Covariation of Deep Antarctic Pacific Oxygenation and Atmospheric CO₂ during the Last 770 kyr

The Southern Ocean (SO) is believed to play a central role in modulating the ocean–atmosphere CO₂ exchange over glacial cycles, thereby adjusting the greenhouse effect on global climate [1–3]. The SO may exert a substantial control on the atmospheric partial pressure of CO₂ (pCO₂) by (1) facilitating the upwelling and exposure of deeply sequestered CO₂ and nutrient-rich water masses along outcropping isopycnals and (2) storing dissolved inorganic carbon (DIC) in the ocean interior following the export and respiration of biologically fixed carbon. For instance, if export production increased or the rate of vertical exchange decreased in the SO, DIC concentrations in deep waters would increase, in turn resulting in lower atmospheric CO₂ concentrations. Such circumstances could arise through a decrease in circum-Antarctic sea ice cover [4], causing an increase in air–sea gas exchange [5], and/or from enhanced or southward shifting southern hemisphere westerlies, which would accelerate overturning circulation in the SO [6, 7].

At present, vertical exchange in the SO critically involves the lower and upper cells of the global meridional overturning circulation (MOC); these are approximately divided by the 27.6 kg m⁻³ isopycnal [7, 8]. From the nutrient- and CO₂-rich lower cell, deep waters are transported upward along the isopycnal surface, cropping out south of the polar front (PF) in the Antarctic Zone (AZ) of the SO (Figures 1(a) and 1(b)). Carbon dioxide leakage is modulated by physical processes via changes in westerlies and sea ice cover [7, 9, 10]. The upwelling waters are then transported southward where they sink to form the Antarctic bottom water (AABW), which brings biologically fixed carbon into the ocean interior (Figure 1(b)). During the Last Glacial Maximum (LGM), the combination of increased phytoplankton productivity—triggered by enhanced inputs of iron-bearing dust—and weakened vertical exchange in the SO would have reduced the leakage of CO₂ to the atmosphere, strengthening the overall efficiency of the “organic carbon pump” (OCP) [11–13].

Although physical and biological processes in the SO could have had a significant impact on ocean–atmosphere CO₂ exchange in glacial–interglacial (G–IG) cycles [2, 6, 14], there are gaps in existing reconstructions that hinder a more comprehensive understanding of these processes. First, very little evidence has been reported for OCP-driven changes in CO₂ storage in the deep SO from before the last interglacial, especially close to the 800 kyr time range covered by the reliable and comparable atmospheric CO₂ record from EPICA Dome C (EDC) [15]. In addition, published reconstructions from sites located at the edges of the AZ and Subantarctic Zone (SAZ) have shown the overall impact of OCP in the SO on atmospheric CO₂ across both regions with mixing of waters [10, 16]. However, very few records exist from the central AZ, which would provide the opportunity to independently capture the response of the AZ carbon reservoir to upwelling/air–sea exchange and photosynthetic production. Furthermore, the Pacific Ocean dominates the world ocean volumetrically (it is three times larger than the Atlantic), so variations in its dissolved nutrient and gas (O₂ and CO₂) inventory would have potentially significant impacts on the carbon cycle [17–19]. To date, however, records from the Pacific sector of the SO remain sparse. Moreover, significant variations in vertical stable carbon isotope (δ¹³C) gradients of DIC in the SO are found on G–IG and millennial timescales [20]. These gradients are influenced by the storage of respiratory CO₂ in the deep ocean, but they are also susceptible to large disequilibrium effects that may cause δ¹³C to be decoupled from DIC storage, as the air–sea equilibrium of carbon isotopes is still an order of magnitude slower than that of DIC itself [21]. Since dissolved oxygen (O₂) is consumed stoichiometrically during the respiration of sinking organic matter, it can provide a robust constraint on carbon storage in the deep AZ, while being replenished an order of magnitude faster than CO₂ during air–sea exchange.

Here, we present new geochemical data that show variations in the oxygenation of the deep SO. The data are derived from sediment core ANT34/A2-10, which was retrieved from the central AZ (i.e., the midlatitude AZ) area in the southern Pacific (Figure 1), and they chronicle deep SO carbon storage over the last 770 kyr. Our records cover multiple G–IG cycles and so offer a welcome opportunity to investigate covariations between the size of AZ carbon reservoirs and atmospheric CO₂ concentrations in the EDC that covers a similar time frame.

Gravity core ANT34/A2-10 (125°35$′$31$″$W, 67°02$′$10$″$S, 4216.6 m water depth) was retrieved from the Pacific sector of the central AZ, south of the present-day position of the Antarctic PF (Figure 1(a)), by R/V Xuelong during the “34th Chinese National Antarctic Research Expedition” cruise. This site is bathed in Antarctic bottom water (AABW), which formed through shelf processes and deep convection [22]. The sediment sequence with a total length of 4.54 m consists mostly of gray clay silt without turbidite, diatom, and sponge sequences, with discontinuous carbonate-bearing intervals and ice-rafted debris. The core was subsampled at 2 cm intervals, and a total of 277 samples were prepared for the next set of analyses.

AMS¹⁴C dating was performed on samples from two sediment layers in the core at Beta Analytic Lab, Miami, USA (Table 1). For one sample, the analysis was performed on organic carbon, specifically the acid insoluble organic carbon component of the bulk sample. In the second sample, the analysis was performed on inorganic carbon in the form of biogenic carbonate, with the sample consisting of fossil tests of the planktonic foraminifer species Neogloboquadrina pachyderma (sin) that were manually picked from the 150–250 μm size fraction. The AMS¹⁴C ages were converted into calendar ages using Calib 8.2 [23] with a $△R$ value of 639 years according to ¹⁴C years of known age marine samples in the Antarctic Pacific [24].

X-ray fluorescence (XRF) scanning of this core was performed with an Avaatech XRF Core Scanner at Tongji University. Data were obtained at a resolution of 1 cm over an area of 1.2 cm² directly at the split core surface of the archive half, and we were able to determine the relative contents of 29 elements. The core surface was covered with 4 mm thick SPEXCerti Prep Ultralene1 foil to avoid contamination of the XRF measurement unit. Although the XRF data were unable to provide the percentages of elements directly, they show a positive correlation (linear logarithmic relationship) with the elemental concentrations measured via inductively coupled plasma mass spectrometry (ICP–MS) [25, 26]. In recent studies, elemental concentrations have been normalized to concentrations of Ti to eliminate the influence of terrigenous material and the “dilution effect” associated with deposition [27–29]. The application of normative Mn/Ti calculations assumes that sedimentary Ti has a detrital origin and that the compositions of the Ti-bearing phases of the terrigenous material (detrital Mn/Ti) remained constant in space and time [18].

The concentrations of selected elements (aluminium (Al), titanium (Ti), manganese (Mn), germanium (Ge), and barium (Ba)) were performed on samples taken every 4 cm (a total of 139 samples) at the Key Laboratory of Marine Sedimentology and Environmental Geology, First Institute of Oceanography, Ministry of Natural Resources, Qingdao, China. Samples were measured using inductively coupled plasma optical emission spectroscopy (ICP–OES; Thermo Fisher X Series II) via solution nebulization after mixed acid digestion (HF–HClO₄) under pressure. Precision was better than 5% for replicate measurements. The authigenic fraction of U was estimated by normalization to Al as follows: $aU=Utot−U/Allithogenic×Al$⁠, with $U/Allithogenic=1.5×10−6$ [30, 31]. The biogenic fractions of Ba and Ge (bioBa and bioGe, respectively) are useful proxies for integrated export production [30] and are calibrated as follows: $BioBa=Batot−Ba/Allithogenic×Al$⁠, with $Ba/Allithogenic=0.0065$ [18, 31]; $BioGe=Getot−Ge/Allithogenic×Al$⁠, with $Ge/Allithogenic=0.0000199$ [30, 31]. Samples were normalized using the upper-crust compositions of McLennan [31], as study on the core in the Weddell Sea of the SO [30]. The calculation uses Al normalization which is commonly used to minimize dilution effects. The calculations of normative Mn/Ti, aU, Ba/Ti, and bioBa assume that the composition of the Al- and Ti-bearing phases of the terrigenous material remained constant in space and time [27].

Our sediment core spans the last seven G–IG cycles, starting from Marine Isotope Stage (MIS) 19 (Figure 2). Due to the absence of biogenic carbonate (required to develop continuous foraminiferal oxygen isotope (δ¹⁸O) stratigraphy), we established an age model by (a) using AMS¹⁴C dates for the past 27 kyr and (b) correlating XRF-derived Ba/Ti ratios with the LR04 δ¹⁸O stack for the older part of the core [32]. In recent studies [28, 33, 34], bioBa has been used to generate age models in sediment cores from the Indian Ocean and Pacific Ocean sectors of the Antarctic during the late Quaternary, based on the close coupling between the bioBa and LR04 curves. BioBa is also used as a primary productivity proxy in many paleoceanographic studies [35–39]. Given the high and low productivity during interglacial and glacial periods south of the PF [11, 14, 40], respectively, variations in primary productivity are thus closely tied to climatic change over G–IG cycles. Assuming that sedimentary Ti is of detrital origin, Ba abundance normalized to Ti yields an estimate of the sedimentary concentration of bioBa [35, 41, 42], which serves as a tool to reconstruct a reliable chronological framework for the core. Our downcore record of Ba/Ti ratios displays a strong visual correspondence with the LR04 δ¹⁸O stack, display pronounced G–IG cyclicity based on the main features (highs and lows), which are used as age control points. Twenty-eight stratigraphic tie points were chosen in the study core ANT34/A2-10 (including the AMS¹⁴C dates, Table 2) for detailed comparison with the LR04 δ¹⁸O stack (Figure 2(a)). During MIS 2 and MIS 1, where two AMS¹⁴C dates are available, the close correlation between the Ba/Ti record and the LR04 δ¹⁸O stack (Figure 2(b)) supports the reliability of the age framework reconstruction. By comparing the Ba/Ti curve with the LR04 δ¹⁸O stack graphically, we conclude that core ANT34/A2-10 spans the past 773 kyr and therefore MIS 1–19 (Figure 2). The time-averaged sedimentation rates for the core range from 0.18 to 2.75 cm/kyr, based on the ages assigned at the individual tie points (Figure 3(g)). The mean sedimentation rate for the entire core is 0.59 cm/kyr.

To verify the reliability of the chronological framework, we examined the consistency of the magnetic susceptibility (MS) in this age model with the dust concentration of the EPICA Dome C (EDC) ice core [43]. As atmospheric dust indicators, MS in SO deep sea sediment could be used for the graphic comparison to reconstruct the reliable age model [44, 45]. The comparison revealed essentially a consistency in the variation between the MS in this framework and the EDC dust concentration (Figure 2(b)). In particular, the dust concentration peaks also correspond to the MS peaks in the study core. To quantify the margin of error between our age model and a known reliable age model such as the EDC ice core, we differenced the age of each peak of the MS from the dust concentration in the EDC. The results show a margin of error of -8.00 to 7.75 kyr between the age model of ours and the EDC (Table 3), an error that will have no significant impact on our study of the variability in the G–IG cycles on the orbital scale. This consistency of indicators from another method for age model reconstruction and the margins of error provides validation and strong constraint on the reliability of this chronological framework.

The sedimentary geochemistry of Mn is dominated by the redox control of its speciation, with higher oxidation states (Mn (III) and (IV)) occurring as insoluble oxyhydroxides in well-oxygenated environments and the lower oxidation state (Mn (II)) being much more soluble in oxygen-depleted settings [46]. Manganese enrichment due to the presence of Mn oxide coatings is observed in deep sea sediments where low accumulation rates of organic matter allow oxic conditions to persist to great sediment depths. In sediments that are deposited in anoxic settings, there is no accumulation or recycling of Mn oxides, and the Mn concentrations of buried deposits are low and entirely controlled by their insoluble detrital fraction. In addition to oxides, Mn enrichment via Mn carbonates can occur in sediments. However, the carbonate content of our core sample is extremely low, and Mn enrichment due to the presence of Mn carbonate is unlikely. The presence of Mn concentrations in excess of what can be expected from the detrital fraction therefore suggests that the host sediment must have accumulated under oxygenated bottom waters.

Uranium is well mixed in the ocean and behaves conservatively in oxygenated sea water as U(VI) within the highly soluble uranyl carbonate complex (UO₂(CO₃)₃⁴⁻) [47]. One of the main mechanisms of removal is through diffusion into reducing surface sediments [48]. In suboxic conditions, soluble U(VI) is reduced to insoluble U(IV) [49], resulting in a concentration gradient between bottom water and sediment pore water. Along this gradient, dissolved U then diffuses into the sediment, leading to an enrichment of authigenic or excess U (aU) in the sediment.

Sedimentary redox conditions can be influenced by two major factors: (1) the rain of organic matter to the sea floor, which consumes dissolved oxygen as it is respired, and (2) the rate of oxygen supply from the input of younger waters with higher oxygen content. The use of multiple proxies for redox conditions (in our case, Mn/Ti, XRF ln(Mn/Ti), and aU) may compensate for specific limitations associated with individual proxies and thus provide information about the overall redox state of the sediment. Combining the redox state proxies with paleoproductivity proxies (in our case bioBa and bioGe) allows these processes to be disentangled. It follows that relative changes in oxygen concentration at the water–sediment interface can also be inferred. If sediments become more reducing with no observed increase in paleoproductivity (i.e., no increased oxygen demand owing to the respiration of organic matter), it can be inferred that the change in redox state is due to a reduced oxygen supply to the bottom water.

All paleoproductivity proxies are associated with large uncertainties, so a multiproxy approach may help to compensate for specific limitations associated with the individual proxies. The formation and downward flux of barite (BaSO₄) in seawater is tightly linked to integrated surface productivity [50]. Precipitation of bioBa (or excess barite) is thought to accompany phytoplankton decay [51] and may therefore represent a reliable proxy for integrated organic carbon export from the photic zone. BioBa dissolves under conditions of sulfate (SO₄) depletion by microbial SO₄ reduction [52]. An estimated 30% of the bioBa produced in the water column is preserved in sediments under oxic/suboxic conditions [30].

Diatoms are the main primary producers in the SO ecosystem. Because Ge has a variety of geochemical properties that are very similar to those of Si, Ge is often used as a tracer for the biochemical Si cycle [53, 54]. Indeed, studies have shown that the uptake of dissolved Si by diatoms in seawater is accompanied by the uptake of Ge, which is then deposited to the seafloor along with the siliceous shells of diatoms [55].

In core ANT34/A2-10, the downcore records of all three redox-state proxies (Mn/Ti, XRF ln(Mn/Ti), and aU) show a high degree of consistency over G–IG cycles, signaling more reducing conditions during each glacial period, which rapidly become more oxidizing thereafter (Figure 3). There are similar features in the records of bioBa and bioGe, which both show increases in interglacial periods and decreases in glacial periods (Figures 3(e) and 3(f)), suggesting variations in the export of organic matter to the sea floor over G–IG cycles [56]. Because of the greater solubility of oxygen in cold waters, the inference that the deep AZ was more poorly oxygenated during glacial periods indicates that the solubility effect on oxygenation was overwhelmed by larger changes in the utilization of oxygen in the ocean interior [56]. In addition, oxidative “burn-down” occurs following a sharp reduction in the rates of sedimentation and/or organic carbon delivery to the seafloor, as diagenetic dissolution of authigenically precipitated metals during renewed pore water O₂ exposure could affect the distribution of trace metals [57, 58]. A comparison between records of sedimentation rate and redox state proxies in core ANT34/A2-10 clearly shows that downcore changes in redox state were not controlled by variations in sediment accumulation (Figure 3(g)). The consistency between the three redox proxy records and constraints from records of Antarctic temperature and sedimentation rate again firmly support the idea that the development of more reducing conditions during periods of decreased organic carbon supply to the seabed, which can only be reasonably driven by the decreased oxygen content of overlying bottom waters [16, 59]. This is in turn associated with increased storage of respired carbon in the abyssal central AZ during glacial periods.

During glacial intervals, we found that the lower oxygen concentrations in the central Antarctic Pacific are consistent with the low benthic foraminiferal δ¹³C values located at AABW in the SAZ (Figure 4(a)) [60, 61], which indicates that the sequestration of greater amounts of respired carbon in the deep ocean relative to the interglacials may be a widespread occurrence in the SO. The very negative deep water δ¹³C values during glacial periods are also present in the benthic foraminifera from the Pacific (below 2000–2500 m) [62–64], Atlantic [65, 66], and the Indian oceans [66, 67], suggesting that the deep-water reservoir stored more carbon in the glacial ocean. This large-scale phenomenon is consistent with the global stacked benthic foraminiferal δ¹³C record (Figure 3(d)) [68].

The expansion of Antarctic sea ice cover (Figure 4(k)) [43] and northward movement and/or weakening of Southern Hemisphere (SH) westerlies during glacial periods (Figure 4(l)) [69] are believed to have the potential to significantly attenuate the ventilation and oxygen supply of the ocean’s interior [1, 7, 13], which is consistent with the discovery of old isolated glacial water masses in the North and South Pacific, as well as in the South Atlantic [19, 70–72]. This provides evidence for enhanced sequestration of respired carbon at depth and the drawdown of pCO₂^atm (Figure 4(d)).

During the subsequent glacial terminations, there were consistent shifts in the records of all three redox-state proxies indicating improved oxygenation of bottom waters (Figure 3). Our data show greater oxygenation of deep waters in the central AZ during the deglacial intervals when atmospheric CO₂ concentrations also rose rapidly (Figure 4(d)), concurrent with a greater rain of organic particles to the sea floor (Figures 4(f) and 4(g)). The records imply that the decrease in deep ocean carbon storage in the central AZ during the deglacial was controlled largely by enhanced ventilation. Thus, the oxygenation proxy data are in line with the release of deeply sequestered remineralized carbon to the atmosphere during G–IG cycles from 770 ka. These observations for the deglacial are consistent with the higher benthic foraminiferal δ¹³C values of Lower Circumpolar Deep Water in the SAZ (Figure 4(a)) [60, 61] and reports of increased oxygenation in the northern AZ and SAZ during the last glacial periods [9, 10]. This suggests that carbon was expelled from the deep SO during the deglacial terminations leading to improved deep ocean oxygenation.

The sedimentary records of bioBa and bioGe from core ANT34/A2-10 are interpreted as indicating lower organic matter export from the surface ocean during glacial periods (Figures 4(f) and 4(g)), consistent with measurements elsewhere in the Antarctic SO south of the PF [10, 14, 73]. Decreased productivity during glacial intervals may have been driven by a reduction in the supply of iron from deep waters [14]. Indeed, vertical mixing and upwelling rather than atmospheric transport determine the supply of iron to the surface of the AZ region [74]. During subsequent glacial terminations, significant rises in export production coincide with intervals of strengthened SH westerlies and decreased sea ice cover (Figures 4(k) and 4(l)), which are accompanied by summer sea surface temperature overshoots and abrupt disappearances of winter sea ice [75]. These factors could lead to an increase in the nutrient supply to the euphotic zone and pulses of export production. Combined with records in the northern Antarctic Atlantic [14], the systematic and repeated glacial-to-interglacial rises in export production in the AZ region (Figures 4(f) and 4(g)) indicate a robust pattern of enhanced SO ventilation during interglacials.

In contrast to the major decreases in northern AZ export in the early stages of glaciation (during the late part of interglacials) [14], the decreases in central AZ productivity occur between peak interglacial conditions and later in the glacial progression, coinciding approximately with the entire interval of atmospheric CO₂ decline in the glaciation (Figure 4(d)). The latter half of these decreases overlaps with the dramatic increases in productivity in the SAZ (Figure 4(h)). The observed decreases in atmospheric CO₂ concentration of ~40 parts per million (ppm) could be explained using numerical model simulations of SAZ iron fertilization [76–78]. In addition, carbon sequestration induced by weakened SO ventilation might contribute continuously to the residual CO₂ decline into each glacial period, from each interglacial peak to glacial maximum. Throughout the glaciation, the process of weakened ventilation associated with AZ productivity is continuous, whereas the intervals of high productivity in the SAZ are concentrated mainly in the second half of the glaciation (Figures 5(a) and 5(b)). These work together to cause the expansion of the deep SO carbon reservoir and glaciation in G–IG cycles. Mixing occurs in the upper modern ocean across the fronts separating the AZ and SAZ [79], which may explain the inconsistency between the central and northern AZ productivity records. Surface nutrients in the northern AZ might have been further depleted by iron fertilization in the SAZ, but the supply of central AZ iron, which is driven by changes in SO ventilation, could be independent of processes in the SAZ.

The above observations demonstrate the potential significance of the repeated changes in oxygenation in G–IG cycles, with specific implications for the role of the AZ OCP in regulating atmospheric CO₂ concentrations [10, 14]. However, it remains to be clarified whether the observed G–IG variations in AZ OCP efficiency resulted primarily from changes in export productivity (allowing oxygen levels to decrease owing to organic carbon remineralization in the ocean interior) or from changes in SO ventilation (causing carbon loss to the atmosphere with direct oxygen gain in the ocean interior).

Our data show that the oxygenation of deep waters in the central AZ is generally correlated with export productivity over the past 770 kyr (Figure 4), ruling out a dominant control of local surface ocean productivity on oxygenation and respired carbon levels in the deep central AZ Pacific. We therefore conclude that variations in the level of oxygenation and OCP efficiency in the central AZ could only be driven primarily by physical “ventilation” processes (i.e., overturning circulation, mixing, and/or air–sea gas exchange). The oxygen levels in the abyssal central AZ, which varied in parallel with atmospheric CO₂ levels during the several G–IG cycles, provide strong evidence for a significant impact of SO ventilation on the sequestration of respired carbon in the deep ocean and atmospheric CO₂ concentrations in G–IG cycles. In addition, increased ventilation in the AZ region could lead to an increase in the local dominance of southern-sourced deep waters and an “improvement” in their ventilation state in the SAZ and therefore partially influence the efficiency of the SAZ OCP and deep carbon storage [9].

Our results show that physical “ventilation” processes in the central AZ might exert the dominant controls on deep carbon storage and the efficiency of the AZ OCP in G–IG cycles during the last 770 kyr. Together with the SAZ OCP efficiency, which is driven primarily by dust-induced biological productivity [80], they control the OCP of the SO and had synergistic impacts on deep ocean carbon sequestration and atmospheric CO₂ levels. Although there are two opposing theories about the SO’s role in controlling past concentrations of atmospheric CO₂ that focus on physical and biological processes [6, 12, 14, 20, 81, 82], studies on the oxygenation of the deep SAZ and the deep northern AZ have in part helped to reconcile this contradiction on millennial timescales [9, 10]. However, our records from the central AZ have the potential to coordinate the dominant physical and biological processes in the AZ and SAZ, respectively, on G–IG timescales (Figure 5).

Regardless of the efficiency of the OCP, the “pump” concept represents the transport of carbon from sea surface to deep carbon reservoir. When considering the net transmission of respired carbon from the deep ocean to the atmosphere, we propose that a new metaphor, the “carbon venting valve,” is used to depict the physical ventilation processes in the AZ that act in opposite directions. The function of the SO in modulating atmospheric CO₂ changes in G–IG cycles is performed by both the AZ’s “carbon venting valve” and the SAZ’s OCP (Figure 5). In deglacial intervals, for example, the weakening of the Atlantic Meridional Overturning Circulation owing to freshwater inputs in the northern Atlantic (Figure 4(j)) [83–85] can lead to coupled changes in ocean and atmospheric circulation (Figure 5(c)) [81, 86]. This causes enhanced upwelling waters to crop out south of the PF in the AZ [1, 87] and a reduction in the supply of airborne dust to the SAZ owing to glacial erosion and the trapping of glacial flour in proglacial lakes [88]. Thus, enhanced iron limitation in the SAZ and upwelling in the AZ, coordinating the input of carbon to the deep sea as well as its release by deep ventilation, would have led collectively to the reduced amounts of carbon storage in the SO through biological and physical processes, respectively. The strength and the efficiency of carbon sequestration in the deep SO were controlled by the coordination between changes in biological export productivity in the SAZ and physical/dynamical changes in the AZ (Figure 5) [9, 10]. Our records emphasize that although the biological OCP in the SAZ drowns the flux of carbon to the deep ocean, providing opportunities for carbon sequestration, the physical “carbon venting valve” in the AZ will ultimately determine the efficiency of carbon sequestration by ventilation-induced CO₂ outgassing. Consequently, the biological “pump” in the SAZ and the physical “valve” in the AZ acted together synergistically, but dominated at different intervals over G–IG cycles, to repeatedly switch the SO between carbon sink and carbon source, thereby modulating the atmospheric CO₂ over the last 770 kyr.

We reconstructed levels of deep ocean oxygenation and export production in the central AZ since 773 ka B.P. using redox-sensitive metal and export production proxies extracted from core ANT34/A2-10 from >4000 m water depth, which currently sits in AABW. Our results have generated the following conclusions:

• (1)

  Our observations in the central AZ Pacific show that changes in the oxygenation of the deep AZ are in line with the changes in the amount of deeply sequestered remineralized carbon released to the atmosphere during G–IG cycles over the last 770 kyr. Oxygen concentrations were lower, and more carbon was stored in the deep AZ during glacial periods. During glacial terminations, carbon storage in the deep AZ was reduced, owing primarily to enhanced ventilation

• (2)

  The systematic and repeated glacial-to-interglacial rises in export production in the AZ region indicate a robust pattern of enhanced SO ventilation during interglacial periods. In addition to decreases in atmospheric CO₂ levels caused by iron fertilization in the SAZ during the latter half of the glacial progression, decreases in productivity in the central AZ suggest that weakened SO ventilation led to increased carbon sequestration in the deep AZ. Furthermore, this might have contributed continuously to the residual decline in CO₂ levels going into each glacial period, from each interglacial peak to glacial maximum

• (3)

  Changes in deep ocean oxygenation and biological OCP efficiency in the central AZ might be driven primarily by physical “ventilation” processes (i.e., overturning circulation, mixing, and/or air–sea gas exchange). Abyssal oxygenation in the central AZ, which covaried with atmospheric CO₂ levels during the several G–IG cycles, provides strong evidence for SO ventilation having had a significant impact on the sequestration of respired carbon in the deep ocean and atmospheric CO₂ levels on G–IG timescales

• (4)

  We suggest that the OCP in the SAZ and the physical ventilation processes in the AZ (the “carbon venting valve”) together performed the SO’s role in impacting atmospheric CO₂ concentrations in G–IG cycles. The SAZ’s biological “pump” and the AZ’s physical “valve” acted together synergistically but dominated at different intervals over G–IG cycles by turns, to repeatedly switch the SO between carbon sink and carbon source, thereby modulating the atmospheric CO₂ over the last 770 kyr

Data is available on Pangea.de. The link is https://doi.pangaea.de/10.1594/PANGAEA.933285.

The authors declare no conflicts of interest regarding the publication of this paper.

We thank the 34th Chinese Antarctic Expedition cruise members and the Chinese Arctic and Antarctic Administration for retrieving the sediment core. We are grateful to Prof. Yanguang Liu and Zhihua Chen for helpful discussions, to Aimei Zhu, Lianhua He, and Yazhi Bai for the geochemical measurements, and to Bingbin Qin for the figure output. This work was supported by the Impact and Response of Antarctic Seas to Climate Change (IRASCC2020-2022-01-03-02), the Basic Scientific Fund for National Public Research Institutes of China (Grant nos. 2019Q09, 2019S04, and 2017Y07), the National Natural Science Foundation of China (Grant nos. 41976080, 42076232, 41406220, and 42006075), and the Taishan Scholars Project Funding (Grant no. TS20190963).