The initial rise in atmospheric CO2 during the last deglaciation was likely driven by input of carbon from a 13C-depleted reservoir. Here we show that high resolution benthic foraminiferal records from the mid-depth Brazil Margin display an abrupt drop in δ13C during Heinrich Stadial 1 (HS1) that is similar to but larger than in the atmosphere. Comparing the Brazil Margin results to published records from the North Atlantic, we are unable to account for the South Atlantic δ13C data with conservative mixing between northern and southern component water masses. Rapid input of abyssal water from the Southeast Atlantic could account for deglacial δ13C anomalies at the Brazil Margin but it would require a reversal in deep water flow direction compared to today. A new mid-depth water mass may explain similar HS1 δ13C values in both the North and South Atlantic, but contrasting oxygen isotopic values between the two basins do not support such a scenario. Instead, it appears that δ13C behaved non-conservatively during the deglaciation, possibly reflecting the input of carbon from an isotopically depleted source.
 The rise in atmospheric carbon dioxide between the Last Glacial Maximum (LGM) and the Holocene was first recognized over 30 years ago [Neftel et al., 1982], yet the underlying mechanisms responsible for the CO2 change remain unclear [Sigman and Boyle, 2000; Sigman et al., 2010]. High resolution ice core records indicate that the initial 30 ppmv increase in CO2 from ~17 to 16 kyr BP coincided with a decrease in the δ13C of atmospheric CO2 of 0.3‰ [Schmitt et al., 2012]. The synchroneity of the signals suggests that the release of carbon from a 13C-depleted reservoir into the atmosphere was a key initiator of the last deglaciation.
 Carbon isotope minima were widespread phenomena in the surface and mid-depth ocean during the last deglaciation [Oppo and Fairbanks, 1989; Ninnemann and Charles, 1997; Spero and Lea, 2002; Curry et al., 1988; Peck et al., 2007]. The largest δ13C anomalies occurred in the mid-depth North Atlantic. South of Iceland, δ13C decreased by ~1‰ at water depths from 1200 m to 2300 m [McManus et al., 1999; Rickaby and Elderfield, 2005; Thornalley et al., 2010]. Similar anomalies occurred at 1300 m in the western tropical Atlantic [Zahn and Stuber, 2002] and 1100 m in the eastern subtropical Atlantic [Zahn et al., 1997]. A variety of mechanisms have been proposed to explain the carbon isotope minima, including regional brine formation [Dokken and Jansen, 1999; Thornalley et al., 2010; Waelbroeck et al., 2011], greater incursion of southern source intermediate water [Rickaby and Elderfield, 2005], and weakening of the Atlantic meridional overturning circulation [Zahn et al., 1997]. The latter mechanism appears to be consistent with evidence for reduced export of Pa from the North Atlantic relative to the LGM [Gherardi et al., 2009].
 The Brazil Margin cores used in this study are ideally located to evaluate the evolution of Atlantic water masses during the deglaciation. In the modern Southwest Atlantic, the cores span the transition between Antarctic Intermediate Water (AAIW), Upper Circumpolar Deep Water (UCDW), and North Atlantic Deep Water (NADW) (Figure 1). During the LGM, the core locations were influenced by Glacial Antarctic Intermediate Water (GAAIW), Glacial North Atlantic Intermediate Water (GNAIW), and Glacial Antarctic Bottom Water (GAABW) [Curry and Oppo, 2005]. Below we show that the Brazil Margin δ13C time series display a negative excursion similar to the record of atmospheric δ13C during the deglaciation. We then discuss whether the Brazil Margin data can be explained by conservative mixing between northern and southern waters.
 The analyses presented in this paper are based on six cores from 1200 m to 2500 m water depth that were retrieved during the KNR159-5 cruise to the Brazil Margin [Curry and Oppo, 2005]. Cores discussed in this paper include 36GGC, 17JPC, 78GGC, 33GGC, 42JPC, and 30GGC (Table 1). Each core was sampled at 4 or 5 cm intervals, and the resulting samples were then freeze-dried, washed through a 150 µm sieve, and dried at 40°C.
Table 1. Locations and Water Depths of Brazil Margin Cores Used in This Papera
Water Depth (m)
All cores were retrieved during R/V Knorr cruise KNR159-5.
 Planktonic foraminifera (Globigerinoides ruber and Globigerinoides sacculifer) were picked from the >250 µm size fraction of each sample for accelerator mass spectrometry (AMS) radiocarbon dates. Radiocarbon dating was carried out at the Keck Carbon Cycle Accelerator Mass Spectrometry laboratory at University of California, Irvine, where the samples underwent a 10% leach using 0.01 N HCl to ensure removal of any modern 14C. The foraminifera were then hydrolyzed in 85% phosphoric acid and the resulting CO2 was combined with hydrogen and iron powder at 560°C to create graphite. The graphite was analyzed using AMS to obtain 14C results.
 Modern reservoir ages along the coast of southeastern Brazil are 407 ± 59 years (1σ), equivalent to a ΔR of 7 ± 59 years [Angulo et al., 2005]. For age calibration purposes, we used a ΔR 0 ± 200 years (1σ) to account for unknown changes in reservoir age in the geologic past. Calendar ages were calibrated using Calib v.6.0 (http://calib.qub.ac.uk/calib/). The uncertainties in calibrated calendar ages, typically ±300 years (1σ), are due primarily to our assumed uncertainty in ΔR.
2.2 Stable Isotopes
 Benthic stable carbon and oxygen isotopic analyses were based on individual tests of Cibicidoides species from the >250 µm size fraction. Analyses were run on a Finnigan MAT 253 triple-collector gas source mass spectrometer coupled to a Finnegan Kiel automated carbonate device at the University of Michigan's Stable Isotope Laboratory. Isotope values were corrected to Vienna Pee Dee Belemnite using National Bureau of Standards (NBS) 19 (n = 78, δ13C = 1.94 ± 0.04‰, δ18O = −2.19 ± 0.07‰) and NBS 18 (n = 12; δ13C = −5.00 ± 0.04‰, δ18O = −22.98 ± 0.06‰). Samples for cores 42JPC and 36GGC were run at Woods Hole Oceanographic Institution [Oppo and Horowitz, 2000; Curry and Oppo, 2005].
 In addition to NBS 18 and NBS 19, we used the Atlantis II standard (δ18O = 3.42‰; Ostermann and Curry, ) to constrain the “heavy” end of the oxygen isotope spectrum. Atlantis II standards run during this study yield a mean δ18O of 3.47 ± 0.08‰ (n = 90), within one-sigma error of the 3.42‰ value presented in Ostermann and Curry . Atlantis II δ18O values run during analysis of samples for core 17JPC averaged 3.52‰. In this instance, we subtracted 0.1‰ from the unknowns to compensate for the higher than normal δ18O results.
3.1 Age Models
 Age models for 17JPC, 78GGC, 33GGC, and 30GGC are shown in Figure 2. The age models for 36GGC and 42JPC are presented in Sortor and Lund  and Hoffman and Lund , respectively. Each core has sedimentation rates of 2–3 cm/kyr during the Holocene with higher rates during the deglaciation and LGM, ranging from 5 cm/kyr to 35 cm/kyr (Figure 2). High accumulation rates for 78GGC and 33GGC suggest the core locations were sediment drift sites during the deglaciation and LGM. The cores generally lack modern core tops, with the most recent material dating to between 0.5 kyr BP and 3 kyr BP. Nonzero core top ages may be due to a lack of sediment deposition, erosion by deep currents, or an artifact of the coring process.
 The radiocarbon results illustrate clear age reversals in cores 78GGC, 17JPC, and 30GGC. Deglacial age reversals have been documented in other cores from the Brazil Margin; Sortor and Lund  used stable isotopic data to conclude that reversals in 36GGC were due to deep burrowing. A similar phenomenon appears to have occurred in 78GGC, where anomalously low benthic δ18O values from 69 cm to 81 cm coincide with an interval of young radiocarbon ages (Figure 3). The anomalous δ18O points have values ranging from 3.5‰ to 3.7‰, indicating they originated from the 35 cm to 45 cm stratigraphic interval. Radiocarbon ages of ~13.5 kyr BP in the disturbed section support this interpretation. We exclude these data from the age model for 78GGC. Note that the section of the core with the large decline in δ13C (discussed below) occurs prior to the disturbed section. Two additional isolated reversals prior to HS1 (at 145 cm and 165 cm) are also excluded from the age model for 78GGC (Figure 2b).
 In the case of 30GGC, there are two possible age models; one that includes the age at 41 cm and one that excludes it (Figure 2d). This age is not a reversal per se but would require a drastic change in sedimentation rate. Since there is a clear reversal at 39 cm, we chose to use the age model that excludes both points (dashed black line). Large age reversals in 17JPC at 50 cm, 58 cm, 66 cm, and 82 cm were excluded from the age model for this core. Age models for 78GGC, 17JPC, and 30GGC were produced by linear interpolation between the remaining calendar ages. Age inconsistencies in 33GGC are subtler than the other cores; many are within error of adjacent dates in the stratigraphy (Figure 2c). As a result, we used a polynomial fit to construct the age model rather than linear interpolation between individual points.
3.2 Stable Isotopic Time Series
 Benthic foraminiferal δ18O time series for the Brazil Margin are shown in Figure 4. The contrast between LGM and Holocene δ18O values is ~2.0‰ at water depths below 2000 m compared to ~1.8‰ above 2000 m, consistent with results spanning the full water column range at the Brazil Margin [Curry and Oppo, 2005; Lund et al., 2011]. The new radiocarbon constrained records indicate the timing of the LGM to Holocene shift varies by water depth (Figure 4). Above 2200 m, benthic δ18O began to decrease at 17.5–18.0 kyr BP whereas below 2200 m, the δ18O decrease occurred after 17.0 kyr BP, similar to the depth-dependent timing noted at other Atlantic sites [Waelbroeck et al., 2011].
 Benthic δ13C in the deepest cores (2300–2500 m) and shallowest core (1300 m) increased by 0.5–0.8‰ between the LGM and Holocene, while those in between (1800–2100 m) show little glacial-interglacial difference (Figure 4). The most striking feature of the δ13C records is the pronounced negative excursion near the beginning of Heinrich Stadial 1 (HS1; 14.5–17.5 kyr BP). The decrease is largest between 1600 m to 2100 m water depth (~0.5‰), with smaller excursions at 2300 m and 2500 m (~0.2‰). The one exception where δ13C increased monotonically during the deglaciation is at 1300 m water depth.
 The two high resolution records at 1800 m and 2100 m have δ13C histories similar to the atmosphere (Figure 5). Each record displays a pattern of change that resembles a leaning “W”, including an abrupt drop in δ13C near ~18 kyr BP, a partial recovery by 14 kyr BP, a modest decrease at 12–13 kyr BP, and finally a gradual increase into the mid-Holocene. The magnitude of the oceanic δ13C anomalies is ~0.2‰ larger than in the atmosphere. For each benthic δ13C record, the maximum rate of decline occurred at 17.8 kyr BP, assuming the regional offset in surface water reservoir age (ΔR) was 0 ± 200 years (1σ) (Figure 5). If instead ΔR was 400 years, the maximum rate of δ13C change would have happened at 17.3 kyr BP. By comparison, the maximum rate of δ13C change in the atmosphere occurred at approximately 17.2 kyr BP. If the age models for these records are correct, it appears that the rapid decrease in δ13C at the Brazil Margin either led or was synchronous with the atmosphere.
 The strong correspondence between the Brazil Margin and atmospheric δ13C records suggests they are linked by a common mechanism. One possibility is that the Brazil Margin δ13C anomalies reflect changes in the composition of northern component water or its proportion relative to southern component water. Alternatively, input of abyssal water from the South Atlantic could have driven the δ13C signal. In either case, the apparent changes in circulation represented by the benthic δ13C records must have also led to outgassing of 13C-depleted carbon from the ocean in such a way that both the oceanic and atmospheric δ13C records had a similar pattern. Below, we first review LGM water mass properties at the Brazil Margin to provide context for the deglaciation. We then evaluate the different circulation mechanisms that could potentially account for the deglacial anomalies.
4.1 LGM Water Mass Proportions
 During the LGM, Brazil Margin δ18O and δ13C pairs from 1800 m to 4000 m water depth fall on a mixing line between GNAIW and GAABW (Figure 6a). Assuming δ13C values of 1.4‰ for GNAIW [Curry and Oppo, 2005] and −0.2‰ for GAABW [Hoffman and Lund, 2012] (Table 2), the δ13C results at 1800 m can be explained by a mixture of 25% GAABW and 75% GNAIW (Table 3). Given that δ13C is a non-conservative tracer, it is important to validate these proportions using δ18O. We do so by estimating the predicted δ18O at each depth given the δ13C-based proportions and the δ18O end-member values for GNAIW and GAABW. Using a δ18O of 4.2‰ for GNAIW [Curry and Oppo, 2005] and 4.9‰ for GAABW [Hoffman and Lund, 2012], the predicted δ18O at 1800 m is 4.38‰, indistinguishable from observed value of 4.40‰ (Table 3). The predicted and observed LGM δ18O values at 2100 m, 2300 m, and 2500 m are also in good agreement.
Table 2. Stable Isotope Values for LGM End-Members in the Atlantica
End-member values for GAAIW and GNAIW are based on isotope data presented in Curry and Oppo . End-member values for GAABW are based on the results from Hoffman and Lund .
Table 3. LGM Proportions of SCW and NCW at the Brazil Margina
Water depth (m)
% GAAIW or GAABW
Proportions are estimated using δ13C because of the large range in the end-member δ13C values. The resulting proportions are then used to estimate the expected δ18O at each water depth.
 Between 1100 m and 1600 m, the isotopic data generally fall between GNAIW and GAAIW (Figure 6a). Water at 1600 m depth is composed of ~50% GAAIW and ~50% GNAIW while water at 1300 m appears to consist largely of GAAIW. The offset between observed and estimated δ18O for the shallower cores is larger than for the deeper sites (Table 3). This is most likely due to the small contrast in δ18O values for GAAIW (4.3‰) and GNAIW (4.2‰) relative to the uncertainty in mean LGM values at each water depth (±0.05‰). Nevertheless, the Brazil Margin data suggest that δ13C at mid-depths acted as a largely conservative tracer during the LGM, consistent with results from deeper in the water column [Hoffman and Lund, 2012].
4.2 Northern Component Influence During HS1
 An apparent reorganization in water mass structure occurred during HS1 (Figure 6b). Between 1800 m to 2300 m water depth, δ18O decreased 0.4–0.6‰ while δ13C decreased 0.2–0.5‰. In comparison, δ18O of southern component water (SCW) decreased by ~0.2‰ while its δ13C increased ~0.2‰ [Hoffman and Lund, 2012]. Northern component water (NCW) underwent the largest isotopic shift, with δ18O and δ13C decreasing by ~1.0‰ and ~0.7‰, respectively. Assuming HS1 end-member δ18O values of 3.1‰ for NCW and 4.6‰ for SCW, the HS1 results from 1800 m to 2300 m water depth can be explained by a mixture of approximately 60% SCW and 40% NCW. Compared to the LGM proportions (Table 3), this represents a substantial reduction in the influence of NCW at the Brazil Margin. Unlike the LGM, however, the HS1 δ18O-δ13C pairs do not fall on a mixing line between NCW and SCW. Projecting the HS1 δ18O results to the mixing line, we estimate that δ13C for the mid-depth Brazil Margin sites should range from 0‰ at 2500 m to 0.3‰ at 1300 m, about 0.2–0.4% lower than observed (Figure 6b).
 Given that the Atlantic circulation was probably not in steady state during the deglaciation, a full accounting of isotopic records in the Southwest Atlantic should consider transient changes in NCW and SCW. Unfortunately, surface water reservoir age variability of up to several hundred years in the South Atlantic and more than 1000 years in the North Atlantic [Waelbroeck et al., 2001] precludes reliable quantification of the effect of transient end-member variability on the Brazil Margin records.
4.3 Input of Abyssal Water
 The Brazil Margin δ13C results imply that water depleted in 13C invaded the mid-depth South Atlantic during the deglaciation. Because the abyssal Southwest Atlantic was characterized by low δ13C during the LGM, input of water from the abyss could account for the abrupt changes at mid-depth. The flow path would presumably involve upwelling south of the Antarctic Polar Front [Marshall and Speer, 2012; Anderson et al., 2009] and the subsequent formation and northward advection of a new mid-depth water mass. Just prior to HS1, GAABW in the Brazil Basin had a δ13C of −0.2 ± 0.2‰ [Hoffman and Lund, 2012]. Accounting for the HS1 δ13C excursions would require 40 ± 10% of the water at mid-depth to be replaced by GAABW, assuming no dilution by mixing during its transit to the Brazil Margin.
 Input of GAABW would also influence benthic foraminiferal δ18O. During the LGM, GAABW had a δ18O of 4.9 ± 0.1‰ [Hoffman and Lund, 2012]. Rapid addition of 40 ± 10% GAABW would cause δ18O to increase by 0.2 ± 0.05‰, yet the records at 1800 m and 2100 m instead show either no change or a small decrease from the LGM to HS1 (Figure 7). If GAABW was the source of the δ13C anomalies, the associated δ18O signal must have been offset by warming or the input of 18O-depleted water from the surface ocean.
 Results from the Cape Basin suggest abyssal water in the Southeast Atlantic had a δ13C of −0.9 ± 0.1‰ and δ18O of 4.2 ± 0.1‰ [Ninnemann and Charles, 2002; Waelbroeck et al., 2011]. Although benthic δ13C in this region may be influenced by epibenthic decay of organic matter [Mackensen et al., 1993] and the δ18O results may be biased by the use of common acid bath instrumentation [Ostermann and Curry, 2000; Hodell et al., 2003], the potential influence of this isotopically distinct water mass must also be considered. Accounting for the Brazil Margin δ13C anomalies would require ~25% of the water at mid-depths to originate from the Southeast Atlantic. Adding this proportion would cause only a slight (<0.1‰) decrease in δ18O, consistent with observations (Figure 7). However, the net advective flow at 1–3 km water depth in the modern South Atlantic is eastward from the Brazil Basin to the Cape Basin, followed by entrainment in the Antarctic Circumpolar Current (ACC) [Sloyan and Rintoul, 2001]. Assuming a similar circulation pattern during the deglaciation, the isotopic signal would need to be transmitted from the Southeast Atlantic via the ACC around Antarctica with little or no dilution by mixing. Although we cannot rule out this possibility, it appears to be an unlikely explanation of the large and abrupt δ13C signal at the Brazil Margin.
4.4 A New Mid-Depth Water mass
 Carbon isotopic results from the North and South Atlantic suggest a single 13C-depleted water mass occupied mid-depths during HS1. In the western tropical North Atlantic, a region where large changes in reservoir age are unlikely, δ13C dropped abruptly at approximately 18 kyr BP, reached a minimum value of ~0.6‰ during HS1, and then recovered slowly into the mid-Holocene [Zahn and Stuber, 2002] (Figure 8). This pattern is very similar to that observed at the Brazil Margin. Homogeneous δ13C values are observed throughout the mid-depth western Atlantic during HS1 [Oppo and Curry, 2012]. Indeed, vertical profiles from the North and South Atlantic show that the mid-depth records are characterized by a δ13C of 0.6 ± 0.1‰ (Figure 9a,b). This pattern is strikingly different than the LGM where the clear north-south contrast in δ13C reflects the dominant influence of GNAIW in the North Atlantic. In the South Atlantic, GNAIW manifests itself as δ13C maximum near 1800 m water depth.
Table 4. Locations of Published High-Resolution Records From the North Atlantic Used in This Paper
 Unlike δ13C, benthic δ18O profiles from the North and South Atlantic are very different during HS1. At the Brazil Margin, δ18O ranges from 3.8‰ at 1300 m to 4.5‰ at 2500 m (Figure 9d). In the North Atlantic, δ18O values are lower throughout the same depth interval, spanning from 3.1‰ to 4.1‰ (Figure 9c). At the Brazil Margin, HS1 δ18O values decreased by 0.1‰ to 0.5‰ relative to the LGM, whereas in the North Atlantic, the δ18O decrease is much larger, ranging from 0.5‰ to 1.0‰. The results suggest the δ18O anomaly originated in the North Atlantic, perhaps due to influx of 18O-depleted melt water. The different δ18O histories in the two basins are also apparent in the time series in Figure 8. Throughout the LGM and deglaciation, benthic δ18O in the western tropical Atlantic is consistently 0.5‰ to 0.8‰ lower than at the Brazil Margin even though the δ13C history in the two locations is nearly identical. The δ18O data highlight the existence of two separate water masses at mid-depth, in contrast to the δ13C results. Given that δ18O is a conservative water mass tracer [Lund et al., 2011], it appears that another process is necessary to account for the δ13C minima.
 What could cause a similar δ13C history in the North and South Atlantic? Air-sea gas exchange with a 13C-depleted atmosphere would cause the δ13C in the source regions for northern and southern component water to decrease. If this were the case, the signal must have somehow been amplified given the larger oceanic δ13C anomaly (Figure 5). Temperature-driven amplification [e.g., Broecker and Maier-Reimer, 1992] is unable to account for the larger oceanic change because such a process would yield opposite δ13C signals in the ocean and atmosphere [Spero and Lea, 2002; Ninnemann and Charles, 1997].
 The timing of the Brazil Margin δ13C anomalies also appears to be inconsistent with an atmospheric driver. The mean ventilation age at 2000 m water depth in the modern Southwest Atlantic is approximately 400 years [Gebbie and Huybers, 2012]. If enhanced Southern Ocean upwelling and formation of a new water mass created the δ13C minima, the shift in atmospheric δ13C should lead mid-depth records by several hundred years. Instead, the oceanic records either lead or are synchronous with the atmosphere (Figure 5). An increase in surface ocean ΔR at the Brazil Margin in excess of 500 years is necessary for the atmosphere to feasibly lead the oceanic signal. Large shifts in ΔR seem unlikely at the latitude of the Brazil Margin where subtropical convergence causes a ΔR of 0 years today [Angulo et al., 2005]. Given age model uncertainties, establishing the exact phasing of the atmospheric δ13C relative to 2000 m water depth will require the development of high-resolution planktonic δ13C records from the Brazil Margin. Regardless of the relative timing, however, the magnitude of the oceanic anomalies indicates the atmosphere was not the primary driver of common δ13C variability in the North and South Atlantic.
4.5 Non-Conservative Behavior of δ13C
 Our discussion of water masses during the LGM and HS1 is based on the assumption advection and diffusion are the primary factors that influence benthic foraminiferal δ13C. Although benthic δ13C at the Brazil Margin acted conservatively during the LGM and Holocene [Hoffman and Lund, 2012], it may have acted non-conservatively during the deglaciation. The enigmatic nature of the Brazil Margin isotopic results and the apparent ubiquity of the HS1 δ13C anomaly in the Atlantic suggest that another source of carbon may have been involved. Biological remineralization of marine organic carbon is an unlikely culprit because it would cause δ13C to be lower than expected from water mass mixing rather than higher (Figure 6b). Evaluating other potential carbon sources will require high-resolution reconstructions from a range of locations to determine whether the isotopic anomalies are consistent with an alternative reservoir.
 Mid-depth benthic foraminiferal records from the Brazil Margin display abrupt negative δ13C anomalies similar to the atmosphere during the last deglaciation. The magnitude of the Brazil Margin δ13C signal is larger than in the atmosphere, similar to the pattern in published δ13C records from the North Atlantic. These findings strongly point towards the ocean as a source of the deglacial atmospheric δ13C anomaly.
 Broadly speaking, the coherence between the Brazil Margin and atmospheric δ13C records can be explained in one of two ways. One possibility is that changes in the oceanic circulation simultaneously altered the water mass distribution in the Southwest Atlantic and influenced the outgassing of 13C-depleted carbon from the abyssal ocean. Alternatively, input of 13C-depleted carbon from another source could have moved both the upper ocean and atmosphere towards more depleted values. We evaluate the first possibility by comparing the Brazil Margin results with stable isotopic constraints for Atlantic water masses during the LGM and HS1. During the LGM, stable isotopic results from mid-depths can be explained in terms of conservative mixing between northern and southern component waters. During HS1, however, we are unable to easily account for the Brazil Margin results with a simple two end-member mixing model. A new mid-depth water mass characterized by low δ13C is apparently required.
 Given that GAABW had low δ13C values during the LGM, it is logical invoke input of water from the abyss as a source of 13C-depleted water. By mass balance, the δ13C anomalies would require a large fraction of the water at mid-depths to be pure GAABW from the Southwest Atlantic. Input of this quantity of GAABW would also cause δ18O to increase, yet little change is observed. The influence of GAABW could be minimized via warming or the input of 18O-depleted water, but it is unlikely that such compensation would perfectly offset the abyssal signal. Water from the deepest Southeast Atlantic could be a feasible source of low δ13C but the flow path required (from the Cape Basin to the Brazil Basin) is opposite that observed in the modern South Atlantic.
 Input of abyssal water may help explain the Brazil Margin results but it does little to reconcile conflicting stable isotope data in the North and South Atlantic. Similar absolute δ13C values throughout the Atlantic imply that a single water mass dominated the mid-depths during HS1. However, δ18O results from the same cores show a strong meridional gradient in δ18O, with isotopically depleted values in the north and comparatively enriched values to the south. Given that δ18O is a conservative tracer, these data suggest two distinct water masses influenced the mid-depth Atlantic during HS1. Benthic foraminiferal δ13C, on the other hand, is a non-conservative tracer that is influenced by advection, diffusion, and the input of carbon from other sources. Given that biological remineralization of marine organic matter does not appear to be a viable explanation of the δ13C anomalies, another source of carbon external to the ocean-atmosphere system may have been involved.
 We would like to thank Lora Wingate for performing stable analyses. We are also grateful to John Southon for his oversight of the radiocarbon analyses. We would like to thank Jamie Hoffman, Rachel Franzblau, Rachel Seltz, and Elliot Jackson for sample processing. We are grateful to the WHOI core lab for sample collection and archiving and to Bill Curry and Delia Oppo for the opportunity to work on the Brazil Margin cores. This work was supported by NSF grant OCE-1003500.