Human Gene Targeting by Adeno-Associated Virus Vectors Is Enhanced by DNA Double-Strand Breaks

Abstract

The use of adeno-associated virus (AAV) to package gene-targeting vectors as single-stranded linear molecules has led to significant improvements in mammalian gene-targeting frequencies. However, the molecular basis for the high targeting frequencies obtained is poorly understood, and there could be important mechanistic differences between AAV-mediated gene targeting and conventional gene targeting with transfected double-stranded DNA constructs. Conventional gene targeting is thought to occur by the double-strand break (DSB) model of homologous recombination, as this can explain the higher targeting frequencies observed when DSBs are present in the targeting construct or target locus. Here we compare AAV-mediated gene-targeting frequencies in the presence and absence of induced target site DSBs. Retroviral vectors were used to introduce a mutant lacZ gene containing an I-SceI cleavage site and to efficiently deliver the I-SceI endonuclease, allowing us to carry out these studies with normal and transformed human cells. Creation of DSBs by I-SceI increased AAV-mediated gene-targeting frequencies 60- to 100-fold and resulted in a precise correction of the mutant lacZ reporter gene. These experiments demonstrate that AAV-mediated gene targeting can result in repair of a DNA DSB and that this form of gene targeting exhibits fundamental similarities to conventional gene targeting. In addition, our findings suggest that the selective creation of DSBs by using viral delivery systems can increase gene-targeting frequencies in scientific and therapeutic applications.

Adeno-associated virus (AAV)-mediated gene targeting is the precise alteration of a specific chromosomal site by an AAV vector. AAV gene-targeting vectors contain inverted viral terminal repeats and genomic DNA sequences homologous to the chromosomal target except for the genetic change to be introduced (18, 35). Up to 1% of normal human cells can undergo gene targeting by AAV vectors (17), and the sequence changes introduced include 1- and 2-bp substitutions, small deletions, and insertions of up to 1.5 kb (17-19). While these targeting frequencies are 3 to 4 orders of magnitude higher than can be achieved by conventional gene targeting in the same human cell types (3, 31, 47), they are still too low for most therapeutic applications, especially in clinical settings where vector delivery may not be optimal, and even 1% targeting frequencies may be difficult to achieve. Thus, an improved understanding of the mechanism of AAV-mediated gene targeting and the development of methods for increasing targeting frequencies are essential if therapeutic gene targeting is to succeed.

Conventional gene targeting and AAV-mediated gene targeting may involve distinct mechanisms, given their different frequencies and the unique topology of the AAV vector genome. The efficient nuclear delivery of single-stranded AAV targeting constructs, even in primary cells resistant to transfection, presumably contributes to the high targeting frequencies observed. In contrast, conventional gene targeting by double-stranded DNA constructs may be limited by transfection efficiencies, nuclear delivery of targeting constructs, and/or unwinding of double-stranded DNA to allow pairing. There could also be mechanistic differences after pairing with homologous chromosomal targeting sequences has occurred. The AAV-mediated reaction likely involves only three DNA strands rather than the four-stranded intermediate created when transfected plasmid constructs pair with the chromosome. This idea is supported by previous studies showing that the majority of AAV vector genomes remain single stranded after entering cells (37) and that double-stranded versions of the AAV vector genome created by packaging dimers do not contribute to gene targeting (18). Another important factor is the structure of the AAV inverted terminal repeats, which can pair to form T-shaped hairpins that may bind cellular factors important for gene targeting.

The mechanism of conventional gene targeting has been extensively studied through analysis of the effects of double-strand breaks (DSBs) on targeting frequencies. Early studies in Saccharomyces cerevisiae showed that DSBs present in targeting plasmids stimulated Rad52-dependent homologous recombination between the plasmid and the chromosome (28, 29) and led to the DSB model of homologous recombination (46). Similar experiments with linearized plasmids in vertebrate cells demonstrated enhanced recombination between plasmids (25) and increased gene-targeting frequencies (43). Later studies showed that DSBs present at chromosomal target loci also increase gene-targeting frequencies (5, 6, 34, 41). In addition, disruption of vertebrate genes encoding homologues or paralogues of yeast recombination proteins in the Rad52 epistasis group decreased DSB-induced homologous recombination and conventional gene targeting (7, 11, 22, 30, 33). Thus, homologous recombination in yeast and conventional gene targeting in vertebrate cells can occur by a similar mechanism involving DSBs. Here we have studied the effects of DSBs on AAV-mediated gene targeting, both to improve our understanding of the targeting reaction and to develop methods for enhancing the process.

MATERIALS AND METHODS

Plasmids and DNA analysis.

Foamy retrovirus vector plasmid pCnZPNO contains a nucleus-localized lacZ gene and bacterial promoter (AccI-StuI fragment of pPD16.43) (12), mouse phosphoglycerate kinase (PGK) promoter (NotI-HindIII), neomycin phosphotransferase (neo) gene with a bacterial Tn5 promoter (BamHI-Esp3I fragment of pSV2neo) (44), and p15A replication origin (SspI-Bst1107I fragment of pACYC184) (4) in a deleted foamy vector backbone with a cytomegalovirus (CMV) promoter (pCBglKPac) (48). pCnZ1450+22PNO was made from pCnZPNO by annealing 5′ phosphorylated oligonucleotides 5′-GATCATTACCCTGTTATCCCTA-3′ and 5′-GATCTAGGGATAACAGGGTAAT-3′ and by insertion at a lacZ BclI site to create an I-SceI recognition site at bp 1450 of the lacZ gene (bp 1 being at the first ATG 3′ of the multiple cloning site in pPD16.43) and was confirmed by DNA sequencing. I-SceI-expressing murine leukemia virus (MLV) vector plasmid (pLSceISHD) was constructed by inserting the I-SceI gene (EcoRI-BamHI fragment of pCMV-I-SceI3xnls [26]) into control vector plasmid pLXSHD (27) digested with the same enzymes. Retroviral vector helper plasmid pCI-VSV-G was obtained from Garry Nolan. AAV2 vector plasmid pA2nZ3113 contains the 5′ portion of the lacZ gene from pCnZPNO (NotI-EcoRI fragment of pCnZPNO) in a backbone based on pTRbSN (21) that lacked a polyadenylation signal. AAV2 vector plasmid pA2RHbSN contains an RSV promoter, a hygromycin phosphotransferase gene, an intron, and a simian virus 40 polyadenylation signal inserted into the pTRbSN backbone (21). Plasmid DNAs were purified using a plasmid maxi kit (Qiagen Inc., Valencia, Calif.). Genomic DNAs were isolated, and Southern blots were performed by standard techniques (39). Southern blot hybridization signals were quantified by PhosphorImager analysis (Molecular Dynamics, Sunnyvale, Calif.).

Cells and cell culture.

All cells were grown at 37°C in 5% CO₂ in Dulbecco's modified Eagle's medium containing 4 g of glucose per liter (Gibco/Invitrogen, Carlsbad, Calif.), 10% heat-inactivated fetal bovine serum, penicillin, and streptomycin. Primary, normal male human fibroblasts (MHF2) were obtained from the Coriell Institute for Medical Research (Camden, N.J.) (catalog no. GM05387). HT-1080 human fibrosarcoma cells (32), Phoenix-GP cells (24), and 293T cells (10) have been described previously.

HT-1080 and MHF2 cells containing proviral target sites were generated by transduction with foamy virus vector CnZ1450+22PNO and selection with G418 (0.3 mg of active compound per ml) until all cells in control dishes had detached (10 to 12 days). Drug-resistant clones of HT-1080 cells were isolated with cloning rings. G418-resistant polyclonal MHF2 populations were derived from >10⁴ independent transduction events, as determined by seeding dishes with dilutions of transduced cells, selecting in G418 the next day, and counting the number of G418-resistant colonies.

Vector preparations.

Concentrated, helper-free foamy virus vector preparations of CnZP1450+22PNO were made by transient transfection of 293T cells with pCnZP1450+22PNO and helper plasmids, and the titer was determined by counting the G418-resistant colonies of transduced HT-1080 or MHF2 cells present after serial dilution of infected cell populations as described previously (49). MLV vectors LSceISHD and LXSHD were made by transient transfection of Phoenix-GP cells with pCI-VSV-G and vector plasmids pLSceISHD and pLXSHD, respectively (1:1 ratio), replacing the culture medium 16 and 48 h later, harvesting conditioned medium after 16 h of exposure to cells, and filtering through a 0.45-μm-pore-size syringe filter. These preparations were then concentrated 50- to 100-fold by centrifugation (51), and the titer was determined using histidine-free Dulbecco's modified Eagle's medium containing 5 mM l-histidinol for selection as described above. Transduction with MLV vectors was performed in the presence of 4 μg of Polybrene (Sigma-Aldrich Corp., St. Louis, Mo.)/ml. AAV vector AAV2-nZ3113 (serotype 2) was made by transfection of 293T cells with pDG (14) and pA2nZ3113, Benzonase treatment of cell lysates, purification by iodixanol step gradient and heparin affinity column (HiTrap; Amersham Biosciences, Uppsala, Sweden) (52), and salt removal with a HiTrap desalting column. AAV vector titers were based on the amount of full-length single-stranded vector genomes detected by alkaline Southern blot analysis (21). AAV vector AAV2-RH was made the same way but using vector plasmid pA2RHbSN instead of pA2nZ3113.

Generation of genomic DSBs and gene targeting.

DSBs were generated at target site loci by seeding clonal HT-1080 or polyclonal MHF2 cells containing the target site provirus (CnZ1450+22PNO) at 5 × 10⁵ cells/10-cm-diameter dish on day 1, infecting with MLV vector LSceISHD (or control vector LXSHD) on day 2 (multiplicity of infection [MOI] of 1), and changing the culture medium on day 3. On day 4, the cells were seeded for AAV-mediated gene targeting and a portion was saved for genomic DNA isolation. Gene-targeting assays were performed by seeding 5 × 10⁴ cells/well in 24-well dishes, infecting with AAV2-nZ3113 on day 5, transferring 0.25 and 99.75% of the cells to separate 10-cm-diameter dishes on day 6, replacing the medium every 3 days until day 14, and then staining the 0.25% dish with Coomassie brilliant blue G and the 99.75% dish with 5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside (X-Gal) (40). The total number of viable cells per original well was determined by colony counts of the 0.25% dish (values ranged from 2.4 × 10⁴ to 5 × 10⁴ viable cells/original well). β-Galactosidase-positive (β-Gal⁺) foci were counted on the 99.75% dish, and targeting frequencies were expressed as the number of β-Gal⁺ foci/10⁵ viable cells seeded.

To isolate targeted HT-1080 cell lines, live cells were stained for β-gal activity with the Detectagene Green CMFDG lacZ gene expression kit (Molecular Probes, Eugene, Oreg.) in the presence of chloroquine to inhibit endogenous β-Gal activity and sorted with a Vantage SE cell sorter (Becton Dickinson, Franklin Lakes, N.J.) 7 days after infection with AAV2-nZ3113.

Shuttle vector rescue in bacteria.

Rescue of CnZ1450+22PNO shuttle vector sequences was performed as described previously (38), except for the following modifications. Genomic DNAs (10 μg) containing integrated CnZ1450+22PNO proviruses were digested with SexAI, extracted with phenol and chloroform, and precipitated with ethanol. DNA fragments were resuspended in 269.5 μl of H₂O and brought to 300 μl with 30 μl of 10× ligation buffer and 0.5 μl of T4 DNA ligase (New England Biolabs, Beverly, Mass.) at a concentration of 400 U/μl. Ligations were incubated at 15°C overnight, extracted with phenol and chloroform, and precipitated with ethanol. The DNA pellets were resuspended in 5 μl of H₂O, and Escherichia coli strain DH10B (15) was transformed by electroporation with ∼8 μg (4 μl) of DNA. Growth of transformed bacteria was selected on agar containing 50 μg of kanamycin/ml and spread with 80 μg of X-Gal and 80 μg of isopropyl-β-d-thiogalactopyranoside (IPTG).

RESULTS

DSB formation by I-SceI in human cells.

HT-1080 cells containing recognition sites for the I-SceI homing endonuclease were generated by transduction with foamy retrovirus vector CnZ1450+22PNO, which contains a CMV-driven nuclear lacZ target gene interrupted by a 22-bp insertion that includes the 18-bp I-SceI site (Fig. 1). Like other retroviral vectors, foamy virus vectors integrate and establish long terminal repeat-flanked, proviral genomes at multiple locations in host chromatin (23, 36). The CnZ1450+22PNO vector also expresses neomycin phosphotransferase (neo) from both mammalian PGK and prokaryotic Tn5 promoters, allowing selection in both mammalian and bacterial cells with G418 and kanamycin, respectively. The p15A plasmid origin was included to allow replication of rescued target sites as bacterial plasmids. Several transduced, G418-resistant HT-1080 clones were screened by Southern blot hybridization, and one containing a single vector provirus was used for subsequent experiments (HT-1080/CnZ1450+22PNO).

FIG. 1.

Vectors used in the study. Maps of foamy retrovirus target site vector CnZ1450+22PNO, AAV targeting vector AAV2-nZ3113, and MLV vectors LXSHD (control) and I-SceI-expressing LSceISHD are shown with the viral long terminal repeats (LTR), CMV, PGK, Tn5, and simian virus 40 (SV40) promoters, lacZ, hisD, and neo genes, nuclear localization signals (nls), and the p15A replication origin. Arrows indicate transcription start sites. The probe used in Southern blot analysis as well as relevant restriction enzyme sites and the I-SceI site in CnZ1450+22PNO are shown.

MLV vector LSceISHD was used to express the I-SceI endonuclease and the selectable marker histidinol dehydrogenase (hisD) (Fig. 1). HT-1080/CnZ1450+22PNO cells were infected with LSceISHD or control vector LXSHD (lacking I-SceI) at an MOI of 1 transducing unit/cell and selected for hisD expression. Genomic DNA isolated from cells that were transduced with LSceISHD contained a fraction of target sites that were resistant to in vitro cleavage by I-SceI (see Fig. 3A), indicating prior cleavage and inaccurate repair of the DSB at the I-SceI recognition site. These experiments demonstrate that expression of I-SceI in human cells is not lethal and show that the target site is accessible to enzyme digestion in this cell clone.

FIG. 3.

DSBs increase AAV-mediated gene-targeting frequencies in HT-1080 cells. (A) Southern blot of genomic DNAs from HT-1080/CnZ1450+22PNO cells containing a single copy of the CnZ1450+22PNO provirus, which were digested with BglII and incubated with I-SceI in vitro or exposed to I-SceI in vivo by infection with MLV vector LSceISHD and selection of transduced cells. DNAs were probed for 3′ lacZ sequences, and the positions of size standards (in kilobases) are shown on the left. (B) HT-1080/CnZ1450+22PNO cells transduced by MLV vector LSceISHD (+) or LXSHD (−) were infected with AAV2-nZ3113 at the indicated MOIs (vector particles/cell) and assayed for AAV-mediated gene targeting by staining infected cell cultures for β-Gal expression (see Materials and Methods). Gene-targeting frequencies are shown as β-Gal⁺ foci/10⁵ cells. The black and gray columns represent results from two independent experiments. (C) Southern blot analysis of HT-1080 clones targeted with AAV2-nZ3113. Genomic DNAs were purified from parental, untargeted HT-1080/CnZ1450+22PNO cells and five targeted clones isolated by fluorescence-activated sorting of cells that contained lacZ target sites corrected by AAV-mediated gene targeting in the presence of I-SceI. These DNAs were digested in vitro with BglII and I-SceI or BglII alone (far right lane) and probed for 3′ lacZ sequences. The positions of size standards (in kilobases) are shown on the left.

DNA DSBs increase lacZ gene targeting by AAV vectors.

HT-1080/CnZ1450+22PNO cells containing lacZ I-SceI target sites and either I-SceI-expressing vector LSceISHD or control vector LXSHD were infected with AAV targeting vector AAV2-nZ3113, which contains a nonfunctional 5′ fragment of the lacZ gene homologous to the target locus but lacking the I-SceI recognition site (Fig. 1). Infected cells were transferred to 10-cm-diameter dishes and allowed to grow for 7 to 10 days, and then lacZ gene correction was analyzed by sorting or staining for β-Gal activity or rescue of target sites as bacterial plasmids (Fig. 2). Because both the target site and targeting vector lack functional lacZ genes, the presence of β-Gal activity indicated that the mutant lacZ gene in the proviral target site had been corrected by AAV-mediated gene targeting.

FIG. 2.

Experimental design. A schematic view of the gene-targeting experimental protocol is shown with the relevant vectors (Fig. 1) and selection (G418 or l-histidinol) used.

Gene-targeting frequencies were approximately 100-fold higher in cells expressing I-SceI and increased with the AAV vector MOI (Fig. 3B; compare at the MOI of 6 × 10⁴). No β-Gal⁺ foci were observed in the absence of I-SceI and the AAV targeting vector. A few β-Gal⁺ foci were observed when I-SceI was expressed in the absence of the AAV vector (Fig. 3B, column 3). Presumably these events represent I-SceI digestion at the target site and inaccurate DSB repair to form a mutant, yet functional lacZ gene.

Accurate removal of the I-SceI recognition sequence at targeted lacZ genes.

We used a live-cell stain for β-Gal activity to sort single cells containing corrected lacZ genes after infection with LSceISHD and AAV2-nZ3113. These cells were expanded and analyzed by Southern blotting to determine whether the target locus had undergone rearrangement (Fig. 3C). The BglII restriction enzyme cuts once within the target site provirus (Fig. 1) and once in flanking genomic DNA to generate a 13-kb junction fragment when hybridized to a 3′ lacZ probe. An intact I-SceI site in the target can be identified by further digestion in vitro with I-SceI to produce a 2.6-kb fragment. Five of five β-Gal⁺ clones contained I-SceI-resistant 13-kb junction fragments, indicating that the I-SceI sites were absent and the genomic target site locus had not undergone major rearrangements. To further assess gene-targeting fidelity, we rescued target sites as bacterial plasmids from these five clones and also from unsorted cells infected with LSceISHD and AAV2-nZ3113 (four additional plasmids from blue β-Gal⁺ bacterial colonies). In each case the rescued plasmids were not rearranged based on restriction digests, and DNA sequencing revealed wild-type lacZ sequence where the I-SceI site had been located (data not shown). We also sequenced plasmids from several control (β-Gal⁻, white) bacterial colonies and found intact I-SceI sites. These data demonstrate that AAV-mediated DSB repair is usually accurate; however, rare alterations might be generated that are beyond the sensitivity of our experiments.

DSBs increase lacZ gene-targeting frequencies in normal human fibroblasts.

We studied gene targeting in HT-1080 cells because they are immortal, near-diploid, and human. However, their transformed state may have influenced how DSBs are handled. Therefore, we performed a similar experiment with normal human fibroblasts, in which DNA repair processes and cell cycle checkpoints should be intact. Low-passage human fibroblasts were transduced with foamy retroviral vector CnZ1450+22PNO containing the lacZ I-SceI target site (Fig. 1), and a polyclonal G418-resistant population of cells was selected (see Materials and Methods). The distribution of target site locations in a polyclonal population averages differences in targeting frequencies caused by position effects or other unique characteristics of individual cell clones. This polyclonal population was then infected with the I-SceI-expressing vector LSceISHD or control vector LXSHD and the AAV2-nZ3113 targeting vector. Judged on the basis of the number of β-Gal⁺ foci obtained, the presence of I-SceI increased gene-targeting frequencies 60-fold (MOI, 6 × 10⁴; Fig. 4A), demonstrating that DSBs also stimulate AAV-mediated gene targeting in normal human cells. In comparison to the results obtained with HT-1080 cells, expression of I-SceI in the absence of the AAV2-nZ3113 targeting vector produced fewer β-Gal⁺ foci in fibroblasts (column 3 in Fig. 3B and 4A). This difference may reflect more accurate DSB repair by nonhomologous end joining in normal cells, due to the presence of intact cell cycle check points and/or DNA repair pathways that can be compromised in transformed HT-1080 cells.

FIG. 4.

DSBs increase AAV-mediated gene-targeting frequencies in normal human fibroblasts. (A) Polyclonal human fibroblasts containing CnZ1450+22PNO lacZ target sites were transduced by MLV vector LSceISHD (+) or LXSHD (−), infected with AAV2-nZ3113 at the indicated MOIs (vector particles/cell), and assayed for AAV-mediated gene targeting by staining infected cell cultures for β-Gal expression (see Materials and Methods). Gene-targeting frequencies are shown as β-Gal⁺ foci/10⁵ cells. Values are the means and standard errors of three experiments. (B) Southern blot analysis of DSB levels in normal human fibroblasts expressing I-SceI. Genomic DNAs were purified from the normal human fibroblasts used for the experiment shown in panel A after transduction with LSceISHD (in vivo I-SceI +) or LXSHD (in vivo I-SceI −). Samples were digested in vitro with SpeI with (+) or without (−) in vitro I-SceI as indicated and probed for 3′ lacZ sequences. The positions of size standards (in kilobases) are shown on the left.

Gene-targeting frequencies based on plasmid rescue of target sites.

Gene-targeting frequencies as determined by β-Gal activity in transduced cells might be underestimates when target sites are not transcribed, and the percentage of corrected target sites cannot be precisely determined from β-Gal⁺ focus counts, since individual foci have different cell numbers, indicating they arose at different times (18). The ability to rescue lacZ target sites as bacterial plasmids allowed us instead to analyze targeting frequencies by counting blue and white bacterial colonies grown on agar containing X-Gal. The percentage of blue colonies recovered is a measure of the fraction of corrected lacZ genes in the cell population, since each rescued plasmid represents a single target site, rescue does not require a functional lacZ gene, and the presence of the prokaryotic lacZ promoter should ensure adequate expression in E. coli.

Genomic DNA was isolated from normal human fibroblasts containing CnZ1450+22PNO proviral target sites and I-SceI-expressing vector LSceISHD that had been infected with the AAV2-nZ3113 targeting vector at an MOI of 3 × 10⁴, 6 × 10⁴, or 12 × 10⁴ vector particles/cell. Under these conditions, more than 99% of the β-Gal⁺ foci were due to I-SceI activity (Fig. 4A). Plasmids containing target sites were rescued from SexAI-digested genomic DNA samples by circularization with DNA ligase, bacterial transformation, and growth on agar containing kanamycin, X-Gal, and IPTG. Table 1 shows the number of blue and white colonies scored for each MOI tested and the percentages of target sites that were repaired by gene targeting. The calculated targeting frequency correlated with our prior estimates based on β-Gal activity in transduced cells and was approximately 1% at all three MOIs. Target sites in nine blue colonies were sequenced and determined to be the wild type. Plasmids from three white colonies were sequenced as controls and contained intact I-SceI sites.

TABLE 1.

Bacterial colony assay of lacZ target sites rescued from normal human fibroblasts^a

AAV2-nZ3113 MOI	No. of bacterial colonies		% Targeted loci
	Blue	White
0	0	1,705	<0.06
3 × 104	11	1,168	0.93
6 × 104	18	1,709	1.04
12 × 104	18	1,846	0.97

^aGenomic DNA was prepared from polyclonal human fibroblasts containing CnZ1450+22PNO target sites that had been transduced by LSceISHD and infected with AAV2-nZ3113 at the indicated MOIs and used for plasmid rescue in bacteria (see Materials and Methods). Blue (β-Gal⁺) and white (β-Gal⁻) bacterial colonies were scored, and the percentages of blue colonies recovered are the percentages of targeted loci.

Fraction of DSBs undergoing AAV-mediated gene targeting.

We characterized the number of DSBs present in cells at the time of AAV vector infection and estimated the efficiency of gene targeting at these sites. Figure 4B shows a Southern blot of DNA prepared from normal human fibroblasts containing I-SceI target sites and transduced with control vector LXSHD or I-SceI-expressing vector LSceISHD. Samples were digested with SpeI to excise the 4.0-kb lacZ target, which can be further cleaved to a 1.7-kb product when digested with I-SceI (Fig. 1). Cells expressing I-SceI contained a faint 1.7-kb band (after digestion with SpeI) that was absent from control cells (Fig. 4B; compare lanes 1 and 3), demonstrating that approximately 5% of target sites contained DSBs at the I-SceI recognition sequence in vivo (as determined by PhosphorImager analysis). Since 1% of all target loci were corrected at the highest MOI tested (Fig. 4A and Table 1), approximately one in five DSBs was repaired by AAV-mediated gene targeting at this MOI. While it is possible that targeting frequencies can be further increased with higher MOIs or other manipulations, there may be a theoretical limit for gene targeting that corresponds to the percentage of cells with target site DSBs. As with HT-1080 experiments, I-SceI-expressing fibroblasts also contained a portion of target sites that were resistant to I-SceI cleavage in vitro, demonstrating prior cleavage and inaccurate DSB repair in vivo (Fig. 4B, lane 4).

Nonhomologous integration of AAV vectors.

In addition to transduction by gene targeting, AAV vectors can transduce cells by random integration (16, 38, 50) or persist as monomeric and concatemeric double-stranded circular episomes (8). We estimated random integration frequencies of the AAV2-nZ3113 lacZ targeting vector in fibroblasts containing an induced DSB by subtracting the number of episomal and concatemeric vector genomes from the total number visualized on Southern blots (using the same DNA as described in Table 1) (Fig. 5). Episomal forms and concatemers were detected by digestion in the vector genome with BssSI to produce specific fragments depending on the type of vector-vector junction (head-tail, head-head, or tail-tail; Fig. 5B, lanes 1 to 4). The vector-chromosome junction fragments specific for integrated vector genomes produce fragments with sizes that differ depending on the location of BssSI sites in flanking genomic DNA. The polyclonal lacZ target sites produce a diffuse signal above 6 kb. The total number of vector genomes was determined by digestion with MscI, which cuts twice within the vector genome (Fig. 5B, lanes 5 to 8) and produces a fragment distinct from the lacZ target sites. Based on quantitation of these different fragments, random integration frequencies ranged from 1.7 to 4.6% of infected cells (total vector genomes minus episomal and concatemeric genomes). These integration frequencies are similar to those observed in previous studies (17, 20, 35). To determine whether DSBs increase random integration, we used an AAV vector containing a hygromycin phosphotransferase expression cassette (AAV2-RH), which allowed us to more accurately measure random integration frequencies by selecting for stably transduced colonies. There was no significant difference in the numbers of hygromycin-resistant colonies when cells were infected with AAV2-RH in the presence or absence of an induced DSB (data not shown), demonstrating that nonhomologous integration frequencies were not significantly increased by a single DSB.

FIG. 5.

Nonhomologous integration of AAV vectors. (A) Diagram of AAV vector concatemer forms and restriction enzyme sites used for Southern blot analysis as described for panel B. Vector orientations are indicated by the arrows. Inverted terminal repeats (ITRs) are indicated as black or white boxes. The vector-vector junction can contain 1 to 2 ITRs (9), so predicted sizes are accurate to within about 200 bp. The probe binds to sites indicated by the black bars underneath the diagrams. Note that the probe does not hybridize to the tail-tail junction fragment but detects flanking head-head or head-tail junctions from the same concatemer. (B) Southern blot of genomic DNAs from polyclonal human fibroblasts containing CnZ1450+22PNO target sites that had been transduced by LSceISHD and infected with AAV2-nZ3113 at the indicated MOIs (using the same samples as described in Table 1). DNA samples were digested with BssSI (lanes 1 to 4) or MscI (lanes 5 to 8) and hybridized to an 800-bp ClaI fragment from the 5′ end of the lacZ gene. The positions of size standards (in kilobases) are shown on the left. ss, single stranded.

DISCUSSION

An understanding of the mechanism of AAV-mediated gene targeting is the first step toward manipulating the process and achieving higher targeting frequencies. Our finding that DSBs stimulate AAV-mediated gene targeting demonstrates that the mechanism can involve a DSB and the host factors necessary for its repair. By analogy with other gene-targeting experiments in yeast and vertebrate cells (5, 6, 13, 41), these factors include those involved in DSB repair by homologous recombination, such as the Rad52 epistasis group. Repair of DSBs by AAV-mediated gene targeting can be explained by two related mechanisms (Fig. 6). After a DSB is generated, 5′ to 3′ exonuclease degradation at either side of the chromosomal break creates single-stranded DNA tails (45) that can pair with the homologous single-stranded AAV vector. Once pairing has occurred, the break in the chromosomal strand complementary to the AAV vector can be repaired by DNA synthesis, using the vector genome as a template. The other chromosomal strand can be repaired either by a second round of DNA synthesis using the newly repaired complementary strand as a template or by a double crossover and exchange of a portion of the homologous vector genome. DSBs could act at several levels to stimulate this process, including facilitating pairing by exposing chromosomal single strands, recruiting proteins that carry out the necessary reactions, and/or allowing DNA repair synthesis at the break.

FIG. 6.

Possible mechanisms of DSB repair by AAV vectors. A chromosome containing a DSB that was processed to leave single-stranded 3′ tails and the AAV targeting vector genome with ITRs are shown pairing and undergoing gene targeting by two different pathways.

Relatively low gene-targeting frequencies were observed in these experiments compared to those seen in previous studies in which frequencies of 0.1 to 1% were obtained, even in the absence of an induced DSB (17, 20, 35). This result might have been due to the type of mutation present in the lacZ target sites. Targeted insertion of DNA sequences by AAV vectors is ∼10-fold more efficient than corrections requiring deletions of target site sequences (references 17 and 20 and data not shown). Since the I-SceI recognition site present in the target locus is a 22-bp insertion, correction requires a deletion of a target site sequence and is expected to occur at a relatively low frequency in the absence of a DSB. It is also possible that the HPRT and COL1A1 genomic loci analyzed previously were efficiently targeted because they were hotspots for DSB formation.

One possible interpretation of our findings is that all gene targeting requires preexisting genomic DSBs. AAV-mediated gene targeting can occur in the absence of an introduced DSB under optimal conditions at single-copy chromosomal loci in approximately 1% of human cells (17, 35). Since the haploid genome size is 3 × 10⁹ bp and an AAV vector genome is about 4,000 bp, approximately 7,500 DSBs/haploid genome are necessary for a DSB to fall within homologous sequence in 1% of cells [(1 DSB/4 × 10³ bp) (3 × 10⁹ bp/genome)/100]. The necessary number of DSBs could be significantly lower if they need not be located exactly within the target homology. For example, if a DSB located within a 40-kb region that includes the vector homology can stimulate strand invasion, then only 750 DSBs/haploid genome are necessary to initiate recombination at a random locus and achieve a targeting frequency of 1%. Although these DSB estimates seem large given the observation that cell cycle arrest can occur with a single unrepaired DSB in yeast (1), the human cells we used were clearly able to tolerate at least a transient DSB, which presumably was repaired before or during replication. Thus, it is possible that multiple DSBs can be tolerated for a portion of the cell cycle in normal human cells and that all gene targeting requires a DSB. Further experiments are necessary to characterize the number of DSBs in normal human cells and to determine whether and where DSBs must be present in target loci to influence gene-targeting frequencies.

We used integrating viral vectors to introduce target loci and to express the I-SceI endonuclease. This system has advantages over other approaches based on transfection of these elements, since the target locus has the predictable structure of a single-copy integrated provirus and the endonuclease can be delivered to many cell types, including those resistant to transfection. Viral delivery of these vectors allowed us to efficiently introduce a specific DSB and directly visualize the unrepaired chromosomal ends by Southern blot analysis of genomic DNA. Thus, we were able to quantify for the first time the percentage of target sites containing a DSB in primary human cells, where normal DNA repair pathways and cell cycle check points are functional. Our approach should prove useful in future studies of DSB repair in normal human cells, which may process DSBs differently than transformed cells or cells from other species.

While we used a transgene and the I-SceI endonuclease to study gene targeting, the same strategy could be applied more generally by using engineered proteins consisting of DNA binding motifs linked to the FokI endonuclease domain (2, 42). These customized proteins can cleave at specific chromosomal sites and stimulate gene targeting at the desired genomic locus. Alternatively, pharmacological and/or genetic manipulation of the host factors involved in DSB repair and homologous recombination may improve targeting frequencies. Given that AAV vectors can efficiently infect many cell types from different mammalian species by both in vivo and ex vivo delivery, these approaches hold promise for achieving even higher gene-targeting frequencies in scientific and therapeutic applications.