Introduction

Compared to the primitive mantle, most normal mid-ocean ridge basalts (N-MORB) are characterized by lower abundance ratios of highly incompatible elements to moderately incompatible elements, and lower 87Sr/86Sr, higher 143Nd/144Nd and 176Hf/177Hf ratios1,2. These geochemical signatures define N-MORB as depleted3,4. It is widely accepted that depleted N-MORB are derived from mantle sources that were modified during ancient melting events3,4. In contrast, basalts that erupted at intraplate settings, including ocean islands, large igneous provinces (LIP), oceanic plateaus, and intracontinental basalt provinces, generally exhibit higher (highly incompatible elements)/(moderately incompatible elements) ratios, along with higher 87Sr/86Sr, lower 143Nd/144Nd and 176Hf/177Hf ratios, indicating an enriched geochemical signature3,4,5,6. However, not all hotspot-related lavas are enriched. Many hotspot settings also produce lavas with depleted geochemical characteristics, for examples, Hawaii7,8,9, Galápagos10,11, Iceland12,13,14, the Azores15,16, Heard Islands (Kerguelen)17, Loranchet Peninsula (Kerguelen)18, Ninetyeast Ridge (Kerguelen)19,20,21, and the Cretaceous Comei LIP (Kerguelen)22,23,24.

The origin of the depleted component in hotspot magmas remains debated6,9,16,25,26,27. For lavas with compositions resembling N-MORB, particularly in plume-ridge interaction settings, the prevailing hypothesis suggests incorporation of N-MORB or its source into ascending mantle plumes7,10,17,26,27. Examples include Iceland28, Hawaii7,29, Galápagos27, and Loranchet Peninsula (Kerguelen)18. Conversely, for lavas with compositions distinct from N-MORB, the depleted component is considered intrinsic to the hotspot8,9,30,31,32. This component may derive from recycled oceanic lower crust gabbros, as proposed for Iceland14,33, or from garnet- and clinopyroxene-rich ancient mantle melting residues34, as suggested for the Kerguelen hotspot19,20 and Hawaiian-Emperor chain26. Determining whether depleted components in hotspot magmas originate from entrained depleted upper mantle, recycled crustal materials, or an ancient depleted plume component is challenging using trace elements and radiogenic isotopes alone. This is because both depleted upper mantle and plume components result from intramantle melting processes. Additionally, partial melting and extraction of incompatible element-rich melts from subducted oceanic crust in subduction zones can produce depleted garnet- and clinopyroxene-bearing residues (eclogites)35, which, when melted with ambient mantle, would yield geochemical features similar to ancient intramantle differentiation7,10,17,26,27. In all cases, ultradepleted Hf isotope values can develop, either in garnet-bearing residual mantle36,37,38 or residual eclogite39,40. However, 18O/16O ratios, sensitive to rocks-hydrosphere interactions, provide a diagnostic tool for distinguishing recycled gabbros from a depleted upper mantle or plume component41,42. Recycled gabbros typically exhibit variable, often low δ18O values, while intramantle differentiation yields a limited δ18O range42,43,44,45. This distinction suggests that depleted primary magmas with light oxygen isotope signatures more likely originate from recycled crustal materials, rather than mantle differentiation processes. Although low-δ18O basalts have been documented in Iceland, Hawaii, the Azores, and Canaries, their origin remains contentious, with interpretations divided between derivation from recycled hydrothermally altered gabbroic oceanic crust41,46,47,48,49,50,51 and crustal contamination by altered rocks52,53,54. Notably, no systematic study has yet specifically investigated the nature of the depleted component in hotspot mantle sources through oxygen isotope analysis.

The Kerguelen hotspot, active at the Kerguelen Archipelago55 and the Heard and McDonald islands56,57, is sustained by the Kerguelen mantle plume and represents one of Earth’s largest and longest-lived LIPs, originating from the Cretaceous Comei LIP58,59. It has generated two major features: (1) the Cretaceous Kerguelen LIP58,59, and (2) the 5000-km-long Ninetyeast Ridge, Earth’s longest volcanic lineament60 (Fig. 1a). The plume exhibits depleted components in Heard Islands17, Loranchet Peninsula (Kerguelen)18, the Ninetyeast Ridge19,20,21 and the Cretaceous Comei LIP22,23,24, formed after a prolonged interval. Understanding the isotopically depleted end-members of this long-lived plume is essential for assessing mantle geochemical heterogeneity, as they may sample multiple geochemical domains over time6,9. Unlike other hotspots, oxygen isotope data from Kerguelen remain limited61, leaving substantial gaps in understanding its mantle sources, crustal recycling, and the origin of depleted components.

Fig. 1: Overview of the spatiotemporal evolution of the Kerguelen Plume and Comei LIP, with gabbro and diabase outcrops in the Cona area.
figure 1

a Etopo1 topography showing the Eastern Indian Ocean and adjacent areas and Kerguelen plume-related magmas (revised after refs. 63,64). Source: NOAA National Centers for Environmental Information. b Simplified geologic map showing spatial extent and distributions of Cretaceous igneous rocks of the remnant Comei LIP (revised after refs. 23,65). c Gabbro and d Diabase outcrops in the Cona area.

Full size image

The Tethyan Himalaya, situated along the northern margin of the Himalaya orogen, represents the northern passive margin of Greater India in paleogeographic reconstructions62. The Comei LIP, located in the eastern Tethyan Himalaya (Fig. 1a), comprises eroded remnants of a LIP linked to Cretaceous Kerguelen mantle plume activity22,24,63,64,65. Zircon U–Pb dating reveals magmatic activity between 147 and 115 Ma, peaking around 132 Ma, with a clustering at 134–130 Ma, for Comei LIP65,66,67,68. This timing aligns with magmatism in the Bunbury Basalt and Naturaliste Plateau in southwestern Australia, paleogeographically adjacent regions63,64,65. The Comei LIP is dominant by basaltic lavas, diabase sills and dikes, gabbroic intrusions, and minor ultramafic and felsic intrusions (Fig. 1b)23,24,65,66,67,68. Its volcano-sedimentary sequences include the Lower Cretaceous Sangxiu and Lakang formations, exposed in the northern and southern regions, respectively (Fig. 1b). Geochemically, the mafic rocks primarily exhibit ocean island basalt (OIB)-like enriched signatures22,23,24,66,68, though depleted N-MORB-like intrusions are identified in the Cona22 and the Niangzhong areas24. Some basaltic lavas show transitional compositions between OIB and N-MORB, with flat to mildly enriched rare earth element (REE) patterns22,24,68.

To explore the nature of the depleted component in hotspot magmas and its implications for long-term mantle evolution and crustal recycling, this study integrates zircon O–Hf–U–Pb isotopes, whole-rock major and trace elements, and Nd isotopes from N-MORB-like gabbros in the Cona area of the Comei LIP, linked the Kerguelen mantle plume (Fig. 1a, b). Whole-rock geochemistry for enriched OIB-type diabases from the same area were also analyzed for comparison. We identified ultradepleted Hf and low δ¹⁸O values (3.97–4.80‰) negatively correlating with ages for N-MORB-like gabbros, suggesting recycled gabbroic oceanic crust as the depleted component in the Comei LIP.

Results and discussion

Whole-rock geochemical compositions of Cona gabbros and diabases

Whole-rock geochemical analysis of the Cona gabbros reveals relatively homogeneous major, trace, and radiogenic isotope compositions (Supplementary Data S1). The samples contain ~49 wt.% SiO2, 6–7 wt.% MgO, low TiO2 (<2 wt.%), and moderate Cr (120–150 ppm) and Ni (70–80 ppm) levels (Supplementary Fig. S1; Supplementary Data S1). They exhibit a typical N-MORB-like REE and trace element distribution patterns, with light REE depletion relative to middle and heavy REE, except for moderate enrichment in Rb, Ba, Th, U, K, and Pb (Fig. 2a, b). But they show elevated Sc/Nb and Y/Nb ratios compared to N-MORB (Fig. 2c). In contrast, the Cona diabases show broader compositional variation (Supplementary Data S1), with higher SiO2 (47–54 wt.%), TiO2 (>3.5 wt.%), and lower MgO (4–6 wt.%), Cr (40–150 ppm), and Ni (15–70 ppm) than the gabbros (Supplementary Fig. S1; Supplementary Data S1). Their REE and trace element distribution patterns are characteristic of OIB, except for variable depletion or enrichment in Rb, Ba, K, and Sr (Fig. 2a, b). They also exhibit more enriched Nd isotopic composition (Fig. 2d; Supplementary Data S2). These geochemical features align with the published data (Fig. 2), confirming the samples as representative N-MORB- and OIB- like mafic rocks of the Comei LIP.

Fig. 2: Selected whole-rock geochemical features for the Cretaceous igneous rocks from the Comei LIP (Kerguelen).
figure 2

a Chondrite-normalized REE patterns for Comei LIP mafic rocks, with OIB and N-MORB values for comparison. b Primitive mantle-normalized trace element patterns for Comei LIP mafic rocks, with OIB and N-MORB values for comparison. c Sc/Nb vs. Y/Nb ratios, sensitive to garnet and clinopyroxene, in Comei LIP igneous rocks. d εNd(t) versus Nb/U ratio for Comei LIP igneous rocks. The dark gray curved arrow in (d) illustrates the compositional trend of crustal assimilation by felsic rocks. Symbol meanings: dark green circles, N-MORB-like Cona gabbros; dark purple squares, OIB-like Cona diabases (this study); light green circles, N-MORB-like samples (literature); light purple squares, OIB-like samples (literature); light yellow triangles, transitional N-MORB- to OIB-like samples; light blue diamonds, OIB-like samples with lithospheric contamination; gray crosses, felsic rocks; dark blue star, N-MORB. Data sources: Chondrite and OIB, ref. 1; N-MORB, ref. 2; Whole-rock and zircon data for Comei igneous rocks, Supplementary Data 6 and 7119. Abbreviations: CMLIP Comei LIP, TS this study.

Full size image

Zircon U–Pb–Hf–O isotope compositions for Cona gabbros

Zircon is rare in the gabbros, with only several to a few hundred grains typically extracted from 3–5 kg of hand specimens. Zircons from gabbro sample 15XZ132, which contains the most zircon grains suitable for U–Pb–Hf–O isotope analyses among our samples, exhibit diverse crystal forms with length-to-width ratios of 1:1 to 2:1. Most zircons are euhedral to sub-euhedral, displaying oscillatory, sector, or other zonings in CL images (Supplementary Fig. S2). Combined with Th/U ratios >0.1 (Supplementary Data S4), these features indicate a magmatic origin69. Secondary ion mass spectrometer (SIMS) U–Pb dating of 22 zircon spots yielded 206Pb/238U ages of 126.7–137.9 Ma, with a weighted mean age of 132.6 ± 1.5 Ma (Fig. 3a; Supplementary Data S4). The age distribution is overdispersed (MSWD = 2.0) for a single magmatic batch, but the data suggest an emplacement age of ~133 Ma. Two spots yielded older ages of 474 Ma (Spot #4) and 406 Ma (Spot #15) (Supplementary Data S4). The syn-magmatic zircon age aligns with the peak magmatic activity of the Comei LIP65, indicating crystallization during gabbro emplacement.

Fig. 3: Zircon U–Pb–Hf–O isotope features of Cretaceous igneous rocks from the Comei LIP.
figure 3

a Zircon U–Pb isotopic data for N-MORB-like gabbros from Cona area. Error ellipses represent 1σ; MSWD—mean square of weighted deviates. b Histogram of zircon δ18O values for the N-MORB-like gabbros from the Cona area, showing a quasi-normal distribution with an average value below the typical mantle zircon range (light gray area)70. c Zircon δ18O vs. 206Pb/238U age for N-MORB-like gabbros from the Cona area. Green points represent single spots, with 2 standard deviation bars indicating repeated measurements. Light gray area denotes the typical mantle zircon range. Light green arrow highlights the first-order negative covariation trend. (d) εHf(t) vs. 206Pb/238U age for zircon grains in the Cretaceous igneous rocks from the Comei LIP. Symbols follow Fig. 2.

Full size image

Oxygen isotope analyses of 22 syn-magmatic zircon spots yielded uniformly low δ18O values, ranging from 3.97‰ to 4.80‰, with an average of 4.38‰ ± 0.09 (Fig. 3b, Supplementary Data S3). These values, follows a quasi-normal distribution, are markedly lower than the typical mantle zircon (δ18O = 5.3 ± 0.6‰)70. Similar to 206Pb/238U ages, the δ18O variation (MSWD = 1.8) is slightly overdispersed for zircons crystallized from a homogeneous magma in a closed system.

Lu-Hf isotope analysis of 20 syn-magmatic zircon spots revealed εHf(t) values ranging from 13.7 to 22.8, averaging 17.9 ± 1.3 (Fig. 3c; Supplementary Data S5). These exhibit the most depleted Hf isotope compositions among Comei LIP mafic rocks, many even exceeding the depleted mantle (ultradepleted) (Fig. 3d). Like U–Pb and O isotopes, the Hf isotope variation (9 epsilon) is too overdispersed for crystallization from a homogeneous magma in a closed system.

The origin of the overdispersion of O-U-Pb-Hf isotopes and light O isotopes

The overdispersion of O–U–Pb–Hf isotopes and light O isotopes in the Cona gabbro sample likely has a geological origin, as analytical issues are ruled out by standard data quality (See “Methods” for detail). The zircon grains probably crystallized at different times (yielding varied ages) from magmas of differing compositions (resulting in heterogeneous Hf–O isotopes) and were later captured together, indicating an open-system mixing process before or during the emplacement and cooling of the parental magma. In principle, this open-system process, combined with light O isotopes71, could arise from: (1) meteoric water interaction with basaltic magma in a long-lived system; (2) assimilation of altered crustal rocks with light O by basaltic magma; or (3) recharge mixing among basaltic magmas with variable δ18O depletion and heterogeneous Hf isotopes. A long-lived magma system is plausible, given the Comei LIP’s duration of at least ten million years before Cona gabbro formation22,24,63,64,65.

Direct water addition to magma is unlikely due to limited H₂O transportability and solubility in ductile rocks and silicate melts71,72. Water typically migrates away from hot magma bodies during liquid/vapor transitions, as shown by studies on low-δ¹⁸O alteration near ignimbrite deposits73. Mass-balance constraints further exclude this process, as even highly ¹⁸O-depleted meteoric water (e.g., δ¹⁸O of –19‰ in Yellowstone)74 would require ~20 wt.% H₂O to shift normal δ¹⁸O magma (+6‰) below 0‰71, exceeding maximum water solubility in magmas. Thus, direct meteoric water-magma interaction is ruled out as the cause of light O isotopes.

The Cona gabbros show minor enrichment in large-ion lithophile elements (Fig. 2B), indicating slight late-stage alteration. However, low LOI values (<2 wt.%) suggest limited impact on bulk composition. Zircons, resistant to alteration69 as shown by their ultradepleted Hf isotopes, confirm that their low oxygen isotope values are not due to hydrothermal alteration. Crustal contamination, particularly via dissolution of felsic or mafic igneous rocks, is a plausible mechanism for introducing light oxygen into magma systems71. However, felsic crustal contamination typically elevates incompatible trace elements (e.g., light REE) and enriches whole-rock Nd isotopes, given the compositional contrast with basaltic magmas from depleted mantle sources75. In contrast, the Cona gabbros and other N-MORB-like mafic rocks in the Comei LIP show uniform light REE depletion (exceeding N-MORB; Fig. 2a) and consistently high εNd(t) values (Fig. 2d), with no correlation with contamination-sensitive ratios like Nb/U (Fig. 2d). These features rule out substantial felsic crustal contamination. Additionally, zircon δ¹⁸O values show no correlation with εHf(t) (Supplementary Fig. S3), a proxy for crustal contamination or magma mixing76. Notably, even zircons with ultradepleted Hf isotopes exhibit δ¹⁸O values below normal mantle zircons (Supplementary Fig. S3), excluding felsic crustal contamination as the cause of low δ¹⁸O values. An alternative contamination mechanism could involve hydrothermally altered mafic rocks with light oxygen isotopes but chemical compositions resembling the parental magma of the Cona gabbros, analogous to cannibalization processes in felsic systems like Yellowstone71. Such contamination would subtly perturb geochemical signatures, with oxygen isotope dynamics operating on dual timescales: short-term cycles (~1–10 kyr) triggered by caldera collapse produce immediate post-collapse eruptions of the lowest-δ¹⁸O magmas, followed by gradual δ¹⁸O recovery as the depleted reservoir is exhausted, while superimposed long-term trends (1–10 Myr) reflect a secular δ¹⁸O decrease due to cumulative cannibalization–a feature in contrast to the observation in Cona gabbros (Fig. 3C). Crucially, while hydrothermally altered country rocks (low-δ¹⁸O contaminants) form at magma chamber margins, zircons remain resistant to hydrothermal alteration and retain their pre-collapse δ¹⁸O signatures. These unmodified zircons persist as normal-δ¹⁸O xenocrysts or cores within newly formed low-δ¹⁸O magmas–a feature not observed in the Cona gabbros (Fig. 3c; Supplementary Fig. S2). This framework explains why our documentation of increasing zircon δ¹⁸O values at million-year timescales fundamentally conflicts with cannibalization models. Studies on olivine oxygen isotopes also suggest that ongoing assimilation would progressively lower magma δ¹⁸O values, evident by the clear positive correlation between olivine forsterite content and δ¹⁸O values52,54,77.

Having excluded hydrothermal reactions and assimilation-fractional crystallization processes, the remaining explanation for the light oxygen isotopes and O–U–Pb–Hf overdispersion is recharge mixing of heterogeneous basaltic magmas from the mantle. The recharged magmas must have progressively higher δ¹⁸O values, approaching normal mantle-like compositions, while maintaining ultradepleted Hf isotopes. This implies a mantle component with light oxygen isotopes, ultradepleted Hf, and high Hf content (resisting dilution by other sources), gradually consumed during melting alongside mantle sources with normal oxygen isotopes and lower Hf content. The nature of this component is discussed below.

Nature of the low-δ18O ultradepleted component: recycled gabbroic oceanic crust

The δ18O values of N-MORB-like gabbros from the Comei LIP are consistently ~1‰ lower than typical mantle zircon (Fig. 3b). During peridotite melting, pyroxenes (with higher δ18O) preferential participate, typically resulting in basaltic melts slightly higher in δ18O relative to their mantle source78. Fractional crystallization can increase whole-rock δ18O (δ18OWR) by up to 1‰, but zircon δ18O (δ18OZrc) remains stable due to proportional increase in Δ18OWR-Zrc with increasing abundance of high δ18O minerals (e.g., quartz and feldspar). Thus, δ18OZrc values markedly below mantle levels cannot arise from melt-mineral differentiation during mantle melting or magma evolution45. The sub-continental lithospheric mantle is a potential low-δ¹⁸O reservoir79,80,81,82,83, invoked to explain low-δ¹⁸O features in kimberlites84 and mantle zircons70,85. Eclogites (δ¹⁸O as low as 2‰)82 and metasomatized mantle rocks (e.g., phlogopite-bearing lherzolites or mica-amphibole-rutile-ilmenite-diopside suite, with δ¹⁸O as low as 4.4‰ in clinopyroxene and 2.4‰ in ilmenite)79,80,81,83 are the only sub-continental lithospheric mantle lithologies with δ¹⁸O markedly below mantle values. However, assimilation of phlogopite-bearing lherzolites or mica-amphibole-rutile-ilmenite-diopside suite is unlikely, as these are potassium-rich and have enriched Nd isotopes, contrasting with the Cona gabbros (Fig. 2). Additionally, Comei LIP mafic rocks influenced by sub-continental lithospheric mantle exhibit arc-like trace element patterns and enriched radiogenic isotopes23,24,68. Contaminated mafic rocks also show low Nb/U ratios and εNd(t) values, approaching to felsic rocks in εNd(t) vs. Nb/U plots (Fig. 2d), distinct from Cona gabbros and other N-MORB-like mafic rocks (Fig. 2). In principle, If the Cona gabbro primary melt interacted with eclogites, it could acquire low-δ¹⁸O and ultradepleted Hf isotopes due to garnet’s radiogenic Lu decay36,37,38,39,40. However, low-δ¹⁸O eclogites are rarer than high-δ¹⁸O ones82,85, making selective interaction with only low-δ18O regions unlikely86. Thus, the low-δ18O features of the Cona gabbros are best explained by derivation from a sub-continental lithosphere source.

A potential source for the low-δ18O signature in mantle plume-derived rocks is the assimilation of low-δ18O material from the outer core87. However, this would result in lower Mn/Fe ratios due to the core’s high Fe content, which is not observed in the Cona gabbros (Supplementary Fig. S4a). Byerly et al.88 proposed that low-δ18O mantle sources could originate from deep mantle domains undergoing fractional crystallization during the Hadean. However, the Cona gabbros lack of obvious negative Zr anomalies and exhibit high Hf/Sm ratios (Supplementary Fig. S4b, c), inconsistent with deep crystallization processes89. Furthermore, Ouyang et al.86 argue that deep mantle fractional crystallization, dominant by bridgmanite and Ca-perovskite, would increase δ18O in residual melt, ruling it out as the cause of low-δ18O in Comei mafic rocks.

Low-δ18O components in intraplate mafic rocks may instead originate from recycling of subducted oceanic lithosphere41,70,86,90, which acquires low-δ18O signatures in the gabbroic layer through high-temperature water-rock interactions42,45. This is supported by mantle eclogite, the metamorphic equivalent of the subducted oceanic crust82, which display δ18O values (2–12‰) comparable to altered oceanic crust43,44, indicating minimal δ¹⁸O change during subduction91,92. Beyond light oxygen and ultradepleted Hf isotopes, the Cona gabbros exhibit higher Sc/Nb and Y/Nb ratios than N-MORB (Fig. 2b, c), suggesting a mantle source enriched in Sc and Yb. Similar trace element and Hf isotopic features in Ninetyeast Ridge lavas have been attributed to intramantle melting of garnet- and clinopyroxene-rich plume component19,20. However, this cannot explain the low-δ18O values in the Cona gabbros. Instead, garnet- and clinopyroxene-rich components likely derived from partial melting of subducted oceanic crust under eclogitic conditions35, persisting in the mantle before incorporation into hotspot magmas6. Garnet decay further contributes to ultradepleted Hf isotopes36,37,38,39,40.

Thus, the low-δ¹⁸O component in the Comei LIP mantle source is most likely derived from recycled gabbros of subducted oceanic crust (Fig. 4). These gabbros likely melted with surrounding mantle to form the depleted Cona gabbros (Fig. 4a), as direct melting of recycled components alone cannot produce MgO-rich and SiO2-poor gabbros, and would yield more enriched incompatible trace elements93,94. Gabbroic (eclogitic) components melt more readily than ambient mantle and are rapidly consumed once melting begins95, consistent with the gradual increase in δ¹⁸O observed in Cona zircons. The lack of similar temporal variation in Hf isotopes is due to higher Hf concentrations in gabbro/eclogite-derived melts compared to ambient mantle melts96 as subducted eclogite-derived melts would have 3–5 times higher Hf concentrations than ambient mantle-derived melts35, making Hf isotopes less sensitive to change.

Fig. 4: Schematic diagram illustrating the generation of N-MORB-like gabbros with light oxygen isotope compositions from Cona area of the Comei LIP in the context of plate tectonics and mantle plume.
figure 4

a Schematic cross-section of the mantle showing major mantle structures and locations of potential chemical reservoirs with variable oxygen isotopic, radiogenic isotopic, trace elemental compositions. These include a heterogeneous large low shear velocity province (LLSVP) (indicated by different shades of dark red, including mantle plumes), ultra-low velocity zones (ULVZs, red) and subducted oceanic lithosphere transporting recycled surface materials with variable oxygen isotope composition (e.g., gabbros, blue) into the mantle (Revised after ref. 6). Upper mantle peridotite, N-MORB, and OIB whole-rock δ18O values are from refs. 41,42,120, respectively; arc lavas whole-rock δ18O values are from refs. 78,108. Whole-rock δ18O values for Cona gabbros area were calculated from zircon using ref. 121. Abbreviations: EM-I, enriched mantle I; EM-II enriched mantle II, HIMU high μ (U/Pb). Mantle plume compositions reflect contributions from multiple mantle reservoirs, including recycled crustal material. b Penrose model of oceanic crust based on Oman ophiolite, showing estimated alteration temperature and bulk rocks δ18O data (modified from ref. 82). Whole-rock δ18O ranges for the N-MORB-like Cona gabbros are included for comparison.

Full size image

This study provides robust evidence that the depleted component in the Kerguelen hotspot system originates from recycled gabbroic oceanic crust (Fig. 4). Combined with previously identified ancient intramantle melting residues19,20, these findings reveal multiple depleted components within the Kerguelen mantle plume sources. These components, preserved and transported over geological timescales, emerge as depleted magmas during both the LIP and subsequent hotspot stages, despite the substantial temporal gap between them. This long-term preservation and reemergence of depleted materials highlight the heterogeneous nature of the deep mantle and its dynamic evolution (Fig. 4)6. Our results underscore the importance of recycled oceanic crust in shaping mantle plume geochemical diversity, offering critical insights into mantle convection, recycling, and Earth’s long-term chemical differentiation (Fig. 4).

Implications for the difficulties in identifying recycled gabbroic components

Hotspot lavas are thought to originate from a complex mixture of materials, including lower mantle components, recycled oceanic lithosphere (sediments, basaltic crust, serpentinites, metasomatized lithosphere), depleted lithosphere, recycled continental crust and lithosphere from delamination and subduction processes, and even core contributions (Fig. 4a). Among which, the substantial volume of recycled oceanic lithosphere in plume sources94,97 primarily explains the variable geochemical signatures of their products compared to MORB (Fig. 4a)5,6,98, particularly for oxygen isotopes41,42.

The Penrose model of oceanic crust, based on the Oman ophiolite, shows that lower crustal gabbros and upper crustal basalts/sediments deviate in δ¹⁸O values in opposite directions from typical peridotites (δ18O = 5.5 ± 0.2‰) (Fig. 4b), helping to distinguish subducted oceanic components41. Gabbroic layers exhibit distinct light δ¹⁸O signatures (Fig. 4b), unique trace elements like positive Sr and Eu anomalies, and Sc enrichment due to plagioclase and clinopyroxene accumulation99. Gabbros also make up a larger volume of oceanic crust than the upper layers (Fig. 4b)44. These features theoretically enable the detection of recycled gabbros in mantle plume products, as 18O-depleted rocks are known to be subducted and occasionally identified as high-pressure mantle xenoliths82. Consequently, we expect to find evidence of their oxygen isotope signatures in hotspot basalts derived from sources where such rocks are entrained42. However, this remains exceptionally challenging14,41,49,86,90, not only because low-δ¹⁸O signatures are rare and no OIB suites exhibit consistently low-δ¹⁸O values100, but also due to ongoing debates over whether these low-δ¹⁸O features are original.

While a mantle source of low δ18O, attributed to recycled hydrothermally altered oceanic crust gabbros, has been proposed for basaltic magmas from the Iceland51, the Hawaii101, the Azores46,47, and the Canaries48,49,50, others argue that crustal contamination by hydrothermally altered rocks is responsible for low δ18O basaltic magmas from the Hawaii52,53 and the Azores54. Evidence supporting a mantle origin includes: (1) no correlation between δ18O and olivine forsterite content in Azores basalts46; (2) limited ranges of SiO2 and 3He/4He ratios in some Iceland basalts102; (3) absence of contamination indicators like elevated Cl/K ratios or correlations between low δ18O and Pb–Nd–Sr–He isotopes in Iceland basalts100; (4) low δ18O with elevated Nb/B and variable δ11B in Azores basalts47; (5) no correlation between 187Os/188Os and fractionation indices (e.g., MgO) in Os-rich Canary Island samples48; and (6) low δ18O olivines with high 3He/4He ratios in Iceland basalts103. Conversely, evidence for crustal contamination includes: (1) correlations between decreasing δ18O and 3He/4He in Hawaiian basalts41; (2) abrupt loss of low δ18O signatures in Mauna Kea lavas at the submarine-to-sub aerial transition without radiogenic isotope changes52; (3) correlations between δ18O and enrichment indices (e.g., K2O/TiO2, La/Sm) and differentiation indices (e.g., Mg#, CaO/Na2O) in Iceland basalts104; (4) decoupling between normal olivine and low-δ18O matrix in Iceland basalts53,105; (5) positive correlations between olivine δ18O and forsterite content in Hawaiian52,77 and Azores basalts54; and (6) low-δ18O olivines in the western Galapagos, where the lithosphere is thickest106.

Overall, the geochemical evidence remains inconclusive, and interpretations of the same observations often diverge. For example, the positive correlation between forsterite content and δ18O is considered strong evidence for crustal contamination in OIB but is interpreted as evidence for interactions between recycled oceanic crust-derived melts and normal mantle-derived components in continental basalts, kimberlites, and komatiites84,86,90. Nevertheless, geological evidence from extensively studied regions like Hawaii and Iceland strongly supports the dominant role of crustal contamination42. Iceland, in particular, is unique due to its abundance of low δ18O magmas and thick crust (up to 40 km), which has undergone extensive hydrothermal alteration involving low δ18O meteoric water at high latitudes. Notably, light oxygen isotopes are absent in alkaline lavas from areas distal to Iceland’s rift zones100. This suggests that the 18O-depleted mantle hypothesis requires substantial evidence, especially given the extreme 18O depletion observed in hydrothermally altered crustal rocks42. Similarly, the sharp disappearance of low δ18O signatures in Mauna Kea lavas at the submarine-to-sub aerial transition, without changes in radiogenic isotope compositions, further challenges the mantle origin hypothesis52.

In summary, no definitive example of hotspot basaltic rocks with low-δ18O signatures conclusively attributed to recycled gabbroic oceanic crust currently exists. Moreover, all reported low-δ18O basalts are geochemically enriched, as shown by light REE enrichment over heavy REE (Fig. 5), and often isotopically enriched, making it difficult to entirely rule out crustal contamination. In contrast, the Cona gabbros to the best of our knowledge represent the first example of a geochemically depleted, even ultradepleted source with low-δ18O signatures (Fig. 5), offering a unique perspective on this debate. Our study, combining oxygen isotopes, trace elements, and radiogenic isotope data, clarifies why gabbro contributions are elusive. First, trace element masking: While gabbros are geochemically distinctive, their trace element concentrations are relatively low compared to sediments and basaltic crust99, making their signatures hard to detect unless they dominate the source96. Even when they do, distinguishing gabbroic contributions is challenging, as seen in debates over high Sr/Nd ratios in Icelandic basalts14,107. This complexity arises because basaltic crust, despite some enrichment, has a composition similar to its upper-mantle source, especially after subduction zone melting and melt extraction35, which makes it easily masked. Second, oxygen isotope homogenization: Although gabbros can exhibit low δ¹⁸O values (+1.7‰), their average ranges (+4.0 to +8.0‰) often overlap with mantle values (Fig. 4b)100, making their detection difficult. Mixing with other oceanic crust components further homogenizes their isotopic signatures, as seen in adakites and high-Mg andesites108. Third, contamination: Low-δ¹⁸O signatures in gabbro-derived melts can be masked by contamination from the sub-continental lithospheric mantle or crust during ascent75,84, hindering identification. Thus, detecting recycled gabbro in hotspot sources is most feasible when the low δ¹⁸O recycled gabbroic (eclogitic) components and ambient depleted mantle dominant the plume sources, and the melt remains uncontaminated during ascent. These conditions are rarely met, explaining the scarcity of gabbroic contributions in plume and subduction zone settings41,108,109. These challenges underscore the significance and value of our discovery.

Fig. 5: δ¹⁸O versus (La/Yb)N for light-oxygen hotspot mafic rock samples that are ascribed to be acquire the light-oxygen features from recycled gabbroic oceanic crust in their plume sources (Supplementary Data S8).
figure 5

To ensure data reliability, we retained only oxygen isotope data obtained through laser fluorination and SIMS in situ analyses of olivine, glass, and melt inclusions. The whole-rock (WR) oxygen isotope compositions were calculated from olivine data follow ref. 42. Data points with δ¹⁸O values above mantle range were retained, as other olivines from the same samples exhibit below-mantle values. Notably, the Cona gabbros represent the only known occurrence of light-oxygen magmas exhibiting light REE depletion relative to heavy REE. The subscript N in (La/Yb)N denotes chondrite-normalized values. Error bars represent 2σ.

Full size image

Methods

Zircon grains were separated by conventional magnetic and heavy liquid techniques, and further purified by hand-picking under a binocular microscope for further U-Pb, O, and Lu-Hf isotopic analyses.

To investigate the U–Pb–Hf–O isotope compositions of zircon grains, we employed in situ analytical techniques, specifically, field-emission scanning electron microprobe, SIMS, and laser ablation multi-collector inductively coupled plasma mass spectrometry. In addition, we analyzed major and trace elements, as well as Nd isotope compositions of whole-rock samples, using X-ray fluorescence spectrometry, inductively coupled plasma mass spectrometry, and multi-collector inductively coupled plasma mass spectrometry, respectively.

Field-emission scanning electron microprobe

Zircon cathodoluminescence imaging was performed using a TESCAN MIRA3 Field-emission scanning electron microprobe at the State Key Laboratory of Lithospheric and Environmental Coevolution, University of Science and Technology of China. The instrument was operated at an accelerating voltage of 10 kV, with a beam current of 15 nA, and a working distance of 13 mm.

SIMS

In-situ oxygen isotope analyses of zircon were analyzed using a Cameca IMS 1280-HR at the State Key Laboratory of Isotope Geochemistry, Guangzhou Institute of Geochemistry, Chinese Academy of Sciences, following the method of ref. 110. A Cs⁺ primary ion beam was rastered over a 10 μm area, with a 20 μm analysis spot. Oxygen isotopes were measured in multi-collector mode, with magnetic field control via an NMR probe. Instrumental mass fractionation was corrected using the Penglai zircon standard (δ¹⁸O = 5.31 ± 0.10‰)110, normalized to Vienna Standard Mean Ocean Water. Internal precision was typically ±0.20‰ (2σ), and external precision for repeated analyses of the Penglai zircon was ±0.34 to ±0.41‰ (2 SD). Qinghu zircon yielded δ¹⁸O values (5.46 ± 0.07‰, n = 19) consistent with its recommended value (5.40‰ ± 0.20‰)111. Oxygen isotope data are listed in Supplementary Data S3.

After oxygen isotope analyses, U-Pb isotopic compositions of zircon grains were also conducted using the Cameca IMS 1280-HR. Operating and data processing followed procedures described by ref. 112. The diameter of the analysis spot is 20 μm. Samples were calibrated against the Plěsovice zircon (206Pb/238U = 0.05369, 337.1 Ma)113. A long-term uncertainty of ±1.5% (1 RSD) for 206Pb/238U was applied, and errors during this study averaged ±1% (1 RSD). Common Pb was corrected using the 204Pb method, assuming surface contamination. Mean ages are reported with 95% confidence intervals, and discordant ages (>±10%) were discarded. Data were processed using Isoplot/Ex3114. Secondary standard zircon Qinghu yielded a mean 206Pb/238U age of 159.6 ± 1.3 Ma (n = 13), consistent with its reference value (159.5 ± 0.2 Ma)111. U-Pb data are available in Supplementary Data S4.

Laser ablation multi-collector inductively coupled plasma mass spectrometry

Zircon Hf isotope analyses were performed using a Neptune Plus multi-collector inductively coupled plasma mass spectrometer coupled with a RESOlution S155 laser system at the State Key Laboratory of Lithospheric and Environmental Coevolution, University of Science and Technology of China, Hefei. The analytical conditions included a laser spot diameter of 43 μm, a repetition rate of 8 Hz, and an energy density of 3.5 J/cm². A mixture of helium (375 ml/min) and argon (~0.9 L/min) served as the carrier gas. Each analysis comprised a 15-s background measurement followed by 35 s of data acquisition. To correct for isobaric interference of ¹⁷⁶Lu on 176Hf, the 176Lu/175Lu ratio of 0.02655 was utilized, following the method outlined by ref. 115. and the isobaric correction model proposed by ref. 116. Yb mass fractionation corrections were applied using the βYb value, determined by fitting βYb = k × βHf with the MUNZ zircon standard during the analytical session117. Data processing was conducted offline using the LAZrnHf-Calculator@HFUT_v1.4.11.4116. The accuracy and precision of the measurements were monitored using zircon standards 91500, Qinghu, and Penglai. The corrected 176Hf/177Hf ratios obtained were 0.282306 ± 8 for 91500 (n = 7), 0.283018 ± 14 for Qinghu (n = 9), and 0.282919 ± 10 for Penglai (n = 5), consistent with published values110,111,118. All Hf isotope data are provided in Supplementary Data S5.

X-ray fluorescence spectrometry

The major element analysis of whole rock at Wuhan SampleSolution Analytical Technology Co., Ltd. is performed using X-ray fluorescence spectrometry with a Rigaku Primus II XRF analyzer. The analytical procedure follows the national standard GB/T 14506.28-2010, ensuring consistent and reliable results. The analysis begins with sample preparation, where a representative rock sample (~0.9 g) is calcined or ignited and mixed with ~9.0 g of a Lithium Borate Flux (a 50–50% mixture of Li2B4O7 and LiBO2). This mixture is then fused in an auto fluxer at temperatures ranging from 1050 °C to 1100 °C, resulting in a flat molten glass disc that is allowed to cool and is subsequently analyzed for major elements. The data obtained are processed using specialized software and calibrated with certified reference materials, achieving analytical precision better than ±1–2%. Two standards, plagioclase amphibolite GSR-15 and granitic gneiss GSR-14, are analyzed simultaneously to monitor the analytical quality. Major element data are listed in Supplementary Data S1.

Inductively coupled plasma mass spectrometry

Trace element analysis of whole rock samples was conducted using an Agilent 7700e inductively coupled plasma mass spectrometry at Wuhan SampleSolution Analytical Technology Co., Ltd., Wuhan, China. The sample preparation procedure was as follows: (1) Sample powder (200 mesh) was placed in an oven at 105 °C and dried for 12 h; (2) 50 mg of dried sample powder was accurately weighed and placed in a Teflon bomb; (3) 1 ml of HNO₃ and 1 ml of HF were slowly added to the Teflon bomb; (4) The Teflon bomb was sealed within a stainless steel pressure jacket and heated at 190 °C in an oven for over 24 h; (5) After cooling, the Teflon bomb was opened, and the solution was placed on a hotplate at 140 °C and evaporated to incipient dryness, followed by the addition of 1 ml of HNO₃ and evaporation to dryness again; (6) 1 ml of HNO₃, 1 ml of MQ water, and 1 ml of internal standard solution (1 ppm In) were added, the Teflon bomb was resealed and heated again at 190 °C for over 12 h; (7) The final solution was transferred to a polyethylene bottle and diluted to 100 g with 2% HNO₃ for analysis. Two standards, plagioclase amphibolite GSR-15 and granitic gneiss GSR-14, are analyzed simultaneously to monitor the analytical quality. Trace element data are listed in Supplementary Data S1.

Multi-collector inductively coupled plasma mass spectrometry

Whole-rock Nd isotope analysis was conducted at Guizhou Tongwei Analytical Technology Co., Ltd. At first, whole-rock samples were dissolved, and Nd isotopes were separated and purified using TRU Resin (50–100 µm) and LN Resin (50–100 µm) from Eichrom Technologies. Then, Nd isotope compositions were measured on a Nu Instruments Plasma 3 multi-collector inductively coupled plasma mass spectrometry, with the JNdi standard used for monitoring, yielding a 143Nd/144Nd ratio of 0.512115 ± 4 (2 SD, n = 7). The United States Geological Survey reference materials BHVO-2 and BCR-2 were processed alongside the samples for quality control, and the results for the 143Nd/144Nd ratios (0.5126360 and 0.5129766) were consistent with recommended values, ensuring reliable measurements within acceptable analytical errors (2σ). Nd isotope data are listed in Supplementary Data S2.