Current perspectives in zebrafish reverse genetics: Moving forward



Use of the zebrafish as a model of vertebrate development and disease has expanded dramatically over the past decade. While many articles have discussed the strengths of zebrafish forward genetics (the phenotype-driven approach), there has been less emphasis on equally important and frequently used reverse genetics (the candidate gene-driven approach). Here we review both current and prospective reverse genetic techniques that are applicable to the zebrafish model. We include discussion of pharmacological approaches, popular gain-of-function and knockdown approaches, and gene targeting strategies. We consider the need for temporal and spatial control over gain/loss of gene function, and discuss available and developing techniques to achieve this end. Our goal is both to reveal the current technical advantages of the zebrafish and to highlight those areas where work is still required to allow this system to be exploited to full advantage. Developmental Dynamics 237:861–882, 2008. © 2008 Wiley-Liss, Inc.


The zebrafish, unlike most vertebrate organisms, lends itself remarkably well to forward genetic approaches. The ease of keeping many zebrafish in a small space, the large brood size, the diploid genome, and the reasonably short life cycle, render large-scale genetic screens practical and economical. In 1996, an entire issue of the journal Development was dedicated to the description of two large-scale forward genetic screens in the zebrafish. This work from the labs of Nusslein-Volhard in Tubingen, and Driever and Fishman in Boston, was greeted with much fanfare, and to a large degree was responsible for the launch of zebrafish as a respectable vertebrate model organism that could take a place alongside the “big three” of Xenopus, mouse, and chick. Over the last decade, many articles have discussed the advantages of zebrafish for forward genetics (see recent article by Lieschke and Currie,2007). Here we discuss how the zebrafish model is also studied effectively using the reciprocal, and often complementary, approach of reverse genetics.

The power of forward genetics is its unbiased nature. Forward genetics is driven by phenotypes: mutagens are employed to produce random changes in the DNA and the phenotypic consequences are assayed. By contrast, genes drive reverse genetics: take a gene of interest, find a way to introduce an altered version or to disrupt the endogenous gene, and assay the phenotypic consequence. This approach has become the stalwart of developmental biology studies in systems where forward genetics is unfeasible. In particular, reverse genetic techniques have complemented the powerful embryological and biochemical approaches developed in Xenopus laevis to provide remarkable insights into vertebrate developmental processes. Indeed, it could be argued that the lack of forward genetics has driven Xenopus researchers to be uncommonly creative in developing novel approaches to study gene function. The rapid establishment and continued growth of reverse genetics in zebrafish has undoubtedly benefited from the expertise of Xenopus researchers, particularly those who embraced the optically clear zebrafish embryo as a complement to their tadpoles.

The main criticism that can be leveled at the reverse genetic approach is its reliance upon available knowledge of candidate genes. Unlike the unbiased approach of randomly mutagenizing the entire genome, reverse genetics typically focuses on a single gene or small group of genes. However, previous limitations in the availability of candidate genes are fast disappearing in this genomic age; we can now know essentially every gene, and in the case of zebrafish a full genome sequence is in sight. Rather than sequence information being limiting, it is more likely to be our own lack of imagination that prevents us from exploring the importance of particular sequences until forward genetics points the way. On the other hand, educated predictions can allow the reverse geneticist to effectively investigate the redundant gene functions that forward genetics may never be able to reveal.

Here we review the many reverse genetic approaches that are in common use in the zebrafish. We have attempted to point out both the strengths and the limitations of each approach. We have also considered methodologies that are currently in their infancy, as well as some that are still lacking from the zebrafish lexicon, with the expectation that these missing techniques will be established in the years to come.


The use of pharmacological reagents to study biology is not a true reverse genetic approach, because it does not rely on specific knowledge of an organism's individual genes. Nevertheless, reagents that were once understood only at the level of their phenotypic consequence can now be utilized with a clear understanding of the molecular pathways they disrupt or interact with. For example, LiCl has long been known to cause dorsalization and radialization of embryos (Kao et al.,1986), but we now understand that this is a consequence of its interaction with the Wnt signaling pathway (Klein and Melton,1996). Similarly, ingestion of cyclopamine, a corn lily–derived alkaloid, has been known for 50 years to cause a cyclopic phenotype in the offspring of grazing sheep, but we now know this is a consequence of cyclopamine's ability to block Sonic Hedgehog signaling (Cooper et al.,1998). Retinoic acid (RA) was initially described as a teratogenic agent (e.g., Kochhar,1973), but we now know many details of the natural synthesis, metabolism, and signaling pathway of endogenous RA. As a variety of reagents are available to manipulate the RA signaling pathway, this set of reagents provides a useful example of the power of the pharmacological approach and is further considered below.

The RA signaling pathway is relatively simple, as RA functions as a nuclear steroid hormone (Fig. 1A; reviewed by Mark et al.,2006). In brief, RA is a ligand for the RA Receptors (RARs), which dimerize with Retinoid Receptors (RXRs) to form heterodimers that are constitutively bound to well-defined RA response elements (RAREs) in the regulatory sequences of target RA-responding genes. In the absence of RA ligand, the receptors recruit co-repressors to actively silence target gene expression. When ligand becomes available, the binding of RA to its receptor leads to the replacement of co-repressors by co-activators and the induction of target genes. Signaling is modulated by the degradation of RA into hydroxylated polar derivatives catalyzed by Cyp26 class cytochrome P45 enzymes. Experimentally, the RA pathway can be activated by providing exogenous RA or effectively disrupted using pharmacological reagents that act at the level of RA synthesis, reception, or degradation (Fig. 1C,E,F). As zebrafish embryos are small, externally fertilized, and develop in a simple aqueous solution, treatment with any of these low molecular weight pharmacological reagents is very simple, requiring only that the molecules be dissolved into embryo medium. The lipophilic nature of RA and its antagonists allows rapid crossing of the plasma membrane, but also causes low solubility in aqueous solutions. For this reason, RA is normally dissolved at high concentrations in a carrier solution of DMSO and ethanol, then this solution is added at much lower concentration to the aqueous embryo medium. As both ethanol and DMSO themselves have phenotypic consequences at high concentration, an appropriate control for RA treatment is an equivalent concentration of carrier alone.

Figure 1.

Pharmacological and antisense-mediated inhibition of the Retinoic Acid (RA) pathway in zebrafish. A: RA biosynthetic and degradation pathways during cell–cell signaling. Retinaldehyde dehydrogenase (Raldh2 family, in zebrafish Aldh1a2) catalyzes the last step in the biosynthesis of RA, whereas Cyp26 class cytochrome P45 enzymes degrade it into more polar, inactive compounds. Changes in gene expression result from the binding of the RA Receptor (RAR) to RA. Various known pharmacological inhibitors of the RA pathway are highlighted in gray. B–J: Lateral view, anterior to the left, of 30 hr post-fertilization (hfp) zebrafish embryos with impaired RA signaling. B: Wild type embryo. C: Wild type embryo treated with RA starting at 8 hpf. D: nlsi26 embryo carrying a mutation in the aldh1a2 gene. E: Wild type embryo treated with the RA Receptor inhibitor BMS493, starting at 8 hpf. F: Wild type embryo treated with the Aldh1a2 inhibitor DEAB, starting at 8 hpf. G: Wild type embryo injected at the 1-cell stage with a morpholino against aldh1a2. H: gyrrw716 embryo carrying a mutation in the cyp26a gene. I: Wild type embryo injected at the 1-cell stage with a morpholino against cyp26a. J: Wild type embryo injected at the 1-cell stage with three morpholinos against cyp26a, cyp26b, and cyp26c. Asterisk, head defects; arrow, heart edema; arrowhead, tail bending; bracket, tail shortening. Scale bar = 100 μm.

The biologically active form of retinoic acid is all-trans RA, a derivative of Vitamin A. Treatment of embryos with physiological concentrations of RA (generally considered to be in the order of 10−7M) produces a well-characterized suite of phenotypic consequences (Durston et al.,1989; Conlon,1995; Fig. 1C). The rate-limiting step in RA synthesis is catalyzed by retinaldehyde dehydrogenase (RALDH) enzymes. During early zebrafish embryogensis, RALDH activity is supplied by the aldh1a2 gene. Two different mutations of zebrafish aldh1a2 have been described, neckless (nls; Fig. 1D) and no-fin (nof; Begemann et al.,2001; Grandel et al.,2002). These mutants display a range of phenotypes including disruptions of the posterior hindbrain, fins, pancreas, and cardiac edema (Stafford and Prince,2002; Fig. 1D). Several pharmacological reagents are currently available that disrupt RA signaling at different levels in the pathway. Reagents that can disrupt RAR function include BMS493 and AGN193109, both of which act as inverse agonists of all three RA-receptor (RAR) subforms (Klein et al.,1996; Wendling et al.,2000; Dupe and Lumsden,2001; Stafford and Prince,2002; Linville et al.,2004; Fig. 1E). These molecules bind tightly to RARs, blocking RA ligand from binding, and stabilize co-repressor interactions to maintain target genes in their repressed state. By preventing RA from functioning, each of these reagents produces similar phenotypes to the nls and nof mutants (Stafford and Prince,2002; Linville et al.,2004; Fig. 1D). Unfortunately, a significant problem with these and similar reagents is that supplies may be limited as the pharmaceutical companies that originally designed the molecules do not provide them on a commercial basis. The more widely available reagents citral and disulfiram are, respectively, a competitive inhibitor of the aldehyde dehydrogenase-mediated oxidation of retinol and retinal to RA (Connor and Smit,1987; Chen et al.,1995; Kikonyogo et al.,1999), and an inhibitor of the activity of aldehyde dehydrogenase class 1 enzymes (Vallari and Pietruszko,1982; Veverka et al.,1997); these two reagents suffer from the disadvantage that they are likely to display a comparatively wide substrate specificity. By contrast, another easily obtained reagent, DEAB, is a competitive, reversible inhibitor antagonist that acts as a specific substrate of retinaldehyde dehydrogenases (Russo et al.,1988,2002). Interestingly, DEAB treatment produces phenotypes that are more severe than the null aldh1a2 mutants (Begemann et al.,2004; Fig. 1F), possibly reflecting the presence of maternally deposited RALDH activity, or a yet to be identified zebrafish aldh gene. This finding illustrates one of the useful features of pharmacological reagents; they can often be used to block signals that rely on multiple gene products, thus revealing the redundant functions that forward genetics cannot show.

In genetic terms, null lethal mutations reveal the earliest requirement for a given gene: only conditional mutations can begin to unveil later requirements of that same gene. By contrast, one of the other great advantages of pharmacological reagents is that they can be applied at very specific times during embryonic development, thus revealing stage-specific effects. For example, our group has made use of timed treatments of zebrafish embryos with the BMS493 inhibitor to determine that the RA-sensitive phase of pancreatic beta-cell development is completed by 13 hr post-fertilization (Stafford and Prince,2002).

Many additional pharmacological reagents are widely used in the zebrafish system. These include SU5402, an inhibitor of FGF receptor function (Poss et al.,2000), cyclopamine (inhibitor of the Sonic hedgehog pathway; Cooper et al.,1998), the gamma-secretase inhibitor DAPT (which blocks the Notch pathway; Geling et al.,2002), and SU5614 and SU1498 (which inhibit VEGF/Flk-1 tyrosine kinase signaling; Liang et al.,2001). In general, it should be noted that use of these pharmacological reagents is only as reliable as the biochemical characterization of their functions. As there is a significant likelihood that pharmacological reagents may exert pleiotropic effects, it is critical to be confident in the specificity of any given reagent if accurate conclusions are to be drawn regarding the underlying biology of its effect. For example, although SU5402 has high FGF receptor binding affinity, it can also weakly bind other tyrosine kinases such as the VEGF receptor and block their function (Mohammadi et al.,1997). For many such reagents there is a concentration-dependent specificity, with high concentrations interacting with a subset of the substrates not recognized at lower concentrations. Thus, the effective concentration in embryos also needs to be carefully controlled. In addition, the stability of the reagent may be difficult to establish, and some pharmacological reagents such as SU5402 have notoriously transient effects.

Other ways to exploit pharmacological reagents include high-throughput “small-molecule” screens. For example, screens can be performed by identifying phenotypes of interest in embryos exposed to panels of small molecules. The Zon lab (Murphey et al.,2006) screened a library of more than 16,000 compounds to discover 14 new compounds that influence the cell cycle; in this type of experiment the zebrafish is essentially being used as a drug discovery tool to help identify small molecules of potential interest. In a related screen, prostaglandin E2 was identified as a regulator of haematopoietic stem cell homeostasis (North et al.,2007), revealing how a reciprocal approach allows use of previously characterized small molecules to identify pathways important in a particular biological process. Use of these approaches will likely expand in the future as small-molecule libraries increase in size, and molecule characterization improves. Zebrafish are also being developed as “sentinels” to monitor aquatic pollution, for example by generating transgenic animals able to respond to contaminating concentrations of toxins or pharmacological agents present in water (reviewed by Carvan et al.,2000).


In zebrafish, as in some other model organisms, one of the best-established approaches to test gene function is to over-express the gene of interest by microinjection of nucleic acids (Cho et al.,1991; Kelly et al.,1995a,b; Toyama et al.,1995; Nikaido et al.,1997; Koos and Ho,1998). In this approach, a normal or altered version of the gene product of interest is expressed at a time and/or location where it would not normally be found. The resultant phenotype can provide information about the normal functions of the gene, although it has to be remembered that a straightforward gain-of-function reveals functional capacity, i.e., what the gene can do, rather than directly revealing the normal function of the gene. In general, up to three variants of the gene of interest can be tested independently for function: wild type, constitutively active, and dominant-negative forms (e.g., Koos and Ho,1998). Of these, the last variant is designed to interfere and block the activity of the endogenous gene product in a competitive manner, providing in an ideal situation, loss-of-function information about the gene of interest.

In all cases, constructs are microinjected into zebrafish embryos at the 1- to 4-cell stage of development, either as plasmid DNA or more frequently in the form of synthetic, 5′-capped mRNA. When injected in a plasmid vector, constructs start to be expressed after mid-blastula transition, which in zebrafish occurs at about 3.5 hr post-fertilization, and the expression is invariably mosaic (i.e., some but not all of the cells in the embryo will express the construct). By contrast, mRNA injection leads to a far more homogenous distribution of the protein, with translation initiating shortly after the time of injection. As all cells are cytoplasmically connected at early stages, microinjections at or before the 4-cell stage lead to a fairly even distribution of the injected mRNA throughout all embryonic cells. However, as in practice the mRNA often becomes asymmetrically localized, it is common to trace the distribution of the gene product within the embryo. This can be achieved by co-injecting an easily detectable tracer such as GFP or LacZ (e.g., Bruce et al.,2003). A more direct method is to tag the molecule of interest, for example with a series of Myc-epitopes (e.g., McClintock et al.,2001), to allow direct antibody detection of the ectopic protein. One potential caveat of this approach is that epitope tags may interfere with normal protein function, which has led some groups to look for alternative methods to follow over-expression of gene product in the embryo. One such method makes use of the viral T2A self-cleaving peptide, the tertiary structure of which interferes with peptide bond formation between conserved glycine and proline residues, thus allowing the stoichiometric translation of two unfused protein products from a single messenger RNA (Donnelly et al.,2001; de Felipe et al.,2006). This method has recently been successfully employed in zebrafish to produce two independent fluorescent proteins with different sub-cellular localization domains from a single transcript (Provost et al.,2007).

Although the gain-of-function method has proven to be invaluable in dissecting early developmental events (e.g., axis determination, organizer formation, neural induction, etc.), and the roles that different pathways play in these processes (e.g., Wnt, Nodal, BMP, etc.), it has limited applications when studying processes occurring after the initial 24 hr of development (Kelly et al.,1995a,b; Toyama et al.,1995; Nikaido et al.,1997). This is due primarily to the gradual degradation of injected mRNAs, coupled with the halving of mRNA concentration per cell that accompanies each cell division. The resultant rapid reduction in gene product concentration limits the use of mRNA injection approaches to the analysis of early gene functions. This is unfortunate, as numerous genes are used again and again at different developmental stages. To overcome these problems, zebrafish transgenic technologies have been developed that allow the induction of a gene in a controlled spatial or temporal manner, by using inducible or tissue-specific promoters to drive gene expression (see below for a more detailed discussion of transgenic methodology). While in many cases these approaches have proven extremely useful, they do necessitate a significant time investment for the generation of stable transgenic lines.

Other approaches to overcome the limitations associated with gene over-expression have focused on non-transgene methods to provide temporal control of gene product availability. These approaches can, in general, be categorized by whether they operate by preventing mRNA from being translated or by interfering with proper activity of the protein being tested. A conceptually appealing, but technically challenging, approach is to block translation until the stage of interest by the use of photo-removable (caging) groups, which can be attached to the mRNA to render it inactive and then removed in vivo by photo-illumination with light of a specific wavelength. Ando and colleagues (Ando et al.,2001,2004) reported a method in which full-length, in vitro synthesized mRNA was allowed to react with the caging agent 6-bromo-4-diazomethyl-7-hydroxycoumarin (Bhc-diazo) and, after purification, injected into 1-cell stage embryos. Embryos were then raised in the dark until the appropriate developmental stage was reached, and photo-mediated uncaging of mRNA achieved by illuminating whole or part of the embryo with UV light, allowing the mRNA to become accessible to the translation machinery (Ando et al.,2004). Although this technique was successfully used to study several developmentally important genes, such as engrailed2a, lhx2, and six3 (Ando et al.,2001,2005), it has yet to gain broad popularity among the zebrafish community. This is perhaps because the Bhc-diazo caging agent is not commercially available and its synthesis is complex and susceptible to by-product contamination. Unfortunately, even when caging reagent is available, it has not worked well in all hands. However, the chemical synthesis of Bhc-diazo has been licensed to a commercial vendor (Waco Industries, Japan), which should enable wider testing of the efficacy of this approach in the future.

The second level at which one can gain temporal control over the production of a functional protein during over-expression experiments is by regulating protein activity itself. One of the first strategies used at this level of regulation made use of the characteristic hormone-mediated translocation of the Estrogen and Glucocorticoid nuclear hormone receptors from the cytoplasm to the nucleus (Kolm and Sive,1995). This translocation is dependent on the hormone-binding domain of these receptors, which can be fused to any protein to confer its hormone-dependent cytoplasmic or nuclear sub-localization properties. This technique is particularly well suited to the study of transcription factors, as they will be retained in the cytoplasm until they encounter the hormone. After addition of the hormone to the media, the fusion protein translocates to the nucleus and activates target gene transcription. While this approach has been frequently used in the Xenopus field, for unknown reasons it has only rarely been used in the zebrafish (Kolm and Sive,1995).

More recently, two new exciting techniques have emerged in the field of molecular biology that may provide a handle on the temporal regulation of protein availability in the embryo. These techniques make use of temperature-sensitive inteins and small molecule/protein tags to control protein stability. Inteins, or “protein introns,” are small proteins embedded within larger ones that have the ability to autonomously self-excise post-translationally while regenerating the intact host protein (Paulus,2000). Low-fidelity PCR combined with a yeast-based screen was used to identify temperature-sensitive alleles of the S. cerevisiae Sce VMA intein, which were unable to splice out from the host protein at the non-permissive temperature of 30°C, but restored protein activity at the permissive temperature of 18°C, irrespective of the type of protein or cellular background (Zeidler et al.,2004). Since zebrafish tolerates this temperature range well, it may be feasible to inject mRNA coding for a chimeric protein carrying one of these inteins and then induce excision at any time during development by transferring the embryos for a short period of time from their normal growing temperature of 28°C to the permissive 18°C temperature.

The second approach uses a binary system consisting of a protein tag capable of directing any fused protein for degradation by the proteasome and a small and very specific molecule that serves as an antidote to prevent this degradation (Lampson and Kapoor,2006). Since the degradation rate by the proteasome exceeds the rate of synthesis and folding of nascent proteins, tagged proteins are unable to reach a mature, functional state. Currently, the most commonly used degradation tags make use of polypeptides involved in the binding of rapamycin and its analogs: the 89–amino acid FRB subdomain of FRAP/mTOR (Target Of Rapamycin) and its binding partner, the 107-residue immunophilin FKBP12. In a screen designed to identify mutant FRB domains with increased binding specificity for the more inert rapamycin analog MaRAP, one particular variant (FRB*) was selected because it seemed to render proteins unstable; FRB*-fused proteins were degraded when expressed in cultured cells unless rapamycin or MaRAP was added to the media (Stankunas et al.,2003). The feasibility of this system for in vivo work was recently demonstrated in an elegant study looking at the role of GSK3 during palate closure in the mouse embryo (Liu et al.,2007). Because the stabilization of FRB* was shown to be dependent on FKBP12 (Stankunas et al.,2003), additional efforts were made to identify alternative ligand-dependent destabilizing domains based on FKBP12 itself. Screening of a PCR-generated mutant FKBP12 library fused to YFP led to the identification of a destabilizing variant (L106P), which can be stabilized by the presence of a small molecule called Shield (Banaszynski et al.,2006). Temporal control of the activity of several classes of kinases, GTPases, and transmembrane glycoproteins fused to L106P was then used to demonstrate its general ability to affect protein stability in cell culture (Banaszynski et al.,2006). Although this technique has yet to be tested in vivo, both in mouse and in zebrafish, it opens the possibility of studying events in late embryogenesis that would be impossible to address using standard over-expression approaches.


While attempts to disrupt zebrafish gene function using RNA interference are yet to show real promise (but see below), antisense based approaches have been singularly successful in this organism. At the 2000 International Zebrafish Development and Genetics meeting, a talk by Robert Ho clarified the problem of non-specific effects of dsRNA injection (Oates et al.,2000; Zhao et al.,2001), but the very next presentation rendered the community optimistic once again about the possibilities of “knockdowns,” as Steve Ekker formally unveiled “morpholinos” (Ekker,2000; Nasevicius and Ekker,2000). Morpholinos are stabilized antisense oligonucleotides, which due to their altered backbone are not a nuclease substrate and are, therefore, very resistant to breakdown in vivo. In addition, due to their small size and/or altered chemistry, they do not trigger the immune response of the organism. Morpholinos have been designed to block translation by binding to mRNA transcripts close to or upstream of the translational start. Unlike antisense RNAs, the morpholino does not render the target mRNA a substrate for endogenous nucleases, but instead acts by steric hindrance to block ribosome entry and hence prevent protein production. This approach does not work if the morpholino is complementary to sequences within the open reading frame, presumably because the ribosome can displace the morpholino once translation is fully underway. Translation blocking morpholinos leave the target gene's transcript intact, and the transcript's localization can continue to be monitored by in situ hybridization, often as a useful component of phenotypic analysis. Morpholinos have also been designed to target splice acceptors or donors (e.g., Draper et al.,2001), thus blocking normal splicing and leading either to altered transcripts that lack an exon, or to loss of transcripts via nonsense-mediated decay. In fact, splice-blocking morpholinos were amongst the first to be designed, as a potential therapy for forms of thalassemia in which beta-globin transcripts are mis-spliced (Suwanmanee et al.,2002). More recently, morpholinos have also been shown to be effective in blocking micro RNA (miRNA) maturation, thus preventing the degradation of target mRNA by RNA-induced silencing complexes (RISCs, see below) (Kloosterman et al.,2007; Woltering and Durston,2008). The success of morpholino technology is reflected in a new searchable on-line database that allows investigators to look for morpholinos by name, target gene, anatomical structures affected, phenotype produced, and dose effect (Knowlton et al.,2008).

While problems with delivery of morpholinos in vivo have unfortunately rendered them of limited use in patients, morpholino technology was successfully used for some years in tissue culture, with the nucleotides being “scrape-loaded” into cells in a procedure that functions by damaging the cell membrane. Janet Heasman, a pioneer of antisense RNA knockdown techniques in Xenopus, recognized that morpholino reagents could potentially function in embryos. Her group successfully demonstrated the potential of morpholinos to study embryo development in a study of Xenopus β-catenin function (Heasman et al.,2000). At the same time, Steve Ekker, a fellow University of Minnesota faculty member, was exploring the use of morpholinos in the zebrafish, and the first reports of morpholino use in zebrafish followed shortly thereafter. Early confirmations that morpholino knockdown or “morphant” phenotypes effectively phenocopy mutant phenotypes (Nasevicius and Ekker,2000) quickly led to the use of the technology to confirm new mutant gene identification (e.g., the mutants off-road, neckless, and giraffe disrupt frizzled3a, aldh1a2, and cyp26a, respectively; Begemann et al.,2001; Emoto et al.,2005; Wada et al.,2006; Fig. 1G,I), to explore the functions of genes where no mutant exists (e.g., tbx6; Goering et al.,2003), and to investigate redundant gene functions by co-injection of morpholinos targeted against more than one gene (e.g., duplicate genes hoxb1a and hoxb1b, or the triplicates cyp26-a, -b, and -c; McClintock et al.,2002; Hernandez et al.,2007; Fig. 1J). Just like other oligonucleotide-based reagents, morpholinos are introduced into embryos by microinjection at early developmental stages. As morpholinos require some buffering, 1× Danieau buffer is the recommended carrier. Useful information on how to store the reagents and prepare them for injection can be found on the web site of their manufacturer, GeneTools LLC (Philomath, OR). We have found that the recommendation to warm the reagents to 65°C before use often improves efficacy.

Morpholino design follows similar rules to general primer design, as the bases follow the usual Watson-Crick pairing rules with GC-rich sequences binding more tightly than AT-rich sequences, and the usual requirements to avoid palindromes, primer-dimer formation, and high sequence similarity to other genes. GeneTools, LLC have significant experience in primer design and offer this service to researchers ordering their reagents. In view of the significant expense of morpholinos, it is wise to be careful not only with morpholino sequence design but also in the initial sequencing of the gene to be targeted. As a word of warning, we have discovered discrepancies with published sequences both from the genome and from published cDNAs. These may reflect sequencing errors but can also be a consequence of allelic differences between different strains of zebrafish. In particular, introns and UTRs are more likely to demonstrate allelic differences than are coding sequences. Before ordering a new morpholino it is our policy to isolate and sequence the region of interest of the target gene from our own strains of fish. Conveniently, morpholinos tend to have high specificity, a 4-bp change in the usual 25-bp length sequences will reduce target:morpholino binding by more than 80% (Nasevicius and Ekker,2000).

Despite general high specificity, morpholinos can sometimes show non-specific effects. For example, injection of high concentrations of morpholino can lead to defects such as neural death, small heads and eyes, somite and notochord defects, and eventually craniofacial dysmorphology (reviewed by Ekker and Larson,2001). To some degree, this may reflect the presence of residual contaminants in a morpholino preparation, but it has recently been demonstrated that such phenotypes can often be a consequence of sequence-specific “off-targeting” effects (Robu et al.,2007), which are nevertheless independent of the intended gene target. Interestingly, the off-targeting effect, which can occur with both 5′ and splice-blocking morpholinos, is mediated through a p53-dependent cell death pathway. While the details of the pathway remain somewhat indistinct, it has been clearly demonstrated that the cell death phenotype, but no other non-specific off-targeting phenotypes, can be blocked by co-injection of a p53 morpholino (Robu et al.,2007). As long as care is taken to confirm that specific cell death phenotypes are not missed as a consequence, this new p53 co-injection approach will be particularly helpful in studies of neural and cranial development.

While p53-mediated off-targeting is difficult to control for, GeneTools LLC will design 4-bp mis-match oligonucleotides to use as more general specificity controls. However, the expense of such control reagents has driven many researchers to make use of unrelated sequences for control injections. While such control injections, whether of a mis-match or an unrelated sequence, have become the norm, they essentially control only for non-specific morpholino injection effects, which unlike sequence-specific off-targeting seem to be very rare. Therefore, it is important to provide positive evidence that the phenotypes observed are due to the specific targeting of the gene of interest. In short, appropriate control experiments are needed to ensure: (1) that the gene of interest is effectively knocked-down, and (2) that phenotypic consequences are due to specific knockdown of the gene of interest rather than of related genes. A great advantage of using splice-blocking morpholinos is that efficacy of the morpholino can be monitored simply by PCR analysis of the target gene's transcript. By contrast, the ideal way to monitor efficacy of a translation-blocking morpholino is to assay protein levels of the target gene using a specific antibody. However, because antibodies are only rarely available, a variety of other methodologies have been developed to demonstrate effective knockdown. These include in vitro assays, where translation efficiency of the target gene in the presence of specific versus control morpholinos is assayed, and in vivo tests where translation efficiency of an exogenously supplied tagged version of the target gene is assayed in the presence of specific versus control morpholinos. The latter approach seems preferable as it is at least within the normal physiological environment, and use of GFP-tagged variants allows rapid assays. However, the criticism remains that knockdown of an exogenous transcript may not necessarily reflect knockdown of the endogenous transcript.

To address the second point, that phenotypic consequences are due to specific knock-down of the gene of interest, a commonly used and convincing approach is to demonstrate that two or more independent morpholino sequences produce equivalent knockdown phenotypes. Thus, rather than using mismatch control morpholinos, it seems that funds are better spent on the purchase of additional specific morpholinos to the gene of interest. In many cases, although two morpholinos will produce equivalent phenotypes, one will do so at a lower concentration implying greater efficacy. A related observation is that two unrelated morpholinos when co-injected will often work synergistically to produce the strongest possible phenotype, providing a second rationale to generate more than one morpholino to each gene of interest.

Arguably, the most convincing morpholino control of all is a phenotypic rescue, as this approach directly addresses the issue of specificity. For some genes, the knockdown phenotype can be effectively reversed by restoring the gene product of interest via co-injection of an mRNA not recognized by the morpholino. For example, the hoxb1a morphant phenotype can be rescued in this way (McClintock et al.,2002). Although such rescue experiments are compelling, it should be remembered that there are a large number of genes where the “rescue,” which essentially involves a gain-of-function experiment, causes a whole new set of phenotypes and thus the result is uninformative. This is particularly true for genes with highly localized expression patterns or in cases where outcome is sensitive to precise protein levels. In cases where mRNA expression at early stages causes a gastrulation phenotype, it may be possible to avoid this by activating expression of the gene of interest only at a later stage. For example, we were able to rescue the cdx4/kugelig mutant phenotype using a transgenic line where cdx4 was placed under control of a heat-shock promoter (Skromne et al.,2007), but establishment of a transgenic line is of course a time-consuming exercise. In other cases, mosaic mis-expression of a heat-shock-inducible construct may be informative, thus avoiding establishment of a transgenic line (O'Hara et al.,2005).

Despite the power and popularity of the morpholino knockdown approach, as with mRNA injection, a major limitation is the difficulty of introducing morpholinos at later stages of development, or in targeting them to specific regions of the organism. While morpholinos should be more stable than introduced mRNAs, they still suffer from a dilution effect at each cell division, and as a consequence are unlikely to block gene function efficiently much beyond the 48-hr post-fertilization stage; the precise stage will depend on transcript levels of the gene of interest. In chick, morpholinos labeled with a lissamine side-chain have been successfully electroporated into tissue to block gene function (Taneyhill et al.,2007). Recently, methodology to electroporate lissaminated morpholinos into single cells of the zebrafish has also been described (Cerda et al.,2006), although whether gene knockdown is efficiently achieved in these cells remains to be demonstrated. As this approach allows only one or a few cells to be targeted, it is most likely to prove useful for the study of genes that function cell-autonomously in fairly superficial cells, such as those of the retina or early neuroepithelium. Methodology has also been established to electroporate morpholinos into regenerating limb blastemas, and this approach has been successfully used to reveal roles for several different genes in the regeneration process (Thummel et al.,2006a,2007). In the long term, establishing more general methodology to achieve gene knockdown throughout a zebrafish tissue of interest will be extremely useful, but such methods may not rely on morpholinos.

While morpholinos do lend themselves uncommonly well to studies of early development, another potential limitation is in their capacity to knock down maternal gene function. If a gene product has already been translated before introduction of a translation-blocking morpholino, it will clearly not be possible to fully block function of that gene. However, in cases where maternal transcripts, but not protein, are present at the time of injection, gene function can be effectively disrupted; several maternal genes have had their functions disrupted by morpholinos, including sox21 and beta-catenin-2 (Argenton et al.,2004; Bellipanni et al.,2006). Further, a comparison of the phenotypic consequences resulting from translation- versus splice-blocking morpholinos can be informative regarding the different roles of maternal versus zygotic gene expression.

In principle, other modified oligonucleotides should also be useful for knock down of gene function in zebrafish. In particular, negatively charged Peptide Nucleic Acids (ncPNAs) have been reported to function with potency and specificity comparable to morpholinos (Urtishak et al.,2003; Wickstrom et al.,2004). Although Wickstrom and colleagues successfully used PNAs to block function of several different genes, these reagents have not achieved the same level of general use as morpholinos. This may soon change, however, with a recent report describing a method to inactivate or “cage” ncPNAs using a photolabile group to render them inactive until activated by light (Dmochowski and Tang,2007; Tang et al.,2007). In this novel approach, an antisense ncPNA to the targeted gene and a short sense 2′-OMe-RNA strand are linked via a 1-(5-(N-ma- leidomethyl)-2-nitrophenyl)ethanol N-hydroxysuccinimide ester photocleavable linker. The linker aids in the formation of a hybrid stem-loop structure and inactivation of the ncPNA. Upon cleavage of the photo-cleavable linker, the hybrid-stem loop structure dissociates releasing the ncPNA, which can then bind and prevent translation of its target mRNA. Using a similar strategy, another group has recently reported the successful caging of a ntl morpholino with a photo-labile group and its uncaging and activation with UV light (Shestopalov et al.,2007). In principle, this approach provides a switch to turn off gene expression at any point during development. Whether this technique will become a standard for the spatial and temporal control of gene activity will depend on how easily available these photo-cleavable reagents become.


One intriguing and poorly characterized method for disrupting gene function in zebrafish takes the novel approach of targeting mRNA for degradation rather than engineering a loss-of-function version of the gene as in more classical DNA “knock-out” methods. This method, which is based on the use of endogenous or engineered ribozymes (RNA enzymes) with ribonuclease activity, aims to eliminate gene function by creating non-functional mRNA incapable of being translated. In general, a short antisense sequence complementary to the 5′ coding region of the gene is used as a “guide” for homing in ribonucleases that will cleave the mRNA of interest, leaving it non-functional (Fig. 2). In ribozyme-mediated “knockdown,” the RNA sequence normally used by a ribozyme to identify its target substrate is replaced with the antisense guide for the gene of interest (Xie et al.,1997; Walker et al.,2001; Fig. 2A). In contrast, in the external guide sequence (EGS) technique, the antisense guide is designed to bind to the target mRNA such that together they form a complex of hairpin structures that resemble a tRNA molecule (Pei et al.,2007). This structure is then recognized by the endogenous ribonuclease (RNase) P, which then cleaves the mRNA in a manner analogous to the maturation of tRNA from its precursor molecule (Liu and Altman,1996; Rangarajan et al.,2004; Pei et al.,2007; Fig. 2B).

Figure 2.

mRNA inactivation by ribozymes and external guided sequences. A: Schematic representation of Ribozyme-mediated mRNA inactivation. A guide sequence complementary to the target mRNA (black structure) is incorporated at the appropriate location in a Ribozyme (gray circle) to guide the recognition and cleavage of the target mRNA (gray line). Cleavage of the mRNA (arrow) results in degradation of the molecule. B: Structures formed due to base pairing in naturally occurring tRNAs (left structure) and External Guided Sequences (EGS) paired to their target mRNA (gray; right structure). Both structures are recognized and processed by RNase P (arrows). While tRNA cleavage results in its maturation, cleavage of the target mRNA results in degradation of the molecule.

The success of these techniques depends, to a great extent, on the identification of good guide sequences. Two important parameters that need to be taken into consideration are the location of the sequence relative to the translation initiation site and the secondary structure of the mRNA molecule. In order for the ribonuclease to produce a non-translatable mRNA molecule, the cleavage has to occur after the Ribosomal entry site (i.e., Kozak sequence), but before or shortly after the translation initiation site (ATG). Equally important, the target site has to be located in a portion of the mRNA that lacks secondary structure, preferably within a loop. These structures can be identified by analyzing the mRNA sequence using programs such as RNAdraw (Matzura and Wennborg,1996). Once guide sequence candidates have been identified, they are cloned into the appropriate expression vectors to generate the ribozymes or the external guide sequences. Delivery of these constructs into the embryo is performed following standard injection protocols (Xie et al.,1997; Walker et al.,2001; Pei et al.,2007).

Although the feasibility of these methods has been demonstrated in zebrafish, two important issues having to do with selectivity and reproducibility remain to be addressed. So far, this technique has only been tested on a single gene, no tail (ntl; Xie et al.,1997; Walker et al.,2001; Pei et al.,2007). This gene was selected as a target for obvious reasons; it encodes a transcription factor (T-box) and its disruption causes a severe tail-truncation phenotype (Schulte-Merker et al.,1992,1994; Halpern et al.,1993). Whether other genes encoding secreted factors, signal transduction components, enzymes, and so on, can also be knocked-down with these techniques remains to be shown. Even for ntl, the ribozyme-mediated mRNA cleavage gave knockdown efficiencies that were disappointingly low and highly variable. One group reported that only 7% of injected embryos gave the truncated phenotype characteristic of no tail homozygous mutant embryos, 12% gave partial phenotypes, and the remaining 81% were phenotypically wild type (Xie et al.,1997). Similar success rates are reported by another group (Walker et al.,2001). Knockdown efficiency was improved when external guide sequences were used to cleavage the ntl mRNA, with about 24 to 35% embryos injected displaying the no tail phenotype (Pei et al.,2007). Assuming these methods prove broadly applicable and their efficiency can be increased, they could potentially serve as a cheap alternative for reducing or eliminating gene function in zebrafish.


RNA interference, or RNAi, has become one of the most powerful reverse genetic methods to transiently and specifically knock down gene function in eukaryotes. Not only has it been applied in single-gene studies, but has also provided new tools to dissect gene function at the genome-wide level. Besides being the thrust behind novel gene knockdown and knockout strategies in many classical animal models (Martin and Caplen,2007), RNAi has allowed the development of loss-of-function approaches in species such as rats and goats that are not yet amenable to homologous recombination techniques (Dann et al.,2006; Golding et al.,2006). While the development of in vitro and in vivo RNAi methodologies for the transient or stable manipulation of gene expression levels in mammalian and invertebrate systems has seen an explosion in the last couple of years (Kennerdell and Carthew,1998; Misquitta and Paterson,1999; Caplen et al.,2001; Elbashir et al.,2001), its broader use in amphibians and fish species has not been adapted as quickly as one might have anticipated. This is probably due, at least in zebrafish, to the coincidental discovery that some forms of RNAi have non-specific and deleterious effects in the development of the embryo (Oates et al.,2000; Zhao et al.,2001) and that morpholino-modified antisense oligonucleotides can easily and specifically be used to knock down gene function in this organism (Nasevicius and Ekker,2000).

The phenomenon of RNAi was initially discovered as the mechanism mediating translational repression of genes involved in the transition from the first to the second larval stages of life in C. elegans (Lee et al.,1993; Wightman et al.,1993). Since then, RNAi has been shown to play roles in diverse biological processes including transcriptional gene silencing and antiviral responses (Plasterk,2002; Zamore,2002). These silencing mechanisms are induced by 21–23-nucleotide- (nt) long double-stranded RNA (dsRNA). Two types of 21–23 nt dsRNAs can be found in any given eukaryotic cell, the microRNAs (miRNAs) and the small interfering RNAs (siRNAs). While these two types of molecules are indistinguishable at the biochemical and molecular level, they can be distinguished by their biogenesis, targets, and silencing mechanism (reviewed in Bartel,2004; Fig. 3). Starting with their biogenesis, miRNAs are generated as single-stranded RNA precursors that fold into complex stem-loop structures. Precursors are then cleaved by the nuclear RNase III endoribonuclease Drosha into ∼60–70-nt stem-loop dsRNA intermediates. Maturation is completed in the cytoplasm by Dicer, a second RNase III endoribonuclease, which cleaves these ∼60–70-nt intermediaries into the smaller 21–23-nt miRNAs. In contrast, siRNAs originate as long heteroduplex dsRNAs that Dicer cleaves into small 21–23-nt fragments. In Drosophila, two different dicer enzymes process the miRNAs and siRNAs and, based on their double-stranded intermediates, pair them with specific Argonaute (Ago) proteins: miRNAs are paired with Ago1 and Ago2, while siRNAs associate almost exclusively with Ago2 (Tomari et al.,2007). Argonautes are small RNA-binding proteins that lie at the core of the RNA-induced silencing complexes (RISCs) and are responsible for their silencing activity (Carmell et al.,2002). Even though Ago proteins are very similar, important differences do exist between them. For example, Ago2 is a robust RNA-directed RNA endoribonuclease, whereas Ago1 is not (Forstermann et al.,2007). Therefore, miRNA- or siRNA-loaded Ago2-RISC can mediate RNAi by targeting mRNA for degradation, while miRNA-Ago1 RISCs would only repress protein translation (Forstermann et al.,2007). Whether similar mechanisms operate in vertebrates remains to be seen, as in these organisms repression of mRNA translation or accelerated mRNA decay are more frequently observed than guided mRNA cleavage (Valencia-Sanchez et al.,2006). Lastly, miRNAs and siRNAs also differ in the target genes that they repress. While miRNAs target genes in regions of the genome outside their locus, siRNAs tend to target the locus that generates them, such as a viral or transposon sequence. Based on this, it has been proposed that miRNAs' primary function is in gene regulation, while siRNAs play a more prominent role as a defense mechanism against foreign DNA (Plasterk,2002; Bartel,2004).

Figure 3.

miRNA and siRNA biogenesis. A–G: miRNA biogenesis. A: miRNA gene transcription and folding into a Pri-miRNA hairpin structure. B: Cleavage by Drosha to generate a Pre-miRNA with a 2-base pair 3′ overhang. C: Nuclear export of Pre-miRNA to the cytoplasm. D: Loop cleavage by Dicer and generation of miRNA:miRNA* duplex with 2-base pair 3′ overhangs. E: Unwinding of the miRNA:miRNA* by a Helicase. F: Loading of miRNA into RNA-induced silencing complexes (RISC). G: Silencing of target mRNAs by miRNA-RISC complex. 1–4: siRNA biogenesis. 1: Cleavage of long dsRNA into short fragments by multiple rounds of Dicer activity. 2: Unwinding of siRNA duplexes by Helicase activity. 3: RISC loading of mature siRNA. 4: Target silencing by siRNA-RISC complex in the nucleus or cytoplasm (not shown). Entry points of synthetically generated dsRNA, siRNA or shRNA are indicated in gray.

In general, there are three types of RNA molecules that can be used to experimentally knock down gene function by RNAi: long dsRNAs (>100 nt), short dsRNAs (20–80 nt, similar to the siRNAs) and short hairpin RNAs (shRNAs, similar to the miRNAs). Upon delivery, these molecules will enter the miRNA/siRNA pathway at different levels to be processed into 21–23-nt dsRNAs (Fig. 3). Because of the small size of the active molecules, it is important to minimize off-target recognition events by determining regions unique to the gene to be knocked down. This can be done using a variety of computational approaches (Kim et al.,2006; Sethupathy et al.,2006; Watanabe et al.,2006). For large regions, sense and antisense RNA strands are synthesized and subsequently annealed to generate dsRNA. These long dsRNAs can then be directly delivered into the experimental system, or be digested with recombinant endoribonucleases (Dicer or RNase III) before delivery, to generate a pool of 21–23-nt dsRNAs. Chemical synthesis can also be used to generate short dsRNAs from annealed, sense, and antisense strands. An alternative to the short dsRNAs are the shRNAs, palindromic 70–100-nt-long single-stranded RNAs that fold into a stem-loop structure or hairpin and mimic the structure of immature miRNAs. These shRNAs can be generated in bacterial expression systems or be chemically synthesized. When designing short dsRNAs and shRNAs, additional parameters need to be considered to ensure that the RNA will be functionally active including assessing the complementarity of the 5′ end of the RNA to its target sequence (the “core” or “seed” element, the 2–8 nt at the 5′ end of the sense strand), its G/C content and the nucleotide distribution (Kim et al.,2006; Sethupathy et al.,2006; Watanabe et al.,2006).

In zebrafish, long and short dsRNAs have been tested for RNAi activity with conflicting results. In a variety of zebrafish adult- and embryonic-derived cell lines, long dsRNAs have been shown to be highly effective in inducing RNA interference and producing knockdown phenotypes, demonstrating that zebrafish cells possess a functional RNAi machinery and that cell culture systems are a feasible method for studying gene loss-of-function in zebrafish (Gruber et al.,2005). However, in intact embryos long dsRNAs have generated mixed and often contradictory results, with some groups claiming that dsRNA induces specific RNAi-mediated gene knockdown (Wargelius et al.,1999; Li et al.,2000; Hsieh and Liao,2002; Acosta et al.,2005), while others argue that they cause non-specific defects (Oates et al.,2000; Zhao et al.,2001; Gruber et al.,2005). Most of the genes targeted in these studies were selected because their effect on development could be compared to the phenotype of the mutant condition (e.g., zf-T/no-tail, tbx16/spadetail, pax2.1/no-istmus). These phenotypes were also compared to negative controls, dsRNAs generated against genes not present in zebrafish (e.g., egfp). Despite having positive and negative controls, the phenotypes claimed to be specific by some groups (Wargelius et al.,1999; Hsieh and Liao,2002), have been questioned as non-specific by others (Gruber et al.,2005). Non-specific phenotypes include general growth retardation, truncated tails, loss of brain and eye structures, and enlarged heart cavities (Oates et al.,2000; Zhao et al.,2001; Gruber et al.,2005). It is worth noting that these phenotypes are remarkably similar to those seen in embryos that lack maternal and zygotic Dicer function (Giraldez et al.,2005), which may indicate a competition between the dsRNA and the endogenous miRNAs for components of the RNA silencing machinery. Another hypothesis that has been put forward is that the non-specific effects caused by the dsRNAs are due to the activation of the interferon-mediated antiviral response (Oates et al.,2000; Gruber et al.,2005), as activation of this pathway has been reported after injection of dsRNA into fish embryos (Kumar and Carmichael,1998; Collet and Secombes,2002; Jensen et al.,2002; Collet et al.,2004). Regardless of the mechanism responsible for the non-specific effects, it is clear that long dsRNAs are not a reliable tool for studying gene function in zebrafish embryos.

Somewhat more promising results have been obtained when using short dsRNAs. One particularly well-controlled experiment addressing the substrate requirements of the let-7 miRNA in zebrafish showed that short (72 nt) dsRNAs could cause specific developmental defects, and that these defects could be rescued by co-injecting a non-hydrolysable oligonucleotide that paired with the let-7 dsRNA and prevented its activity (Kloosterman et al.,2004). However, another group's attempts to knock down the zf-T/no tail gene were less successful. While they could partially recapitulate the zf-T/no tail phenotype and avoid non-specific effects using 20-nt-long dsRNAs, a pool of RNase III–digested double-stranded zf-T/no tail dsRNA resulted in non-specific developmental defects similar to those seen with undigested dsRNA (Liu et al.,2005). Finally, a third group has used short 21-nt dsRNA to knock down the zebrafish dystrophin gene (Dodd et al.,2004). Although the authors claim their technique successfully eliminated gene function, the embryonic phenotypes are more similar to the non-specific effects of injection of long dsRNAs (Oates et al.,2000; Zhao et al.,2001; Gruber et al.,2005) than to the phenotype of sapje mutant embryos, which carry a mutation in the dystrophin gene (Bassett et al.,2003).

In zebrafish, shRNAs have not yet been tested for their ability to induce an RNAi response in live embryos. However, an intriguing experimental approach has been used to knock down gene activity using shRNA-based reagents. One group has reported that, by embedding the sequence of an shRNA within the intron of a gene, they could reduce the expression of a second target gene (Ying and Lin,2006). Here, the authors used a construct consisting of an RFP gene that has been split up into two exons by an intervening intron that carried an shRNA against EGFP. When this construct was expressed in transgenic embryos expressing EGFP, the green fluorescence of the EGFP decreased as the red fluorescence of the RFP increased (Ying and Lin,2006). Subsequently, this approach was used to address the role of miRNAs in the silencing of the fragile mental retardation-1 gene in zebrafish, the gene responsible for 99% of the human cases of Fragile X syndrome (Lin et al.,2006). Although the authors report changes in the neuronal populations of experimental embryos that are consistent with the defects seen in people afflicted with this condition (Lin et al.,2006), additional changes are also apparent in these embryos. In particular, these fish have gross morphological head abnormalities that are similar to the neuronal defects observed in dsRNA-injected embryos (Oates et al.,2000; Zhao et al.,2001; Gruber et al.,2005). Unfortunately, without appropriate controls it is impossible to determine if the neuronal changes are a specific and primary defect versus merely a secondary consequence of the head malformation. In the future, both siRNA and shRNA reports will need to include positive and negative controls showing that the constructs used are specific for the targeted gene, the phenotype observed is not caused by off-target (non-specific) effects, and that it can be rescued by competitors that block the RNAi response.

While dsRNA has been clearly ruled out as an RNAi method for inducing gene specific knockdown in zebrafish embryos, it remains a promising technique to study gene function in zebrafish cell culture assays (Gruber et al.,2005). On the other hand, it is difficult to say at this point if smaller siRNA and shRNA molecules will one day be deemed effective gene knockdown alternatives to, say, morpholino oligonucleotides. Issues that need to be addressed regarding RNAi technology include, but are not limited to, ensuring efficient and specific gene targeting events with minimal stimulation of host responses and preventing potential down-regulation of non-targeted transcripts by hyper-activation of the RNAi machinery. A systematic and well-controlled analysis of the phenotypes caused by expression of these molecules in a variety of conditions will be needed before discarding RNAi technology as a useful approach to study gene function in zebrafish.


Transgenesis (or the introduction of exogenous genes into the genome) can be exploited for a variety of tasks including analysis of regulatory elements, gene over-expression, tracing of cellular lineages, and mutagenesis. In contrast to other gene over-expression techniques, transgenesis can more readily be used to gain spatial and/or temporal gene expression control, removing the embryo-to-embryo variability associated with the injection of molecules into fertilized eggs. In recent years, the ease of generating stable transgenic fish has improved markedly. New techniques based on the use of enzymatic approaches, transposons, and retroviruses have dramatically increased the efficiency of permanently incorporating foreign genes into the zebrafish genome. As comprehensive reviews on this subject have recently been published (Grabher and Wittbrodt,2007; Kawakami,2007; Kikuta et al.,2007; Korzh,2007; Sivasubbu et al.,2007), we limit our discussion to a brief comparative description of each methodology, highlighting, with examples, some of the most recent technological advances in each field.

The first method devised for the generation of transgenic lines utilized injection of linear or circular DNA plasmids or bacterial artificial chromosomes (BACs) into fertilized eggs (Stuart et al.,1988, 1990; Culp et al.,1991; Lin et al.,1994; Amsterdam et al.,1995; Jessen et al.,1998,1999). This approach not only has low germ line transmission efficiency (5–20% transgene transmission to the F1 generation; Stuart et al.,1988; Culp et al.,1991), but also has the tendency of integrating the transgenes as concatemers of many tandem copies (Stuart et al.,1988; Culp et al.,1991; Cretekos and Grunwald,1999), which, as the fish ages or as the transgene is passed through several generations, are liable to become methylated and silenced (Gibbs et al.,1994; Thummel et al.,2006b). As a consequence of these limitations, new methods have gradually become more popular over the last decade. One of the methods used to increase the frequency of germ line transmission relies on flanking the transgene construct with the 18–base pair recognition site of the S. cerevisiae homing endonuclease I-SceI (Thermes et al.,2002). Co-injection of the construct and the I-SceI meganuclease into fertilized eggs increases the frequency of germ line transmission to approximately 30% (Thermes et al.,2002). This method has two additional advantages: the presence of the I-SceI sites brings the number of integration events per genome close to one and also eliminates the formation of concatemers (Thermes et al.,2002).

Another method used for the generation of transgenic lines uses the mobile DNA elements known as transposons. Mobilization involves the recognition of short inverted DNA repeats flanking the transposon by a transposase, which excises the transposon from one DNA location and inserts it in a different DNA region via recombination events (Largaespada,2003). Since the only sequences required for transposition are the short inverted repeats, almost any DNA molecule up to 10 Kb can be loaded into a transposon and injected into fertilized eggs (Balciunas et al.,2006). The transposase is provided to the cell by co-injecting it with the transposon on a separate DNA vector or as synthetic mRNA (Kawakami,2007; Korzh,2007; Sivasubbu et al.,2007). Although numerous transposons have been tested in zebrafish as transgenesis vectors (Fadool et al.,1998; Kawakami et al.,1998,2000; Raz et al.,1998; Kawakami and Shima,1999; Davidson et al.,2003), only two are currently in use: the Sleeping Beauty and Tol2 transposons (reviewed in Kawakami,2005,2007; Korzh,2007; Sivasubbu et al.,2007). Nowadays, Tol2-mediated transgenesis allows up to 50% of the injected fish to transmit the transposon insertion to their progeny (Kawakami et al.,2004), integrating most of the time as single insertions (Kawakami et al.,2000,2004). Furthermore, a large community effort has led to an increasing number of tools for the generation, manipulation, and delivery of transgenes using the Tol2 system, including site-specific recombination-based cloning kits for the modular assembly of constructs (Kwan et al.,2007; Villefranc et al.,2007), universal vector systems for the analysis of putative promoter and enhancers (Fisher et al.,2006b), and high cargo-capacity mini-Tol2 transposons (Balciunas et al.,2006). These advances have made transposon-based technology one of the most versatile methods for the analysis of genes and their regulatory elements, mutagenesis, and gene and enhancer trap screens (Kawakami,2007; Korzh,2007; Sivasubbu et al.,2007).

Pseudotyped retroviral vectors have also been used for transgenesis in zebrafish (Amsterdam and Becker,2005; Kikuta et al.,2007). The retroviral vector consists of the genome and two genes derived from the Moloney murine leukemia virus (MoMuLV) and the envelope glycoprotein from the vesicular stomatitis virus (VSV). This combination of viral components is used because the MoMuLV genome can easily accommodate DNA insertions, whereas the VSV envelope protein with its broad host range allows the infection of zebrafish cells. In general, high-titer retrovirus is required for independently infecting cells at the mid-blastula stage of embryonic development (Chen et al.,2002). Only those integration events occurring in the primordial germ cells will be passed on to the progeny, but because the germ lines are mosaic, only 1 to 20% of the offspring will be transgenic (Gaiano et al.,1996a; Linney et al.,1999; Chen et al.,2002; Ellingsen et al.,2005). Each transgenic animal may carry as many as 10 to 15 independent insertions, which then need to be screened in the F2 generation (Amsterdam et al.,1999; Chen et al.,2002). Because preparation of virus at sufficiently high titers can be challenging, retroviruses have been used primarily in mutagenesis, gene, and enhancer trap screens. The reasons why not all vectors will produce high virus titers are not completely clear or predictable, thus forcing the researcher to test multiple constructs before finding one that works (Chen et al.,2002). Despite this caveat, retroviruses have proven themselves invaluable to the zebrafish community in insertional mutagenesis and enhancer-trap screens (Gaiano et al.,1996b; Amsterdam et al.,1999; Golling et al.,2002; Amsterdam and Hopkins,2004; Ellingsen et al.,2005).

One use of transgenic approaches is to test the function of putative transcriptional regulatory sequences. Transient transgenic zebrafish embryos, into which DNA has been microinjected, can be used to rapidly obtain an idea of the capacity of specific pieces of regulatory DNA to drive reporter expression. However, the mosaicism of transient transgenics generated by simple injection of linearized DNA requires that a large number of specimens be analyzed before significant results can be obtained. To facilitate the measurement of transient transgenic reporter output, a “dual fluorescence” strategy has been developed to allow direct comparison of the output from a regulated transgene with that of a ubiquitously expressed transgene (Szeto and Kimelman,2004). By contrast, generation of stable transgenic lines was traditionally much more time-consuming, but arguably provided more reliable enhancer analysis, as well as valuable analytical tools. A good example of this was the generation of several islet1-GFP transgenic lines; these have allowed dissection of the regulatory elements driving islet1 branchimotor neuron expression (Higashijima et al.,2000; Uemura et al.,2005), and have also produced a useful genetic screening device for identification of mutants affecting branchimotor neuron migration (Wada et al.,2006) and axon pathfinding (Tanaka et al.,2007). In the last decade, numerous transgenic zebrafish lines have been generated (Shafizadeh et al.,2002) and this number is now expected to rise due to the dramatic increases in efficiency of transient and stable transgenesis afforded by Tol2-based systems (for example, see Burket et al.,2007). The Tol2 system has not only speeded up the generation of stable transgenic lines, but has also rendered analysis of transient transgenics a much more efficient and reliable process, as nicely demonstrated by a recent analysis of RET regulatory elements (Fisher et al.,2006a).

An obvious reverse genetic application of transgenesis, as previously mentioned, is to over-express genes of interest. As many gene products have roles at multiple stages, and these may well be obscured by phenotypes or even lethality associated with early global gain-of-function, there has been considerable interest in developing transgenes that can be regulated either temporally or spatially. For example, the use of a heat-shock promoter has proven extremely useful in allowing temporal control of stable transgene expression (Halloran et al.,2000; Skromne et al.,2007). Recently, by use of a modified soldering iron, it has been shown that heat-shock can also be achieved in a local fashion, enabling some spatial control of heat-shock-inducible transgenes, at least in superficial structures (Hardy et al.,2007). At least three additional genetic methods to regulate gene expression are currently being exploited to afford tightly regulated control: the Gal4-UAS system, the tetracycline (Tet-On) system, and the Cre/loxP system. Each of these methods can be used to activate gene expression in a regulated manner. This allows controlled gain-of-function, but also has broader applications such as expression of antimorphs to down-regulate gene function. Similarly, permanent activation of a reporter gene in a specific cell type allows the progeny of those cells to be followed providing efficient lineage tracing. The use of the Gal4-UAS system in zebrafish dates back to the late 1990s (Scheer and Campos-Ortega,1999). This approach exploits the yeast transcriptional activator GAL4 to drive expression of a gene of interest via its specific cis-binding sites known as upstream activating sequences (UAS). By generating “driver” lines where GAL4 is expressed in specific tissues, and crossing them with “effector” lines carrying UAS driven transgenes, expression of effectors can be targeted to any time or location for which a driver is available. In recent years, as transgenesis efficiency has improved, there have been efforts to produce more driver lines, generally using the more potent GAL4-VP16 fusion as the transcriptional activator of choice (Koster and Fraser,2001). For example, a pilot enhancer trap screen for new driver lines using a Tol2 vector carrying both GAL4-VP16 and a linked UAS:eGFP has recently been reported (Davison et al.,2007). This approach allowed 15 new stable transgenic lines to be produced that express both eGFP and GAL4-VP16 in reproducible patterns across a variety of tissues. Importantly, these new drivers were shown capable of activating UAS-regulated transgenes. A similar approach that focused on the nervous system identified 21 new neural specific drivers (Scott et al.,2007).

The Tet-On system again relies on a driver and an effector gene, but in this case temporal control is also dependent upon drug treatment. Huang et al. (2005) developed a heart-specific Tet-On system by using the cardiac myosin light chain 2 promoter to drive the reverse Tet-controlled transactivator (rtTA). rtTA was in turn used to drive expression of GFP via an rtTA-responsive element. This system is only activated when the stable Tet derivative doxycycline (Dox) is added to the medium, with a delay of approximately 7 hr for GFP detection. Transient GFP expression was achieved either by coinjecting separate plasmids carrying rtTA driver and GFP responder, or more efficiently by using a single plasmid carrying both (Huang et al.,2005). In principle, as with the Gal4-UAS system, stable transgenic driver lines can be established expressing rtTA in a variety of locations, and these in turn can be used to drive expression of any gene placed under control of the rtTA-response element. Currently, the Tet-On system lags behind the Gal4-UAS system, and its further development will require establishment of convenient driver lines that do not suffer from undue “leakiness.” The advantage of this system is that it affords an additional level of control via drug delivery. However, while gene activation in response to Dox is fairly rapid, inactivation in response to Dox removal is very slow and thus less likely to be useful.

The third methodology, the Cre/loxP system, is very widely used in mouse, in particular to allow conditional knockouts or to activate reporters for use in lineage tracing, but the approach can equally be used to facilitate conditional gain-of-function. In 2005, it was shown that Cre recombinase, introduced by mRNA microinjection, can successfully excise the dsRED gene from a CMV-loxP-dsRED2-loxP-EGFP transgene, demonstrating that the Cre/loxP system could be successfully applied in zebrafish (Langenau et al.,2005); in the absence of Cre function, this trangene expressed dsRED2 under control of the ubiquitous CMV promoter, while in the presence of Cre, the dsRED2 gene was excised thus placing GFP under the control of the CMV promoter to produce a switch from red to green fluorescence. This group went on to generate a rag2-loxP-dsRED2-loxP-EGFP-mMyc transgene, in which introduced Cre brings the myc oncogene under rag2 promoter control, to induce Myc activity in T cells and induce acute lymphoblastic leukemia. This use of Cre allowed conditional rag2-mMyc transgenic fish to be raised to adulthood and propagated via normal mating, whereas a non-conditional transgene produced severely diseased fish that were very difficult to maintain (Langenau et al.,2005; Feng et al.,2007). A similar approach was used to study the oncogenic properties of kRASG12D in zebrafish (Le et al.,2007). The success of this approach opens the way to using loxP flanked transgenes in conjunction with Cre-driver transgenic lines to conditionally activate transgenes specifically in domains where Cre is expressed (Pan et al.,2005). Other groups have generated a transgenic line carrying a heat-shock-inducible GFP-tagged Cre, and found that the expressed transgene was both compatible with viability and capable of excising chromosomally integrated loxP sites (Thummel et al.,2005). Disappointingly, although recombination was efficient, the pCMV-EGFP reporter generated by Cre/loxP excision of a stop sequence was inactivated in this particular case (Thummel et al.,2005). However, use of zebrafish-specific ubiquitous promoters rather than virally derived sequences such as CMV is likely to prevent such problems in the future (Burket et al.,2007). The Cre/loxP system is potentially very versatile (Nagy,2000), and once again an investment in developing Cre driver transgenic lines would provide an invaluable resource. The continued improvements in transgenesis technology make this a very feasible goal.


Disruption of a biological function by the inactivation of a gene product via mutation has genuine advantages over other gene inactivation methods. Foremost, mutations are permanent and can be used to study biological processes in embryos, larvae, or adults. Mutations can, in principle, eliminate, reduce, or modify the function of a gene. Furthermore, a null mutation is the only means to guarantee the complete loss-of-function of a gene. The numerous advantages that mutants have over other knockdown methods are such that, despite being costly, laborious, and time-consuming, numerous groups conduct routine small- and large-scale mutagenesis screens to identify mutants defective in a variety of biological processes (Mullins et al.,1994; Driever et al.,1996; Haffter et al.,1996; Henion et al.,1996; Fadool et al.,1997; Golling et al.,2002; Dosch et al.,2004; Wagner et al.,2004; Muto et al.,2005; Sadler et al.,2005).

Two approaches are emerging as complementary alternatives to forward genetic screens for the generation of loss-of-function alleles: targeted gene inactivation by homologous recombination and targeted detection of lesions after random chemical, transposon, or retroviral mutagenesis (discussed in Targeted Lesion Detection After Random Chemical or Insertional Mutagenesis). These methods can be used to (1) target genes of interest identified in genome projects, large-scale microarrays, morpholino or in situ hybridization screens; (2) expand the allele repertoire of previously characterized genes; or (3) target genes that, due to small size or redundancy, would not show an obvious phenotype in a typical forward genetic screen.

In mouse and other mammals, the method of choice for the generation of loss-of-function alleles of a gene of interest involves the targeted inactivation of such a gene by homologous recombination (Doetschman et al.,1987,1988; Capecchi,1989; Thompson et al.,1989; McCreath et al.,2000; Lai et al.,2002). In this method, the DNA recombination machinery of the cells is exploited to replace the gene of interest with a construct that modifies or eliminates the activity of the targeted gene (Capecchi,1989). While homologous disruption in mouse is carried out in embryonic stem (ES) cells (Doetschman et al.,1987,1988; Capecchi,1989; Thompson et al.,1989), successful inactivation of genes in mammals where ES cells are not available has been achieved by targeting the genes in fibroblast donor cells and subsequently transferring their nuclei to enucleated oocytes (McCreath et al.,2000; Lai et al.,2002).

Although targeted gene inactivation has not yet been achieved in zebrafish, significant advances towards the development of this methodology have been made over the past years. Zebrafish cells with ES-like properties have been isolated (Sun et al.,1995) and after cultivation and transplantation they have been shown to contribute to the germ-line (Ma et al.,2001), even after several culture passages (Fan et al.,2004; Fan and Collodi,2006). Although the frequency of germ-line chimera contribution is low (approximately 2–4%; Ma et al.,2001; Fan et al.,2004; Fan and Collodi,2006), the use of morpholino-based germ-line replacement methods will increase the efficiency of donor cell contribution to the germ-line (Ciruna et al.,2002). Nuclear transfer of genetically modified culture cells has also been achieved in zebrafish, using long-term cultured embryonic fibroblast cells (Lee et al.,2002). Although the overall rate of germ-line transmission is similar to that using ES-like cells (2% success rate), this approach demonstrated that nuclei from cultured cells can be reprogrammed to support embryonic development. Because terminally differentiated cells can be maintained in culture for longer periods of time than ES-like cells, they provide a longer window of time for various genetic manipulations. Thus, while slow, progress in this field has been significant, leaving the zebrafish community eagerly anticipating the day when this method becomes part of the zebrafish reverse genetic toolbox.


In contrast to targeted gene inactivation by recombination, where lesions in the gene of interest are generated in a targeted manner, target-selected gene inactivation is based on the systematic search for lesions in a gene of interest by analyzing the genomic DNA of randomly mutagenized individuals. In one way, targeted lesion detection is more akin to forward genetic screens than to homologous gene disruption methods: both target-selected mutagenesis and forward genetic screens utilize a population of randomly mutagenized individuals as starting material. However, while candidate mutants are selected on the basis of phenotypic abnormalities in forward mutagenesis screens, in the reverse genetic approach lesions are selected on the basis of nucleotide changes or DNA insertions in the gene of interest. This approach has clear advantages and disadvantages; it allows the immediate identification of the lesions, yet the functional consequences of such lesions cannot be predicted except in the rare event that the lesion generates a premature stop codon (nonsense mutation). This is the exact opposite situation to the results obtained in forward genetic screens, where the phenotypic changes used for selection assure that the function of a gene has been compromised, yet the phenotype tells little about the identity of the gene. Because forward genetic and targeted lesion detection screens make use of the same start-up material and provide complementary information about genes and biological processes, they can be combined into one large screening project (Wienholds et al.,2002).

In general, targeted lesion detection after mutagenesis consists of the generation of a mutagenized library followed by the systematic screening of lesions in the genes of interest. The two most frequently used methods in zebrafish to generate a library of mutagenized individuals, chemical mutagenesis using the alkylating agents ethyl methanosulfonate (EMS) or N-ethyl-N-nitrosourea (ENU) (Mullins et al.,1994; Solnica-Krezel et al.,1994; van Eeden et al.,1999; Lekven et al.,2000; Wienholds et al.,2002,2003b), and insertional mutagenesis using transposons or retroviruses (Raz et al.,1998; Amsterdam et al.,1999; Golling et al.,2002; Davidson et al.,2003; Ellingsen et al.,2005), can be used as a starting point for target-selected mutagenesis. For chemical mutagenesis, both EMS and ENU are highly effective in introducing random mutations in post-meiotic germ cells in zebrafish, but only ENU is an effective mutagen of pre-meiotic germ cells (Mullins et al.,1994; Solnica-Krezel et al.,1994; van Eeden et al.,1999). Similarly, high numbers of retroviral insertions, between 10 and 15, can be achieved in founders after infection, which can then be transmitted to individual F1 progeny (Amsterdam et al.,1999; Chen et al.,2002). However, there is bias for the retroviruses to integrate in 5′ non-coding regions of genes, including the promoter and first intron (Amsterdam and Becker,2005; Kikuta et al.,2007). On the other hand, the number of transposon insertions per founder tends to be lower than for retroviruses, in the single digits (Balciunas et al.,2004; Kawakami et al.,2004; Parinov et al.,2004), but insertions tend to be more randomly distributed through the genome (Kawakami,2005,2007). Regardless of the mutagenesis method used, the success of the screen depends on the quality of the mutagenized library. To increase the chances of obtaining mutants in a given set of animals, it is important to have high mutagenesis efficiency and a large pool of animals for screening. A third factor influencing the success of the screen is target gene size; the larger the target gene, the higher the chances of a lesion in that region.

The initial step in identification of lesions is the amplification of genes of interest using a PCR-based approach. If the mutagenesis was carried out using a transposon or a retrovirus, one can use a combination of vector- and gene-specific primers, which, if the vector landed within the gene, will result in the amplification of a DNA fragment. These DNA fragments can then be sequenced to identify the integration site. An alternative method involves digesting the genomic DNA with a restriction enzyme that cuts close to the 5′ ends of the vector, followed by the addition of linkers to these fragments. Linker-mediated PCR can then be performed with one primer specific to the vector and the other primer specific to the linker (Wu et al.,2003; Ellingsen et al.,2005). Further amplification can be achieved by nested PCR using a second pair of primers internal to the first primer pair. Subsequently, amplified products containing a region of the target gene can be identified by Southern blotting using the target gene as a probe. Vector insertion sites can then be identified by sequencing and aligning the amplified PCR products with zebrafish genomic databases (Ellingsen et al.,2005; Sivasubbu et al.,2006).

The relative simplicity by which affected genes can be identified using retroviral vectors has led Gollinger and colleagues at Znomics, Inc. (Portland, OR) and Burgess and colleagues at the NIH (Wang et al.,2007) to create libraries of thousands of retroviral insertions, which have been mapped to current genome databases. These libraries can then be used to identify specific provirus insertions in or near a gene of interest, and to recover the insertion from cryopreserved sperm. Although the presence of an insert in or near a gene does not guarantee that the gene is necessarily affected, chances of finding an insertion that impairs gene expression can be increased by choosing insertions within the coding or 5′ regulatory regions of a gene, the most favored retrovirus insertion site (Amsterdam and Becker,2005; Kikuta et al.,2007).

PCR and DNA sequencing are also the methods of choice for the identification of chemically induced lesions in zebrafish. In fact, the first target-selected mutagenesis screen carried out successfully in zebrafish relied on a live library of chemically mutagenized individuals, nested PCR, and direct sequencing for the identification of mutations in the rag1 gene (Wienholds et al.,2002), and more recently, mutations in the nanos and piwi genes (Draper et al.,2007; Houwing et al.,2007). However, the identification of a nonsense mutation with a likely loss-of-function phenotype by this method requires a very large number of sequencing reactions, which can be costly and time-consuming. For this reason, alternative screening methods have been developed to quickly identify individuals carrying nucleotide substitutions.

One increasingly popular method for the identification of point mutations is called Targeting Induced Local Lesions IN Genomes or TILLING (Fig. 4; Colbert et al.,2001; Till et al.,2003). This method has already been shown to quickly and reliably identify chemically induced point mutations in the zebrafish (Wienholds et al.,2003a). In TILLING, PCR is used to amplify the gene of interest in wild type and mutant individuals, and to label each product with a different fluorescent label. DNA heteroduplexes between wild type and mutant PCR fragments are formed by mixing, denaturing, and re-annealing the DNA products generated during the PCR amplification step. These DNA heteroduplexes are then incubated with celery derived CelI endonuclease, which recognizes and cleaves mismatches in the heteroduplex DNA generated by small nuclear polymorphisms (SNPs) and point mutations (Oleykowski et al.,1998). After purification, labeled digested fragments are separated and visualized on slab gel sequencers. Fragments that are generated due to the presence of SNPs will be present in all individuals. In contrast, novel point mutations will generate unique fragments in only a few individuals (Fig. 4). These sequence changes can then be confirmed by direct sequencing of the original amplified product and the genomic DNA (Wienholds et al.,2003b; Draper et al.,2004; Wienholds and Plasterk,2004; Sood et al.,2006).

Figure 4.

Gene inactivation by target-selected mutagenesis by TILLING (Targeting Induced Local Lesions IN Genomes). A: General flow-through diagram for the inactivation and selection of target genes using chemical mutagenesis and TILLING in zebrafish. Important steps in the TILLING procedure are explained in more detail in B. B: Identification of local lesions in targeted genes. Gene of interest is amplified using PCR and primers that are labeled at their 5′ end with different fluorescent compounds (triangles, hexagons). Products are heat denaturated and slowly reannealed to generate double-stranded DNA containing mismatches due to polymorphisms or mutations (hybrid DNA). DNA is then treated with CelI nuclease, which cleaves the DNA at unpaired bases. DNA is once again denatured, run on an Acrylamide gel, and scanned. Numerous cleavage products resulting from DNA polymorphisms are frequently seen. Rare cleavage products, the likely result of mutagen-induced base pair substitutions, are seldom seen. Additional rounds of selection and sequence analysis are required to confirm the presence of a mutation.

As target-selected mutagenesis relies on the identification of nucleotide changes in the gene of interest rather than on phenotypic changes, the functional consequences of such lesions cannot be directly predicted. It has been shown that retroviral integration in coding, intronic, or regulatory regions of a gene can lead to functional disruption of the gene in some cases (Amsterdam et al.,1999; Golling et al.,2002), but not others (Ellingsen et al.,2005). Single nucleotide changes may be silent, with no phenotypic consequence. For the identification of loss-of-function alleles, one aims to identify retroviral insertions, nonsense or splice site mutations in the coding region of the gene, close to its 5′ end. However, the remaining pool of retroviral insertion or missense mutations is also likely to contain genes with functional changes. These other alleles could represent the full spectrum of protein activities including partial loss of function (hypomorphs), dominant-negative (antimorphs), or novel functions (neomorphs). While predicting which missense mutations might give rise to proteins with a functional consequence is challenging, some of these mutations have nonetheless proven useful in gene function characterization (Chasman and Adams,2001; Ng and Henikoff,2001). This is particularly true if the null allele is lethal or sterile as characterization of the gene's function may be carried out using weaker alleles.


The reverse genetic techniques of gain-of-function by mRNA injection and gene knockdown by morpholino injection have become standard for the zebrafish system. These approaches have proven to be extremely powerful despite some inherent limitations. One important technical challenge to address in the future of zebrafish reverse genetics is the achievement of conditional knockdowns (or knockouts) of specific genes. Efforts to achieve homologous recombination in zebrafish ES cells that can then reconstitute the germ line have not yet been fully rewarded; although it should be remembered that to date the mouse remains the only organism where targeted knockouts can be achieved, with even the closely related rat proving intractable. Thus, while efforts to develop zebrafish knockout technology should certainly continue, it is also important to develop alternative approaches. Excitingly, new and straightforward technologies have recently been developed that make the generation of zebrafish transgenic lines a relatively quick and simple process, achievable by any zebrafish lab. This major step forward will enable increased use of the various transgenic strategies to achieve regulated gene expression. In addition to allowing conditional gain-of-function, this approach opens the way to the use of inducible transgenes to express RNAs capable of gene knock down. The zebrafish scientific community has always been a particularly collaborative one, and the continued sharing of new techniques and resources will allow the most efficient use of the emerging reverse genetic tools for this organism.


We thank Steve Johnson for suggesting this topic for review and Robert Ho for his support during the process of writing it. We also thank Cheng Huang for critically reading the manuscript, and three anonymous reviewers for their insightful criticisms and suggestions. I.S. is a fellow at the UM Neuroscience Center.