Southern Ocean bottom water cooling and ice sheet expansion during the middle Miocene climate transition

The middle Miocene climate transition (MMCT, ~14.5–13.0 Ma) was associated with a significant expansion of Antarctic ice, but the mechanisms triggering the event remain enigmatic. We present a new clumped isotope (∆47) bottom water temperature (BWT) record from 16.0 Ma to 12.2 Ma from Ocean Drilling Program (ODP) Site 747 in the Southern 15 Ocean, and compare it to existing BWT records. We show that BWTs in the Southern Ocean were ~8–10°C during the middle Miocene greenhouse, and thus considerably warmer than today. Nonetheless, bottom water δ18O (calculated from foraminiferal δ18O and ∆47) suggests substantial amounts of land ice throughout the interval of the study. Our dataset demonstrates that BWTs at Site 747 decreased by ~3–5°C across the MMCT. This cooling preceded the stepped main increase in global ice volume, and appears to have been followed by a transient bottom water warming starting during or 20 slightly after the main ice volume increase. We speculate that a regional freshening of the upper water column at this time may have increased stratification and reduced bottom water heat loss to the atmosphere, counteracting global cooling in the bottom waters of the Southern Ocean and possibly even at larger scales. Additional processes and feedbacks required for substantial ice growth may have contributed to the observed decoupling of Southern Ocean BWT and global ice volume.


Introduction 25
During the Cenozoic Era (the last 65 Myr), Earth's climate transitioned from a state of expansive warmth with very limited ice to colder conditions and permanent ice sheets at the poles (Zachos et al., 2001). The middle Miocene climate transition (MMCT, ~14.5-13 Ma) represents one of the main steps of Cenozoic cooling. A substantial increase in benthic foraminiferal oxygen isotope ratios (δ 18 O) during the MMCT has been interpreted to reflect a combination of decreasing bottom water temperatures (BWTs) and ice sheet expansion (increasing bottom water δ 18 O) occurring in the Southern Hemisphere (Lear et 30 al., 2015;Lewis et al., 2007). A roughly coeval decrease in atmospheric pCO2 of ~100-300 ppm was estimated based on https://doi.org/10.5194/cp-2020-157 Preprint. Discussion started: 18 December 2020 c Author(s) 2020. CC BY 4.0 License.
The carbonate clumped isotope (∆47) paleothermometer is based on the measured abundance of 13 C-18 O bonds relative to 65 their stochastic distribution Schauble et al., 2006), and is independent of the isotopic composition of the parent water from which the carbonate grew (e.g., Eiler, 2011). On the basis of current knowledge, other environmental variables such as pH and salinity appear to be of minor importance for measured ∆47 values over the range of natural variation (Tripati et al., 2015;Watkins and Hunt, 2015). When applied to foraminiferal calcite, the method also does not show detectable species-specific vital effects (Grauel et al., 2013;Meinicke et al., 2020;Modestou et al., 2020;Peral et al., 70 2018;Piasecki et al., 2019;Tripati et al., 2010). Consequently, the ∆47 thermometer holds great promise for reconstructing accurate BWTs from benthic foraminiferal tests (e.g., Leutert et al., 2019). The ∆47 thermometer has been previously applied to middle Miocene benthic foraminifera from ODP Site 761 in the Indian Ocean yielding results that are in good agreement with Mg/Ca BWTs from the same site Modestou et al., 2020). However, there are intervals with very low data density in the middle Miocene record from Site 761, limiting its informative value for understanding the drivers of the 75 MMCT. At the same time, middle Miocene BWTs in other key regions such as the Southern Ocean remain poorly constrained. Here, we present ∆47-based BWTs measured on benthic foraminiferal calcites from ODP Site 747 located on the Kerguelen Plateau in the Southern Ocean (Fig. 1). We compare our new absolute BWT record to previous BWT estimates for the middle Miocene, and interpret the BWT records in the context of middle Miocene glaciation and CO2 drawdown.  Schlich et al., 1989). At present, the site is situated south of the Polar Front and is bathed by Circumpolar Deep Water (CDW) with a temperature of ~1-2°C (Belkin and Gordon, 1996;Billups and Schrag, 2002). The middle Miocene geographic position of Site 747 relative to Antarctica was similar to today (e.g., Abrajevitch et al., 2014). The clumped 85 isotope record generated in this study covers the depth interval from 62.64 m below sea floor (mbsf, Sample 747A-7H-5, 14-16 cm) to 85.36 mbsf (Sample 747A-9H-8, 75-77 cm) in Hole 747A. 191 samples (15-20 cm 3 , mostly calcareous nannofossil ooze with foraminifera) were taken continuously with a mean temporal resolution of ~20 kyr. We slightly rescaled the originally assigned shipboard sample depths to account for core expansion (Schlich et al., 1989), similar to previous studies focusing on the middle Miocene section of Hole 747A (e.g., Abrajevitch et al., 2014;Majewski and Bohaty, 2010). 90 https://doi.org/10.5194/cp-2020-157 Preprint. Discussion started: 18 December 2020 c Author(s) 2020. CC BY 4.0 License. and 761 are ~1700 m and ~2200 m, respectively Schlich et al., 1989). Annual mean temperatures at these depths are shown in (a) and (b). Temperatures from the 2013 World Ocean Atlas (Locarnini et al., 2013) visualized with Ocean Data View (Schlitzer, 2019). Inset map with paleogeographic reconstruction from the plate tectonic reconstruction service of the Ocean Drilling Stratigraphic 95 Network (http://www.odsn.de).

Sample material
Each sediment sample was freeze-dried, washed over a 63 µm sieve, oven-dried at 50°C and then dry-sieved into different size fractions. We mainly picked tests of Cibicidoides mundulus from the 250-355 µm size fraction for our measurements. 100 For samples with low abundances of benthic foraminifera in this size fraction, the >355 µm size fraction was also included.
The interval from ~16.0 Ma to ~15.3 Ma was additionally complemented with measurements on Cibicidoides wuellerstorfi. Middle Miocene benthic foraminifera (and more specifically Cibicidoides) from Site 747 were previously described as well preserved (e.g., Abrajevitch et al., 2014;Billups and Schrag, 2002). Our examination appears to confirm the impression of relatively good preservation of middle Miocene Cibicidoides at Site 747 (see Figs. S1 and S2). Similar to other studies (e.g., 105 Billups and Schrag, 2002;Gottschalk et al., 2016;Yu and Elderfield, 2008), C. mundulus and C. wuellerstorfi are grouped into a single genus. We note that some of the analysed specimens of C. mundulus and C. wuellerstorfi closely resemble the sensu lato morphotype of the respective species (shown in Fig. 2 of Gottschalk et al., 2016 Prior to isotope analysis, we cracked the picked specimens and ultrasonicated the test fragments in deionized water (3×30 seconds) and methanol (1×10-30 seconds) to remove adhering sediment. Test fragments were rinsed once between 110 each ultrasonication step and at least three times at the end of the cleaning. The cleaned test fragments were subsequently oven-dried at 50°C.

Stable isotope measurements and data processing
Low abundances of carbonate ions containing both 13 C and 18 O isotopes require stringent analytical procedures and comparably large sample sizes to obtain clumped isotope temperatures that are precise enough for Cenozoic ocean 115 temperature reconstructions. We achieve the necessary precision by averaging over ~30-40 clumped isotope values measured on small (~100 μg) carbonate samples Hu et al., 2014;Meckler et al., 2014;Schmid and Bernasconi, 2010). Results from adjacent samples are pooled to achieve this number of measurements (e.g., Grauel et al., 2013;Rodríguez-Sanz et al., 2017), due to the generally low abundance of mono-specific benthic foraminifera (allowing for only 1-5 individual measurements per sample, Fig. S3b). Producing a low-resolution clumped isotope temperature record 120 with this approach yields higher-resolution δ 18 O and δ 13 C time series in parallel.
Clumped isotope measurements were performed using two Thermo Scientific MAT 253 Plus mass spectrometers at the University of Bergen, Norway, and one Thermo Scientific MAT 253 mass spectrometer at ETH Zurich, Switzerland. All mass spectrometers were coupled to Thermo Fisher Scientific Kiel IV carbonate preparation devices. CO2 gas was extracted from a carbonate sample with phosphoric acid at a reaction temperature of 70°C, as reported in Schmid et al. (2012). A 125 Porapak trap included in each Kiel IV carbonate preparation system was kept at -20°C to remove organic contaminants from the sample gas (Schmid et al., 2012). Between each run, the Porapak trap was baked out at 120°C for at least one hour for cleaning. Every measurement run included a similar number of samples and carbonate standards. Four carbonate standards (ETH-1, ETH-2, ETH-3 and ETH-4) with different isotopic compositions and ordering states were used for monitoring and correction of the results (see Appendix A for details). External reproducibilities (one standard deviation) in corrected 130 ∆47 values of ETH-1, ETH-2, ETH-3 and ETH-4 were typically between 0.030 ‰ and 0.040 ‰ (Table S2). External reproducibilities (one standard deviation) for δ 18 O and δ 13 C values of the same standards (given relative to VPDB) were 0.03-0.10 ‰ and 0.02-0.06 ‰, respectively.
We converted the sample ∆47 values (averages over ~30-40 separate measurements each) into temperature using a calibration based on various recent datasets from core top-derived foraminifera, corrected with the same carbonate standards 135 as used in our study (Eq. (2) of Meinicke et al. (2020)): (1) This combined calibration has been recommended for foraminifer samples (Meinicke et al., 2020). We note that the 140 individual datasets in this compilation (Meinicke et al., 2020;Peral et al., 2018;Piasecki et al., 2019) are all in good agreement with a travertine-based calibration (Kele et al. (2015), recalculated by Bernasconi et al. (2018)) spanning a wider temperature range (6-95°C). For consistency, previously published ∆47-based ocean temperatures from ODP Sites 761 (Modestou et al., 2020) and 1171 (Leutert et al., 2020) originally based on the travertine calibration were recalculated with the calibration equation of Meinicke et al. (2020). We propagated analytical and calibration uncertainties in ∆47-based 145 temperatures (as described in the supporting information of Huntington et al. (2009)), and report combined uncertainties as 68 % and 95 % confidence intervals. ∆47-based temperatures were used in combination with benthic foraminiferal δ 18 O values to calculate δ 18 Obw (reported relative to VSMOW) with Eq. (9) of Marchitto et al. (2014).

Age models
We revised the Hole 747A age model by integrating six magnetostratigraphic tie points (Abrajevitch et al., 2014;Majewski 150 and Bohaty, 2010) on the GTS2012 timescale (Gradstein et al., 2012), three benthic foraminiferal δ 13 C-based tie points associated with the "Monterey" carbon-isotope excursion (using the nomenclature of Holbourn et al. (2007)), and one peak warm event visible in benthic foraminiferal δ 13 C and δ 18 O   (Table S1). For the δ 13 C-based tie points, we used the high-resolution isotope stratigraphies of IODP Sites U1335, U1337 and U1338 in the eastern equatorial Pacific Ocean Kochhann et al., 2016) as reference (Fig. 2c). In addition, we included a hiatus at the core 155 break between Cores 7H and 8H, identified by previous studies (e.g., Majewski and Bohaty, 2010). δ 18 O and δ 13 C time series of Sites 747, 761, 806, U1335, U1337 and U1338 are shown in Fig. 2 with isotope-based age tie points for Site 747 as black crosses. The age model for Site 761 is from Leutert et al. (2020). For Site 806, we utilized a previously published orbitally tuned age model from ~14.1 Ma to ~13.3 Ma ; ages for the older and younger parts of the Site 806 record (~16.4-14.1 Ma and ~13.3-12.3 Ma) were derived by assuming the same mean sedimentation rate as between 160 ~14.1 Ma and ~13.3 Ma.

Benthic foraminiferal δ 18 O and δ 13 C values
The isotope record of Site 747 (Fig. 2) displays features typical of middle Miocene sequences, including the stepped increase in benthic δ 18 O across the MMCT and the pronounced δ 13 C maxima associated with the "Monterey" carbon isotope 165 excursion (e.g., Holbourn et al., 2007Holbourn et al., , 2014Kochhann et al., 2016;Vincent and Berger, 1985). From ~16.0 Ma to ~15.3 Ma, we analysed stable isotope compositions of both C. mundulus and C. wuellerstorfi, allowing for a direct assessment of species-specific effects on the isotopic compositions of these two different epifaunal species (Fig. 2a and c). δ 18 O values https://doi.org/10.5194/cp-2020-157 Preprint. Discussion started: 18 December 2020 c Author(s) 2020. CC BY 4.0 License. measured on C. mundulus and C. wuellerstorfi appear indistinguishable, whereas a consistent offset of up to ~0.5 ‰ exists between the δ 13 C values of these species at Site 747. Similar δ 13 C offsets between C. mundulus and C. wuellerstorfi have 170 been previously observed for the sub-Antarctic Atlantic during the Quaternary (Gottschalk et al., 2016). Our δ 13 C values from the middle Miocene underscore the need to carefully examine inter-species offsets in δ 13 C before combining different species to produce a single δ 13 C curve.
https://doi.org/10.5194/cp-2020-157 Preprint. Discussion started: 18 December 2020 c Author(s) 2020. CC BY 4.0 License. Southern Ocean  this study), ODP Site 761 in the eastern Indian Ocean Lear et al., 2010;Modestou et al., 2020), ODP Site 806 in the western equatorial Pacific Lear et al., 2015) as well as IODP Sites U1335, U1337 and U1338 in the eastern equatorial Pacific Ocean Kochhann et al., 2016). Correlation tie points for Site 747 (this study) are visualized with black crosses. We only plot δ 18 O and δ 13 C values from Sites 747 and 761 that were measured on the species C. mundulus (mun) and C. wuellerstorfi (wuel). In contrast to Site 747, offsets in both δ 18 O and δ 13 C between 180 these species appear minimal at Site 761 . We note that we also use ∆47 values from other benthic foraminiferal species from Site 761 (see Modestou et al. (2020) for details), as no species-specific vital effects on benthic foraminiferal ∆47 have been observed (Modestou et al., 2020;Piasecki et al., 2019). For Site 806, we show δ 18 O values of Cibicidoides spp. (cibs) , in addition to δ 18 O and δ 13 C measured specifically on tests of C. mundulus and C. wuellerstorfi . δ 18 O, δ 13 C and ∆47 at Sites 747 and 761 were measured several times per sample in this study and Modestou et al. (2020). See Fig. S3 for ∆47 values and 185 number of replicate measurements for each sediment sample.

Revised estimates of bottom water temperature for the middle Miocene
As expected, the ∆47 signal (Fig. S3a) is much noisier in comparison to δ 18 O and δ 13 C (Fig. 2), necessitating an averaging of ∆47 over many adjacent samples before interpreting the data in terms of calcification temperature. We have visualized our   747 (this study) 761 (Modestou et al., 2020) 747  806  (b)  (Modestou et al., 2020) in the same way as our results from Site 747 (e.g., temperature calibration, smoothing) to optimize comparability of BWTs from these two middle Miocene reference sites. Similar absolute 220 BWTs at Sites 747 and 761 during the Miocene may be expected from their similar present-day temperature ranges (~1-2°C; Other available Mg/Ca-based BWT records covering the MMCT do not show the same features. Similar to Site 806, Mg/Ca ratios were also measured on the infaunal species O. umbonatus at Site 761 . This approach yields BWTs 235 that are within uncertainty of those from ∆47 measured at the same site (Fig. S4), regardless of whether or not the Mg/Cabased BWTs have been corrected for changes in saturation state (Modestou et al., 2020). However, Mg/Ca-based BWT estimates from that site have been deemed less reliable than those from Site 806, due to unusual and variable pore water chemistry at Site 761 , and are compromised by reduced data density in crucial intervals (as noted above).
In comparison to Sites 761 and 806 where an infaunal foraminiferal species has been used , Mg/Ca 240 records from Southern Ocean Sites 747 (Kerguelen Plateau) and 1171 (South Tasman Rise) were measured on the epifaunal species C. mundulus Shevenell et al., 2008). These Mg/Ca-based BWT records also do not show the same temperature pattern as our ∆47-derived BWT record from Site 747 or the Mg/Ca-derived BWT record from Site 806 (Fig. S4). The observed discrepancies suggest additional non-thermal controls on Mg/Ca and/or ∆47, which may be related to seawater chemistry during test precipitation and/or post-depositional alteration, such as dissolution.
Seawater chemistry does not appear to significantly influence ∆47 signatures in foraminifera over the range of natural variation (e.g., Tripati et al., 2015;Watkins and Hunt, 2015), whereas Mg/Ca signatures can be affected by changes in seawater Mg/Ca (Evans and Müller, 2012) and carbonate ion saturation (Elderfield et al., 2006;Yu and Elderfield, 2008). On the timescales considered here, the latter is more likely to be important. The relatively few Mg/Ca-based BWTs from Site 747 can be directly compared to our BWTs based on ∆47 from the same site (Fig 4a).  Fig. 4b and c). Mg/Ca-based temperatures from Site 747 were measured on foraminiferal tests of the epifaunal species C. mundulus; compared to infaunal foraminifera, this species lives in more direct contact with bottom water, and may thus be more prone to saturation state-related effects (Elderfield et al., 2006;Lear et al., 2015). The observation of diverging Mg/Ca-and ∆47-255 based BWTs in times of increased dissolution supports the interpretation of a possible saturation state effect on the Mg/Ca signatures of C. mundulus (see Fig. S5 for sensitivity study). For Site 1171, we do not have constraints on saturation state variability.
In addition to saturation state effects, variable dissolution itself (Fig. 4b and c) could have influenced foraminiferal Mg/Ca and/or ∆47 signatures. For planktic foraminifera, dissolution controlled by bottom water saturation has the potential to 260 significantly lower initial Mg/Ca signatures and thus also the estimated ocean temperatures in certain burial settings (e.g., Regenberg et al., 2014). Dissolution may also impact the Mg/Ca signatures of benthic foraminiferal tests, although the tests of benthic foraminifera appear generally denser and more resistant to dissolution than those of planktic foraminifera (e.g., Berger, 1973;Pearson et al., 2001). The effects of dissolution on benthic foraminiferal Mg/Ca have thus received little attention. Similarly, dissolution effects on benthic foraminiferal ∆47 signatures have not yet been specifically assessed. While 265 there is currently no evidence for a significant dissolution effect on foraminiferal ∆47 (e.g., Breitenbach et al., 2018;Leutert et al., 2019) or variable dissolution of benthic foraminiferal calcite at Site 747 during the interval of this study (Fig. S2), a potential effect of dissolution cannot be fully ruled out. We interpret ∆47-based temperatures as unaffected by dissolution in the absence of indications otherwise, but note that this aspect warrants further study. https://doi.org/10.5194/cp-2020-157 Preprint. Discussion started: 18 December 2020 c Author(s) 2020. CC BY 4.0 License.

Regional and global implications
δ 18 Obw signatures have been widely used to infer the evolution of global ice volume during the Neogene (e.g., Billups and Schrag, 2002;Lear et al., 2010Lear et al., , 2015Modestou et al., 2020;Shevenell et al., 2008). We calculated δ 18 Obw values from measured benthic foraminiferal δ 18 O in combination with ∆47-based BWTs (Fig. 5e). Benthic foraminiferal δ 18 O values of 280 the taxon Cibicidoides were averaged over the same intervals as have been used for ∆47 averaging. For Site 747, we used the δ 18 O values from this study (measured on C. mundulus and C. wuellerstorfi), whereas the foraminiferal δ 18 O values for Site 761 were compiled from existing studies Lear et al., 2010;Modestou et al., 2020). Due to comparably large random errors in our ∆47-based BWT estimates, the propagated uncertainties in δ 18 Obw are also large.
However, systematic biases in our δ 18 Obw estimates may be smaller compared to other methods (e.g., paired benthic 285 foraminiferal Mg/Ca and δ 18 O measurements) because ∆47 signatures seem to be largely independent of complicating environmental parameters and/or foraminiferal vital effects (e.g., Leutert et al., 2019;Peral et al., 2018;Piasecki et al., 2019;Tripati et al., 2015;Watkins and Hunt, 2015). The earliest interval of this study (>15.6 Ma) is characterized by δ 18 Obw values ranging from approximately -0.1 ‰ to 1.3 ‰. For the later MCO (15.6-13.9 Ma), our estimates of δ 18 Obw range from around -0.3 ‰ to 0.7 ‰, and increase to ~0.3-1.0 ‰ after the MMCT. All reconstructed δ 18 Obw values are consistently higher than 290 expected for minimal ice (i.e. -0.89 ‰ according to Cramer et al. (2011)). Therefore, our results suggest the presence of substantial ice sheets on Antarctica throughout the warm MCO, similar to previous estimates of middle Miocene δ 18 Obw (e.g., Modestou et al., 2020). This interpretation is robust towards utilizing different suggested δ 18 O-temperature relationships (Fig. S6). However, we cannot exclude the presence of short-lived (orbital-scale) minima in global ice volume during peak MCO interglacials (e.g., Levy et al., 2016), which may not be visible in the ∆47-based records due to their 295 temporal resolution.
Our Southern Ocean bottom water proxy record can be compared to evidence of past ice sheet variability from the Antarctic margin, in addition to orbital parameters. The stepped main increase in benthic δ 18 O starting between 13.8 Ma and 14.0 Ma (Fig. 5d) is associated with growing ice sheets (increasing δ 18 Obw, Fig. 5e) and occurs in an interval of low seasonal contrast over Antarctica (declining eccentricity, decreasing amplitude variations in obliquity, Fig. 5a), as pointed out by Holbourn et 300 al. (2005). The inferred ice volume increase recorded at Site 747 is supported by evidence for an episode of maximum ice sheet advance (MISA-4) recorded in the ANDRILL (AND)-2A drill core from the western Ross Sea, Antarctica (Levy et al., 2016). Similarly, an earlier period of maximum ice sheet advance documented in the Ross Sea around 14.7-14.6 Ma (MISA-3) corresponds to a maximum in δ 18 Obw at Site 747 (suggesting larger global ice volume). The early MCO is characterized by several intervals of peak warmth at the ANDRILL Site (PW-3 to PW-5) (Levy et al., 2016). Unfortunately, this interval is 305 not sufficiently covered by our Site 747 record to draw any conclusions; further proxy records are required to clarify the impact of Antarctic warming during the PW episodes on Southern Ocean bottom waters and at larger scales.  (Modestou et al., 2020) 747 (this study) 761 (Modestou et al., 2020) 747 (this study) 761  Age ( Levy et al., 2016). ∆47-based BWTs (Modestou et al., 2020; this study) and upper ocean temperatures (Leutert et al., 2020) are shown with 68 % confidence intervals. TEX86-based temperatures (Leutert et al., 2020) are based on the subsurface calibration of Ho and Laepple (2016). Site 761 benthic δ 18 O values are from Holbourn et al. (2004). Eccentricity 315 and obliquity are from Laskar et al. (2004).
Our observations suggest a decoupling between global ice volume and BWT in the Southern Ocean ( Fig. 5c-e).
Significantly, the stepped increase in benthic foraminiferal δ 18 O (and δ 18 Obw) around 14.0-13.7 Ma (~0.5-1.0 ‰, interpreted as cryosphere expansion) occurs clearly (roughly 0.5 Myr) later than the BWT drop. A similar decoupling of global ice 320 volume and BWT was discussed based on the Mg/Ca record of Site 806 , and suggested to be related to feedbacks and thresholds controlling Antarctic ice growth. It is possible that the bottom water temperature signal at Site 747 reflects changes in Southern Ocean hydrography, which might also have been transferred to the deep Pacific. Majewski and Bohaty (2010)  foraminiferal ∆47 data, we compare our BWT record to ∆47-and TEX86-based upper ocean temperature records from Site 1171 located on the South Tasman Rise at slightly lower latitudes (Fig. 5b and c;Leutert et al. (2020)). Acknowledging the caveat of substantial geographical distance between these sites, we find no evidence for a strengthening in the vertical 330 temperature gradient at high southern latitudes at that time. To the contrary, the bottom water warming starting during or slightly after the stepped increase in global ice volume diminishes the temperature gradient. Therefore, we relate the observed increase in the vertical δ 18 O gradient at Site 747 primarily to a freshening of the upper ocean water column relative to bottom waters associated with Antarctic ice sheet expansion, in line with the interpretation of Majewski and Bohaty (2010). An upper ocean freshening across the MMCT was also reconstructed at Site 1171 (Leutert et al., 2020). At high 335 southern latitudes, salinity has a large effect on stratification (e.g., Kuhnert et al., 2009). Similar to Leutert et al. (2020), we hypothesize that a Southern Ocean freshening concurrent with Antarctic ice sheet expansion may have decreased convective vertical mixing resulting in a shielding of upper ocean waters from comparably warm deeper waters. This stratification mechanism may have influenced Southern Ocean BWTs during the MMCT, explaining the transient bottom water warming and the partially opposing trends of upper ocean temperature and BWT. An increase in stratification starting between 340 13.5 Ma and 14 Ma is supported by an increase in dissolution at that time (Figs. 4 and S5), which may be related to reduced ventilation and an increase in CO2 storage in the deep ocean.

Conclusions
We constrain the middle Miocene BWT evolution at Site 747 in the Southern Ocean with clumped isotope thermometry.
Similar to existing BWT reconstructions from lower latitude sites, we find that Southern Ocean BWTs were substantially 345 warmer than today, despite the presence of ice sheets on Antarctica. The observed discrepancies between ∆47-and Mg/Cabased BWTs may be caused by changes in deep water carbonate ion saturation, but further Mg/Ca and ∆47 measurements are needed to conclusively test this hypothesis. We cannot fully rule out a dissolution effect on benthic foraminiferal ∆47, although there is currently no evidence for such an effect. Taken at face value, our ∆47 values indicate pronounced shifts in Southern Ocean BWTs, which resemble observations at equatorial Pacific Site 806, and a long-term decrease of ~3-5°C 350 across the MMCT. Comparison of changes in BWT and δ 18 Obw indicates a significant degree of decoupling and a more complicated sequence of events surrounding the MMCT than previously appreciated based on benthic δ 18 O alone. These findings suggest the involvement of additional feedbacks and thresholds in middle Miocene ice growth and/or regional effects on middle Miocene BWTs at Site 747. We hypothesize that a possible factor could be shifts in the vertical density structure of the Southern Ocean. Reconstructed BWTs may in part reflect changes in heat transport between upper and deep 355 ocean, induced by growing ice sheets on Antarctica. Independent higher-resolution BWT records from additional locations in and outside the Southern Ocean would allow for examining the spatial scale of the changes observed at Site 747 as a basis for better understanding the drivers of the MMCT.

Appendix A: Clumped isotope methodological details
Clumped isotope data are presented in the conventional ∆47 notation, which is defined as follows (e.g., Eiler, 2007;360 Huntington et al., 2009): R i are the measured abundance ratios of mass i relative to mass 44. R i* represent the stochastic abundance ratios calculated 365 from the bulk isotope composition of the sample (δ 18 O and δ 13 C).
All (clumped) isotope measurements were carried out in micro-volume mode. At the University of Bergen (UiB), we followed the long-integration dual-inlet (LIDI) protocol (Hu et al., 2014;Müller et al., 2017), whereas the measurements at ETH Zurich were performed via repeated cycles of alternating reference and sample gas measurements (Meckler et al., 2014;Rodríguez-Sanz et al., 2017). For data processing, we used the community software "Easotope" (John and Bowen, 2016). 370 2013; He et al., 2012;Meckler et al., 2014) and a conversion into the absolute reference frame (Dennis et al., 2011). For the conversion into the absolute reference frame, we utilized replicate measurements of three (UiB) respectively four (ETH Zurich) different correction standards from a window of ±12-40 standards around the sample replicate. At UiB, we used the carbonate standards ETH-1, ETH-3 and ETH-4 for correction from October 2016 to December 2016; ETH-2 was used for 375 monitoring during this interval. From August 2018 to June 2019, ETH-1, ETH-2 and ETH-3 were used for correction and ETH-4 for monitoring. For the measurements carried out at ETH Zurich, ETH-1, ETH-2, ETH-3 and ETH-4 were all included in the correction procedure. The accepted ETH standard values are from Bernasconi et al. (2018). These ETH standard values were determined using an acid fractionation correction of +0.062 ‰ (Defliese et al., 2015). Measured δ 18 O and δ 13 C values were drift-corrected based on three (UiB) respectively four (ETH Zurich) different correction standards 380 (with scale "stretching" only applied for δ 18 O at UiB and for both δ 18 O and δ 13 C at ETH). All isotope data were calculated with the Brand correction parameters (Daëron et al., 2016). Further details on analytical and data processing methods can be found elsewhere (Leutert et al., 2019;Piasecki et al., 2019).
We excluded three clumped isotope measurements as outliers, based on their offset of more than four standard deviations (4×0.037 ‰, estimated from the long-term mean reproducibility of all standards) from the mean. Raw standard and sample 385 measurement data are included in the supplement and will be available on EarthChem at the time of publication.

Data availability
The data from this paper are archived in the supplement. In addition, the final temperature data will be published at Pangaea (https://doi.pangaea.de/10.1594/PANGAEA.923258) and the full raw data on the EarthChem Database (a link to the EarthChem dataset will be provided prior to publication).

Competing interests 395
The authors declare that they have no conflict of interest.