Phenotypic evolution of broomcorn millet (Panicum miliaceum) in ancient Turpan, Xinjiang, China

Abstract

Broomcorn millet ( Panicum miliaceum L.) is a typical dryland cereal and one of the oldest cultivated crops. Unraveling its phenotypic changes during dispersal is crucial for understanding the adaptive evolution of broomcorn millet across different agricultural environments. In this study, modern weedy and cultivated broomcorn millet samples were used as training material. Six morphological features (length, width, thickness, perimeter, area, and length/width ratio) of caryopsis were measured, and a high-precision discrimination model between cultivated and weedy broomcorn millet was established by integrating morphological statistics and machine learning methods. The results show that width is the most parsimonious and effective single morphological marker for distinguishing the two types. This model was then applied to desiccated broomcorn millet remains unearthed from three sites in Turpan, Xinjiang (Shengjindian, Astana, Tuyoq), spanning approximately 1500 years from the Western Han to the Southern Song Dynasties (2nd c. BC-13th c. AD). Ancient samples were classified, outliers removed, and a diachronic analysis of phenotypic trends was conducted. The results indicate that broomcorn millet individuals with cultivated-like features increased in frequency, while those with weedy-like features decreased, reflecting a transition from extensive to intensive oasis agriculture. Over the long term, the area, perimeter, and width showed increasing trends but with low goodness of fit, overall exhibiting stable fluctuations, consistent with the post-domestication diversification stage. The increase in size was mainly reflected in width and area, suggesting that artificial selection primarily acted on grain filling. This study applies machine learning to ancient broomcorn millet remains to understand phenotypic responses during dispersal, an approach also applicable to other small-seeded cereals.