Heritable Targeted Gene Disruption in Zebrafish Using Designed Zinc Finger Nucleases

The zebrafish has proven to be an excellent model for vertebrate development and disease 1 due to its rapid development, transparent embryos, and its relatively facile forward genetics. Techniques for reverse genetic approaches in zebrafish, however, are limited to mRNA knockdown strategies using modified antisense oligomers (morpholinos) 2 and TILLING for point mutations in a locus of interest. Both strategies have shortcomings and are limited in scope: morpholinos are active for only the first few days of development, hence their transient effect fails to impact on juvenile or adult phenotypes, while TILLING is both time-consuming and less effective for intron-rich genes with small exons because the chances of obtaining null or hypomorphic alleles are reduced 3. Conventional “gene targeting,” a powerful technique for gene disruption in mouse embryonic stem cells 4, often requires positive-negative selection with cytotoxic drugs 5, which is inapplicable in the context of a vertebrate embryo.

The resistance of the genome in most metazoa to targeted sequence alteration 5 has recently been overcome through the use of zinc finger nucleases (ZFNs). Initially developed to cleave DNA in vitro6, ZFNs are a fusion between the cleavage domain of the type IIS restriction enzyme, FokI, and a DNA recognition domain containing 3 ormore C₂H₂ zinc finger motifs. The heterodimerization at a particular position in the DNA of two individual ZFNs in precise orientation and spacing leads to a double-strand break (DSB) in the DNA 7. DSB repair pathways operate in all eukarya, and include non-homologous end-joining (NHEJ, the direct ligation of the two DNA ends, frequently with loss or gain of small amounts of DNA sequence), as well as homology-directed repair 8. Importantly, Carroll and colleagues 9 exploited NHEJ-based repair of a ZFN-induced DSB to disrupt a gene in Drosophila and an intergenic region in C. elegans10, thus establishing the potential of this technology in reverse genetic applications, recently expanded to mammalian tissue culture cells 11. Here, using golden/slc24a5 (gol) and no tail/Brachyury (ntl) as test loci, we investigated the feasibility of using designed ZFNs for reverse-genetic applications in zebrafish.

The gol pigment locus encodes a transmembrane protein, Slc24a5 12, and was chosen for an initial test of ZFN-driven editing because the lack of embryonic pigmentation caused by homozygous viable gol mutations make scoring for ZFN activity by loss-of-function straightforward 13. The C₂H₂ zinc finger is a versatile DNA binding motif capable of recognizing a broad range of sequences 14, and efforts over the past 14 years have yielded a large number of zinc fingers aimed at recognizing investigator-specified DNA triplets (see Supplementary Information online). As a consequence, simple in silico analysis can illuminate a high number of potential zinc finger proteins (ZFPs) for any given target gene. We describe here an experimentally straightforward two-step approach to identify those particular ZFPs that will yield robust ZFNs for in vivo gene disruption, and its application to the zebrafish gol and ntl genes.

We analyzed in silico the cDNA sequence of the gol gene and assembled a panel of 4-finger ZFPs (see Materials and Methods, Supplementary Information, and Supplementary Fig. 1 online) directed against a range of distinct positions in the gol locus (Supplementary Table 1 online). To limit the number of ZFN pairs to be tested in vivo, the affinity of the ZFPs for their target sequence was measured using an ELISA assay (Supplementary Fig. 2 online). Next, to determine which of these ZFPs will form pairs most likely to be active in zebrafish gene disruption, we developed a budding yeast-based system to rapidly measure ZFN activity (Fig. 1A–B; Supplementary Fig. 3 online; a detailed description is provided in Supplementary Information online).

In building this system, we made use of the well-established fact that while NHEJ operates relatively inefficiently in budding yeast, a DSB induced between two short direct repeats spaced by a heterologous sequence of up to 25 kb leads to resection of the entire intervening DNA, and the restoration of chromosome integrity via the single-strand annealing of the two repeats 15. We disabled the open reading frame for a secreted form of alpha-galactosidase (encoded by the MEL1 gene) with the gol sequence and placed this construct into a specific position of the budding yeast genome. Experiments using a control construct, in which the DSB between the direct repeats was induced by the well-studied yeast endonuclease, HO, validated the dynamic range and low background of the system in reporting the efficiency of DSB induction by the nuclease (Fig. 1C–D). We then screened the panel of gol-targeting ZFNs for DNA cleavage activity in a budding yeast reporter strain carrying their target sequence (Fig. 1A). This analysis identified active ZFN pairs directed against exons 4, 8, and 9 of the gol gene (Fig. 2A) – we focused on the exon 4 and 9-targeting ZFNs in subsequent experiments (Fig. 2B, Supplementary Fig. 4 online).

Following screening in budding yeast, the ZFNs identified as active in that system were tested for native gene disruption activity at their intended target locus in zebrafish: the ZFN expression constructs were transcribed and capped in vitro, and the resulting mRNAs were injected into 1-cell zebrafish embryos heterozygous for the gol^b1 mutation 12. For the majority of our experiments we used ZFNs carrying obligate-heterodimer forms of the FokI endonuclease domain 16; these inhibit ZFN homodimerization and thus improve specificity by limiting cleavage at off-target sites recognized by ZFN homodimers 16. At 2 days post fertilization (dpf), gol^b1 heterozygotes normally have dark eye pigmentation (Fig. 2C, upper left); however, we detected clones of unpigmented cells in the eye, potentially representing loss of gol function on the non-gol^b1 chromosome, in up to 32% of ZFN-injected, but not in control, embryos (Table 1 and Fig. 2C). A clear positive correlation was observed between extent of mosaicism and amount of ZFN mRNA injected (Table 1). At the highest doses of ZFNs, over 30% of injected embryos exhibited pigmentation mosaicism in the eye (Table 1). Also at the highest ZFN doses, a small fraction of ZFN-injected embryos exhibited mild to moderate developmental defects, such as a bent body axis or mild head necrosis. However, even at the highest concentrations tested, the majority of embryos showing pigment mosaicism were indistinguishable from wild-type embryos (69–90% depending on the ZFN pair tested) (Table 1).

To investigate whether these phenotypes were the consequence of somatic loss-of-function mutations resulting from repair of ZFN-induced DSBs by NHEJ, we genotyped the gol locus in injected embryos. We used PCR to amplify a 312 bp region including the ZFN target site, with total genomic DNA from individual fish with gol clones serving as the input for each PCR reaction. Consistent with the phenotypes observed above, we found a broad range of loss-of-function alleles in the genomes of ZFN-injected animals, but not in controls (data not shown). These included deletions ranging in size from 7 to 65 bp and insertions of 4 to 12 bp (Fig. 2D). Taken together, these experiments demonstrated that a single injection of ZFN-encoding mRNA into an early zebrafish embryo can result in the efficient mutation of the locus targeted by the ZFNs to yield alleles that confer loss-of-function phenotypes.

To confirm the generality of the observations above, we next targeted the no tail/Brachyury (ntl) gene, a major regulator of early embryogenesis. The choice of target gene was in large part inspired by the fact that the study of early development is a major direction of biological investigation in zebrafish as a model system. The ntl gene encodes an essential T-box transcription factor required for proper mesoderm formation; zebrafish embryos homozygous for a null ntl mutation lack a notochord and a tail 17. To test whether ZFNs can induce somatic mutations in the ntl gene, we followed the same scheme described for the gol-targeting ZFNs (ZFN design, ELISA in vitro binding assay, and in vivo yeast-based activity assay) (Fig. 3A; Supplementary Table 2 and Fig. 5 online). For these experiments, we focused our design efforts on exon 2, the location of a well-studied null allele of the gene (Fig. 3B).

Using lead proteins (Supplementary Fig. 6 online) identified in the yeast-based assay (Fig. 3A) for our initial effort in ntl disruption, we injected mRNA encoding high-fidelity, obligate-heterodimer 16 ZFNs into 1-cell embryos from a cross between wild-type and ntl^b195 heterozygous fish. Since half the injected embryos carry the ntl^b195 null allele, we hypothesized that these embryos would reveal ZFN-driven mutations of the wild-type allele on the other chromosome. Strikingly, we observed that 16–27% of injected embryos displayed a ntl-like phenotype (Fig. 3D), either mimicking the null phenotype 17 (Fig. 3C) or a less severe phenotype typical of a hypomorphic allele, ntl^b487 (Supplementary Fig. 7 online). We noted that even high ZFN mRNA doses were well-tolerated by the embryos (Fig. 3D).

We genotyped individual ntl-like embryos at the ZFN target stretch using a rapid assay based on a mismatch-sensitive endonuclease 16 that allows one to measure the fraction of chromatids carrying mutations relative to the wild-type sequence. This analysis showed that the fraction of mutated chromatids ranged from 3 to 17% in the majority of embryos (Supplementary Fig. 8 online). To further analyze the types of mutations induced, we sequenced a 226 bp region surrounding the DSB site from ZFN-injected embryos, and observed a broad range of deletions and insertions precisely at the targeted locus (Fig. 3E). The mutant sequences shown in Fig. 3E appear to represent alleles generated by NHEJ-induced repair of the ZFN-cleaved chromatid, because chromatids bearing the ~1.5 Kb ntl^b195 insertion (see Fig. 3B) are too large to be amplified and analyzed.

Given our published data on the feasibility of generating single-step biallelic genome editing events using ZFNs in human cells 18, we analyzed whether wild-type embryos injected with high-fidelity ntl-targeting ZFNs displayed ntl mutant phenotypes. In these experiments, we found that juvenile and adult fish had posterior tail truncations (20% [3/15] and 21% [12/58] among stocks injected with 1 ng and 5 ng ZFN Pair 2, respectively, and 7.7% [4/52] among stocks injected with 5 ng ZFN Pair 3) (Fig. 4A). We sequenced the ntl locus surrounding the ZFN-cleavage site from a small posterior tissue sample from each of three tailless embryos and found that ntl mutant-bearing amplicons represent a substantial fraction of the total (Sample 1, 5/25 (20%) ntl-bearing chromatids, 2 different alleles; Sample 2, 3/30 (10%) ntl-bearing chromatids, 1 allele; Sample 3, 8/29 (28%) ntl-bearing chromatids, 4 different alleles) (Supplementary Fig. 9 online).

Interestingly, when ntl-targeting ZFNs carrying wild-type FokI cleavage domains were injected into wild-type embryos, a high frequency of embryos exhibiting a ntl phenotype were observed; sequencing a 226 bp region surrounding the DSB site revealed that each of 3 representative embryos carried between 64–81% disrupted ntl alleles (Table 2; Supplementary Fig. 10 online). Given that a single wild-type ntl allele is sufficient for a normal phenotype, these data suggest that ZFNs can induce biallelic disruption of a target gene locus after mRNA injection into the 1-cell embryo.

To demonstrate that ZFNs can effectively induce mutations in the germline, wild-type embryos injected with ntl-targeting high-fidelity 16, obligate-heterodimer ZFNs were raised to sexual maturity and screened. We found that ZFN-injected fish breed normally and give clutches of similar size as when we intercross our non-ZFN injected wild-type stocks. Eggs from ZFN-injected females were fertilized in vitro with sperm from a male heterozygous for the ntl^b195 allele. Of 7 females analyzed, 4 generated ntl progeny (Table 3; Fig. 4B) at frequencies ranging between 1–13% as gauged by this complementation cross (Table 3). To measure the frequency of gametes carrying gene disruptions, we directly genotyped the chromatid provided to the progeny (both wild-type and ntl) by four of the founder mothers, and found that the germline carried mutations at frequencies ranging from 5–32% (Table 3). Direct sequencing confirmed these estimates and revealed that three founders carried at least two new alleles, and one founder carried at least one (Fig. 4C). In addition to founders identified using complementation crosses and chromatid genotyping (Table 3), we have identified an additional 2 carriers (germline transmission rates of 33% and 53%) using complementation crosses and an additional 5 carriers (germline transmission rates of 6–25%) by analyzing haploids generated from founder females. Taken together, the data show that over 60% of ntl ZFN-injected founders (11/18) carry mutations at the ntl locus, with an average germline frequency of 20%.

To determine the level of specificity of ZFN function, an experimentally determined DNA binding consensus was derived for each ZFN (see Materials and Methods, Supplementary Fig. 11 online for experimental details). This information was then used to search the zebrafish genome for sites showing the highest similarity to their intended target site (Supplementary Fig. 11 online). To experimentally determine whether ZFNs cleave at these alternative “off-target” sites, we directly analyzed the top 5 potential off-target loci for each ntl-targeting ZFN pair. We used PCR to amplify the cognate chromosomal loci from phenotypically ntl progeny embryos (Table 3) carrying ZFN-induced mutations at the ntl gene, and analyzed them by direct sequencing, loss-of-RFLP, and mismatch endonuclease assays. All the chromatids we analyzed were wild-type (Supplementary Fig. 11 online). These results confirm that ZFNs can be used to specifically and efficiently create heritable mutant alleles at loci of interest.

Finally, to show that ZFN-induced alleles can be propagated in subsequent generations, we outcrossed two ntl ZFN-injected females (Females C and D, Table 3) and recovered heterozygous carrier individuals at frequencies consistent with the germline frequency of the mutation in each founder female (Table 3), indicating that heterozygosity for the new induced alleles was completely viable. As stocks reached sexual maturity, fish were intercrossed to siblings and backcrossed to founders. We recovered ntl mutant phenotypes in the expected ratios and observed no non-ntl phenotypes among F2 progeny. While our analysis has failed to reveal ZFN-induced mutations at loci other than ntl, in the hypothetical possibility that an undesired secondary mutation was occurring at a rate as high as 10%, it would rapidly segregate from the desired ntl mutation even after one generation.

Our data show ZFNs to be highly effective tools for inducing mutations at a specified locus in the zebrafish, the rapid production of embryos and adult animals with corresponding loss-of-function phenotypes (Tables 1 and 2) and adults carrying new alleles (Table 3). To serve this purpose, ZFNs can be injected as mRNA, thus allowing only transient nuclease expression, which we have shown to be sufficient to introduce a stable genetic mark on the targeted locus. The experiments in the present work describe the successful disruption of 3 distinct positions at 2 separate genetic loci and the use of ZFN injection into wild-type embryos to yield phenotypes expected from a loss of targeted gene function (Table 2). In addition, we show that ZFNs can be used to generate novel germline mutations with high specificity of action (Table 3; Supplementary Fig. 11 online).

The C₂H₂ zinc finger motif 19 has been engineered to recognize many distinct DNA sequences 14,20,21 (see also Supplementary Information online), and this has paved the way for the design of ZFNs used in efficient genome editing at a broad range of investigator-specified endogenous loci, including a mutation hotspot in a human monogenic disease locus 18. As our data show, the combination of this versatile platform with a rapid yeast-based assay for the identification of maximally active engineered ZFNs, open fundamentally novel avenues of experimentation in the zebrafish, and suggest that ZFN technology may be widely applied to many organisms for which reverse genetics strategies are currently unavailable.