Unraveling transcription factor functions through integrative inference of transcriptional networks in Arabidopsis thaliana yields novel regulators involved in reactive oxygen species stress signaling

Abstract

Gene regulation is a dynamic process in which transcription factors (TFs) play an important role to control spatiotemporal gene expression. While gene regulatory networks describe the interactions between TFs and their target genes, our global knowledge about the complexity of TF control for different genes and biological processes is incomplete. To enhance our global understanding of regulatory interactions in Arabidopsis thaliana, different regulatory input networks capturing complementary information about DNA motifs, open chromatin, TF binding and expression-based regulatory interactions, were combined using a supervised learning approach, resulting in an integrated gene regulatory network (iGRN) covering 1,491 TFs and 31,393 target genes (1.7 million interactions). This iGRN outperforms the different input networks to predict known regulatory interactions and has a similar performance to recover functional interactions compared to state-of-the-art experimental methods like yeast one-hybrid and ChIP-seq. The iGRN correctly inferred known functions for 681 TFs and predicted new gene functions for hundreds of unknown TFs. For regulators predicted to be involved in reactive oxygen species stress regulation, we confirmed in total 75% of TFs with a function in ROS and/or physiological stress responses. This includes 13 novel ROS regulators, previously not connected to any ROS or stress function, that were experimentally validated in our ROS-specific phenotypic assays of loss- or gain-of-function lines. In conclusion, the presented iGRN offers a high-quality starting point to enhance our understanding of gene regulation in plants by integrating different experimental data types at the network level.