Apolar fractions were analyzed on an Agilent 7890A gas-chromatograph flame ionization detector (GC-FID) to determine concentrations of the n-alkanes. Samples were run in splitless mode with an inlet temperature of 250 ∘C and inlet flow rate of 26.5 mL min-1, using hydrogen as the carrier gas. Separation was achieved using a 5 % phenyl methyl siloxane column (HP-5; 60m×320µm×0.25µm), with a flow rate of 4.6 mL min-1. The oven temperature program was as follows: 70 ∘C for 2 min, increasing at 17 ∘C min-1 to 130 ∘C, then increasing at 7 ∘C min-1 to 320 ∘C, and finally holding at 320 ∘C for 15 min. Compound identification was performed by comparison to a standard mixture of C21–C40 n-alkanes injected during each run and by confirming compound identification for a subset of samples via GC mass spectrometry, following the methods detailed in Keisling et al. (2017). Concentrations of each n-alkane were determined using an external squalene calibration curve. Here, we examine leaf wax distributions using the average chain length (ACL; Bray and Evans, 1961; Bush and McInerney, 2013), calculated for the terrestrial (C27 to C33) n-alkanes (Eq. 1): 1ACL=∑Cn⋅n/∑Cn.

The polar fractions were filtered through a 0.45 µm PTFE filter in 99:1 hexane : isopropanol (v/v). Isoprenoid glycerol dialkyl glycerol tetraethers (iGDGTs) and brGDGTs were analyzed using an Agilent 1260 ultra-high-performance liquid chromatograph (UHPLC) coupled to an Agilent 6120 single quadrupole mass selective detector (MSD) and following the methods detailed by Hopmans et al. (2016), which separates the 5- and 6-methyl brGDGT isomers. Briefly, separation is achieved using a silica precolumn with two BEH HILIC columns in series (2.1×150 mm, 1.7 µm; Waters^®). Samples were eluted using hexane (solvent A) and hexane : isopropanol (9:1 v/v; solvent B) in the following program: 18 % solvent B for 25 min, a linear increase to 35 % solvent B over 25 min, then a linear increase to 100 % solvent B for 30 min. The column temperature was 30 ∘C and the flow rate was 0.2 mL min-1. Compounds were ionized using atmospheric pressure chemical ionization, and the MSD was run in single ion monitoring (SIM) mode for the GDGT core lipids.

Lake El'gygytgyn sediments are dominated by brGDGTs with only small concentrations of iGDGTs present (Fig. S4 in the Supplement) (Holland et al., 2013; D'Anjou et al., 2013; de Wet et al., 2016; Keisling et al., 2017; Daniels et al., 2021). Thus, we use the methylation of branched tetraethers index (Eq. 2), considering only the 5-methyl isomers (MBT5ME′), to reconstruct past temperature (De Jonge et al., 2014). 2MBT5ME′=Ia+Ib+IcIa+Ib+Ic+IIa+IIb+IIc+IIIa We explored the use of several lacustrine brGDGT calibrations to convert MBT5ME′ values to temperature. The first is based on a suite of lakes in East Africa that span an elevation transect to capture temperature gradients (Russell et al., 2018; Eq. 3). This calibration is to mean annual air temperature. 3MAAT=-1.21+32.42⋅MBT5ME′(RMSE=2.44∘C)

More recently, Zhao et al. (2021) developed an in situ calibration to summer (JJA) water temperature using sediment trap data from a site in southern Greenland (Eq. 4). 4JJAwatertemp.=-1.82+56.06⋅MBT5ME′(RMSE=0.58∘C)

Raberg et al. (2021) explored relationships between compound fractional abundances (FAs) within structural groups based on methylation number, methylation position, cyclization number and environmental parameters using a previously published globally distributed data and adding new sites from the high latitudes. They recommend using their “Meth” calibration for general use in lake sediments (Eq. 5). 5MAF∘C=92.9±15.98+63.84±15.58×fIbMeth2-130.51±30.73×fIbMeth-28.77±5.44×fIIIaMeth2-72.28±17.38×fIIbMeth2-5.88±1.36×fIIcMeth2+20.89±7.69×fIIIaMeth2-40.54±5.89×fIIIaMeth-80.47±19.19×fIIIbMeth(RMSE=2.14∘C) Raberg et al. (2021) also provide a “Full” calibration (Eq. 6) with the highest R2 and lowest RMSE in their data and suggest this calibration may be appropriate to apply at sites with good conductivity or pH control. 6MAF∘C=-8.06±1.56+37.52±2.35×fIaFull-266.83±98.61×fIbFull2+133.42±19.51×fIbFull+100.85±9.27×fIIa′Full2+58.15±10.09×fIIIa′Full2+12.79(±2.89)×fIIIaFull(RMSE=1.97∘C) We also use a mean annual air temperature calibration from Feng et al. (2019) which was developed using lake sediments and several decades of instrumental climate data from Tiancai Lake, China (Eq. 7). 7MAAT∘C=14.74-34.46×fIIIa+27.49×fIIa-35.56×fIIb-60.36×f1a-95.91×f(Ib)(RMSE=0.18∘C) We also apply the global Bayesian calibration for lacustrine brGDGTs, known as BayMBT, using the baymbt_predict() MATLAB function (Martínez-Sosa et al., 2021). Following the approach of Martínez-Sosa et al. (2021), we estimated the prior value for our dataset by calculating a mean temperature using the Russell et al. (2018) calibration; we also used 10 ∘C for the standard deviation.

Our new brGDGT data from Lake El'gygytgyn document relative temperature changes, revealing several important aspects of orbital and long-term climate variability throughout the MPT. The MBT5ME′ record shows that temperatures varied at Milankovitch and sub-Milankovitch timescales and generally follows changes noted in the global benthic oxygen isotope stack (Lisiecki and Raymo, 2005) and insolation (Fig. 3). This observation agrees with prior studies of Lake El'gygytgyn (e.g., de Wet et al., 2016), although it is evident that the relative importance of obliquity and precession-pacing has varied temporally (Figs. 3 and 4). Our MBT5ME′ temperature record supports the observation of Melles et al. (2012) that congruency between peak precession, obliquity, and eccentricity together with inter-hemispheric teleconnections can generate superinterglacial periods such as MIS 11 and 31. The brGDGT data generally support this interpretation during MIS 31, as well as during the alignment of Northern Hemisphere summer perihelion and obliquity during the warm MIS 21 (Huybers and Wunsch, 2005), but we note that several other intervals are as warm in the MBT5ME′ record despite differences in the orbital configurations. Furthermore, we note that the inverse is also true for temperatures at Lake El'gygytgyn whereby congruent minima in summer insolation, obliquity, and relatively low eccentricity resulted in very cold conditions during MIS 28 and MIS 22 (Fig. 3).

We observe no long-term trend in the MBT5ME′ data (Fig. 3), which contrasts with cooling trends observed in the global benthic δ18O stack (Lisiecki and Raymo, 2005; Clark et al., 2006; Li et al., 2004), Mg/Ca-derived temperatures of Atlantic deep waters (Sosdian and Rosenthal, 2009), and some sea surface temperature records from the northern high latitudes (e.g., Martínez-Garcia et al., 2010; Lawrence et al., 2009; McClymont et al., 2008). At Lake El'gygytgyn, Francke et al. (2013) report changes in orbital frequencies of grain size distributions before and after the MPT, while other proxies at Lake El'gygytgyn do not exhibit any long-term trends across the MPT interval (Wennrich et al., 2016). These other proxies do not directly record temperature yet appear to agree with the brGDGT data in suggesting relatively stable long-term conditions across the MPT. The lack of MPT cooling at El'gygytgyn is difficult to explain given the expansion of Northern Hemisphere ice sheets at that time. It could imply that the climate at the study site is not representative of the pan-Arctic region, and indeed, there is considerable spatial variability in climate change across the Arctic (Daniels et al., 2021; Tulenko et al., 2020). Alternatively, it may suggest that Arctic cooling was not the critical driver of intensified ice sheet growth, implicating a strong role for the regolith removal hypothesis (Clark and Pollard, 1998; Yehudai et al., 2021) or Southern Hemisphere (i.e., Antarctic) cooling and ice sheet expansion (Ford and Raymo, 2020).

Although we see no overall trend in the brGDGT data, there is a notable increase in leaf wax ACL values reflecting a significant climate-driven ecological change across the MPT (Fig. 2), most likely indicating aridification of the study region. A long-term aridification trend over the past ∼1 Myr is similarly noted from the Chinese Loess Plateau (Zhao et al., 2018; Zhou et al., 2018; Wu et al., 2020). At El'gygytgyn, aridity is in part controlled by the amount of moisture being sourced from the nearby high-latitude seas. A sedimentary record from the Northwind Ridge in the Western Arctic Ocean indicates a transition from seasonal sea ice to perennial sea ice around 1 Ma, which could explain the observed trend toward drier conditions (Dipre et al., 2018). The contrast between enhanced sea ice coverage and stable temperatures at El'gygytgyn points to strong geographic variations in climatic cooling across the MPT, particularly between the marine realm and northeast Russia.

A characteristic feature of the MPT is the appearance of longer glacial cycles, expressed in the frequency domain of many paleoceanographic and continental records as increasing power in the quasi-100 kyr band, although there is no fundamental change in orbital forcing across the MPT. In the Arctic, precession and obliquity are key drivers of warm-season temperatures because of the large changes in peak summer insolation and changes in summer duration. Indeed, spectral analysis of the Lake El'gygytgyn MBT5ME′ record reveals significant obliquity and precession cycles over the MPT (Fig. 4), with the obliquity signal being the strongest prior to ∼1.0 Ma (Fig. 4). Increased variability at longer wavelengths is observed after ∼900 ka but at an ∼82 kyr cyclicity, which may be twice an obliquity signal rather than eccentricity. However, we note that the interval studied here may be too short to fully evaluate changes in longer orbital frequencies occurring during the MPT. Prior studies of the Lake El'gygytgyn drill core report similar spectral results but with some notable differences in the relative strength of various frequencies. Grain-size data, reflecting climate-dependent (glacial–interglacial) clastic sedimentation processes at Lake El'gygytgyn, also display a strong obliquity cycle throughout the MPT. Yet in contrast to the brGDGT data, the strongest obliquity signature is noted in the younger part of the record from 950 to 670 ka (Francke et al., 2013). The grain-size data also exhibit a strong precession cycle from 1100 to 900 ka (Francke et al., 2013), in agreement with the brGDGT data. The authors note that changes in grain size at Lake El'gygytgyn are not directly coupled to changes in global ice volume.

The effect of MPT ice sheet expansion on spectral signatures would naturally be strongest in direct proximity to where the ice sheets develop, namely North America, Greenland, and Fennoscandia. In northeast Siberia, however, ice sheets have not been as prominent a feature of the Pleistocene environment (Brigham-Grette et al., 2013), and it is not entirely clear how the 100 kyr ice sheet influence is transferred to Lake El'gygytgyn. Through model simulations, Melles et al. (2012) indicated that temperature variability at Lake El'gygytgyn is decoupled from Greenland Ice Sheet growth and decay. Instead, precession variations here are more likely connected with regional insolation differences or latitudinal climatic teleconnections (Francke et al., 2013), which may help explain the lack of brGDGT-inferred cooling across the MPT.

Interestingly, the Lake El'gygytgyn brGDGT data exhibit the strongest spectral power at ∼10.9 kyr cyclicity (Fig. 4), which could reflect a half-precession signal (Verschuren et al., 2009). The signal is strongest between 850 and 1030 ka and is apparent, for example, in the structure of the substages of MIS 21. Sub-orbital climate variations have been noted at Lake El'gygytgyn but not discussed in detail (e.g., Wennrich et al., 2016), as most proxies are dominated by the Milankovitch frequencies. Nonetheless, half-precession has been observed during the MPT in the northern high latitudes. Haneda et al. (2020) report half-precession variability in the Kuroshio current of the western North Pacific during MIS 19. Likewise, Ferretti et al. (2010) report that during MIS 21, foraminifera δ18O varied strongly at the 10.6 kyr wavelength at Site U1313 (Fig. 1; this Site is a revisit of DSDP 607) and a number of other sites across the North Atlantic. Their finding supports a previous identification of Milankovitch harmonics in the North Atlantic (Wara et al., 2002). In that region, the isotope signal was driven by variations in temperature and the strength of Atlantic Meridional Overturning Circulation (AMOC).

The presence of half-precession variability in the North Atlantic has been attributed to non-linear feedbacks due to orbital forcing within the North Atlantic region, related to the different timescales of ice sheet dynamics, deep ocean convection, and moisture feedbacks (Wara et al., 2002). It has alternatively been linked to tropical hemi-precession (e.g., Verschuren et al., 2009) being transmitted to the high latitudes via oceanic and atmospheric teleconnections (Ferretti et al., 2010). In the Northwest Pacific, Haneda et al. (2020) suggest a combination of AMOC variation and equatorial insolation forcing generated a signal of half-precession during MIS 19. Half-precession (9.2–12.7 kyr variance) is also apparent in Late Pleistocene thermocline temperatures from the Western Equatorial Pacific, resulting in sub-orbital variability in the east–west gradient across the equatorial Pacific (Jian et al., 2020). The thermocline water temperature gradient plays a controlling role in the dynamics of the El Niño Southern Oscillation and Walker circulation in the tropical Pacific, which in turn have been linked to climate in the Bering Strait region, mainly through their influence on the position and strength of the Aleutian Low pressure system (Niebauer, 1988). Considering that Bering Strait climatology can exert a strong influence on temperatures at Lake El'gygytgyn (Nolan et al., 2013), this atmospheric teleconnection may be the most direct explanation for the occurrence of half-precession in the El'gygytgyn MBT5ME′ record.

In addition to potentially influencing sub-orbital climate dynamics at El'gygytgyn, strengthening of Walker circulation between 1.17 and 0.9 Ma has been hypothesized to be a key driver of the mid-Pleistocene transition (McClymont and Rosell-Melé, 2005). Walker cell intensification likely resulted in a westward shift and deepening of the Aleutian low, thereby lowering air and sea surface temperatures over the Bering Sea and potentially contributing to enhanced upwelling and expanded sea ice (Worne et al., 2021). In contrast to the potential half-precession sensitivity of Lake El'gygytgyn MBT5ME′ to tropical and sub-tropical Pacific dynamics, we do not see a clear effect on Walker circulation changes across the MPT in the MBT5ME′ temperature record presented here, potentially because changing position of the Aleutian low can result in variable temperature responses at El'gygytgyn (Nolan et al., 2013). Furthermore, McClymont and Rosell-Melé (2005) speculate that intensification of Walker circulation increased winter precipitation in the Arctic, contributing to ice sheet expansion. This mechanism contrasts with the vegetation change and moisture reduction across the MPT suggested by pollen and leaf waxes at El'gygytgyn, potentially because of locally expanded sea-ice or lower continental temperatures suppressing evaporation in Lake El'gygytgyn moisture source regions. We note that the half-precession signal in the brGDGT record (Fig. 4) gets stronger between approximately 1.1 and 0.9 Ma, while Walker circulation was intensifying prior to the first skipped interglacial. Future climate modeling could elucidate how shifts in Walker circulation may have influenced sub-orbital variations in Pacific climate variability, including its influence at Lake El'gygytgyn.

In the following discussion, we take a closer look at glacial–interglacial variability noted in our new Lake El'gygytgyn biomarker records.

In the interval spanning MIS 34 to 26, we observe a strong correspondence between the MBT5ME′ temperatures and global benthic foraminifera oxygen isotope (δ18O) stack (Lisiecki and Raymo, 2005) (Fig. 3). The coldest periods in the MBT5ME′ record align closely with glacial Facies A in the El'gygytgyn sediment profile (Fig. 3), as well as with biomarker-inferred expansions of sea ice beyond the marginal ice zone of the Bering Sea (Fig. 6; Detlef et al., 2018) indicating cooling across the Bering region during the MIS 32, 30, 28, and 26. While precession and half-precession are apparent in the evolutionary power spectrum (Fig. 4), the temperature was primarily paced at 41 kyr over this interval, as seen in alignment between the smoothed brGDGT temperature record and the obliquity history (Fig. S9). This could reflect the importance of summer duration on Arctic summer temperature or lake water temperature. Alternatively, the 41 kyr pacing may be driven by changes in glacial/interglacial pCO2 variations, for which there is not a well-resolved record over this time span, but it likely tracked global climate conditions at 41 kyr pacing (Berends et al., 2021).

During MIS 31, the MBT5ME′ record shows somewhat surprising results in comparison to other Lake El'gygytgyn proxies (Fig. 3). In the El'gygytgyn record, superinterglacial periods (interglacials that are exceptionally warm and wet) were previously identified based on the presence of Facies C, a sedimentary unit characterized by weakly laminated reddish oxidized sediments, high Si/Ti ratios, and low organic matter content (Melles et al., 2012). Facies C is interpreted as reflecting an oxygenated water column, high aquatic productivity, seasonal lake ice, and generally warm conditions (Melles et al., 2012). Pollen spectra from Facies C intervals typically corroborate this interpretation, exhibiting a temporary appearance of birch, alder, or other trees (Melles et al., 2012). The MBT5ME′ data show a temperature increase of ca. 5–8 ∘C (using the BayMBT calibration; Martínez-Sosa et al., 2021) compared to the glacial stages immediately preceding and following. However, MIS 29, 27, and 21 were as warm or warmer than MIS 31 in the MBT5ME′ record, yet Facies C is absent during these intervals. The differing redox conditions seen in Facies C may have impacted brGDGT distributions, suppressing the warming signal during MIS 31. However, previous empirical evidence suggests that oxic conditions would most likely accentuate MIS 31 warmth (Buckles et al., 2014; Martínez-Sosa and Tierney, 2019), which is not what we observe. We cannot fully exclude other microbial factors that may impact brGDGT producers, but we can infer that other meteorological variables contributed to the presence of Facies C. Based on mineral characteristics, Wei et al. (2014) demonstrated that precipitation is a key driver of Facies C formation. This may imply that MIS 29, 27, and 21 were relatively arid in comparison to MIS 31. The pollen record indicates that cool conifer forest was present around Lake El'gygytgyn during MIS 31 and that higher precipitation characterized this superinterglacial (Zhao et al., 2018; Lozhkin and Anderson, 2013; Melles et al., 2012). Additionally, it is possible that MIS 31 was windier than other interglacials, thereby decreasing water column stability and increasing ventilation and nutrient availability in the surface waters. Indeed, high Si/Ti values, a proxy for diatom productivity, are noted in the Lake El'gygytgyn record during MIS 31 (Melles et al., 2012).

In the interval from MIS 34 to 26, the n-alkane ACL record has the highest resolution during MIS 31. At this time, a shift to lower ACL values is noted (Figs. 2 and 5), potentially driven by wetter conditions. During the other interglacials in this interval (MIS 33, 29, and 27), lower ACL values generally occur during peak interglacial conditions. However, low sample resolution during MIS 27 hampers full evaluation of this relationship, and higher variability in ACL values is noted during MIS 33, 29, and 27 compared to MIS 31. In the pollen record, MIS 29 is characterized by Alnus and Betula pollen while MIS 27 is characterized by an increase in larch, dwarf birch, and alder pollen (Zhao et al., 2018). Our MBT5ME′ record differs from the pollen record in that MIS 29 appears warmer compared to MIS 27, whereas in the pollen record, MIS 27 is interpreted as the warmer and wetter of these interglacials (Zhao et al., 2018). The Lake El'gygytgyn pollen record displays a sharp increase in cold steppe pollen at ∼975 ka indicating the onset of glacial MIS 26 (Zhao et al., 2018). This dramatic change appears to be reflected in the ACL record with the largest shift in our record initiating at ∼963 ka, when ACL values increase significantly (Fig. 2).

In the Lake El'gygytgyn brGDGT record, MIS 25 is comparatively cooler than other interglacial periods (Fig. 3). Summer insolation during MIS 25 reaches its highest value of the study interval, yet reconstructed temperatures are similar to the preceding and following glacial stages when insolation and obliquity were lower. MBT5ME′ values average ∼0.25, or 6 ∘C on the BayMBT calibration. Following a brief increase in n-alkane ACL values at the termination of MIS 26, ACL again declines during the MIS 25 obliquity maximum circa 953 ka, coinciding with the expansion of trees and shrubs in the area as noted in the pollen record (Zhao et al., 2018). Whereas Zhao et al. (2018) infer warm conditions during MIS 25, the cool brGDGT-inferred temperatures suggest that the vegetation change was rather driven by a combination of longer growing seasons and increase in the moisture balance. During MIS 25, the El'gygytgyn Si/Ti ratio is slightly elevated (Fig. 3), suggesting an increase in lake productivity. In the absence of increased summer temperatures, the change in productivity was likely caused by a combination of longer growing season, increased runoff, or increased windiness promoting mixing of the water column. Regionally, the weak MIS 25 warming agrees well with alkenone-derived SSTs from ODP 882 in the North Pacific (Martínez-Garcia et al., 2010), which also exhibits a relatively cool MIS 25. At Site U1343 in the nearby Bering Sea, paleoceanographic changes are complex. A low abundance of sea ice biomarkers points to ice-free conditions during MIS 25 (Fig. 6g; Detlef et al., 2018), while high opal accumulation rates (Kim et al., 2014), the presence of ice-marginal diatom species (Worne et al., 2021), and a high nutrient upwelling index (Worne et al., 2020) indicate a peak in marginal sea ice conditions, increased wind strength, and a longer sea ice melt season. If these changes in wind strength and seasonality extended over Arctic Asia, it would help explain the limnological and vegetation changes observed at El'gygytygyn.

Numerous MPT studies note a global cooling trend with increasing sea ice extent culminating in an anonymously cool or “skipped” MIS 23 interglacial and a strong glacial period during MIS 22 (Head and Gibbard, 2015, and references therein), in effect giving rise to the first long glacial cycle of the Late Pleistocene. The lack of warming at El'gygytgyn during MIS 25 obscures the nature of the first long glacial cycle from MIS 24–22 that is apparent in the δ18O history (Lisiecki and Raymo, 2005; Clark et al., 2006). From MIS 24 to 22, there is a strengthening of sub-orbital temperature variations, with two short-lived warming excursions bracketing a brief cold period at 905 ka despite peaks in both obliquity and Northern Hemisphere summer insolation (Fig. 3). The evolutionary power spectrum shows that the strength of the 41 kyr climate variability weakens from MIS 25–22, whereas the sub-Milankovitch frequencies become more prominent (Fig. 4), suggesting a breakdown of the climate dynamics that dominated previously. The lower-resolution ACL data are in closer agreement with the benthic isotope stack, exhibiting an increasing trend from MIS 25 to 22 and an abrupt decrease at the beginning of MIS 21 (Fig. 3).

Overprinting the increase in ACL values are a series of glacial–interglacial oscillations. For reasons that are not clear, pollen is absent in Lake El'gygytgyn sediments during MIS 23 (Zhao et al., 2018) yet n-alkanes are present. During the peak of MIS 23, ACL values are high and are higher than those of the previous glacials, suggesting conditions that were either arid or more glacial-like (Fig. 2). Temperatures during MIS 22 were particularly cold and arid, characterized by some of the lowest MBT5ME′ values and the highest ACL values of the record (Fig. 3). These cool conditions were associated with particularly low productivity, seen in the Si/Ti ratio (Fig. 3). There was a two-phase termination of MIS 22, with an abrupt warming at 880 ka followed by a brief return to cold conditions just prior to the final warming event. The two-phase deglaciation is also apparent in the doublet of Facies A during MIS 22 (Fig. 3). The deglacial warming is approximately synchronous with the increasing obliquity, leading the deglacial decrease in δ18O (Fig. 3).

While MIS 29, 27, and 21 all show MBT5ME′ values as high as the superinterglacial MIS 31, the most pronounced warm period of the brGDGT record is MIS 21 (Fig. 3). Warming in the MBT5ME′ starts early in MIS 22 and continues to ∼868 ka when peak interglacial conditions are noted. Temperatures increased by at least 4 ∘C and possibly up to 10 ∘C at the start of MIS 21 as seen in the BayMBT calibration. The comparatively rapid warming recorded at Lake El'gygytgyn from MIS 22 to MIS 21 agrees with Pacific records (Fig. 6; Martínez-Garcia et al., 2010; McClymont et al., 2013; Kender et al., 2018; Lattaud et al., 2019). From 865–845 ka shrubby vegetation communities, including stone pine, birch, and alder, returned to the region, while n-alkane ACL decreased, thereby suggesting warm and wet conditions (Zhao et al., 2018). The reemergence of tree and shrub communities during MIS 21 is also reflected in the loess sediments of northeast China (not shown; Zhou et al., 2018) suggesting a widespread shift to milder conditions. Despite the notable warmth and a slight increase in aquatic productivity seen in the Si/Ti ratios, MIS 21 is not identified as a superinterglacial interval based on the Si/Ti or lithofacies records (Melles et al., 2012).

Interglacial conditions remained relatively warm for approximately 40 kyr during MIS 21–20. The return to glacial conditions at El'gygytgyn differs notably from the global benthic isotope composite (Lisiecki and Raymo, 2005); whereas the δ18O data indicate a gradual cooling or gradual development of ice sheets culminating in full glacial conditions around 814 ka, the brGDGT data indicate an earlier and more rapid cooling to glacial conditions around 835 ka, which was then followed by relatively stable or even warming climate moving into MIS 20 (Fig. 3). This early cooling is also seen in the pollen record, which shows the appearance of open steppe vegetation from 845–810 ka (Zhao et al., 2018). The apparently cool, open-landscape environment in the latter half of MIS 21 occurred despite increasing obliquity, reminiscent of the scenario across MIS 25.