Water, plants, and early human habitats in eastern Africa

Abstract

Water and its influence on plants likely exerted strong adaptive pressures in human evolution. Understanding relationships among water, plants, and early humans is limited both by incomplete terrestrial records of environmental change and by indirect proxy data for water availability. Here we present a continuous record of stable hydrogen-isotope compositions (expressed as δD values) for lipid biomarkers preserved in lake sediments from an early Pleistocene archaeological site in eastern Africa—Olduvai Gorge. We convert sedimentary leaf- and algal-lipid δD values into estimates for ancient source-water δD values by accounting for biochemical, physiological, and environmental influences on isotopic fractionation via published water–lipid enrichment factors for living plants, algae, and recent sediments. Reconstructed precipitation and lake-water δD values, respectively, are consistent with modern isotopic hydrology and reveal that dramatic fluctuations in water availability accompanied ecosystem changes. Drier conditions, indicated by less negative δD values, occur in association with stable carbon-isotopic evidence for open, C₄-dominated grassland ecosystems. Wetter conditions, indicated by lower δD values, are associated with expanded woody cover across the ancient landscape. Estimates for ancient precipitation amounts, based on reconstructed precipitation δD values, range between approximately 250 and 700 mm·y⁻¹ and are consistent with modern precipitation data for eastern Africa. We conclude that freshwater availability exerted a substantial influence on eastern African ecosystems and, by extension, was central to early human proliferation during periods of rapid climate change.

The role of water and ecosystem change in human evolution remains a subject of active debate (1–3), but experts widely acknowledge that these factors likely shaped early human (hominin) foraging strategies and diet (4) approximtely 2.0–1.8 million years ago (Ma). According to marine records, this juncture occurred during an interval of protracted grassland expansion across eastern Africa (1). In contrast, coeval terrestrial records from hominin archaeological sites such as Olduvai Gorge indicate recurrent fluctuations between open-grassland and closed-woodland ecosystems (5). Although such ecosystem fluctuations are commonly interpreted in relation to aridity, grassland expansion is sensitive to multiple factors (6), and proxy signals more closely linked to meteoric waters could strengthen interpretations.

Here we use δD values for lipid biomarkers preserved in lake sediments to reconstruct source-water δD values at Olduvai Gorge during the early Pleistocene. In modern lake basins, precipitation and lake-water δD values correlate strongly with δD values for leaf- and algal-lipids, respectively, after accounting for isotopic fractionation during lipid biosynthesis (7). Present and past source-water δD values reflect the combined influences of vapor-source, transport history, and aridity, ultimately linking local hydrologic patterns to atmospheric and oceanic-circulation dynamics (8).

Sample Locality.

Olduvai Gorge is incised into the eastern margin of the Serengeti Plain in northern Tanzania (Fig. 1). Gorge walls expose a thick sequence of volcaniclastic sediments and tuff accumulated in lake and lake-margin environments (9, 10). Between approximately 2.0 and 1.8 Ma, a perennial saline-alkaline lake (paleolake Olduvai) occupied the center of the closed basin (9–11). Lake-margin stratigraphy suggests paleolake Olduvai responded sensitively to local and regional precipitation patterns during this 200,000-y interval (9–12). Stable carbon-isotope compositions (expressed as δ¹³C values) of leaf-lipids preserved in the lake sediments vary by more than 15‰, suggesting that pronounced ecosystem fluctuations accompanied lake-level fluctuations (5). Here we measure leaf-lipid δD values in a subset of the same lake sediment outcrop. All samples are from locality 80 (Bed I), which accumulated near the depocenter of paleolake Olduvai (10) and was exposed by stream incision during the late Pleistocene (9).

Fig. 1.

Modern precipitation patterns in eastern Africa (66) during the (A) long rains (March to May) and (B) the short rains (October to December), with respect to average monthly precipitation in northern Tanzania (C) (Data courtesy of http://climexp.knmi.nl). In A and B, bold horizontal lines mark the position of the Intertropical Convergence Zone, whereas dashed lines mark the Interoceanic Confluence (13). Target symbols mark the location of Olduvai Gorge (2° 48′S, 35° 06′E). In C, the bold line reflects observed average monthly rainfall (1964–1984); gray envelops variation for average monthly precipitation. The dashed line reflects modeled average monthly rainfall for Tanzania (67). (D) Depositional environments surrounding paleolake Olduvai during the early Pleistocene (9). Contracted (dashed line) and expanded (bold line) lake margins are based on correlated stratigraphic sections (9–12).

Precipitation Patterns in Eastern Africa.

In eastern Africa, precipitation patterns reflect interactions between regional highlands and two convergent boundaries (Fig. 1). The Intertropical Convergence Zone (ITCZ) and closely associated African rainbelt (13) mark north–south convergence of monsoon airstreams. The Interoceanic Confluence (IOC; also referred to as the Congo Air Boundary) marks east–west convergence of air masses derived from the Indian and Atlantic Oceans (14). Both the ITCZ and IOC migrate in response to insolation-driven surface heating patterns (13), yielding two rainy seasons. In eastern Africa, the “long rains” (March to May) account for more than 50% of annual totals (15). Precipitation during the “short rains” (October to December) is more variable but correlates strongly with annual totals (16).

Precipitation δD values (δD_rain) reflect the origins and histories of the related air masses that produce it. In eastern Africa, δD_rain values correlate inversely with precipitation amounts due to the influences of vapor-source, transport history, and “amount” effects (17). Today the Indian Ocean is the primary vapor source to eastern Africa (13); the Atlantic Ocean and continental surface-water evaporation are important secondary sources. Transport history and amount effects result in δD_rain values that differ between the rainy seasons (18). On average, δD_rain values for the long rains are less negative (−20‰) than for the short rains (−28‰); in central eastern Africa, the average annual δD_rain value is approximately −22‰ (19–22).

Leaf-Lipid Apparent Fractionation Factors.

Plant-water δD values reflect soil-water δD values (δD_soil) (7). Isotopic relationships between δD_soil and δD_rain values can be influenced by surface evaporation in arid and semiarid regions (8), but plant-water δD values attenuate evaporative signals by accessing deep soil moisture and increasing productivity during rainy seasons (23).

Soil water is not fractionated significantly during uptake by plant roots, and stem-water δD values generally reflect δD_soil values (23). In contrast, leaf-water δD values can vary markedly from δD_soil values as a result of transpiration (24). The relative importance of stem-water vs. leaf-water during lipid biosynthesis remains unclear (25, 26), as does biosynthetic phenology. These factors could account for differences in apparent fractionation between source-waters and lipids (ɛ_lipid/water = [(δD_lipid + 1,000)/(δD_water + 1,000)] − 1, expressed in permil (‰)) among living plants when grouped according to plant functional type (PFT).

We reevaluate a global compilation of published leaf-lipid δD values for living plants (SI Appendix) to determine representative ɛ_lipid/water values for different PFTs relevant to this study. We target subtropical and tropical taxa in clades most representative of plants in eastern Africa since the early Pleistocene (27, 28). Leaves of subtropical and tropical plants commonly contain the leaf-lipid hentriacontane (nC₃₁) (29, 30), and nC₃₁ is also abundant in lake sediments from Olduvai Gorge (5). Therefore, we use published δD values for nC₃₁ (δD₃₁) to determine ɛ_lipid/water values applicable to sedimentary δD₃₁ values from Olduvai Gorge.

We define PFTs in terms of photosynthetic pathway and growth habit: C₃ woody plants, C₃ herbs, and C₄ graminoids. We determine representative ɛ_lipid/water values—termed ɛ_31/model values—from published δD₃₁ values using modeled annual δD_rain values () (21, 22). Measured annual δD_rain values rarely accompany published leaf-lipid δD values but, when available, generally coincide with values (7, 21).

Collectively, ɛ_31/model values average −124‰ (ɛ_average). Among individual PFTs the median ɛ_31/model value for C₄ graminoids is most negative (−146‰ ± 8‰, 95% confidence interval; n = 51). The median ɛ_31/model value for C₃ herbs (−124‰ ± 10‰; n = 24) is more negative than for C₃ woody plants (−109‰ ± 8‰; n = 84).

We calculate “landscape” apparent fractionation factors (ɛ_landscape) using ɛ_31/model values and relative PFT abundances (Fig. 2) estimated from δ¹³C values for nC₃₁ (δ¹³C₃₁) in lake sediments (5). Leaf-lipids represent relative PFT abundances in basins ranging from small lakes (31) to expansive river systems (32), despite differences in basin morphology, transport, and burial processes (33). We multiply each ɛ_31/model value by relative PFT abundance (i.e., f_woody, f_herb, and f_gram for woody plants, herbs, and gramminoids, respectively) to calculate ɛ_landscape values:

Fig. 2.

Schematic depiction of ɛ_31/model values as a function of reconstructed ecosystem. (A) We use δ¹³C₃₁ values to estimate relative abundances for three different PFTs: C₄ graminoids, C₃ herbs, and C₃ woody plants. (B) We relate relative PFT abundances to ecosystem (5) according to United Nations Educational, Scientific, and Cultural Organization terminology (68). (C) We calculate “landscape” apparent fractionation factors (ɛ_landscape) for deuterium by mass balance.

Finally, we apply ɛ_landscape values to sedimentary δD₃₁ values to reconstruct δD_soil values.

The relative influences of biochemical, physiological, and environmental processes on ɛ_lipid/water values are difficult to account for in interpretations of δD₃₁ values. For instance, our ɛ_31/model value for C₃ woody plants does not distinguish by canopy height, despite observed differences between modern trees and shrubs (SI Appendix). Similarly, annual δD_rain values do not distinguish seasonality. In eastern Africa, annual and rainy-season δD_rain values can differ in excess of 20‰ (SI Appendix), although >85% of annual precipitation occurs during rainy seasons (15). Thus, to the extent that plant growth takes place in rainy seasons, annual δD_rain values can overestimate δD_soil values during lipid biosynthesis.

Although propagated uncertainty in ɛ_landscape values (SI Appendix) represents a substantial portion of the variability in modern δD_rain values across central eastern Africa (approximately 60‰), sedimentary δD₃₁ values capture space- and time-integrated signals that attenuate variability of individual plants or species (7, 31). Thus, uncertainty in ɛ_landscape values largely reflects ecosystem-scale differences in transpiration and phenology (7), which are at least partially accounted for by ɛ_31/model values, provided living plants are representative of their ancient counterparts. Despite some significant shortcomings in the state of the art, ɛ_landscape values provide a useful interpretational framework to account for biological and physical influences on leaf-lipid δD values—a factor often overlooked in hydrologic reconstructions.

Algal-Lipid Apparent Fractionation Factors.

Aquatic photosynthetic organisms acquire hydrogen for lipid biosynthesis from ambient waters (7). Therefore, algal δD values reflect lake-water δD values—which integrate precipitation, groundwater, runoff, and evaporation—as modified by biosynthetic fractionation. Biosynthetic fractionation, in turn, incorporates biological and physical factors (7). Culture studies indicate that algal ɛ_lipid/water values vary between species (34), but space- and time-integration seems to minimize these effects in sediments (7).

Field studies indicate that algal ɛ_lipid/water values vary in relation to salinity and must be accounted for when interpreting algal-lipid δD values (35). The response of algal ɛ_lipid/water values to salinity is markedly consistent at 0.9‰ ± 0.2‰ ppt⁻¹ (7). Heptacosane (nC₁₇) is a general biomarker for algae (36) and is abundant in lake sediments from Olduvai Gorge. Modern studies establish an apparent fractionation between freshwater and nC₁₇ equal to −172‰ (7), and we use this value to determine algal ɛ_lipid/water values at different salinities—termed ɛ_lake values:

We construct a basic lake-water evaporation model to constrain ɛ_lake values in the past (SI Appendix). Briefly, we estimate the total solute load for paleolake Olduvai on the basis of stratigraphic evidence for maximum lake area (approximately 200 km²) (10) and lake level (approximately 5 m) (12) during the early Pleistocene and fossil evidence for minimum salinity (approximately 20 ppt) (9). Then, we infer changes in lake level from changes in sedimentary total organic carbon (_%TOC) because these values covary in many modern lakes in eastern Africa (37). Next, we assume a conservative solute balance and use estimates for paleolake volume to estimate salinities during lake contraction. Finally, we apply ɛ_lake values to sedimentary δD₁₇ values to reconstruct lake-water δD values (δD_lake). Salinity estimates for paleolake Olduvai range from approximately 20 to 100 ppt, resulting in ɛ_lake values that vary by up to 88‰ from freshwater algal ɛ_lipid/water values.

Biosynthetic processes responsible for the influence of salinity on δD₁₇ values are unclear (7), and not all possible mechanisms result in linear relationships. Still, ɛ_lake values range between −158‰ and −84‰ for paleolake Olduvai, highlighting the importance of salinity when interpreting δD_lake values from sedimentary δD₁₇ values.

Results

Sedimentary δ¹³C₃₁ values range from −36.3‰ to −21.4‰, with an average value of −27.8‰ (Fig. 3). Sedimentary δD₃₁ values range from −148% to –132‰ and correlate weakly with δ¹³C₃₁ values (r² = 0.11) and _%TOC (r² = 0.08). Reconstructed δD_soil values show an increased isotopic range of 54‰, from −38‰ to +16‰ (Fig. 4).

Fig. 3.

Relationships between leaf- and algal-lipid δD values and ancillary proxy data for Olduvai Gorge. (A) Sedimentary δ¹³C₃₁ values (5). (B) Dust fluxes into the Arabian Sea (gray line) and closed- or moist-habitat bovids by percentage of total bovids (black circles) (1, 69). (C) Sedimentary δD₃₁ values (black circles) alongside ɛ_ecosystem values (gray line). (D) Stratigraphic evidence for lake levels (9–12). Horizons that contain faunal evidence for low (down arrows) or high (up arrows) lake levels are also shown (12). (E) Sedimentary total organic carbon. (F) Sedimentary δD₁₇ values (black circles) alongside ɛ_lake values (gray line).

Fig. 4.

Leaf- and algal-lipid δD values and reconstructed source waters with respect to sedimentary _%TOC. (A) Measured δD₃₁ (hollow circles) and reconstructed δD_soil values (black circles) with respect to _%TOC. For reference, a gray line marks the linear regression for values based on the average ɛ_31/model value of −124‰. (B) Measured δD₁₇ (hollow circles) and reconstructed δD_lake values (black circles) with respect to _%TOC. A gray line marks the linear regression for values based on a freshwater ɛ_17/water value of −172‰.

Sedimentary δD₁₇ values range from −150‰ to −30‰ (Fig. 4). Measured values correlate strongly with δ¹³C₃₁ values (r² = 0.83) and _%TOC (r² = 0.86). Reconstructed δD_lake values show a relatively smaller the isotopic range, from +3‰ to +59‰, but still correlate strongly with δ¹³C₃₁ values (r² = 0.88). Interestingly, sedimentary δD₁₇ values correlate weakly with sedimentary δD₃₁ values (r² = 0.11), but reconstructed δD_lake values correlate strongly with reconstructed δD_soil values (r² = 0.84).

Interpretations and Discussion

Precipitation in Eastern Africa.

Historical precipitation patterns serve as a framework for interpreting reconstructed hydrologic patterns over timescales of 10¹ to 10³ thousand years (38), although regional tectonism and the intensification of zonal atmospheric (Walker) circulation during the early Pleistocene could weaken this interpretational link (1, 13). Modern δD_rain values reveal a regional meteoric waterline (RMWL) for eastern Africa (Fig. 5):

Fig. 5.

Modern isotopic relationships for precipitation and lake waters in eastern Africa. We plot a RMWL (bold line) based on modern monthly δD_rain and δ¹⁸O_rain values and single precipitation events (19–22, 39, 40): . For reference, we also plot the global meteoric waterline (dashed line). A large black circle marks the point defined by a single reconstructed δD_soil and published δ¹⁸O_rain value reconstructed from minerals in closely associated lake-margin sediments (SI Appendix). Propagated uncertainties are shown (SI Appendix). For reference, a cross marks the δD value based on the average ɛ_31/model value of −124‰. Modern δD_lake and δ¹⁸O_lake values yield an LEL for eastern Africa (48–53): . A large black square marks the point defined by a single reconstructed δD_lake and published δ¹⁸O_lake value reconstructed from authigenic clays in associated lake sediments (SI Appendix), whereas the overlying cross marks the δD value based on a freshwater ɛ_17/water value of −172‰.

Today, annual δD_rain values in eastern Africa range from approximately −30‰ and −10‰, whereas monthly δD_rain values range from approximately −50‰ and +30‰ (19–22, 39–41). We pair reconstructed δD_soil values with published δ¹⁸O_rain values derived from lake and soil-carbonate minerals in closely associated lake (12) and lake-margin (41) sediments (SI Appendix). Reconstructed δD_soil–δ¹⁸O_rain values plot within the range of modern precipitation data for eastern Africa and show strong agreement with the RMWL (Fig. 5).

A wide range in δD_soil values (and, by inference δD_rain values) indicates pronounced changes in vapor source, transport history, amount of precipitation, or a combination thereof. Today, changes in vapor source account for up to 10‰ variability in δD_rain values (40). Sea-surface temperatures in vapor-source regions influence δD_rain values by approximately 1‰ °C⁻¹ (42), accounting for up to 5‰ variability during the early Pleistocene (43). Transport history is difficult to constrain (44) but can account for up to 10‰ variability today. Together, changes in vapor source and transport history can account for nearly half of the variability (25‰) in reconstructed δD_soil values for eastern Africa. If so, changes in the amount of precipitation account for the remaining variability (29 of 54‰) in reconstructed δD_soil values.

The relationship between modern δD_rain values and amount of precipitation is difficult to evaluate in many tropical regions because of sparse measurements, but available data suggest comparable effects at seasonal, annual, and interannual timescales (44, 45). In eastern Africa, available rainy-season δD_rain values correlate inversely with precipitation at a slope of −0.125‰ mm⁻¹ (SI Appendix). If modern sensitivity is representative for the past, and changes in vapor-source and transport history were important, then a 29‰ range in δD_soil values translate to 225 mm precipitation range; the full 54‰ range translates to 415-mm precipitation range.

Historical data for annual δD_rain values (−22‰) (19, 39) and mean annual precipitation (MAP) (550 mm) (9, 15) provide a local reference point from which to project reconstructed δD_soil values. Because rainy seasons account for approximately 85% of MAP in northern Tanzania (15), we use an amount effect for rainy-season months to reconstruct MAP in the past:

MAP estimates for the full range of δD_soil values are from ∼700–250 mm. This range is consistent with reconstructions based on pollen spectra (∼750 mm) (46) and soil carbonates (<400 mm) (47) during wetter and drier intervals, respectively. Woody cover strongly covaries with MAP today in eastern Africa (6):

We find a similar relationship between our estimates of MAP and woody cover (5). For instance, in the modern calibration, MAP of 700 mm yields an f_woody value of 0.84, which is consistent with a δ¹³C₃₁-derived (−36.3‰) f_woody value of 0.90 (5). These observations suggest amount effects influenced δD_rain values more than changes in vapor-source or transport history at Olduvai Gorge and highlight the importance of using ɛ_landscape values to reconstruct hydrologic patterns from leaf-lipid δD values.

Lake-Water Evaporation in Eastern Africa.

Loss of lighter isotopic species (H₂¹⁶O) during evaporation progressively enriches residual lake waters in DH¹⁶O and H₂¹⁸O. In eastern Africa, lake-water δD values (δD_lake) and δ¹⁸O values (δ¹⁸O_lake) define an isotopic trajectory—called a local evaporation line (LEL)—with a slope that is lower than that of the RMWL (8). LEL slopes are primarily a function of relative humidity (h); in general, very low h values (e.g., 0.25) result in slopes close to 4, whereas higher h values result in slopes closer to 6 (8). Modern δD_lake and δ¹⁸O_lake values yield an LEL for eastern Africa:

Modern δD_lake values range from approximately −30‰ in humid regions of eastern Africa to +80‰ or higher in extremely arid regions (48–53).

The LEL defines source-water composition at its intersection with the RMWL. For modern waters in eastern Africa, LEL and RMWL intersect at a source-water δD value of −22‰ (Fig. 5), which closely matches historical data (19–22). In closed basins, lake waters derive primarily from precipitation (54).

Tracing Isotopic Hydrology at Olduvai Gorge.

To compare modern and ancient lake-waters, we pair reconstructed δD_lake values with published δ¹⁸O_lake values that were determined from authigenic clays (12) in associated sediments (SI Appendix). Reconstructed δD_lake–δ¹⁸O_lake values show close agreement with the modern LEL (Fig. 5). Further, reconstructed δD_lake and δD_soil values strongly correlate, suggesting lake-water compositions shifted largely owing to changes in precipitation. Evaporation rates decrease at high salinity because of the decreased activity of water in high ionic-strength solutions. As a result, potential evaporation can exceed lake-water evaporation by up to 100-fold (54). Thus, although reconstructed δD_lake values vary only slightly more than δD_soil values, changes in source-water and amount of precipitation would have been accompanied by large changes in potential evaporation. Reconstructed lake evaporation relative to meteoric input (E/I) based on our data suggest higher evaporation during intervals of reduced precipitation (E/I = 2.9) than during increased precipitation (E/I = 1.3) and are consistent with historical and modeled E/I values for eastern Africa (SI Appendix).

Water Availability and Ecosystem Dynamics.

Reconstructed δD_rain and δD_lake values reveal strong relationships between water and carbon-isotopic data for ecosystem type. Lower δD_rain and δD_lake values, which reflect increased MAP and decreased evaporation, respectively, correspond with increased woody cover (f_woody = 0.90). Although the organic carbon-derived indicators we use to determine ɛ_landscape and ɛ_lake values may be codependent (55), reconstructed values for f_woody and lake level are consistent with independent indicators for ecosystem type and paleolake level (Fig. 3). Much like today (6), aridity was a dominant control on ecosystem change in eastern Africa during the Pleistocene.

Contrasting proxy records have fueled debate about the pace and patterns of environmental change in eastern Africa during the Pleistocene. Pollen and fossil abundance records suggest expansion of arid-adapted species beginning near 2.0 Ma and culminating around 1.8 Ma (56, 57). Marine dust-flux records and soil-carbonate δ¹³C values also suggest shifts toward more arid conditions around 1.8 Ma (1), although geomorphic evidence suggests regionally wetter conditions (1).

Lipid biomarkers from Olduvai Gorge point to rapid changes in plants and water between approximately 2.0 and 1.8 Ma, and we suggest this environmental variability both influenced and can reconcile proxy records. For instance, increased seasonality can lead to C₄ graminoid expansions (58) but can also lead to unrepresentatively positive C₄-like δ¹³C values in soil carbonates (59). Similarly, rapid wet-to-dry transitions can simultaneously produce both increased dust and elevated lake-levels (60). Over the past several million years, modulation of marine dust-flux records from the Arabian Sea has been tightly coupled with orbital eccentricity, resulting in distinct intervals of exceptionally high-amplitude variability during orbital-eccentricity maxima (1). We hypothesize that high-amplitude, orbital precession-paced environmental variability, as opposed to gradual or stepwise aridification, characterized eastern Africa during the early Pleistocene.

Water and Early Human Evolution.

In semiarid regions, precipitation primarily determines water availability (15). Today, water availability shapes primate behaviors through its influence on vegetation and resource distributions (4, 61). For example, regions with MAP <700 mm do not support chimpanzee populations (62). Water likely shaped behavioral adaptations in the genus Homo (63). Our evidence for dramatic variability is consistent with water as a strong selective pressure in human evolution (1, 2). However, thermoregulatory and dietary constraints function at microhabitat scales (64), and many hominin fossil sites—including Olduvai Gorge (9)—are associated with ephemeral or saline water sources (65). Our reconstructions of precipitation and lake-chemistry indicate that, even during maximum lake expansion, lake waters at Olduvai Gorge were likely not potable (12). Groundwater-fed freshwater springs could have aided hominin existence and proliferation (65).

Conclusions

This study presents a continuous record of δD values for lipid biomarkers from lake sediments at Olduvai Gorge that were deposited during a key juncture in human evolution, ∼2.0–1.8 Ma. We pair sedimentary leaf-lipid δD values with corresponding δ¹³C values to account for physiological and environmental influences on reconstructed precipitation δD values. We use a basic lake-water evaporation model to account for the influence of salinity on algal-lipid δD values and reconstructed lake-water δD values. Sedimentary leaf- and algal-lipid δD values show a weak relationship, but “corrected” values correlate strongly. We compare reconstructed precipitation and lake-water δD values with isotopic data for environmental waters in modern eastern African to estimate ancient precipitation amounts and evaporative losses, respectively. Our results indicate Olduvai Gorge received approximately 250 mm of MAP during arid intervals and ∼700 mm during wetter intervals. Given the magnitude and variability in water availability revealed by our reconstructions, we hypothesize freshwater springs were important for hominin subsistence in highly variable environments.

Materials and Methods

Lipid Extraction and Purification.

Lipid biomarkers in lake sediments were extracted and separated according to Magill et al. (5).

Isotopic Analysis.

Lipid δD values were measured by gas chromatography–isotope ratio monitoring–mass spectrometry. Hydrogen gas of known δD value was used as reference. Samples were coinjected with internal standards of known δD values (nC₄₁, androstane, and squalane). Uncertainty (1σ SD) for nC₄₁ was 4‰.

Age Model.

Our age model is constrained using previously published ⁴⁰Ar/³⁹Ar, magnetic stratigraphy, and tuff correlation dates (5).