Introduction

The zebrafish Danio rerio has emerged as a powerful model system in which to study numerous aspects of development. The characteristics of small size, external fertilization, and transparency during embryogenesis allow for easy manipulation and observation of developmental processes. Like other model systems, large-scale genetic screens have been performed to identify key genes responsible for the development of such organs as the skeleton and heart [1, 2]. There are certain drawbacks to genetic screens though, for example some mutations may lead to zygotic lethality. Conditional perturbations have been employed to bypass early requirements of developmental processes. Chemical screens are an example of a conditional approach.

Zebrafish are well-suited for use in chemical screens. Small molecules are typically added to the aquatic environment in which they live, allowing absorption into the fish without the need for invasive and time-consuming injection. Zebrafish embryos can be arrayed in 96-well plates with several embryos per well, compounds can be added using pins or pipettes, and the plates can be screened either by eye or use of high-throughput video techniques [3, 4]. Today, numerous chemical libraries exist for this purpose, from small collections of characterized compounds to larger libraries consisting of tens of thousands of compounds of uncharacterized function.

Classic genetic screens have been remarkably fruitful in identifying genes that govern essential developmental processes. Many of the lessons learned from these screens can be used as guiding fundamentals for chemical screens in zebrafish, such as scoring obvious phenotypes and the use of reporter genes as an output. However, there are some challenges unique to a chemical screening approach. In early genetic screens, the identification of the gene responsible for the phenotype was the starting point for laborious downstream investigation of gene function. However, with the advancement of genetic tools, progress in gene annotation, and development of methods for analyzing gene function, the path from mutation to function has become more straightforward over the last decade. This is less true for small molecules where bridging the gap from compound identification to mechanism of action remains a significant challenge. In this review, we discuss several recent screens highlighting unique aspects of screen designs as well as methods that have been used to identify the mechanisms of action of the resulting compounds.

Screening Approaches

General Considerations

Most chemical screens accomplished to date rely on the visual observation of individual zebrafish embryos within a 96-well plate. An important consideration when optically screening, whether manually or in an automated fashion, is embryo orientation. A potential advantage of early screening is that prior to inflation of the swim bladder at approximately 72 h post-fertilization, de-chorionated embryos lie on their side at the bottom of the wells with little spontaneous movement. Certain organs, such as the heart, can be readily observed while in this position. Zebrafish are well-suited to screens for cardiovascular phenotypes. Cardiac development proceeds rapidly with a beating heart at 24 h after oocyte fertilization. However, cardiac function is not required for survival during the first 5 days as the embryo can subsist on simple diffusion. This characteristic allows for the study of early development despite severe developmental defects in both heart morphology and physiology without the embryo dying. These traits have allowed models of several human cardiovascular diseases to be developed in zebrafish including dilated cardiomyopathy [5], vascular lipid deposition [6], long QT syndrome [7], aortic coarctation [8], and arrhythmic right ventricular cardiomyopathy [9].

Another consideration is the number of embryos per well. There is a potential tradeoff between throughput and phenotypic certainty. While many (up to ten) embryos per well can increase the certainty of a given phenotype, especially if the drug effect is not 100% penetrant, it also increases the amount of labor. A final consideration when setting up a small molecule screen is ease of readout. While visual examination of phenotypes is sometimes preferred, this analysis may be time-consuming and laborious. The technology to use fluorescent markers in zebrafish is readily available, and complements the embryonic feature of transparency. Organs such as the heart or blood vessels are easily visualized in transgenic fish that express fluorescent proteins driven by the organ specific promoters. An added advantage of using such markers is that changes in expression levels may be read by automated video microscopy, allowing for high-throughput screening.

Screens Targeting Specific Developmental Processes

While there is considerable discussion in the literature about the use of small molecule screens for drug discovery, many of the compounds that have been identified to date are better suited as probes of specific developmental processes. These probes can be powerful tools for use in whole animal studies (Fig. 1).

Fig. 1
figure 1

Approaches to identify the target of small molecules identified in chemical screens. Small molecule targets can be identified using: (1) previously known binding partners of the small molecule. (2) Structure–activity relationship (SAR) approaches to look for similar molecules with known mechanisms. (3) The similarity of the chemotype to a known genetic phenotype. (4) Purification of the small molecule binding partner. (5) Systems biology approaches to compare large data sets in order to cluster the data into functional units

Full size image

The first published small molecule screen in zebrafish focused on identifying perturbations of four developmental processes: formation of the central nervous system, the cardiovascular system, the otic structure similar to the ear, and pigmentation [10]. A library of 1,100 compounds was screened, and molecules were identified that caused specific defects each of these processes. In fact, 1% of the library compounds were found to affect at least one of these developmental systems. Several nervous system phenotypes were observed, such as increase in the size of the hindbrain and a duplication of neural folds. Similarly, interesting phenotypes were found in the cardiovascular system included alterations in cardiac morphology and an increase in the ratio of atrial to ventricular contractions. Pigmentation phenotypes included one molecule that inhibited all pigment production, and one that preferentially inhibited pigmentation of neural crest-derived melanocytes. Finally, one compound prevented otolith formation in the otic placode. Importantly, by adding or washing away compounds at specific times, investigators were able to determine critical time windows for compound effects, demonstrating an exquisite temporal control that could not be accomplished using traditional genetic approaches. This study demonstrated the feasibility of small molecule screens for developmental phenotypes, and also established a template for future screens.

Similarly, two groups performed chemical screens for compounds that alter vascular patterns in the developing eye [11, 12]. Alvarez et al. used a small library of compounds known to affect angiogenesis, identifying one compound, LY294002, a PI3 kinase inhibitor. This compound prevented the formation of new vessels, without disturbing existing vessels, and had no measurable effect on visual acuity. In a larger screen (2,000 compounds) Kitambi et al. identified small molecules that affected vessel morphology. Two of these molecules (enalapril maleate and zearalenone) cause a widening of the diameter of the vessels while three (pyrogallin, albendazole, and mebendazole) cause a loss of vessels. These screens were noteworthy as they demonstrated a small molecule approach could be used to examine very subtle phenotypes such as the development or size of single blood vessels.

The development and homeostasis of hematopoietic stem cells was the subject of a chemical screen in zebrafish conducted by North et al. [13]. These investigators screened 2,357 compounds and assayed the number of hematopoietic stem cells by in situ hybridization at 36 h post-fertilization. The classic challenge of identifying mechanistic pathways was obviated by careful examination of the 77 compounds identified. Ten of these compounds affected the prostaglandin pathway. Further work demonstrated that prostaglandin E2 (pgE2) administration results in expansion of multipotent hematopoietic progenitors. Importantly, this compound identified in zebrafish was able to improve graft potency in a murine transplantation model. This work has resulted in the use of pgE2 in clinical trials as a graft enhancer in humans (clinicaltrial.gov identifier NCT00890500) and highlights the use of chemical screening in zebrafish as a relevant tool of discovery in translational therapeutics.

Interestingly, some of these studies discovered phenotypes similar to specific syndromes or diseases, even though the focus was on the use of small molecules to dissect developmental processes. The utility of the developmental probes identified in these studies is highly dependent on understanding their molecular and cellular mechanisms of action, which can be challenging to determine. One solution to this difficulty is to design screens that probe a specific pathway. Screens designed this way should yield compounds that are active in the pathway of interest.

Screens Targeting Known Signaling Pathways

While the previous examples focused upon perturbations of developmental processes, other screens are aimed at disrupting specific signaling pathways within zebrafish. When embarking on this type of small molecule screen in a whole organism, it is worthwhile to consider whether a simpler biological system might achieve the desired goal. Protein or cell-based assays may offer several advantages, depending on the particular assay, including the ease of scale-up, smaller format, cheaper materials, and higher throughput. However, whole organism approaches have their own advantages that can include the delivery of higher content information such as cell-type specific readouts, the study of cellular function in the native context, the ability to detect pro-drugs, or active metabolites, and the fact that the cells are not transformed. The choice of organism is also important. For instance, if screening for an early developmental cardiac phenotype such as cardioblast migration, one might choose Drosophila as it has a primitive heart tube structure and is amenable to high-throughput screening techniques [14, 15]. However, if one is more interested in higher order cardiac structure, for instance valve development, the zebrafish model system would be more appropriate, as it has structures that resemble human valves [16]. A final advantage to using a whole organism approach is that compounds that display organ specific toxicity may be missed in screens that rely on a protein or cell-based readouts. Compound toxicity is an important problem in many drugs of therapeutic potential. For instance, life-threatening arrhythmia can be caused by drugs that initially appear promising as therapeutic candidates. Zebrafish appear to be a reliable detector for such arrhythmias that would be missed in a simpler screen [17].

One study that directly explored whole organism versus cell assays was a screen for cell cycle inhibitors undertaken in zebrafish embryos [18]. Investigators screened a library of 16,320 compounds for alterations in phospho-histone H3 expression patterns, and identified 14 compounds that were not previously known to have cell cycle activity. They followed up these results in both zebrafish and human cell culture proliferation assays, showing that six of these compounds had activity in both human and zebrafish cell culture assays, one appeared to be zebrafish specific with activity only in zebrafish cells, three were serum inactivated in the cell culture assays, and four showed activity only in the zebrafish whole organism assay and neither cellular assay. This last group may target signals that are not reproduced in the culture model, such as unique cell–cell interactions, or may be pro-drugs that are converted in vivo to an active metabolite. Of note, because this last group of compounds did not have activity in either zebrafish or human cell culture lines, the authors could not be sure that these were not zebrafish specific. On balance, it would appear that the effect of performing this screen in zebrafish, as opposed to a cell-based system, included identification of one to four false-positive results (a zebrafish specific compound), but also at least four and perhaps seven molecules that would have been missed in a cell culture screen. These results argue that there are tradeoffs between simple biologic systems and whole organism screens that may offset the challenges of the latter.

In a similar study where the authors chose to use zebrafish rather than a simpler cellular assay, Torregroza et al. synthesized and screened a small library of compounds that were based around a flavone lead known to antagonize TGF-β binding to the TGF-β receptor II [19]. Screening of this small library revealed several active compounds that elicited developmental defects. One compound in particular yielded severe body axis defects in early development and at lower doses affected cardiovascular development. Further analysis showed no change in phospho-Smads 2 and 3 or phospho-Smads 1 and 5. Instead, the compound appears to modulate a Smad-independent TGF-β pathway, SAPK/JNK. In this example, the use of the whole organism assay enabled the discovery of interesting, relevant off-target effects of a library compound that would have been missed by a simpler cellular assay, for instance, one that looked only at phospho-Smad levels. Here, the complexity and context of the whole organism screen appeared to work to the advantage of the investigators.

A BMP pathway screen was based on the observation that mutants in the BMP pathway have dorsoventral axis defects [20]. No inhibitors of the BMP pathway had yet been described, therefore Yu et al. initiated a screen for compounds resulting in dorsalization. Screening a 7,500-member library resulted in one hit, which investigators named dorsomorphin. Subsequent experiments revealed that dorsomorphin acts by inhibiting the BMP type I receptors ALK2, ALK3, and ALK6, and revealed a role for BMP signaling in iron homeostasis. A more potent analogue of dorsomorphin was later identified [21] and was found to ameliorate a mouse model of the congenital disorder fibrodysplasia ossificans progressiva [22]. This result is further evidence for the utility in using zebrafish as a first-order screen to identify compounds of therapeutic interest for human disease.

Finally, an interesting transgenic model was developed to search for modulators of the FGF pathway [23, 24]. The dual-specificity phosphatase 6 (Dusp6) is both regulated by, and serves as a regulator of FGF signaling. Therefore, investigators created a transgenic animal of which the dusp6 promoter drives a destabilized GFP as a reporter of FGF signaling. A library of 5,000 compounds was screened and one compound, BCI, was identified as an enhancer of dusp6::GFP expression. Examination revealed that BCI binds to and inhibits the Dusp6 phosphatase, preventing ERK dephosphorylation and leading to upregulation of FGF. It is unclear whether this screen could have been carried out as successfully in a cell culture assay. One anticipated advantage of the whole organism fluorescent transgenic reporter might have been identification of compounds that caused expansion of reporter expression beyond its usual boundaries, but this did not occur. The one hit identified resulted in a brighter signal within the usual pattern. Additionally, GFP intensity, which is not usually considered a quantitative reporter of gene expression, in this case served as a reliable and screenable reporter of gene expression.

Suppressor Screens of Mutations and Disease Models

Animal models of disease can be powerful tools to examine pathophysiology. Further, these models can potentially be used as tools for therapeutic discovery, assuming they faithfully re-capitulate the disease in question. In theory, the therapeutic promise of compounds discovered in disease model screens could reduce the need for a detailed understanding of mechanism of action, since many clinically useful drugs still have no clearly defined mechanism of action. Here, we discuss chemical screens performed on zebrafish models of human disease. These screens are unbiased with respect to small molecule mechanism and are akin to genetic suppressor screens. The first zebrafish chemical screen for suppression of a disease phenotype was performed in a model of aortic coarctation [8]. Specifically, hypomorphic mutations in the gene hey2 cause the gridlock phenotype, in which blood flow is disrupted due to defects in vessel formation. The screen identified two compounds, both of which caused a remarkable reversion in the gridlock mutant to the wild-type phenotype. The primary hit proved to activate the VEGF pathway, and further experiments demonstrated that upregulation of the VEGF pathway was sufficient to rescue the gridlock phenotype. This study is an early example of the power of high-throughput screening aimed at ameliorating a mutant phenotype, and demonstrates the utility of zebrafish pharmacologic screens for understanding disease processes and discovering potential new therapies. Recently, the efficacy of a gridlock suppressor in promoting new blood vessel formation has been demonstrated in a mouse hindlimb ischemia model, providing further evidence that small molecules discovered in zebrafish screens can be translated to mammalian systems [25].

Since the original gridlock screen, multiple human diseases have been modeled in zebrafish, and screens to search for suppressors of the disease phenotypes have been performed. These screens have varied in how they model human disease.

For instance, a model of aminoglycoside-induced hearing loss was developed based on the hair cells present in zebrafish neuromasts which resemble mammalian inner ear hair cells [26]. Treatment with the antibiotic neomycin effectively kills these cells. A diverse 10,960-compound library was screened for protection of hair cells from neomycin-induced death. Two compounds, both benzothiophene carboxamides, were identified. Importantly, these compounds did not inhibit neomycin’s antibacterial effect, which could allow them to be useful in a clinical setting.

A different approach was used to study polycystic kidney disease (PKD). Zebrafish mutants have been isolated in forward genetic screens that harbor lesions in genes known to be important in PKD pathology. In addition to the cysts present in the zebrafish kidney, these mutants also display curved body phenotypes. Because of the challenge in scoring changes in embryonic kidney cysts, body curvature was used as a surrogate to screen a small library of known compounds for suppressors of the pkd2 mutant phenotype [27]. Of six compounds that suppressed body curvature, two also prevented kidney cyst formation, and revealed a role for histone deacetylase in PKD pathogenesis.

A zebrafish model for leukemia, developed by overexpression of a known human oncogene AML-ETO, mimics many aspects of the disease, including changes in hematopoietic precursor differentiation. A library was screened using in situ expression of the gene gata1 as an output [28]. The COX-2 inhibitor nimesulide was identified. Further experiments implied that AML1-ETO acted to disregulate hematopoiesis through a pathway that included COX-2, prostaglandin E2, and β-catenin.

These studies highlight one of the great advantages and goals of chemical screening in zebrafish: the promise of discovering novel therapeutics to alleviate in vivo disease phenotypes. Even if compounds identified are not therapeutics per se, they may be helpful in examining the pathophysiology of disease.

Identifying Lead Compound Mechanisms

While genetic screens can rely on the mapping of a heritable locus in order to identify a disrupted gene, one of the major challenges in a chemical screen is elucidating the mechanism of action for a resulting hit. Some of this difficulty can be mitigated during the initial screen design. For example, BCI, identified in the screen for enhancement of dusp6 promoter::GFP reporter was expected to act upon a component of the FGF pathway, and was found to inhibit Dusp6 itself. Similarly, the dorsoventral axis screen was anticipated to discover modifiers of the BMP pathway, which it did. However, pathway agnostic disease screens cannot rely on this luxury. Some of the approaches taken to understand a compound’s mechanism of action are described in this section.

Known Compounds

Many libraries are designed to contain small molecules whose binding partners and mechanisms are known. This eases the discovery process, as an investigator need only to confirm that the compound is acting through that pathway in the zebrafish model. For example, in a screen that looked for chemical modulators of zebrafish regeneration, the glucocorticoid beclomethasone was identified as a potent inhibitor of fin regeneration [29]. Beclomethasone is known to act through the glucocorticoid receptor in other systems to upregulate target genes, including FKBP5, GILZ, and SOX9b. The investigators demonstrated upregulation of these target genes in beclomethasone-treated fish, and more importantly showed that knockdown of the glucocorticoid receptor restored regenerative capability in the presence of beclomethasone.

Similarity to Known Mutant Phenotypes

Compound mechanisms may also be identified by comparison of the drug-treated phenotypes to known zebrafish mutants. DTAB is a small molecule that causes specific anterior-posterior defects in zebrafish embryos during development [30]. The shortened axis of DTAB-treated fish resembles that observed in mutations in components of the retinoic acid signaling pathway. Analysis demonstrated that DTAB works by upregulating the retinoic acid signaling pathway through specific retinoic acid receptors.

However, just because a molecule phenocopies a known mutation, does not always imply it affects the same pathway. For example, concentramide is a compound that causes the cardiac ventricle to form in the center of the atrium [31]. This phenocopies the mutant heart and soul (has) which is due to a mutation in the PKC-lambda gene. However, concentramide and the mutation of has appear to operate through different signaling pathways to interfere with early heart field fusion.

Structural Similarities and Structure–Activity Relationships

Once a compound has been identified in a screen, hypotheses about its mode of action may be generated by examination of its structure. Databases such as the similarity ensemble approach are available that allow searches based on structural similarities that can identify similar compounds whose mode of action is known [32]. Testing with the known compound can then aid in determination of mechanism. Structurally similar compounds with unknown mechanisms of action can be used to define structure–activity relationships (SAR) leading to compounds with greater potency, better specificity and/or less toxicity than the first compound. This approach has been used successfully to identify compounds that gave better specificity than the original hits [11, 33, 34].

Affinity Purifications

Another approach to investigating a compound’s mechanism of action is identification of binding partners using affinity chromatography. In this approach, a chemical linker is attached to the compound of interest, allowing it to be immobilized on a solid support. Zebrafish lysates are then incubated with the solid support, followed by washing then elution of the binding partner. Identification can be accomplished using a variety of mass-spectrometry and proteomic methodologies. This strategy was employed successfully by investigators studying compounds that alter zebrafish pigmentation [35, 36]. One unique advantage of their approach was that they started with a triazine-based library that provided a built-in affinity handle, allowing them to quickly perform purifications. Using this approach, they were able to identify a compound that bound to a mitochondrial ATPase, and demonstrate that modulation of the ATPase resulted in changes in pigmentation in vivo.

The Next Generation of Screens

The screens discussed above have focused largely on defined biological pathways or structural models of human disease. The stage has now been set to exploit the power of zebrafish pharmacological screens to study more complex phenotypes including physiologic parameters and behaviors. The development of high-throughput assays for complex behaviors presents unique challenges, but two recent reports have demonstrated the feasibility of this approach. The first used motion analysis software to define movement during wake and sleep cycles in zebrafish, and then screened approximately 4,000 unique compounds for modifiers of movement behavior [37]. The second screen defined a complex behavior, the photomotor response (PMR), a robust behavioral response to flashes of light [4]. A library of 14,000 compounds was screened for modifiers of the response. In both cases, original high-throughput screening platforms and analysis programs were developed. The shear amount of data generated made analysis challenging, so both groups shared a system to create barcodes, allowing for hierarchal clustering of the data into groups of phenotypically related molecules. This proved to be a powerful tool in predicting modes of action for unknown compounds that produced specific behavioral changes. For example, in modifiers of the PMR response one phenocluster (which caused a “slow-to-relax” behavior) contained acetylcholinesterase (AChE) inhibitors, along with two structurally unrelated and uncharacterized compounds. Hypothesizing that these two molecules may also be AChE inhibitors because of their presence in the phenocluster, Kokel et al. found that these compounds did indeed have the predicted activity in vitro and in vivo. Cluster analysis of compounds identified in the screen for circadian movement revealed interesting details on pathways involved in this process, including a role for the inflammatory pathway in setting normal daytime movements and the finding that inhibitors of the potassium channel Ether-a-go-go-Related Gene (ERG) increase waking activity at night in zebrafish.

In addition to behavioral screens, other physiologic parameters are becoming accessible for screening as well. For instance transparency and rapid development of the heart, possibly combined with fluorescent reporters allow rapid measurement of heart rate and rhythm (arrhythmia). Many drugs of therapeutic potential have been taken off the market due to their liability to act as ERG channel blockers, which can lead to prolongation of the QT interval and life-threatening arrhythmias. Small-scale screens have already been performed to establish zebrafish as a model in which to analyze heart rate, ERG function, and to screen drugs that may affect both of these parameters [17, 38]. Commercial zebrafish assays are now available that screen for liability to prolong the QT interval, highlighting the potential of physiologically based chemical screening.

Conclusions

Numerous chemical screens have been performed in zebrafish since the first developmental screen described 10 years ago. These screens have resulted in the identification of important probes of organismal processes, along with several compounds of potential therapeutic value. Indeed, prostaglandin E2 is in early-stage clinical trials for use in graft enhancement. Emerging efforts into more complex behavioral and physiologic phenotypes will exploit the power of whole organism phenotypic screening in the years to come.