Inferring cell division from cell shape

Abstract Cell division patterns shape developing organs, but their systematic analysis remains limited by the technical difficulty of existing detection methods. Here, we present a practical method for inferring recent daughter-cell pairs from static cell shapes. Although cell divisions have long been inferred by expert judgment, the accuracy of this approach has rarely been quantified. Using live imaging of Arabidopsis leaf primordia, we evaluated 49 features from 7,573 neighboring cell pairs, including 1,114 direct daughter pairs and 6,459 non-daughter pairs. A simple junction-angle criterion correctly inferred 1,030 daughter pairs with 94.1% precision and 92.5% recall, whereas symmetry-based indices performed poorly. Temporal analysis revealed that junction angles were initially high after division, decreased during growth, and were altered by subsequent divisions, thus, separating daughter pairs from non-daughter pairs. This method also achieved 98.4% precision and 97.2% recall in an independent shoot apical meristem live-imaging dataset. Application to the pointed-tip Arabidopsis mutant rpl4d-3 and two non-model species further supported its potential applicability for analyzing division patterns in diverse leaf morphologies. For community use, we provide FIJI/ImageJ and Qt-based graphical user interface implementations. Overall, this study quantitatively validates empirical geometric cues and provides a practical framework for inferring daughter pairs from static tissue images.