InsectExpress reveals layered regulatory architecture underlying cross-species tissue expression in insects

<title>Abstract</title> <p>How genomic sequence encodes tissue-resolved gene expression across deep evolutionary distances remains poorly understood outside mammals. Here we developed InsectExpress, a multimodal framework that integrates promoter-centred DNA sequence, RNA stability features and ESM-2 protein embeddings to predict expression across 14 tissues in 12 insect species spanning six orders. Under stringent gene-family-aware evaluation, InsectExpress achieved an overall Pearson correlation of 0.680 ± 0.012 across independent seeds, outperforming multilayer perceptron and vanilla Enformer baselines, and generalised to two external species, Spodoptera frugiperda 𝑟=0.542 and Helicoverpa armigera 𝑟=0.612. We then used the model to ask what makes this transfer possible. Cross-order promoter analysis revealed lineage-biased deployment of core promoter elements, whereas tissue-specific promoter analysis identified transcription factor motifs recurrent across multiple orders, indicating deep conservation of tissue-defining regulatory programmes. Across 138,524 ortholog pairs, coupling between protein evolution and tissue-resolved expression evolution was strongest in chemosensory genes, but weak or absent in housekeeping and immune classes. Together, these results support a layered model of insect gene regulation in which conserved tissue programmes are superimposed on order-specific promoter syntax. More broadly, InsectExpress provides a transferable framework for comparing sequence-to-expression relationships across deeply diverged animal lineages.</p>