Changes in North Atlantic deep-water oxygenation across the Middle Pleistocene Transition

The oxygen concentrations of oceanic deep-water and atmospheric carbon dioxide (pCO2) are intrinsically linked through organic carbon remineralization and storage as dissolved inorganic carbon in the deep sea. We present a high-resolution reconstruction of relative changes in oxygen concentration in the deep North Atlantic for the past 1.5 million years using the carbon isotope gradient between epifaunal and infaunal benthic foraminifera species as a proxy for paleo-oxygen. We report a significant (>40 micromole per kilogram) reduction in glacial Atlantic deep-water oxygenation at ~960 thousand to 900 thousand years ago that coincided with increased continental ice volume and a major change in ocean thermohaline circulation. Paleo-oxygen results support a scenario of decreasing deep-water oxygen concentrations, increased respired carbon storage, and a reduction in glacial pCO2 across the Middle Pleistocene Transition. Description Deep concentration Between about 1.25 million and 800,000 years ago, the climate system went through a major change during a period called the Middle Pleistocene Transition. Was the inventory of dissolved oxygen in the ocean affected by this episode? Thomas et al. show that oxygen concentrations in glacial deep North Atlantic waters suffered a stepped reduction about 900,000 years ago, coincident with reductions in the concentration of glacial atmospheric carbon dioxide and global ice volume. —HJS Glacial deep-sea oxygen concentrations, fell along with ice volume and atmospheric pCO2, around 900 thousand years ago.

D uring the past 800 thousand years (kyr) glacial carbon dioxide concentrations (pCO 2 ) in Earth's atmosphere aver-aged~190 parts per million (ppm) by volume (1), lower than the preindustrial value by~90 ppm (1,2). Aside from ice core records, atmospheric pCO 2 records are fragmentary but suggest that glacial pCO 2 may have been higher by 20 to 40 ppm prior tõ 1000 to 800 thousand years ago (ka) during the early Pleistocene (3)(4)(5)(6)(7). Such a drop in glacial atmospheric pCO 2 concentrations is one of the proposed causes of increased continental ice volume during glacial periods across the Middle Pleistocene Transition (MPT) (8,9), which occurred between 1.25 and 0.64 million years ago (Ma) (10). During the MPT ice sheets grew larger and the duration of glacial cycles increased from primarily 41-kyr oscillations before the MPT to quasi-100-kyr cycles afterward (8,10,11). Lowering glacial pCO 2 across the MPT most likely involved increased carbon storage in the deep ocean through enhanced biological CO 2 uptake (12)(13)(14) and/or reduced CO 2 exchange between the atmosphere and surface ocean (12,15,16) especially in the Southern Ocean (17,18). These processes would not only increase carbon storage in the deep ocean but would also reduce the deep-sea concentration of dissolved oxygen.
We applied a paleoproxy to estimate past changes in oxygen concentration (hereafter expressed as [O 2 ]) in deep water of the North Atlantic. The proxy uses an empirical calibration between the [O 2 ] and the carbon isotopic composition difference (Dd 13 C) between surface and deep-dwelling benthic foraminifera, as first proposed by McCorkle et al. (19,20) and recalibrated by Hoogakker et al. (21) 1; supplementary materials). We measured the epibenthic species Cibicidoidies wuellerstorfi that provides a record of deepwater carbon isotopic composition (d 13 C) (21,24), whereas the infaunal species Globobulimina affinis provides a record of d 13 C near the oxic-anoxic boundary in sediment porewaters (21,24,25). The carbon isotope gradient (Dd 13 C cib-aff ) expresses the difference between the two.
The main assumption of the Dd 13 C cib-aff derived [O 2 ] method is that the carbon isotope gradient between the sediment water interface and the oxic-anoxic boundary is controlled by deep-water [O 2 ] and hence total dissolved inorganic carbon (DIC) released during aerobic respiration. Addition of excess DIC through anaerobic processes such as nitrate reduction or sulfate reduction serves to lower d 13 C at the oxic-anoxic boundary (20,26)  Low organic carbon concentrations (<1%) at Site U1385 leads to relatively low rates of organic carbon oxidation and a deep sulfatemethane transition at~50 meters below the seafloor (mbsf) (27). The Iberian Margin nearsurface sediment column demonstrates low rates of sulfate reduction in the upper tens of centimeters (28), suggesting that Dd 13 C at Site U1385 is governed predominantly by the amount of aerobic respiration controlled by deepwater [O 2 ]. According to recommendations of Jacobel et al. (26), we compare the paleo-[O 2 ] estimates derived from Dd 13 C cib-aff with other redox/oxygenation proxies: U/Ca, U/Mn, and percent C 26 OH/(C 26 OH + C 29 ). We also as-sess relative changes in paleoproductivity that may lead to variable organic carbon fluxes using several proxies: Dd 13 C cib-uvig , Uvigerina spp. abundance, and sedimentary Ba/Al ratios (see supplementary text, sections 3 and 4).
The cores from different holes at Site U1385 were spliced to produce a composite section that extends from 0 to 166. 5 (Fig. 1, fig. S1, and table S1) and the d 13 C value of DIC averages~1‰ (Fig. 1). During glacial stages the relative proportion of northern-sourced deep water decreased at Site U1385 as the fraction of southern-sourced deep water increased (30)(31)(32)(33).
Hoogakker et al. (21) applied the Dd 13 C cib-aff method to estimate [O 2 ] for the past 160 kyr using piston core MD95-2042 (3146 mbsl), which is slightly deeper than nearby Site U1385 but has very similar modern-day oxygen concentrations (~245 mmol kg −1 ). They concluded that deep-water [O 2 ] was lower during the last two glacial stages relative to today, with the lowest values (~100 mmol kg −1 lower) recorded during cold stadial conditions associated with Heinrich events. With Site U1385, we extend the [O 2 ] record of MD95-2042 beyond the last glacial cycle to span the past 1.5 Ma (Fig. 2 S2) (25).
We assess the reliability of the Dd 13 C cib-aff proxy using a transect of core-top sediment recovered from the Iberian Margin during Cruise JC089 (22) S6) but remain higher than those centered on 900 ka.
The relationship between changes in deepwater [O 2 ] and atmospheric pCO 2 arises from the biological carbon pump through the production and consumption of organic matter (18,(37)(38)(39). Whereas many marine processes are involved in pCO 2 variations on glacialinterglacial time scales [see (40) for a review], changes in the efficiency of the ocean's softtissue pump directly affect dissolved deep-water oxygen through the regeneration of respired carbon. Weaker overturning circulation during glacial stages leads to a more efficient biological pump (i.e., an increase in the ratio of regenerated versus preformed nutrient content in the interior ocean) (18) by stemming the ventilation "leak" of CO 2 from the ocean to the atmosphere and increasing its oceanic residence time (41). At the same time a stronger biological pump (i.e., an increased rate of organic carbon exported through greater productivity in the surface ocean) (42) 4 3− ] according to stoichiometric ratios involved in organic matter respiration (38,45).
ODP Site 1267 in the Southeast Atlantic (Fig. 1) provides evidence of an increase of 0.5 mmol kg −1 in [PO 4 3− ] at~960 ka (Fig. 3E)  The change in nutrient content of the deep South Atlantic also corresponds to a major change in thermohaline circulation (THC) inferred from changes in neodymium isotopes (14,46,47). Neodymium isotope (e Nd ) records from North Atlantic DSDP Site 607/ V30-97 (Figs. 1 and 2F) (47) and ODP Sites 1267 (14), 1088, and 1090 (46) in the Southeast Atlantic ( Figs. 1 and 3, F and G) document significant reduction in NADW contribution and/or an increased influence of AABW (14) between 950 and 900 ka (14,46). These THC changes occurred at times of reduced upwelling and degassing of southernsourced deep water under expanded sea ice cover in the Southern Ocean, thereby Thomas (48,49). An increase in deep water residence time results in decreased oxygen concentrations through organic matter oxidation with or without attendant changes in carbon flux from the surface (41,50). Lastly, a change in the proportion of northern-versus southern-derived deep water masses (16,51,52) bathing Site U1385 can affect [O 2 ]; for example, NADW has [O 2 ] values that are~50 mmol kg −1 higher than those of AABW and 75 mmol kg −1 higher than Circumpolar Deep Water ( Fig. 1 and table S1) (53).
Changes in source areas of NADW formation could substantially affect [O 2 ] in the deep North Atlantic. CLSW is a well-ventilated water mass that distributes oxygen throughout the entire North Atlantic (54) including Site U1385 on the Iberian Margin that today is bathed by~45% CLSW (29). Oxygen saturation of CLSW is highly sensitive to winter conditions in the Labrador Sea through bubblemediated air-sea transfer associated with intensive winds, cooling, and deep convection (55). The Labrador Sea has been described as a "trapdoor" through which the flux of oxygen ventilates much of the deep Atlantic basin (55). During Heinrich stadials of the last glaciation, the Labrador Sea was covered by extensive sea ice which would have reduced the ventilation of the deep Atlantic by CLSW (21,56). Sea ice also expanded over the Nordic Seas (57-59) (source areas of DSOW and ISOW) (53) and the Irminger Sea where these water masses mix to form NADW (53,60,61).
The decrease in glacial deep-water [O 2 ] at 960 ka coincides with an increase in continental ice volume and lowered sea level, inferred from changes in d 18 O of seawater at Site 1123 in the Southwest Pacific (51, 62, 63) (Figs. 2G, 3B, and 4A). Available evidence suggests Northern Hemisphere glaciation intensified during MIS 22 [(51) and references therein], which resulted in changes to Atlantic deep-water circulation (52).
The timing of the initial decrease in deepwater [O 2 ] at~960 ka during MIS 26 is associated with a short-lived excursion to very negative e Nd values at Site 607 (Fig. 2F) (47). This event is interpreted as a result of extensive weathering and erosion of the North American craton between MIS 27 to 25 (~980 to 950 ka) and the possible transition from terrestrial-to marine-terminating ice sheets in the Northern Hemisphere (64). MIS 26 also represents the first time the polar front shifted to a zonal position south of Site 980/981 in the Northeast Atlantic (55.5°N, 14.7°W) as in-ferred from an increase in the percent of the polar planktonic foraminifera Neogloboquadrina pachyderma (percent NPS), a proxy for cold surface temperatures ( fig. S10) (65, 66). These precursor changes heralded the major changes in ice volume, thermohaline circulation, and pCO 2 associated with the "900-ka event" during MIS 24 to 22 in less highly resolved records than those of Sites U1385 and 980/981 (3, 4, 6, 10, 14, 46, 47, 62, 65). Thomas  3-] (green squares) (14). (F and G) ODP Sites 1088 (mid-depth) and 1090 (deep) e Nd , respectively (46); interglacial (orange circles) and glacial (blue circles) maxima.
The deep-water oxygen depletions may have been driven by increased freshwater inputs and ice rafting to the source areas of NADW formation. The lowest [O 2 ] values during the last glacial cycle are associated with cold millennial events in the North Atlantic ( fig. S11B) (21). Millennial [O 2 ] depletion events occur throughout the entire record and are most common toward the latter part of glacial cycles and during terminations. These times are also associated with enhanced sea ice extent in the North Atlantic (67), cold stadial periods, and increases in ice-rafted detritus (IRD) at both the North Atlantic Site U1308 (68) (Fig. 1 and figs. S11A and S12) and the Labrador Sea Site U1302/03 ( fig. S13) (69). A close connection exists today between North Atlantic deep-water oxygenation and winter surface conditions in the Labrador, Irminger, and Nordic Seas (55)(56)(57)(58). We suggest that enhanced surface stratification and reduced deep-water convection in NADW source areas may have caused episodic reduction of North Atlantic deepwater ventilation, leading to reduced deepwater oxygen and increased carbon storage (70). Incursions of glacial southern-sourced deep water into the North Atlantic associated with weaker NADW production (51,52,61,71,72) could also partly account for [O 2 ] changes observed at Site U1385 over the past 1.5 Ma given the [O 2 ] difference (~50 mmol kg −1 ) between northern and southern-sourced deep water today ( Fig. 1 and table S1).
The changes in deep-water [O 2 ] at Site U1385 could equally be driven by surface processes in the Southern Ocean such as changes in productivity (13), surface stratification (15,73), vertical mixing, and sea ice extent. These changes would be transmitted to the deep sea through an expanding southern-sourced deep water mass such as Lower Circumpolar Deep Water (74) in which oxygen was considerably reduced during the last glacial period (26,36,44). Reduction in ventilation due to circulation changes, Southern Ocean stratification, and sea ice expansion would have contributed to the inferred increase in deep-sea carbon storage over the MPT (14-16, 18, 73, 75-78).
Decreased glacial deep-water [O 2 ] and increased deep-sea carbon storage across the MPT have implications for atmospheric pCO 2 . The Dd 13 C cib-aff proxy of [O 2 ] at Site U1385 is consistent with changes in pCO 2 measured directly in ice cores and blue ice in Antarctica (1, 6, 7) and inferred indirectly from boron isotope data (3)(4)(5) (Fig. 4B). Prior to~900 ka, minimum glacial pCO 2 values were~24 ppm higher than they were afterward (3,(5)(6)(7). If applied to the whole Atlantic Ocean influenced by an expanding southern-sourced deep water mass, Hoogakker et al. (21) estimated at least 15% of the reduction in atmospheric pCO 2 during the last glacial maximum (LGM) could be accounted for by the increases observed in their respired carbon pool. Furthermore, over the same period, the Pacific Ocean is reported to have had old, oxygen-depleted deep-water (26,43,44). The reconstructed glacial [O 2 ] values reported here for the MPT are at least 20 mmol kg −1 lower than the LGM (21), implying greater carbon storage than during the LGM. It remains to be seen from other basins whether this was a global phenomenon.
In summary, inferred changes in North Atlantic deep-water [O 2 ] for the past 1.5 Ma reveal a significant (>40 mmol kg −1 ) reduction in glacial deep-water [O 2 ] at~900 ka suggesting increased storage of respired carbon, which is consistent with a drawdown of glacial atmospheric pCO 2 values (Fig. 4B) (1,3,4,6,7). The inferred change in [O 2 ] is supported by trace metal records (16) and nutrient (14) proxy records in other Atlantic sites associated with a critical change in glacial THC at~900 ka (46) (Fig. 3, C to G). The close association between [O 2 ] depletions and IRD events (68,69,79) suggests that increased stratification and sea ice cover in NADW source regions (56-59) reduced the oxygen supply to much of the deep North Atlantic (55). In addition, northward expansion of southern-sourced deep water into the North Atlantic and processes in the Southern Ocean (e.g., productivity, surface water stratification, vertical mixing, and sea ice extent) also contributed to reduced ventilation associated with a major change in deepwater circulation (46). Our results support a set of internally consistent changes in Atlantic deep water beginning at~960 ka across the MPT, which included a decrease in oxygen concentrations, increased nutrient concentrations (14), and storage of respired carbon that led to a reduction in glacial pCO 2 (3-7, 47) and an associated increase in global ice volume (62). Thomas  A B