Main

Arctic sea ice plays an important role in the climate system. It insulates the ocean from heat loss and largely determines the Arctic surface albedo1. Modern sea-ice cover is reducing rapidly, and most simulations predict at least one ice-free month in summer before the year 20502. A changing sea-ice cover can trigger positive feedback processes that influence regional and global climate, which need to be well understood. Identifying the last time Arctic sea ice disappeared is therefore important for anticipating future changes as well as providing targets for testing climate model performance3. The Last Interglacial (LIG; the Eemian, Marine Isotope Stage 5e (MIS5e) ~129–115 thousand years before present (ka bp)) is a particularly well-suited period to study ongoing Arctic sea-ice change as terrestrial circum-Arctic summer temperatures were ~4–5 °C degrees warmer than today4, and boreal forests extended to the Arctic Ocean coastline5. However, the extent of Arctic sea ice during this time is debated and hinders a comprehensive understanding of the LIG climate. So far, climate models and proxy records have mostly suggested a relatively extensive perennial sea-ice cover6,7, although some studies have hinted at a more reduced ice pack or a seasonally ice-free Arctic Ocean8,9,10. The uncertainty in sea-ice reconstructions is due in part to limited proxy-record coverage across the Arctic Ocean. In particular, there has been a dearth of records from the difficult-to-reach central Arctic Ocean, where the thickest sea ice persists. In this Article, we investigate sediment cores from this key area and other parts of the Arctic Ocean, using distribution patterns of the classic subpolar foraminifera species Turborotalita quinqueloba to demonstrate seasonally ice-free conditions during the LIG.

The ecology and role of T. quinqueloba

At high northern latitudes, the planktic foraminifer T. quinqueloba (genetic type IIa11) is considered a subpolar species thriving in areas that are largely ice free or bordering the sea-ice edge, predominantly in the Arctic–Atlantic gateways of the Barents Sea and Fram Strait12,13,14,15,16,17. A key observation is that T. quinqueloba abundance coincides with incoming Atlantic waters, while the East Greenland Current, composed of cold, low-salinity surface waters exiting the Arctic, contains very few T. quinqueloba15 (Extended Data Figs. 1 and 2). By contrast, the Arctic specialist Neogloboquadrina pachyderma (genetic type I11) has a much broader distribution, tolerating the polar extremes of the central Arctic Ocean, even under perennial sea ice, and occurs throughout the Nordic Seas and Fram Strait–Barents Sea region, including the East Greenland Current (Extended Data Figs. 1 and 2)15,16,17,18. A clear ecological boundary occurs northwest of Svalbard, at the border of the central Arctic Ocean province, beyond which resident populations of T. quinqueloba do not occur today and where N. pachyderma dominates (Extended Data Fig. 3)17,18. This boundary coincides with Atlantic Water sinking to depths of 200–500 m below the 100–200-m-thick Arctic halocline, in turn capped by the cold and fresh polar mixed layer19 (5–50 m thick; Fig. 1 and Extended Data Fig. 4). Rare living T. quinqueloba have been found at latitudes up to 86° N, but such individuals are considered advected non-residents, their main habitat and reproducing area occurring southwards18. Presumably, T. quinqueloba disappears in the central Arctic either because its preferred Atlantic Water habitat and associated food source is no longer at the surface, and/or it cannot tolerate the cold, fresh polar mixed layer and extensive sea-ice cover.

Fig. 1: Study area and core sites.
figure 1

Overview of marine sediment cores and ice cores mentioned in the text, together with the bathymetric map of the Arctic Ocean53 and mean ice-edge positions during September and January for the period 1981–201054. Yellow circles indicate core sites with new microfossil data. White circles indicate sites with microfossil data that is used in the data compilation (Extended Data Table 1). White squares indicate sites used for comparison with stratigraphic and geochemical data, these sites either have no planktic assemblage data from the >63 μm fraction (sites 16, 27) or calcareous microfossils are not well preserved (sites 29, 31). Grey circles are ice-core sites. The modern pathway of Atlantic-derived ocean currents is indicated with red arrows; the colour transparency change signifies subduction of Atlantic Waters beneath the 100–200-m-thick Arctic halocline. NAC, Norwegian Atlantic Current; WSC, West Spitsbergen Current; EGC, East Greenland Current; CAO, Central Arctic Ocean; FS, Fram Strait; BS, Barents Sea; ESS, East Siberian Sea; LS, Laptev Sea. The ice edge is defined as locations with a 50% probability of ice occurring at a concentration of 15% or greater54. Figure adapted from ref. 1 under a Creative Commons license CC BY 4.0.

Full size image

On the basis of the preceding associations, T. quinqueloba is often considered a reliable proxy for Atlantic Water influence in the sub-Arctic North Atlantic and has been used for reconstructing the position of the sea-ice edge, the Arctic front20,21 and even the extent of Arctic Atlantification in the Fram Strait22. Preferences for particular temperature and salinity conditions have been shown in a broad sense; for example, in the Fram Strait, T. quinqueloba occurs in temperatures of 0.7–3.4 °C and salinities of 34–35 psu (ref. 17). In the North Atlantic, peak abundances occur broadly at water temperatures of 4–7 °C and salinities of 34.8–35.1 psu (ref. 23), supporting a general intolerance of low-salinity polar surface waters (Extended Data Figs. 1, 2 and 5). Importantly, living T. quinqueloba abundance in the water column shows a predictable seasonal pattern, with one to two peaks per year, demonstrating that T. quinqueloba occurrence is also strongly coupled to the primary productivity cycle, in addition to temperature and salinity regimes12,14,15,24. The abundance of N. pachyderma also shows a seasonal signal, as observed in the Nordic Seas24. However, it appears to have greater potential to overwinter beneath sea ice, perhaps having evolved a stage of reduced metabolism or ability to harvest ice-stored organic matter that T. quinqueloba lacks14.

Invasion of the Arctic Ocean by T. quinqueloba

Our new planktic foraminifer assemblage records from five sites across the central Arctic Ocean (AO16-10PC, AO16-9PC, AO16-8GC, LRG12-3PC, 96/12-1pc) show that T. quinqueloba is largely absent from the modern central Arctic Ocean, in agreement with previous research25 (only rare individuals were found in near-surface sediments of core AO16-10PC). By contrast, sediment intervals dated to MIS5 (MIS5a to MIS5e) consistently record multiple abundance peaks of T. quinqueloba that account for 30–60% of the planktic foraminifera assemblage (Fig. 2). The age resolution of the MIS5 interval is ~2,000 yr cm–1 in all five records. We compile these results with all previously reported T. quinqueloba LIG occurrences from the Arctic and northern North Atlantic (Fig. 3 and Extended Data Table 1). Age models for the new cores rely on lithostratigraphic correlation to dated sedimentary sequences on the Lomonosov Ridge, in particular to the classic sediment cores LRG12-3PC, ACEX and 96/12-1pc26,27,28 (Extended Data Fig. 6). We show that in all central Arctic cores where calcareous microfossils are well preserved, T. quinqueloba occurs throughout the entire MIS5 interval, in conjunction with N. pachyderma (Figs. 2 and 3 and Extended Data Fig. 9). At the base of the MIS5 interval, interpreted to represent MIS5e, the T. quinqueloba abundance is generally highest and most consistent, reaching up to 60% of foraminifera assemblages. In all cores, T. quinqueloba abundances are low in the mid part of the MIS5 interval (MIS5b–5d), where N. pachyderma dominance returns. At the top of the MIS5 interval (MIS5a), T. quinqueloba abundance increases again in some cores (Fig. 2). Extremely rare occurrences (1–2 individuals) of another subpolar species, Globigerinita uvula, also occur in cores AO16-8GC (MIS5e) and LRG12-3PC (MIS5e and 5a). Neither Neogloboquadrina incompta nor Globigerina bulloides was observed; nor were other species.

Fig. 2: Foraminiferal data.
figure 2

a–f, Downcore variation of planktic foraminifera assemblage composition in the MIS5 interval in the central Arctic Ocean from key cores 96/12-1pc (a), AO16-9PC (b), LRG12-3PC (c), AO16-8GC (d) and AO16-10PC (e). Red colours indicate T. quinqueloba abundance; blue colours indicate N. pachyderma abundance. Star symbols indicate very rare occurrences of Globigerinita uvula. For stratigraphic context, see Extended Data Fig. 6. f, Global benthic δ18O stack for MIS555, with marine isotope substages indicated. g, Global benthic δ18O stack for the past 270 kyr.

Source data

Full size image
Fig. 3: Data compilation and interpretation.
figure 3

Compilation of new and existing T. quinqueloba occurrences (>1% of the seafloor assemblage) during modern times and the LIG, together with observed and reconstructed sea-ice conditions, with bathymetric map in the background53. a, Modern T. quinqueloba occurrences and sea-ice cover. Data are from published databases56,57 combined with results from this study. Sea-ice extent from 2020 is from NSIDC58. b, T. quinqueloba occurrences and sea-ice cover during the LIG. The extent of the LIG summer minimum sea-ice extent is inferred on the basis of our new T. quinqueloba constraints, assuming this species requires largely ice-free summers. This is a maximal estimate for the summer sea-ice minimum since there are currently no observations (sediment cores) north of the Canadian Arctic Archipelago region that comprise LIG sediments (shown by the diagonal hatched region)59,60,61,62,63,64,65,66,67,68,69,70. Figure adapted from ref. 53 under a Creative Commons license CC BY 4.0.

Full size image

Interpreting Arctic Ocean ice conditions and oceanography

The T. quinqueloba invasion documents a remarkable transformation of the central Arctic Ocean to subpolar-type conditions during the LIG. Extending the T. quinqueloba ecological constraints back in time, we propose that oceanic conditions comparable to those currently found in the Fram Strait/Barents Sea had propagated into the central Arctic Ocean. This implies that vast areas of open water and immensely reduced summer sea ice characterized the central Arctic Ocean. These conditions allowed seasonal primary production maxima supporting prolific T. quinqueloba populations. The spatial distribution of sediment cores with maximal T. quinqueloba abundances allows us to delineate a boundary of the summer sea-ice minimum during the LIG. Under the assumption that sea-ice retreat during the LIG occurred in a manner similar to the pattern of modern ice retreat, sea ice could not have extended farther than ~500 km from the Greenland–Canada coast (Fig. 3). Modern observations and model projections indeed identify the coastal zones bordering Canada–Greenland as the ‘last ice area’ for retreating sea ice, as a result of the configuration of Arctic winds and currents29. Moreover, simulations of sea ice also showed that the spatial pattern of sea-ice retreat during the LIG largely mimicked that of modern sea-ice retreat7. The northward expansion of boreal forest right up to the Arctic coastline with reconstructed temperatures 9 °C higher than today, as documented in Western Beringia5, also makes the persistence of multi-year sea ice in this region highly unlikely. Ice-free conditions were also proposed near Alaska on the basis of the finding of extralimital (non-native) molluscs10. Even if some sea ice just north of Canada/Greenland survived the summer melt, the LIG summer sea-ice minimum area could not have constituted more than 600,000 km2, which is well below the 1,000,000 km2 limit used as a threshold for an ice-free Arctic Ocean (hatched area in Fig. 3). The uncertainty about ice persisting in the region just north of Canada/Greenland exists because it remains a major gap for sediment coring and proxy data, implying that our reconstruction represents a maximal estimate of the extent of the sea-ice minimum. Since T. quinqueloba populations are highly seasonal24 and N. pachyderma remained present during the LIG, relatively extensive winter ice could have persisted. In the absence of any specific proxy for winter sea-ice extent, we do not attempt to delineate the winter margin for the LIG. The Laptev and East Siberian Sea shelf regions lack constraints due to the poor quality of calcareous microfossil preservation there, a consequence of carbonate-corrosive waters derived from Siberian river discharge30,31.

Multiple processes could lead to environmental conditions allowing T. quinqueloba to occupy the central Arctic Ocean. For this to happen, a stratification change involving shoaling of the halocline and/or its erosion would be required. Since T. quinqueloba prefers near-surface waters with Atlantic Water-influenced salinities and open waters in the summer to sustain a seasonal phytoplankton-based diet, we suggest that the Atlantic layer in the central Arctic Ocean shoaled to a water depth of ~50 m. This resulted in the northward propagation of the T. quinqueloba habitat, allowing it to invade the central Arctic (Fig. 4). Currently, in the southern Barents Sea, ongoing shoaling of the Atlantic Water and weakened stratification are indeed observed, and these are considered key processes contributing to the ‘Atlantification’ of the Arctic Ocean32. In the northern Barents Sea, a decrease in sea ice was shown to result in a weakened ocean stratification and increased upward fluxes of heat and salt33, similar to observations in the Eurasian basin34. A class of simple conceptional models suggests that the depth of the Arctic halocline is mobile and highly dependent on freshwater perturbations where an increased freshwater input in a silled basin such as the Arctic Ocean induces a shallowing of the halocline35. This implies that the freshwater input from melting of surrounding continental ice sheets during the LIG may have triggered a shallowing of the halocline, allowing Atlantic Waters to reach the productive photic zone and creating the environmental conditions suited for T. quinqueloba. Since the Greenland Ice Sheet was strongly reduced during at least one or multiple interglacials36,37, the prerequisite for such a scenario might well have been in place during the LIG. Another mechanism that could invoke the required stratification change is the erosion of the halocline, induced by a temperature increase of the inflowing Atlantic Water. A first-order approximation based on modern observations suggests that, given a fixed salinity of about 35 psu, an increase of the Atlantic layer temperature to about 6–7 °C would be needed to reach the same density as the bottom of the halocline and thus be conducive to mixing (Extended Data Fig. 8). Interestingly, peak occurrences of T. quinqueloba in the modern ocean are found exactly at these environmental conditions (Extended Data Fig. 5). However, this assumes that the halocline temperature at the LIG is similar to its modern value.

Fig. 4: Conceptual cartoon.
figure 4

Illustrative model depicting ocean stratification and sea-ice conditions (upper 500 m) in the modern Nordic Seas–Arctic Ocean region (a) and their hypothesized transformation during the LIG (b). The transect is located along the 5° E meridian, from 70° N to the North Pole. The occurrences of T. quinqueloba and N. pachyderma in the water column (determined from plankton tows in the modern case, inferred from sediments for the LIG case) are indicated with small black and white circles, respectively. Their relative abundances at the seafloor (summarized from existing and new data presented here) are indicated by pie charts.

Full size image

Once the optimal environmental conditions were in place, T. quinqueloba could theoretically have entered the Arctic Ocean from either the North Atlantic or the Pacific Ocean. Modern observations show that planktic foraminifera are not resident in, and do not migrate through, the Bering Strait, possibly because it is too shallow (sill depth ~53 m) and too acidic to support the calcifying plankton communities38, making the Atlantic–Arctic corridor the most logical route for the subpolar invaders. Interestingly, our LIG oceanographic reconstruction appears to reflect conditions similar to those expected to develop in response to ongoing Atlantification of the Arctic22. For example, once sea-ice export through the Fram Strait ceases (around the year 2050), it is expected that the Arctic planktic communities in the Fram Strait will abruptly shift towards non-sea ice and Atlantic species, and that these changes will propagate into the central Arctic12. Since T. quinqueloba percentages remain high in the top of the MIS5 interval, this would suggest that similar conditions also persisted through MIS5a, which appears consistent with prolonged interglacial warmth reconstructed from lake proxy data in northern Finland39.

Wider perspectives on LIG sea-ice conditions and climate

Previously, the Arctic occurrence of T. quinqueloba and its relation to reduced sea-ice conditions during the LIG had been suggested on the basis of a single core site just north of Greenland8. This interpretation was later challenged, with the suggestion that there was no extensive sea-ice loss but that open-water conditions were restricted in the form of local polynyas6. However, there is to our knowledge no evidence of T. quinqueloba being associated with polynya habitats in the Arctic40. The invasion of the Arctic basin by T. quinqueloba demonstrated in this study clearly renders the polynya hypothesis unlikely and corroborates basin-wide reduced sea-ice concentrations. This finding contrasts with the extensive, perennial sea-ice-cover reconstructions based on the absence of the open-water indicator brassicasterol and low concentrations of the biomarker IP25 (ref. 6) (Extended Data Fig. 7). We would argue that the presence of microfossils in the sedimentary record is currently a more robust proxy that is less ambiguous to interpret compared with the absence of biomolecules since the sources and sedimentary fate of the latter are still relatively unknown and under investigation41. In particular, the absence of IP25 remains difficult to interpret since this could be due to either perennial or ice-free conditions, or may be the result of low diatom production or degradational processes, especially in offshore deep-water settings as investigated here42. By contrast, modern environmental conditions of subpolar foraminifera are readily observable.

Our findings substantiate several studies that have hinted at reduced ice conditions but lacked conclusive, direct evidence. On land, extralimital molluscs were found at almost every LIG locality along the Alaska coast10. From this, it was suggested that winter sea ice did not expand south of Bering Strait, that the Bering Sea was annually ice free and that the Arctic Ocean ice cover was perhaps only seasonal at the time10. In addition, a well-known feature in multiple ice-core records from Greenland (NEEM, NGRIP, GIPS2 and Camp Century) is that stable oxygen isotope values from the LIG are at least 2.5‰ higher compared with present day, implying surface air temperatures that are 8 ± 4 °C warmer than during the last millenium43. Climate simulations that respond solely to greenhouse gases and orbital forcing have failed to capture these δ18O and temperature anomalies4,44. However, a recent isotope-enabled climate modelling study demonstrated that sea-ice loss was capable of producing the magnitude of the δ18O anomaly43.

Unifying this proxy-based evidence for a seasonally ice-free ocean with sea-ice reconstructions from climate models remains difficult due to the high levels of uncertainty involved in sea-ice simulations. Climate models produce LIG sea-ice conditions that range widely; most models simulate a relatively extensive, perennial sea-ice cover, whereas some simulate very little ice or ice-free conditions7,9. This wide spread can be expected since many general circulation models even have difficulty accurately capturing the recent changes in Arctic sea ice that have occurred during the past decades45,46, not to mention the subtleties of ocean stratification changes. Regardless, an ice-free Arctic Ocean is much more compatible with the marked Arctic warming that occurred during the LIG than is an Arctic Ocean with an extensive, perennial ice cover4,47. Data compilations indeed show that high-latitude sea surface and air temperatures were warmer than present during the LIG, with global sea level estimated at 5–10 m higher than today48,49,50. Furthermore, an ice-free Arctic can provide the missing link for the longstanding puzzle of why climate models have underestimated sea and air temperatures during this period9. This is supported by sensitivity experiments that demonstrate a strong response of Arctic temperatures to sea-ice retreat51. It is further consistent with the simulation of the Hadley Centre Global Environment Model version 3, which produced an excellent match to summer Arctic temperature proxies while simulating a seasonally ice-free Arctic Ocean9. In addition, while modern Atlantification is currently restricted to the eastern Eurasian basin and model outputs diverge with regard to future stratification changes52, our results demonstrate the potential for panarctic Atlantification occurring in a world of 0.5–1.5 °C above Preindustrial.

In the context of rapidly decreasing sea ice, an important goal of palaeoclimate studies is to identify the last time the Arctic became ice free. The invasion of the Arctic basin by T. quinqueloba provides the first firm evidence for extensive seasonally ice-free conditions across the interior Arctic Ocean where perennial sea ice remains in place today. Therefore, we put the LIG forward as the youngest and perhaps most relevant geological period for studying a seasonally ice-free Arctic Ocean, which is expected to occur this century. This information provides a useful constraint that can be included in ocean and climate models, leading to an improved understanding of Arctic sea ice in the climate system.

Methods

Core collection and core logging

Piston cores AO16-9PC and AO16-10PC were collected during the 2016 SWEDARCTIC expedition with icebreaker Oden71. While at sea, the bulk density, magnetic susceptibility and compressional wave velocity were measured at a 1 cm downcore resolution on whole cores using the Geotek Multi-Sensor Core. Measurements of X-ray fluorescence (XRF) were obtained with an ITRAX XRF core scanner at the Department of Geological Sciences, Stockholm University. The archive half of the core was scanned using a molybdenum tube set at 55 kV and 50 mA with a step size of 2 mm and an exposure time of 25 seconds. To remove background noise and produce analytically reliable counting statistics, the elemental data were normalized by zirconium, a conservative element commonly found in weathering-resistant minerals72. To optimize the original core photos, contrast, brightness and saturation were increased (Extended Data Fig. 6). This optimization changes dark brown to red colours and beige to yellow colours. Outliers from the multi-sensor core logger and XRF measurements, related to the edges of core sections (top and bottom), were removed.

Chronology and stratigraphy

The chronology (MIS1 to MIS6) of the new sediment cores AO16-10PC and AO16-9PC is derived through lithostratigraphic correlation to the well-dated sediment sequences found on the Lomonosov Ridge (Extended Data Fig. 6). Correlation is based on physical properties (bulk density, Mn/Zr variation) in combination with visual observations (for example, sediment colour). AO16-10PC and AO16-9PC were correlated to sediment core AO16-8GC and LRG12-7PC (Extended Data Fig. 6). This way, the stratigraphy of these cores could be fitted into the existing stratigraphic framework of the Lomonosov Ridge since core AO16-8GC was previously tied to LRG12-3PC73. LRG12-3PC is a key core because its chronology is constrained by nannofossil biostratigraphy27. In particular, the chronology of LRG12-3PC is anchored on the occurrence of two calcareous nannofossil taxa that have globally calibrated biozonations; Pseudoemiliania lacunosa, a species that went extinct during MIS12, and Emiliania huxleyi, a species that evolved during MIS874. Importantly, the nannofossil biostratigraphy of LRG12-3PC provided evidence that strongly supported previously developed age models that were established for the classic drill core ACEX26 and core 96/12-1pc28 (Extended Data Fig. 6). Other than AO16-10PC and AO16-9PC, all presented cores within this study have age models that were developed in previous studies, and their corresponding references can be found in Extended Data Table 1. While the age model adopted here is based on the most widely accepted paradigm regarding the Arctic chronostratigraphy—that glacial/interglacial periods have distinct and generally readily identifiable facies (especially sediments predating MIS7)73—it should be acknowledged that alternative age models based on Uranium decay series have been proposed75. These alternative age models would attribute the T. quinqueloba peaks to interglacials between MIS7 and MIS1376,77,78. From a climatological perspective, MIS5e is the most logical candidate for a seasonally ice-free Arctic Ocean as it is considered the most intense interglacial of the past 800 kyr, closely followed by MIS11c79. MIS7 is an interglacial of intermediate intensity, and MIS9 is considered fairly intense79. MIS13 is one of the weakest interglacials in the past 800 kyr (ref. 79). Therefore, even if our evidence for a seasonally ice-free ocean would be later attributed to an older (colder) interglacial, this would strongly imply that MIS5e would have been seasonally ice free as well.

Microfossil analysis

In sediment cores AO16-10PC, AO16-9PC, LRG12-3PC and AO16-8GC, samples of 1 cm thickness (5 cm3) were obtained throughout the respective MIS5 intervals. In addition, a series of 2-cm-thick samples that were used for an initial general assessment of microfossil content were also analysed. The samples were sieved over a 63-μm-mesh sieve, dried in an oven (<40 °C), weighed and transferred to glass vials. Counting of planktic foraminifera was performed on the dried >63 μm fraction. For core 96/12-1pc, pre-existing samples (63–125 μm) were analysed. Analysis of the >63 μm fraction is important because Arctic populations of some species (such as T. quinqueloba) are small and might be absent in larger size fractions57,80. Representative samples containing at least 300 tests were obtained by sediment splitting, performed using a microsplitter (ASC Scientific MS-1). Counting and identification of foraminifers was performed using a Leica M205C microscope. The census work is based on the taxonomic framework of ref. 12, which recognizes a single species name for N. pachyderma and T. quinqueloba, without distinguishing constituent morphotypes81. Special attention was given to avoid counting the lightly calcified Nps-5 N. pachyderma morphotype as T. quinqueloba82,83,84. Our taxonomy was cross-checked and verified using scanning electron microscopy to study the wall textures (Extended Data Fig. 9). Scanning electron microscope (SEM) imaging was carried out using a JSM-7000 F SEM for T. quinqueloba (accelerating voltage = 10 kV, working distance = 9.9 mm) and a TM 3000 for N. pachyderma. Both SEMs are housed in the Department of Materials and Environmental Chemistry, Stockholm University. The specimens were mounted on sticky carbon discs and gold coated before analysis (2 × 60 seconds, applied current of 20 mA). We found no evidence for carbonate dissolution within the MIS5e interval, and since the thin-shelled T. quinqueloba is more prone to dissolution than N. pachyderma14, the interpretations would remain the same even if differential dissolution had been at play.

Different morphotypes of N. pachyderma were observed but grouped into the same category. Benthic foraminifera contained a very small percentage of the total foraminifera abundance (<1% of the planktic plus benthic population). The most common benthic foraminifers include Oridorsalis tener, Bolivina arctica, Epistominella arctica, Stetsonia horvathi and Cibicides wuellerstorfi. A literature review was performed to document previously reported T. quinqueloba occurrences in the Arctic and Northern Atlantic oceans; these sampling sites and their respective references are found in Extended Data Table 1.

We also examined the cores LRG12-7PC, AO16-8GC and LRG12-3PC in terms of recording the first appearance of the key species E. huxleyi. A 10 cm sampling was carried out on the top 1.0–1.5 m of each core. In some cases, the sampling resolution was increased to 5, 3 and 2 cm in dark-coloured units and intervals close to biozonal boundaries. Smear slides were prepared from unprocessed sediments and were examined under a polarizing light microscope (Zeiss Axio Scope.A1) at ×1,000 and ×1,250 magnifications.