The trans-regulatory landscape of gene networks in plants

Abstract

The effector domains of transcription factors play a key role in controlling gene expression; however, their functional nature is poorly understood, hampering our ability to explore this fundamental dimension of gene regulatory networks. To map the trans-regulatory landscape in a complex eukaryote, we systematically characterized the putative effector domains of over 400 Arabidopsis thaliana transcription factors for their capacity to modulate transcription. We demonstrate that transcriptional effector activity can be integrated into gene regulatory networks capable of elucidating the functional dynamics underlying gene expression patterns. We further show how newly characterized domains can enhance genome engineering efforts and reveal how plant transcriptional activators share regulatory features conserved across distantly related eukaryotes. Our results provide a framework to systematically characterize the regulatory role of transcription factors at a genome-scale in order to understand the transcriptional wiring of biological systems.

Systematic identification of transcriptional activator domains from non-transcription factor proteins in plants and yeast

“…We aimed to systematically discover previously uncharacterized ADs derived from non-TF proteins in two model eukaryotic systems, A. thaliana and S. cerevisiae . In previous work, we have shown that AD predictors derived from fungal data can accurately predict ADs in plant TFs and that plant ADs function in yeast (10). Here, we leveraged this result to predict activators in plant and yeast proteins, followed by high-throughput experimental validation in yeast.…”

“…Besides these candidates in Arabidopsis, we found AD containing genes in the chloroplast (19), plasma membrane (15), mitochondria (13), golgi (9), endoplasmic reticulum (9), peroxisome (2), vacuole (2) and extracellular space (9). In yeast there were candidates in the endoplasmic reticulum (12), vacuole (14), bud neck (11), mitochondria (10), the vacuolar membrane (3) and multiple non-nuclear organelles (26). Overall, 90 Arabidopsis and 101 yeast non-nuclear genes with ADs are non-cytosolic and are targeted to a specific organellar compartment, raising the question whether they can be relocated to the nucleus to modulate transcription.…”

“…In this work, we will refer to all short protein sequences that function in AD assays as ADs, regardless of their native function. New high-throughput methodologies have helped characterize the regulatory activity of transcriptional effector domains en masse in yeast, human, and fly models (3)(4)(5)(6)(7)(8)(9), and these approaches are beginning to be implemented to study plants (10). Still, most studies have largely biased their focus on TFs, leaving other nuclear proteins and their potential role in transcription understudied.…”

“…yeast, human, etc.). We previously reported that a machine learning model trained on data from a large library of synthetic activators from yeast can correctly predict ADs in plant TFs (10); however, it is still unclear how applicable and scalable these models are in plant systems, necessitating further evaluation of more plant ADs predicted by yeast models. A larger set of validated plant ADs would allow the comparison of sequence features in plant ADs with observations in other well-studied eukaryotes.…”

The balance of acidic and hydrophobic residues predicts acidic transcriptional activation domains from protein sequence

“…Computationally annotating activation domains would allow studies of how paralogous transcription factors diversify after duplication and enable evolutionary comparisons of domain shuffling. Comprehensive lists of transcription factors with activation domains could improve gene regulatory networks by adding signs to the connections inferred from genome binding data (Hummel et al, 2022) or by distinguishing direct and indirect connections inferred from genetic perturbations. Predicting activation domains is a key step towards building models that predict how mutations in activation domains modulate activity, which, in the long term, could classify patient mutations in activation domains as benign or pathogenic (Richards et al, 2015; Starita et al, 2017).…”