Evidence for global cooling in the Late Cretaceous

The Late Cretaceous ‘greenhouse’ world witnessed a transition from one of the warmest climates of the past 140 million years to cooler conditions, yet still without significant continental ice. Low-latitude sea surface temperature (SST) records are a vital piece of evidence required to unravel the cause of Late Cretaceous cooling, but high-quality data remain illusive. Here, using an organic geochemical palaeothermometer (TEX86), we present a record of SSTs for the Campanian–Maastrichtian interval (~83–66 Ma) from hemipelagic sediments deposited on the western North Atlantic shelf. Our record reveals that the North Atlantic at 35 °N was relatively warm in the earliest Campanian, with maximum SSTs of ~35 °C, but experienced significant cooling (~7 °C) after this to <~28 °C during the Maastrichtian. The overall stratigraphic trend is remarkably similar to records of high-latitude SSTs and bottom-water temperatures, suggesting that the cooling pattern was global rather than regional and, therefore, driven predominantly by declining atmospheric pCO2 levels.

O ne of the warmest climates of the past 140 million years occurred in the early Late Cretaceous (late Cenomanian-early Turonian, between 95 and 90 Ma) [1][2][3][4] , with ice-free polar regions 5 , tropical sea surface temperatures (SSTs) greater than 35°C (ref. 2) and shallow latitudinal temperature gradients 6,7 . The interval following this (late Turonian through Maastrichtian, B90 to 66 Ma) is considered to have been a period of significant global cooling, possibly driven by a combination of declining pCO 2 levels and opening ocean gateways 1,4,5,8,9 . Although general trends in Late Cretaceous climate evolution are relatively well established 1,4,5 , these inferences are largely based on either bulk fine-fraction carbonate or benthic foraminiferal stable isotope data, representing mixed (fine fraction) or bottom-water temperature records. The rate and structure of Late Cretaceous SST cooling is poorly constrained, as most reconstructions are limited to short, fragmentary and low-stratigraphic-resolution planktonic foraminifera d 18 O records 6,7,[9][10][11] . Furthermore, the recognition of early diagenetic recrystallization of planktonic foraminifera at the sea floor, or shortly after burial 12 , has led to the rejection of many estimates of Cretaceous low-latitude SSTs that were anomalously cool compared with equivalent modern latitudes (the so-called cool tropics paradox). The shortcomings of carbonate-based palaeothermometry can, in some cases, be circumnavigated by using organic palaeotemperature proxies, such as TEX 86 (TetraEther indeX of tetraether consisting of 86 carbon atoms) 13 , which we apply here to a critical, but poorly quantified, interval of climate change-the Campanian-Maastrichtian.
The TEX 86 palaeothermometer provides estimates of mean annual SST 13 , independent of initial seawater chemistry and, compared with carbonate microfossils, is less subject to the modifying effects of diagenesis 13 . The TEX 86 proxy is based on the observed relationship between the ratio of different thaumarchaeotal membrane lipids (tetraether) and the mean annual temperature of the seawater in which the organisms lived [13][14][15][16] . However, TEX 86 is still a relatively new method and development of the proxy is ongoing. For example, recent studies have suggested that TEX 86 values may be influenced by archaeotal and thaumarchaeotal biology (taxonomy, diversity changes, physiology, life habitats) 15,16,17 and oceanographic setting (for example, water depth), which has led to discussions about how best to apply the temperature calibration in studies of past climate 13,14,18,19 .
Our samples come from the Shuqualak-Evans borehole in Mississippi (USA) that comprises a B240 m-thick sequence of shelfal, hemipelagic sediments. Deposition during the Cretaceous occurred at a palaeolatitude of B35°N 20 , on a broad shelf bordering the subtropical western North Atlantic to the east and the proto-Caribbean region to the south (Fig. 1 14,19,21 . TEX 86 H has been widely viewed as the most appropriate calibration for past greenhouse climates and is used here for discussion of our estimates of Cretaceous SSTs, as our measured values of TEX 86 are high and the study site is a low-latitude setting 14,21 . However, the estimates we provide should be considered to be maximum values, as a recent study 18 14 , and the magnitude of change in the Maastrichtian should be interpreted with caution.

Discussion
The maximum TEX 86 H -based SST estimates of 28-35°C from the Shuqualak-Evans borehole suggest that Late Cretaceous climate was consistently warmer at B35°N than at present (average SSTs at 35°N in the modern ocean are B20°C), and this conclusion is broadly true even using the TEX 86 L calibration ( Supplementary  Fig. 2). These estimates of low-latitude SST are consistent with recently published TEX 86 data from Israel 22 , which suggest a range of SSTs from B23 to 33°C (using TEX 86 H ) during the Campanian-Maastrichtian at a palaeolatitude of 5°to 15°N. Compared with the data from Shuqualak-Evans, the data from Israel display a greater range of SSTs and lower maximum values, which may be the result of deposition within an upwelling system. In contrast to these TEX 86 -based estimates of SSTs, a previous study based on d 18 O palaeothermometry of mixed-layer-dwelling planktonic foraminifera from Blake Nose (western North Atlantic B30°N palaeolatitude) suggested SSTs of 19-25°C 7 , which are equal to, or cooler than, present day SSTs at the equivalent latitude. It has been suggested that early diagenetic recrystallization of planktonic foraminifera could account for some of the lower temperatures reported from the Coniacian-Maastrichtian of Blake Nose 7 . Comparison with the TEX 86 data (using either TEX 86 H , TEX 86 L or BAYSPAR) from Shuqualak-Evans lends some support to this explanation for the mid to late Campanian (for example, R. calcarata zone). Furthermore, the highest SSTs we reconstruct (35°C using TEX 86 H ) are broadly consistent with (sub)tropical temperature estimates from other Mesozoic and Paleogene greenhouse intervals. For example, lowto mid-latitude mid-Cretaceous (Cenomanian-Turonian) and early Paleogene (Eocene) data from exceptionally well-preserved foraminiferal d 18 O and TEX 86 suggest that SSTs of 33°C and above were typical in these greenhouse climate regimes [23][24][25] .

29
GDGT index 2 (TEX 86 H ) 14 TEX 86 (log scale) R.c. Note that TEX 86 is plotted on a logarithmic scale. The Abathomphalus mayaroensis and Racemiguembilina fructicosa Planktonic Foram Zones cannot be assigned in the Shuqualak-Evans core: A. mayaroensis has not been recorded, probably due to environmental and/or palaeogeographical constraints and base R. fructicosa is recorded in the same horizon as base Pseudoguembelina hariaensis, likely due to the very low Maastrichtian sedimentation rate and not because of a hiatus, since all the nannofossil zones are present. C. plummerae, Contusotruncana plummerae; D. a., Dicarinella asymetrica; G. a., Globotruncana aegyptiaca; G. e., Globotruncanita elevate; G. g., Gansserina gansseri; G. havanensis, Globotruncanella havanensis;

G.havanensis
The addition of our new data to existing high-quality Late Cretaceous SST records from the equatorial 2,24 and North Atlantic 26 , South Atlantic 6,7 , Tanzania 12,25 and tropical Pacific 27 provides an overview of the spatial and temporal evolution of SSTs for this time interval (Fig. 3 and Supplementary Fig. 3). Although limited by the number of available records, this compilation suggests a very shallow latitudinal temperature gradient during the Turonian, which steepened in the Campanian through Maastrichtian. Similar SSTs at 0°and 35°N in the North Atlantic during the Turonian-earliest Campanian may be due, in part, to the relative isolation and restriction of the basin, which was only connected by shallow and/or restricted gateways to the South Atlantic, Pacific and Tethys oceans at that time. However, the similarity between North Atlantic SSTs and estimates from the southern Tethys 22 may suggest that North Atlantic temperatures were not remarkably different from other ocean basins at low latitudes, although, clearly, more good-quality data from the Tethys and Pacific are required to fully validate this hypothesis.
The similarity between Campanian-Maastrichtian low-(this study) and high-latitude 6,7 SST trends and the global benthic foraminiferal d 18 O record during the Campanian-Maastrichtian 4,5 indicates that the Campanian cooling, evident in all datasets, was not solely a high-latitude phenomenon, but represents a global event. This cooling coincided with, and may have been related to, reconfiguration of oceanic gateways 28,29 and hence deep, intermediate and shallow ocean circulation 4,29-31 . However, for significant deep-and surface-water cooling to occur across a wide range of latitudes, in both upwelling and nonupwelling settings, we suggest that declining atmospheric pCO 2 levels 32 , possibly due to decreasing ocean crust production 33 , were the ultimate driver of this long-term climate evolution. The changing tectonic configuration may have led to slight differences in the timing and pattern of change in different regions and water depths, a hypothesis that could be tested with improved agemodels for all critical sites and improved estimates of the timing of key tectonic events. The steepening of latitudinal temperature gradients during the Late Cretaceous is consistent with predictions from climate modelling 34 that suggest that the latitudinal gradient is strongly dependent on pCO 2 levels, with shallower gradients at higher pCO 2 . Short-term variability in Maastrichtian benthic foraminifera d 18 O has previously been interpreted as representing repeated reversals of deep-ocean circulation from low-to high-latitude sources 10 . However, the existence of broadly synchronous trends in our SST data, and temperature-indicative changes in calcareous nannofossil Our new data provide a critical addition to the understanding of climate evolution, from extreme warmth during the mid-Cretaceous to the termination of greenhouse conditions at the end of the Eocene. The data demonstrate that the transition from the so-called 'supergreenhouse' conditions of the mid-Cretaceous (Aptian-Turonian) to the cooler greenhouse world of the later Cretaceous and early Paleogene, occurred through gradual global cooling, rather than rapid, stepped changes, and that cooling was not confined to high latitudes. A similar transition has also been documented for the Eocene, before the switch to icehouse-mode climates in the Oligocene, albeit with much lower magnitude cooling at low latitudes 23,35,36 . The long-term cooling trend at high latitudes in the Eocene was likely caused by the opening of the Tasman Gateway during the early to middle Eocene, rather than a simple decrease in atmospheric pCO 2 alone, which would have led to more substantial cooling of (sub)equatorial surface waters than is observed 36 . It has been suggested that, during the late Eocene, declining pCO 2 levels eventually crossed a critical threshold (of B750 p.p.m.) at the Eocene/Oligocene (E/O) boundary, which allowed the rapid growth of the Antarctic ice sheet and a stepped climate change into an icehouse state [37][38][39] . Why, then, did significant continental ice-sheet growth occur at the E/O boundary, but not during the late Cretaceous, given the cooler Late Campanian and Maastrichtian water temperatures (surface and deep) and cooling at both high and subequatorial latitudes 5 ? Proxy reconstructions and models do not provide an unambiguous picture of the temporal evolution of Late Cretaceous-early Paleogene atmospheric pCO 2 levels and trends 40 , although, overall, the latest Cretaceous appears to have been characterized by lower pCO 2 levels than the mid-Cretaceous. It is therefore unclear whether the absence of continental scale glaciation was because pCO 2 did not fall sufficiently far, or because the pCO 2 threshold limit for ice-sheet growth was lower during the Late Cretaceous compared with the Eocene, due to different baseline conditions, such as ocean gateway configurations. Modelling of the E/O boundary event 37 suggests that open tectonic gateways around Antarctica are not necessarily required for initiation of continental scale glaciation, but they can exert a control on the amount of pCO 2 decline needed for glaciation. In the case of the E/O boundary, modelling suggests that, for major glaciation to occur with a closed Drake Passage, the pCO 2 threshold would be B140 p.p.m. lower than in a scenario with the Drake Passage open. During the Late Cretaceous, both the Drake Passage and the Tasman Gateway were closed and, thus, it is likely that the threshold for glaciation was oB600 p.p.m.V. Understanding why a major ice sheet was not initiated in the Late Cretaceous during an interval of marked global cooling, given some superficial similarities to Eocene climate trends and absolute values, is an intriguing challenge, which, if addressed through additional data and modelling, could provide valuable insights into the long-term controls on cryosphere development during greenhouse and 'doubthouse' conditions, climate sensitivity to changing pCO 2 , and the plausibility of glacioeustatic sea-level change during the Late Cretaceous and Early Paleogene.

Methods
Core location and palaeogeography. The Shuqualak-Evans core is from Shuqualak, Mississippi, USA (32°58 0 49 00 N, 88°34 0 8 00 W) and was sampled for TEX 86 from a depth of 9.45 m down to 251.46 m, spanning the Santonian/ Campanian boundary interval through to the uppermost Maastrichtian. Figure 1 shows a model of the palaeogeographic evolution of North America during the latest Cretaceous. In these reconstructions and others (for example, ref. 20) the Shuqualak-Evans borehole was situated on a broad shelf bordering the North Atlantic Ocean and Gulf of Mexico during the latest Cretaceous. Surface ocean circulation reconstructions (summarized in ref. 31) suggest that this location was likely not influenced by waters of the Western Interior Seaway and this is supported by calcareous nannofossil assemblage components and abundances, which suggest an open ocean water mass.
Biostratigraphy and age-model. The age-model for the Shuqualak-Evans core is based on integrated calcareous nannofossil 41 and planktonic foraminifera 42 biostratigraphic datums (Supplementary Fig. 1  GDGT extraction and analysis. Samples were solvent extracted using the technique previously published by Schouten et al. 13,43 Approximately 6 g powdered sample was ultrasonically extracted using one time methanol, three times dichloromethane (DCM)/methanol (1:1, v/v) and three times DCM. All extracts were combined and dried under a continuous N 2 flow at 40°C. Any water remaining in the samples was removed by passing the extracts (dissolved in DCM/methanol (3:1, v/v)) over a column containing anhydrous Na 2 SO 4 . Extracts were split into polar and apolar fractions by column chromatography, using hexane/DCM (9:1, v/v) and DCM/methanol (1:1, v/v) sequentially as the eluents and Al 2 O 3 as the stationary phase. The polar extract containing the targeted GDGTs was dissolved in hexane/propanol (99:1, v/v) and then filtered through a PTFE (polytetrafluoroethylene) 0.45 mm filter. After drying down, the samples were redissolved in a certain volume of hexane/propanol (99:1, v/v), which depends on the weight of each polar fraction. All 48 samples were analysed in the Department of Earth Sciences at UCL on an Agilent 1200 series HPLC attached to a G6130A single-quadrupole mass spectrometer. The analytical protocol followed is as described in Schouten et al. 43 . The abundance of both isoprenoid and branched GDGTs was measured in selective ion monitoring mode. Ion peaks of the respective GDGTs were integrated to determine the relative abundance of each molecule in the sample. These abundances were then used to determine the TEX 86, TEX 86 H (GDGT-index 2), and TEX 86 L (GDGT-index 1) indices 13,14 . These indices are defined as follows: where Cren' represents the crenarchaeol regioisomer.  Fig. 3. Nonetheless, in Supplementary Figs 2 and 3 we show both the TEX 86 H and TEX 86 L data (where possible) and the calculated SSTs to illustrate the potential range of values. We also present in Supplementary Fig. 2 SSTs calculated using the recently developed BAYSPAR approach 19 . The BAYSPAR model considers how the relationship between TEX 86 and temperature varies spatially and considers uncertainties in the modern SST-TEX 86 relationship. Critically, the stratigraphic trends for Shuqualak-Evans are near identical, irrespective of which proxy is used and, thus, our conclusions regarding the temporal evolution of the direction of Late Cretaceous climatic and latitudinal gradient change remains valid, even if absolute values are harder to constrain.
Recent work, based on an analysis of modern water-column GDGT abundance profiles, the core-top calibration dataset and a compiled Paleogene dataset, suggests that TEX 86 H may be less appropriate than TEX 86 L for use at sites where the water depth was approximately shallower than 1000 m (such as Shuqualak). This is likely due to variations in the export dynamics of individual GDGT compounds with depth, and an apparent temperature/water depth bias in the core-top calibration dataset 18 . In both the modern core-top calibration dataset and the Paleogene dataset, sites deposited in o1,000 m of water exhibit low GDGT-2/GDGT-3 ([2]/ [3]) ratios and high offsets between SSTs calculated by TEX 86 H and TEX 86 L (DH-L). We have been able to obtain the raw GDGT data for the sites on Demerara Rise, which were thought to have been deposited at water depths of o1500 m 44 . The GDGTs from these sites exhibit low [2]/ [3] ratios and high DH-L, which Taylor et al. 18 suggest is characteristic of water depths o1,000 m. We therefore contend that it is appropriate in Supplementary Fig. 3 to compare Demerara Rise and Shuqualak-Evans using either TEX 86 H or TEX 86 L (as the water depth of all sites was likely about, or shallower than, 1000 m). The application of the TEX 86 L calibration to our data from Shuqualak-Evans suggests SSTs of B28°C for the earliest Campanian and B20°C for the Campanian-Maastrichtian transition, which are some B7-8°C lower than the estimates of the TEX 86 H model, and much closer to modern SSTs at comparable latitudes, which perhaps is surprising given that Late Cretaceous pCO 2 levels are thought to have been higher than present (B600 to 800 p.p.m. versus 280 p.p.m. for the preindustrial modern) 40 . Furthermore, the use of TEX 86 L suggests almost no temperature gradient between 5°and 60°absolute palaeolatitude during the Turonian-Santonian, which also seems unlikely. The issue of how best to calculate temperatures from GDGT data is ongoing and it may be that a calibration based on suspended particulate organic matter may overcome some of the issues described above 16,18 .
Repeated analysis of an in-house standard and selected samples suggest that analytical reproducibility of TEX 86 is better than ± 0.009, in line with previous studies 45 that suggest an analytical error for TEX 86 index of ±0.01. In our data, we estimate that the error on SST estimates associated with analytical precision is o±0.4°C for the TEX 86 H calibration, and o±0.6°C using TEX 86 L . Analytical error is far less than the standard error associated with the core-top calibrations, which for TEX 86 H is ± 2.5°C, and for the TEX 86 L calibration is ± 4.0°C 21 .
BIT and MI indices. The GDGTs analysed for the TEX 86 palaeotemperature proxy are mainly produced by marine thaumarchaeota. However, the same GDGTs are also produced by terrestrial soil organisms and methanotrophs. GDGTs of terrestrial origin can be washed into the marine realm by rivers, potentially biasing SST reconstructions. Apart from isoprenoid GDGTs, which are used for the TEX 86 techniques, terrestrial organisms also produce branched GDGTs 46 . Branched GDGTs are typical of terrestrial organisms, but they do not occur among marine thaumarchaeota 46 . Thus, the branched GDGTs are used to quantify the terrestrial GDGT contamination in marine sediment samples by calculating the Branched and Isoprenoid Tetraether (BIT) index 46 . The BIT index is based on the ratio of branched GDGTs to the isoprenoid GDGT crenarchaeol 46 . In our study, the measurement of these branched GDGTs was included in the analytical protocol. There is no significant relationship between BIT and TEX 86 in our data (Supplementary Fig. 4) and the BIT index is between 0.05 and 0.15 ( Supplementary Fig. 2). This is well below the recommended threshold of 0.2, above which soil microbial contamination may be problematic 47 . Thus the SST estimates made in this study are probably not altered by an influence of terrestrial GDGTs.
GDGTs produced by methanotrophs within the sediments can distort the TEX 86 signal and lead to erroneous estimates of palaeotemperature. The degree of influence of methanotropic archaea can be estimated using the Methane Index (MI) 17 . Normal marine sediments have values o0.3, whereas sediments influenced by high rates of methane production have values 40.5. The interval from 0.3 to 0.5 marks the transition between the two environments 17 . The MI values from Shuqualak vary from B0.1 to 0.25 ( Supplementary Fig. 2), suggesting normal marine conditions and a lack of GDGT production by methanotrophs. We therefore conclude that methanotrophic production of GDGTs has not impacted adversely on our TEX 86 records.
Palaeotemperatures from foraminiferal oxygen isotopes. In Fig. 3 and Supplementary Fig. 3, we show estimates of palaeotemperature based upon the oxygen-isotopic (d 18 O) composition of a global compilation of benthic foraminiferal data 4 , selected low-latitude planktonic foraminiferal datasets 12,25 and mixed-layer-dwelling planktonic foraminifera from southern high-latitude sites (DSDP Site 511 and ODP Site 690) 6,7 . Calcareous dinoflagellate cysts from ODP Site 690 have similar d18O values to planktic foraminifera 48 but are not included as a suitable temperature calibration for Cretaceous dinoflagellates is not available. Note for the Turonian-age samples from Tanzania, we have recalculated the SST range using the typical range of planktonic d 18 O values measured ( À 4 to À 5 %) 25 . In order to apply a consistent approach to calculating temperatures from d 18 O of foraminiferal calcite, we have recalculated temperatures from the original published oxygen-isotopic datasets. We used equation (6)  We used a d w value of À 1.27, which includes a correction of À 0.27 for the conversion of Vienna Standard Mean Ocean Water to Vienna PeeDee Belemnite and an ice volume value of À 1%. For calculation of SSTs from planktonic foraminifera, we applied an additional latitudinal salinity correction 50 to d w , using the palaeolatitudes of DSDP Site 511 and ODP Site 690 in the Late Cretaceous (58°S and 67°S, respectively) and equation (7): Surface water d w correction ¼ 0:576 þ 0:041x ð ÞÀ 0:0017x 2 À Á þ 1:35Ã10 À 5 x 3 À Á where x ¼ absolute palaeolatitude between 0°and 70°. Palaeolatitudinal positions for each site were using the ODSN Plate reconstruction software (http://www.odsn.de/odsn/services/paleomap/paleomap.html).