Miocene restriction of the Pacific-North Atlantic throughflow strengthened Atlantic overturning circulation

Export of warm and salty waters from the Caribbean to the North Atlantic is an essential component of the Atlantic Meridional Overturning Circulation (AMOC). However, there was also an active AMOC during the Miocene, despite evidence for an open Central American Seaway (CAS) that would have allowed low-salinity Pacific waters to enter the Caribbean. To address this apparent contradiction and to constrain the timing of CAS closure we present the first continuous Nd isotope record of intermediate waters in the Florida Strait over the past 12.5 million years. Our results indicate that there was no direct intermediate water mass export from the Caribbean to the Florida Strait between 11.5 and 9.5 Ma, at the same time as a strengthened AMOC. After 9 Ma a strong AMOC was maintained due to a major step in CAS closure and the consequent cessation of low-salinity Pacific waters entering the Caribbean.

O f the 30 ± 1.5 Sverdrup (Sv, 1 × 10 6 m 3 s −1 ) of water transported via the Florida Current today, 13 Sv continue to flow northwards as the Gulf Stream 1 , thus constituting a major contribution to the Atlantic Meridional Overturning Circulation (AMOC), which has a mean deep volume transport of 18 ± 2.5 Sv at 24°N 2 . Most model experiments suggest that an open Central American Seaway (CAS) allowed low-salinity waters from the equatorial Pacific to enter the Atlantic at mid-latitudes resulting in an overall weaker AMOC (e.g. [3][4][5][6], and a possible link between CAS closure and changes in North Atlantic Deep Water production has been widely discussed (e.g. [7][8][9][10]. The timing of CAS closure is thus of crucial importance but available reconstructions are controversial, with much of the controversy hinging on the definition of closure 11 . On the basis of geochemical and geochronological data and paleomagnetic reconstructions, recent studies have argued that the CAS essentially disappeared completely as early as 15 Ma [12][13][14] , although allowing that transient and shallow seaways may have persisted until 10 Ma 14 . This, however, contradicts biostratigraphic evidence for a gradual shoaling from 15 Ma onwards, with the seaway still 1000-m deep at 12 Ma 15 . The appearance of a significant and permanent surface water salinity difference between the eastern equatorial Pacific and the Caribbean after~4.7 to 4 Ma 16-20 points towards a later CAS closure, although changes in sea-surface properties could alternatively be explained by a switch from predominantly El Niño-like to predominantly La Niña-like conditions 21,22 . What constitutes 'closure' is also under debate where the fossil record is concerned. If shallow seaways with sufficiently high currents, similar to the Indonesian throughflow 23 , were present in the Pliocene CAS, this may have prevented migration of species between North and South America 20 and could explain the divergence of marine molecular DNA on either side of the CAS after 3 Ma 20 and the fossil evidence for the Great American Biotic Interchange 24 from 2.6 Ma onwards 25 . If the currents in the CAS were not strong, or if the seaways were transient, then this could explain why a recent study combining fossil and molecular data found waves of terrestrial organism dispersal at~20 and 6 Ma and divergence of Atlantic and Pacific marine organisms at~23 and 7 Ma 26 . In the light of the ambiguous evidence for the timing of CAS closure, it is pertinent to examine the progressive restriction of water mass exchange via the CAS in more detail and to compare these results with model predictions of the consequences of variations in Pacific water export to the Caribbean and the Atlantic. The radiogenic Nd isotope composition of past seawater has been used extensively to reconstruct changes in the mixing and provenance of water masses (cf. ref. 27 ). The longest record of the Gulf Stream ε Nd evolution from 850 m water depth in the northwest Atlantic extends back to 8.5 Ma and shows a shift towards less radiogenic compositions between 8.5 and 5 Ma, which was interpreted as a reduction in the advection of intermediate-depth Pacific water to the western North Atlantic 28 , a conclusion supported by model results 29 . For deep waters, sedimentary ε Nd records from the Caribbean extending back to 18 Ma show that the irreversible shift away from Pacific-like compositions already occurred earlier, at 10.7 Ma 10 . However, in apparent contrast to most model predictions (ref. 5 and references therein), episodes of increased Pacific through-flow in the Middle to Late Miocene Caribbean (between~12.5 and 10 Ma) were shown to have coincided with increased AMOC strength 10 .
In this contribution, we specifically compare the evolution of the export of intermediate-depth Pacific waters to the western North Atlantic to records of AMOC strength [30][31][32] . We present a continuous new 12.5 Myr seawater ε Nd record obtained from sedimentary foraminifera coatings of Ocean Drilling Program (ODP) Site 1006 (658 m water depth) located on the leeward slope of the Great Bahama Bank (GBB) (Fig. 1 of the origin of western North Atlantic (Gulf Stream) waters during the Miocene-Pliocene.

Results
Seawater and detrital Nd-isotope signatures. Comparison of surface sediment and water samples from the vicinity of ODP Site 1006 confirms that uncleaned planktonic foraminifera reliably recorded the local bottom water ε Nd signal 33 (Supplementary Figs. 1 and 2). The oldest part of the new seawater record of Site 1006 reveals highly radiogenic ε Nd(t) values of up to −4.6 prior to 11.5 Ma dropping sharply to −8.1 at 10.8 Ma (Fig. 2). Seawater ε Nd(t) values then became slightly more radiogenic again between 9.5 and 8 Ma (−6.6 to 7.3 ε Nd(t) ) before decreasing to −8.8 ε Nd(t) by 6. New Miocene data were also produced to extend the existing foraminiferal and detrital ε Nd(t) records of Site 1000 34 and now allow direct comparison of the long-term changes at the two intermediate-depth sites (Fig. 2, Supplementary Fig. 3). The early part of the Site 1000 seawater record is significantly more radiogenic than that of ODP 1006 with ε Nd(t) values reaching −2.6 to −3.1 between 13.6 and 12.1 Ma (Fig. 2). This is followed by an ε Nd(t) decrease to −7.8 between 12.1 Ma and 8.4 Ma and a slight increase to −6.6 between 7.3 and 6.8 Ma. Unlike the Site 1006 record, the detrital fraction of Site 1000 sediments shows a distinct step-change from ε Nd(t) values between −2.2 and −7.6 prior to 9.4 Ma to between −8.3 and −10.5 after 8.4 Ma ( Supplementary Fig. 3).

Discussion
Prior to 12.1 Ma, seawater ε Nd(t) at Sites 1006 and 1000, as well as at deep Caribbean Sites 998 and 999 10 , was within a range from −4.6 to −2, reflecting a strong Pacific throughflow into the western North Atlantic (Fig. 2). Pacific waters could have been transported to Site 1006 either via the Florida Strait or, as shown in model simulations 6 , through entrainment in the Atlantic gyre, or a combination of both. Some contribution of radiogenic Nd derived from partial dissolution of volcanic ashes may be expected in this tectonically highly active region but a previous study 10 found no correlation between ε Nd signatures and volcanic ash mass accumulation rates at Caribbean Sites 999 and 998. An overall less radiogenic seawater ε Nd(t) signal at Site 1006 compared to Sites 998, 999 and 1000 is consistent with decreasing ε Nd signatures with increasing distance from the CAS, given that the Pacific signal was continuously mixed with Atlantic waters with a less radiogenic ε Nd signature 28,35,36 . Likewise, any particulate material from the Pacific transported across the CAS would more likely have been deposited in the vicinity of Site 1000 than at Site 1006, which explains why the detrital fraction at Site 1000 was on average~2.5 ε Nd units more radiogenic than at Site 1006 prior to 9 Ma ( Supplementary Fig. 3). Partial dissolution of volcanic particles along the path of the recirculated gyre waters or a marked change in the Southern Atlantic end-member composition would be further possible sources of radiogenic Nd, but there is no evidence to support this (see Supplementary Discussion).
After 12.1 Ma, seawater ε Nd(t) at Caribbean Site 1000 became progressively less radiogenic by more than 3 ε Nd units to reach −7.8 by 8.4 Ma. This decrease is similar to the overall trend at two deep Caribbean Sites, which together with increasing sediment accumulation rates was attributed to a decrease in the inflow of more radiogenic and corrosive Pacific waters entering the basin via the closing CAS 10 . Our new data show that the composition of the intermediate-depth Caribbean was strongly impacted by CAS closure in the Late Miocene, as reflected by a marked drop to less radiogenic detrital ε Nd signatures after 9 Ma. The ε Nd(t) and 87 Sr / 86 Sr signatures of the detrital material ( Fig. 3) suggest that the main source of the material delivered to Site 1000 between 14 and 9 Ma was the Mexican volcanic belt. After 9 Ma, the supply of  10,34 , and of ferromanganese crust Blake (BM 1963.897)) 28 . Samples have been corrected to ε Nd(t) to take into account the ingrowth of 143 Nd. All previously published data of the Blake crust and ODP Sites 998 and 999 are presented as thick lines that include the 2σ external reproducibility of the analyses. For clarity, the error bars for the new data are excluded in this figure, but are shown in Supplementary  Fig. 3. The Blake record only starts at 8.5 Ma. The vertical lines indicate the periods of time when Caribbean and Florida Strait seawater ε Nd signatures diverged between 11.5 and 9.5 Ma and converged again between 9.5 and 9 Ma. The detrital composition at ODP Site 1000 changed after 9 Ma (Supplementary Fig. 3)  Orinoco sediments played the most important role, despite evidence that uplift and mountain building in the Northern Andes from~12 Ma actually re-routed the drainage system in northern South America away from the Caribbean and towards the tropical Atlantic 37 . The shift in detrital material composition at Site 1000 therefore indicates a major change in the overall Caribbean subsurface circulation pattern. We find that an initial decrease in seawater ε Nd at Site 1006 occurred earlier than at the Caribbean Sites and reached −8.1 at 10.8 Ma. This marked divergence from the Site 1000 record suggests that there was a change in the regional hydrography between~11.5 and 9.5 Ma and that Atlantic waters with a less radiogenic Nd isotope composition dominated the signal at Site 1006. Continuous sedimentation of the Santaren Drift since 12.4 Ma 38 argues against tectonic uplift of a sill in close vicinity of Site 1006 being responsible for the divergence between the Caribbean and Florida Strait signals. The disconnection was more likely caused by a hydrographic barrier caused by the dominance of the circum-tropical current between the Pacific and Caribbean over the intra-Caribbean current connecting the Northern and Southern Caribbean, as was suggested on the basis of nannofossil assemblages 39 . Whatever the exact cause of the disconnection was, the evidence from the seawater ε Nd records clearly shows that there was no significant export of radiogenic intermediatedepth Pacific waters to the western North Atlantic during this time interval.
The convergence of the Caribbean and Florida Strait ε Nd records between 9.5 and 8 Ma documents the full hydrographic reconnection of the two intermediate-depth Sites (Fig. 2). ODP Site 1000 is located in the Pedro Channel, which crosses the Northern Nicaraguan Rise 40 . According to tectonic reconstructions 41 , the Nicaraguan Rise was fully accessible for northward flow only after 9 Ma, which likely explains the convergence of all seawater ε Nd records between 9.5 and 8 Ma. Moreover, the essentially identical ε Nd compositions of a Blake Ridge record 28 support an unrestricted export of Caribbean water into the western North Atlantic during this time interval (Fig. 2). Between 8 and 6.8 Ma, the Florida Strait record again diverges somewhat from the Caribbean records (Fig. 2), which may also have been a result of a hydrographic barrier. However, during this period, the deep Caribbean seawater ε Nd composition remained constant 10 , and the detrital composition of Site 1000 had already shifted towards that of Orinoco sediments, suggesting that the influence of Pacific intermediate waters was at most minor and any observed changes originated from the Atlantic. On the basis of nannofossil assemblages, it was proposed that the north and south Caribbean basins were separated between 8.35 and 3.65 Ma 39 , but the seawater ε Nd records do not support this interpretation given that Sites 998 and 1000 were similar to the south Caribbean Site 999 record for the majority of this time interval 10,34 .
The vast majority of modelling simulations find that an enhanced supply of low salinity Pacific waters to the North Atlantic reduces the production of Northern Component Water (NCW, pre-cursor of North Atlantic Deep Water (NADW)) and hence the strength of the AMOC (ref. 5 and references therein). However, Newkirk and Martin 10 found apparently contradicting evidence that the interval of high Pacific water throughflow into the deep Caribbean between ca. 12.4 and 9.5 Ma corresponded to an increase in the production of NCW, as estimated from δ 13 C gradients between the Atlantic and the Pacific 31 . Lear et al. 32 also found evidence for a proto-NADW between 12.5 and 10.5 Ma on the basis of cooler benthic Mg/Ca temperatures and higher δ 13 C signatures. Our new record from Site 1006 now shows that this interval of high NCW production [30][31][32] (Fig. 4) occurred at a time when there was little or no export of intermediate waters from the Pacific to the western North Atlantic and thus, importantly, the low-salinity Pacific waters present in the Caribbean at that time 10 could not be transferred to the high northern latitudes of the Atlantic (Fig. 5). Following the same reasoning, lower NCW production between 9.5 and 9 Ma may be attributed to the reestablishment of the Pacific-North Atlantic connection, either directly via the Florida Strait or indirectly via the Atlantic gyre (Figs. 4 and 5). These findings support model predictions that changes in the supply of Pacific waters to the North Atlantic affected the AMOC and highlights the importance of intermediate water mass exchange.
According to the results of an ε Nd -enabled ocean-atmosphere general circulation model, the Atlantic became the dominant source of intermediate-depth Caribbean waters once the CAS had shoaled to between 500 and 200 m 6 . Our new records constrain the timing of this significant change away from a Pacific supply tõ 9 Ma, therefore providing support to studies that found a major step in CAS closure during the Late Miocene 12-14 . However, the new data do not support a complete closure of the CAS by Atlantic-pacific Δδ 13 C (‰)  Fig. 4 Comparison of seawater Nd isotope records with estimates of Northern Component Water production. Seawater ε Nd(t) signatures of Caribbean and Western Atlantic sites (this study 10,34 ) together with two different estimates of Northern Component Water (NCW) production based on δ 13 C gradients between the Atlantic and the Pacific (upper black curve, ref. 32 ; lower three curves showing NCW production (thin black curve) and ±1 sigma uncertainty (thin grey lines), with a hiatus and unreliable estimates before 10.7 Ma omitted, ref. 30 ) for the period between 12.5 and 8 Ma. Vertical dashed lines show episodes of weak NCW production when the ε Nd(t) records indicate strong Pacific-western North Atlantic throughflow 10 Ma 14 and instead allow for a shallow seaway or seaways until at least 9 Ma and possibly much later.

Methods
Regional setting. ODP Site 1000 is today bathed by diluted Antarctic Intermediate Water (AAIW) 42 that can be traced by its salinity minimum as far as 20°N in the North Atlantic 43 . ODP Site 1006 is located in the Santaren Channel and receives a mixture of waters from the Caribbean and recirculated gyre waters 44,45 (Fig. 1 Fig. 2). All samples were freeze-dried and weighed and an aliquot of 2.5 g of each sample was separated and ground using an agate pestle and mortar. The remaining sample aliquot was wet sieved, oven dried and separated into size fractions ready for microscope work. Approximately 35 mg of mixed planktonic foraminifera were picked (mostly Globigerinoides ruber and Neogloboquadrina dutertrei) from the >350-μm size fraction for the extraction of the bottom water Nd isotope composition from the ferromanganese coatings of the uncleaned foraminifera 51 . The shells were crushed between two glass plates to open the chambers, and were then ultrasonicated with alternate rinses of distilled (MQ) water and ethanol to remove any detrital particles. A final visual check for removal of particles was performed under the microscope before dissolution of the foraminifera samples in 0.5 M HNO 3 .
To extract the detrital fraction, the 2.5 g powdered aliquots were leached following the procedures fully described in refs. 52,53 . Samples were rinsed with distilled (MQ) water, followed by a decarbonation procedure using an acetic acid/ sodium acetate buffer (with a pH around 4). A hydroxylamine hydrochloride solution was used as a leaching agent for removing the Fe-Mn oxyhydroxide fraction, and was buffered with acetic acid/sodium hydroxide to reach the pH value of 4. This initial leach solution was removed and an additional 12 h of leaching with newly added leaching solution was carried out to completely remove any authigenic coatings. After a triple rinse with MQ water and oven drying, an aliquot of 0.05 g of the remaining detrital fraction underwent total dissolution via hotplate digestion using a mixture of HNO 3 , HF and HCl. Hydrogen peroxide and HClO 4 were included in the procedure for the removal of organic material.
Further separation of Nd and Sr and purification of the sample solutions was carried out ion chromatographically. All samples were put through cation exchange columns (AG50W-X12 resin, mesh 200-400 µm) 54 to collect rare earth elements (REEs) and, in the case of the detrital samples, the Sr cut also was collected. Thereafter, Nd was separated from the other REEs (Ln Spec resin columns, mesh 50-100 µm) 55 and Sr was separated from Rb and other interfering elements (Sr spec resin column, mesh 50-100 µm) 56 .
Nd and Sr isotope analysis. Isotope ratios were measured on a Nu Instruments Multiple Collector Inductively Coupled Plasma Mass Spectrometer (MC-ICPMS) at GEOMAR. The 143 Nd/ 144 Nd results were mass bias corrected to a 146 Nd/ 144 Nd ratio of 0.7219 and normalized to the accepted JNdi-1 standard 143 Nd/ 144 Nd value of 0.512115 (ref. 57 ). For downcore samples, ε Nd(t) values were calculated by correcting the measured ε Nd(0) for ingrowth of 143 Nd using an Sm/Nd ratio of 0.139 for uncleaned foraminiferal calcite 58 and 0.109 for detrital material, based on the average upper crustal ratio 59 . The external reproducibility (2σ) was between 0.21 and 0.40 ε Nd units for the detrital samples. The external reproducibility (2σ) for the measurements of the foraminifera samples was higher (0.98 ε Nd units) due to smaller sample sizes that required a time resolved measurement method.
After correction for Kr and Rb interferences, Sr results were mass bias corrected to an 86 Sr/ 88 Sr ratio of 0.1194 and normalized to the widely accepted 87 Sr/ 86 Sr value of 0.710245 for the NIST SRM987 standard. The 87 Sr/ 86 Sr results were corrected for ingrowth of 87 Sr using an 87 Rb/ 87 Sr ratio of bulk earth of 0.09 (ref. 60 ). The external reproducibility (2σ) was between 0.000019 and 0.000022.  Total procedural blanks for Nd chemistry were ≤360 pg and for Sr chemistry were ≤4600 pg. In both cases, these blanks were below 1% of the total sample size and therefore considered insignificant.

Data availability
Data generated during this study are available in the PANGAEA database (https://doi. org/10.1594/PANGAEA.904251).