A Deep-Learning Atlas of XPO1-Mediated Nuclear Export at Proteome Scale

Abstract

Exportin 1 (XPO1/CRM1) is the principal nuclear export receptor for cargos bearing hydrophobic nuclear export sequences (NESs). Dysregulation of XPO1-dependent export is implicated in cancer, neurodegeneration, and other diseases, yet a comprehensive view of XPO1 function remains limited by the poor reliability of sequence-based NES prediction. Existing predictors are largely derived from a small set of XPO1-cargo structures and are therefore biased toward canonical docking geometries, limiting their ability to detect NESs that engage XPO1 through noncanonical pocket-occupancy patterns. We hypothesized that deep learning-based structural modeling could overcome this limitation by directly sampling binding geometries. Using AlphaFold 3, we modeled full-length cargo-XPO1-RanGTP complexes for more than 4,000 human proteins and identified over 3,000 previously uncharacterized, high-confidence NESs. Integration of AlphaFold predictions with unsupervised structural geometry analysis and experimental validation identified both canonical NESs and noncanonical sequence patterns exhibiting atypical anchor-residue usage, expanding the structural language of XPO1-recognized NESs. Groove-resolved contact maps further revealed helix rotation within the export groove as a regulatory feature that can rewire pocket usage without altering the core NES sequence, enabling PTM- and cofactor-sensitive tuning of export strength. This exportome atlas resolves many previously ambiguous or unidentified NESs in disease-associated proteins and across major cellular systems, including centrosome organization, mRNA processing, ubiquitin signaling, kinase networks, ribosome quality control, and macroautophagy. We further identified recurrent NES-NLS tandem motifs encoded in primary sequence, suggesting coordinated regulation of nucleocytoplasmic transport. Together, our deep learning-based exportome atlas, integrated with NLS maps and accessible through a web-searchable resource, defines an expanded and regulatable code of nuclear transport at proteome scale and offers a framework for dissecting nuclear trafficking and its dysregulation in human disease.