The Mediterranean Sea hydrography is typically characterized by a negative net precipitation-evaporation balance across the basin. As a result of this deficit, the Mediterranean Sea thermohaline circulation system (Med-THC) presents an anti-estuarine pattern, where the inflowing fresh and cold Atlantic surface waters entering through the Strait of Gibraltar, progressively transform into saltier surface waters, that eventually sink to intermediate depths at the Levantine Sea as Levantine Intermediate Water (LIW). The LIW contributes to preconditioning deep water convection both in the Adriatic Sea and the Aegean Sea, where the Eastern Mediterranean Deep Water (EMDW) is formed. EMDW and LIW outflow together through the Strait of Sicily, hereafter called Eastern Mediterranean Source Waters (EMSW), contributing also to the convection of deep waters at the Gulf of Lions to produce Western Mediterranean Deep Water (WMDW). Both WMDW and EMSW eventually outflow into the Atlantic Ocean through the Strait of Gibraltar forming the so-called Mediterranean Outflow Water (MOW), thus closing the Med-THC system (Fig. 1a)1,2,3,4.

Fig. 1: Neodymium and salinity modern values of the Mediterranean Sea.
figure 1

a Map of the study area in the Mediterranean Sea. Modern surface and deep hydrology of western and eastern Mediterranea Sea represented in the map is based on ref. 2,3,4. Basic map of the Mediterranean Sea obtained from Ocean Data View. The green star represents the studied core NDT-6-2016 (1) and red stars represent the other cores discussed in this manuscript located in both the western and eastern Mediterranean Sea: 2) SU92-33 and 3) RECORD2341, 4) MD01-2472 and 5) MD90-91739, 6) MS27PT42 and 7) BC0751. Violet circles stand for the stations of the obtained present-day seawater εNd values. Blue arrows indicate deep waters (Western -WMDW- and Eastern -EMDW- Mediterranean Deep Waters), red arrows represent Levantine Intermediate Water (LIW) and black ones the surface Modified Atlantic Water (MAW). b Map of the seawater neodymium distribution in the Mediterranean Sea based on previous acquired data from stations (violet circles): B, C, 5, 51, 64;47 8, 9, 3, 2;49 GeoB7718-3, GeoB7716-1;44 2160. c Map of the salinity distribution in the Mediterranean Sea based on Medatlas II acquired data.

Accordingly, the Mediterranean Sea and the Atlantic Ocean behave as a closely coupled system connected through the Strait of Gibraltar and therefore, changes in the Atlantic hydrography could induce changes in the Med-THC and vice versa. In this regard, some studies have hypothesized on the potential role that past climate changes might have exerted in modifying MOW physico-chemical properties, that by extension could have induced changes in the Atlantic Meridional Overturning Circulation5,6,7. Several proxies show evidence that the intensity of the Med-THC system is highly sensitive to past climate oscillations at different timescales and in some particular periods8,9,10,11. In particular, last deglacial sea level rise and the associated increase of Atlantic fresh-water inflow through the Strait of Gibraltar promoted surface water stratification, that caused a generalized weakening of deep water convection at Gulf of Lions11,12. Increased water column stratification also promoted poorly-oxygenated conditions at depth that consequently led into the formation of the last Organic Rich layer at the deep western Mediterranean basin (15–8.9 kyr Before Present or hereafter BP, 1950 as the commencement date of the age scale) (Cacho et al. 2002; Pérez-Asensio et al. 2020; Rogerson et al. 2008).

The development of the last Organic Rich layer coincided with the onset of the so-called African Humid Period (from ~15 to 6 kyr BP) driven by the intensification of the African Monsoon that led to increased runoff from north African riverine systems into the Eastern Mediterranean (E-Med) basin13,14,15,16. It has been argued that generalized freshening of surface waters through runoff promoted a stronger sea surface density stratification that consequently weakened intermediate and deep water convection of the E-Med prior to the Sapropel formation, although such pre-conditioning phase has been largely debated8,17,18. Interestingly, the Atlantic Meridional Overturning Circulation slowdown associated with the Younger Dryas (YD; from 12.95 to 11.65 kyr BP) involved worldwide changes in precipitation patterns19,20,21. In particular, a weakening of the monsoon precipitation in North Africa caused an important decrease of freshwater supply toward the Mediterranean Sea14,17,22,23, while overall cold and arid climate conditions predominated in the E-Med region24,25,26. Although evidence is inconclusive, some studies have proposed that cold/arid conditions during the YD could have favoured a partial reactivation of the intermediate and/or deep water convection in the E-Med13,17,27.

The maximum reduction of the E-Med intermediate/deep water convection was reached during the early Holocene, with the deposition of an organic-rich sediment layer known as the last Sapropel (S1)8,10,13,18,28,29,30,31,32. General consensus is that the S1 occurred between 10.8 and 6.1 kyr BP31, although several authors have proposed that the end of S1 was asynchronous in the water column, with intermediate depths in the E-Med being “re-ventilated” between 7.7 and 7 kyr BP33,34,35. The S1 formation is attributed to the sum of 1) strong surface ocean stratification resulting from increased influx of fresh-water by river systems, that led to the establishment deep-basin anoxic conditions and 2) increased export productivity resulted in enhanced organic matter fluxes to the deep sea floor8,17,31,32,33,34,35,36,37,38. The potential impact of the S1 in the water exchange between the E- and W-Med sub-basins has also been previously explored39,40,41,42,43. However, no quantitative estimates of changes in the eastern-western water exchange through the Strait of Sicily during the deglacial and Holocene periods are available to date.

In this work we present a continuous record of quantitative estimations of eastern-western water exchange since the last deglaciation, including the YD and the S1 events. Estimations are based on neodymium isotope (εNd) measurements on planktic foraminifera Fe-Mn crusts, which has been shown to represent well the Nd isotopic signature of the waters masses where foraminifera shells are deposited44,45. Present day E-Med seawater εNd values are typically higher than W-Med εNd values (EMDW = −7.1 ± 1.6 n = 19, 2σ, LIW = −6.6 ± 1.9 n = 19, 2σ, WMDW = −9.1 ± 1.3; n = 12; 2σ), therefore establishing the basic ground for the use of Nd isotopes as useful tracer of past changes in Med-THC (Fig.1b)44,46,47,48,49.

Gravity core NDT-6-2016 collected during the Next Data expedition (CNR-URANIA R/V, 2016) was recovered at the transition area between the W-Sicily channel and the southern Tyrrhenian Sea (38°0'26,60” N and 11°47'44,84“E) at 1066 m of water depth (Fig. 1a–c). The core depth is below the present-day interphase between WMDW and the EMSW outflowing through the Strait of Sicily, thus a suitable location to evaluate changes in the intermediate-deep water exchange between the two main Mediterranean sub-basins49. εNd results are complemented with other published εNd records from different regions of the Mediterranean Sea (Fig. 1a)39,41,42,50,51. Through non-linear isotopic end-member modelling we also perform estimates of the mixing proportions between the outflowing EMSW and the WMDW during the late deglaciation and Holocene periods.

Results and discussion

The εNd record from W-Sicily exhibits remarkable changes along the late deglaciation and Holocene periods (the last ~15 kyr cal. BP). The highest εNd values are recorded during the YD (between −5.6 ± 0.4 and −6.3 ± 0.3, 2σ; Fig. 2), while the lowest εNd values are predominant during the S1 interval (from −7.8 ± 0.3 to −8.4 ± 0.3, 2σ; Fig. 2). The pre-S1 and the post-S1 (between ~11.6 and 10.5 and ~7–6 kyr, respectively)52 were characterized by intermediate values ranging from −7.0 ± 0.4 to −7.4 ± 0.6 and from −7.3 ± 0.3 to −7.5 ± 0.3 (2σ), respectively (Fig. 2; Supplementary Table 1). In order to ensure that the measured foraminiferal εNd values truly reflect changes in the local hydrography without external contributions of Nd coming from terrigenous sediments and/or atmospheric dust deposition53,54,55, we have conducted additional εNd analysis in the detrital fraction < 63 μm. Our lithogenic data show considerably less radiogenic values on the detrital fraction with rather constant values throughout the studied period (from −11.58 ± 0.2 (2σ) to −12.06 ± 0.2 (2σ)) (Supplementary Table 2 and Supplementary Fig. 1). Any potential impact of detrital fraction into the pore water εNd composition should therefore be expressed as deviations towards more negative εNd values in the foraminifera samples. Our εNd results on lithogenic fraction are generally compatible with the εNd signature of the continental margins at the studied region46 and also agree particularly well with a detrital εNd record from a nearby location at the Gulf of Sirte56 (Supplementary Fig. 1). Then, these data, support that our εNd values obtained from the lithogenic fraction represent well the sedimentological signal of the studied area. Consequently, we can confidently discard any noticeable contribution from terrigenous components because 1) measured lithogenic εNd is far more unradiogenic (between 4 and 6 ε units) than the observed foraminiferal εNd values obtained in this study and 2) our lithogenic εNd record shows unchanged values through the studied period and therefore, could not explain the observed changes in our foraminifera εNd record. We have also discarded any potential Nd remobilization in the sediments induced by oxygen depleted pore waters57,58,59, since no evidence of sapropel layers and/or low oxygen sediments exist at the W-Sicily record. In addition, the εNd value of the uppermost sample of the record (Fig. 2, −7.9) is within range of present day seawater εNd values in the region for depths between 2400 and 400 m, thus, supporting the validity of our results60 (Fig. 2b). Nevertheless, although we are confident that contribution of reactive/labile phases from sedimentary sources is negligible, we cannot completely exclude this rare diagenesis process. In this regard, a recent study has shown that to a great extent, the most reactive/labile neodymium is released at the surface ocean, well above the thermocline61. In addition, an external source of new radiogenic εNd values during prolonged periods of time should have been recorded elsewhere, but there are no evidences for it in the published records of authigenic or lithogenic neodymium39,40,41,50,62. These observations provide strong support that our foraminifera εNd results represent with confidence past seawater εNd signatures, and thus, the foraminiferal εNd record from core NDT-6-2016 truly represents changes in the regional oceanography for the last 14 kyr BP.

Fig. 2: Neodymium isotopes records from Mediterranean Sea.
figure 2

a Records of Neodymium isotopes in planktic foraminifera from NDT-6-2016 (this study), SU92-33 and RECORD2341, MD01-2472 and MD90-91739, MS27PT42 and BC0751 (violet circles, blue diamonds, orange six pointed stars, light blue squares, brown five pointed stars, light green upward triangles and dark-green downward triangles, respectively). The 14C-calibrated dates used for each chronology are colored according to the corresponding record. In the sediment core NDT-6-2016, violet circles represent 14C-calibrated dates and yellow circles represent tie points. Violet shading correspond to the uncertainties of εNd results (2sd) of the studied core NDT-6-2016. Error bars reflect the other plotted records uncertainties reported in the bibliography. Grey-blue bar represents YD and light-brown bar, the S1. b Modern εNd seawater values for the Tyrrhenian Sea area60 (circles) and εNd seawater from uppermost sample of NDT-6-2016 (studied core). Dashed line indicate the calculated modern average εNd seawater value of the Tyrrhenian Sea area through the stations 4, 7, 8, and 9. Error bars reflect uncertainties associated to εNd results (2sd). c Map of the study area in the Mediterranean Sea with the discussed sites location.

Enhanced EMSW outflow during the Younger Dryas

The remarkably high εNd values recorded at the west flank of Sicily channel during the YD are one of the most outstanding features recorded for the studied period (from −5.6 ± 0.4 to −6.3 ± 0.3, 2σ; Fig. 2; Supplementary Table 1). We interpret these εNd values as the result of either or a combination of 1) an enhanced EMSW contribution through the Strait of Sicily into the study area, provided that W-Med sourced waters, such as WMDW, are typically represented by more negative εNd values (Fig.2)44,46,47,48,49, 2) a shift towards more radiogenic composition in the EMSW end-members, i.e., EMDW and LIW. Since outflowing EMSW comprises a mix between LIW and EMDW49, any relative change in the εNd endmember composition of either of these two water-masses could have also induced variations in the W-Sicily εNd signature, independently of any changes in the volume of water exported.

This later hypothesis would be in good agreement with previous data from the deep Levantine Sea, showing the presence of a more radiogenic EMDW during the YD than today and attributed to the intensification of Nile river discharge during the African Humid Period (Fig. 2, Nile slope, MS27PT)42. Since εNd values of Nile river sediments range from −7 to 246,47, it is very likely than an increase of river discharge would have favoured the formation of higher radiogenic EMDW. In addition, changes in the EMDW εNd signature could have also been induced through changes in the relative contribution of both Adriatic Sea and Aegean Sea waters to the EMDW. The Adriatic Sea shows relatively constant εNd values along the studied period (Fig.2, MD90-917)39, which are consistently more negative than those found in the Levantine Sea (Fig. 2, MS27PT, BC07)42,51. A recent study has shown that modern Aegean Sea water εNd signature is more positive than Adriatic Sea water εNd signature (−5.9 ± 0.5, n = 6, 1σ versus −7.4 ± 0.15, n = 3, 1σ, respectively)63, thus increased relative proportions of Aegean/Adriatic waters into the EMDW cannot be discarded as a potential source for the observed YD anomaly. Relatively high εNd values of EMDW, however, were recorded at the Levantine Sea for the whole deglacial period and early Holocene (Fig. 2)42, while at our W-Sicily site the radiogenic εNd anomaly is restricted to the YD period. Consequently, changes in the EMSW end-member composition cannot solely account for our observations. Nevertheless, despite the generally good correspondence between the W-Sicily εNd record with an analogous εNd record from the eastern flank of Corsica, the high εNd values during the YD at our study site contrast with low εNd values recorded at Corsica channel (core MD01-2472 in Fig. 2). It is possible that the relatively low temporal resolution of εNd record from Corsica could have compromised the identification of this relatively short event. Furthermore, the outflowing EMSW progressively mix with the underlying less radiogenic WMDW, modifying its εNd signature towards more negative values during its north-westward path49. In this scenario it would be reasonable to expect a modified (more negative) εNd signal at the Corsica site during the YD. Thus, we propose that the εNd anomaly recorded at our site during the YD likely resulted from the combined effect of 1) the presence of more radiogenic EMSW end-members and 2) the increased contribution of EMSW into the W-Med over this period.

By considering that εNd values in the W-Sicily site are the result from both changes in EMSW volume export as well as changes in EMSW εNd end-member composition we present a quantitative estimate of the proportion of EMSW reaching the W-Sicily location through the Strait of Sicily for the studied period. Percentages of EMSW have been estimated applying a simple three endmember isotopic mixing model, using our W-Sicily εNd record in combination with previously published records as water mass εNd end-members for WMDW (SU92-33)41, EMDW (MS27PT) 42and LIW (BC07)51 (Supplementary Table 3)41,42,51. It must be noted that we used εNd data from the Levantine Sea as EMDW endmember instead of data from the Adriatic Sea area (Adriatic Deep Water, ADW). As EMDW originates from different mixing proportions of both Aegean and Adriatic waters, using ADW alone as the EMDW endmember would not be an accurate representation of this water mass2,4. In order to account for the uncertainties associated with mixing estimates as well as intrinsic analytical errors in our and previously published Nd isotope ratios, we have implemented a Monte-Carlo approach of the mixing equation (see details in methods section). Furthermore, since proportions of LIW and EMDW contributing to the EMSW in the past are unknown, different scenarios in the relative mixing proportions between LIW and EMDW have been considered (LIW = EMDW, LIW = 2*EMDW, and EMDW = 2*LIW; see details in methods section). Previous calculations of EMSW proportions using potential temperature, salinity and oxygen (20–40%)49 agree with our present-day estimations of EMSW contribution at W-Sicily (~30%; Fig.3), thus, supporting the robustness of the mixing model.

Fig. 3: Water-masses distribution in the Mediterranean Sea.
figure 3

a Distribution of intermediate and deep water masses in the Mediterranean Sea during both the YD and S1 and the location of the main cores discussed in this work. b Percentages of Eastern Mediterranean Source Water (EMSW) expressed as % contribution. Different scenarios corresponded with different proportions of EMDW and LIW; the first scenario characterized by the same contribution of each water mass, the second one considers a LIW contribution that doubles that of EMDW and the third one takes into account a double contribution of the EMDW with respect to the LIW (yellow, light-green and dark-green lines, respectively). Shading represents the range of uncertainty in the reconstruction (95%), based on analytical errors. Grey-blue bar represents the YD and light-brown bar the S1. Current percentages of EMSW are calculated using present-day seawater εNd values from WMDW, LIW, EMDW47 and South Tyrrhenian Sea49 and are indicated with rhombs (see supplementary Table 3). c Average values of EMSW export in the studied location for each period discussed in this work. Error bars reflect the uncertainties associated to the average values calculated for each period (YD, n = 9; pre-YD, n = 9; post-YD, n = 15; S1, n = 33; 2sd). Error bar for present-day values correspond to the uncertainty in the reconstruction (95%), based on analytical errors.

Mixing estimates calculations further support a larger proportion of EMSWs arriving into W-Sicily during the YD (Fig. 3b, c), as compared to the pre- and post-YD (in average: YD, 57 ± 5%, n = 9; pre-YD, 23 ± 4%, n = 9; post-YD, 33 ± 5%, n = 15; Fig. 3a–c), Our mixing estimates are supported by previous results from an ocean-biogeochemical model experiment and proxy compilation17, that show a partial re-ventilation of the E-Med basin during the YD. This situation could be associated with increased cooling and aridity conditions in the Aegean Sea that could have favoured an intense deep water convection in the area (Fig. 4g)24,25. Enhanced deep water convection in the Aegean Sea during the YD is also coherent with the higher εNd values recorded in W-Sicily.

Fig. 4: Comparison between marine records from the Mediterranean region.
figure 4

From the top to the base: (a) Zr/Al from U1396. b, c Low oxygen benthic foraminiferal taxa percentages from intermediate record HER-GC-UB6 and deep MD95-204367 respectively. d, e Grain-size measurements from M864 and MD01-247287 respectively. f Neodymium isotope records from NDT-6-2016 (this study). g Semi-desert taxa percentages obtained from SL15224,88. h Ba/Al profiles from MP50PC34. Grey-blue bar represents the YD and light-brown bar the S1.

Interestingly, the higher εNd values recorded during the YD at our site are in direct concurrence with maximum values in sediment grain-size reported from both a shallower and distal core located at the eastern flank of Corsica (at 501 mwd, Fig. 4e)13 and also from an even further west and deeper core in the Alboran Sea (at 914 mwd, Fig. 4d)64. These grain size records suggest an increase in the currents velocity at intermediate depths in the W-Med during the YD, that is also synchronous with increased speed of Mediterranean waters outflowing the Strait of Gibraltar as reported in contourite deposits of the Gulf of Cadiz6,65,66 (Fig. 4a), suggesting a direct connection between the EMSW flow and MOW. In addition, enhanced YD EMSW outflow occurred simultaneously with the arrival of more oxygenated waters at intermediate depths in the Alboran Sea (Fig. 4c), whereas the deep basin was characterized by poorly-oxygenated conditions as a result of weakened W-Med deep water convection at this time (Fig. 4b)67. We hypothesize that enhanced YD EMSW outflow 1) could have induced changes in the W-Med deep-water convection that led to a re-oxygenation at intermediate depths (~900 m) and 2) could also be linked to the intensification of MOW currents through the Strait Gibraltar6.

Weakening in the westward flow of Eastern Mediterranean Source Waters during the last Sapropel

In contrast to the high εNd values described during the YD in our W-Sicily record, our results show that the S1 period is identified by relatively low εNd values, (Fig. 2; from −8.4 ± 0.3 to −7.8 ± 0.3, 2σ), i.e., between ~2 and 3-unit lower than for the YD period (between −5.6 ± 0.4 and −6.3 ± 0.3, 2σ). This relatively low εNd values are interpreted as a response of a weaker outflow of EMSW into the W-Sicily during the S1 (Fig. 3a). This interpretation is supported by the low percentages of EMSW obtained throughout the end-member mixing model estimates (in average: 16 ± 6%, n = 33; Fig. 3b, c), i.e., approximately one half of EMSW exports estimated for the pre- and post-S1 (in average: pre-S1, 33 ± 5%, n = 15; post-S1, 29 ± 2%, n = 9; Fig. 3b, c).

Our relative low εNd values recorded at W-Sicily also agree with previous εNd results from the Sardinia channel measured in deep-sea corals at 414 m, both of them showing a decrease in the εNd records from the pre-S1 to the S1 (Fig. 2)41. Thus, εNd data from both W-Sicily and Sardinia channel support moderate EMSW outflow during the pre-S1 but, particularly show an important reduction in the export during the S1 period at both locations. All these indicators are consistent with a weaker EMSW outflow during the S1, which presumably resulted in a larger volume of low εNd WMDW occupying intermediate depths at the study site (Fig. 3a). The end of the S1 period is shown as a transition in the εNd values from −8.4 ± 0.3 to −7.5 ± 0.3 units (2σ) (Fig. 2). Quantitative mixing estimates indicate that this transition accounted for an increase in the EMSW contribution from 4 ± 8% to 28 ± 8% at the W-Sicily (Fig. 3b). These results confirm an increase in the export of EMSW through the Strait of Sicily at the end of the S1, likely indicating the end of the deep water stagnant conditions that prevailed in the E-Med during the S1 interval. Interestingly, the estimated decrease in the EMSW outflow agrees well with an intense weakening of MOW current intensities (Fig.4e) described at the Gulf of Cadiz6,65,66. This situation supports the previously argued control of E-Med intermediate/deep water convection on the MOW intensities and its impact in the Atlantic Meridional Overturning Circulation7,68.

The end of the S1 stagnation occurred between 7.5 and 7 kyr BP according to our age model52, suggesting that the interconnection between the E-W Mediterranean basins was re-established ~1 kyr before the end of the S1 period around 6.1 kyr BP31. If this observation was confirmed, it would support that the re-activation of intermediate and deep water circulation might have not occurred synchronously across the E-Med basin, but instead, intermediate waters would have been re-ventilated before any other deeper water masses in the basin. This hypothesis is coherent with previous evidences for an earlier S1 re-ventilation between 7.7 and 7 kyr BP, both in the Adriatic and the Aegean Sea (Fig. 4h)34,35,38,69,70. Although this earlier re-ventilation at intermediate depths might have not affected the deep basin, our records indicate that it would have been strong enough to re-establish the intense water exchange between the E-Med and W-Med.

Summary and conclusions

εNd data from sediment core NDT-6-2016 at 1066 m provide quantitative information on the proportion of EMSW exported towards the W-Med basin, during the last deglacial/Holocene periods. The highest percentages of EMSW are identified at W-Sicily during the YD (57 ± 5%) indicating that an enhanced intermediate-deep water interconnection prevailed between the E-Med and the W-Med. We speculate that this situation responded to the combined effect of enhanced convection in the Aegean and Levantine basins in response to a regional aridification phase and a weaker western deep water formation that led to a deeper expansion of the EMSW into the western basin. We propose that enhanced westward flow of EMSW also corresponds with the previously described stronger currents at intermediate depths of the W-Med. This situation would have induced changes in the W-Med deep water convection promoting re-oxygenation of intermediate layers (~ 900 m) and ceasing the stagnation initiated during the last Organic Rich layer formation in the westernmost part of the Mediterranean Sea. Estimated proportions of EMSW for the S1 period provide the minimum values of the record (16 ± 6%), supporting a substantial reduction of the EMSW outflow through the Strait of Sicily. We propose that the outflow of EMSW through the Strait of Sicily at 7.5 and 7 kyr BP is related to an earlier convection enhancement in the Adriatic and Aegean Seas, that marked the end of the stagnant conditions at intermediate depths of the E-Med. The described evolution of EMSW flow at W-Sicily, vigorous during the YD and weak during the S1, agrees well with changes in MOW current intensities at the Gulf of Cadiz. These results highlight the relevance of the E-Med intermediate/deep water convection in controlling the intensity of MOW.

Materials and Methods

Core description

The gravity core NDT-6-2016 (~4 m long) was sampled every cm and consists of homogeneous silty-clay sediments with no sedimentary irregularities along the whole sequence. Neodymium isotopes were analysed every 2–6 cm for the interval between 1.20 and 1.62 m and at lower resolution (10–30 cm) for the rest of the core.

Radiocarbon dates and age model

The chronological framework was established by twenty 14C dates analysed in monospecific planktic foraminifera samples (> 250 μm, Globoconella inflata). Radiocarbon ages were calibrated using MARINE20 calibration71. In addition to the 14C dates of the lower part, three tie points were added based on an adjusted alignment between the studied δ18O record from Globigerina bulloides and the well-dated reference NGRIP isotope record72,73.The age model52 was constructed using the Bayesian statistics software Bacon74.

Neodymium isotope measurements

A total of 32 samples were selected for neodymium isotope analyses. For each sample, 20 to 30 mg of mixed planktic foraminifera from the > 212 μm fraction were handpicked, except for those samples where planktic foraminifera were scarce and only 4 to 12 mg were available. Each sample was processed individually and the specimens were crushed with extreme care to avoid the over-crushing using two glass slides under the microscope. Each sample was cleaned with Ultra-Pure water (Resistivity > 18 MΩ cm) as many times as necessary to remove the clay fraction, and then rinsed with methanol to eliminate particles that could have remained attached to the vial75,76. Between each Ultra-Pure water and methanol cleanings, samples were sonicated to re-suspend the clay fraction. Sample dissolution was carried out by adding 0.5 ml of Ultra-Pure water and 0.1 ml of acetic acid in each sample. Then, they were sonicated until completely dissolved. Finally, all samples were centrifuged to avoid possible undissolved particles and transferred to clean vials to carry out the sample purification. In addition, Nd isotope ratios were measured on five bulk sediment samples on the < 63 μm size fraction. Around 150 mg of dry sediment samples was taken and grinded to obtain a very fine dust. In order to extract the detrital component in the bulk sediments a sequential leaching protocol was applied77. Leaching steps consisted in 1) organic matter removal, 2) carbonate leaching and 3) authigenic components removal. After these steps, digestion of detrital samples was performed following78. Samples for Nd isotope analysis on foraminifera coatings and terrigenous fraction (< 63 μm) were treated and purified by standard column chromatography methods at the Laboratori d’Isòtops Radiogenics i Ambientals (LIRA) of the Universitat de Barcelona (UB). In the case of foraminifera coatings chromatography consisted in two steps79; 1) rare elements (REE) were separated from sample matrix and collected using a Tru-spec resin chemistry and 2) neodymium was isolated from other REE by means of the Ln-Spec resin chromatography. For terrigenous fraction we followed an elution procedure in a controlled-pressured vacuum system80. Measurements of εNd were performed using a Plasma 3 Multi Collector Inductively Coupled Plasma Mass Spectrometer (Nu Instruments-AMETEK) at Centres Científics i Tecnològics of the Universitat de Barcelona (CCiTUB). Procedural blanks were systematically corrected and were always found to be negligible (between 26 and 45 pg of Nd). The 143Nd/144Nd instrumental mass bias was corrected using the exponential law81, assuming a reference 146Nd/144Nd ratio of 0.7219. Furthermore, the JNdi-1 standard was analysed before and after each sample, and an additional sample-standard bracketing normalization of the mass bias corrected ratios was carried out, using the reference 143Nd/144Nd value of 0.512115 ± 0.000007 for this standard82. Nd isotopic composition is expressed as εNd = ([(143Nd/144Nd)sample/(143Nd/144Nd)CHUR] − 1)*104, where CHUR is the Chondritic Uniform Reservoir and represents a present-day Earth value of (143Nd/144Nd)CHUR = 0.51263883. Uncertainties of the εNd values correspond to the external reproducibility of the JNdi-1 analysis performed throughout each measurement session (2sd, n = 12–20), except for samples with internal standard error larger than 2sd. Uncertainties ranged between 0.3 and 0.5 εNd-units, depending on the Nd amounts in the samples. The accuracy of the method has been demonstrated through multiple analysis of the AMES II standard (0.511972 + /− 0.000012, 2 SD = 23 ppm, n = 79), and results were not significantly from those previously published84,85,86. In addition, 4 independently replicated sample analysis have been conducted at core depth “147 cm”, from foraminifera picking, cleaning, sample purification and analysis. Our replicated sample results range from −7.7 to −8.0 (mean εNd = −7.8 ± 0.3, n = 4; 2σ), with a standard error of ± 0.2 (Supplementary Table 4).

Estimation of water masses mixing using Nd isotopes

Water mass mixing estimations were performed using our W-Sicily εNd record in combination with three previously published W- and E-Med εNd records. Isotopic mixing equations require previous knowledge on the isotopic composition (as well as elemental abundance) of the different end-members considered. In the case of water masses mixing in the ocean, it is also necessary to consider time-evolving end-member Nd isotope composition into the mixing model. For this reason, in this work we have defined the Nd end-member composition based on existing εNd records in the area of study from sediment core SU92-33 (WMDW), MS27PT (EMDW) and BC07 (LIW)41,42,51. The estimates of EMSW export were carried out according to the following equation:

$$\varepsilon _{{{{{\rm{Ndmix}}}}}}=\frac{(({{{{{\rm{\varepsilon }}}}}}_{{{{{{\rm{Nd}}}}}}\,{{{{{\rm{WMDW}}}}}}}* [{{{{{\rm{Nd}}}}}}]_{{{{{\rm{wmdw}}}}}}* {{{{{\rm{f}}}}}}_{{{{{\rm{WMDW}}}}}})+({{{{{\rm{\varepsilon }}}}}}_{{{{{{\rm{Nd}}}}}}\,{{{{{\rm{LIW}}}}}}}* [{{{{{\rm{Nd}}}}}}]_{{{{{\rm{Liw}}}}}}* {{{{{\rm{f}}}}}}_{{{{{\rm{LIW}}}}}})+({{{{{\rm{\varepsilon }}}}}}_{{{{{{\rm{Nd}}}}}}\,{{{{{\rm{EMDW}}}}}}}* [{{{{{\rm{Nd}}}}}}]_{{{{{\rm{Emdw}}}}}}* {{{{{\rm{f}}}}}}_{{{{{\rm{EMDW}}}}}}))}{(([{{{{{\rm{Nd}}}}}}]_{{{{{\rm{Wmdw}}}}}}* {{{{{\rm{f}}}}}}_{{{{{\rm{WMDW}}}}}})+([{{{{{\rm{Nd}}}}}}]_{{{{{\rm{Liw}}}}}}* {{{{{\rm{f}}}}}}_{{{{{\rm{LIW}}}}}})\left.\right]+([{{{{{\rm{Nd}}}}}}]_{{{{{\rm{Emdw}}}}}}* {{{{{\rm{f}}}}}}_{{{{{\rm{EMDW}}}}}}))}$$

where εNdmix correspond to our measured εNd values from core NDT-6-2016, whereas εNd WMDW, εNd LIW, εNd3 EMDW are the εNd values from cores SU92-33, MSPT2 and BC07 respectively. [Nd] corresponds to the neodymium concentration of WDWM, EMDW as well as LIW and f is the relative proportion of each water-mass, according to the mixing mass balance constraint that f1 + f2 + f3 = 1 (Fig. 3a). In order to overcome heterogeneous age sampling among the different records utilized we have performed linear interpolation in the εNd values from cores SU92-33, BC07, and MS27PT to match the ages samples at our study site (Supplementary Table 3). Unfortunately, there is no way to estimate reliably past seawater Nd concentrations to date. Thus, to account for the uncertainties derived from the lack of reliable past-seawater Nd concentration estimates as well as to account for analytical uncertainties, we have implemented a Monte-Carlo approach of the mixing equation. The Monte-Carlo approach estimates the most likely resulting scenarios from a set of 10.000 simulations of the mixing equation using pseudo-random number generation for every single variable considered. This means random numbers within the 2σ analytical uncertainty for each Nd isotope value considered at the 4 core sites. For the Nd concentrations we have considered typical uncertainties in present day seawater measurements of Nd concentrations (see below). The dataset resulting from running 10.000 simulations is then assessed statistically, and the mean values and standard error values are obtained for the different mixing proportions estimated. As a work around to the problem of having 3 unknowns with 2 equations we have considered different scenarios in the mixing proportions between the eastern Mediterranean water masses contributing to the EMSW, i.e., LIW and EMDW. The first scenario is defined by the same relative contribution of LIW and EMDW (i.e. fLIW = fEMDW), the second scenario the contribution of the LIW doubles that of the EMDW (i.e., fLIW = 2*fEMDW) and the third scenario considers that the proportion of EMDW doubles that of the LIW (i.e. fEMDW = 2*fLIW) (Supplementary Table 5). Nd concentrations used for WMDW, LIW and EMDW have been 23.05 ± 1.05, 27.8 ± 1.6, and 28.25 ± 3.35 respectively, based on present-day Nd concentrations measured in each water-mass47.