The Pliocene epoch (5.33–2.58 million years ago, Ma) is an interval with particular relevance to understanding Earth's modern climate, as it is the most recent time in Earth's history that the atmospheric carbon dioxide (CO2) concentration (pCO2) has been comparable to today's (>400 parts per million, ppm) (NOAA Global Monitoring Laboratory, 2025). Because of the Pliocene's elevated CO2 levels and its similar continental configuration to today's Earth, this epoch, and especially the Mid-Piacenzian Warm Period (mPWP, ∼3.3–3.0 Ma), is often considered to be a paleo-analog for near-future climate conditions under moderate warming scenarios (Burke et al., 2018). During this time, mean annual surface temperatures were 2–4 °C warmer, sea levels were elevated, and ice sheet extent was reduced compared to preindustrial conditions, particularly in the northern hemisphere (Haywood et al., 2013).

Better constraining Pliocene climate is additionally important to strengthening our understanding of the conditions that allowed large, continental-scale ice sheets grow in the Arctic region at the very end of the Pliocene. Over the past ∼50 Myr since the early Eocene Climatic Optimum, Earth has experienced a long-term cooling trend and transitioned from a warm greenhouse climate state to an icehouse, in which the high latitudes are covered by large ice sheets. Although Antarctica experienced continental-scale glaciation since the early Oligocene (ca. 34 Ma) (Lear et al., 2008; Zachos et al., 2001), large ice sheets did not start to grow in the northern high latitudes until the intensification of Northern Hemisphere glaciation (iNHG) ca. 2.72 Ma, as evidenced by the appearance of ice-rafted debris (IRD) in North Atlantic marine sediments (Blake-Mizen et al., 2019; Flesche Kleiven et al., 2002; Jansen et al., 2000; Thiede et al., 2011).

Attempts to identify the primary driver(s) of the intensification of Northern Hemisphere glaciation have generated several plausible hypotheses. Some researchers have postulated that tectonic processes, like the closure of the Panama Seaway, could have led to glaciation by intensifying the Gulf Stream, increasing moisture transport to the high latitude north, and strengthening the formation of North Atlantic Deep Water (NADW) (Haug and Tiedemann, 1998; Lunt et al., 2008b). However, tectonic processes were likely too slow to cause iNHG, and models have found that the closure of the seaway likely was not a major forcing mechanism for late Pliocene glaciation; instead, models suggest that a decrease in atmospheric CO2 concentrations to 280 ppm was needed to drive glaciation on Greenland (Lunt et al., 2008a, b; Tan et al., 2017).

This hypothesis – that a reduction in greenhouse forcing lowered global temperatures and drove the iNHG – is supported by a growing body of literature. Paleoclimatic proxy data suggests that atmospheric CO2 decline coincided with late Pliocene glaciation (Rae et al., 2021; de la Vega et al., 2020) and that global cooling was occurring in all ocean areas other than the Indo-Pacific warm pool (IPWP) around 4 Ma and continuing through the iNHG (Clark et al., 2024). This global cooling signal is consistent with climate models, which have shown that reducing CO2 levels in the atmosphere would result in cooling across both the northern and southern hemispheres (Broccoli and Manabe, 1987; Stap et al., 2018).

One other prominent hypothesis exists to explain the primary driver of iNHG: that regional climate change in the Southern Hemisphere ultimately drove ice sheet growth in the north. In this scenario, growth of the Antarctic ice sheet would fuel the positive ice-albedo feedback effect, leading to regional cooling concentrated around the high-latitude Southern Hemisphere (McKay et al., 2012). Past a certain threshold, ice sheet expansion could have altered global ocean and atmospheric circulation, culminating in global temperature reductions large enough to lead to Northern Hemisphere glaciation (Woodard et al., 2014).

The alkenone-based reconstruction of Ocean Drilling Program (ODP) Site 1090's sea surface temperature (SST) of Martínez-Garcia et al. (2010a, b) is a critical piece of evidence informing this hypothesis (Fig. 1). This record features cooling over parts of the late Pliocene that outpace much of the rest of the world, a possible indication of high-latitude Southern Hemisphere climate cooling leading up to iNHG, and it has been utilized in several studies positing that regional forcing could have played a significant role in driving iNHG (McClymont et al., 2023; Woodard et al., 2014; Burton et al., 2024).

Two aspects of the Martínez-Garcia data set appear anomalous, necessitating a re-examination of this alkenone-based SST record. The first of these is the offset between this record and an alkenone-based record produced by Etourneau et al. (2010a, b), also at Site 1090, during the 800 ka leading up to iNHG, with the Etourneau record indicating SSTs up to 5 °C warmer than Martínez-Garcia et al. (2010a) (Fig. 2). The second aspect is the unusually high C37:4 content reported by Martínez-Garcia et al. (2010a). While the C37:4 alkenone is associated with the presence of sea ice (Liao et al., 2023b; Wang et al., 2021), the Martínez-Garcia et al. (2010a) reconstruction reveals only moderate (>7 °C) temperatures in the Pliocene, suggesting the need for an investigation to explain this anomalous result. We show below that improved analytical methods lead to significantly different results from data published by Martínez-Garcia et al. (2010a).

The location of ODP Site 1090 (42°54.5^′ S, 8°54.0^′ E, Fig. 1) is optimal for studying a variety of oceanographic and climatic phenomena through time, including Antarctic ice-sheet development, Antarctic Circumpolar Current (ACC) frontal boundary movements, and changes in the contribution of NADW to the Southern Ocean (Shipboard Scientific Party, 1999). Located on the southern flank of the Agulhas Ridge at a water depth of 3702 m, ODP Site 1090 contains one of the most southerly records recovered of late Pliocene/early Pleistocene climate. Today, the site sits within the sub-Antarctic Zone of the ACC, between the Subantarctic Front (SAF) and Subtropical Front (STF), and north of the southernmost Polar Front (PF) (Reid, 1989). The SST of Site 1090 has been strongly influenced by meridional shifts in the ACC, driven by the northward migration of the PF, SAF, and STF. Biomarker, isotope, and microfossil data suggests these fronts may have migrated anywhere from ∼5° equatorward during the Mid-Pleistocene Transition (de Castro et al., 2025) to upwards of 7° latitude during cold stadials in the Pleistocene, resulting in SST changes of up to 6 °C between glacial/interglacial periods (Bard and Rickaby, 2009; Becquey and Gersonde, 2002).

Five holes were drilled at ODP Site 1090. The composite of the cores drilled from these holes produces a continuous stratigraphic section to ∼4 Ma, preceded by a late Miocene/early Pliocene hiatus (Shipboard Scientific Party, 1999, Billups et al., 2004). Sedimentation rates estimated by the Shipboard Scientific Party ranged from 10–35 mMyr-1 prior to the late Miocene/early Pliocene hiatus (1999).

Alkenone paleothermometry is a well-calibrated and reliable paleo-proxy for open-ocean SST (Herbert, 2003; Müller et al., 1998). Alkenones – lipid biomarkers produced by certain species in haptophyte class Prymnesiophyceae – are 37-, 38-, and 39-carbon chained (C37, C38, C39) ketones (Marlowe et al., 1984). The relative unsaturation of the C37 ketone, represented by the U37K′ and U37K indices (Table S1), linearly decreases with increasing temperature in all but the warmest (>24 °C) ocean waters (Müller et al., 1998; Tierney and Tingley, 2018). Because of this relationship, the global distribution of alkenones, and the exceptionally high preservation potential of alkenones, they are one of the most highly utilized proxy biomarkers for near-surface temperatures in the past open ocean (Prahl et al., 1988; Sikes et al., 1991). In this study, we utilize the global core-top based calibration of Müller et al. (1998) to derive SST. Because residuals between core-top U37K′-SST and satellite mean annual SST are miniscule in the southeast Atlantic, we expect U37K′ to act as a proxy of mean annual temperature at ODP 1090 (Tierney and Tingley, 2018).

We sampled the sediment core from ODP Site 1090 approximately every 5 cm, beginning at ∼50 m below the seafloor (MBSF, hole D), corresponding to about 55.3 m composite depth (MCD), and ending at ∼61 MBSF This sample range and resolution was selected to overlap with and extend the Martínez-Garcia et al. (2010a) late Pliocene SST record, spanning from the end of the Pliocene ca. 2.58 Ma to the hiatus in the sediment core at an estimated 4.3 Ma which prematurely ends the Pliocene record.

Sediment samples were frozen overnight and freeze-dried for 24 h prior to homogenization using a mortar and pestle. Approximately 8–10 g of sediment were mixed with sand (previously baked to prevent contamination by organic compounds) to improve solvent flow and packed into 34 mL cells with glass fiber filters. Alkenones were extracted from samples using 100 % Dichloromethane (DCM) in an Accelerated Solvent Extraction system (Dionex ASE 350). Every round of 24 samples extracted on the ASE included one blank extraction of pre-combusted sand to check that no contamination was introduced during the extraction process.

Although the calibration equations chosen for both studies are nearly identical (Tables 1 and S1), they differ in one key aspect: Martínez-Garcia et al. (2010a, b) utilized the U37K alkenone unsaturation index (Brassell et al., 1986), which includes the tetra-unsaturated alkenone (C37:4): 1U37K=[C37:2-C37:4][C37:2+C37:3+C37:4] However, Etourneau et al. (2010a) used the U37K′ index (Prahl et al., 1988), which excludes C37:4: 2U37K′=[C37:2][C37:2+C37:3] Sites where C37:4 is present in the largest quantities experience at least seasonal sea ice cover – unlike ODP Site 1090 – leading some to suggest this ketone may have utility as a sea ice proxy but perhaps should not be included as a temperature proxy specifically (Liao et al., 2023b; Wang et al., 2021).

Extracted samples were dried in a TurboVap blowdown evaporator at 35 °C under constant flow of N2. Samples were rehydrated in DCM and glass pipetted into 4 mL vials. After initial gas chromatography with flame ionization detection (GC-FID) results conducted on the total lipid extract yielded chromatograms with noisy baselines, likely due to other organic residues that can co-elute with the ketones we seek to measure, we conducted compound purification. First, we conducted polar-apolar fractionation via flash column silica gel chromatography using hexanes and DCM as solvents. This was followed by a silver nitrate-impregnated silica gel column chromatographic procedure for unsaturation-saturation separation, which utilized DCM and ethyl acetate as solvents (D'Andrea et al., 2007). Example chromatograms showing GC-FID analysis before and after this cleanup procedure can be found in the Supplement (Fig. S1). After cleanup, samples were again dried at 35 °C under a mini evaporator before being spiked with 210 µL of an internal lab standard (hexatriacontane [C36 alkane] and heptatriacontane [C37 alkane], 2 µgmL-1 in toluene). 50 µL of each sample extract was transferred to its own 2 mL glass vial with a 300 µL insert. Samples were capped and kept refrigerated until ready for alkenone measurement.

There are multiple challenges that must be considered to ensure reliable analyses at ODP Site 1090: low biomarker concentrations and a complex sediment matrix. While total C37 alkenone concentrations (C37total) are highly variable throughout the global ocean, it is often present in concentrations >0.1 µgg-1 sediment; at sites with very low C37total, such as ODP Site 1090 (where our average C37total was ∼0.02 µgg-1 sediment), it can prove difficult to differentiate the biomarker signal from coeluting compounds and baseline noise resulting from other matrix components present in the sediment extract using the typical GC-FID method. Therefore, there is a need to not only attempt to reproduce Martínez-Garcia et al. (2010a)'s SST record, but also to unambiguously assess whether the C37:4 ketone is present in measurable quantities at ODP Site 1090. To ensure the precision and accuracy of our results, we not only measured alkenone concentrations using traditional gas chromatography (GC) techniques but also confirmed our findings using high performance liquid chromatography (HPLC). Sixteen samples were randomly selected to undergo both GC-FID and HPLC analysis.

Each collected datum was individually checked for quality control in three ways. The total measured concentration of the C37 ketone (C37 total) was found for each sample; data from samples with undetectable or especially low C37 total (<0.0012, |mean-stdev|) were removed from the final SST record. We calculated the U37K′ height to area ratio of the GC peaks, as large (>10 %) deviations from the peak area-based index can indicate co-elution of ketones with other compounds, and samples found to have large height deviations from area were also removed. Finally, we compared the U37K′ area to the U38MEK (the 38-carbon chain-length methyl-ketone based unsaturation index) area, which follow a linear relationship in modern sediments (Novak et al., 2022) as a final check to identify anomalous analyses. In total, seven samples were rejected from the final SST reconstruction, which consists of 352 samples.

Previous work by Martin et al. (2024) has found that silver-nitrate purification may modify the C37 alkenone distribution and thus the unsaturation indices, especially the U37K index, by preferentially retaining alkenones with more double bonds. We conducted additional analyses to determine whether and to what extent our column chromatography-based cleanup methods may be biasing our measurements of 37-, 38-, and 39-carbon chained ketones. Approximately 1 mL of our in-house alkenone standard was pipetted into eight 4 mL vials and evaporated under a constant N2 stream on a hot plate set to 35 °C. Six vials of this standard were reconstituted in 1 mL of hexane before undergoing the previously described cleanup procedures; the additional two vials were left as a control. This procedure was also repeated using the Greenland standard (D'Andrea and Huang, 2005), which, unlike the in-house standard, contains a large quantity of C37:4 methyl-ketones.

Experimental evaluation showed little to no evidence of cleanup biasing our results. The average SST of all experimental samples (N=13) – our typical lab standard (n=5), the evaporated and re-constituted standard (n=2), and the cleaned standards (n=6) – were within a few hundredths of a degree of each other, both using our typical DB-1 method and utilizing the 10 µL sample volume required by ODP Site 1090 alkenone measurements (Fig. S2). Our experimental results using the Greenland standard yielded similar results, with negligible change in our derived SSTs following silica gel and silver nitrate cleanup (Fig. S3). We performed a paired t test (two-tailed) to formally evaluate whether SST estimates before and after the cleanup procedure differ significantly and confirmed that there is no significant difference: our in-house standard samples produced a p value of 0.68, and the Greenland standard yielded a result of 0.93. While this experiment was conducted using standards and is not a direct comparison to the sediment's chemical environment, the consistent behaviour of the alkenones across both standards suggests that, when executed correctly, the cleanup procedures are unlikely to significantly impact the alkenone unsaturation index of our samples.

In their study, Martínez-Garcia et al. (2010a, b) elected to utilize the culture-based temperature-U37K calibration of Prahl and Wakeham (1987): 4U37K=0.033(SST)+0.043 To establish a more direct comparison between our data, we estimated the approximate SST Martínez-Garcia et al. (2010a) would have reported had they utilized the simplified unsaturation index (U37K′). We first inferred their U37K from their SST data, and then we used their reported %C37:4 to estimate U37K′: 5inferredU37K′=U37K+(%C37:4×0.01)1-(%C37:4×0.01) Then, we applied the SST calibration of Müller et al. (1998; Eq. 3) to reconstruct temperature from our estimated values of U37K′.

Compared to the SST record of Martínez-Garcia et al. (2010a), our SST estimates record only a gentle cooling trend leading up to iNHG (Fig. 3), made apparent by the gentler cooling and warmer SSTs evident in the 400 kyr moving average of our reconstruction, in addition to the gentler slope of our linear regression. In addition, both our new high-resolution dataset and the lower-resolution SST data of Etourneau et al. (2010b) indicate that SSTs were on average 1.22 °C warmer during the late Pliocene at ODP Site 1090 than Martínez-Garcia et al. (2010a) estimate. Minimum temperatures in the interval spanning ca. 2.8–3.2 Ma are often 1–2 °C warmer in our record, but some differences were even greater – at one comparable depth interval, the SST difference between records exceed 6 °C. These larger sample-specific differences suggest that the offset between our records cannot be attributed solely to calibration uncertainty or inter-laboratory effects, and that the observed SST differences are analytically meaningful. There is little to no indication of anomalously cold events at ODP Site 1090 uniquely preceding iNHG (Figs. 3 and 4). We believe that most of the differences between data sets result from the very difficult sample matrix presented by Site 1090 and the optimization of sample purification and strategic use of HPLC in the present study.

Both our GC-FID and HPLC results indicate that C37:4 was not present in reliably measurable quantities in ODP Site 1090 sediments. As C37:4 is present only in low concentrations in most marine sediments, including those of Site 1090, its measurement is subject to disproportionate risk of overestimation due to co-eluting compounds and baseline noise (Brassell et al., 1986). The overutilization of less abundant compounds, such as the tetra-unsaturated alkenone, in proxy calibrations can dampen trends and obscure important relationships, risking the introduction of bias in analyses (Raberg et al., 2021). Therefore, it is advisable to positively identify measurable C37:4 in marine sediments before utilizing the U37K index in preference to the U37K′ index.

Our record of late Pliocene ODP Site 1090 SSTs does not show a cooling trend indistinguishable from those at other open ocean sites, including those in the Northern Hemisphere and low latitudes (Fig. 4). In fact, when comparing the Site 1090 SST record with those of ODP Site 846 (eastern equatorial Pacific; Kimble et al., 2024; Herbert et al., 2020) and IODP Site U1313 (northern Atlantic; Naafs et al., 2020), the cooling trend is the most gradual at ODP Site 1090; while the slopes of the best fit lines of SSTs from ODP 846 and IODP U1313 are 0.0011 and 0.0019 °Ckyr-1, respectively, it is only 0.0009 °Ckyr-1 at ODP Site 1090. This more gradual cooling trend is likely influenced by the more variable SSTs observed at ODP Site 1090, particularly prior to 3600 ka, which dampens the expression of long-term trends, especially in SST reconstructions where we expect, and observe, non-monotonic cooling. Notably, the cooling trend observed at IODP Site U1313, in the North Atlantic, is approximately double that of the other sites – a possible indication that more Northern Hemisphere cooling preceded iNHG, which challenges the hypothesis that Southern Hemisphere cooling was a primary driver of northern glaciation (McKay et al., 2012; McClymont et al., 2023; Woodard et al., 2014).

In addition, the relatively muted cooling at Site 1090 likely reflects regional oceanographic controls, rather than a simple response to large-scale cooling, as its location within a frontal transition zone makes it particularly sensitive to changes in ocean circulation. Site 1090 (∼43° S) lies between the SAF and STF, where SSTs are highly sensitive to frontal migration, wind-driven circulation, and water-mass mixing. Latitudinal shifts in Southern Ocean fronts associated with changes in westerly winds and meridional temperature gradients during the late Pliocene (e.g., Martínez-Garcia et al., 2010b; Hillenbrand et al., 2009) could cause the site to experience alternating influence of subtropical and subantarctic waters. Such variability would act to buffer or modulate the local expression of global cooling, producing a more gradual and variable SST trend compared to regions more directly coupled to large-scale radiative and thermocline forcing. Furthermore, Site 1090 is located just north of the SAF in the modern ocean (Fig. 1); had the location of this front been more poleward in the warmer-than-modern Pliocene (Hillenbrand et al., 2009; McClymont et al., 2016), Site 1090 may still have been located in the envelope between the SAF and STF throughout this transition period, potentially buffering regional SSTs from the global trend.

The SST record of ODP Site 1090 shares orbital-scale features with records from other regions of the global ocean, highlighting its relevance within the broader context of global climatic patterns (Fig. 4). To further explore these shared characteristics, spectral analyses were conducted using the multitaper method Fourier transform. This analysis identifies the dominant periods of variability in the SST records of ODP Site 1090, ODP Site 846, and IODP Site U1313, enabling assessment of the commonality of these orbital-scale features at the 95 % confidence interval (Fig. 5). Our spectral analysis reveals that all three sites had dominant periodicities in the obliquity band (41 kyr cycles). Both ODP 846 and 1090 showed significant power in the precession bands (23 kyr cycles), but no site showed power in the eccentricity band (100 kyr cycles) that reached the 95 % significance threshold. Coherence between the SST records of each site and the average eccentricity, obliquity (tilt) and precession (ETP) is strongest in the obliquity band for all sites (Fig. 5). This demonstrates that Earth's changes in obliquity acted as a significant control on temperature globally during the late Pliocene.

The revised SST record of ODP Site 1090 supports the hypothesis that global SST records displayed only a modest long-term cooling trend throughout the late Pliocene. Intriguingly, interglacial SST values of the late Pliocene were almost as high as in the preceding million years (Fig. 3). This contrasts with the suggestion that asynchronous global cooling, caused by a regional temperature forcing like the ice-albedo feedback in the Antarctic, led to the iNHG (e.g., McKay et al., 2012; McClymont et al., 2023; Woodard et al., 2014). Had an orbitally paced, regional forcing caused the expansion of the Antarctic ice sheet and driven global cooling leading to the iNHG, we would expect to see cooling signals unique to the ODP 1090 record compared to more northerly sites. However, all sites' SST reconstructions reveal similarly paced warm- and cool-intervals (Fig. 4), hinting that a global mechanism drove this variability in temperature.

Our results are consistent with a significant body of compelling research supporting the hypothesis that synchronous cooling, driven by decreasing atmospheric carbon dioxide concentrations, allowed for Northern Hemisphere ice sheets to grow under favorable orbital configurations. Earth system models have demonstrated that lowered atmospheric CO2 would have led to synchronous global cooling (Lunt et al., 2008a; Willeit et al., 2015). Similarly, a recent synthesis of global paleo-temperature records showed that changes in long-term cooling and variability during the past 4.5 million years is likely paced by geologically driven changes in Earth's carbon cycle (Clark et al., 2024). DeConto et al. (2008) utilized isotope-capable global climate/ice-sheet models to show that the atmospheric CO2 threshold below which glaciation occurs is higher (∼750 ppm) in the Southern Hemisphere compared to the high latitude north (∼280 ppm). Other researchers similarly found that lowering atmospheric CO2 concentrations to 280 ppm controlled the late Pliocene glaciation of Greenland in a fully coupled atmosphere–ocean general circulation model and an ice-sheet model (Lunt et al., 2008a). Proxy data confirms that this carbon dioxide threshold likely was not crossed until the late Pliocene, ∼2.7–2.8 Ma, and that decreasing levels of atmospheric CO2 in the late Pliocene likely corresponded with cooling global temperatures (Bartoli et al., 2011; de la Vega et al., 2020; Stap et al., 2018). Similarly, recent analyses of global temperature change during the past 4.5 million years by Clark et al. (2024) show that no region of the ocean displays an accelerated degree of cooling from the mean global cooling trend occurring throughout the Pliocene; this pattern is apparent even with the inclusion of the previously published, likely erroneously cold SST estimates from ODP 1090 (Martinez-Garcia et al., 2010b).

Analyses using models from Phase 2 of the Pliocene Model Intercomparison Project (PlioMIP2) and paleoclimate proxy data from the PlioVAR working group conducted by Burton et al. (2024) indicate that at most sites, late-Pliocene SST changes were forced by CO2. Although they also utilize the SST record by Martínez-Garcia et al. (2010a), they nevertheless found that changes in SST at ODP 1090 were found to be primarily driven by CO2 forcings, and non-CO2 forcings were dominant only at North Atlantic sites. However, the researchers also note that there is a significant, negative relationship between FCO2 (the proportion of the total Pliocene minus preindustrial climate change that is due to CO2 forcing; Burton et al., 2023) and data-model agreement at ODP 1090, with this agreement ranging from 0.35 °C (CESM2) to -2.98 °C (NorESM1-F) with a multi-model mean of -1.63 °C, indicating that the paleo-record suggested temperatures were on average colder than the model output (Burton et al., 2024). Using our revised SST estimates, which are on average ∼0.8 °C warmer than the Martínez-Garcia et al. (2010a) record over MIS KM5c (3.205±0.01 Ma), reduces, but does not eliminate, the data–model discrepancy at this site. The magnitude of disagreement would be approximately halved relative to that reported by Burton et al. (2024). We emphasize, however, that Site 1090 is located in a region strongly influenced by Southern Ocean frontal dynamics and latitudinal shifts of the Antarctic Circumpolar Current. Such processes are unlikely to be fully resolved in model simulations and could contribute to both the relatively high variability in the proxy record and persistent model–data offsets. Nevertheless, our results highlight the importance of accurately constrained regional proxy records for evaluating model performance in dynamically complex settings.

Why might atmospheric CO2 concentrations have fallen to levels that allowed brief Northern Hemisphere glaciations throughout the Pliocene, and later the extensive growth of Northern Hemisphere ice sheets? Changes in high-latitude oceanographic processes tied to the global carbon cycle may have driven both global temperature changes and glaciation (Herbert et al., 2010). Reconstructions of neodymium isotope ratios and carbonate ion saturation indicate that ocean circulation shifted in the late Pliocene, providing more water to the Pacific sourced from the Southern Ocean. They also suggest that respired carbon storage into the deep Pacific increased during the late Pliocene, possibly due to strengthened deep-water formation and biological-pump efficiency in the Southern Ocean due to the expansion of sea-ice (Jian et al., 2023). Continued proxy-studies regarding carbon sequestration, carbonate production, and the biologic pump in the late Pliocene may help to further explain how atmospheric CO2 levels changed leading to iNHG.