Earth’s mantle composition revealed by mantle plumes

Mantle plumes originate at depths near the core−mantle boundary (~2,800 km). As such, they provide invaluable information about the composition of the deep mantle and insight into convection, crustal formation, and crustal recycling, as well as global heat and volatile budgets. In this Review, we discuss the effectiveness and challenges of using isotopic analyses of plume-generated rocks to infer mantle composition and to constrain geodynamic models. Isotopic analyses of plume-derived ocean island basalts, including radiogenic (Sr, Nd, Pb, Hf, W, noble gas) and stable isotopes (Li, C, O, S, Fe, Tl), permit determination of mantle plume composition, which in turn generate insight into mantle plume origins, dynamics, mantle heterogeneities, early-formed mantle reservoirs, crustal recycling processes, core−mantle interactions and mantle evolution. Nevertheless, the magmatic flux, temperature, tectonic environment and compositions of mantle plumes can vary. Consequently, plumes and their melts are best evaluated along a spectrum that acknowledges their different properties, particularly mantle flux, before making interpretations about the interior of the Earth. To provide insight into specific mantle and plume processes, future work should document correlations across elemental and isotopic data sets on the same sample powder, coordinate targeting sampling strategies, and refine stable isotopic fractionation factors through experiments. Such work will benefit from collaboration across geochemical laboratories, as well as among geochemists, mineral physicists, seismologists and geodynamicists. The mantle of the Earth influences many dynamic processes such as crust formation, recycling and mantle convection. This Review describes modern isotopic methods used to characterize plume-derived basalts and gain insight into the composition of the mantle.


Review article
This Review focuses on the geochemistry and geodynamics of deep plumes that form well-studied oceanic islands (such as Hawai'i, Réunion, Kerguelen and Galápagos, among others). New perspectives on the geochemical composition of the mantle have emerged from advances in analytical techniques and isotopic systems, enabling the resolution and quantification of small isotopic variations in OIB samples. These techniques and isotopic systems shed light on the presence, source and age of heterogeneities in the deep mantle.
In this Review, we discuss developments in the isotopic analysis of mantle plume-fed OIBs and their effectiveness as an essential tool for documenting mantle heterogeneity (Fig. 1b). We describe the fundamentals of mantle plumes, including their origin, physical and dynamic behaviour and their compositional variations, with an emphasis on contributions from distinct mantle reservoirs. We illustrate the utility of different isotopic systems and recent advances in their analysis for documenting the contributions made by subduction and crustal recycling to the plume source, identifying early-formed reservoirs in the mantle and their role in Earth evolution and quantifying the extent of convective mixing of mantle reservoirs. Finally, we explore the importance of buoyancy flux for understanding plumes, as well as the need for coordinated, interdisciplinary approaches in future mantle plume research.

Fundamentals of mantle plumes
The strength of upwelling flux, the composition and the duration of magmatic activity have been used to classify mantle plumes 14,16,17 (Fig. 1a and Box 1). Primary plumes come from between 1,000-km and 2,800-km depth (extending to the CMB) and are defined as having a seismic velocity anomaly of at least −1.5%, a linear and age-progressive volcanic chain, a large igneous province (LIP) at the oldest end of the volcanic chain, a strong magmatic flux and high 3 He/ 4 He or other indications of compositions distinct from those of MORBs 14,15,[18][19][20] . Primary plumes are the focus of this Review.

Introduction
The mantle has an important role in many of the most dynamic systems of the Earth. Comprising 84% of the volume of the Earth, the mantle is integral to the formation of mid-ocean ridges, convergent margins and intraplate volcanism, which, in turn, drive the formation and recycling of oceanic and continental crust [1][2][3][4][5][6] . The mantle also contributes to water and heat budgets of the Earth through radioactive decay and thermochemical convection. Seismic wave velocity variations reflect differences in temperature and/or composition in the lower mantle (660−2,890-km depth) [7][8][9][10][11] , suggesting that the lower mantle, or a part of it, is a reservoir for compositionally distinct material, including recycled and potentially ancient primitive components. Thus, understanding many fundamental Earth processes relies on documenting the composition and spatial heterogeneity of the mantle of the Earth today and its evolution throughout geological time.
At shallow mantle depths (<100 km), upwelling causes the upper mantle to melt through adiabatic decompression, leading to the generation of mid-ocean-ridge basalts (MORBs), an important source of information about the composition of the upper mantle 12 . By contrast, mantle plumes occur in all tectonic settings (Fig. 1a) and originate at different depths, from shallow upper-mantle upwellings (≪660 km) 13 to deep sources at the core-mantle boundary (CMB, ~2,800 km) 14 . Consequently, mantle plumes provide invaluable insight into the geochemistry of the deep interior of the Earth.
Once plumes rise above 1,000 km 15 , they form narrow, roughly cylindrical pipes of anomalously hot mantle that upwell and melt partially to form ocean island basalts (OIBs) on oceanic plates. Mantle plume melting occurs at sublithospheric depths, requiring a mantle source that is either hotter than the upper mantle, or enriched in rocks and/or fluids that lower the solidus temperature. Thus, mantle plumes provide the heat and fertile mantle components necessary for intraplate magmatism and to generate OIBs (Fig. 1b). Plume-generated oceanic islands usually sample the mantle without chemical interference from the continental crust or subcontinental lithospheric mantle, providing a geochemical window into the composition of the Earth's mantle and its evolution over time, including the sources of deeply rooted plumes.
Previous articles have discussed progress in seismic imaging, mantle flow modelling, plate tectonic reconstructions and geochemical analyses of deep mantle plumes and their roles in Earth processes 16 . ) on a tomography slice from the lowest mantle at 2,800-km depth, extracted from the SEMUCB-WM1 model 36 , generated using the SubMachine online tool 271 . Red, warmer, regions indicate the seismically slow Pacific and African-Atlantic large low-shear-velocity provinces (LLSVPs); blue, cooler, regions indicate the seismically faster parts of the lowermost mantle. The mean depth anomaly within a 500-km diameter circle around a mantle plume was calculated with the MiFil volume method, after filtering out small-scale features such as seamounts and islands 20 . The MiFil volume is a proxy for the bathymetric swell generated by the underlying mantle plume, which is assumed to correlate with the magnitude of its flux. The circle diameters correspond to the magnitude of buoyancy flux defined by the key on the right of the figure. The coloured circles correspond to plumes with geochemical data shown in Figs. [2][3][4][5] and Supplementary Figs. 1-9. The white circles indicate mantle plume sites not included in subsequent figures. The grey circles indicate mantle plumes without flux estimates. Note the geographic proximity of mantle plumes to LLSVPs and how mantle plume buoyancy flux varies globally by orders of magnitude. b, Schematic cross-section of the mantle illustrating major mantle structures and locations of potential chemical reservoirs, including a heterogeneous LLSVP (indicated by different shades of orange, including mantle plumes), ultra-low velocity zones (ULVZs) (red) and subducted oceanic lithosphere (blue) that transports recycled surface materials into the mantle (such as sediments shown in brown). The reservoirs are defined by isotopic and trace element data from plume-generated ocean island basalt. Mantle mineral assemblages are shown from shallower to deeper mantle depths and consist of olivine (Ol), pyroxene (Py), garnet (Gt), ringwoodite (Rw), bridgmanite (Bm) and ferropericlase (Fp). The insets associated with each reservoir list the isotopic systems that are used to identify and trace the reservoir (Supplementary Table 2) and the processes that can be examined using these isotopic systems on material associated with each reservoir (EM-I, enriched mantle I; EM-II, enriched mantle II; HIMU, high μ (U/Pb). The morphologies of the plumes, LLSVPs and ULVZs (whose thickness has been exaggerated to be visible at the scale of the diagram) are informed broadly by seismic studies 15,272 and geodynamic results 34,273 . The slab morphologies are guided roughly by seismic studies 145,274 and geodynamic results 31,32 . Mantle plume compositions reflect contributions from multiple mantle reservoirs, including recycled crustal material.

Review article
in the past decade focus on the interplay between the recycled crust and a layer of primordial iron-rich material with an excess density of about 1% [29][30][31][32] . Alternatively, dense and reduced mantle material could pool at the CMB, contributing to the formation of LLSVPs 33 . Laboratory experiments 34,35 and numerical simulations can reproduce the formation of large thermochemical plumes observed in tomographic models 36 . In some models, both the LLSVP and the ambient mantle supply material to plumes rising from the boundary between them 37 .

Review article
velocities, which are best explained by an ~10%-density increase in the material 22,38 . The ULVZs are 10-40 km high regions that extend laterally up to ~1,000 km along the CMB 39 . Models for the origin of ULVZs consider the effects of core-mantle reactions 40

Mantle plumes and large igneous provinces
Geodynamic models and analogue experiments indicate that when plumes begin to rise through the mantle, upwelling material takes the form of a mushroom, with a large, semicircular head trailed by a narrow tail 58,79,295,296 . Initial melt production is high, owing to the large volume of the plume head. When the plume head melts, it produces large igneous provinces (LIPs) on land (for example, Deccan and Siberian Traps), below sea level (such as Shatsky, Ontong−Java and Manihiki oceanic plateaus) or both (Kerguelen). This LIP phase is usually brief (1−3 Myr), but can last longer and involves large mantle volumes. Part a of the figure shows the minimum and maximum cross sections of relative mantle volumes required to partially melt to form several LIPs 297 . The LIP phase of plume activity is followed by a reduction in melt production, which yields oceanic island basalts (OIBs) sourced from the plume tail; this phase often lasts million years 69,70,253,298,299 .
The sheer size of plume heads requires a large volume of material from the lower mantle. Indeed, mantle plumes are associated with low seismic velocities, reflecting the elevated temperature of the ascending mantle material 18,102 . As a result of findings from seismic studies of the lower mantle 9,300-302 mantle plume models have evolved from the classic mushroom-shaped, thermal model toward more complex thermochemical structures. The internal dynamics of thermochemical plumes reflect the interplay between positive (thermal) and negative (compositional) buoyancy forces. The symmetrical geometry is lost 34,254,273 because some parts of the plume conduit sink 253 , whereas others ascend slowly.
The geochemical signatures of LIP phases reflect the complex relationship between compositional heterogeneity and the internal structures of mantle plumes. Components sampled by LIP melts include the plume material itself, upper mantle and even lithospheric material. Such signals are often more strongly observed in LIPs emplaced on continents than in submarine settings. Geochemical contributions to LIPs can be identified using alteration-resistant, so-called immobile elements, which include Th, Ti, Yb and Nb. During mantle melting, these elements behave similarly and predictably, making them useful as proxies both for common mechanisms that affect mantle-derived melts and for different mantle and/or crustal components 303 . Elevated Th/Nb ratios in LIP melts (see the figure, part b) reflect increased crustal contamination. The TiO 2 /Yb ratio increases with greater depth of melt generation (more melting in the garnet region of the mantle, at depths >80 km) [304][305][306] ; this ratio also increases as the partial melting extent decreases, a signature frequently observed in the OIB (plume tail) phase, when considerably less melt is being generated than during the LIP stage.
The Kerguelen mantle plume serves as an example of a hybrid LIP, consisting of continental flood basalts in the early stages (>120 Ma), followed by an oceanic plateau , then a plume trail , the Kerguelen Archipelago (30 Ma−now) and Heard Island 305 . Kerguelen lavas preserve signatures of multiple components whose origin varies from the lower mantle to the crust, providing a valuable cross-section of the mantle reservoirs supplying the Kerguelen plume. Kerguelen melts are predominantly within the oceanic field (red circles, see the figure, part b), overlapping with other LIP fields and extending slightly into the transitional area between MORB-dominant and OIB-dominant LIP melts 87 . A few older samples also overlap with the EM-OIB and lithospheric fields, the latter of which is interpreted as crustal contamination of LIP melts, a phenomenon commonly observed in continental settings 303 44,45 . The high density of ULVZs inhibits their convective stirring, and only the most vigorous plumes can entrain thin tendrils of the ULVZ material by viscous coupling 30 .
Mineral physics research has explored ways in which mantle heterogeneity can affect phase relations, density and elastic moduli through changes in bulk composition [46][47][48] , water content 49,50 and redox state 33,51 . Although accessory phases can vary in type and abundance, the dominant phase in the lower mantle is bridgmanite, an Mg-silicate perovskite 52 . The strength of the structure of bridgmanite could encourage the formation of narrow channels of upwelling and downwelling, thereby stabilizing deep-rooted mantle upwellings from near the CMB 53 .
Geochemical research on mantle plumes reveals complex spatial variations in the composition of OIBs (Fig. 2). The best-studied example is Hawai'i, where the distinct isotopic compositions of Loa-trend and Kea-trend volcanoes indicate a bilateral compositional asymmetry, or compositional stripes in the plume 54 . Specifically, the southwestern Loa chain consists of more enriched material, and the northeastern Kea chain has more depleted signatures. It has been proposed that the Loa chain is sourced in the Pacific LLSVP and Kea lavas are supplied by the ambient lower mantle [55][56][57] , which reflects the location of the plume along the northern boundary of the Pacific LLSVP. The compositionally distinct materials from the LLSVP and the lower mantle only mix minimally between the CMB and the surface, where they are expressed as distinct signatures in erupted lavas 55,58,59 . Similar bilateral compositional patterns have been recognized in other LLSVP-sourced plumes, including the Samoan, Galápagos, Easter and Marquesas Islands in the Pacific 55,60-64 and Tristan−Gough in the Atlantic 65 . This proposed origin model for plumes 55 has led to additional nuances and suggestions for models. In the Marquesas, the sizes of compositional heterogeneities are much smaller than in Hawai'i and are generated from low degree partial melting plumelets that originate from a large dome structure in the mantle, under Polynesia 61 . Rather than originating at the LLSVP−lower mantle interface, a drifting Hawaiian plume could have become anchored on the LLSVP, resulting in the entrainment of the LLSVP material after that event 66 . This model would explain the abrupt increase in plume flux along the Hawaiian chain after the bend in its path. The compositional zonation in Hawai'i could also reflect a major change in absolute plate motion 67 .

Buoyancy flux of plumes
The strength, or buoyancy flux, of a mantle plume influences the partial melting regime of its corresponding oceanic island system 68 . Tholeiitic and alkalic basalts are produced by higher or lower degrees of partial melting, respectively. Thus, whether tholeiitic or alkalic basalts dominate an oceanic island provides insight into the buoyancy flux and the mantle potential temperature of a plume system. For example, the Hawaiian chain represents a rare occurrence of high-volume magmatism that produces tholeiitic basalts over 85 Myr. Other examples include Réunion, Iceland, Galápagos and Easter Islands 20,69 , systems that also have elevated buoyancy flux and/or mantle potential temperature 68,[70][71][72][73][74] . Iceland is the only oceanic island that follows a silica-saturated tholeiitic evolution trend 12 ( Supplementary Fig. 1). Alkalic lavas, formed from lower degrees of partial melting, dominate some plumes with intermediate fluxes, such as the Cook−Austral chain 75 and Society Islands 76 (Fig. 1a). Generally, however, most intraplate oceanic volcanoes produce dominantly alkalic melts 77 , suggesting that high degrees of partial melting might be uncommon in some mantle plumes, at least during the Cenozoic.
Furthermore, plume strength could be increased by proximity to an LLSVP, as proposed for Hawai'i 66,78 . The volume of magma produced along the Hawaiian chain has increased over the past 45 Myr, which contrasts with the decrease through time according to mantle plume models [79][80][81] . Moreover, the highest-flux plumes, including Hawai'i, overlie mega-ULVZ structures at the CMB 39 , implying that lower-mantle structures, the strength of a plume, its temperature and its melting conditions could all be linked 66 .

Influence of plate motion and plate boundaries
The compositional structure of a plume is recorded by its volcanoes in different ways depending on the lithospheric thickness, proximity to a spreading ridge, plate motion vectors and the geometry of the compositional structure within the plume conduit itself. Lithospheric thickness can influence the depth and degree of melting, spreading ridges can re-direct upper-mantle flow and the direction of plate motion relative to the compositional cross-section of the plume determines how plume compositions are expressed at the surface. For example, since ~3−6 Ma at Hawai'i, both the LLSVP boundary and current Pacific plate motion vector 82 have been oriented along NW-trending strikes, which yield two parallel chains of compositionally distinct volcanoes 55,83 . By contrast, in the Galápagos, the orientation of the LLSVP boundary is oblique to the motion of the plate, resulting in early-erupted, enriched material from the LLSVP being buried beneath younger, more depleted material from the ambient mantle 64 . Additionally, shifts in plate motion direction could shear mantle plume stems and generate ephemeral changes in the spatial distribution of the erupted, compositionally distinct material 67 .
When plumes interact with adjacent spreading centres, plume material can be incorporated into the return flow of the ridge, which increases the entrainment of depleted upper mantle into plumegenerated lavas as observed in the Azores, Galápagos and Cretaceousage Hawaiian eruptions, among other near-ridge plumes [84][85][86] . The near-zero lithospheric thickness of ridge-proximal plumes could also increase plume melting and affect the magmatic architecture of ocean islands formed near the ridge [87][88][89] . Near-ridge plumes, including Iceland, Galápagos, Kerguelen and Easter, produce abundant tholeiitic lavas that require higher degrees of partial melting than other mantle plumes 68,71,72,84,87,90 . In plumes near subduction zones, additional flexure of the plate could generate melt and/or encourage entrainment of the upper-mantle material, as proposed for Samoa 91,92 and for volcanism that cannot be plume-related in the Emperor Seamounts near the Aleutian Arc 93 .

Oceanic island volcanism
Owing to the lack of interaction with continental material, ocean islands with deep plume sources provide the most direct view and best representation of lower-mantle compositions 66,81 .

A geochemical overview of OIBs
Oceanic island volcanoes feature various volcanic rock types, including picro-basalts, rhyolites, basanites, phonolites and trachytes (Supplementary Fig. 1). Nevertheless, the dominant rock type is basalt. This Review focuses on shield-stage basaltic magmas that form lavas of tholeiitic or alkalic compositions because they have experienced the least differentiation and contamination en route to the surface. Another approach to studying mantle heterogeneity is to analyse melt inclusions, which enables the detection of smaller isotope heterogeneities than is possible with the bulk analysis of erupted lavas 94

Review article
capturing different aspects of the magmatic system than either whole-rock or mineral-phase analysis. They represent a snapshot of melt packets that existed before melt amalgamation, magma mixing and substantial fractional crystallization. Melt inclusion analyses are analytically challenging and vulnerable to many processes that disturb or distort the composition of the post-entrapment melts, generating data less representative than whole-rock or mineral-phase analysis because of the small size of the inclusions. Therefore, this Review focuses exclusively on isotopic analyses of rock samples that integrate the composition of melts generated by large mantle volumes (many cubic kilometres) 96 (Box 2). The compositional diversity of OIBs (especially compared with the relative homogeneity of MORBs 3,12 ) reflects the lower degrees of partial melting of OIBs compared with MORBs. However, melting systematics can only explain a small fraction of the observed variations 3,97 . Characteristic features of major and trace element compositional variations in OIB include enrichment in the light rare-earth elements (REEs) compared with the primitive mantle (which refers to the composition of the Earth's mantle soon after core formation), depletion in heavy REEs relative to middle and light REEs and near-constant ratios of the most incompatible elements (the left side of Supplementary  Fig. 2 (refs. 98-100)). Incompatible elements are those whose charge or ionic radius is too large to substitute into mantle minerals; some of the most incompatible elements in mantle lithologies include Ce, Nb, Zr, U, Th, Ta, La, Sm and Pb. The REE patterns of OIBs suggest that most OIB melts are generated primarily in the garnet stability field (>80-km depth) 101 , which reflects the higher (100−200°C) mantle temperatures of plumes 102 . By contrast, most MORBs begin melting at shallower levels where garnet is no longer stable (<80 km), inducing no substantial middle-to-heavy REE fractionation in the generated magmas.
In some cases, the concentrations of major element oxides and trace elements correlate with the dominant isotopic trends displayed by OIBs 73,103 . Variations in major and trace elements are typically generated by different processes (such as the degree of melting and magmatic differentiation) from those that produce variations in isotopic compositions (for example, compositional differences in time-integrated mantle sources). Thus, correlations between major and trace element concentrations and isotopic compositions in OIB might reflect mixtures of compositionally distinct sources melting to variable extents and/or the fact that the compositionally distinct materials in the mantle differ in their major and trace element signatures as well as their isotopic compositions 96 .
The five compositional endmembers define the known range of isotopic compositions in OIB and thus in the mantle sampled by plumes. Except for PREMA, these endmembers are chemically and isotopically distinct from each other and are thought to come from separate chemical reservoirs in the mantle, resulting from processes generating unique chemical signatures that evolved over time. This section considers the geochemical signatures of these mantle reservoirs (Supplementary Table 1), their origins and the insights they provide into the composition and evolution of the mantle.
The HIMU endmember. The HIMU mantle endmember is named for its high U/Pb (commonly referred as μ) source ratio 2 . The extreme U/Pb ratios required to generate HIMU Pb isotopic signatures (Fig. 2a) indicate that the original mantle source had high initial U/Pb (μ) and Th/Pb (ω) 109 . The mechanism that is most often used to explain the removal of Pb required for the HIMU signature to occur is the dehydration of ancient, subducted oceanic crust, which is then stored in the mantle for 1.5−2 Gyr before being incorporated into mantle plumes 1,2,106,109-112 . Alternative explanations include the addition to the plume source of carbonatitic silicate melts from recycled oceanic crust 73,113 , ancient carbonate-metasomatized sub-continental lithospheric mantle 114,115 or recycled Archean marine carbonates 116 . Of the major mantle endmembers sampled by plumes, HIMU has a limited distribution, is the least voluminous (Figs. 1 and 3a,b) and is the dominant component in only a few small ocean islands mostly in the Pacific ocean (such as Cook−Austral, St. Helena, Mangai, Tubuai and Rurutu).

Enriched mantle endmembers.
On the basis of OIB isotopic systematics, two distinct enriched mantle endmembers (EM-I and EM-II) can be identified 2 . There is little consensus on the origin of the EM-I endmember 117,118 ; however, hypotheses include the subduction and recycling of various near-surficial and surficial materials 109,110 , including delaminated subcontinental lithosphere 119 , depleted MORB material 120 and/or pelagic sediments 121 or Archean carbonates [122][123][124] , which are incorporated into the deep mantle. By contrast, the EM-II reservoir exhibits contributions from continental crust materials 2,73,109,125-127 , although alternative models have been proposed, including the incorporation of ancient metasomatized oceanic lithosphere into the EM-II reservoir 128

An introduction to isotopes and isotopic analysis
Isotopes of a chemical element have the same number of protons in their nucleus but differ in their number of neutrons. The relative abundance of one isotope to another is measured by a mass spectrometer. Isotopes can be classified on the basis of their stability and formation pathways.

Radiogenic isotopes
Radioactive isotopes have unstable atomic nuclei that emit radiation and spontaneously transform into another isotope, the radiogenic product. Their decay rate is characterized by its half-life (T 1/2 ), which is the time for half of a given amount of the parent isotope to decay. Time is primarily responsible for differences in radiogenic isotopic compositions in geological settings. Isotopes with T 1/2 > 100 Myr, such as 87 Rb and 147 Sm, are still decaying today (see the figure). The resulting variations in the abundances of daughter products take millions to billions of years to develop and are measured relative to a stable isotope of the same element ( 87 Sr/ 86 Sr and 143 Nd/ 144 Nd) using mass spectrometry (Box 3). Such long-lived isotopic systems are used for geochronology and fingerprinting sources of materials. Conversely, radioactive isotopes with T 1/2 < 100 Myr have now fully decayed and are extinct, providing information about events that happened early in the history of the Earth.
Neodymium provides an example of the differences between short and long half-life isotope systems. Neodymium has two radiogenic isotopes ( 142 Nd and 143 Nd) produced by the decay of two radioactive isotopes of Sm ( 146 Sm and 147 Sm) with very different half-lives. Therefore, the two isotopic systems display isotopic variations that reflect distinct processes: the 146 Sm− 142 Nd system tracks the presence of early Earth material and the 147 Sm− 143 Nd system documents long-term source changes. Although the 147 Sm− 143 Nd system has been analytically accessible since the 1980s, the variations in 142 Nd caused by 146 Sm decay are so small that measuring 142 Nd/ 144 Nd is almost at the limit of analytical capabilities of modern mass spectrometers and has only been achievable since 2000.
Unlike trace elements (defined as elements that are not structural components of major mantle minerals and have abundances < 0.1 wt%), radiogenic isotopic compositions are not affected by the degree of partial melting or the crystallization history of a melt. Therefore, radiogenic isotopic ratios (such as 87 Sr/ 86 Sr, 143 Nd/ 144 Nd, 206 Pb/ 204 Pb, 4 He/ 3 He or 176 Hf/ 177 Hf) of a basalt represent the composition of the material that melted to produce the basalt, providing time-integrated information about its origin. By contrast, variations of extinct radiogenic isotopic systems (such as 142 Nd/ 144 Nd, 182 W/ 184 W and 129 Xe/ 130 Xe) indicate the presence of material that differentiated from the rest of the mantle during accretion.

Stable isotopes
Isotopes with a stable nucleus have invariant isotopic abundances. Some of these isotopes, especially light elements, can be fractionated by physical, biological and sometimes chemical processes and are used to trace fractionation processes rather than the long-term evolution of the lava source. Examples include atmospheric weathering (O), groundwater interaction (H, O and C), sediment recycling (Li and Tl) and redox changes (Fe).

Noble gases
All noble gases have radiogenic and stable isotopes. For many radioactive systems (Supplementary Table 2), parent-daughter fractionation by gas loss within the lifetime of the radioactive isotope generates radiogenic noble gas isotope signatures over time. For example, 3 He is primordial, because it is a stable, non-radiogenic isotope whose abundance was established during accretion; any primordial isotopes lost to the atmosphere are not replaced in the noble gas budget of the Earth. Primordial noble gas isotopes are no different from the stable, non-radiogenic normalizing isotopes used in other decay systems (for example, 204 Pb). Accordingly, the less commonly used notation 4 He/ 3 He is consistent with radiogenic isotope convention, but helium isotopic ratios are usually reported as the inverse, 3 He/ 4 He. In contrast to other isotopic systems, however, 4 He is continuously produced by radioactive decay of 235 U, 238 U and 232 Th. Unradiogenic helium isotopic ratios sample reservoirs that have experienced less degassing (transport of gas from the mantle to the atmosphere associated with mantle processing by partial melting and volcanism). A less-degassed mantle reservoir has relatively high 3 He/ 4 He (or low 4 He/ 3 He) because it has retained more of its primordial He, and the impact of radiogenic ingrowth of 4 He is muted compared with the rest of the mantle. Furthermore, a less-degassed reservoir is not necessarily primordial mantle. Regassing (transport of atmospheric gases into the mantle via subducted slabs) of He is negligible as it is light enough to escape to the atmosphere. Atmospheric contamination during or after sample formation affects Ne, Ar, Kr and Xe isotopic compositions measured in all ocean island basalts. Corrections are needed to determine the mantle source composition and to assess whether regassing has affected the mantle being sampled in any given analysis.  Long-lived radioactive isotope (T 1/2 > 10 8 years) Short-lived extinct radioactive isotope (T 1/2 < 10 8 years)

Review article
The prevalent mantle. Most OIB isotopic arrays, especially major mantle plumes, have one endmember anchored at the centre of the global isotopic OIB data field, the intermediate area defined as PREMA 2 . The PREMA reservoir represents the average ambient lower mantle (Figs. 2 and 3b) and is compositionally distinct from the upper mantle. The high 3 He/ 4 He ratios relative to MORB 107 suggest that PREMA samples a less-degassed reservoir than other parts of the mantle 129,130 . PREMA might represent primitive mantle if that reservoir has non-chondritic Sm/Nd, Nb/U and Ce/Pb 6,56,131 . Alternatively, the PREMA reservoir could be a well-mixed combination of a less-processed, less-degassed mantle with recycled components 2,132,133 . Hawai'i, Iceland and the Galápagos all exhibit substantial PREMA contributions 56,64 .
The depleted mantle source in OIBs. The major and trace element characteristics of strongly depleted OIB lavas (Supplementary Figs. [1][2][3][4][5] are consistent with a mantle source that has a multimillion-year-long history of melt extraction and is therefore depleted in the most incompatible trace elements 134 . Although the depleted component in OIBs is easily masked by mixing with melts from enriched heterogeneities in the mantle source, analysis of melt inclusions that record contributions from depleted melts before magma mixing and homogenization shows that it is a component of some Azores lavas 135 . Rare, depleted signatures in Hawaiian lavas and xenoliths also support a depleted component in mantle plumes [136][137][138] . In Iceland, the Azores and the Galápagos, the depleted signature is attributed to their near-ridge settings, in which higher degrees of mantle melting occur under a thin lithosphere 64,104,[139][140][141][142] .
Uncertainty persists regarding the origin of depleted mantle compositions in OIB 86,104,143 , leading to several potential hypotheses. First, these compositions could comprise a large proportion of the lower mantle, meaning that the DMM or an early-enriched reservoir could be a major matrix component in mantle plumes 131 . Second, they could represent the melting of an entrained sheath of depleted upper-mantle material. Finally, they could reflect large degrees of melting of the plume source, as expected underneath thin lithosphere. By contrast, multivariate statistical data analysis suggests that the apparent overlap of MORB-OIB data trends in 2D isotope ratio diagrams does not translate consistently to multidimensional isotope data space 143 . Therefore, the compositional variations displayed by MORB-OIB could be controlled by smaller-scale, regional domains, rather than a limited number of global-scale reservoirs 143 . Such major differences in the interpreted magnitude and dynamics of compositional reservoirs highlight the need for a more nuanced understanding of these important parts of the mantle and their roles in mantle plumes.

Subduction and mantle heterogeneity
Early in the development of mantle plume theory, plumes were thought to be supplied exclusively by the primitive lower mantle and depleted upper mantle 144 . On the basis of geochemical data, the idea that the source of OIBs might contain ancient oceanic crust was proposed in the 1980s (ref. 110). This hypothesis posits that oceanic plates are recycled via subduction into the lower mantle, in which they are subsequently sampled by plumes; this proposed cycle of downwelling and upwelling provided an important step forward in our understanding of mantle plumes and their role in the global plate tectonic system. Geochemical evidence acquired since the hypothesis was proposed has confirmed that recycled oceanic crust contributes to ocean island mantle sources 1,2,106,109,111,112 . Consistent with the recycling model, seismic data 145 and geodynamic models 32 document the transport of subducted plates to the lower mantle.

Recycled material in OIBs
The recycled oceanic crust consists of various materials, including basaltic crust and the overlying sedimentary pile (Fig. 1b). Although basaltic crust has a chemical composition that is, within an enrichment factor, not substantially different from that of the upper-mantle source, the composition of the sedimentary material can be highly diverse. The average composition for global subducting sediment 146 is similar to that of continental crust but is chemically distinct from the mantle or basaltic material. Thus, even minuscule contributions of sediment to the source of OIBs can be detected using geochemical methods. Atmospheric and spallation-generated 10 Be concentrations, Li, B and Fe isotope compositions as well as U-series disequilibrium measurements demonstrate that sediments survive subduction processing to mantle depths [147][148][149][150][151] , confirming that the inclusion of recycled sediments in the mantle plume source is a viable model to account for some of the heterogeneities. Consequently, to understand mantle plumes and their Nd/ 144 Nd in ppm deviation from the terrestrial standard ( JNdi-1, AMES or νNd-b) measured in ocean island basalt samples 103,194,202,[284][285][286][287][288][289][290] . Baffin Island samples are included because they have the highest 3 He/ 4 He ratios measured to date 291 . The error bars indicate 2SE (standard error when the sample was measured once) or 2SD (standard deviation when samples were measured several times), and the number of measurements is indicated near the symbol. External reproducibility (2SD) is estimated by repeatedly measuring a standard reference material during the same analytical session as the samples (light grey, ~5 ppm; dark grey 1.1 ppm based on 10 measurements of JNdi-1 (ref. 290)). Red outlines indicate values with an SD that is significant compared with external reproducibility (Réunion 202 and Samoa 194 ). b, As in part a but for μ 182 W (μ 182 W is 182 W/ 184 W measured in ocean island basalt (OIB) samples and reported in ppm deviation relative to terrestrial standard, Alfa Aesar [204][205][206][207] ). External reproducibility is ~4−5 ppm; large deficits in OIB samples can be resolved clearly. Samples from Baffin Island are not included because 182 W results are controversial; the positive μ 182 W could be an analytical artefact 292,293 . Small variations in extinct isotopic systems represent either potential contributions from early-formed material or core interaction with the plume source. c, 129 Xe/ 130 Xe in mantle sources for Iceland 160 , the Samoa Rochambeau Rift 169 and Galápagos (accumulated gas 163 ). Error bars are 1SD following convention in the noble gas literature. The depleted mantle source (solid grey line) is 129 Xe/ 130 Xe = 7.8 for mid-ocean-ridge basalt (MORB) and well gases 159,161,162,208 . The atmospheric source (dashed grey line) is 129 Xe/ 130 Xe = 6.496 (ref. 294), which is distinctly lower than 129 Xe/ 130 Xe determined for OIB and MORB mantle sources. Noble gas systematics highlight the compositional differences between the MORB and OIB mantle sources. d, Step-crush data with 1SD error bars are shown for Iceland, Samoa and North Atlantic MORB 160,169,208,210 . Neither post-eruptive atmospheric contaminated mantle (illustrated by the trends toward atmospheric values in the step-crush data) nor incorporation of regassed atmospheric Xe into the mantle can explain the OIB data arrays in 129 Xe/ 130 Xe− 3 He/ 130 Xe space. The low 129 Xe/ 130 Xe values in OIB mantle sources are thus interpreted to reflect a lower I/Xe ratio during the lifetime of 129 I compared with the upper mantle 208,210 . Short-lived isotopic decay products indicate OIB sample early-formed reservoirs and provide insight into early Earth evolution processes. Concerted efforts to measure μ 142 Nd, μ 182 W and Xe isotopes in the same samples are needed to better understand what processes generated these early-formed signatures and how they have been preserved in the mantle.

Review article
sources fully, it is essential to document the geochemical composition of the subducted oceanic crust and the associated sediments that are transported into the lower mantle.
As sediments and oceanic crust are subducted into the mantle, they undergo distinct compositional changes. Specifically, the subducted material is dehydrated or melts and it loses a large proportion of the fluid-mobile elements (such as Li, B and Cs), although retaining refractory elements such as REE and high field strength elements [152][153][154] . These changes enable the identification and quantification of the contributions of recycled oceanic crust and sediments to mantle sources 152,155 .
The presence of sediments in the OIB source could explain some of the specific trace element characteristics of EM-I and EM-II basalts. The two canonical ratios, Ce/Pb and Nb/U, generally have constant values in all mantle-derived magmas (25 ± 5 for Ce/Pb and 47 ± 10 for Nb/U 98,131 ). In EM-I and EM-II OIBs ( Supplementary Fig. 4), these ratios deviate from their universal mantle values towards lower values typical of continental crust (3.7 for Ce/Pb and 4.4 for Nb/U 109,118,126,127,156 ), providing some constraints on the origin of isotopic enrichment in the OIB source.

Characterization of recycled material in the mantle
Radiogenic isotopes are more effective than trace elements for detecting and quantifying sedimentary material in OIB sources. For example, 6% of sediment in the mantle source shifts the Sr isotopic composition of Samoan basalts from 0.704 to ~0.720; such high values have not been measured in any other OIB to date 127 . Most EM-II basalts have much less radiogenic Sr isotope ratios than Samoan lavas, corresponding to a sediment contribution of <2% in their sources. Nevertheless, most OIBs with radiogenic Sr coupled to unradiogenic Nd and Hf must originate from a source that contains some recycled sedimentary material 157 .
The most common model used to explain the coupling of unradiogenic Sr isotopic ratios with highly radiogenic Pb isotopic signatures observed in HIMU basalts ( Fig. 2d and Supplementary Fig. 9) is the presence of old (~2 Ga) oceanic crust in the source that has lost some of its Pb during subduction. This interpretation is supported by the elevated Ce/Pb of most HIMU basalts (typically >30 (ref. 118)), which reflects the loss of Pb to subduction fluids. Furthermore, the relationship between Nd and Hf isotopic ratios in OIBs ( Fig. 2c and Supplementary Fig. 8

Box 3
Analytical precision in determination of radiogenic isotopic ratios 142 Nd/ 144 Nd and 182 W/ 184 W ratios are mainly determined using thermal-ionization mass spectrometry (TIMS) techniques that allow high-precision analysis, but they also require considerable time and effort to achieve successfully. Analytical artefacts can produce small isotopic variations in TIMS measurements, which can be linked to factors such as isotopic ratio fractionations not following the exponential law 307 , the rate of mass fractionation being too high for dynamic measurements 308 , mixing of different sample reservoirs on the filaments 307,308 , non-mass-dependent isotopic variations caused by the chemical separation protocol 287,289,293 and the need to improve the determination of O isotope composition for TIMS measurements of W isotopes in the oxide form 232 . Deviations in the 142 Nd/ 144 Nd and 182 W/ 184 W systems are small; therefore, they are expressed in μ-notation (deviation in ppm relative to the average ratio measured in the terrestrial standard).
High-precision isotopic ratio measurements have revealed small differences between the terrestrial standards commonly used in the different laboratories 287,309 , as exemplified by the variation of a few ppm in 142 Nd/ 144 Nd ratios measured in LaJolla, JNdi-1 and AMES Nd. A multimass-step acquisition scheme enables all isotopic ratios of the element to be determined in the dynamic mode, with subsequent application of quality-control criteria 289 .
When publishing isotopic data, analytical uncertainty is reported as 2SD (standard deviation, calculated from repeated measurements of the same sample) or 2SE (standard error, in which 2SE = 2SD√N, where N is the number of measurements). The SE corresponds to the internal error when the sample is measured once. The scientific community has debated the significance of a deviation relative to its analytical precision extensively, without a clear resolution. Regardless of the choice of 2SD or 2SE, publications claiming high-precision isotopic measurements should thoroughly describe the analytical protocols used and all data associated with the measurement and should also present the results for international, cross-calibrated reference standard materials. A rigorous analysis requires that an appropriate number of duplicate, replicate and total blank measurements be performed and their results reported in the publication.
To achieve high-precision isotopic OIB measurements, it is important to measure only the original, magmatic composition of the basalt, which is achieved primarily through appropriate sample preparation procedures. For example, Rb−Sr, U−Pb and stable Li isotopic systems are susceptible to perturbation by seawater alteration, and Tl and Pb isotopic measurements are sensitive to the presence of ferromanganese precipitation. These secondary products must be removed by physical separation and cleaning followed by careful, systematic acid leaching procedures 183,[310][311][312][313][314][315][316][317][318] . Finally, older OIB samples (>7−10 Ma) must be age-corrected for in situ decay since their eruption 66,78 . Age correction requires measurements of the elemental parent−daughter ratio of the unleached sample. For the U−Th−Pb system in particular, care must be taken to estimate the primary U concentration in submarine OIB properly, as U is susceptible to secondary alteration. For example, primary U can be estimated using the Th/U of unaltered samples 66,319 .
In OIB studies, it is difficult to compare data on the same samples when measured using different instruments, methods and/or analytical laboratories. These comparisons require geochemists to apply a normalization scheme on the basis of published values for standards and reference materials 55,81 . It is essential that the same standard and reference values are used; otherwise, spurious correlations can occur.

Review article
from recycled basalts and sediments with ages ranging from Archean to present day 121,155,158 .
Heavy noble gas (Ne, Ar, Kr and Xe) isotopic ratios in mantle rocks are also sensitive tracers of the subducted material. Atmospheric heavy noble gases, which have distinctive elemental ratios and isotopic compositions, are incorporated into the mantle through subduction [159][160][161][162][163][164] ; however, this regassing transport is likely to have been inefficient before 2.5 Ga 165,166 . Once they have been corrected to account for shallow post-eruptive atmospheric contamination, the Xe isotopic signatures of OIB mantle sources are dominated by regassed atmospheric Xe 133 . This observation clearly shows that there is a surface-derived material in the OIB mantle source because the atmospheric Xe isotopic signature reflects sources and processes that are distinct from mantle Xe 167,168 . Likewise, Kr isotopes in Iceland and Galápagos samples indicate strong regassing of atmospheric Kr 163 . Regassed atmospheric signatures are present even in samples with high 3 He/ 4 He and primitive, solar-like Ne isotopes 160,163,169 .
The oceanic crust also carries distinctive signatures from the surface into the mantle. For example, Li is incorporated into oceanic crust during hydrothermal alteration and serpentinization. Of the stable isotope systems commonly used in the geosciences, Li has the largest relative mass difference between its two isotopes ( 6 Li and 7 Li). As such, low-temperature, aqueous processes result in extreme fractionation 150,170,171 and, consequently, Li isotopes can trace processes that involve fluid mobilization, such as subduction (Fig. 2e,f). During the subduction process, Li is transported to the mantle wedge as oceanic slabs are dewatered and metamorphosed 149,172 . Variations in Li isotopic values (δ 7 Li > 4; Li isotopes are measured as the ratio of 7 Li to 6 Li, which is normalized to the ratio of a NIST Li carbonate standard; that value is then scaled by a factor of 1,000 to yield the δ 7 Li value) consistent with subduction alteration have been measured in some OIBs from Hawai'i, Cook−Austral, St Helena and Azores 173-176 , providing evidence for the recycling of subduction-altered material into the OIB source. By contrast, MORBs exhibit a relatively uniform Li isotopic composition 171,177 (δ 7 Li = 3.5 ± 1.0‰). There is also a measurable difference between Li isotopic compositions of HIMU OIB (δ 7 Li = 2.5−8.5‰) and EM-I OIB (δ 7 Li = 0.5−4.5‰), which reflects the diversity of subducted components that contributes to these mantle reservoirs 175,176 .
Isotopic systems with large natural fractionations can provide good estimates of the type of the subducted surface material present in OIB sources. For example, stable isotopes of thallium ( 205 Tl and 203 Tl) are fractionated up to 35 epsilon units (ε 205 Tl is the deviation in parts per ten thousand of 205 Tl/ 203 Tl relative to a reference value) and exhibit large concentration contrasts between geochemical reservoirs. Pelagic sediments have ε 205 Tl and Tl concentrations [Tl] up to +5 and 1,000 ng g −1 , respectively 178 ; low-temperature altered oceanic crust has ε 205 184 . In some cases, however, the Tl isotopic composition of OIB is ambiguous, requiring careful consideration when interpreting results and using this isotopic system as a mantle tracer 181,183 .
The subducted material also carries information about previous redox conditions at the surface of the Earth. In specific circumstances, such as the extreme surface conditions during the Archean before the great oxygenation event at ~2.5−2.4 Ga 185 , mass-independent fractionation of stable isotopes such as S can occur, creating isotopic anomalies. These signatures are then incorporated into sedimentary materials and basalts at the surface. For example, negative S isotopic anomalies have been detected in lavas from Mangaia 186,187 and in basalts from Pitcairn 123 . These observations indicate that the HIMU (Mangaia) and EM-I (Pitcairn) mantle sources include material that was present at the surface of the Earth during the Archean, either in the form of Archean basaltic crust for Mangaia or in the form of Archean sedimentary material for Pitcairn.
Isotopic tracers that enable the characterization of recycled material in the OIB mantle source continue to be developed. For example, Ce is redox-sensitive 188 . Therefore, measurements of the long-lived 138 La− 138 Ce system can constrain the timing of pelagic sediment recycling into the mantle because no Ce anomalies are expected in water columns before the great oxygenation event at ~2.5−2.4 Ga 189,190 . Collectively, many isotopic and trace element systems (Box 2, Fig. 1b and Supplementary Figs. [2][3][4][5] indicate that material previously at the surface of the Earth is present in the source of OIBs. However, not all isotopic characteristics of OIB can be explained by recycling surficial material into the plume source alone. Additional types of material must be present and the influence of several large-scale processes must also be considered, including the physical proximity of plumes to LLSVPs, the depth and source of plume magmatism and core−mantle interactions.

Early-formed reservoirs
The OIB mantle source is a heterogeneous mixture of recycled surface materials, deep mantle material that separated early (>4.45 Ga) and experienced less degassing than the upper mantle, and potentially distinct remnant materials that formed early in history of the Earth. Determining the age and the nature of early-formed materials is important for understanding the structure and dynamics of the lower mantle, especially given the spatial correlation of major mantle plumes with LLSVPs and ULVZs 21,23,25,191,192 .

Tracing early-formed reservoirs
Processes such as metal-silicate differentiation, magma ocean crystallization and degassing of volatiles during energetic accretionary impacts influenced the geochemistry of the interior of the early Earth by fractionating groups of elements with properties different from each other. Because these events only occurred for relatively short periods close to 4.5 Ga, the best tools to understand them and track their contributions to the OIB mantle are short-lived radionuclide systems, such as 182 Hf− 182 W, 146 Sm− 142 Nd, 129 I− 129 Xe and 244 Pu− 131−136 Xe, with half-lives between <10 Ma and ~100 Ma. Variations in the radiogenic products of some short-lived isotope systems are small (such as μ 142 Nd ~ 15 ppm and μ 182 W ~ 30 ppm, where the notation indicates deviations in parts per million relative to the terrestrial reference), making them difficult to detect 193,194 (Fig. 4 and Box 3). Volatile elements pose additional challenges, owing to the loss of magmatic gas from samples and atmospheric contamination during and after eruption. Nevertheless, progress has been made measuring μ 142 Nd, μ 182 W and 129 Xe− 130 Xe variations in OIBs, as well as interpreting the importance of these anomalies for mantle history and dynamics.
Trace element abundances and their ratios are also powerful tools for quantifying fractionation processes in the early Earth. For example, mass balance calculations using (Nb, Ta)/U and 143 Nd/ 144 Nd

Review article
isotope ratios 131 demonstrate that continental crust and present-day depleted mantle could not have originated from the primitive mantle as previously thought [195][196][197] . This conclusion is supported by the delicate measurements of a radiogenic excess of 142 Nd (+7.9 ± 1.9 ppm) in the mantle relative to its building blocks 198,199 . Therefore, the mantle of the Earth could be slightly depleted in incompatible elements and characterized by higher Sm/Nd than chondrites. The Sm/Nd fractionation could be inherited from the accretion stage when protocrust of the Earth, enriched in incompatible elements and formed in planetesimals, was lost to space during collisional events 198 , removing the need for an early-enriched reservoir to have been preserved in the deep mantle 200 .
Many stable isotope ratios of major elements can preserve early-formed isotopic signatures because their values were not reset on a global scale (Box 2). Stable isotope fractionation has been constrained through high pressure-temperature experiments that simulate early differentiation events 201 to predict the isotopic signatures in mantle sources and determine stable isotope fractionation factors that enable distinguishing potential stable isotopic heterogeneities of early-formed reservoirs.

Terrestrial magma ocean relics
Small 142 Nd anomalies measured in some OIBs (−8 to +6 ppm; Fig. 4a) from Réunion 202 and Samoa 194 suggest that early-formed reservoirs might be preserved in the deepest part of the mantle. Owing to the short half-life of 146 Sm (~103 Ma), the variations measured in 142 Nd/ 144 Nd must reflect Sm/Nd fractionation that took place during the first few hundred million years of Earth history. Both Sm and Nd are lithophile REEs and were excluded from the metal phase during core−mantle differentiation 203 ; therefore, early fractionation of Sm/Nd must have occurred exclusively through silicate differentiation, providing evidence for crystallization of a terrestrial magma ocean. Most OIBs are also characterized by negative 182 W anomalies (μ 182 W down to −25 ppm) [204][205][206] , which could also represent the remnant of an early terrestrial magma ocean (Fig. 4a). However, core−mantle interaction could affect the 182 W signal, making it difficult to interpret 44,207 .
The short-lived 129 I− 129 Xe system provides additional support for the preservation of early-formed isotopic heterogeneity (measured as 129 Xe/ 130 Xe ; 129 Xe is produced by the decay of short-lived 129 I, and 130 Xe is not radiogenic). Mantle Xe isotope compositions can be broken down into component contributions from accretion (chondritic Xe), radioactive decay and atmospheric regassing 133,167 . Limited measurements of mantle-derived samples from OIBs, MORBs [160][161][162][163]166,[208][209][210] and back-arc basin basalts 169 as well as volcanic and continental well gases 211,212 yield precise estimates of mantle source Xe compositions, corrected for shallow atmospheric contamination. Ratios of 129 Xe/ 130 Xe from Iceland 160 , the Samoan Rochambeau Rift 169 and Galápagos 163 are low compared with those of the depleted mantle 159,161,162 . Differential incorporation of atmospheric Xe with low 129 Xe/ 130 Xe into the OIB mantle cannot account for these OIB 129 Xe/ 130 Xe signatures (Fig. 4c,d). Thus, a low I/Xe ratio must have been established in the OIB mantle within the first ~100 Myr of Earth history, and its signature preserved despite ~4. 45 Gyr of convection 160 . The low I/Xe could reflect inefficient degassing of the deeper parts of the magma ocean, or low I abundances in the early-accreted materials.
The paired I−Pu−Xe decay system provides additional insight into early magma ocean history. Initially, catastrophic outgassing would have transported Xe out of the terrestrial magma ocean. After closure, the products of 129 I decay and 244 Pu fission would have been retained in the silicate Earth. Because 129 I and 244 Pu decay at different rates, the 129   Review article 129 Xe produced by 129 I decay and the subscript denotes 136 Xe from Pu fission 160 ) can be used to calculate a closure age that marks the end of open system magma ocean outgassing 213 . The Iceland and Samoan Rochambeau Rift samples exhibit low 129 Xe * / 136 Xe Pu ratios compared with the MORB mantle. If the whole mantle had an initially homogeneous I/Pu ratio, then low 129 Xe * / 136 Xe Pu ratios in OIBs would indicate a late closure age for the OIB mantle relative to the MORB mantle, because less of the shorter-lived 129 I would remain at the onset of Xe retention in the mantle. A more realistic scenario might be that the mantle had an initially heterogeneous I/Pu, and regions of low I/Pu reflect the limited accretion of volatile-rich materials into the OIB mantle 133,160,211 .
A relatively dry OIB mantle could have contributed to inefficient mixing in both the terrestrial magma ocean and the solid mantle throughout Earth history 164 . Thus, the OIB I−Pu−Xe signature not only records heterogeneity in the early mantle but also provides insight into the mechanisms that preserve heterogeneities formed during the early magma ocean stage.

The effect of the core on the OIB source
Early core formation and its subsequent evolution have likely played an important role in controlling radiogenic and stable isotope variations in OIBs. The metallic core of the Earth physically separated from the mantle during the first few tens of million years of Earth history [214][215][216] , trapping a substantial proportion of the light elements (~10% of Si, O, S, C, N and H 217,218 ). The incorporation of these light elements into the core could have changed the isotopic composition of the mantle on a bulk scale 219 ; however, such fractionation would not generate mantle heterogeneities if the core formed when the entire mantle and core were in equilibrium. For example, whole mantle−core equilibrium 220 has been proposed to explain the fractionation of the mantle in silicon isotopes relative to chondrites 221,222 . By contrast, evidence from diamond inclusions suggests that Fe isotopic heterogeneities exist in the deep mantle of the Earth. Cullinan-like, Large, Inclusion-Poor, Pure, Irregular and Resorbed (CLIPPIR) diamonds from Letseng, Lesotho, which originate from depths of 360−750 km, exhibit heavy Fe isotopic signatures (δ 56 Fe = 0.79−0.90‰) that lie outside the near-0‰ range of known mantle compositions or expected reaction products that occur at these depths 151 . High pressure and temperature experiments suggest that core formation on its own cannot account for such a large shift in Fe isotopic ratios 223 . Rather, these data provide evidence for the subduction of surface material characterized by light isotopic ratios into the lower mantle 151 .
Owing to the lack of temporal control on stable isotope signatures, it is plausible that the Fe stable isotope fractionation observed in the mantle reflects the cumulative effects of several processes. These processes likely include both core−mantle differentiation and the subtle, but systematic, heterogeneity in the convecting mantle caused by billions of years of subduction. To identify and untangle all the possible fractionation mechanisms within the mantle, more experiments need to be conducted in relevant pressure, temperature and compositional space.
Core−mantle interactions. The extremely high temperatures at the CMB can cause mantle minerals that are in direct contact with the liquid core to be in chemical equilibrium with that liquid. As the core cools and the composition of the outer core changes because of the ongoing crystallization of the inner core, some elements might diffuse across the CMB as they are exsolved from the liquid outer core. Thus, the composition of the core could be changing progressively, increasing mantle heterogeneity as the core evolves 224 . Grain-diffusion experiments found that siderophile elements diffuse through MgO at a high enough rate to transport those elements across geological length scales (tens of kilometres) over 4.5 Ga, demonstrating that grain-boundary diffusion is an efficient pathway for core−mantle interactions 225 .
Geochemical signatures of several highly siderophile elements suggest that the lower mantle was polluted with core material before being entrained into mantle plumes. The resulting signatures include elevated Fe/Mn ratios, such as those in Hawaiian lavas 226 , and radiogenic Os isotope enrichments detected in several mantle plumes 227,228 . Despite further investigations [228][229][230] , no other isotopic signature indicative of element transport across the CMB was identified until high-precision μ 182 W measurements could be performed 193,231,232 . Small negative anomalies in μ 182 W (−25 ppm) have been measured in OIBs 44,[204][205][206] (Fig. 4b).
In mantle-derived samples, μ 182 W shifts from positive values in Hadean− Archean samples (4.3−2.7 Ga) to negative values in modern samples. This observation suggests that the W signature in OIBs could reflect a time-integrated core contribution of W to the mantle. The mechanism of the interaction between the outer liquid core and the mantle remains uncertain, but current work is focused on the exsolution of the Si−Mg−Fe oxide 207 and diffusive exchange across the CMB between foundered oxidized, oceanic crust and the outer core 44,45 .
Core−mantle interactions could also manifest in the I−Xe system. If iodine were more strongly siderophile than Xe at core formation pressures and temperatures 17,233 , the core would have had elevated I/Xe and I/Pu compared with the mantle. In this case, the core would supply high 129 Xe/ 130 Xe and 129 Xe * / 136 Xe Pu to the mantle. However, OIBs sample a reservoir with low 129 Xe/ 130 Xe and 129 Xe * / 136 Xe Pu relative to the rest of the mantle, suggesting that the core might have acted as a sink for lower mantle I during accretion with no subsequent transfer of radiogenic Xe back to the mantle over time 17 . The core has been proposed as the source of high 3 He/ 4 He in the OIB mantle [234][235][236][237] . Negative μ 182 W anomalies are broadly associated with elevated 3 He/ 4 He 44,[204][205][206]238 in some lavas from Hawai'i, Samoa, Iceland and the Galápagos, suggesting that core contributions could supply both negative μ 182 W and high 3 He/ 4 He to material in the lower mantle. However, the highest OIB 3 He/ 4 He ratios are associated with modest negative μ 182 W, and the strongest negative μ 182 W anomalies are associated with only moderate 3 He/ 4 He.
One proposal to explain the relationship between He and W isotopes in OIB is that their anomalies only persist in mantle domains least affected by crustal recycling, because recycled crust contributions overwhelm He and W isotopic signatures from the core material 238 . This hypothesis is consistent with Th enrichment observed in OIB samples with low 3 He/ 4 He, as recycled crust has high Th abundances, which decays radioactively producing 4 He (ref. 129). Nevertheless, the most negative μ 182 W anomalies measured to date are accompanied by strong indications of recycling in Kr and Xe isotopic compositions at Fernandina Island, Galápagos 163 . Similarly, there is a pronounced recycling signature in the Kr and Xe isotopes at Iceland 160,163 , where some of the highest 3 He/ 4 He ratios have been detected. Additional measurements of W and noble gases from samples with thoroughly characterized radiogenic isotopic compositions are needed to resolve the relationship among core-hosted signatures, the contribution from recycled crustal material and primitive mantle domains.

Constraints on lower-mantle seismic structures
Variations in the 182 Hf− 182 W, I−Pu−Xe and 146 Sm− 142 Nd isotopic systems in OIBs require that plumes sample early-formed reservoirs, likely from the deepest part of the mantle (Fig. 4). Although few lavas have been Review article measured for all of these isotopic systems, the existing data provide insight into the nature of early-formed mantle reservoirs. The ULVZs are speculated to be the source of the high 3 He/ 4 He and the most negative μ 182 W in OIBs 74,204,239,240 . If core−mantle interactions supply high 3 He/ 4 He and the most negative μ 182 W signatures 44,206,207 , then the ULVZs might have formed through interactions with the core. An alternative mechanism to explain variable μ 182 W anomalies in OIB is early silicate differentiation that modified the Hf/W ratios of mantle reservoirs, which subsequently remained largely isolated from the rest of the convecting mantle 231,241,242 .
The LLSVPs, described as thermochemical piles, might contain relics of magma ocean crystallization after the moon-forming giant impact 5,53,243,244 . Variations in μ 142 Nd measured in some OIBs suggest that remnants of magma ocean crystallization could be preserved in the deep mantle 202 . If the moon-forming giant impact occurred ~4.4−4.35 Ga, as suggested by both terrestrial and lunar samples [245][246][247] , then the 182 Hf− 182 W system was already extinct. Therefore, the mantle 182 W isotope composition would be unchanged by the collision, explaining the lack of correlation between 142 Nd/ 144 Nd and 182 W− 184 W in OIB lavas (except for Réunion lavas 206 ). Alternatively, LLSVPs might have formed from the accumulation of the subducted material 248,249 or from the sinking of dense, reduced material 33,51 . It is likely that LLSVPs do not result from a single process, but incorporate material from primordial and early events, ongoing convection and recycled subducted material 38,133 .

Mantle mixing and convection
Linking the spatiotemporal geochemical variations of plume-derived lavas to the heterogeneous structure of the deep mantle requires an understanding of the internal dynamics of plumes, which depend on their rheology, composition and excess temperature. In a purely thermal plume, assuming a Newtonian rheology, the morphology is controlled by the viscosity contrast between the hot plume and the colder ambient mantle. If the viscosity is strongly temperature-dependent, the plume develops a mushroom shape, with a large head and a narrow tail 79 . For a constant viscosity, the plume shape is finger-like 250,251 . Moreover, the internal flow across a purely thermal plume depends on the viscosity contrast between the hot axial part of the conduit and its colder periphery; the vertical velocity is largest at the plume axis and decreases exponentially with the square of the radial distance from the axis 252 . Such a velocity profile generates zones with high strain rates, in which passive geochemical heterogeneities get stretched into filaments 58,59 .
The lower mantle, however, is likely to be compositionally heterogeneous, which raises the question of how heterogeneous material entrained by a plume is deformed during upwelling and how the plume morphology and flow across the conduit are modified by those heterogeneities. Both laboratory experiments 34,35,253 and numerical simulations 27,29,31,32,244,254 have explored global convection and plume dynamics in a heterogeneous mantle. Compositional heterogeneities are often simulated as intrinsically denser material 33 , representing either eclogitic recycled crust, which is denser than the surrounding pyrolytic mantle 255 , or Fe-enriched, relatively primordial material 29,31 ,256 . Rheological heterogeneities are often simulated as more viscous domains 53,244 , because of a silica enrichment 53,257 , an increase in the mineral grain size 258 and/or reduced water content 164,259,260 . These studies indicate that variations in density and/or rheology affect convective mixing efficiency by promoting the long-term preservation of deep mantle heterogeneities.
Numerical simulations of global-scale mantle convection (Fig. 5) suggest that after 4.5 Ga of convection, part of the recycled oceanic crust accumulates along the CMB and forms large piles, whereas the remainder is dispersed in the mantle as small streaks 261 . The primordial material, which is intrinsically more viscous than the surrounding mantle, partially survives as distinct blobs, and a fraction of the ancient FeO-rich basal layer can be preserved by incorporation into the denser basal piles. Numerical simulations 262 of mantle plumes carrying finite size (30−40-km radius) rheological heterogeneities that are 20−30 times more viscous than the surrounding rocks indicate that these heterogeneities can resist stretching because they rotate during their ascent within the plume conduit. Such material could preserve and transport a distinct isotopic fingerprint from the deep mantle to the surface.
For thermochemical plumes, the subtle balance between positive thermal buoyancy and negative compositional buoyancy induces oscillatory behaviour 34 and complex internal dynamics, because some parts of the conduit might sink, whereas other parts ascend 253 . Furthermore, the idea that isotopic zonation in the plume conduit preserves the large-scale zonation in the mantle source region might not be accurate for thermochemical plumes because, under certain conditions, compositionally denser material rises preferentially at the plume axis 263 . However, if chemical heterogeneity is a passive component of lower-mantle structures (that is, if it has little effect on physical parameters such as density), and if lower-mantle structures differ mainly in their thermal properties, then isotopic zonation could potentially be preserved in the plume conduit 263 .

Importance of mantle flux
The Hawaiian-Emperor chain has been used as the basis for many mantle plume models 264 . However, data accumulated from ocean islands worldwide suggest that Hawai'i might be the exception for plume behaviour, rather than the type model. For example, the Hawaiian system has the highest buoyancy flux and mantle potential temperature of any terrestrial plume 20,265,266 (Fig. 1a), and younger segments of the Hawaiian chain record much higher melt flux than most other plumes 80,265,267 . Furthermore, the correlation among buoyancy flux, mantle potential temperature and elevated 3 He/ 4 He supports the inference that the Hawaiian plume results from dynamic processes rooted in the deep mantle 268 . Thus, the extent to which the Hawaiian plume can serve as the mantle plume archetype should be questioned; however, its importance as an accessible and well-studied but extreme example of intraplate volcanism cannot be minimized.
Instead, OIBs can be effectively evaluated using a range of variables, such as plume strength and temperature, source mantle composition and melting conditions. These variables can be assessed across a mantle plume spectrum, based primarily on magma flux 20 . The Hawaiian-Emperor chain defines the mantle plume endmember with the strongest magma production. Samoa, Iceland and Galápagos are also classified close to the high-production end of the spectrum. They all have high 3 He/ 4 He ratios in some of their erupted lavas and elevated buoyancy fluxes 74 , as well as slow velocity zones and an LLSVP at their source 239 . The other end of the spectrum is represented by weaker plumes, with potentially shallower sources, which do not exhibit multiple mantle geochemical components nor have mantle potential temperatures substantially hotter than MORBs. Examples at this end of the spectrum include Ascension, Cobb−Eickelberg seamounts on the Juan de Fuca ridge and the Bowie−Kodiak (also called Pratt−Welker) seamount chain, which have not been associated with a lower-mantle shear wave seismic tomography anomaly that extends to Review article the CMB 15 . They also have isotopic compositions that are nearly indistinguishable from MORB, in part owing to plume−ridge interactions (such as Bowie−Kodiak and Cobb−Eickelberg 269,270 ), and are among the coolest mantle plumes 268 .
Between the strong Hawaiian-type plumes and the weak Ascension-type systems are those that most closely emulate the classic model, such as Louisville, Kerguelen, Caroline, Easter, Réunion and Tristan. These plumes possess a voluminous head that formed an LIP, a plume tail that generated an age-progressive volcanic chain and a buoyancy flux between ~0.5 Mg s −1 and 2 Mg s −1 that wanes over time. This flux-based plume spectrum provides a systematic reference framework for comparing mantle plumes and plume chemistry, which will prevent inappropriate comparisons between vastly dissimilar systems.

Summary and future perspectives
The study of mantle plumes, their sources and chemical heterogeneity in the mantle has generated important hypotheses and ideas about major mantle processes such as convection, recycling of crustal materials and mantle residence times, as well as the nature of interactions between various reservoirs throughout the Earth system. It is no coincidence that advances in analytical geochemistry capabilities have occurred alongside the increasing sophistication of mantle plume models. The improvement in precision, sensitivity and resolving power of mass spectrometers has opened areas of the periodic table for analysis, including small isotopic anomalies, short-lived isotopic systems and small mass-dependent fractionations that reflect early differentiation processes, recycling of the subducted material into the mantle and evidence for material preserved from the earliest history of the Earth.
The focused development of new isotopic and elemental analysis methods provides insight into large-scale planetary processes and compositional evolution, offering opportunities for future discoveries. Specifically, systems such as I−Pu−Xe, 182 W− 184 W and 142 Nd/ 144 Nd have the potential to resolve core-mantle interactions and to document the preservation and sampling of early-formed reservoirs in the mantle. However, laboratories rarely have the capabilities to analyse both rare noble gases and low-abundance W and Nd isotopes, necessitating collaborative efforts to generate insights from different isotopic systems on the same set of samples.
Improved analytical precision for the more commonly analysed isotopic systems (such as Nd, Pb and Hf isotopes) is essential to continue advancing the characterization of geochemical components in OIBs and examining differentiation processes throughout mantle history. This objective includes targeting melt inclusions, in which extreme compositions from melted mantle heterogeneities can be captured before melt homogenization. Future work should also focus on exploring correlations across elemental and isotopic data sets. A database of high temperature and pressure isotopic fractionation factors will be essential to understand the processes affecting different isotopic systems.
Some of the greatest insights in mantle geochemistry have come from targeted sampling strategies, in which isotopic systems are applied to locations most likely to carry imprints of specific processes. Again, such studies require collaborations between laboratories and researchers, both for access to the advanced analytical techniques and to avoid analysing sample powders in isolation from their geological contexts.
Collaborations among geophysicists, geochemists and geodynamicists are key to resolving important questions about how the transport of heterogeneities from the mantle reservoirs of the Earth to the surface is controlled by partial melting, plume−lithosphere interactions and plume buoyancy forces. Debate persists regarding how geochemical components are entrained, mixed, stretched, stalled and melted during mantle transport and ultimately expressed in erupted lavas. This uncertainty propagates into the models that explain geochemical variations in erupted lavas, which rely on the interpretation of spatial patterns and time-integrated signatures in those lavas. A better understanding of mantle geodynamics, along with how lithology affects the melting and mixing of magmas, is needed to provide better constraints for understanding the source, evolution and preservation of geochemical heterogeneities in the mantle.
Finally, much is still unknown about the composition of seismically imaged mantle heterogeneities, in both the shallow and deep mantle, and how they relate to chemical heterogeneities ( Supplementary  Fig. 10). The spatial differences in OIB compositions, both between different plume systems and over time at individual plumes, suggest that there are systematic variations in mantle geochemical domains on many scales that are currently poorly understood. Resolving uncertainties in mantle geodynamics and melt homogenization processes will help determine whether these differences reflect distinct tectonic histories. Clearly, much remains to be learnt about mantle plumes and the composition of the Earth. It is likely that the greatest advances will emerge from cross-disciplinary studies in diverse fields such as experimental petrology, mineral physics, numerical geodynamics, seismology and geochemistry.

Data availability
Figure 2a-f and the Supplementary figures were constructed from a combined data set of precompiled files for oceanic island groups from GeoRoc and several curated data sets 64,81,86 . New data were downloaded from the GeoRoc geochemistry database in October 2021 and include data from Azores, Easter and Salas y Gomez Islands, Iceland, Kerguelen and St Helena. Primary GeoRoc data selection criteria were geological setting (Ocean Island), selection of ocean island chain, type of material (whole rock) and type of rock (volcanic rock). GeoRoc data from the initial search were combined with additional data downloaded from GeoRoc in July 2020 and 2021, some of which is presented in ref. 86. These data include Samoa, Cook−Austral Islands, Pitcairn−Gambier, Easter, Galápagos, Society and Mauritius (see supplementary information in ref. 86 for a full list of references). New Pitcairn and Society trace element concentration and isotope composition data from ref. 118 were added to the GeoRoc compilation, along with data from the Galápagos from ref. 64. Hawaiian-Emperor data were taken from ref. 81. The total number of samples in the compiled data set is 19,824 and most isotopic data are post-1990. The format of each of these precompiled files was standardized and imported into R, a free open-source statistical computing application for analysis and plotting. All data sets except those downloaded in October 2021 were renormalized to the same standard values to ensure comparability 81 . For major element and isotope plots, no filters were used on the data set to assess data quality, which varied between laboratories, instrumentation, methods and detection limits over the past 40-50 years (much of these metadata are not included in the GeoRoc database or are inconsistently included and therefore difficult to apply across such a varied data set). For trace element plots, a filter of SiO 2 >55 wt% and total alkalis (Na 2 O+K 2 O) <8 wt% was applied to remove highly silica-undersaturated samples or lavas that were produced by anomalously low degrees of partial melting. This filter Review article removes samples with heavily enriched incompatible trace element concentrations, which would skew the average results presented in the extended trace element spider diagram.