## Introduction

The concentration of atmospheric carbon dioxide (pCO2, expressed in partial pressure µatm), as directly measured from air trapped in ice cores, has had a major influence on climate over the past 800 thousand years (ka)1. During this period, pCO2 and temperature oscillated together between stable bounds every 100 ka2. In the past two centuries, the rise of pCO2 has broken those bounds from the pre-industrial values, previously only ranging between ca. 180 to 280 µatm, to the 410 µatm of today3. This rapid rise in pCO2 causes concern that climate, particularly temperature, will accordingly change. In order to better understand how changes may occur, reconstructing longer trends in pCO2 over the geologic record could offer context for evaluating the direction and magnitude of climate change.

Many proxies have been developed for reconstructing past pCO2 and applied with mixed success over the past several decades4. One method for studying past pCO2 makes use of the stable carbon isotopic fractionation due to CO2-fixation (Ɛp), where biomass of photoautotrophs becomes increasingly depleted in 13C as pCO2 increases due to kinetic discrimination by the CO2-fixing enzyme Rubisco5,6,7. Ɛp can be derived from the δ13C of photoautotrophic biomass, recorded in sedimentary organic matter, and the δ13C of inorganic CO2 derived from the carbonate in the shells of planktonic foraminifera8.

Although pCO2 has been shown to be one of the dominant physiological control on the δ13C of photoautotrophic biomass9, studies on Ɛp in algae have shown that other factors may influence this value, primarily growth rate10 and cell size and shape11, as well as minor influences such as light, and temperature12,13,14,15. These additional influencing factors on Ɛp are considered in pCO2 reconstructions via the catchall term b16, described in the equation17 as:

$${\text{CO}}_{{{2}[{\text{aq}}]}} = b/ \, (\upepsilon_{{\text{f}}} - \upepsilon_{{\text{p}}} )$$
(1)

where Ɛf is the maximum isotopic fractionation due to CO2-fixation via the enzyme Rubisco, which has shown a sum range from 25 to 28‰17,18,19. It should be noted that the very few in vivo Rubisco fractionation studies have much lower values20,21, which Wilkes and Pearson22 suggest there may be due to multiple stages of fractionation instead of the singular Rubisco fractionation step. Several other studies have expanded on Eq. (1) for specific consideration, particularly in calculating b, e.g. instantaneous cell growth rate accounting for differences in photoperiod23,24 and CO2 fixation rate25.

Using the knowledge obtain from culture studies26,27, the measurement of Ɛp in algal biomarkers preserved in the geologic record can be used to reconstruct past pCO2. These biomarkers are almost exclusively alkenones, long-chain unsaturated methyl and ethyl n-ketones produced by haptophytes8,28,29. Although this proxy has generated a large number of pCO2 records30,31,32, there are several limitations, such as the exceptionally low Ɛf recorded for the alkenone-producer Emiliania huxleyi of 11‰20, a potential insensitivity of this proxy at low CO2 levels24,33, and difficulties in constraining the b factor over time34. One other limitation is the fact that alkenones first commonly appeared in the geologic record ca. 45 million years (Ma) ago35, prohibiting pCO2 reconstructions prior to this time.

As an alternative, the isotopic fractionation of general phytoplankton biomarkers, compounds that are produced by a multitude of species, may offer some solutions to the limitations of the alkenone pCO2 proxy such as more spatial ubiquity and temporal longevity. This general biomarker approach has been poorly explored; however, though there are some examples of this being applied to phytane, a diagenetic product of omnipresent chlorophyll-a, for periods extending beyond the alkenone record, i.e. in the Cretaceous36,37,38 and in a Phanerozoic compilation39. However, this general biomarker approach has not been extensively tested in laboratory cultures or present-day environments.

For modern studies of the general biomarker approach, naturally-occurring phytoplankton communities are necessary to mimic the widespread contributors to general phytoplankton biomarkers, as opposed to the typical single-species approach of laboratory cultures. Mesocosm experiments may offer more natural environmental conditions and communities, though none have been conducted on general phytoplankton biomarkers for pCO2 reconstructions. Alkenones and particulate organic carbon (POC) have been explored in one mesocosm experiment using natural communities, i.e. under three pCO2 conditions in a contained area for ca. 21 days40. These authors suggested the minor changes they observed in δ13C values for alkenones and POC indicate that fractionation is not primarily controlled by CO2 concentrations but instead by algal growth rate and carbon-uptake mechanisms. However, these experiments are inherently difficult to set-up, reproduce, and control.

Here we expand this new approach to testing pCO2 response in natural phytoplankton communities, by analyzing the response of isotopic fractionation in general phytoplankton biomarkers across a CO2 gradient at a naturally occurring CO2 seep. CO2 seeps, which consistently bubble CO2 into the surrounding environment and thus have very high CO2 concentrations near the seep, have hardly been explored for biological studies due to the assumed high sulfide concentrations, toxic to many organisms, typically associated with volcanic degassing41. However, Hall-Spencer et al.42 used these extremely high pCO2 environments for ocean acidification experiments, which lead to studies at other seep sites, i.e. Italy43, Papua-New-Guinea44, New Zealand45, and Japan46. The new approach was initially tested with a 3-point transect (high, mid, and control pCO2) of a marine CO2 seep site on Shikine Island, Japan, covering a range of CO2 concentrations that offer an analogue for past oceans39. However, this specific site proved to have confounding factors where the imprint of CO2 on Ɛp measured in general biomarkers of surface sediment was masked by extreme weather events (i.e. typhoons) that caused sediment transport.

Here, we more thoroughly explore this new approach at a different marine CO2 seep system approximately 30 m into Levante Bay at Vulcano Island, Italy, a location with much more stable weather conditions than Japan. We collected surface sediments in a high-resolution 16-point transect from high CO2 towards ambient CO2 values. Here, we analyzed the Ɛp of several general phytoplankton biomarkers, compounds that have been virtually unstudied in modern phytoplankton communities, deposited in surface sediments and tested their response to the CO2 gradient at sixteen sites throughout the bay.

## Results

For this study, we collected surface sediments in May and October from close to the seep site (ca. 3 m distance) to a control site unaffected by the seep47 at a constant depth of ca. 1.5 m at the time of sampling (Fig. 1). The δ13C of DIC measured in seawater collected in May from the bay does not show notable change over the gradient of CO2 (Table S1), which confirms that lack of change noted in the literature48. For this reason, we averaged the δ13C of DIC measured in our study with that of Cornwall et al.49 across all sites (0.7‰ ± 0.4‰ s.d.) and assumed this to be representative for the bay region.

Analysis of the polar fractions of the lipid extracts obtained from the surface sediments showed the same biomarker lipids in similar distributions throughout the transect from the CO2 seep to the control sites (e.g. biomarker distributions at Site 5, near the CO2 seep, and the control site, Site 16, are shown in Fig. 2). These biomarker lipids include: loliolide, phytol, even carbon numbered C10-C16 fatty alcohols, C30 alkane-1,15-diol, C32-17β(H),21β-hopanol, and sterols, such as cholesta-5,22E-dien-3β-ol, cholesterol, 23-methylcholesta-5,22dienol, campesterol, stigmasterol, and β-sitosterol (Fig. 2). Consistently, the most abundant among these compounds were loliolide, cholesterol, and phytol, as were also observed at the CO2 seep site in Japan50.

All three biomarkers show a steady increase in δ13C values over the transect from the CO2 seep towards the control site (Fig. 3; Table S2). The exceptions are the more depleted δ13C values at Site 2 and Site 9, where we observed some minor gas bubbling in the sediment, suggesting the release of small amounts of CO2 at these sites. Over the transect from Site 1 (the seep) to Site 16 (the control), the δ13C of loliolide ranges from − 27.4 to − 21.6‰ (Fig. 3A). From the seep to around Site 10, the δ13C of loliolide fluctuates between ca. − 27 and − 25‰, followed by a prominent increase from ca. − 25 to − 22.5‰. For the sites sampled in both May and October, there appears to be consistency between the two seasons, i.e. in Site 5 (− 25.2‰ for both seasons) and the control site (− 22.3‰ in May and − 22.6‰ in October), though Site 14 shows a spread of 2‰ between seasons. The δ13C of cholesterol shows a smaller but more consistent shift over the transect, ranging from − 26.3 to − 21.2‰ with a 1‰ difference between the two seasons (Fig. 3B). Phytol shows the largest shift, ranging from − 28.4‰ at the seep site to − 20.4‰ at the control site (Fig. 3C). There is a relatively consistent increase in the δ13C of phytol over the entire transect, except for a small decrease at Site 9, where we observed minor additional gas bubbling in the sediment. The δ13C of phytol shows minor variation between seasons (ca. 0.5%), except for the control site which showed a difference of 1.4‰.

## Discussion

The three most abundant biomarkers, loliolide, cholesterol, and phytol, are all derived from phytoplankton and represent broad phytoplankton groups51,52,53. Composition of the diatom assemblages and cyanobacteria in this bay are further described in Johnson et al.43 All become increasingly enriched in 13C over the transect from high CO2 concentrations near the seep to the control Mediterranean values. The observed isotopic depletion that occurs with increasing CO2 concentrations matches theory5,6,54. Furthermore, this pattern closely follows the results observed at Shikine Island, i.e. a consistent depletion δ13C of the same biomarkers with increasing proximity to the CO2 seep50, but here offered in a 16-point transect instead of the 3-points at the Japan site. Given that CO2 was the major variable over the transect in Italy, as well as Shikine Island, this strongly suggests that CO2 concentrations indeed have a strong impact on isotopic fractionation of general phytoplankton biomarkers, suggesting their potential as a pCO2 proxy.

Although the general trends between the two CO2 seep sites are similar, there is a difference in the magnitude and consistency in isotopic changes between the two sites. In the Shikine Island study, loliolide showed the largest isotopic shift over the transect (− 7.9‰) as compared with phytol (− 5.2‰) and cholesterol (− 5.2‰). However, in the Vulcano Island surface sediments, phytol had the most pronounced isotopic shift (− 8.0‰) as compared with loliolide (− 5.8‰) and cholesterol (− 5.1‰). Furthermore, the changes in loliolide over the Vulcano Island transect are more variable compared with the consistent trends in isotopic values observed in phytol and cholesterol. Here, we will explore these differences.

The δ13C profile of loliolide at Vulcano Island (Fig. 3A) has the least consistent trend among the three biomarkers, fluctuating between − 27.4 and − 25.0‰ from Site 1 (the seep) to Site 13. Loliolide is derived from the major xanthophyll fucoxanthin and is considered a biomarker for diatoms, especially in the absence of haptophyte algae51,55, based on its predominance at sites with substantial diatom communities, although some other non-diatom species also produce fucoxanthin56. Light microscopy analysis of selected sediments across the transect showed that Site 2 contains nearly no diatom frustules, Site 5 had abundant centric diatoms as well as some pennate diatoms, while Site 9 is characterized by a great diversity especially among pennate diatoms though with relatively low overall abundance, and Site 13 and Site 16 (control site) had both high abundance and high diversity of both centric and pennate diatoms (Stoll H. and Mejía Ramírez L. M., personal communications). Decreased diversity in increased proximity to the seep has previously been observed in periphytic diatom assemblages at this site43, though with a drastic increase abundance in chlorophyll-a by ca. fivefold from Site 6 to 16. Johnson et al. suggest that the increase abundance but decreased diversity is due to some diatoms benefitting from increasing CO2 through a reduction in the energetic costs of their CCMs43. The different composition of diatoms at each site, particularly between centric and pennate diatoms, may explain why we observe a high δ13C variability in loliolide. Different species may have slightly different isotopic fractionation due to e.g. different cell geometry and morphologies11 or different bicarbonate pumping strategies that has been observed in diatom species57,58,59. This concept may be further supported by the stronger increase in δ13C values observed between sites 13 and 16, where the higher diversity of species may yield a more robust overall δ13C signal through averaging biosynthetic differences among species. This complexity in the signal of loliolide may weaken the potential of this biomarker for past pCO2 reconstructions.

The δ13C profile of cholesterol (Fig. 3B) showed a more consistent decline over the transect than loliolide, though with a smaller difference in absolute values than phytol and loliolide from the seep towards the control. Because cholesterol is produced by all eukaryotes, such as phytoplankton or by heterotrophs, which modify ingested sterols52,60, terrestrial input, in addition to the algal input, can potentially dilute the autochthonous isotopic signal. However, the lack of terrestrial triterpenoids and long-chain (> C22) even carbon number fatty alcohols (Fig. 2) suggest minimal input of terrestrial biomarkers in the bay. Another explanation for the smaller isotopic change is that the cholesterol has contributions from heterotrophs, which produce cholesterol by modifying ingested phytoplanktonic sterols. Although this does not yield large isotopic fractionation61, the zooplankton often have stronger mobility than their photoautotroph counterparts; they may consume phytoplankton from various locations (and consequently various CO2[aq] concentrations) throughout the bay. This idea is supported by the notable δ13C differences in cholesterol between the two seasons, where the offsets are not consistently in one direction. Based on these observations the δ13C of cholesterol must be considered carefully when used in reconstructing past CO2 concentrations.

The δ13C profile of phytol had the most robust trend across the transect (Fig. 3C) with an δ13C enrichment of ca. 8‰ from the seep to the control. Phytol, derived from chlorophyll-a, is found in all oxygenic photoautotrophs53. Terrestrial input may affect the signal of phytol but, as discussed above, there is no evidence for this here. Based on these results, phytol shows the greatest sensitivity to the CO2 gradient, and thus the most promise for reconstructing past pCO2. The phytol results from Shikine, Japan50 likewise show great promise for reconstructing past pCO2.

To test the validity of using the δ13C of the general biomarkers to estimate past pCO2, we used phytol, the most promising of the various general phytoplankton biomarkers explored here with the most consistent trend and the strongest δ13C shift over the gradient. We calculated the stable carbon isotopic photosynthetic fractionation (Ɛp) using the δ13C of phytoplankton biomass (δp) and the δ13C of CO2d):

$$\upepsilon_{{\text{p}}} = 1000 \, \cdot \, \left[ { \, \left( {\delta_{{\text{d}}} + 1000} \right) \, / \, \left( {\delta_{{\text{p}}} + 1000} \right) \, {-}{ 1}} \right]$$
(2)

The δp is calculated from the offset between phytol and biomass, which is 3.5‰ ± 1.3 standard deviation based on the average of 23 representative marine phytoplankton species grown in cultures39. The δd is calculated from the δ13C of DIC (0.7‰ ± 0.4‰ s.d.) correcting for temperature and pH (Table S1). The mean annual sea surface temperature for Vulcano Island (20.2 ºC ± 0.5 °C s.d.; https://www.seatemperature.info) was used to calculate the temperature-dependent carbon isotopic fractionation of CO2[aq] with respect to HCO362. The pH gradient, ranging from 5.5 pH near the vent to 8.2 pH in the control63, was used to define the relative contribution of different inorganic carbon species to the measured DIC64 (Table S1). Uncertainty was calculated using Monte Carlo simulations which consider the culmination of each individual parameter with its associated uncertainty, as described by Witkowski et al.39, here including δ13C of phytol ± 0.5‰ s.d., offset between biomass and phytol ± 1.3‰ s.d., δd ± 0.4‰ s.d., and T °C ± 0.5 °C (Table S2). This uncertainty has an equal effect on the final uncertainties in calculated εp, i.e. 0.1‰ error in the δd will lead to a 0.1‰ error in εp39.

Phytol-derived Ɛp ranges from 22.2 to 8.2‰ ± 1.4‰ s.d. (Fig. 4A) and shows a consistent decline in fractionation from the seep towards the control site. This includes Site 2 where measured δ13C values are higher than at Site 1, but Ɛp now shows the expected trend of more fractionation closer to the vent. This is attributed to the strong shift in pH between these two sites (5.5 pH at the vent and 6.25 pH at Site 263) which we have here corrected for. The highest Ɛp value of 22.2‰ near the seep is approaching maximum isotopic fractionation due to CO2-fixation (Ɛf), which has been shown to range between 25 and 28‰ in laboratory cultures18, but still does not quite reach full expression of Ɛf. This is somewhat unexpected given the constant bubbling of CO2 at this site and thus very high CO2 concentrations, i.e. up to ca. 3× modern CO2[aq]43.

Several possibilities may explain why the full expression of Ɛf has not been reached. For one, given the relatively small area of the bay, it is possible that surface sediment has moved around the bay over time due to tidal actions and bottom water currents, which dampens the overall signal by allochthonous organic matter transported from area’s outside of the bay, as also inferred for Shikine Island50. Furthermore, algae are unlikely to grow and deposit in precisely same location and given that the impact of the CO2 seep noticeably changes over tens of meters43, this likely leads to some mixed signal among sites, resulting in a suppressed signal. Another alternative is that the calculated Ɛf of the phytoplankton community in Levante Bay may be lower than that inferred from the many culture studies11,17,65. Indeed, several recent studies show that Ɛf of the different Rubisco types may be lower than previously assumed66.

In order to see how well Ɛp of phytol can reconstruct CO2[aq], we estimated CO2[aq] and pCO2 from the δ13C of phytol using the equation adapted from the high plant model5 for algae7, and described in Eq. (1)17, where b reflects species carbon demand per supply8 and Ɛf reflects the maximum isotopic fractionation due to CO2-fixation. The value of b is a complicated catchall for factors influencing isotopic fractionation such growth rate and cell-size67, light intensity and membrane leakiness24,68, further complicated due to the multitude of sources for general phytoplankton biomarkers. Studies have suggested an empirical average 170‰ kg µM−1 ± 43 kg µM−1 s.d. for b based on a compilation of δ13C values of bulk organic matter in marine surface sediments, as well as some limited phytol studies39,50. Furthermore, we use an average Ɛf for phytoplankton species of 26.5‰ ± 1.5‰ uniform distribution39 based on the 25 to 28‰ range observed in laboratory cultures69. As described above, uncertainty was calculated using Monte Carlo simulations, considering each individual parameter with its associated uncertainty, as described by Witkowski et al.39. Here, we include the uncertainties associated with Ɛp plus the new additional uncertainties associated with b ± 43 kg µM−1 s.d., Ɛf ± 1.5‰ uniform distribution, T °C ± 0.5 °C s.d., and sea surface salinity ± 1‰ s.d.

The resulting phytol-based CO2[aq] values range from 9.3 to 39.4 µM (Fig. 4B). The highest value of 39.2 µM (+ 20.6/− 11.0 µM) is near the vent at Site 1, dropping to 23.7 µM (+ 7.1/− 5.2 µM) at Site 2, then to 14.3 µM (+ 3.0/− 2.7 µM) at Site 3, before gently declining to 9.6 µM (± 1.8 µM) at the control Site 16. If we calculate the pCO2 from CO2[aq] using Henry’s Law constant K0, which considers salinity and temperature70, the resulting pCO2 reconstruction range from 280 to 1,182 µatm (Fig. 4C). The highest pCO2 values were reconstructed for the sites closest to the seep, Site 1 at 1,200 µatm (+ 636/− 333 µatm) and Site 2 at 728 µatm (+ 222/− 158 µatm), while the remainder of the transect showed fairly ambient values from Site 3 at 438 µatm (+ 95/− 80 µatm) to the Site 16 control at 294 (+ 56/− 50 µatm).

Comparison of CO2[aq] estimates with those reported for sites63 equivalent of our Site 2, 9, and 16 (30 µM ± 7, 14 µM ± 1, and 12 µM ± 1, respectively; Fig. 4B), show that these estimates agree within uncertainty, suggesting that our approach yields reasonable estimates. Only at the control site there is a slight underestimation of CO2 concentrations. One possible explanation is an incorrect assumption for the b value. However, this seems unlikely given that (i) b values would need to be increased beyond any known b value thus observed to account for this underestimation, and (ii) this would lead to even higher past pCO2 estimations which are based on b values compiled from laboratory cultures and natural experiments39. A more likely explanation is the change in phytoplankton community over the bay, where the control community is dominated by high affinity CCM species as observed for macroalgae49. Given that these species actively pump bicarbonate under low CO2 conditions, this may explain the lessened Ɛp, yielding lower CO2 estimations. This effect has also been observed in the mesocosm experiments with different CO2 concentrations40, especially if there is limited carbon dioxide leakage from cells. Recent studies have shown lower sensitivity of Ɛp to CO2 in laboratory cultures and in glacial-interglacial reconstructions caused by the upregulation of phytoplankton CCMs24,33, which suggest using this Ɛp based proxy with caution in reconstructing low-CO2 worlds. In contrast, the proxy seems to do well in estimating pCO2 concentrations similar to some of the higher concentrations that have been reconstructed over the past 455 Myr39, suggesting it may be applicable for past greenhouse worlds.

## Conclusion

We tested three general phytoplankton biomarkers in surface sediments in a transect from a naturally occurring CO2 seep located in Levante bay, Vulcano Island, Italy, towards the open Tyrrhenian Sea. The δ13C of the biomarkers showed a distinct increase with increasing distance from the CO2 seep, in agreement with the idea that CO2 concentrations have a strong control on isotopic fractionation. In particular, the δ13C of phytol yielded a strong and consistent trend throughout the transect, and the agreement between estimated and measured CO2 concentrations demonstrates the promise of this biomarker for paleo pCO2 reconstructions. Our results show that CO2 seep environments may prove a useful testing ground for new CO2 proxies.

## Materials and methods

### Sample site

Levante Bay (Fig. 1) is located on the northeast of Vulcano Island, an Aeolian Island north of Sicily. Volcanic activity on the island started in the upper Pliocene71, where the cooling of magmatic and hydrothermal fluid mixing into the crater fumeroles is believed to have created the pocket of CO2, which outgasses into the bay72. Located at ca. 1 m depth at 38.41694° N 14.96° E, the main underwater venting gas field outputs ca. 3.6 tons of gas per day73. This gas is composed of 97–98% CO2 and ca. 2% H2S63. The sea water temperature63 of ca. 19.7 °C and salinity43 of ca. 38‰ is homogenous throughout the small bay. Currents are mostly wind-driven, with minimal tidal range (< 40 cm) and depths throughout the entire bay, and thus all sample sites, ranged between 1 and 2 m. Precipitation varies throughout the year, with the dry months (May–August) averaging 16 mm/month and the wet months (October–January) averaging 87 mm/month. The input of CO2 gas intensely influences the geochemical composition of the bay’s waters, as seen by the strong pH gradient starting at the seep to across the bay from pH 5.5 to 8.2 in April and from pH 6 to 8 in September. For more details on the geochemistry, see Boatta et al.63.

## Materials

Samples were collected in 23–24 May and 16–17 October of 2017. A preliminary study was conducted in May using one site with a high CO2 concentration, two sites with a middle CO2 concentration, and one control site (i.e. not affected by the CO2 seep) as defined in Johnson et al.43, where seawater was collected for the δ13C of dissolved inorganic carbon (DIC) and surface sediments were collected for the δ13C of biomarker lipids. Seawater for DIC analysis was collected by overfilling glass vials and adding mercury chloride (0.5%) before sealing the vials closed with Apiezon M grease and securing the stopper with rubber bands. Surface sediments were collected by diving, scooped into geochemical bags, and immediately frozen; once back in the lab, these sediments were freeze-dried and kept refrigerated. All surface sediments were collected in triplicate at each site within a square of 2 by 2 m. The same sediment sampling method was used again in October, when a higher-resolution transect of sixteen sites was collected (Fig. 1). Given that the results of the δ13C of DIC collected in May was homogenous throughout the bay (see Table S1), as also revealed by another study in this region48, seawater samples were not collected in October.

### Methods

The δ13C of DIC of seawater collected in May was measured on a gas bench coupled to an isotope ratio mass spectrometer (IRMS) in duplicate. Samples were prepared using 100 µL of 85% H3PO4 then flushed with He. Seawater (500 µL) was injected to each vial, left to react for 1 h, and then the headspace was measured. Standards prepared with 0.3 mg of Na2CO3 and 0.4 mg of Ca2CO3 were flushed with He, injected with 100 µL of 85% H3PO4, and reacted for 1 h. The standards were run at the start and end of each sequence, as well as every six runs.

Sediments were freeze-dried and homogenized using a mortar and pestle. Sediments were then extracted using a Dionex 250 accelerated solvent extractor at 7.6 × 106 Pa at 100 °C using dichloromethane (DCM): MeOH (9:1 v/v). Extracts were transferred to centrifuge tubes to be refluxed with 1 N KOH in MeOH and the resulting base hydrolyzed extracts were neutralized to pH 5 using 2 N HCl in MeOH. The hydrolyzed extract was separated into apolar (hexane: DCM, 9:1 v/v), ketone (DCM), and polar (DCM: MeOH, 1:1 v/v) fractions, respectively, over an alumina column. Polar fractions were silylated with pyridine: N,O-Bis(trimethylsilyl) trifluoroacetamide (1:1 v/v) and heated for 1 h at 60 °C. The δ13C values of loliolide, cholesterol, and phytol were corrected for the addition of three C atoms in the trimethylsilyl group using the known δ13C value of BSTFA (− 32.2‰).

Silylated polar fractions were then injected on gas chromatography-flame ionization detector (GC-FID) to determine relative abundances and general quality of chromatography before analyzing it on a gas chromatography–mass spectrometer (GC–MS) to identify compounds and on gas chromatography-isotope ratio-mass spectrometer (GC-IRMS) to measure the isotopic composition of specific compounds. GC-FID, GC–MS, and IRMS instrumentation all had starting oven temperatures of 70 ºC ramped at 20 ºC/min to 130 ºC and then ramped at 4 ºC/min to 320 ºC for 10 min. Separation was accomplished using a CP-Sil 5 column (25 m × 0.32 mm; df 0.12 μm) with He carrier gas. System performance on all three instruments was conducted daily using the same in-house mixture of n-alkanes and fatty acids. Additional standards were run on the IRMS using perdeuterated n-alkanes (C20 and C24) with known δ13C values (− 32.7 and − 27.0‰, respectively) and were limited to uncertainty within the standard of ± 0.5‰; if outside this range, the machine was conditioned until it was within this limit. The IRMS was also oxidized regularly, with a daily oxidation of 10 min, backflushed with He for 10 min, and purged for 5 min; a shorter version of this sequence was conducted in post-sample seed oxidation, which includes 2 min oxidation, 2 min He backflush, and 2 min purge conditioning line and a longer version of this sequence was conducted at the end of each week with 1 h oxidation, 1 h He backflush, and 10 min purge conditioning line.