Quest for safety at AAValon

In this issue of Blood, Li and colleagues show an exhaustive, long-term safety assessment of adeno-associated virus (AAV) vector integration in mouse liver.¹  Massive amounts of histopathologic, vector integration, and transcriptomic data provide a highly detailed view of the genomic impact of AAV vector integration in vivo. AAV vector integrations are not associated with tumor formation in these settings.

AAV-based vectors are promising tools for the treatment of several genetic diseases. Indeed, depending on the AAV serotype, a single systemic administration allows massive gene transfer, enough for entire organs to obtain therapeutic levels of transgene expression from AAV genomes in episomal form.²  Despite the lack of a professional integrase, AAV vectors are still able to integrate into the host genome at measurable levels³  and, in a mouse model of lysosomal storage disease, these levels were enough to induce a significantly higher incidence of hepatocellular carcinoma (HCC).⁴  Indeed, vector integration analysis showed that 4 HCCs harbored integrations targeting Rian, a nonprotein-coding gene within the complex imprinted Dlk1-Dio3 region, implicated in HCC^5,6  and other cancers.⁷  The finding of a Common Insertion Site (CIS) in different tumors is an important indication that HCCs were triggered by AAV vector–mediated insertional mutagenesis, similar to the mechanisms found int retrovirus and transposon-based oncogene-tagging screenings in mice⁸  and γ-retroviral vector (γRV)–based clinical trials.⁹ 

Thus the risk of insertional mutagenesis, one of the major hurdles of gene therapy with integrating vectors, also applies to AAV vectors. This has implications for the AAV vector–mediated gene therapy field and mandates a thorough assessment of the safety of AAV vector integration.

In the article from Li et al in this issue, 2 groups well known in AAV gene therapy and retroviral integration combined their expertise to perform a very detailed and long-term in vivo safety assessment of an AAV vector proposed for the therapy of hemophilia B.¹⁰  Therapeutic or high AAV vector doses did not increase the incidence of HCC when given systemically to a large number of wild-type mice. Molecular analysis on HCCs and surrounding healthy liver tissue provide a vector integration landscape composed of > 1000 integrations, as demonstrated by vector copy number measurements. A whole transcriptome analysis was preformed in HCC samples. The massive amount of data obtainedwas analyzed by powerful and sophisticated approaches developed and optimized to study vector integration profiles at specific chromatin features, types of targeted genes, CpG islands, and other genomic features. Thus, using a standardized bioinformatic platform it is possible to visually compare the AAV vector, lentiviral vector, and γRV integration profiles. The strength of the study is that coupling the whole transcriptome analysis to the integration site data obtained from HCCs, it was possible to measure the impact of AAV integration on the expression of genes targeted in these tumors in vivo. Importantly, genes in neighboring AAV vector integrations did not display aberrant levels of expression. Overall, the level of detail reached by this analysis sets a new standard for AAV vector safety studies in mice.

It is also possible also to appreciate some peculiar challenges in the study of AAV vector safety that instead do not appear to be a problem with retroviruses. For example, vector integration levels are difficult to estimate as AAV vector genomes coexist both in episomal (mostly) and integrated forms that cannot be easily distinguished. Also, if the integration frequency as determined by Li and colleagues is ∼ 1/1500, can we expect the HCCs to be marked? It is difficult to answer this question because in the complex tumor microenvironment, expanding tumor cells are mixed with bystander cells marked by integrated and/or episomal AVV genomes and unmarked cells, rendering vector marking measurements problematic. How is it possible, then, to distinguish the integrations in tumor cells (if any) from the bystanders? In the HCCs in this study, 1-2 integrations are represented by high numbers of sequencing reads compared with other integrations from the same sample. Could these be the ones that form tumor cells? Ultimately however, the relative contribution of each integration site should be addressed experimentally.

Why in some studies does AAV appear to be genotoxic while in others it does not? The AAV vector from Donsante et al⁴  contained a viral-derived cytomegalovirus early enhancer/chicken β actin promoter driving the expression of the human β-glucuronidase, while Li et al used a cellular promoter chimera composed of the apolipoprotein E enhancer/alpha1-antitrypsin promoter driving Factor IX expression. These differences may be an important to considered in genotoxicity assays to confirm that the use of cellular promoters provides an added safety value to retroviral vectors.¹¹  Other not mutually exclusive variables may be related to an enhanced susceptibility to insertional mutagenesis of the disease model on the mouse strain or environmental factors that may trigger chronic liver damage.¹²  All these factors play an important role in HCC formation and could play a role in the selection of hepatocytes with genotoxic AAV integrations.

It is expected that in the quest for safer AAV vectors, future studies will involve in vivo testing and validation of optimized vector constructs and dissection of the role of genetic and environmental variables in genotoxicity.