This paper deals with the origin of enriched MORB independent from any hot spot activity. Indeed, MORB enrichment was readily attributed to a ridge/hot spot interaction and in absence of identified neighboring hot spot, to more questionable processes (e.g., incipient plume or plume activity residue). More recently, the existence of enriched MORB away from any identifiable hot spot was attributed to different origins (i.e., recycled oceanic crust and/or enriched mantle after subduction metasomatism). Within this frame, we present here a new set of geochemical analyses of major and trace elements and Sr, Nd and Pb isotopes on samples collected by submersible on both intersections of the 15°20′N fracture zone with the spreading axis of the Mid-Atlantic Ridge. This area is characterized by an increasing enrichment of the lava compositions from north to south through the fracture zone. Results show that the geochemical enrichment observed with a different intensity on both sides of the fracture zone is linked to the 14°N topographic and geochemical anomaly. Our modeling shows that both trace element and isotopic compositions are consistent with a binary mixing between the regional depleted MORB mantle source and a recycled OIB/seamount, as previously proposed to explain the observed enrichment at 14°N. This model can also account for other enriched MORB i.e., the 18°–20°S region of the Central Indian Ridge, illustrating that it does not represent an isolated and local process. On the basis of our results and on the DMM isotopic evolution, the age of the recycled OIB/seamount is estimated to be ∼250 Ma, suggesting a recycling within the upper mantle. Considering the huge number of ocean islands and seamounts upon the ocean floor, their recycling into the upper mantle is a plausible process to produce enriched MORB.
 Over the last 3 decades, the increasing number of oceanographic expeditions together with the development of analytical techniques and the understanding of the mantle-derived basalt composition have allowed us to evaluate the chemical structure of the Earth. In particular, the numerous studies made on mid-ocean ridges basalts (MORB) have led to a classification in groups mainly based on trace element enrichment. Historically, the most abundant ones are depleted and were named “normal” MORB (N-MORB). Compared to basalts from other tectonic environments, N-MORB are characterized by low abundances in incompatible elements, low radiogenic Sr and Pb and high radiogenic Nd (and Hf) isotope ratios [e.g.,Hofmann, 1997]. Conversely, enriched MORB (E-MORB [Sun et al., 1979; Schilling et al., 1985]) retains their tholeiitic composition but are characterized by more elevated incompatible trace element abundances, trace clement ratios and more radiogenic compositions in Sr and Pb isotopes than N-MORB [Saunders et al., 1988]. E-MORBs have long been considered as a rare expression of MORB, but numerous spreading ridge investigations have demonstrated that MORB enrichment does not represent isolated anomalies [Macdougall and Lugmair, 1986; Zindler and Hart, 1986; Leroex et al., 1992; Cousens et al., 1995; Castillo et al., 1998; Niu et al., 1999]. Therefore, these observations highlight the nature and heterogeneity of the upper mantle beneath mid-ocean ridges.
 Although the existence of E-MORB has been known for a long time [Schilling et al., 1985], their origin proves difficult to determine within the binary model of enriched hot spot mantle versus depleted MORB upper mantle. Although some E-MORBs are easily explained by hot spot activity in the vicinity [Schilling, 1973; Schilling et al., 1985], the origin of numerous E-MORB found far away from any obvious plume still remains enigmatic. Along the Mid-Atlantic Ridge, several examples of hot spot-unrelated E-MORB have been described, among others at 23°N [Donnelly et al., 2004], 14°N [Bougault et al., 1988; Staudacher et al., 1989; Dosso et al., 1991, 1993; Bonatti et al., 1992; Hémond et al., 2006], or 33°S [Michael et al., 1994]. To explain such enrichment, most of the authors have brought up the presence of an enriched component drowned in the “normal” depleted MORB mantle source (DMM), and several possible origins were proposed: old oceanic crust or sediments [Staudacher et al., 1989], an embryonic mantle plume associated with the triple plate junction [Bougault et al. 1988; Dosso et al., 1991], relics of subcontinental mantle [Bonatti et al., 1992], or an unidentified passively embedded chemical heterogeneity in the mantle [Michael et al., 1994]. Nevertheless, none of these have obtained unanimous agreement to explain hot spot-unrelated E-MORB genesis. To account for the origin of E-MORB along the Mid-Atlantic Ridge,Agranier et al.  suggested the dispersion of the “South Hemisphere anomalous mantle” (i.e., the DUPAL anomaly [Dupré and Allègre, 1983; Hart, 1984]). Donnelly et al. explained that the enrichment observed in one basalt from the Mid-Atlantic Ridge near Kane fracture zone (23°N MARK area) could result from the presence beneath the ridge of a peridotite previously metasomatised by crust-derived enriched melts during subduction. Finally, using trace element concentrations,Hémond et al. justified the presence of E-MORB at 14°N by recycling beneath oceanic ridges of previously subducted alkali basalts.
 In this paper, we present a study of samples recovered on both northern and southern sides of the 15°20′N fracture zone (FZ) during the cruise MODE 98 (R/V Yokosuka, 1998). These samples gave us the opportunity to better constrain the extent of the topographic and geochemical anomaly described at 14°N by Dosso et al. [1991, 1993], Bonatti et al. , and Hémond et al. . By using binary mixing modeling on new trace element and radiogenic isotope (Sr, Nd and Pb) analyzes, this study shows the plausible implication of recycled OIB-type material to the formation of E-MORB not only at 14°N, but also at a global scale.
 The northern segment of the 15°20′N FZ is a complex area with numerous fractures and discrepancies. This segment globally shows slow spreading ridge characteristics, such as a deep axial valley ranging in depth from 3900 m to 4700 m at the intersection with the fracture zone. The ridge flanks are markedly asymmetrical [Cannat et al., 1997]. Ultramafic rocks outcroping north of the 15°20′N FZ are serpentinized peridotites. They have extremely depleted harzburgite-like compositions and represent a high partial melting degree mantle residue [Cannat et al., 1992]. Peridotitic rocks are covered with a thin basaltic layer (<100 m) locally missing. Basalts crop out in the ridge axis as pillow lavas, tubes or breccia and present fresh glass rims, which are covered by a thin sedimentary layer (<10 cm). This attests a fairly recent volcanic activity [Cannat et al., 1997].
 South of the 15°20′N FZ, harzburgites tectonically crop out in the internal ridge flanks. Basalts are also erupted along the ridge axis, but no sediments are observed, suggesting a higher magmatic activity compared to the northern segment. They are crosscut by decimeter to meter large fissures revealing recent extensive tectonic activity [Cannat et al., 1997]. The entire segment south of the 15°20′N FZ is characterized by a high topographic bulge centered at 14°N [Dosso et al., 1993].
 In summary, basalts are located near the center of the ridge and seem to result from recent eruptions whereas peridotites and gabbros are exposed on the ridge flanks by a conjugate fault set. Uneven topography is attributed to peridotitic massif uplift across the oceanic crust, associated with intrusive gabbroic bodies, suggesting highly heterogeneous lithosphere, both compositionally and mechanically, near the 15°20′N FZ (<60 km [Escartín and Cannat, 1999]). Away from the fracture zone, the crust is thicker, associated with axis parallel faults and abyssal hills. No peridotite has been dredged, which is consistent with a more homogenous and magmatic crust [Escartín and Cannat, 1999].
 Samples were picked up using the Shinkaï submersible during the Japanese oceanographic cruise MODE 98 on board the R/V Yokosuka (1998). The samples were collected from the spreading axis and from the intersection massifs between the ridge and the transform fault of both northern (dives 415, 417, 418, 419, 420 and 421) and southern (dive 428; Figure 1) segments.
3. Analytical Techniques
 Major element contents were obtained by in situ analyses on basaltic glasses by electron probe Camebax SX-50 in Brest, and trace element concentrations using laser ablation sector-field ICP mass spectrometer (laser 213 nm New Wave coupled with Element 2 ICPMS) at the Max-Planck-Institut für Chemie of Mainz (methodology byJochum et al. ). Sr and Nd fractions were separated using standard ion exchange techniques adapted from Richard et al.  and White and Patchett . Analyses were performed using a Triton T1 thermal ionization mass spectrometer (TIMS) at the Institut Universitaire Européen de la Mer of the University of Brest; Pb fractions were separated following Manhès et al. and were analyzed using a Nu Instrument HR multicollector ICP-MS at the École Normale Supérieure of Lyon.
 Analyses were done on small basaltic glass chips (2–5 mm) carefully selected under a binocular microscope. All chips with apparent sign of alteration or Mn crust were removed. Selected glass chips were leached with 10 mL of HCl 6 N at 50°C during 24 h. Then, the samples were rinsed twice with distilled water and leachates were saved and analyzed to quantify the effects of the leaching process. Average element concentrations in leachates relative to those of samples do not exceed 8% for trace elements.
 Blanks for Sr, Nd and Pb isotope compositions are respectively 250, 90 and 80 pg and are thus negligible compared to the amount of Sr, Nd and Pb concentrations measured here. Sr (NIST-987), Nd (JNdi-1) and Pb (NIST-981) isotope ratios of reference materials were run with each sample batch. Average results are 0.710241 ± 9 (n = 5, ±2σ) for 87Sr/86Sr ratio in NBS-987 reference material, 0.512101 ± 9 (n = 4, ±2σ) for 143Nd/144Nd ratio in JNdi-1 reference material and 16.9327 ± 5, 15.4865 ± 6 and 36.684 ± 2 (n = 10, ±2σ) respectively for 206Pb/204Pb, 207Pb/204Pb and 208Pb/204Pb ratios in NIST-981 reference material. All these results are consistent with those previously published and available online in theGeological and Environmental Reference Materials (GeoReM, http://georem.mpch-mainz.gwdg.de/) database.
4.1. Major Elements
 Major element data show that our samples are all tholeiitic in composition. It was suggested that major elements in E-MORBs may reveal some variations relative to N-MORBs (e.g., increasing content in Al2O3observed in E-MORBs from the MARK area [Donnelly et al., 2004]). This is not the case for the 15°20′N FZ area. No systematic variations can be identified in major elements versus parameters such as latitude or distance to the 15°20′N FZ (e.g., Mg# ranges from 41 to 47), except for Na8.0 average which decreases from 2.8 for the northern segment to 2.3 for the southern segment.
4.2. Trace Elements
 Extended trace element patterns (normalized to primitive mantle) [McDonough and Sun, 1995] are shown by Figure 2. Basalts from the northern segment are slightly enriched in LREE compared to HREE, with average (La/Sm)N and (Dy/Yb)N ratios of 1.10 and 1.07 respectively. 418R3 is the only sample that presents higher ratios, with (La/Sm)N ∼1.23 and (Dy/Yb)N∼1.15. Enrichment for LREE is about 20 to 30 times C1 chondrite contents and less than 20 times for HREE. These samples are therefore intermediate in compositions between N-MORB and more enriched MORB such as basalts from the MARK area or from the MAR 14°N anomaly. Extended trace element patterns show slight enrichment in the most incompatible elements with normalized concentrations ranging from 6 to 10 times primitive mantle contents [McDonough and Sun, 1995]. All samples present negative Pb anomalies, which are characteristic of suboceanic mantle [Hofmann, 1988], but also high positive anomaly in Nb-Ta, and negative anomaly in Sr. Generally speaking, the northernmost samples (e.g., 415R4) are less enriched than samples near the fracture zone (e.g., 418R3;Figure 3), and follow the geochemical gradient described by Dosso et al.  from 17°10′N to 14°N (Figure 3).
 Samples patterns from the south of the 15°20′N FZ are sub-identical, indicating that samples probably come from the same magmatic body. These basalts are significantly enriched in LREE with average (La/Sm)N ratios of ∼2. (Dy/Yb)Nratios are slightly higher than those from the northernmost samples (1.25 versus 1.13). Enrichments are also described for strongly incompatible element. Nb-Ta concentrations are 40 times that found in primitive mantle, twice the contents of the most enriched sample from the north segment (418R3).
4.3. Sr, Nd and Pb Isotope Data
 New isotope data presented in this study are plotted in Figure 2 together with isotope data from Atlantic MORB [Agranier et al., 2005]. Considering average isotopic compositions for N-MORB from the North Atlantic of 0.70240, 0.51322, 18.116, 15.463 and 37.68 respectively for Sr, Nd and Pb isotopic ratios (averages calculated using MORB away from any hot spot; data are fromAgranier et al. ), all of our samples presented in this study are more radiogenic in Sr and Pb (and less in Nd) than those of North Atlantic N-MORB. MORB Sr ratios from the northern segment increase from 0.70249 to 0.70265 toward the 15°20′N FZ. The most radiogenic values (∼0.70276) are measured on E-MORBs located south of the fracture zone. The dispersion observed in87Sr/86Sr ratios for 415R4, 415R5, 417R1 and 419R4 while they are on the mantle array in Nd-Pb and Pb-Pb spaces is likely to be related to seawater interaction. The fact that NAUFARA-007-005 and CHR0077-006-145 samples fromAgranier et al.  behave like these four samples also display good evidence for seawater contamination as well. 143Nd/144Nd ratios from northern segment samples decrease from the 15°20′N FZ to the north. Lower values (∼0.512950) are observed in the southern segment samples with isotopic ratio comparable to those of hot spot-related E-MORB (35°N–45°N region related to the Azores hot spot [Turner et al., 1997]; Equatorial MAR contaminated by the Sierra-Leone mantle plume [Schilling et al., 1994]).
 MORBs from the northern segment range from 18.3647 to 18.5668 for 206Pb/204Pb ratio, from 15.4828 to 15.5075 for 207Pb/204Pb ratio and from 37.889 to 38.138 for 208Pb/204Pb ratio. Conversely, E-MORBs from the southern segment are characterized by higher isotopic lead ratios, with values about 19.23, 15.56 and 38.86 respectively for206Pb/204Pb, 207Pb/204Pb and 208Pb/204Pb. The global increase observed in lead isotope ratios from north to south is in agreement with the progressive source enrichment through the 15°20′N FZ, as determined by previous trace element and isotope composition analyses (Figure 3).
5.1. Enrichment Origin
 The statistical study published by Arevalo and McDonough shows that Atlantic MORBs compared to MORBs from Indian and Pacific oceans are in average more enriched in very incompatible element (e.g., LREE, Ti, Ta, Nb, …) as well as less incompatible elements (e.g., HREE). To account for this observation, the authors suggested that this enrichment is likely related to smaller degrees of partial melting and/or greater extends of fractional crystallization due to smaller spreading rates, even though the presence of prominent recycled source component or variable proportion of pyroxenite in the Atlantic mantle source is also envisaged. Nevertheless, for the most enriched MORB, it is now commonly accepted that the genesis of E-MORB does not only occur because of smaller degrees of partial melting but requires an enriched source contribution [Dosso et al., 1991, 1993; Michael et al., 1994; Niu and Batiza, 1997; Donnelly et al., 2004; Hémond et al., 2006; Nauret et al., 2006]. Recent studies on MAR basalts propose that Atlantic MORB heterogeneities have two origins: (1) a long-wavelength character revealing the influence of hot spot mantle (e.g., Iceland), and (2) a pervasive shorter wavelength character explained by MORB source contamination with old subducted recycled lithosphere (i.e., the dispersion of the DUPAL anomaly [Dupré and Allègre, 1983; Agranier et al., 2005]). The idea of a hot spot interaction to account for the 14°N anomaly was rapidly abandoned, the nearest hot spot, the Cape Verde islands, being located 1800 km away. Bougault et al.  followed by Dosso et al. proposed that the 14°N E-MORB may reflect an interaction with an embryonic mantle plume associated with the triple plate junction, but alternative origins of the enriched mantle source was suggested by the same authors shortly after [Dosso et al., 1993]. It was also demonstrated that E-MORBs from the East Pacific Rise (EPR) could be attributed to small discrete blobs of variably fractionated material, with more radiogenic signatures, dispersed as heterogeneities with a size and distribution related to spreading rate at fast spreading ridge [Allègre et al., 1984; Niu et al., 1999]. Thus, it appears that E-MORBs do occur far from any hot spot and without any relationship with deep mantle plumes [Donnelly et al., 2004]. Therefore, when no hot spot is involved, mixing processes between some N-MORB source and some enriched component is the most adequate explanation accounting for the MORB enrichment variability [Staudacher et al., 1989; Dosso et al., 1991, 1993; Michael et al., 1994; Donnelly et al., 2004; Hémond et al., 2006]. This hypothesis is first suggested by the Na8.0 variation. Na8.0 refers to values of Na2O corrected from low-pressure fractionation to a common MgO content of 8 wt.% and is considered as an index of the partial meting degree [Klein and Langmuir, 1987]. This parameter is also known to increase with the axial depth, explaining why samples north to the 15°20′N FZ have higher values than those located to the region of the 14°N topographic high previously described (2.8 versus 2.3 in average; Table 1). As proposed by Klein and Langmuir , such depth/chemistry relationship may reflect source heterogeneity in major element composition. The mixing hypothesis is also suggested by the isotope data. The correlations described between our samples isotope ratios indicate that the observed MORB compositional variability probably derived from a binary mixing. Even if the mixing hypothesis was early proposed [Bougault et al., 1988; Staudacher et al., 1989; Dosso et al., 1991, 1993], the petrological and geochemical processes leading to the formation of enriched sources were not really discussed until recently by Donnelly et al. .
Table 1. Localization, Major Elements, Trace Elements and Isotopic Data of Samples From the 15°20′N Fz Area Analyzed in This Studya
 These authors underlined that such enrichment requires a low partial melting degree stage to account for the trace element ratios whereas oceanic ridges are mostly characterized by high melting degrees. To account for the enrichment of a mildly enriched basalt (MARK area), authors have developed a model where the MORB mantle source is contaminated by a metasomatised peridotite, which was previously enriched by low degree partial melts during subduction. Without refuting this model, Hémond et al. proposed another explanation for E-MORB genesis. Following the observation ofMcKenzie et al. , they argued that the recycling of alkali basalts composing the great majority of ocean islands and submarine seamounts beneath oceanic ridges could represent an alternative to the metasomatic model proposed by Donnelly et al. . Both models were developed based on trace element characteristics of E-MORB, and particularly the recurrent occurrence of Nb-Ta positive anomaly.Niu et al. estimated that Nb-Ta positive anomaly in oceanic basalts indicates that the sources of these basalts are recycled oceanic lithosphere that had previously undergone subduction zone dehydration which preferentially transferred the Th and U (versus Nb and Ta) to the mantle wedge, leaving a “residual” lithosphere enriched in Nb and Ta. Melting of such enriched lithosphere could therefore enrich the surrounding mantle peridotites and finally produce E-MORB at spreading ridge axis, corroborating the metasomatism model proposed byDonnelly et al. . Nevertheless, Nb-Ta positive anomaly is also characteristic in overall OIB [e.g.,Willbold and Stracke, 2006], probably inherited from the process described by Niu et al. , as recycled oceanic lithosphere is thought to be the major source of OIB [Willbold and Stracke, 2006]. In addition, based on the calculation of trace element variability of the 14°N MAR samples, Hémond et al. suggested that seamounts can pass through the subduction (in spite of the fractionation related to subduction) without being significantly modified in their trace element composition, even for the more “mobile” element such as Rb and U. Thus, recycling OIB and/or seamounts beneath mid-ocean ridge would also be consistent with the genesis of E-MORB.
 The main limitation of the latter model is that Hémond et al.  did not extrapolate it to the isotopes. Conversely, Donnelly et al. used them to demonstrate that the recycling of the oceanic crust cannot explain the isotopic composition of E-MORB, since recycled oceanic crust is expected to have a DMM isotopic composition, which is not radiogenic enough to constitute a suitable source of E-MORB. Based on this assumption, they developed a metasomatic model described above which creates the enrichment necessary to account for the trace element and the isotopic compositions of E-MORB. Similarly toDonnelly et al. , Cooper et al. considered the origin of Azores E-MORB as related to the melting of a peridotite previously contaminated by melts deriving from the recycling of an old oceanic crust. They suggested that the highly radiogenic Sr composition of E-MORB could be inherited from the hydrothermal alteration of the recycled oceanic crust prior to its subduction. Authors also explain the unradiogenic Nd values by aging effects, but such effects are not only restricted to Nd isotope ratios, and would also affect other isotopic systems. Moreover, this model does not explain the highly radiogenic Pb composition usually observed in E-MORB. Although Pb isotopic compositions can increase due to the seawater alteration, this assumption is far from systematic. Indeed, even Fe-Mn crust contributes to increase Pb isotope ratios [von Blanckenburg et al., 1996; Abouchami et al., 1997], Vidal and Clauer  demonstrated that Sr isotopes can show evidences of seawater contamination whereas Pb isotopes do not (this is also the case in this study for some of our samples, Figure 3). The major argument against Donnelly et al.'s  model (and also the one of Cooper et al. ) concerns the depth of the first melting stage. In both models, low F melting of the subducting slab is believed to occur at greater depths than those usually considered for arc-magma genesis (i.e., ∼110 km [e.g.,Hattori and Guillot, 2003; Rüpke et al., 2004]). This assumption may be problematic for two reasons: First, melting of subducting slab requires particular conditions that are far from systematic in subduction zones. Iwamori demonstrates that slab melting is restricted to young slab (<10 My), mainly producing particular arc-magmas such as adakites. Such melting generally occurs in the forearc and is thus localized at low depth (<80 km [Iwamori, 1998]). Second, melting is unlikely to occur at depths greater than ∼150 km. Melting in subduction is associated to the breakoff of hydrous minerals (i.e., serpentine) composing the slab and/or the overlying mantle wedge that release an important amount of water leading to its partial melting [e.g., Hattori and Guillot, 2003; Rüpke et al., 2004]. Even small amount of water can be retained in the slab at depth greater than 150 km, the pressure is likely too high and the amount of water too low to produce slab-derived melts. Therefore, while these observations are not sufficient to rule out the metasomatism model ofDonnelly et al. , they allow us to consider further the OIB recycling model of Hémond et al. . Nevertheless, to validate the model of Hémond et al. , isotopic compositions must be taken into account, particularly considering that even if OIB can be estimated as homogeneous in trace element composition, this assumption is not true for isotopes.
5.2. OIB Recycling Model
5.2.1. Formation of Seamounts: On-Axis Versus Off-Axis Origin
 The major feature that has to be taken into account before considering the model of seamount recycling is the origin of the seamounts. Indeed, the viability of the model critically depends on how do the seamounts form and do all of them display enriched composition that could contribute to produce E-MORB after their recycling. Basically, seamounts can be related to: (1) a hot spot activity; (2) a mid-ocean ridge activity; (3) an intraoceanic subduction zone. Although seamounts deriving from arc magmatism are clearly different in composition compared to the DMM, their number is very limited [Koppers and Watts, 2010] and thus their contribution to the mantle heterogeneity can be considered as negligible.
 Volcanic islands and seamounts formed away from tectonic boundaries (so-called intraplate volcanism, e.g., Hawaii-Emperor chain, Society Islands, Réunion, Cape Verde) are commonly considered as the surface expression of deep mantle plume activity [Morgan, 1971]. In this “off-ridge” setting [Watts et al., 2006; Koppers and Watts, 2010], such intraplate magmatism leads to the formation of large volcanic islands (locating the present-day hot spot activity) which become seamounts as the plate motions move them away from the “stationary” parent plume, forming linear age-progressive seamount trails. Basalts related to hot spots (so-called OIB for oceanic island basalts) differ from MORB in their chemical composition. Unlike the MORB, which are tholeiites thought to be formed by high degrees of partial melting (>10%), OIB are mainly silica-undersaturated alkali basalts deriving from low degrees of partial melting and which are enriched in incompatible elements and radiogenic isotope ratios relative to MORB [e.g.,White and Hofmann, 1982; Zindler and Hart, 1986]. As they were believed to derive from a deep source (i.e., the D″ layer at the Core-Mantle boundary [Morgan, 1971]), OIB offered a unique opportunity to understand deep mantle processes. Therefore numerous studies focused on them over the last three decades [e.g., White and Hofmann, 1982, White, 1985; Zindler and Hart, 1986; Hart et al., 1992; Stracke et al., 2005; Willbold and Stracke, 2006, 2010]. These studies have demonstrated that OIB do not derive from a unique source and that they arise from the mixing of approximately five mantle reservoirs (DMM, EM-1, EM-2, HIMU and FOZO/PREMA/PHEM/C) defined by their isotope ratios [Zindler and Hart, 1986], leading to a scheme nicknamed “the mantle zoo” [Stracke et al., 2005]. Compared with the DMM, presumed to approximate the source of the MORB, EM-1 (Enriched-Mantle-1) is characterized by high87Sr/86Sr, very low 144Nd/143Nd and higher 208Pb/206Pb at a given 206Pb/204Pb, properties that are thought to be inherited from the recycling of pelagic sediments in the source of magmas. OIB that fall into this category are Pitcairn, Kerguelen, Walvis Ridge, Tristan da Cunha and Gough. OIB from EM-2 source (Enriched-Mantle-2) are more radiogenic in Sr and Pb and less radiogenic in Nd than EM-1 and are associated to the recycling of terrigenous sediments eroded from the continental crust and deposited on the seafloor. Society, Samoa and Marquesas Islands are representative of EM-2 type OIB. HIMU (for HIghμ, i.e., high 238U/204Pb) type OIB includes St Helena and some of the Cook-Austral chain basalts (others belong to the FOZO type, see below) and are characterized by extremely high Pb isotope ratios. They are also less radiogenic in Sr compared to EM-types and are likely to be formed by the recycling of old, subducted, dehydrated oceanic crust. Compared to HIMU, the last classification, called FOZO (for FOcal Zone [Hart et al., 1992; Stracke et al., 2005]) was attributed to OIB that were less radiogenic in Pb at a given Sr and Nd isotopic composition. FOZO is mainly represented in basalts from Cook-Austral chain [seeStracke et al., 2005], and it is also considered to be common to most basalts. The idea of a common component mixed in the various sources of OIB, first called PREMA (for PREvalent MAntle [Zindler and Hart, 1986]), has evolved to FOZO, PHEM (for Primitive HElium Mantle [Farley et al., 1992]) and C (Common Component [Hanan and Graham, 1996]), but all of them differ from each other only in detail [Hofmann, 2003]. As the name and the origin of this component have no influence on the proposed model, we will simply refer to FOZO hereafter.
 As their chemistry strongly differs from the DMM, the recycling of OIB may introduce significant amount of enriched material in the mantle. Nevertheless, intraplate oceanic seamounts do not all originate from plumes. Even, several studies have shown that this type of seamount is predominant in the seafloor compared to OIB related seamounts [Batiza, 1982; Hofmann, 1997; Watts et al., 2006; Koppers and Watts, 2010]. They usually occur as isolated edifices (near-ridge seamounts, e.g., along the EPR), clusters (Rano-Rahi seamount field, near the EPR) or lines (the Cameroon Line in the Atlantic ocean or the Line Islands and the Pukapuka ridge in the Pacific Plate). As only few data are available on this type of seamounts, their origin remains poorly constrained,Hofmann arguing that most of these are melting anomalies likely coming from the upper mantle. Following this idea, the hypothesis of small-scale sublithospheric convection [Ballmer et al., 2007] has been developed to explain linear seamount chains that do not display age progression (such as Pukapuka ridge or Line Islands) or seamount clusters (Rano-Rahi seamount field). Recently, on the basis of the elastic thickness, a parameter that is sensitive to whether a seamount is formed nearby of faraway from a mid-ocean ridge,Watts et al.  and Koppers and Watts estimated that the largest part of the world's seamounts are not related to plume and was formed close to mid-ocean ridge (i.e., “on-ridge” type seamount). Studies focused on these “on-ridge” seamounts have shown that they are geochemically different from N-MORB, carrying slightly enriched MORB-like signatures [Batiza and Vanko, 1984; Zindler et al., 1984; Graham et al., 1988; Niu et al., 1997, 2002]. In reality, non-OIB seamounts display large variations in composition, ranging from N-MORB-like signature to OIB-like signatures (Figure 4). Particularly, numerous EPR seamounts have isotopic compositions that are highly comparable with those assumed for the DMM. However, most of the “on-ridge” seamounts exhibit slightly to strongly enriched compositions and plot along the DMM-FOZO array. This indicates that such type of seamounts derives from the mixing of small-scale heterogeneities with the surrounding depleted upper mantle and thus also incorporates a significant amount of enriched material in the mantle when they are recycled.
 Nevertheless, although we show that “on-ridge” type seamounts are slightly enriched compared to the DMM, both types of seamounts do not have the same potential to produce E-MORB after their recycling. As illustrated above, “on-ridge” seamounts differ from “off-ridge” seamounts mainly on their number, their size and their chemistry.Hillier and Watts  estimate a total number of ∼3 million seamounts with a height >0.1 km of which only 200,000 are clearly identified. Considering these numbers, only a few of them are clearly documented enough to constrain their origin. However, Watts et al.  and Koppers and Watts recently estimated that at least 60% of the identified seamounts were formed in an “on-ridge” setting (estimate made usingWessel's database of ∼15,000 seamounts with a height >1.5 km). Although “off-ridge” seamounts are less abundant, they are significantly more enriched in incompatible elements and more radiogenic in Sr and Pb isotopes. Among all “on-ridge” seamounts plotted inFigure 4, only a few reaches the radiogenic enrichment usually described in OIB (e.g., the Cameroon line or some samples from the Line Islands). All others more closely resemble E-MORB rather than OIB. “Off-ridge” seamounts are also higher in size and volume [Batiza, 1982; Wessel, 2001]. For instance, the almost totality of the structures that form the Hawaiian-Emperor seamount chains are higher than 4 km, while “on-ridge” seamounts are smaller (likely <1.5 km in average) suggesting that the largest amount of enriched material that derives from the recycling of seamounts is related to OIB.
5.2.2. Trace Element Modeling
 The trace element modeling reproduced in this study is nearly the same as developed in Hémond et al. . This model was used to constrain the Sr, Nd and Pb concentrations in the enriched component, data that are required for isotope modeling. Parameters used in this model and their references are given in Table 2 and 3 (see also Table S1 in the auxiliary material for complete parameter values used for the modeling including partition coefficients, average compositions in trace elements and isotopes for OIB, etc.).
Table 3. Results of Trace Element and Isotopic Modeling
Mid-Atlantic Ridge (14°N)
Central Indian Ridge (18°–20°S)
Modeled Recycled OIB/Seamount Composition
Modeled Recycled OIB/Seamount Composition
 The assumption of Hémond et al.'s modeling is that E-MORB from the 14°N anomaly in the MAR derive from the melting of an enriched source which can be modeled as a solid binary mixing process between the DMM and enriched component (i.e., a recycled OIB/seamount). For purpose of modeling, the trace element composition of the recycled OIB/seamount is calculated using the non-modal batch meting equation fromShaw  and the solid binary mixing model developed by Faure . The non-modal batch melting equation defined byShaw is turned to express the trace element concentration of the E-MORB source (C0) as a function of the trace element composition of the E-MORB (CEL), the partial melting degree (F), the source bulk distribution coefficient (D0) and the melt partition coefficient (P):
The most enriched E-MORB (sample 428R5) was chosen for the melt composition (CEL). It makes sense to use this sample in this purpose as this sample comes from the area where the contamination by the enriched component is the strongest. As postulated by Hémond et al. , a relatively high partial melting degree (F = 18%) was used because of the enhanced fertility caused by the addition of the enriched component within the DMM.
 The same expression as (1)is used to constrain the trace element composition of the N-MORB source (i.e., the DMM):
Here, CDMM is the trace element composition of the DMM and CN-MORBcorresponds to an average of N-MORB trace element composition (N-MORB are here defined as MORB with (La/Sm)N < 1 and 87Sr/86Sr < 0.7027). D0 and P are the same as in (1) and F = 10%, following the model of Hémond et al. . Then, the composition of the enriched component (i.e., the recycled OIB/seamount; CEC) drowned in the surrounding, more depleted, mantle source (i.e., the DMM; CDMM) is deduced from the solid binary mixing equation given in Faure :
where f corresponds to the proportion of the enriched component in the mixture.
 In Figure 5, the trace element spidergrams of two calculated enriched components (red lines) are reported and compared with the OIB fields [Willbold and Stracke, 2006]. The “0.05:1” and “0.15:1” lines represent the PM-normalized trace element compositions of the modeled recycled OIB/seamount calculated forf = 5% and 15%, respectively (see Table 3 for the compositions). Contrary to Hémond et al. who chose the abundance patterns of the recycled OIB/seamount “so that the final model product matches the observed, average E-MORB at 14–15°N exactly,” we calculate here the trace element composition of the enriched component at different mixing proportions with respect to the OIB fields. In other words, we consider in our calculation that the modeled enriched component is consistent with “natural” OIB when at least 90% of the PM-normalized values plot within OIB field. Even if mixing proportions were chosen to stay within the OIB field,Figure 5demonstrates that chondrite-normalized trace element patterns are consistent with “natural” OIB composition (i.e., incompatible element enrichment, positive Nb-Ta anomaly, HREE depletion). Furthermore, these mixing proportions (5 to 15%) are close to the abundance of recycled material present in north Atlantic MORB source (≤10% [Cooper et al., 2004]). Our estimates can be closer (6 to 11%) to those of Cooper et al. by considering only modeled patterns of which 100% of the chondrite-normalized values plot within OIB field.
where ICm, ICEC and ICDMM are respectively the isotopic compositions of the mixture, the enriched component and the DMM, CEC and CDMM are the elemental concentrations in both components and fis the proportion of the enriched component in the mixture. Results of trace element modeling were used for Sr, Nd and Pb concentrations, as well as the mixing proportions. Isotope ratios of the mixture and the DMM were determined by the same way as for the trace elements: isotope ratios of the most enriched E-MORB (sample 428R5) were chosen as ratios of the mixture, and DMM isotope ratios correspond to the regional average of N-MORB isotope ratios. Other parameters are summarized inTable 2 and Table S2. Results are presented in Figure 6 and Table 3.
 Contrary to the results given by the trace element modeling, variations in the mixing proportions do not lead to significant changes in the modeled enriched component composition. The modeling results are ∼0.70292 for 87Sr/86Sr, ∼0.51287 for 143Nd/144Nd ratio, ∼19.564 for 206Pb/204Pb ratio, ∼15.594 for 207Pb/204Pb ratio and 39.22 for 208Pb/204Pb ratio. Compared to OIB fields, the calculated enriched component appears incompatible with both EM end-members. Sr and Nd isotopes are consistent with HIMU, but Pb isotopes are too low to fit with the extremely radiogenic values characterizing HIMU. However, all isotope ratios are comparable with the FOZO isotope compositions, which are characterized by lower Pb isotopic composition at given Sr and Nd. Considering that both trace element and isotope modeling reach the same results, the recycling of OIB seems to be a reasonable model to explain the E-MORB genesis at 14°N on the MAR.
5.3. Does This Happen Somewhere Else?
 Although our model can explain the MORB enrichment in the 14°N region in the MAR, its validity at a global scale (whether this model can account for MORB enrichment in other ridge environment or not) has to be verified. To test this hypothesis, we used samples from the Central Indian Ridge (CIR) at 18°S [Nauret et al., 2006], as in many points, they resemble to the samples from the 14°N anomaly in the MAR: (1) Samples are enriched in incompatible elements ((La/Sm)N> 1), (2) they display Nb-Ta positive anomaly, (3) they are highly radiogenic in Sr and Pb and isotopically depleted in Nd compared to the DMM.
Nauret et al. suggested that “the CIR MORB, between 18° and 20°S are generated by partial melting of a heterogeneous source consisting of an enriched component and a normal, depleted upper-mantle peridotite.” These authors also highlighted that “the nature of the enriched component is a matter of speculation.” They finally noticed that this enrichment is not related to the presence of a hot spot, as the E-MORB composition is distinct from the Réunion plume composition, which is the closest hot spot in the area. Instead,Nauret et al.  considered that the enriched component could result from the metasomatic event proposed by Donnelly et al. , even if enrichment by recycling ancient alkali basalt is also considered.
 The model established on trace elements and isotopes for the samples from the 14°N region in the MAR was extrapolated to the 18°–20°S region of the CIR. Calculations were made using sample GNT-Pl1-1 since it corresponds to the most enriched sample. Results are presented inFigure 7 and Table 3 (all plots are available in Figure S1). Trace element composition of the calculated enriched component is comparable to OIB compositions using mixing proportions varying between 1% and 5%. Applying these results on the isotope modeling gives an enriched component more radiogenic in Sr (∼0.70525), less radiogenic in Nd and 206Pb (∼0.51270 and ∼18.955 respectively) and almost identical in 207Pb/204Pb and 208Pb/204Pb (∼15.618 and ∼39.19) compared to the one calculated for the 14°N anomaly in the MAR. As illustrated in Figure 7, this enriched component is consistent with common EM 2 basalts, demonstrating that the model of OIB recycling can also explain MORB enrichment in this region.
 More generally, several studies reported the existence of E-MORB along mid-ocean ridges of which at least a part presents identical geochemical characteristics (incompatible element enrichment, Nb-Ta positive anomaly, isotopically enriched in Sr and Pb, depleted in Nd) to E-MORB from 14°N in the MAR. This is the case, for example, of MORB sampled along the MAR at 23°N [Donnelly et al., 2004] and 33°S [Michael et al., 1994], as well as for those located at 11°20′N along the EPR [Niu et al., 1999]. Although our model was not tested in these cases due to incomplete data set, the existence of these MORB with geochemical characteristics as listed above (particularly the Nb-Ta positive anomaly) allows us to suppose that our modeling is expected to work on these regions too.
 Finally, we notice that the mixing proportions determined from the CIR samples are notably the same as those calculated for the 14°N anomaly (globally <10%). This result is compatible with previous results obtained by Cooper et al. who estimated that the materials responsible for the genesis of E-MORB in the North Atlantic are present in abundances of 2–5% in average. However, these abundances are used by Cooper et al. to explain the pollution of upper mantle at a global scale while ours explain smaller heterogeneities along mid-ocean ridges. This means that the percentage of the recycled OIB/seamount in the mantle source of MORB could be greater if required without being in opposition to the model ofCooper et al. [2004, 2009]. In any case, the small amount we calculated together with the generally small scale heterogeneity that represent E-MORB we studied along the MAR and the CIR suggest that the enriched component represents a very small volume (maybe a few cubic kilometer scale) drowned in the surrounding “normal” mantle. This statement is compatible with the recycling of OIB or seamount which are below the sea level and thus small sized when they subduct.
5.4. Hyperbolic Versus Straight Mixing Trends
 As developed by Langmuir et al. , mixing relationships in isotope spaces form linear or hyperbolic trends depending on the denominators in the isotope ratios and on the elemental ratio of both mixing components: if the denominators in isotope ratios are identical (as in the lead isotopes, i.e., 204Pb), the mixing trend is a straight line. Else, the mixing trend depends on the ratio of elemental concentrations in both mixing components:
Choosing for example Sr for x and Nd for y, it becomes
in which a and bare the mixing end-members. If K = 1, the mixing trend is also a straight line. Else, it is an hyperbola. InFigures 6 and 7, plotting the model results with the data shows essentially straight lines, even in Nd-Sr, Sr-Pb and Nd-Pb isotopic projections where hyperbolic trends are likely to occur. As described above, this feature can be explained by a K value of ∼1, showing that the ratios of elements (i.e., Sr/Nd, Sr/Pb and Nd/Pb) are approximately the same in both end-members. This seems to be paradoxical, since our model involves a mixing between the DMM and an enriched OIB source.Figure 8 presents histograms showing the frequency and variation of element ratios (Sr/Nd, Sr/Pb and Nd/Pb) in MORB and OIB. This figure illustrates the broad overlap in the range of Sr/Nd, Sr/Pb and Nd/Pb ratios in MORB and OIB. We found average ratios of 12.8, 247 and 20.1 for Sr/Nd, Sr/Pb and Nd/Pb ratios in MORB, respectively. Average ratios calculated for the OIB approximate those of MORB, leading to K values close to 1 (calculated here as the ratios in MORB divided by those in OIB). As an example, a virtual mixing between sediments and the DMM leads to a K that can be higher than 20 (calculated using the Nd/Pb ratios of the GLOSS from Plank and Langmuir ). Therefore, as element ratios in MORB and in OIB are similar, a mixing between these two components will form curvilinear trends rather than hyperbolas in isotope spaces.
5.5. Time Constraints
 Several studies try to constrain the age of the heterogeneities responsible for E-MORB genesis [e.g.,Dosso et al., 1999; Cooper et al., 2004; Donnelly et al., 2004]. Using mantle isochrons and Monte-Carlo simulations on Sr, Nd and Pb isotope systems,Dosso et al.  give an age of 250 Ma for the heterogeneity beneath the 14°N region. The same age is also given for the enriched component beneath the Azores [Dosso et al., 1999; Cooper et al., 2004]. Cooper et al. extrapolate this age to all enriched components present in the North-Atlantic MORB source (excluding deep hot spot), arguing that all of them represent recycled products subducted along the western margin of Pangea. In their study,Donnelly et al.  also calculated an age of ∼300 Ma based on mantle isochrons, but also showed that this age does not necessarily represent a specific event. Instead, they assume that this age results from the continuous formation and destruction of the enriched mantle source and depends on multiple parameters such as the decay constant of the radioactive decay system or the incompatibility of the daughter element during melting. They finally demonstrated that the mantle isochron ages could approximate the residence time of the enriched reservoir in the mantle if its mass represents a few percent of the system.
Figure 9 shows the evolution of the 206Pb/204Pb and 208Pb/204Pb compositions with time, calculated for both modeled recycled OIB/seamount (14°N in the MAR and 18°–20°S region of the CIR) and compared to the evolution of the DMM. Considering that the Pb isotopic composition of the recycled OIB/seamount cannot be more depleted than the DMM, a maximum age is given by the intersection between both evolution lines. Pb isotope systems give apparent ages ranging from 266 Ma to 295 Ma for the recycled OIB/seamount beneath the 14°N anomaly in the MAR, which is close to the ∼250 Ma estimated by Dosso et al.  and Cooper et al. . Age calculated for the recycled OIB/seamount in the CIR region ranges from 298 Ma to 332 Ma. Using extremely depleted isotopic compositions for the DMM (D-DMM inFigure 9), ages measured for recycling OIB in both regions are ∼400–550 Ma. Tackley demonstrated that the time scale of complete mantle overturn (i.e., travel down to core-mantle boundary and back to surface) approximates 400–500 Myr. This suggests that recycling OIB underwent a maximum of one “cycle” within the mantle before reappearing beneath the ridge. Considering that the 400–500 Myr age is obtained for an extremely (anomalously) depleted mantle composition, an age of ∼250 Ma seems more reasonable, suggesting that the recycled OIB/seamount likely never entered the deep mantle (Figure 10).
5.6. Reconciling the OIB/Seamount Recycling Model and the Oxygen Isotope Composition of MORB
 Recently, Cooper et al.  have published a compilation of available laser fluorination δ18O data for MORB form the Atlantic, Indian and Pacific ridges and have used it to constrain the nature and percentage of enriched material within the upper mantle globally. They revealed that the δ18O compositions of MORB from each ocean basin fit a normal distribution, with identical means and standard deviations (δ18Omean ∼5.50‰) and with a total range of 5.25–5.8%. In addition, they showed that the oxygen isotope data correlate with radiogenic isotope and trace element. Based on this, Cooper et al.  suggested the DMM to have a δ18O composition of ∼5.25‰ and the average measured value of 5.5‰ to reflect a global contamination of the upper mantle by recycling crustal material in proportion ranging from 5 to 10% (taking into account that these estimates may vary depending on the sedimentary contents in the recycled material). This model is much in favor of Donnelly et al.'s  model rather than ours, since ones can assume that the OIB/seamount recycling model cannot account for the apparent correlation between oxygen isotope data and radiogenic isotope or trace element, because OIBs display the same range in δ18O composition as MORB [Eiler et al., 1996; Eiler, 2001]. Eiler et al.  have reported the δ18O composition of olivine from basalts belonging to each subtype of OIB (i.e., HIMU, EM 1, EM 2 and others including some of the FOZO group). Their results show no evidence for δ18O values outside the range for MORB, except for EM 2 type OIB, of which anomalously high δ18O compositions reflect the presence of subducted sediments in their mantle source.
 Therefore, taking into account the results of Cooper et al. , trends in δ18O plotted versus trace elements or isotope data are not consistent with the OIB/seamount recycling model unless the OIB that are recycled are all EM 2. However, this assessment can be addressed by at least two reasons: First, although δ18O data are clearly of a certain interest in the study of mantle heterogeneities, the correlations proposed by Cooper et al.  are not always well defined (this is particularly true for Pacific samples which exhibit a large range of δ18O compositions at a given Sr and Pb isotopic composition). Although the weakness of the correlations may be related to the lack of high resolution oxygen data such as those published by Cooper et al. [2004, 2009], these correlations have to be considered with caution and are not robust enough to rule out the OIB/seamount recycling model. Moreover, as oxygen isotope data are available neither in the 14°N region of the MAR nor in the 18–20°S of the CIR, there is no evidence for a possible application of Cooper et al.'s  model in these regions. That is not ruling out the model of Cooper et al. , but without a denser database in oxygen isotopes for MORB, one could admit that correlations may occur at large scale (ocean scale) while they do not at more restricted scale (not more than few hundred kilometers). Second, the model defended by Cooper et al. is not incompatible with ours as both could actually happen synchronously in different localities, i.e., a recycled seamount could represent the enriched material in some areas while other types of enriched materials could contribute to form E-MORB in others. The main difference between the two models resides in that the recycling of the oceanic crust may account for large scale heterogeneities (e.g., the global enrichment in oxygen isotopes of the upper mantle at ocean scale, as postulated byCooper et al. ) while the recycling of OIB/seamounts are more likely to form very restricted MORB enrichments (no more than hundred kilometers).
5.7. Recycled Seamounts: More Than a Process for Local MORB Enrichment?
 We show in this paper that recycling OIB can be a viable process to explain the genesis of E-MORB at 14°N in the MAR, but it is also susceptible to explain E-MORB genesis along other ridges. As demonstrated above, E-MORB from the 18°S region of the CIR, and in regions potentially presenting similar geochemical features (MORB at 23°N and 33°S in the MAR and those from 11°20′N in the EPR) could be explained by this process. This suggests that the recycling of OIB is not an isolated process.Wessel  and Hillier and Watts  identified slightly less than 15,000 seamounts with a height >1.5 km, and almost a half of them resides on the Pacific plate. Although the size of a seamount is negligible compared to the volume of the mantle, calculations made by Hillier and Watts  show that overall identified seamounts cover more than 18% of the oceanic seafloor area. A virtual addition of these numerous seamounts allows a thickening of the oceanic crust of about 180 m. Considering these estimates and those of Hofmann  which showed that 20 km3 of oceanic crust are recycled per year, we can calculate that seamounts represent approximately 2% of the recycled material. Taking into account that the values given by Hillier and Watts  are minima estimates, i.e., using only identified and measured edifices and that the authors expect around 3 million seamounts with a height >0.1 km, the proportion of seamounts in the recycled material can significantly increase. Thus, considering the trace element budget, the isotopic composition of the mantle and the 5–10% estimation for the total recycled material in the upper mantle with seamounts representing ∼2%, it is clear that the recycling of seamount can hardly be considered as a significant contributor to the upper mantle heterogeneity. Nevertheless, it has to be considered as a plausible mechanism to generate (local) MORB enrichment.
 New trace element and isotope data have demonstrated that MORB enrichment from north and south of the 15°20′N FZ can find their origin in the geochemical anomaly centered at 14°N and associated with a high topographic bulge.
 Isotope ratios confirm and valid the enrichment model established on trace element by Hémond et al. : the E-MORB origin can be attributed to a binary mixing between N-MORB source and OIB/seamount material recycled through subduction. The age of the recycled OIB/seamount is estimated to 250 Ma.
 Finally, we show that the model described here can be applied to E-MORB from other ridge segments, and is thus unlikely to be an isolated process. Considering the great number of seamounts on the ocean floor, their recycling may represent a significant process to account for MORB enrichment.
 M.U., C.H. and P.N. thank Claire Bollinger and Marcel Bohn for their help during the laboratory and analytical works, and Philippe Telouk for his help during lead isotope measurements at the ENS of Lyon. We also thank Catherine Chauvel, Al Hofmann, Charlotte Fillon and Morgane Ledevin for their help during the revision of the manuscript. K.P.J. thanks Brigitte Stoll for its contribution during trace element measurements at the Max-Plank-Institut für Chemie of Mainz. We also thank the Editor Joe Baker and the two anonymous reviewers and Matthew Jackson for their comments that helped to significantly improve this manuscript.