Oxygen depletion recorded in upper waters of the glacial Southern Ocean

Oxygen depletion in the upper ocean is commonly associated with poor ventilation and storage of respired carbon, potentially linked to atmospheric CO2 levels. Iodine to calcium ratios (I/Ca) in recent planktonic foraminifera suggest that values less than ∼2.5 μmol mol−1 indicate the presence of O2-depleted water. Here we apply this proxy to estimate past dissolved oxygen concentrations in the near surface waters of the currently well-oxygenated Southern Ocean, which played a critical role in carbon sequestration during glacial times. A down-core planktonic I/Ca record from south of the Antarctic Polar Front (APF) suggests that minimum O2 concentrations in the upper ocean fell below 70 μmol kg−1 during the last two glacial periods, indicating persistent glacial O2 depletion at the heart of the carbon engine of the Earth's climate system. These new estimates of past ocean oxygenation variability may assist in resolving mechanisms responsible for the much-debated ice-age atmospheric CO2 decline.

T he Southern Ocean is widely considered to be critical to global nutrient and carbon cycling, including over glacialinterglacial time scales 1 . As an area of incomplete nutrient utilization, it is a major source of CO 2 to the atmosphere today 2 . At present, old (CO 2 À and nutrient-rich and relatively O 2 À depleted) deep waters upwell along most of the Antarctic continental margin 3,4 (Fig. 1), release CO 2 into, and recharge O 2 from surface waters before they down-well in distinct areas, such as the Weddell and Ross seas, to form Antarctic Bottom Water (AABW). In the glacial Southern Ocean, strengthening of the biological pump due to enhanced iron supply 5,6 , increased stratification 7 , and expanded sea-ice cover 8 , were among the dominant players in reducing atmospheric CO 2 by B90 p.p.m.V. Each of these mechanisms could counterbalance the increased O 2 solubility due to lower glacial temperatures, leading to a reduction in the O 2 concentration of the seawater. Since the Southern Ocean is thought to have reduced its CO 2 leakage during glacial periods 1 , it provides an ideal location to search for evidence of deoxygenation linked to CO 2 sequestration in the upper ocean.
During the last glacial period, deep waters surrounding Antarctica were less ventilated, and older than today (relative to the atmosphere) 9 . A recent quantitative O 2 proxy study based on benthic foraminiferal d 13 C indicates that decreased ventilation linked to a reorganization of glacial ocean circulation and a strengthened global biological pump significantly enhanced the ocean storage of respired carbon in the deep North Atlantic 10 . Early box-models hypothesized very low-oxygen levels in the high latitude Southern Ocean 11,12 . Proxies did not paint a clear picture for bottom-water O 2 concentrations in the glacial Southern Ocean 13 . Only a few studies on marine sediment cores south of the APF have found evidence for substantially lowered bottom water O 2 concentrations. There, authigenic uranium concentrations were elevated in sediments deposited during glacial Marine Isotope Stages (MIS) 2 and 6 (refs 14,15). By contrast, another study highlighted a transient stagnation event during the early stage of the last interglacial (MIS 5e) 16 .
Bottom water or porewater redox proxies cannot capture upper ocean O 2 levels far from the continental shelf, so there is scant constraint on upper ocean oxygenation conditions in vast tracts of the open ocean 13 . A novel proxy, the I/Ca composition of marine carbonates, especially planktonic and benthic foraminiferal tests, has demonstrated its potential to reconstruct paleo-oxygenation levels in both the upper ocean 17-20 and bottom waters 21 , respectively. The thermodynamically stable forms of iodine in seawater are iodate (IO 3 À ) and iodide (I À ) 22 .
The total concentrations of IO 3 À and I À are relatively uniform in the world ocean at around 0.45 mmol l À 1 due to the residence time of B300 kyr (ref. 23), supported by a more recent compilation of iodine concentrations in global rivers 24 . Therefore, the total iodine concentration in the global ocean likely remained invariant over the duration of a glacial termination (B6 kyr).
Iodate is taken up by marine organisms as a micronutrient in surface waters 25 , but its concentration does not increase during the aging of deep waters 26,27 , in contrast to those of the major nutrients nitrate and phosphate, probably due to the low I/C org ratio of plankton 25 . Iodine speciation is strongly redox sensitive. IO 3 À is completely converted to I À when oxygen is depleted 28 . Because IO 3 À is the only chemical form of iodine that is incorporated into the structure of carbonate 17 , calcareous tests precipitated closer to an oxygen minimum zone (OMZ) will record lower I/Ca and vice versa. An OMZ is defined by O 2 o20 mmol kg À 1 in the Pacific Ocean and O 2 o50 mmol kg À 1 in the Atlantic Ocean 29 .
In this paper, we use recent planktonic foraminifera and modern water column data to establish typical I/Ca values for the presence of an OMZ or O 2 -depleted water. On the basis of this proxy development, the down-core record of planktonic foraminifera I/Ca obtained at site TC493/PS2547 indicates the persistent presence of oxygen-depletion in the upper waters of high latitude Southern Ocean during the last two glacial periods.

Results
Site selection. We measured I/Ca values on eleven planktonic foraminiferal species in modern to Holocene samples, and in one sample from a previous interglacial (Supplementary Table 1 and Supplementary Figs 1 and 2). We chose sites from well-oxygenated areas (for example, the North and sub-Antarctic South Atlantic), and sites located beneath OMZs, including Ocean Drilling Program (ODP) Sites 658, 709, 720 (Site 720: last interglacial samples), 849 and 1242. First, we use these data to further establish the foundations of the I/Ca proxy. Subsequently, we focus on an I/Ca down-core record on Neogloboquadrina pachyderma sinistral deposited during the last two glacial cycles in two sediment cores (PS2547 and TC493) recovered from the same location (71°09 0 S, 119°55 0 W, water depth 2,096 m) on a seamount in the Amundsen Sea ( Fig. 1) 30 . The excellent carbonate preservation at this site 30 provides a unique window to reconstruct past upper ocean conditions south of the APF. Site TC493/PS2547 is currently bathed by Circumpolar Deep Water (CDW), which is overlain by a layer of Antarctic Surface Water (AASW), or Winter Water 31-33 , and is located on the edge of the average modern summer sea-ice limit 34 (Fig. 1). During the Last Glacial Maximum (LGM), the sea-ice boundaries within the Southern Ocean shifted significantly northwards 35,36 . Thus, it is highly likely that site TC493/PS2547 was located within the permanent sea-ice zone during past glacial periods 34 .
Age model and glacial polynyas. The sediments of core TC493/ PS2547 consist mainly of foraminiferal ooze and sandy mud, with N. pachyderma (s) tests forming the primary carbonate component 30  This scenario is consistent with the occurrence of the benthic foraminifera species Epistominella exigua, which is adapted to highly episodic phytodetritus supply 40 .
I/Ca in foraminifera. I/Ca values in the modern and late Holocene samples are lower than B2.5 mmol mol À 1 at sites with O 2 minima o70 mmol kg À 1 in the upper ocean (0-500 m) (Fig. 2a).   depth profiles at OMZ sites from the Eastern Equatorial Pacific (EEP) 28 and the Arabian Sea (station N8) 48 and at a well-oxygenated high-latitude site near the Weddell Sea (station PS71/179-1) 54 .

Discussion
A tremendous amount of work has been devoted to developing foraminiferal proxies for temperature and pH, using global calibrations derived from core-top samples (for example, the Mg/ Ca seawater temperature proxy 41 ). Low I/Ca ratios of planktonic foraminifera unambiguously reveal the presence of low-oxygen waters, but a global calibration approach cannot establish planktonic foraminifera I/Ca as a linearly quantitative proxy for the continuum of dissolved O 2 concentration. Due to the stepwise nature of redox reactions 42 , quantitative IO 3 À reduction does not occur before the dissolved oxygen is depleted to a certain threshold value, triggering nitrate reduction 43 . IO 3 À concentrations at water depths matching planktonic foraminiferal habitats are often not in equilibrium with the in situ O 2 concentrations, and O 2 contents which are sufficiently low to initiate major IO 3 À reduction may be detrimental to many species 44 . Instead, the I/Ca (recording the in situ IO 3 À concentration) is determined by the depth habitat of the foraminifera and the upper ocean IO 3 À mixing gradient. This mixing gradient is largely controlled by the surface water IO 3 À concentration and the depth of the IO 3 À reduction zone 28 . Nonetheless, a planktonic foraminifera proxy that semiquantitatively approximates dissolved O 2 concentrations, indicative of the presence of an OMZ, can still be highly valuable for the paleoceanography community.
Before interpreting the down-core record from site TC493/ PS2547, we identify the characteristic I/Ca signals for modern OMZs.  (2) the normal open ocean type (for example, in a well-oxygenated water column), with relatively high surface water values and even higher subsurface values. A threshold O 2 concentration will cause complete IO 3 À reduction in the subsurface, and there may be a surface water IO 3 À threshold concentration below which complete IO 3 À reduction is likely to happen in the water column. Combined with modern water column IO 3 À and O 2 data, the I/Ca values measured on modern and late Holocene planktonic foraminifera consistently indicate that I/Ca o2.5 mmol mol À 1 is equivalent to a surface water IO 3 À concentration of o0.25 mmol l À 1 , thus providing a marker for the presence of oxygen-depleted water with a subsurface O 2 concentration o20-70 mmol kg À 1 (Fig. 2a-c). Modern surface water IO 3 À concentrations are influenced by productivity and the presence of a subsurface OMZ 25,28 . To visualize this relationship, we compiled surface water IO 3 À concentrations from the literature and plotted them against the minimum O 2 concentrations in the subsurface water (Fig. 2b). The IO 3 À concentration broadly increases with the minimum O 2 concentration when the surface water IO 3 À concentration is 40.25 mmol l À 1 (Fig. 2b). This correlation is likely a reflection of surface productivity versus subsurface respiration, because lower productivity leads to lower iodine uptake in surface water and less oxygen consumption by subsurface organic matter decomposition. In areas with a strong OMZ and near-zero O 2 values, the surface water IO 3 À concentrations are below 0.25 mmol l À 1 (Fig. 2b) 17,20 . Using this K d value, an IO 3 À concentration oB0.25 mmol l À 1 results in I/Ca values oB2.5 mmol mol À 1 in calcite. This estimate is consistent with modern I/Ca at OMZ Sites 658, 849 and 1242, as well as the last interglacial I/Ca value at Site 720 (Fig. 2a). Therefore, a surface water I/Ca value o2.5 mmol mol À 1 indicates that a pronounced subsurface O 2 minimum exerted the dominant control on the upper ocean IO 3 À profile. This I/Ca threshold value does not seem to depend on foraminiferal species (Fig. 2a).
The O 2 threshold for maintaining an OMZ-type IO 3 À profile is useful for the paleoceanographic application of the planktonic I/Ca proxy. At O 2 concentrations o20 mmol kg À 1 , microbial processes become dominant 29 , and IO 3 À likely would be completely reduced to I À since the reaction is biologically mediated 45 (for example, ODP Sites 1242, 720 and 849 in Fig. 2a,c). ODP Site 658 is located at the northern edge of a shallow pocket of distinctively low-oxygen water with mean O 2 concentrations of B70 mmol kg À 1 in the upper 200 m (ref. 46), which may be sufficiently low to generate an OMZ-type iodate profile. Three species of planktonic foraminifera analysed at ODP Site 1242 show exceptionally low I/Ca ratios around 0.5 mmol mol À 1 , corresponding to an IO 3 À concentration of B0.05 mmol l À 1 . Such a low IO 3 À concentration is comparable to that reported for a location where an extreme hypoxic event occurred 47 . Moreover, this low IO 3 À concentration implies that IO 3 À reduction should occur shallower than at Site 849 and at two sites with classic OMZ-type IO 3 À profiles (Eastern Equatorial Pacific 28 and Arabian Sea 48 ; Fig. 2c). A comparison of the O 2 profiles of these sites reveals that the O 2 threshold needs to be 450 mmol kg À 1 to achieve a shallower IO 3 À reduction at Site 1242. Therefore, we suggest that I/Ca values lower than B2.5 mmol mol À 1 indicate O 2 minima o20-70 mmol l À 1 . This O 2 range cannot be further narrowed down with the available information, and we refer to this range as the O 2 threshold for an OMZ-type IO 3 À profile. However, the threshold behaviour of IO 3 À reduction (relative to O 2 ) in subsurface waters does not necessarily lead to step changes in down-core records of planktonic I/Ca. This is because planktonic foraminifera typically record the IO 3 À mixing gradient in the top part of water column, above the O 2 -depleted zone where rapid step changes in IO 3 À concentrations occur. Low planktonic I/Ca values may be driven by shoaling of O 2 -depleted water, and/or by increasing productivity, both of which could change gradually over time.
The available data from modern and late Holocene planktonic foraminifera suggest that the I/Ca ratio acts as a robust (paleo-) proxy for determining the signature of O 2 -depletion in the upper ocean (Fig. 2). At site TC493/PS2547, I/Ca was high (5-7 mmol mol À 1 ) during the Holocene and interglacial MIS 5 when compared with the lowest values (o2 mmol mol À 1 ) characterizing peak glacial periods MIS 2 and 6 ( Fig. 3). Changes in salinity, temperature and foraminiferal habitat, most likely, are not the main drivers for this record (Supplementary Discussion). The glacial I/Ca values of N. pachyderma (s) are best explained by the presence of a water mass with a dissolved O 2 content o70 mmol kg À 1 close to, i.e., above or near, this site (Figs 2 and  3). We reiterate that the low I/Ca does not necessarily imply O 2depleted seawater within the foraminiferal habitat.
At present, CDW wells up to a water depth of approximately 250-300 m in the Amundsen Sea 31 and has O 2 concentrations notably lower than the top 200 m of the water column (Fig. 2c). Although the interpretation of absolute values of planktonic d 13 C is far from straightforward in the seasonal ice zone (for example, disequilibrium from seawater 49 ), it is reasonable to assume that CDW had a strong influence on the local water column during glacial periods, as its upwelling along the continental margin was probably responsible for the opening of the glacial polynyas. The CDW upwelling at site TC493/PS2547 today partly originates from Pacific Deep Water (PDW) moving southward from the equator, with a low-oxygen and high nutrient signature  Fig. 1) 50,51 . d 30 Si data from fossil diatoms and sponges indicate higher silicic acid concentrations in the Pacific sector of the Southern Ocean during the LGM, which further imply that either the southward transport of PDW was more efficient or PDW was less ventilated than today 52 . So glacial CDW was likely more O 2 depleted than during interglacials, and upwelling of this water contributed to the glacial I/Ca signal at site TC493/PS2547. The oxidation of I À to IO 3 À is thought to take from a few months up to 40 years 53 . Long-distance transport of welloxygenated deep water with low IO 3 À concentrations (o0.25 mmol l À 1 ) has not been documented in the modern ocean, but this scenario should be tested with further work on I À oxidation kinetics. Today our site is bathed in CDW transported from a Pacific OMZ and the interglacial I/Ca values at site TC493/PS2547 do not show any remnant signal of the OMZ from the Pacific Ocean. On the basis of the knowledge about iodine speciation change in modern ocean, we interpret the observed glacial I/Ca values as a local signal, in principle, indicating the presence of a water mass with low O 2 and low IO 3 À vertically or horizontally close to the planktonic foraminiferal habitat.
In the setting of site TC493/PS2547 a coherent conceptual model for N. pachyderma (s) recording the presence/absence of O 2 -depletion needs to integrate changes in productivity, sea-ice extent and the opening/closing of polynyas on time scales of glacial to seasonal cycles (Fig. 4). Although the polynyas complicate the interpretation of the proxy data, their presence arguably provides the only window for sufficient accumulation of planktonic microfossils to record upper ocean conditions during glacial periods at such high latitudes.
The modern O 2 profile at site TC493/PS2547 is defined by equilibration with the atmosphere at 0-250 m, and CDW influence below 250 m, as shown by the distinctively low O 2 concentrations (Fig. 2c). With O 2 above the threshold for complete IO 3 À reduction in the entire water column, the IO 3 À profile at site TC493/PS2547 should be similar to those at other high latitude locations, for example, site PS71/179-1 at 69°31 0 S and 0°3 0 W in the Weddell Sea 54 (Fig. 2d). An interglacial scenario of relatively high seasonal productivity, high O 2 and surface water IO 3 À (40.3 mmol l À 1 ) concentrations (Fig. 4a), is well described for the modern Atlantic sector of Southern Ocean 54 .
Relative to the interglacial periods, the Southern Ocean experienced expanded sea-ice cover during glacial periods, and was less ventilated 9,36 . A more dynamic seasonal sea-ice cycle during ice ages would have increased water column stratification. Increased winter sea-ice formation (spatially and volumetrically) may have generated waters dense enough to sink ultimately to the bottom of the ocean 55 . On the other hand, melting of thicker sea ice during glacial-time summers in the seasonal sea-ice zone would have strengthened the halocline (not considering the influence of polynyas). So, the glacial seasonal stratification was likely stronger than today. These factors overall should have lowered the glacial O 2 concentrations in the Southern Ocean (Fig. 4b). At site TC493/PS2547, glacial I/Ca demonstrate that the IO 3 À profile was OMZ-like with complete IO 3 À reduction near the foraminiferal habitat (Fig. 2d). However, the dynamics of polynyas must be considered when interpreting the location of the low O 2 water mass, and the means by which the signal was recorded by N. pachyderma (s).
Without a polynya above site TC493/PS2547, glacial phytoplankton productivity under perennial sea-ice cover would have been relatively low due to the scarcity of light 34 , and planktonic foraminifera depending on algae could not flourish. The water column would have been relatively poorly ventilated and strongly stratified during these times, creating the ideal environment for developing low O 2 conditions and an OMZ-type IO 3 À profile (Fig. 4b). The episodic opening of a polynya re-established primary production (mainly by diatoms) and thus a planktonic foraminiferal habitat, vertical mixing and oxygenation in, at least, the uppermost part of the water column (Fig. 4c). While overall glacial-time production was reduced 30,34 , the planktonic foraminifera preserved in the glacial sediments probably recorded transient I/Ca changes in the water column associated with polynya-induced peaks in glacial productivity. Modern open ocean productivity pulses do not lower IO 3 À concentrations to o0.25 mmol l À 1 in oxygenated water (Supplementary Discussion) 54 , thus the glacial I/Ca signal is most likely driven by changes in O 2 and not productivity.
The likely short-lived nature of glacial polynyas makes it difficult to envisage that very brief plankton blooms alone could produce a utilization-driven O 2 depletion in a cold, welloxygenated Southern Ocean. For the same reason, it is difficult to imagine that the vertical mixing cells restricted by the size of the polynya could rapidly oxygenate voluminous nearby waters outside of the polynya, if most of the sea-ice covered areas were O 2 -depleted. The more likely scenario is that the O 2 concentrations in the deep and abyssal Southern Ocean were generally lower during glacial periods than during interglacial periods. Upwelling of a more O 2 -depleted CDW in the generally stratified upper ocean was mainly responsible for the IO 3 À reduction at site TC493/PS2547 (Fig. 4b), while the episodic opening of polynyas created habitable conditions for planktonic foraminifera to record the deoxygenation in the upper ocean (Fig. 4c). We suggest that the I/Ca proxy should be used as a local proxy, in principle. However, it is probably a reasonable speculation that this record (Fig. 3) shows oxygenation changes integrated over a regional volume of water (e.g. CDW).
The timing of glacial deoxygenation and deglacial reoxygenation at site PS2547 shows potential linkages to global climate changes (Fig. 3). The appearance of OMZ-type I/Ca values (oB2.5 mmol mol À 1 ) during past glacial periods coincided with the lowering in atmospheric pCO 2 level below the long-term mean value 56 . Identical timing was reported for a strongly stratified Antarctic Zone coincident with pCO 2 decrease under the same threshold value (225 p.p.m.) in the Atlantic sector of the Southern Ocean 57 . Stronger stratification may be the common driving force for the productivity change (ODP Site 1094) and oxygenation change (PS2547/ TC493) in the Antarctic zone. Furthermore, during the last interglacial period, the recovery of N. pachyderma (s) I/Ca values is offset from the d 18 O trend, with peak I/Ca occurring about 10 kyr after the peak d 18 O (Fig. 3), an observation worthy of future investigation.
Our I/Ca results build on other evidence 52,58,59 to make a stronger case for lower oxygen concentrations in CDW (and very likely PDW) during glacial periods. Altogether with the reconstructed O 2 content of deep waters in the glacial North Atlantic 10 , these observations seem to allude to large scale deoxygenation in the glacial global ocean interior 60 . Future work providing quantitative reconstructions of bottom water O 2 concentrations in the Southern Ocean, especially south of the APF, and in other major ocean basins will shed new light on the mechanisms of sequestering atmospheric CO 2 during ice ages.

Methods
Foraminifera cleaning. Sediments were sampled from the split core sections and wet sieved. Approximately 40 tests of N. pachyderma sinistral were picked from the 200-250 mm size fraction of each sample. The cleaning procedure followed the Mg/Ca protocol of Barker et al. 61 . Cleaned glass slides were used to gently crack open all chambers. Clay particles were removed in an ultrasonic water bath. After adding NaOH-buffered 1% H 2 O 2 solutions the samples were heated in boiling water for 10-20 min to remove organic matter. Calcareous microfossils were then thoroughly rinsed with de-ionized water. Reductive cleaning was not applied because contribution of iodine from Mn-oxides is deemed negligible 19 .
ICP-MS measurements. The cleaned samples were dissolved in 3% nitric acid, and diluted to solutions with 50 p.p.m. Ca for analyses. Iodine calibration standards were freshly prepared also with 50 p.p.m. Ca. 0.5% tertiary amine solution (Spectrasol, CFA-C) was added to stabilize iodine within a few minutes after the sample dissolution. The measurements were performed immediately after that to minimize potential iodine loss. The sensitivity of iodine was tuned to above 80 kcps for a 1 p.p.b. standard. The precision for 127 I is typically better than 1%. The long-term accuracy is guaranteed by frequently repeated analyses of the reference material JCp-1 (ref. 17). The detection limit of I/Ca is on the order of 0.1 mmol mol À 1 . The I/Ca measurements were performed using a quadrupole ICP-MS (Bruker M90) at Syracuse University.