AAV vectors

AAV vectors are gene-delivery vehicles built from adeno-associated virus, a small, non-enveloped parvovirus that infects humans without causing known disease. Nearly every licensed in vivo gene therapy uses one, including treatments for inherited retinal dystrophy, spinal muscular atrophy, and haemophilia. The vector's virtues and its constraints come from the same source: a genome barely large enough to hold one gene, a capsid the immune system recognises, and a payload that mostly sits outside the chromosomes rather than joining them.

§How it works

Wild-type AAV carries about 4.7 kilobases of single-stranded DNA between two inverted terminal repeats, encoding two gene cassettes — rep, for replication, and cap, for the capsid. It cannot complete its own life cycle without a helper virus such as adenovirus or herpes simplex, which is why it was first noticed in 1965 as a contaminant in adenovirus preparations rather than as a pathogen in its own right.

To make a vector, everything between the terminal repeats is deleted and replaced with the therapeutic cassette: a promoter, the gene, a polyadenylation signal. The rep and cap functions are supplied separately during manufacturing, so the resulting particle can enter a cell and deliver its cargo but cannot replicate. Once inside, the single-stranded genome is converted to double-stranded DNA and circularises into episomes that persist in the nucleus, transcribing for years in non-dividing tissue. A small fraction integrates at semi-random sites; wild-type AAV's preference for the AAVS1 locus depends on Rep protein, which vectors lack.

Cell entry requires glycan attachment factors that vary by serotype plus a shared protein receptor, AAVR, identified in 2016.¹ The discovery explained why some cell types resist transduction regardless of capsid choice.

Because the vector reaches only body cells and does not enter the germ line at any appreciable rate, AAV therapy sits squarely on the somatic side of the boundary that defines Human germline editing. Outside medicine, AAV is the standard delivery route for Optogenetics constructs in neuroscience and, less legitimately, the vector class most often discussed in connection with Gene doping.

§Serotypes and tropism

The natural AAV serotypes differ chiefly in their capsid surface, and therefore in which tissues they enter. AAV2 was the first characterised and remains common in the eye. AAV8 and AAV5 favour the liver. AAV9 crosses the blood-brain barrier in newborns well enough to underpin an approved systemic therapy for spinal muscular atrophy. AAV1, AAV6, and AAVrh74 are used for muscle.

Capsid engineering aims to do better than nature. Directed evolution, DNA shuffling, and peptide-display libraries have produced capsids with sharply improved central nervous system delivery in mice.² Machine-learning models fitted to library-screening data now propose viable capsid variants directly, an application of AI protein design to the delivery problem rather than to the therapeutic protein. The best-known cautionary tale is AAV-PHP.B, which crosses the murine blood-brain barrier with startling efficiency and then turned out to depend on a receptor, Ly6a, present in some inbred mouse strains and absent in primates.³ Its failure to translate is the standard example of why capsid performance in one species predicts little about another — the same species-gap problem that recurs throughout Targeted drug delivery.

Mouse capsids are not human capsids

Every claim about a new capsid's tropism should be read with the question: in which species, and does the receptor it exploits exist in people? Several widely cited enhanced capsids do not work in non-human primates at all.

§Development history

1. 1965
   Discovery
   AAV is identified as a small contaminating particle in adenovirus preparations and shown to require a helper virus.
2. 1984
   First recombinant vector
   Researchers demonstrate that the viral genes can be removed and replaced with a transgene between the inverted terminal repeats.
3. 2006
   The immune barrier appears
   In a haemophilia B trial, factor IX expression rises then falls as capsid-specific T cells destroy transduced hepatocytes, defining the field's central obstacle.
4. 2012–2017
   First approvals
   Glybera is licensed in Europe and later withdrawn; Luxturna is approved in the United States for RPE65 retinal dystrophy.
5. 2019
   Systemic dosing at scale
   Zolgensma is approved for spinal muscular atrophy, delivering AAV9 intravenously at doses orders of magnitude above earlier local injections.
6. 2020–2025
   Dose-limiting toxicity
   Deaths from acute liver failure and thrombotic microangiopathy in high-dose systemic trials, including for X-linked myotubular myopathy and Duchenne muscular dystrophy, force reassessment of dose ceilings.

§Limitations

Cargo. The 4.7 kb ceiling excludes many important genes outright — dystrophin, most CRISPR–Cas9 constructs once guides and regulatory elements are added, and the still larger dCas9 fusions used for Epigenome editing. Workarounds include micro-genes such as the truncated dystrophins used in Duchenne programmes, dual-vector systems that reconstitute a protein from two halves via split inteins, and pairing AAV with smaller nucleases. Each adds inefficiency. Muscle-directed constructs expressing follistatin, discussed under Myostatin inhibition, are small enough to fit and have been used in both clinical and unsanctioned settings.

Pre-existing immunity. Because AAV circulates naturally, a substantial fraction of adults — depending on serotype and region, roughly a third to more than half — carry neutralising antibodies that inactivate the vector before it reaches its target. Those patients are screened out of trials, which distorts both eligibility and evidence.

Redosing. A treated patient develops high antibody titres against the capsid, so a second dose of the same serotype is ineffective. Switching serotypes helps only partially because of cross-reactivity. Strategies under investigation include plasmapheresis, IgG-cleaving bacterial proteases, and B-cell-depleting or tolerising regimens; none is routine practice as of 2026.

Durability. Episomal genomes are not replicated with the chromosome, so dividing cells dilute them out. This matters most in children, whose livers and muscles grow substantially after treatment, and it underlies the declining expression seen in some haemophilia programmes. Editing the genome rather than adding an episome avoids the problem, which is one reason Base editing and Prime editing payloads are attractive despite the packaging squeeze.

Manufacturing. Vector is produced by transient transfection of HEK293 cells or in insect cells with baculovirus, then purified. Yields are modest, a large fraction of particles are empty capsids that contribute immunogenicity without therapeutic effect, and cost of goods for a systemic dose is a meaningful share of the price of the finished therapy discussed under Somatic gene therapy.

§Risks

The immunology cuts several ways. Innate sensing of vector DNA drives complement activation; high systemic doses have produced complement-mediated thrombotic microangiopathy, acute kidney injury, and hepatotoxicity. Adaptive responses against the capsid destroy transduced cells, a pattern first documented in a 2006 haemophilia B trial in which factor IX expression rose and then fell as capsid-specific T cells cleared the transduced hepatocytes.⁴ Responses against the transgene product itself can occur when a patient has never made the protein and does not recognise it as self. Where a vector delivers a nuclease rather than a gene, editing errors add a second, independent risk channel.

Deaths have occurred. Several patients died of hepatobiliary failure in a trial of a high-dose AAV therapy for X-linked myotubular myopathy, and in 2025 acute liver failure in recipients of a Duchenne muscular dystrophy therapy prompted regulatory action and a re-examination of systemic dose ceilings. Prophylactic corticosteroids, complement inhibitors, and weight-capped dosing are all responses to this pattern, and none of them is a solution.

Insertional mutagenesis is a lower but non-zero concern. Hepatocellular carcinoma driven by vector integration has been reported in mice; the human evidence remains equivocal, and long-term follow-up requirements reflect that uncertainty. Comparison with other delivery routes clarifies the trade-offs:

§Outlook

AAV is unlikely to be displaced soon for durable gene addition to non-dividing tissue: nothing else delivers a gene to photoreceptors or motor neurons and keeps it expressing for years. Its share of new programmes is nonetheless shrinking, because Lipid nanoparticles are better suited to the hit-and-run logic of genome editing, can be redosed, and are cheaper to make. The most consequential open questions are whether capsid engineering can produce a vector that reaches muscle or brain at doses low enough to avoid complement activation, and whether any of the antibody-clearing strategies makes redosing routine. A negative answer to both would leave AAV where it is now — extraordinarily effective for a small number of tissues and diseases, and a poor fit for anything requiring a second dose or a large payload. Delivery, not editing chemistry, has been the binding constraint on genetic medicine for thirty years, and it remains one for both ex vivo cell products and any future application to Gene therapy for aging.

§See also

• Somatic gene therapy
• Lipid nanoparticles
• CRISPR–Cas9
• Base editing
• Gene therapy for aging
• Casgevy
• Optogenetics
• Off-target effects in genome editing

§References