There has been a lot of excitement about the potential applications of zinc finger nucleases (ZFNs), as these enzymes allow highly specific, targeted genome modification in live cells. ZFN-mediated gene correction is more specific than the viral integration methods currently used in gene therapy, and two ZFNs are currently in clinical trials for treating HIV and cancer. However, two new papers by Pattanayak et al.1 and Gabriel et al.2 raise questions about the purported specificity of these genome modification tools. They show that ZFNs, in addition to cleaving at their desired sites, can also have unexpected cleavage effects in vivo that cannot be predicted using conventional in silico analyses. These findings could have important consequences for the safe use and optimization of ZFNs in gene therapy. We asked four experts to comment on the implications of this study.
These two studies1,2 are important and timely, providing a sobering assessment of the reality of ZFN-mediated gene targeting. Although it is disappointing (but not surprising) to realize the inherent imperfections of this method, there are several ways we could avoid off-target effects and optimize targeting in human cells. For example, screening in vitro1 or in a model cell line2 is better than in silico predictions of all the possible ZFN cleavage sites on the basis of biophysical and biochemical principles, although not all the potential sites will be cut in a given cell type, especially if the ZFN concentration is optimized. Therefore, the most relevant cell types should be used for gene targeting, and the targeted cells should be analyzed before conducting important biological experiments or putting the targeted cells into patients.
It is unlikely that we will be able to completely avoid off-target events in the near future, even if the specificity of ZFNs or related technologies such as homing endonucleases or transcription activator–like effector nucleases (TALENs) is further improved. However, for most applications, we will be able to select correctly targeted cells without needing to ensure correct targeting in every cell. In fact, current gene targeting technologies using homologous recombination rely on the ability to select and expand rare cells that have been correctly gene targeted, because the homologous recombination rate in nontransformed human cells is still low (1 × 10–4 or less), even when aided by ZFNs. Therefore, it would be desirable to use cell types that can be substantially expanded in culture, such as stem cells. Human pluripotent stem cells make it feasible to select a correctly targeted clone3,4. With the improved efficiency of gene targeting achieved by ZFNs and other tools, and the improved capacity for whole-genome DNA sequencing of several selected clones, we should be confident that it will be possible to derive precisely edited human cells for basic research and gene therapy.
Professor of Medicine, Stem Cell Program in the Institute for Cell Engineering, The Johns Hopkins University School of Medicine, Baltimore, Maryland, USA.
ZFNs are reagents that allow precise gene targeting and offer an advantage over gene therapy vectors that are prone to semirandom genomic integration. Given that the majority of the genome is transcribed5 and thus may produce either protein-coding or regulatory transcripts, it is of the utmost importance that only the intended genomic sequence is targeted.
Two recent manuscripts analyzed the off-target effects of ZFNs and show that even these precise tools can cause unintended genomic modification1,2. The papers overlap in their analysis of a ZFN targeting the chemokine receptor gene CCR5 that has already entered clinical trials as an anti-HIV therapy. Using an in vitro DNA-binding assay, Pattanayak et al.1 identified 37 sites in the human genome cleaved by the CCR5 ZFN. Analysis in intact human cells, however, showed that only ten of the sites were actually modified in K562 cells. Using the same cell type and ZFN but a different approach, whereby an integrase-deficient lentiviral vector cassette can be 'trapped' in a ZFN-induced double-stranded DNA break, Gabriel et al.2 mapped the genomic sites at which the ZFN was active. They showed that 93.8% of the insertion events were at the CCR5 locus. Interestingly, in only one case did both studies identify the same off-target site, CCR2. Given that in vitro DNA binding does not fully predict in vivo binding and given the high homology between the two genes, the finding that the CCR2 locus was affected is not surprising.
Collectively, these studies offer guidance for future ZFN design and use and suggest that in silico analyses may be best suited toward reagent optimization rather than prediction of off-target effects. Because ZFNs can bind with some degree of mismatching, off-target effects can occur, which may be minimized by using lower ZFN concentrations and optimizing binding to the on-target site1,2. As this field of research progresses, factors that determine ZFN target specificity will become better defined. Finally, direct testing of ZFN reagents that have optimized architectures in a nonbiased manner in the cell type of interest will be crucial for future ZFN clinical applications.
Regents Professor and Andersen Chair in Transplantation Immunology, Department of Pediatrics, Division of Blood and Marrow Transplantation University of Minnesota, Minneapolis, Minnesota, USA.
Pattanayak et al.1 and Gabriel et al.2 have reported two different approaches to identifying off-target cleavage sites of ZFNs. These papers invite comparison, as they both examine, in the same cell line, a pair of ZFNs that cleave at the CCR5 locus, and one might predict that they would identify similar or overlapping groups of off-target sites. Unexpectedly, the papers do not seem to agree, but because they use different analyses and because off-target sites are sufficiently rare, this is perhaps not so surprising.
Gabriel et al.2 used an integrase-deficient lentiviral vector to tag CCR5 ZFN target sites in K562 cells and then mapped the sites using linear amplification-mediated polymerase chain reaction (LAM-PCR) and sequencing approaches. By contrast, Pattanayak et al.1 screened a library of 1 × 1011 DNA sequences to identify those that were cleaved by the CCR5 ZFNs in vitro. They then examined 34 potential cleavage sites found in the human genome by deep sequencing them from transfected K562 cells expressing the ZFNs. Differences in the sequences of the CCR5 ZFNs, and in their concentrations in the cell lines, may partially account for the differences in identified off-target sites1,2.
For clinical applicability, an unbiased analysis of cleaved sites following exposure to ZFNs in a way that most closely mimics the desired application is important. It is unclear what would be gained by in vitro analysis of a DNA library, when off-target cleavages that occur in endogenous DNA that is subject to dynamic epigenetic modification are more relevant to clinical studies. Therefore, the analysis by Gabriel et al.2, and a similar one recently reported6, seem to model potential clinical application more faithfully. The latter report, in which ZFNs were expressed in mice6, allows collection of in vivo safety data over many months, which can supplement the molecular analyses described in these two reports. An issue that remains to be addressed is when during clinical development off-target sites need to be thoroughly analyzed. Off-target sites that threaten cell viability will be identified in early experiments, but the clinical effects of rare events with subtle effects may be more difficult to identify.
Investigator, Howard Hughes Medical Institute, Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, USA.
Since the first report of using ZFNs for targeted genome modification in mammalian cells, they have been known to cause cellular cytotoxicity7. First-generation assays for cellular toxicity used cell viability and immunofluorescence of DNA double-strand breaks to measure the specificity of ZFNs. These effects arose from ZFNs cutting at specific off-target sites in the genome, but the identity of these sites was not known.
Importantly, Pattanayak et al.1 and Gabriel et al.2 have now developed different approaches to identify some of these sites. Interestingly, using the same ZFNs in the same cell line and excluding the obvious CCR2 site, the two groups identified a non-overlapping set of sites. Thus, the identification of the complete set of off-target sites, even for the well-studied pair of ZFNs used in these studies, is not complete. Both studies highlight that previous approaches using in silico prediction methods are of limited utility, and claims of ZFN specificity on the basis of such in silico studies should be reevaluated. Moreover, the strategy each group used to validate their off-target sites was focused on the creation of small insertions and deletions rather than the rarer, but potentially more problematic, gross chromosomal rearrangements. Thus, future studies will need to focus on both the identification of a more comprehensive set of off-target sites and the frequency with which ZFNs create gross chromosomal rearrangements.
Ultimately, the functional significance of off-target effects of ZFNs rather than the frequency of these events will need to be assessed, as the simple creation of double-strand breaks, which are caused by many widely used therapies, does not preclude the clinical use of these enzymes. Finally, Pattanayak et al.1 discuss that precise expression of ZFNs (the 'Goldilocks' effect) is essential to maximize on-target activity while minimizing off-target activity. Small molecules have already been used to regulate ZFN protein expression with this concept in mind, and this is just one approach for titrating ZFN expression so it's 'just right'8.
Associate Professor, Department of Pediatrics, Divisions of Cancer Biology, Hematology/Oncology and Human Gene Therapy, Stanford University, Stanford, California, USA.
K.H. is a co-inventor on a pending patent using zinc finger nucleases to do genome editing in hemophilia.
About this article
Molecular Therapy (2013)