ELONGATE DENDRITIC phytoliths as indicators for cereal identification and domestication: exploring a 3D morphometric approach

ROSALIE MADELEINE HERMANS^1,2*Mao Li³William H. Brightly⁴Timothy J Gallaher⁵Wouter Smagghe^6,7Hannah Lee⁵Leticia Arco⁸Lara Stas⁹Perseverence Savieri⁹Luc Vrydaghs²Karin Nys²Christophe Snoeck²Caroline A. E. Strömberg^10,5

• ¹Vrije University Brussels, Brussels, Belgium
• ²Archaeology, Environmental Changes & Geo-Chemistry research group, Elsene, Belgium
• ³Donald Danforth Plant Science Center, St. Louis, United States
• ⁴The University of Sheffield School of Biosciences, Sheffield, United Kingdom
• ⁵University of Washington Department of Biology, Seattle, United States
• ⁶Center for Plant Systems Biology, Vlaams Instituut voor Biotechnologie, Ghent, Belgium
• ⁷Department of Plant Biotechnology and Bioinformatics, Ghent University, 9000, Belgium
• ⁸Artificial Intelligence Lab, Department of Computer Science, Vrije Universiteit Brussel, Brussels, Belgium
• ⁹Core Facility for Support for Quantitative and Qualitative Research, Vrije Universiteit Brussel, Brussels, Belgium
• ¹⁰Burke Museum of Natural History & Culture, University of Washington, Seattle, United States

The phytolith (plant silica) morphotype, ELONGATE DENDRITIC, is used to indicate the presence of domesticated grasses (cereals) from the Pooideae subfamily, such as wheat and barley, in the archaeological record, but related wild taxa also produce ELONGATE DENDRITIC that closely resemble those of cereals. By examining the morphometric traits of ELONGATE DENDRITIC in a diverse set of extant Pooideae taxa, we evaluate its effectiveness as a proxy for cereal domestication and identification. We investigated the occurrence of ELONGATE DENDRITIC across a wide range of Pooideae taxa and generated 3D meshes of phytoliths using confocal microscopy. From these meshes, we extracted geometric morphometric and topological traits, which served as input for machine learning (ML) models to assess the taxonomic resolution of ELONGATE DENDRITIC. Regression models and linear discriminant analyses (LDAs) were applied to test for links between morphometric traits, domestication status, and ploidy level. Our results show that ELONGATE DENDRITIC occurrence is likely an ancestral trait within Pooideae, with high levels largely confined to Triticeae (wheat, barley, rye) and Avena (oats). Machine learning applied to 3D phytolith traits captured meaningful taxonomic patterns, with more reliable identification at broader taxonomic levels than at finer ones. However, the approach requires further refinement before it can be robustly applied to archaeological samples. Regression models and LDA demonstrated that while domestication significantly influences morphometric variation, ploidy level does not, although further study is warranted. These findings offer important guidance for archaeologists and biologists studying crop domestication. By integrating 3D morphometrics, topological data analysis, and ML, this study introduces a new approach to quantitative phytolith identification. Continued expansion of reference datasets, coupled with methodological refinement, will be essential for improving identification at finer taxonomic levels and unlocking the full potential of ELONGATE DENDRITIC in the study of domestication and cultivation practices.