Deep learning uncovers conserved regulatory logic and element dosage underlying stable gene expression in grasses

Area of Science:

• Genomics and Evolutionary Biology
• Plant Molecular Biology
• Computational Biology

Background:

• Gene expression programs are conserved across grass species (Poaceae).
• Rapid evolutionary turnover of cis-regulatory element (CRE) sequences complicates understanding of expression stability.
• Mechanisms maintaining stable gene expression despite noncoding divergence are unclear.

Purpose of the Study:

• To investigate the conservation of regulatory logic mapping DNA sequence to chromatin accessibility (CA) across grass species.
• To identify high-impact candidate CREs (cCREs) using a novel computational approach.
• To elucidate the regulatory architectures and evolutionary dynamics of CREs in grasses.

Main Methods:

• Generation of chromatin accessibility (CA) maps and transcriptomes for five orthologous tissues across five Poaceae species.
• Training deep-learning models to predict CA from DNA sequence, testing cross-species predictive accuracy.
• Development and application of Basenji-HMM, integrating deep learning and Hidden Markov Models, for cCRE identification.
• Integration of Basenji-HMM with motif scanning and footprinting analyses.

Main Results:

• Deep-learning models trained in one species accurately predicted CA in another (e.g., rice), with cross-species correlations >0.70.
• Basenji-HMM identified nucleotide-resolution cCREs, revealing conserved regulatory logic despite noncoding divergence.
• cCRE abundance around orthologous genes is conserved across species, correlating with conserved CA and gene expression, even with low sequence orthology (27-36%).

Conclusions:

• The regulatory logic linking DNA sequence to chromatin accessibility is conserved across grass species.
• A 'regulatory equivalence' model is supported, where compensatory CRE turnover maintains gene dosage and expression stability.
• This provides a framework for understanding how gene expression is stabilized across diverse, rapidly evolving genomes.