Machine learning-guided leaf margin classification and discovery that BcPHB promoter selection rewires a BcPHB-BcLMI1 module to shape leaf margins in Brassica campestris.

The morphological diversification of leaf margins reflects the evolutionary reshaping of conserved developmental programs. However, fully deciphering the underlying molecular basis remains challenging due to difficulties in phenotype quantification and the incomplete characterization of complex regulatory networks. In this study, we established a computer-vision framework to decode leaf lobe trait and identified the HD-ZIP III factor PHABULOSA (BcPHB), combining genome-wide association studies (GWAS) and selective sweep analysis in Brassica campestris. We found that promoter variation defines two major haplotypes, BcPHBHapI and BcPHBHapII, driving divergent expression levels associated with entire and lobed leaves. Functionally, BcPHB knockdown lines (BcPHBami) exhibited lobed margins, whereas complementation of lobed accessions with the BcPHBHapI promoter partially restored the entire-leaf phenotype. Similarly, overexpression of LATE MERISTEM IDENTITY 1 (BcLMI1), a conserved regulator of leaf lobation, phenocopied the BcPHBami lines. Mechanistically, BcPHB and BcLMI1 form a reciprocal negative feedback loop that stabilizes margin patterning. Furthermore, domestication-driven selection of BcPHBHapI facilitated the transition from ancestral lobed leaves to cultivated entire forms. This study reveals a pivotal developmental module and illustrates how regulatory evolution drives morphological diversification in crops.