Algal biomarkers as a proxy for pCO2: Constraints from late Quaternary sapropels in the eastern Mediterranean

Records of carbon dioxide concentrations (partial pressure expressed as pCO2) over Earth’s history provide trends that are critical to understand our changing world. To better constrain pCO2 estimations, here we test organic pCO2 proxies against the direct measurements of pCO2 recorded in ice cores. Based on the concept of stable carbon isotopic fractionation due to photosynthetic CO2 fixation (Ɛp), we use the stable carbon isotopic composition (δ13C) of the recently proposed biomarker phytol (from all photoautotrophs), as well as the conventionally used alkenone biomarkers (from specific species) for comparison, to reconstruct pCO2 over several Quaternary sapropel formation periods (S1, S3, S4, and S5) in the eastern Mediterranean Sea. The reconstructed pCO2 values are within error of the ice core values but consistently exceed the ice core values by ca. 100 μatm. This offset corresponds with atmospheric disequilibrium of present day CO2[aq] concentrations in the Mediterranean Sea from global pCO2, equivalent to ca. 100 μatm, although pCO2 estimates derived from individual horizons within each sapropel do not covary with the ice core values. This may possibly be due to greater variability in local CO2[aq] concentration changes in the Mediterranean, as compared with the global average pCO2, or possibly due to biases in the proxy, such as variable growth rate or carbon-concentrating mechanisms. Thus, the offset is likely a combination of physiological or environmental factors. Nevertheless, our results demonstrate that alkenoneand phytol-based pCO2 proxies yield statistically similar estimations (P-value = 0.02, Pearson’s r-value = 0.56), and yield reasonable absolute estimations although with relatively large uncertainties (± 100 μatm).


Introduction
The atmospheric partial pressure of CO 2 (pCO 2 , expressed in µatm) has a significant impact on Earth system dynamics, including its climate-influencing role as a greenhouse gas.
Assessing changes in pCO 2 over geological timescales may help us better understand current climate changes and predict the near future. Proxies, methods for quantitatively reconstructing past conditions, make it possible to look beyond the scope of direct measurements and provide secular trends in pCO 2 (Beerling and Royer, 2011). Although there are continual improvements of these pCO 2 proxies (Hollis et al., 2019), their accuracy remains largely uncertain and their incongruities increase with geologic time (Foster et al., 2017). Terrestrial proxies have been used to reconstruct pCO 2 over the past several hundred million years but often with unconstrained uncertainties, in part due to the limitations of the proxies and the effects of local carbon cycling that can occur in heterogenous terrestrial environments (e.g., Hollis et al., 2019). Marine proxies for pCO 2 reconstructions, on the other hand, tend to have more constrained homogenous signals and more continuous records but do not span as far back in time (Royer, 2013).
The marine-based pCO 2 proxy using the stable carbon isotopic fractionation associated with CO 2 fixation (Ɛ p ) has been developed over the past several decades. Ɛ p relies on the kinetic isotope fractionation that occurs as algae capture CO 2 in their environment for photosynthesis, where the CO 2 -fixing enzyme Rubisco more rapidly incorporates 12 C over 13 C into photoautotrophic biomass (Farquhar et al., 1982(Farquhar et al., , 1989Popp et al., 1989;Hayes et al., 1990). This fractionation results in a lower 13 C content (δ 13 C) of biomass than the inorganic carbon source that was fixed during photosynthesis. Increasing the availability of CO 2 has been shown to increase fractionation, and thus a positive relationship between pCO 2 and Ɛ p is generally observed (e.g., Popp et al., 1989;Jasper and Hayes, 1990;Freeman and Hayes, 1992).
Although sedimentary bulk organic matter can be used to reconstruct past δ 13 C values of the algal biomass component (e.g., Hayes et al., 1999), species-specific algal biomarkers have been the primary target for decades (e.g., Jasper et al., 1994;Pagani, 2002). In the latter case, the δ 13 C value of each biomarker is corrected for its offset from the δ 13 C value of biomass (δ p ): δ p is then used to reconstruct Ɛ p , together with the δ 13 C value of dissolved CO 2 in the photic zone (δ d ) derived from e.g., planktic foraminiferal carbonate and corrected for the carbon isotopic fractionation of CO 2(aq) with respect to HCO 3 -: The relationship between CO 2[aq] and Ɛ p is complex and several models have been applied. The most common and simplified equation is based on the theory first developed for higher plants (Farquhar et al., 1982(Farquhar et al., , 1989) and subsequently modified for marine algae (Popp et al., 1989;Jasper and Hayes, 1990;Jasper et al., 1994;Bidigare et al., 1997): In this equation, the apparent observed fractionation Ɛ p is subtracted from the maximum potential fractionation for CO 2 fixation (Ɛ f ) and related to CO 2 via the catch-all term b, a term considering fractionation factors other than CO 2 such as growth rate, cell geometry, membrane permeability to CO 2 , and the boundary layer thickness dependent on temperature, pH, and salinity (Rau et al., 1996;Laws et al., 1997;Popp et al., 1998;Bolton et al., 2016;Stoll et al., 2019). These combined parameters are equivalent to dissolved CO 2 conditions for the algae during its growth. Dissolved CO 2 may then be converted to atmospheric pCO 2 concentrations via the Henry's Law constant (K 0 ) using temperature and salinity (Weiss, 1974).
Most studies that reconstruct pCO 2 from Ɛ p have used long-chain alkenones (Jasper and Hayes, 1990;Pagani et al., 1999Pagani et al., , 2005Pagani, 2002;Zhang et al., 2013), biomarkers produced by a select group of haptophytes (Volkman et al., 1980). Due to the selectivity of the biomarkers, the difference between the δ 13 C values of biomarker and biomass can be relatively well-constrained based on laboratory cultures of the specific alkenone-producing species (Riebesell et al., 2000). However, due to the fairly recent evolutionary history of alkenone producers, pCO 2 reconstructions are largely limited to last 45 Myr (Brassell, 2014).
There are also some complicating factors with Ɛ p -based pCO 2 reconstructions, such as the evolutionary development of carbon concentrating mechanisms (CCMs) which actively pump bicarbonate in many marine phytoplankton (e.g., Laws et al., 1997), in contrast to the principle assumptions that Ɛ p is based on passive diffusion of CO 2[aq] (e.g., Bidigare et al., 1997). The role and influence of CCMs has continued to be explored over the past two decades (Rost et al., 2003;Bach et al., 2013;Bolton et al., 2016). Although some studies suggest that CCMs are weakly expressed in haptophytes (e.g., Reinfelder, 2011), the most recent studies contrarily suggest that CCMs may limit the use of this proxy during periods of low pCO 2 (Badger et al., Stoll et al., 2019), i.e. when aqueous CO 2 concentrations fall below ca. 7 μmol L −1 (Badger, 2020). This proxy is further complicated by the nature of the catch-all term b which may vary over space and time (Zhang et al., , 2020, making it difficult to constrain this parameter, and consequently pCO 2 reconstructions, over long timescales. The Ɛ p -based pCO 2 proxy has recently been reevaluated for general phytoplankton biomarkers, compounds derived from a multitude of marine algal species (Witkowski et al., 2018(Witkowski et al., , 2019(Witkowski et al., , 2020. There has been minimal proxy development research on Ɛ p from general algal biomarkers, with the exception of some paleo-pCO 2 applications of chlorophyll a products, including its porphyrin core (Popp et al., 1989;Freeman and Hayes, 1992) and the diagenetic product of its phytol side-chain phytane (Bice et al., 2006;Sinninghe Damsté et al., 2008;van Bentum et al., 2012;Naafs et al., 2016), which should have the same δ 13 C value as phytol given that there are no additions or loss of C during diagenetic transformation. Because chlorophyll a is the vital light harvesting pigment in all photoautotrophs, it includes eukaryotic algae, cyanobacteria, and plants in both marine and terrestrial environments, offering greater spatial and temporal ubiquity throughout the geologic record as compared with its species-specific counterparts, i.e. alkenones limited to ca. 45 Ma (Brassell, 2014). At the same time, chlorophyll a and its products offer more specificity than bulk organic matter; bulk organic matter raises concerns of isotopic heterogeneity with different organisms contributing different types of preserved organic matter e.g., carbohydrates, proteins, and lipids, each with distinct δ 13 C values (Hayes, 1993) and distinct diagenetic changes to those δ 13 C values e.g., via carbohydrate sulfurization . Given that chlorophyll a rapidly breaks down and is not prone to lateral transport, phytol and its diagenetic products are likely deposited and buried close to their source organism. In our recent reevaluation of phytane over the Phanerozoic, we thus focused our efforts on open marine settings with minimal terrestrial input in order to limit the source of organisms to primarily algae (Witkowski et al., 2018). Phytane shows similar estimates to other pCO 2 proxies over the Phanerozoic and offers the longest marine-based pCO 2 record currently available (Witkowski et al., 2018). However, although some modern studies across naturally occurring high CO 2 gradients have recently been conducted (Witkowski et al., 2019(Witkowski et al., , 2020, the accuracy of this proxy has not been tested on shorter geologic timescales with smaller variability in pCO 2 nor compared with the more commonly applied alkenone-based pCO 2 proxy. The GC-irMS Isolink II combustion reactor was daily oxidized for 15 min, He backflushed for 10 min, and purged for 5 min. Every analysis ended with a 2 min post-sample seed oxidation and, in addition, 1 h oxidation sequences were run once week. All samples were run in at least duplicate. Silylated phytol was corrected for the three additional C atoms in the trimethylsilyl group by using the pre-determined δ 13 C value of BSTFA (-32.2‰).
The δ 13 C value of carbonate in the surface-dwelling planktic foraminifera Globigerinoides ruber was measured at 1 cm resolution over the same core as the organic compounds.
The mean external reproducibility was less than ± 0.05‰.

Estimating pCO 2 from the δ 13 C values of phytol and alkenones
Several studies on eastern Mediterranean sapropel organic matter have shown a dominant input of mainly marine algal biomarkers, in particular the long-chain unsaturated ketones, alkanediols, loliolide, and sterols (ten Haven et al., 1987bHaven et al., , 1987aBouloubassi et al., 1999), supporting a phytoplankton origin for phytol. Indeed, we observe similar biomarkers in our sediments and minimal terrestrial input (e.g., odd-over-even long-chain n-alkanes and triterpenoids). The precise contributions of different species to the phytoplankton pool are difficult to define, but evidence of calcareous nannoplankton, diatoms, and dinoflagellate cysts are common during sapropel periods (e.g., Giunta et al., 2006).
Ɛ p was calculated from the δ 13 C value of organic matter (δ p ) and the δ 13 C value of dissolved CO 2 in the photic zone (δ d ) using Eq. 1. δ p was calculated from the δ 13 C value of each biomarker corrected for its offset from the δ 13 C value of biomass. For phytol we used an average offset of 3  (Jasper and Hayes, 1990;Schouten et al., 1998;Riebesell et al., 2000;Laws et al., 2001;van Dongen et al., 2002), with many of these studies conducted under the same conditions for both phytol and alkenones, and some even from the same organism. The δ d is derived from the high-resolution record of δ 13 C values of the surfacedwelling planktic foraminifera Globigerinoides ruber from the same core (Supplementary   Tables S1 and S2). δ d was then corrected for temperature-dependent carbon isotopic fractionation of dissolved CO 2 with respect to HCO 3using the equation from Mook et al.
(1974) and Weiss (1974): For temperature (expressed in K), we calculated sea surface temperature (SST) from based on the alkenones reported here and using the global core top calibration (Müller et al., 1998). The -based SSTs ranged from 17.7 °C to 23.5 °C (Table 1), in agreement with those previously reported for sapropels (Emeis et al., 2003).
Ɛ p values derived from phytol ranges from 9.0‰ to 15.0‰ over all sapropels, ranging from 13.4‰ to 13.7‰ in S1, 9.0‰ to 11.5‰ in S3, 10.5‰ to 15.0‰ in S4, and 9.8‰ to 11.1‰ in S5 (Table 1). Ɛ p values derived from alkenones ranges from 10.4‰ to 13.2‰ over all sapropels, ranging from 11.3‰ to 12.4‰ in S1, 10.9‰ to 11.3‰ in S3, 12.2‰ to 13.2‰ in S4, and 10.4‰ to 11.6‰ in S5 (Table 1). The amalgamation of error propagation was calculated using Monte Carlo simulations in which each individual parameter with its associated uncertainty was included, as described by Witkowski et al. (2018), and expressed as 1 s.d. (68%; Supplementary Tables S1 and S2). Parameter uncertainties included the δ 13 C value of the biomarkers (0.5‰), the δ 13 C value of the carbonates (0.1‰), SST (2 °C), and the offset between the δ 13 C value of biomass from each biomarker (1.3‰ for phytol; 0.9‰ for alkenones), culminating to an uncertainty in Ɛ p values of ca. ± 1.4‰ for both phytol and alkenones. When compared within the same sediments and thus time periods, there is a striking similarity between these two proxies (Fig. 1).
Individual Ɛ p values calculated from the δ 13 C value of phytol (derived from the whole phytoplankton community) and those calculated from the δ 13 C value of alkenones (derived from species-specific producers) yield statistically similar values (P-value = 0.005, Pearson's r-value = 0.645). This suggests that isotopic fractionation is similar between haptophyte algae and other phytoplankton, or possibly that they represent a similar source, i.e. that haptophyte algae dominate the overall phytoplankton pool. There are several data points which lay just outside the one-to-one line between Ɛ p values derived from alkenones vs phytol, all from the onset of sapropels S1 and S4. At the onset, changes in sea surface salinity and nutrient input associated with a large freshwater input from the African continent, including the Nile (Lourens et al., 1996;Rohling and De Rijk, 1999), likely influenced the overall phytoplankton community, in which the phytol-producing species (the overall photoautotrophic community) may have differed from alkenone-producing species in average cell size or growth rates.
To reconstruct pCO 2 from Ɛ p , Eq. 2 was used. A b value of 170 ± 43‰ kg µM -1 has been used for phytol based on a compilation of 18 studies of the δ 13 C values of modern surface sediment organic matter (see Witkowski et al., 2018) is calculated from temperature and salinity (Weiss, 1974), in which SST is derived from the alkenone-based U K′ 37 temperature proxy measured in the same sapropel layer (

Comparison of reconstructed pCO 2 with ice core data
Past global atmospheric pCO 2 recorded in ice core gas bubbles (Petit et al., 1999;Pépin et al., 2001) have values in glacial-inception pCO 2 of 226 µatm at ca. 84 ka (same timing as S3) and 234 µatm at ca. 107 ka (same timing as S4) and have values in the interglacial period of pCO 2 of 265 µatm at ca. 10 ka (same timing as S1) and 271 µatm at ca.
124 ka (same timing as S5). Our individual proxy estimations are just within error of this ice core data (Fig. 2). However, individual pCO 2 estimations based on alkenone-and phytol- been shown to be a mutable variable , any major changes to the b value during sapropel deposition or among the four different sapropels may explain the lack of correlation with the ice core data.
CCMs in phytoplankton could also affect the mechanisms of the proxy, especially given that this is a period of low pCO 2 . In order to supplement CO 2 under insufficient levels of pCO 2 , many phytoplankton have been shown to have developed CCMs which actively pump HCO 3 − at the active site of Rubisco (e.g., Raven and Beardall, 2014;Kottmeier et al., 2016). This differs from the diffusive model used here (Eqs. 1 and 2), which is based on the assumption that dissolved CO 2[aq] (only) passively enters the algal cell, a concept observed in laboratory cultures where CO 2 availability is high relative to cellular carbon demand (Francois et al., 1993;Rau et al., 1996). Active uptake is a concern given the substantial δ 13 C difference between bicarbonate (0‰) and CO 2 (-8‰) (Mook et al., 1974) and because they can decouple the amount of intracellular CO 2 from outside the cell. Results of a statistical multilinear regression model, that quantitatively considers factors influencing Ɛ p values in cultures of alkenone-producing algae, suggests that there is lower sensitivity of Ɛ p to pCO 2 than proposed by the diffusive model (Stoll et al., 2019). CCMs have been invoked to explain the muted response of pCO 2 reconstructed from the δ 13 C values of alkenones as compared with ice core pCO 2 data, as well as pCO 2 reconstructed from the δ 11 B of foraminifer shells (Badger et al., 2019) when aqueous carbon dioxide concentrations fall below 7 μmolL −1 (Badger, 2020).
Finally, it has been proposed that stable carbon isotopic fractionation is impacted by a ratelimiting step upstream of Rubisco under excess photon flux, rather than fractionation of Rubisco, thus changing the sensitivity of fractionation to CO 2 changes (Wilkes and Pearson, 2019).
Apart from issues that may arise from the proxy itself, local variability may be a possible explanation for this difference between the individual proxy estimates and the individual ice core data. In other words, the proxies may reflect changes in their local environment. Dissolved CO 2 concentrations are more likely to vary locally over time, especially in a semi-enclosed Mediterranean Sea, as compared with the more homogeneous atmospheric pCO 2 . This perhaps explains the high variability within S4 (Fig. 2), where the standard deviation for the individual pCO 2 estimations are ca. 57 µatm. When this S4 data is removed from the overall dataset, a notably improved correlation between the biomarker pCO 2 reconstructions with the ice core data can be seen (phytol: P-value = 0.075, Pearson's rvalue = 0.509; alkenones: P-value = 0.028, Pearson's r-value = 0.606). These local offsets may be caused by the many influences on CO 2[aq] in the Mediterranean Sea, such as the cyclic influence of freshwater input from the Nile that may change alkalinity, temperature, nutrient availability, and other seawater components. Local changes could affect the δ 13 C values of the CO 2 via periodic deep-water convection (Melki et al., 2010), causing the mixing of 13 Cdepleted CO 2 from below the chemocline in the otherwise-stratified Mediterranean water column during sapropel formation (Küspert, 1982). This effect on the δ 13 C values of CO 2 used by phytoplankton is, however, not observed in the planktic foraminifera signal as it remains fairly constant (Supplementary Tables S1 and S2).
With these possible issues in mind, there is likely some combination of factors to explain why individual data points differ between the ice core data and the estimated organic proxy calculations. When data of the individual layers are combined per sapropel to obtain a clear view on general trends, a consistent offset of the Ɛ p -based pCO 2 estimations and the pCO 2 from ice core data by ca. 100 µatm is observed for all sapropels (Fig. 2) which, as discussed above, behaves differently compared to other sapropels. Hence, if this offset is taken into consideration, both phytol and alkenone proxies based on Ɛ p seem to yield reasonable pCO 2 estimations in the late Pleistocene to Holocene.

Conclusions
The δ 13 C values of a potential biomarker for pCO 2 , phytol, as well as the δ 13 C values of its established biomarker counterpart, alkenones, were used to calculate photosynthetic isotopic fractionation (Ɛ p ) and estimate pCO 2 from Quaternary Mediterranean Sea sapropels. Phytoland alkenone-based pCO 2 values yielded similar estimations, i.e. 300 µatm to 450 µatm for phytol and ca. 330 µatm to 390 µatm for alkenones. These values overestimate global atmospheric pCO 2 by ca. 100 µatm, which corresponds with the enhanced dissolved CO 2 concentrations in the Mediterranean Sea due to its high alkalinity. Given this disequilibrium consideration, the Ɛ p proxy for reconstructing pCO 2 seems to reflect CO 2 concentrations during Quaternary sapropel formation in the Mediterranean. Although these results are favorable, there is a lack of correlation between changes in the individual reconstructed pCO 2 values from the two biomarkers and individual pCO 2 values from ice core data, most notably in S4. Importantly, the ranges for the phytol-and alkenone-based pCO 2 estimates are much larger than that observed in the ice core pCO 2 values, which largely explains this lack of covariation. This larger variability in range for the proxies may be due to higher local variability in the semi-enclosed Mediterranean, e.g., influencing dissolved CO 2 and the b factor, as well as potential influences from carbon concentrating mechanisms.