Thermocline state change in the eastern equatorial Pacific during the late Pliocene / early Pleistocene intensification of Northern Hemisphere glaciation

The late Pliocene/early Pleistocene intensification of Northern Hemisphere glaciation (iNHG) ∼ 2.5 million years ago (marine isotope stages, MIS, 100–96) stands out as an important tipping point in Earth’s climate history, which strongly influenced oceanographic and climatic patterns including trade wind and upwelling strength in the eastern equatorial Pacific (EEP). The thermocline depth in the EEP, in turn, plays a pivotal role in the Earth’s climate system: small changes in its depth associated with short-term climate phenomena such as the El Niño–Southern Oscillation can affect surface-water properties and the ocean–atmosphere exchange. However, thermocline dynamics in the EEP during the iNHG still remain unclear. While numerous studies have suggested a link between a thermocline shoaling in the EEP and Northern Hemisphere ice growth, other studies have indicated a stable thermocline depth during the iNHG; consequently, a causal relationship between thermocline dynamics and ice-sheet growth has been excluded. In light of these contradictory views, we have generated geochemical (planktic foraminiferal δ18O, δ13C and Mg /Ca), sedimentological (sand accumulation rates) and faunal (abundance data of thermocline-dwelling foraminifera) records for Ocean Drilling Program Site 849 located in the central region of the EEP. Our records span the interval from ∼ 2.75 to 2.4 Ma (MIS G7–95), which is critical for understanding thermocline dynamics during the final phase of the iNHG. Our new records document a thermocline shoaling from ∼ 2.64 to 2.55 Ma (MIS G2–101) and a relatively shallow thermocline from ∼ 2.55 Ma onwards (MIS 101–95). This indicates a state change in thermocline depth at Site 849 shortly before the final phase of the iNHG. Ultimately, our data support the hypothesis that (sub-)tropical thermocline shoaling may have contributed to the development of large Northern Hemisphere ice sheets.

at ~2. 5 Ma as documented by the first three large-amplitude (~1 ‰ in the benthic δ 18 O record) glacial-interglacial cycles ) that indicate substantial waxing and waning of ice sheets (Lisiecki and Raymo, 2005). At that time, ice rafting became widespread across the North Atlantic Ocean (Shackleton et al., 1984;Naafs et al., 2013). This so-called "intensification of Northern Hemisphere Glaciation" (iNHG) represents the most recent major tipping point in Earth's climate history. It strongly influenced oceanographic and climatic patterns worldwide, affecting, for example, the amount of 5 biological production in the Eastern Equatorial Pacific Ocean (EEP) (Etourneau et al., 2010;Jakob et al., 2016) -a region that exerts a strong influence on the Earth's climate system through its effects on the global carbon and nutrient cycles (e.g., Schlitzer, 2004;Takahashi et al., 2009).
Today, the shallow depth of the thermocline in the EEP plays a pivotal role in the evolution of short-term climate phenomena including the El Niño-Southern Oscillation (Fedorov et al., 2004;Ma et al., 2013). Its position in the water 10 column thus bears important consequences for the Earth's climate system. In general, proxy records and modeling results consistently document a long-term shoaling of the thermocline in the EEP and other (sub-)tropical upwelling regions throughout the Plio-Pleistocene (Wara et al., 2005;Fedorov et al., 2006;Dekens et al., 2007;Steph et al., 2010;Ford et al., 2012). However, the dynamics of the thermocline in the EEP and its potential links to the iNHG have yet remained enigmatic. Some studies have inferred that thermocline depth reached a critical threshold at ~3 Ma, which allowed trade 15 winds to deliver cool waters from below the thermocline to the surface in (sub-)tropical upwelling regions such as the EEP (Fedorov et al., 2006;Dekens et al., 2007). The timing led to the hypothesis that thermocline shoaling and the development of the "EEP cold tongue" (Wyrtki, 1981) was a necessary precondition for iNHG via a strengthening of the tropical Pacific Walker Circulation and, at the same time, a reduction of poleward heat transport (Cane and Molnar, 2001). Other studies, however, identified fundamental shifts in thermocline depth only prior to ~3. 5 Ma (Wara et al., 2005;Steph et al., 2010;20 Ford et al., 2012), which would imply that thermocline depth in the EEP did not play an important role in the development of large-scale glaciation in the Northern Hemisphere.
In light of these contradictory views, and to ultimately shed new light on potential links between low-latitude thermocline dynamics and high-latitude ice-sheet build-up, we have investigated thermocline state changes during the iNHG in the EEP by integrating new with previously published proxy records from Ocean Drilling Program (ODP) Site 849 (Fig. 25 1). We use planktic (both sea-surface-and thermocline-dwelling) foraminiferal geochemical (δ 18 O, δ 13 C and Mg/Ca) proxy records in combination with sedimentological (sand-accumulation rates) and faunal (abundance data of thermocline-dwelling foraminiferal species) information to reconstruct thermocline depth for the final phase of the late Pliocene/early Pleistocene iNHG from ~2.75 to 2.4 Ma (MIS G7-95). Clim. Past Discuss., https://doi.org/10.5194/cp-2017-151 Manuscript under review for journal Clim. Past Discussion started: 20 November 2017 c Author(s) 2017. CC BY 4.0 License.

The Eastern Equatorial Pacific
The EEP has considerable relevance for the Earth's atmospheric and marine carbon budget (Toggweiler and Sarmiento, 1985;Takahashi et al., 2009), and at the same time exerts strong control on oceanographic and climatic circulation patterns (Fedorov and Philander, 2000;Pennington et al., 2006). Today, as a part of the tropical Pacific Walker Circulation, 5 westward-blowing trade winds in the EEP induce a year-around upwelling of cold waters from below the thermocline to the surface. This results in a thin, nutrient-enriched and relatively cold (~23 °C [Locarnini et al., 2013]) mixed layer, the socalled "EEP cold tongue" (Wyrtki, 1981) (Fig. 1a). In today's oceans, the EEP upwelling system supports more than 10 % of the global biological production (Pennington et al., 2006). The relatively high primary productivity results in exceptionally high sedimentation rates that amount to up to 3.0 cm kyr -1 during the past 5 Myr (Mayer et al., 1992;Mix et al., 1995). 10 The EEP stands out as an ideal natural laboratory for studying the dynamics of the tropical thermocline: Owing to the shallow depth of the thermocline in the present-day EEP upwelling system (~50 m [Wang et al., 2000]) ( Fig. 1b), even small changes in its depth can affect surface-water properties such as temperature or nutrient content that can be ideally reconstructed from sediments underneath the "cold tongue". Thereby, the high sedimentation rates allow for the acquisition of proxy records at high temporal resolution compared to outside the EEP upwelling zone. 15

ODP Site 849
To reconstruct changes in thermocline depth in the EEP we focused on sediments from ODP Leg 138 Site 849 (coordinates: 0°11′N, 110°31′W; present-day water depth: 3851 m [Mayer et al., 1992]) (Fig. 1). This site has been selected because of (i) its position within the equatorial "cold tongue" west of the East Pacific Rise in the open ocean that makes it less prone to continental influence compared to sites east of the East Pacific Rise (Mix et al., 1995); (ii) the good preservation of 20 foraminifera (Jakob et al., 2016(Jakob et al., , 2017 despite the fact that the present-day water depth at Site 849 is close to the lysocline (Adelseck and Anderson, 1978;Berger et al., 1982); and (iii) high sedimentation rates (2.7 cm kyr -1 for our study interval [Jakob et al., 2017]) with continuous sedimentation (Mayer et al., 1992).
crassaformis and the surface-dwelling species G. ruber, as well as sand-accumulation rates were generated from the full sample set (temporal resolution: ~750 yr). From these measurements surface-to-thermocline δ 18 O, δ 13 C and Mg/Ca gradients 15 were calculated. Abundance counts of both deep-thermocline-(G. crassaformis) and intermediate-thermocline-dwelling species (G. menardii and G. tumida) were conducted every 20 cm (temporal resolution: ~7.5 kyr).
For G. crassaformis, we have generated new δ 18 O, δ 13 C and Mg/Ca records, except for the interval from ~2.65 to 2.4 Ma for which δ 13 C values have already been published by Jakob et al. (2016). For this purpose, an average of 15 individuals per sample was picked from the 315-400 µm fraction. Tests for δ 18 O and Mg/Ca analyses were cracked, 20 homogenized and split into two subsamples. For G. crassaformis, both sinistral-and dextral-coiling specimens occur at Site 849; geochemical data were preferentially measured on tests with the most abundant coiling direction (sinistral). Sinistralcoiling specimens, however, do not occur continuously across our study interval. Thus, dextral-coiling tests were used for some intervals (in particular from ~2.46 to 2.43 04 mcd] and from ~2.51 to 2.50 ).
Measurements of geochemical parameters (δ 13 C, δ 18 O, Mg/Ca) separately on sinistral-and dextral-coiling specimens from 25 the same samples have demonstrated that the reconstructed values are independent of the coiling directions (see also Jakob et al. [2016]).
The number of G. crassaformis individuals in the investigated size fraction (315-400 µm), was, however, too low to allow for geochemical analyses in specific intervals. These intervals are from 2.68 to 2.65 .65 mcd corresponding to MIS G3) and from 2.73 to 2.69 Ma (77.02-75.79 mcd corresponding to MIS G7-G5). We decided not to 30 fill these gaps by using another size fraction because (i) this might have biased our Mg/Ca and stable-isotope records due to ontogenetic effects (Elderfield et al., 2002;Friedrich et al., 2012), and (ii) the abundance of G. crassaformis specimens <315 For G. ruber, the previously published Mg/Ca-based SST and δ 18 O datasets of Jakob et al. (2017) were augmented by δ 13 C data, except for the interval from ~2.65 to 2.4 Ma for which δ 13 C values have already been published by Jakob et al. (2016). For this purpose, an average of twelve specimens was picked from the 250-315 µm size fraction. An overview of all 5 geochemical datasets evaluated in this study is presented in Table 1.

Foraminiferal preservation
The preservation of planktic foraminiferal (G. crassaformis and G. ruber) tests used for geochemical analyses was examined by Scanning Electron Microscope (SEM) images of selected specimens from both glacial and interglacial intervals. Close-up 10 views were taken using a LEO 440 SEM at the Institute of Earth Sciences, Heidelberg University.

Stable-isotope analyses
Oxygen and carbon isotopes of G. ruber and G. crassaformis were analyzed using a ThermoFinnigan MAT253 gas-source mass spectrometer equipped with a Gas Bench II at the Institute of Geosciences, Goethe-University Frankfurt. Values are reported relative to the Vienna Pee Dee Belemnite (VPDB) standard through the analysis of an in-house standard calibrated 15 to NBS-19. The precision of the δ 18 O and δ 13 C analyses is better than 0.08 ‰ and 0.06 ‰ (at 1σ level), respectively. δ 13 C values of G. ruber reported herein have been adjusted for species-specific offset from equilibrium precipitation by the addition of + 0.94 ‰ (Spero et al., 2003), while measured δ 18 O values of G. ruber have been considered to approximate equilibrium precipitation (Koutavas and Lynch-Stieglitz, 2003) and therefore were not adjusted. For G. crassaformis we report the values measured on this species because laboratory investigations on δ 13 C and δ 18 O fractionation for G. 20 crassaformis are still lacking.

Mg/Ca analyses
Samples for Mg/Ca analyses of G. crassaformis were carefully cleaned to remove clay minerals, organic material and readsorbed contaminants following the protocol of Barker et al. (2003). Analyses were carried out with an Agilent Inductively Coupled Plasma-Optical Emission Spectrometer 720 at the Institute of Earth Sciences, Heidelberg University. Reported 25 Mg/Ca values were normalized relative to the ECRM 752-1 standard reference value of 3.762 mmol/mol (Greaves et al., 2008). To ensure instrumental precision, an internal consistency standard was monitored at least every 20 samples. Based on replicate measurements, a standard deviation for Mg/Ca of ± 0.02 mmol/mol (corresponding to ± 0.12 °C) is obtained. To identify possible contamination by clay particles or diagenetic coatings that might affect foraminiferal Mg/Ca ratios (Barker et al., 2003), elemental ratios of Al/Ca, Fe/Ca and Mn/Ca were screened (see Sect. 5.2). 30 Clim. Past Discuss., https://doi.org /10.5194/cp-2017-151 Manuscript under review for journal Clim. Past Discussion started: 20 November 2017 c Author(s) 2017. CC BY 4.0 License.

Paleotemperature reconstruction
Species-specific conversions of Mg/Ca to temperature for G. crassaformis have only been calibrated based on samples from the Atlantic Ocean (Anand et al., 2003;Regenberg et al., 2009;Cléroux et al., 2013). These equations yield the same trend and amplitudes when applied to G. crassaformis Mg/Ca values from Site 849, but differ with regard to absolute values (Fig.   2). We converted Mg/Ca ratios of G. crassaformis into temperature following the species-specific equation of Cléroux et al. 5 (2013) for two reasons: (i) The selected calibration is based on specimens with a grain size of 355-425 µm, which matches the size fraction used in our study (315-400 µm) better than the other equations that are available; and (ii) the Mg/Ca range of G. crassaformis from Site 849 (~1-3 mmol/mol) fits best to the calibration range (~1-2 mmol/mol) of the equation of Cléroux et al. (2013) compared to the other equations.
Because the equation of Cléroux et al. (2013) is based on oxidative and reductive cleaning of foraminiferal tests, 10 while only oxidative cleaning has been applied to Site 849 samples, measured Mg/Ca values had to be reduced by 10 % (Barker et al., 2003). The pooled uncertainty in the temperature record (analytical error of ± 0.02 mmol/mol [corresponding to ± 0.12 °C] and calibrational error of ± 0.82 °C [Cléroux et al., 2013]) is ± 0.94 °C.

Abundance counts
Abundance counts of deep-thermocline-dwelling (G. crassaformis, both sinistral-and dextral-coiling specimens) and 15 intermediate-thermocline-dwelling (G. tumida and G. menardii) species (Fig. 1b) were generated on the >250 µm size fraction of Site 849 samples. The >250 µm size fraction was split down to at least 400 individual planktic foraminifera using a microsplitter. Abundance counts were converted to mass-accumulation rates using linear sedimentation rates and dry bulk density data (calculated from high-resolution GRAPE density shipboard measurements [IODP JANUS database; Mayer et al., 1992]) following the methodology described in Jakob et al. (2016). 20

Sand-accumulation rates
The sand-accumulation rates (SAR) available for the MIS G1-95 (~2.65-2.4 Ma) interval from Site 849 as presented in Jakob et al. (2016) were extended back to MIS G7 (~2.75 Ma) in this study. They were calculated using linear sedimentation rates, dry bulk density data (calculated from high-resolution GRAPE density shipboard measurements [IODP JANUS database; Mayer et al., 1992]), and the portion of the >63 µm sand fraction following the approach described in Jakob et al. 25 (2016).

Foraminiferal test preservation at Site 849
Shipboard investigations on core catchers had originally reported a poor preservation of planktic foraminiferal tests at Site 849 (Mayer et al., 1992). However, these observations do not apply to the samples from our study interval. Instead, SEM images of both glacial and interglacial G. crassaformis and G. ruber specimens show well-preserved fine features such as 5 delicate spines, pore channels and a layered wall structure, and a lack of secondary calcite on test surfaces (Fig. 3). This indicates that test preservation is consistently sufficient for the acquisition of high-quality geochemical data for both species throughout the study interval. A low planktic foraminiferal fragmentation index at Site 849 further confirms this interpretation (Jakob et al., 2017), indicating that dissolution has not considerably affected foraminiferal tests.

Assessment of contamination and diagenetic effects on Mg/Ca ratios of G. crassaformis 10
Al/Ca, Fe/Ca and Mn/Ca ratios >0.1 mmol/mol are typically considered to indicate the presence of detrital clay, Fe-rich coatings and Mn-rich overgrowth, respectively, and therefore hint at foraminiferal test contamination that might have biased measured Mg/Ca ratios (Barker et al., 2003). In most of our samples the content of Al was below the detection limit, arguing against the presence of detrital clay. With the exception of few samples, Fe/Ca values also commonly do not exceed the critical value of 0.1 mmol/mol and therefore indicate no contamination by Fe-rich overgrowth (Fig. 4a). The lack of a 15 statistically significant correlation between Mg/Ca and Fe/Ca ratios (r 2 = 0.17, p <0.01) further supports this finding.
Measured Mn/Ca ratios, however, were above the 0.1 mmol/mol threshold and therefore might indicate Mn-rich overgrowth on the analyzed tests (Fig. 4b). Our data show a weak correlation between Mg/Ca and Mn/Ca ratios for G. crassaformis samples (r 2 = 0.44, p <0.01) that indeed might indicate a bias of measured Mg/Ca ratios due to a Mn-rich overgrowth. We note, however, that SEM images argue against the existence of any kind of overgrowth (Fig. 3). 20 Diagenetic overgrowth usually has a Mg/Mn ratio of ~0.1 mol/mol (Barker et al., 2003, and references therein). If an unrealistically high Mg/Mn ratio of 1 in the diagenetic overgrowth is assumed, this might change temperature estimates by on average ~0.5 °C; however, we note that this would not affect the overall shape of the Mg/Ca-based temperature record ( Fig. 4c, d). Although we cannot completely rule out the possibility of diagenetic changes in the G. crassaformis tests because of significantly enriched Mn/Ca ratios, we conclude with reasonable certainty that early diagenetic overprinting (if 25 existing) has no significant impact on the interpretation of our data.

Stable-isotope and Mg/Ca records of G. crassaformis and G. ruber at Site 849
In the tropics, where seasonal climate variability is low, geochemical signatures of the investigated foraminiferal species are typically considered not to be seasonally biased (Lin et al., 1997;Tedesco et al., 2007). Therefore, all data presented in this study are inferred to reflect mean annual conditions. 30 Ma) (Fig. 5b). Lowest values correspond to interglacials and highest values are associated with glacials. With a mean value of 2.01 ‰, δ 18 O values of G. crassaformis (this study) are on average 2.78 ‰ higher than those of G. ruber (Jakob et al., 2017), indicating cooler and/or more saline waters in the depth habitat of G. crassaformis (bottom of the thermocline) compared to G. ruber (surface waters) ( Fig. 1b; see Sect. 3 for details). In theory, a deeper depth habitat of G. crassaformis 5 than of G. ruber should result in a lower δ 13 C signature in G. crassaformis tests than in those from G. ruber since the amount of organic matter remineralization and the associated release of light 12 C into the surrounding water typically increases with water depth (Deuser and Hunt, 1969;Kroopnick, 1985). Indeed, by ~1.86 ‰ lower δ 13 C values are recorded for G. crassaformis (mean: 0.46 ‰) than for G. ruber (mean: 2.32 ‰) (Jakob et al., 2016; this study) (Fig. 5c). In general, δ 13 C values of G. ruber and G. crassaformis fluctuate between minima of 1.36 ‰ and 0 ‰ during glacials and maxima of 2.97 ‰ 10 and 0.97 ‰ during interglacials, respectively (Jakob et al., 2016; this study) (Fig. 5c). It has been shown in a former study that the glacial-interglacial foraminiferal δ 13 C pattern at Site 849 during iNHG is strongly controlled by the amount of primary productivity in the Southern Ocean, which is highest during interglacials (high δ 13 C) and lowest during glacials (low δ 13 C) (Jakob et al., 2016).
Mg/Ca values of G. crassaformis vary between 0.82 mmol/mol and 3.02 mmol/mol (this study), being on average 15 1.43 mmol/mol lower than Mg/Ca ratios of G. ruber from the same site and time interval (Jakob et al., 2017) (Fig. 5d).

Geochemical evidence
Variations in the vertical temperature gradient within the upper water column allow to effectively monitor shifts in thermocline depth with a small temperature difference between surface and thermocline waters indicating a deep thermocline and vice versa (e.g., Steph et al., 2009;Nürnberg et al., 2015). Absolute values of surface-to-thermocline temperature gradients can be reconstructed from the Mg/Ca-based temperature gradient derived from surface-dwelling (e.g., G. ruber) 25 and thermocline-dwelling (e.g., G. crassaformis) species. In addition, the δ 18 O gradient of the same species provides information on relative changes in the surface-to-thermocline temperature gradient since the effect of global ice volume and salinity (i.e., the two other main factors beside temperature that affect foraminiferal δ 18 O [Ravelo and Hillaire-Marcel, 2007]) should be identical for planktic species from the same locality.
On longer time scales, the δ 18 O gradient decreased by ~1 ‰ from ~2.64 Ma to 2.55 Ma (MIS G2-101), and remained relatively constant throughout the remainder of our study interval (~2.55-2.38 Ma; MIS 101-95) (Fig. 6b). The Mg/Ca-based temperature gradient shows the same overall pattern as the δ 18 O gradient (i.e., a ~5 °C increase from ~2.64 Ma 5 to 2.55 Ma, and a constant value of ~20 °C from ~2.55 to 2.38 Ma) (Fig. 6b). We interpret the decrease in the δ 18 O gradient and the increase in the Mg/Ca-based temperature gradient from ~2.64 Ma to 2.55 Ma to represent a shoaling of the thermocline. From that time onwards, constant δ 18 O and Mg/Ca-based temperature gradients suggest that the thermocline remained relatively shallow throughout the final phase of the late Pliocene/early Pleistocene iNHG until ~2.38 Ma.

Faunal and sedimentological evidence 10
Based on our data, the overall abundances of the deep-thermocline-dwelling species G. crassaformis at Site 849 were relatively low when the thermocline in the EEP was relatively deep (i.e., prior to ~2.55 Ma as indicated by the δ 18 O and Mg/Ca-based temperature gradients) (Fig. 6c). More specifically, G. crassaformis was present in substantially reduced numbers only or even completely absent between ~2.73 and 2.64 Ma (MIS G7-G2) (with the exception of MIS G4 [~2.68-2. 69 Ma]). Since ~2.64 Ma, i.e., when the thermocline started to shoal, relative G. crassaformis abundances increased 15 markedly from typically <5 % to ~10-35 % such that representatives of this taxon are present continuously throughout the remainder of the study interval. The increase of G. crassaformis abundances occurs at the expense of the intermediatethermocline-dwelling species G. menardii and G. tumida, which decline from ~30-75 % to only ~5-65 % (Fig. 6c). This observation suggests that prior to the final phase of the late Pliocene/early Pleistocene iNHG, when the thermocline was relatively deep, the living conditions for the deep-thermocline-dwelling species G. crassaformis were unfavorable. At the 20 same time, low G. crassaformis abundances allowed intermediate-thermocline-dwelling species (such as G. menardii and G. tumida) to dominate the planktic foraminiferal assemblages.
In general, G. crassaformis reaches highest abundances in oxygen-depleted waters (Jones, 1967;Kemle von Mücke and Hemleben, 1999), while other environmental factors including temperature, salinity or nutrient availability are markedly less important (Cléroux et al., 2013). The oxygen content of deep waters (and accordingly the abundance of G. crassaformis) 25 is typically closely coupled to surface-water productivity (with a lower oxygen level and higher G. crassaformis abundances in deeper waters when surface-water productivity is high) . We therefore hypothesize that primary productivity at Site 849 prior to ~2.64 Ma (MIS G2) was relatively low (with the exception of MIS G4). Low primary productivity led to a reduction of organic matter remineralization and therefore oxygen consumption in deeper waters. An elevated oxygen content in deeper waters implies, in turn, low G. crassaformis abundances. Such a scenario is supported by 30 the SAR-based primary productivity record from Site 849 (Jakob et al., 2016; this study), which clearly indicates low productivity rates prior to ~2.64 Ma when the thermocline was relatively deep compared to after ~2.64 Ma when the thermocline became shallow (Fig. 6b, d).
As outlined above, on longer timescales primary productivity at Site 849 appears to be coupled to variations in thermocline depth. The observed long-term trend in primary productivity is overprinted by a clear glacial-interglacial cyclicity particularly during the final phase of iNHG (MIS 100-96); the position of the thermocline, however, remained constant along the obliquity (i.e., 41-kyr) band (Fig. 6b, d). Notably, the accumulation rates of G. crassaformis, G. menardii and G. tumida perfectly capture the pattern derived by the SAR-based productivity record (Fig. 6c, d). This suggests that the 5 abundances of deep-and intermediate-thermocline-dwelling foraminiferal species are coupled to the strength of the biological pump and can be used as a tracer for primary productivity at Site 849.
The relatively high δ 13 C gradient for MIS G4-G2 as it emerges from our data (Fig. 6d) confirms the interpretation of Jakob et al. (2016) for MIS G1-95 that low-amplitude productivity changes prior to MIS 100 were mainly driven by the nutrient content within the upwelled water mass as long as the thermocline was relatively deep (as reflected by a low 10 thermocline-to-surface temperature gradient prior to ~2.55 Ma [MIS 101];Fig. 6b,d). High-amplitude productivity changes resulting from upwelling intensification played an important role from MIS 100 onward (Jakob et al., 2016). At the same time, the thermocline became relatively shallow (as reflected by a large thermocline-to-surface temperature gradient since ~2.55 Ma [MIS 101] at Site 849; Fig. 6b, d). This is unlikely a coincidence, but rather suggests that the thermocline depth reached a critical threshold at that time. Prior to this, relatively warm and nutrient-poor waters from above the thermocline 15 have upwelled in the EEP. Consequently, primary productivity rates were low. When the thermocline became shallow enough, however, trade winds could deliver cooler, nutrient-enriched waters from below the thermocline to the surface, which allowed primary productivity to increase. Such a scenario is supported by the observation that lower SSTs prevailed at Site 849 during glacials since MIS 100 (Jakob et al., 2017) (Fig. 5d).
Our findings imply a marked effect of long-term thermocline state change on primary productivity in the EEP. In 20 particular, when thermocline shoaling reached a critical threshold, primary productivity increased, thereby removing CO 2 from the surface ocean. Considering that the EEP contributes substantially to global biological production in the present-day oceans, the observed coupling between thermocline state change and primary productivity is of major importance for the Earth's climate; thus, it may have favored the early development of large ice sheets in the Northern Hemisphere.

Conclusions 25
We integrate new with previously published foraminiferal-based geochemical and faunal records and with sedimentological data for ODP Site 849 (~2.  to reconstruct changes in thermocline depth in the EEP during the late Pliocene/early Pleistocene iNHG. Our data document a shoaling of the thermocline at Site 849 from ~2.64 to 2.55 Ma, while it remained relatively shallow until ~2.38 Ma, implying that major changes in thermocline depth in the EEP occurred shortly before the final phase of the late Pliocene/early Pleistocene iNHG (i.e., prior to MIS 100-96). This finding, which is in line 30 with former studies (Fedorov et al., 2006;Dekens et al., 2007), supports the hypothesis that (sub-)tropical thermocline shoaling was a precondition to allow the development of large ice sheets in the Northern Hemisphere (Cane and Molnar, Clim. Past Discuss., https://doi.org/10.5194/cp-2017-151 Manuscript under review for journal Clim. Past Discussion started: 20 November 2017 c Author(s) 2017. CC BY 4.0 License. 2001). At the same time, our new data contradict studies that have documented substantial shifts in thermocline depth in the EEP only prior to ~3. 5 Ma (Wara et al., 2005;Steph et al., 2010;Ford et al., 2012). Our new records also suggest a low primary productivity during times of a relatively deep thermocline prior to ~2.64 Ma (MIS G2). In turn, the relatively shallow thermocline associated with low SSTs after ~2.55 Ma allowed primary productivity to increase during prominent iNHG glacials (MIS 100-96). 5

Data availability
All data reported will be made available upon publication of this paper via the open access PANGAEA database (www.pangaea.de).

Author contributions
The project was designed by OF, KAJ, and JP. KAJ and CS carried out Mg/Ca analyses; JF and KAJ performed stable-10 isotope analyses. KAJ wrote the manuscript with input from all co-authors.   this study) and temperature (red; this study) gradients between the surface-dwelling species G. ruber and the thermocline-dwelling species G. crassaformis as a proxy for relative thermocline depth. (c) Relative abundances of G. crassaformis (light 5 blue; this study) and G. menardii and G. tumida (dark blue; this study) together with their mass-accumulation rates (green; this study). (d) δ 13 C gradient between G. ruber and G. crassaformis as an indicator for upwelling strength (purple; Jakob et al. [2016], this study) together with sand-accumulation rates as an indicator for primary productivity (Jakob et al. [2016], this study). Gaps in the δ 18 O, δ 13 C and Mg/Ca gradients for 2.68-2.65 Ma (MIS G3) and 2.73-2.69 Ma (MIS G7-G5) are due to a lack of G. crassaformis specimens in the investigated (315-400 µm) size fraction. 10 Clim. Past Discuss.