Neogene biomarker record of vegetation change in eastern Africa

The δ¹³C_C31 values from 24 Ma to 10 Ma come from the Site 235 and 241 cores in the Somali Basin and are very depleted and nearly invariant, ranging from −33.9‰ to −32.2‰ (Fig. 2), indicative of C₃ ecosystems. The range of δ¹³C_C35 values before 10 Ma is similarly small, from −33.3‰ to −31.3‰ (Fig. S4 and Dataset S5). The other n-alkane homologs, C₂₉ and C₃₃, have a pooled range of −34.0‰ to −30.8‰. Given the similarity in δ¹³C values of the four homologs through time and the relative geographic proximity of Sites 235 and 241, we combine data from both sites into a single Somali Basin record (Fig. S4). Combining the two sites into a single record is further supported by clay mineralogy of the Somali Basin sediments, which indicates a similar sediment provenance of the Horn of Africa and possibly Central Africa for both sites (67). All values from Somali Basin samples are indicative of C₃ ecosystems in eastern Africa before 10 Ma. The limited range within each homolog from 20 Ma to 10 Ma suggests that the δ¹³C values likely define endmember values for C₃ ecosystems in eastern Africa. The persistently low δ¹³C values of the n-C₃₅ homolog are particularly diagnostic, indicating no isotopic evidence for C₄ vegetation in east Africa from 24 Ma to 10 Ma.

The average molecular distribution of n-alkanes for pre-10-Ma samples has a n-C₃₁ maximum, followed in order of abundance by n-C₂₉, n-C₃₃, n-C₂₇, and n-C₃₅ (Fig. S5). Although the data do not preclude minor amounts of rainforest vegetation, the n-alkane distribution does not support widespread rainforests in eastern Africa, nor do measured δ¹³C_C31 values, which do not reach the low values characteristic of rainforest vegetation (see figure 6a in ref. 64 for reference n-alkane distributions). Rather, the molecular distributions and δ¹³C_C31 values reflect a combination of nonrainforest trees, shrubs, herbs, and C₃ grasses. Relative contributions of these C₃ plant types cannot be ascertained from the molecular distribution data. Two possible exceptions to this are the molecular distributions from samples from 19.18 Ma (Site 241, sample KU357) and 1.15 Ma (Site 228, sample KU315), which have distributions similar to modern rainforest plants (Fig. S6). However, the δ¹³C values are −32.4‰ and −25.2‰, respectively, neither of which reflect modern rainforest plants values of −35.6 ± 2.8‰ (corrected to preindustrial atmospheric δ¹³C values) (64).

The absence of C₄ vegetation at the regional scale before 10 Ma based on n-alkane carbon isotope and molecular distribution data is supported by vegetation reconstructions from paleobotanical records of predominantly forest or wooded ecosystems in Ethiopia, Kenya, and Uganda (15–17, 19, 27).

Notable exceptions from two sites in Kenya suggest minor amounts of C₄ grasses may have been present locally before 10 Ma (52, 68, 69). At these sites, isotopic evidence comes from pedogenic carbonate δ¹³C (δ¹³C_PC) values, where C₃-dominated ecosystems have δ¹³C_PC values ranging from −15‰ to −8‰, mixed C₃−C₄ ecosystems range from −8‰ to −2‰, and C₄-dominated ecosystems range from −2‰ to +4‰, after correction to preindustrial atmospheric δ¹³C values.

A single pedogenic carbonate value from the Hiwegi Formation (17.8 Ma) on Rusinga Island in Lake Victoria has a δ¹³C_PC value of −5.7‰ (68). However, the corresponding δ¹³C value from coexisting soil organic matter is −24.3‰, consistent with water-stressed C₃ vegetation or minor amounts (ca. 15%) of C₄ vegetation. There is an enrichment in ¹³C between soil organic matter and pedogenic carbonate that is a function of kinetic processes (diffusion of CO₂ in soils) and equilibrium isotopic exchange during carbonate precipitation that results in a theoretical enrichment of +14‰ to +17‰ (35). The observed enrichment factor of +18.6‰ for the Hiwegi Formation sample falls slightly outside this theoretical value. The other δ¹³C_PC values that suggest C₄ vegetation existed before 10 Ma come from the Tugen Hills (Baringo Basin) in the central Kenyan Rift Valley (52, 69). Five pedogenic carbonates from the ∼15.5-Ma sediments in the Muruyur Formation have δ¹³C_PC values ranging from −7.4‰ to −5.2‰. Only one of the five samples (MU 453, δ¹³C_PC = −7.4‰) has a corresponding organic matter value with an enrichment value within the theoretical range of +14–17‰, making it the most reliable value from this period (70). Two δ¹³C_PC values around −6.5‰ from the ∼10.5-Ma section of the Ngorora Formation, also in the Tugen Hills, indicate minor amounts of C₄ vegetation. A corresponding organic matter δ¹³C value from one of the samples is −24.3‰, which falls within the range of water-stressed C₃ vegetation. Even if all pre-10-Ma pedogenic carbonate values >−8‰ are accepted and considered with other paleovegetation data from Rusinga and the Tugen Hills, the picture that emerges from both sites can be summarized as wooded to forested environments with minor amounts of C₄ vegetation present. The early presence of minor amounts of C₄ vegetation in eastern African is entirely consistent with an emergence of the C₄ photosynthetic pathway in the late Eocene or early Oligocene (71, 72). Despite an early emergence, C₄ grasses remained minor constituents of global ecosystems until their late Miocene expansion (73).

Fort Ternan (14 Ma) is the final middle Miocene site that warrants discussion, because well-preserved grass macrofossils were found there (18) and 54% of total pollen comes from grasses (27). Pedogenic carbonate values range from −14.9‰ to −9.8‰, leading to the conclusion that C₃ grasses were present at the site (74). Fort Ternan illustrates that, without plant fossils, whether from pollen, phytoliths, or macrofossils, C₃ grasses are invisible when viewed solely through an isotopic lens. The well-documented lack of pollen and the rarity of plant macrofossils in the Neogene of eastern Africa demonstrate the need to find other proxies that will enable us to investigate the Cenozoic history of grasses. Phytoliths could play a pivotal role; they have proven to be robust indicators of grasslands in the Cenozoic in other parts of the world (75). Other proxies for detecting C₃ grasses, where stable isotope methods are silent on the matter, include grass cuticles (76) or, possibly, n-alkane distributions (64, 65).

The onset of C₄ grassland expansion in eastern Africa occurs at ∼10 Ma. It is marked by a sharp 3.1‰ increase in the δ¹³C value of the n-C₃₅ alkane that occurs between samples dated at 10.08 Ma and 9.65 Ma (Fig. 3). This is the first occurrence, to our knowledge, of a δ¹³C value greater than −30‰ in the n-alkane record. Enrichment in ¹³C occurs across all other alkane homologs with deceasing magnitude from C₃₃ to C₂₉ (Fig. S4), which follows the predicted pattern based on chemotaxonomic differences in n-alkane production observed in modern C₃ and C₄ plants in Africa (63) and unequivocally dates the onset of expansion of C₄ grasses in eastern Africa.

A remarkable consequence of the appearance of C₄ grass in eastern Africa is the immediate and, in some cases, dramatic dietary response of large herbivores (36). The n-alkane δ¹³C values reveal a synchroneity between the onset of expansion of C₄ grasses and dietary shifts in proboscideans (Fig. 3) and other large herbivore lineages (36). Most striking is the shift from C₃-dominated to C₄-grazing diets in equids documented in Kenya between 9.9 Ma and 9.3 Ma. Other large mammals that followed suit by 9 Ma include bovids and rhinocerotids. Equids only arrived in Africa around 10.5 Ma, so there is no dietary record in eastern Africa for them in the earlier Neogene (59). Proboscideans, which include modern elephants and their ancestors, are an Afrotherian order for which dietary records exist from the early Miocene to present. Fig. 3 illustrates the dietary evolution of proboscideans from eastern and central (i.e., Chad) Africa over the past 17 My based upon carbon isotope data from their teeth (Dataset S7, compiled primarily from ref. 77). Small amounts of C₄ vegetation are present in proboscidean diets beginning at 9.9 Ma. The dominant early and middle Miocene proboscideans, Deinotheriidae and Gomphotheriidae, had brachydont molars with transverse lophs for shearing vegetation (78, 79). Although the inclusion of some C₃ grass in their diets cannot be ruled out, their craniodental morphology indicates they were browsers. If C₃ grasses were abundant at this time, proboscideans did not capitalize on this dietary resource and instead waited until the appearance of C₄ grasses at ∼10 Ma to begin grazing. In fact, as one of the dominant early Neogene herbivores, it is possible that browsing proboscideans maintained or even increased grasslands, as their extant browsing relatives, Loxodonta africana, do today (80).

The earliest purported hominins, Sahelanthropus tchadensis and Orrorin tugenensis, appeared in the late Miocene (7–6 Ma) (Fig. 3D). By that time (6.5 Ma), the Somali Basin n-alkane data show an increase to 19% C₄ vegetation on the landscape, or a 2.6‰ increase in δ¹³C_C31 values above average early Neogene C₃ values based on the linear regression equation for all data (Fig. S7C). An even greater increase of 4.6‰ occurs in δ¹³C_C35 values by this time. These two early possible hominins emerged amid a period of major change in faunal communities and diets driven by changes in ecosystem structure associated with the expansion of C₄ grasses. The dietary shift toward C₄ grazing was well underway in many mammalian lineages (36, 43), but there are not yet any isotopic analyses of teeth to ascertain the diets of these early hominids.

However, extensive isotopic data from Plio-Pleistocene hominin teeth provide a clear picture of diets in different lineages, temporal trends, and dietary niche partitioning between Pleistocene genera (91–96). Two early species (4.4–4 Ma), Ardipithecus ramidus and Australopithecus anamensis, are anomalous for their C₃-dominated diets (−11.6‰ to −8.5‰). After 3.8 Ma, Australopithecines and Kenyanthropus platyops show a dramatic increase in isotopic range and, presumably, dietary breadth, with δ¹³C values ranging from −13‰ to −2.7‰. Australopithecus bahrelghazali from Toros-Menalla (Chad, central Africa) shows the earliest evidence, to date (3.6 Ma), of a hominin with a consistent, mostly C₄-based diet (range −4.4‰ to −0.8‰). It is possible that the dietary evolution of central African hominins differed from those in eastern Africa, as is the case with the southern African hominins (97). In eastern Africa, the diet of Paranthropus evolves to become dependent on C₄-based foods from 2.7 Ma to 1.5 Ma (92). Overall, the increase in C₄-based foods in eastern African hominin diets mirrors the increasing prevalence of C₄ biomass on the landscape. Our genus, Homo, does not follow this trend.

The isotopic range of Homo diet remains variable and broad (−9.9‰ to −2.2‰, mean = 6.1‰), indicating dietary flexibility from ∼2.4 Ma to 1.3 Ma (95). Microwear studies suggest dietary differences between Homo species based on tooth wear, supporting the broad isotopic range (98). An important observation, then, is that Homo diet remained flexible during a time of rising environmental inflexibility imposed by increasing aridity and C₄ vegetation on the landscape. One critical element of hominin diet that remains unknown, and is especially important for Homo, is when meat became an important food source. Carbon isotopes do not provide information on trophic level; thus isotope-based reconstructions of hominin diet reflect some combination of plant foods and of higher trophic level foods such as insects, mammals, or aquatic fauna.

The plant wax data from the Somali Basin, along with pedogenic carbonate and Gulf of Aden (Site 231) plant wax data, show that continued increases in the proportion of C₄ grasses on the landscape ultimately influenced hominin diet, possibly from the inception of the lineage. The shift to more open environments presented hominins with new challenges as to how they acquired their food, avoided predators, and managed higher heat loads. The challenges posed by the changing environmental conditions were met with changes to diet (C₄-based foods and, eventually, meat), the innovation of tools, and, ultimately, increased social complexity such as the cooperative acquisition of resources through foraging, scavenging, or hunting.

The Neogene plant wax record provides the opportunity to place the expansion of C₄ grasslands in eastern Africa into a global context. C₄ grassland expansion has been investigated most intensively in the Siwaliks of South Asia (48, 51, 53), the Americas (49, 54, 99–101), southern Africa (102, 103), and, to a lesser degree, China (104, 105), the Mediterranean (106), and the Middle East (85, 86). The onset of expansion at 10 Ma in eastern Africa precedes by at least 2 My that of well-dated terrestrial records from the Siwaliks (∼8 Ma), North and South America (∼8–6 Ma), and South Africa (∼8–7 Ma).

An important feature of the eastern Africa biomarker record is that it captures the transition from isotopically invariant C₃-dominated ecosystems during the first part of the Neogene to the period of dynamic change after 10 Ma (Fig. 2). The expansion of C₄ grasslands in eastern Africa differs markedly from other regions around the world in its earlier onset and the monotonic increase in C₄ grasses at the regional scale. The mode of expansion differs from the step-like change that occurred in the Siwaliks, arguably the most well documented case of grassland expansion (51). It also differs from the Great Plains of North America, which shows possible minor amounts of C₄ vegetation as early as ∼23 Ma but an expansion beginning around 7 Ma, based on the pedogenic carbonate record (54). A North American vegetation reconstruction from n-C₃₁ alkane data from a Gulf of Mexico core (DSDP Site 94) also suggests minor amounts of C4 vegetation by 9.1 Ma or possibly earlier (49). Two marine sediments cores (Ocean Drilling Program Sites 1081 and 1085) spanning ∼10 degrees of latitude from southern Africa both indicate the onset of expansion at ∼7.5 Ma based on n-C₃₁ alkane data, lagging eastern Africa by ∼2.5 My (102, 103). A Neogene vegetation record from western Africa could provide information on mechanisms behind the expansion in Africa because it would show whether or not expansion occurred synchronously across eastern and western Africa. Clearly, there is the need for continued efforts to produce well-dated records of C₄ expansion to identify additional regional, continental, or global patterns. This includes characterizing the vegetation and climatic conditions before the expansion, which were not uniform globally. Improved spatial and chronological resolution will greatly enhance our understanding of the driving mechanisms of grassland expansion.

Commonly invoked causes for C₄ grassland expansion include a late Miocene decrease in atmospheric CO₂ concentrations (pCO₂) or increases in aridity, seasonality of precipitation, fire, and herbivory (56, 107–109). These factors are not mutually exclusive; therefore, debate continues about the relative importance of each variable (110). Most are difficult to quantify in the geologic record, and all continue to benefit from new or improved proxy methods for reconstruction in deep time. This includes characterizing fire frequency and intensity, identifying C₃ grass abundance, quantifying seasonality and aridity, and refining pCO₂ estimates.

Age−depth models were developed for the four DSDP cores using calcareous nannofossil occurrence data (117, 118) and volcanic ashes (60) documented in the cores. We used updated calcareous nannofossil biozonations for the Neogene and the ages assigned therein, which primarily derive from astronomically tuned cyclostratigraphy (61, 62). Biozonations come from top (T) or bottom (B) horizons in the cores. Miocene biozones have an average age uncertainty of ± 0.02 My, whereas, for the Plio-Pleistocene, it is ± 0.007 My. Data are presented in Dataset S1 and Fig. S1.

Marine core sediment samples were taken by the curatorial staff at the International Ocean Discovery Program’s Kochi Core Center in Shikoku, Japan. Samples were taken from undisturbed sections of the core where possible. Site and core information are provided in Dataset S3, along with sample locations, calculated depths, and lithological descriptions. Lithological descriptions are condensed from Initial Reports of the DSDP (82, 119, 120). Sample ages were calculated using sample depth to linearly interpolate an age based on adjacent biohorizons in the age−depth models developed at each site (Dataset S1 and Fig. S1). The minimum age uncertainty for each sample is equal to the uncertainty of the associated biohorizon, with uncertainty increasing with sample distance from a biohorizon. The maximum age uncertainty can be estimated as half of the total time represented in the biozone to which the sample belongs. Interpolated ages are presented in Dataset S1.

The outer surface of sediment samples was cleaned with a Dremel to remove possible modern contamination, including drilling mud, which was visually easy to differentiate from sediment, based on color. Samples were lyophilized overnight (−70 °C, ∼10 millitorr) to remove water from sediments. Samples were then solvent-rinsed with dichloromethane (DCM), placed in a mortar and pestle, and crushed to a powder. Between samples, the mortars and pestles were washed, rinsed with deionized water, and solvent-rinsed with methanol and DCM.

Lipids were extracted from 9.1 g to 57.9 g (median: 35.1 g) of powdered samples with organic solvents (9:1 DCM:methanol) using a Dionex Accelerated Solvent Extractor (ASE) 350 packed into 60-mL extraction cells. We found that samples required four 10-min static cycles at 100 °C with a flush volume of 150% to completely extract soluble lipids. An internal standard was added to the ASE extract (2,000 ng of 5α-androstane) for later quantification of lipids.

The total lipid extract (TLE) was transferred to 4-mL vials, and four pilot samples were separated by solid-phase extraction on silica gel columns (∼0.5 g solvent-rinsed silica gel, 230–400 mesh) and analyzed on a gas chromatograph (GC; details in the following section, Biomarker GC Mass Spectrometry and Isotopic Analyses) to check for the presence of sulfur and unsaturated compounds in the aliphatic fraction, which contained the n-alkanes. Sulfur is common in anoxic sediments and interferes with compounds during measurement on a GC. Both sulfur and minor amounts of unsaturated compounds were present in the aliphatic fraction, necessitating additional sample purification detailed in the following two paragraphs.

TLEs were separated by solid-phase extraction on silica gel columns. The aliphatic fraction (F1) was eluted with 4 mL of hexane, the ketone and ester fraction (F2) was eluted with 4 mL of DCM, and the polar fraction (F3) was eluted with 4 mL of methanol.

To remove sulfur, TLEs were treated with copper. Oxidation of copper to copper sulfide removes sulfur from the sample. DCM- and hexane-rinsed copper wool was added to each F1. Samples were then placed in a thermowell at 50 °C and reacted overnight. The following day, the copper-treated F1 (F1Cu) fractions were loaded directly onto silver ion-impregnated silica get columns (Ag-ion Si gel) in hexane to separate saturated and unsaturated aliphatic compounds. The saturated fraction (F1Cu_S) was eluted in a total of 4 mL hexane, followed by elution of the unsaturated fraction (F1Cu_US) with 4 mL ethyl acetate. Ag-ion Si gel column chromatography was done under low light conditions (columns were shrouded in aluminum foil in a hood with room lights off) to avoid oxidative conditions in the columns.

The n-alkanes in the saturated, aliphatic fraction (F1Cu_S) were quantified and characterized on an Agilent GC (Agilent 7890A GC) equipped with both a mass selective detector (5975C MSD) and flame ionization detector (FID). One microliter of sample dissolved in 100 μL hexane was injected into a multimode inlet injector at 60 °C (0.1 min hold) which was then ramped to 320 °C at 900 °C/min and held for the duration of the analysis. Initial GC oven temperature was set at 60 °C (1.5 min hold) and ramped to 150 °C at 15 °C/min, then to 320 °C at 4 °C/min. A helium-purged microfluidics device downstream of a DB-5 column (30 m length, 250 μm i.d.) quantitatively split the GC flow to the MSD and FID detectors. Compound identification was done with comparison of mass spectra and retention times to authentic standards, and n-alkane quantification was done by peak integration on the mass 57 ion from the MSD. The n-alkane concentrations were calculated based upon the peak area of internal standard of known concentration added to the TLE. We correct for mass-dependent changes in ionization efficiency (or response) using the standard mixture of C₈−C₄₀ n-alkanes. The response factor correction results in more accurate concentrations of long-chain alkanes.

Carbon isotope ratios of n-alkanes were analyzed using a Trace Ultra GC coupled to a Thermo Delta V isotope ratio mass spectrometer through a combustion interface (Thermo IsoLink) at the Lamont-Doherty Earth Observatory Stable Isotope Lab. All sample injections (1–4 μL) were interspersed with injections of molecular mixtures with known isotopic values (Mixes A4, A5, and F8 supplied by Arndt Schimmelmann, Indiana University, Bloomington, IN) that were used for correction of carbon isotope values. The carbon isotope ratio is expressed using delta notation, where δ¹³C = (R_sample/R_standard – 1) and R= ¹³C/¹²C. We report the δ¹³C values of odd-numbered n-alkanes C₂₇ to C₃₅, where uncertainties for these values incorporate both analytical uncertainty and the uncertainty in converting from the laboratory reference gas scale to the Vienna Pee Dee Belemnite (VPDB) scale (Dataset S5) (111).

Terrestrial plant wax molecular distributions are commonly characterized by two metrics, the abundance weighted average chain length (ACL) and carbon preference index (CPI). ACL is calculated as ${\displaystyle \sum }_{\mathit{i}=\mathit{a}}^{\mathit{b}}i{\mathrm{C}}_{\mathit{i}}/{\displaystyle \sum }_{\mathit{i}=\mathit{a}}^{\mathit{b}}{\mathrm{C}}_{\mathit{i}}$, where ${\mathrm{C}}_{\mathit{i}}$ is the concentration of the molecule with chain length $i$ ($i$= 29, 31, 33, 35). The range of ACLs is 29.6–31.5 (Dataset S4).

n-alkanes show a characteristic odd-over-even preference resulting from their biosynthetic pathway. This preference is formalized and calculated, using the CPI (121), as $\left[{\mathrm{C}}_{27}+2\left({\mathrm{C}}_{29}+{\mathrm{C}}_{31}+{\mathrm{C}}_{33}\right)+{\mathrm{C}}_{35}\right]/2\left({\mathrm{C}}_{28}+{\mathrm{C}}_{30}+{\mathrm{C}}_{32}+{\mathrm{C}}_{34}\right)$ for n-alkanes. CPI values in plants have a wide range (∼0.04–99), although most yield CPIs > 2 (122). The range of CPIs in these samples was 3.5–11.0 (Dataset S4).