inventory of moderately siderophile elements, including W (e.g., ref. 41). If the silicate fractionation event occurred prior to the dominant phase of late accretion, the late-accreted material would need to eventually be uniformly mixed into the fractionated mantle domain(s) in the same proportion as in the accessible upper mantle so as to retain the negative offset in μ¹⁸²W from the BSE. This in turn requires a mixing process that could introduce the late-accreted materials to both the kimberlite source reservoir(s) and the ambient mantle without mixing away the unique kimberlite source characteristics. Despite this caveat, this type of process also remains a viable means to explain the kimberlite data.

Late accretion to Earth may have been dominated by the addition of a limited number of Pluto mass bodies (45). This would have resulted in an initially nonuniform or “grainy” distribution of siderophile elements within the mantle (46). The W-rich (∼200 ppb) and low-μ¹⁸²W value (∼−190) nature of bulk chondritic materials means that grainy late accretion to the mantle provides another mechanism for the creation of one or more mantle reservoirs characterized by negative μ¹⁸²W values (e.g., ref. 35). For example, if the HSE abundances and μ¹⁸²W value of 0 of the bulk mantle resulted from the addition of 0.65 wt % of late-accreted materials with bulk chondritic compositions, grainy addition of 0.9 wt % to a discrete, isolated portion of the mantle would result in a μ¹⁸²W value of ∼−6 (mixing parameters are provided in SI Appendix, Table S5). An excess of late-accreted materials is appealing to account for the source of some kimberlites because the process would have had negligible impact on lithophile element abundances and, hence, the subsequent evolution of the Sm-Nd and Lu-Hf isotopic systems.

One predicted collateral effect of excess late-accreted materials in a mantle domain is a ∼40% excess in HSE compared to estimates for the BSE (SI Appendix, Table S5). Although the limited HSE data for kimberlites suggest they are derived from a mantle source with broadly similar HSE abundances to the BSE (e.g., refs. 47 and 48), estimating the HSE abundances in the mantle source of kimberlites is difficult both because of the uncertain distribution coefficients of the HSE during kimberlite melting and because of the chance of contamination by HSE-rich materials in the lithospheric mantle (5, 48). Hence, this aspect is currently not possible to test. This model must also be considered viable although not unique.

If there was a common mantle source reservoir for primitive kimberlites as suggested by refs. 8 and 9, all three processes discussed to create and preserve the anomaly require that the source reservoir be globally distributed and long-lived. This is most strongly suggestive of a lower mantle reservoir. Some prior studies have suggested a link between the mantle source(s) of certain kimberlites and seismically observed deep mantle structures such as Large Low Shear Velocity Provinces (LLSVP; e.g., refs. 9, 25, and 49). One of two known LLSVPs currently underlies Africa and is a potential source for young kimberlites from southern Africa (9, 25). Although previous studies have shown that LLSVPs could have remained stable for the past 250 to 300 Mya, the longevity of LLSVPs and whether their positions migrate in time relative to surface features remains poorly constrained (e.g., refs. 50 and 51). Consequently, it is not possible to assess their potential as source reservoirs to kimberlites >300 Ma old.

Summary

Kimberlites previously identified to originate from a “primitive” mantle source reservoir are characterized by average μ¹⁸²W and μ¹⁴²Nd values of −5.9 ± 3.6 ppm (2SD, n = 13) and +2.7 ± 2.9 (2SD, n = 6), respectively. These results do not require but are permissive of a single worldwide reservoir, as suggested by prior studies. Possible explanations for the modestly negative μ¹⁸²W values include the ancient transfer of W from the core to the mantle source reservoir, creation of the source reservoir as a result of early silicate fractionation, an overabundance of late-accreted materials in the source reservoir, or a combination of these. Regardless of the details of its formation, the data support the conclusion that at least some kimberlites are derived from one or more deep mantle reservoirs that developed early in Earth history and remained isolated from the convective vigor of the accessible upper mantle throughout most of Earth history. The isotopic distinctiveness and required long-term isolation characteristics of this type of mantle reservoir suggest a location deep in the lower mantle where viscosity is higher than in the mantle transition zone and asthenosphere, thus preventing extensive physical exchange with other mantle components.

Materials and Methods

Major and Trace Element Concentration Measurement.

Major element analyses of Monastery and Premier-P samples were performed at the Geological Survey of Canada (Ottawa) using a Phillips Pananalytical PW1400 instrument on fused discs using wavelength dispersive methods. Ni and Cr concentration data obtained by X-ray fluorescence (XRF) pressed powder pellet were used in preference to inductively coupled mass spectrometry (ICP-MS) data. Calibration was performed using natural rock reference materials using the “accepted values” of ref. 52. Quality control was ensured using repeat analyses of United States Geological Survey reference material PCC-1 and an in-house kimberlite “standard,” K2WI.

Major element analyses of Aikhal A4 and Premier-C samples were performed at Franklin and Marshall College (Pennsylvania) using a Phillips Panalytical PW 2404 XRF. The concentrations were determined by comparison with geochemical rock standards, data for which are reviewed in refs. 52 and 53.

Thorium, U, and Yb data, along with other trace elements, were determined using a PerkinElmer Sciex Elan 6000 ICP-MS at Durham University. Chemical procedures, instrumental conditions, and repeatability for in-house kimberlite standard K2WI are described in ref. 54. In brief, ∼100 mg of sample were digested in concentrated HNO₃ and HF at ∼150 °C for 2 d, followed by repeat dry-downs in concentrated HNO₃. Samples were dissolved in HNO₃ and analyzed with ICP-MS. The kimberlite samples are enriched in incompatible trace elements and exhibit compositions typical of archetypal kimberlites worldwide (SI Appendix, Fig. S4).

¹⁴³Nd/¹⁴⁴Nd and ¹⁷⁶Hf/¹⁷⁷Hf Isotopic Composition Measurement.

Between 70 and 90 mg samples (Aikhal A4, Premier-C, and PREM-1) were dissolved at high pressure in “Parr style” vessels overnight at 150 °C and then refluxed with nitric acid to remove fluorides (all three samples yielded clear solutions with no residue visible after centrifugation). After removing a small aliquot for trace element analysis by quadrupole ICP-MS (Q-ICP-MS), Hf and Nd were extracted using Eichrom TRU and LN resins (for Nd) and LN resin (Hf). Isotopic compositions were measured on a Nu Plasma multicollector ICP-MS. Instrumental mass bias was corrected by internal normalization to ¹⁴⁶Nd/¹⁴⁴Nd = 0.7219 and ¹⁷⁹Hf/¹⁷⁷Hf = 0.7325 using the exponential law, and ¹⁴³Nd/¹⁴⁴Nd and ¹⁷⁶Hf/¹⁷⁷Hf are reported relative to La Jolla Nd = 0.511860 and JMC475 = 0.282160, respectively. External precisions (2 std dev) are ±0.000030 and ±0.000015, based on results for rock standards. Sm/Nd and Lu/Hf ratios for age corrections were calculated from high-precision trace element data obtained by Q-ICP-MS for a split of each sample solution; external precisions on ¹⁴⁷Sm/¹⁴⁴Nd and ¹⁷⁶Lu/¹⁷⁷Hf are ±2%, and duplicate analyses on a small dataset by isotope dilution methods suggest accuracy better than 2% for Sm-Nd and 3% for Lu-Hf. Monte Carlo simulations suggest that these uncertainties are a very minor component of the overall uncertainty budget in age-corrected epsilon values which are dominated by the measurement uncertainty on the isotope ratio.

Tungsten Concentration Measurement.

Tungsten concentrations were determined by isotope dilution using a ¹⁸²W spike. Approximately 100 mg of sample were digested and equilibrated with a ¹⁸²W-enriched spike in 1 mL of concentrated HNO₃ and 5 mL of concentrated HF in a capped perfluoroalkoxy (PFA) vial heated at ∼150 °C for 3 d. After digestion, samples were dried down, and 2 mL of 6 M HCl were added to convert the sample to chloride form, then dried down. The dried residues were dissolved in 0.5 M HCl–0.5 M HF, and W was separated using an anion-exchange column chemistry similar to that discussed in ref. 55. Concentrations were measured using a Neptune Plus multicollector ICP-MS at the University of Maryland (UMd).

Tungsten Isotopic Composition Measurement.

Between 0.3 and 3 g of sample powder was processed to obtain ∼1,000 ng of W. Samples were digested in PFA vials with between 12 and 36 mL of a concentrated 1:5 mixture of HNO₃ and HF for at least 5 d at 150 °C. After sample dry-down, the samples were treated twice with concentrated HNO₃ and several drops of H₂O₂ and dried down to remove organics. The dried residues were then redissolved in 8 M HCl-7 M HF and dried down again. After redissolution in 0.4 M HCl–0.5 M HF, W was separated using an anion-exchange column chemistry as described in ref. 56. The first or second stage column separations were repeated to reduce Ti/W abundance ratios to <1. The total W recovery including sample digestion was between 74 and 88% for all kimberlite bulk samples.

Tungsten isotopic compositions were measured by thermal ionization mass spectrometry in negative ionization mode (N-TIMS) using a Thermo Fisher Triton at the UMd. Purified W was loaded onto a Re ribbon filament and dried. After the sample was dried, it was covered with an activator solution containing ∼5 μg of La and Gd each and dried. The measurement method of N-TIMS analysis was reported in ref. 57. ¹⁸⁶W¹⁶O₂¹⁸O and ¹⁸⁷Re¹⁶O₂¹⁸O were measured with every line of analysis allowing per-integration oxide interference corrections. Mass fractionation was corrected using ¹⁸⁶W/¹⁸³W = 0.92767 or ¹⁸⁶W/¹⁸⁴W = 1.98594 (58). The in-house Alfa Aesar laboratory W reference material was measured repeatedly. The long-term uncertainties in μ¹⁸²W of our laboratory standard measurements were ± 3.3 (2SD, SI Appendix, Table S2). All measured ¹⁸³W/¹⁸⁴W ratios were identical within uncertainties to the average standard data (SI Appendix, Table S2).

¹⁴²Nd/¹⁴⁴Nd Isotopic Composition Measurement.

Approximately 50 mg of sample powder was dissolved in capped Savillex beakers placed on a hot plate overnight. Sample treatment after initial dissolution and Nd separation procedures follow the method described in ref. 21. Total Nd analytical blanks for this procedure were 100 pg and are negligible for the sample sizes used. Following purification, the amount of Nd separated was determined using a small aliquot of the sample measured by signal height comparison in a Thermo Fisher iCAP Q Q-ICP-MS at Carnegie’s Earth and Planets Laboratory (EPL). Approximately 700 ng of Nd was loaded onto outgassed Re filaments in 3M HCl and, after drying, covered in 1 μL of 0.1 M H₃PO₄. The sample solution was confined to the middle of the filament by depositing two wax ridges melted from the edge of sheet of Parafilm©. After drying, the filament was heated to a dull red to burn off the wax and oxidize the Nd on the filament. Nd isotopic compositions were measured on a Thermo Fisher Triton XT mass spectrometer at EPL. The measurement procedures followed those described in ref. 59 using double filaments to produce Nd⁺ ions. This four-step dynamic routine allowed calculation of all isotope ratios, except for ¹⁵⁰Nd, using ion intensities from several mass steps in order to cancel out faraday cup inefficiencies. Successful runs include 485 to 900 cycles that provide 970 to 1,800 dynamic ratios for ¹⁴²Nd/¹⁴⁴Nd calculated from 8-s integrations at signal sizes of ¹⁴⁴Nd between 2.8 to 3.8 × 10⁻¹¹ amps. Data were fractionation corrected to ¹⁴⁶Nd/¹⁴⁴Nd = 0.7219 assuming exponential mass dependency of the fractionation. The data also were corrected for temporal variability in fractionation by linear interpolation between two consecutive measurements of ¹⁴⁶Nd/¹⁴⁴Nd. Equations for this temporal correction are given in ref. 21. Potential interferences from Ce and Sm are monitored at ¹⁴⁰Ce and ¹⁴⁷Sm. Measured ¹⁴⁰Ce/¹⁴⁴Nd ratios in the samples ranged from 2 to 53 ppm requiring corrections to ¹⁴²Nd/¹⁴⁴Nd of 0.3 to 6.6 ppm. Measured ¹⁴⁷Sm/¹⁴⁴Nd was below 7 ppm for all samples. Data for all Nd isotope ratios and interferences for both standards and samples are reported in SI Appendix, Tables S3 and S7.

Evaluation of Degree of Crustal Contamination.

Kimberlite melts are likely to be influenced by crustal contamination during ascent. Crustal contamination is commonly reflected in elevated Al₂O₃, Pb, and heavy rare earth element contents such as Yb (60, 61). The degree of crustal contamination can be evaluated using ln(Mg/Yb) versus ln(Si/Al) (SI Appendix, Table S1 and Fig. S5A). Samples with low ln(Si/Al) and ln(Mg/Yb) are most likely to have been affected by crustal contamination. Although termed “primitive,” samples LQ-7, Monastery, DRY-1, Premier-C, and Premier-P are characterized by comparatively low ln(Si/Al) and ln(Mg/Yb) ratios among the primitive kimberlites and could have been affected by minor crustal contamination. Despite this, the μ¹⁸²W values of these samples are in good agreement with the values for the other primitive kimberlites, within the analytical uncertainties, and also for the mantle-derived ilmenite megacryst PREM-1 from Premier which could not have been affected by crustal contamination. Additionally, there is no correlation between the ln(Si/Al) or ln(Mg/Yb) ratios and W concentrations (SI Appendix, Fig. S5 B and C) or μ¹⁸²W values for the primitive kimberlite samples. This indicates that the original W systematics of the primitive kimberlites were not modified by crustal contamination. In particular, one of the primitive kimberlites from Drybones, central Slave Craton, Canada, DRY-1 has a negative μ¹⁸²W value of −4.2, whereas the crustal rocks of the Slave craton formed in the Archean are characterized by positive μ¹⁸²W values of between 7.8 and 13.4 (33). This supports an interpretation that crustal contamination, if present, had a negligible effect on the W isotopic composition of sample DRY-1.

Data Availability

All study data are included in the article and/or SI Appendix.