rAAV-Mediated Tumorigenesis: Still Unresolved After an AAV Assault

The risk of oncogenesis mediated by vector-induced insertional mutagenesis during therapeutic gene transfer has received much attention in recent years. Any nucleic acid, regardless of how it is delivered, can cause insertional mutagenesis if it integrates into the genome. The two parameters that define this risk are integration site preference and frequency of integration. Recombinant adeno-associated viral (rAAV) vectors have been shown to be safe and efficacious in early gene therapy clinical trials,^1,2,3,4 although such vectors do integrate into the genome at a low but measureable rate (0.1 to 1% of transduction events) in animal models.^5,6 In this issue of Molecular Therapy, Rosas and colleagues test various conditions that are hypothesized to facilitate rAAV integration so as to determine whether these events can lead to an increased rate of oncogenesis.⁷

The issue of oncogenesis in rAAV-mediated gene therapy became more than just a theoretical concern when tumors were identified in neonatal β-glucuronidase-deficient mice treated with rAAV.⁸ Four of these tumors were found to contain integration sites within the imprinted Dlk1-Dio3 locus on mouse chromosome 12qF1, which led to the development of hepatocellular carcinoma (HCC) in a manner independent of β-glucuronidase status.⁹ To confirm causality, an rAAV vector designed to integrate into the Dlk1-Dio3 locus via homologous recombination was shown to be able to replicate the HCC phenotype.¹⁰

Although rAAV-mediated integration is much lower than that seen with either retroviral/lentiviral or transposon vectors, it is much greater than that observed with plasmid or adenoviral vectors. Moreover, it has been shown that rAAV proviral genomes preferentially integrate near or within transcriptionally active genes.^11,12,13 There is a long-standing debate regarding whether these integrations significantly enhance the risk of oncogenesis,^{8,9,11,14,15,16} which has been exacerbated by the existence of a number of confounding variables in previous studies that have yet to be fully controlled for (for discussion see ref. 14).

In the new study, the authors generated a self-complementary AAV (scAAV) vector that included promoter and enhancer sequences but lacked coding or poly­adenylation sequences (CBA-null), which was specifically designed to enhance readthrough transcription into neighboring genes. Integration of this vector was compared to that of a more typical gene transfer vector that included coding and polyadenylation sequences (scAAV-CMV-GFP-pA). To further promote vector-induced oncogenesis, a subset of mice was treated with camptothecin and/or underwent partial hepatectomy. Camptothecin is a DNA double-strand break–inducing agent that should increase proviral DNA integration,¹⁷ whereas a surgical partial hepatectomy drives hepatocellular regeneration, which has several implications for potential integration (described below). Two murine strains were evaluated: C3H/HeJ mice, which develop HCC at a high frequency (30–50% of males)¹⁸; and severe combined immunodeficient (SCID) mice, which lack the catalytic subunit of DNA-PK.¹⁹ The latter mice exhibit impaired B- and T-cell maturation resulting in a lack of cell-mediated and humoral-adaptive immune responses, as well as impaired DNA double-strand break repair. Both of these para­meters have been shown to increase the rate of AAV integration.²⁰ This combination of factors is akin to a pilot in a flight simulator trying to land a plane while various safety overrides are systematically inactivated.

So, was there a safe landing? In short, yes, although perhaps with some turbulence. In all but one condition, rAAV sequences did little to promote tumorigenesis (Figure 1). A large number of mice (n ≈ 25 per group) were used to deduce significant increases in tumor incidence in each test group. SCID mice produced very few tumors; the same can be said for female C3H mice and mice treated with camptothecin. However, male C3H mice, which are prone to develop HCC, did show an accelerated and statistically significant increase in tumor incidence in the liver following delivery of either scAAV-CMV-GFP-pA or CBA-null rAAV vectors. Tumors from CBA-null-treated animals contained identifiable rAAV integration sites, suggesting that rAAV accelerated the tumor phenotype by inducing additional oncogenic events. Importantly—and boding well for prevention of toxicity—no tumors were identified in any of the 12 additional tissues examined. This was consistent with previous studies in which tumor sites were limited to the liver.¹⁰ However, the absence of tumors in nonhepatic tissues may reflect low levels of transduction in other tissues following intravenous infusion of rAAV8 vectors.

Figure 1.

Overview of the rate of oncogenesis and the identified integrated genes. rAAV, recombinant adeno-associated virus; SCID, severe combined immunodeficient.

Another arm of the study yielded inconsistent and inconclusive results as to whether partial hepatectomy had an effect on tumor progression. Partial hepatectomy can enhance tumorigenesis both by enabling clonal expansion of cells with oncogenic integrations, and by inducing cell division and corresponding DNA replication that could promote viral integration. However, a two-thirds partial hepatectomy results in all hepatocytes undergoing one or two cell divisions resulting in the loss of ~90–95% of rAAV genomes.²¹ If performed at an early time point (16 hours post-rAAV infusion in this case), a two-thirds partial hepatectomy might decrease the total number of integration events. However, it should be noted that in the current study the degree of liver regeneration was probably less robust, as less than 50% of the liver was removed. Nonetheless, in control mice injected with either saline or AAV-CMV-GFP, partial hepatectomy led to a decrease in tumor incidence, whereas a small increase in tumor frequency was observed in CBA-null–treated animals.

These conflicting results are also reflected in the vector genome analysis. The mean vector copy number (per diploid genome) in tumor samples was higher in tumors from post-hepatectomy CBA-null–treated animals than in tumors from the AAV-CMV-GFP–treated control groups, suggesting that CBA-null integration was more often associated with tumor formation. However, the mean vector copy number in tumors from AAV-CMV-GFP control mice post-hepatectomy and post–camptothecin + hepatectomy indicated less than one copy of vector genome per cell, possibly due to a “hit-and-run” phenom­enon. In a hit-and-run event the evidence of an integration event can disappear if the region surrounding the proviral AAV genome is lost for any reason (e.g., recombination). Second, the number of nonmalignant cells contained within the tumor will affect the integration signal within a malignancy. Interestingly, the vector copy number reported at 2 weeks post-transduction was ~100 times higher than the amount at the end of the study, suggesting that the C3H/HeJ liver undergoes substantial regeneration and loss of episomal DNA relative to a normal liver even without surgical partial hepatectomy. This effect is therefore partly redundant with the partial hepatectomy, where vector copy number decreased ~10 times from control mice treated with the same vectors but without induced liver regeneration.

The integration locations are also of particular interest. Four integration events occurred in the Hras promoter region, whereas three were in intron 8 of Sos1, leading to activation of downstream, but not upstream, Sos1 exons. Since the transactivation domain is at the beginning of the Sos1 gene, the presumed novel gene product would be devoid of this regulatory domain and remain constitutively active. Two events occurred at the Fgf3 gene promoter in an antisense orientation, nevertheless leading to Fgf3 activation. Finally, single integration events took place near or within the Fgf10, Raf1, Safb2, Ctnnb1, and CopZ1 genes. The induction of cell division could influence genes such as Fgf3 and Fgf10, which are typically silenced in the adult liver and presumably have a closed chromatin state, to become actively expressed and have the potential to be exposed to rAAV integration events.

Does this mean that these are hot spots of oncogenesis? Considering that the authors did not detect multiple integration sites per tumor, it appears that these sites are particularly prone to integration events. However, the C3H mice, which already commonly develop HCC, do so through activating mutations at codon 61 of the Hras gene.²² Thus, it would be relevant in this study to sequence Hras in these tumor samples to determine if compounding mutations were present or required for transformative growth. The appearance of Hras mutations in C3H mice further suggests that Hras is permissive in this model to DNA damage, and that integration at this site is less likely to be a function of the viral vector sequence. Without a comprehensive evaluation of the effect of gene activation, it remains unclear whether the integration sites identified in this study are indeed oncogenic, or whether additional integration events took place that were not detected by linear amplification–mediated polymerase chain reaction. Furthermore, some integration events may have prompted local genomic rearrangements during DNA repair. This could disrupt endogenous genes while eliminating any trace of the offending viral vector. This situation may account for the lack of a detectable integration event in the majority of tumors (85 of 102) isolated from rAAV-infected mice. The advent of RNA sequencing technologies and their diminishing cost will help enable discovery of some of these “missing” integration events in the future.

Does this study settle the question of whether rAAV is carcinogenic? In some respects, it diminishes the fear of rAAV-mediated toxicity. The only mice that exhibited an increased tumor burden were those that were already predisposed and those that have been shown to develop tumors in up to 80% of males, depending on additional treatments.²³ This raises the issue of whether rAAV should be used to treat disorders for which the disease pathogenesis is already oncogenic (e.g., viral hepatitis). Several variables can potentially influence the rate of oncogenesis, most notably the sex and age of the mice (e.g., neonates vs. adults); the transgene sequence and choice of promoter; the vector serotype, dose, and delivery route; the genetic background and environmental living conditions. Even if one could control for these variables, it is not clear how transferable these results are to human clinical studies. However, this study encourages the continued study of “safe harbor” sites where integration events have little effect on the transcription of neighboring genes²⁴ or are directed into known benign genomic locations such as the widespread rRNA loci.^25,26 Additionally, this study has added to the list of genomic locations at which integration sites should be monitored and avoided.

It is estimated that there are 300 billion hepatocytes in the human liver, and if one assumes a vector dose in which 10% of the hepatocytes are transduced, even with an integration rate of 0.1%, a single individual will have ~30 million hepatocytes with at least one integration event—not a trivial number. However, there is no way, at present, to correlate the number of these events with the risk of oncogenesis. Overall, it seems logical that proviral integration will provide some risk, but overall animal studies continue to suggest that this risk is relatively low. It appears that additional flight simulations are required to ensure a safe landing for rAAV delivery in the future.