Targeted Gene Therapies: Tools, Applications, Optimization

Abstract

Many devastating human diseases are caused by mutations in a single gene that prevent a somatic cell from carrying out its essential functions, or by genetic changes acquired as a result of infectious disease or in the course of cell transformation. Targeted gene therapies have emerged as potential strategies for treatment of such diseases. These therapies depend upon rare-cutting endonucleases to cleave at specific sites in or near disease genes. Targeted gene correction provides a template for homology-directed repair, enabling the cell's own repair pathways to erase the mutation and replace it with the correct sequence. Targeted gene disruption ablates the disease gene, disabling its function. Gene targeting can also promote other kinds of genome engineering, including mutation, insertion, or gene deletion. Targeted gene therapies present significant advantages compared to approaches to gene therapy that depend upon delivery of stably expressing transgenes. Recent progress has been fueled by advances in nuclease discovery and design, and by new strategies that maximize efficiency of targeting and minimize off-target damage. Future progress will build on deeper mechanistic understanding of critical factors and pathways.

THE NEED FOR GENE THERAPIES

Gene therapies are emerging as a technically sophisticated but practical approach to treatment of human disease. The term gene therapy is broadly used to describe modifications of the genome that restore function of a defective essential gene or abolish function of a disease gene. The initial strategy was to provide cells with a functional version of a defective gene, much like transgenic mice were generated to carry genes of interest. More recently, targeted strategies for gene therapy have been developed that enlist a cell's own repair pathways to correct or disrupt a target gene.

A wide spectrum of diseases are candidates for treatment by targeted gene therapies. Many devastating human diseases are monogenic disorders, caused by mutations in a single gene. Such genetic deficiencies can now be treated, in principle and increasingly in practice, by correcting the mutation that causes disease. Acquired diseases, including cancer and some infectious diseases, are also candidates for targeted gene therapies. The deregulated cell proliferation typical of cancer cells is often the result of mutations that increase expression or activity of oncogenes, and the course of disease could be slowed or arrested by therapies that inactivate those oncogenes. Some infectious diseases — most notably AIDS — are caused by pathogenic microorganisms that take up residence in the nucleus. These and other acquired diseases could be treated by gene disruptions designed to inactivate the infectious agent, or protect cells from infectivity.

TARGETED GENE CORRECTION OR DISRUPTION THERAPIES

Targeted gene therapies enlist a cell's own DNA repair pathways to correct or disrupt a disease gene. The first step is cleavage of the chromosome at a site very near the mutation using rare-cutting endonucleases that recognize long (15-30 bp) DNA sequence motifs predicted to be unique in the targeted genome (Figure 1). Cleavage at the recognition site may then initiate homologous recombination (HR) driven by donor templates that are supplied along with the targeting endonuclease, to effect targeted gene correction. Alternatively, cellular factors associated with the nonhomologous end joining (NHEJ) pathway may create short deletions or insertions that inactivate gene function, to effect targeted gene disruption.

Figure 1. Targeted gene therapies and transgene-based therapies.

Targeted gene therapies (left) enlist the cell's DNA repair pathways to correct or disrupt a gene of interest. In the first step, the chromosome is cleaved by a rare-cutting endonuclease, to generate a double-strand break (DSB) as shown, or a DNA nick. DNA repair by homologous recombination with an exogenous donor template can result in targeted gene correction, at either a DSB or nick. A DSB can also undergo repair via the nonhomologous end-joining pathway, creating a short deletion that inactivates gene function to effect targeted gene disruption. Transgene-based therapies (right) provide a therapeutic gene typically utilizing viral vectors. Integration of the transgene is typically not targeted to a specific locus and can result in insertional mutagenesis. A color version of this figure is available online.

To initiate targeted gene correction, a cell is provided with two essential tools: a nuclease that creates a site-directed DNA lesion, and a DNA template for gene correction. The lesion — either a double-strand break or single-strand break — engages the cellular DNA repair machinery, which then carries out HR with the template DNA (Figure 1, left). This alters the chromosomal DNA sequence to correct the mutation in the target gene in situ. Targeted gene correction presents considerable advantages over methods of gene therapy based on delivery of a transgene that supplies an intact copy of the gene. In targeted gene correction, the corrected gene remains in its natural location, under control of its own promoter and enhancer, and no exogenous gene needs to integrate into the genome, diminishing the danger of insertional mutagenesis.

In most eukaryotic cells, the spontaneous frequency of HR at any given locus is low, but can be increased by DNA cleavage. The ability of DNA cleavage to induce HR was first demonstrated in experiments using the homing endonuclease (HE) I-SceI, in yeast (Nickoloff et al., 1986; Plessis et al., 1992), then plant cells (Puchta et al., 1993), and subsequently mammalian cells (Rouet et al., 1994b). This pioneering research led the way to detailed molecular analysis of the mechanisms of HR and NHEJ, central to current targeted gene therapies. Proof of principle that site-directed DNA cleavage was applicable to correcting specific genes was provided by experiments that demonstrated the ability of zinc finger nucleases (ZFNs) to elevate HR frequency in frog oocytes (Bibikova et al., 2001), Drosophila (Bibikova et al., 2003) and human cells (Porteus and Baltimore, 2003).

Like targeted gene correction, targeted gene disruption also depends upon nuclease cleavage. However, it is carried out in the absence of an exogenous donor template using the NHEJ repair pathway, and generally results in short deletions (Figure 1, right). The potential for repair of a targeted DSB to create small deletions was first evident upon analyses of lesions created by I-SceI (Lukacsovich et al., 1994). This observation opened the way to targeted gene disruptions, which are now being applied in therapeutic contexts. For example, the CCR5 gene encodes a co-receptor essential for HIV-1 infection of T lymphocytes. Targeted disruption has been used to eliminate CCR5 expression, thereby rendering cells resistant to infection by HIV-1 (see Cannon and June, 2011). This application of gene therapy by gene disruption is now in a phase I/II clinical trial (http://clinicaltrials.gov/ct2/show/NCT01252641). Other applications in infectious disease include targeting the latent viral genome to prevent viral replication (Aubert et al., 2011). Gene disruption could also be used to treat dominant monogenic diseases, such as collagen deficiencies (Chamberlain et al., 2004) or keratin genodermatoses (Bowden, 2011).

TRANSGENIC APPROACHES TO HUMAN GENE THERAPY

Targeted gene therapies contrast with approaches that seek to overcome a deficiency by delivery to the cell of a transgene that encodes a functional gene product. This is sometimes referred to as “gene replacement therapy”, even though the defective gene is not in fact replaced but its function complemented by a gene that expresses the missing gene product. This approach to gene therapy has been validated in a number of clinical trials, but it has clear limitations.

One of the earliest genetic disease treated by delivery of a transgene was the severe combined immunodeficiency (SCID) caused by deficiency in adenosine deaminase (ADA). ADA-deficiency affects fewer than 1 in 10⁵ live births worldwide, but it is the second most common cause of SCID (Buckley et al., 1997). ADA functions in purine metabolism. In its absence dATP accumulates, inhibiting ribonucleotide reductase, diminishing the dNTP pool and impairing cell proliferation. This has particularly severe consequences for lymphocytes, and causes “combined” immunodeficiency affecting both the B cell and T cell compartments.

ADA-deficiency can be treated by enzyme replacement therapy (i.e, providing the missing enzyme by infusion), but this does not fully restore immune function. To provide a functional gene, retroviral vectors were engineered to transduce the ADA gene into T cells (Blaese et al., 1995). Three studies of ADA gene therapy, including over 30 participants, have proven the effectiveness of the treatment (Fischer et al., 2011; Gaspar et al., 2011). This represents a significant advance, despite the fact that the numbers of patients treated is of necessity small as the condition is rare.

The most common severe combined immunodeficiency, SCID-XI, is caused by deficiency in the common γ chain shared by six different cytokine receptors, and encoded by the IL2RG gene on the X chromosome. SCID-XI can be treated by hematopoietic stem cell transplant, but appropriate donors cannot always be found. SCID-XI has been successfully treated by gene therapy with retroviral vectors (Hacein-Bey-Abina et al., 2010). Of the first 10 patients involved in the retroviral gene therapy trial, all but one were successfully treated for the genetic deficiency. Unfortunately, four of 10 patients developed T cell leukemia due to insertion of the retrovirus at the LMO2 proto-oncogene. The leukemia was effectively treated in three of those four patients by chemotherapy, but one of the patients succumbed to the malignancy (Hacein-Bey-Abina et al., 2008).

RARE-CUTTING ENDONUCLEASES, NATURAL AND ENGINEERED

Targeted gene therapies have emerged as potentially safer alternatives to therapies that require delivery of transgenes. There has been a recent explosion of progress in targeted gene therapies fueled largely by development of endonucleases that recognize and cleave specific sites with considerable specificity and efficiency. Three classes of rare-cutting endonucleases are currently used in gene targeting: the naturally occurring Homing Endonucleases and two classes of engineered chimeric proteins, the Zinc Finger Nucleases, and the TAL Effector Nucleases (Figure 2). The long target site recognized by theses enzymes (15-30 bp) provide considerable sequence specificity to DNA cleavage, even in larger genomes. The haploid human genome is 3 × 10⁹ bp in length, so a nuclease that recognizes a 16 bp sequence is predicted in principle to cleave approximately one site, on average (4¹⁶ = 4.3 × 10⁹). In practice, however, rare-cutting endonucleases are not truly sequence-specific. Significant off-target activity has been documented, and one of the continuing challenges has been to minimize harmful effects associated with rare-cutting nucleases.

Figure 2. Domain structures of HEs, ZFNs and TALENs.

Diagrams of domain structures of the rare-cutting endonucleases currently in use for gene therapies. Homing endonucleases (HEs) may be monomeric or dimeric, with two active sites that cleave top and bottom DNA strands (arrows). Zinc finger nucleases (ZFNs) are engineered to include domains that recognize 3 bp regions of DNA (circles), joined by linkers and fused to the FokI restriction nuclease domain, to cleave the central region on the top and bottom DNA strands (arrows). TALE nucleases (TALENs) are engineered to include domains that recognize a single DNA base pair (cylinders), joined by linkers and fused to the FokI restriction nuclease domain, to cleave the central region on the top and bottom DNA strands (arrows). A color version of this figure is available online.

Homing Endonucleases

Homing endonucleases (HEs) are naturally occurring sequence-specific endonucleases that recognize and cleave long sequence motifs in double-stranded DNA (Figure 2). They are encoded by introns, and promote intron mobilization by cleaving the recipient gene (Belfort and Roberts, 1997; Guhan and Muniyappa, 2003). HEs were first shown to activate HR in experiments that employed I-SceI cleavage at a reporter site to monitor the outcomes of DNA DSB repair (Rouet et al., 1994a). They have since emerged as promising tools for gene targeting due to their natural ability to recognize long DNA sequence motifs. In this context they are frequently referred to as meganucleases, reflecting their long DNA recognition motifs, not their molecular size. In fact, enzymes in this family tend to be relatively small (30 kD).

To date, hundreds of HEs have been identified in Archaea, Eubacteria and Eukarya and their viruses (Barzel et al., 2011). HEs are classified into five families based on their conserved active site core motif. The best characterized is the LAGLIDADG family. This family includes both homodimeric enzymes such as I-CreI, that target palindromic or near-palindromic consensus motifs, and monomeric enzymes, such as I-SceI and I-AniI, composed of two subdomains that target non-palindromic motifs. Members of the LAGLIDADG family display considerable DNA recognition specificity and have been the main focus of genome engineering applications.

HEs have proven amenable to modification driven by in silico analyses. This can be supplemented by in vitro directed evolution and selection or high throughput assays to optimize DNA cleavage (Chen and Zhao, 2005; Doyon et al., 2006; Scalley-Kim et al., 2007; Jarjour et al., 2009). Computation-based redesign was first accomplished for I-MsoI (Ashworth et al., 2006). Redesign is still time consuming and predicted variants require considerable experimental validation, so HE therapeutic targets have been chosen in part by virtue of proximity to a promising variant HE recognition motif.

Considerable engineering has focused on the normally dimeric enzyme, I-CreI. Mutations in the XPC (Xeroderma pigmentosum C) gene lead to a predisposition to skin cancers due to an inability to repair ultraviolet light-induced DNA damage, and I-CreI has been successfully engineered to generate a heterodimeric derivative that induces efficient targeted recombination in the XPC gene (Redondo et al., 2008). I-CreI has also been engineered to produce a monomeric derivative that targets the RAG-1 gene, which encodes a component of the transposase complex that catalyzes V(D)J recombination in B cells and T cells (Grizot et al., 2009).

Genome mining has enriched the repertoire of HEs and their cognate target sequences. To take advantage of the reservoir of natural diversity, the enormous public sequence databases were searched to create a library containing an estimated 416 unique target ranges for 684 HEs that can be freely accessed at http://homebase-search.tau.ac.il/ (Barzel et al., 2011). Structural conservation permits well-studied HEs to serve as scaffolds onto which specificity determinants identified in such libraries may be grafted, thereby generating novel specificities. For example, grafting of residues predicted to determine target site specificity onto the I-AniI scaffold has created an array of new specificities (Szeto et al., 2011). Engineering of I-OnuI has resulted in a derivative that targets the monoamine oxidase B gene (MAO-B), which encodes a component of the mitochondrial outer membrane important in neurotransmitter metabolism and a potential target in neurodegenerative diseases, such as Parkinson disease (Takeuchi et al., 2011). Engineering has also built on a proprietary database, largely I-CreI-derived, to generate monomeric enzymes that cleave herpes simplex virus to reduce viral load in cultured cells (Grosse et al., 2011).

Modular Engineered Nucleases: ZFNs and TALENs

Two classes of rare-cutting endonuclease, zinc-finger nucleases (ZFNs) and transcription activator-like effector nucleases (TALENs), consist of modular DNA-binding domains fused to the nuclease domain of FokI, a type IIS restriction endonuclease (Figure 2). The modularity of these enzymes enables new specificities to be readily engineered by joining DNA recognition domains.

ZFNs

ZFNs (Figure 2) are artificial proteins composed of the DNA binding domain of a zinc finger protein fused to the FokI nuclease domain (Kim et al., 1996; Porteus and Carroll, 2005; Klug, 2010). Specificity is conferred by the zinc finger, a self-contained domain composed of an α-helix which makes contacts with the DNA target, stabilized by two cysteines and two histidines, coordinated by a zinc ion (hence the name). A single zinc finger recognizes a specific 3 bp DNA sequence, and ZFNs can be engineered to recognize longer sequence motifs by fusing individual zinc finger domains. Endonucleolytic activity is the property of the FokI endonuclease domain, which is joined to the C-terminus of the zinc finger modules by a short linker. FokI only cleaves as a dimer, so (unless a recognition motif is palindromic) a heterodimer must be engineered for each DNA half-site, with a FokI monomer linked to each subunit. DNA binding by the zinc fingers then enables FokI dimerization and DNA cleavage in the spacer region between the target DNA sequence motifs.

ZFNs have been used very effectively in genome engineering to enable reverse genetics in a variety of model organisms, as well as in therapeutic contexts (reviewed by Urnov et al., 2011). Utility of ZFNs for gene therapy was established in experiments showing that ZFNs could greatly stimulate reporter gene targeting in human cells (Porteus and Baltimore, 2003). Targeting of a natural chromosomal site, the human IL2RG locus, was reported two years later (Urnov et al., 2005). More recently, ZFNs have been applied to genomic modifications of human embryonic stem cells (Hockemeyer et al., 2009; Zou et al., 2009).

TALENs

TALENs (Figure 2) consist of sequence-specific TAL DNA binding domains fused to the FokI nuclease (Bogdanove and Voytas, 2011). The transcription activator-like (TAL) modules derive from plant pathogenic bacteria and normally function to bind specific promoters and activate transcription in the nucleus of the host plant. Some TAL effectors are virulence factors, but the function of most is unknown.

TAL effectors recognize DNA via a central domain, composed of tandem repeats, typically 34 amino acids long. Each repeat binds to one base pair, with specificity determined by two amino acid residues (Boch et al., 2009; Moscou and Bogdanove, 2009). Most natural TAL effectors recognize sequences 15-19 bp in length. The TAL effector domain has been successfully customized to create targeted transcription factors. These have been used to activate genes in plants, their natural hosts (Morbitzer et al., 2011). They have also been adapted to function in human cells, after substitution of a mammalian nuclear localization signal and the transcription activation domain from VP64 (Zhang et al., 2011).

The apparently simple code that TAL effectors use for DNA recognition has spurred interest in developing TAL derivatives that cover the spectrum of specificities desired for genome engineering. The first TALENs were constructed only recently (Christian et al., 2010); nonetheless the readiness with which they can be customized and their efficacy in genome engineering has since been demonstrated in multiple contexts, including gene targeting in human iPS cells (Hockemeyer et al., 2011).

Advantages and applications of the platforms for nuclease-targeted gene therapy

HEs, ZFNs and TALENs are in current use in a great variety of applications, ranging from genomic engineering in model organisms to clinical trials for treatment of human disease. Some of those applications are outlined in Figure 3, and a more detailed list is included in Supplementary Table 1. The list is by no means exhaustive, and the field is currently moving so rapidly that no list can be complete.

Figure 3. Targeted genome engineering applications using HEs, ZFNs and TALENs.

Applications of HEs, ZFNs and TALENs in genome engineering. These nucleases have been employed to target a variety of organisms, cell types, viruses, and genes implicated in human diseases, as shown. ZFNs have been used the most extensively in genome engineering applications, but HEs and TALENs are rapidly catching up. A color version of this figure is available online. A more detailed list of the genes and organisms targeted by these endonucleases is presented in Supplementary Table 1.

The size of the target repertoire is one critical criterion for utility of nucleases used for genome engineering. The repertoire of natural HE sites is relatively small, although genome mining along with in silico design and directed evolution has expanded it considerably. Nonetheless, developing a nuclease to cleave near a specific mutation, as is necessary in targeted gene therapies, may prove to be time-consuming and expensive. The modular ZFNs and TALENs have the advantage that they can in principle be designed to cleave nearly any site. TALEN engineering is greatly facilitated by the considerable specificity of DNA binding by TAL modules, especially the ability of individual TAL modules to recognize DNA with neighbor-independent specificity. The recent momentum in customizing TALENs suggests that these may prove to be especially useful in expanding the repertoire of potential targets.

Specificity of cleavage is a second key criterion. None of the targeting nucleases developed thus far exhibits absolute sequence-specificity. As discussed in detail in later sections, improvements in specificity have been achieved by a variety of strategies for nuclease redesign and selection, as well as by improved methods for analysis of off-target cleavage in vivo.

Another important criterion for most applications is the ease of delivery to the target cell. When viral vectors are used for nuclease delivery, the size of the gene that encodes the nuclease may become an important consideration. HEs are small (approximately 30 kD), and can be encoded by relatively short genes (approximately 900 bp). ZFNs are somewhat larger, as cleavage requires dimerization of two FokI nuclease domains, each linked to a set of modules designed to recognize one of the two half-sites for binding. Each subunit of a ZFN heterodimer that recognizes an 18 bp recognition sequence is about 30 kD, and the heterodimer is encoded by two genes each approximately 900 bp in length. TALENs are significantly larger. Each 34 residue TAL repeat module recognizes only a single base pair of DNA, and the engineered polypeptide must include approximately 100 residues of additional native TAL sequence N- and C-terminal of the repeats. This means that recognition of 9 bp of DNA requires a monomer of about 500 residues in length, linked to the 200 residue monomeric FokI endonuclease domain for a total of 700 residues. Thus, the predicted size of a functional TALEN dimer is more than 150 kD, encoded by more than 4 kb of DNA. This begins to approach the limits of some viral vectors, although other methods of nuclease delivery, particularly transfection of mRNA (discussed below), appear to offer good alternatives to viral vector-based gene expression.

Alternatives to Nucleases: Using DNA to Target DNA

Several approaches to gene targeting have been reported that rely on the annealing of short oligonucleotides to specify the site of the lesion. Specificity can be determined either by Watson-Crick base pairing or, in the case of triplex forming oligonucleotides (TFOs), by formation of a non-canonical triple helical structure (Mukherjee and Vasquez, 2011). Additionally, the targeting oligonucleotide can itself be used as the donor molecule for gene correction or two oligonucleotides can be used, one to target the lesion and one as donor. DNA-based targeting has been used to correct a mutation in the human dystrophin gene, thus restoring function in a Duchenne muscular dystrophy (DMD) model (Kayali et al., 2010). A similar approach using either TFO or non-TFO oligonucleotides has corrected β-globin mutations (Chin et al., 2008; Lonkar et al., 2009).

Both TFO and non-TFO oligonucleotides can be used for genome modifications alone (reviewed in Aarts and te Riele, 2011) or when conjugated to a DNA damaging agent (de Piedoue et al., 2007; Kim et al., 2007b; Majumdar et al., 2008). An early example of oligonucleotide-directed DNA damage utilized an ¹²⁵I-TFO to make site-directed DSBs that were repaired in vitro by NHEJ (Odersky et al., 2002). Subsequent work has focused on use of psoralen-conjugated oligonucleotides to target interstrand crosslinks (Majumdar et al., 2008; Liu et al., 2009).

CELLULAR PATHWAYS OF DNA BREAK REPAIR

The DNA lesion created by rare-cutting endonucleases can be repaired by distinct repair pathways. HR with an exogenous donor template will yield gene correction, while NHEJ can yield small deletions that disrupt gene function. These pathways both maintain genomic integrity and cell viability in the face of endogenous DNA damage.

Homologous Recombination

HR repairs a DSB by utilizing a homologous DNA template. This aspect of HR renders it “error-free” when the appropriate template is used. HR frequently repairs breaks during the S and G2 phases of the cell cycle when the newly replicated sister chromatid provides an ideal proximal template for DNA repair.

The mechanism of HR has been studied in considerable detail (Heyer et al., 2010). A DSB first undergoes 5’-end resection to leave free 3'-ends (Figure 4, left). The ends become coated with the single-strand DNA (ssDNA) binding protein, RPA; and the strand-annealing protein RAD51 then displaces RPA from the ssDNA, forming a nucleoprotein filament with a free 3’-end that can invade a homologous template. This results in the formation of a D-loop, a critical intermediate. Many accessory factors are required for this step, and for subsequent removal of RAD51 necessary for progression of HR, as discussed in detail elsewhere (Heyer et al., 2010). After D-loop formation, three unique paths are available for repair: Holliday junction (HJ) formation and resolution, break-induced recombination (BIR), and synthesis dependent strand annealing (SDSA), as described below.

Figure 4. Alternate pathways of DNA DSB repair.

Homologous recombination (left) and nonhomologous end joining (right) are the two major modes of repair at a DNA DSB. In order to enter the homologous recombination pathway, a DSB must undergo extensive 5’-end resection to leave free 3'-ends that can invade the repair template, to form a D-loop (left). In the predominate homologous recombination sub-pathway in somatic cells, synthesis-dependent strand annealing, the invading 3’ end then primes limited replication using the homologous DNA as template. The newly extended end is then released from the D-loop and anneals with the other end of the processed DSB. Synthesis-dependent strand annealing transfers relatively short stretches of DNA sequence. After ligation, the resulting heteroduplex is resolved either by the mismatch repair pathway or replication and segregation to generate a corrected gene. Alternatively, a DNA DSB may enter the nonhomologous end-joining pathway (right), in which DNA ends are bound by the Ku70-Ku80 heterodimer. Limited processing occurs, which may be accompanied by small insertions or deletions, and generates DNA ends that can be ligated by the XRCC4-LIG4 complex. The factors that play important roles at various stages of these DSB repair pathways are shown in boxes. A color version of this figure is available online.

Holliday junction formation requires that the second exposed 3’-end also invades the D-loop. This creates an intermediate that can be resolved by cleavage, resulting in physical exchange of genetic material between two DNA duplexes. HR is typically error-free. If it occurs between equivalent positions on sister chromatids, as it normally does in somatic cells, no genetic changes result. However, the identity of the donor for HJ formation is critical. If HR occurs between repetitive sequences, it can result in translocations and other chromosomal rearrangements (Colnaghi et al., 2011). Recombination between homologous regions of two different chromosomes can cause loss of heterozygosity (Blackburn et al., 2004; Larocque et al., 2011). Both loss of heterozygosity and chromosomal rearrangements are potentially oncogenic (O'Keefe et al., 2010; Thompson and Compton, 2010), and this poses one safety concern for targeted gene therapy. Somatic cells have evolved mechanisms that inhibit formation of HJ (Lorenz and Whitby, 2006; Larocque et al., 2011).

BIR can occur when the free 3’-end that created the D-loop primes replication to the end of the chromosome (Llorente et al., 2008). BIR can involve multiple rounds of strand invasion and template switching (Smith et al., 2007). In yeast, BIR results in a high frequency of point mutations (Deem et al., 2011), perhaps due to replication by error-prone polymerases.

SDSA occurs when the free 3’ end that created the D-loop primes limited replication, thereby copying genetic information from the template. The newly extended end is then released from the D-loop and anneals with the other end of the processed DSB. SDSA transfers relatively short stretches of sequence and cannot result in chromosomal rearrangements. The cellular mechanisms that inhibit HJ formation in somatic cells may tend to favor SDSA as a pathway for gene correction. One important consideration in targeted gene correction is the length of sequence transferred in the course of repair — the conversion tract — as this determines the necessary proximity of the targeting endonuclease site to the disease-associated mutation. SDSA may be limited in the length of the conversion tract (Taghian and Nickoloff, 1997; Elliott et al., 1998; Larocque and Jasin, 2010), although it may be possible to extend conversion tract length by altering the balance of cellular repair factors.

Nonhomologous End Joining

NHEJ is the predominant pathway of DSB repair in mammalian cells. In NHEJ, two dsDNA ends are ligated together (Figure 4, right). Depending on the nature of the break and the processing that occurs prior to joining, NHEJ can lead to small deletions or insertions. NHEJ can also (though less frequently) join the ends of two distant DSBs resulting in large deletions, inversions and/or chromosomal translocations (reviewed in Mladenov and Iliakis, 2011). The short deletions that accompany NHEJ can be used to promote targeted gene disruption (e.g.; Holt et al., 2010).

NHEJ occurs via two distinct mechanisms; classical NHEJ and alternative, or backup, end joining. Classical NHEJ is a well-characterized pathway that has been thoroughly reviewed (e.g. Mahaney et al., 2009; Mladenov and Iliakis, 2011). In NHEJ, the Ku70-Ku80 heterodimer binds the DNA ends of a DSB where no, or limited, 5’-end resection has occurred. The Ku heterodimer then recruits and activates the DNA-dependent protein kinase catalytic subunit (DNA-PKcs). The XRCC4-LIG4 complex completes the process by ligating the ends together.

Alternative end joining is a distinct pathway from classical NHEJ (Wang et al., 2003; Bennardo et al., 2008). Importantly, alternative end joining has been shown to be responsible for the majority of translocation induced by two DSBs produced concurrently by either I-SceI or ZFNs (Simsek and Jasin, 2010; Simsek et al., 2011) and has been implicated in the formation of oncogenic translocations (reviewed in Nussenzweig and Nussenzweig, 2010). Alternative end joining could, therefore, be a source of chromosomal rearrangements following DNA cleavage, especially if an endonuclease with significant off-target cleavage is used.

Pathway Choice at DSBs: HR vs. NHEJ

The use of the HR and NHEJ pathways is tightly regulated and the “choice” of pathways utilized at a DSB can play a critical role in the success of targeted therapy (Symington and Gautier, 2010). The relative lengths of G1 and S phases of the cell cycle are one determinant, as factors that promote HR are elevated in S phase, and longer G1 phase may favor NHEJ. The structure of the resected DSB is also critical. MRN (MRE11/RAD50/NBS1) is one of the first complexes to bind a DSB. In conjunction with CtIP, MRN can initiate 5’-end resection (Limbo et al., 2007; Sartori et al., 2007). Complexes containing either the EXO1 or DNA2 endonuclease can then further excise, exposing the free 3’ end required for HR (Nimonkar et al., 2011). DSBs that do not have a free 3’ end will be repaired by NHEJ. Resection, and therefore pathway choice, is regulated by multiple factors including CDK, which may enforce the absence of HR in G1, as well as γH2AX, CtIP, BRCA1 and 53BP1 (e.g.Huertas and Jackson, 2009; Yun and Hiom, 2009; Bouwman et al., 2010; Bunting et al., 2010; Coleman and Greenberg, 2011; Helmink et al., 2011). Most homing endonucleases leave a short (2-5 bp) 3′ overhang at the cut site (Chevalier and Stoddard, 2001), while the FokI endonuclease used in ZFNs and TALENs leaves a short 5’ overhang. It remains to be seen whether a difference in polarity of the overhang affects the choice and efficiency of the repair pathway used at a DNA break.

TARGETED GENOME ENGINEERING

Genome engineering in cellular models or model organisms can be carried out by variations on the strategies for gene therapy outlined above, leading to site-directed mutagenesis or precise DNA deletion or transgene insertion (Figure 5). ZFNs have been used to construct a human embryonic stem cell model of Parkinson's disease by introducing specific mutations into the endogenous α-synuclein gene (Soldner et al., 2011). ZFNs have also been used for site-specific insertion of transgenic DNA up to 8kb in length at the human IL2RG locus (Moehle et al., 2007), and to tag the endogenous histone H3.3 variants in ES cells, thereby enabling genome wide profiling of this histone mark in different cell types (Goldberg et al., 2010). Targeted genome engineering can also be adapted to generate large DNA deletions, by simultaneous cleavage of a chromosome at two sites to promote deletion of the entire region between the breaks (Figure 5). ZFNs have been used in this sort of application, to create deletions ranging from several hundred to 15 Mbp in human cells (Lee et al., 2010). These are only a few examples of the possibilities offered by rare-cutting nucleases for targeted genome engineering (for a more complete list see Supplementary Table 1).

Figure 5. Variations of targeted genome engineering.

This diagram illustrates some of the outcomes that can result when a targeted DNA double-strand break (DSB) is repaired by homologous recombination or nonhomologous end joining. Homologous recombination (left) requires a donor homologous at least in part with the cleaved allele. Depending on details of that donor, it may be used to effect gene correction, site-directed mutagenesis, or site-specific insertion driven by matching arms of homology. Nonhomologous end joining (right) will occur in the absence of donor DNA. If cleavage is followed by excision, genetic information may be lost (or gained) at the site of the break, and gene disruption can result. Two DSBs occurring simultaneously on the same chromosome can cause loss of the intervening stretch of DNA, resulting in a defined deletion. Targeted cleavage may also lead to site-specific insertion of a transgene at a defined DSB. A color version of this figure is available online.

NUCLEASE AND DONOR DELIVERY

Targeted gene therapies require delivery to the cell of a gene expressing the initiating endonuclease, and therapies dependent upon gene correction also require a donor template. Optimization of delivery of these essential components is an ongoing and active area of research. Viruses evolved to deliver nucleic acids to cells and they are currently vectors of choice for many kinds of gene therapy (Mitchell et al., 2010). Considerable understanding of the advantages and drawbacks of viral vectors emerged from applications of these vectors to transgene delivery. Viruses require specific cell surface receptors for infection, and this natural tropism promotes the use of viral vectors in therapy directed toward some cell types. It is also possible to alter viral packaging to modify natural tropism (“pseudotyping”). However, viral vectors also pose specific drawbacks that are difficult to overcome, including limited cargo capacity, potential for insertional mutagenesis and immunogenicity.

Retrovirus and Lentivirus Vectors

Retroviral vectors have been extensively used to introduce DNA to cells due to their high efficiency of stable delivery and broad cellular tropism, but pose the risk of insertional mutagenesis. Retroviruses are diploid, and a single virion carries two RNA molecules, each 7-10 kb in length. Retroviral vectors for gene delivery need not carry the gag, pol and env genes which are essential for virion replication, but the genome must include cis-regulatory sequences for genome packaging, replication, integration and transgene transcription, which reduces the effective capacity to 5 – 8 kb. Retroviral vectors have been used in applications requiring stable expression of a transgene because integration of a DNA copy of the RNA genome is an essential stage in the retroviral life cycle. Retrovirus integration is not site-specific, but tends to favor regions near promoters, possibly because the open chromatin structure promotes integration. This creates the risk of insertional activation, the most publicized downside to retrovirus vectors. Insertional activation of proto-oncogenes has been documented in independent gene therapy trials treating diseases including X-linked severe combined immunodeficiency (Hacein-Bey-Abina et al., 2008) and chronic granulomatous disease (Stein et al., 2010). The frequency of these events varies and likely depends on multiple factors including cell type and viral vector.

To minimize the risk of insertional mutagenesis, vector engineering has sought to limit viral integration and activation of adjacent genes by shifting transcriptional activation from the viral LTR promoter to an internal promoter, by separating the promoter from the LTRs with insulator sequences (Shaw and Kohn, 2011), or by directing transgene integration to “safe-harbor” sites (DeKelver et al., 2010; Gaj et al., 2011; Gersbach et al., 2011; Papapetrou et al., 2011). Much of this effort has recently focused on lentiviruses, a genus of retrovirus that establishes infection accompanied by a long latent period. Lentiviruses can infect either dividing or nondividing cells, and establish stable chromosomal integrants. Integration-deficient lentivirus vectors (IDLV), engineered to carry mutations in integrase, exhibit greatly diminished chromosomal insertion, enabling genes or gene products to be supplied only transiently (Yanez-Munoz et al., 2006; Wanisch and Yanez-Munoz, 2009; Matrai et al., 2011). IDLVs have been successfully used for delivery of the nuclease and/or donor template to affect both targeted gene correction and gene disruption (e.g. Cornu and Cathomen, 2007; Lombardo et al., 2007).

Adeno-associated virus vectors (rAAV)

Adeno-associated virus vectors (rAAV) derive from a parvovirus, AAV, with a small single-stranded DNA genome (4.7 kb). AAV requires either a helper virus (e.g. adenovirus, herpes simplex virus) or chemical stimuli to enable efficient replication (Mitchell et al., 2010). Replication proceeds via a circular double-stranded DNA intermediate that can integrate in the chromosome as a single copy or multimers, or form nuclear episomes. There has been recent clinical success in using rAAV to deliver therapeutic genes to treat Leber's congenital amaurosis, a disease of the retina (Maguire et al., 2008) and Parkinson's disease, which affects the central nervous system (Christine et al., 2009). In both retinal and nervous tissue, gene transfer is to immunologically privileged sites comprised of nondividing cells that may favor rAAV episome maintenance. However, rAAV-driven gene therapy of other tissues, such as impaired clotting due to Factor IX deficiency (Manno et al., 2006), has been accompanied by an immune response to viral proteins.

The small size of rAAV precludes delivery of large genes by this vector, such as the gene encoding dystrophin, deficient in Duchenne muscular dystrophy. The small genome also prohibits incorporation of small genes with relatively long regulatory sequences, such as β-globin, a therapeutic target in sickle cell anemia (see Yannaki et al., 2010; Pichavant et al., 2011). Defining minimal sequences that will yield sufficiently regulated expression and still be deliverable by current viral vectors has proven difficult (Yannaki et al., 2010). Some effort has been devoted to the design of minigenes that might provide necessary gene products in trans by encoding only a fraction of the defective host polypeptide. However, such polypeptides may be immunogenic, as evident in experiments that attempted to treat muscular dystrophy with a minigene version of the very large (79 exon) dystrophin gene (Mendell et al., 2010; Moore and Flotte, 2010).

rAAV vectors can also be used to introduce defined sequence changes at homologous chromosomal loci, in the absence of nuclease cleavage (Russell and Hirata, 1998). For example, rAAV was successfully used to inactivate dominant negative mutations in the COL1A1 and COL1A2 collagen genes in mesenchymal cells from individuals with osteogenesis imperfecta (Chamberlain et al., 2004; Chamberlain et al., 2008). However, even though rAAV-mediated gene editing does not require nucleolytic cleavage, efficiency is 100-fold increased by introduction of a DSB in the chromosomal target (Porteus et al., 2003; Metzger et al., 2010).

Vector-free delivery: DNA and mRNA

Delivery of DNA on a vector that cannot replicate has the potential to enhance safety. Plasmids (pDNA) can be delivered to cells in a variety of ways, with the most common being microinjection, electroporation, and lipofection (Gao et al., 2007). The limited lifetime of pDNAs within the nucleus may not only be adequate but also provide a real advantage for genome editing applications, where a limited window of expression of the targeting nuclease may help to minimize off-target cleavage. pDNA exhibit a relatively low frequency of integration, which diminishes but does not abolish the risk of insertional mutagenesis.

Minicircle DNAs are another promising emerging technology. Minicircle DNAs are plasmids that have been stripped of any unnecessary backbone sequences and are generated by a site-specific recombination process. Several studies have already highlighted their superiority in gene delivery, as they are not immunogenic and appear not to be subject to transcriptional silencing by host cells (Chen et al., 2003; Zhao et al., 2010).

mRNA transfection offers another approach that guarantees a brief window of gene expression, with minimal potential for genomic damage. First shown to be effective over 20 years ago (Malone et al., 1989), mRNA transfection has since been very effectively used in immunotherapies that supply autologous cells with antigen receptors that recognize markers on hematopoietic tumor cells (Van Tendeloo et al., 2001; Barrett et al., 2011); and to generate iPS cells by reprogramming diverse cell types (Warren et al., 2010). This may prove to be an effective means for delivery of genes expressing engineered nucleases (particularly TALENs, which are larger than ZFNs or HEs) or factors that modulate repair pathways or chromatin structure, as discussed below. The “hit-and-run” aspect of mRNA transfection makes it especially attractive for delivery of targeting nucleases.

Donor template delivery typically requires the use of vectors, either plasmid or viral, that pose some of the same risks as the vector-borne nuclease delivery mechanisms. The use of short (≤100 nt) single-stranded oligonucleotides as donor templates for nuclease-targeted gene correction could bypass some of these risks. Single-stranded oligonucleotides were initially shown to be effective templates for targeted DSB repair in yeast (Storici et al., 2003). More recently, they have been shown to function efficiently as donor templates for nuclease-targeted gene editing in human cell lines (Radecke et al., 2006; Majumdar et al., 2008; Liu et al., 2010; Chen et al., 2011). One possible concern in a therapeutic context is the potential for sequence insertion between the ends of a targeted DSB (Radecke et al., 2010). The mechanism of transfer of genetic information from a single-stranded oligonucleotide to the chromosome is still undefined, and understanding of mechanism may identify ways to avoid such unwanted outcomes. In any event, this approach will certainly be useful in applications in which the ability to rapidly produce a large number of sequence variants is crucial.

OPTIMIZATION OF TARGETED GENE THERAPIES

Optimization of safety and efficiency of targeted gene therapies is most critical in applications that pertain to human health. In these applications, cells may be scarce and any unintended genetic alteration may have devastating consequences. Optimization of endonuclease activity and specificity should increase efficiency and reduce the danger of off-target breaks. In addition, the modulation of genetic and epigenetic factors should enhance the frequency of the desired outcome, whether correction or disruption. Possible strategies are described below.

Off-Target Cleavage Jeopardizes Safety

DSBs lead to genomic instability, so the potential for off-target cleavage presents an important safety concern for the use of customized endonucleases. Assays of affinity of protein binding to synthetic duplexes in vitro provide one measure of nuclease specificity. Such analyses have show that some but not all bases in a DNA binding motif are critical for protein/DNA interaction. However, there is not an absolute correspondence between binding affinity and cleavage (Thyme et al., 2009), so while this analysis is highly suggestive, it does not robustly predict off-target cleavage either in vitro or in vivo.

Dose-dependent toxicity of ZFNs provided an early alert to the potential of rare-cutter nucleases to carry out off-target cleavage resulting in cell death (Alwin et al., 2005; Porteus, 2006; Cornu et al., 2008; Pruett-Miller et al., 2008). A variety of strategies have been implemented to assay off-target cleavage in vivo. These include cytological profiling with established markers of double strand breaks (γ-H2AX, TP53BP1), assessment of rates of random integration, and karyotyping to assay gross chromosomal alterations. By these different measures, expression of ZFNs was found to be associated with considerable off-target cleavage activity (Perez et al., 2008; Gabriel et al., 2011).

Improved methodologies for identification of unintended target sites on a genomewide scale will also contribute to improving the safety of these enzymes. For example, toxicity of I-SceI, an HE, initially appeared relatively low, suggesting that this class of rare-cutter might be more amenable to genomic engineering (Porteus, 2006). However, this was subsequently challenged by experiments that measured genome integration of an adeno-associated virus in I-SceI-treated cells, which revealed a number of noncanonical cleavage sites (Petek et al., 2010). Insertion of integration-deficient lentiviral vectors (IDLVs) has similarly been used to tag sites of ZFN cleavage, thus marking transient and otherwise undetectable DSBs (Gabriel et al., 2011). Specificity has also been tested by coupling systematic identification of target sites in vitro with genomewide sequence analyses that assay mutagenesis at cognate cellular sites (Pattanayak et al., 2011).

Improving Target Specificity

ZFNs have been subject to several sorts of refinements to improve specificity and diminish off-target cleavage. The first ZFNs were generated by modular assembly, linking fingers that recognize specific triplet DNA sequences into a multifinger peptide to target specific endogenous loci (Kim et al., 2009). However, many zinc fingers proved to not exhibit strict functional modularity; instead, sequence-specificity was influenced by neighboring fingers, resulting in off-target cleavage.

This necessitated development of alternative strategies to identify ZFNs that recognized a target site with robust specificity. Some strategies involve structure-based redesign of protein/DNA contacts (Alibes et al., 2010). Others rely on bacterial selection systems to identify finger combinations that work well together (Maeder et al., 2008; Pruett-Miller et al., 2008) or improve ZFN specificity (e.g. Guo et al., 2010). Selections may start either from a large random library of ZFNs, or from archived pools pre-selected to recognize specific sequences (Maeder et al., 2008; Kim et al., 2011; Sander et al., 2011; Urnov et al., 2011). These strategies have led to a new generation of ZFNs that exhibit significantly less toxicity than their predecessors (Miller et al., 2007; Szczepek et al., 2007; Handel et al., 2009; Pruett-Miller et al., 2009; Gupta et al., 2011; Ramalingam et al., 2011).

ZFNs and TALENs both use the Fok I catalytic domain for nuclease cleavage. Some off-target effects may be attributed to the mechanistic properties of the Fok I domain (Halford et al., 2011). The Fok I nuclease domain must dimerize to cleave. Enzymes are designed with the intention that the ZFN domains bind DNA half-sites contained on a short contiguous target, and direct Fok I cleavage specifically to that region. However, interaction of a DNA-bound subunit with a free subunit may stimulate cleavage, making it difficult to avoid off-target cleavage at half-sites. Another source of off-target cleavage is homodimerization, which enables recognition of a palindromic site composed of two half-sites. Modification of the architecture of the dimer interface has been carried out to prevent homodimerization of ZFN subunits (Miller et al., 2007; Szczepek et al., 2007; Sollu et al., 2010; Doyon et al., 2011; Ramalingam et al., 2011).

Although specificity testing of TALENs and HEs has not been as extensive as for ZFNs, these classes of enzymes seem to have clear therapeutic potential with less re-engineering. In gene editing experiments, TALENs performed as well as ZFNs with regard to efficacy and showed greater specificity. In a study targeting endogenous loci in yeast, no evidence of off-target site was found using whole genome sequencing (Li et al., 2011). In human stem cells, only two of 19 potential off target sites were found to be occasionally disrupted but much less frequently than the intended target (Hockemeyer et al., 2011). Perhaps most significantly, in a side-by-side comparison of ZFNs and TALENs targeting the CCR5 and IL2RG genes, both nucleases showed similar gene disruption activities but TALEN showed much reduced cytotoxicity and off-target cleavage at the CCR2 locus (Mussolino et al., 2011).

Improving Safety: Initiation of Targeted Gene Correction by Nicks, not DSBs

The ability of DNA nicks (single-strand breaks) to initiate homologous recombination was first established in yeasts (Strathern et al., 1991; Arcangioli, 1998). These results were extended to mammalian cells in experiments that showed recombination could be stimuated by derivatives of the RAG1/RAG2 nuclease that can generate nicks, but not DSBs (Lee et al., 2004). More recently, DNA nicks have been shown to initiate targeted gene correction nearly as efficiently as DSBs, and they appear to have much more desirable safety properties. Initial experiments showed that a derivative of the monomeric HE, I-AniI, mutated at one if its two active sites, could promote targeted gene correction of either an episomal or chromosomal reporter (McConnell Smith et al., 2009). Subsequently, use of a reporter that can measure either targeted gene correction or disruption resulting from a DSB or nick at a single site (Certo et al., 2011) permitted direct comparison of the efficiency and safety of these two kinds of DNA lesions. Nicks were shown to initiate targeted gene correction nearly as efficiently as DSBs, but to cause orders of magnitude less NHEJ, a very desirable safety profile for gene correction (Davis and Maizels, 2011).

The HEs, ZFNs and TALENs in current use for targeted gene therapies can be readily converted from enzymes that cut both DNA strands to enzymes that nick DNA. Monomeric HEs can be converted, by a point mutation in one of their two active sites, a conversion first reported for I-SceI (Niu et al., 2008). This should be applicable to most, if not all, monomeric LAGLIDADG HEs. FokI, which provides the endonuclease domain for the modular engineered ZFNs and TALENs, can similarly be redesigned as a nickase (Sanders et al., 2009) that may be incorporated into the current ZFN and TALEN platforms (Maeder et al., 2008; Cermak et al., 2011; Li et al., 2011). Thus, the use of nicks rather than DSBs should be generally applicable for efficient and safe DNA targeting.

Improving Efficiency: Modulation of Repair Factors

Modulation of repair factor levels or activity has clear potential to enhance the efficiency of targeted gene correction. For example, enhanced expression of RAD51 paralogs has been shown to increase not only the frequency of HR but also the length of gene conversion tracts (Nagaraju et al., 2006; Nagaraju et al., 2009; Ordinario et al., 2009). Longer repair tracts could extend the range of an endonuclease engineered to target a specific gene.

siRNA knockdowns also have the potential to stimulate targeted gene correction or enhance its safety. A recent screen of siRNAs targeting more than 19,000 human genes identified 64 genes that affected the frequency of HR between chromosome and exogenous target (Delacote et al., 2011). Even if modulation of levels of a single factor has only a modest (several-fold) effect, judicious combinations may provide sufficiently improved efficiency to enable applications in therapeutic contexts, where initial cell numbers may be limiting. Repair pathway modulation does have the potential to perturb global genome stability (Moynahan and Jasin, 2010), and careful monitoring will be necessary to ensure that gains in efficiency are not compromised by loss of safety.

Given the tremendous potential of stem cells in clinical uses of TGE (Narsinh and Wu, 2010; Rahman et al., 2011), understanding details of DNA repair in these cells is of particular importance. Several recent studies have demonstrated that NHEJ in human embryonic stem cells (hESCs) is predominantly independent of ATM and DNA-PKcs. This is in contrast to NHEJ in hESC-derived neural progenitors and astrocytes (Adams et al., 2010a; Adams et al., 2010b). This alternative regulation of NHEJ could be the basis of elevated levels of NHEJ observed in hESCs (Fan et al., 2011). Consistent with these findings, levels of HR have been observed to be reduced in hESCs cell lines relative to differentiated cell lines (Fung and Weinstock, 2011). Interestingly, hESCs and iPS cells have a relatively large proportion of cells in S phase (Fan et al., 2011; Ghule et al., 2011), the cell cycle phase that favors HR. Perhaps the alternative regulation of DNA repair indicated by the ATM and DNA-PKcs independence of NHEJ is necessitated by a requirement for both rapid proliferation (short G1) and efficient DNA repair (NHEJ).

Modulation of Chromatin Structure

Chromatin structure may affect the choice and efficiency of repair pathway at the DSB. Studies comparing DSB repair efficiency in condensed vs. relaxed chromatin regions showed that DSBs induced in heterochromatic regions have slower repair kinetics than those induced in euchromatic regions (Cowell et al., 2007; Kim et al., 2007a; Goodarzi et al., 2010). The chromosomal genes targeted for correction or disruption are likely to vary in terms of DNA accessibility, so the modulation of chromatin structure holds considerable promise for enhancing the efficiency and possibly safety of targeted gene therapy.

Treatment of cells with histone deacetylase inhibitors can diminish the effects of repressive chromatin but this may be accompanied by side effects that are especially undesirable in therapeutic contexts (Mehnert and Kelly, 2007). An alternative approach is to identify specific factors that may respond to downregulation by specific siRNAs. Supporting this approach, one factor identified and vetted for functionality in a recent genomewide screen for siRNAs that promote targeted gene correction has proven to be involved in chromatin remodeling (Delacote et al., 2011).

It may also be possible to recruit chromatin-associated proteins to a lesion to favor a specific DNA repair pathway. Chromatin-associated proteins are rapidly mobilized to and from DNA breaks and include histone modifying enzymes, chromatin-remodeling factors, and DNA methyltransferase enzymes (Polo and Jackson, 2011). Interestingly, DSB modification of H3K56 occurs in a biphasic manner, initially undergoing rapid deacetylation by histone deacetylases (HDACs) 1 and 2 (Miller et al., 2010) that may promote NHEJ, followed by acetylation to favor HR. It may therefore be possible to modulate histone acetylation to tip the balance toward HR or NHEJ, depending on the application. One specific target in this pathway may be the histone deacetylase SIRT6, which has been shown to accumulate at DSBs induced by HEs, where it is necessary for stabilization of PKcs, a key factor in NHEJ (McCord et al., 2009).

Chromatin modifiers can be targeted to specific genomic sites upon expression as chimeras fused to DNA binding domains of factors that recognize specific sequence motifs (Cummings et al., 2007; Cummings et al., 2008; Soutoglou and Misteli, 2008; Yabuki et al., 2009). This approach is conceptually analogous to fusion of a nuclease domain to a DNA binding domain to engineer a ZFN or TALEN. It offers the potential to maintain the delicate balance of repair necessary for genome integrity while producing the desired outcome at the target site.

Controlling epigenetic modifications of delivery vectors may also enhance targeted gene correction. As discussed above, pDNA and viral vectors have been used to deliver the targeting nuclease and repair template to the target cell. pDNA can assemble into nucleosome-containing minichromosomes (Reeves et al., 1985) and undergo methylation that leads to intracellular silencing (Hong et al., 2001). Similarly, viral vectors, including rAAVs and IDLVs, are chromatinized following cell entry (Okada et al., 2006; Kantor et al., 2009). Therefore, strategies to counter epigenetic silencing will have to be developed, such as the use of insulator sequences (reviewed in Raab and Kamakaka, 2010).

Future perspectives

Targeted gene therapies are still in the early stages of development, but hold significant therapeutic promise. Progress will depend upon developing endonucleases for exquisitely specific cleavage in the context of the entire genome, enhancing the efficiency of HR, and minimizing the potential for DNA damage and translocation. Future advances in all these areas will depend on deeper understanding of the molecular mechanisms that underlie genomic stability and instability in human cells.