Dynamic storage of glacial CO2 in the Atlantic Ocean revealed by boron [CO32−] and pH records

Abstract The origin and carbon content of the deep water mass that fills the North Atlantic Basin, either Antarctic Bottom Water (AABW) or North Atlantic Deep Water (NADW) is suggested to influence the partitioning of CO2 between the ocean and atmosphere on glacial–interglacial timescales. Fluctuations in the strength of Atlantic meridional overturning circulation (AMOC) have also been shown to play a key role in global and regional climate change on timescales from annual to millennial. The North Atlantic is an important and well-studied ocean basin but many proxy records tracing ocean circulation in this region over the last glacial cycle are challenging to interpret. Here we present new B/Ca-[CO 3 2 − ] and boron isotope-pH data from sites spanning the North Atlantic Ocean from 2200 to 3900 m and covering the last 130 kyr from both sides of the Mid-Atlantic Ridge. These data allow us to explore the potential of the boron-based proxies as tracers of ocean water masses to ultimately identify the changing nature of Atlantic circulation over the last 130 kyr. This possibility arises because the B/Ca and boron isotope proxies are directly and quantitatively linked to the ocean carbonate system acting as semi-conservative tracers in the modern ocean. Yet the utility of this approach has yet to be demonstrated on glacial–interglacial timescales when various processes may alter the state of the deep ocean carbonate system. We demonstrate that the deep (∼3400 m) North Atlantic Ocean exhibits considerable variability in terms of its carbonate chemistry through the entirety of the last glacial cycle. Our new data confirm that the last interglacial marine isotope stage (MIS) 5e has a similar deep-water geometry to the Holocene, in terms of the carbonate system. In combination with benthic foraminiferal δ13C and a consideration of the [CO 3 2 − ] of contemporaneous southern sourced water, we infer that AABW influences the eastern abyssal North Atlantic throughout the whole of the last glacial (MIS2 through 4) whereas, only in the coldest stages (MIS2 and MIS4) of the last glacial cycle was AABW an important contributor to our deep sites in both North Atlantic basins. Taken together, our carbonate system depth profiles reveal a pattern of changing stratification within the North Atlantic that bears strong similarities to the atmospheric CO2 record, evidencing the important role played by ocean water mass geometry and the deep ocean carbonate system in driving changes in atmospheric CO2 on these timescales.

The origin and carbon content of the deep water mass that fills the North Atlantic Basin, either Antarctic Bottom Water (AABW) or North Atlantic Deep Water (NADW) is suggested to influence the partitioning of CO 2 between the ocean and atmosphere on glacial-interglacial timescales. Fluctuations in the strength of Atlantic meridional overturning circulation (AMOC) have also been shown to play a key role in global and regional climate change on timescales from annual to millennial. The North Atlantic is an important and well-studied ocean basin but many proxy records tracing ocean circulation in this region over the last glacial cycle are challenging to interpret. Here we present new B/Ca-[CO 2− 3 ] and boron isotope-pH data from sites spanning the North Atlantic Ocean from 2200 to 3900 m and covering the last 130 kyr from both sides of the Mid-Atlantic Ridge. These data allow us to explore the potential of the boron-based proxies as tracers of ocean water masses to ultimately identify the changing nature of Atlantic circulation over the last 130 kyr. This possibility arises because the B/Ca and boron isotope proxies are directly and quantitatively linked to the ocean carbonate system acting as semi-conservative tracers in the modern ocean. Yet the utility of this approach has yet to be demonstrated on glacial-interglacial timescales when various processes may alter the state of the deep ocean carbonate system. We demonstrate that the deep (∼3400 m) North Atlantic Ocean exhibits considerable variability in terms of its carbonate chemistry through the entirety of the last glacial cycle. Our new data confirm that the last interglacial marine isotope stage (MIS) 5e has a similar deep-water geometry to the Holocene, in terms of the carbonate system. In combination with benthic foraminiferal

Introduction
Over the last 3 million years, Earth's climate has oscillated between interglacial conditions, much like the pre-industrial, and relatively cold glacials, when global sea-levels were ∼130 m lower and global temperatures ∼4 • C cooler than the pre-industrial (Crowley, 2000;Grant et al., 2014). For at least the last 800 kyr the glacial-interglacial climate cycles varied together with, and were amplified by, changes in atmospheric CO 2 concentrations (low/high during glacials/interglacials; Bereiter et al., 2015;Broecker and Denton, 1989). The most recent glacial termination (Termination I; T1) occurred <20 kyrs ago and, although a wealth of proxy records are available for this important interval, the forcing and amplifying mechanisms responsible for the repeating glaciation and deglaciation cycles in the Pleistocene remain far from fully understood (e.g. Kohfeld and Ridgwell, 2013). However, because the oceans are the only large exogenic carbon storage reservoir that can respond on such rapid timescales, it is highly likely that changes in ocean carbon storage play a major role in (sub)millennial-scale changes Site map and section. a: Locations of the cores used in this study, ODP Sites 980 and 999, and IODP Sites U1308 and U1313 present a depth, latitude and longitude transect of the North Atlantic. Flow of modern deep water is indicated by the arrows -after Raymo et al. (2004) modified from Yu et al. (2008), dashed lines indicate Intermediate water pathways over the Caribbean sill (depth ∼1.8 km), map was created using Ocean Data View (Schlitzer, 2009). Abbreviations are: NGS, Norwegian-Greenland Sea; DSOW, Denmark Strait Overflow Water; ISOW, Iceland Sea Overflow Water; WTRO, Wyville Thomas Ridge Overflow; LSW, Labrador Seawater (together make up NSW); and SSW, Southern Sourced (deep) Water. Also shown are the other published cores sites used in this study. b: A section plot of the Atlantic using GLODAPv2 data (Key et al., 2015) for [CO 2− 3 ], also showing the sites used in this study, the water mass divides between NADW and SSW are clearly visible between high and low [CO 2− 3 ], respectively. (For interpretation of the colours in the figure(s), the reader is referred to the web version of this article.) in atmospheric CO 2 and hence climate (Ahn and Brook, 2008;Martinez-Boti et al., 2015). Thus, it is important to determine where in the deep ocean atmospheric CO 2 is sequestered during glacial intervals in order to glean further insight to what mechanism(s) is (are) responsible for changing the relative partitioning of CO 2 between the atmosphere and the ocean. Among the many suggested mechanisms for glacial ocean CO 2 storage (e.g. Archer et al., 2000;Brovkin et al., 2012;Hain et al., 2010;Sigman et al., 2010) the expansion of Dissolved Inorganic Carbon (DIC)-rich [CO 2− 3 ]-poor Antarctic Bottom Water (AABW) into the North Atlantic basin is suggested to account for up to ∼50% of the CO 2 decline observed associated with the transition from the last interglacial to the last glacial maximum (Brovkin et al., 2012;Hain et al., 2010;Skinner, 2009) while an increase in the C-storage in the deep ocean particularly through changes in nutrient utili-sation and stratification in the Southern Ocean is suggested to be the other major process enhancing whole ocean storage (Burke and Robinson, 2012;Rae et al., 2018;Toggweiler, 1999). Changes in the patterns of North Atlantic circulation (see Fig. 1) are supported by many proxy records (e.g. carbon and oxygen isotopes, and Cd/Ca and Mg/Ca ratios of benthic foraminiferal calcite, e.g. Curry and Oppo (2005); Gebbie (2014), as well as neodymium isotopes (ε Nd ) (e.g. Böhm et al., 2015;Howe et al., 2016) with the standing hypothesis being that expansion and contraction of AABW accounts for a substantial component of the oceanatmosphere glacial-interglacial CO 2 variability Skinner, 2009), because more of this carbon-rich water locked away in the deep ocean isolates CO 2 from the atmosphere. Determining the magnitude of this AABW expansion/contraction is key, because different proxy and modelling solutions demand dif- b: ε Nd profiles from sites in similar locations red ODP 980 (Crocker et al., 2016), green IODP 1313 (Lang et al., 2016), blue NEAP8K (Yu et al., 2008)  ferent glacial solutions, and rely on many mechanisms to lower atmospheric CO 2 Kohfeld and Ridgwell, 2013;Martinez-Boti et al., 2015;Skinner, 2009). Similarly, the extent to which sustained northern deep water formation may inhibit the volumetric proportion of AABW (and thus its assumed carbon storage potential) could be crucial to understanding the sensitivity of ocean circulation to large changes in climate (Gebbie, 2014). Most published studies tracing the geometry of NADW over the last 20 kyr and longer have utilised δ 13 C in benthic foraminiferal calcite (e.g. Curry and Oppo, 2005). Although this work has greatly improved our understanding of the changing patterns of circulation there is a degree of ambiguity associated with the use of δ 13 C in this way. As δ 13 C is a non-conservative tracer, its preformed (surface) water signal can be modified during transport to the deep ocean. This is primarily achieved by conservative mixing as well as by the acquisition of organic, remineralised carbon (Lynch-Stieglitz, 2003). In addition, the end member compositions of preformed nutrients and thus, δ 13 C likely vary, through time, in particular as a result of variable air-sea gas exchange in the high latitude surface ocean (Lynch-Stieglitz et al., 1995). In the modern ocean Southern sourced water (SSW) is typically characterised by low δ 13 C (i.e. rich in preformed phosphate and other nutrients) and occurs at >2500 m depths in the North Atlantic during glacial maxima (Fig. 2a).
An expansion of southern sourced deep water during the glacial is consistent with neodymium (ε Nd ) data from the Atlantic (Fig. 2b), where all records appear to shift towards what are typically considered to be southern sourced values of ∼−8 ε Nd , which in sites deeper than 2500 m is also coincident with the δ 13 C minima. ε Nd is a more conservative tracer than δ 13 C, but the potential for ε Nd change in endmembers, particularly in the North Atlantic, complicates the picture; changes in endmember composition could drive change in Nd isotopic composition at all depths even without large circulation changes or mixing of slow moving water masses (Howe et al., 2016, Fig. 2b). Furthermore, while ε Nd tracks deep water source area and pathway, it is not a sensitive proxy for detecting changes to water mass carbon content or changes in transport path without associated changes to water mass geometry or source (Gebbie, 2014;Howe et al., 2016). Other proxies commonly used to establish and deconvolve water mass provenance and flow rates, such as Cd/Ca, Pa/Th and sortable silts can also suffer from ambiguities in their interpretation, due to localised effects or geochemical interactions (e.g. Marchitto, 2013;Thomas et al., 2006). Despite a broad understanding of the potential changing patterns of ocean circulation in the North Atlantic over the last glacial cycle, there is a pressing need for additional proxy based information to reveal the full evolution of ocean circulation with respect to climate over this time interval, particularly in the form of nutrient-type proxies which can trace biological utilisation and the carbon content of the ocean interior.
The boron content and isotopic composition of benthic foraminiferal tests has potential as an additional tracer of deep water circulation and deep water carbon content (Yu et al., 2008;Yu and Elderfield, 2007). The direct link to the carbonate system that underlies both the B/Ca-CO 2− 3 proxy (Yu and Elderfield, 2007;see below), and the boron isotope-pH proxy (Rae et al., 2011) provides a unique insight into ocean carbon storage, has fewer controlling influences than δ 13 C, and because the carbonate system is semi-conservative (i.e. is changed through both conservative mixing and internal ocean processes such as respiration and carbonate production), in the modern ocean it also makes a good tracer of water mass structure on an ocean basin scale (see Fig. 1b). Existing boron proxy data suggests that low [CO 2− 3 ] (high CO 2 and hence low pH) water invaded the Atlantic at depths greater than 2800 m during the last glacial cycle around 70 kyrs ago (Yu et al., 2016). As southern sourced water in the Atlantic is known to have remained low in carbonate ion concentration throughout the last glacial period from qualitative measurements (e.g. preservation, Hodell et al., 2001) and chemical proxy measures (e.g. [CO 2− 3 ], Yu et al., 2014), these data are consistent with δ 13 C-based interpretations of SSW expansion into the deep Atlantic at all latitudes ( Fig. 2; Hodell et al., 2008;McManus et al., 1999;Raymo et al., 1989;Sarnthein et al., 1994). During the deglaciation, modern-style circulation resumed as chemical stratification collapsed (Yu et al., 2008), but when and how that stratification first developed across the North Atlantic basin remains uncertain, partly because of the limitations on depth, spatial and temporal coverage of existing proxies records and partly because of the uncertainties associated with ε Nd and δ 13 C proxy reconstructions (Fig. 2).
Here we report records of B/Ca and δ 11 B in benthic foraminifera from a series of sites across the North Atlantic Ocean from Ocean Drilling Program (ODP) Sites 980 and 999 and Integrated Ocean Drilling Program (IODP) sites U1313 and U1308. Glacialinterglacial CO 2 change is affected by both changes in water mass geometry and changes to water mass chemical properties, both of which impact on the ocean carbonate system. By utilising a multiproxy approach that incorporates new boron proxy data and combining them with published data from these sites we undertake a more complete examination of this problem that accounts for mixing as well as changes to composition from endmember changes and carbon cycle changes. This allows us to quantify the evolution of carbon storage and ocean circulation in the region over the last glacial cycle.

Samples and age model
To reconstruct the carbonate system of the deep North Atlantic Ocean over the last glacial cycle, sediment samples were taken from ODP Site 980 (2184 m depth), IODP Site U1313 (3426 m depth) and IODP Site U1308 (3871 m depth) covering the last 150 kyrs. Site 980 was targeted because it is currently bathed in intermediate water sourced from the north (Fig. 1). Sites U1308 and U1313 were chosen because they are currently bathed in NADW, in the eastern and western abyss of the North Atlantic, respectively. New data are also presented from ODP 999 (2839 m depth), a site in the Caribbean Sea currently bathed in Antarctic Intermediate Water (AAIW) because of the shallow sill depth (∼1800 m) separating the Caribbean and Atlantic basins (Fig. 1).
The cores studied were already heavily sampled so our samples were taken outside of the primary splices. For Sites 980 and U1308 physical properties (L * and magnetic susceptibility) were then used to tie our sample depths (∼10 cm spacing) to the site splices and their appropriate age models (Hodell et al., 2008;McManus et al., 1999).

Brief analytical and sampling methods
Sediment samples from core material were washed with deionised water and sieved to >63 μm, to separate fine and coarse fractions. Individual foraminifera (Plano-convex specimens of Cibicidoides wuellerstorfi, sometimes known as Cibicides wuellerstofi, hereafter C. wuellerstorfi) were hand separated from the 212-500 μm size fraction and measured for boron isotopes and B/Ca ratios using established protocols at the University of Southampton (Henehan et al., 2013;Rae et al., 2011). Detailed methods and carbon system calculations are described in the Supplementary Methods, section 3.
New δ 18 O and δ 13 C of C. wuellerstorfii were generated for Site U1313 and these data are consistent with the sample ages developed in Lang et al. (2016) by tuning this secondary splice to the age model of Naafs et al. (2013) from the primary splice at U1313. All cores are thus tied to the common chronology of the LR04 stack (Lisiecki and Raymo, 2005), the top 40 kyrs of ODP 980 is an exception and uses an updated independent chronology (Crocker et al., 2016). For further details see section 2.3 in the Supplementary Methods.
Hydrographic data for the sites (used for carbonate system calculations) were gathered from the GLODAP dataset (Key et al., 2004), and are presented in Table 1 in the Supplementary Methods.

Results
New records of B/Ca and δ 11 B data from (I)ODP Sites 980, 999, U1313 and U1308 are shown in Fig. 3. The secular evolution of both the B/Ca and δ 11 B records are similar among sites. Given the size of the uncertainty envelopes, the changes in benthic δ 11 B appear muted when compared to the precision on each individual data point. Only U1313 shows well resolved changes in δ 11 B from Holocene values over the last glacial cycle indicating that long term changes of less than ∼0.1 pH units are unlikely to be resolvable in δ 11 B-pH. In contrast, our more highly resolved B/Ca ratio data, captures orbital changes over the glacial cycle and these are particularly evident at the two deepest water sites (U1313 and U1308).
Despite the larger relative uncertainty on δ 11 B-pH calculated core top pH and [CO 2− 3 ] data from all sites are in good agreement with measured carbonate data (Lauvset et al., 2016, shown in  ∼118 ± 20 μmol/kg, and pH ∼7.95 ± 0.05), owing to a NADW bathing all depths (as is the case today). Note that due to the pressure and temperature effects on the δ 11 B-pH proxy, the depth profile of δ 11 B is not equivalent to that of pH (see Fig. 3a vs. 4).
The linear transformation of B/Ca to CO 2− 3 (equation (2), Supplemental material), however ensures that the patterns in the CO 2− 3 data are very similar to raw B/Ca (Fig. 3b vs. Fig. 4). A comparison of the two data types, despite the limitations of our δ 11 B-pH record, shows a good agreement (Fig. 4, r 2 = 0.46 see Supplemental Fig. 1). This supports the assertion of Yu et al. (2010a) that when used in conjunction, these two B-based proxies represent a powerful proxy toolbox, with the analytically more straightforward B/Ca proxy capable of resolving small and high frequency changes and the more labour intensive δ 11 B-pH providing a robust framework by which to interpret changes in the ocean carbonate system.

Patterns of circulation in the North Atlantic over the last 150 kyr
To reconstruct changing stratification in the North Atlantic over the last glacial cycle, we compare our new estimates of bottom water carbonate ion derived from B/Ca ratios and pH from δ 11 B for Sites 980, 999, 1313 and 1308 to existing data from the North Atlantic (Yu et al., 2008), Caribbean Site VM28-122 (3623 m), Ceara Rise (RC16-59; Broecker et al., 2015), and South Atlantic (MD07-3076, Gottschalk et al., 2015, TN057-21, Yu et al., 2014 in a B/Ca time series in Fig. 5 and depth profiles in Fig. 6. Note that for VM28-122, as for Site 999, its location in the Caribbean Sea means that water mass ingression is controlled by sill depth -the effective depth of exchange with the Atlantic is therefore ∼1800 m. To first order, the depth-latitude transect of the sites shows that they were all bathed by a high [CO 2− Fig. 3. δ 11 B and B/Ca data. Boron isotope data from (I)ODP sites, a: red: 980, b: blue: U1313, dark blue: U1308 (dotted uncertainty envelope for clarity). External 2SD uncertainty is shown by the shading and dotted lines. c, d: B/Ca from the same sites and also ODP 999 in the Caribbean (C) revealing more structure due to the higher resolution. e, f: shows the LR04 stack of benthic isotopes (Lisiecki and Raymo, 2005) and Site specific δ 18 O: ODP 980 (Crocker et al., 2016), IODP U1308 (Hodell et al., 2008) and IODP U1313 (this study). b, d and g) of an enlarged low-[CO 2− 3 ] deep water mass shoaling to between 2800 and 3400 m -consistent with previous findings based on δ 13 C and B/Ca for the LGM (Curry and Oppo, 2005;Gebbie, 2014;Yu et al., 2008). These full glacial characteristics, with a strong chemical gradient at ∼2500 m are indicative of greater carbon storage in the deep ocean below a steepened chemocline than during the Holocene.
To a first order the carbonate system appears to be primarily depth-dependent in the North Atlantic, with the western basin (this study, Yu et al., 2010a;Broecker et al., 2015) following the same trends as a compilation of records for the eastern basin (Yu et al., 2008) for a given depth during the last 40 kyr ( Fig. 5a and b).
For the last glacial cycle, three distinct "modes" of carbonate system depth stratification are evident, each associated with a characteristic circulation pattern in the Atlantic (Fig. 6) (Broecker et al., 2015;Gottschalk et al., 2015;Yu et al., 2014Yu et al., , 2008). An interglacial "overturning mode", seen during the Holocene and in MIS5e with little vertical structure in [CO 2− 3 ] and pH (LOESS fit lines in Fig. 6, see supplemental materials section 2.3 for details); a "weak glacial stratification mode" with enhanced AABW penetration confined below 3400 m, seen during MIS5a-d and MIS3 (Figs. 6c and 6e); and a "full glacial mode", wherein the interface between northern and southern sourced water is shallowest, situated no deeper than 2800 m in the LGM and MIS4 featuring and the strongest vertical gradients in reconstructed carbonate chemistry ( Fig. 6b and 6d).
While both of our deep water sites show reductions in pH and carbonate in the cold periods of the last glacial cycle, the amplitude of change is greater (to lower values) in the eastern basin (Site U1308) than in the western basin (Site U1313), especially during MIS4 and the LGM/MIS2 (Fig. 5). We attribute this observation to the greater depth of Site U1308 (by ∼400 m) and therefore greater influence of AABW.

Controls on Atlantic carbonate system variations
Although the carbonate system is predominantly quasi-conservative in the modern ocean (see Fig. 1) and its properties strongly determined by water mass source, there is no guarantee that this characteristic of the system is maintained in the geological past. Given that the carbonate system tracers will react differently to δ 13 C to changes in biological modification and/or carbon flux the two proxy systems can be used together to elucidate both changes in water mass structure/geometry and non-conservative effects (e.g. Yu et al., 2008). Fig. 7 shows a cross plot of δ 13 C and B/Ca derived [CO 2− 3 ] from (I)ODP Sites 980, 999, U1308 and U1313. During the Holocene (solid circles) and MIS5e all available North Atlantic tracer data cluster around [CO 2− 3 ] ∼120 μmol/kg and δ 13 C ∼1h. This is consistent with one water mass bathing all these sites -NADW. Through the glacial cycle the sites diverge with ODP 980 and IODP U1308 following a trend line indicative of CO 2 efflux (ODP 980) and influx (IODP U1308, dashed blue "CO 2 evasion" line). This suggests that at these two East Atlantic Sites the carbonate system remains a quasi-conservative tracer over the glacial cycle, suited to identifying the influence of shallow, wellventilated GNAIW and deep, isolated AABW respectively in this basin. At Site 980 (considered our northern endmember), glacial pH also increased as expected from lower atmospheric CO 2 and a rapidly ventilated and shallow overturning cell (Fig. 4a). U1308 on the other hand shows no significant pH change in our record, Fig. 4. pH and carbonate data from individual Sites. Paired pH (total scale) and carbonate data from a: ODP 980, b: IODP U1313 and c: IODP U1308. d: Ice core CO 2 (Bereiter et al., 2015). In all sites the pH and carbonate reconstruction follow similar patterns adding confidence to the use of B/Ca as a carbonate proxy, although the pH data does not always have the required resolution to show the smaller changes revealed by the higher resolution B/Ca. Deep water on both sides of the Atlantic reaches SSW compositions of [CO 2− 3 ] at times during the glacial cycle. Shallower sites record high carbonate, high pH intermediate water throughout the glacial period. All diamonds = boron isotope derived pH and external 2SD, all circles = independent B/Ca derived [CO 2− 3 ] and external 2SD. but without a southern pH-endmember this is impossible to interpret (either as mixing, SSW or accumulation of respired carbon). We therefore limit most of our interpretation at this stage to the B/Ca-[CO 2− 3 ] record alone, which shows well resolved variation.
At U1313, the gradient in δ 13 C-[CO 2− 3 ] space is significantly steeper than the trend seen at U1303 and 980, representing much less conservative behaviour, as seen from the larger range of δ 13 C values. This evolution through the last glacial cycle shows that U1313 is either influenced by a change in southern sourced endmember composition (black diamond, in Fig. 7; Gottschalk et al., 2015;Yu et al., 2014) or exhibits an increased build-up of remineralized carbon along the water pathway (shown by black dashed 'biology' line, after Yu et al., 2008). U1313 also shows the largest resolvable changes through time in δ 11 B-pH (Fig. 4), which may also signal accumulation of extra carbon in the local deep water. This behaviour is most likely attributable to southern sourced water mass divergence north of the Romanche and Vema fracture zones before reaching sites U1313 and U1308. Our δ 13 C data point to a build-up of remineralized carbon in the western basin relative to the eastern basin. Although other processes may be contributory (e.g. carbonate compensation), these combined data sets suggest that both a change in water mass geometry and differential acquisition of carbon in eastern and western basins are required to explain the carbonate system patterns we have reconstructed over the last glacial cycle in the deep Atlantic Ocean.

Atlantic carbonate system depth stratification linked to CO 2 storage
In order to examine the temporal evolution of stratification of the carbonate system in the North Atlantic basin we define and plot a stratification index [CO 2− 3 ] * : the difference between [CO 2− 3 ] at Site 980 ([CO 2− 3 ] 980 ) and the available deep water sites (([CO 2− 3 ] 980 − [CO 2− 3 ] U1313, U1308 and RC16-59 ); Fig. 8). Site 980, is interpreted to have been continually bathed in northern-sourced waters during both glacials and interglacials throughout the Late Pleistocene ( Fig. 7 and Raymo et al., 1990). [CO 2− 3 ] * is then interpreted as the relative influence of northern and southern sourced waters in the deep part of the basin and/or the relative increase in respired carbon storage. High values of [CO 2− 3 ] * are therefore associated with enhanced carbonate system stratification between shallow and deep, and/or increased deep ocean carbon storage. Deep water entering the South Atlantic from either the Southern Ocean or the Pacific has a low [CO 2− 3 ] composition such as shown at sites TN057-21 at 41 • S, 7.8 • E 4981 m and MD07_3076 at 44 • S, 14 • W, 3770 m (Gottschalk et al., 2015;Yu et al., 2014), which is also true of Pacific Ocean water, as seen  (Broecker et al., 2015). b: compilation from the Eastern Atlantic basin, ODP 980, IODP 1308, other sites <2800 m pale red, pale grey >2800 m (BOFS and NEAPS Sites from Yu et al., 2008), black, South Atlantic (Gottschalk et al., 2015;Yu et al., 2014) c: 150 kyrs of data from the whole Atlantic basin (see above for colour scheme), the grey bars in a-c show modern Atlantic deep water composition, and the dotted line indicated modern southern sourced water composition. Pale shading indicates the range of values from the South Atlantic TN057 and MD07_3076 (Gottschalk et al., 2015). d: LR04 for glacial-interglacial reference (Lisiecki and Raymo, 2005). Shallow sites form an upper, high carbonate cell during the LGM and deeper sites a low carbonate cell. Depth stratification is eroded following the resumption of a strong AMOC following the last deglaciation (panels a and b) and the penultimate deglaciation (panel c).
by site RR0503-83 (Allen et al., 2015), see yellow oval in Fig. 7. Any excursion to higher values of [CO 2− 3 ] * (the down direction on Figs. 8 and 9), marks an increase in carbonate system stratification of the basin and greater influence of low [CO 2− 3 ] waters at depth. We characterise this low [CO 2− 3 ] water as having a southern source, particularly evident in the eastern Atlantic basin, with low δ 13 C, very low [CO 2− 3 ] and potential low pH (i.e. a relatively high DIC water mass). This strong influence of SSW influence at depth is evident from MIS5a-d through MIS2, with the most prominent intervals being MIS2, MIS4, MIS5d and MIS6 where deep water  ] is between 40 and 90 μmol/kg lower than northern sourced water (NSW) values (solid diamonds in Fig. 7, and Fig. 8).
In the western Atlantic basin, the steeper gradient of δ 13 C to  ] seen in Fig. 7 strongly implicates remineralized, biologically utilised carbon as being the driver of the lower [CO 2− 3 ] * of this basin. 3 ] data were calculated in the R statistical software, using a smoothing function which takes account of all depths but with decreasing weighting according to the cube of their distance in the depth domain (for more details see supplemental information section 3.3), we use data from the Holocene, MIS2 and MIS3 for these lines and apply them to other time slices which share 'modes' of circulation.
The temporal evolution of carbonate ion stratification in the North Atlantic bears some similarity to the depth stratification of δ 13 C ( δ 13 C, Fig. 8a). However, there are more resolvable differences in [CO 2− 3 ] than in δ 13 C between the sites especially during the MIS3 and 5a-d. The carbonate ion gradient plot suggests strong stratification throughout much of the last glacial cycle as opposed to merely during MIS2 and MIS4 as suggested by δ 13 C (Fig. 8b). The [CO 2− 3 ] and δ 13 C data are complementary for the most-part, but minor or short-lived SSW excursions may be more easily resolved in the [CO 2− 3 ] dataset because they will form a larger relative change. Regardless of processes represented in the proxy values, there is a good correlation between [CO 2− 3 ] * in the deep north Atlantic cores and atmospheric CO 2 (Fig. 9, crosscorrelation (r 2 ) of 0.59, similar to the 0.46 from Southern Ocean δ 13 C, Hodell et al., 2003). This correlation supports a contribution to CO 2 change throughout the last glacial cycle by changes in the stratification of the carbonate system in the Atlantic resulting from the expansion of the SSW cell. The residual ([CO 2− 3 ] * not contributing to CO 2 change) is shown in the lower panel of Fig. 9, and at lower CO 2 (glacial) values is generally close to analytical precision (<10 μmol/kg) and thus insignificant. At higher CO 2 levels (interglacial), the [CO 2− and ODP 999 (dark yellow hollow circles). Also shown are shaded endmember compositions for the Southern Ocean (grey oval, Yu et al., 2014) and the Pacific (yellow oval, Kerr et al., 2017). Holocene and last glacial data are highlighted by solid points (black-bordered circles and diamonds respectfully), the glacial northern endmember is assumed to be denoted by the cluster of glacial ODP 980 points in the upper-right corner. Linear fit lines are shown for ODP 980, IODP U1313, and IODP U1308 (solid coloured lines) and calculated lines for CO 2 addition/evasion (pale blue, dotted) and biological utilization (black, dotted). Sample gradients for alkalinity and temperature change are also shown, after Yu et al. (2008). other non-Atlantic processes dominating the relatively minor CO 2 changes in the Holocene and MIS5. Furthermore, during glacial onset U1308 shows a large change in [CO 2− 3 ] * but generally has a weaker correlation to CO 2 over the whole cycle than the West Atlantic (r 2 = 0.38 and 0.62 respectively). During termination 1 (and 2) both eastern and western basin [CO 2− 3 ] * indices show a strong relationship with CO 2 as would be expected from the large and fast collapse in the deep ocean carbon storage reservoir that occurred at these times.

Conclusions
Using relatively low temporal resolution δ 11 B data combined with higher resolution B/Ca, we find that we can divide Atlantic circulation change throughout the last glacial cycle into three main modes: (i) An interglacial "overturning mode", seen in the Holocene and MIS5e; (ii) a "weak glacial mode" with enhanced AABW penetration confined to below 3400 m, seen in MIS5a-d and MIS3, with a well ventilated circulation in the upper cell, shown by , shows a correlation, suggesting the deep Atlantic is an important C-store at least partially moderating G-IG CO 2 change, the gradient is −0.47. The r = 0.77 (r 2 = 0.59) is similar, but greater than that found from δ 13 C gradients, overall the western Atlantic [CO 2− 3 ] * (1313) has a better correlation with CO 2 than the eastern Atlantic (1308, r 2 = 0.62 vs. 0.38). The residuals from this relationship are shown in the lower panel. All values are <25 μmol kg −1 (2SD = 20.6 μmol kg −1 ), which is similar in size to the square root addition of analytical and calibration uncertainty of 2 individual measurements (14.1 μmol kg −1 ), although there is an increase towards higher CO 2 values suggested that deep ocean processes are less strongly linked to interglacial climate states.
high carbonate ion and high pH at a depth of ∼2200 m; and (iii) a "full glacial mode", with the interface between NSW and SSW at no less than 2800 m in the LGM and MIS4. It is likely that CO 2 drawn down into the surface Southern Ocean during cold intervals and stored in an expanded deep ocean southern sourced cell was a quantitatively important process (Fig. 5). This hypothesis is well supported by the correlation observed between atmospheric CO 2 and deep ocean carbonate parameters in the North Atlantic (Fig. 9), and hence the increased volume of high DIC low [CO 2− 3 ] water. In addition we find that changes in the eastern basin occur early in the glacial cycle, alongside the stadial-interstadial changes in MIS5. The western basin [CO 2− 3 ] content exhibits greater correlation with atmospheric CO 2 during the coldest intervals compared to the eastern basin as the water there appears to accumulate more respired carbon. Improved spatial coverage of data from all ocean basins is however needed to further investigate this pattern and to better understand the capacity of the oceans to drawdown and release of carbon, a process of key importance for ocean-atmosphere interactions in our warming future. Boron-based proxies demonstrate semi-conservative behaviour over the last glacial cycle which, in concert with either non-conservative or fully conservative tracers such and δ 13 C and ε Nd , makes them a powerful constraint on the results and mechanisms of carbon cycle change.