Here, we present a systematic approach for developing physiologically relevant, sensitive and specific in vivo assays for interpreting variation in human pathology. Transient genetic manipulation via microinjection of WT and mutant human mRNA and morpholino (MO) antisense oligonucleotides harness the tractability of the developing zebrafish embryo to rapidly assay pathogenic mutations, especially, but not exclusively, in the context of human developmental disorders.
Here, we present methods for the development of assays to query potentially clinically significant nonsynonymous changes using in vivo complementation in zebrafish. Zebrafish (Danio rerio) are a useful animal system due to their experimental tractability; embryos are transparent to enable facile viewing, undergo rapid development ex vivo, and can be genetically manipulated.1 These aspects have allowed for significant advances in the analysis of embryogenesis, molecular processes, and morphogenetic signaling. Taken together, the advantages of this vertebrate model make zebrafish highly amenable to modeling the developmental defects in pediatric disease, and in some cases, adult-onset disorders. Because the zebrafish genome is highly conserved with that of humans (~70% orthologous), it is possible to recapitulate human disease states in zebrafish. This is accomplished either through the injection of mutant human mRNA to induce dominant negative or gain of function alleles, or utilization of morpholino (MO) antisense oligonucleotides to suppress genes to mimic loss of function variants. Through complementation of MO-induced phenotypes with capped human mRNA, our approach enables the interpretation of the deleterious effect of mutations on human protein sequence based on the ability of mutant mRNA to rescue a measurable, physiologically relevant phenotype. Modeling of the human disease alleles occurs through microinjection of zebrafish embryos with MO and/or human mRNA at the 1-4 cell stage, and phenotyping up to seven days post fertilization (dpf). This general strategy can be extended to a wide range of disease phenotypes, as demonstrated in the following protocol. We present our established models for morphogenetic signaling, craniofacial, cardiac, vascular integrity, renal function, and skeletal muscle disorder phenotypes, as well as others.
The functional interpretation of genetic information and assignment of the predictive clinical value to a genotype represents a major problem in medical genetics and is becoming increasingly poignant with the accelerating technical and economic feasibility of genome-wide sequencing. Therefore, it is necessary to develop and implement new paradigms to test the pathogenicity of variants of unknown significance (VUS) detected in patients. These assays must then be accurate, time- and cost-efficient, and harbor the potential to catalyze a transition to clinical utility.
While the mouse has been traditionally the tool of choice in the field of human disease modeling, zebrafish are emerging as a scientifically and economically favorable surrogate. Unlike the mouse, zebrafish biology allows easy and timely access to all developmental stages, aided by optical clarity of embryos which allows for real-time imaging of developing pathologies.1 The relatively recent generation of mutant zebrafish lines has provided additional testing and modeling options, employed by many in functional studies, but this technology continues to be limited (reviewed in 1,38 ). Not only are genetic mutants with knock-ins of specific mutations laborious to attain, they are also not amenable to medium- or high-throughput analysis for the testing of a range of mutations within a single gene. Importantly, a single suite of tests can provide information not only for the pathogenic potential of alleles, but also for the direction of effect at the cellular level (e.g. loss of function vs. gain of function), which is critical for informing mode of inheritance in families, especially when small human pedigrees harbor limited information on the mode of genetic transmission. For further comparison of the uses of available mouse and zebrafish models, see Table 1.
We also note that there are inherent limitations to the zebrafish model system. Although D. rerio have rapid initial development of organ systems, sexual maturity requires approximately three months. Because of this, prenatal and pediatric-onset disorders are the most amenable to this transient expression model. While ideal for conducting large chemical compound screens, the use of genetic mutants is not feasible for the systematic testing of thousands of nonsynonymous variants that contribute to, and continue to be detected in pediatric disorders.
The complementation tests described here take advantage of this experimental tractability, high degree of homology, and preservation of function between human and zebrafish proteins, particularly so for molecules necessary for conserved developmental processes. Figure 1 outlines the testing and identification strategy for various allele effects. Both loss of function (LOF) and dominant assays can be performed. For LOF, the experiment begins with the suppression of the gene of interest with a morpholino knockdown, and assaying for phenotypes that might be relevant to the clinical phenotype under investigation. Suppression can be achieved either by blocking translation by targeting a MO at or near the translational start site of the zebrafish locus (translation blocker morpholino; tbMO) or by interfering with splicing by placing a MO on a splice junction, typically inducing either inclusion of an intron or aberrant exon skipping (splice blocking morpholino; sbMO).
Subsequently, capped mRNA from the orthologous human transcript is introduced and quantifiable rescue of the phenotype is measured. Once the assay is established, candidate mutations in the human message can be introduced and assayed for their ability to rescue the MO-induced phenotype at the same efficiency as WT human mRNA. Conversely, for candidate dominant alleles, human mRNA (but not MO) is introduced with an expectation that WT human mRNA will not grossly affect zebrafish anatomy and physiology, whereas introduction of test mutations that have a dominant effect will induce phenotypes analogous to those observed in the human clinical condition. This experiment can be fine-grained further to dissect whether the dominant effect occurs by a gain of function (GOF) or a dominant-negative mechanism by blending WT and mutant human mRNA; for GOF events, addition of WT human mRNA is expected to be irrelevant, whereas for dominant-negative alleles, blending of WT and mutant mRNA should alter the severity of the phenotype induced by mutant message. In all cases, we recommend that all combinations of injections (MO with WT human mRNA vs. morpholino with mutant human mRNA etc. be performed, preferably within the same clutch of embryos (see Figure 1). Interpretation is as follows:
For LOF tests:
For dominant tests:
Contingency plan:
The methods described here represent a general protocol applicable to the assay of nonsynonymous changes associated with a variety of human genetic disease phenotypes (Table 2, Figure 3). Our approaches have proven useful to evaluate the potential impact of variation on disease phenotypes, and to help dissect disease mechanisms (such as the contribution of dominant negative mutations to Bardet-Biedl Syndrome, a primarily autosomal recessive disorder 17). To date, through the development of the presented decision tree, we have modeled at reasonable cost and time in excess of 200 genes causally associated with genetic disorders, to an excess of 1,000 alleles.
Although not discussed in detail here, we have also shown that these methods are appropriate to model other types of genetic lesions, such as copy number variants (CNVs) as well as and genetic interactions. Analyses of such events are beyond the scope of the present method description, although they fundamentally rely on the same principle of systematic testing of candidate genes (including pairs of genes injected concurrently) to determine induction or exacerbation of clinically-appropriate phenotypes. For example, to elucidate which of the 29 genes in the 16p11.2 CNV might be relevant to the observed microcephaly observed in patients with duplications of a 660 kb genomic segment, mRNAs corresponding to each of the 29 genes within the segment were injected and head size measurements were conducted at 2 dpf and 5 dpf, revealing a major contribution of a single transcript, KCTD13.21 Additionally, we have used this model to assay genetic interactions of genomic lesions in patients with both Bardet-Biedl syndrome and Hirschsprung disease.22 By comparison of MO suppression of the causal genes of the two clinical identities separately and simultaneously, we were able to identify the resulting phenotype as being a synergistic interaction rather than merely additive severity.
Despite having established high sensitivity (98%) and specificity (>82%) for variants contributing to ciliopathies17, we do not yet have sufficient data to determine if these are generalizable to all phenotypic readouts in zebrafish models. To do so, a large number of alleles, predicted genetically to be either benign or pathogenic, must be tested within each phenotypic category. This will be particularly important for the implementation of such assays in the clinical setting, wherein functional interpretation of VUSs can inform diagnosis and management only if a robust understanding of false positives and false negatives can accompany the delivery of such results to physicians and patients. Nonetheless, these methods can contribute significantly toward a better understanding of the landscape of human genetic disease. We anticipate that these models will not only serve as a foundation for improved interpretation of clinical genetic information, but will also be employed as useful models to conduct therapeutic screens. In vivo data may also be compared to in silico computational predictions from sources such as PolyPhen23, SIFT24, SNPs&GO25, or MutPred26 to show concordance. Note that in a previous study of prediction databases SNPs&GO and MutPred were found to be the most accurate, with accuracies reaching only 0.82 and 0.81, respectively.27
Although we have outlined the robustness of these methods for a subset of pediatric anatomical defects (Table 2, Figure 3), certain phenotypes are less tractable by these methods. Some exceptions notwithstanding, there are three main classes of disorders not amenable to our protocol. Adult-onset disorders (such as Parkinson’s disease) represent a challenge to model in an embryonic system. Slow progression degenerative phenotypes (such as frontotemporal dementia) may require more time than the seven dpf window of MO activity to produce a phenotype. Other gene knockdown technologies such as RNAi and siRNA are available to interfere with or degrade the gene target, but it has been shown that none are as specific, stable, non-toxic, or long-lasting as MOs28, thus also limiting the timeframe for phenotyping. Third, some vertebrate structures, such as the mammalian lung, do not have a sufficiently orthologous structure in the zebrafish embryo. We have also provided a suggested contingency plan for investigation of those cases in which WT human mRNA injection leads to a phenotype, though we caution this is an unusual and undesirable situation.
Certain disease phenotypes may then require a greater degree of abstraction and surrogacy. It is possible that gene function has diverged sufficiently to weaken the phenotypic similarity between model and true phenotype, or that zebrafish physiology inherently complicates the effects of the induced disease. In such cases, we suggest further dissection of the produced phenotype prior to dismissal. We have produced some successful examples in which problematic phenotypes for this assay have been modeled in zebrafish embryos. For example, mutations in TCF8, a gene associated with Fuchs corneal dystrophy (FCD), were assayed using our protocol by using gastrulation defects as a surrogate phenotypic readout based on the known roles of this transcript in early development.29 In other instances, such as adult-onset muscular dystrophy caused by mutations in DNAJB6, we were able to generate myofiber phenotypes in 5dpf embryos despite the fact that humans are bereft of appreciable muscle pathology in their first three to four decades of life.19
In addition to the transient mutant models presented here, others have also taken advantage of this transient system to model human disease in a variety of body systems. In one example, retinitis pigmentosa was modeled in zebrafish by the knockdown of the gene RP2, resulting in retinal cell death and decreased retinal lamination. Rescue with human wild-type mRNA resulted in development of all three layers of retinal lamination, while four out of five mutant mRNAs failed to rescue.30 Although this model of a human sensory disorder is based on a morphological phenotype, it is also possible to assay response to stimuli such as acoustic startle or prepulse inhibition.47
Recently, a zebrafish model was used to investigate Alzheimer disease pathogenesis through amyloid precursor protein.31 The authors showed that gene knockdown caused impaired axonal outgrowth of motor neurons, which could be rescued with human mRNA. This model has proven to be especially informative, as mouse models display only subtle phenotypes (single knockdown) or postnatal lethality (double knockdown). The ability to evaluate the zebrafish embryos in vivo throughout development helped to discern the pathogenic effect of reduced amyloid precursor protein, as well as provided direct evidence that the protein requires both an extracellular and intracellular domain for proper function. Other notable models include that of additional muscular dystrophies32, Diamond Blackfan anemia33, Axenfeld-Reiger syndrome (ocular and craniofacial development)34, inflammatory bowel disease35 (antibacterial activity), Parkinson disease36 (neuron and locomotion loss) , and seizure37(hydrocephalus and hyperactivity).
More common are mutant zebrafish lines that have been shown also to recapitulate a human disease phenotype. Reviewed in 1,38, models include leukemia, melanoma, dilated cardiomyopathies, Duchenne muscular dystrophy, and many others.
The authors have nothing to disclose.
We acknowledge support from a Duke University Dean’s Summer Research Fellowship (A.N.), American Heart Association (AHA) fellowship 11POST7160006 (C.G.), National Institutes of Health (NIH) grants R01-EY021872 from the National Eye Institute (E.E.D.), R01HD04260 from the National Institute of Child Health and Development (N.K.), R01DK072301 and R01DK075972 from the National Institute of Diabetes Digestive and Kidney Disorders (N.K.), and the European Union (Funded by EU 7th FP under GA nr. 241955, project SYSCILIA; E.E.D., N.K.) NK is a Distinguished Jean and George W. Brumley Professor.
Reagent | |||
Phusion High-Fidelity DNA Polymerase | NEB | M0530S, M0530L | |
DpnI restriction endonuclease | NEB | R0176L, R0176S | |
Max Efficiency DH5α competent cells | Invitrogen | 18258-012 | |
Big Dye Terminator | Applied Biosystems | 4337455 | |
mMESSAGE mMACHINE Kit | Invitrogen | AM1340, AM1344, AM1348 | |
Morpholino | Gene-Tools | n/a | |
1-phenyl-2-thiourea (PTU) | Sigma Aldrich | P7629 | Prepare as 0.003% PTU in embryo media |
Paraformaldehyde (PFA) | Sigma Aldrich | P6148 | For embryos that must be fixed prior to phenotyping, prepare as 4% |
Tricaine methane sulfonate | Western Chemical | N/A | For anesthetization and euthanasia |
Equipment | |||
PTC-225 Tetrad Thermal Cycler | BioRad | Any equivalent thermal cycler | |
Nano Drop 2000 spectrophotometer | Thermo Scientific | ||
SMZ 745T Stereomicroscope | Nikon | ||
AZ100 Stereomicroscope | Nikon | ||
DS Fi1 Digital Camera | Nikon | For color/fluorescent imaging | |
DS QiMC Digital Camera | Nikon | For black/white imaging | |
Advanced Resarch 3.2 Imaging Software | NIS- Elements |