Royal Society Publishing

A 40-million-year history of atmospheric CO2

Yi Ge Zhang , Mark Pagani , Zhonghui Liu , Steven M. Bohaty , Robert DeConto

Abstract

The alkenone–pCO2 methodology has been used to reconstruct the partial pressure of ancient atmospheric carbon dioxide (pCO2) for the past 45 million years of Earth's history (Middle Eocene to Pleistocene epochs). The present long-term CO2 record is a composite of data from multiple ocean localities that express a wide range of oceanographic and algal growth conditions that potentially bias CO2 results. In this study, we present a pCO2 record spanning the past 40 million years from a single marine locality, Ocean Drilling Program Site 925 located in the western equatorial Atlantic Ocean. The trends and absolute values of our new CO2 record site are broadly consistent with previously published multi-site alkenone–CO2 results. However, new pCO2 estimates for the Middle Miocene are notably higher than published records, with average pCO2 concentrations in the range of 400–500 ppm. Our results are generally consistent with recent pCO2 estimates based on boron isotope-pH data and stomatal index records, and suggest that CO2 levels were highest during a period of global warmth associated with the Middle Miocene Climatic Optimum (17–14 million years ago, Ma), followed by a decline in CO2 during the Middle Miocene Climate Transition (approx. 14 Ma). Several relationships remain contrary to expectations. For example, benthic foraminiferal δ18O records suggest a period of deglaciation and/or high-latitude warming during the latest Oligocene (27–23 Ma) that, based on our results, occurred concurrently with a long-term decrease in CO2 levels. Additionally, a large positive δ18O excursion near the Oligocene–Miocene boundary (the Mi-1 event, approx. 23 Ma), assumed to represent a period of glacial advance and retreat on Antarctica, is difficult to explain by our CO2 record alone given what is known of Antarctic ice sheet history and the strong hysteresis of the East Antarctic Ice Sheet once it has grown to continental dimensions. We also demonstrate that in the Neogene with low CO2 levels, algal carbon concentrating mechanisms and spontaneous biocarbonate–CO2 conversions are likely to play a more important role in algal carbon fixation, which provides a potential bias to the alkenone–pCO2 method.

1. Introduction

Knowledge of the partial pressure of atmospheric CO2 (pCO2), beyond direct measurements of atmospheric [1] and ice-core gas compositions over the last 800 000 years [24], derives from a range of techniques including stomatal indices of fossil leaves (e.g. [59]), the carbon isotopic compositions of palaeosol carbonate nodules (e.g. [1012]), the boron isotopic compositions of shallow-dwelling foraminifera (e.g. [1317]) and the stable carbon isotope compositions of algal biomarkers, such as alkenones (e.g. [1824]). Most Cenozoic CO2 reconstructions indicate substantially higher CO2 values in the Early Eocene with a broad decrease during the Middle Eocene into the Neogene. However, a twofold variation in absolute pCO2 concentrations persists among the available proxies during key climatic intervals [25].

Following the initial development of alkenone-based CO2 reconstructions for the Pleistocene and Holocene [26,27], a revised methodology was applied for long-term CO2 reconstructions for the Middle Eocene to the Pleistocene [1820,22,23,28]. More than 20 marine localities have been used for Cenozoic reconstructions. However, the published Cenozoic alkenone–CO2 record represents a stacked record from various ocean localities (e.g. [28]). This potentially introduces bias into CO2 trends and magnitudes because temperature, nutrient conditions, seasonality and thermocline depth can impact algal growth conditions and contribute to variability in reconstructed CO2 estimates. Thus, temporal CO2 patterns from composite records could have trends specific to individual sites and not representative of changes in long-term global atmospheric conditions [21].

In this study, we present a continuous alkenone-based CO2 record from one marine site for the past 40 million years to limit variability introduced by composite results. With this new record, we attempt to reconcile the discrepancy of CO2-level reconstructions between alkenone and other proxies and explore CO2 forcing in the context of global climate change since the late Middle Eocene. Also discussed is a preliminary assessment of the potential bias related to algal carbon concentrating mechanisms (CCMs) and spontaneous bicarbonate–CO2 conversion, when using the alkenone–pCO2 method to estimate the CO2 levels in the Cenozoic.

2. Background

(a) The alkenone–pCO2 method

(i) Diffusion versus active carbon uptake models of carbon transport

Alkenones occur as a suite of long-chained (C37−C39) unsaturated ethyl and methyl ketones produced by a few haptophyte algae in the modern ocean [29,30]. Today, the predominant sources of alkenones derive from Emiliania huxleyi (originating in the Late Pleistocene [31]) and Gephyrocapsa oceanica (originating in the Pliocene [32]). Sedimentary alkenones that pre-date the Pliocene were probably produced by haptophytes related to the family Noelaerhabdaceae and genera Reticulofenestra and Dictyococcites [3337].

Alkenone-based pCO2 reconstructions require paired measurements on stable carbon isotope composition of the di-unsaturated C37 methyl ketone (δ13C37:2) and carbonates. Differences between the δ13C of algal carbon and carbonate represent the total carbon isotope fractionation (εp) that occurs during algal growth. εp derived from δ13C37:2 values (εp37:2) is calculated by the following equation: Embedded Image2.1where δorg is the carbon isotopic composition of the algal cell estimated from the δ13C of alkenone (δ37:2). The isotopic difference between δorg and δ37:2 is 4.2‰, following earlier studies [26,27,38]: Embedded Image2.2where δCO2(aq) is the δ13C value of dissolved CO2(aq) approximated from the δ13C values of shallow-dwelling foraminifera assuming isotopic and chemical equilibria among all inorganic carbon species, atmospheric CO2 and foraminifera calcite tests (e.g. [26]).

The alkenone–CO2 method is based on the assumption that the transport of inorganic carbon (dissolved CO2) across the cell membrane to the site of carbon fixation within the algal cell predominantly occurs by diffusion. For diffusion transport, the total stable carbon isotope fractionation, which occurs during marine photosynthesis, is described by the following equation [39,40]: Embedded Image2.3where Ce represents the ambient concentration of CO2(aq) ([CO2(aq)]), Ci is the intracellular [CO2(aq)], εf is the total carbon isotope fractionation that occurs during photosynthesis and εt is the carbon isotope fractionation associated with carbon transport. Most alkenone-based pCO2 reconstructions have adopted a range of εf values from 25 to 28‰ (e.g. [1824]), consistent with the majority of algal growth experiments [41,42].

Chemostat incubations for two strains of alkenone-producing haptophyte algae and two diatoms, under continuous light- and nitrate-limited conditions show that εp linearly varies with [CO2(aq)], specific growth rate (μ) and cell geometry [4244], consistent with a predominantly diffusion carbon transport model. By contrast, dilute batch cultures of E. huxleyi grown under nutrient-replete conditions and variable irradiance yield lower εp37:2 values, minor response to [CO2(aq)] changes [45] and an irradiance effect on the magnitude of εp37:2 [46]. Such results suggest that assumptions of a strictly diffusion carbon uptake model may not be valid under some nutrient and light conditions. For example, low [CO2(aq)] chemostat experiments for diatom Phaeodactylum tricornutum resulted in a nonlinear response of εp to μ/[CO2(aq)], indicating that a CCM was triggered when the supply of CO2(aq) became limiting to growth (in this case, approx. less than 7 μmol kg−1 [44,47]). CCMs are known to be available to a broad range of algal species [4851]. However, whether or not ancient alkenone-producing algae were characterized by CCMs is presently unknown.

(ii) Simplified model of diffusion carbon uptake and palaeoclimate reconstructions

Various physiological factors have been shown to impact Ci (equation (2.3)) [52]. However, because Ci is difficult to constrain, Jasper & Hayes [26] recast equation (2.3) as Embedded Image2.4where the term b represents a range of physiological variables that encompass the combined effects of growth rate, cell geometry and other potential factors.

Surface water analysis of δ13C37:2 and seawater chemistry show a strong relationship between b in equation (2.4) and the concentration of soluble phosphate [21,53]. Dissolved reactive phosphate ([Embedded Image]) is a key macronutrient that often covaries with other biolimiting trace-metal micro-nutrients, such as iron, zinc and cobalt [53], providing a link between [Embedded Image], growth rate and the value of εp.

Calibration of b with respect to [Embedded Image] using all available data (see [21]) results in the following relationships assuming εf=25‰: Embedded Image2.5

(iii) Uncertainties in pCO2 calculations

If carbon assimilation is primarily determined by diffusion transport, εp is well described by equation (2.3). However, various other parameters, in addition to CO2(aq), are known to play a role in the expression of εp. Some of these parameters influence εp through a direct impact on Ci, including growth rate, irradiance, cell membrane permeability and cell size. CO2 estimates are further compromised by uncertainties in phosphate concentrations used to estimate the physiological-dependent term b, as well as the potential of regionally different b versus [Embedded Image] relationships relative to the global calibration. Assumptions regarding the value of enzymatic isotope fractionation (εf), as well as knowledge of sea surface temperatures (SSTs)—generally obtained from various geochemical proxies—further contribute to CO2 uncertainties [21]. Finally, other factors, such as CCMs, probably arise when the intracellular concentration of CO2 becomes limiting. CCMs can impact the value of εp by transporting different carbon species with specific carbon isotopic compositions, and altering Ci relative to diffusive flux. Also, isotope fractionations associated with active transport mechanisms are likely distinct.

(b) Site location

Ocean Drilling Program (ODP) Site 925 (4°12.25N, 43°29.33W, 3042 m water depth) is located on Ceara Rise in the western equatorial Atlantic Ocean (figure 1) [55]. Modern surface waters at this site are characterized by high mean annual SST (approx. 27.5°C) and low nutrient levels (surface [Embedded Image] approx. 0.14 μmol l−1 [56]). The Middle Eocene to Pleistocene sedimentary succession recovered at Site 925 consists of pelagic carbonate oozes and chalks, primarily foraminifer-bearing nannofossil oozes with minor amounts of clay. All samples used in this study were taken from ODP Hole 925A. Previously reported Plio-Pleistocene pCO2 data are based on Hole 925C and 925D [22].

Figure 1.

Global plate reconstruction shows the location of ODP Site 925 at 40, 20 Ma and the present, as well as a detail of the northward movement of Site 925 since 40 Ma [54]. Continents are shown as tectonic terrains rather than shorelines.

Palaeogeographic reconstructions indicate that this site has moved northward over the past 40 Ma, from just south of the equator during the Late Eocene, to approximately 4°N today (figure 1) [54]. Site 925 was right on the equator around 30 Ma. Available mass accumulation rates of biogenic barium and reactive phosphorus suggest stable productivity and nutrient conditions from 36.9 to 32.7 Ma [57]. Calcareous nannofossils indicate weak to moderate tropical upwelling history [55] consistent with a minor air–sea CO2 disequilibrium expressed today, with modern surface waters enriched by approximately 22 ppm [58].

3. Material and methods

(a) Lipid extraction and analysis

Approximately 40–100 g of sediment (dry weight) were used for biomarker analysis. Samples were freeze-dried and extracted with dichloromethane/methanol (2 : 1, v/v) using an Accelerated Solvent Extractor (ASE 300, Dionex) at 120°C and 10.3 MPa. Total lipid extracts (TLEs) were concentrated under a stream of purified N2 using a Zymark Turbovap II, and then separated into three fractions using silica gel chromatography. TLEs were separated into compound classes using ashed Pasteur pipettes loaded with approximately 4 g deactivated silica gel (70–230 mesh), and sequentially eluted with 2 ml hexane, 4 ml dichloromethane and 4 ml methanol to obtain aliphatic, aromatic and polar fractions, respectively.

The dichloromethane fraction containing ketones was subject to basic hydrolysis and silver nitrate column chromatography to purify long-chain alkenones. Relative abundances of di- and tri-unsaturated methyl ketones were analysed on a Thermo Trace 2000GC equipped with an Rxi-1 ms column (60 m×0.25 mm×0.25 μm), a programmed temperature vaporization (PTV) injector and a flame ionization detector with He as the carrier gas. The oven was set to 90°C upon sample injection, held for 1 min, ramped up at 20°C min−1 to 280°C, then 2°C min−1 to 320°C. Temperature was held constant at 320°C for 30 min. The relative abundance of C37 : 2 and C37 : 3 alkenones was used to compute Embedded Image, a proxy developed to reconstruct ancient SSTs [59,60]. Analytical precision, determined through multiple analyses of an in-house alkenone standard, represents an uncertainty of ±0.005 Embedded Image units (see electronic supplementary material).

The methanol fraction containing glycerol dialkyl glycerol tetraethers (GDGTs) was further purified by passing through activated alumina dissolved in dichloromethane/methanol (1 : 1, v/v). The eluting fraction was dried under pure N2 stream, then dissolved in an azeotrope of hexane/isopropanol (99 : 1, v/v) and filtered through an ashed 0.7 μm glass microfibre filter. Analyses of GDGTs were conducted following a slightly modified method described by Hopmans et al. [61]. Compound identification and relative abundance analyses were determined using an Agilent 1200 high-performance liquid chromatography/atmospheric pressure chemical ionization mass spectrometer (HPLC/APCI-MS). Separation of GDGTs was achieved on an Alltech Prevail cyano column (150 mm, 2.1 mm I.D., and 3 μm grain size) kept at 30°C. The following solvent polarity gradient was used, based on A: hexane/isopropanol (99 : 1, v/v), and B: hexane/isopropanol (90 : 10, v/v): 100% solvent A for 5 min, then solvent B was linearly increased from 0% at 5 min, to 7.4% at 40 min, and maintained from 40 to 50 min. After each sample analysis, the column was back flushed with 100% solvent B for 14 min, followed by 10 min equilibration with normal phase flow of 100% solvent A. Mass spectrometric identification and quantification were achieved using an Agilent 6130 ion-trap mass spectrometer coupled to the HPLC by an APCI interface. Ion scans were set to m/z 1200–1500. Quantification was based on peak intensities in the mass chromatogram of the [M+H]+ ions. Relative abundance of GDGTs was used to calculate TEX86, a proxy employed to estimate past SSTs [62]. Repeated measurements of an in-house laboratory standard showed the analytical precision represented ±0.01 TEX86 units (see electronic supplementary material).

(b) Carbon isotope measurements

Carbon isotopic compositions of C37:2 alkenones were analysed on a Thermo Finnigan MAT 253 mass spectrometer interfaced with a Trace GC Combustion III (GC-IRMS) equipped with a PTV injector and a J&W Scientific DB-1 capillary column (60 m×0.25 mm×0.25 mm), using He as a carrier gas with a flow speed of 2.0 ml min−1. GC temperature was held at 60°C for 1 min, increased to 320°C at 15°C min−1 and held isothermally for 35 min. Carbon isotopes are reported relative to the VPDB standard based on an in-house reference gas calibrated to the OzTech standard (δ13C=−40.61‰). A C20 n-alkane standard was injected daily to determine the analytical accuracy of the carbon isotope measurements, yielding an uncertainty of ±0.2‰. Embedded Image and δ13C37:2 data reported for the Eocene–Oligocene climate transition [21] and the Pliocene–Pleistocene [22] have been previously published (see electronic supplementary material).

(c) Foraminifera stable isotope measurements

Splits of samples were wet sieved with deionized water at 63 μm. Surface-dwelling planktic foraminifera, including Globigerinoides ruber, ‘Globigerinoides ruber s.1.’ and Globigerinoides altiaperturus, were picked from the 250–300 μm sieve fraction in the Neogene section, and epifaunnal benthic foraminifera (Cibicidoides spp.) were picked from greater than 250 μm sieve fraction throughout the entire Site 925 study section. Carbon stable isotope analysis of the picked foraminiferal samples was performed using a Europa GEO 20–20 mass spectrometer at the University of Southampton. Samples were reacted with phosphoric acid at 70°C using an automatic carbonate preparation system ‘CAPS’ in line with the mass spectrometer. Typically, five to eight benthic specimens from each sample were combined and analysed together, and 15–20 planktic specimens were analysed together. All values are reported relative to the VPDB scale, with an external analytical precision estimated at 0.08‰ for δ13C (see electronic supplementary material).

(d) Chronology

Age determinations for ODP Hole 925A are based on biostratigraphy planktic foraminifera [63,64] with 75 age control points based on first and last occurrence data (see electronic supplementary material) over a span of 40 million years. All data are calibrated to the geomagnetic polarity timescale of [65] and the integrated timescale of Berggren et al. [66].

4. Results

(a) Temperature

SSTs based on Embedded Image and TEX86 indices were reconstructed for the past 40 Ma using the linear calibration of Conte et al. [67] and the reciprocal calibration of Liu et al. [68], respectively (figure 2a). Embedded Image SST reconstructions indicate relatively constant temperatures of approximately 28°C since the Late Eocene, except for an approximately 2°C cooling during the Pliocene–Pleistocene interval. However, 28.5°C represents maximum temperatures expressed by the Embedded Image calibration (Embedded Image value reaches 1), and therefore warmer temperatures above approximately 28°C cannot be determined. By contrast, maximum temperatures associated with the TEX86 proxy are unconstrained. Reconstructed TEX86 records exhibit much larger fluctuations compared with Embedded Image with temperatures of approximately 30°C in the Late Eocene that fall to approximately 22°C during the Late Pliocene—more than 5°C cooler at the same locality today. However, the TEX86-based temperatures since Pliocene might suffer from additional GDGT inputs, as indicated by high BIT values (mean value is 0.6; electronic supplementary material) [69].

Figure 2.

Time series of ODP Site 925. (a) Embedded Image- and TEX86-based SST reconstruction. (b) Adjusted carbon isotopes from foraminiferal stable isotope measurements to proximate δ13C of surface water dissolved inorganic carbon. (c) Carbon isotopes of alkenones; (d) εp37:2 calculated from (b) and (c); (e) pCO2 calculated from (d), using both Embedded Image- and TEX86-based SST estimates, and assuming [Embedded Image] values were 0.20, 0.14 (lower limit) and 0.39 μM (upper limit). Refer to electronic supplementary material to see details of constructing (b,d,e) and their related uncertainties; one CO2 estimate at 35.52 Ma gives a value more than 2400 ppm, which is off the chart and therefore not shown in the following figures.

In order to estimate CO2, we do not favour one SST proxy over the other for the entire 40 million year record from Site 925. Instead, we selected the highest temperature estimates from both Embedded Image and TEX86 records to synthesize a composite temperature profile. Differences between Embedded Image and TEX86 records are considered to reflect the uncertainty of SST reconstructions at each sample, which is further used in the computation of εp37:2 and pCO2. We also realize that different calibrations for Embedded Image- and TEX86-SST conversion could introduce additional uncertainties, but they are secondary compared with the difference between the two proxies themselves.

(b) Carbon isotopes and εp37:2

Planktonic foraminifera are partially to fully recrystallized in the Upper Eocene–Lower Oligocene interval of Site 925. δ13C measurements on the presumed surface water dwellers Turborotalia spp. show only small differences (0.1–0.5‰) from the benthic Cibicidoides spp. within the same sample, suggesting that Turborotalia either maintained a thermocline habitat or is affected by diagenetic alteration [21]. However, agreement between Cibicidoides δ13C values at Site 925 and an Eocene–Oligocene Southern Ocean compilation of δ13C values from ODP Sites 689, 733, 744 and 748 suggests adequate preservation of δ13C values [21]. Consequently, Pagani et al. [21] reconstructed surface δ13C by measuring Cibicidoidis δ13C values, assuming a constant offset (0.94‰) between surface and deep waters δ13CDIC.

Miocene mixed-layer foraminifera are better preserved than Palaeogene specimens. Globogerinoides altiaperturus were measured in the Early Miocene and G. ruber s.1. were selected in the Mid- to Late Miocene interval. A δ13C offset of 0.94‰ with an ascribed uncertainty of ±0.3‰ [70] was applied to both G. ruber and G. altiaperturus δ13C values to estimate surface water δ13CDIC (figure 2b). The planktic foraminifera approach in the Miocene is accompanied by benthic foraminifera Cibicidoides δ13C analyses and the applications of assumed surface–deep water isotopic offsets. Results show that the two independent approaches are within 0.20‰ from each other (see electronic supplementary material). Carbon isotope compositions on the shallow-dwelling Globogerinoides sacculifer since 5 Ma have been previously reported (figure 2b) [22].

δ13C37:2 values are more 13C depleted during the Palaeogene relative to the Neogene (figure 2c). Calculation of δ13C of surface seawater aqueous CO2 is by converting the mixed-layer δ13CDIC (figure 2b), using the composite temperature estimates and associated uncertainties from organic geochemical proxies (§4a), assuming equilibrium isotopic fractionations between different carbonate species. Aqueous CO2 δ13C and δ13C37:2 are further used to calculate εp37:2.

Figure 2 shows that variability in δ13C37:2 is predominantly responsible for the character of εp37:2 trends. εp37:2 values exhibit a long-term decreasing trend, with the majority of εp37:2 varying between 21 and 18‰ during the Late Eocene to Early Oligocene, 16–13‰ in the Late Oligocene/Miocene and 13–11‰ during the Pliocene/Pleistocene (figure 2d).

(c) pCO2 estimates

We assume a value of 25‰ for εf [41,42,44], and a range of [Embedded Image] from 0.14 to 0.39 μM, representing the present-day phosphate concentrations from 0 to 100 m near Site 925 [56], with 50 m (0.20 μM) to be the most representative depth for alkenone production. Estimates of pCO2 were then calculated assuming air–sea equilibrium and proxy temperatures (the composite temperature, see §4a) using Henry's law (figure 2e). The uncertainty of pCO2 is constrained by the variability of temperature and phosphate estimates. The maximum pCO2 is calculated by using the upper limit of the εp, highest phosphate level and highest temperature estimates; whereas the minimum of pCO2 is a result of the lower limit of εp, lowest phosphate and lowest temperature (see §4a,b for the determination of uncertainties associated with temperature and εp). The resulting uncertainty of pCO2 estimates averages 38%, much larger than an error propagation model through Monte Carlo procedure (11%) with assumed uncertainty of different parameters used in pCO2 calculations [20].

5. Discussion

(a) Comparisons between Site 925 and published Cenozoic pCO2 records

Our new CO2 results from Site 925 (figures 3 and 4) broadly agree with previously published composite alkenone–pCO2 record for the Cenozoic [28]. Nevertheless, systematic differences are apparent with the new record indicating higher CO2 levels after approximately 32 Ma (figure 3). This discrepancy can be explained, in part, by regional differences in oceanography and algal growth conditions, as expressed by variable εp values. For example, the Miocene average εp at Site 925 is approximately 15‰, significantly higher than the mean value of approximately 10‰ at Deep Sea Drilling Project (DSDP) Site 588 [19,20]. Furthermore, new SST estimates (e.g. [19,20]) are used in this study, and these new Embedded Image- and TEX86-based temperature records are considerably warmer during the Miocene to Early Pliocene, compared with foraminifer δ18O-based estimates (6–8°C, DSDP Site 588 [22]). Warmer SST estimates increase calculated pCO2 levels [22].

Figure 3.

A comparison of alkenone-based pCO2 composite from multiple marine sites as compiled in the study of Pagani et al. [28] and ODP Site 925 record since the Late Eocene. Antarctic glaciation thresholds (approx. 750 ppm) and Northern Hemisphere glaciation threshold (approx. 280 ppm) deduced from climate models [71] are marked by dashed lines.

Figure 4.

Comparison between different proxy-based pCO2 estimates for the past 40 Ma. Boron isotope, palaeosol and stomata data are summarized by Beerling & Royer [25], with additional data from Foster et al. [15].

Finally, comparison of our new record with recent CO2 reconstruction efforts, with putatively improved techniques and higher quality data [25], suggests broad agreement with boron isotope-pH [13,15,16], stomata [59] and palaeosol estimates [1012], although palaeosol estimates have inherently larger uncertainties (figure 4).

(b) Palaeoclimate implications

(i) The late Middle Eocene to Early Oligocene

Coupled climate–ice sheet model simulations of the Eocene/Oligocene climate transition indicate that sudden nonlinear jumps in ice volume can be triggered by the congruence of favourable orbital cycles when pCO2 concentrations decrease to 2.5–3 times pre-industrial levels (approx. 750 ppm; figure 5) [71,73]. Independent proxy reconstructions of pCO2 across the Eocene–Oligocene transition, including those estimated by the boron isotope-pH methodology [16] and alkenone-based CO2 records from Site 925 [21], support these model results and suggest CO2 began to decrease immediately prior to the Oi-1 event and declined during the climate transition, probably falling below the approximately 750 ppm threshold.

Figure 5.

Climate and atmospheric CO2 history for the past 40 Ma. Benthic δ18O from Zachos et al. [72]. Major warming (red arrows) and cooling (blue arrows) events are labelled. Red bars indicate brief history of Antarctic and Northern Hemisphere ice sheets. Antarctic glaciation thresholds (approx. 750 ppm) and Northern Hemisphere glaciation threshold (approx. 280 ppm) deduced from climate models [71] are marked by dashed lines.

Compared with a multi-site approach, oceanography and algal growth conditions might be less variable in a single site. However, the Late Eocene–Early Oligocene CO2 record at Site 925 may still carry features that are specific to this locality. Although no major tropical upwelling has been detected throughout the entire record by calcareous nannofossil assemblages [55], plate reconstructions indicate that Site 925 tectonically moved from the Southern to the Northern Hemisphere and was positioned on the equator around 30 Ma (figure 1). This transit would have altered the expression of local sea surface CO2, particularly at the equator where the convergence of trade winds drives upwelling of CO2-rich deep waters [74]. A distinct spike in εp values and reconstructed pCO2 is evident in the Middle Oligocene (approx. 29.5 Ma, figure 2d), representing about 10–30% increase in CO2 from the background level. The magnitude of present-day equatorial upwelling, however, is represented by a seawater CO2 enrichment in the western equatorial Atlantic by approximately 30 ppm—about 9% deviation from atmosphere–seawater equilibrium [58]. This potentially implies that equatorial upwelling only partially contributed to the Middle Oligocene pCO2 spike recorded in Site 925, or that upwelling intensity was higher and/or upper water-column stratification was less intense during the Oligocene.

(ii) The Late Oligocene to Early Miocene

Climatic behaviour during the Late Oligocene/Early Miocene interval is particularly difficult to explain given existing CO2 records. After a period of gradual cooling following the onset of glaciation on Antarctica near the Eocene–Oligocene climate transition at approximately 34 Ma, an approximately 0.8‰ decrease in benthic δ18O values (ODP 1218, eastern equatorial Pacific [75]) suggests a period of substantial warming and/or deglaciation during the latest Oligocene (approx. 27–23 Ma; figure 5). However, our corresponding CO2 records indicate a long-term decrease in atmospheric carbon dioxide through the Late Oligocene (figure 5). Isotopic evidence for warming and/or partial deglaciation at the end of the Oligocene is not simply an artefact of stacked benthic δ18O records from different marine sites [72,75] given that the δ18O record from ODP Site 1218 spans the entire Oligocene and unambiguously supports the presence of a negative oxygen isotope excursion during this time [75]. Thus, the cause for this negative δ18O trend remains unresolved in relation to our new CO2 record.

Another apparent decoupling between CO2 and climate occurs near the Oligocene–Miocene boundary (approx. 23 Ma), represented by a transient, positive benthic foraminiferal δ18O excursion (greater than 1‰) interpreted as a period of substantial glaciation (known as the Mi-1 event) [76]. Our records suggest invariant CO2 concentrations during this apparent glaciation/deglaciation, defying our current understanding of the necessary forcing required to drive Antarctic ice sheet variability.

Assuming approximately 2°C of cooling in the deep sea [77], approximately 0.5‰ of the 1‰ δ18O shift at Mi-1 must have been driven by an increase in ice volume. If continental ice on Antarctica had an average isotopic composition of −40‰, as indicated by isotopic modelling [71], then more than 22×106 km3 of ice—roughly equivalent to the entire present-day East Antarctic Ice Sheet (EAIS)—must have accumulated within 400 000 years. However, CO2 levels during the Oligocene appear low enough to have already maintained a fully glaciated Antarctica according to ice sheet simulations [78]. Moreover, the recovery phase of Mi-1 is even more enigmatic because models require substantially higher CO2 levels—at least two times higher than the formation threshold of the EAIS (approx. 1500 ppm) to cause substantial ice sheet retreat [71,78]. Coupled climate–ice sheet simulations show that orbital forcing alone cannot cause Antarctic deglaciation once the ice sheet expands over the continent [73], and requires increases in greenhouse gas concentrations to more than four times pre-industrial levels to cause substantial ice retreat [78]. It is possible that the resolution of our new CO2 record is too low to resolve large fluctuations within the body of the Mi-1 event. Still, exiting the glacial period would require substantially higher CO2 levels that are not detected in our record. Alternatively, polar climate sensitivity to CO2 might have been much higher than simulated in climate models, or our understanding of Antarctic ice physics is incomplete and ice sheet stability, potentially reflected by the ‘Oi’ and ‘Mi’ events (e.g. [75]), is far more sensitive and dynamic than indicated in modelling studies. Another possibility is that the assumption that benthic δ18O variability is representative of global temperature/ice volume is flawed. For example, Nd isotope records suggest alternating deep water sources during the Palaeogene [79,80], with the potential to alter the temperature, salinity and δ18O value of deep waters. Different deep water sources during the Late Oligocene and the Mi-1 event could have played a role in the magnitude of the observed benthic δ18O fluctuations.

Apparent discrepancies between proxy records of ice volume and CO2 could also result if alkenone–CO2 estimates are simply incorrect or biased for some time intervals. Various theoretical and empirical exercises have been performed to assess non-CO2 factors that impact algal carbon isotope fractionation and uncertainties in alkenone–pCO2 estimations [50,52]. For example, higher ambient CO2 levels during the earlier part of the Cenozoic could have contributed to a reduction in bicarbonate uptake via β-carboxylation, resulting in εf values greater than 25‰ [21], as well as inhibiting CCMs. The isotopic impact of CCM upregulation during low CO2 would arguably minimize εp and reconstructed CO2 variability, which is discussed in greater detail in §5c.

(iii) The Mid- to Late Miocene

Very low Miocene pCO2 concentrations that characterize published alkenone (figure 3) [19,20] and boron isotope [81] CO2 records have puzzled the palaeoclimate community for nearly a decade. If atmospheric CO2 was a key parameter forcing cryosphere expansion during the Neogene, then major Northern Hemisphere glaciation should have arguably occurred some 20 million years earlier than the accepted age of approximately 2.7 Ma (figure 3) [71,82,83]. Low pCO2 levels are also difficult to reconcile with the well-documented warmth of the Middle and Late Miocene, which was characterized by SSTs significantly higher than today [84]. For example, subtropical east Pacific (ODP 1010) and northeast Pacific (ODP 1021) have been found to be at least 12°C warmer at 12 Ma relative to today, and 5°C warmer relative to the Early Pliocene [84]. Prior alkenone and boron isotope reconstructions have both indicated that the global warmth of the Middle Miocene climate optimum (MMCO, approx. 17–14 Ma) and the subsequent expansion and stabilization of Antarctic ice sheets (Middle Miocene climate transition (MMCT) approx. 14 Ma) were associated with relatively invariant pCO2 [20,81] and slightly higher pCO2 during ice expansion (figure 3) [20]. Not surprisingly, climate models applying published alkenone and boron isotope-based CO2 records [85,86] cannot simulate Middle Miocene climate signals as determined by proxy records [19,81].

Intriguingly, our new record implicates a more important role for pCO2 in the climatic variability of the Early to Middle Miocene. The partial pressure of carbon dioxide rises to 400–500 ppm during the climatic optimum of the Middle Miocene followed by an approximately 100 ppm decline during MMCT (figure 5), which better agrees with recent stomatal index [6] and boron isotope estimates (figure 4) [15]. Comparing the Site 925 record with the earlier alkenone–pCO2 record from DSDP Site 588 [19,20], we found that the isotopic fractionation εp37:2 is larger (15‰ versus 10‰) and the SST estimate is higher by 6–8°C, both of which could have contributed to higher Miocene CO2 levels in the new assessment. Although we cannot rule out the possibility that CO2 air–sea disequilibrium at Site 925 contributed to higher local CO2 estimates, three independent methods agree that pCO2 during the MMCO was higher, and then declined during the MMCT cooling, implying that CO2 was closely linked to major Miocene climate events. Importantly, Miocene CO2 levels now appear higher than Pliocene and Pleistocene concentrations, consistent with the appearance of Northern Hemisphere glaciation during the Pliocene (figures 3 and 5).

(c) Assessing the effect of bicarbonate–CO2 conversion

The capacity for active carbon transport is common among marine and fresh water algae [48,49,51]. This is because ribulose-1,5-bisphosphate carboxylase/oxygenase, the primary carboxylase, has a relatively low affinity for CO2 and, for most algal species, is less than half saturated under current CO2 levels [51]. CCMs increase carbon availability and reduce rates of photorespiration that impact rates of carbon fixation [51].

Whether or not CCMs were an important aspect of ancient alkenone producers remains speculative. Whereas some organisms, such as diatoms, show highly efficient and active CCMs [48,87], CCMs in modern E. huxleyi and other coccolithophores appear weakly expressed [53,88]. Physiological parameters of extinct alkenone-producing species are unknown. Since CCMs are energetically expensive, ancient haptophytes probably lacked operational CCMs under the significantly higher CO2 levels of the past [89]. However, the presence of CCMs could have become more pronounced during the Neogene due to the considerable decline in atmospheric CO2 (figures 36). If this perspective is valid, then quantitative CO2 reconstructions could be compromised during periods of relatively low pCO2 levels.

Figure 6.

Comparison between the spontaneous bicarbonate–CO2(aq) conversion (green symbols) and the CCM potential of alkenone producers (orange bands) for the past 45 Ma. Embedded Image−CO2(aq) conversion (r/rk) based on the model of Riebesell et al. [90] and Wolf-Gladrow et al. [91], using seawater pH and SST data provided by boron isotope studies [13,15,16,18], and a constant radius of 2.67 μm for alkenone-producing haptophyte algae. Computation of CCM potential is based on the model of [44,50], using previously published cell size [21,37,89,92] and existing palaeo-CO2 estimates [18,21,28]. CCM potential ([(μ/Ce)/(P/C)]100−100) represents an arbitrary, qualitative scale of CCM activity.

Various theoretical models have been developed to understand the factors that would drive upregulation of CCMs [47,50,90,91]. One model assumes that active transport occurs when diffusive flux achieves a minimum threshold of intracellular CO2 necessary for adequate growth. Minimum diffusive flux estimates depend on ambient [CO2(aq)] and constraints imposed by cell size. Assuming spherical geometry, minimum [CO2(aq)] can be defined as [47,90,91] Embedded Image5.1where Fin is the diffusive influx of CO2(aq), DT is the diffusion coefficient of CO2, r is the ‘surface area equivalent’ spherical cell radius [91], rk is the reacto-diffusive length (i.e. length of the boundary layer where HCOEmbedded Image has the opportunity to convert to CO2(aq)), and the term (1+r/rk) represents the contribution of extracellular, uncatalysed and spontaneous Embedded Image−CO2 conversion to the total supply of CO2 [47,90].

The term rk is calculated by Embedded Image5.2where k is the rate constant for conversion of Embedded Image and H2CO3 to CO2(aq). An approximation for the temperature dependence of DT is given by [93] Embedded Image5.3with activation energy Ed=19 510 J mol−1, gas constant R=8.3143 J K−1 mol−1 and TK is temperature in kelvin. For the rate constant k, Gavis & Ferguson [94] demonstrated that Embedded Image5.4where k1 and k2 are the rate constants of formation of CO2 from Embedded Image and H2CO3, respectively, and [OH] is the hydroxyl ion concentration. At 25°C, k1=8500 m3 mol−1 s−1 and k2=3×10−5 s−1 [94].

Given the equilibrium constant for seawater Kw=[H+][OH], [OH] can be calculated if seawater pH is known. At atmospheric pressure [95], Embedded Image5.5Therefore, the portion of fixed CO2 contributed by the conversion of bicarbonates in the geological past can be calculated if seawater temperature (TK), pH, salinity (S) and the cell size (r) of ancient alkenone producers are constrained.

Ocean pH estimates and SST reconstructions are available for the Eocene–Oligocene climate transition [16], Middle Miocene [15], Pliocene [18] and the Pleistocene [13], and ancient cell sizes for assumed alkenone producers (e.g. Reticulofenestra, Dictyococcites and Cyclicargolithus) can be estimated using coccolith size measurements [21,37,89,92]. However, the available coccolith length data and boron-based pH estimates are not derived from the same marine sites. Cell sizes of assumed alkenone producers show large spatial variability. For example, Pagani et al. [21] demonstrated that Reticulofenestrid coccoliths from the Southern Ocean sites (e.g. ODP 1090) were two times as large as those in the tropical Atlantic (ODP 925, 929) during the Eocene–Oligocene transition. This apparent cell size variability of alkenone producers hampers the application of available coccolith data to the Embedded Image−CO2 conversion evaluation of the Cenozoic. Therefore, given that cell size data for the past 40 million years are not available for Site 925, we assume a constant cell radius of 2.67 μm, which is the average value estimated from coccolith dimensions for the Cenozoic [21,37,89].

Boron isotope-derived seawater pH and coeval Mg/Ca-based SST, with the assumption of a constant salinity of 35 and 2.67 μm cell radius of alkenone producers, enables an assessment of the impact of bicarbonate–CO2 conversion on alkenone–pCO2 estimates during the Cenozoic. Temporal patterns in the ratio r/rk indicate that the percentage of fixed carbon derived from bicarbonate conversion is generally low, of the order of 1.4–2.3% (figure 6), with a trend of elevated bicarbonate contribution by the Pleistocene (approx. 40% more relative to the Late Eocene; figure 6). Since we assume constant cell size, this increase is primarily due to an approximately 0.5 pH drop from approximately 7.7 during the latest Eocene [16] to 8.2 today. Alternatively, CCMs could have become increasingly important [50,96] to counter low CO2. However, accurate quantitative correction for the effect of Embedded Image−CO2 conversion on alkenone–pCO2 estimates will require more detailed site-specific data on cell size of the alkenone producers and pH history of the region.

An alternative model to estimate CCM activity [44,50] considers both diffusion and active uptake of inorganic carbon to describe the nonlinear culture results of P. tricornutum: Embedded Image5.6Here, β is a constant equivalent to the ratio of CO2 loss (leakage by diffusion) to carbon fixation, P is membrane permeability, C is cell carbon content and εt is the isotopic fractionation associated with the inorganic carbon diffusion back from the cell to the surrounding seawater. As detailed in Laws et al. [50], the value of εp becomes insensitive to μ/[CO2(aq)] when Embedded Image5.7Empirical relationships between P and C show its dependence on the cell radius: P/C=0.285 r−1. Minimum [CO2(aq)] that would trigger active transport occurs when the term (1+β) approaches unity, or leakage is zero. Laws et al. [50] calculate that εp for E. huxleyi would become a nonlinear function of μ/[CO2(aq)] as photoperiod growth rate exceeds 1.1 d−1 for a cell radius of 2.6 μm, comparable with the culture results of E. huxleyi [42].

Growth rates in the natural environment for E. huxleyi and other alkenone-producing haptophytes are generally low, below 1 d−1, ranging from 0.1 to 1 d−1 [53,97]. Cell dimensions of alkenone-producing algae since Middle Eocene [18,21,37,89] used in conjunction with estimates of pCO2 provide a means to identify geologic intervals when upregulation of CCMs was potentially necessary. For example, cell radii for alkenone producers range from approximately 4 to 2.4 μm during the Eocene–Oligocene transition [21], and cell radii were about 50% smaller during the Early Miocene [37]. Knowledge of cell radius allows calculation of the ratio P/C [50] and the ratio μ/[CO2(aq)] can be estimated from alkenone-based CO2 reconstructions and a range of modern haptophyte growth rates, assuming alkenone-based CO2 estimates are valid even if CCMs were active. Using pCO2 data from earlier results [19,20,28], figure 6 shows that for low to moderate growth rates, the operation of CCMs is expected during the Miocene, particularly during the climatic optimum of the late Early Miocene, suggesting that the alkenone–CO2 estimates may be compromised during this time interval. Obviously, this result is speculative given the necessary assumptions. For example, CO2 concentrations used in this exercise derive from alkenone measurements and if CCMs were active, then these CO2 values would appear artificially lower than actual concentrations and further exaggerate the potential influence of CCMs.

6. Conclusions

An alkenone-based pCO2 record spanning the past 40 million years is presented for ODP Site 925—representing the first long-term Cenozoic pCO2 record constructed from a single site. Facilitated by improved methodology and careful consideration of assumptions, this record provides refined pCO2 estimates from a site that is characterized by limited long-term variability in oceanographic conditions. This record, therefore, reflects our most up-to-date effort to better constrain the Cenozoic history of atmospheric CO2. This new record confirms predictions from climate models regarding the role of CO2 in the cryosphere evolution in both hemispheres at several key time intervals (e.g. Eocene–Oligocene transition, MMCT, Plio-Pleistocene). This record also suggests that Miocene CO2 levels were higher than earlier estimates, which better reconciles a long-standing data–model discrepancy for the Miocene. However, outstanding issues remain which include the presumed warming or deglaciation in the Late Oligocene and the abrupt and transient oxygen isotope excursion at Mi-1, both of which challenge our understanding of CO2 climate and ice sheet sensitivity.

We also evaluate spontaneous HCOEmbedded Image−CO2(aq) conversion, which is one variable that could impact CO2 estimates from the alkenone–pCO2 method. We conclude that the contribution from HCOEmbedded Image−CO2 conversion to the carbon fixed by alkenone-producing algae is minor, although this potential has increased by 40% from Late Eocene to the Quaternary. Finally, an exercise to evaluate potential CCM activity in haptophyte algae suggests that low pCO2 levels in the Neogene are more likely to trigger CCMs.

Funding statement

This study was supported by NSF AGS P2C2 award 1203163 to M.P. and R.D.

Acknowledgements

We thank Dan Lunt for his editorial handling, Rich Pancost and an anonymous reviewer for their thoughtful reviews.

Footnotes

References

View Abstract