A comprehensive Arabidopsis transcription factor binding atlas reveals pervasive positional and syntactic organization of their DNA binding

Transcription factor (TF) binding underlies gene regulation, yet the determinants of TF-DNA interactions across plant genomes remain incompletely understood. Here, we present a comprehensive atlas of TF binding in Arabidopsis thaliana, generated by reanalyzing 1,157 publicly available ChIP-seq and DAP/ampDAP-seq datasets using a unified processing framework. After stringent curation, 681 high-quality experiments were retained, providing binding information for 425 TFs across 42 families and 23 structural classes. Using this resource, we show that TF binding predictability varies widely across TFs, with family identity explaining substantially more variation in predictive performance than structural class. Among the tested TF binding site (TFBS) prediction models, deep learning approaches achieved the highest overall predictive accuracy, while classical position weight matrices remained competitive and readily interpretable. Beyond motif recognition, we uncover widespread organizational principles of TF binding. TFBS positioning relative to transcription start sites is strongly family-dependent and more concentrated near promoters in vivo. Moreover, preferred spacing between homotypic TFBS is pervasive across TF families, indicating that binding-site syntax is a general organizational feature of TF binding in plants. Comparison of in vivo and in vitro binding profiles further identifies a shared core of sequence-driven binding alongside in vivo-enriched sites associated with non-canonical or composite sequence features. Together, these results highlight the interplay between intrinsic DNA recognition, higher-order binding syntax, and cellular context in shaping TF binding landscapes, while providing a broadly useful resource for the plant community.