AAV vector integration and insertional oncogenesis: what it changes for pharmacovigilance

On 6 November 2026 the FDA announced that it is investigating a serious risk of neoplasia, including malignancy, in patients treated with Zolgensma (onasemnogene abeparvovec) and its intrathecal counterpart Itvisma. The communication includes a sentence the gene therapy field has been bracing for since the spring: the agency has received "confirmed reports of product-related insertional oncogenesis involving the GLI1 … gene" [1].

Recombinant adeno-associated virus (AAV) vectors have been used clinically for more than two decades on the understanding that they are predominantly non-integrating. That understanding has not been overturned. What has changed is that integration-driven oncogenesis has moved from a theoretical, label-level caveat to a documented, if rare, human event, described in three separate programmes within twelve months. For pharmacovigilance teams, that shifts a set of questions from the risk-management plan's "important potential risk" column into live signal management.

This article sets out what is established, what is not, and what safety teams supporting AAV programmes should be doing now.

What the Zolgensma and Itvisma cases show

FDA's description is specific. Post-marketing reports of tumour formation after Zolgensma include tumours of the central and peripheral nervous systems, the spinal and paraspinal regions, and the kidney. Three cases involved tumours associated with the GLI1 oncogene. In two of those, molecular analysis of tumour tissue showed vector DNA integrated into the host genome, with the promoter region of the onasemnogene abeparvovec construct inserted upstream of or within GLI1. The mechanism of GLI1 overexpression in the third case remains under investigation [1].

FDA has applied the risk to both products, because Zolgensma and Itvisma contain the same vector and construct, and it states that it is monitoring other similar AAV programmes. There have been no reported cases of neoplasia after Itvisma to date. The agency's conclusion on benefit-risk is unchanged: the overall benefits of both products continue to outweigh their potential risks for their approved uses, while it evaluates whether regulatory action is appropriate, and whether the current prescribing information and long-term follow-up studies are adequate [1].

The day before, Novartis provided a statement through SMA Europe: more than 6,000 patients have been treated with Zolgensma across clinical trials, managed access programmes and commercial use; since first approval in 2019 the company has received nine reports of tumours in previously treated patients; two of those cases to date have shown evidence of vector DNA integration into the patient's DNA; both patients were treated and continue to be monitored. Novartis notes that the potential risk of tumours due to vector integration is already in the prescribing information, and that health authorities were notified in each case [2].

Two points deserve emphasis. First, the label already carried the risk, described as theoretical: the US prescribing information for both products refers under Warnings and Precautions to a "theoretical risk of tumorigenicity due to integration of AAV vector DNA into the genome" [1]. The new information is not that the hazard was unanticipated; it is that it has been observed. Second, nine tumour reports across more than 6,000 treated patients is not an incidence rate: spontaneous reports have no denominator discipline, under-reporting is certain, and the background tumour rate in this population is not zero. The figures tell us the events occur, not how often.

The other two human programmes

The FDA communication explicitly cites two published reports of AAV integration-driven oncogenesis in other programmes [1].

RGX-111 (PLAG1). A five-year-old boy with severe MPS I (Hurler subtype) developed a neuroepithelial tumour four years after intracisternal magna administration of an AAV9 vector carrying IDUA. Molecular analysis showed clonal integration of rearranged AAV vector elements into PLAG1 and expression of a chimeric AAV-PLAG1 transcript. The tumour was resected successfully and the child has continued to show advanced cognitive function for his age, indicating mitigation of the underlying disease [3]. REGENXBIO disclosed the case on 28 January 2026, when FDA placed RGX-111 on clinical hold and extended the hold to RGX-121 on the grounds of product, population and risk similarity [5]. The child had also received a cord-blood stem-cell transplant at four months of age before the gene therapy, a confounder that has to be weighed in the causality assessment [3].

FBX-101 (GLI1). A three-year-old boy with infantile Krabbe disease received intravenous FBX-101, an AAVrh.10 vector carrying GALC, at 3×10¹³ vector genomes/kg, 29 days after a second myeloablative umbilical-cord-blood transplantation. Twenty-six months later, surveillance MRI found an incidental L2–L3 paraspinal mass, diagnosed as a GLI1-altered neoplasm. Long-read sequencing identified a clonal 1,038-bp insertion in GLI1 intron 1 present in 44% of reads. The integrant was a subgenomic vector fragment — partial AAV2 ITR, CMV enhancer, chicken β-actin promoter, exon 1 and a splice donor — with no GALC transgene and no polyadenylation signal, raising the possibility that a promoter-only species arising during manufacturing was the integrating substrate. Allele-specific hypomethylation confined to the integrant allele established the integration as the proximate cause of GLI1 activation. The patient remains asymptomatic [4].

Both integrants were truncated, rearranged subgenomic structures that retained a CAG promoter-enhancer and a splice donor, and both activated a developmental transcription factor [4]. That shared mechanistic pattern is the most actionable finding in the published literature for vector designers.

A related but separate signal sits alongside these. On 24 August 2026 FDA placed RGX-121 on clinical hold after an expanded MRI monitoring plan, introduced following the RGX-111 hold, found a small nodule or small cystic mass on spine MRI in five of 48 CAMPSIITE participants dosed three to six years earlier. Investigators deemed the findings non-serious and radiologists considered them likely benign; no brain nodules or masses were found; and because spine MRI is not routinely performed in MPS, the background prevalence of such findings is unknown [6]. Intensified surveillance generates findings of uncertain significance, so a monitoring plan needs a pre-specified adjudication pathway.

The cases that were investigated and cleared

Causality assessment cuts both ways, and the two haemophilia examples are the field's best worked examples of an integration analysis that exonerated a product.

Etranacogene dezaparvovec (Hemgenix). A HOPE-B participant was diagnosed with hepatocellular carcinoma about one year after dosing. Integration analysis found vector DNA in a very small fraction of cells with randomly distributed, non-dominant integration sites and no evidence of clonal expansion, while whole-genome sequencing of the tumour showed chromosomal and gene alterations typical of HCC and independent of vector integration. The patient had pre-existing risk factors, including hepatitis B and C and steatotic liver disease. The published conclusion was that the HCC was highly unlikely to have been caused by the gene therapy [8].

Valoctocogene roxaparvovec (Roctavian). A participant in the phase 1/2 study was diagnosed with a grade 2 parotid gland acinar cell carcinoma in year six. Target enrichment sequencing of tumour and adjacent healthy parotid tissue found low integration frequency (8.25×10⁻⁵ per diploid cell), no excess of integrations in tumour tissue, no shared insertions between tumour samples, and no insertion plausibly driving tumorigenesis [13]. A separate case of B-cell acute lymphoblastic leukaemia reported in the programme was also assessed as unrelated to the study drug [24]. Integration profiling in liver biopsies from five treated participants found that roughly 99% of vector genomes persisted episomally, with polyclonal, low-abundance integration and no evidence of clonal outgrowth [14].

The methodological lesson is that a malignancy after AAV therapy is an assessable event, not an unanswerable one — provided tissue is available.

The animal and natural-history evidence

The human cases did not arrive without warning.

Neonatal mice. AAV integration at the Rian locus within the imprinted Dlk1-Dio3 region was linked to hepatocellular carcinoma in neonatal mice in 2007 [11]. The most comprehensive follow-up study found that integrations clustered within a small genomic window in Rian, in or near miR341, correlated significantly with HCC, and that vector dose, enhancer/promoter selection and the timing of delivery were all determinants of HCC incidence [9]. No human orthologue of miR341 has been identified, and the mouse finding has not translated to older mice without pre-existing liver disease, dogs or non-human primates.

Haemophilia A dogs. Nine dogs were followed for up to 10 years after AAV8 or AAV9 vectors expressing canine factor VIII. Analysis of liver samples from six dogs identified 1,741 unique integration events and expanded cell clones in five dogs, with 44% of integrations near genes involved in cell growth. Two dogs showed gradually rising factor VIII levels from about four years after treatment, plausibly reflecting clonal expansion of transduced cells. No dog developed a tumour or altered liver function. The authors concluded that the findings support clinical development while emphasising long-term monitoring for genotoxicity [7]. It remains the clearest demonstration that integration can confer a proliferative advantage in a large animal without producing cancer.

Wild-type AAV2 in human liver cancer. Clonal wild-type AAV2 insertions were found in 11 of 193 human hepatocellular carcinomas, in known cancer driver genes — CCNA2 (four cases), CCNE1 (three), TNFSF10 (two), TERT (one) and KMT2B (one) — with overexpression of the targeted genes. Nine of the 11 tumours arose in non-cirrhotic liver and six in patients with no known risk factor [10]. This is natural infection, not gene therapy, and clinical vectors lack the wild-type rep gene, but it establishes that AAV sequences can act as insertional mutagens in human cells.

FDA's 2021 advisory committee. At the Cellular, Tissue, and Gene Therapies Advisory Committee meeting on 2–3 September 2021, oncogenicity due to integration was one of the central topics. The briefing document stated plainly that animal studies suggest a potential for oncogenicity but that, at that time, AAV integration had not been associated with cancer in humans or non-rodent species. The committee concluded that animal models have limited predictive value, that study designs should recapitulate the intended clinical use and run long enough to detect tumours, that integration and clonal expansion analysis methods need standardisation, and that an arbitrary upper limit on total vector genome or capsid dose was not appropriate [12]. Five years later, the first clause of that briefing statement no longer holds.

The examples at a glance

AAV against the integrating vectors

The lentiviral comparison matters because it is the field's reference class for insertional oncogenesis, and because the differences are as instructive as the similarities.

AAV's risk profile is different in kind, not merely in degree: fewer integration events, in cells less able to propagate them. That is why the events are rare, and not a reason to treat them as impossible.

Regulatory long-term follow-up: FDA and EMA

A practical consequence: the five-year AAV window was designed around persistence and transgene expression, not around latency to a solid tumour. In the reported cases the interval from dosing to diagnosis was 26 months and four years, against a product licensed since 2019. Five years is not obviously wrong, but it is no longer obviously sufficient, and sponsors should expect that question.

Pharmacovigilance implications

Signal detection. Malignancy after gene therapy is a designated medical event in all but name: it should be flagged at intake, not left to disproportionality analysis. Spontaneous reporting systems detect these events poorly, because latency is long, the treating oncologist is rarely the treating gene therapy centre, and the paediatric populations involved are small and geographically dispersed. Practical measures include a standing query across the relevant MedDRA SMQs and neoplasm HLGTs, targeted follow-up questionnaires triggered on any neoplasm term, and reconciliation against LTFU study data, registries and managed-access programmes rather than the safety database alone. Anything that could be a tumour — including an incidental imaging finding — needs a documented adjudication route.

Causality assessment. Integration-site analysis is now a causality tool, and the published cases define what good looks like: long-read sequencing of tumour tissue, demonstration of clonality and abundance, characterisation of the integrant's structure, RNA evidence of a chimeric transcript, comparison of target-gene expression against a reference cohort, exclusion of alternative oncogenic drivers, and, where available, epigenetic evidence tying activation to the integrant allele [3][4]. The counterpart is the exculpatory pattern: low integration frequency, random distribution, no enrichment in tumour versus adjacent tissue, no shared insertion across tumour samples, and a mutational profile typical of the tumour type [8][13]. Confounders must be assessed explicitly, not waved away: two of the three index cases involved prior myeloablative conditioning, transplantation or immunosuppression [3][4].

Long-term follow-up and registries. Sponsors should re-examine whether their LTFU protocol can actually detect the event: whether the duration matches plausible latency, whether assessments include imaging of the most heavily transduced tissues where justified, whether retention will keep paediatric patients enrolled into adolescence, and whether registries such as the WFH Gene Therapy Registry, which follows participants annually for 15 years [22], extend reach beyond the sponsor's own cohort.

RMP and REMS. For AAV products, "tumorigenicity due to vector integration" has typically sat in the EU RMP as an important potential risk with routine pharmacovigilance plus an LTFU study. The characterisation has changed, and so should the documentation: an upgrade to an identified risk should be considered where the evidence supports it, with the LTFU study described as an additional pharmacovigilance activity and the effectiveness of educational materials reassessed. In the US, FDA has said it is evaluating whether regulatory action is appropriate [1]; sponsors should prepare labelling and risk-communication options rather than wait. Class-level questions are unavoidable: FDA has stated it is monitoring other similar AAV programmes, and a sponsor whose construct shares a CAG promoter-enhancer, a splice donor, a high dose or a paediatric population should be able to explain its own position before being asked.

Tissue sampling and biobanking. This is the most under-appreciated action and the cheapest to fix. None of the causality assessments above would have been possible without tumour tissue. Protocols, informed consent and LTFU plans should provide for tumour tissue collection and banking, adjacent normal tissue as comparator, baseline and interval surrogate samples such as peripheral blood, and a pre-agreed analytical laboratory with validated integration-site methodology. Sponsors should also write this into their safety-event response plan, so that a treating centre is told what to preserve on the day a tumour is found.

Labelling, DSUR and PSUR. Where a label describes the risk as theoretical, that wording is no longer accurate for the affected products and should be reviewed for others. Product-specific surveillance wording already exists as a precedent: the Hemgenix US prescribing information directs annual abdominal ultrasound and alpha-fetoprotein testing for five years in patients with pre-existing HCC risk factors [23]. Periodic reports should carry a dedicated section on malignancy and integration: cumulative case listings with latency, the integration analyses performed and their results, confounders, LTFU enrolment and retention figures, literature on the class, and an explicit benefit-risk statement. DSURs for programmes in development should address the implications of the published cases for the construct in question, even where no case has occurred. Expect regulators to ask what has changed in the development safety profile since 6 November 2026.

Vector design and quality interfaces. Both published integrants were truncated, rearranged subgenomic species retaining a promoter-enhancer and a splice donor [3][4]. The FBX-101 authors argue for splice-donor-free cassette design and attention to vector purity [4]; the RGX-111 investigators argue for the lowest feasible dose and tissue-specific promoters [3]. These are CMC and non-clinical decisions with direct pharmacovigilance consequences, and they are a reason for safety to sit in product-design discussions rather than receive the construct as a given.

What PV teams should do now

Established, uncertain, and unknown

It is established that AAV vector DNA can integrate into the human genome, that integration can activate an oncogene, and that this has now been demonstrated in tumour tissue from patients in at least three programmes [1][3][4]. It is established that integration occurs at low frequency, that the large majority of vector genomes remain episomal, and that clonal expansion can occur without transformation [7][14].

It is uncertain how often integration-driven oncogenesis occurs, because there is no reliable denominator and no systematic genomic surveillance of treated patients. It is uncertain how much of the risk is attributable to vector design, dose, age at dosing, target tissue or co-interventions such as myeloablative conditioning, whether a five-year window captures the relevant latency, and how far the findings generalise across capsids, routes and constructs.

What is not in doubt is that the benefit side remains substantial. Zolgensma treats a disease that was, until recently, the leading genetic cause of infant death, and FDA's position is that benefits continue to outweigh risks for approved uses [1]. The right response to these cases is better evidence — tissue, analysis, follow-up and transparent communication — rather than retreat.

References

1. FDA: Investigating Serious Risk of Malignancy Linked to Zolgensma Gene Therapy, 6 Nov 2026
2. SMA Europe: Information regarding recent Zolgensma safety concerns, including the Novartis statement, 5 Nov 2026
3. Ahrens-Nicklas RC et al. Neuroepithelial Tumor with AAV Integration after Intracisternal Magna Vector Delivery. N Engl J Med 2026;394:2126-2133
4. Vo JN et al. Integration-Associated Paraspinal Neoplasm after AAV Gene Therapy. N Engl J Med 2026;395:1555-1558
5. REGENXBIO: Regulatory update on ultra rare MPS programs, 28 Jan 2026
6. REGENXBIO: Regulatory update on RGX-121 for MPS II, 24 Aug 2026
7. Nguyen GN et al. A long-term study of AAV gene therapy in dogs with hemophilia A identifies clonal expansions of transduced liver cells. Nat Biotechnol 2021;39:47-55
8. Schmidt M et al. Molecular evaluation and vector integration analysis of HCC complicating AAV gene therapy for hemophilia B. Blood Adv 2023;7:4966-4969
9. Chandler RJ et al. Vector design influences hepatic genotoxicity after adeno-associated virus gene therapy. J Clin Invest 2015;125:870-880
10. Nault JC et al. Recurrent AAV2-related insertional mutagenesis in human hepatocellular carcinomas. Nat Genet 2015;47:1187-1193
11. Donsante A et al. AAV vector integration sites in mouse hepatocellular carcinoma. Science 2007;317:477
12. FDA Cellular, Tissue, and Gene Therapies Advisory Committee, 2-3 September 2021: briefing document on toxicity risks of AAV vector-based gene therapy
13. Symington E et al. Long-term safety and efficacy outcomes of valoctocogene roxaparvovec gene transfer up to 6 years post-treatment. Haemophilia 2024
14. Russell CB et al. Recombinant AAV integration profiles in nonhuman primates and gene therapy participants after treatment with valoctocogene roxaparvovec. Hum Gene Ther 2025;36:945
15. FDA: Investigating Serious Risk of Hematologic Malignancy Following Skysona (elivaldogene autotemcel)
16. FDA: Approves required labeling changes for increased risk of hematologic malignancy following treatment with Skysona
17. bluebird bio: Updated findings from reported case of AML in the LentiGlobin for sickle cell disease programme, 2021
18. FDA: Long Term Follow-up After Administration of Human Gene Therapy Products, Guidance for Industry, January 2020
19. Cancer cases in bluebird's gene therapy trials. Nat Biotechnol 2024
20. EMA: Guideline on follow-up of patients administered with gene therapy medicinal products (EMEA/CHMP/GTWP/60436/2007)
21. Regulation (EC) No 1394/2007 on advanced therapy medicinal products
22. World Federation of Hemophilia Gene Therapy Registry (NCT04883710)
23. Hemgenix (etranacogene dezaparvovec) US prescribing information, DailyMed
24. BioMarin: Molecular characterisation of recombinant AAV5 encoding FVIII after human administration (reported malignancy cases)

Safety Strategy · Pharmacovigilance and safety risk management · Vaud, Switzerland

---

This article is for information only and is not medical or regulatory advice.

Related services: signal management and benefit-risk assessment, risk management (EU RMP, US REMS, aRMMs) and regulatory filings including DSURs.

Follow Safety Strategy on LinkedIn (opens in a new tab) for new articles and updates. Back to all news