Abstract
Adeno-associated virus (AAV) vectors are widely used for in vivo gene delivery to the central nervous system (CNS), muscle, and retina, but many clinically used capsids show limited potency in human tissues, necessitating high systemic doses that increase cost and toxicity risk. Here, we summarize recent capsid-engineering strategies designed to improve on-target delivery and reduce vector dose requirements. For CNS applications, receptor-informed engineering-such as capsids targeting transferrin receptor 1 (TfR1) or alkaline phosphatase (ALPL)-has produced large gains in blood-brain barrier (BBB) penetration and cross-species translation. In the retina, intravitreal (IVT) performance improves through fine-tuning of heparan sulfate proteoglycan (HSPG) interactions to facilitate inner limiting membrane (ILM) traversal, while suprachoroidal and laterally spreading subretinal vectors expand posterior-segment coverage. For muscle, next-generation myotropic and liver-detargeted capsids enable uniform skeletal and cardiac transduction at substantially lower intravenous doses. We compare directed evolution, rational design, and machine-learning (ML) approaches, highlighting how these methods increasingly converge by integrating structural hypotheses, in vivo selections, and multi-trait computational optimization. Quantitative benchmarks across tissues demonstrate that engineered capsids routinely deliver multi-fold improvements in potency and biodistribution relative to natural serotypes. Collectively, these advances outline a translational path toward safer, lower-dose AAV gene therapies with improved precision and clinical feasibility.