Protein targeting to extracellular vesicles and exosomes remains poorly understood, with few generalizable sequence determinants beyond classical signal peptides and transmembrane domains. This work reports the discovery of a previously uncharacterized short N-terminal targeting signal associated with exosome localization, identified using a deterministic dynamical geometric embedding framework (PhaseOS). A single latent embedding coordinate reproducibly discriminates exosome-annotated proteins in the CAFA6 dataset. Through systematic perturbation, phase rotation, masking, positional jitter, reduced-alphabet remapping, and targeted motif knockout and insertion experiments, the signal is localized to an 8–12 residue N-terminal window exhibiting strong phase sensitivity and short-period amphipathic structure. Reduced-alphabet log-odds analysis reveals a small family of degenerate residue grammars, exemplified by patterns including HGCP, HGPC, HCGP, and HCPG, which are enriched in independently annotated UniProt exosome proteins. These findings support the existence of a phase-encoded amphipathic targeting mechanism distinct from canonical signal peptides. The identified signal, termed the Amphipathic Phase EXosome (APEX) phase code, suggests that vesicle targeting can be mediated by short, phase-encoded amphipathic helices and demonstrates the utility of dynamical geometric embeddings for uncovering latent biophysical targeting rules not captured by conventional sequence models.
Kevin Shepheard (Wed,) studied this question.