Isolation of Recombinant Adeno-Associated Virus Vector-Cellular DNA Junctions from Mouse Liver

Abstract

Recombinant adeno-associated virus (rAAV) vectors allow for sustained expression of transgene products from mouse liver following a single portal vein administration. Here a rAAV vector expressing human coagulation factor F.IX (hF.IX), AAV-EF1α-F.IX (hF.IX expression was controlled by the human elongation factor 1α [EF1α] enhancer-promoter) was injected into mice via the portal vein or tail vein, or directly into the liver parenchyma, and the forms of rAAV vector DNA extracted from the liver were analyzed. Southern blot analyses suggested that rAAV vector integrated into the host genome, forming mainly head-to-tail concatemers with occasional deletions of the inverted terminal repeats (ITRs) and their flanking sequences. To further confirm vector integration, we developed a shuttle vector system and isolated and sequenced rAAV vector-cellular DNA junctions from transduced mouse livers. Analysis of 18 junctions revealed various rearrangements, including ITR deletions and amplifications of the vector and cellular DNA sequences. The breakpoints of the vector were mostly located within the ITRs, and cellular DNA sequences were recombined with the vector genome in a nonhomologous manner. Two rAAV-targeted DNA sequences were identified as the mouse rRNA gene and the α1 collagen gene. These observations serve as direct evidence of rAAV integration into the host genome of mouse liver and allow us to begin to elucidate the mechanisms involved in rAAV integration into tissues in vivo.

Adeno-associated virus (AAV) has been extensively explored as a potential vector for gene therapy (15). The advantages of the use of AAV-based vectors are that they transduce both dividing and nondividing cells and achieve long-term expression of therapeutic genes with no apparent adverse effects. We and others have investigated the feasibility of transferring the human coagulation factor IX (hF.IX) gene into mouse liver by recombinant AAV (rAAV) vectors (21, 31). In our previous study, we demonstrated long-term therapeutic levels of hF.IX in mouse plasma following portal vein injection of an AAV-EF1α-F.IX vector (21). However, the mechanism of persistent expression of therapeutic gene products from the liver by rAAV vectors has not been fully understood. A possible interpretation of sustained expression is rAAV vector integration into the host genome. Latent infection of wild-type AAV (wtAAV) results in site-specific integration into the AAVS1 region of human chromosome 19 (14, 28). The inverted terminal repeat (ITR) sequence is the only vector element necessary for integration (38), but the efficiency and site specificity of wtAAV integration into the host genome relies on the viral Rep proteins and the presence of the target sequence, i.e., AAVS1 (2, 9, 16, 18, 33). Although it has been demonstrated that rAAV vectors devoid of a Rep expression cassette integrate with low efficiency and with a lack of specificity into the host genomes of some dividing cells in vitro, little is known about the integration of rAAV vectors into nondividing cells (35). Recently several groups have suggested, on the basis of Southern blotting data, that rAAV vectors integrate into the genomes of mouse liver and muscle cells (4, 8, 10, 30). More recent studies by Miao et al. have demonstrated evidence of rAAV vector integration into mouse liver on the basis of pulsed-field gel electrophoresis and fluorescence in situ hybridization analysis of metaphase hepatocytes (20). The present study was undertaken to further establish integration of rAAV vectors into the host genome. Detailed Southern blot analysis of the molecular forms of rAAV in transduced mouse liver was performed. In addition, we present direct evidence for integration by isolation of rAAV vector-cellular DNA junctions from transduced mouse liver.

MATERIALS AND METHODS

AAV-EF1α-F.IX vector.

AAV-EF1α-F.IX vector (Fig. 1A) was produced based on plasmid pV4.1e-hF.IX (21) and purified as outlined elsewhere with an adenovirus-free system (12, 17) with a modification. The vector preparation was further purified by including a nuclease digestion step of the crude cell lysate and by using two successive CsCl gradients followed by dialysis. The physical vector titer was determined by a quantitative dot blot assay (12).

FIG. 1.

Map of rAAV vectors. (A) AAV-EF1α-F.IX; (B) AAV-EF1α-GFP.AOSP. The locations of restriction enzymes and probes are depicted. EF1α-P, the human polypeptide elongation factor 1α gene enhancer-promoter; hF.IX cDNA, human coagulation factor IX cDNA; pA(hGH), the human growth hormone gene polyadenylation signal; δEF1α-P, truncated EF1α-P; Lac-P, bacterial lac operon promoter; GFP, the enhanced green fluorescent protein gene; pA(β-gl), the human β-globin gene polyadenylation signal; Amp^r, the bacterial ampicillin resistance gene; Ori, plasmid origin of replication; A, AlwNI; B, BamHI; Bg, BglII; C, ClaI; E, EcoRI; H, HindIII; N, NotI; P, PmeI; Xa, XbaI; Xo, XhoI.

AAV-EF1α-GFP.AOSP vector.

The AAV-EF1α-GFP.AOSP vector was constructed to serve as a rAAV shuttle vector for the isolation of rAAV vector-cellular DNA junctions. The vector plasmid pAAV-EF1α-GFP.AOSP encodes the following elements between two ITRs: a hybrid promoter comprising a truncated EF1α enhancer-promoter (δEF1α-P) and the bacterial lac operon promoter (lacp), a multiple cloning sequence (MCS), the enhanced green fluorescent protein (GFP) transgene, a poly(A) signal derived from the human β-globin gene, the bacterial ampicillin resistance (Amp^r) gene, and a plasmid origin of replication (Ori) (Fig. 1B). A 1.6-kb stuffer fragment from the coding sequences of the bacterial lacZ gene was inserted outside the two ITRs to stabilize this plasmid. The construction of pAAV-EF1α-GFP.AOSP was done as follows. A 228-bp fragment of the human β-globin gene poly(A) signal was inserted at the unique StuI site of pEGFP (Clontech, Palo Alto, Calif.), downstream of the GFP coding sequence, to create pEGFP.βglpA. The 3′ ITR was removed from pV4.1e-hF.IX by double digestion with PvuII and BbrPI and inserted at the unique AflIII site of pEGFP.βglpA, just downstream of Ori, to create pV3EGFP.βglpA. A 1.8-kb HindII fragment of the lacZ gene was excised from pAAV-LacZ (12) and inserted at a unique PvuII site between the 3′ ITR and lacp, to create pV3EGFP.βglpA.spc. We also created pV4.1δe-hF.IX, a pV4.1e-hF.IX derivative with a truncated EF1α-P (δEF1α-P), by removing a 1.3-kb SpeI/MunI fragment from the EF1α-P sequences of pV4.1e-hF.IX. The 5′ ITR and δEF1α-P were excised from pV4.1δe-hF.IX by double digestion with PvuII and EcoRI, and the resulting 1.3-kb fragment was blunted and inserted into the unique PvuII site of the pEGFP.βglpA, creating pV5eEGFP.βglpA. A 1.5-kb fragment comprising the 5′ ITR and the δEF1α-P–lacp hybrid promoter was excised from pV5eEGFP.βglpA with a PvuII/PstI double digestion and ligated with a PvuII/PstI fragment of pV3EGFP.βglpA.spc comprising the GFP gene, β-globin poly(A), Amp^r, Ori, 3′ ITR, and the 1.6-kb stuffer. This resulted in the construction of pAAV-EF1α-GFP.AOS. To construct pAAV-EF1α-GFP.AOSP, we inserted a unique PmeI site in frame at the SalI site of the MCS located between lacp and the GFP gene. PmeI cuts the mouse genome infrequently, and incorporation of this site was used to reduce the background in the integrant plasmid library. As can be seen in Fig. 2A, without the PmeI digestion step, the library would be contaminated with many plasmids that do not harbor integration junctions.

FIG. 2.

(A) Schematic structure of possible forms of rAAV in tissue (center row), plasmids possibly rescuable by BamHI digestion and religation (top row), and plasmids possibly rescuable by BamHI digestion and religation following PmeI digestion and CIP treatment (bottom row). Examples of possible forms in tissue are episomal circular rAAV forms (monomers circularized at the ITRs and aberrant rAAV monomers lacking both BamHI and PmeI sites), rAAV provirus with three tandem repeats forming a head-to-tail concatemer, and a tail-to-tail junction from either episomal or integrated rAAV forms. PmeI digestion can remove GFP⁺ episomal forms and head-to-tail circular molecules derived from inner repeats of rAAV concatemers. B, BamHI; P, PmeI site. The straight and zigzag lines represent rAAV vector and mouse genomic DNA sequences, respectively. (B) Construction of a rAAV vector-cellular DNA junction fragment library from mouse liver DNA transduced with AAV-EF1α-GFP.AOSP. High-molecular-weight DNA was isolated from transduced liver. Seven micrograms of liver DNA was digested with PmeI and treated with CIP. DNA (2.25 μg) was directly used to transform E. coli to assess contamination of episomal circular forms of rAAV insensitive to PmeI digestion, and 0.75 μg of DNA was analyzed by gel electrophoresis. Three micrograms of the above-mentioned DNA was digested with BamHI and self-ligated with T4 DNA ligase. A junction fragment library was made by transforming E. coli with 2.25 μg of the above-mentioned DNA, and the remaining 0.75 μg was electrophoresed to analyze the DNA. Colonies were classified as GFP⁺ (green) and GFP⁻ (white) by UV excitation. The white colonies were subjected to the screening procedure for identification of integrant candidates.

Contamination of wtAAV in the vector preparations was assessed by a modification of the replication center assay (31). Briefly, 293 cells were seeded in 10-cm-diameter dishes and infected with 10⁹ particles of a rAAV vector in the presence of adenovirus type 2 (multiplicity of infection, 10). The cells were harvested 24 h later, and 1/20 of the freeze-thaw cell lysates were used to infect 293 cells in six-well plates in the presence of adenovirus for sequential amplification of wtAAV. The cells were then incubated for 24 h and transferred to nylon membranes, followed by denaturation and neutralization. The membranes were hybridized with a ³²P-labeled rep-cap gene probe and analyzed by autoradiography. We detected no wtAAV in the vector preparations at a sensitivity of 10³ particles of wtAAV in 10⁹ particles of rAAV.

Animal procedures.

All the animal experiments were performed according to the National Institutes of Health guidelines. Six- to eight-week-old female C57BL/6 mice were purchased from Charles River Laboratories (Boston, Mass.). AAV-EF1α-F.IX, at doses of 2.7 × 10¹¹, 5.5 × 10¹⁰, and 1.1 × 10¹⁰ particles/animal, were injected by portal vein (PV) or intravenous (i.v.) injection (n = 4) into C57BL/6 mice. Direct liver (DL) injection of C57BL/6 mice was performed with 2.7 × 10¹¹ particles of AAV-EF1α-F.IX per mouse (n = 4). The PV injection procedure was as previously described (21). For i.v. infusion, 200 μl of rAAV vector solution was given through the tail vein. The DL injection procedure was as follows. The animals were anesthetized with an intraperitoneal injection of ketamine and xylazine, and the liver was exposed through a ventral incision. Two hundred microliters of the rAAV vector solution was injected directly into the hepatic parenchyma at six different sites of the left lateral lobe with a handheld 27-gauge needle connected to an infusion pump (model 2000; Instech Laboratories, Plymouth Meeting, Pa.) at a flow rate of 15 to 20 μl/min. Bleeding after injection was controlled by compression with forceps, and the peritoneal cavity was closed with 5-0 Dermalon (Davis+Geck, Manati, Puerto Rico). The mice were bled from the retro-orbital plexus several times up to 10 months, and plasma hF.IX levels were measured with an hF.IX-specific enzyme-linked immunosorbent assay (32). One mouse from each group (PV, i.v., and DL) was sacrificed for tissue collection 3 months postinjection.

For the isolation of rAAV-cellular DNA junctions from transduced mouse livers, four C57BL/6 mice received PV injections of 1.2 × 10¹¹ particles of AAV-EF1α-GFP.AOSP vector. A single mouse was sacrificed 1 week and 1, 5, and 9 months postinjection, and liver DNA was isolated for analysis.

Southern blot analysis.

Total DNA was extracted from liver and other tissues as described by Sambrook et al. and analyzed by Southern blotting (26). The copy number standards (the number of double-stranded rAAV genomes per diploid genomic equivalent) were prepared by spiking an equivalent number of the vector plasmid molecules into total DNA extracted from naive mouse liver.

Isolation of rAAV vector-cellular DNA junctions.

Construction of the rAAV vector-cellular DNA junction fragment library is shown schematically in Fig. 2B. Seven micrograms of high-molecular-weight DNA extracted from AAV-EF1α-GFP.AOSP vector-transduced liver was digested with PmeI at 37°C for 4 h. A single PmeI site is present in AAV-EF1α-GFP.AOSP, located between lacp and the GFP gene, 19 bp upstream of the unique BamHI site. The PmeI-digested DNA was treated with calf intestinal phosphatase (CIP; New England Biolabs, Beverly, Mass.) at 50°C for 1 h to prevent ligation of free DNA ends, including the PmeI ends. After the CIP was heat inactivated at 65°C for 1 h, the DNA was extracted by phenol-chloroform and precipitated by ethanol. To assess the level of contamination by episomal circular intermediates of rAAV (6) insensitive to PmeI digestion, 2.25 μg of the purified DNA was used to transform Escherichia coli by electroporation. The remaining DNA was subjected to the plasmid rescue scheme described below or analyzed for integrity by agarose gel electrophoresis. Rescue of the rAAV vector-cellular DNA junctions involved BamHI digestion and self-ligation followed by transformation of E. coli. Three μg of PmeI and CIP-treated DNA was digested with BamHI at 37°C for 4 h. BamHI cleaves the vector at a single site, which is located just downstream of the PmeI site. After digestion, the DNA mixture was incubated at 65°C for 20 min, and the DNA was self-ligated with T4 DNA ligase (New England Biolabs) at 16°C overnight in 700 μl of reaction mixture. The DNA was extracted with phenol-chloroform, precipitated by isopropanol (34), and resuspended, and 2.25 μg was used to transform the bacteria by electroporation. The remaining 0.75 μg was analyzed for integrity by electrophoresis. The transformed bacteria were plated on Luria-Bertani agar plates containing ampicillin (50 μg/ml). Expression of GFP in bacterial colonies in these libraries was analyzed by long UV excitation (365 nm) on an electronic transilluminator (Ultra-Lum, Carson, Calif.). GFP-negative colonies were isolated, and plasmid DNA was extracted and analyzed by restriction digestion analysis and DNA sequencing. Nucleotide sequences were determined by dideoxy chain termination reactions. All bacterial transformations were done with E. coli DH10B (ElectroMAX DH10B cells; GIBCO BRL, Gaithersburg, Md.), and the electroporation was performed with a 0.1-cm=path=length cuvette and an Electro Cell Manipulator 600 (BTX, San Diego, Calif.) under conditions of 17.5 kV/cm and 186 Ω. These conditions resulted in efficiencies of ≥10¹⁰ transformants/μg of pUC19.

RESULTS

Delivery of rAAV to the liver by DL or PV injection is more efficient than delivery by i.v. injection.

C57BL/6 mice received different doses of AAV-EF1α-F.IX via three different routes (PV, i.v., or DL injection). When the highest dose of the vector (2.7 × 10¹¹ particles/animal) was injected into the liver either directly or via the PV, levels of approximately 1,000 to 2,000 ng of hF.IX/ml (20 to 40% of normal levels in human plasma) were observed for up to 10 months (Fig. 3). The PV and DL routes appeared to be equally efficient and approximately 10-fold better than the i.v. route. A fivefold-lower dose of the vector (5.5 × 10¹⁰ particles/animal) administered via PV also resulted in therapeutic levels of hF.IX (200 to 250 ng/ml). Southern blot analysis of liver DNA isolated from animals injected by these three routes confirmed that PV or DL administration was more efficient at delivering the vector genome than the i.v. route. As shown in Fig. 4A and C, the livers of animals injected with 2.7 × 10¹¹ particles of AAV-EF1α-F.IX via either the PV or the DL route contained approximately one vector genome (double-stranded rAAV genome) per diploid genomic equivalent whereas a 10-fold-lower copy number was observed following i.v. administration of vector (Fig. 4B). None of the other tissues examined (lung, heart, kidney, intestine, brain, spleen, peritoneum, and leg muscle) demonstrated the presence of the vector sequences (sensitivity, 0.01 copy/diploid genome).

FIG. 3.

Levels of hF.IX in plasma of C57BL/6 mice following administration of three doses (2.7 × 10¹¹, 5.5 × 10¹⁰, and 1.1 × 10¹⁰ particles/animal) of AAV-EF1α-F.IX via PV or i.v. injection and a dose (2.7 × 10¹¹ particles/animal) of the same vector via DL injection. Plasma samples were collected over time and assayed for hF.IX (n = 4 in each group). A single mouse in each group injected via the PV (2.7 × 10¹¹), i.v. (2.7 × 10¹¹), and DL (2.7 × 10¹¹) routes was sacrificed 3 months postinjection. The hF.IX levels at the 8- and 10-month time points (IV) (2.7 × 10¹¹) represent the results from a single mouse.

FIG. 4.

Southern blot analysis to determine vector copy numbers in tissues from mice injected with 2.7 × 10¹¹ particles of AAV-EF1α-F.IX via three different routes: PV (A), i.v. (B), and DL (C). The mice were sacrificed 3 months postinjection, and tissues (left lateral lobe of the liver, uninjected liver in the case of DL, lung, heart, kidney, intestine, brain, spleen, peritoneum, and leg muscle) were analyzed. Twenty micrograms of total DNA extracted from each tissue was digested with ClaI and HindIII, electrophoresed, and hybridized with probe A. Vector copy number standards were 20 μg of naive mouse liver DNA spiked with an equivalent number of the vector plasmid molecules and described as copies/cell (the number of double-stranded rAAV genomes per diploid genomic equivalent).

Southern analysis of rAAV in liver.

Detailed Southern blot analysis allowed the characterization of rAAV vector forms in transduced liver tissue. Restriction enzymes that cut at the 5′ (ClaI), center (EcoRI), and 3′ (HindIII) portions of the rAAV genome were selected. The sites of the enzymes and the probes used for this analysis are shown in Fig. 1. Digestion with EcoRI can distinguish linear monomer and multimer (concatemer) forms, while digestion with ClaI or HindIII can distinguish tail-to-tail, head-to-head, or head-to-tail forms. Since others demonstrated that rAAV genomes are converted from low-molecular-weight episomal forms to high-molecular-weight DNA 4 to 8 weeks postinjection (4, 8, 10, 20, 30, 37), we analyzed DNA samples extracted from animals sacrificed 3 months postinjection to elucidate the rAAV forms responsible for persistent transgene expression.

Digestion of transduced liver DNA with EcoRI, which cuts in the center of the vector, revealed bands representative of concatemers (Fig. 5). Circular monomer forms could generate the same bands as concatemers when cut with a single cutter; however, we assume that they are a small population, since most of the DNA remained high-molecular-weight DNA when uncut or digested with an enzyme that does not cut the vector (Fig. 5). To characterize these concatemers further, DNA was digested with ClaI or HindIII and hybridized with probe A. The results revealed that the vectors formed head-to-tail, head-to-head, and tail-to-tail concatemers (Fig. 5). However, the bands representative of head-to-head and tail-to-tail concatemers tended to be fainter than those of head-to-tail concatemers, indicating, as previously shown (20), that the vectors predominantly formed head-to-tail concatemers. The high-molecular-weight signals observed from undigested DNA did not change when the DNA was digested with a noncutter (KpnI or AlwNI), despite showing a normal smear of genomic DNA as determined by ethidium bromide staining (data not shown). The observed DNA smear by single-cutter digestion was suggestive of vector integration into the host genome (Fig. 5). This was also observed in a sample of liver DNA isolated from a mouse injected 5 months earlier with AAV-EF1α-GFP.AOSP vector. Digestion of this sample with a single cutter (BglII or AlwNI), followed by hybridization with probe B, revealed head-to-tail concatemers and a background smear (data not shown). However, the data described above cannot exclude large episomal concatemers with various deletions or rearrangements.

FIG. 5.

Southern blot analysis to determine rAAV vector forms in the livers from C57BL/6 mice injected with a dose of 2.7 × 10¹¹ particles of AAV-EF1α-F.IX via PV or DL route. The time of sacrifice, the route of vector administration, and the enzymes used are shown above each lane. Twenty micrograms of total DNA was digested with selected restriction enzymes, electrophoresed on a 0.8% agarose gel along with vector copy number standards, transferred to a nylon membrane (Duralon UV; Stratagene), and hybridized with ³²P-labeled probe A. The membrane was washed and exposed to film at −80°C for 1 to 3 weeks. Copy number standards (the number of double-stranded rAAV genomes per diploid genomic equivalent) were prepared by spiking an equivalent number of the vector plasmid molecules into 20 μg of total DNA extracted from naive mouse liver. The predicted fragment lengths of concatemeric rAAV forms with intact ITRs are (i) 5.0 kb of ClaI, HindIII, or EcoRI digests for head-to-tail forms; (ii) 8.5 and 5.2 kb of HindIII and EcoRI digests, respectively, for head-to-head forms; and (iii) 9.0 and 4.7 kb of ClaI and EcoRI digest, respectively, for tail-to-tail forms. The fragment lengths of copy number standards are 8.1 (ClaI), 4.3 (HindIII), 5.4 and 2.4 (EcoRI), 8.1 (KpnI), and 8.1 (AlwNI) kb. KpnI and AlwNI do not cut the vector genome. The solid arrowheads indicate head-to-tail forms, while the open arrowheads show head-to-head or tail-to-tail forms. The bands indicated by arrows were presumed to represent double-stranded linear-monomer rAAV vector forms, because the same pattern was observed from the DNA extracted from rAAV stocks, in which only a linear-monomer form was observed by alkaline gel electrophoresis (data not shown). The three bands in rAAV stocks showing the same pattern as in these figures (data not shown) were presumed to represent artificially reannealed single-stranded rAAV genome in the process of DNA extraction, since these bands were digestible with enzymes that cut the vector. M, months. For the locations of enzyme sites and the probe, see Fig. 1.

Examination of the band sizes in Fig. 5 indicated that the head-to-tail, head-to-head, and tail-to-tail concatemers were shorter by a few hundred base pairs than predicted, suggesting that deletions had occurred. In order to determine if the deletions were within the ITRs, we digested genomic DNA with enzymes that were expected to excise internal vector fragments of decreasing size, as shown in Fig. 6A. If deletions occurred more frequently in or near the ITRs, one would expect digestion with SrfI or NotI to result in a low level of hybridization to the predicted internal fragment and hybridization to a smear. The smear represents junction sequences similar to what was observed when DNA was digested with a single-cutting enzyme. In contrast, digestion with the enzyme PpuMI, which excises a shorter fragment from the internal vector sequences, would result in a higher level of hybridization to the predicted internal fragment and no hybridization to a smear, because the PpuMI sites would be intact. As shown in Fig. 6B, there was a gradual increase in the hybridization intensity of the predicted band as the location of the restriction enzyme was further removed from the ITRs. This observation, in addition to the decreased hybridization to a smear, suggested that deletions at the ends of the vector, including the ITRs, occurred frequently.

FIG. 6.

Analysis of ITR deletions by Southern blotting. (A) Expected vector fragments released from AAV-EF1α-F.IX by various enzyme digestions. (B) Twenty micrograms of liver DNA of C57BL/6 mice injected with AAV-EF1α-F.IX via PV or DL route (3 months postinjection) was digested with the indicated enzymes and hybridized with probe C. Decreased hybridization to a predicted fragment and increased hybridization to a smear in SrfI and NotI digests suggest deletions of the ITRs and their flanking sequences.

To confirm the observation of deletions within the ITRs, DNA was digested with BglI, which cleaves the vector five times: within the ITRs and at three internal positions (Fig. 7A). When the DNA was hybridized with probe A, an unexpected discrete band of 4.4 kb was observed, in addition to the expected bands of 2.1 and 2.4 kb (Fig. 7B). A smear beginning at 4.4 kb also hybridized to the probe. The presence of the 4.4-kb band and the smear suggests that deletions have occurred in the ITRs, which resulted in the elimination of the BglI sites, as illustrated in Fig. 7C.

FIG. 7.

Further analysis of ITR deletions by Southern blotting. (A) Locations of the BglI sites in AAV-EF1α-F.IX. The expected fragments obtained by BglI digestion are also shown. (B) Twenty micrograms of liver DNAs from C57BL/6 mice injected with a dose of 2.7 × 10¹¹ particles of AAV-EF1α-F.IX were digested with BglI and hybridized with probe A. The time of sacrifice and the route of administration are indicated above each lane. M, months. (C) Schematic representation of deletion of two BglI sites in the ITR. The 4.4-kb band indicates joining of two BglI fragments of 2.4 and 2.1 kb due to deletions of the ITR sequence involving BglI sites.

Direct evidence of rAAV integration and isolation of vector-cellular DNA junctions.

Southern blot analysis suggests that rAAV vectors integrate into the host genome of mouse liver. However, other interpretations are possible, such as the presence of high-molecular-weight concatemeric episomes. Therefore, to directly demonstrate rAAV vector integration, we developed a rAAV shuttle vector which facilitated the recovery of vector-cellular DNA junctions as plasmids in bacteria. The salient features of this vector are as follows: (i) an Amp^r gene and a plasmid origin of replication to allow rescue of junction fragments as plasmids in E. coli; (ii) a GFP gene expressed from a hybrid promoter, allowing expression of GFP in both mammalian and bacterial cells; (iii) disruption of the GFP gene by BamHI digestion, used to screen, in E. coli, for potential plasmids containing integration junctions; (iv) a unique and rare restriction enzyme site, PmeI, incorporated 5′ of the BamHI site and used to remove plasmids that did not contain junctions from the plasmid pool. These included circular episomal rAAV forms and plasmids generated from self-ligated internal repeats of head-to-tail concatemers (Fig. 2A). A dose of 1.2 × 10¹¹ particles of AAV-EF1α-GFP.AOSP vector was injected into the portal vasculature of four adult C57BL/6 mice, and DNA was extracted from the transduced livers. Since Southern blot analysis revealed that the DNA samples collected 1 week and 1 month postinjection were considerably contaminated with low-molecular-weight vector forms (data not shown), we selected the DNA samples collected 5 and 9 months postinjection for the isolation of junctions.

Recently Duan and colleagues reported that following transduction of mammalian cells in vivo and in vitro, rAAV vectors are converted to double-stranded head-to-tail monomeric circular forms (6). Thus, we anticipated that such circular intermediates might be present in the DNA preparations isolated from transduced mouse liver and that they would contribute to the plasmids isolated from E. coli. GFP-expressing (green, or GFP⁺) colonies could be eliminated as not containing potential junctions; however, non-GFP-expressing (white, or GFP⁻) colonies would be selected as potentially harboring plasmids with junction fragments. Predigestion of genomic DNA with PmeI was included in the scheme as a way of reducing circular intermediates from the pool; however, deletions from the circular intermediates at the PmeI site would compromise this step. With this in mind, an initial experiment was performed to analyze the circular intermediates that would be present in the DNA extracted from the liver harvested 5 months postinjection. Transformation of E. coli with 1 μg of undigested total DNA generated 469 colonies (345 GFP⁺ and 124 GFP⁻ colonies). Restriction enzyme digestion and sequencing of plasmids recovered from these bacterial colonies revealed that most of these plasmids were circularized double-stranded monomer forms of AAV-EF1α-GFP.AOSP vector with various deletions involving both vector ends and the internal sequences (data not shown). The circular intermediates with a functional GFP gene formed green bacterial colonies, whereas the GFP⁻ circular intermediates had a disrupted GFP gene, resulting in white colonies. The PmeI digestion of liver DNA that was incorporated in the isolation scheme was designed to remove these contaminants. However, PmeI digestion was not expected to remove all of the background colonies arising from rAAV circular intermediates because many GFP⁻ intermediates had deletions involving the PmeI site. When we transformed bacteria with 2.25 μg of DNA that had been digested with PmeI, 2 GFP⁺ and 215 GFP⁻ colonies were isolated, indicating that PmeI digestion is effective in eliminating the GFP⁺ circular intermediates but less effective in removing the altered GFP⁻ intermediates.

Isolation of rAAV-cellular DNA junctions was attempted by first digesting the DNA extracted from transduced mouse liver (5 months postinjection) with PmeI followed by treatment with CIP. This was followed by BamHI digestion and religation. The ligated material was subsequently introduced into E. coli by electroporation, and 2 GFP⁺ and 270 GFP⁻ colonies were obtained. Plasmids containing rAAV vector-cellular DNA junctions were expected to produce GFP⁻ colonies. Therefore, we isolated 183 independent GFP⁻ colonies and extracted plasmid DNA from each. A procedure for obtaining plasmid DNA for putative rAAV-cellular DNA junctions was developed. The criteria for bona fide junction fragments included the following: (i) the plasmid DNA must contain a single BamHI site; (ii) the plasmid DNA must not contain a PmeI site; (iii) the plasmid DNA must not contain the XbaI site located 5′ of the BamHI site (Fig. 1B); (iv) the plasmid DNA must generate a vector-specific 746-bp fragment when digested with BamHI and XbaI; and (v) the plasmid DNA must generate several vector-specific fragments when digested with BglI or TaqI. Most of the altered GFP⁻ intermediates could be eliminated by criteria 1 to 3, and contaminating unrelated plasmids could be excluded by criteria 4 and 5. Using these criteria, we selected 17 of 183 plasmids as potential candidates for containing rAAV-cellular DNA junctions from transduced mouse liver harvested 5 months postinjection. Detailed restriction digests and sequencing of these 17 candidates revealed that 5 represented vector-vector junctions with various rearrangements (data not shown), 3 showed recombinations between vector sequences and AAV helper plasmid or adenovirus helper plasmid sequences (data not shown), and the remaining 9 possessed true rAAV vector-cellular DNA junctions (Fig. 8A and B). We also isolated another 9 rAAV-cellular junctions and a recombination between rAAV and adenovirus helper plasmid from 6.75 μg of liver DNA of an AAV-EF1α-GFP.AOSP-injected mouse sacrificed 9 months postinjection (Fig. 8B). Of these 18 rAAV-cellular DNA junctions, the presumptive mouse genomic DNA sequences flanking the rAAV vector were screened for homologies in GenBank by an advanced BLAST search. The search revealed that the cellular sequences in clone J121 possessed 100% homology to the mouse α1 (XVIII) collagen (COL18A1) gene (GenBank accession no. U34612) and clone J192 had cellular sequences that were 100% homologous to the mouse 45s pre-rRNA gene (GenBank accession no. X82564). The integration sites were located in an intron of the α1 collagen gene and the 28s rRNA transcribed region. Another clone harbored partial homology to the mouse putative chloride channel protein CLC 6 gene (J242; positivity, 120 of 129 [93%]), and another was partially homologous to the mouse α-N-acetylglucosaminidase gene (J288; positivity, 63 of 69 [91%]), but no other apparent homology was found in the remaining sequences.

FIG. 8.

(A) Structures of nine rAAV vector-cellular DNA junctions isolated from transduced mouse liver taken 5 months postinjection. The vector structures 5′ of the BamHI site are not known because of the nature of the strategy. The bold lines and thin lines represent rAAV vector and cellular DNA sequences, respectively. The vertical bars on the lines indicate BamHI sites. Vertical bars across the lines show the ITR, with the strokes roughly representing the lengths of the deleted ITR. The reversed letters show the vector sequences in an inverted orientation. The other nine junctions isolated from the transduced mouse liver taken 9 months postinjection are not shown here; all of them showed simple joining of rAAV and cellular DNA sequences of various sizes. (B) Structures around each junction at the DNA sequence level (for the junction J104, see Fig. 9). Unrearranged vector sequences around the junctions (the ITR and its flanking sequences) are shown, along with the junction sequences isolated from the transduced liver tissue. The numbers above the sequences begin at the 5′ end of the intact vector. The ITR sequences are depicted in two possible orientations (flip and flop). Flanking sequences derived from mouse DNA are shown in lowercase. The arrows show the breakpoints. Nucleotides underlined with duplicated lines indicate residues shared by the vector and the flanking sequences at a junction site, which made it impossible to determine the exact breakpoint. (C) Distribution of the rAAV vector-cellular DNA junctions. One dot represents one break in recombination events. When one junction clone contained more than one breakpoint in the vector sequences, each breakpoint is included.

An alternative process for obtaining plasmids containing presumptive rAAV vector-cellular DNA junctions might be intermolecular ligation between a linear double-stranded rAAV genome and one of the BamHI-BamHI fragments of genomic DNA which are present in large numbers in the DNA preparation after BamHI digestion. This linear double-stranded rAAV would have a BamHI site in the MCS and a free vector end on the other side after BamHI digestion. Although such intermolecular ligations between noncomplementary DNA ends are unfavored, it might be possible for this system to trap not only integrants but also genomic DNA not associated with rAAV provirus. To address this issue, we performed a reconstitution experiment in which we tested an intermolecular ligation between BamHI-XhoI pAAV-EF1α-GFP.AOSP fragments (BamHI site–GFP–pA–Amp^r-Ori–3′ ITR–stuffer–5′ ITR–δEF1-P–XhoI site) and genomic DNA fragments digested with BamHI and XhoI (e.g., BamHI-BamHI, BamHI-XhoI, and XhoI-XhoI genomic fragments). The molecular ends, excluding BamHI, were all CIP treated before ligation. In brief, naive mouse liver DNA spiked with pAAV-EF1α-GFP.AOSP plasmid at 10 copies/diploid genomic equivalent (about 30-fold higher than the copy number observed in the AAV-EF1α-GFP.AOSP-transduced liver DNA [data not shown]) was digested with XhoI, treated with CIP, and then digested with BamHI, followed by religation in the same manner. E. coli was transformed as described above and generated 14 GFP-negative colonies. Diagnostic enzyme digestions revealed that 2 of 14 colonies contained pAAV-EF1α-GFP.AOSP sequence-derived plasmids while the others were contaminating laboratory plasmids unrelated to pAAV-EF1α-GFP.AOSP. The sequencing of these two plasmids derived from the vector sequences demonstrated that neither of them contained sequences unrelated to pAAV-EF1α-GFP.AOSP. Thus, we concluded that such intermolecular ligations, even if they actually occurred, were negligible for the analysis.

As shown in Fig. 8B and C, the rAAV vector-cellular junction sites in the vector were predominantly located within the ITRs, and both flip (D-A′-C-B-A) and flop (D-A′-B-C-A) ITR orientations were found. Within the ITRs, the junction sites were scattered throughout and there appeared to be no preferential sites for breakage, although the number of junctions analyzed was limited. Recombinant AAV vector-cellular junctions were also observed in the internal rAAV sequences near the ITRs (J104, J196, J242, and J270). Fifteen integrants (J16, J30, J121, J134, J166, J196, J216, J236, J242, J270, J278, J288, J299, J305, and J313) could be explained by a simple crossover between rAAV and cellular DNA sequences, whereas the other three harbored rearrangements that included a 4-bp duplication of ITR sequence near the junction (J192), a 1.9-kb duplication of the vector sequence including an ITR (J175), and a scrambled alignment of the vector and cellular DNA sequences in both direct and inverse orientations (J104 [Fig. 9]), where a stretch of genomic sequence of about 0.2 kb (Fig. 9A, U1), and vector sequence in an inverted orientation (Fig. 9A, δEF1α-P) was repeated twice and interposed between the vector sequences and another unknown 2.5-kb genomic sequence (Fig. 9A, U2). When we looked for homology between the vector sequences and the targeted cellular sequences in the 18 clones, none shared significant homology. The four recombination events between rAAV vector and helper plasmid sequences (J1, J78, J123, and J228), which were isolated from vector-injected mouse liver DNA, were as follows. In J1, the vector Ori sequence was recombined with the Flp recognition target sequence artificially incorporated in an AAV helper plasmid. J78 had a recombination event between the vector Ori and the E4 gene of adenovirus helper plasmid. In J123, the palindromic region of the 3′ ITR was recombined with the Ori of adenovirus helper plasmid. J228 had a recombination between the vector Ori and the Amp^r of adenovirus helper plasmid. These were illegitimate recombinations. Since the flanking helper plasmid sequences were terminated at an adjoining BamHI site due to the nature of the plasmid rescue strategy, we could not determine the complete structures of these molecules.

FIG. 9.

Detailed map of a junction clone, J104, which showed a scrambled alignment of the vector and cellular DNA sequences. (A) The whole structure of the junction fragment isolated. Bold lines indicate the vector sequences, and thin lines represent two kinds of unknown cellular DNA sequences (U1 and U2). Reversed letters on the vector lines indicate inverted orientation of the vector sequences. (B) Nucleotide sequences at each recombination site surrounded by brackets (junctions 1 to 5) in panel A. The vector sequences are in uppercase, while cellular DNA sequences are in lowercase. The nucleotides indicated by double-underlined boldfaced letters indicate a residue or residues shared at junction sites. Portions of unrearranged rAAV vector sequences around the junction sites are shown, along with isolated junction sequences with numbers beginning at the 5′ end of the intact vector. (C) Possible structure of an integration intermediate of J104. The boldfaced solid line and dashed lines represent rAAV vector and cellular DNA sequences, respectively. The vector sequences were presumed to have associated with the host DNA at three points of two DNA strands (P1 and P2 in one DNA strand and P3 in another). The vector and cellular DNA sequences were broken at each site (P1, P2, and P3), and then DNA repair and replication occurred, as shown by a thin line with an arrowhead. In the second round of DNA replication of the bottom cellular DNA strand, DNA polymerase skipped several nucleotides at both junctions P1 and P2 (see the sequencing results). The black boxes on the bold line are ITRs. The structures outside the ITRs in this integration intermediate are not known.

In summary, we isolated 18 rAAV vector-cellular DNA junctions from transduced mouse livers. This supports the Southern blot data suggestive of vector integration and serves as direct evidence of in vivo integration of rAAV into the host genome. The vector sequences were illegitimately recombined with the host genome predominantly in the ITR, which is similar to what has been observed for rAAV provirus in dividing cells in vitro (25, 38).

DISCUSSION

rAAV is a promising vector for gene therapy because it transduces nondividing cells with high efficiency in vivo to direct stable long-term expression of transgene products in animals (4, 8, 10–12, 21, 30, 31, 37). Gene transfer to the liver by rAAV vectors is a relatively recent observation, and the mechanisms underlying transgene persistence in the liver after rAAV vector transduction have not been completely elucidated, although previous works by others suggested integration of rAAV into the host genome (4, 8, 10, 20, 30, 37). The goal of this study was to isolate junction fragments between rAAV vector sequences and genomic DNA from mouse livers in order to obtain direct evidence of rAAV vector integration, which had not been achieved by others. By elucidating the molecular structure of the integrated and nonintegrated forms, the molecular basis of vector DNA transduction can begin to be more fully understood.

In this study, we first performed Southern blot analysis to elucidate the forms of rAAV vectors in transduced liver tissue. While the result is similar to that recently described (20, 31), that is, they are mainly head-to-tail high-molecular-weight concatemers suggestive of vector integration, the unique aspect of our study is the demonstration, by Southern blotting, of deletions in the ITRs and their flanking sequences in the concatemers, which was later confirmed by DNA sequencing of the isolated vector-cellular junctions. Site-specific integration of concatemeric wtAAV genome is a common feature in a latent infection in dividing human cells (14, 28), whereas rAAV integrates in vitro as a single copy or at low copy numbers, as opposed to a tandem repeat, and without specificity (22, 25, 38). Although the mechanism by which rAAV concatemerizes in vivo but not in vitro is not yet clear, a possible interpretation of this discrepancy is that the in vitro assays used proliferating cell lines whereas the majority of cells in adult mouse liver are presumed to be quiescent, where the cellular machinery responsible for integration may differ from that in dividing cells.

The head-to-tail forms had been interpreted to represent integrated rAAV vector forms, since head-to-tail tandem arrays are typically associated with integrated wtAAV in latent infection while head-to-head and tail-to-tail forms are generated during productive infection (19). However, this study and a recent study by Duan et al. (6) demonstrated the presence of episomal circular head-to-tail monomer forms of rAAV in transduced animal tissue. We observed a considerable number of such episomal circular intermediates even in samples taken 5 and 9 months postinjection. The circular intermediates also possessed ITR deletions of various sizes, similar to the observation in the head-to-tail junctions of integrated wtAAV (references 6 and 21a). Therefore, PCR-based assays designed to detect head-to-tail junctions cannot be used to demonstrate integration, and the structure of the ITRs analyzed by PCR are not representative of the ITRs of integrated rAAV vector.

It is not clear how efficiently rAAV vectors integrate into the host genome in vivo. However, taking into consideration that only up to 5% of hepatocytes are transduced by rAAV (13, 20, 31, 36) and that the efficiency of plasmid rescue is low even when stably transduced cell lines are used (up to 100 transformations per μg of DNA by Rutledge and Russell [25]), the fact that we could isolate junctions at rates of 4 and 1.3 junctions/μg of DNA suggests that the majority of the high-molecular-weight signal originated from integrated vector forms.

Vector-cellular DNA junction sequences of wtAAV and rAAV proviruses have been extensively examined in immortalized cell lines in vitro (2, 3, 7, 9, 16, 19, 24, 25, 28, 38). Their common features are as follows: (i) the junctions are preferentially located in or near the ITRs, (ii) no complete ITRs are observed at any junctions, (iii) duplication or deletion of a short stretch of the ITR sequence sometimes occurs, (iv) target cellular DNA and vector sequences never share significant homology, (v) both flip and flop orientations of the ITRs are observed, and (vi) the vector and/or flanking cellular DNA sequences are occasionally amplified in both direct and inverted orientations. Despite such advances in the elucidation of the structure of wtAAV or rAAV vector-cellular DNA junctions in dividing cells in vitro, analysis of junctions of rAAV integrants in nondividing cells in vitro or in vivo has been hindered by technical difficulties in the isolation of junctions. Recently Wu et al. first demonstrated that rAAV integrated in nondividing neurons in vitro and in rat brain in vivo by using Alu PCR (35). They characterized a junction isolated from a nondividing neuron, which showed a simple crossover at the palindromic region of the ITR without significant rearrangement. In the present study, we isolated 18 junctions from mouse livers by a plasmid rescue technique. It is noteworthy that all of these 18 junctions isolated from mouse liver shared some of the six above-mentioned features of junction sequences, and all the features were observed in at least 1 of the 18 junctions. The ITR sequence has been reported to possibly be unstable in E. coli (19, 27); therefore, we cannot totally exclude the possibility that the junction sequences we observed contained prokaryote-induced deletions. However, as long as two intact ITRs were not placed close to each other, we have not seen any ITR deletions in DH10B cells under standard culture conditions. We also used another strain of E. coli for cloning unstable DNA (SURE strain; Stratagene) to isolate episomal circular intermediates and integrants. Since we failed to retrieve rAAV-cellular DNA junctions from this strain, which was presumably due to lower transformation efficiency of SURE in our system (approximately 1 × 10⁹ to 2 × 10⁹ transformants/μg of pUC18 DNA) compared to that of DH10B (≥1 × 10¹⁰ transformants/μg of pUC19 DNA), we could not directly compare retrieved vector-cellular DNA junction sequences in these two strains. However, comparison of vector-vector DNA junction sequences in these two strains. However, comparison of vector-vecctor junction sequences of circular episomal forms isolated by SURE and DH10B cells revealed no fundamental differences in the junction sequences involving the ITRs.

The junction sequences of J104 were of special interest in that repetition of both the vector and the flanking genomic sequences and interposition of genomic sequences in provirus were observed. Similar molecular events have been previously reported in human Detroit 6 cells latently infected with wtAAV (3, 19). Although the repeated stretch in J104 did not include the ITR, which might be recognized as an origin of replication by cellular enzymes (19), it may have been created by DNA amplification at the junction site. This amplification could occur during the integration process rather than after integration, since minor differences were detected at each junction site of these two repeat units (Fig. 9B). The probability of different DNA sequence alterations occurring precisely at each rAAV vector-cellular DNA junction of the repeat (the junctions of the 3′ end of U1 and the 5′ end of the inverted δEF1α-P [Fig. 9A]) is predicted to be small if the amplification took place after integration. We assume that the vector and genomic DNAs formed a complex intermediate during the integration process, where the vector DNA associates with host DNA at three different sites (Fig. 9C). It is not clear at this time whether the two unknown cellular DNA sequences of 0.2 and 2.5 kb reside on the same DNA strand or two different chromosomes.

wtAAV targets the AAVS1 sequence in human cells, while rAAV presumably integrates randomly. In this study, we demonstrated that 2 of 18 genomic sequences targeted by rAAV were identified as genes (the mouse α1 collagen gene in one and the mouse 45s pre-rRNA gene in another) which are transcribed in the liver (23). This observation is reminiscent of recent studies on hot spots of retrovirus integration (5, 29), which had been assumed to occur randomly. Shih et al. reported that Rous sarcoma virus favors actively transcribed genes and CpG-rich islands for integration (29). Discovering whether rAAV vectors target transcriptionally active genes awaits further characterization of more genomic sequences targeted by rAAV vectors in vivo.

In conclusion, we have demonstrated direct evidence of rAAV integration into the host genome of mouse liver by isolating and characterizing rAAV vector-cellular junctions. This observation provides an explanation for the sustained levels of gene expression observed with rAAV vectors. Further elucidation of the integration mechanism will be important in assessing the effects of integration on the host.