Introduction

Oxygen isotope data from benthic foraminifera (δ18Ob) record changes in Cenozoic climate. The overall pattern of change suggested by pioneering δ18Ob studies1 and subsequent composites of higher resolution work2,3,4,5 reveals pronounced global warmth during the early Eocene followed by long-term cooling and growth of polar ice sheets. Ice sheet growth was likely conditioned by slowly declining atmospheric carbon dioxide (CO2) levels6,7 (Fig. 1a) and geologically sudden threshold-breaching jumps in response8,9,10. The Cenozoic increase in Earth’s land ice budget was strongly diachronous between hemispheres. Large ice sheets were first sustained on Antarctica around 34 million years ago (Ma)11,12,13 (Fig. 1b). The transition to bipolar glacial conditions occurred much later13. A Greenland Ice Sheet may have developed by 11−7 Ma (refs. 14,15), even though isolated glaciers may have existed during the middle Eocene16, but it was not until about 2.6 Ma that extensive ice sheets waxed and waned over North America and Eurasia17 (Fig. 1b). Thus, for about 30 million years, nearly half of Cenozoic history, Earth’s ice age history was primarily written by the growth and decay of Antarctic ice sheets. This unipolar icehouse climate state (Fig. 1b) presents an opportunity to examine a future-relevant high CO2 climate system with Antarctic-only ice sheet variability, but it is understudied compared to the bipolar icehouse climate state, especially the great ice ages of the late Pleistocene.

Fig. 1: Contrasting indications of the astronomical pacemaker of Antarctic glacial-interglacial cycles during the high CO2 mid-Cenozoic unipolar icehouse.
figure 1

a atmospheric carbon dioxide (pCO2) reconstructions with 2σ uncertainties7, from phytoplankton (cyan triangles) and boron isotopes (blue circles). b marine benthic foraminiferal oxygen isotope (δ18Ob) composite record CENOGRID5 with 10-point (black) LOESS smoothing, relevant climate events and glacial states shown. c contrasting astronomical pacing of two marine δ18Ob records (ODP sites 926 and 1264, refs. 19,29) 26−18 Ma. d power spectra for the data shown in c. Vertical black lines denote bandwidths used throughout this paper. e Antarctic (ice-proximal) records AND-2A, CRP-2/2A, and DSDP Site 270 with colours representing simplified lithology from refs. 18,22,23: light grey mudstone (often including ice-rafted debris), dark grey sandstone, green diamictite, brown sandstone, and sinusoidal lines disconformities. The question mark in DSDP Site 270 denotes uncertain lithological age. Site locations shown in Figure S1.

Full size image

A mechanistic understanding of Antarctic ice sheet (in)stability must incorporate astronomical forcing. Palaeoclimate records from locations both proximal and distal to Antarctica show unequivocal imprints of astronomical forcing18,19,20,21. Ice-proximal records reveal direct evidence of glacial advances and retreats over the drill sites that are linked to the Antarctic hinterland, yet the same glacial advances cause hiatuses. Distal δ18Ob records facilitate long-term continuous evolutions of astronomical pacing of global ice volume and temperature yet could be affected by preservation issues or low sedimentation rates, smoothing out higher-frequency signals. The lack of a coherent picture from Antarctic ice-distal and ice-proximal records—information that should be complementary—is particularly acute for the late Oligocene-to-early Miocene interval (herein referred to as the Oligo-Miocene, ~28−20 Ma). Glacio-marine sedimentary sequences recovered from the Ross Sea (Fig. S1) show unequivocal evidence of multiple episodes of ice sheet advance and retreat during the Oligo-Miocene (Fig. 1)18,20,22,23 that were astronomically, mainly obliquity paced18,24. A recent review of ice-proximal records from multiple Antarctic locations25 highlights a wealth of past climate data (e.g., refs. 26,27) that support dynamic oceanic environments related to ice age variability. During short intervals of the Oligo-Miocene, both obliquity (41 thousand years [kyr]) and short eccentricity (110-kyr) cycles have been observed at different locations (e.g., refs. 18,24,28) and may thus ultimately drive ice sheet variability in the hinterland of those localities24,28.

Records from deep-sea sediments at ice sheet-distal sites complement ice-proximal records by providing the astronomical pacemaker of the globally integrated cryosphere system, rather than individual ice margins or sheets, over prolonged periods of time. Deep-sea sediment records are typically continuous over long periods of time and can therefore pick up differences in ice sheet variability across changes in astronomical configuration. Yet, the astronomical imprint on the pre-Quaternary deep-sea δ18Ob record is far from clear cut. Published records for the Oligo-Miocene are especially puzzling because they show pronounced spatial and temporal variability both in amplitude and response to astronomical forcing19,21,29,30,31 (as synthesized in Fig. 1). This arrhythmia in the δ18Ob record (Fig. 1c, d), unless explained by markedly different signal fidelity between sites, presents a major challenge to interpreting palaeoclimate because it contradicts the fundamental principles of oxygen isotope stratigraphy: on timescales longer than the mixing time of the ocean (~1.5 kyr), the ice-volume signal encoded in δ18Ob must be common to all records of sufficient fidelity32. Here, we present a δ18Ob record of exceptional fidelity from the North Atlantic Ocean (Supplementary Data 1) and use it as a benchmark to address the arrhythmia identified in the published δ18Ob records and to evaluate the astronomical pacing of the early Antarctic ice ages in synthesis with Antarctic proximal records.

Results and Discussion

A δ18Ob record from Site U1406

At Site U1406, clay-rich nannofossil oozes were recovered from a sediment drift at 3.8-km water depth on the Newfoundland margin in the Northwest Atlantic Ocean hosting well-preserved calcareous microfossils33 (Figure S1). Before now, δ18Ob data for most of the Cenozoic from this climate-sensitive region (the North Atlantic Ocean) were sparse because of large gaps in the geological record attributable to discontinuous coring, hiatuses, and condensation horizons34,35. Our δ18Ob record spans approximately 26.4 to 21.8 Ma across the Oligocene-Miocene Transition, a critical interval to the debate about the sensitivity of the early Antarctic ice sheet to astronomical forcing and the mechanisms controlling the evolution of those ice ages18,19,20,31,36. With a temporal resolution of generally 1–3 kyr, our record is the best-resolved δ18Ob time series available from anywhere in the world for the Oligo-Miocene interval. Our δ18Ob record was constructed using exceptionally well-preserved benthic foraminifer Cibicidoides mundulus (Methods) from a sequence characterised by rates of sedimentation between 1 and 3 cm kyr−1 and placed on an astronomically tuned age model that is independent of our δ18Ob data (Supplementary Data 2, SI Astronomical Tuning, Data Quality Control).

We recognise three main intervals with distinct climatic imprints in our δ18Ob record from Site U1406 (Fig. 2a). Our record begins close to the transition between the ‘middle’ Oligocene glacial interval (MOGI, ~28−26.3 Ma), when high δ18Ob values indicate a cold deep glacial climate state with high-amplitude glacial-interglacial cycles19. There follows an interval of inferred long-term deglaciation through the late Oligocene (‘Late Oligocene Warming’ LOW; ~25.7−23.7 Ma) related to warming and/or continental subsidence in West Antarctica19,37,38 within which three abrupt, high amplitude (up to ~1 ‰) decreases in δ18Ob are revealed at ~25.5, ~25.1, and ~24.4 Ma (Fig. 2a, arrows). An interval of high-amplitude variability in δ18Ob with a pronounced overall transient increase delineates the rapid major changes in glacial state of Antarctica during the Oligocene-Miocene Transition (OMT; ~23.7−22.6 Ma). High-amplitude variability, including a fourth prominent abrupt decrease in δ18Ob at ~22.1 Ma, is also characteristic of the early Miocene interval (EMI, ~22.6−21.8 Ma). Three of these four events initiated when δ18Ob was close to peak values for the data set (≥2.1‰ ± 0.17 ‰, Fig. 2). Two of those documented at Site U1406 (at ~25.5 and ~22.1 Ma) occur at the start of maxima in ~405-kyr Eccentricity Cycles 63 and 55 and are also present (although less well resolved) in the δ18Ob records of sites 926 and 1264 (Fig. 2b, c)19,29. The δ18Ob decreases identified at ~25.1 and ~24.4 Ma are not distinct at sites 926 and 1264 (Fig. 2b, c)19,29.

Fig. 2: Our δ18Ob record from IODP Site U1406 compared to the records from sites 926 and 1264 and the astronomical series.
figure 2

a δ18Ob from Site U1406 (bright red, this study). EMI (Early Miocene Interval), OMT (Oligocene-Miocene Transition), LOW (Late Oligocene Warming), and MOGI (Mid-Oligocene Glacial Interval). Arrows indicate abrupt decreases in δ18Ob at Site U1406. b and c δ18Ob at sites 926 (dark red)29 and 1264 (yellow)19, respectively. d and e La2004 astronomical solutions41 for obliquity and eccentricity, respectively, with obliquity instantaneous amplitude and long-term eccentricity modulations of ~405 kyr and ~2.4 Myr. Site locations shown in Fig. S1.

Full size image

A strong globally consistent response of δ18Ob to eccentricity

To assess the astronomical forcing of Oligo-Miocene climate, we calculated the obliquity variance (VObl) and eccentricity variance (VEcc) (Methods) for selected δ18Ob records and our δ18Ob record from the North Atlantic (SI Data Quality Control). Both VObl and VEcc show pronounced changes over the Oligo-Miocene study interval (Fig. 3). We highlight two main observations on the temporal and spatial variability in VEcc. First, VEcc almost always exceeds VObl in all records (Fig. S2b). The same result is also clearly seen in the Middle-to-Late Pleistocene of the δ18Ob record LR0439 (Fig. S3d, e) when ~100-kyr pacing dominated glacial-interglacial cycles40. Second, throughout our Oligo-Miocene study interval, we find that VEcc is remarkably similar among all records and strongly resembles the long-term (~2.4-million years [Myr]) modulation of the La2004 eccentricity solution (Fig. 3b)41. The analysed time interval spans most of ~2.4-Myr Eccentricity Cycle 12 through to the first half of Eccentricity Cycle 9 (Fig. 3c). The highest consistently recorded VEcc values of ~0.010−0.014 ‰2, indicative of high-amplitude ~110-kyr eccentricity cycles in δ18Ob, occur during ~2.4-Myr Eccentricity Cycles 12 and 10 in the La2004 astronomical series41. During ~2.4-Myr Eccentricity Cycle 11, concurrent with the LOW interval, the response of δ18Ob to eccentricity is modest with VEcc values ranging between ~0.004 and 0.008 ‰2 (Fig. 3b). The higher VEcc values during ~2.4-Myr Eccentricity Cycles 12 and 10 (Fig. 3b) are associated with generally higher absolute δ18Ob values (colder/more ice) during the MOGI, OMT and EMI (Fig. 3a). Thus, our analysis reveals a relationship wherein cooler more deeply glaciated Oligo-Miocene climate states are coupled with a stronger imprint of Earth’s eccentricity than warmer less glaciated climate states. This result is consistent with the one established more recently in the Pleistocene39 when ice sheets in the Northern Hemisphere dominated glacial-interglacial climate variability (Fig. S3).

Fig. 3: A globally coherent response to eccentricity and inconsistent response to obliquity in the Oligo-Miocene δ18Ob record.
figure 3

a δ18Ob of ODP/IODP sites U1406 (bright red; presented here), 926 (dark red)29, 1090 (purple)30, 1218 (blue)21, and 1264 (yellow)19 during EMI (Early Miocene Interval), OMT (Oligocene-Miocene Transition), LOW (Late Oligocene Warming), and MOGI (Mid-Oligocene Glacial Interval). Blue shading indicates colder intervals (roughly equivalent to the MOGI/LOW transition and OMT interval) with high-amplitude variability in δ18Ob on frequencies corresponding to eccentricity. b eccentricity variance (VEcc) of the δ18Ob records shown in a), the composite δ18Ob record20,47 and the La2004 eccentricity solution41 shown in (c) with long-term eccentricity modulations of ~405 kyr and ~2.4 Myr offset for clarity. Grey line denotes maximum VEcc values in the 5.0−1.2 Ma interval of the LR04 δ18Ob stack39 (see Fig. S3). d obliquity variance (VObl) of the δ18Ob records shown in a), the composite δ18Ob record20,47, and La2004 obliquity solution (La2004, dashed)41 shown in (e) with amplitude modulation. Grey line denotes maximum VObl values in the 5.0−3.1 Ma interval of the LR04 δ18Ob stack39 (see Fig. S3). Note that the y-axes of VObl and VEcc are scaled differently to allow site-to-site comparisons (see Fig. S2b for plot with identical y-axes). Site locations shown in Fig. S1.

Full size image

Site-to-site variability in the response of δ18Ob to obliquity

In contrast to the globally consistent picture seen in VEcc, we see striking site-to-site variability in VObl for most of the studied Oligo-Miocene time interval (Fig. 3d). Before we can consider environmental explanations for this observation, we must first assess processes that may bias δ18Ob records and artificially introduce site-to-site variability, especially in higher frequency bands (here VObl). These include low sedimentation rates, under-sampling, the effects of stratigraphic discontinuities, age models that inadvertently tune power into or out of the obliquity bandwidth, and foraminiferal species and preservation effects on δ18Ob (summarized below, Supplementary Data 3, and SI Data quality control). The record from Site 1090 is from a mix of genera and species30 and it is not sampled at high enough resolution to assess VObl between 22.5−20.5 Ma (Fig. 3b, d, S2f). The record from Site 1218 is also multi-species and is too sparsely resolved to resolve obliquity-paced variability from ~23.2 Ma and younger. At Site 926, the δ18Ob record was produced using a mixture of Cibicidoides species in the interval between ~26.5−25 Ma (ref. 29) but uses only Cibicidoides mundulus between 25.0−17.9 Ma and is a particularly useful comparison to our record because of its temporal resolution, site location and water depth (see below). The record from Site 1264 is derived from a single species (Cibicidoides mundulus) and sampled at high resolution, but, because sedimentation rates are low (often ≤ 1cm kyr−1, Fig. 3b, d, S2g), it is likely that the obliquity-paced signal is damped or even removed by sediment mixing (bioturbation).

Our record from Site U1406 provides a benchmark against which to assess Oligo-Miocene variability in VObl because (i) it benefits from a strong astrochronology independent of δ18Ob (Supplementary Information), (ii) was developed using unusually well-preserved, mono-specific samples of Cibicidoides mundulus, and (iii) comes from a site with relatively high sedimentation rates (~1‒3 cm kyr−1) (Methods) that is sampled at high resolution (generally 1−3 kyr sample spacing) to resolve obliquity cycles throughout. We draw three major findings by comparing our record to published records from other sites. First, records from all sites, including U1406, show a long-term minimum in VObl within the LOW ~25.4−24.4 Ma (Fig. 3d) that indicates little response to modest obliquity forcing (Fig. 3e). Second, the striking site-to-site variability in VObl that characterizes the older and younger parts of our Oligo-Miocene record (~25.8–25.4 and ~24.4−22.6 Ma) becomes especially pronounced when a strong response to modulated obliquity forcing emerges at some sites, but not others, during minima in VEcc and in the ~2.4 Myr amplitude modulation of eccentricity (Fig. 3b, c). One example occurs close to the MOGI-LOW transition (25.8–25.4 Ma; Fig. 3d) when baseline δ18Ob values are high and the global imprint of ice volume changes on δ18Ob was large19,42. We attribute the higher VObl seen in our North Atlantic record from Site U1406 to a combination of higher sedimentation rates, higher temporal resolution compared to the other records available (Fig. S2f), and the use of a single species (whereas a mix of species was used at sites 926 and 1218) (Supplementary Data 3). A second example of pronounced site-to-site variability in VObl occurs between the culmination of LOW and the OMT (~24.4−22.6 Ma), when δ18Ob values and the imprint of ice volume changes on δ18Ob ranged between the lowest and highest values in the whole of our record (Fig. 3a, d)19. Here, sedimentation rates and temporal resolution among the records (Fig. S2f, g) are more similar across sites, but VObl in the North Atlantic Site U1406 is two to four times lower than in the equatorial Atlantic (Site 926) and sub-polar South Atlantic (Site 1090) despite obliquity tunings underpinning the age models at all three sites (Fig. 3d).

The offset in VObl between U1406 and 926 between ~24.4−22.6 Ma is especially significant because the record from Site 926 was also produced using monospecific foraminifera in this interval. Therefore, the inconsistency between these two records in this interval cannot be readily explained by taxonomic biases and instead suggests an environmental control. The most likely explanation for the observed contrasting VObl is lower amplitude changes in bottom water temperature (BWT) at the mid-northern-latitude Site U1406 than at the equatorial Site 926. Consequentially, the influence of the Deep Western Boundary Current (DWBC) on the mid-depth Oligo-Miocene western North Atlantic Ocean may have been substantially weaker than today, because sites U1406 and 926 lie in very similar water depths (~3.8 km and 3.6 km, respectively) and are both bathed today by the same water mass (DWBC). Southern Component Waters (SCW) may have influenced changes in BWT at Site 926 more than at Site U1406, possibly only during obliquity minima as demonstrated by increased corrosiveness31,36,43. Before ~24.4 Ma, Antarctic margin cryosphere processes influenced BWT more similarly at both sites. The reduced influence of SCW at Site U1406 from 24.4 Ma onwards is part of a reorganization in ocean circulation, as shown by deep-water flow speed increases in the Southern Ocean in the latest Oligocene44,45,46. Such change in ocean circulation has also been hypothesized to be one of the causes of the major Antarctic glaciation at the Oligocene-Miocene Transition44, in addition to a favorable astronomical configuration29,31,41 and a decreasing pCO2 concentration during the late Oligocene7.

The culmination of this major Antarctic glaciation at the Oligocene-Miocene Transition was proceeded by high-latitude climate cooling, a change in tectonic setting 19,37,38, and an expansion of the Antarctic ice sheet into marine settings18,20,25,38. The margins of these ice sheets were, at times, astronomically paced18. To establish the sensitivity of these margin to orbital forcing, Levy et al. (ref. 20) calculated obliquity sensitivity (SObl), a ratio between VObl in δ18Ob (climate response in the obliquity bandwidth) and VObl in the astronomical solution (climate forcing by obliquity), using a composite δ18Ob record47 for the interval between ~34 and 5 Ma. This revealed an abrupt increase in SObl at ~24.5 Ma (Fig. 3d), which is only apparent in some of the δ18Ob records (Fig. S2c), but not in ours. We attribute the SObl jump20 visible in the δ18Ob megasplice at ~24.5 Ma to the tie point between the two spliced records (1218 and 1090)21,30 with contrasting SObl (Fig. 3, S2c, S4). Our analysis of all δ18Ob records shows markedly different obliquity-paced changes among sites. Therefore, we suggest that the mechanism(s) that drove high-latitude ice volume and BWT variability in the late Oligocene were less sensitive to changes in obliquity than indicated by a subset of δ18Ob records and some ice-proximal sedimentary records.

Eccentricity pacing of the Antarctic Ice Sheet

The site-to-site congruence in the deep-sea δ18Ob record in the eccentricity bandwidth is unmistakable within the Oligo-Miocene interval (Fig. 3b) and strongly suggests a common cause that is imprinted similarly in all records. We attribute this result to the waxing and waning of ice sheets on Antarctica because the ice volume component in δ18Ob is globally uniform on astronomical time scales and because ice sheets did not develop extensively in the Northern Hemisphere until the latest Pliocene17. Eccentricity pacing of the Oligo-Miocene Antarctic ice sheets is consistent with a record of seawater δ18O (δ18Osw) change for the OMT48, which suggests a transient sea-level fall of up to ~50 meters sea-level equivalent (msle) at the Oligocene-Miocene Transition. This is similar to the up to 30−40-msle estimates from inverse modelling of 1-D ice sheets49 and backstripping estimates50. Our analysis reveals major glacial-interglacial cycles in Antarctic climate and ice sheet size during the late MOGI into early LOW and from peak OMT into EMI (Fig. 2), which were predominantly paced by eccentricity between 26 and 20 Ma. Obliquity pacing of larger volumes of Antarctic ice may have occurred when Earth’s orbit was circular (i.e., no eccentricity)42 and occurred regularly much later, e.g., around the Miocene Climate Optimum (18−13 Ma)20,51. Throughout this time, ice-proximal evidence shows that eccentricity-modulated precession also still paced volumetric changes in land-based ice52,53.

The amplitude of the eccentricity-paced variations in δ18Ob suggest waxing and waning of a large proportion of the Antarctic ice sheet. We note that absolute δ18Ob values and ice-proximal sedimentological evidence22,25 indicate sufficiently large ice volumes during the OMT climate event that the Antarctic ice sheet transiently expanded over the continental shelf into the marine realm. However, the high-amplitude δ18Ob variations together with palaeotopographic and erosion rate reconstructions for the entirety of Antarctica in the Oligo-Miocene54,55 (see below) emphasise regular waxing and waning of the large land-based ice sheets on Antarctica at other times (not only the OMT climate event). Moreover, strong eccentricity pacing of Oligo-Miocene ice ages, together with strengthened evidence for abrupt decreases in δ18Ob registering glacial termination events (note arrows in Fig. 2a), suggests ice sheet hysteresis behaviour that was much weaker than currently modelled for Antarctica56,57,58,59 but strong enough to allow ~110-kyr (not ~20-kyr or 41-kyr) ice age cycles. Further analysis of this problem is merited to understand why the largely land-based early Antarctic ice sheet, despite its pole-centred configuration, shows cyclical behaviour akin to the late Pleistocene Laurentide Ice Sheet (LIS). As for the LIS, the early Antarctic ice sheet clearly underwent sustained growth over successive insolation cycles (Fig. 3). Presumably, the abrupt Antarctic ice age termination events (Fig. 2) were driven by an astronomically paced increase in radiative forcing and lagged glacial bedrock isostatic rebound, but, while the LIS contributed to its own demise by advancing deep into the mid-latitudes60,61,62, the same process is not possible on Antarctica.

The Middle to Late Miocene transition to an obliquity-paced Antarctic ice sheet, perhaps more vulnerable to ocean melting than before20, was likely partly caused by the ice sheet itself as it advanced and eroded the Antarctic continental shelf63. Antarctic palaeotopographic maps and offshore sediment accumulation rates54,55 reveal where the Antarctic ice sheet was most active in the mid to late Cenozoic. Estimated offshore long-term erosion rates point to a dynamic East Antarctic Ice Sheet (EAIS) during the Oligo-Miocene, especially in the low-lying Recovery, Aurora, and Wilkes subglacial basins54,55. In contrast, West Antarctic erosion rates appear to have been relatively subdued during the Oligo-Miocene compared to the last ~14 Myr54. Smaller ice sheets with marine margins in West Antarctica and on the Antarctic Peninsula should not be discounted22,64 and would have contributed to transient changes in sea-level. Large parts of West Antarctica, however, lay above sea level during the late Oligocene and early Miocene54 providing nucleation points for ice caps to form65. We suggest, therefore, that the congruent behaviour of VEcc seen in globally distributed δ18Ob records during the Oligo-Miocene was mainly driven by waxing and waning of land-based Antarctic ice sheets and included East Antarctica due to the amplitude of the δ18Ob changes. This suggestion is consistent with palaeotopographic maps and offshore sediment accumulation rates54,55.

The direct effect exerted by Earth’s eccentricity on insolation is too weak to drive major changes in climate directly. As a result, the strong response of the Antarctic climate-cryosphere system on eccentricity time scales must originate from a nonlinear response to eccentricity-modulated precession forcing41. Results of a detailed trace element study48 of benthic foraminifera from Site 926 across the OMT interval suggest an amplifying mechanism rooted in the carbon cycle and reconstructions of the sensitivity of the Antarctic ice sheet for that interval66 are consistent with those of coupled climate-ice sheet model experiments57. One mechanism that may explain amplification of eccentricity-paced climate forcing is the destabilization of methane gas hydrates after a sufficient drop in sea level like at the OMT climate event67. This negative feedback mechanism could drive a relatively quick glacial termination which is visible in these global δ18Ob records (e.g., at ~22.1 Ma, Fig. 2). Another mechanism may occur at low latitudes and involves the monsoon-driven modulation of global marine organic carbon burial rates, resulting in eccentricity-paced oscillations in atmospheric CO268. Further work is needed to explore the influence of low-latitude processes on astronomically paced changes in polar air temperature during the Cenozoic and to improve our understanding of hysteresis behaviour of the early Antarctic ice sheets.

Methods

Isotope geochemistry

Discrete samples of approximately 20 cm3 were taken at a sample spacing of 2−4 cm between 54.76 and 155.08 m CCSF-M using the revised splice69. Half of these samples (n = 1446; 54.76−107.97 and 146.50−155.08 m CCSF-M) were processed at the University of Southampton's Waterfront Campus, National Oceanography Centre Southampton (UoS-NOCS), and the other half were processed at the University of Leipzig (UoL, n = 467; 108.00−126.85 m CCSF-M), University of Frankfurt (UoF, n = 396; 126.89−145.65 m CCSF-M), and University of Heidelberg (UoH, n = 353; 116.08−145.90 m CCSF-M) using standard methods. Samples were oven-dried (40−50°C), washed over a 63-μm sieve, and oven-dried again. Weights were recorded between every processing step, and the wt.% coarse fraction (>63 μm) was determined for every sample (Supplementary Data 1). To obtain a robust isotopic value for each sample level and minimize isotopic variability attributable to ontogenetic effects, three to eight of the best preserved Cibicidoides mundulus specimens were selected from the 125–250 μm sieve fraction and transferred into reaction vials. We used C. mundulus because the isotopic offset from seawater equilibrium values is well documented for this epifaunal species70,71,72. Where insufficient C. mundulus were present in the 125–250 μm fraction, additional individuals were picked from the 250–355 μm size fraction. Stable oxygen isotope data (Supplementary Data 1) were generated using a Thermo Fisher Scientific MAT 253 mass-spectrometer coupled to a Thermo-Finnigan Kiel IV Carbonate Device in Southampton, Heidelberg, and Leipzig, and coupled to a Gasbench II in Frankfurt. International standards (NBS-18 and NBS-19) and in-house quality control standards were used to calibrate δ18Ob across laboratories, yielding a reproducibility for δ18Ob in the range of 0.05−0.08‰ (±1σ) over the 30-month measurement period.

Site selection and statistical data processing

Published qualitative observations suggested that obliquity pacing was strongly expressed in δ18Ob records from the western equatorial Atlantic Ocean at Ocean Drilling Program (ODP) Site 926 (Fig. 1c, d)29,31 and the eastern sub-Antarctic Atlantic at ODP Site 109030. In contrast, short (~110 kyr) eccentricity was identified as the primary pacemaker of the δ18Ob record from the eastern subtropical South Atlantic Ocean (Site 1264), where no strong obliquity signals were detected (Fig. 1c, d)19. These three sites (ODP sites 926, 1090, and 1264)19,29,30 fall along a latitudinal transect from the equatorial to South Atlantic Ocean. A fourth δ18Ob record from the equatorial Pacific Ocean (Site 1218)21 combined a varying mix of both eccentricity and obliquity signals.

To quantify the obliquity and eccentricity pacing, we calculated the obliquity variance (VObl) and eccentricity variance (VEcc) of all records. First, we removed substantial visual outliers from all δ18Ob records (i.e., from sites 926, 1090, 1218, 1264, and U1406). Next, records were linearly interpolated to 4-kyr, approximately equivalent to the coarsest continuous sampling resolution available, which is more than sufficient to statistically capture the obliquity cycle73. Subsequently, the δ18Ob records were detrended using a high-pass filter preserving frequencies >1 cycle Myr−1. We used Astrochron20,74 to compute VObl and VEcc as well as obliquity sensitivity (SObl) over frequencies of 25 ± 2 cycles Myr−1 and 9 ± 3 cycles Myr−1 that correspond to the 40-kyr obliquity cycle and the ~110-kyr eccentricity cycle, respectively, across 1-Myr-long sliding windows. We also computed VObl and VEcc of a composite δ18Ob record47 comprising data spliced together from ODP sites 926, 1090, and 1218, from which obliquity sensitivity was also computed by ref. 20.

Carbonate content calibration

Bulk carbonate content strongly covaries with the natural logarithm of calcium over potassium counts, i.e., ln(Ca/K), calculated from XRF core scanning data69 (Fig. S5a). ln(Ca/K) was calibrated to coulometric CaCO3 concentration using an exponential fit (Fig. S5a). We used both shipboard33 and shore-based coulometry data (wt.% CaCO3) generated at UoS-NOCS, using a CM5015 coulometer equipped with an AutoMate automated analysis device (Supplementary Data 4). Three samples were selected to evaluate measurement precision: These samples were all run three times, yielding mean wt.% CaCO3 contents of 34.9 wt.%, 11.3 wt.%, and 17.8 wt.% with a precision at 1σ of 0.35 wt.%, 0.25 wt.%, and 0.47 wt.%, respectively.

Phase calculations and wavelet analysis

We obtained coherency and phase estimates between CaCO3 content and δ18Ob (Fig. S5b) on the initial magnetostratigraphic age model75 using the SPECTRAN software76 and a Parzen window across 33% of the data series. Phase and coherency estimates were calculated in the obliquity bandwidth of 23 ± 2 cycles Myr−1. The computational settings for phase and coherency estimates included 2σ error bars and yielded coherency confidence levels of 80% and 95%.

Morlet wavelet analysis77,78 were used for time-frequency analysis in both depth and age domain, after linear interpolation to 2 cm (XRF-based CaCO3 in depth) and 4 kyr (XRF-based CaCO3 and δ18Ob in age) and detrending using a high-pass filter preserving frequencies >0.04 cycle m−1 (preserving cycles with a period <25 m) or >1 cycle Myr−1.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.