USING AUTOMATED, QUANTITATIVE IMAGING ANALYSIS OF FOSSIL GRASS SILICA PHYTOLITH TO ELUCIDATE EVOLUTION OF TEMPERATE GRASSLANDS

With nearly 4,000 species, Pooideae is one of the largest subfamilies in the Poaceae (grasses). It is thought to have radiated in the northern hemisphere starting sometime during the Late Cretaceous–Paleocene, and its members now dominate vast expanses of primarily northern temperate- and high-altitude grasslands. Key to dominance in these cooler environments is the unique tolerance of many pooids to both temporary and prolonged periods of cold. Phylogenetic studies indicate that the suite of cold tolerance traits evolved in a stepwise manner starting well before the diversification of Pooideae, initially as a general stress response, with difference pooid lineages acquiring mechanisms to withstand long-term cold during the late Eocene-Oligocene global cooling. Paleobotanical data in the form of grass silica short cell phytoliths (GSSCP) suggest that pooids were dominant in Earth’s first grassy habitats (woodlands–grasslands), which spread in North America and Turkey during the Oligocene–early Miocene. However, counter to what might be expected, these pooids coexisted with palms and spiral gingers, pointing to much milder climates than typically associated with Pooideae, and to vegetation that lacked a modern analog.

To better understand the ecological role of pooids in early grass-dominated habitats, we use quantitative analysis of GSSCP 3-D morphology of modern and Oligo-Miocene pooid GSSCP. Our work has previously demonstrated that this approach can robustly classify fossil phytoliths into grass subclades (e.g., subtribe), although precision varies depending on subclade. To do so, we have created a database of nearly 2,400 3-D GSSCP models from 45 Pooideae species and 70 species early-diverging grasses, Bambusoideae and Oryzoideae. The models were developed from confocal images of GSSCP stained with fluorescent dye, which were converted to 3-D surface models that can be quantified and analyzed using 3-D geometric morphometrics. We then use machine learning and linear discriminant analysis to develop a framework in which we can place fossil GSSCP in Oligo-Miocene samples from the Great Plains of North America and Turkey, allowing us to better circumscribe the ecologies of the grasses that made up the earliest grassland communities.