Adeno-associated virus vector integration

Abstract

Adeno-associated virus (AAV) vectors efficiently transduce various cell types and can produce long-term expression of transgenes in vivo. Although AAV vector genomes can persist within cells as episomes, vector integration has been observed in various experimental settings, either at non-homologous sites where DNA damage may have occurred or by homologous recombination. In some cases, integration is essential for the therapeutic or experimental efficacy of AAV vectors. Recently, insertional mutagenesis resulting from the integration of AAV vectors was associated with tumorigenesis in mice, a consideration that may have relevance for certain clinical applications.

Introduction

Adeno-associated virus (AAV) vectors are able to transduce a wide range of dividing and non-dividing cell types, which has made these vectors an important tool for gene therapy. A major advantage of AAV vectors is the long-term expression of transgenes that can be obtained after in vivo gene delivery. In many cases, AAV vectors integrate into the host cell genome, which can be important for certain applications, but can also have unwanted consequences. Current concepts of AAV vector integration are reviewed, including the mechanisms of integration, the sites of integration and the potential clinical consequences.

The integration of wild-type AAV

AAV is a helper-dependent parvovirus with an approximately 4.7 kb, single-stranded, linear DNA genome that contains two open reading frames encoding the replication and capsid proteins (rep and cap, respectively) [1,2]. The AAV genome includes inverted terminal repeats (ITRs) that fold into characteristic T-shaped hairpins, which are the only sequences required in cis for the replication and packaging of the AAV genome. Co-infection with a helper virus such as adenovirus or herpesvirus is required for the replication of AAV and for productive infection. In the absence of a helper virus, AAV can produce a latent infection in which the viral genome persists in infected cells. AAV genomes have been recovered from lung tissue [3-5], as might be expected given the respiratory route of infection. Remarkably, AAV has also been detected in other tissues derived from humans and non-human primates, including bone marrow, brain, spleen, colon, heart, liver, lymph node and kidney [5-7], suggesting that the virus spreads throughout the body during infection. While the presence of these latent genomes is consistent with integrated proviruses, and in some cases Southern blot analysis has suggested the occurrence of integration [6], episomal genomes may also persist for prolonged periods.

An interesting feature of wild-type AAV is that the genome can integrate in a site-specific manner in a region on the long arm of chromosome 19 (19q13-qter), termed the AAVS1 site [1,8,9]. Although large regions of homology do not exist between the AAVS1 site and the AAV genome, AAVS1 contains an active Rep binding element and the AAV Rep protein (a DNA helicase) is required for site-specific integration of the vector. Thus, Rep-mediated replication at AAVS1 may promote DNA repair processes that allow the proviruses to integrate [10]. Approximately 0.1% of infecting wild-type AAV genomes integrate at AAVS1 [1]. Although accounting for the integration of a small fraction of viral DNA, this mechanism could nonetheless play a role in the natural history of the virus, and AAV proviruses have been identified in AAVS1 after natural infection. In two individuals, sequence analysis of the junctions between the vector and chromosome sequences demonstrated that integration had occurred in testicular tissue [11].

Synthetic AAV vectors have also been designed to integrate at the AAVS1 site by the same Rep-dependent mechanism. However, the rep gene is toxic and its inclusion in the vector significantly reduces the viral packaging capacity. Progress with rep-containing vectors will not be discussed. The wild-type AAV genome can also integrate at sites in the genome other than AAVS1, as demonstrated by the identification of a provirus on chromosome 1q31.1 in human tonsillar tissue and in multiple different chromosomes in vitro [3,12]. These events were presumably Rep-independent and are analogous to the integration of rep-deficient vectors described in this review.

Nuclear processing of AAV vector genomes

Most AAV vectors contain a transgene expression cassette between the ITRs that replaces the rep and cap genes. Vector particles are produced by the co-transfection of cells with a plasmid containing the vector genome and a packaging/helper construct that expresses the rep and cap proteins in trans [13]. During infection, AAV vector genomes enter the cell nucleus and can persist in multiple molecular states. One common outcome is the conversion of the AAV genome to a double-stranded circular episome by second-strand synthesis or complementary strand pairing [1,2,14-17]. The episomes can be converted to high-molecular-weight concatamers that are observed as tandem repeats [2,18,19]. These multimeric concatamers are believed to be formed by the recombination of monomer genomes and can provide long-term transgene expression, particularly in non-dividing cells [19-22]. AAV vectors can also integrate at non-homologous sites in the host genome, either as single-copy proviruses or concatamers [2,23-25]. Approximately 0.1% of vector genomes that enter cells integrate in this manner [1,26]. Although the frequency of integration by AAV vectors is significantly reduced compared with that of retroviral vectors, integration is sufficiently high to produce many integrated vector proviruses in most experiments. In some cases, both episomal and integrated vector genomes appear as high-molecular-weight DNA, and can be challenging to distinguish. Transgenes can be expressed from either episomal or integrated vector genomes. AAV vectors can also integrate at specific sites in the genome by homologous recombination (see the Homologous recombination section) [27].

Non-homologous vector integration

Early studies of AAV vector integration used marker genes to select for cells that contained integrated proviruses. Southern blot, PCR and fluorescence in situ hybridization analysis of transduced human cell lines suggested that AAV vectors integrated at apparently random genomic sites [28-32]. In subsequent studies, AAV shuttle vectors were used to rescue integrated proviruses in the form of bacterial plasmids in order to sequence vector:chromosome junctions. Russell and colleagues analyzed 9 integration sites isolated from HeLa cells [23,33] and 977 sites from unselected human fibroblasts [34], and Kay and colleagues analyzed 18 sites from the livers of wild-type mice [35] and 347 sites from the livers of mice with hereditary tyrosinemia, in which transduced hepatocytes selectively proliferated [36,37]. More recently, Inagaki et al characterized approximately 1000 integration sites from the liver, heart and skeletal muscle of mice [38], and Han et al sequenced 5 sites from mouse bone marrow [39]. Taken together, these studies represent the bulk of data on the sequenced AAV vector integration sites and have provided important insights into the AAV vector integration process. PCR and Southern blots also suggested that AAV vector integration occurs in various tissues, including the brain, heart, skeletal muscle, liver, kidney and testicular tissue, although no provirus junctions were identified in these studies [40,41].

The integration of AAV vectors occurred preferentially at specific sites in the genome. Surprisingly, 3 to 8% of integrated vector proviruses were present in the ribosomal DNA repeats that encode ribosomal RNAs [34,37]. This is not a property of other mammalian vectors, but a similar phenomenon has been observed with crustacean and insect transposons. A preference was also noted for the integration of AAV vectors at CpG islands and within 1 kb of transcription start sites [34,37,38], and this preference was also observed with retroviral vectors based on murine leukemia virus [42-44]. Segmental duplications, satellite DNA and palindromes were other preferred integration sites of AAV vectors [34,38]. In addition, integration ‘hot spots’ were identified where multiple independent proviruses were present [34,37]. In mouse hepatocytes, 53 to 62% of AAV vector integrations occurred in active genes [36,37], in contrast to 39% in cultured fibroblasts [34,37]. These data suggest that the non-homologous integration of AAV vectors is not random, and that the preferential integration at certain sites presumably reflects specific mechanisms of integration.

AAV vectors do not provide the proteins required for integration, and therefore the entire mechanism is dependent on host factors. The lack of a vector-encoded nuclease suggests that a key integration step is the availability of free chromosomal ends that can ligate to the AAV genome. AAV vectors were demonstrated to integrate at chromosomal double-strand breaks generated by the endonuclease I-Sce I or by treatment with etoposide and γ -irradiation [45]. Increasing the level of DNA damage increased integration, suggesting that the formation of double-strand breaks was the rate-limiting step. Thus, the preferred integration sites of AAV vectors may represent chromosomal regions prone to double-strand breaks or other forms of DNA damage. For example, CpG islands undergo methylation and mutagenesis that could promote DNA repair [46-48]. In addition, ribosomal DNA repeats, segmental duplications and palindromes may be involved in recombination events that create free chromosomal ends [49-51]. Therefore, integration hot spots may represent regions of genomic instability [34,37]. Furthermore, transcription start sites are susceptible to damage during the unwinding and exposure of bases that accompany the initiation of transcription [52]. In the case of hepatocytes, in which transcription had a greater effect on integration than in cultured fibroblasts, this may reflect the low proliferation rate of hepatocytes, as the lack of DNA replication would increase the proportion of double-strand breaks that result from transcription. Thus, the available evidence is consistent with a model in which AAV vectors integrate at pre-existing chromosomal breaks that provide free DNA ends for the non-homologous end-joining pathway. Importantly, there is no evidence that the vectors themselves create breaks.

The structure of integrated vector proviruses can vary significantly. Typically, bases are deleted from the vector genome termini and the hairpins of inverted repeats are favored at junction sites [23,24,34,35,37,45]. Rearrangements, deletions and insertions are present in the flanking chromosomal DNA. Genomic deletions of > 1 × 10⁶ bp and insertions as large as 500 bp have been identified [34,37]. Translocations may also occur at provirus sites in which the left and right termini of the vector are ligated to different chromosomes [34,37]. These translocations are associated with 1 to 2% of all AAV vector provirus insertions [34,37]. While the genetic changes identified at AAV provirus sites are concerning, DNA double-strand breaks are likely to result in mutations and it is possible that similar chromosomal abnormalities would have occurred in the absence of AAV vectors. A key question in the field is whether AAV vectors change the spectrum of chromosomal mutations that arise during normal cellular repair of DNA damage. In one study, AAV vectors did not increase the background mutation frequency of the cellular HPRT gene [45], suggesting that AAV vectors do not have a significant mutagenic role.

Consequences of non-homologous integration of AAV vectors

The integration of an AAV vector results in insertional mutagenesis that can alter the expression of chromosomal genes. Possible genotoxic effects include the inactivation of genes at the integration site and the dysregulation of neighboring genes as a consequence of enhancers and promoters present in the vector. The most concerning possibility is the malignant transformation of a transduced cell. The majority of experimental data have demonstrated that AAV vectors are safe in this regard. For example, the incidence of liver tumors in mice (n = 695) that received different AAV vector preparations at between 6 and 8 weeks of age was not increased compared with controls [53]. In addition, there was no increase in tumor formation after infection with AAV vectors in a study in p53-deficient mice that are prone to developing malignancies [54]. However, several studies have reported increased liver tumor incidence in mice that received AAV vectors. In some cases the tumors were likely to have been caused by vector-encoded transgenes, such as the oncogenic HBV X gene present in the woodchuck hepatitis virus post-transcriptional regulatory element [55], or the toxicity of short hairpin RNA that increased hepatocyte division [56]. However, increased tumorigenesis was observed in mice with a deficiency in ornithine transcarbamylase that received an AAV vector expressing lacZ, which should not have been an oncogenic transgene [57]; the tumor tissues contained between 0.023 and 23 copies of the vector per cell [57]. Vector:chromosome junctions were not identified in this study so it is unclear if or where the vector genomes integrated.

More convincing evidence for AAV-induced tumorigenesis is derived from experiments with an AAV vector containing a CMV enhancer, a chicken β-actin promoter and a human β-glucuronidase gene [58]. Newborn mice with mucopolysaccharidosis VII (resulting from mutations in β-glucuronidase) injected with the AAV vector had an increased risk of developing tumors, predominantly hepatocellular carcinoma [58]. Although vector genomes were present in tumor tissues, integrated provirus junctions were not identified. A follow-up study demonstrated that non-mutant mice injected with the same vector also had an increased incidence of hepatocellular carcinoma (33 to 56% compared with 8.6% in controls), suggesting that tumor formation was a feature of the AAV vector itself rather than the mutant mouse strain [59]. In this latter study, integrated AAV vector proviruses were observed in four different tumors from four different mice, and all four integrated AAV vectors mapped to a 6-kb region of chromosome 12. Microarray analysis demonstrated nearby chromosomal transcripts were upregulated, in particular one containing several small nucleolar RNAs (Rian) and another containing several microRNAs (Mirg), which could have influenced the expression of target genes responsible for malignant transformation. Although these genes were not previously associated with hepatocellular carcinoma, the occurrence of four independent tumors containing vector proviruses in the same locus (but not in the surrounding normal liver tissue) suggests that insertional mutagenesis by AAV vectors can lead to cancer in some situations [59]. Further research will be required to determine which factors were important for tumorigenesis in this model, including the age and species of the animals used and the types of vector elements such as the enhancer, promoter or transgene. Whether a similar phenomenon can occur in humans is a crucial question, particularly as a clinical trial is underway with an AAV-based liver-directed gene therapy [60] (ClinicalTrials.gov identifier: NCT00515710). AAV vectors also integrate into tissues other than the liver, including the brain, heart, skeletal muscle, kidneys and testes [38,40,41], thus the possibility that other types of tumors could develop following AAV vector delivery must also be considered.

Homologous recombination

AAV vectors can integrate by homologous recombination, which is termed AAV-mediated gene targeting [27]. In order to promote homologous recombination, a vector that contains homology arms derived from genomic chromosomal DNA that flank the genetic modification being introduced is used. Different types of mutations can arise as a consequence of recombination such as deletions, substitutions and insertions, including the insertion of transgenes and selectable markers [27,61-65]. Under optimal conditions, 0.1 to 1.0% of the total cell population exposed to an AAV targeting vector undergoes gene targeting [27], making this method more efficient than conventional, transfection-based techniques by several logs. Although the mechanism of AAV-mediated gene targeting is not completely understood, the available evidence suggests that the single-stranded vector genome pairs and recombines with homologous chromosomal sequences. Analogous parvoviral vectors based on the mouse minute virus package only one vector strand and exhibit a strand preference in targeting, which argues against a double-stranded vector genome intermediate [66]. Also, double-strand breaks present at the target locus enhance gene targeting by 60- to 100-fold, implicating the double-strand break repair mechanism in recombination [67,68]. For example, RNAi-mediated silencing of RAD54L or RAD54B (important proteins involved in homologous recombination) reduced gene targeting frequencies, and partial silencing of XRCC3 (a Rad51-related protein) abolished targeting [69]. In contrast, the silencing of the DNA-dependent protein kinase catalytic subunit, which is involved in non-homologous DNA end-joining, did not significantly reduce the frequency of AAV-mediated gene targeting [69]. These studies suggest that the same homologous recombination mechanisms known to be involved in conventional gene targeting are important for AAV-mediated gene targeting.

The high frequency and accuracy of AAV-mediated gene targeting make this method of gene targeting useful for various different applications. Several investigators have used AAV vectors to introduce mutations into transformed cell lines and study the function of oncogenes [70-81]. AAV vectors have also been used for in vivo gene targeting in the mouse liver, including the correction of mutations in a lacZ transgene, the β-glucuronidase gene and the fumarylacetoacetate hydrolase (Fah) gene [82,83]. Up to 1 in 1 × 10⁴ hepatocytes underwent gene targeting in these experiments, which is a sufficiently high frequency to cure hereditary tyrosinemia in mice with Fah mutations, but is too low for most therapeutic applications. Ex vivo gene targeting may be a more realistic application at present. AAV vectors have been used to disrupt dominant-negative mutant collagen genes in mesenchymal stem cells from patients with osteogenesis imperfecta. Using a promoter-trap strategy, 30 to 90% of cells containing the selectable marker gene were targeted, and those with the mutant allele disrupted produced healthy collagen [61,62]. In principle, these bone-forming mesenchymal stem cells could be transplanted into patients and produce normal bone. A major advantage of therapeutic gene targeting is the potential to reduce or eliminate the risks of insertional mutagenesis. In the case of AAV-targeting vectors, random integration can still occur, but potential genotoxic effects can be minimized by designing vectors without enhancers or promoters.

Conclusion

Although AAV vectors are often assumed to transduce cells as episomes, chromosomal integration occurs at significant frequencies and can produce the long-term expression of transgenes. Vector integration can occur at non-homologous locations that are sites of DNA damage, or at specific sites via homologous recombination. As with other integrating vectors, insertional mutagenesis is a concern with AAV vectors, and this should be considered in clinical applications.