Tetraspanin-hijacking sorting motif and AI-driven predictive modeling enable functional nuclear delivery across the blood-brain barrier via engineered extracellular vesicles

초록

Introduction Extracellular vesicles (EVs) represent promising therapeutic delivery vehicles due to their biocompatibility, barrier-crossing abilities, and capacity to evade endolysosomal sequestration. However, efficient EV engineering remains limited by incomplete understanding of cargo sorting mechanisms. Here, we developed a Natural Nanoparticle (NNP) engineering platform centered on a PTGFRN-derived transmembrane Sorting Motif, termed the NNP Sorting Motif (NSM), and established an AI-driven predictive platform for rational EV engineering. We elucidated NSM's tetraspanin-hijacking mechanism and validated its versatility across diverse cargo types, including fluorescent proteins, growth factors, and innate immune checkpoint blockers. Furthermore, we demonstrated functional nuclear delivery of therapeutic enzymes to brain parenchymal cells across the blood–brain barrier (BBB). Methods We screened EV-enriched transmembrane scaffolds and quantified sorting efficiency by immunoblot-based EV-to-cell enrichment metrics. NSM variants were generated by mutagenesis, and critical residues were mapped by leucine-scanning analysis. Tetraspanin dependency (CD9, CD81, CD63) was validated using shRNA knockdown and CRISPR/Cas9 knockout, with rescue experiments guided by multimer docking. In silico predictions of NSMtetraspanin interactions were cross-validated with wet-lab experiments. An AI predictive model was developed from a curated library of experimentally validated NSM-cargo sequences. NSM versatility was assessed using eGFP (fluorescent protein), EGF (growth factor), and SIRPα (innate immune checkpoint blocker). For CNS delivery, we engineered transferrin receptor-targeting peptide-displaying EVs carrying Cre recombinase. BBB penetration was assessed by intravital imaging, and functional nuclear delivery to brain parenchymal cells was evaluated in transgenic reporter mice. Results PTGFRN-derived constructs yielded the highest EV cargo enrichment, with the transmembrane domain alone sufficient. The V7I/T11A-optimized sequence (NSM) significantly enhanced sorting. Leucine scanning identified essential residues (G6, S10, G14, G21) forming a glycine-zipper-like motif. In silico structural modeling predicted direct NSM interactions with tetraspanins CD9, CD81, and CD63; these predictions were experimentally validated through CRISPR/Cas9 knockout and shRNA knockdown, which significantly reduced cargo loading without impairing EV biogenesis. This cross-validation between computational and wet-lab approaches confirmed a multi-tetraspanin hijacking mechanism. The AI model achieved high concordance between predicted and observed incorporation rates. NSM-engineered EVs successfully incorporated diverse functional cargos: eGFP-EVs confirmed precise membrane display and luminal encapsulation; EGF-EVs induced EGFR phosphorylation in keratinocytes; SIRPα-EVs exhibited specific CD47 binding and enhanced macrophage-mediated phagocytosis, demonstrating innate immune checkpoint blockade. For CNS-targeted delivery, intravenously administered transferrin receptor-targeting peptide-displaying EVs demonstrated efficient BBB penetration as confirmed by intravital imaging, reaching brain parenchymal cells including neurons. Cre recombinase-loaded EVs distributed broadly throughout brain tissue following systemic administration and achieved functional nuclear delivery to neurons, as evidenced by tdTomato reporter activation across multiple brain regions including cerebral cortex, hippocampus, and striatum in transgenic mice. Conclusion NSM enhances EV engineering through direct engagement with multiple tetraspanins (CD9, CD81, CD63), validated by in silico-wet lab cross-validation. Combined with AI-driven predictive modeling, this platform enables rational cargo optimization and functional nuclear delivery of therapeutic enzymes across the BBB to brain parenchymal neurons, establishing a versatile strategy for next-generation CNS-targeted EV therapeutics.