Contracting eastern African C4 grasslands during the extinction of Paranthropus boisei

The extinction of the Paranthropus boisei estimated to just before 1 Ma occurred when C4 grasslands dominated landscapes of the Eastern African Rift System (EARS). P. boisei has been characterized as an herbivorous C4 specialist, and paradoxically, its demise coincided with habitats favorable to its dietary ecology. Here we report new pedogenic carbonate stable carbon (δ13CPC) and oxygen (δ18OPC) values (nodules = 53, analyses = 95) from an under-sampled interval (1.4–0.7 Ma) in the Turkana Basin (Kenya), one of the most fossiliferous locales of P. boisei. We combined our new results with published δ13CPC values from the EARS dated to 3–0 Ma, conducted time-series analysis of woody cover (ƒWC), and compared the EARS ƒWC trends to regional and global paleo-environmental and -climatic datasets. Our results demonstrate that the long-term rise of C4 grasslands was punctuated by a transient but significant increase in C3 vegetation and warmer temperatures, coincident with the Mid-Pleistocene Transition (1.3–0.7 Ma) and implicating a short-term rise in pCO2. The contraction of C4 grasslands escalated dietary competition amongst the abundant C4-feeders, likely influencing P. boisei’s demise.

Pedogenic carbonates at depths greater than 30 cm in soils with relatively high respiration rates incorporate CO 2 of decaying organic matter derived from surface vegetation during soil development (5)(6)(7)(8)(9). Vegetation-derived CO 2 δ 13 C values are incorporated into soil organic matter (subscript SOM) and pedogenic carbonate nodules (subscript PC) preserved in paleosols, which are used to quantitatively measure the relative amounts of C 3 and C 4 biomass on the land surface during soil carbonate formation (10). Surface biomass CO 2 δ 13 C values are generally comparable to δ 13 C SOM values (11), whereas δ 13 C PC values are enriched by +14-17‰ relative to biomass-derived CO 2 (9). Cerling and others (12) derived a regression equation to calculate the fraction of woody canopy cover (ƒ WC ) based on measured δ 13 C SOM values of modern African and Australian soils, for categorizing habitat structure based on UNESCO classifications of African vegetation. Vegetation structures have been reconstructed from the many δ 13 C PC -based studies of eastern African Plio-Pleistocene fossil hominin sites (12).
Stable oxygen isotopic values of pedogenic carbonates (δ 18 O PC ) are controlled by soil pore water δ 18 O values, temperature-dependent isotopic fractionation during carbonate formation, evaporation rates and soil moisture (5,12). At depths greater than 30 cm, soil pore water δ 18 O values approximate those of meteoric water (5,9,14). Soil pore water δ 18 O values can change from expected meteoric water δ 18 O values due to evaporation and season of carbonate formation (9,15). δ 18 O PC values are widely used to estimate rainfall source, soil moisture, and temperature in continental environments (16). However, Fox and others (17) caution that when utilizing δ 18 O PC to detect changes in temperature and aridity over geologic time it is impossible to deconvolve evaporative effects of soil water δ 18 O and soil temperature changes without knowing either meteoric water δ 18 O values or air/soil temperatures in the past. Moreover, changes in source rains can produce δ 18 O PC variations when temperature and rainfall amounts are constant (16)(17).
There are several limitations with δ 13 C PC and δ 18 O PC methods for inferring vegetation structure and climatic conditions. Obviously pedogenic carbonate isotopic analysis is limited to characterizing carbonate-bearing paleosols, which form with negative water budgets (5,7,18) and thus typically characterize environments that are relatively dry, at least seasonally. Fossil soils preserved in the Plio-Pleistocene Turkana Basin are dominated by paleo-Vertisols, formed under a dry season of four or more months and 250-1000 mm of annual moisture (19). The formation of pedogenic carbonate nodules is not common along perennial river channels due to ample water availability; moreover, soils tend to occur on stable land surfaces (18,(20)(21). Pedogenic nodules primarily form in the warm season and differentially record times of soil dewatering (23)(24) and therefore may underrepresent conditions during cool and wet seasons (25). Pedogenic nodule formation times also vary, averaging 10 1 -10 3 of years (22) and therefore potentially dampen extreme variations in vegetation structure with changing paleohydrological and paleogeography (e.g., river channels migrate laterally across the landscape and/or lake margins transgress and regress) as well as with seasonal changes. Pedogenic carbonate δ 13 C PC reflect the vegetation and soil water conditions of less than a square meter during nodule formation (5,26). Sampling paleosols laterally yields variability in δ 13 C PC values (27)(28)(29)(30) as would be expected in African savanna ecosystems (19,27) with transitioning vegetation structures influenced by the local hydrological configuration, geography, and geomorphology as well as regional and global factors such as pCO 2 , rainfall and temperature. Sampling a small spatial scale may over-represent microenvironments due to limited paleosol preservation.

B. Field sampling and age control
At the outcrop (Supplementary Figures S1-S4), we identified pedogenic carbonate nodules within paleosols by criteria of Retallack (31) and applied to the Turkana Basin by Wynn (19,27). We sampled pedogenic carbonates in the preserved calcic horizon of the paleosol at a minimum of 30 cm below the contact with the overlying stratum (7) and excavated back from the vertically exposed surface by approximately 50 cm. Calcite nodules were extracted from within individual peds. Since most Turkana fossil soils are paleo-Vertisols, they show vertic features and slickensided surfaces; we chose calcite nodules that exhibited slickensides and/or were adjacent to slickenslided surfaces. Although abundant in the formation deposits, we did not include calcareous rhizoliths in this study due to isotopic alteration by shallow cementation (32) and groundwater (33).
Age control of pedogenic carbonate samples was determined with the established chronostratigraphic framework and scaled with linear sedimentation relative to dated tuffs (34)(35). The Brunhes-Matuyama Boundary occurs within the Chari Member approximately 5 m below the stratigraphic level of the Silbo Tuff (36) and was further used to scale collected samples relative to the ~500-kyr depositional hiatus (37). We utilized the date of 0.77 Ma for the Brunhes-Matuyama Boundary (38).

C. Data compilations and treatments
We combined our new δ 13 C PC and δ 18 O PC values from Nariokotome and Ileret to those reported in the Turkana Basin from the Nachukui and Koobi Fora Formations, respectively, from 3-0 Ma (data from 19,[27][28][29][30][40][41]. Utilizing methods of Cerling and others (12), we subtracted 14‰ from the δ 13 C PC values to convert to the isotopic equivalent of organic carbon (δ 13 C om ) and used the equation: ƒ WC = {sin[-1.06688 -0.08538(δ 13 C om )]} 2 to generate estimates of fraction woody canopy cover for classification into UNESCO categories of African vegetation. These categories were taken from White (42) and have the following δ 13 C PC value ranges of pedogenic carbonates (12): 1) forest: continuous stand of trees at least 10-m tall with interlocking crowns (δ 13 C PC : >-11.5‰), 2) woodland/bushland/shrubland: woodland is an open stand of trees at least 8-m tall with woody cover exceeding 40% and a field layer dominated by grasses; bushland is an open stand of bushes usually between 3-and 8-m tall with woody cover exceeding 40%; and shrubland is an open or closed stand of shrubs up to 2-m tall (δ 13 C PC : -11.5 to -6.5‰), 3) wooded grassland: land covered with grassland and has 10-40% tree or shrub cover(δ 13 C PC : -6.5 to -2.3‰), and 4) grassland: land covered with herbaceous plants with less than 10% tree and shrub cover (δ 13 C PC : <-2.3‰).
Plio-Pleistocene eastern African environments preserve evidence for mosaic vegetation structures and habitat heterogeneity across relatively small spatial extents. Compilations of δ 13 C PC values show large ranges (43)(44) especially for those locations that were sampled across synchronous units (28,(45)(46)(47). δ 18 O PC are also highly variable in eastern African environments (30). We analyzed ƒ wc trends for each sampling location and/or geologic formation with simple exponential smoothing (α=0.1) to dampen extreme variations and determine central tendency shifts in the woody cover during the evolutionary history of P. boisei. Locations/geologic formations that have been sampled for δ 13  We compiled data from locations that provided a record across the MPT interval in order to detect relative changes in vegetation structures and to minimize impacts from oversampled single locations and intervals. These included the Awash Basin, the Turkana Basin, Tugen Hills, and Olduvai (Oldupai) Gorge. Simple exponential smoothing (α=0.1, 0.3, 0.6) and Loess (locally estimated scatterplot smoothing) (48) were applied to the compiled EARS ƒ WC record to estimate changes in the central tendency of vegetation structures through time. We utilized the Bayesian change point algorithm (49) of a 5-point running mean of the EARS ƒ WC record to detect significant changes in the longterm C 4 trend. Probabilities (posterior) of a change point being selected from the model were generated to detect significance (49). We compared three different methods of time-series analysis to detect temporal changes in the ƒ WC record due to the wide data scatter and differential sampling resolution. Due to differences in rainfall across the EARS (50) and current δ 18 O PC data availability, we limited the δ 18 O PC data compilation and simple exponential smoothing (α=0.1) methods to those from the Turkana Basin. Statistical analyses were conducted with Acycle v2.2.

D. Turkana Basin and EARS pedogenic carbonate isotopic records
δ 13 C PC values (n=53, 95 analyses) averaged -5.2 ± 1.7‰. These results indicate an average fraction of woody canopy cover (ƒ WC ) of 30%, ranging from 4-62% (Supplementary Fig. S5). Vegetation structural categories present during the MPT include woodlands, grassy woodlands, wooded grasslands, and grasslands. Forests are not indicated with these data. δ 18 O PC values (n=53, 95 analyses) have an average of 0.3 ± 0.4‰ (Supplementary Fig. S5). The majority of the MPT interval data points reported here are derived from the Nachukui Formation due to differential preservation (Supplementary Data). At Ileret, the 500-kyr depositional hiatus in the Koobi Fora Formation (37) does not capture the full record of vegetation structure during the MPT interval ( Supplementary  Fig. S5). The Shungura Formation in the Lower Omo Valley to the north of the Turkana Basin shows a trend toward relatively lower δ 13 C PC and δ 18 O PC values at the start of the MPT interval but has not been sampled later than 1.2 Ma (Supplementary Fig. S7) (30). Notably, the two parallel sections spaced ~1 km apart at Nariokotome demonstrate comparable δ 13 C PC and δ 18 O PC values ( Supplementary Fig. 4), and both show excursions to lower values at 1 Ma ( Supplementary Fig. S5).
The three coeval geologic formations of the Turkana Basin (Koobi Fora, Nachukui) and Lower Omo River Valley (Shungura) preserve differential depositional settings through time due to a number of factors including regional tectonic events and basin infilling, evolving hydrological configuration, and geographic and geomorphologic differences across subregions, amongst others (51)(52). Differences in depositional setting have been shown to influence δ 13 C PC values in the Plio-Pleistocene EARS (28,30). Lacustrine environments as preserved in the Nachukui and Koobi Fora formations had 10-20% less ƒ WC than those in riverine environments in the Shungura (Supplementary Fig. S7). We collected pedogenic carbonate samples from alluvial paleosols found adjacent to the margin of paleo-Lake Silbo (37,51) Our finding of relatively more C 3 vegetation on the margins of paleo-Lake Silbo is the opposite of predicted by depositional setting (28,30).
Individual EARS basins show a wide range of ƒ WC estimates ( Supplementary Fig. S7) and also differential sampling resolution. For example, Malawi ƒ WC record (53) near the early Pleistocene P. boisei site of Malema show evidence for persistent C 3 vegetation from the late Pliocene to the middle Pleistocene, yielding ƒ WC estimates between 40-60%, but the record lacks the MPT interval entirely. The Shungura Formation preserving many P. boisei fossils shows a fluctuating trend through time but is lacking the majority of the MPT interval. Two subsets of the compiled EARS ƒ WC record from Olorgesailie and Kanjera South yield evidence for pure grasslands but are potentially oversampled in single locales and time horizons. Olorgesailie was sampled may times across one paleosol/time horizon dated to 0.99 Ma (n=61) (45) and may have included phreatic components according to additional analyses reported by Levin in the Supplementary Information of Potts and others (54). Kanjera South on the Homa Peninsula was sampled many times across one paleosol/time horizon dated to 2.0 Ma (n=22) (55). Neither of these locations offers sampling resolutions for relative changes in vegetation structures through time.
The three different time-series analyses of the compiled EARS record all yield excursion peaks at 1 Ma toward C 3 -dominated vegetation structures. Based on the three generated curves, the EARS C 3 excursion equates to between a 10% and 20% reduction in woody cover within wooded grasslands.

E. P. boisei's fossil sites and estimated time of extinction
P. boisei fossils have been discovered throughout the EARS (Fig. 1) (56). Fossil sites that have also been sampled for δ 13 C PC include the Turkana Basin (Nachukui and Koobi Fora formations) of northern Kenya, the Lower Omo River Valley (Shungura Formation) in southern Ethiopia, Olduvai (Oldupai) Gorge in Tanzania, the Chesowanja site near the Tugen Hills, and the Karonga Formation at the site of Malema in Malawi. Although no δ 13 C PC values are yet reported, P. boisei has also been recovered from Konso and Peninj. P. boisei fossils have not yet been discovered in the Awash Basin, Kanjera South, or Olorgesailie, but those sites have been sampled for δ 13 C PC analysis.
The discovery of the P. boisei partial skeleton, OH-80, at Olduvai (Oldupai) Gorge in Tanzania provides its current last appearance datum of 1.34 Ma (57). The LAD is not likely the exact time of extinction (58), but it provides a lower limit of the extinction interval, which may be significantly later in time due to taphonomic factors (59). P. boisei's time of extinction is estimated as "at some point before 1.0 Ma" (56, pg. 109) and more recently as "just over 1 Ma" (60, pg. 23202). We use the interval of ~1.3-1.0 Ma as the time of P. boisei's extinction.

F. EARS hominins and T. oswaldi specimen identification references and fossil image credits
Dietary isotopic values shown in Fig. 1 and listed in Supplementary Data include only those reported to species-level designations for Paranthropus (P. aethiopicus, P. boisei) and Theropithecus (T. oswaldi). Genus Homo specimens were separated into two groupings, Early Homo and H. erectus, after (61). Early Homo includes specimens designated as H. habilis, H. rudolfensis, or Homo indet. Photo thumbnails of hominin skulls shown in Figures 1 and 7