Ecosystem variability and early human habitats in eastern Africa

Abstract

The role of savannas during the course of early human evolution has been debated for nearly a century, in part because of difficulties in characterizing local ecosystems from fossil and sediment records. Here, we present high-resolution lipid biomarker and isotopic signatures for organic matter preserved in lake sediments at Olduvai Gorge during a key juncture in human evolution about 2.0 Ma—the emergence and dispersal of Homo erectus (sensu lato). Using published data for modern plants and soils, we construct a framework for ecological interpretations of stable carbon-isotope compositions (expressed as δ¹³C values) of lipid biomarkers from ancient plants. Within this framework, δ¹³C values for sedimentary leaf lipids and total organic carbon from Olduvai Gorge indicate recurrent ecosystem variations, where open C₄ grasslands abruptly transitioned to closed C₃ forests within several hundreds to thousands of years. Carbon-isotopic signatures correlate most strongly with Earth’s orbital geometry (precession), and tropical sea-surface temperatures are significant secondary predictors in partial regression analyses. The scale and pace of repeated ecosystem variations at Olduvai Gorge contrast with long-held views of directional or stepwise aridification and grassland expansion in eastern Africa during the early Pleistocene and provide a local perspective on environmental hypotheses of human evolution.

Climate-dependent ecosystem characteristics, such as habitat and water availability, likely influenced natural selection during human evolution (1). For example, woody plants may have influenced thermoregulatory and dietary adaptations in hominins and other terrestrial mammals since the Pleistocene (2–4) about 2.6 Ma. Unfortunately, in many cases, reconstructions of ecosystem characteristics and climate at hominin archaeological sites are limited by poor preservation and coarse temporal resolution. Moreover, discontinuities are common in terrestrial sediment sequences. As a result, much of the environmental context of human evolution has been interpreted based on regional and global conditions reconstructed from marine records (5, 6).

The role of savannas in human evolution remains a subject of debate (5–8). This debate stems, in part, from the historically imprecise definition of savanna for modern and ancient ecosystems and the difficulties of estimating plant community compositions—particularly woody cover—from sediments. Recently, the work by Cerling et al. (7) estimated plant community compositions based on present day relationships between woody cover and carbon-isotope compositions for soil carbonates and soil organic matter (SOM). This approach offers insights into ecosystem structures at hominin archaeological sites (e.g., Omo-Turkana Basin), but it is limited to environments supporting ancient soils (paleosols).

Here, we extend the approach in the work by Cerling et al. (7) to include lipid biomarkers archived in lake sediments deposited between about 2.0 and 1.8 Ma at an important hominin archaeological site—Olduvai Gorge. In addition to including key junctures in human evolution (8), this time interval is associated with important changes in tropical climate, including strengthening of east–west (Walker) atmospheric circulation across the Indian and Pacific Oceans (9). Weakened Walker circulation before about 2.0 Ma was similar to conditions projected to accompany the continued rise in greenhouse gas concentrations during the coming century (10). To examine connections among ocean and atmospheric circulation, regional climate, and plant community composition, we also compare our organic carbon signatures to reconstructions of polar ice volume and sea-surface temperatures (SSTs) in the Atlantic and Indian Oceans.

Background

Site Descriptions.

Olduvai Gorge is just south of the equator in northern Tanzania (2° 48′S, 35° 06′E), where it cuts across a 50-km rift platform basin to expose a 2.0 million y sequence of lake and river sediments (Fig. 1). The basin formed on the western margin of the East African Rift System in response to extension tectonics and the growth of a large volcanic complex (11, 12). During the early Pleistocene, the basin area covered an estimated 3,500 km² and included a saline–alkaline lake near its center (12). Sediments from this central lake are composed primarily of reworked volcanic material and air-fall tuffs (11). For our study, we use samples recovered from outcrop exposures near the center of the paleolake that preserve a stratigraphic record of continuous deposition (11).

Fig. 1.

Present day precipitation patterns during long rains (A; March to May) and short rains (B; October to December) (75). Bold horizontal lines mark the ITCZ. The IOC (B; dashed line) is prominent during short rains, and its eastward displacement correlates with higher seasonal precipitation (66). Targets show Olduvai Gorge’s location (●, ODP site 662; ○, ODP site 722). (C) Depositional environments at Olduvai Gorge during the early Pleistocene about 2.0–1.8 Ma. Expanded and contracted lake levels are reconstructed based on detailed stratigraphic correlations (11, 12, 43).

Today, annual precipitation patterns at Olduvai Gorge and surrounding regions of eastern Africa are defined by monsoon circulation by two major convergence zones (13)—the Intertropical Convergence Zone (ITCZ) and the Interoceanic Confluence (IOC). The ITCZ marks convergence of regional trade winds, whereas the IOC marks zonal confluence of water vapor derived from the Atlantic and Indian Oceans (Fig. 1). Seasonal migrations of the ITCZ and the IOC result in an annual cycle consisting of two rainy seasons separated by arid conditions that last from May to September. Long rains (March to May) produce the largest proportion of total annual precipitation; short rains (October to December) are more variable but also correlate with total annual precipitation (14). Today, Olduvai Gorge experiences mean annual precipitation (MAP) of about 550 mm; several independent proxy archives suggest that MAP ranged between about 400 and 900 mm during the early Pleistocene (15–17).

We compare our data with coeval records for alkenone-derived SSTs from the eastern Atlantic and western Indian Oceans (18–20) (Fig. 1). Ocean Drilling Program (ODP) site 662 (1° 23′S, 11° 44′W, 3,824 m water depth) is in the eastern Atlantic Ocean in the Gulf of Guinea. ODP site 722 (16° 37′N, 59° 48′E, 2,028 m water depth) is in the northwestern Indian Ocean. Today, SSTs at both sites are sensitive to monsoon-driven seasonal upwelling (21–23), but surface sediment calibrations indicate that alkenone signals reflect mean annual SSTs (22, 23).

Sedimentary Organic Matter.

Organic matter in lake sediments derives from bacteria, algae, and plants (24), and it integrates contributions from the surrounding watershed (25–28). In contrast, organic matter in soils depends more strongly on local sources and preservation (29, 30), and it can vary spatially over scales of 10 m² or less (25).

Biomarkers are molecular fossils that have structures with biological specificity (31). Those biomarkers from plants and algae carry isotopic signals derived from terrestrial and aquatic sources, respectively. For instance, long straight-chained hydrocarbons, such as nonacosane (nC₂₉) and nC₃₁, occur abundantly in leaves (32) and indicate sedimentary inputs from terrestrial plants (33). Although somewhat less specific, certain shorter-chained hydrocarbons (e.g., nC₁₇ and nC₁₉) indicate contributions from algae and cyanobacteria (31).

Carbon Isotopes in Leaves, Biomarkers, and Soil Organic Matter.

Photosynthetic carbon-isotopic fractionation is defined between atmospheric carbon dioxide (δ¹³C_CO2) and leaf tissues (δ¹³C_leaf), and its magnitude varies with plant functional type and water availability (34, 35):

We note that ɛ-values are expressed in permil, which is a unit of parts per thousand. Variability in ɛ_CO2/leaf, which is approximately equivalent to negative Δ_leaf values (widely used in ecological literature) (34), is well-constrained for different photosynthetic pathways (35), but these modern relationships are not easily applied to ancient plants because of limited preservation of leaf tissues through time. Relatively recalcitrant leaf lipids, such as nC₃₁, afford an opportunity to circumvent this challenge provided that fractionation between δ¹³C_leaf and nC₃₁ (δ¹³C₃₁) during biosynthesis can be documented for subtropical and tropical plants (36).

We evaluated carbon-isotopic relationships between soil organic matter (δ¹³C_SOM), leaf tissues, and leaf lipids using published data for 64 plants species and 288 soils from nearly 300 tropical and subtropical localities (Fig. 2). For C₃ plant soil systems, we find that nC₃₁ is ¹³C-depleted by about 7‰ (n = 45) with respect to leaf tissue, whereas SOM is ¹³C-enriched by about 2‰ (n = 184) relative to leaves. Thus, in C₃ ecosystems, there is an isotopic difference (ɛ_SOM/31) between nC₃₁ and SOM equal to about 9‰:

Fig. 2.

Histograms and box-and-whisker plots for ɛ_SOM/31 from published δ¹³C_SOM and δ¹³C₃₁ values from subtropical and tropical regions (SI Appendix). Published data are binned at 0.5‰ intervals (counts) based on site averages. Values for δ¹³C_SOM (dark gray) and δ¹³C₃₁ (light gray) are plotted relative to δ¹³C_leaf. Values of ɛ_SOM/31 equal about 9‰ in both C₃ and C₄ ecosystems. Individual boxes contain interquartile ranges (IQRs). Bold vertical lines within boxes mark median values. ○, mean values. Horizontal whiskers mark minimum and maximum values, except values outside of 1.5 IQR (●). Notch half-width values indicate confidence in differentiating median values.

In C₄ plant soil systems, both nC₃₁ and SOM are ¹³C-depleted with respect to leaf tissue by about 10‰ (n = 19) and 1‰ (n = 104), respectively. Accordingly, ɛ_SOM/31 in C₄ ecosystems also equals about 9‰. Therefore, we apply ɛ_SOM/31 of 9‰ to estimate δ¹³C_SOM from δ¹³C₃₁ values, thus extending to plant waxes the predictive capabilities of δ¹³C_SOM for estimating woody cover (7) (Fig. 3). We caution that the value of 9‰ for ɛ_SOM/31 derived here is based on data from xeric woodlands and scrublands, tropical deciduous forests, and C₄ grasslands, and it may not be representative for other ecosystems (36).

Fig. 3.

Stable carbon-isotope compositions for SOM (δ¹³C_SOM; bottom x axis) and nC₃₁ (δ¹³C₃₁; top x axis) relative to woody plant cover (A), ecosystem (B), and plant community composition (C). The published relationship between woody plant cover and δ¹³C_SOM is described by the following function, where −13 ≤ δ¹³C_SOM ≥ −31‰ (10): . Here, we relate δ¹³C_SOM to δ¹³C₃₁ using a value of 9‰ for ɛ_SOM/31. The relationship between woody plant cover and δ¹³C₃₁ is, thus, described by the following function, where −22 ≤ δ¹³C_SOM ≥ −40‰): . Ecosystem definitions adhere to African plant communities according to UNESCO terminology (40).

The fractional abundance of C₃ plants is inversely related to C₄ plants in tropical ecosystems (25, 37). However, the relationship between woody cover (f_woody) and C₄ abundance is nonlinear (7), because herbaceous C₃ plants can occur in both open and wooded ecosystems (38). As a result, for δ¹³C_SOM values between near-total C₃ composition (−30‰) and negligible C₃ cover (−14‰), we follow the approach in the work by Cerling et al. (7) to estimate f_woody (where −13 ≤ δ¹³C_SOM ≥ −31‰):

Structural Classification for Ancient Ecosystems.

The broad functional definition for savannas as a continuous herbaceous understory with irregular distributions of trees or bushes does not account for differences in f_woody (39). Therefore, we adopt definitions for African plant communities defined by the United Nations Educational, Scientific and Cultural Organization (UNESCO) (40). According to this classification scheme, (i) forests display continuous tree cover (>10 m) with interlocking crowns and poorly developed understory, (ii) woodlands—including bush/shrublands—display open or closed stands of shrubs or trees (up to 8 m) with at least 40% woody plant cover and understory with grasses and other herbs, (iii) wooded grasslands display 10–40% woody plant cover and well-developed groundcover with grasses and other herbs, (iv) grasslands display less than 10% woody plant cover and well-developed groundcover with grasses and other herbs, and (v) deserts display sparse groundcover and sandy, stony, or rocky substrate. UNESCO does not distinguish between forests and woodlands in terms of woody plant cover, but here, we consider forests to display greater than 80% woody plant cover (7, 39).

The evolutionary implications of orbital forcing on environmental change were recognized over a century ago (41) but remain controversial today (3–6). Marine sediments just off African shores document orbital rhythms in terrestrial inputs during the Pleistocene (5), and a growing number of terrestrial sequences hint at a similar pacing for environmental changes in eastern Africa (42, 43). Marine sequences are often indirectly or too poorly constrained in time to infer relationships between key junctures in human evolution and terrestrial conditions or change (6). Lithostratigraphic patterns in lake margin sediments at Olduvai Gorge reveal five episodes of lake expansion between ca. 1.85 and 1.74 Ma, suggesting that lake level changes may have tracked orbital precession (43). Here, we use sedimentary organic matter signatures in high-resolution and temporally well-constrained lake sediments to evaluate the magnitude and timing of ecosystem changes associated with lake level changes.

Results and Discussion

Ecosystem Change and Woody Cover.

We observe repeated shifts in δ¹³C₃₁ values between about −36‰ and −20‰ (Fig. 4), which track closely with orbitally paced lake margin lithostratigraphic patterns, suggesting that ecosystem changes at Olduvai Gorge were also influenced by orbital cycles (Fig. 5). Total organic carbon (TOC) δ¹³C values (δ¹³C_TOC) show a smaller isotopic range of about 9‰ and correlate tightly with δ¹³C₃₁ values (r² = 0.86); δ¹³C_TOC values follow terrestrial plant inputs but are attenuated, likely by algal or macrophytic inputs. Taken together, δ¹³C₃₁ and δ¹³C_TOC values suggest rapid local ecosystem shifts between closed C₃ woodlands and open C₄ grasslands. These changes were comparable with extreme events, such as the greening of the Sahara about 120,000 y ago, that accompanied the dispersal of modern humans out of Africa (44).

Fig. 4.

Biomarker and isotopic signatures for organic matter preserved in lake sediments at Olduvai Gorge. (A) Leaf lipid δ¹³C values for nC₃₁ (δ¹³C₃₁). (B) TOC δ¹³C values (δ¹³C_TOC). (C) TOC percentages (_%TOC). (D) Diagrammatic depiction of relative lake level changes at Olduvai Gorge based on lake margin lithostratigraphy (43). (E) Ratios of algal lipids (nC₁₇ + nC₁₉) relative to algal and terrestrial plant lipids (nC₁₇ + nC₁₉ + nC₂₉ + nC₃₁). Higher values reflect relatively increased algal inputs (31). (F) Ratios of pristane (Pr) to phytane (Ph). In lake sediments, values above two generally reflect a dominance of terrestrial plant inputs (31). (G) Ratios of macrophytic lipids (nC₂₃ + nC₂₅) relative to macrophytic and terrestrial plant lipids (nC₂₃ + nC₂₅ + nC₂₉ + nC₃₁; that is, P_aq). Higher values reflect relatively increased macrophytic inputs (50). (H) Values of δ¹³C for the macrophytic lipid nC₂₅ (δ¹³C₂₅). Horizontal gray bands highlight periods of time characterized by δ¹³C₃₁ values higher than about −28‰ (i.e., open grassland and wooded grassland ecosystems).

Fig. 5.

Time series for global, regional, and local proxy indicators during the early Pleistocene. (A) Global benthic oxygen-isotopic composite (19). Gray lines connect original data, and the bold black line shows a five-point smoothing. (B and C) Alkenone-derived SST estimates for the western Indian Ocean (SST₇₂₂) and eastern Atlantic Ocean (SST₆₆₂), respectively (18–20). Gray lines connect original data, and bold black lines show five-point smoothing. (D) Calculated orbital precession (ω_p). Values for ω_p equal the product of calculated eccentricity (e) and the sine function of the longitude of the perihelion (ω). (E) Leaf lipid δ¹³C values for nC₃₁ (δ¹³C₃₁; ○ connected by bold black lines) and TOC δ¹³C values (δ¹³C_TOC; gray lines). F, forest; G, grassland; W, bush/shrubland and woodland; WG, wooded grassland. Ecosystem definitions adhere to African plant communities according to UNESCO terminology (40) based on δ¹³C₃₁ values (Fig. 4). Vertical gray bands highlight periods of time characterized by δ¹³C₃₁ values higher than about −28‰ (i.e., open grassland and wooded grassland ecosystems). Hominin taxonomic grades and fossil occurrences are illustrated using information in the work by Wood (76); skulls denote taxonomic grades that have been identified at Olduvai Gorge.

Carbon-isotope evidence for pronounced ecosystem shifts at Olduvai Gorge contrasts with previous reconstructions for eastern Africa that proposed that ecosystems were stable at local to regional scales in the early Pleistocene (15, 45, 46) and generally lacked closed woodlands near hominin archaeological sites since 6 Ma (7). Dramatic and rapid changes in δ¹³C₃₁ values highlight ecosystem instability in this region; furthermore, δ¹³C₃₁ values indicate that closed woodlands dominated local landscapes for up to 20% of the time.

Differences in the interpretation of ancient ecosystems, in addition to problems related to savanna heterogeneity, can stem from inherent proxy biases (6). For instance, δ¹³C_SOM values can overrepresent C₄ inputs as a result of ¹³C enrichment during organic matter decomposition (47). Values for δ¹³C₃₁ can also overrepresent C₄ inputs as a result of inorganic carbon (e.g., bicarbonate) assimilation by macrophytes, although δ¹³C₃₁ values in arid environments are more likely biased to wet conditions (when plants synthesize most leaf lipids) (48). Although specific mechanisms responsible for differences among ecosystem proxies cannot necessarily be reconciled here, we suggest that carbon-isotopic signals for leaf lipids can complement δ¹³C_SOM values, which otherwise can skew to C₄—and thus, arid—signals on seasonal and longer timescales.

Biogeochemical Variability at the Ecosystem Scale.

Organic matter in lake sediments incorporates inputs from both aquatic and terrestrial photosynthetic organisms and can vary in proportion with productivity and deposition in a lake and the surrounding watershed. At Olduvai Gorge, δ¹³C_TOC values correlate strongly with TOC (%TOC; r² = 0.84) but show no clear relationship with algal vs. terrestrial plant inputs (Fig. 4):

The abundance ratio of pristane vs. phytane is constantly near 2.5, suggesting that organic carbon deposition was dominated by terrestrial plant input (31). These observations corroborate strong correlation between δ¹³C_TOC and δ¹³C₃₁ values and suggest that δ¹³C_TOC values primarily follow terrestrial plant inputs.

Macrophytes contain abundant intermediate-chain n-alkanes (e.g., nC₂₃ and nC₂₅) but limited long-chain homologs compared with most terrestrial plants (49, 50). Thus, relative abundances of nC₂₃ and nC₂₅ vs. nC₂₉ and nC₃₁ (that is, P_aq) can provide estimates for macrophytic vs. terrestrial plant inputs (50):

In lake sediments from Olduvai Gorge, δ¹³C₃₁ values correlate weakly with both P_aq (r² = 0.17) and nC₂₅ δ¹³C values (r² = 0.19), suggesting that macrophytes did not significantly contribute to nC₃₁ inputs. P_aq and P_alg are measures of relative abundance of molecules. They do not directly indicate a proportion of biomass from different organisms; instead, these ratios are useful for identifying organic facies or correlations with other biogeochemical signals.

Mechanisms of Ecosystem Change.

Atmospheric CO₂ concentrations (pCO₂), temperature, seasonality, and water availability are potential determinants of C₃ vs. C₄ plant abundance (38). Since the middle Pleistocene, records of pCO₂ correlate strongly with polar ice volume changes (51, 52), which were obliquity paced before ∼1 Ma (5). Lake sediment δ¹³C_TOC values for Olduvai Gorge correlate weakly with reconstructed polar ice volumes (r² = 0.16) based on marine oxygen-isotopic records (9, 19), suggesting that ecosystem changes in this region did not track 41,000-y glacial cycles. If polar ice volume is a representative proxy for pCO₂ during the early Pleistocene, then local ecosystem changes were not exclusively tied to pCO₂ changes. This conclusion contrasts with suggestions of a dominant role for pCO₂ in southern African ecosystems during the early Pleistocene based on speleothem carbonate δ¹³C values (53), but it is in agreement with marine oxygen-isotopic evidence for eastern African climate sensitivity to polar ice volume only after 1.0 Ma (54). Values for δ¹³C_TOC correlate strongly with precession (ω_p) and thus, do not support temperature as primary determinant of ecosystem change, because ω_p negligibly influences mean annual temperatures (55). Similarly, paleosol carbonates indicate that mean annual temperatures varied by less than about ±5 °C at and around Olduvai Gorge during the early Pleistocene (56, 57). Strong correlation between δ¹³C_TOC and ω_p also suggests that biotic (e.g., herbivory) or abiotic disturbances, such as fire, were not primary determinants of ecosystem change, although they may have served as feedback mechanisms that accelerated changes. In agreement with a variety of other studies (58, 59), we suggest that changing C₃ and C₄ plant abundances at Olduvai Gorge varied with orbital precession in response to water availability.

Mechanisms of Hydroclimate Change.

Cycles of about 21,000 y are common in a variety of hydroclimate proxy records in eastern Africa since the Pliocene (5), and δ¹³C_TOC values correlate strongly with ω_p (r² = 0.61) in single-factor regression (SI Appendix). Although specific mechanisms responsible for these cycles remain unclear, the timing and magnitude of local and regional hydroclimate changes are consistent with theoretical effects of ω_p on monsoon strength (60) (that is, higher summer insolation would enhance land–ocean temperatures contrasts, resulting in stronger monsoons and increased precipitation).

Insolation alone cannot account for the magnitude of hydroclimate change in eastern Africa (61). Previous reconstructions based on pollen and oxygen-isotope compositions of soil carbonates suggest that MAP fluctuated between ∼400 and 800 mm at Olduvai Gorge and surrounding regions during the early Pleistocene (15–17, 43). However, in climate simulations, insolation variability accounts for MAP fluctuations of less than 200 mm and mostly affects long rains (60, 61). Thus, precipitation amounts in the past must have been impacted by multiple factors, the same as they are today (13).

Today, precipitation in eastern Africa responds sensitively to SSTs in the Indian Ocean and Atlantic Ocean (62). In particular, intensifications of short rains (up to 200 mm) accompany coordinated warm and cold SST anomalies in the western Indian Ocean and eastern Atlantic Ocean (63–65), respectively, as a result of transcontinental surface pressure gradients across Africa and monsoon displacement of the IOC from west to east (66). Partial regressions reveal that SSTs for ODP sites 662 (SST₆₆₂) and 722 (SST₇₂₂) are significant (P < 0.001) secondary predictors that are statistically independent of covariance with ω_p, and the combination of ω_p, SST₆₆₂, and SST₇₂₂ explains 73% of the variability in δ¹³C_TOC values in a multiple regression model. During the early Pleistocene, both SST₆₆₂ and SST₇₂₂ show strong ω_p and 41,000-y (obliquity) periodicity (20), but only SST₆₆₂ shows a consistent relationship with monsoon-driven upwelling (20). Because upwelling in the eastern Atlantic correlates positively with monsoon strength during late boreal summer (20–22), we suggest that obliquity-paced cooling in the eastern Atlantic Ocean and monsoon strengthening (and therefore, stronger westerly winds) resulted in more frequent eastward displacements of the IOC and intensification of short rains in eastern Africa.

Ecosystems and Hominin Evolution.

Fossil evidence for pronounced aridification and faunal turnover in eastern Africa between about 2.0 and 1.8 Ma has sparked hypotheses linking the emergence and dispersal of the genus Homo to climate-driven ecosystem change (2–6). Fossil evidence for cranial expansion in premodern Homo (e.g., H. erectus sensu lato) has been linked to irregular resource distributions (67), and our carbon-isotopic data are consistent with enhanced ecosystem variability as a context for encephalization (Fig. 5). During the early Pleistocene, strong ecosystem preferences are not apparent between transitional (e.g., H. habilis) and archaic (e.g., Paranthropus boisei) hominins (68); however, isotopic and fossil data suggest that transitional species accessed a broad spectrum of dietary resources compared with archaic species (68–70). Among primates, quality (i.e., energy density) of dietary resources correlates strongly with brain size (67). Assuming that dietary resources were primarily unrelated to technological innovations by transitional species (4, 68), we hypothesize that ecosystem variability favored hominin species with large brains that allowed for versatile foraging strategies and dietary diversity.

Conclusions

This study presents high-resolution biomarker and δ¹³C records of ecosystem variability from lake sediments at Olduvai Gorge that were deposited during an interval of pronounced shifts in vertebrate community and global climate reorganization about 2.0–1.8 Ma. Values of δ¹³C₃₁ indicate rapid and repeated ecosystem restructuring between closed C₃ woodlands and open C₄-dominated grasslands. Our δ¹³C records reveal coupled fluctuations between ecosystem and precession. Additional variability is explained by differences in SST between the Atlantic and Indian Oceans. These observations suggest aridity-controlled, as opposed to carbon dioxide- or temperature-controlled, woody plant cover in eastern Africa during the early Pleistocene. We conclude that highly variable ecosystems accompanied the emergence and dispersal of the genus Homo. Our study also builds on soil data to construct an interpretive framework for ecosystem reconstruction based on leaf lipids.

Materials and Methods

Reconstructions for Polar Ice Volume.

We use the approach in the work by Bintanja and van de Wal (71), which assumes that a global composite record of benthic δ¹⁸O values (LR04 Stack) accurately traces polar ice volume during the early Pleistocene (19).

Multiple Regression Analysis.

We used rank transformation (72) and Fourier cross-correlation (73) to compare nonlinear, time-shifted, and unevenly sampled proxy records (SI Appendix). Because ω_p correlates most strongly with δ¹³C_TOC in single-factor analyses (r² = 0.61), we constrain multiple regression models for δ¹³C_TOC values to always include ω_p. This approach is justified, because the influence of ω_p on climate in eastern Africa is well-supported by theory and other observations (10, 61). Because ω_p is correlated with other factors, care is required in evaluating additional influences on δ¹³C_TOC values. We assessed the influence of other factors on δ¹³C_TOC values using multivariate partial regression models that account for covariance between ω_p and other factors. Partial regression models with SST₆₆₂ (partial r² = 0.17) and SST₇₂₂ (partial r² = 0.11) as secondary predictors are the only models with notable explanatory power for δ¹³C_TOC values. The following equation shows the δ¹³C_TOC relationship shared with ω_p, SST₆₆₂, and SST₇₂₂ (r² = 0.73):

Lipid Extraction and Purification.

Freeze-dried and ground lake sediments were Soxhlet-extracted with dichloromethane:methanol (9:1 vol/vol) for 12 h. Total lipid extracts were then separated into apolar and polar fractions over alumina with hexanes and methanol, respectively. Apolar molecules were separated into saturated and unsaturated fractions over 5% (wt/wt) silver nitrate-impregnated alumina with hexanes and dichloromethane, respectively. Finally, unsaturated apolar compounds were separated with a 5-Å molecular sieve to isolate n-alkanes.

Isotopic Analysis.

All δ¹³C_TOC values were measured after decarbonation of ground lake sediments with excess 2N hydrochloric acid (74). Residual materials were combusted in an elemental analyzer, and δ¹³C_TOC values were measured in a ThermoFinnigan Delta+ XP. Standard reference materials of known δ¹³C values, including polyethylene foil (NIST 8540), were used throughout sample runs to ensure accuracy. SD (1σ) for NIST 8540 equaled 0.1‰.

Compound-specific δ¹³C values were measured by GC–isotope ratio monitoring–MS (Delta+ XP; ThermoFinnigan) with a combustion interface. Carbon dioxide gas of known δ¹³C value was used as an external standard; internal standards (i.e., nC₄₁, androstane, and squalane; Schimmelmann Standards) were used throughout sample runs to ensure accuracy. SD (1σ) equaled 0.3‰.