Adeno-associated Vector Toxicity—To Be or Not to Be?

Vectors based on adeno-associated virus (AAV) are currently in use in numerous gene therapy clinical trials with reports of benefit for patients, for example in treating hemophilia or inherited blindness. An AAV1 vector (Glybera) has recently been approved as the first gene therapy to achieve marketing authorization in Europe. More than 120 clinical trials using AAV have shown vector safety, with a transient and asymptomatic hepatitis as the most severe side effect. In comparison to the widespread use of AAV vectors in clinical, preclinical, and basic research, there has been limited study of the interaction of the parental virus in humans. Aside from reports suggesting an association between AAV infection early in pregnancy and spontaneous abortions,¹ AAV infection has so far not been associated with illness. However, in a recent issue of Nature Genetics, Nault et al. report the presence of AAV insertions in a total of five different cancer driver genes in 11 of 193 patient samples of hepatocellular carcinoma (HCC) tumors compared with their matched nontumor liver tissue.² Although the presence of AAV sequences in tumor samples may not be very surprising given the high prevalence of AAV infection in humans, the validity of the conclusion for an active role of AAV in tumor formation—in particular, the extrapolation of these results to AAV vectors—, has to be challenged.

AAV is a nonenveloped, single-stranded DNA virus, with a biphasic life cycle, with productive replication in the presence of a helper virus and establishment of a latent infection in its absence.³ Cell culture experiments pointed toward viral genome integration with preference for a specific region on human chromosome 19 (AAVS1), now known as a safe harbor for genetic engineering.³ Of the three initially discovered serotypes (AAV1, AAV2, AAV3), AAV2 was the first to be vectorized. Current systems of AAV vector production use vector genome and packaging signals (ITRs, inverted terminal repeats) built on AAV2, whereas the capsid is chosen depending on the desired cell type or tissue preference and/or antigenic reactivity.³ At least 13 human and nonhuman primate AAV serotypes have been identified so far, and rational-design and directed evolution–based strategies have further expanded the “AAV toolbox,” which comprises a growing number of naturally occurring and engineered capsids into which vector genomes either of the single-stranded (4.5 kilobases, natural) or self-complementary (2.3 kilobases, artificial) design are encapsidated.^3,4

The prevalence of AAV infection in the human population is high, reaching >70% for serotype 2.⁵ Latent infection is established, with liver, bone marrow, spleen, and gut appearing as predominant tissues.⁶ To date, no human pathology has been associated with AAV infection. On the contrary, a protective role of AAV against human papillomavirus–induced carcinomas,⁷ as well as induction of antiproliferative gene clusters upon AAV infection, was reported.⁸ Nault et al. sought to identify factors contributing to the development of HCC in nonfibrotic livers, a rare subset of this disease with frequencies of around 5% of all HCCs.² Upon screening telomerase reverse transcriptase (TERT) promoter sequences, commonly altered in HCC, they detected a 208–base pair (bp) sequence of the wild-type AAV2 3′-ITR upstream of the start codon of TERT. The insertion was accompanied by a 16-bp deletion of the human sequence and a further 7-bp insertion of unknown origin. TERT mRNA was elevated in this HCC sample compared with nonmalignant tissue of the same liver, consistent with a role for telomerase in malignant transformation. Prompted by this finding, researchers screened a total of 193 HCCs and their matched nontumor liver tissue—but no liver biopsy samples from healthy subjects—for the presence of AAV. The majority of these tissues (147/193) lacked AAV sequences. However, in 11 of the 193 HCC tumors, AAV insertions were found in five different cancer driver genes (TERT (1/11), cyclin A2 (CCNA2) (4/11), cyclin E (CCNE1) (3/11), tumor necrosis factor superfamily member 10 (TNFSF10) (2/11) and lysine-specific methyltransferase 2B (KMT2B) (1/11)).

The authors' interpretation of these data requires scrutiny. In total, 11 HCCs were found to harbor deleted AAV sequences, although only 6 of these 11 were found in patients lacking other known risk factors such as hepatitis virus B or C infection or alcohol consumption (Supplementary Table 2 in the article by Nault et al.). Notably, 4 of these 6 HCCs presented AAV-independent HCC-related mutations. Thus, only 2 HCCs were identified with AAV as the single detected “mutation.” It is not shown in detail whether these 2 HCCs also showed significant HCC gene–related RNA overexpression (Figures 1b and 2b in the article by Nault et al.). Thus, it remains to be determined whether AAV is simply a passenger mutation rather than a driver mutation in HCC formation.

The extrapolation of the findings to suggest AAV-related genotoxicity must also be challenged. A complete virus genome was detected in none of the samples. Indeed, the fragments with homology to AAV were all rather small. Furthermore, no distinct integration pattern emerged, with insertions appearing upstream of a promoter sequence (1/11), within an intron (5/11), within an exon (2/11), or within the 5′- or 3′–untranslated region (3/11). Thus, in the absence of a complete AAV genome and a more unifying pattern, it appears a bit exaggerated to attribute any genotoxicity to AAV rather than to other foreign DNA (if any) integrated into these regions.

Two recent publications have confirmed preference for AAV2 to integrate into the AAVS1 region on chromosome 19, which is caused by the accumulation of Rep binding sites and terminal resolution site (trs)-like motifs within this region.^9,10 It is thus surprising that no AAV sequences were found within the AAVS1 region in either the HCCs or the matched nonmalignant tissue,² even though wild-type AAV has an active Rep integrase targeting this site.

Finally, the implications of these results for the use of AAV vectors in the clinic appear quite limited. Current AAV vectors are “gutless,” i.e., devoid of all known viral open reading frames, leaving the ITRs as the sole viral sequences in cis. The wild-type AAV2 3′-ITR is may be the only common denominator in the reported HCC integrations, as it was detected in 10 of the 11 cases.² This sequence, however, is not present in the commonly used AAV vector genomes, as during the cloning procedure for AAV's vectorization parts of the wild-type 3′-ITR were replaced by the 5′-ITR.¹¹

Furthermore, AAV vector genomes tend to persist episomally as concatemeric structures.³ Although numerous groups have demonstrated the safety of AAV vectors despite rare integration events,^{12,13,14,15,16,17} a handful of preclinical studies have suggested a genotoxic potential for AAV vector integration within hepatic tissue under specific circumstances.^18,19,20 A thorough analysis of experimental variables identified neonatal injection of a high dose of hepatotropic AAV8 vectors in combination with a strong enhancer/promoter sequence as critical parameters,¹⁹ arguing against an intrinsic ITR-related genotoxicity. The sensitivity of neonatal mice to insertion at Rian, identified as an integration hot spot, was correlated with accessibility of Rian at this stage of development. Insertion at Rian per se—even in neonatal mice—did not appear to be genotoxic on its own, but rather required use of a strong promoter/enhancer.¹⁹ Of further note, integrations that were found to be associated with HCC in mice clustered at the Mir341 locus in Rian, which lacks a human orthologue.¹⁹

None of the other existing preclinical and clinical AAV studies has demonstrated any severe adverse events.^{13,17,21,22,23} Associated integration site studies also did not find any relevant integration preferences in HCC-susceptible gene regions. Accordingly, recently presented integration data from different biopsy specimens from nonhuman primate tissues (liver, spleen, adrenal gland) after intravenous administration of clinical AAV5-cohPBGD vectors (unpublished clinical data) revealed no unwanted side effects or integration at these loci.²⁴ To conclude, we do not aim to challenge the authors' findings of partial wild-type AAV2 sequences in 11 out of 193 cases of human HCCs. However, we believe that the implications of the findings are significantly overestimated, at least with regard to AAV vectors, whose genome sequences are profoundly distinct from that of wild-type AAV2. The vast majority of preclinical and all clinical studies have shown the genome-wide distribution of rare AAV vector integrations without any genotoxicity or preference for the previously reported HCC-associated genes, emphasizing the very limited genotoxic risk of AAV vectors.