Knockdowns versus knockouts: why the phenotypic difference?
University of Bristol, BSc Biochemistry Literary Project, 2015–16
Supervisor: Dr. David Stephens
This project is my own work except where indicated. All text, figures, tables, data or results which are not my own work are indicated and the sources acknowledged.
The antisense knockdown technology, the morpholino oligomer (MO), has been a staple of investigation into development in the model organism zebrafish (Danio rerio) and others such as the African clawed frog (Xenopus laevis and Xenopus tropicalis) for over 15 years. Despite concerns since their inception over off-target effects, MO studies in which these concerns have been minimised with careful controls as recommended by previous reviews have produced a wealth of phenotypic information regarding various gene knockdowns (morphants). However, concerns have again resurfaced in the last few years, which have gestured especially to the rising volume of data displaying the failure of many genetic knockouts (generated using new technologies such as zinc-finger nucleases (ZFN), clustered regularly-interspaced short palindromic repeats (CRISPR) and transcription activator-like effector nucleases (TALEN)) to phenocopy the knockdown phenotype generated by MO studies. Advocates of such concerns have declared such mutants to be the ‘gold standard’ of research into gene function, and called for non-phenocopying MOs to be discarded as false examples of a loss of function (LoF). In this review we critically reevaluate the arguments used by the advocates of such concerns in the light of the emerging fields of developmental plasticity and epigenetics, in order to both suggest possibilities for the reasons behind the morphant-mutant phenotypic incongruence and justify the continued use of MOs as a useful technology in this field to further probe the mechanisms of this hitherto largely unexplained phenomenon particularly.
List of abbreviations
CRISPR, clustered regularly-interspaced short palindromic repeats; CRISPRi, CRISPR interference; LoF, loss-of-function; MO, morpholino oligomer; p53, cellular tumor antigen p53; RNAi, RNA interference; siRNA, short interfering RNA; TALEN, transcription activator-like effector nuclease; ZFN, zinc-finger nuclease.
Antisense interference technologies first began being used within developmental biology in the late 1980s, and were mainly in the form of short (18–22 mers) lengths of DNA that targeted specific mRNAs for degradation by RNase H and nucleases by forming DNA-RNA hybrids . The weaknesses of antisense DNA as an interference tool soon became apparent, as off-target effects and toxicities were widespread , and they also could not reliably produce a loss-of-function (LoF) phenotype due to the degraded mRNA transcripts being continually replenished [3, 4]. By the late 1990s, Gene Tools, LLC  had begun large-scale manufacture of the next generation of antisense oligomer technology: morpholino oligomers (MOs). MOs differ from DNA and RNA in that morpholine rings replace (deoxy)ribose rings, and phosphorodiamidate linkages replace phosphorodiester linkages [Fig. 1] (see Summerton & Weller (1997)  for a more detailed description of MO structure and properties). These structural properties bolster the stability of MOs, making them more resistant to degradation by nucleases . In particular, an enormous step forward was enabled by the non-charge-carrying backbone (compared to the negative charge of the sugar-phosphate backbones of DNA and RNA). This lack of charge protected MOs from non-specific (and potentially toxic) interactions with other cellular components . They are usually ~25-mers, and are engineered to be a precise antisense match to a region of a gene. Coming onto the scene, MOs broadened the developmental biology toolbox with the ability to block translation (by targeting regions around the first ATG), stop correct splicing of precursor mRNA (by targeting a slice donor or acceptor site), and interfere with transcription-regulating microRNAs, all using the same technology [3, 8, 9]. Additionally, spatiotemporal control over MO activity soon became available by use of an MO caged with a photolabile group [10, 11] [Fig. 2].
Currently, however, use of morpholino oligomers, along with other antisense knockdown technologies (such as short interfering RNA (siRNA)), is under more scrutiny than ever before, and debate has flared up on the topic in such high-impact journals Development and Developmental Cell, as well as the dedicated journal Zebrafish, over the course of 2014 and 2015 discussing the growing number of CRISPR (clustered regularly-interspaced short palindromic repeats) or TALEN (transcription activator-like effector nuclease) generated mutants which do not recapitulate the phenotypes observed in the corresponding morphant (MO-mediated knockdown organism) [12–17]. Particularly, the scathing paper by Kok et al. (2015)  has put many MO-derived data in the defence stand, with the result of a real and looming worry that such data are likely to face an extremely tough challenge in convincing reviewers of their worth from here onwards [15–18].
Here we argue that, while the limitations of MOs are well-documented, the incongruence between MO-mediated knockdowns and TALEN- or CRISPR-mediated knockouts ought to be the least condemning issue in this sense, and that this should be spurring more rather than less interest in the use of MOs as an investigative tool in genetics (as well as — even perhaps with primacy — the infant and yawning field of epigenetics). Imprecise targeting, p53-mediated apoptosis upregulation , the variability in effects depending on the MO concentration administered , and the inability to simply assess data from multiple MO studies due to the lack of standard procedure  are all serious problems in the field, but the concluding recommendations made by Kok et al. for ‘the need for a genetic approach as the definitive determination of gene function’ and ‘broader community-wide and editorial guidelines that require an observed MO-induced phenotype to be validated in embryos bearing mutations in the same gene, after which a MO could then be reliably applied for subsequent functional studies’  strikes us as both premature and foolish, for reasons which we will develop in this review. As Dr. Jon D. Moulton makes clear in his account of the Strategic Conference of Zebrafish Investigators meeting in January of 2015 at Asilomar in Pacific Grove, California, comments by Dr. Nathan Lawson (which Kok et al. would appear to agree with) advocating for the use exclusively of “validated” (CRISPR-phenocopying) MOs have caused something of an outcry within the zebrafish community, who ‘admonished Dr. Lawson’s use of words like “rigor” and “gold standard” applied to CRISPRs, asserting that these terms were inflammatory and prejudicial and could inappropriately influence reviewers and study section leaders’ . But, while these polemics are all very well, the scientific case for continued use of MOs even where they do not phenocopy a CRISPR/TALEN/ZFN knockout must also be built. We intend to do this here by reviewing the existing data through a fresh lens — one that takes the kind of dialectical materialism-inspired  developmental plasticity described by Rose (2005)  as central to (re)interpreting the entire problematic of genetics and epigenetics, not least the morphant versus mutant controversy now strangling this particular research community. Aspects of the issue to which we pay particular attention in this review will include the off-target effects of MOs, the lack of widely-implemented standard procedure or controls within the MO-using zebrafish community, and ontogenic compensation mechanisms within the organism.
Why the phenotypic difference? This is a question with answers on two levels.
The first level: experiment
As already mentioned, there are non-specific effects of MOs which have already been characterised. Within the zebrafish example, narrowing of the midbrain and hindbrain of the embryo is reliably seen following injection of MOs, but not with the corresponding mutants. Ekker & Larson (2001)  first identified this effect as non-specific, and it has since been shown by Robu et al. (2007)  to be due to transcriptional upregulation of cellular tumor antigen p53 and resultant p53-induced neural apoptosis, as well as a resultant upregulation of downstream factors such as p21. This is also found as a result of the use of other antisense interference technologies, such as gripNAs and siRNAs, and with MOs is observed with both translation blockers and splice-site-targeting oligos. The mechanism behind p53 being upregulated following treatment with MOs (even when the target gene is not involved in cell survival) so far remains unknown, but some details are available. For example, the level of p53 activation is dependent on the MO sequence, as MOs with different sequences but the same gene target affect p53 activation to different degrees (although the reasons for this are still up for debate, as Robu et al. report that MOs with this off-target effect “do not exhibit any overt primary sequence similarity to repeated elements such as rRNA genes or the zebrafish mitochondrial genome”). However, the extent to which this problem can be seen as a condemnation of MO use is limited. It has been demonstrated that coinjection of a p53 MO with the problematic MO in question attenuates the neural death induced by the latter. Moreover, this also alleviates the neural death induced by gripNAs, perhaps offering a general antidote to the off-target effects associated with antisense knockdown technologies. As prescribed by Robu et al., a good tool for this purpose should be “effective, innocuous, and specific”, and indeed coinjection with a p53 MO would appear to tick all three of these boxes. It does not carry with it any significant defects of its own (due to it not being necessary for ordinary development in either fish or mammals), and does not affect the efficacy of the inhibition of the gene or splice site targeted by the other MO. This is also an essential experimental step to take if one is to differentiate between off-target neural death due to p53 upregulation and neural death in similar areas due to specific, on-target effects, such as are seen when cytochrome C activity is reduced via MO-mediated interference . The only limitation to the use of a p53 MO as an adjunct to MO experiments seems to be that it would make it difficult to use MOs to study developmental processes in which p53-mediated apoptosis is involved.
A key piece of evidence in Kok et al.’s attempted onslaught on antisense knockdown technologies is their appraisal of a megamind morphant. megamind is a long non-coding RNA (sometimes referred to as linc:birc6), knockdown of which has been shown to cause hydrocephaly in zebrafish (Ulitsky et al., 2011) . Kok et al. produced a megamind null mutant (megamindum209), the deletion in which also deleted the sequences targeted by the MOs used by Ulitsky et al. (2011). However, upon injection of this mutant with the aforementioned MO, hydrocephaly was still observed — a result that ought not to occur if the morphant phenotype were only due to on-target effects. As a brief aside, it is worth noting that this (megamindun209) is the only example cited of an MO-induced phenotype in zebrafish embryos that lack the MO-targeted part of the gene. They therefore argue that the conclusions of Ulitsky et al. are incorrect, and the observations of the latter are not due to megamind knockdown but to off-target effects, and they extend this hypothesis to a number of other existing morphants for which they found poor phenotypic correlation with their ZFN-, TALEN- and CRISPR/Cas9-generated mutants.
On brief inspection this appraisal may seem fair, but a less superficial evaluation of the data reveals that there are large differences between the experimental procedures of the two studies, not least the dosage of the MO used. The dosage used by Kok et al. to produce the phenotype in the megamind null mutant was 20 ng/embryo, which is four times greater than the dosage used by Ulitsky et al. Additionally, Lin et al.  report a non-morphological behavioural phenotype (along with the disappearance of the wild-spliced RT-PCR gel band) following injection of either a splice-targeted or conserved-region-targeted MO at only 1 ng/embryo. This clearly demonstrates an effect on RNA processing caused by an MO used at an appropriate dose. Kok et al. do not even consider testing a series of lower dosages, which is worth bearing in mind when weighing up their conclusions, as off-target effects are particularly likely at high concentrations and overdosing is likely to be a large factor in bringing about the discrepancies noted above. This was something recognised even over eight years ago(!) in the review by Eisen & Smith (2008) , which set out guidelines for MO use in an attempt to find solutions to these same problems. In the review, they recommend generating a dose-response curve (injecting a range of concentrations) for at least two non-overlapping MOs targeting the same gene, recording the resultant phenotype of each trial, and then finally coinjecting the two MOs at respective concentrations which, individually, would not elicit a phenotype, so as to be sure that the phenotype observed in the coinjection (matching the phenotype observed with the single-MO morphants) is due to on-target effects. It is also of importance to carefully control the injection volumes in each trial and to repeat each experiment at least three times. This repetition is to reveal any inconsistency that may be due, for example, to the difficulty in injecting precise and reproducible volumes of MOs (and indeed at precise and reproducible spatiotemporal coordinates within the embryo and its development), particularly in small embryos. Variations in injected MO concentrations can produce markedly different phenotypes. For example, the observations (and hence conclusions) made by Hetheridge et al. (2012) and Phng et al. (2015), though regarding the role of the same formin-like protein fmnl3 in zebrafish development, were vastly different [26, 27]. Upon MO-mediated translation-blocking of fmnl3, Hetheridge et al. (2012) observed major defects in ISV formation, whereas Phng et al. (2015) using the same technique observed no such thing. The fact that Hetheridge et al. (2012) found that their phenotype was rescued upon injection of the human gene implies that the defects they observed were truly as a result of fmnl3 knockdown, so the incongruence between the two sets of results is puzzling. However, as Stainier et al. (2015)  has commented, an explanation may lie in the fact that Phng et al. (2015) injected at 10 ng/embryo whereas an injection concentration was not even reported by Hetheridge et al. (2012). The difference in phenotype is likely due to a difference in administered MO concentrations. In this vein, we advise, following Eisen & Smith (2008), that accurate and detailed information regarding the experimental protocol followed must be published as part of any report that makes use of MOs, so that reliable comparisons between data can be made. In addition, such dose-dependent effects of MOs can even be of use to investigators, as it allows graduated knockdown of a gene. As in the case of graded MO-mediated knockdown of activin B which elucidated its role as a morphogen , these intermediary semi-knockdown phenotypes can provide more rich information than could be revealed by a genetically-achieved blanket LoF. Frustratingly, Kok et al. did not carry out p53 MO coinjection either, which would have limited off-target effects even at such a high MO dosage as was used, despite this also having been recommended in the guidelines set out by Eisen & Smith (2008).
Yet another limitation of the data generated by Kok et al. is that for only as few as 4 of the 48 mutants was the mutant thoroughly demonstrated to be a null. This raises the question that a proportion of their mutants may be hypomorphs, as admitted as true fact by the authors for one of their lines, and this could be one reason for the observed stronger phenotypes in the corresponding morphants. In their paper, Kok et al. themselves even mention several reasons why frameshift mutations generated by making severe lesions in exonic DNA might not give rise to a LoF allele, such as activation of cryptic splice sites, or exon skipping. Further possibilities include lesion-workaround mechanisms such as RNA editing and ribosome frameshifting , and also initiation of translation at a non-AUG codon . We would recommend, in addition to the standards proposed by Eisen & Smith (2008), that when using mutant lines as a complementary approach to morphant analysis, robust testing is carried out to ensure that any RNA or protein relating to the “knocked out” gene are not being produced.
These above factors constitute one level of an answer to the question of phenotypic difference, that of ensuring that the appropriate experimental controls are employed so as to standardise an experimental procedure that is not well understood. From here, results can be more usefully evaluated, analysed and compared.
The second level: hypothesis
To refer back yet again to Eisen & Smith (2008) , “it is also important in interpreting experiments that make use of antisense oligonucleotides to consider the mechanisms by which the reagent of choice acts.” This can be supplemented by an insightful quote from a commentary on the morphant versus mutant question on the blog of BioMed Central Biology: “both methods might be reliable (repeatable) in the results observed in this case; the trick is to try to be sure about what question you want to ask, and what question a particular method is actually going to help answer” . This is a crucial point. Assuming a mutant is a null, expression is halted at the level of DNA and the gene does not persist, whereas with MOs (as well as siRNAs, gripNAs etc.), expression is halted at the level of mRNA and both the gene and the mRNA persist (i.e. the mRNA is not degraded as it is with RNase H-dependent RNA interference (RNAi) techniques). It is entirely plausible then, that if mRNA has some function(s) surplus to its role as the substrate of translation, this may not be perturbed by antisense interference, while expression of the gene at the protein level is. Indeed, this has been found to be the case regarding the knockdown of the mRNA encoding the vegetal cortex protein VegT in X. laevis . It would appear that the VegT mRNA has a role in the localisation of another vegetal cortex RNA, Vg1, as the knockdown of VegT results in the release of Vg1 mRNA from the vegetal cortex, along with a reduction in levels of the Vg1 protein. The mRNAs encoding Bicaudal C and Wnt11 also lose their localisation within the vegetal cortex under these conditions. Crucially, this does not occur when the knockdown technique used is a translation-inhibiting MO, i.e. when the VegT mRNA is not degraded. From these observations, we can learn that MOs may potentially be a useful tool for studying protein function without inadvertently interfering (as a knockout would) with other poorly-understood mechanisms that certain DNAs or mRNAs may take part in. However, we must also be aware that, as the embryo attempts to compensate for loss of the protein, mRNA levels may actually increase, and this will have its own consequences that must be taken into consideration.
The difference between the modes of action of MO-mediated knockdown and a genetic knockout must also be paid specific attention when considering the time point(s) during development at which they act. Knockout embryos are bred — they lack the gene in question from the beginning of their development — whereas MO injections knock a particular gene product down in a comparatively acute manner. Particularly when analysing this difference through the lens of developmental plasticity and ontogeny, the relevance of this difference becomes clear. The concept of developmental plasticity, as defined by Rose (2004) , is “the capacity of a living system to adapt to experience and environmental contingencies, and to compensate for deficiencies.” Rose also notes that “[t]his capacity is augmented by the functional redundancy present in all organisms. Redundancy assists stability; it means that there may be many alternative routes that the cell and the organism can adopt during development which can lead to an essentially identical end-point.” An appropriate metaphor for this concept would be that of graceful degradation  within the field of computer modelling — the phenomenon of a large and complex system remaining stable and functioning regularly when a small number of components are removed. The biochemical geneticist Henry Kacser who formulated the mathematical proof for this property, once referred to it (in his case, operating within the study of metabolism) as ‘molecular democracy’ , where stability does not reside in the individual protein components but in the web of connections — the system, or structure if you will — itself. In this model, the components are secondary to their position within the structure, and the component occupying a particular node within the structure is contingent on the dialectic between genetic material and the environment at a given moment during development [20, 21]. In terms of the morphant versus mutant problem, this concept lends itself to the hypothesis of compensation which has been proposed by several morphant-mutant comparative studies [35–38], reviews , and is beginning to attract attention in popular science media too [31, 39, 40].
Compensation is the upregulation of the synthesis and/or activity of a protein to remedy a blockade of a biochemical process caused by a deficiency in one of its key factors. This could function via the upregulated protein directly replacing the deficient one by merit of them sharing certain properties required for the function needing to be fulfilled, or it could also function by means of upregulating synthesis and/or activity of other proteins so as to divert the process via a different route using these other, available, proteins, the end result being similar to or even identical to the wild-type phenotype. For example, in mice, upregulation of either of the muscle membrane-associated proteins utrophin or dystrophin compensates for the genetic loss of the other, with the effect of a mild phenotype compared to a double knockout . A second mouse study found a similar functional redundancy between the histone deacetylases HDAC1 and HDAC2, which when deleted independently result in no observable phenotype, but when deleted in combination give rise to abnormal chromatin structure, DNA damage and even embryonic death. Most importantly, HDAC1 is upregulated in the absence of HDAC2 . The studies which have identified compensation as a likely cause of a discrepancy between morphant and mutant phenotypes have proposed that it occurs in the mutants, but not in the morphants [35–38]. What justifies this proposition? Again it comes down to the difference between the time points of an embryo’s development at which each technique functions, and the level (DNA, mRNA or protein) at which interference occurs, as noted above. A knockout embryo is without a particular gene or its product from its inception and remains this way, and so has an opportunity to, if it can, compensate from the earliest possible stage of its development onwards and hence is more likely to develop a successful compensation strategy. An MO-mediated knockdown embryo, on the other hand, experiences an acute disruption to a gene product anywhere up to the 8 cell stage, and so may not be able to compensate sufficiently due to it having already committed to particular developmental pathways. This is related to the fact that a knockout embryo also does not possess the DNA or the mRNA corresponding to the missing gene product, while a MO-mediated knockdown embryo, as already discussed, has both, which may have consequences for signalling mechanisms which might otherwise induce compensation. Indeed, Rossi et al. (2015), one of the first papers that proposed compensation as a reason for the poor phenotypic correlation between morphants and mutants, makes the following observation while investigating compensatory mechanisms active between the zebrafish vascular endothelial growth factors Aa (Vegfaa) and Ab (Vegfab): “qPCR analysis showed that vegfab, a paralogue of vegfaa, was upregulated in mutants but not morphants. Additionally, blocking Vegfaa function using a dominant negative approach also failed to trigger vegfab upregulation, placing the signal triggering compensation upstream of protein function” [emphasis added] . This would appear to implicate mRNA or perhaps even the DNA itself in a signalling mechanism to trigger compensation, and therefore would go some way to offering an explanation for why this compensation is not seen in morphants, as the persistence of the DNA and mRNA conceals the requirement for compensation and so the embryo commits to an incomplete developmental path, resulting in an aberrant phenotype.
While investigation into compensation and plasticity is relatively young, already a number of examples have been discovered, many through comparisons of MO-mediated knockdowns and genetic mutants. Rossi et al. (2015), in addition to vegfaa/vegfab, also investigated the zebrafish endothelial extracellular matrix gene egfl7. Consistent with the hypothesis of compensation and plasticity laid out above, they found that egfl7 mutants do not show an obvious phenotype whereas egfl7 morphants show serious vascular defects. What makes this study particularly useful is that they also tested the sensitivity of the mutants to Egfl7 knockdown and found that they were less sensitive than wild type zebrafish, which argues against both residual protein function within the mutants and off-target effects of the MO as determining factors. They identified a set of genes and proteins which were upregulated in the mutants but not in the morphants, including several other extracellular matrix genes that can rescue eglf7 morphants, implicating them in compensatory functions. Interestingly, these genes were not found to be upregulated in zebrafish subjected to egfl7 CRISPRi, a technique that blocks transcript elongation, and these zebrafish showed similar phenotypes to the morphants. Again, this would appear to place the compensatory mechanisms upstream of protein function, and even upstream of transcription. Hu et al. (2016) found a further example in the two liver-enriched gene 1 (leg1) homologues leg1a and leg1b . Leg1 is a glycosylated secretory protein that makes up part of an anti-stress pathway that protects liver development in zebrafish. Using MOs, leg1a had previously been shown to be required for liver bud outgrowth , but Hu et al. found that zygotic leg1azju1 mutants do not show the morphant phenotype and additionally the maternal-zygotic leg1azju1 mutant show a milder phenotype than the MO-mediated knockdown. After eliminating the possibility of off-target effects of MOs causing the discrepancy, they found that leg1b expression is heightened in the leg1azju1 mutant compared to the wild-type, implying that upregulation of leg1b expression happens as a result of compensation by the organism for deleterious mutations in leg1a. Novodvorsky et al. (2015) summarise that such mechanisms may account for the fact, uncovered by a large-scale mutagenesis study in zebrafish, that only 6% of disruptive mutations in protein-coding genes produce a phenotype discernible from the wild-type. They highlight that a gene’s role in development is therefore ironically not easily solved within a genetic paradigm, as even mutations which are clearly disruptive to a gene do not often result in the loss of the function commonly attributed to that gene .
As Rossi et al. (2015) conclude, compensatory networks seem to be active strictly against deleterious mutations rather than any genetic deviation , thus allowing for variation when genetic lesions are not deleterious. But even when it is active, such plasticity is not, of course, limitless. Every gene (and indeed every phenotype) can only tolerate a certain degree of stress before the organism becomes overwhelmed and things fall apart, resulting in an aberrant phenotype or even death . Revisiting the web analogy, certain nodes on the web may be more vulnerable when the protein occupying it is knocked down, perhaps due to the protein’s large interactome (and hence its precise involvement in many processes) which increases the difficulty of compensation for its loss. Identifying such proteins may identify the most effective targets for gene therapies. The occasions on which morphants do phenocopy mutants (or perhaps it would be more appropriate here to speak of the mutants phenocopying the morphants) are examples of the mutant being unable to sufficiently compensate for the knockout of a gene and the loss of its products, resulting in a mutant with the same aberrant phenotype as observed in the morphant, and hence may provide useful insight into how dispensable (or not so) particular genes are, so as to aid the development of more effective gene therapies that bolster these genes particularly.
Plasticity, and in addition the epigenetic mechanisms that facilitate it, are relatively recent areas of biochemistry to come under the lens, and so at this time very little is known about them. However, by investigating circumstances in which these processes can be interfered with, such as in the cases of MO-mediated knockdowns which do not phenocopy mutants, we can start to build up a picture of how the mechanisms that allow a gene to ‘know’ that another gene has been deleted and then compensate for the loss function. This is the primary reason why MOs as an interference technique must not be thrown out of the biochemist’s toolbox, and the discrepancies between morphants and mutants be paid attention to in order to identify modifier genes . No matter how blinding it may be to the current genetic reductionist paradigm, to close curtains on one of the few windows through which light is coming is to leave us in the dark.
 Corey, D. R. & Abrams, J. M. (2001) Morpholino antisense oligonucleotides: tools for investigating vertebrate development. Genome Biol., 2, reviews 1015.1–1015.3
 Walder, R. Y. & Walder, J. A. (1988) Role of RNase H in hybrid-arrested translation by antisense oligonucleotides. PNAS, 85, 5011–5015
 Summerton, J. (1999) Morpholino antisense oligomers: the case for an RNase H-independent structural type. Biochim. Biophys. Acta, 1489, 141–158
 Eisen, J. S. & Smith, J. C. (2008) Controlling morpholino experiments: don’t stop making antisense. Development, 135, 1735–1743
 Gene Tools, LLC. Available at: <https://www.gene-tools.com> Accessed 4 January 2016
 Summerton, J. & Weller, D. (1997) Morpholino antisense oligomers: design, preparation, and properties. Antisense Nucleic Acid Drug Dev., 7, 187–95
 Hudziak, R. M., Barofsky, E., Barofsky, D. F., Weller, D. L., Huang, S. B. & Weller, D. D. (1996) Resistance of morpholino phosphorodiamidate oligomers to enzymatic degradation. Antisense Nucleic Acid Drug Dev., 6, 267–272
 Draper, B. W., Morcos, P. A. & Kimmel, C. B. (2001) Inhibition of zebrafish fgf8 pre-mRNA splicing with morpholino oligomers: a quantifiable method for gene knockdown. Genesis, 30, 154–156
 Kloosterman, W. P., Lagendijk, A. K., Ketting, R. F., Moulton, J. D. & Plasterk, R. H. A. (2007) Targeted inhibition of miRNA maturation with morpholinos reveals a role for miR-375 in pancreatic islet development. PLoS Biol., 5, e203
 Shestopalov, I. A., Sinha, S. & Chen, J. K. (2007) Light-controlled gene silencing in zebrafish embryos. Nat. Chem. Biol., 3, 650–651
 Wang, Y., Wu, L., Wang, P., Lv, C., Yang, Z. & Tang, X. (2012) Manipulation of gene expression in zebrafish using caged circular morpholino oligomers. Nucleic Acids Res., 40, 11155–11162
 Kok, F. O., Shin, M., Ni, C.-W., Gupta, A., Grosse, A. S., van Impel, A., Kirchmaier, B. C., Peterson-Maduro, J., Kourkoulis, G., Male, I., DeSantis, D. F., Sheppard-Tindell, S., Ebarasi, L., Betsholtz, C., Schulte-Merker, S., Wolfe, S. A. & Lawson, N. D. (2015) Reverse genetic screening reveals poor correlation between Morpholino-induced and mutant phenotypes in zebrafish. Dev. Cell, 32, 97–108
 Lawson, N. D. (2016) Reverse Genetics in Zebrafish: Mutants, Morphants, and Moving Forward. Trends Cell Biol., 26, 77–79
 Schulte-Merker, S. & Stainier, D. Y. R. (2015) Out with the old, in with the new: reassessing morpholino knockdowns in light of genome editing technology. Development, 141, 3103–3104
 Stainier, D. Y. R., Kontarakis, Z. & Rossi, A. (2015) Making sense of anti-sense data. Dev. Cell, 32, 7–8
 Blum, M., De Robertis, E. M., Wallingford, J. B. & Niehrs, C. (2015) Morpholinos: Antisense and Sensibility. Dev. Cell, 35, 145–149
 Morcos, P. A., Vincent, A. C. & Moulton, J. D. (2015) Gene Editing Versus Morphants. Zebrafish, 12, 319
 Moulton, J. D. (2014) Morpholinos, mutants and phenocopying. Gene Tools LLC, [blog] 11 February. Available at: <http://www.gene-tools.com/content/morpholinos-mutants-and-phenocopying> Accessed 22 February 2016
 Robu, M. E., Larson, J. D., Nasevicius, A., Beiraghi, S., Brenner, C., Farber, S. A. & Ekker, S. C. (2007) p53 Actication by Knockdown Technologies. PLoS Genet., 3, e78
 Lewontin, R. & Levins, R. (2009) The Dialectical Biologist. Delhi: Aakar Books
 Rose, S. P. R. (2005) Lifelines: Life Beyond the Gene. London: Vintage
 Ekker, S. C. & Larson, J. D. (2001) Morphant technology in model developmental systems. Genesis, 30, 89–93
 Baden, K., Murray, J., Capaldi, R. A. & Guillemin, K. (2007) Early developmental pathology due to cytochrome c oxidase deficiency is revealed by a new zebrafish model. J. Biol. Chem., 282, 34839–34849
 Ulitsky, I., Shkumatava, A., Jan, C. H., Sive, H. & Bartel, D. P. (2011) Conserved function of lincRNAs in vertebrate embryonic development despite rapid sequence evolution. Cell, 147, 1537–1550
 Lin, N., Chang, K.-Y., Li, Z., Gates, K., Rana, Z. A., Dang, J., Zhang, D., Han, T., Yang, C.-S., Cunningham, T, J., Head, S. R., Duester, G., Dong, P. D. & Rana, T. M. (2014) An evolutionarily conserved long noncoding RNA TUNA controls pluripotency and neural lineage commitment. Mol. Cell, 53, 1005–1019
 Hetheridge, C., Scott, A. N., Swain, R. K., Copeland, J. W., Higgs, H. N., Bicknell, R. & Mellor, H. (2012) The formin FMNL3 is a cytoskeletal regulator of angiogenesis. J. Cell Sci., 125, 1420–1428
 Phng, L.-K., Gebala, V., Bentley, K., Philippides, A., Wacker, A., Mathivet, T., Sauteur, L., Stanchi, F., Belting, H.-G., Affolter, M. & Gerhardt, H. (2015) Formin-Mediated Actin Polymerization at Endothelial Junctions Is Required for Vessel Lumen Formation and Stabilization. Dev. Cell, 32, 123–132
 Piepenburg, O., Grimmer, D., Williams, P. H. & Smith, J. C. (2004) Activin redux: specification of mesodermal pattern in Xenopus by graded concentrations of endogenous activin B. Development, 131, 4977–4986
 Pan, T. (2013) Adaptive translation as a mechanism of stress response and adaptation. Annu. Rev. Genet., 47, 121–137
 Kozak, M. (2002) Pushing the limits of the scanning mechanism for initiation of translation. Gene, 16, 1–34
 Bell, G. (2015) Reproducibility and morpholinos: different methods, different answers. On Biology, [blog] 23 July. Available at: <http://blogs.biomedcentral.com/on-biology/2015/07/23/reproducibility-morpholinos-different-methods-different-answers> Accessed 9 February 2016
 Heasman, J., Wessely, O., Langland, R., Craig, E. J. & Kessler, D. S. (2001) Vegetal localization of maternal mRNAs is disrupted by VegT depletion. Dev. Biol., 240, 377–386
 Denning, P. J. (1976) Fault tolerant operating systems. ACM Computing Surveys, 8, 359–389
 Kacser, H. & Burns, J. A. (1979) Molecular democracy: who shares the controls? Biochem. Soc. Trans., 7, 1149–1160
 Rossi, A., Kontarakis, Z., Gerri, C., Nolte, H., Hölper, S., Krüger, M. & Stainier, D. Y. R. (2015) Genetic compensation induced by deleterious mutations but not gene knockdowns. Nature, 524, 230–233
 Hu, M., Bai, Y., Zhang, C., Liu, F., Cui, Z., Chen, J. & Peng, J. (2016) Liver-Enriched Gene 1, a Glycosylated Secretory Protein, Binds to FGFR and Mediates an Anti-stress Pathway to Protect Liver Development in Zebrafish. PLoS Genet., 12, e1005881
 Novodvorsky, P., Watson, O., Gray, C., Wilkinson, R. N., Reeve, S., Smythe, C., Beniston, R., Plant, K., Maguire, R., Rothman, A. M. K., Elworthy, S., van Eeden, F. J. M. & Chico, T. J. A. (2015) klf2ash317 Mutant Zebrafish Do Not Recapitulate Morpholino-Induced Vascular and Haematopoietic Phenotypes. PLoS ONE, 10, e0141611
 Won, M., Ro, H. & Dawid, I. B. (2015) Lnx2 ubiquitin ligase is essential for exocrine cell differentiation in the early zebrafish pancreas. PNAS, 112, 12426–12431
 Heil, M. (2015) Gene knockout: loss of a gene can be compensated by another gene. Max Planck Institute for Heart and Lung Research, [online] 21 July. Available at: <https://www.mpg.de/9331068/knockout-knockdown> Accessed 12 March 2016
 Hesman Saey, T. (2015) Mutation-disease link masked in zebrafish. ScienceNews, [online] 13 July. Available at: <https://www.sciencenews.org/article/mutation-disease-link-masked-zebrafish> Accessed 2 February 2016
 Deconinck, A. E., Rafael, J. A., Skinner, J. A., Brown, S. C., Potter, A. C., Metzinger, L., Watt, D. J., Dickson, J. G., Tinsley, J. M. & Davies, K. E. (1997) Utrophin-dystrophin-deficient mice as a model for Duchenne muscular dystrophy. Cell, 90, 717–727
 Hagelkruys, A., Lagger, S., Krahmer, J., Leopoldi, A., Artaker, M., Pusch, O., Zezula, J., Weissmann, S., Xie, Y., Schöfer, C., Schlederer, M., Brosch, G., Matthias, P., Selfridge, J., Lassmann, H., Knoblich, J. A. & Seiser, C. (2014) A single allele of Hdac2 but not Hdac1 is sufficient for normal mouse brain development in the absence of its paralog. Development, 141, 604–616
 Chang, C., Hu, M., Zhu, Z., Lo, L. J., Chen, J. & Peng, J. (2011) liver-enriched gene 1a and 1b encode novel secretory proteins essential for normal liver development in zebrafish. PLoS One, 6, e22910
Originally published at www.academia.edu.