Editing of the human genome to correct disease-causing mutations is a promising approach for the treatment of genetic disorders. Genome editing improves on simple gene replacement strategies by effecting in situ correction of a mutant gene, leading to restoration of normal gene function under the control of endogenous regulatory elements, and reducing risks associated with random insertion into the genome. Gene-specific targeting has historically been limited to mouse embryonic stem (ES) cells. The development of zinc finger nucleases (ZFNs) has permitted efficient genome editing in transformed and primary cells previously thought intractable to such genetic manipulation¹. In vitro, ZFNs have been shown to promote efficient genome editing via homology-directed repair (HDR) by inducing a site-specific double-strand break (DSB) at a target locus^2–4, but it is unclear whether ZFNs can induce DSBs and stimulate genome editing at a clinically meaningful level in vivo. Here we show that ZFNs are able to induce DSBs efficiently when delivered directly to mouse liver, and when co-delivered with an appropriately designed gene targeting vector, stimulate gene replacement through both homology-directed and homology-independent targeted gene insertion at the ZFN-specified locus (referred to here as gene targeting). Importantly, the level of gene targeting achieved is sufficient to correct the prolonged clotting times in a mouse model of hemophilia B, and remained persistent following induced liver regeneration. Thus, ZFN-driven gene correction can be achieved in vivo raising the possibility of genome editing as a viable strategy for the treatment of genetic disease.

Viral vector-mediated transfer of the wild-type (w.t.) copy of a gene that is defective in disease (gene replacement therapy) has been successfully performed in a variety of animal models, and humans^5–9. However, disadvantages of gene replacement include risks related to insertional mutagenesis^10–12, and loss of endogenous regulatory signals that control gene expression. Gene-specific targeting in mouse induced pluripotent stem (iPS) cells has highlighted the potential to overcome these challenges through ex vivo correction of a disease-causing mutation¹³, but the majority of genetic diseases affect organ systems where ex vivo manipulation of target cells is not feasible. One such organ is the liver, the major site of plasma protein synthesis including the blood coagulation factors. A model genetic disease for liver gene therapy is hemophilia B, which is caused by deficiency of blood coagulation factor IX (F.IX), encoded by the F9 gene. Most affected individuals have circulating F.IX levels below 1% of normal (5,000 ng/mL), but restoration to ~5% activity (250 ng/mL) converts severe hemophilia B to a mild form¹⁴. The majority of mutations in the F9 gene are distributed across the coding sequences for exons 2–8 (Figure 1a)¹⁵. Thus, specific targeting of any single mutant allele would not allow complete coverage of the wide spectrum of mutations found in the human population. However, ZFN-mediated targeting of a promoterless therapeutic gene fragment ^2,16 (i.e. a partial cDNA preceded by a splice acceptor site) into the first intron of F9 would allow for splicing of a w.t. coding sequence with exon 1, leading to expression of functionally active F.IX and rescue of the defect caused by most mutations. We thus sought to investigate whether ZFNs combined with a targeting vector carrying the w.t. F9 exons 2–8 could induce gene targeting in vivo and correct a mutated F9 gene in situ.

We designed ZFNs (F9 ZFNs) targeting intron 1 of the hF9 gene (Supplementary Figure 1) and confirmed their capacity to introduce a DSB at the intended targeted site (Figure 1b) and stimulate genome editing by homology-directed repair (HDR) in human cells (Figure 1c, d). This ZFN pair was found to be highly active, driving up to 45% small insertions and/or deletions (indels) characteristic of DSB repair by non-homologous end-joining (NHEJ) and ~17–18% of alleles stably carrying the NheI restriction site diagnostic of repair by HDR using a homologous donor template designed to insert a novel restriction enzyme site in the F9 locus. Similar results were obtained in the Hep3B human hepatocyte line (Supplementary Figure 2). For in vivo evaluation we generated a humanized mouse model of hemophilia B (HB) since the F9 ZFNs target a site present in hF9 intron 1 absent from the murine gene. We constructed an hF9 mini-gene¹⁷ under control of a liver-specific enhancer and promoter¹⁸, which mimics a previously identified mutation (Y155stop)¹⁹ that results in the absence of circulating F.IX protein. We knocked in this mini-gene at the mouse ROSA26 locus²⁰, confirmed its genotype (Figure 2a), and showed that the resulting transgenic mice had no detectable circulating hF.IX (Figure 2b). We then crossed these mice (hereafter referred to as hF9mut mice) with an existing mouse model that has a deletion of the murine F9 gene²¹ to generate hF9mut/HB mice to test ZFN-driven gene correction activity in vivo (Figure 3a).

To deliver the F9 ZFNs to the liver we generated a hepatotropic adeno-associated virus vector, serotype 8 (AAV8-ZFN) expressing the F9 ZFNs from a liver-specific enhancer and promoter¹⁸. To test the cleavage activity of the F9 ZFNs in vivo we injected hF9mut mice with AAV8-ZFN through the tail vein and isolated liver DNA at day 7 post-injection. Cleavage activity was measured via the Surveyor Nuclease (Cel-I) assay²² which determines the frequency of the small insertions and deletions (indels) characteristic of DSB repair by NHEJ. We observed mutation frequencies ranging from 34 to 47%, demonstrating that coupling of the F9 ZFNs with AAV8-mediated delivery promotes highly efficient genome modification in mouse liver (Figure 2c), results which were confirmed by direct sequencing of the target locus (Supplementary Figure 3).

To correct the mutated hF9 gene in situ, we generated an AAV donor template vector (AAV8-Donor) for gene targeting, with arms of homology flanking a corrective, partial cDNA cassette (“SA – wild-type hF9 exons 2–8 – polyA”) (Figure 3a). Having first established that we could readily detect HDR in vitro (Supplementary Figure 4) we co-injected hF9mut/HB mice at day 2 of life by intra-peritoneal (I.P.) injection with AAV8-ZFN + AAV8-Donor, AAV8-Mock + AAV8-Donor, or AAV8-ZFN alone. (Note that I.P. injection in neonatal mice is less efficient than tail vein injection in adult mice [compare Cel-I results in Figure 3b to those in Figure 2c], but is used since it leads to higher survival rates). At week 10 of life we extracted liver DNA to assay for gene replacement at the hF9 locus via HDR. Using primers that hybridize to the chromosome outside of the donor homology arms, generating a larger amplicon for a targeted allele, (Figure 3a, primers P1/P3, Figure 3b, upper panel) we observed HDR only in mice receiving both the donor and F9 ZFNs with targeting efficiencies in the 1–3% range (Figure 3b, upper panel). We confirmed HDR using alternative primers that hybridize to sites outside the donor homology arms and within the inserted cassette, respectively (Figure 3a, primers P1/P2 and Figure 3b, middle panel). Thus co-delivery of ZFNs and a donor template using AAV vectors leads to HDR in vivo.

To determine if ZFN-mediated gene targeting results in production of circulating hF.IX, we injected hF9mut mice at day 2 of life I.P. with AAV8-ZFN alone, AAV8-Mock + AAV8-Donor, or AAV8-ZFN + AAV8-Donor. Plasma hF.IX levels for mice receiving ZFN alone or Mock+Donor averaged <15 ng/mL (the lower limit of detection of the assay), while mice receiving ZFN+Donor averaged 116–121 ng/mL (corresponding to 2–3% of normal) (Figure 4a), significantly greater than mice receiving ZFN alone and mice receiving Mock+Donor (p≤0.006 at all time points, 2-tailed T-test, Supplementary Figure 5). Importantly, in individual mice, the amount of circulating hF.IX directly correlated with the detected level of gene targeting via HDR (Supplementary Figure 6).

To confirm stable genomic correction, we performed partial hepatectomies (PHx). Levels of hF.IX persist after hepatectomy following genome editing (Figure 4a), whereas an episomal AAV vector expressing hF.IX (AAV-hF.IX, Figure 4b) revealed markedly reduced hF.IX expression, since extra-chromosomal episomes are lost during liver regeneration²³ (Figure 4b). Control mice receiving ZFN alone or Mock+Donor continued to average <15 ng/mL post-hepatectomy (Figure 4a) (p≤0.01 at all time points, 2-tailed T-test, Supplementary Figure 5).

To ensure hF.IX expression did not result from random donor integration into the genome, we injected wild-type mice (lacking the hF9mut mini-gene) at day 2 of life I.P. with AAV8-ZFN alone, AAV8-Mock + AAV8-Donor, or AAV8-ZFN + AAV8-Donor. Importantly, plasma hF.IX levels for mice in all groups averaged < 15, 30, and 30 ng/mL, respectively (Figure 4c), indicating the majority of hF.IX expression in ZFN+Donor treated hF9mut mice came from specific gene correction. PCR targeting assays in these w.t. control mice were negative, indicating amplicons used to quantify HDR were target gene specific (Supplementary Figure 7).

To determine if ZFN-mediated gene targeting would provide circulating hF.IX levels sufficient to correct the HB phenotype, we injected hF9mut/HB mice at day 2 of life I.P. with AAV8-ZFN alone, AAV8-Mock + AAV8-Donor, or AAV8- ZFN + AAV8-Donor. Plasma hF.IX levels for mice receiving ZFN alone again averaged < 15 ng/mL. Mice receiving Mock+Donor averaged < 25 ng/mL, and mice receiving ZFN+Donor had significantly higher hF.IX levels (p≤0.04 at all time points compared to Mock+Donor, 2-tailed T-test, Supplementary Figure 5), averaging 166–354 ng/mL (3–7% of normal circulating levels) (Figure 5a). A titration of AAV-Donor showed degree of correction was dependent on AAV-Donor dose (Supplementary Figure 8). To assay whether the HB phenotype was corrected, we assayed activated partial thromboplastin time (aPTT), a measure of clot formation kinetics that is markedly prolonged in hemophilia. The average aPTTs for wild-type mice (n=5) and HB mice (n=12) were 36 seconds and 67 seconds, respectively (Figure 5b). Mice receiving Mock+Donor (n=3) averaged 60 seconds, while mice receiving ZFN+Donor (n=5) had significantly shortened aPTTs, averaging 44 seconds (p=0.0014 compared to Mock+Donor, 2-tailed T-test). Clotting times for ZFN+Donor and wild-type mice were not significantly different (p=0.086, 2-tailed T-test) (Figure 5b). Together, these data demonstrate clinically significant correction of the coagulation defect in hemophilia B via direct in vivo delivery of ZFNs to mediate permanent correction of the genome in mouse hepatocytes.

To begin to evaluate the specificity of this approach, we used a SELEX-based method²² to identify the top 20 potential off-target sites for the F9 ZFNs within the mouse genome. Cel-I assays performed at each of these sites were unable to detect cleavage in 19 out of 20 (lower limit of detection 1%). At the 20th site, located in an intergenic region at mouse chromosome 9qE3.1, we detected cleavage at 1/10 the frequency seen at the F9 target site (Supplementary Figure 9). Thus the specificity of the hF9 ZFNs are comparable to the CCR5-specific ZFNs by this analysis²².

To further investigate the specificity of the ZFN approach we employed LM-PCR and 454 pyrosequencing to detect sites of AAV vector integration genome wide²⁴. Comparison of ZFN+ Donor and Mock + Donor mice revealed similar distributions of AAV integration sites across the mouse genome (Supplementary Figure 10), which, consistent with previously reported data, favored genes^24,25, but not oncogenes as integration sites. We next validated the prediction from in vitro studies²⁶ that a ZFN-induced DSB would capture the AAV vector itself, by employing a direct PCR approach using primers that anneal to the hF9mut locus and the AAV8 ITR (Supplementary Figure 11). This assay confirmed AAV integration at the ZFN target site in ZFN + Donor mice but not in the Mock + Donor mice. Finally, pre-clinical evaluation of toxicity in injected and control mice showed no effects on growth or weight gain in either hF9mut or wild-type mice (n=43) over 8 months of observation (data not shown), and no changes in LFTs at 4, 29, and 32 weeks following injection (Supplementary Figure 12), suggesting that the treatment was well-tolerated.

Studies showing that ZFNs can efficiently mediate gene correction through the introduction of site-specific DSBs, and induce HDR in cultured cells have provided important proof-of-concept results for the clinical application of engineered nucleases for diseases affecting cells that can be removed and returned to the patient. However, the necessity to isolate and manipulate cells ex vivo limits application to a subset of genetic diseases. Our results demonstrate that AAV delivery of donor template and ZFNs in vivo induces gene targeting, resulting in measurable circulating F.IX levels. This therapeutic strategy is sufficient to restore hemostasis in a mouse model of hemophilia B, thus demonstrating genome editing in an animal model of a disease. Clinical translation of these results will require optimization of correction efficiency and a thorough analysis of off-target effects within the human genome, an issue that we have begun to monitor. Together these data demonstrate that AAV-mediated delivery of ZFNs and a donor template effect persistent and clinically meaningful levels of genome editing in vivo, and thus can be an effective strategy for targeted gene disruption or in situ correction of genetic disease in vivo.