Geological records provide key context to current assessments of the consequences of rising atmospheric CO₂ on ice sheet stability and oceanic temperatures (Foster et al., 2017; Golledge, 2020; Zachos et al., 2008). The Oligocene to Miocene time interval has been proposed to represent a nonlinear transition between the “greenhouse” and “icehouse” stages of Earth history (Miller et al., 1991; Zachos et al., 2001) useful to evaluate the Earth system climate sensitivity to the hypothesized progressive CO₂ drawdown of the Cenozoic (Deconto et al., 2008; Zhang et al., 2013). However, the long term decline in CO₂ estimated by existing proxy records contrasts with the rather stable climatic state with multimillion year warming (e.g. Late Oligocene Warming) and cooling (e.g. Mi-1 glaciation) trends interpreted from deep ocean (Cramer et al., 2011; Lear et al., 2000), and surface ocean records (Guitián et al., 2019; Liu et al., 2009; O'Brien et al., 2020) and variable Antarctic Ice sheet volume and sea level (Lear et al., 2004; Liebrand et al., 2017; Miller et al., 2020).

The long term pCO₂ trends from the Oligocene to early Miocene are derived from the sensitivity of marine algae to pCO₂ based on the carbon isotopic fractionation in organic matter during photosynthesis (εp) of marine phytoplankton (Henderiks and Pagani, 2008; Pagani et al., 1999, 2000, 2005; Super et al., 2018; Zhang et al., 2013) while published δ11B based CO₂ estimates cover the latest Oligocene into early Miocene (younger than 24 Ma). The algae isotopic fractionation can be reconstructed in the past from sediments by the analysis of δ13C of the organic lipids and reconstruction of the δ13C of the DIC in the seawater from which biomass was produced. Fractionation (εp) is predicted to be higher when CO₂ availability is high relative to cellular carbon demand. A decrease in atmospheric CO₂ and consequently in CO₂ of the surface ocean should therefore lead to a decrease in εp at any given site. However, in addition to CO₂, the εp in phytoplankton is affected by physiological factors such as the rate of carbon fixation, which may vary over time in a given location due to variations in temperature or the supply of light (Rau et al., 1996; Stoll et al., 2019).

One approach to evaluate the relative contribution of physiological factors vs CO₂ is to produce εp records from sites of widely contrasting oceanographic setting, where the CO₂ signal may be expected to be common to both locations but the environmental factors affecting the fractionation such as nutrient availability might not be expected to change in unison. In this study, we produce a new long-term record of εp over the Oligocene to Miocene time interval at two new, widely separated locations: IODP Site1406 in the subtropical North Atlantic off the Newfoundland coast, and ODP 1168 in the Southern Ocean off Tasmania. We also increase the resolution of determinations at the equatorial Atlantic ODP 925. The existing εp-based CO₂ estimations for the Oligocene are derived from ∼ 1 million year resolution measurements from two sites (Site 925 and 516) on the South American margin of the equatorial and South Atlantic; in the early Miocene an additional North Atlantic record (Site 608) provides data (CenCO2PIP Consortium, 2023).

Our new < 1 million year resolution εp records from these two mid- latitude locations allow us to directly compare εp with estimates of SST derived from alkenones extracted from the same samples, since unlike very warm tropical locations, the U37k′ index still retains sensitivity to temperature in the mid-latitudes during the Oligocene and early Miocene. Additionally, we compare εp with benthic δ18O available from the same sediments. Variations in benthic δ18O are controlled by changes in both deep-water temperature and deep ocean δ18O_sw which reflects ice volume. We further measure εp and benthic δ18O at approximately 20–30 ky resolution over a series of eccentricity cycles in the early Oligocene at IODP 1406. The dataset allows a robust evaluation of the relationship between εp and climate for this time interval. We further discuss the significance of the observed εp record with the implications for the phytoplankton sensitivity over multimillion year timescales over the Cenozoic.

The carbon isotopic fractionation in phytoplankton during photosynthesis is affected not only by the aqueous carbon dioxide (CO_2[aq]) but also by physiological factors related to the cellular uptake of carbon. Physiological factors were initially modelled from the assumption of diffusive carbon acquisition in phytoplankton cells (Rau et al., 1996), where higher εp could be induced by higher CO_2(aq), lower instantaneous growth rates, or a higher cellular surface area to volume ratio. Both cellular permeability and the carbon isotopic fractionation by the Rubisco enzyme have been assumed to be constant, with Rubisco fractionation typically estimated between 25 ‰ and 29 ‰ for alkenone producers (Pagani, 2014). Traditional attempts to reconstruct pCO₂ from εp have simplified this original diffusive model by relating εp and CO₂ with a single factor b defined to include all physiological parameters affecting the fractionation, and εf representing the fractionation of the Rubisco enzyme (Jasper et al., 1994). 1εp=εf-bCO2[aq] The b-value has been estimated from modern photic zone and culture samples, for which CO_2(aq) is independently known. For sedimentary alkenones, previous pCO₂ calculates have either (1) assumed the modern b-value for that oceanographic setting remained constant in the past (e.g. Zhang et al., 2013), (2) applied modern relationships between b-value and phosphate and a simulated paleo-surface ocean phosphate concentration at the site (Pagani et al., 2011) (3) estimated the difference between the modern b-value at the site and the paleo-setting b-value from productivity proxies (Bolton et al., 2016) or (4) applied variation in the b-value at the site based on proxies for coccolithophore size (Henderiks and Pagani, 2007). Despite the appeal of this approach, a recent re-evaluation of cultures and field observations suggest the b term is not well predicted by growth rate, light or cell size alone in a diffusive model but that additional effects occur from carbon concentration mechanisms (CCM) on carbon uptake at lower CO₂ concentrations, which cause a deviation in the CO₂ dependence from the theoretical hyperbolic relationship (Stoll et al., 2019; Hernández-Almeida et al., 2020). A further challenge to the physical diffusive model is that the Rubisco fractionation in coccolithophores has been measured in-vitro as 11 ‰ rather than 25 ‰ (Boller et al., 2011), suggesting that fractionations larger than 11 ‰ may reflect the operation of additional enzymatic fractionations (Wilkes et al., 2018). The lower Rubisco fractionation implies a lower sensitivity of εp to CO₂ (e.g. as explored in González-Lanchas et al., 2021).

A meta-analysis of experimental culture data (Stoll et al., 2019) suggests that εp features a logarithmic dependence on CO₂, rather than the hyperbolic dependence implied by Rau et al. (1996). This analysis does not resolve the mechanisms for the observed relationship between εp and CO₂, but over the range of CO_2(aq) from 5 to 30 µM, it provides an empirical relationship for interpreting the magnitude of CO_2(aq) change implied by a given εp change. The culture dataset illustrates more broadly how εp is the sum of its dependencies on ln(CO₂), ln(light), and growth rate μi and cell radius: 2εp=2.66ln⁡CO2+2.33ln⁡(light)-6.98μi-1.28radius+6.26 where CO_2(aq) is in µM, light is in µE, growth rate μi is d⁻¹ and radius is in microns (see Stoll et al., 2019 for confidence intervals on the regression).

From this empirical culture calibration, two challenges remain for the estimation of past CO₂ from εp measurements derived from sedimentary alkenones. First, its use would require an estimation of the cell radius, light during the season and depth of alkenone production, and the growth rate. While cell size can be estimated from coccolith length (Henderiks and Pagani, 2007), determining the absolute light and growth rate is rarely possible. Since the equation is a linear sum of these influences, these non-CO₂ variables may be integrated into the intercept (e.g. as in González-Lanchas et al., 2021), where the intercept (I) would decrease with higher growth rates and larger cell sizes and increase with higher light. 3εp=2.66ln⁡CO2+I Yet, as with Eq. (2), there remains the challenge of determining which value should be used for the intercept for past conditions, and whether a constant or variable I is more appropriate for a given site since there are limited proxies for algal growth rate. Recent culture studies document a 0.5 ‰ decrease in εp per 1 °C warming (Torres Romero et al., 2024), and show that this magnitude is identical to the product of εp dependence on growth rate (Stoll et al., 2019) and the modeled temperature dependence of coccolithophore growth rates (Krumhardt et al., 2017) derived from diverse culture and field studies (Fielding, 2013; Sherman et al., 2016; Behrenfeld et al., 2005). This suggests that growth rate, and I, may vary over time at a given location if temperature is variable. Therefore, records of SST from alkenone unsaturation or other proxies provide the opportunity to deconvolve the effects of temperature-driven growth rate variations on εp even when the absolute growth rate is not known.

In this study, given the potential for oceanographic conditions at the studied sites to differ significantly from those in the modern ocean at these locations, and the concomitant high uncertainties in estimating an appropriate b-value for the traditional approach or I for Eq. (3), we do not calculate the past absolute CO₂ concentration from our εp measurements. Instead, we account for the potential influence of temperature-driven growth rate changes on our εp records using alkenone temperature estimates derived from the same samples. Similarly, we evaluate the potential impact of cell size variations on the εp changes. Then, we employ the sensitivity of εp to CO₂ in Eq. (3) to estimate possible relative changes in CO₂ in the case where the other nutrient-stimulated growth rate or light influences on εp were constant during the studied interval at each site evaluating evidence for this assumption.

Biomarkers from sediments of IODP 1406 and ODP 1168 were extracted from 30 g of freeze-dried sediment using an Accelerated Solvent Extractor 350 with CH₂Cl2/MeOH (9:1 v/v) solvent for four static cycles at 100 °C and further silica gel column chromatography protocols for purification of the ketone fraction containing the alkenones (see Guitián et al., 2019 for details).

Alkenone ratios were obtained with a Thermo Scientific Trace 1310 Gas Chromatograph (GC)-FID. Originally, for IODP 1406 and 1168 samples the GC-FID was equipped with a non-polar (60 m × 0.25 mm × 0.25 µm) capillary column (ZB-1ms, Zebron^™) at ETH Zurich and at Lamont-Doherty Earth Observatory from Columbia University (Guitián et al., 2019; Guitián and Stoll, 2021). However, ODP Site 1168 samples older than 22.4 Ma featured more complex chromatograms and a high diversity of compounds. To reduce the effects of coelution, samples were additionally analyzed on a 105 m column RTX-200ms at ETH Zurich, which improved separation of long chain ketones (Rama-Corredor et al., 2018). The following temperature program was used: 1 min at 50°C, temperature gradient of 40 °C min⁻¹ to 200°C and 5 °C min⁻¹ to 300°C, hold for 45 min, and increased to 320°C at 10 °C min⁻¹ and hold for 8 min. Carrier gas was Helium at a flow rate of 1.5 mL min⁻¹. In-house standards and replicates injected at every sequence ensured instrument precision. A subset of IODP Site 1406 and samples younger than 22.4 Ma from Site 1168 were re-measured with the RTX-200 to ensure replicability (Table S1 in the Supplement). Method used for each organic analysis is described in the supplementary material dataset.

Sea surface temperature was calculated from U37k′ ratio using the Bayspline calibration (Tierney and Tingley, 2018) for all samples in IODP 1406 and for the young set (< 23.1 Ma) of samples in ODP 1168. Because for those, we find U37k′ ratios within the analytical uncertainty using both columns, we report the original ZB-1 results for all Site 1406 samples and Site 1168 samples younger than 22.4 Ma/408.22 m depth. The RTX-200 column provided substantially improved resolution of C38 peaks, allowing quantification of C_38:2 and C_38:3 ME peaks. For samples between the ages of 23.1 and 29.1 Ma in ODP 1168 the RTX-200 column still did not sufficiently resolve coelutions on the C_37:3 peaks. Therefore, for this interval we provide temperatures estimated from the U38MEk′ ratio applying the Novak et al. (2022) core top calibration (Table S1). For the ODP 925 equatorial site samples, the C_37:3 methyl ketone is under the detection limit, therefore we further purified and analyzed the extracted organics as in Guitián et al. (2019) to get the temperatures from the TEX₈₆ ratio using the BAYSPAR calibration by Tierney and Tingley (2015).

Compound-specific δ13C measurements were performed on a Thermo Scientific Trace 1310 Gas Chromatograph coupled to a Thermo Scientific GC Isolink II, a Conflo IV, and a Delta V Plus Mass Spectrometer at ETH Zurich. Oxygen was flushed through the combustion reactor for one hour at the beginning of each sequence and seed oxidized for one minute before each injection. Alkenones from ODP Site 1168 and IODP 1406 were analyzed on a GC equipped with a non-polar capillary column (60 m × 0.25 mm × 0.25 µm) (ZB-1ms, Zebron^™) and 5 m guard column. Helium was used as carrier gas flow with 2 mL min⁻¹. GC oven was set to 90 °C, ramped to 250 at 25 °C min⁻¹, to 313 °C at 1 °C min⁻¹ and finally to 320 °C at 10 °C min⁻¹. The GC oven was then maintained isothermally for 20 min. A subset of IODP 1406 samples were measured additionally at the Lamont-Doherty Earth Observatory on equivalent instrumentation but with some modifications for improved sensitivity (Baczynski et al., 2018) and following similar GC procedures, with similar results.

From ODP Site 1168, a subset of samples older than 24.5 Ma featuring more complex chromatograms were rerun on a GC-irMS equipped with a RTX-200 ms column. GC oven was set at 50 °C ramped to 275 °C at 40 °C min⁻¹, to 295 °C at 0.5 °C, hold for 22 min, and finally ramped to 320 °C at 10 °C min⁻¹ and hold for 5 min. Flow rate was 1.5 mL min⁻¹. Comparison of a subset of samples from Site 1406 and ODP 1168 younger than 22.4 Ma showed that δ13C C37:2 were similar on both ZB-1 and RTX-200 columns. We consequently report here δ13C C37:2 from the ZB-1 runs, with the exception of the samples from Site 1168 older than 22.4 Ma. All values are reported here in parts per mil (‰) relative to VPDB (Vienna Pee Dee Belemnite). Sample replicates, in-house alkenone standard (provided by G. O'Neil, Western Washington University, and C. M. Reddy, Woods Hole Oceanographic Institution), and known isotopic mixtures A5 and B4 (supplied by A. Schimmelmann, Univ. of Indiana) were simultaneously measured to determine the analytical accuracy of the measurement and an uncertainty of 0.5 ‰.

Isotopic composition of CO_2[aq] is estimated from the temperature dependent fractionation between DIC and aqueous CO₂ during alkenone production of Rau et al. (1996) based on Mook et al. (1974) and Freeman and Hayes (1992): 4δ13C[CO2]aq=δ13CDIC+23.644-9701.5T We calculate the δ13C DIC from the δ13C measured on the bulk carbonate, which is dominated by calcareous nannofossils, for which previous studies show Reticulofenestra to be the most abundant genera (Guitián et al., 2020). Because there is no divergence of vital effects between small and large coccoliths in the late Oligocene to early Miocene (Bolton and Stoll, 2013), we propose that the offset between coccolith δ13C and DIC is likely to remain constant. We subtract 0.5 ‰ from the δ13C bulk to calculate δ13C DIC, based on average alkenone-producing coccoliths cultured at DIC < 4 mM compiled in Stoll et al. (2019). Support for estimating photosynthetic fractionation from coccolith δ13C is provided by recent culture studies of G. oceanica (Torres Romero et al., 2024). In previous studies, the δ13C DIC has also been estimated from the δ13C of calcium carbonate of benthic foraminifera with the assumption of a constant and known offset between the δ13C DIC of the deep and surface ocean. Although the foraminifera content in Site 1406 and 925 is very low, sediments feature sufficient well preserved benthic foraminifera, mainly epifaunal Cibicidoides spp. in the size range larger than 200 µm. At ODP Site 1168 benthic foraminifera were scarce for picking for isotopes in many intervals and the progressive evolution of water depth at the site may change the δ13C offset between the benthic environment and the surface ocean over time (Exon et al., 2001). For an additional sensitivity test to evaluate the significance of the method of DIC estimation and facilitate comparison to other published εp records calculated from benthic δ13C, we also estimate surface ocean DIC by adding a constant offset of +2 ‰ to the δ13C benthic measurements, following previous Miocene and Oligocene studies (Guitián et al., 2019; Pagani et al., 2011; Zhang et al., 2013).

Bulk carbonate and benthic foraminifera were measured using analytical techniques described in Guitián et al. (2019) with the guidelines from Breitenbach and Bernasconi (2011) for small carbonate samples on a GAS BENCH II Delta V Plus irMS from Thermo Scientific with international (NBS-19 & 18) and in-house carbonate as standards achieving a precision of 0.07 ‰.

Carbon isotopic fractionation (εp), describes the fractionation occurring during photosynthesis when carbon is fixed into algal cellular biomass (δ13C_org) from the ambient aqueous CO₂ (δ13C_CO2aq) (Freeman and Hayes, 1992): 5εp=δ13C[CO2]aq+1000(δ13Corg+1000)-1⋅1000 Organic δ13C is obtained from the δ13C analysis of haptophyte specific alkenone di-unsaturated C_37.2. Culture experiments showed that the lipid organic matter is depleted in ¹³C relative to the whole cell isotopic composition by 4.2 ‰, a correction that needs to be applied (Wilkes et al., 2018; Popp et al., 1998): 6δ13Corg=[(δ13C37:2+1000)⋅((4.2/1000)+1)-1000] Uncertainties were propagated by a full Monte Carlo (n= 10 000) simulation following Tanner et al. (2020).

To compare our new records with previous data spanning the same time interval, we discuss published εp datasets recently compiled by the paleo CO₂ community (CenCO2PIP Consortium, 2023), from DSDP 516 in the South Atlantic (Pagani et al., 2000, 2005, 2011), DSDP 608 in the North Atlantic (Super et al., 2018), and the equatorial site from ODP 925 (Zhang et al., 2013). For these, we ensure that εp for the published records is calculated from biomarker-based paleothermometers. The most recent publications from DSDP 608 and 925 used glycerol dialkyl glycerol tetraethers (GDGT) derived estimations from TEX₈₆. To better compare our results with DSDP 516, where originally temperatures were derived from δ18O of planktic foraminifera for the Miocene section and GDGTs for part of the Oligocene, we have updated the εp calculations using a running averaging of the recent higher resolution GDGT temperature reconstructions from Auderset et al. (2022) at the same site.

For our data of paired εp and alkenone SST, we calculate the shift in εp which is expected from temperature-stimulated growth rates. Using the relationship of -0.48 ‰ (95 % CI = -0.37 to -0.95 ‰) per 1 °C SST (Torres Romero et al., 2024), we adjusted each samples εp absolute value by using the difference between the SST estimated for that sample and the average SST during the studied interval at that site. We complete a similar exercise for cell radius, calculating the deviation in εp only relative to the median cell size, for each point using the culture dependence of εp on cell radius shown in Eq. (2) (Stoll et al., 2019). Biogenic silica (bioSi) was determined on 20 samples from ODP 1168 following methods described previously in Guitián et al. (2020).

From the εp time series we estimate the change in CO₂ relative to the maximum values at 29 Ma, using the adjustment in εp for temperature sensitive growth rate described in the previous paragraph, and Eq. (3) as applied in González-Lanchas et al. (2021), where I reflects the size and light influences on εp and is assumed constant across all time intervals, and the εp dependence on ln [CO₂[aq]] of 2.66 is the 50th percentile estimate of the modern cultures. We then estimate the doubling/halving of CO₂ relative to the CO₂ at the reference age (R) applying the solubility for the measured temperature (Zeebe and Wolf-Gladrow, 2001) which can be reduced to: 7doublingCO2=εpt-εp(R)2.66ln2+log2sol(R)sol(t)

In addition to CO₂, εp may be influenced by changes in cell size and cellular growth rate regulated by light and nutrients. There is no long-term trend in mean coccolith size in these records (Guitián et al., 2020) and estimating the impact of it on the εp records shows a negligible effect on long-term trends (Fig. 5). At discrete time intervals of IODP Site 1406, the correction for the size effect reduces most values older than 27 Ma and produces a steeper decrease on εp towards the Oligocene Miocene transition.

For the statistical model of Eq. (2), it is complex to identify proxy records for any possible effect of nutrient-stimulation of growth rate or changes in the mean light conditions at the depth of growth. In modern spatial gradients in the ocean, these factors are often coupled, so that settings characterized by deep mixing and high nutrient supply rates to stimulate growth, are also characterized by lower mean light levels due to the deeper mixing, both factors lowering εp (Hernández-Almeida et al., 2020).

As one possible nutrient indicator, a higher concentration of biogenic silica (bioSi) in sediments may reflect a higher rate of bioSi delivery to the seafloor due to higher export production produced by siliceous organisms (mainly diatoms) in the ocean (Ragueneau et al., 2000). In the modern ocean, regions with abundant dissolved Si in the photic zone are regions also characterized by higher concentrations of macronutrients such as P and N. However, bioSi is an imperfect indicator of past surface nutrient content because coccolithophores have a minimal Si requirement, and Si remineralization in the ocean does not occur at the same rate as soft-tissue nutrients such as N and P. At IODP 1406, bioSi concentrations generally increase from the Oligocene to earliest Miocene, potentially indicating a gradual increase in the concentration of dissolved Si in surface waters at the site (Fig. 5). If the increase in dissolved silica observed at the North Atlantic is correlated to an increase in dissolved P or N, it could contribute to increase in growth rate, and therefore likely increase in biomass and chlorophyll, which would reduce light in the water column, both factors potentially contributing to the observed long term decrease of εp. However, the actual correlation between bioSi and εp is not that strong (Fig. S2), suggesting that while increased nutrient concentrations could contribute to the long-term evolution of εp, the specific steps of εp decline are less likely to be driven by increased nutrients and growth rate.

The drivers for increasing bioSi burial rates at Site 1406 are not clear. They could reflect a global increase in nutrient delivery or local effects. Important changes in the rate of continental weathering within the Oligocene- early Miocene are often interpreted from the evolution of radiogenic isotopes of Sr, Li and Os (Misra and Froelich, 2012) including the steep rise in 87Sr/86Sr, although the precise origin of the late Eocene and Miocene increase in 87Sr/86Sr remains under discussion (Rugenstein et al., 2019). On a global scale, the nutrient delivery may be conditioned by the riverine supply of P from continental erosion and weathering of P containing minerals. Yet, on the time scales examined in our records, much longer than the residence time of P, the net effect on nutrient concentrations depends on the balance of the supply and the nutrient removal in sediments.

While a significant increase in erosion and weathering and nutrient inventory is one mechanism to contribute to the long term decline in εp via enhanced algal growth rates, an increase in erosion and weathering can itself contribute to a CO₂ drawdown by CO₂ consumption through silicate weathering and enhanced burial of organic carbon in delta regions (Raymo and Ruddiman, 1992). If the biogenic Si increase at 1406 were representative of a global trend, an increase in nutrient supply may have contributed to εp decline through both CO₂ decline and increased nutrient stimulation of phytoplankton growth. A global decline in εp solely from increased weathering and nutrient concentrations without a CO₂ decline would require that in the Oligocene, the nutrient release from silicate weathering was less coupled to carbon burial than in the late Neogene. If the periodically glaciated margin of Antarctica is a major locus for increased erosion and weathering in the Oligocene (Reilly et al., 2002), release of nutrients and radiogenic isotopes may have occurred in the continental margins, but with much less organic carbon burial than the modern Himalaya system due to limited terrestrial biomass on Antarctica and temperature and sea-ice limited oceanic biomass production in the marine regions.

On the other hand, the long term trend of increased bioSi is not observed in the Southern ocean Site 1168 (Fig. 5). The available Miocene bioSi at ODP 1168 is stable with no change across the steep εp drop from the latest Oligocene to early Miocene. The increasing distance of Site 1168 from the coastline with basin subsidence may have decreased the availability of Si from the early Oligocene through the early Miocene, imparting a local effect superimposed on any potential global trend. However, likely not only Si but also other nutrients would decrease with increasing distance from the coast. If the long term trend in εp at both 1406 and 1168 sites were conditioned by increased nutrient availability, faster growth rates, and lower light levels it would require bioSi accumulation rates at Site 1168 to be decoupled from the overall changes in nutrient availability, which we consider less likely. Consequently, we propose that the similarity in trend and magnitude of the long term εp decline in both sites (and in tropical Site 925), is more consistent with a global forcing of εp, which may be most plausibly driven by a significant decrease in atmospheric CO₂ and CO₂aq.

The new εp data from Site 1406 and Site 1168 provide the first records of εp from the early Oligocene to early Miocene with alkenone unsaturation indices as independent estimations of SST for the precise time intervals of εp determination. Since they are biomarkers derived from the same organism, alkenone-derived SST estimates correspond to the same season and growth depth as the alkenone εp determinations. There are two processes which may influence the relationship between temperature and εp. First, higher temperatures lead to higher phytoplankton carbon fixation rates, decreasing εp. Secondly, higher CO₂ would increase εp and through radiative forcing lead to warmer global average temperature and SSTs. The expected relationship between εp and SST from either process could be obscured by a superposition of temperature effects on growth rate and a climatic correlation of εp with mean air temperature.

The long term trends between 30 and 16 Ma based on new εp data at two sites and recalculation of previous εp studies with uniform methods cannot be attributed to a temperature effect on growth rate and εp, nor to changes in the cell size of the alkenone producing community. Both effects are small in magnitude according to the sensitivities observed in cultures and do not alter the long-term trend (Fig. 5). Therefore, the long-term εp decline must have a significant global driver, with the most obvious being a decline in pCO₂.

Although the calculation of absolute CO₂ concentrations from εp in the Oligocene and early Miocene remains challenging, the logarithmic dependence of εp on CO_2[aq] observed in cultures allows us to estimate the relative changes in CO₂ if the sensitivity of εp to CO₂ in the Oligocene were similar to modern cultured species using Eq. (7). If we incorporate a temperature correction and apply the 50th percentile estimate of the modern culture εp dependence on ln [CO_2[aq]] of 2.66, it implies major changes in CO₂ concentrations, with potentially 4 halvings of CO₂ concentration from 29 to 16 Ma (Fig. 8). Modern General Circulation Models (GCM) summarized by the IPCC estimate climate sensitivity as “very likely” in the range of 2 to 5 °C per doubling or halving of CO₂ (Forster et al., 2021), which if representative for the Oligocene to early Miocene, would imply 8 to 20 °C of cooling of earth's mean surface temperature (6 to 15 °C incorporating the lower confidence interval of modern culture εp dependence on ln [CO_2[aq]] of 3.5, which would imply 3 halvings of CO₂). Although ocean is 70 % of the globe and temperature changes are around 1.5-fold less than land temperature (Sutton et al., 2007), such a temperature change of at least 6 °C would be expected to be reflected in paleoceanographic proxies.

Similar to phytoplankton proxy records, the available low resolution leaf gas CO₂ records suggest a decline in CO₂ from the mid to latest Oligocene. However, in contrast to phytoplankton proxy records indicating a significant long term decline in CO₂ from the early Oligocene through mid-Miocene, leaf gas CO₂ proxies suggest higher CO₂ in the early Miocene than the Oligocene due to a positive shift across the OMT. Boron isotope-based CO₂ records from 24 to 18 Ma show significant variability with no clear trend, although the higher density of data around the OMT suggests a CO₂ rise from 23 to 20 Ma which may be consistent with the trend observed in the εp record at Site 1406, which has the highest resolution for this time interval.

The late Oligocene climate and CO₂ paradox has been discussed based on previously published lower resolution εp record from Site 925 (O'Brien et al., 2020). Our new results from two additional sites confirm the steep CO₂ decline through the late Oligocene warming and underscore the paradox. On a global scale, biomarker SST estimates do not show evidence for systematic cooling during the CO₂ decline (Liu et al., 2018; O'Brien et al., 2020; Guitián et al., 2019). If the interpretation of εp as a CO₂ decline is correct, it suggests that climate sensitivity was either significantly weaker so that no appreciable change in global mean surface temperature occurred, or that available paleotemperature estimates reflect a significant misinterpretation of measured biomarker signals. During this time the inferred CO₂ decline also coincides with sequence stratigraphic evidence for ice margin retreat in Antarctica (Levy et al., 2019; Salabarnada et al., 2018), and sea level transgressions inferred from estimates of deep sea δ18O_sw and Mg/Ca records (Miller et al., 2020), suggesting a different relationship between ice expansion and CO₂ than characterized the late Neogene.

For the late Oligocene to early Miocene, the Southern Ocean Site 1168 is the only surface ocean temperature record which exhibits a colder early Miocene coincident with the record of a large magnitude pCO₂ decline. This long term cooling is despite the equatorward drift of the site over this time interval (Guitián and Stoll, 2021). If the ODP Site 1168 temperature trend is more representative of global average temperature trends, whereas the long term alkenone temperature record at Newfoundland Ridge Site 1406 and Site 1404 (Liu et al. 2018) is dominated by variations in the heat transport from the Gulf Stream, then the 1168 temperature trend may reflect the signal of radiative greenhouse forcing. Yet, temperature trends at either site may be subject to both global factors as well as regional temperatures, and with only two sites with temperature records paired to εp proxy records it is difficult to ascertain which, if any, of the sites may better reflect global temperature forcing.

A decoupling was at one time proposed for the late Miocene based on apparent negligible pCO₂ change and substantial cooling of SST (LaRiviere et al., 2012). Revisions of the alkenone carbon fractionation to CO₂ calibration approaches for low pCO₂ periods have refined the record from the last 15 Ma, revealing clear pCO₂ -SST covariation (Stoll et al., 2019; Rae et al., 2021). However, the Oligocene paradox is not easily resolvable from updated calibration of the εp-CO₂ relationship. The late Oligocene paradox arises from an inverse correlation between εp and SST reconstructions in regions other than the Southern Ocean such as the North Atlantic, and a lack of correlation between εp and the global climate signal in benthic δ18O trends. The discrepancies between alkenone and published TEX₈₆ at ODP 1168 suggests continued reevaluation of SST proxy interpretation are needed, along with evaluation of the potential influence of changing surface ocean circulation on SST in some locations such as the North Atlantic. Additionally, the divergence of CO₂ trends among εp and boron isotopes suggest that further interrogation of ocean chemistry and biogeochemical cycles potentially affecting the growth and physiology of alkenone producers and the calculation of CO₂ from boron isotopes, are crucial to reconcile climate sensitivity to CO₂ in the Oligocene to early Miocene.