Determining the Principles behind Tissue-Specific Gene Expression

In our laboratory, we have recently created a genome-wide dataset that includes both Transcription Start Sites (TSSs) and Open Chromatin (OC) regions in Arabidopsis roots, as well as machine learning models trained on these datasets that allow us to: 1) very accurately predict genomic locations that will express highly, and 2) derive from the model those cis-regulatory elements (CREs) that are most predictive of expression for each highly-expressing TSS location. The analogous dataset and modeling concept applied to leaves successfully learns a CRE-based model for leaf TSSs. What distinguishes leaf-specific from root-specific characteristics in endogenous plant promoters, and what do these characteristics reveal about synthetic design principles for tissue-specific promoters? We have taken the first step toward addressing these questions in building a model that can readily distinguish a highly expressed root-specific TSS from a leaf-specific TSS. In this talk, we will explain what we have learned by interpreting this model, and discuss the results of several initial in vitro tests of these concepts.