Adeno-associated virus (AAV) vectors are among the most widely used delivery vehicles in gene therapy, and the conventional wisdom has long held that they are safer than alternatives because their genetic cargo stays separate from a patient’s chromosomes. That is mostly true: the vast majority of AAV vector DNA persists as free-floating circles called episomes rather than stitching itself into your chromosomes. But “mostly” is doing heavy lifting in that sentence. Integration does happen, and as sequencing technology has grown more sensitive, researchers have found it occurs more often and in more consequential genomic neighborhoods than early studies suggested. Understanding where, how often, and under what conditions AAV integrates matters for anyone undergoing gene therapy or following its development.
How Most AAV Vector DNA Behaves After Delivery
When a recombinant AAV (rAAV) vector enters a cell, the single strand of DNA it carries gets converted into a double-stranded form, and in most cases that double-stranded DNA links up with other copies to form long chains called concatemers. These concatemers sit in the nucleus without inserting into your chromosomes. In mouse skeletal muscle, careful quantitative analysis has shown that more than 99.5% of rAAV vector copies remain episomal.1Molecular Therapy. Quantitative Analysis of Adeno-Associated Virus Vector Genome Integration in Mice Work in nonhuman primates paints a similar picture, with roughly 96% of vector genomes in muscle and 93% in liver assembling into concatemers that are predominantly episomal.2Molecular Therapy. Comprehensive Analysis of rAAV Vector Integration Site Distribution in Nonhuman Primates Supporting Safety and Efficacy in Translating Clinical Gene Therapy
The episomal nature of most AAV genomes is actually a double-edged feature. It means the vector is unlikely to disrupt your genes through insertion, which is good for safety. But episomal DNA can be lost when cells divide, because it does not replicate along with the chromosomes. In tissues like muscle and brain, where cells rarely divide in adulthood, episomal AAV can stick around for years. In the liver, where cells turn over more readily after injury or disease, those episomes may gradually disappear, and expression of the therapeutic gene can decline over time.
What Drives the Small Fraction That Does Integrate
Wild-type AAV has an elegant trick: proteins called Rep68 and Rep78 guide the viral genome to a specific landing pad on chromosome 19, known as AAVS1. Both the virus and this chromosomal site share a short recognition sequence that the Rep proteins bind, enabling targeted insertion.3PubMed. A 16bp Rep binding element is sufficient for mediating Rep-dependent integration into AAVS1 This site-specific integration involves an unusual preliminary step: the AAVS1 locus gets amplified before the viral DNA inserts, suggesting the process is more complex than simple cutting and pasting.4PubMed Central. Adeno-associated virus site-specific integration and AAVS1 disruption
Here is the critical distinction for gene therapy: recombinant AAV vectors used in the clinic have had the Rep genes stripped out. Without Rep proteins, the vector loses its ability to home in on AAVS1. Any integration that does occur is essentially random, driven not by the virus’s own machinery but by the cell’s DNA repair systems. Specifically, proteins involved in a repair pathway called nonhomologous end joining (NHEJ) play a role. One key enzyme, ligase IV, is needed for what site-specific integration remains; without it, the ratio tips heavily toward random insertion.5PubMed Central. Adeno-associated virus site-specific integration is mediated by proteins of the nonhomologous end-joining pathway Another repair-related enzyme, DNA-PKcs, is involved in converting AAV genomes into circular episomal forms, particularly in non-dividing cells.6PubMed Central. Host cell DNA repair pathways in adeno-associated viral genome processing
The hairpin-shaped structures at the ends of the AAV genome, called inverted terminal repeats (ITRs), also appear to influence integration. Researchers have proposed that the single-stranded, flexible nature of these ITRs allows them to slip into nicks or breaks in chromosomal DNA through mechanisms that do not follow the classical repair pathways, possibly involving tiny stretches of matching sequence.7Communications Biology. AviTag-seq unifies nucleotide-resolution maps of CRISPR off-targets and AAV vector integrations And once integrated, the ITR structures are frequently mangled: one study found that only about 10% of ITRs at integration sites were intact, with most showing truncations averaging around 103 base pairs instead of the full length.8Nature Biotechnology. Integrated vector genomes may contribute to long-term expression in primate liver after AAV administration
Where in the Genome Integration Tends to Land
Without Rep proteins steering them, rAAV vectors do not insert uniformly across the genome. Instead, integration events cluster in regions that are genetically active. Studies consistently find insertions near active genes, in CpG-rich regions associated with gene regulation, and in areas of open chromatin.9Molecular Therapy. AAV Integration: Implications for Long-Term Genomic Stability This pattern makes intuitive sense: open, actively used stretches of DNA are physically more accessible to incoming fragments than tightly packed, silent regions.
The bias is even more striking in living tissue. A high-resolution study in mouse liver found that promoter regions, the “on switches” for genes, accounted for about 35% of integration events, a two-fold increase compared to what was seen in cell lines grown in a dish. Introns (the non-coding stretches within genes) made up another 42%, while insertions in the vast intergenic deserts between genes dropped to just 7%.7Communications Biology. AviTag-seq unifies nucleotide-resolution maps of CRISPR off-targets and AAV vector integrations In non-dividing liver cells under normal physiological conditions, chromatin accessibility appears to be the dominant factor shaping where vectors land.
Nonhuman primate studies reinforce this. Integration was seen in every AAV-treated animal, with insertions enriched in highly expressed genes, open chromatin, and transcriptionally active ribosomal RNA regions.10Journal of Thrombosis and Haemostasis. Adeno-associated viral vector integration: implications for long-term efficacy and safety That said, the distribution across chromosomes was largely random with respect to chromosome size, meaning no single chromosome was disproportionately targeted.
How Often Does Integration Actually Happen in Humans
Pinning down the rate has been difficult, partly because integration is rare enough that you need very sensitive detection methods, and partly because most clinical studies rely on tissue biopsies that capture only a snapshot. The most direct human evidence comes from liver biopsies taken from hemophilia A patients who received AAV gene therapy. In those samples, collected up to about three years after treatment, vector-genome junctions (the signature of integration) accounted for roughly 43% of sequencing reads, with episomal vector-vector forms making up the rest.11Blood. Adeno-associated viral integration profiles in human liver biopsies following Hemophilia A gene therapy That is a higher proportion than many expected, though the absolute number of unique integration sites was modest: 237 across two biopsies, with about half falling between genes rather than inside them.
A separate analysis of human liver biopsies following treatment with valoctocogene roxaparvovec (a hemophilia A gene therapy) found that the number of genomic integrations was substantially lower than the number of cells actually expressing the therapeutic protein. The integration profiles were polyclonal, meaning they were spread across many different genomic locations without evidence of any single insertion site being amplified, and there was no enrichment near cancer-associated genes.12PubMed. Recombinant Adeno-Associated Virus Integration Profiles in Nonhuman Primates and Gene Therapy Participants after Treatment with Valoctocogene Roxaparvovec
In a xenograft model where human liver cells were expanded in mice after AAV transduction, integration was detected at a frequency of about 1% to 3%. The inserted sequences were often heavily rearranged, and the host DNA at the insertion site frequently had deletions.13PubMed Central. AAV integration in human hepatocytes That frequency was described as “surprisingly high” by the authors, though the model involved expanding cells through division, which could amplify integration events that would otherwise remain negligible in a stable, non-dividing liver.
The Cancer Question From Mouse Models
The most alarming signals about AAV integration have come from studies in neonatal mice. When newborn mice received AAV vectors, researchers found insertions near a genomic region called the Rian locus, which contains regulatory RNAs involved in cell growth. Those insertions led to liver tumors (hepatocellular carcinoma, or HCC) in a striking proportion of treated animals.14PubMed Central. Liver Injury Increases the Incidence of HCC following AAV Gene Therapy in Mice Detailed sequencing confirmed that these tumors contained AAV integrations in the Rian locus at much higher copy numbers than surrounding normal tissue, suggesting the integration itself drove uncontrolled growth.15JCI Insight. Vector design influences hepatic genotoxicity after adeno-associated virus gene therapy
What makes this more nuanced than a simple “AAV causes cancer” story is that the context matters enormously. In adult mice on a regular diet that received the same Rian-targeting vector, only about 5% developed tumors. But when those adult mice were put on a high-fat diet or underwent partial surgical removal of the liver (which forces the remaining cells to divide rapidly), the tumor rate shot up to 100%.16Molecular Therapy. Liver Injury Increases the Incidence of HCC following AAV Gene Therapy in Mice This points to cell division as a major amplifier: when liver cells are replicating frequently, whether because the animal is a newborn, is recovering from injury, or has a fatty liver, the opportunity for integration events to be clonally expanded increases dramatically.
Humans do not have an equivalent of the mouse Rian locus, so the specific mechanism does not translate directly. But the broader principle does: integration near growth-regulating genes, in tissue that is actively dividing, creates a higher-risk scenario. This is one reason researchers pay close attention to pre-existing liver disease in patients considered for AAV-based liver-directed gene therapy. In mouse models of progressive liver disease, conventional AAV transduction was markedly reduced in animals with established disease, suggesting that the diseased liver environment affects both the therapy’s effectiveness and its risk profile.17PubMed. Prevention of Cholestatic Liver Disease and Reduced Tumorigenicity in a Murine Model of PFIC Type 3 Using Hybrid AAV-piggyBac Gene Therapy
What the Human Data Shows About Tumor Risk So Far
The clinical picture has been more reassuring than the mouse experiments would suggest. In the hemophilia A liver biopsies described earlier, histological assessment found no evidence of tumor formation up to three years after treatment. Integration sites were distributed uniformly across chromosomes, with no clustering that would suggest clonal expansion, and no increase in insertions near known cancer genes.11Blood. Adeno-associated viral integration profiles in human liver biopsies following Hemophilia A gene therapy The most common single integration site was found in an intron of a gene called PTPRM, and it appeared only twice in sequencing reads, hardly the signature of a clone taking over.
Similarly, integration profiles from valoctocogene roxaparvovec-treated patients and nonhuman primates showed polyclonal distributions with only a small bias toward open chromatin, and no relative enrichment near cancer-associated genes.12PubMed. Recombinant Adeno-Associated Virus Integration Profiles in Nonhuman Primates and Gene Therapy Participants after Treatment with Valoctocogene Roxaparvovec In the nonhuman primate studies, the largest clonal expansion observed was 29 copies of a single integration, which is notable but did not lead to any pathology.10Journal of Thrombosis and Haemostasis. Adeno-associated viral vector integration: implications for long-term efficacy and safety
These findings do not rule out the possibility of tumor formation over longer time horizons. Three years of follow-up is brief compared to the latency of most cancers. And the number of biopsied patients remains small. Regulatory agencies have taken the position that integration risk assessment should be part of nonclinical safety testing for all AAV gene therapies, reflecting the view that low probability is not the same as zero probability.18PubMed Central. Regulatory Consideration for the Nonclinical Safety Assessment of Gene Therapies
Can Integrated Vectors Contribute to Long-Term Therapeutic Expression
There is an ironic twist to the integration story. While safety discussions focus on the risks, some researchers have begun to ask whether the small fraction of vectors that do integrate might actually be pulling their weight therapeutically. In primate liver, a study using long-read sequencing found that integrated vector genomes could contribute to sustained transgene expression, particularly if episomal copies are gradually lost through cell turnover.8Nature Biotechnology. Integrated vector genomes may contribute to long-term expression in primate liver after AAV administration In other words, the very phenomenon that raises safety concerns might also be part of the reason some gene therapies keep working years after a single dose.
This creates a genuine tension in vector design. Strategies that minimize integration might improve safety but could reduce durability of treatment. Strategies that tolerate or even encourage controlled integration (such as pairing AAV with CRISPR-based tools to guide insertion to a specific safe-harbor locus) could improve long-term expression but introduce different risks.19PubMed Central. CRISPR-mediated Integration of Large Gene Cassettes using AAV Donor Vectors The field is working its way toward designs that balance these competing demands.
How Detection Methods Shape What We Know
Much of what we understand about AAV integration has changed as detection tools have improved. Earlier studies relied on methods that could recover only handfuls of integration junctions. The development of high-throughput integrant capture sequencing marked a dramatic shift, recovering nearly 12 million AAV junctions from a single human cell line and providing five orders of magnitude more data than was previously available.20PubMed Central. High-throughput sequencing reveals principles of adeno-associated virus serotype 2 integration Similar methods applied to AAV serotype 5 yielded more than 660,000 integration junctions, revealing that different viral serotypes have distinct integration preferences.21PubMed Central. Highly divergent integration profile of adeno-associated virus serotype 5 revealed by high-throughput sequencing
The latest generation of tools includes both short-read and long-read sequencing approaches, and the two do not always agree. Short-read methods generate huge volumes of data but can miss complex rearrangements at the integration site. Long-read methods capture the full structure of the insertion, including the truncated ITRs and host-DNA deletions, but produce fewer total reads. One head-to-head comparison found that about 53% of AAV-containing reads from a long-read approach were chimeric (containing both vector and host sequence), compared to just 7% from a short-read approach, a difference driven largely by the physical constraints of short reads.22Molecular Therapy. RAAVioli enables comprehensive analysis of AAV integration sites and vector rearrangements from short- and long-read sequencing Updated bioinformatic pipelines for short-read data have also helped by reducing artifact hotspots that earlier analyses mistook for true integration clusters.23Molecular Therapy Methods & Clinical Development. Comparison and cross-validation of long-read and short-read target-enrichment sequencing methods to assess AAV vector integration into host genome
The practical upshot is that earlier estimates of integration frequency were probably underestimates, and the genomic landscape of integration is more skewed toward active genes than originally thought. As methods continue to improve, the picture is likely to keep evolving.
Chromosomal Damage at Integration Sites
Integration is not always a clean insertion. Early work showed that integrated AAV proviruses can be associated with chromosomal deletions and other rearrangements, and that these events frequently mapped to chromosome 19 even though they did not always land at the wild-type AAVS1 site.24PubMed. Chromosomal effects of adeno-associated virus vector integration More recent long-read studies have confirmed that most inserted rAAV sequences are heavily rearranged and are accompanied by deletions of the surrounding host DNA.13PubMed Central. AAV integration in human hepatocytes In non-dividing cells, this kind of local genomic disruption may not have immediate consequences. But in dividing cells, where DNA damage can be propagated to daughter cells, these structural changes represent a real if still poorly quantified risk.
Germline Transmission and Heritable Changes
One question that comes up whenever a gene therapy vector integrates into DNA is whether those changes could be passed to future generations. Extensive animal studies across mice, rats, rabbits, and dogs have addressed this for AAV delivered to muscle or the liver. The consistent finding is that while vector sequences can be detected in gonadal tissue using sensitive PCR methods, the signal diminishes over time and the DNA does not appear to enter the germ cells themselves. In rabbits, AAV signal was localized to the connective tissue surrounding the testes rather than inside the sperm-producing cells. And in human clinical trial participants who received intramuscular AAV injections, no vector sequences were found in semen.25Molecular Therapy. Lack of Germline Transmission of Vector Sequences Following Systemic Administration of Recombinant AAV-2 Vector in Males
Despite this reassurance, the theoretical risk has not been dismissed entirely. Early ethical frameworks argued that because our understanding of how gene therapy vectors distribute to reproductive tissue was incomplete, every vector should be tested for its potential to integrate into germ cells.26PubMed. Germline alteration by gene therapy: assessing and reducing the risks That principle has been largely adopted: reproductive toxicology assessments remain part of the regulatory package for AAV gene therapies, even though no confirmed case of germline transmission has been documented in humans.
Why AAV for the Brain Is a Different Conversation
Most of the integration concern centers on the liver, because that is where the highest vector doses tend to accumulate after intravenous delivery. In the nervous system, the landscape is different. Neurons are among the most stable, non-dividing cells in the body, which means episomal AAV genomes can persist for extremely long periods without being lost to cell division. AAV vectors have a strong natural affinity for neurons, which is why they have become the vector of choice for many neurological gene therapies.27PubMed Central. Therapeutic AAV Gene Transfer to the Nervous System: A Clinical Reality In this context, integration is less important for therapeutic durability and the risks associated with it are mitigated by the absence of cell division. The flip side is that any integration event in a neuron is essentially permanent, since the cell is never going to be replaced. Whether that matters clinically over a lifetime remains an open question.
Using Integration On Purpose
While much of the discussion treats AAV integration as an unwanted side effect, some researchers are deliberately harnessing it. AAV vectors can serve as donor templates during CRISPR-guided gene editing, where the cell’s own repair machinery inserts the AAV-carried sequence at a precise location cut by the CRISPR system.28Cell Reports. CRISPR-Mediated Integration of Large Gene Cassettes Using AAV Donor Vectors In this scenario, integration is the goal, not the hazard, and the site is chosen in advance to avoid disrupting important genes. This approach has shown promise for correcting genetic defects in both dividing and non-dividing cells, effectively turning AAV’s ability to enter diverse cell types into a delivery advantage while sidestepping the randomness that makes uncontrolled integration risky.
The distinction between controlled and uncontrolled integration is likely to define much of the next decade in gene therapy design. Uncontrolled insertion happens at low frequency but in unpredictable locations shaped by chromatin accessibility and DNA damage. Controlled insertion, guided by programmable nucleases, happens where you want it but introduces its own risks from off-target cuts by the editing machinery. Neither approach is risk-free, and the field is still working out which trade-offs make sense for which diseases.