Using ZFNs to create DSBs in specific zebrafish genes, two groups disrupted three ‘proof-of-principle’ loci that give well-characterized phenotypes when mutated: mutants in kdrl (kinase insert domain receptor like; also known as flk1, kdra, and vegr2 [16]) have specific vasculature defects [17], no tail (ntl) mutant embryos lack a notochord and tail [18, 19] and golden (gol)/slc24a5 mutants lack embryonic melanosomal pigmentation [20, 21]. A schematic of the ZFN recognition sites within the kdrl and ntl genes and the ZFNs engineered to bind them are shown in Figure 1. Although differences in ZFN constructs and methodology distinguish the two studies, the basic experimental design used by both groups was similar and straightforward: mRNAs encoding ZFNs were injected into one-cell zebrafish embryos that were subsequently screened for specific mutations at the intended target locus (Figure 2). To detect somatically induced mutations, ZFN-injected founders were screened 1–2 days post-injection, either phenotypically (after ZFN mRNA injection into heterozygous carriers) or molecularly (by direct sequencing of the ZFN-targeted interval, by screening for loss of a restriction site that lies within the ZFN spacer region, and/or by assaying for cleavage by a mismatch-sensitive endonuclease). The observed mutations were typically small insertions and deletions and characteristic of DSB repair by NHEJ [11, 12].

In order for ZFNs to be a viable reverse genetic strategy in zebrafish, it was essential to show that ZFN-induced mutations can be effectively transmitted through the germ line. Importantly, both studies show that ZFN mRNA-injected zebrafish embryos raised to sexual maturity transmit ZFN-induced alleles to their progeny at high frequency [11, 12]. Both groups report strikingly similar results: 30% (6/20) of kdrl-targeting ZFN mRNA-injected animals transmit new kdrl alleles to their progeny at frequencies ranging from ~8% to 50% [12], and over 60% (11/18) of ntl-targeting ZFN mRNA-injected animals transmit new ntl alleles at frequencies ranging from ~1% to 50% [11]. The fact that some ZFN-injected founders transmit ZFN-induced alleles at the same frequency as a heterozygous carrier demonstrates that mutagenic events can occur very early in development. The high percentage of founders carrying new alleles (30–60%) and the high average germ line transmission rate (20–30%) also demonstrate that it would be relatively easy to identify founders carrying mutations in genes for which there is no previously characterized mutant phenotype. Taken together, the results suggest that it should be sufficient to screen a handful of embryos from a handful of ZFN-injected founders to identify carriers of new disrupted alleles, if ZFN-induced mutagenesis at kdrl and ntl are an accurate indication of ZFN efficacy at other loci in the zebrafish genome. Because repair by NHEJ creates small deletions and insertions of variable length, it is also possible to create novel alleles of varying strength, nicely demonstrated the isolation of a hypomorphic kdrl allele [12]. Importantly, both groups showed that ZFN-injected fish are fertile and breed normally, suggesting that ZFNs do not create overt deleteri-ous effects.

Unwanted off-target mutations are of concern in any mutagenesis procedure, and reagents like ZFNs that cause DSBs are no exception. Thus, it was critical to assess whether undesired off-target effects might complicate mutation recovery or analysis. To reduce the possibility of unintended mutagenesis at alternative sites within the genome, both studies utilized ZFNs containing obligate heterodimer forms of the Fok1 endonuclease cleavage domain; these modified forms greatly diminish ZFN homodimeric cleavage activity and toxicity by reducing the number of possible off-target sites that are recognized and cleaved [22, 23]. Meng and colleagues [12] used three-finger ZFNs in their study; each ZFN recognizes a 9 bp sequence and together the ZFN pair has an 18 bp recognition sequence interrupted by a 6 bp spacer region (Figure 1A). They identified 41 potential heterodimeric and homodimeric off-target sites in the zebrafish genome, differing by 1–4 bp from the intended target site, and used Solexa sequencing technology to analyse mutagenic activity at these alternative sites. Consistent with the use of obligate heterodimer FokI proteins, they detected no cleavage at homodimeric sites, even in severely deformed embryos. Cleavage at heterodimeric sites was limited to only a few sites and was minimal (~1–5%), with on-target cleavage several 100-fold more likely to occur than off-target cleavage [12]. Doyon and colleagues [11] used four-finger ZFNs in their study; each ZFN pair has a 24 bp recognition sequence interrupted by a 6 bp spacer region (Figure 1B). They identified potential off-target binding sites by experimentally determining a DNA binding consensus for each ZFN, which was then used to search the zebrafish genome for putative alternative sites with highest similarity to the consensus. They sequenced the top five heterodimeric alternative sites from embryos carrying new ntl alleles and detected no off-target lesions [11]. With the caveat that both groups could only analyse for disruption at off-target sites predicted from available genome sequence, and thus would not detect mutations or chromosomal rearrangements elsewhere in the genome, the results suggest that if undesired secondary mutations were to occur, they would rapidly segregate from the desired mutation in subsequent generations.

Both of these highlighted zebrafish studies represent the collaborative effort of researchers who specialize in ZFN design and applications and researchers specializing in zebrafish developmental genetics. Thus, a relevant question for the general research community is how feasible it is for individual laboratories to generate ZFNs that reliably target a gene of interest. Protocols, software and web-based tools are available that describe modular-based approaches to ZFN design [24–26]; although a recent publication summarizing the efficacy of modular-based approaches cautions that modular design approaches have high failure rates unless the design includes fingers with well-validated target sequences [27]. Thus, another critical aspect of the recent zebrafish ZFN work is that both groups describe new assays or approaches that can reliably identify ZFNs that will be active in vivo. Doyon et al. [11] describe a simple yeast-based assay that capitalizes on the fact that yeast cells can utilize single-strand annealing very efficiently to repair DSBs with appropriately designed substrates. In the assay, yeast cells carrying a ‘broken’ MEL1 reporter gene (disrupted with the ZFN target site) driven by a constitutive promoter are transformed with ZFNs predicted to recognize the target site. If the ZFNs efficiently bind and cleave the intended target sequence, the resulting DSB is repaired using single-strand annealing between partial MEL1 sequences carried on the same reporter construct on either side of the ZFN target, restoring the MEL1 reporter open reading frame. Because the yeast MEL1 gene encodes a secreted form of α-galactosidase, the activity of which is easily assayed in culture, restoration of constitutive α-galactosidase reporter activity provides a direct readout of the efficiency of the ZFN pair to bind and cleave the target site in vivo. This yeast assay system may be especially useful for testing ZFNs constructed using modular design strategies. Using a different approach, Meng et al. [12] combined design and selection to build ZFNs with efficient in vivo binding and cleavage activity. Using a bacterial one-hybrid system, they first identified single zinc fingers that efficiently bind to each 3 bp subsite of the 9 bp target sequence. Then, they combined the three independent zinc-finger pool to generate a new three-finger ZFP library, from which candidate zinc-finger proteins were selected based upon how well their DNA binding specificity matched that of the intended target site. Bacterial selection systems are thus also very useful for identifying active ZFNs [12, 28].

The two papers highlighted here demonstrate convincingly that transient ZFN expression, supplied by injection of ZFN-encoding mRNAs into the one-cell embryo, can efficiently create targeted mutations in the zebrafish genome via NHEJ-mediated repair of the resulting DSBs. An exciting implication of the work is that mutations need not be restricted to protein-coding genes, but could be extended to include non-coding sequences such as miRNAs or gene regulatory sequence. The success of ZFN-mediated mutagenesis in zebrafish, which does not require either transgenesis or embryonic stem cells, also suggests that ZFN technology might also be used to generate targeted mutations in the genomes of other model organisms. Finally, the challenge now facing the zebrafish community is to establish conditions that promote ZFN-induced DSB repair by the other major repair pathway, homology-directed repair. Once homologous recombination techniques are established in zebrafish, as they have been in other model systems and in cultured cells [2, 7, 29–31], researchers will be able to precisely control gene modification, allowing one to create of zebrafish models of human disease, engineer conditional alleles, and to knock-in fluorescent protein reporters into endogenous loci, to mention only a few of the myriad of exciting possibilities.