Millennial soil retention of terrestrial organic matter deposited in the Bengal Fan

The abundance of organic carbon (OC) in vegetation and soils (~2,600 PgC) compared to carbon in the atmosphere (~830 PgC) highlights the importance of terrestrial OC in global carbon budgets. The residence time of OC in continental reservoirs, which sets the rates of carbon exchange between land and atmosphere, represents a key uncertainty in global carbon cycle dynamics. Retention of terrestrial OC can also distort bulk OC- and biomarker-based paleorecords, yet continental storage timescales remain poorly quantified. Using “bomb” radiocarbon (14C) from thermonuclear weapons testing as a tracer, we model leaf-wax fatty acid and bulk OC 14C signatures in a river-proximal marine sediment core from the Bay of Bengal in order to constrain OC storage timescales within the Ganges-Brahmaputra (G-B) watershed. Our model shows that 79–83% of the leaf-waxes in this core were stored in continental reservoirs for an average of 1,000–1,200 calendar years, while the remainder was stored for an average of 15 years. This age structure distorts high-resolution organic paleorecords across geologically rapid events, highlighting that compound-specific proxy approaches must consider storage timescales. Furthermore, these results show that future environmental change could destabilize large stores of old - yet reactive - OC currently stored in tropical basins.

February 1994 (top of core SO93-96KL) and June 2006. This latter method, first applied to core SO93-96KL 1 makes use of 2-to 15-cm thick, fining-upward, interbedded sand and silt layers found throughout the core and interpreted as tempestites. These units have distinct basal contacts, generally grade from ~80% sand and silt at their base to ~30% near the top, and lack current-induced bedding 1 . These characteristics suggest deposition by settling from suspension clouds that are mobilized and transported by tropical cyclone-induced downwelling and that, upon crossing the deeper water of the SoNG, lose their ability to transport coarser particles 1,2 . Down-core sand and silt concentrations are derived from very-high-resolution laser diffraction particle size analyzer records and positively correlated with the historical cyclone record 1 . The application of this approach to core SO188-336KL was undertaken at the University of Bremen and provided a highresolution age model for the top 358 cm (21 years) 4,8 .

Bulk measurements
The bulk-sediment weight-percent total organic carbon content (TOC) of all samples was analyzed in triplicate on an elemental analyzer coupled to a Finnigan Deltaplus isotope ratio mass spectrometer (EA/IRMS). TOC compositions were determined following fumigation acidification of powdered sample aliquots 9 . These were sealed in a vacuum desiccator with a beaker of 50 mL of 12N HCl, fumigated for 60-72 hours at 60-65°C to remove carbonates, and dried in a separate desiccator for an additional 24 hours prior to measurement. Average precision (2) of replicate measurements are 0.02%.
Major and trace element concentrations (used in Al/Si calculations) were determined at the Service d'Analyse des Roches et des Minéraux (SARM; Nancy, France) by IPC-AES and ICP-MS following LiBO2 fusion 10 of powdered sample aliquots pre-rinsed with milli-Q water to minimize sea salt contributions.
Sr and Nd isotopic compositions were measured on powdered sample aliquots at CRPG (Nancy, France) by Thermal Ionization Mass Spectrometry following carbonate removal via leaching with 10% acetic acid 11 . Nd isotopic compositions are reported as εNd. Average uncertainties (2) of major/trace elemental compositions and of 87 Sr/ 86 Sr and εNd isotopic compositions are better than 2% (relative), 2x10 -5 , and 0.5 ε units, respectively.

Bulk Radiocarbon
Aliquots of powdered samples were weighed into silver capsules to yield between 250 and 400 g of C. The measured TOC was used to estimate the required sample mass. The powdered sample was acidified by HCl fumigation 9 to remove inorganic carbon prior to radiocarbon analysis. The bulk radiocarbon data was acquired at ETH Zurich using the elemental analyzer-accelerator mass spectrometer (EA-AMS) MIcroscale CArbon DAting System (MICADAS) 12 .

Sample preparation for molecular analysis
Sediment samples were freeze-dried and lipids were extracted from powdered sediment (~140-210 g) with a 9:1 (v:v) dichloromethane:methanol (DCM:MeOH) solvent mixture using a microwave-assisted reaction systems (MARS, CEMS corporation). After centrifuging, the solvent extract was decanted and collected. The sediment was solvent-rinsed and centrifuged a minimum of three times. The total lipid extract was concentrated using a Turbovap and saponified using 15 mL of 0.5 M KOH in MeOH and ~150 μL of Milli-Q water. After the solution was heated for 2 hours at 70°C, 15 mL of Milli-Q water and 0.5 g of NaCl were added to the solution. A basic lipid fraction was extracted with hexane (5 x 5 mL rinses). The remaining solution was acidified dropwise to a pH of ~2.5 using 12 N HCl. An acidic lipid fraction was extracted with 4:1 hexane:DCM (5 x 5 mL rinses). The basic and acidic lipid fractions were collected and fractionated separately. They were dried over combusted Na2SO4 and fractionated into compound classes by column chromatography using a stationary phase of 1 g aminopropyl-functionalized silica gel. Five fractions were eluted using 4 mL of hexane The F4 fractions of the basic and acidic lipid extracts were combined into a total F4 fraction containing the fatty acids. The fatty acids were methylated with acidified MeOH of known isotopic composition by adding 15 mL of 95:5 MeOH:HCl to the dried fatty acid fraction. The samples were purged with nitrogen and heated at 70°C overnight, after which the methylation reaction was quenched with 15 mL of Milli-Q water. The fatty acid methyl esters (FAMEs) were recovered using 4:1 hexane:DCM (5 x 6 mL rinses) and dried over combusted Na2SO4. The FAMEs were purified further with a second aminopropyl-functionalized silica gel column. Three fractions were eluted with 4 mL of hexane (F1), 7 mL of 4:1 hexane:DCM (F2, FAMEs), and 15 mL of 1:1 DCM:MeOH (F3, column flush).
The purified FAMEs fractions were screened and quantified on a gas chromatography-flame ionization detector (GC-FID). Saturated FAMEs were further purified by silver nitrate chromatography, which removed unsaturated compounds. Three fractions were eluted from Pasteur pipettes loaded with 0.5 g of silver nitrate impregnated silica gel, where 5 mL of 95:5 hexane:DCM was used to elute F1, 18 mL of 5:1 hexane:DCM was used to elute F2 containing FAMEs, and 5 mL of 1:1 DCM:acetone was used to elute F3. The purity of the saturated FAMEs was reassessed by GC-FID prior to stable carbon isotopic analysis and preparative capillary gas chromatography (PCGC) for compound-specific radiocarbon analysis. Purified saturated FAME fractions were subsampled for stable C isotopic analyses. This study focuses on saturated, even-numbered, straight-chained fatty acids, where the n-CX:0 fatty acid will be referred to as n-CX (x corresponds to the carbon chain length).

Compound-specific stable carbon isotopic analysis
The stable carbon isotopic compositions of the FAMEs were acquired on an HP 6890 GC with a Gerstel CIS-4 programmable temperature vaporizing (PTV) inlet and CP-Sil 5-CB-MS column (0.25 mm i.d. x 0.25 μm phase x 60 m length) coupled via a Finnigan-MAT GCC-III (GC Combustion-III) interface 13 to a DeltaPlus gas isotope ratio mass spectrometer. The GCC-III reference gas was calibrated using a suite of nine extensively analyzed compounds injected repeatedly, resulting in an accuracy and precision averaging better than 0.3‰. Samples were analyzed in triplicate at a minimum, and the associated error represents the standard deviation from the mean.

Compound-specific radiocarbon preparation and analysis
Six individual saturated FAMEs (n-C16, n-C24, n-C26, n-C28, n-C30, and n-C32) were purified and collected using the PCGC method 14 using either an Agilent 7890A or HP 5890 Series II GC coupled to a Gerstel fraction collector. The purified saturated FAMEs fractions were dissolved in either iso-octane or toluene at a concentration that yielded 0.5-1 μg on column per injection. Depending on the total FAME concentration, ~50-150 injections were performed. The compounds were eluted from the PCGC traps with 4 mL of DCM, concentrated under a nitrogen stream, and further purified by 1% deactivated silica gel column chromatography (~3 cm of gel) by eluting 4 mL of DCM. The recovery and purity was checked on a GC-FID, where yields were in the range of ~40-80% of the initial material.
If purified n-C30 and n-C32 FAME concentrations were estimated to yield C masses less than 10-15 μg, these two compounds were combined into a n-C30+32 FAME sample to increase sample size and reduce analytical uncertainty during radiocarbon analyses. The purified FAMEs were dissolved in DCM (~ 250 μL) and loaded into combusted quartz tubes. Samples were dried in each quartz tube under a high-purity nitrogen stream at 37°C, until all solvent was removed. Combusted copper oxide (~150 g) was added to the quartz tube after solvent removal. The samples were frozen in the quartz tube in a dry ice/isopropanol slurry for several minutes before the tubes were evacuated for ~1 minute to < 30 μTorr. The dry ice/isopropanol slurry was replaced with liquid nitrogen, and the quartz tubes were flame-sealed under vacuum. The FAME samples were combusted in flame-sealed quartz tubes at 850°C for 5 hours. The following day, the quartz tubes were cracked under vacuum, releasing the evolved gas. A dry ice/isopropanol slurry was used to trap water that was produced during combustion. The sample CO2 was trapped with liquid nitrogen and manometrically quantified before being trapped using liquid nitrogen and flame-sealed in a pyrex tube for radiocarbon analyses. Radiocarbon measurements of sample-derived CO2 were performed at ETH Zurich between September 2015 and September 2016. The AMS MICADAS system and operation parameters used at ETH Zurich are described by Christl et al. 15 .
Some samples were lost (e.g., sample tube was broken) or contaminated during the radiocarbon preparation and analysis. Carbon masses calculated on the vacuum line were compared to the GC-FID concentrations to identify contamination. Samples 292-302 cm, 905-915 cm, and 1,505-1,515 cm were the first samples that were prepared for radiocarbon and some adverse conditions were noted during their preparation. In the case of sample 905-915 cm, a capillary broke in the preparative fraction collector during the PCGC preparation, which likely led to the observed low sample recoveries, and these samples had higher carbon masses on the vacuum line than expected compared to the GC-FID quantifications. Additional peaks were noted in the GC-FID chromatograms of the PCGC isolated fatty acids from 292-302 cm and 1,505-1,515 cm. These peaks likely contributed to the larger carbon masses on the vacuum line than estimated from the fatty acid quantification on the GC-FID. These three samples do not differ from the remaining samples in the following characteristics: fatty acid distributions, total fatty acid concentrations, TOC values, 137 Cs values, bulk organic 14 C, Al/Si ratios, fatty acid  13 C values, or mean grain size. Therefore, it was concluded that these samples were indeed contaminated, so they were not included in the fatty acid age distribution modeling.

Blank Determination for Compound-Specific 14 C analysis
It is assumed that the preparative GC and vacuum line preparation are the primary sources of 14 C contamination. The purification steps prior to the PCGC isolation are not considered in the following blank assessment. In order to characterize the magnitude and isotopic composition of the blank contribution to samples during the PCGC and vacuum line preparation, two solvent blank PCGC analyses were performed where pure solvent, rather than sample, was injected. These experiments were performed under the same analytical conditions described above for the fatty acid samples, and 110 and 80 injections were completed for the first and second experiment, respectively. The first three traps were opened within a minute of the retention time corresponding to when n-C18 typically elutes. Likewise, the final three traps were opened within a minute of when n-C30 typically elutes. The first three traps were eluted into 4 mL vials that were spiked with 10, 25, and 40 μg of a modern n-C18 FAME standard (Fm = 1.1124). The last three traps were eluted into 4 mL vials that were spiked with 10, 25, and 40 μg of a dead n-C30 FAME standard (Fm = 0.0). The FAME standards and isotopic measurements were provided courtesy of Li Xu (NOSAMS, Woods Hole, MA, USA). After this point, the blank samples were prepared according to the same sample protocol described for the sample fatty acid radiocarbon analyses. Radiocarbon measurements of blank-derived CO2 were performed at ETH Zurich between September 2015 and March 2016. Some blank traps had an anomalous degree of contamination, so they were excluded from the blank calculations. These samples had much high carbon masses than were expected based on the spike concentrations. This contamination was likely introduced through leaks during vacuum line preparation, as indicated by the detection of non-condensable gases, or failing to successfully evacuate and flame-seal sample requiring that the sample be transferred and re-prepared for vacuum line preparation.
The two different blank experiments with different number of injections yielded similar results (Table S6).
Therefore, these two datasets were combined to determine the blank mass and isotopic composition.
Similarly, the magnitude of the dead blank component was calculated graphically using the modern n-C18 FAME data (Fig. S4) and the following rearranged equation written as a function of 1/mMeas: The same uncertainty-weighted Model II regression was applied to calculate mB_Dead from the slope substituting the value for mB_Mod calculated in the previous regression and the known value of the modern n-C18 FAME standard for FmStd. The regression parameters are listed in Table S7 (R 2 = (Table S8).

Blank and methylation correction for fatty acid radiocarbon data
The measured FAMEs fraction modern data are corrected for blank contribution during the PCGC and vacuum line preparation according to the mass balance equations: and where mT and FmT are the true sample mass and fraction modern, respectively, without blank carbon contribution.
The blank-corrected Fm data are further corrected for a single carbon addition during the methylation step according to the following equation: where FmT, FA is the methylation and blank-corrected fatty acid Fm, FmT,FAME is the blank-corrected FAME Fm, FmMeOH is the Fm of the MeOH used during the fatty acid methylation, and n equals the purified fatty acid chain length. In the case where n-C30 and n-C32 were combined into n-C30+32, average chain length (ACL) is substituted for n in equation S10, where ACL is determined according to a concentration weighted average: The reported error for the corrected fatty acid Fm represents the propagated 1 error through blank and methylation corrections. The carbon masses measured on the vacuum line were assigned an error of +/-5%.

Methylation correction for fatty acid 13 C data
The fatty acid stable carbon isotopic data are also corrected for a single carbon addition during methylation according to the following mass balance equation: where  13 CMeOH is the stable carbon isotopic composition of the MeOH used for methylation,  13 CMeas, FAME is the measured FAME stable carbon isotopic composition, n equals the purified fatty acid chain length, and  13 CFA is the methylation corrected fatty acid stable carbon isotopic composition. The reported error reflects the propagated 1 error from the analytical error and the error associated with the  13 CMeOH.

Numerical simulations of fatty acid age structure
The incorporation of bomb carbon into the fatty acids demonstrates that measured fatty acid radiocarbon ages, which are older than the initiation of nuclear weapons testing, mask a mixture of an old component that is relatively insensitive to the atmospheric bomb spike and a fast-cycling component that incorporates bomb carbon. Accordingly, a two-component isotope-mixing model was constructed to quantify the ages and fractional contributions of the fast-and slow-cycling components, which can be expressed as and fFast + fSlow = 1 (S14) where FmFA is the measured fatty acid Fm, FmFast and FmSlow are the Fm of the fast-and slow-cycling components, and fFast and fSlow are the fractional abundances of the fast-and slow-cycling components.
Rather than assigning a discrete age to each of the components, normal (Gaussian) age distributions were used to characterize the fast-and slow-cycling components 18,19 . This approach takes into account that continental reservoirs host organic matter with a smear of ages rather than a single discrete age or a combination of several discrete ages. The age distributions are described by the following probability distribution function: where  is the standard deviation or width of the distribution and  is the mean or center of the age distribution.
The Fm of the two components are expressed as a linear combination of sums of the atmospheric Fm (FmAtm) weighted by the probability distribution function both evaluated at time t: and FmSlow =  t=t0 pSlow(t| Slow,Slow)*FmAtm(t). (S17) The time domain is limited from the sediment deposition year t0 to 100,000 years BP (tmax). This truncates the age distributions at the sediment deposition year (t0) such that all of the fatty acids were biosynthesized before or during the year of sediment deposition. Importantly, organic matter older than 50,000 BP is considered radiocarbon dead, so organic matter older than 50,000 BP is indistinguishable by radiocarbon. In order to account for truncation, the areas of the normal distributions are normalized so that they integrate to 1: and FmSlow = ( tmax t=t0 pSlow(t| Slow, Slow)*FmAtm(t))/ tmax t=t0 pSlow(t| Slow, Slow). (S19) As a result of truncation,  deviates from the average age of the distribution, so it primarily represents the age offset of the distribution center relative to the sediment deposition year t0.
Atmospheric radiocarbon composition, which sets the original fatty acid radiocarbon composition, varies over For each chain length, the calculated fSlow, FmFast(t), and FmSlow(t) corresponding to each combination of fastand slow-cycling ages are substituted into equations S13 and S14 to generate synthetic Fm time series. Finally, the root mean squared error (RMSE) is calculated to determine the fit between the synthetic Fm data and the measured fatty acid data for each chain length. In addition to calculating RMSE for each individual fatty acid, a combined C24-32 RMSE is calculated for each age structure combination in order to identify age structures that best approximate the measured long-chain fatty acid data on a whole. Fast-and slow-cycling age combinations are filtered out if the optimal fSlow was determined to be less than 0 or greater than 1.

Geochemistry and modeling results
The bulk geochemistry shows little variability over the sample set (Table S1). The total organic carbon (TOC; range: 0.39 -0.61%) is positively correlated with Al/Si ratios (range: 0.32 -0.44) (Fig. S5), which suggests similar particle loading as observed in the G-

Bulk organic carbon age distribution
Down-core bulk OC values record a muted bomb spike that is offset below the weighted average long-chain fatty acid Fm (Fig. S8). An additional old OC component, devoid of n-C16 and long-chain fatty acids, explains the translation of bulk OC Fm to lower values. Previously, terrestrial OC in the G-B floodplain and delta was apportioned into ~5% petrogenic carbon, 10-29% refractory biospheric carbon exceeding an average age of 15,000 years, and 66-85% labile biospheric carbon 29, 30 . By construction, the refractory biospheric carbon does not contain long-chain fatty acids because the average age of 15,000 years was derived by extrapolating to 0 g/g n-C24+ fatty acid concentration 29 , and unlike n-alkanes, petrogenic sources do not contribute to the fatty acid inventory 31,32 . Indeed, our model results suggest that long-chain fatty acids are absent in terrestrial OC exceeding 15,000 years because nearly 100% of the fatty acid inventory is deposited in the Bengal sediments within 0 to 2,500-5,000 years of biosynthesis, depending on the best-fitting age distributions.
In order to characterize the age structure of the bulk OC, a mass balance is written in terms of a labile biospheric      inverse measured C mass for blanks spiked with modern n-C18 FAME. The regression lines were calculated using an uncertainty-weighted Model II regression.   Table S2). FAME concentrations are reported relative to gram dry weight of sediment (gdw), and the error bars represent 1 standard deviation.