A phenotype- and potency-gated extracellular vesicle allocation hypothesis for autism.

Autism is a heterogeneous neurodevelopmental condition for which a single extracellular vesicle (EV) product is unlikely to match every biological context. We hypothesized that four development-coded EV candidates should be allocated by convergent phenotype, biomarker, and patient-derived functional response rather than diagnosis alone. EXO2, a cord-blood-plasma EV candidate, was tested only in an immune-inflammatory endophenotype characterized by regression or severe irritability along with objective inflammatory activity, after excluding pain, sleep disorder, epilepsy, infection, and other mimics. EXO3, a dental-pulp mesenchymal-stromal-cell EV candidate, was tested in a synaptic-developmental endophenotype with prominent language impairment or intellectual disability, but only if the patient-derived neurons show improved neurite, synaptic, and network function. EXO1, a Wharton-jelly mesenchymal-stromal-cell EV candidate, was tested in a circuit-plasticity endophenotype dominated by abnormalities in social communication, sensory, and repetitive behaviors using a concordant neuronal-network assay. R-EXO, an umbilical-cord-tissue mesenchymal-stromal-cell EV candidate, remained an exploratory option for a neurovascular-barrier/extracellular matrix endophenotype. Regression, language delay, or behavior alone are not selection biomarkers. Each lot must pass identity, purity, safety, and phase-specific potency gates. Sequential use of EXO2, followed by EXO1, EXO3, or, in exceptional cases, R-EXO is predicted to outperform ungated or simultaneous use when inflammation masks a residual developmental mechanism. This is a falsifiable preclinical framework and does not support clinical administration.