4.1. Climatic and Environmental Changes Across ETM2 in the Arctic Ocean
 Prior to the onset of ETM2, both the biomarker concentrations and δ13C values show only minor variations, suggesting that environmental conditions were relatively stable (Figures 3a and 3b). Furthermore, no isorenieratane is detected in these sediments implying that the water column was not euxinic within the photic zone. This is also supported by relatively stable assemblages of typical open marine dinoflagellate cysts in this interval [Sluijs et al., 2009]. Relatively high TOC concentrations and the presence of “sulfur-bound” organic molecules in these sediments points toward a relatively productive paleoenvironment and low bottom water oxygen concentrations, which is in agreement with previous observations [Sluijs et al., 2008; Stein et al., 2006].
 The synchronous drop in δ13C at ∼368.9 mcd of both the specific biomarkers and TOC confirms that the CIE in TOC is not caused by changes in the composition of the bulk organic matter, but is linked to the injection of 13C-depleted carbon into the global exogenic carbon pool. During the recovery of ETM2, the δ13C of the biomarkers no longer track the δ13CTOC profile. The δ13CTOC record shows a smooth return to background δ13C values from 368.8 mcd, while the δ13C profiles of the biomarkers abruptly move toward more positive values at 368.6 mcd (Figure 3b). Additionally, directly after the CIE, at ∼368.7 mcd, there is a sharp increase in concentrations of phytane, C25 HBI and C35 hopane, which is followed by the development of photic zone euxinia (PZE) as indicated by the presence of isorenieratene derivatives. Possibly, enhanced productivity contributed to the development of PZE conditions in this interval, as was suggested for ETM2 and the PETM in the Arctic Ocean [Sluijs et al., 2006, 2009; Stein et al., 2006]. The increase in biomarker concentrations may also be explained by an increase in export production. However, Knies et al.  investigated the response of marine productivity to variations in nutrient supply to the Cenozoic Arctic Ocean using nitrogen isotopes. For ETM2 they found evidence for an increase in primary production rates even after correcting for the higher burial efficiency caused by the euxinic conditions. Furthermore, abundances of the freshwater tolerant dinoflagellate species that peak synchronously with isorenieratane concentrations are also regarded as indicators for nutrient-rich conditions [Sluijs et al., 2005, 2009]. Although the timing with these dinoflagellate peaks is not perfectly synchronous, an increase in primary productivity could explain the increase in biomarker concentrations and the positive isotope shift in the specific biomarkers at 368.6 mcd, as an increase in primary productivity can lead to increased growth rates and decreased isotopic fractionation [Jasper and Hayes, 1990; Laws et al., 1995; Bidigare et al., 1997; Popp et al., 1998a, 1998b]. Therefore, although the lipids obviously had to be exported from the surface ocean to settle on the seafloor, the increase in concentrations in our view was at least partially related to increased productivity. This would imply that during this interval regional effects control the biomarker records, whereas TOC in this case more accurately tracks the δ13C evolution of the exogenic carbon pool.
 At ∼368.48 mcd, biomarker concentrations decrease and isorenieratane is below detection limit (Figure 3a), suggesting that the PZE conditions ended. Subsequently, all biomarker isotope values return to ‘background’ pre-ETM2 values and continue to track the δ13CTOC signal (Figure 3b).
4.3. Estimating Isotopic Fractionation Across ETM2
 Based on the measured stable isotopic composition of S-bound phytane, C25HBI and C35 hopane, we estimated the average carbon isotopic fractionation of photoautotrophs, and changes therein. Averaged δ13C values were calculated for three time intervals: the pre-ETM2 interval (369.60–368.94 mcd), the CIE of ETM2 (368.84–368.72 mcd), and the post-ETM2 interval (368.20–368.04 mcd). Average biomarker δ13C values for these three periods were used to estimate the isotopic fractionation (ɛp):
where δp is the δ13C value of the total organic carbon of the organism and δd is the δ13C value of the carbon substrate. To obtain δp, a correction must be made for the isotopic offset between the biomarker lipid and cell biomass. Schouten et al.  and Oakes et al.  reported, based on culture experiments and literature study of a range of different algae, that phytol is ∼6‰ depleted relative to the total algal biomass. For C25 HBIs a depletion of 6.6‰ relative to biomass was reported by Schouten et al.  for the diatom Rhizosolenia setigera, whereas Massé et al.  found similar carbon isotopic compositions of the C25 HBIs and phytol in Haslea ostrearia. This suggests an isotopic offset of ca. 6‰ for both phytane and C25 HBI. Results from culture experiments of the cyanobacterium species Synechocystis revealed an isotopic offset of 8.4‰ for bishomohopanol [Sakata et al., 1997].
 Values for δd can be obtained from the carbonate shells of planktonic foraminifera using the following equation:
The δ13C of planktonic foraminifera (δ13Cpf) represents the δ13C composition of the primary carbon in CaCO3. The term between brackets describes the isotopic effect associated with the equilibrium exchange reaction between CO2aq and HCO3− as reported by Mook et al. , which only depends on temperature (T in degrees Kelvin).
 Unfortunately, foraminiferal carbonate is absent in ACEX sediments [Sluijs et al., 2008, 2009]. Instead, we used the δ13C values of the surface-dwelling genus Acarinina reported for ETM2 at Walvis Ridge [Lourens et al., 2005; Stap et al., 2010]. Although this induces one factor of uncertainty, the δ13Cpf of ∼2‰ for the period prior to ETM2 compares quite well with those of stacked carbonate isotope records, as well as with modeling studies for the Early Eocene [Hayes, 1999; Berner, 2006; Berner and Kothavala, 2001], suggesting that this assumption is reasonable. We do not believe that in the semi-enclosed Arctic Basin additional effects, such as the input of recycled CO2 from anoxic deep waters play a large role. Van Breugel et al.  demonstrated that in an anoxic marine system the effect of recycling of respired CO2 on the δ13C of phytoplankton lipids is negligible. Sea surface temperatures (SSTs) used in equation (2) were obtained from the oxygen isotopes of the same foraminiferal records from Walvis Ridge and a TEX86 measurement during ETM2 [Stap et al., 2010].
 The calculated ɛp values for the preexcursion interval show remarkably high carbon isotope fractionation factors of ca. 21–22‰ and ca. 14.5‰ for marine algae and (cyano)bacteria, respectively (Table 3). The lower value determined for (cyano)bacteria is consistent with the smaller carbon isotopic fractionation by cyanobacteria in comparison to algae [Hayes, 2001; Popp et al., 1998b]. During ETM2, ɛp values increase even further by 1–2‰. For all three biomarkers this results in ɛp values that lie close to the maximum fractionation of photoautotrophic organisms, i.e., 25–28‰ for the Rubisco enzyme of autotrophic eukaryotes [Bidigare et al., 1997; Goericke et al., 1994; Popp et al., 1998b] and 16–22‰ for autotrophic cyanobacteria [Sakata et al., 1997, and references therein].
Table 3. Estimations for pCO2 Based on the δ13C Composition of S-Bound Phytane, C25 HBI, and C35 Hopane
|Biomarker||δ13Ca (‰)||Δδb (‰)||δpc (‰)||δ13Cpfd (‰)||SST-WRe (°C)||SST-AOf (°C)||δdg (‰)||ɛph (‰)||ɛfi (‰)||K0j (mol L−1 atm−1)||pCO2k (ppmv)|
|b = 160||b = 200||b = 240|
|Preexcursion Interval (369.60–368.94 mcd)|
|S-bound phytane||−33.9 ± 0.4||6||−27.9||2||18.5||19||−8.7||19.7||25||0.03431||900||1100||1300|
|S-bound phytane||−33.9 ± 0.4||6||−27.9||2||18.5||19||−8.7||19.7||27||0.03431||650||800||950|
|S-bound HBI||−33.3 ± 0.3||6||−27.3||2||18.5||19||−8.7||19.1||25||0.03431||800||1000||1200|
|S-bound HBI||−33.3 ± 0.3||6||−27.3||2||18.5||19||−8.7||19.1||27||0.03431||600||750||900|
|S-bound C35 hopane||−31.5 ± 0.2||8.4||−23.1||2||18.5||19||−8.7||14.7||20||0.03431||900||1100||1350|
|Excursion Interval (368.84–368-72 mcd)|
|S-bound phytane||−36.7 ± 0.4||6||−30.7||0||21.5||23||−10.4||21.0||25||0.0307||1300||1650||1950|
|S-bound phytane||−36.7 ± 0.4||6||−30.7||0||21.5||23||−10.4||21.0||27||0.0307||850||1100||1300|
|S-bound HBI||−36.7 ± 0.2||6||−30.7||0||21.5||23||−10.4||20.9||25||0.0307||1300||1600||1900|
|S-bound HBI||−36.7 ± 0.2||6||−30.7||0||21.5||23||−10.4||20.9||27||0.0307||850||1100||1300|
|S-bound C35 hopane||−35.1 ± 0.2||8.4||−26.7||0||21.5||23||−10.4||16.8||20||0.0307||1600||2000||2400|
|Postexcursion Interval (368.20–368.04 mcd)|
|S-bound phytane||−33.6 ± 0.2||6||−27.6||1.5||17.5||18||−9.3||18.8||25||0.0353||750||900||1100|
|S-bound phytane||−33.6 ± 0.2||6||−27.6||1.5||17.5||18||−9.3||18.8||27||0.0353||550||700||850|
|S-bound HBI||−33.1 ± 0.2||6||−27.1||1.5||17.5||18||−9.3||18.2||25||0.0353||650||850||1000|
|S-bound HBI||−33.1 ± 0.2||6||−27.1||1.5||17.5||18||−9.3||18.2||27||0.0353||500||650||800|
|S-bound C35 hopane||−32.5 ± 0.3||8.4||−24.1||1.5||17.5||18||−9.3||15.1||20||0.0353||900||1150||1400|
 The magnitude of ɛp is mainly determined by the carbon fixation enzyme and carbonate concentration mechanism, which in turn can be affected by factors such as the amount of available CO2 in the water column ([CO2aq]), growth rate, light intensity, and species-specific factors such as cell geometry [e.g., Jasper and Hayes, 1990; Laws et al., 1995; Popp et al., 1998a, 1998b; Cassar et al., 2006]. Thus, in principle, the observed increase in biomarker ɛp values during ETM2 can be caused by increased levels of [CO2aq], but there are several additional factors which may be potentially responsible for this. The most important ones are a decrease in specific growth rates, a change in cell geometry, a change in light intensity, and the carbon uptake mechanism [Bidigare et al., 1997; Laws et al., 1995; Popp et al., 1998a, 1998b; Rau et al., 1996; Burkhardt et al., 1999]. It is unlikely that cell geometry has changed on this relative short time interval of the ETM2 for both the marine algae and (cyano)bacteria. Moreover, we also use S-bound phytane which is a biomarker not specific for only one group of organisms, but is contributed by many different species of marine algae and cyanobacteria. Furthermore, all available information indicates that productivity increased rather than decreased during ETM2 (see above), which theoretically should lead to a decrease of ɛp values. To avoid the imprint of growth rate changes on fractionation, we only used δ13C values before, and directly after the interval where several lines of evidence, including elevated biomarker concentrations, indicated elevated productivity (see section 4.1).
 Another important aspect to consider is the carbon uptake mechanism used by autotrophs during photosynthesis. Many photosynthetic organisms have evolved mechanisms to actively take up CO2 or HCO3− (a so-called carbon concentrating mechanism or CCM) in order to overcome the deficiency of the enzyme Rubisco in low-CO2/high-alkaline environments and this mechanism will impact a reduced isotopic fractionation [Giordano et al., 2005]. In our case, however, the time of ETM2 most likely belonged to a high-CO2/low-pH world, considering the large input of 13C-depleted carbon, making it unlikely that they need a CCM. Furthermore, isotopic modeling which incorporates active transport shows that ɛp is still a function of growth rate and CO2 under nutrient limitation (though this function is different under light limitation [Cassar et al., 2006]). Finally, the very negative biomarker δ13C values suggest that the organisms that made the lipids likely did not use a CCM, which has also been previously suggested for diatoms that biosynthesize HBI isomers [Schouten et al., 2000].
 Growth experiments of aquatic algae indicate that light-limitation may also have a potential effect on isotopic fractionation [Burkhardt et al., 1999; Cassar et al., 2006]. However, at this latitude it is likely that phytoplankton thrived only during summer in full light conditions, particularly with the absence of ice at this time. The only change in light conditions could appear when the water column is more stratified and fresher during ETM2, resulting in increasing light intensity and an increase in the magnitude of isotopic fractionation. However, the time of highest stratification, i.e., when isorenieratene derivatives are present, is some time after the CIE. In contrast, this interval is marked by slightly enriched 13C values for the different biomarkers. This suggests that light limitation cannot explain the isotopic fractionation patterns we observe. We, therefore, mostly attribute the increase in ɛp to a substantial increase in seawater CO2 concentration ([CO2aq]), in turn caused by elevated atmospheric pCO2 levels during ETM2.
4.4. A First Attempt to Estimate pCO2 for ETM2 Using Carbon Isotopic Fractionation Factors
 For alkenone-producing haptophytes the relationship between [CO2aq] and ɛp is relatively well constrained [Pagani et al., 2002, and references therein]. Therefore, stable carbon isotopic fractionation records using long-chain alkenones are frequently used for pCO2 reconstructions [e.g., Andersen et al., 1999; Benthien et al., 1999; Pagani et al., 1999, 2002, 2005; Pagani, 2002; Bijl et al., 2010; Palmer et al., 2010]. However, Popp et al. [1998b] also reported a relation between [CO2aq], growth rate and cell dimension for certain diatoms and cyanobacteria, although again other factors such as light intensity may play a role as well [Burkhardt et al., 1999; Cassar et al., 2006]. This would imply that the carbon isotope composition of specific marine algal biomarkers, other than alkenones, may also be applicable for pCO2 reconstructions. Indeed, ancient pCO2 levels were determined by Freeman and Hayes  using the carbon isotopic fractionations of sedimentary porphyrins [Popp et al., 1989]. Furthermore, variations in the offset between carbonate and organic matter isotopic composition have been applied as paleo-pCO2 proxy to reconstruct the expected drawdown in atmospheric CO2 during the late Cenomanian oceanic anoxic event [Jarvis et al., 2011]. Their trend in isotopic fractionation is remarkably consistent with previously estimated Cretaceous pCO2 values using the δ13C values of the specific marine biomarkers (S-bound) phytane and C35 hopane [Bice et al., 2006; Sinninghe Damsté et al., 2008].
 Here we follow the approach of Freeman and Hayes , Bice et al. , and Sinninghe Damsté et al.  to provide estimates of pCO2 during the early Eocene ETM2 interval. Large uncertainties and assumptions which are associated with this approach will be discussed below. Our goal here is merely to present estimates of the atmospheric CO2 concentrations and changes therein, which potentially can give insight in the changes of pCO2 levels across an Eocene hyperthermal, and provide a method which can be used at other environmental settings where similar isotopic biomarker records can be obtained.
4.4.1. Calculation of pCO2 Estimates
 In order to reconstruct the atmospheric CO2 concentrations across ETM2 from carbon isotopic fractionation factors, we assume that the relationship between ɛp and [CO2]aq, based on the calibration of δ13C composition of alkenones, is also applicable for δ13C values of other biomarkers produced by photoautotrophic organisms, in this case S-bound phytane, C25 HBI and C35 hopane. If so, then the degree of isotopic fractionation (ɛp) in a cell can in theory be related to CO2 concentrations using the following equation [Bidigare et al., 1997]:
where b is the sum of species-specific factors and reflects the carbon demand of the cell. Atmospheric pCO2 concentrations can then be estimated from the [CO2(aq)] values using Henry's law.
 For haptophyte algae it has been shown that b displays a strong positive correlation with phosphate concentrations [Andersen et al., 1999; Benthien et al., 2002; Bidigare et al., 1997; Pagani et al., 2002], and thus, if phosphate concentrations were known then b, and thereby [CO2(aq)], could be estimated. A similar relation, with different b values, is observed for other algae [Popp et al., 1998b] and we assume here that b values of these algae also depend on nutrients such as phosphate. However, it is difficult to predict the PO4 concentrations of Arctic surface waters, especially considering the stratified conditions during ETM2. Andersen et al.  reported an inverse relationship between the bulk nitrogen isotopes and phosphate concentrations in equatorial and south Atlantic core top sediments. They used this relationship to reconstruct b and in turn the pCO2 levels using their calibration of sedimentary δ13C alkenones. As an approach to constrain the b value for equation (3), we applied this relationship to the Early Eocene Arctic Ocean by using the nitrogen isotope values measured by Knies et al. , leading to average phosphate concentrations of 1.25 μmol/L prior to ETM2 to 1.5 μmol/L just at the onset of ETM2. Depending on the calibration, this leads to a b value ranging between 160 to 240. The pCO2 estimates obtained using the approach outlined above are illustrated in Figure 4. Here we plotted the ɛp-CO2 relationship of the algal biomarkers for the three time intervals at an intermediate b value of 200. The error bars include uncertainties in SST (±1°C) and δ13Cpf (±0.5‰), in addition to the analytical errors. To illustrate the importance of b, we varied this parameter over a range of 160–240 (Table 3 and Figure 4). One has to bear in mind, though, that our pCO2 estimates are based on the ɛp-[CO2aq] relationship originally calibrated for δ13C alkenones [Pagani et al., 2002, and references therein]. In addition, we assume that the δ13Cpf from Walvis Ridge is a representative of that in the Arctic Ocean during the ETM2. The propagated uncertainty stemming from these assumptions is difficult to quantify and is further discussed below.
Figure 4. Estimations of the atmospheric CO2 concentrations for the pre-ETM2 interval (green symbols), CIE of ETM2 (red symbols), and the post-ETM2 interval (blue symbols) using (a) the average δ13C values of C35 hopane (diamonds) using a maximum fractionation level (ɛf) of 20‰, (b) the average δ13C values of phytane (circles) and C25 HBI (triangles) with an ɛf of 25‰, and (c) the average δ13C values of phytane (circles) and C25 HBI (triangles) using an ɛf value of 27‰. Error bars include variations in SST and δ13Cpf of 1°C and 0.5‰, respectively, in addition to analytical errors. The gray shaded areas give the range of b (160–240) with b = 200 as intermediate value. Note that the uncertainty of the pCO2 estimates increases with higher ɛp values. Minimum pCO2 values using this approach are 590, 860, and 520 ppmv for the pre-ETM interval, the CIE of ETM2, and the post-ETM2 interval, respectively. This is at least two to three times preindustrial pCO2 levels (blue dotted line).
Download figure to PowerPoint
 For all three periods, the estimated pCO2 values are practically similar using three independent biomarkers and all suggest that pCO2 values were at least 2× preindustrial values, i.e., the minimum pCO2 estimates (considering all the uncertainties). Furthermore, when using the intermediate b value, the estimated pCO2 values are 800 to 1100 ppmv (3 to 4 times preindustrial values) for the preexcursion interval and 1100 to 2000 ppmv (4 to 7 times preindustrial values) for the CIE of ETM2 (see Table 3). Thus, pCO2 levels during ETM2 may have been 300 to 800 ppmv higher than prior to the ETM2.
4.4.2. Uncertainties, Caveats, and Future Outlook
 Clearly, our estimated pCO2 values are all associated with large uncertainties as indicated by the large error bars in Figure 4, and we caution that they should not be taken at face value. As mentioned before these pCO2 estimates are relying on a number of assumptions: (1) the δ13C composition of the DIC (δ13Cpf in equation (2)) of the Arctic Ocean surface waters equals the surface water δ13C of DIC of the subtropic SE Atlantic Ocean at Walvis Ridge during the Early Eocene; (2) the relationship between ɛp and [CO2]aq, based on the calibration of δ13C composition of alkenones, is also applicable for δ13C values of other biomarkers produced by photoautotrophic organisms, in this case S-bound phytane, C25 HBI and C35 hopane; and (3) the b value of photoautotrophs other than haptophyte algae are also related to nitrogen isotopic compositions. Since these assumptions have not yet been tested, it is not possible to estimate potential errors they introduce in the pCO2 estimates, but clearly they will have a large impact. Furthermore, there are a number of uncertainties associated with estimations of the isotopic fractionation factors as discussed in section 4.3. For example, an uncertainty in δ13Cpf may arise due to diagenesis and vital effects, and may be in the order of 0.5‰. An uncertainty of that magnitude will result in an equal uncertainty of 0.5‰ in ɛp. In turn, this will result in a significant error of the pCO2 estimations, which will be higher with higher ɛp values. The error caused by uncertainties in SST estimates is twofold. An increase of 1°C causes ɛp to increase with ∼0.12‰. The second uncertainty is in the sensitivity of the Arctic SST on the pCO2 estimates as the solubility of CO2 is higher under lower seawater temperatures. In comparison with uncertainties in δ13Cpf, an uncertainty in SST does not result in a large error in the pCO2 estimates (10–100 ppmv per 1°C) and depends on the amount of [CO2]aq. Thus, uncertainties in SST cause a relatively minor effect on our pCO2 estimates. Nevertheless, our ‘background’-ETM2 pCO2 estimates are in agreement with other estimates using proxy data [Demicco et al., 2003; Lowenstein and Demicco, 2006] and modeling [Berner and Kothavala, 2001; Pagani et al., 2006a; Zeebe et al., 2009] for the early/middle Eocene.
 Clearly, further research constraining the viability of this approach is needed. Especially, a good calibration between biomarker δ13C, ɛp and pCO2 based on modern microorganisms other than haptophytes, is essential to gain better insight in the factors that influence isotopic fractionation as discussed in the previous section. These calibrations are needed to test the assumptions that are at the base of our pCO2 reconstructions. Another way to test the reliability of our fractionation model is to compare estimated pCO2 using existing δ13C records of organic biomarkers with better-constrained pCO2 conditions during past intervals, such as the last glacial cycles. As a first step, we used the δ13C of biomarkers of C25 HBIs in Holocene sediments of the Arabian Sea [Schouten et al., 2000] to estimate preindustrial pCO2 levels using our method. We arrive at values between 250 to 300 ppmv, which compares favorably well with preindustrial pCO2 values (Table 4).
Table 4. The pCO2 Estimates Inferred From Diatom Biomarkers From Holocene Arabian Sea Sediments Sampled at Different Sitesa
|Site||SST (°C)||δ13C (‰)||δd (‰)||Biomarker||δ13C (‰)||ɛp (‰)||b (kg μmol−1 L)||[CO2aq] (μmol kg−1)||pCO2 (ppmv)|