Enhanced Terrestrial Carbon Export From East Antarctica During the Early Eocene

Terrestrial organic carbon (TerrOC) acts as an important CO2 sink when transported via rivers to the ocean and sequestered in coastal marine sediments. This mechanism might help to modulate atmospheric CO2 levels over short‐ and long‐ timescales (103–106 years), but its importance during past warm climates remains unknown. Here we use terrestrial biomarkers preserved in coastal marine sediment samples from Wilkes Land, East Antarctica (∼67°S) to quantify TerrOC burial during the early Eocene (∼54.4–51.5 Ma). Terrestrial biomarker distributions indicate the delivery of plant‐, soil‐, and peat‐derived organic carbon (OC) into the marine realm. Mass accumulation rates of plant‐ (long‐chain n‐alkane) and soil‐derived (hopane) biomarkers dramatically increase between the earliest Eocene (∼54 Ma) and the early Eocene Climatic Optimum (EECO; ∼53 Ma). This coincides with increased OC mass accumulation rates and indicates enhanced TerrOC burial during the EECO. Leaf wax δ2H values indicate that the EECO was characterized by wetter conditions relative to the earliest Eocene, suggesting that hydroclimate exerts a first‐order control on TerrOC export. Our results indicate that TerrOC burial in coastal marine sediments could have acted as an important negative feedback mechanism during the early Eocene, but also during other warm climate intervals.


of 14
The erosion and subsequent burial of TerrOC in marine sediments is determined by a variety of climatic and tectonic controls (Eglinton et al., 2021). Steep river catchments are typically characterized by enhanced TerrOC export (e.g., Hilton, 2017;Kao et al., 2014). However, climate also exerts a key control on TerrOC export Hilton, 2017) and there is a close link between mean annual run-off and TerrOC export in modern river systems (Hilton, 2017;Wang et al., 2020). Although TerrOC export and burial may help to modulate atmospheric CO 2 over geological timescales, the importance of this process in the past is unknown and the majority of work has been on modern erosive systems (e.g., Galy et al., 2015;Hilton, 2017;Wang et al., 2020).
The early Eocene Climatic Optimum (EECO; 53.3-49.1 Ma) is characterized by high temperatures (∼10°-16°C warmer than pre-industrial; Inglis et al., 2020), an intensified hydrological cycle (Carmichael et al., 2016), and lacked continental-scale ice sheets (Francis & Poole, 2002). As such, it can serve as a natural laboratory to understand how TerrOC export and burial operates when it is significantly warmer and wetter than present. Here we use samples recovered from the East Antarctic margin (Integrated Ocean Drilling Program (IODP) Leg 318; IODP Site U1356; ∼67°S paleolatitude; Escutia, Brinkhuis, Klaus, Bendle, et al., 2011) to determine the amount of TerrOC transported along the land-sea continuum during the earliest Eocene (54.3-51.5 Ma), including one of the three DeepMIP intervals (i.e., the Early Eocene Climatic Optimum; 53.3-49.1 Ma; Hollis et al., 2019). A combination of bulk and molecular proxies are used to fingerprint the type of OC in the East Antarctic margin. We then apply a mixing model approach to quantify TerrOC burial along the East Antarctic margin. We also develop new leaf wax δ 2 H records and compare these alongside isotope-enabled modeling simulations to determine the relationship between the hydrological cycle and TerrOC export during the early Eocene.

Site Description
Site U1356 (modern coordinates: 63°18.6138′ S, 135°59.9376′ E) is located ∼300 km off Wilkes Land, Antarctica, at the transition between the continental rise and the abyssal plain. Current water depth is 3,992 m (Escutia, Brinkhuis, & Klaus, 2011a; Figure 1). Samples were obtained from lithological Unit XI (∼948-1,000 mbsf) that consists of bioturbated hemipelagic early Eocene claystones. Occasional laminated siltstone and sandstone interbeds indicate sporadic gravity flows or bottom current activity reaching the site . Lithological Unit XI is dated as early Eocene based upon a combination of published dinoflagellate cyst (dinocyst) biostratigraphy and magnetostratigraphy Tauxe et al., 2012).

Bulk Isotope Geochemistry
Total organic carbon (TOC) measurements were carried out at MARUM, University of Bremen. Approximately 10 cm 3 of sediment (n = 360) was freeze-dried and ground to a fine powder using an agate mortar. Total carbon (TC), total organic carbon (TOC), and total sulphur (TS) were measured using a LECO CS-200 Carbon-Sulphur analyzer. Approximately 65 mg of the homogenized sample was weighed in a ceramic cup and burnt in an induction furnace. The resultant CO 2 and SO 2 were then measured with a nondispersive infrared detector to provide a measure of the sedimentary TC and TS content. To determine the TOC content, sediments (∼65 mg) were decalcified using 12.5% hydrochloric acid (HCl) to remove carbonate species, followed by a subsequent rinse of the residue with de-ionized water. Samples were subsequently analyzed as described above. Total inorganic carbon (TIC) was determined by subtracting TOC from TC. All data are reported in weight percent (wt. %) dry sample.
Bulk carbon isotope analyses of the OC fraction (δ 13 C org ) were carried out at MARUM, University of Bremen at 2 cm resolution throughout Cores 318-U1356A-103R to −106R (n = 368). The bulk samples were decalcified with 1 N HCl solution, followed by a subsequent rinse of the residue with de-ionized water. The sample was then combusted using the oven of an HERAEUS CHN-Analyzer. The δ 13 C value was subsequently measured with a Finnigan MAT Delta E Mass-spectrometer. Measurements were calibrated using a house standard (milled and decalcified organic-rich surface sediment from the German Wadden Sea), itself calibrated to international standards. Repeatability of the house standard is typically ±0.15‰.
For a subset of earliest Eocene samples (n = 11), we performed urea adduction on the N1 fractions to separate cyclic (i.e., non-adduct) and aliphatic (i.e., adduct) hydrocarbons. To achieve this, 200μl of hexane, acetone and urea (10% in MeOH) were added to the saturated hydrocarbon fraction. The sample was frozen for ca. 60 minutes until urea crystals formed. Solvent was then removed under a stream of N 2 and extracted (x5) with ca. 1ml of n-hexane (non-adduct fraction). The urea crystals were then dissolved in 500 μl of MeOH and 500 μl of water and the adduct fraction was extracted (x5) with ca. 1ml of n-hexane. The adduction procedure was repeated on the adduct fraction once more to ensure all non-adduct material was removed (Inglis et al., 2019).

GC and GC-MS Analysis
N1 fractions were redissolved in hexane and analyzed using a gas chromatograph (GC; Shimadzu 2010) with a flame ionization detector (FID) and hydrogen as carrier gas at constant pressure (190 kPa). Separation of the different compounds was achieved using the following column: length: 60 m, diameter: 0.25 mm, film thickness: 0.25 μm, coating: 100% dimethyl-polysiloxane. The gas chromatograph temperature program increased from 50° to 120°C at 30°C min −1 , then 120°-310°C at 5°C min −1 , with a final isothermal time of 20 min at 300°C. Compound identification was confirmed by GC-MS (Shimadzu QP2010-Plus Mass Spectrometer (MS) interfaced with a Shimadzu 2010 GC) using the same column described above. Compounds were identified using retention times and mass spectra. The abundance of n-alkanes and hopanes are normalised to TOC. An in-house external standard comprised of 20 n-alkanes (ranging from C 16 to C 37 ) was run every 10 samples to monitor analytical performance.

GC-C-IRMS Analysis
Compound specific hydrogen isotopic compositions were determined for a subset of samples from the earliest Eocene (n = 11) and EECO (n = 5). We analyzed long-chain (C 27 , C 29 ) n-alkanes via gas chromatography-combustion-isotope ratio mass spectrometry (GC-C-IRMS) using a ThermoFisher Trace GC Ultra coupled to a Ther-moFisher Scientific Delta V Isotope Ratio MS. The GC column used was a 30 m × 0.25 mm i.d. fused silica column with ZB1 stationary phase. The H 3 factor was measured daily allowing isotope values to be corrected for protonation reactions occurring within the ion source of the mass spectrometer (Sessions et al., 1999). The H 3 factor was typically below 5 ppm mV −1 and had a rate of change of less than 0.1 day −1 . The GC program was as follows: starting temperature 70°C, rising at 10°C/min to 300°C, at which point the oven temperature was held stable for 8 min, giving a total analysis-time of 32 min. Triplicate runs of each sample were performed. The average standard error of the mean (SEM) value for a triplicate measurement was typically 2-7‰. Each individual sample was co-injected with sacrificial compounds consisting of n-pentadecane and ethyl caprylate to condition the reactor. Measured isotope values were normalised by comparing the instrument's response to a suite of n-alkanes with a known isotopic composition (B5 mix; supplied by A. Schimmelmann, Indiana University, USA) injected before and after each triplicate of sample runs. The root mean square error determined by replicate measurements of the standard across the course of analyses was between 4 and 7‰. Normalised results are reported in standard per mil (‰) notation as δ 2 H values relative to Vienna Standard Mean Ocean Water (VSMOW) and calculated against calibrated H 2 gas, introduced directly into the ion source.

Lipid Biomarker Proxies
We use biomarkers ratios to assess the source and maturity of organic matter. The average chain length (ACL) expresses the average number of carbon atoms for the long-chain odd-carbon numbered n-alkanes (Pancost & Boot, 2004). The ACL is defined for n-alkanes using the following equation (Eglinton & Hamilton, 1967): The carbon preference index (CPI) reflects the dominance of odd-carbon-numbered relative to even-carbon-numbered n-alkane homologs (Bray & Evans, 1961). Modern sediments exhibit high CPI values (>3-30; Diefendorf & Freimuth, 2017). This value decreases over time due to diagenesis, approaching values of unity (=1) in mature rocks and oils. The CPI is defined using the following equation (Bray & Evans, 1961): The hopane ββ/(αβ+ββ) ratio is used to evaluate changes in sediment maturity. Modern sediments exhibit high hopane ββ/(αβ+ββ) ratios (up to 1) whereas thermally mature rocks and oils are characterized by lower hopane ββ/(αβ+ββ) ratios (Mackenzie et al., 1980), although there can be exceptions (e.g., acidic peats; Inglis et al., 2018). With increasing maturation, C 31 to C 35 hopanes also undergo isomerization at the C-22 position. An increase in 22S/(22R + 22S) values is characteristic of increasing thermal maturation (Mackenzie et al., 1980).

Mass Accumulation Rates
Biomarker mass accumulation rates (MARs; in ng/cm 2 /kyr −1 ) are calculated using linear sedimentation rate (LSR; cm/kyr), dry bulk density (ρ; g/cm 3 ; Dadey et al., 1992) and biomarker abundance (ng/g dry sediment): Linear sedimentation rates (LSRs) are calculated between the age tie-points. To avoid age inversions (i.e., within Core 105R-1) we constructed LSRs based on a selection of tie points rather than incorporate all bio-and magnetostratigraphic datums in the age model (see Supporting Information S1).

Continental Shelf Area Calculations
We calculate continental shelf area using the Eocene-Oligocene (ca. 34 Ma) paleotopography published elsewhere , Hochmuth, Paxman, et al., 2020Paxman et al., 2019). The same calculations are also conducted using Bedmap2 (Fretwell et al., 2013) for the present day. Continental shelf area for the Eocene is defined as being from the 0 m contour (e.g., shore of an ice free Antarctica) to the −1000 m bathymetric contour ( Figure S1 in Supporting Information S1). We calculate continental shelf area for the region directly adjacent to Wilkes basin and for the whole Antarctic continent in order to understand potential (Terr)OC sequestration.

Modeling Simulations
δ 2 H wax is usually employed as a proxy for the isotopic composition of precipitation (δ 2 H w ). However, this integrates a combination of climatic changes including temperature, rainfall amount and characteristics, and atmospheric vapor transport. Here we use modeled δ 2 H w obtained via the isotope-enabled Community Earth System Model version 1.2 (iCESM1.2; Zhu et al., 2019Zhu et al., , 2020 to compare with our proxy reconstructions and to aide climatic interpretation of δ 2 H w (Lee et al., 2007;Schmidt et al., 2007). The iCESM1.2 simulations were performed following protocols of the Deeptime Model Intercomparison Project (Lunt et al., 2017(Lunt et al., , 2021 with the Eocene paleogeography and vegetation (56.0-47.8 Ma; Herold et al., 2014) and atmospheric CO 2 levels of ×1, ×3, ×6, and ×9 the preindustrial value (284.7 ppmv). The different atmospheric CO 2 levels span the range of proxy-derived CO 2 estimates for the early Eocene (Anagnostou et al., 2020). We also carried out a single pre-industrial simulation to isolate the non-CO 2 component on δ 2 H w (i.e., changes due to paleogeography, vegetation, aerosols, and lack of continental ice sheets). All simulations were run at a horizontal resolution of 1.9° × 2.5° (latitude × longitude) for the atmosphere and land, and a nominal 1° for the ocean and sea ice. Of particular relevance to this study is the capability of iCESM1.2 to simulate the transport and transformation of oxygen and hydrogen isotope ratios (δ 18 O and δ 2 H) in the climate system (Brady et al., 2019). Seawater δ 18 O and δ 2 H in the simulations were initialized from constant values of −1.0‰ and −8.0‰, respectively, to account for the absence of ice sheets in a hothouse climate (Hollis et al., 2019). The simulations were run for more than 2000 years with the surface climate and water isotopes close to equilibration. The simulations capture the magnitude of Eocene global warmth, the reduction in the meridional gradient of sea-surface temperature, and the overall values and distribution of marine planktic foraminifera δ 18 O. Readers are referred to Zhu et al. (2019Zhu et al. ( , 2020 for details of the experimental setup, equilibration state, and assessment of the simulation results.

Bulk Geochemistry
Total organic carbon (TOC) content ranges between 0.1% and 1.5% (Figure 2). There is a gradual increase in TOC values between 54.4 and 53.5 Ma (Core 106R to 105R) followed by a gradual decrease thereafter (53.5-52.7 Ma, Core 104R to 101R). Bulk δ 13 C org values range between −25.7 and −28.1‰ (Figure 2). The early Eocene record from Site U1356 exhibits several δ 13 C org variations. These δ 13 C org cycles and excursions have maximum values of up to −26‰ (in one case −25‰). The average δ 13 C org values of −26 to −28 ‰ are typical for terrestrial organic matter (Hayes, 1993), although this value may be different in the Eocene . The cyclicity in the δ 13 C org record is relatively pronounced in the lower part of the studied section (Cores U1356-105R and −106R; Figure 2), where TOC values are high enough (0.5-1.2 wt. %) to provide robust data. In the upper part of the early Eocene section (e.g., Cores 103R and −104R), TOC contents are often lower than 0.5 wt. % and yield a lower fidelity δ 13 C org signal.

Biomarker Distributions and Mass Accumulation Rates
The N1 fraction contains a range of n-alkanes (C 17 to C 33 ) and is dominated by long-chain homologs (C 27 to C 33 ). The abundance of long-chain n-alkanes ranges between ∼1 and 250 ng/g (average: 92 ng/g; Figure 2d). Long-chain n-alkane (plant-derived OC; Figure 3d) mass accumulation rates (MARs; in ng/cm 2 /kyr −1 ) increase by two orders-of-magnitude between the earliest Eocene (∼54 Ma) and the onset of the EECO (∼53 Ma; Figure 3). CPI values (average: 2.8; Figure 4a) indicate that long-chain n-alkanes have retained a biological odd-over-even predominance. In combination with high ACL values (∼27-30; Supplementary Information), this confirms they are derived from terrestrial higher plants (Diefendorf & Freimuth, 2017).
The apolar fraction also contains a range of bacterial-derived hopanes (C 27 to C 32 ) and hopenes (C 27 to C 30 ). The abundance of hopanes ranges from ∼1 to 191 ng/g (average: 54 ng/g) and exhibits a significant linear correlation with long-chain n-alkane abundance (r 2 = 0.81, p < 0.001; Figure S2 in Supporting Information S1). Hopane (soil-and peat-derived OC; Fig Our analyses are combined with previously published Branched-versus-Isoprenoid Tetraether (BIT) values . BIT values range between 0.16 and 0.62 (average: 0.35) and exhibit higher values (>0.4-0.5) during the EECO (Core 105R to 102R; Supplementary Information) compared to the earliest Eocene (Core 106R).

Long-Chain n-alkane δ 2 H Values
Long-chain n-alkane δ 2 H (δ 2 H wax ) values were determined for a subset of samples (n = 15) deposited during the earliest Eocene ( (Figure 5a). Both the C 27 and C 29 n-alkanes co-vary and exhibit a positive linear relationship (r 2 = 0.78, p < 0.001), indicating that they are derived from a similar source (i.e., higher plants; Figure S3 in Supporting Information S1).

Enhanced TerrOC Burial on the East Antarctic Margin During the EECO
The export of terrestrial OC along the terrestrial-aquatic continuum can be evaluated by analyzing the distribution of lipid biomarkers within the Wilkes Land sediment samples. The hydrocarbon distribution is dominated by long-chain n-alkanes with an odd-over-even predominance. High CPI values (∼3; Figure 4a) together with the dominance of long-chain n-alkanes (relative to shorter-chain homologs) confirms that they are derived from the epicuticular wax of terrestrial higher plants (Eglinton & Hamilton, 1967). The hydrocarbon assemblage also  , (b) total organic carbon (TOC; g/cm 2 /kyr −1 ) mass accumulation rate (MAR), (c) hopanioid MAR (ng/cm 2 /kyr −1 ), (d) long-chain n-alkane MAR (ng/cm 2 /kyr −1 ). Note that (a-c) are on a logarithmic scale. Black segments indicate intervals of no core recovery. contains abundant bacterial-derived hopane biomarkers. Hopanes can be produced in terrestrial and marine environments by a range of bacteria (Talbot & Farrimond, 2007). However, the positive linear relationship between longchain n-alkanes (terrestrial higher plants) and hopanes (r 2 = 0.81; p < 0.001; Figure S2 in Supporting Information S1) indicates that here, both are sourced from the terrestrial biosphere.
Long-chain n-alkane (plant-derived OC; Figure 4d) and hopane (soil-derived OC; Figure 4c) mass accumulation rates (MARs; in ng/cm 2 /kyr −1 ) increase by two orders-of-magnitude between the earliest Eocene (∼54 Ma) and the onset of the EECO (∼53 Ma). This significant increase indicates enhanced burial of plant-and soil OC during the early Eocene, with highest accumulation rates during the EECO. Note that we performed the same calculations assuming a constant LSR (1.82 cm/year, i.e., the average LSR rate) and find similar MARs during the EECO (∼53 Ma; Figures S4-S5 in Supporting Information S1). Delivery of TerrOC into the marine realm is consistent with the high abundance of terrestrial palynomorphs (i.e., pollen and spores; (Contreras et al., 2013;Pross et al., 2012) and other terrestrial biomarkers (including high BIT indices; Bijl et al., 2013;Pross et al., 2012;Figure 3a) in the same sediments. These findings are analogous to those observed in modern (high-latitude) environments, where large river systems in northern Eurasian and North American (Guo et al., 2004;Hilton et al., 2015;Semiletov et al., 2011;Sparkes et al., 2015) are capable of transporting large quantities of TerrOC from soils into the marine realm.
Input of acidic peat into the marine realm is consistent with the occurrence of Stereisporites spores in the Site U1356 sediments (Pross et al., 2012; Figure  S6 in Supporting Information S1) that can be attributed to the Sphagnaceae family, which today only comprises the genus Sphagnum (commonly known as peat moss). Despite the low abundances registered, the occurrence of Stereisporites spores in the Site U1356 sediments appears noteworthy for various reasons. First, Stereisporites spores are relatively rare or absent in early Cenozoic sediments (see Inglis et al., 2015 and ref. therein). Second, spores  . Enhanced poleward moisture transport to the East Antarctic margin during the early Eocene). (a) δ 2 H wax estimates inferred from the C 27 n-alkane (orange) and C 29 n-alkane (blue). Error bars represent ±1σ x̅ (following Polissar & D'Andrea, 2014). (b) pollen-corrected proxy-derived δ 2 H w estimates inferred from the C 27 n-alkane (orange) and C 29 n-alkane (blue). Error bars indicate ±1σ propagated errors of the analytical error and the fractionation uncertainty (following Feakins, 2013). Here we use the standard deviation from modern plant type groups (i.e., C 3 angiosperms, C 3 gymnosperms, ferns; Sachse et al., 2012) to calculate the fractionation uncertainty (see Open Research). are generally strongly underrepresented in sporomorph assemblages from marine sediments due to the selective nature of sporomorph transport as a function of transport distance (Contreras et al., 2014;Moss et al., 2005). As such, even their occurrence in low numbers (e.g., ∼1% of total sporomorph assemblage) can have substantial ecological significance and points toward the existence of peatlands and/or forested mires within the Site U1356 catchment area. The input of acidic peat into the marine realm is also consistent with the results of biome-based modeling suggesting that ∼15%-20% of Antarctica was capable of supporting wetland and peat formation during the Eocene (DeConto et al., 2012).

Quantifying TerrOC Burial on the East Antarctic Margin During the EECO
Lipid biomarkers indicate that a considerable fraction of the OC pool on the early Eocene Wilkes Land continental margin is derived from the terrestrial biosphere. However, the relative proportion of terrestrial versus marine OC must be known in order to quantify TerrOC burial (Weijers et al., 2009). In modern settings, bulk OC δ 13 C values (δ 13 C org ) can discriminate between marine OC ( 13 C-enriched) and vascular plant and/or soil OC ( 13 C-depleted). However, marine OC can be relatively 13 C-depleted during the Eocene (up to 7‰ relative to modern;  due to enhanced fractionation under higher CO 2 (Freeman & Hayes, 1992).
Here we use BIT indices  to characterize the fraction of terrestrial versus marine OC and employ a two end-member mixing-model: where f Terrestrial is the terrestrial OC fraction and X Sample is the BIT index for the sediment sample (following Weijers et al., 2009). The marine-(X marine ) and terrestrial end-member (X terr ) for BIT is 0 and 0.91, respectively . The BIT index can reach a theoretical value of 1. However, BIT values in globally distributed soil samples average 0.91 (Weijers et al., 2006). Using our mixing-model approach, we estimate that the proportion of terrestrial OC in the Wilkes Land sediments ranges between 20% (minimum) and 65% (maximum). As the BIT index captures the proportion of soil OC (rather than soil and plant OC), this approach may underestimate the proportion of terrestrial OC in the Wilkes Land sediments. The BIT index has been shown to be controlled strongly by crenarchaeol-rather than brGDGT-concentrations, and may lead to a decoupling between BIT values and other terrestrial OC tracers (Smith et al., 2012). However, the fractional abundance of crenarchaeol is stable throughout the early Eocene , arguing that BIT is largely controlled by changes in brGDGT abundance. The BIT index can also be influenced by in-situ marine or fluvial brGDGT production. However, brGDGTs reflect typical soil-distributions, with no abnormal Isomerization Ratio or #Rings tetra values (Sinninghe Damsté, 2016). This suggests negligible marine in situ or river production) and is consistent with other early Eocene sites in the southwest Pacific Ocean (e.g., Bijl et al., 2021).
The proportion of TerrOC can then be multiplied alongside TOC MARs to provide a first-order approximation of TerrOC sequestration along the Wilkes Land margin. This first requires an estimate of the Wilkes Land continental shelf area (118,675 km 2 ). Here we assume that the early Eocene shelf was similar to that of the latest Eocene (Paxman et al., 2019) and was significantly smaller than the present day continental shelf area (170,731 km 2 ). We also assume that the sediment package thickness (and thus, %TerrOC) is uniform. Assuming that 20%-65% of the OC pool in the Wilkes Land sediments is terrestrial-derived (see above), this implies the burial of 20-72 PgC of TerrOC the depositional interval (54.3 and 51.5 Ma). If we assume similar climate, vegetation distributions and transport mechanisms across the entire Antarctic shelf area (3,947,850 km 2 ), the potential mass of TerrOC sequestered is much larger (∼410-1,300 PgC) and equates to a burial flux of 0.14-0.46 TgC/yr. These estimates are relatively modest compared to highly erosive mountainous river catchments (e.g., Taiwan; >8 TgC/yr; Kao et al., 2014) and large river systems (e.g., the Mackenzie River; 2.2 (+1.3/−0.9) TgC/yr; Hilton et al., 2015). To refine these estimates, additional studies at different localities are required. Additional seismic data may help to improve estimates of offshore sediment thickness, whilst the application of binary (or ternary) mixing models based on other geochemical proxies (e.g., C/N ratios, n-alkane/alkenone ratios) will help to refine our estimates of TerrOC burial (see Weijers et al., 2009). Until then, our estimates should be treated with caution.
Although these estimates incorporate various uncertainties, our work highlights the potential importance of continental margins as depocentres for TerrOC burial during the early Eocene. This is consistent with High Arctic continental shelf sediments deposited between the Palaeocene-Eocene Thermal Maximum (PETM; ∼56 Ma) and Eocene Thermal Maximum 2 (ETM2; ∼53 Ma), which are dominated by long-chain n-alkyl lipids and low hydrogen indices (Boucsein & Stein, 2009;Weller & Stein, 2008). Terrigenous input-which is strongly correlated with TerrOC export in modern settings (e.g., Galy et al., 2015)-also increases during the EECO, including in New Zealand (e.g., Mead Stream; Nicolo et al., 2007;Slotnick et al., 2012) and the North Iberian margin (Payros et al., 2015). Enhanced TerrOC burial may be further promoted by larger, flatter drowned shelf areas during the Eocene (Bowen, 2013;Sømme et al., 2009). Taken together, this suggests an important role for continental shelves in storing terr(OC) during the Eocene. This is especially important given the evidence for low (terr)OC burial in open ocean Eocene environments (Olivarez Lyle & Lyle, 2006).

Hydroclimate Regulates TerrOC Export During Early Eocene
Enhanced TerrOC burial in coastal marine sediments could be driven by expansion of the terrestrial biosphere during the EECO. However, TerrOC export is largely controlled by precipitation (Hilton, 2017) and the capacity of rivers to erode and transport TerrOC . Here we explore whether hydroclimate regulates OC export from the terrestrial biosphere. To assess this, we use the hydrogen isotopic composition (δ 2 H) of leaf wax (C 27 and C 29 n-alkane) biomarkers in a subset of samples deposited during the earliest Eocene (54.3-53.9 Ma) and EECO. C 27 and C 29 n-alkane δ 2 H values average −190 and −195‰, respectively, and are comparable to those reported from other early Eocene-aged sites from similar latitudes (Speelman et al., 2010). The isotopic composition of precipitation (δ 2 H w ) is estimated by assuming an apparent fractionation factor ( 2 Hɛ wax/w ) which integrates a range of variables (e.g., soil-and/or leaf water evapotranspiration, biosynthetic effects, vegetation type). Theoretical models have incorporated evaporative 2 H-enrichment of soil water into predictions of 2 ɛ wax/w (Konecky et al., 2019;Smith & Freeman, 2006). However, in settings with extensive evaporation, woody plants take up rainwater opportunistically via different rooting strategies (Fan et al., 2017), resulting in minimal xylem water 2 H-enrichment. Although leaf water 2 H-enrichment is more variable (Daniels et al., 2017;Feakins et al., 2016;Kahmen et al., 2013), it is significantly reduced in warm and wet climates (Feakins et al., 2016). As the East Antarctic margin was characterized by elevated temperatures (>16° ± 5°C) and high precipitation (>1,300-1,600 mm/year; Pross et al., 2012), the impact of leaf water 2 H-enrichment is likely minimal.
Changes in the plant community can also influence the apparent fractionation between δ 2 H wax and δ 2 H w values.
To account for the impact of vegetation isotope effects upon leaf wax δ 2 H values, we developed a pollen-corrected fractionation factor ( 2 Hɛ wax/w-corr ) using a three endmember mixing model (following Feakins, 2013). This includes C 3 gymnosperms, C 3 dicots and pteridophytes (i.e., ferns) and is defined as: where ɛ is based on the mean fractionation factor in modern plant types (Sachse et al., 2012) and f is based upon the percentage of pollen in adjacent samples (Contreras et al., 2013;Pross et al., 2012). Our pollen-corrected fractionation factor ranges between −107 and −113‰ and is higher than assumed in some Eocene studies (−130‰) (Handley et al., 2012;Pagani et al., 2006) but similar to an updated global data set of 2 Hɛ wax/w in modern systems (e.g., −121‰ for C 29 n-alkane; McFarlin et al., 2019). When our pollen-corrected fractionation factor is applied to our leaf wax data set, we obtain δ 2 H w-corr values which average −82‰ (C 27 n-alkane; Figure 5b) and −83‰ (C 29 n-alkane; Figure 5b). These values are 2 H-enriched relative to modern values (−99‰; based on iCESM1.2 pre-industrial x1 CO 2 simulation; see also below). There are several reasons for this. First, local air temperatures will yield more 2 H-enriched water vapor (a temperature effect) (Poulsen et al., 2007). Second, under warmer climates, decreased rainout efficiency at the low-latitudes (due to a reduction in vertical moisture transport) will result in more 2 H-enriched precipitation at the mid-to-high latitudes (e.g., Pagani et al., 2006). Our pollen-corrected δ 2 H w values exhibit considerable variability in the earliest Eocene interval (ca. 30-40‰; Figure 5). This may represent changes in the hydroclimate regime. However, it could also record changes in OC source region. Indeed, sporomorph distributions indicate that TerrOC is derived from both the lowlands and higher-altitude hinterlands (Pross et al., 2012). Although leaf waxes typically provide a spatially integrated perspective, the spatial and temporal integration may vary by compound type and/or change through time. Future work using other compound classes (e.g., long-chain n-alkanoic acids) may help to differentiate between lowland (proximal) and highland (distal) OC sources (see Feakins et al., 2018;Hemingway et al., 2016).
Our pollen-corrected δ 2 H w values indicate higher average δ 2 H w values (e.g., −66‰ for the C 29 n-alkane) during the EECO compared to the earliest Eocene (e.g., e.g., −90‰ for the C 29 n-alkane; Figure 5b). This is consistent with an increase in poleward moisture transport in response to higher CO 2 and higher temperatures. To confirm that the higher average δ 2 H w values are due to higher CO 2 and/or higher temperatures, we calculate the precipitation-amount weighted δ 2 H w in isotope-enabled CESM simulations with a range of atmospheric CO2 concentrations (×1, ×3, ×6, and ×9 the preindustrial level; see Section 2.7). The modeled δ 2 H w exhibits a monotonic increase with the CO 2 concentrations in the simulations with values of −106‰ in ×1 CO 2 , −96‰ in ×3 CO 2 , −91‰ in ×6 CO 2 and −81‰ in ×9 CO 2 (Figure 6b). This δ 2 H w sensitivity to CO 2 levels in isotope-enabled climate models has been attributed to warmer temperatures and/or the associated reduction in isotopic rainout at low-latitudes (Poulsen et al., 2007;Speelman et al., 2010). The qualitatively consistent results between model simulations and proxy data supports our interpretation that the observed increase in proxy-inferred δ 2 H w values between the earliest Eocene and EECO (∼20-40‰) is attributable to a CO 2 -induced warming. There is also close data-model agreement in the absolute δ 2 H w values ( Figure 6), supporting our data-based reconstructions.
The impact from non-CO 2 boundary conditions on δ 2 H w can also be evaluated by comparing iCESM1.2 Eocene ×1 simulations against the pre-industrial simulation. This shows that the combined non-CO 2 boundary conditions (i.e., paleogeography, vegetation, aerosols, and removal of continental ice sheets) decrease δ 2 H w by ∼7‰. This implies that non-CO 2 boundary conditions exert a small secondary control on δ 2 H w values at our site. The iCESM1.2 simulations (x3, x6 and x9 PI CO 2 simulations) also exhibit high MAP estimates (∼1310, 1520, and 1720 mm/year, respectively; Figures 7c and 7d) during the early Eocene. These are higher than previous simulations (Carmichael et al., 2016) Pross et al., 2012). Slightly lower values (800-1,350 mm/ year) have been reconstructed for Site U1356 based on climofunctions applied to detrital geochemistry (Passchier et al., 2013) and likely reflects the differences in the catchment sourcing. Figure 6. Reconstructed and simulated δ 2 H w on the East Antarctic margin during the early Eocene: (a) δ 2 H w-corr estimates inferred from the C 27 n-alkane (light gray) and C 29 n-alkane (dark gray). Dashed line represents median value. Box represents 1st and 3rd interquartile range. Whiskers represent 95% confidence interval. (b) model-derived δ 2 H w estimates inferred from iCESM1.2. Open triangle represents pre-industrial x1 CO 2 simulation. Closed circles represent Eocene simulations at four different CO 2 levels (x1, x3, x6, and x9 pre-industrial CO 2 ).

Figure 7.
Long-term hydroclimate variability along the East Antarctic margin during the early Eocene. (a) mean annual precipitation (MAP) estimates for the temperate rainforest biome based on bioclimatic analysis (Pross et al., 2012), (b) MAP estimates for the paratropical rainforest biome based on bioclimatic analysis (Pross et al., 2012), (c) and (d) MAP estimates obtained via iCESM1.2 for four different CO 2 levels (x1, x3, x6, and x9 pre-industrial CO 2 ). Whiskers represent precipitation estimates during summer (JJA) and winter (DJF) months. Black segments indicate intervals of no core recovery.
Taken together, our results reveal that EECO was characterized by enhanced poleward moisture transport relative to the earliest Eocene (∼54 Ma; Figures 5 and 6). The EECO also coincides with an interval of enhanced delivery of terrestrial OC into the marine realm (Figure 3), implying a causal link between the hydrological cycle and TerrOC burial at our site. Crucially, TerrOC burial can help to remove the atmospheric CO 2 and could serve as a negative feedback to decrease the surface temperature during intervals of global warmth. Compared to the silicate weathering feedback (which operates on the timescales of a million years), the organic carbon cycle operates over shorter timescales. As such, it could also be an important negative feedback during shorter-term warming events (e.g., hyperthermals).

Conclusions
The transport and burial of biospheric organic carbon in coastal marine sediments is an important carbon sink in modern settings. However, the relative importance of this feedback in past warm climates remains a major gap in our understanding. Using a multi-proxy approach, we find a significant increase in the accumulation rate of terrestrial biomarkers in early Eocene coastal marine sediments from offshore East Antarctica. As the early Eocene Antarctic continent was covered by a vast terrestrial OC reservoir (>10 15 PgC), the transport and subsequent burial of biospheric organic carbon in coastal marine sediments could have acted as a key CO 2 sink. This study highlights the importance of the terrestrial biosphere during past warm climates and its potential role as a negative feedback to stabilize Earth's surface temperature.