Introduction
There is clear evidence that anthropogenic influences have started to change the Arctic environment1. Constraining the extent and mechanisms of past environmental and oceanic changes in the Arctic region, one of the most sensitive responders to global climate change1, 2, 3, is crucial to better understand both natural and man-made variability. Until recently, however, our understanding of Arctic oceanography before 1 Myr, including the intensification of Northern Hemisphere glaciation (INHG) at 2.7 Myr (refs 4–6), was very limited3, 7, 8. This was mainly due to the technical difficulties that prevented drilling operations in an ice-covered ocean7. The central Arctic sediments recovered in the summer of 2004 during the Arctic Coring Expedition (ACEX; Integrated Ocean Drilling Program, Leg 302) near the North Pole on the Lomonosov ridge (87°5 N, 137° E; 1,250 m water depth) now provide, for the first time, a continuous archive from which Neogene changes of Arctic oceanography and climate can be reconstructed3, 7. The ACEX sediments show clear evidence of ice-rafted transport over the past 15 Myr, but otherwise the sedimentologically monotonous sequence of essentially abiogenic, detrital sediments prevents further microfossil-based paleoceanographic reconstructions3, 7. We report here the Nd isotopic evolution of Arctic Intermediate Water (AIW) recovered from metal-oxide coatings on sediment particles9, 10. The Nd isotopic composition of sea water reflects differences in the isotope composition of the rocks of the surrounding continental land masses (that is, low 143Nd/144Nd reflects old continental crust, whereas high 143Nd/144Nd characterizes young mantle-derived rocks), which are introduced into sea water through weathering processes11. Given that the oceanic residence time of Nd is of the order of 600–2,000 years, Nd isotopes can be used as a quasi-conservative circulation tracer in the open ocean11.
Today, the AIW occupies water depths between 
200 and 1,500 m and is predominantly formed through cooling of the saline surface waters of the North Atlantic Drift (NAD) in the Greenland–Iceland–Norwegian (GIN) Sea12, 13, 14 (Fig. 1). This circulation pattern results in a modern 
Nd (ref. 15) of -10.5 for AIW (P. S. Andersson et al., manuscript in preparation), which we reproduced from the leaches of coatings of various Arctic surface sediments, thus confirming the reliability of the leaching method to extract the past Nd isotope composition of sea water from these sediments10 (see the Supplementary Information).
Figure 1: Schematic map of the high northern latitude seas, ocean circulation and glacial ice sheet distributions.
Light blue arrows indicate the main modern surface currents12, 13, 14: the Gulf Stream, the NAD and the Norwegian Current (NC), which carry an Nd signal of -12.5 to the GIN Sea14(P. S. Andersson et al., manuscript in preparation). Crosses represent regions where modern intermediate water forms, and modern intermediate depth circulation is shown by dark blue arrows12, 13, 14. AIW is, at present, formed in the GIN Seas, and dense brine contributions from the Arctic shelves (dashed arrow) are negligible (P. S. Andersson et al., manuscript in preparation). AIW has a modern Nd of -10.5 (bold value at the ACEX core location, indicated by the star; P. S. Andersson et al., manuscript in preparation). (Nd isotope ratios are expressed as Nd values, corresponding to the measured 143Nd/144Nd normalized to the chondritic uniform reservoir (0.512638), multiplied by 10,000 (ref. 15)). In contrast, the dense water inputs from the Labrador Sea are responsible for NADW having an Nd signature of -13.5 (bold value in North Atlantic)11, 14, 42. Shown in red are the potential supply regions of Nd with a positive Nd signature in the Arctic: the 'Icelandic basalts' and the Putorana basalts, as reflected in the distribution of sediments with >40% smectite content in the Kara Sea49 (the location of the Putorana basalts themselves is indicated by the red 'PB'). Generally during glacial stages, intermediate water formation from North Atlantic surface-water sources shifted south of the GSR40, and brine formation in the Arctic shelf regions was enhanced. The white regions indicate ice distributions for particular times during the last glacial cycle: the purple outline indicates the extension of the Fennoscandian ice sheet at 90 to 80 kyr and the green outline indicates the extension between 22 and 15 kyr, the Last Glacial Maximum47. The borders of the Laurentide ice sheet (dashed blue outline), only a crude approximation, are not relevant for this work.
In pronounced contrast to the modern situation, the ACEX record shows that AIW, with the exception of Late Quaternary interglacial periods, had a significantly more positive Nd value than today throughout the past 15 Myr (Figs 2 and 3). The large and distinct Nd variability presented here documents that AIW formation was subject to major changes both on Myr (Fig. 2) and millenial (Fig. 3) timescales. Below, we discuss the relationship of these changes to tectonic and climatic processes observed in the Arctic region over time.
Figure 2: Arctic Intermediate Water evolution from the Middle Miocene to the present.
(i) NADW Nd data from Fe–Mn crusts in the North Atlantic compiled from various publications11, 32, 33 (open circles and blue envelope). (ii) AIW Nd from ACEX sediment leaches with their 2
external error, illustrating the evolution of AIW before the Quaternary (green circles). The stratigraphic framework of the sediments has been presented previously3, 7. Yellow circles represent core top samples, and the green box indicates the range of Late Quaternary data shown in detail in Fig. 3. Black diamonds are bulk sediment digestion data. (iii) Compiled global benthic foraminiferal 
18O data4 showing the climatically important final shifts (16–11 Myr, and post 3 Myr) of the transition from the 'greenhouse' to the 'icehouse' world. (iv) Periods of enhanced Icelandic plume activity21, 22, 23.
Figure 3: Late Quaternary evolution of AIW.
AIW Nd obtained from sediment leaches of core PS2185 (yellow circles). Core PS2185 (87°31.9 ' N; 144°22.9 ' E; 1,051 m water depth) was recovered at a location neighbouring the ACEX drill site on the Lomonosov ridge and has a very detailed and reliable high-resolution age model for the latest Quaternary45. ACEX Nd data from the Late Quaternary (only measured for some samples older than 200 kyr and not shown here; see text) show similar amplitudes of variation with respect to glacial–interglacial changes in sediment properties7. The 2 external error bars for the Nd data shown are the same as in Fig. 2. Glacial–interglacial cycles are indicated by comparison with the globally stacked benthic foraminiferal 18O data4.
The 'Neogene' record (15 to 2 M yr)
The Nd of AIW became significantly (by 2 Nd units) more positive between 15 Myr and 12 Myr (Fig. 2), followed by a 10 Myr period (until 2 Myr), during which there is no indication for tectonic forcing of AIW circulation (that is, through changes of Arctic sea ways). Because the Pacific and Arctic Oceans were isolated from each other before 5.5 Myr (ref. 16), the interpretation of the ACEX Nd isotope record before this time is constrained by the fact that there are only two sources with positive enough Nd that can have influenced the isotopic composition of AIW: the 'Icelandic basalts' (that is, the basaltic group associated with both Iceland and the Greenland–Scotland ridge, GSR)17, 18 and the Putorana flood basalts of Siberia19, 20 (Fig. 1). It is suggested here that the Putorana flood basalt source, through an indirect control of GSR (Fig. 1) subsidence, offers the best explanation for the early part of the record.
Our preferred scenario is that the opening of the GSR 'gateway', which began well before 15 Myr (refs 21–23), allowed greater penetration of warm, saline North Atlantic surface waters into the GIN seas, thus enhancing heat flow and ultimately moisture supply to higher latitudes of northern Eurasia via the Westerlies (consult ref. 24). This enabled extensive growth of northern Eurasian 'marine-based' ice (either sea ice or floating ice shelves), beginning at 15 Myr. The most likely mechanism to explain the addition of Putorana-sourced Nd to AIW (that is, at depth) is thus a substantial production of dense brines, formed through salt rejection during this sea-ice formation in the Eurasian shelf regions, in particular the Kara Sea region12, 25(P. S. Andersson et al., manuscript in preparation; Fig. 1). Such brines are only produced in relatively small quantities by sea-ice formation in the modern Arctic Ocean26. Brines are, however, extensively generated in the polynyas of the modern Southern Ocean, which form through katabatic winds blowing seawards off the edge of the Antarctic ice sheets and ice shelves27, 28. In the modern Southern Ocean, brine formation processes are mainly responsible for the formation of dense Antarctic Bottom Water27, 28, demonstrating the potential of this mechanism to generate deep waters from saline surface waters in high latitudes. In the case of the Arctic Ocean, a prerequisite for the enhancement of brine production at 15 Myr was the opening of the Fram Strait to the North Atlantic at about 17.5 Myr (ref. 29), which allowed the establishment of open-ocean salinities. An alternative explanation simply invoking enhanced inputs of radiogenic Nd via rivers draining the flood basalts into the Arctic Basin is unlikely owing to the missing mechanism for subduction of a freshwater Nd isotope signature to AIW depths.
As a consequence of the above scenario, it is suggested that the pronounced increase in global benthic 18O from the 'mid-Miocene climatic optimum' (15–17 Myr) to around 10 Myr (ref. 4; Fig. 2) may reflect, at least in part4, 30, 31, the initial growth of significant Eurasian Shelf/Arctic Ocean ice. This is consistent with the record of continuous ice-rafted debris deposition observed in the ACEX core itself3, 7.
It has been shown that in the modern ocean, exchange of Nd between sea water and the Icelandic basalts can alter the Nd isotopic composition of bottom waters14, 18. However, besides the considerations that modern deep water flows southward across the GSR and the large distance of the GSR from the Arctic Ocean, two other observations argue against an involvement of the Iceland basalts in the change of AIW Nd signature between 15 and 12 Myr (Fig. 2): (1) the Nd isotope composition of the North Atlantic11, 32, 33 does not vary with fluctuations in the volcanic activity of the Iceland basalts, as inferred from the history of the GSR plume activity21, 22, 23, and (2) deep-water exchange through Fram Strait was probably still restricted before 12 Myr (refs 23,34). Moreover, even with an open Fram Strait, such as today, GSR-influenced bottom waters are not transferred into the Arctic Basin (P. S. Andersson et al., manuscript in preparation). This leaves the Putorana basalts as the only likely source of radiogenic Nd for AIW, as discussed above.
Volumetric constraints on water mass exchange
Mass balance calculations, assuming a simple two end-member mixing between brine water (Nd=+2; typical for the Putorana basalt signature19, 20) and North-Atlantic-sourced intermediate water (Nd=-10; typical of the NADW record between 16 and 8 Myr (refs 11,32,33); Fig. 2), imply the contribution of large proportions (up to 30%) of brine-water-sourced AIW by 12 Myr. To achieve such mixing ratios, a severe concomitant restriction of Atlantic NAD-sourced intermediate water input (with its negative Nd signature) must have existed. The Nd isotope data clearly indicate that intermediate water exchange between the North Atlantic and the Arctic was limited throughout the period from 12 to 2 Myr (Fig. 2). Applying the end-member isotope compositions listed above, a reduction in the 6 Sv modern flux of North-Atlantic-sourced intermediate water to the Arctic35 to 3 Sv, together with a brine-sourced flux of 0.6 Sv would arrive at an Nd of -8 for AIW. There is evidence that past restrictions of North Atlantic water flow into the Arctic were even more severe than this 50% estimate36, 37. However, regardless of the exact values, brine formation fluxes of this order of magnitude (0.6 Sv) have been observed in modern interglacial oceans26, 27, 28, 38, 39, and demonstrate that the suggested magnitude of past-intensified brine formation on the Arctic Shelves together with a concomitantly restricted North Atlantic inflow is realistic.
The most likely cause of such a reduction in North Atlantic water inflow was a southward shift in the location of surface-water subduction in the North Atlantic to a position south of the GSR (that is, not in the GIN Seas). This situation is similar to that predicted for the glacial periods of the Quaternary40 (described below), the main difference being that, according to our data, this circulation pattern was a generally stable oceanographic configuration between 12 and 2 Myr. It is important to note that, in contrast to the Nd isotope signature of North Atlantic Deep Water (NADW) and its precursors in the North Atlantic11, 32, 33 (Fig. 2), the AIW Nd was remarkably stable between 4 and 2 Myr. This contrast implies (1) that the INHG at 2.7 Myr (refs 4–6) was not driven by changes in the Arctic Ocean, and (2) that the decrease in the North Atlantic Nd record must reflect changes in contributions from non-radiogenic Labrador Sea sources41, 42. This is consistent with the INHG being largely restricted to the Laurentide glaciers at this time, which increased the input of non-radiogenic Nd to the Labrador Sea, but obviously did not have a strong influence on the Nd isotope composition of waters reaching the GIN seas and thus AIW.
The 'Quaternary' record (2 Myr to present)
The significant 'excursion' to Nd values as positive as -6 around 2 Myr (Fig. 2) is ascribed to the growth of the first major North Eurasian ice sheets, which grounded on large areas of the Arctic shelves, including the Kara Sea area, thus transferring increased amounts of Nd with a positive isotope signature into the Arctic Ocean. Analogous to the earlier Neogene part of the record, this 'excursion' is explained by enhanced brine-water production resulting from increased sea-ice formation at the edges of the ice sheets and concomitantly restricted North-Atlantic-sourced intermediate-water input. However, the magnitude of the Nd variation of this 'event', as well as the amplitude of variations during the subsequent glacial–interglacial cycles, demonstrates that this mode of circulation intensified after 2 Myr.
The only other available Arctic sea-water Nd isotope data for the past 5 Myr, recovered discontinuously from Fe–Mn micronodules of various water depths in the Canadian Basin, also indicate a major decrease in Nd isotope composition around 2 Myr (ref. 8), supporting the contention that the oceanographic stability in the Arctic Basin of the preceding 10 Myr was disrupted at this time. Unfortunately, both the uncertainties in the age models of the sediments from which these data were obtained43 and the different water depths that they reflect8 limit the use of these data for direct comparison to the data presented here. Indeed, accurately dating Arctic samples is still a major challenge, and whereas the Neogene ACEX chronology used here is considered robust44, the Quaternary ACEX stratigraphy is not as well defined. For this reason, the latest Quaternary record (the past 400 kyr; Fig. 3) was obtained from sediment samples of a neighbouring core on the Lomonosov ridge (PS2185), for which a robust age model is available45.
The 'Quaternary' Lomonosov ridge data presented here indicate that at 2 Myr a threshold of Arctic glaciation and oceanography was reached, resulting in a Nd isotope variability with a large (4 Nd-unit) amplitude on millennial timescales (Figs 2 and 3). These variations reflect switches between an 'interglacial mode' of AIW circulation, characterized by the modern configuration (Fig. 1), and a 'glacial mode'. During the 'glacial mode', AIW was controlled by enhanced input of shelf waters from brine sources, in particular the Kara Sea, and a coeval strongly restricted input from North Atlantic sources. This configuration is essentially identical to the circulation described previously for the 'Neogene' record. That is, in this 'glacial mode,' Atlantic-sourced intermediate water formed further south in the North Atlantic40, and thus had limited influence on the Arctic Basin owing to the barrier of the GSR, a restriction that was enhanced during the glacial Quaternary drops in sea level. This restricted Atlantic water inflow also caused the Quaternary Arctic Ocean to be less well ventilated during glacials, consistent with observations of cyclic Mn enrichments in Arctic sediments46.
An important feature of the changes in Late Quaternary AIW formation is evident in the Nd isotope record during Marine Isotope Stage 3, between 59 and 24 kyr: AIW formation seems to have changed to the 'interglacial mode' 30 kyr before the Last Glacial Maximum (15–22 kyr; Fig. 3). This unexpectedly early shift can be explained by the fact that the Kara Sea shelf region was not covered by an ice sheet after 50 kyr (Fig. 1; ref. 47), thus inhibiting brine formation, and the consequent input of Nd with a positive Nd signature to intermediate depths, in this area. The striking correspondence in timing of the change in AIW signature and the independently derived terrestrial glaciation record47 lends strong support for the proposed mechanism of interaction between variable dense brine production on the Arctic shelves and changes in North-Atlantic-sourced intermediate-water inputs to explain AIW variability.
The Nd isotope data and the scenario presented here provide the first framework for a Neogene oceanographic and climatic history of the Arctic. Although the history of the Arctic Ocean is far from resolved through this one record, the large Nd isotope variability observed confirms the high sensitivity of the Arctic Ocean to climate forcing mechanisms on all timescales.
Methods
The sea-water-derived Nd in the metal-oxide coatings of the sediment was extracted by leaching, following the protocols in refs 9,10, but modified in the following ways: (1) a buffered acid leach step was omitted as the Arctic sediments are virtually devoid of a carbonate component; (2) the leach solution to extract the oxide phase (a dilute reducing/complexing solution of hydroxylamine.HCl and acetic acid, buffered to pH=4) was diluted 10-fold compared with that of ref. 10 to avoid any contamination caused by leaching of clays. The leachate solution was separated from the sediment, evaporated in double-distilled concentrated HNO3 and then run through a standard ion-exchange procedure to separate and purify the Nd for analysis (AG 50W-X12 resin for cation separation; di-(2-ethylhexyl)phosphate resin for rare-earth element separation). The procedural blank was negligible. The Nd isotopic composition was measured at the mass spectrometry facility of IFM-GEOMAR using a Thermo Triton thermal ionization mass spectrometer, correcting for instrumental fractionation using 146Nd/144Nd=0.7219 (ref. 15). The standard deviation of 27 runs of a Nd solution (SPEX), diluted to the same concentrations as the samples, resulted in the stated 2 external reproducibility (0.50 Nd units), although the internal error given by replicate runs of the JNdi-1 standard48 was better (0.3 Nd units; n=39). A suite of Arctic core top samples replicated the modern Arctic Deep Water Nd signature within error (see the Supplementary Information), demonstrating that a pure sea-water signal was extracted with the applied protocol. To further assure that no Nd was leached from the clays, several Nd and Sr isotopic analyses of complete bulk sediment dissolutions were made, after the sediments had undergone leaching, for comparison with the leached Sr and Nd isotope compositions10. In all cases, the potential clay-derived Nd contribution was less than our stated measurement error (see the Supplementary Information), corroborating the extraction of a pure sea-water signal.
Author contributions
Samples given to M.F. (ACEX core) and taken by R.F.S. (PS2185) were analysed by B.A.H. at the mass spectrometry facility of IFM-GEOMAR run by A.E. All authors contributed equally to the discussion and interpretation of the results.
This article has a related Backstory in this issue of Nature Geoscience.
