Neogene ice volume and ocean temperatures: Insights from infaunal foraminiferal Mg/Ca paleothermometry

,


Introduction
The cryosphere affects climate through changing ocean/atmosphere heat transport and biogeochemical carbon cycling. Accurate records of the cryosphere and corresponding paleoceanographic conditions are required to understand how these processes contributed to the evolution of Earth's climate through the Cenozoic. Records of ice-rafted debris and continental weathering suggest that semipermanent Antarctic continental-scale glaciation initiated at the Eocene-Oligocene climate transition (EOT),~34 Ma [Zachos et al., 1992;Scher et al., 2011]. The history of the development of continental-scale ice sheets in the Northern Hemisphere is more controversial, but Northern Hemisphere glaciation probably occurred later [Edgar et al., 2007;DeConto et al., 2008]. Iceberg-derived dropstones have been documented in the Norwegian Greenland Sea from EOT time but are likely sourced from isolated coastal mountain outlet glaciers on Greenland [Eldrett et al., 2007], whereas major Pleistocene Northern Hemisphere glaciation apparently began around 2.6 Ma [Raymo et al., 1989;Bailey et al., 2013]. The Cenozoic benthic foraminiferal oxygen isotope compilation delineates the nonlinear transition from the greenhouse world of the early Cenozoic to today's icehouse world [Shackleton and Kennett, 1975;Miller et al., 1987;Zachos et al., 2001Zachos et al., , 2008Cramer et al., 2009;Mudelsee et al., 2014]. This record reflects deep-sea temperatures in addition to global ice volume. Three δ 18 O "steps" have been interpreted as significant ice sheet growth events, as the climate system crossed key thresholds [Zachos et al., 1996[Zachos et al., , 2001[Zachos et al., , 2008DeConto et al., 2008]. The earliest "step" at the EOT is recognized as reflecting the continent-wide glaciation on Antarctica [Zachos et al., 1992 (MMCT), around 14 Ma [Holbourn et al., 2005]. Backstripping estimates suggest that sea level fell between 53 and 69 m between 16.5 and 13.9 Ma [John et al., 2011]. Glacial landforms provide evidence of larger-thanmodern ice sheets overriding the Transantarctic Mountains during the middle Miocene [Denton and Sugden, 2005]. The MMCT has been linked to ice sheet advance and a transition from a polythermal to a dry-based Antarctic ice sheet [Lewis et al., 2007], as levels of CO 2 decreased and orbital configurations favored low seasonality at high latitudes [Holbourn et al., 2005[Holbourn et al., , 2007Foster et al., 2012]. Following the MMCT, the Antarctic ice sheet may have been less sensitive to changes in high-latitude radiative forcing, with ice volume controlled more by variations in moisture transport [Holbourn et al., 2013a]. The most recent step in the δ 18 O record largely reflects the Plio-Pleistocene intensification of glaciation in the Northern Hemisphere. However, the dual control of temperature and ice volume on foraminiferal δ 18 O means that significant questions regarding the extent of continental glaciation and the relationship between the crysophere and ocean overturning circulation through the Neogene remain.
Benthic foraminiferal Mg/Ca paleothermometry has shown great promise in deconvolving the temperature and δ 18 O sw signal from foraminiferal δ 18 O records, although it is not without its complications [Lear et al., 2000[Lear et al., , 2010Schrag, 2002, 2003;Martin et al., 2002;Shevenell et al., 2008;Cramer et al., 2011;Bohaty et al., 2012]. Chiefly, these are as follows: (1) variations in Mg/Ca temperature sensitivity between foraminiferal genera necessitating genus-or species-specific temperature calibrations [Lear et al., 2002], (2) the influence of carbonate saturation state at low saturation [Elderfield et al., 2006], (3) uncertainties in seawater Mg/Ca reconstructions [Coggon et al., 2010], and (4) the relationship between seawater Mg/Ca and foraminiferal calcite Mg/Ca [Hasiuk and Lohmann, 2010;Ries, 2004;Cramer et al., 2011;Evans and Müller, 2012]. The last two complications do not compromise the use of Mg/Ca paleothermometry on short (<1 Myr) timescales, making the proxy ideally suited to examining relative temperature changes over geologically short time intervals Elderfield et al., 2012;Mawbey and Lear, 2013]. However, over longer timescales they limit the reconstruction of absolute temperatures, which are required to estimate absolute ice volume from δ 18 O. Previous efforts to address these issues have been informative, but progress has been limited by a paucity of available Mg/Ca records [Cramer et al., 2011;Evans and Müller, 2012]. The saturation state influence is of particular importance for epifaunal foraminifera, which more directly experience changing bottom water mass chemistry.
Here we present a record of benthic foraminiferal Mg/Ca for the past 17 Ma and attempt to minimize or address these factors in order to calculate absolute bottom water temperature (BWT) and δ 18 O sw for the late Neogene. All records are from one species, the shallow infaunal dwelling Oridorsalis umbonatus, for which core top calibrations exist [Lear et al., 2002;Rathmann et al., 2004;Healey et al., 2008;Brown et al., 2011;Tisserand et al., 2013]. We minimize the influence of the bottom water carbonate saturation state [Elderfield et al., 2010] by using an infaunal species living within the upper few centimeters of the seafloor sediment, and by careful site selection, avoiding sites below the lysocline where foraminifera may experience postmortem dissolution. We also present a new Mg/Ca record from the ice-free early Paleogene, for which bottom water temperatures can be independently calculated from δ 18 O, representing a time of far lower seawater Mg/Ca than today. We combine this record with published core top Mg/Ca data to better constrain the relationship between seawater Mg/Ca and foraminiferal Mg/Ca for this species. We find that Oridorsalis umbonatus has a low sensitivity to changes in seawater Mg/Ca, implying a strong biological control on its calcification process. We then interpret our Neogene Mg/Ca and δ 18 O records in terms of absolute BWT and δ 18 O sw and find strong support for greater than modern ice volume immediately following the Middle Miocene Climate Transition and a change in ocean overturning circulation associated with late Neogene global cooling.

Advantages of Infaunal Foraminifera in Mg/Ca Paleothermometry
Benthic foraminiferal Mg/Ca is primarily controlled by changes in bottom water temperature [Rosenthal et al., 1997;Lear et al., 2002;Marchitto et al., 2007]. However, the degree of carbonate saturation (ΔCO 3 2À ) impacts benthic foraminiferal Mg/Ca at low saturation [Elderfield et al., 2006;Rosenthal et al., 2006]. Initial work focused on the epifaunal taxon Cibicidoides, which lives above, at, or very close to the seafloor in direct contact with bottom water [Jorissen et al., 2007]. Mg/Ca in infaunal benthic foraminifera might be a more reliable paleotemperature proxy, because pore waters are buffered to some extent against changes in carbonate saturation [Elderfield et al., 2010]. Oxidation of sedimentary organic matter tends to decrease the pore water carbonate saturation state. Once pore waters become undersaturated, dissolution of carbonate grains buffers against further decreases in saturation state [Zeebe, 2007].
To address the saturation state problem in the Pleistocene, a Mg/Ca temperature record based on the infaunal genus Uvigerina from Ocean Drilling Program (ODP) Site 1123 (3.3 km water depth on Chatham Rise, east of New Zealand) has been used to calculate Pleistocene changes in δ 18 O sw . Remarkable agreement with independent sea level proxies provides strong support for the validity of this approach [Elderfield et al., 2012;Rohling et al., 2014]. We take the same approach using Oridorsalis umbonatus, an extant shallow infaunal species, found globally through the Late Cretaceous-Cenozoic, for which core top calibrations exist [Lear et al., 2002;Rathmann et al., 2004;Healey et al., 2008;Brown et al., 2011;Tisserand et al., 2013]. Core top and down core records suggest that this species, like Uvigerina species, is less sensitive to changes in bottom water saturation state than epifaunal species [Brown et al., 2011;Mawbey and Lear, 2013;Foster et al., 2013]. However, the trace metal composition of O. umbonatus appears sensitive to changes in carbonate saturation state in some instances, so we cannot assume that O. umbonatus is consistently and fully buffered from changes in bottom water saturation state in its microhabitat. For example, Mg/Ca in O. umbonatus increases by 0.28 mmol/mol across the EOT at ODP Site 1218 [Lear et al., 2004], which is surprising given the widespread cooling associated with the major Antarctic glaciation [Liu et al., 2009]. However, at ODP Site 1218 the EOT is also associated with extensive (~1.2 km) deepening of the calcite compensation depth, a lithologic change from radiolarite to nannofossil oozes and chalks, and thus an increase in pore water buffering capacity of this deep ocean site (paleowater depth~3800 m) [Coxall et al., 2005;Lyle et al., 2002]. Possibly, the very low carbonate saturation state at this site influenced the benthic foraminiferal Mg/Ca through a combination of effects during test precipitation and postmortem dissolution (many tests, especially from samples taken from the upper Eocene radiolarites contain dissolution pits).

Site Selection
We cannot assume that all infaunal foraminifera, living at various depths within the sediment from close to the surface to >10 cm deep [Jorissen et al., 2007] inhabit pore waters that are fully buffered with respect to carbonate saturation state, so sites for infaunal Mg/Ca paleothermometry should be selected carefully. The presence of abundant, well-preserved planktonic foraminifera is one indication that sediment can buffer pore waters against significant decreases in carbonate saturation state. Sediment pore water chemistry under regions of vigorous upwelling may be complicated by temporal variations in organic matter degradation, and pore waters thus may not be fully buffered [Rathmann and Kuhnert, 2008]. In addition, hiatuses in sedimentation or transient dissolution events may disrupt the buffering capacity of the pore waters [Mawbey and Lear, 2013]. Therefore, we selected carbonate-rich samples with abundant planktonic foraminifera. In order to maximize the potential dissolution capacity of the sediments, we have selected samples from paleowater depths of less than 3 km.
Our new benthic foraminiferal Mg/Ca and δ 18 O data between~17 and 12 Ma span the MMCT at Ocean Drilling Program (ODP) Site 806, presently at 2521 m water depth on the Ontong Java Plateau in the Pacific Ocean (0°19.1′N 159°21.7′E) (Figure 1). Carbonate contents for these sediments are typically between 90 and 95% [Kroenke et al., 1991]. Detailed stable isotope records from~13.3 to 14.1 Ma on an orbitally tuned age model are presented in Holbourn et al. [2013b]. We combine our Site 806 record with previously published O. umbonatus data from upper Miocene to Pleistocene strata from the same site [Lear et al., 2003]. We also present new benthic foraminiferal Mg/Ca data from the upper Paleocene to lower Eocene from ODP Site 690, present water depth 2914 m, on Maud Rise in the Southern Ocean (65°9.629′S, 1°12.30′E) ( Figure 1). The sediments typically contain between 75 and 95% carbonate and have abundant planktonic foraminifera [Barker et al., 1988]. These strata show sedimentation rates of~1-2 cm/kyr and are estimated to have been deposited at a paleowater depth of 1900 m [Barker et al., 1988;Thomas and Shackleton, 1996]. Therefore, we assume that the pore waters were buffered with respect to carbonate saturation state. Our age model for Site 806 uses nannofossil biostratigraphic events [Nathan and Leckie, 2009] and planktonic foraminiferal events [Kroenke et al., 1991] assigned to ages provided in Wade et al. [2011] ( Figure S1 in the supporting information  Table S1 in the supporting information.

Trace Element Analysis
The Site 806 MMCT samples were cleaned using the same protocol as that used for the upper Miocene to Pleistocene record of Lear et al. [2003]. Between three and ten individuals of the benthic foraminifer O. umbonatus were picked from the 250-355 μm size fraction and crushed between glass plates to open all chambers. Test fragments were cleaned using a protocol to remove clays, metal oxides, and organic matter [Boyle and Keigwin, 1985]. Between the clay removal and reductive steps the samples were examined under a binocular microscope, and noncarbonate particles were removed using a fine paintbrush. Samples were dissolved in trace metal pure 0.065 M HNO 3 and diluted with trace metal pure 0.5 M HNO 3 to a final volume of 350 μL. Samples were analyzed at Cardiff University on a Thermo Element XR ICP-MS against standards with matched calcium concentration to reduce matrix effects [Lear et al., 2002]. All data for a sample were rejected if Al/Ca exceeded 50 μmol/mol, Mn/Ca exceeded 100 μmol/mol or Fe/Ca exceeded 200 μmol/mol. B/Ca data were rejected if the sample boron intensity signal was less than 10 times that of the blank. Long-term precision as determined by analyzing an independent consistency standard during each run is~1% and 5% (relative standard deviation) for Mg/Ca and B/Ca respectively. The ODP Site 690 Paleocene-Eocene samples were cleaned without the reductive step and analyzed using a Varian Vista ICP-OES at Cambridge University, with a precision of <1% for Mg/Ca. To achieve consistency between data sets, 9.1% was subtracted from the Mg/Ca data from ODP Site 690 . Trace metal data are available in Tables S1 and S2 of the supporting information.

ODP Site 806 Middle Miocene Record
Our oxygen isotope stratigraphy displays the global~1‰ shift to heavier values across the Middle Miocene Climate Transition and is consistent with the shorter duration, high-resolution record of Holbourn et al.

ODP Site 690 Paleocene-Eocene Record
Our O. umbonatus Mg/Ca record from ODP Site 690 increases as benthic foraminiferal δ 18 O decreases ( Figure 4a). Maximum Mg/Ca values occurred around 52.8 Ma, close to Eocene Thermal Maximum 2, since no data were included from the Paleocene-Eocene Thermal Maximum (Figure 4b) [Zachos et al., 2008]. We compile the Cibicidoides and Nuttallides δ 18 O data from Kennett and Stott [1990] and Thomas and Shackleton [1996], using the species offsets in Katz et al. [2003], correlated to the record in Littler et al.  (Figure 4c), within the error of the 0.12 ± 0.01 mmol/mol/°C (2 s.e., r 2 = 0.9) sensitivity of a previous calibration based on a core top O. umbonatus and C. pachyderma data set [Marchitto et al., 2007;Lear et al., 2010]. The value for the oldest sample at 57.1 Ma is perhaps slightly higher than expected from temperature alone ( Figure 4b). This is likely caused, at least in part, by natural variability in the Mg/Ca record, and the interspecies interpolation process (compare Figures 4a and 4b), although we cannot rule out the possibility of higher seawater Mg/Ca in this part of the record. However, the slope in Figure 4c is unchanged if this sample is omitted.

Constraining the Sensitivity of Mg/Ca in O. umbonatus to Changes in Seawater Mg/Ca
Seawater Mg/Ca (Mg/Ca sw ) values have increased from around 1.5 mol/mol in the early Cenozoic to today′s value of 5.2 mol/mol [Coggon et al., 2010;Dickson, 2002;Horita et al., 2002]. The residence times of Ca and Mg in seawater (~1 Myr and~10 Myr, respectively) [Broecker and Peng, 1982] means that potential changes in Mg/Ca sw need to be taken into account when calculating absolute BWT through the Cenozoic. Earlier studies assumed that benthic foraminiferal Mg/Ca was linearly related to seawater Mg/Ca [Lear et al., 2000;Billups and Schrag, 2003], but more recent culture studies on a range of calcifying organisms suggest that the relationship is better described by a power law dependency (equation (1)) [Ries, 2004], where Mg/Ca cc istheMg/Caofthecalcite,Mg/Ca sw is the Mg/Ca of seawater, and F is a factor that could include a linear or exponential temperature dependency. A culture study of the planktonic foraminifer Globigerinoides sacculifer did not produce a clear relationship between Mg/Ca cc and Mg/Ca sw [Delaney et al., 1985]. Culture studies of the larger, high-Mg calcite, symbiont-bearing benthic foraminiferal species Amphistegina lobifera, Amphistegina lessonii,andHeterostegina depressa found values for H of 0.8, 0.7-0.8, and 0.44, respectively [Segev and Erez, 2006;Raitzsch et al., 2010;Mewes et al., 2014]. The shallow water, low-Mg calcite foraminifer Ammonia aomoriensis has an H value of 0.7 [Mewes et al., 2014]. While it is clear that there are species differences, it is not known whether these values for H would be similar for other benthic foraminiferal species such as O. umbonatus, which calcifies in deep ocean waters at lower saturation states.  [Lear et al., 2003], closed symbols are from this study.

10.1002/2015PA002833
An alternative approach to constraining the relationship between foraminiferal Mg/Ca and seawater Mg/Ca is to use fossil foraminifera that precipitated their testsinseawaterwithdifferentMg/Cavaluesthanmodern. This approach assumes that the Mg/Ca of the fossil foraminifera has not been diagenetically altered. Benthic foraminifera are less porous and thus less susceptible to postdepositional alteration than planktonic foraminifera [Edgar et al., 2013]. The species Oridorsalis umbonatus has minute pores (e.g., much smaller than Nuttallides truempyi) and thus is among the least susceptible benthic species , but we acknowledge this potential uncertainty. This approach also requires knowledge of the ancient seawater Mg/Ca, BWT, and the species-specific Mg/Ca temperature sensitivity. BWT can be estimated using   [ Dickson, 2002]), this provides an ideal interval to assess the relationship between foraminiferal Mg/Ca and seawater chemistry [Lear et al., 2002;Billups and Schrag, 2003;Cramer et al., 2011;Evans and Müller, 2012]. Cramer et al. [2011] normalized published multispecies benthic foraminiferal Mg/Ca records to a single species, then applied both a "water depth correction" and a "calcite compensation depth" correction for the potential influence of the carbonate saturation state. This approach assumes a linear relationship between benthic foraminiferal Mg/Ca and carbonate saturation state without a threshold, which is unlikely [Yu and Elderfield, 2008].   Here we attempt to improve upon these estimates. First, for our Paleogene samples, we use Mg/Ca and δ 18 O data from the same site to reduce the uncertainty associated with BWT estimates. Second, we use a range of samples from peak greenhouse conditions (52-57 Ma) rather than sparse data from~48 Ma [Lear et al., 2000;Evans and Müller, 2012]. This improves our confidence in the assumption of an ice-free world. Third, rather than combining epifaunal and infaunal Mg/Ca records, we use only the infaunal species O. umbonatus. Finally, we do not introduce an ad hoc correction for carbonate saturation state [Cramer et al., 2011] but focus on ODP Site 690, a carbonate-rich site at an intermediate paleowater depth, which shows relatively minor dissolution even during the extreme dissolution of the Paleocene-Eocene Thermal Maximum [Thomas, 1998;Zeebe and Zachos, 2007;Kelly et al., 2012]. We therefore assume that O. umbonatus precipitated its calcite test in pore waters buffered against changes in carbonate saturation state at ODP Site 690 in the Paleogene.
We assume that the O. umbonatus Mg/Ca temperature Mg/Ca sw relationship has one of two forms, depending on whether the Mg/Ca temperature relationship is best described as an exponential or linear relationship [Ries, 2004]: Values for A, B, C, m, and H can be calculated by solving each equation simultaneously to satisfy both the Paleocene-Eocene data set ( Figure 4) and the Mg/Ca temperature relationship in core top samples, for which Mg/Ca sw = 5.2. This approach is sensitive to the choice of Mg/Ca temperature calibration [Cramer et al., 2011].
The first O. umbonatus Mg/Ca temperature calibration is based on core top data from the North Atlantic, offshore Hawaii, Gulf of California, Sea of Okhotsk, Southern Ocean, and Little Bahama Banks [Lear et al., 2002]. An exponential curve fit through the data spans a temperature range of 0.8 to 9.9°C (equation (4)).
Mg=Ca ¼ 1:528 Â exp 0:09 Â BWT ðÞ The Lear et al. [2002] core top data set used foraminifera that had been oxidatively and reductively cleaned, whereas the Rathmann et al. [2004] data set is based on laser ablation analyses of noncleaned foraminifera. Direct comparison of the absolute values is therefore not trivial, yet the similarity in temperature sensitivities is encouraging. However, a study of core top O. umbonatus Mg/Ca from deep ocean sites with temperatures <4°C [Healey et al., 2008] estimated a temperature sensitivity more than double that of Lear et al. [2002] (equations (6) and (7)). It seems more than likely that this sensitivity is high because of the influence of low carbonate saturation states on Mg/Ca [Healey et al., 2008].
Mg=Ca ¼ 0:988 ± 0:08 Â exp 0:252 ± 0:036 Â BWT ðÞ  [Lear et al., 2002], excluding one obvious outlier (Table S3 in the supporting information). We thus maximize the temperature range of the core top calibration data set (À0.8 to 9.9°C), while minimizing the influence of carbonate ion effects. The resulting Mg/Ca temperature relationship can be cast as an exponential or linear equation (equations (9) and (10)), referred to as "NS-LBB-exp" and "NS-LBB-lin," respectively. However, we add a note of cau  Lear et al., 2010], perhaps suggesting a limit to the buffering capacity of clay-rich sediments.
In calculating the constants in equations (2) and (3), the temperature variable is relatively well constrained because the modern calibrations are based on measured temperatures, and we use interpolated δ 18 O measurements from ODP Site 690 for the Paleocene-Eocene data set, assuming ice-free conditions (section 2.3, Figure 4). We adjust the 690 Mg/Ca record by À9.1% to account for the different cleaning technique relative to the core top samples . Echinoderm Mg/Ca ratios have been used to estimate Mg/Ca sw at 53 Ma as 1.7 mol/mol [Dickson, 2002], although subsequent reinterpretation of these data using a power law dependency has revised this to 1.3 mol/mol [Hasiuk and Lohmann, 2010]. Our calculations assume that  seawater Mg/Ca was invariant over the duration of the Paleocene-Eocene data set; we compensate for this simplification by using a range of Paleocene-Eocene Mg/Ca sw of 1.0, 1.3, and 2.5 mol/mol to calculate the constants in equations (2) and (3) (Tables 1 and 2). We also calculated the calibration constants after omitting the oldest data point described above (57.1 Ma). The values are within the error of the constants presented in Tables 1 and 2 and are provided in the supporting information.
Using the Lear et al. [2002] Mg/Ca temperature calibration, an H value of 0.2-0.5 is predicted, depending on the assumed value of Mg/Ca sw in the Paleocene-Eocene (Table 1). Reconciling the Mg/Ca temperature sensitivities of the NS-LBB-exp, NS-LBB-lin and the Lear et al. [2010] calibration with our Paleocene-Eocene data set requires O. umbonatus to be more or less insensitive to Mg/Ca sw (H ≈ 0). This would suggest that this deep ocean, low-Mg calcite foraminifer has an effective strategy for controlling the Mg 2+ concentration at the site of biomineralization, such that test Mg/Ca is only weakly related to Mg/Ca sw . While there is a range of potential biomineralization processes that may influence test Mg/Ca, the exact mechanism remains unknown [Erez, 2003;Dawber and Tripati, 2012;Nehrke et al., 2013;Mewes et al., 2014]. Further study is required to test this finding, but if H = 0, the Paleocene-Eocene data can be combined with the core top data to extend the Mg/Ca temperature calibration into warmer waters and better constrain the Mg/Ca temperature sensitivity (Figure 4d). The O. umbonatus Mg/Ca data from the Brazilian Margin are compatible with this relationship [Tisserand et al., 2013] (Figure S2). A compilation of the NS-LBB calibration data with the ODP Site 690 data can be fit equally well with an exponential (equation (11)) or linear (equation (12)) relation. The slope of the linear fit is within the error of a previous calibration based on core top O. umbonatus and C. pachyderma [Lear et al., 2010]. Stated errors in equations (11) and (12) Figure 3) [Lear et al., 2010] are~0.4 mmol/mol higher than those of time-equivalent samples at ODP Site 806 (Ontong Java Plateau). The magnitude of this offset is somewhat puzzling, given the similarity in δ 18 O between the two sites ( Figure 3). We consider two possibilities to explain this offset. One possibility is that the higher Mg/Ca at Site 761 is caused by diagenetic carbonate overgrowth. The low sedimentation rates at this site (~0.3 cm/kyr, compared to~2.5 cm/kyr at ODP Site 806) may have favored the formation of Mn carbonate coatings [Jarvis et al., 2001]. Measured Mn/Ca typically falls between 100 and 200 μmol/mol at Site 761, compared to 0-100 μmol/mol at ODP Site 806. However, the Site 761 Mn/Ca record shows an overall increase across the MMCT, whereas the Mg/Ca shows an overall decrease ( Figure S3). This does not rule out a diagenetic contribution per se, because the influence of the coatings would be imprinted onto an unknown primary signal, but it makes it more unlikely. Mn can exist in different mineral phases (e.g., different Mn oxide minerals, Mn adsorbed onto calcite, Ca-Mn carbonates, or Mn-Ca-Mg carbonates) [Boyle, 1983]. Therefore, samples with Mn/Ca below the proposed threshold of 100 μmol/mol [Boyle, 1983] may be considered free of Mn-related contamination, but it is not necessarily true that all samples with Mn/Ca above this threshold are contaminated with respect to Mg/Ca. Pacific ODP Site 1218 provides an example of a site with significant foraminiferal Mn contamination. The ODP Site 1218 record from 20 to 40 Ma displays large variations in Mn/Ca values in reductively cleaned foraminifera, from <100 μmol/mol to >1000 μmol/mol ( Figure S4). Despite this order of magnitude variation in Mn content, there is no relationship between the Mn/Ca and Mg/Ca records, and the highest Mn/Ca values are associated with the lowest Mg/Ca ratios. However, the origin of the Mn at Site 1218 is likely different from that at Site 761. Authigenic (Mn x ,Mg y )CaCO 3 coatings on Pleistocene foraminifera at Caribbean Site 999 resulted in measured Mn/Ca similar to those observed at Site 761 (<500 μmol/mol) [Schmidt et al., 2006]. Assuming a 10:1 Mn/Mg molar ratio of the authigenic carbonates, the authors demonstrate that less than 0.05 mmol/mol of their measured Mg/Ca could originate from such coatings. Even if we arbitrarily assume a tenfold Mg enrichment in the authigenic carbonate relative to that observed in the Panama Basin [Pedersen and Price, 1982], the Mn coatings could not account for more than 0.2 mmol/mol of the Paleoceanography 10.1002/2015PA002833 measured Mg/Ca. We note that there is no offset in C. mundulus δ 13 C between ODP Sites 806 and 761 ( Figure S5). We, therefore, discount this explanation for the intersite offset in Mg/Ca.  (Figure 3). This is consistent with the hypothesis that ODP Site 761 pore waters have a higher degree of carbonate saturation than those at ODP Site 806 . ,one would expect a corresponding offset in B/Ca~11 μmol/mol, in excellent agreement with our record (Figure 3). However, we do not observe an intersite offset in O. umbonatus Sr/Ca, and the observed intersite offset in O. umbonatus Li/Ca is twice the expected magnitude, albeit in the expected direction (higher Li/Ca at ODP Site 761). This may point to uncertainties in core top ΔCO 3 2À calibrations, or additional, unknown processes at work.
The reasons for the hypothesized increased pore water saturation state at ODP Site 761 are unknown. However, we note that the pore water [Mg] and [Ca] profiles at this site are atypical for pelagic marine sediments, in that their concentrations remain relatively constant from the sea floor down to~300 m below seafloor [De Carlo, 1992]. As the organic carbon content of the sediment is low, these trends are unlikely to be caused by enhanced dissolution due to remineralization of organic matter. Instead, these profiles may reflect unusual flushing of the sediments by (oversaturated) seawater, perhaps aided by the unusually low sedimentation rates [De Carlo, 1992]. Hence, it is plausible that the ODP Site 761 Mg/Ca might be biased by the unusual pore water chemistry at the site.

Estimating Neogene BWT and δ 18 O sw
We use our calibration equations (Tables 1 and 2) and a record of Mg/Ca sw from fluid inclusion, calcite vein and echinoderm proxies [Dickson, 2002;Horita et al., 2002;Coggon et al., 2010] to calculate absolute BWT from our ODP Site 806 Mg/Ca record. We fit a third-order polynomial curve through the compiled Mg/Ca sw data and use an uncertainty window of ±0.5 mol/mol to encompass most of the proxy data ( Figure 5). Each BWT scenario is then used in conjunction with the δ 18 O record from the same samples and the Cibicidoides quadratic δ 18 O paleotemperature equation of Marchitto et al. [2014] to calculate δ 18 O sw . The result is a suite of alternative calculated records for the evolution of deep Pacific bottom water temperature and δ 18 O sw since 17 Ma (Figure 6). The range in predicted temperature and δ 18 O sw results from a combination of the following: Figure 5. Third-order polynomial curve fit through compiled seawater Mg/Ca proxy records based on fluid inclusions, calcite veins, and echinoderms [Dickson, 2002;Horita et al., 2002;Coggon et al., 2010]. The grey envelope represents the ±0.5 mol/mol uncertainty window used in our temperature calculations.

10.1002/2015PA002833
(1) differences between Mg/Ca temperature calibrations, (2) the 1.0-2.5 mol/mol range of assumed Paleocene-Eocene Mg/Ca sw used to calculate the constants in Tables 1 and 2, and (3) the ±0.5 mol/ mol envelope on the Mg/Ca sw curve ( Figure 5). Refining the Mg/Ca temperature calibration is key to improving estimates of absolute temperature and δ 18 O sw over long (Myr) timescales ( Figure 6). The Pleistocene portion of our record is at too low a resolution to capture the amplitude of glacialinterglacial cycles. Based on the expected range of Pleistocene δ 18 O sw [Duplessy et al., 2002;Elderfield et al., 2012], our preferred linear Mg/Ca temperature calibration is that of Lear et al. [2010] and our preferred exponential Mg/Ca temperature calibration is that of Lear et al. [2002] (Figure 6). We show the full range of absolute BWT and δ 18 O estimates based on these two calibrations (Figure 7). The uncertainty window on these absolute estimates results from the set of constants corresponding to the 1.0-2.5 mol/mol range assumed for late Paleocene seawater Mg/Ca (Tables 1 and 2) and the ±0.5 mol/mol envelope on the Neogene Mg/Ca sw curve ( Figure 5). After calculating the maximum and minimum temperatures produced from different combinations of these parameters through time, an additional ±1°C uncertainty envelope was added to the record to include sample reproducibility and calibration errors [Elderfield et al., 2012].

Neogene Ice Sheet Evolution and Ocean Cooling
The low-resolution δ 18 O and Mg/Ca records from ODP Site 806 (Figure 2) capture the broad trends in late Neogene Pacific deep water temperature and δ 18 O sw (Figure 7). The records display three noteworthy features, which we discuss in more detail below. First, bottom waters cooled between 15.5 and 14.5 Ma (following the Miocene Climatic Optimum), but there was apparently no significant cooling associated with major ice growth steps between 14 and 12 Ma. Second, δ 18 O sw has increased by~0.5‰ overall since~15 Ma (prior to the MMCT), with an interval of heavy δ 18 O sw between 12 and 8 Ma that could be interpreted as reflecting bipolar glaciation or the presence of a relatively warm salty water mass at Site 806. We acknowledge that these absolute estimates of δ 18 O sw are poorly constrained. Improved core top calibrations will significantly reduce the uncertainties (Figure 6). The third feature is the major deep water cooling since 5 Ma. The Plio-Pleistocene drop in sea level is likely within the uncertainty of this broad reconstruction and cannot be inferred from this low-resolution record.

Neogene Decoupling of Deep Water Temperature and Ice Volume
The decoupling of deep water/high-latitude temperature and ice volume appears to be a consistent feature throughout the late Neogene (Figure 7). This supports our understanding of the cryosphere as a highly thresholded system, with ice growth occurring once the balance between ice accumulation and ablation is favorable [DeConto et al., 2008]. 4.4.2. Greater than Modern Ice Volume Following the MMCT Assuming that the δ 18 O sw signal can be interpreted in terms of global ice volume, our record suggests greater than modern ice volume following the Middle Miocene Climate Transition. The Pleistocene δ 18 O sw -sea level  Tables 1 and 2. Equations based on Lear et al. [2002] are shown in red, on Lear et al. [2010] in green, the NS-LBB linear calibration in blue, and the NS-LBB exponential calibration in black. Scenarios that assume early Eocene seawater Mg/Ca = 1.0 are marked by long dashes, seawater Mg/Ca = 1.3 by solid lines, and seawater Mg/Ca = 2.5 by dotted line.

10.1002/2015PA002833
calibration [Fairbanks and Matthews, 1978] indicates that the~1‰ increase in δ 18 O sw between 14.5 and 12.5 Ma is equivalent to a~90 m sea level fall. Put into context, the modern Antarctic ice sheet contains approximately 60 m sea level equivalent ice, and during the Last Glacial Maximum (LGM), sea levels were approximately 120 m lower than today, with around~14 m attributed to a larger Antarctic ice sheet [Denton and Hughes, 2002;Briggs et al., 2014]. The amount of ice that can be supported on Antarctica depends not only on climate parameters but also on paleotopography and paleogeography. Drilling off the East Antarctic coast has revealed a dynamic Miocene history for the Antarctic ice sheet [Escutia et al., 2005;Passchier et al., 2011]. Age model uncertainties prevent us from correlating intervals of East Antarctic ice sheet advance to our δ 18 O sw records, but diamictite textures indicating offshore grounding lines provide evidence for a greater ice sheet extent than today during the middle Miocene [Passchier et al., 2011]. Paleogeographic reconstructions demonstrate that Antarctic land area has decreased since the earliest Oligocene, with early Oligocene climate and topography supporting an Antarctic ice sheet~35% larger than today . Furthermore, geomorphological evidence indicates that a thicker-than-modern ice sheet, capable of cross-cutting valleys, overrode the Transantarctic Mountains in the middle Miocene [Denton et al., 1984;Denton and Sugden, 2005]. These lines of evidence suggest that Middle Miocene Antarctic paleogeography and climate were able to support a far larger ice sheet than today. However, it is surprising that the high absolute δ 18 O sw values persisted and even slightly increased throughout the late Miocene (Figure 7).   (Figures 7b and 7d). The uncertainty envelope is derived from the range in calibration constants in Tables 1 and 2, the range in seawater Mg/Ca shown by the grey envelope in Figure 5 and a further ±1°C uncertainty to reflect sample reproducibility and calibration error. The solid grey lines represent our favored scenarios, which assume Paleocene-Eocene seawater Mg/Ca =1.3 and use values for Neogene seawater Mg/Ca shown by the solid black line in Figure 5. These records assume constant pore water saturation state. The Pacific zonal sea surface temperature gradient (red line, Figures 7a and 7b) has been calculated from TEX 86 records [Zhang et al., 2014], and the benthic δ 18 O record (blue line) was compiled by Zachos et al. [2008].

10.1002/2015PA002833
We consider two explanations for this unexpected finding. The first is that there was a period of significant Northern Hemisphere glaciation, culminating around 8 Ma, and the second is that the Site 806 δ 18 O sw record includes a significant component of regional salinity change. Middle Miocene paleotopographic reconstructions indicate a high relief (>1500 m) in the northwestern Barents Sea. Ice rafted debris (IRD) fingerprinting has been interpreted in terms of significant glaciation and iceberg calving in the Barents Sea region from 14 Ma [Knies and Gaina, 2008]. Perennial ice has provided a constant source of IRD from all the circum-Arctic shelves since~15 Ma [Darby, 2008], and Scandinavian ice, as inferred from IRD on the Vöring Plateau, initiated around 12.6 Ma [Thiede et al., 1998]. Therefore, there may have been a Northern Hemisphere glaciation component following the MMCT, consistent with the hypothesis that atmospheric CO 2 levels crossed the proposed threshold for Northern Hemisphere glaciation at the Middle Miocene Climate Transition [Foster et al., 2012;DeConto et al., 2008;Holbourn et al., 2013a]. However, evidence for extensive Northern Hemisphere continental ice sheets prior to~7 Ma is lacking [Larsen et al., 1994].
Alternatively, our Site 806 δ 18 O sw record includes a significant regional salinity signal. In the middle Miocene, ODP Site 806 likely lay in the path of Pacific Central Water, the return flow of Circumpolar Deep Water [Holbourn et al., 2013b]. Intersite stable isotope gradients indicate enhanced formation of North Atlantic Deep Water (NADW) during the early Pliocene relative to today, and this relatively warm, salty water mass may have influenced deep Pacific waters [Kwiek and Ravelo, 1999;Ravelo and Andreasen, 2000]. Closure of the Tethys Seaway may have increased Atlantic sea surface salinity and promoted enhanced formation of NADW since the middle Miocene, with consequent cooling of the Southern Ocean [Zhang et al., 2011] and an increased NADW influence on Circumpolar Deep Water [Holbourn et al., 2013b]. We note that the upper Miocene benthic foraminiferal δ 18 O record from ODP Site 806 appears to diverge from the compiled adjusted δ 18 O record of abyssal Pacific Deep Sea Drilling Project Site 77 (paleowater depth~4 km [Cramer et al., 2009] ( Figure S6). We speculate that warmer, saltier waters bathed ODP Site 806, while the Pacific abyss was bathed by a fresher, colder water mass sourced from high latitudes. High-resolution quantitative reconstructions of δ 18 O sw from different ocean basins are required to tease out the global ice volume and regional salinity components of the Site 806 δ 18 O sw record and hence the link between climate sensitivity and ocean overturning regime.

Plio-Pleistocene Deep Water Cooling
Our records display a striking cooling trend of~4-6°C from around 5 Ma to~2.5 Ma, preceding the onset of large-amplitude glacial-interglacial cycles in the late Pliocene [Raymo et al., 1989;Bailey et al., 2013] (Figure 7). The later half of this trend corresponds to a sea surface cooling linked to increased iceberg longevity in the Southern Ocean [Cook et al., 2014]. However, the early phase (between 5 and 4 Ma) of bottom water cooling at Site 806 does not appear to be associated with significant cooling in the Southern Ocean. This cooling might reflect increasing dominance of a cold water mass, rather than a cooling in its source region. In the early Pliocene, north Pacific ODP Site 1018 (2476 m water depth) was apparently influenced by a water mass that was warmer and saltier than modern NADW [Kwiek and Ravelo, 1999]. We speculate that this water mass also influenced ODP Site 806 but became less important between 5 and 4 Ma. The cooling is paralleled by an increase in the tropical Pacific zonal sea surface temperature gradient (Figure 7), which may have been driven by an increased meridional sea surface temperature gradient [Zhang et al., 2014]. The associated thinning of the low-latitude thermocline may have increased climate sensitivity to CO 2 , thus setting the stage for the subsequent glaciation of the Northern Hemisphere [LaRiviere et al., 2012].

Conclusions
Benthic foraminiferal Mg/Ca paleothermometry is a powerful tool for reconstructing bottom water temperatures and δ 18 O sw . Infaunal foraminifera calcify in pore waters that may be buffered from changes in overlying bottom water saturation state. We recommend that paired B/Ca and Mg/Ca records are used to assess this assumption, before Mg/Ca records are interpreted solely in terms of bottom water temperature. We suggest that sediments with abundant planktonic foraminifera (which dissolve more easily than benthics) may have a higher buffering capacity than sediment without or with rare planktonic foraminifera, and/or a high fragmentation index. Paired Mg/Ca and δ 18 O records from the ice-free late Paleocene to early Eocene imply a low sensitivity of O. umbonatus Mg/Ca to seawater Mg/Ca, suggesting that this infaunal deep water species exerts a strong biological control on its calcification process. We provide a range of possible Mg/Ca-BWT-Mg/Ca sw Paleoceanography 10.1002/2015PA002833 calibrations for calculating absolute BWT and δ 18 O sw , encompassing uncertainties resulting from core top calibrations and Cenozoic Mg/Ca sw reconstructions. Uncertainties in reconstructed absolute BWT and δ 18 O sw are large compared to relative changes in BWT and δ 18 O sw as determined over short timescales (<1 Myr). Nevertheless, when applied to an O. umbonatus Mg/Ca record from ODP Pacific Site 806, our calibrations imply greater than modern ice volume following the Middle Miocene Climate Transition. Bottom waters at Site 806 cooled in the Pliocene, as meridional and tropical zonal sea surface temperature gradients increased, thus supporting the idea that a change in ocean hydrographic regime played an important role in Earth'st r a n s i t i o n from a polar to bipolar glaciated world. The apparent decoupling of deep water/high-latitude temperatures and ice volume highlights the sensitivity of the cryosphere to thresholds and internal feedbacks in the climate system.