Deep-sea coral evidence for lower Southern Ocean surface nitrate concentrations during the last ice age

Significance

The concentration of atmospheric carbon dioxide (pCO₂) varies by 80–100 ppm on glacial–interglacial timescales, with lower pCO₂ during the ice ages. In the modern Southern Ocean, the surface nutrients are not fully consumed by phytoplankton, resulting in leakage of deeply sequestered CO₂ to the atmosphere. It has been suggested that more complete nutrient consumption in the Southern Ocean would have caused the lower pCO₂ during the ice ages. Here, we provide the most spatially comprehensive evidence to date in support of the proposal that the entire Southern Ocean was nutrient-depleted during the last ice age relative to modern conditions. These data are consistent with the hypothesis that Southern Ocean changes contributed to the lower atmospheric pCO₂ of the ice ages.

Abstract

The Southern Ocean regulates the ocean’s biological sequestration of CO₂ and is widely suspected to underpin much of the ice age decline in atmospheric CO₂ concentration, but the specific changes in the region are debated. Although more complete drawdown of surface nutrients by phytoplankton during the ice ages is supported by some sediment core-based measurements, the use of different proxies in different regions has precluded a unified view of Southern Ocean biogeochemical change. Here, we report measurements of the ¹⁵N/¹⁴N of fossil-bound organic matter in the stony deep-sea coral Desmophyllum dianthus, a tool for reconstructing surface ocean nutrient conditions. The central robust observation is of higher ¹⁵N/¹⁴N across the Southern Ocean during the Last Glacial Maximum (LGM), 18–25 thousand years ago. These data suggest a reduced summer surface nitrate concentration in both the Antarctic and Subantarctic Zones during the LGM, with little surface nitrate transport between them. After the ice age, the increase in Antarctic surface nitrate occurred through the deglaciation and continued in the Holocene. The rise in Subantarctic surface nitrate appears to have had both early deglacial and late deglacial/Holocene components, preliminarily attributed to the end of Subantarctic iron fertilization and increasing nitrate input from the surface Antarctic Zone, respectively.

Phytoplankton grow in the sunlit surface waters of the ocean, transforming carbon dioxide (CO₂) into organic carbon, and the portion of this organic carbon that sinks out of the surface ocean (“export production”) effectively transfers CO₂ from the surface waters and the overlying atmosphere into the dark, deep ocean. In parallel with carbon, the nutrients required in large quantities by all phytoplankton (the “major nutrients” nitrogen, N, and phosphorus, P) are also exported out of the surface ocean and stored in the deep ocean. The “efficiency” of the biological pump is a global measure of the degree to which marine organisms exploit the major nutrients in the ocean to produce sinking organic matter. A higher efficiency of the biological pump stores more CO₂ in the deep ocean, lowering the partial pressure of CO₂ (pCO₂) in the atmosphere (1).

In today’s Southern Ocean, due to limitation by light and the trace nutrient iron (2, 3), phytoplankton consume only a small fraction of the available N and P, lowering the efficiency of the global ocean’s biological pump, which is manifested regionally as the leakage of deeply stored CO₂ through the Southern Ocean surface and back to the atmosphere (4). Changes in the degree of N and P consumption in the Southern Ocean may have played a role in past changes in atmospheric pCO₂, especially on glacial–interglacial timescales (5). Since the discovery that the ice age pCO₂ was approximately one-third lower than the preindustrial pCO₂ (6), research has been ongoing to reconstruct the biogeochemistry of the Southern Ocean over glacial cycles (5, 7, 8). Relative to interglacials such as the Holocene, reconstructed ice age export production was higher in the Subantarctic Zone (SAZ) but lower in the Antarctic Zone (AZ) (5, 9). Understanding this pattern requires additional biogeochemical information. A complementary constraint, which speaks more directly to air–sea CO₂ exchange, is how the degree of nutrient consumption varied in each of these two zones of the Southern Ocean.

The nitrogen isotopic composition of organic matter has the potential to record the degree of nitrate (NO₃⁻) consumption in Southern Ocean surface waters in the past. During nitrate uptake, phytoplankton preferentially consume ¹⁴N relative to ¹⁵N, resulting in a correlation between the δ¹⁵N of sinking organic matter and the fraction of the nitrate supply that is consumed in the surface ocean, where δ¹⁵N = [(¹⁵N/¹⁴N)_sample/(¹⁵N/¹⁴N)_air] − 1. The early application of this correlation was with bulk sedimentary N (10). However, bulk sedimentary δ¹⁵N can be biased due to diagenetic alteration and contamination by foreign N input, with evidence for major artifacts from these processes in sediment records from both low- and high-latitude sites (11, 12). The δ¹⁵N of organic matter bound within diatom and foraminifera microfossils avoids these issues and has been applied to reconstruct the nitrate consumption in the AZ and SAZ of the Southern Ocean, respectively (12, 13).

To complement isotopic studies of planktonic microfossil-bound N, we have pursued the N isotopic composition of organic matter bound within the carbonate skeleton of deep-sea scleractinian corals (14). Relative to the sedimentary microfossil-based δ¹⁵N records, deep-sea corals have the advantages that (i) they feed on organic matter that derives from the sinking flux out of the surface ocean, the δ¹⁵N of which is forced by mass balance to covary with the degree of nitrate consumption in surface waters (14); (ii) they can be precisely dated with uranium–thorium (U–Th) and radiocarbon methods, with each coral being an independent constraint on past conditions; and (iii) they are found in both the SAZ and AZ and thus can provide a complete picture of the Southern Ocean from a single fossil type, down to the species level. The >250 fossil corals used in this study were collected from the Drake Passage (DP), with collection sites in both the AZ and SAZ, and from the SAZ south of Tasmania (Fig. 1). All of the corals used in this study are of the solitary species Desmophyllum dianthus. The corals were dated by radiocarbon, with ∼50 of them also dated using U–Th techniques (Fig. 2). Prior work dating these corals by both approaches provides a basis for the radiocarbon reservoir age correction (15, 16). The five coral sites in the DP can be separated into an SAZ group (Burwood Bank and Cape Horn, to the north of the Subantarctic Front) and an AZ group (Sars, Interim, and Shackleton Fracture Zone, to the south of the Subantarctic Front, Fig. 1 and Fig. S1).

Fig. 1.

Deep-sea coral sites relative to surface nitrate concentration, ocean fronts, and bathymetry. (A) Southern hemisphere map showing climatological austral summer (December, January, and February) surface nitrate concentration (color scale), fossil coral sites (open, black-rimmed circles), and ocean fronts [white dotted lines, from north to south: Subtropical Front (STF), Subantarctic Front (SAF), Antarctic Polar Front (APF), and Southern Antarctic Circumpolar Current Front (SACCF)]. Shown as open triangles are the locations of two published sediment-based fossil δ¹⁵N records: foraminifera-bound δ¹⁵N at ODP Site 1090 (12) and diatom-bound δ¹⁵N at PS75/072-4 (13). (B) Bathymetric map (color scale) centered on Tasmania showing the deep-sea coral sites (overlapping black circles), austral summer surface nitrate concentration (white contours, in units of micromolar concentration), and the Subtropical and Subantarctic Fronts (yellow lines). (C) As in B but centered on DP and showing its deep-sea coral sites (BB, Burwood Bank; CH, Cape Horn; IN, Interim; SA, Sars; SFZ, Shackleton Fracture Zone).

Fig. 2.

Nitrogen isotopes of Southern Ocean deep-sea corals. (A) All Southern Ocean corals in this study, with the error bars representing analytical precision (1σ) calculated from two to three replicates of the same subsample from each coral septum; and (B–D) Southern Ocean corals grouped into the Tasmanian SAZ (B) and the DP SAZ (C) and DP AZ (D). In B–D, squares indicate corals with U–Th ages, whereas circles denote corals with radiocarbon ages corrected for the reservoir effect. The mean δ¹⁵N histories (bold lines) with error envelopes (2σ; 95% confidence interval) are simulated with Monte Carlo and Kalman filter methods (Materials and Methods).

Fig. S1.

Nitrogen isotopes of corals from Tasmania (A) and Drake Passage (B), with the colors indicating the collection depth of each coral. The error bars (±1σ) are calculated from two to three replicates from the same subsample of each coral septum.

Results and Discussion

Our coral sites cover a wide range of surface nitrate concentration ([NO₃⁻]), from >20 μM in the DP AZ, to ∼15 µM in the DP SAZ, to ∼3 μM in the Tasmanian SAZ (Fig. 1). Our late Holocene (0–5 kyr) average coral δ¹⁵N in the Tasmanian SAZ (10.34 ± 0.99‰, 1σ, n = 37) is higher than that of the DP SAZ (8.60 ± 0.95‰, 1σ, n = 26) and AZ (9.01 ± 1.07‰, 1σ, n = 16), with the DP AZ and SAZ coral δ¹⁵N indistinguishable from one another. These values are consistent with expectations based on the Rayleigh model, given the conditions in these regions (SI Text).

The δ¹⁵N range in corals of similar ages (Fig. 2) is far greater than the analytical uncertainty [1 SD = 0.2‰ (17, 18)]. Complementing this interspecimen variability, analyses from individual coral septa along the growth direction show δ¹⁵N variation of 1–2‰ on what would represent decadal timescales (Fig. S2). Such interspecimen and intraspecimen δ¹⁵N variation may record region-wide changes in the δ¹⁵N of N export, which would be of interest. Alternatively, they may reflect biological (e.g., ontogenetic) variability and/or may result from short timescale changes in deep-particle transport that alter the δ¹⁵N of the food supply to the corals. Measurements of recent corals at individual sites suggest δ¹⁵N variation of as much as 2‰ in coral carbonate-bound N (e.g., as a function of depth) (14), raising the possibility that this degree of variation is inherent to the proxy, although decadal variation in the regional δ¹⁵N of N export may also be contributing to the range in recent coral δ¹⁵N at those sites. Among the Southern Ocean corals studied here (Fig. S2) and others that have been investigated (14), there is no shared trend in δ¹⁵N through individual septa. Thus, the δ¹⁵N variability both within septa and across specimens is best explained by environmental variability, as opposed to an ontogenetic effect, but this does not allow us to distinguish between regional change in sinking N δ¹⁵N (the signal of interest) and local deep-particle dynamics. Further ground-truthing will be required to assess the significance of such short timescale variation in coral δ¹⁵N, which we avoid interpreting here. However, it is worth noting that there may be changes in interspecimen δ¹⁵N variability through the records, for example, greater variability between 15 and 13 kyr; if confirmed, this may yield important insights into past biogeochemical and environmental conditions. Although the inherent high temporal resolution of deep-sea corals raises their potential for studying high-frequency marine N cycle dynamics in the past, it currently complicates the identification of modest millennial-scale changes. We focus here on the large-scale, long-term trends over the past 40 kyr, which, due to the large number of corals analyzed, emerge despite short-term variability. Monte Carlo and Kalman filter methods are used that incorporate all characterized uncertainties (Fig. 2, Datasets S1 and S2, and Materials and Methods).

Fig. S2.

Tasmanian SAZ coral δ¹⁵N variation on single septa along the growth direction. Each symbol indicates one individual coral [filled symbols: late Holocene corals; open symbols: Bølling–Allerød (13–14 kyr) corals].

Antarctic Zone.

In the AZ of the Southern Ocean, diatom-bound δ¹⁵N records generally indicate more complete nitrate consumption during the last glacial period, but the records vary among cores (8), with an opposite change in the Pacific and Atlantic sectors of the AZ (19), and there are concerns of strong effects from diatom species differences (20). Average coral δ¹⁵N is 4–5‰ higher during the Last Glacial Maximum (LGM) than the late Holocene (Fig. 2), consistent with a recent diatom-bound δ¹⁵N record from the Pacific AZ near the Antarctic Polar Front (Figs. 1 and 3), where diatom species changes are minor (13). Using the Rayleigh model, we convert the δ¹⁵N records to the degree of summertime nitrate consumption (the percentage of the wintertime surface nitrate that is consumed over the summer productive period; SI Text). Both records indicate that the surface nitrate consumption was >90%. Assuming no change in the [NO₃⁻] or δ¹⁵N of underlying Circumpolar Deep Water (which is justified by the observation that the modern values in Circumpolar Deep Water are close to the whole-ocean averages), we calculate a summertime surface [NO₃⁻] of <5 μM in the LGM Antarctic (SI Text and Fig. S3). The constriction of the DP causes the AZ in that region to integrate conditions across most of the latitude range of the AZ in other sectors. Thus, the coral δ¹⁵N data provide the strongest evidence to date that the ice age increase in nitrate consumption did not apply solely to sites near the Antarctic Polar Front but rather applied to the full latitude range of the AZ.

Fig. 3.

Comparison of coral δ¹⁵N records with planktonic microfossil δ¹⁵N records from sediment cores, with the δ¹⁵N anomaly calculated by subtracting the modern or core-top δ¹⁵N values from each record. (A) Comparison of the DP AZ coral mean δ¹⁵N record with the PS75/072-4 AZ diatom δ¹⁵N record (13) and (B) comparison of the SAZ coral mean δ¹⁵N records with the ODP Site 1090 SAZ foraminifera δ¹⁵N record (12).

Fig. S3.

Comparison of coral δ¹⁵N records with the sedimentary microfossil δ¹⁵N records as in Fig. 3, and the reconstructed histories of AZ and SAZ summertime surface nitrate concentration. (A) Comparison of the Drake Passage AZ coral δ¹⁵N mean record with the PS75/072-4 AZ diatom δ¹⁵N record (δ¹⁵N anomaly was calculated by subtracting the modern or core-top δ¹⁵N values from each record) (13) and the calculated AZ summertime surface nitrate concentration over the past 40 kyr based on the Rayleigh model (SI Text). (B) Comparison of the SAZ coral δ¹⁵N records with the ODP Site 1090 SAZ foraminifera δ¹⁵N record (12) and the calculated SAZ summertime surface nitrate concentration based on the Rayleigh model and the reconstructed concentration and δ¹⁵N of the nitrate supply to the SAZ (SI Text and Figs. S5 and S6).

Subantarctic Zone.

In the Tasmanian SAZ record (Fig. 2), the average coral δ¹⁵N is 4–5‰ higher during the LGM than the late Holocene, an observation consistent with a foraminifera-bound δ¹⁵N record in the Atlantic SAZ (Fig. 3) (12). The Tasmania SAZ coral δ¹⁵N record is further corroborated by the coral δ¹⁵N compilation from the DP SAZ. In the DP SAZ, no D. dianthus older than ∼17 kyr were found (21). Nevertheless, coral δ¹⁵N also shows a 3.5–4.0‰ decrease from ∼15 kyr to the late Holocene (Fig. 2). In sum, the SAZ deep-sea coral δ¹⁵N data greatly strengthen the N isotopic evidence for more complete nitrate consumption in the ice age SAZ. Combining these results with those from the AZ, both of the major zones of the Southern Ocean were characterized by more complete nitrate consumption during the last ice age.

Questions exist regarding particle dynamics in the Southern Ocean and specifically the role that circulation may play in the δ¹⁵N distribution of deep suspended particles, and this is a concern for paleoceanographic applications of Southern Ocean deep-sea corals. It would seem plausible that deep-sea particle could be exchanged between the AZ and SAZ, leading to an erroneous interpretation from deep-sea corals regarding nutrient conditions in their region of growth; for example, the SAZ coral δ¹⁵N might be argued to respond to AZ nutrient conditions. A range of observations and considerations argues against the dominance of such an effect for the last 40 kyr. First, the ∼2‰ higher δ¹⁵N of modern corals in the Tasmanian SAZ relative to the DP is consistent with local sinking particles dominating the food source to corals in each region (Fig. S4). Given the evidence for low productivity in the AZ during ice ages (9), contamination of the SAZ by AZ particles would be even less likely during the LGM, and the circulation of the Southern Ocean is inconsistent with SAZ particles substantially influencing the AZ. Second, the Tasmanian SAZ and DP AZ coral δ¹⁵N records have coherent differences (Fig. 2B vs. Fig. 2D) that argue against the same history of change being recorded by both regions. Third, the Tasmanian corals are from the northern margin of the SAZ, quite distant from the AZ (Fig. 1), such that most deep particles from the AZ would be decomposed before reaching this far north. Fourth, the deep-sea coral δ¹⁵N records are remarkably consistent with both AZ diatom δ¹⁵N and SAZ foraminifera δ¹⁵N (Fig. 3), and these microfossil-forming plankton are not sensitive to deep suspended particle δ¹⁵N. Southern Ocean N cycling and the controls on the δ¹⁵N of deep suspended particles and deep-sea corals have not been adequately investigated to preclude as-yet-unrecognized complications. Nevertheless, we believe that our interpretation of more complete nitrate consumption across the Southern Ocean during the last ice age is robust.

Fig. S4.

Box plots of coral δ¹⁵N [(A) Tasmania; (B) Drake Passage SAZ (DP-SAZ); (C) Drake Passage AZ (DP-AZ)] for each 5-kyr bin (except for 20–40 kyr) and P values of each two adjacent groups based on Mann–Whitney U test (Wilcoxon rank-sum test).

The SAZ sits downstream of the AZ in the large-scale overturning circulation and today receives a significant fraction of its nitrate from the AZ mixed layer. Because the surface [NO₃⁻] during the LGM was much lower than today in the AZ, the contribution of the AZ surface to nitrate in the SAZ thermocline and surface mixed layer was lower during the LGM. We simulate this effect with a mixing model (Fig. S5 and SI Text), using the reconstructed changes in AZ [NO₃⁻] over the last 40 kyr (Fig. S3). The mixing model also indicates that the δ¹⁵N of the gross nitrate supply to the SAZ would have varied in response to changes in AZ nitrate concentration and δ¹⁵N (Fig. S5). The δ¹⁵N of the nitrate supply to the SAZ is calculated to have been similar during the LGM and in the late Holocene, but with a deglacial maximum (∼1.5‰ higher than the LGM and late Holocene values) (Fig. S6). The estimated rate and δ¹⁵N of nitrate supply to the SAZ are combined with the coral δ¹⁵N reconstruction from the SAZ to yield a preliminary calculation of the history of [NO₃⁻] in the SAZ summer mixed layer (Fig. S3B). As with the AZ, during the LGM, the SAZ surface [NO₃⁻] was <5 μM (Fig. S3B). We caution that our quantitative reconstruction of ice age surface [NO₃⁻] has uncertainties that are poorly characterized. In particular, the SAZ reconstruction relies on changes in nitrate supply that derive from the mixing model, in which the water exchange terms are poorly known today and could also have changed as a function of climate. Numerical modeling frameworks calibrated with modern data have great potential to improve the reconstructions.

Fig. S5.

A schematic showing the impact of AZ processes on the SAZ nitrate concentration and δ¹⁵N and the formulation of this impact with our end-member mixing model. (A) In the modern ocean, the surface nitrate concentration is high in the AZ, and the northward Ekman transport transfers a large amount of nitrate to the SAZ, contributing to 60–70% of the gross nitrate supply to the SAZ. (B) During the LGM, although the AZ surface nitrate has a high δ¹⁵N, the surface nitrate concentration is low in the AZ surface, so the northward Ekman transport thus transfers less nitrate to the SAZ. Under these conditions, the dominant nitrate source in the SAZ is underlying UCDW, which thus has the greatest control on the δ¹⁵N of the net nitrate supply to the SAZ surface.

Fig. S6.

Calculated SAZ thermocline nitrate concentration and δ¹⁵N from our end-member mixing model. These calculated changes in both the concentration and the δ¹⁵N of SAZ thermocline nitrate are included in the Rayleigh-based reconstruction of SAZ summertime surface nitrate concentration plotted in Fig. S3B.

To explain lower ice age pCO₂, a higher degree of nitrate consumption during ice ages has been proposed for both the AZ and the SAZ (4, 22–24). Our coral δ¹⁵N data provide evidence in support of these hypotheses. Remarkably, although the degree of nitrate consumption was higher in both of these regions during the last ice age, the AZ is reconstructed to have hosted much lower biological productivity than during the Holocene, whereas the SAZ hosted higher productivity (Fig. 4) (5, 9). This has led to the interpretation of a circulation-driven ice age reduction in nitrate supply as the ultimate driver of the AZ nitrate consumption change (13), whereas ice age iron fertilization best explains the nitrate consumption increase in the SAZ (Fig. 4D) (12). Given these different mechanisms for changes in surface nitrate, one would not expect identical histories for nitrate consumption in the two regions. Indeed, there are distinctions in the coral δ¹⁵N records from the AZ and SAZ both within the last ice age and since the end of the ice age. Below, we focus on the latter.

Fig. 4.

Comparison of Southern Ocean biogeochemical changes and the history of atmospheric CO₂ concentration over the past 40 kyr. (A) Atmospheric CO₂ (35, 36); (B) average AZ and SAZ δ¹⁵N changes from coral, diatom, and foraminifera records in Fig. 3, with the shading showing ±1σ based on all records; (C) SAZ and AZ export productivity (12, 13); and (D) Southern Ocean dust [iron (Fe)] flux as recorded in the Atlantic SAZ (12).

Deglacial Changes Across the Southern Ocean.

Combined with the foraminifera and diatom δ¹⁵N records (12, 13), our coral δ¹⁵N data allow us to compare the deglacial changes in the Southern Ocean [NO₃⁻] with other relevant environmental changes (Fig. 4). In the AZ, [NO₃⁻] rose from the LGM to ∼12 kyr, suggesting that the AZ contributed to the deglacial rise in atmospheric pCO₂. Export production in the AZ rose during the same period, consistent with the previous suggestion of a deglacial rise in Southern Ocean overturning (10, 13). Interestingly, [NO₃⁻] in the AZ continued to increase throughout the Holocene. This feature suggests a rise in the Southern Ocean overturning in the Holocene that may have contributed to the rise in atmospheric pCO₂ since 8 kyr (25).

In the SAZ, both the coral and foraminifera δ¹⁵N records suggest that there were two episodes of deglacial rise in surface [NO₃⁻] (Figs. 3 and 4). Decreasing iron flux can explain the early to middeglacial increase in [NO₃⁻], consistent with ice age iron fertilization in the SAZ (12, 23). However, the second episode of increase in [NO₃⁻] during the later deglaciation and early Holocene cannot be explained solely by a change in the iron flux, which was low and relatively stable during this period (Fig. 4D). In the iron-limited Southern Ocean, nitrate consumption is widely thought to be modulated by the ratio of rates of iron and nitrate supply (26). Without significant change in the atmospheric iron flux during the late deglaciation and early Holocene, the second episode of deglacial rise in the SAZ [NO₃⁻] is consistent with an increasing supply of nitrate from the AZ into the SAZ, an expected consequence of the reconstructed rise in AZ surface [NO₃⁻]. A complication for the deglaciation is that frontal migration likely also changed the [NO₃⁻] field, shifting lower [NO₃⁻] poleward. This may have countered the deglacial rise in Southern Ocean surface [NO₃⁻], contributing to its apparently gradual nature.

Conclusion

From this first application of deep-sea coral δ¹⁵N, we reconstruct an ice age Southern Ocean without the high [NO₃⁻] or the strong north–south [NO₃⁻] gradient that characterizes the region today. The picture that arises is of an ice age Southern Ocean state radically different from the modern, in which the two major zones (the Antarctic and the Subantarctic) receive most of their nitrate from below and consume almost all of it in the summertime surface mixed layer. This argues strongly for a central role of the biogeochemistry of the Southern Ocean in lowering ice age pCO₂.

At the same time, the data appear to indicate that the Southern Ocean [NO₃⁻] rise of the last deglaciation was incomplete, with additional change through the Holocene (Fig. 4B). Given model calculations of the pCO₂ decline achieved by different Southern Ocean changes (22), the incomplete nature of Southern Ocean biogeochemical change across the early deglaciation suggests that it was not the sole driver of the early deglacial pCO₂ rise. The warming-driven decline in the solubility of CO₂ in the deep ocean, which appears to have occurred early in the deglaciation (27), may thus have been critical in the first deglacial increase in atmospheric pCO₂. There is as-yet no evidence for such a lag of Southern Ocean biogeochemical change relative to Southern Ocean climate for prior glacial terminations (12, 13), but addressing this question requires better age constraints, such as might arise from U–Th dating of deep-sea corals from those terminations.

Materials and Methods

Deep-Sea Coral Collection.

Deep-sea fossil corals in the DP were collected by dredge or trawl (21), whereas a deep submergence vehicle was used to collect coral samples from the Tasmanian seamounts (28). Detailed information on all coral samples used in this study is given in refs. 15 and 16.

Radiocarbon Dating.

From each individual coral, a small (∼40 mg) piece was cut and physically abraded using a Dremel tool to remove the Fe–Mn crust. Samples were then cleaned and radiocarbon dated with methods detailed in refs. 29 and 30. The calendar age was calculated from the measured radiocarbon content and an estimated reservoir age (15, 16, 21, 28).

Uranium–Thorium Dating.

A subset of the corals used in this study (∼20%; shown as squares in Fig. 2) was also subjected to the U–Th dating method (15, 16). Briefly, a piece (0.3–1 g) of coral was cut and physically cleaned using a Dremel tool and then chemically cleaned following ref. 31. After cleaning, the coral was dissolved in nitric acid to dissolve the sample and a mixed ²³⁶U–²²⁹Th spike added. The U and Th were separated, purified, and measured separately by multicollector inductively coupled plasma mass spectrometer. All U–Th ages used in this study have been published in refs. 15 and 16.

Age Model.

The final age model for the three coral δ¹⁵N records is composed of two parts (Fig. 2). Whenever the U–Th ages are available, they are used in the final age model as the calendar age. When U–Th ages are not available, the calendar age is calculated from the radiocarbon age after correction for the water mass radiocarbon content (i.e., reservoir effect), based on the reservoir ages of coral samples for which both U–Th ages and radiocarbon ages have been measured (15, 16).

Nitrogen Isotope Analysis.

The protocol for nitrogen isotope analysis was detailed in refs. 14, 32, and 33. Briefly, 5–10 mg of mechanically cleaned coral septum is ground into coarse powder (with a grain size of a few hundred micrometers) and sonicated for 5 min in 2% (wt/vol) sodium polyphosphate in a 15-mL polypropylene centrifuge tube to remove any detrital material attached to the sample. The sample is rinsed (by filling, centrifugation, and decanting) three times with deionized water (DIW) and reductively cleaned using sodium bicarbonate-buffered dithionite-citrate reagent to remove any metal coatings. After three to four rinses with DIW, the sample is cleaned for 24 h using 10–15% sodium hypochlorite to remove external organic N contamination and again rinsed three to four times with DIW. After cleaning, the sample is dried in an oven at 60 °C and dissolved in 4 M hydrochloric acid. The released organic matter is oxidized to nitrate using a basic potassium persulfate solution. The resulting dissolved nitrate is converted bacterially into nitrous oxide, which is measured for its δ¹⁵N by automated extraction and gas chromatography–isotope ratio mass spectrometry (34). Amino acid standards with known δ¹⁵N (USGS 40 and 41) are included in each batch of samples to correct for the blank associated with persulfate reagent, which is less than 2% of the total N content in an oxidized sample. Each ground sample was processed in duplicate through the entire cleaning and analysis protocol. An in-house coral standard was used in each batch of analysis as quality control and yields a long-term precision better than 0.2‰ (1σ) (32).

To evaluate the δ¹⁵N variation in single septa, multiple samples were cut out along the growth direction of each single septum and analyzed for δ¹⁵N using the above method. The results for these samples are shown in Fig. S2 and discussed in Results and Discussion.

Mann–Whitney U Test.

To evaluate the statistical significance of the coral δ¹⁵N datasets, Mann–Whitney U tests were performed on the δ¹⁵N time series for each 5-kyr bin (except for 20–40 kyr; Fig. S4).

N Isotopes vs. N Content in Corals.

To evaluate the effect of N content in driving the observed trend in our coral δ¹⁵N, we performed a correlation analysis and showed that the coral δ¹⁵N variation cannot be explained by the changes in N content in any of the three records (Fig. S7).

Fig. S7.

Comparison of coral δ¹⁵N and N content, with samples colored based on age. If there is any tendency in N content as a function of age, it is of increasing N content with age, inconsistent with N loss with time. There is a lack a lack of correlation between N content and coral δ¹⁵N (r² values shown in Upper Left).

Monte Carlo and Kalman Filter Simulation.

Monte Carlo simulation and Kalman smoother were combined to obtain a best estimate of the average δ¹⁵N time series on a regular time grid and its corresponding confidence interval. First, M synthetic time series of δ¹⁵N were generated by adding errors to both age and δ¹⁵N measurements. The errors are normally distributed random numbers with zero mean and specified SDs (σ). The SDs for the measured age were assigned to be 500 y (typically larger than the actual measured age uncertainty), and the SDs for the measured δ¹⁵N are from the actual measurement errors (including both analytical error and variation in single coral septa). Then, for each one of the M synthetic time series of δ¹⁵N, a Kalman smoother was applied to obtain the best estimate for δ¹⁵N and its SD on a regular time grid. For each of the best estimates, N time series were generated to represent the probability distribution of this best estimate (by adding normally distributed random numbers with zero mean and SDs to the time series). Thus, outcome was M × N δ¹⁵N series on a regular grid. At each time, the best estimate for δ¹⁵N was obtained from the average of the M × N δ¹⁵N series, and the confidence interval was obtained from the σ of the M × N δ¹⁵N series [plotted as 95% confidence interval (2σ) in Fig. 2].

Calculations of the Summer Surface Nitrate Concentration in the Southern Ocean

Rayleigh Model.

When phytoplankton consume nitrate, they preferentially assimilate ¹⁴N relative to ¹⁵N, leaving the residual nitrate pool enriched in ¹⁵N and yielding a relationship between the δ¹⁵N of the accumulated organic N (integrated product) and the degree of nitrate consumption/nitrate utilization (37). Thus, in ocean regions where surface nitrate is not completely consumed and there is a strong temporal separation of nitrate supply and nitrate consumption, such as the Southern Ocean (38, 39), the δ¹⁵N of the accumulated organic N is an approximate proxy for the degree of nitrate consumption. Over the course of a year, the organic N export/sinking flux should equal the consumed nitrate, so that the δ¹⁵N of the sinking flux can be described by the Rayleigh model’s approximate integrated product equation:

$${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{integrated}}={\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{initial}}+\mathit{ϵ}\times \frac{f\times \ln f}{1-f},$$ [S1]

where ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{initial}}$ is the nitrate δ¹⁵N before nitrate consumption; $\mathit{ϵ}$ is the isotope effect of nitrate consumption (positive value means the product is depleted in ¹⁵N relative to the substrate); and $f$ is the fraction of remaining nitrate. The residual nitrate δ¹⁵N is approximated by the following:

$${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{N}{\mathrm{O}}_3^{-}}={\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{initial}}-\mathit{ϵ}\times \ln f.$$ [S2]

Deep-sea corals feed on organic matter that derives from this sinking flux, and the δ¹⁵N of deep-sea coral-bound organic N has been shown to be a good proxy for the δ¹⁵N of the sinking flux in the modern ocean (14).

Below, we use the Rayleigh model to calculate the surface nitrate concentration in the Southern Ocean over the past 40 kyr, based on our three coral δ¹⁵N datasets and the published foraminifera and diatom δ¹⁵N datasets in Fig. 3 (12, 13). Because diatoms, foraminifera, and corals are three distinct groups of organisms at different trophic levels, we normalize each δ¹⁵N dataset to their modern values by subtracting the modern δ¹⁵N values from each record, yielding the δ¹⁵N anomaly (Fig. 3). We then apply the Rayleigh model to each dataset and compute the surface nitrate concentration for each of the five datasets, taking into account changes in the initial nitrate concentration and δ¹⁵N as described below.

AZ Summertime Surface Nitrate Concentration Since 40 kyr Ago.

In the AZ, both the coral and diatom records show on average 4–5‰ higher δ¹⁵N during the LGM than the late Holocene (Fig. 3A). Applying the Rayleigh model to the AZ surface ocean:

$${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{AZ-integrated}}={\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{UCDW}}+{\mathit{ϵ}}_{\mathrm{AZ}}\times \frac{f_{\mathrm{AZ}}\times \ln f_{\mathrm{AZ}}}{1-f_{\mathrm{AZ}}},$$ [S3]

$${\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{AZ}}=\hspace{0.17em}{\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{UCDW}}\hspace{0.17em}\times \hspace{0.17em}{f}_{\mathrm{AZ}}\hspace{0.17em},$$ [S4]

$${\mathrm{\delta }}^{15}{\mathrm{N}}_{\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]\mathrm{AZ}}={\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{UCDW}}-{\mathit{ϵ}}_{\mathrm{AZ}}\times \ln f_{\mathrm{AZ}}\hspace{0.17em},$$ [S5]

where Upper Circumpolar Deep Water (UCDW) in the AZ is used as the starting nitrate pool [${\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{UCDW}}$ = 33 μM; ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{UCDW}}$ = 5‰ (40)]; ${\mathit{ϵ}}_{\mathrm{AZ}}$ is the isotope effect of nitrate consumption in the AZ surface ocean; $f_{\mathrm{AZ}}$ is the fraction of remaining nitrate (with a modern value of 0.7); ${\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{AZ}}$ is the residual nitrate concentration in the AZ surface; and ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]\mathrm{AZ}}$ is the residual nitrate δ¹⁵N in the AZ surface.

As indicated in Eq. S3, ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{AZ-integrated}}$ is a function of three variables: ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{UCDW}}$, ${\mathit{ϵ}}_{\mathrm{AZ}}$, and $f_{\mathrm{AZ}}$. It is thus important to consider the changes in the other two variables before we attribute the 4–5‰ higher LGM ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{AZ-integrated}}$ to a change in the surface nitrate concentration.

For the AZ, the dominant source of nitrate to the winter mixed layer (which sets the initial nitrate concentration and δ¹⁵N of the summer mixed layer) is mixing with and upwelling of UCDW. Thus, to change ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{initial}}$ in the AZ, one has to change the UCDW nitrate δ¹⁵N. One possible way to do this is to change the mean ocean nitrate δ¹⁵N. However, although there is a paucity of direct evidence as to mean ocean nitrate δ¹⁵N over glacial–interglacial cycles, the existing data suggest that mean ocean nitrate δ¹⁵N changed very little over the past 40 kyr (41, 42). Given the evidence for N fixation feedbacks (11, 33, 43), the ocean nitrate reservoir has likely been tied to that of phosphate (44), and UCDW nitrate concentration is today similar to the mean deep-ocean value. Thus, major changes in UCDW nitrate concentration are also unlikely. In all of the calculations below, we assume that the nitrate concentration and δ¹⁵N in UCDW have remained constant over the past 40 kyr.

A change in ${\mathit{ϵ}}_{\mathrm{AZ}}$ during the LGM is also possible, given the previous finding that ${\mathit{ϵ}}_{\mathrm{AZ}}$ varies as a function of mixed layer depth in the Southern Ocean (45). We can explore the possible contribution of changing ${\mathit{ϵ}}_{\mathrm{AZ}}$ to the 4–5‰ higher LGM ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{AZ-integrated}}$. For a given value of ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{UCDW}}$, a contour plot of ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{AZ-integrated}}$ as a function of ${\mathit{ϵ}}_{\mathrm{AZ}}$ and $f_{\mathrm{AZ}}$ can be generated using Eq. S3 (Fig. S8). It shows that, within the range of observed isotope effect of nitrate assimilation in the modern ocean (45), the ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{AZ-integrated}}$ is more sensitive to changes in $f_{\mathrm{AZ}}$ than in ${\mathit{ϵ}}_{\mathrm{AZ}}$, especially when the surface nitrate consumption is high. The modern AZ has an average $f_{\mathrm{AZ}}$ value of ∼0.7 and a ${\mathit{ϵ}}_{\mathrm{AZ}}$ value of ∼6‰ (45). Because ${\mathit{ϵ}}_{\mathrm{AZ}}$ during the spring-to-fall nitrate drawdown is unlikely to be lower than 4‰ (46), the maximum increase in LGM ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{AZ-integrated}}$ caused by a lower ${\mathit{ϵ}}_{\mathrm{AZ}}$ is less than 2‰ and most likely less than 1‰. When converting the δ¹⁵N anomaly records into the surface nitrate concentration records, we use a ${\mathit{ϵ}}_{\mathrm{AZ}}$ value of 6‰ and assume that ${\mathit{ϵ}}_{\mathrm{AZ}}$ in the AZ has not changed over the past 40 kyr.

Fig. S8.

Contours of the ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{AZ-integrated}}$ as a function of $\mathit{\epsilon }$ and $(1-\mathit{f})$, based on the Rayleigh model (with an initial nitrate δ¹⁵N of 5‰ assumed for the calculations shown in this figure). The downward convergence of isolines shows that ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{AZ-integrated}}$ is less sensitive to the isotope effect of nitrate assimilation at higher surface nitrate consumption, such that it imposes only a modest to minor uncertainty at nitrate consumption higher than 50%.

SAZ Summertime Surface Nitrate Concentration Since 40 kyr.

In the SAZ, the coral and foraminifera records also show 4–5‰ higher ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{integrated}}$ during the LGM than the late Holocene (Fig. 3B). As for the AZ above, we consider here the possibility of changes in the concentration of the nitrate supply, ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{initial}}$, and nitrate assimilation isotope effect, and their role in the 4–5‰ higher LGM ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{integrated}}$ in the SAZ.

In contrast to the AZ, the modern SAZ thermocline water has two primary sources of nitrate and at least three sources of water. First, Ekman transport carries AZ surface water into the SAZ, and this water contains the nitrate remaining from nitrate assimilation in the AZ (18). This nitrate enters the SAZ at the surface, but it is mixed throughout the deep wintertime SAZ mixed layer, including the depths of Subantarctic Mode Water. Second, the SAZ thermocline exchanges waters with the low-latitude upper ocean. This water has a very low nitrate concentration and effectively dilutes the nitrate in the SAZ thermocline (47). Third, diapycnal mixing of the SAZ thermocline with underlying Antarctic Intermediate Water (AAIW) and UCDW incorporates nitrate with a high concentration (33 μM) and relatively low δ¹⁵N (∼5.5‰ in the SAZ). It has been estimated that the Ekman transport contributes to 60–70% of the SAZ thermocline nitrate in the modern ocean, with diapycnal mixing making up the remaining 30–40% (48).

We use a three–end-member mixing model to simulate the mean concentration and δ¹⁵N of the gross nitrate supply to the SAZ thermocline/wintertime mixed layer. Because the low-latitude water contains no nitrate and because we have no reason to expect its contribution of water to vary relative the contribution from UCDW, we combine the low-latitude water end-member with the UCDW end-member. In this way, the three–end-member mixing model is simplified to a two–end-member mixing model. Thus, we use the following proportions of water from the three sources described above: 60% from the Ekman transport and 40% from UCDW (Fig. S5).

Eqs. S6 and S7 describe the mixing results of the SAZ thermocline nitrate:

$${\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{SAZ-thermocline}}={\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{UCDW}}\times f_{\mathrm{AZ}}\times b+{\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{mixedUCDW}}\times \left(1-b\right),$$ [S6]

$${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{SAZ-thermocline}}={\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{AZ}\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}\times \frac{{\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{UCDW}}\times f_{\mathrm{AZ}}\times b}{{\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{SAZ-thermocline}}}+{\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{mixedUCDW}}\times \frac{{\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{mixedUCDW}}\times (1-b)}{{\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{SAZ-thermocline}}},$$ [S7]

where ${\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{SAZ-thermocline}}$ is the nitrate concentration in the SAZ thermocline; $b$ is the SAZ water fraction from the AZ [set to a constant value of 0.6 (48)]; ${\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{mixedUCDW}}$ is the UCDW nitrate concentration after mixing with low-latitude water [set to a constant value of 18 μM (17, 18)]; ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{mixedUCDW}}$ is the δ¹⁵N of the UCDW nitrate input to the SAZ [set to a constant value of 5.5‰ (17, 18)].

This mixing yields a modern SAZ thermocline/winter mixed layer nitrate concentration of ∼20 μM, which is assumed to be drawn down to ∼10 μM by nitrate assimilation in the SAZ (18). Holding these water mixing proportions constant, we estimate the changing SAZ thermocline nitrate concentration from 40 kyr ago to the present (Fig. S6). One clear oversimplification is that summertime AZ conditions are used to calculate the year-round nitrate concentration coming from the AZ. However, during the LGM, the winter-to-summer nitrate concentration decline in the AZ was probably very weak (Fig. S5) such that our approach is a reasonable simplification so far as the effect of the AZ on the SAZ nitrate concentration is concerned.

The reconstructed nitrate concentration for the gross nitrate supply to the SAZ provides the initial nitrate concentration for the Rayleigh model calculation (Eqs. S8 and S9):

$${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{SAZ-integrated}}={\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{SAZ-thermocline}}+{\mathit{ϵ}}_{\mathrm{SAZ}}\times \frac{f_{\mathrm{SAZ}}\times \ln f_{\mathrm{SAZ}}}{1-f_{\mathrm{SAZ}}},$$ [S8]

$${\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{SAZ}=\hspace{0.17em}{\left[\mathrm{N}{\mathrm{O}}_3^{-}\right]}_{\mathrm{SAZ-thermocline}}\hspace{0.17em}\times \hspace{0.17em}{f}_{\mathrm{SAZ}}\hspace{0.17em},$$ [S9]

where ${\mathit{ϵ}}_{\mathrm{SAZ}}$ is the isotope effect of nitrate assimilation in the SAZ [set to a constant value of 8.5‰ (18)]; and $f_{\mathrm{SAZ}}$ is the fraction of residual nitrate in the SAZ surface (with a modern value of 0.5).

The same mixing model also provides a reconstruction of the δ¹⁵N of the gross nitrate supply to the SAZ (Eq. S7 and Fig. S6). Fig. S3B uses this as well as the calculated nitrate concentration described above as the initial nitrate δ¹⁵N and concentration for the Rayleigh model calculation. The δ¹⁵N of the gross nitrate supply to the SAZ from all sources tends to decrease under the LGM case of nearly complete nitrate consumption in the AZ (Figs. S5 and S6). Relative to UCDW, the AZ surface elevates the δ¹⁵N of the nitrate supply to the SAZ. When AZ nitrate concentration is very low, it can no longer play this role, and the δ¹⁵N of the nitrate supply to the SAZ collapses on the δ¹⁵N of UCDW nitrate.

The calculation of the δ¹⁵N of the nitrate supply to the SAZ is more uncertain than that of the concentration of the nitrate supply. A high degree of nitrate consumption in the AZ surface during the LGM would by itself raise the δ¹⁵N of the nitrate to be transported northward into the SAZ. However, the resulting low concentration of nitrate in the summer AZ means that its wintertime mixing with underlying water could erase much of this δ¹⁵N elevation. Accounting for such dilution effects is inherently uncertain. Nevertheless, in the calculations, the AZ is not a major nitrate source to the SAZ during the LGM, and so the δ¹⁵N of this minor nitrate source has little effect on the δ¹⁵N of the nitrate supply to the SAZ at this time.

Similar to the AZ, a change in ${\mathit{ϵ}}_{\mathrm{SAZ}}$ would lead to changes in ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{SAZ-integrated}}$. However, in comparison with the AZ, a given change in ${\mathit{ϵ}}_{\mathrm{SAZ}}$ will contribute less to the higher LGM ${\mathrm{\delta }}^{15}{\mathrm{N}}_{\mathrm{SAZ-integrated}}$ in the SAZ because the average surface nitrate consumption is relatively high (∼50%) even in the modern SAZ (Fig. S8).