1. Genetics and Genomics
  2. Neuroscience
Download icon

A simple and effective F0 knockout method for rapid screening of behaviour and other complex phenotypes

  1. François Kroll
  2. Gareth T Powell
  3. Marcus Ghosh
  4. Gaia Gestri
  5. Paride Antinucci
  6. Timothy J Hearn
  7. Hande Tunbak
  8. Sumi Lim
  9. Harvey W Dennis
  10. Joseph M Fernandez
  11. David Whitmore
  12. Elena Dreosti
  13. Stephen W Wilson
  14. Ellen J Hoffman
  15. Jason Rihel  Is a corresponding author
  1. Department of Cell and Developmental Biology, University College London, United Kingdom
  2. Department of Neuroscience, Physiology and Pharmacology, University College London, United Kingdom
  3. Wolfson Institute for Biomedical Research, University College London, United Kingdom
  4. School of Biological Sciences, Faculty of Science, University of Bristol, United Kingdom
  5. Child Study Center, Yale School of Medicine, United States
  6. Department of Molecular and Cell Biology, James Cook University, Australia
  7. Department of Neuroscience, Yale School of Medicine, United States
Research Article
  • Cited 1
  • Views 4,796
  • Annotations
Cite this article as: eLife 2021;10:e59683 doi: 10.7554/eLife.59683

Abstract

Hundreds of human genes are associated with neurological diseases, but translation into tractable biological mechanisms is lagging. Larval zebrafish are an attractive model to investigate genetic contributions to neurological diseases. However, current CRISPR-Cas9 methods are difficult to apply to large genetic screens studying behavioural phenotypes. To facilitate rapid genetic screening, we developed a simple sequencing-free tool to validate gRNAs and a highly effective CRISPR-Cas9 method capable of converting >90% of injected embryos directly into F0 biallelic knockouts. We demonstrate that F0 knockouts reliably recapitulate complex mutant phenotypes, such as altered molecular rhythms of the circadian clock, escape responses to irritants, and multi-parameter day-night locomotor behaviours. The technique is sufficiently robust to knockout multiple genes in the same animal, for example to create the transparent triple knockout crystal fish for imaging. Our F0 knockout method cuts the experimental time from gene to behavioural phenotype in zebrafish from months to one week.

Introduction

Genomic studies in humans are uncovering hundreds of gene variants associated with neurological and psychiatric diseases, such as autism, schizophrenia, and Alzheimer’s disease. To validate these associations, understand disease aetiology, and eventually inform therapeutic strategies, these genetic associations need to be understood in terms of biological mechanisms. A common approach is to mutate candidate disease genes in cultured cells or animal models. The zebrafish (Danio rerio) is an attractive in vivo model for genetic screens in neuroscience (Tang et al., 2020; Thyme et al., 2019). Indeed, more than 75% of disease-associated genes have an orthologue in the zebrafish genome (Howe et al., 2013), optical translucence allows for whole brain imaging (Ahrens et al., 2013), and behavioural phenotypes can be robustly quantified early in development (Rihel et al., 2010). However, the pace at which new genetic associations are being identified currently far outstrips our ability to build a functional understanding in vivo. In zebrafish, a key limiting factor is the time and space needed to generate animals harbouring a mutation in the gene of interest.

Genome editing using CRISPR-Cas9 has revolutionised our ability to generate zebrafish mutant lines (Hwang et al., 2013). The common strategy is to inject a Cas9/guide RNA (gRNA) ribonucleoprotein (RNP) into the single-cell embryo (Sorlien et al., 2018). The gRNA binds to the targeted region of the genome and Cas9 produces a double-strand break. When joining the two ends, DNA repair mechanisms may introduce an indel by inserting or deleting bases in the target locus (Brinkman et al., 2018). Indels often disrupt protein function, either by mutating sequences that encode essential residues or by introducing a frameshift that leads to a premature stop codon and a truncated, non-functional protein. With this tool, virtually any locus in the zebrafish genome can be disrupted. However, homozygous mutants are only obtained after two generations of adult animals, which typically takes four to six months (Sorlien et al., 2018). This bottleneck places substantial constraints on genetic screens in terms of time, costs, and ethical limits on animal numbers.

Genetic screens would be greatly facilitated by reliably generating biallelic knockouts directly in the injected embryos, termed the F0 generation. The main hurdle is to introduce a deleterious mutation in most or all copies of the genome without causing non-specific phenotypic consequences. Since the first applications of CRISPR-Cas9 in vivo (Chang et al., 2013; Hwang et al., 2013), meticulous optimisation of design (Moreno-Mateos et al., 2015), preparation, and delivery of the RNP (Burger et al., 2016) has improved mutagenesis consistently enough to allow the successful use of zebrafish F0 knockouts in screens for visible developmental phenotypes, such as cardiac development (Wu et al., 2018), formation of electrical synapses (Shah et al., 2015), or distribution of microglia (Kuil et al., 2019). In these applications, disrupting a single locus may be adequate as incomplete removal of wild-type alleles does not impair detection of the phenotype. For example, mutants with an overt developmental defect can be identified even in heterogenous populations where some animals are not complete knockouts (Burger et al., 2016). Similarly, certain phenotypes may be manifest in an animal in which only a subset of the cells carry mutant alleles (Kuil et al., 2019; Shah et al., 2015). However, incomplete conversion into null alleles is potentially problematic when studying traits that vary continuously, especially if the spread of phenotypic values is already substantial in wild-type animals, which is regularly the case for behavioural parameters. Animals in the experimental pool retaining variable levels of wild-type alleles will create overlap between the mutant and wild-type distributions of phenotypic values, reducing the likelihood of robustly distinguishing a mutant phenotype. To ensure a high proportion of null alleles, an alternative strategy is to increase the probability of a frameshift by targeting the gene at multiple loci (Hoshijima et al., 2019; Sunagawa et al., 2016; Wu et al., 2018; Zhou et al., 2014; Zuo et al., 2017). Because rounds of DNA breaks and repair usually occur across multiple cell cycles (McKenna et al., 2016), different F0 animals, cells, or copies of the genome can harbour different null alleles. Targeting multiple loci inflates this diversity of alleles, which is perceived as a potential obstacle for rigorously describing complex phenotypes in F0 knockouts, particularly behavioural ones (Teboul et al., 2017).

We present a simple CRISPR-Cas9 method to generate zebrafish F0 knockouts suitable for studying behaviour and other continuous traits. The protocol uses a set of three synthetic gRNAs per gene, combining multi-locus targeting with high mutagenesis at each locus. The method consistently converts > 90% of injected embryos into biallelic knockouts that show fully penetrant pigmentation phenotypes and near complete absence of wild-type alleles in deep sequencing data. In parallel, we developed a quick and cheap PCR-based tool to validate gRNAs whatever the nature of the mutant alleles. The F0 knockout protocol is easily adapted to generate biallelic mutations in up to three genes in individual animals. The populations of F0 knockout animals generated by the method are suitable for quantitative analysis of complex phenotypes, as demonstrated by mutation of a circadian clock component and by meticulous replication of multi-parameter behavioural phenotypes of a genetic model of epilepsy.

Using standard genetic approaches, the gap from gene to behavioural phenotype in zebrafish often takes half a year. Our F0 knockout method enables this in a week. We believe these methodological improvements will greatly facilitate the use of zebrafish to tackle genetic questions in neuroscience, such as those addressing the contributions of disease-associated genes to nervous system development, circuit function, and behaviour.

Results

Three synthetic gRNAs per gene achieve over 90% biallelic knockouts in F0

What are the requirements for a zebrafish F0 knockout method applicable to large genetic screens studying continuous traits? First, it needs to be quick and reliable. Most techniques so far have used in vitro-transcribed gRNAs (Shah et al., 2015; Wu et al., 2018). In vitro transcription often necessitates the substitutions of nucleotides in the 5'-end of gRNAs, and this can hamper mutagenesis by introducing mismatches with the target locus. Commercial synthetic gRNAs circumvent this limitation (Hoshijima et al., 2019). Second, it needs to be readily applicable to any open-reading frame. Some protocols suggest targeting each gene with one or two synthetic gRNAs designed to target essential domains of the encoded protein (Hoshijima et al., 2019). This strategy requires detailed knowledge of each target, which is likely to be lacking in large genetic screens investigating poorly annotated genes. Third, the method must consistently convert most injected embryos into F0 biallelic knockouts, leaving few or no wild-type alleles within each animal. Complete conversion into null alleles may not be a primary requirement for detection of discrete or overt phenotypes but is a priority when studying continuous traits.

To fulfil these criteria, we chose to maximise the probability of introducing a frameshift by optimising mutagenesis at multiple loci over each gene, as in theory this is a universal knockout mechanism (Figure 1A). In a simple theoretical model (Wu et al., 2018) where frameshift is the sole knockout mechanism and the probability of mutation at each target locus is over 80%, targeting the gene at three to four loci is predicted to be sufficient to routinely achieve over 90% biallelic knockout probability (Figure 1B). While targeting extra loci would increase this probability further, minimising the number of unique RNPs injected reduces both costs and potential off-target effects.

Figure 1 with 2 supplements see all
Three synthetic gRNAs per gene achieve over 90% biallelic knockouts in F0.

(A) Schematic of the F0 knockout strategy. Introduction of indels at multiple loci within the target gene leads to frameshift and premature stop codons and/or mutation of essential residues. (B) Simplified theoretical model of biallelic knockout probability as a function of number of targeted loci, assuming frameshift is the sole knockout mechanism. PKO, probability of biallelic knockout; Pmutation, mutation probability (here, 1.00 or 0.80); Pframeshift, probability of frameshift after mutation (0.66); nloci, number of targeted loci. (C–D) (top) Phenotypic penetrance as additional loci in the same gene are targeted. Pictures of the eye at 2 dpf are examples of the scoring method. (bottom) Unviability as percentage of 1-dpf embryos. (E–F) Proportion of alleles harbouring a frameshift mutation if 1, 2, 3, or 4 loci in the same gene were targeted, based on deep sequencing of each targeted locus. Each line corresponds to an individual animal. (G) 2.5-month wild-type and slc24a5 F0 knockout adult fish (n = 41). (H) 2-dpf progeny from slc24a5 F0 adults outcrossed to wild types (n = 283) or incrossed (n = 313). (I) Example of 3-dpf wild type, mab21l2u517 mutant, and mab21l2 F0 embryos (n = 96/100 injected). See also Figure 1—figure supplement 1.

To test the efficacy of multi-locus targeting to generate functional null mutations, we targeted the pigmentation genes slc24a5 and tyr at different numbers of loci and quantified phenotypic penetrance in individual animals. Homozygous null slc24a5 or tyr zebrafish lack eye pigmentation at 2 days post-fertilisation (dpf), while heterozygous and wild-type siblings are pigmented (Kelsh et al., 1996; Lamason et al., 2005). Additionally, Slc24a5 and Tyr act cell-autonomously, so any unpigmented cells within the eye carries a biallelic null mutation. To generate F0 embryos, we injected at the one-cell stage RNPs (Cas9 protein/synthetic crRNA:tracrRNA duplex) targeting one to four loci per gene. To estimate phenotypic penetrance, eye pigmentation was scored at 2 dpf on a scale from 1 (completely devoid of pigment, i.e. fully penetrant) to 5 (dark as wild types, i.e. no penetrance). The larvae were then followed until 5–6 dpf to quantify any reduced viability in the injected animals, reported as the sum of dead or dysmorphic embryos. Targeting slc24a5 with one or two RNPs generated clutches with low phenotypic penetrance, i.e. most larvae appeared wild-type or had patchy eye pigmentation. Conversely, when three RNPs were injected, 95% (55/58) of larvae were totally devoid of eye pigmentation (Figure 1C). Adding a fourth RNP did not increase the penetrance further. In some cases, the phenotypic penetrance was 100%. For example, using just two RNPs to target tyr yielded 59/59 F0 embryos with no eye pigmentation (Figure 1D). Addition of a third or fourth RNP yielded similar results. In both experiments, the number of unviable embryos remained at tolerable levels but increased when targeting a fourth locus (Figure 1C,D).

Injecting a pre-assembled Cas9 protein/gRNA RNP is more mutagenic in zebrafish than co-injecting Cas9 mRNA and gRNA (Burger et al., 2016). However, discrepancies exist in the literature regarding the optimal ratio of gRNA to Cas9 protein for maximising mutagenesis (Hoshijima et al., 2019; Wu et al., 2018). We tested different gRNA:Cas9 ratios for both slc24a5 and tyr, keeping the amount of the three-gRNA set injected constant at 28.5 fmol while increasing the amount of Cas9 in steps, from 4.75 fmol (1 Cas9 to 6 gRNA) to 28.5 fmol (1 Cas9 to 1 gRNA). For both slc24a5 and tyr, more Cas9 resulted in more larvae without any eye pigmentation, implying that Cas9 should be present at the 1-to-1 ratio with gRNA or even in excess for optimal results (at the 1-to-1 ratio, 63/67 slc24a5 and 71/74 tyr F0 larvae lacked eye pigmentation) (Figure 1—figure supplement 1).

To confirm that the phenotype persists throughout the life of the animal and transmits into the germline, we grew slc24a5 F0 knockout larvae lacking eye pigmentation at 2 dpf to adulthood. All (41/41) adult slc24a5 F0 fish still displayed the golden phenotype (Lamason et al., 2005) at 2.5 months (Figure 1G). Incrossing the slc24a5 F0 adult fish produced clutches of embryos that were all devoid of eye pigmentation at 2 dpf (n = 3 clutches, total 283/283 embryos), while outcrossing them to wild types produced larvae displaying wild-type pigmentation (n = 3 clutches, total 313/313 embryos) (Figure 1H). Unviability in the incross clutches was higher than the outcross clutches, although this difference was not significant (9.6 ± 12.1% vs 1.7 ± 1.5%, p = 0.37 by Welch’s t-test). The F0 protocol thus directly produced phenotypically homozygous knockout animals without the need for breeding, and the mutations were transmitted through the germline.

Finally, we confirmed the efficacy of the protocol by targeting two other developmental genes: mab21l2, which is required for eye development, and tbx16, which encodes a T-box transcription factor. Homozygous mab21l2u517 (Wycliffe et al., 2020) mutants showed microphthalmia (small eye) with a flattened retina, resembling eye defects observed in humans with mutations in its ortholog MAB21L2 (Rainger et al., 2014). 96% (96/100) of larvae injected with three RNPs targeting mab21l2 showed this phenotype (Figure 1I). Homozygous tbx16 loss-of-function mutants display the spadetail phenotype, characterised by a bent tail terminating in a clump of cells (Ho and Kane, 1990). 100% (93/93) of the injected larvae were evident spadetail mutants (Figure 1—figure supplement 2).

For slc24a5, tyr, and tbx16 genes, we obtained strikingly similar results of phenotypic penetrance to that shown in previous work by Wu et al., 2018, but targeting three loci rather than four, which reduces potential off-target effects.

Sequencing of targeted loci reveals the diversity of null alleles in F0 knockout animals

F0 knockout of the developmental genes slc24a5, tyr, mab21l2, and tbx16 consistently replicated homozygous null mutant phenotypes. Does this actually reflect frameshift mutations in most or all copies of the genome? To assess the proportion and diversity of knockout alleles in the F0 larvae generated by the method, we performed deep sequencing of the slc24a5, tyr, and tbx16 loci, as well as most other loci we targeted throughout the study. We collected more than 100,000 reads for 32 targeted loci on 10 separate genes, each in 3–4 individual animals for a total of 123 samples sequenced each at a coverage of 995 ± 631×. We quantified the mutations with the ampliCan algorithm (Labun et al., 2019). At each locus, 87 ± 18% of the reads were mutated (Figure 2A). If a read was mutated, the mean probability that it also carried a frameshift mutation was 65.4%, confirming the absence of bias in frameshift probability (Moreno-Mateos et al., 2015) and verifying the assumption of the theoretical model that 2 out of every 3 mutations induce a frameshift. The same RNP produced the same mutations in different animals more often than expected by chance (Figure 2B; two animals injected with the same RNP shared 3.3 ± 1.6 of their top 10 most frequent indels vs 0.5 ± 0.7 if they were injected with different RNPs), in line with a non-random outcome of DNA repair at Cas9 breaks (Shen et al., 2018; van Overbeek et al., 2016). As expected (Moreno-Mateos et al., 2015; Varshney et al., 2015), shorter indels were observed more often than longer ones, with an overall bias towards deletions (Figure 2C; 57% deletions vs 43% insertions, n = 7015 unique indels). The positions of the deleted nucleotides are normally distributed with a peak at 4 bp before the protospacer adjacent motif (PAM) (Figure 2—figure supplement 1), confirming previous reports (Moreno-Mateos et al., 2015). The diversity of null alleles in individual F0 knockout animals was extensive: at each targeted locus, there were 40.5 ± 27.5 (median ± median absolute deviation) different alleles, which in theory can produce up to hundreds of thousands of different versions of the targeted gene. Importantly, the sequencing data demonstrates the build-up of frameshift probability achieved by multi-locus targeting, in line with the theoretical model (Figure 1B). For all genes targeted (n = 10) and 31/35 of individual animals sequenced, three RNPs were sufficient to achieve over 80% of alleles harbouring a frameshift mutation (Figure 2D). For slc24a5 and tyr, the fourth RNP only marginally augmented this proportion (+ 2.7% for slc24a5; + 4% for tyr) (Figure 1E,F).

Figure 2 with 1 supplement see all
Deep sequencing of loci targeted in F0 embryos.

(A) Percentage of reads mutated (height of each bar, grey) and percentage of reads with a frameshift mutation (orange) at each gene, locus (capital letters refer to IDT’s database), larva (within each gene, the same number refers to the same individual animal; 0 is uninjected control). (B) Number of indels in common when intersecting the top 10 most frequent indels of two samples from different loci or from the same locus but different animals. Black crosses mark the means. *** p < 0.001; Welch’s t-test. (C) Frequency of each indel length (bp). Negative lengths: deletions, positive lengths: insertions. (D) Proportion of alleles harbouring a frameshift mutation if 1, 2, or three3 loci in the same gene were targeted, based on deep sequencing of each targeted locus. Each line corresponds to an individual animal. (E) Sanger sequencing of amplicons spanning multiple targeted loci of slc24a5. Arrowheads indicate each protospacer adjacent motif (PAM), capital letters refer to the crRNA/locus name. (F) Deep sequencing of the top 3 predicted off-targets of each slc24a5 gRNA (A, B, D). Each data point corresponds to one locus in one animal. Percentage of mutated reads at on-targets is the same data as in A (slc24a5 loci A, B, D). See also Figure 2—figure supplement 1.

Introducing a frameshift is a robust, widely applicable strategy to prevent the production of the protein of interest. However, the proportion of alleles harbouring a frameshift mutation is not sufficient alone to generate the high phenotypic penetrance we observed. For example, of the larvae injected with three RNPs targeting slc24a5 or tyr, 3/6 had less than 90% alleles with a frameshift mutation (Figure 1E,F), but phenotypic penetrance was consistently > 94% (score 1 in Figure 1C,D and Figure 1—figure supplement 1). There are multiple reasons why the proportion of alleles carrying a frameshift mutation is a conservative underestimate of null alleles. First, mutations of residues at key domains of the protein can be sufficient for loss of function. Second, large indels may disrupt the sequencing PCR primers' binding sites, preventing amplification of such alleles. Third, the pooled RNPs may also lead to deletion of large sequences that span two loci being targeted (Hoshijima et al., 2019; Kim and Zhang, 2020; Moreno-Mateos et al., 2015; Wu et al., 2018). To test the latter possibility, we Sanger sequenced amplicons from regions that span multiple targeted loci within slc24a5. We identified clear instances where the gene underwent a large deletion between two targeted sites (Figure 2E). As large deletions are highly likely to prevent the expression of a functional protein, they further affirm the efficacy of the three-RNP strategy.

Mutations at off-target loci are a potential concern when using Cas9. We sequenced the top three predicted off-targets in protein-coding exons for each of the three gRNAs of the slc24a5 set, for a total of nine off-targets. There were essentially no mutated reads (< 0.5%) at all but one off-target, for which mutated reads ranged between 15–54% (n = 4 larvae) (Figure 2F). Importantly, 3/4 larvae had few or no reads with a frameshift mutation at this site (0%, 0%, 1.8%), while the fourth had 42% reads with a frameshift mutation. Of the 9 off-targets, the mutated off-target had the lowest number of mismatches with the gRNA binding sequence (2 mismatches vs 3–4 for the other 8 off-target loci) and the worst off-target risk (predicted score of 58 vs 59–87 for the other 8 off-target loci), indicating that mutations may be relatively predictable. These levels of mutagenesis at a single locus are unlikely to be sufficient to produce consistently high proportions of biallelic null alleles, either in individual animals or at the level of the population of F0 mutants. Hence, we do not consider mutations at off-targets to be a major concern in applications where a population of F0 knockout animals is phenotyped.

Overall, our simple protocol involving three synthetic RNPs at 1:1 Cas9 to gRNA ratio takes just a few hours to complete yet consistently achieves > 90% biallelic knockouts. While the method generates a diverse mix of null alleles in the injected animals, frameshift mutations are a universal mechanism which can be deployed on virtually any gene of interest.

Headloop PCR is a rapid sequencing-free method to validate gRNAs

Deep sequencing allows for the quantification of frameshift mutations in F0 animals, but the cost is not always justified simply to confirm sufficient mutagenic activity of gRNAs. As an inexpensive and rapid alternative, we adapted a suppression PCR method called headloop PCR (Rand et al., 2005). We reasoned that suppressing amplification of the wild-type haplotype at a target locus would reveal the presence of mutant alleles. Suppression is achieved by adding to one of the PCR primers a 5' tag that is complementary to the wild-type sequence at the target locus. During PCR, the tag base-pairs to the target sequence in the same strand, directing elongation to form a stable hairpin, which prevents the use of the amplicon as template in the subsequent cycles (Figure 3A). Any indels generated in the target locus will prevent the formation of the headloop and allow exponential amplification. To demonstrate the efficacy of this technique, we used headloop PCR for five targeted loci in slc24a5 that we had also sequenced. No amplification with headloop primers was detected for any of the loci in uninjected embryos, indicating suppression of amplification of the wild-type haplotype (Figure 3B). In contrast, robust amplification of the targeted loci was observed from the F0 embryos injected with highly mutagenic RNPs. Amplification was absent at the locus targeted by a gRNA known to be ineffective (slc24a5 gRNA C, < 2% mutated reads; Figure 3B).

Figure 3 with 3 supplements see all
Headloop PCR is a rapid, sequencing-free method to confirm design of gRNAs.

(A) Principle of headloop PCR. A headloop tag complementary to the target locus is added to one PCR primer (here, to the forward primer). During PCR, the first elongation incorporates the primer and its overhang; the second elongation synthesises the headloop tag. (left) If the template is wild-type, the complementary tag base-pairs with the target locus and directs elongation (hatched sequence). The amplicon forms a hairpin secondary structure, which prevents its subsequent use as template. (right) If the targeted locus is mutated, the tag is no longer complementary to the locus. The amplicon remains accessible as a template, leading to exponential PCR amplification. (B) (left) Target loci (A, B, C, D, G) of slc24a5 amplified with the PCR primers used for sequencing (std, standard) or when one is replaced by a headloop primer (HL). Orange arrowheads mark the 300 bp ladder band. (right) Deep sequencing results of the same samples as comparison (reproduced from Figure 2A, except locus C); u, uninjected control, i, injected embryo. Framed are results for gRNA C, which repeatedly failed to generate many mutations. See also Figure 3—figure supplement 1, 2, 3.

Next, we tested whether headloop PCR could be used in a semi-quantitative manner to estimate the proportion of mutated alleles in F0 knockout embryos. We derived a score for each sample from the standard PCR and headloop PCR band intensities on an agarose gel (Figure 3—figure supplement 1A). This score correlated well with the proportion of mutated reads measured by deep sequencing (Figure 3—figure supplement 1B; r = 0.66, n = 29 samples tested from n = 7 loci). A tentative headloop score threshold at 0.6 could discriminate gRNAs that were less mutagenic (tbx16 gRNA B and tyr gRNA A, both generated < 60% mutated reads). There was a false negative: tbx16 locus D repeatedly produced a low headloop score (< 0.5) while the gRNA generated close to 100% mutated reads. We conclude that headloop PCR can be used in a semi-quantitative manner to confirm that mutagenesis is high at every targeted site, although it may at times be overly conservative.

To test the sensitivity towards small indels, we used headloop PCR to genotype embryos from two stable mutant lines we had available (Figure 3—figure supplement 2). The first allele is a 1–bp deletion followed by a transversion (T>A) in the gene apoea; the second is a 2–bp deletion in the gene cd2ap. Sanger sequencing corroborated 100% of the genotype calls made by headloop PCR. Small deletions are therefore sufficient to prevent the formation of the hairpin.

Importantly, we determined that use of a proofreading polymerase (with 3′→5′ exonuclease activity) was essential for effective suppression (Figure 3—figure supplement 3A), presumably because replication errors made in the target sequence prevent the formation of the hairpin.

These results demonstrate that our adapted headloop PCR method is a simple, sensitive, inexpensive, and rapid approach to verify the mutagenic potential of gRNAs before undertaking an F0 phenotypic screen.

Multiple genes can be disrupted simultaneously in F0 animals

Some applications require the simultaneous disruption of two or more genes. In epistasis analyses, combinations of genes are mutated to resolve a genetic pathway (Michels, 2002). Many traits and diseases are polygenic, with each gene variant contributing a small effect to the outcome. In this case, disrupting multiple genes collectively can reveal synergistic interactions. Mutating a gene can also lead to the upregulation of evolutionary-related counterparts if the mutated transcript is degraded by nonsense-mediated decay (El-Brolosy et al., 2019; Ma et al., 2019). Jointly inactivating evolutionary-related genes may therefore be necessary to overcome genetic robustness.

To test the feasibility of double gene knockout in F0 animals, we targeted pairs of genes that each produce a distinct developmental phenotype when mutated. To compare our method with published work (Wu et al., 2018), we first targeted the pigmentation gene slc24a5 (Lamason et al., 2005) and the T-box transcription factor encoding gene tbx5a, which is required for pectoral fin development (Garrity et al., 2002). Double biallelic knockouts should therefore lack both pigmentation and pectoral fins. Each gene was targeted with a three-RNP set, then the two sets were injected together. Similar to previous results, single gene targeting produced high proportions of knockout animals—100% (37/37) of the slc24a5 F0 larvae had completely unpigmented eyes at 2 dpf and 100% (43/43) of the tbx5a F0 larvae did not develop pectoral fins (Figure 4A). When both genes were targeted in individual animals, 93% (26/28) displayed both phenotypes. This again precisely mirrors results obtained by Wu et al., 2018, but mutating fewer loci in each gene. We replicated this result by targeting a second pair of genes, the pigmentation gene tyr (Kelsh et al., 1996), and the T-box transcription factor encoding gene ta, which is required for tail development (Schulte-Merker, 1995). 100% of the injected embryos exhibited both the no pigmentation and no tail phenotypes (Figure 4B). These experiments demonstrate the feasibility of simultaneously disrupting two genes directly in the F0 animals.

Figure 4 with 2 supplements see all
Multiple gene knockout in F0.

(A) Penetrance of single (slc24a5, tbx5a) and combined (both) biallelic knockout phenotype(s) in F0. Pictures of example larvae were taken at 3 dpf. No pigmentation refers to embryos clear of eye pigmentation at 2 dpf, as in Figure 1C. Pectoral fins were inspected at 3 dpf. (bottom) Unviability as percentage of 1-dpf embryos. (B) Penetrance of single (tyr, ta) and combined (both) biallelic knockout phenotype(s) in F0. Pictures of example larvae were taken at 2 dpf. (bottom) Unviability as percentage of 1-dpf embryos. (C) (left) Pictures of example elavl3:GCaMP6s larvae at 4 dpf. Left was uninjected; right was injected and displays the crystal phenotype. Pictures without fluorescence were taken with illumination from above to show the iridophores, or lack thereof. (right) Two-photon GCaMP imaging (z-projection) of the elavl3:GCaMP6s, crystal F0 larva shown on the left. (inset) Area of image (white box). See also Figure 4—video 1, Figure 4—video 2.

We then assessed the feasibility of generating triple gene knockouts in F0 animals by directly recreating the fully pigmentless crystal mutant. crystal carries loss-of-function mutations in genes mitfa (Lister et al., 1999), mpv17 (D'Agati et al., 2017; White et al., 2008), and slc45a2 (Streisinger et al., 1986), which prevent respectively the development of melanophores, iridophores, and pigmented cells in the retinal pigment epithelium. The crystal mutant therefore lacks dark and auto-fluorescent pigments over the skin and eyes, making it useful for live imaging applications (Antinucci and Hindges, 2016). However, establishing a mutant allele or a transgene onto the crystal background takes months of breeding and genotyping, limiting its use. We therefore tested whether the crystal phenotype could be directly obtained in a transgenic line by targeting slc45a2, mitfa, and mpv17 in Tg(elavl3:GCaMP6s)a13203 (Kim et al., 2017) larvae, which express the calcium indicator GCaMP6s in post-mitotic neurons. We injected three sets of three RNPs, with each set targeting one gene. Targeting three genes simultaneously lowered viability by 4 dpf (50% of injected larvae were unviable). Nonetheless, 9/10 of viable larvae displayed the transparent crystal phenotype (Figure 4C left). The crystal F0 larvae expressing pan-neuronal GCaMP6s were suitable for live imaging under a two-photon microscope. The whole brain and the eyes could be effectively imaged in vivo at single-cell resolution (Figure 4C right, Figure 4—video 1, Figure 4—video 2). This included amacrine and ganglion cells in the retina, which are not normally accessible to imaging in other single-gene knockout lines routinely used for imaging, such as nacre (Lister et al., 1999), due to persistence of pigments in the retinal pigment epithelium (Antinucci and Hindges, 2016). The F0 knockout protocol rapidly produced crystal larvae directly in a transgenic line without the need for crossing.

Continuous traits, including behavioural, can be accurately quantified in F0 knockout animals

With some limited exceptions (Sunagawa et al., 2016), the F0 approach has been constrained to visible developmental phenotypes that can be assessed in individual animals. Continuous traits, for which phenotypic values vary within a continuous range, have rarely been studied using F0 knockouts due to concerns about the incomplete removal of wild-type alleles and diversity of null alleles within and across F0 animals. Both of these issues will potentially dilute the experimental pool with unaffected or variably affected animals, reducing the measurable effect size between experimental and control groups. This would make continuous traits less likely to be reliably detected in a population of F0 knockouts than in a population of stable line mutants, which will all harbour a single characterised mutation in every cell. We therefore tested whether F0 knockouts can recapitulate a variety of known loss-of-function continuous trait phenotypes in larval stages.

We first asked whether a simple mutant behavioural phenotype could be observed in F0 knockouts. trpa1b encodes an ion channel implicated in behavioural responses to chemical irritants such as mustard oil (allyl isothiocyanate). While wild-type larvae show a robust escape response when exposed to this compound, trpa1bvu197 null mutants do not react strongly (Prober et al., 2008). We injected embryos with three RNPs targeting trpa1b and recorded the behavioural response of the F0 knockouts to mustard oil. To control for any non-specific effects of the injection procedure or presence of RNPs on behaviour, control larvae were injected with a set of three scrambled RNPs, which carry gRNAs with pseudo-random sequences predicted to not match any genomic locus. While scrambled-injected control larvae displayed an escape response when mustard oil was added to the water, most (19/22) trpa1b F0 knockout larvae failed to strongly respond (Figure 5A, Figure 5—video 1). Therefore, trpa1b F0 knockouts replicated the established stable trpa1bvu197 loss-of-function mutant behavioural phenotype.

Figure 5 with 1 supplement see all
Dynamic, continuous traits are accurately assessed in F0 knockouts.

(A) Escape response to mustard oil in trpa1b F0 knockouts. (left) Activity (total Δ pixel/second) of scrambled controls and trpa1b F0 knockout larvae at 4 dpf. Pre: 3-min window before transfer to 1 µM mustard oil. Post: 3-min window immediately after. Traces are mean ± standard error of the mean (SEM). (right) Total activity (sum of Δ pixel/frame over the 3-min window) of individual larvae before and after transfer to 1 µM mustard oil. *** p < 0.001 (Δ total activity scrambled vs trpa1b F0); Welch’s t-test. (B) Circadian rhythm quantification in csnk1db F0 knockout larvae. (top) Timeseries (detrended and normalised) of bioluminescence from per3:luciferase larvae over five subjective day/night cycles (constant dark). Circadian time is the number of hours after the last Zeitgeber (circadian time 0 9 am, morning of 4 dpf). DMSO: 0.001% dimethyl sulfoxide; PF-67: 1 µM PF-670462. Traces are mean ± SEM. (bottom) Circadian period of each larva calculated from its timeseries. Black crosses mark the population means. ns, p = 0.825; * p = 0.024; *** p < 0.001; pairwise Welch’s t-tests with Holm’s p-value adjustment. See also Figure 4—video 1.

Next, we tested whether a quantitative molecular phenotype could be accurately probed in a population of F0 knockouts generated by our approach. As in nearly all organisms, zebrafish physiology and behaviour are regulated by an internal circadian (24-hour) clock driven by transcription-translation feedback loops. The periodicity of this clock is in part regulated by the phosphorylation of the Period proteins, which constitutes a component of the negative arm of the feedback loop. Drugs and mutations that interfere with Casein Kinases responsible for this phosphorylation alter circadian period length (Lowrey et al., 2000; Price et al., 1998; Smadja Storz et al., 2013). We therefore targeted casein kinase 1 delta (csnk1db) in the Tg(per3:luc)g1 reporter line, which allows bioluminescence-based measurement of larval circadian rhythms (Kaneko and Cahill, 2005). The circadian period of control larvae injected with scrambled RNPs in constant dark conditions was 25.8 ± 0.9 hr, within the expected wild-type range (Kaneko and Cahill, 2005). In csnk1db F0 knockout animals, the circadian period was extended by 84 min to 27.2 ± 0.9 hr (Figure 5B). To demonstrate that this period lengthening was not due to non-specific or off-target effects, we measured the circadian period of larvae exposed to the pan-casein kinase inhibitor PF-670462. When PF-670462 was added to scrambled RNPs-injected larvae, the period increased more than 8 hr to 34.1 ± 0.9 hr. However, adding the inhibitor to the csnk1db F0 larvae did not further increase the period (34.3 ± 2.7 hr). Therefore, the phenotypic consequences of the casein-kinase inhibitor and csnk1db knockout are not additive, indicating that they influence circadian period length through the same target pathway. This experiment demonstrates that a quantitative molecular phenotype that unfolds over many days and in many tissues can be accurately detected in the population of F0 knockouts generated with our protocol.

If the diversity of null alleles in F0 animals were to produce substantial phenotypic variation, quantitative differences in multi-parameter behaviours would be difficult to assess in populations of F0 knockouts. To test this, we targeted scn1lab, which encodes a sodium channel. In humans, loss-of-function mutations of its ortholog SCN1A are associated with Dravet syndrome, a rare and intractable childhood epilepsy (Anwar et al., 2019). In zebrafish, scn1lab homozygous null mutants display hyperpigmentation, seizures, and complex day-night differences in free-swimming behaviour (Baraban et al., 2013; Grone et al., 2017). As expected, all (91/91) scn1lab F0 knockouts were hyperpigmented (Figure 6A insert). We then video tracked the F0 larvae over multiple day-night cycles and compared the data to behavioural phenotypes collected from scn1labΔ44 mutant larvae. F0 knockouts and scn1labΔ44 homozygous null mutants had similar behavioural changes compared to their wild-type siblings. During the day, both F0 knockouts and scn1labΔ44 homozygotes spent less time active compared to wild types (all three experiments p < 0.001 by two-way ANOVA). At night, F0 knockouts and scn1labΔ44 homozygotes were as active as wild types initially, then showed a gradual ramping to hyperactivity (Figure 6A and Figure 6—figure supplement 1).

Figure 6 with 1 supplement see all
Multi-parameter behavioural phenotypes are closely replicated in F0 knockouts.

(A) Activity (total Δ pixel/second) of larvae across 2 days (14 hr each, white background) and two nights (10 hr each, grey background). Traces are mean ± SEM. (left) Stable scn1labΔ44 mutant line, from 5 to 7 dpf. The drops in activity in the middle of each day is an artefact caused by topping-up the water. (right) scn1lab F0 knockout, from 6 to 8 dpf. This replicate is called scn1lab F0 experiment 1 in B and C. (inset) Pictures of example scrambled-injected control and scn1lab F0 larvae at 6 dpf. (B) Behavioural fingerprints, represented as deviation from the paired wild-type mean (Z-score, mean ± SEM). 10 parameters describe bout structure during the day and night (grey underlay). Parameters 1–6 describe the swimming (active) bouts, 7–9 the activity during each day/night, and 10 is pause (inactive bout) length. (inset) Pairwise correlations (Pearson) between mean fingerprints. (C) Euclidean distance of individual larvae from the paired wild-type or scrambled-injected (scr) mean. Black crosses mark the population means. ns, p > 0.999; *** p < 0.001; pairwise Welch’s t-tests with Holm’s p-value adjustment. See also Figure 6—figure supplement 1.

To test whether scn1lab F0 knockouts also recapitulated finer, multi-parameter details of scn1labΔ44 mutant behaviour, we compared their locomotion across ten behavioural parameters describing down to sub-second scales the swimming bouts and pauses characteristic of larval zebrafish behaviour (Ghosh and Rihel, 2020) (see Materials and methods). To visualise these multi-dimensional traits, we calculated a behavioural fingerprint for each group, defined as the deviation of each mutant larva from its wild-type siblings across all parameters. This fingerprint was similar between F0 knockout larvae and scn1labΔ44 homozygotes (Figure 6B). The two clutches of scn1lab F0 knockouts had highly correlated behavioural fingerprints (r = 0.89), and each correlated well with the fingerprint of the scn1labΔ44 homozygotes (r = 0.86, r = 0.75). We then measured the Euclidean distance between each animal’s behavioural fingerprint and its paired wild-type mean. Unlike scn1labΔ44 heterozygous larvae, which do not display overt phenotypes, scn1labΔ44 homozygotes and both scn1lab F0 knockout clutches were significantly distant from their wild-type counterparts (Figure 6C). The F0 knockout larvae sit in average at greater distances from their wild-type siblings than stable knockout larvae. However, this difference was not significant when comparing effect sizes between experiments (stable knockout wild types vs stable knockout homozygotes: Cohen’s d = 1.57 is not significantly different than scrambled-injected controls vs F0 knockout larvae: d = 2.91 in F0 experiment 1, d = 2.57 in F0 experiment 2; respectively p = 0.18 and p = 0.27). Together, these results demonstrate that diversity of null alleles is not a barrier to measuring detailed mutant behavioural phenotypes in populations of F0 knockouts.

In summary, complex continuous traits, including behavioural phenotypes, can be rigorously measured directly in F0 animals. We demonstrated this by replicating in F0 knockouts the expected lack of escape response to a chemical irritant in trpa1b mutants, by recapitulating the predicted circadian clock phenotype when csnk1db is disrupted, and by phenocopying complex day-night differences in free-swimming behaviour in scn1lab loss-of-function mutants.

Discussion

We have developed a simple and efficient CRISPR-Cas9 F0 knockout method in zebrafish by coupling multi-locus targeting with high mutagenesis at each locus. To validate gene targeting without the need for sequencing, we also adapted a simple headloop PCR method. The F0 knockout technique consistently converts > 90% of injected embryos into biallelic knockouts, even when simultaneously disrupting multiple genes in the same animal. These advances compress the time needed to obtain biallelic knockouts from months to hours, paving the way to large genetic screens of dynamic, continuously varying traits, such as behavioural phenotypes.

Design of F0 knockout screens

Given the rapid pace at which genes are being associated to diseases by large sequencing projects, strategies to accelerate follow-up studies in animal models are vital for these associations to eventually inform therapeutic strategies. We share here some considerations for the design of F0 genetic screens in zebrafish.

The first step is to select gRNAs for each gene that will be tested. Whenever possible, each target locus should be on a distinct exon as this might negate compensatory mechanisms such as exon skipping (Anderson et al., 2017; Lalonde et al., 2017). Asymmetrical exons, i.e. of a length that is not a multiple of three, can also be prioritised, as exon skipping would cause a frameshift (Tuladhar et al., 2019). If the gene has multiple annotated transcripts, one should target protein-coding exons that are common to most or all transcripts. We sequenced the mutations caused by more than 30 individual gRNAs and only one was consistently non-mutagenic. However, the likelihood of selecting non-mutagenic gRNAs may increase as more genes are tested. Hence, we suggest an approach in two rounds of injections (Figure 7A)—a validation round followed by a phenotyping round.

Recommendations for F0 knockout screens.

(A) Suggestion for the design of an F0 screen, based on a three-step process: (1) selection of gRNAs; (2) verification that all gRNAs are mutagenic; (3) phenotyping. During round 1 (step 2), we recommend targeting a pigmentation gene such as slc24a5 to quantify success rate at injections before estimating minimum samples sizes and commencing the screen. (B) Minimum sample size to detect a phenotype at 0.8 statistical power and 0.05 significance level as more knockouts are present in the population of F0 larvae. Based on 10 simulations with 100 animals in each group (scrambled control, F0 knockout). Mean ± standard deviation across the 10 simulations. (left) Minimum sample sizes to detect the lack of response to mustard oil of the trpa1b knockouts. Dashed line indicates the real sample size of the experiment (n = 22, Figure 5A). (right) Minimum sample sizes to detect the lengthened circadian period of csnk1db knockouts. Dashed line indicates the real sample size of the experiment (n = 16, Figure 5B).

In the first round, each gRNA set is injected followed by deep sequencing or headloop PCR to confirm mutagenesis, thereby controlling the false negative rate of a screen. Headloop PCR is cheap, robust, and requires only a single step, which makes it easily adapted to high-throughput screening. No specialist equipment is required, as opposed to qPCR (Yu et al., 2014), high resolution melting analysis (HRMA) (Samarut et al., 2016), or fluorescent PCR (Carrington et al., 2015). Unlike deep sequencing, qPCR, and HRMA, it is also flexible with respect to the size of amplicons and so is sensitive to a wide range of alleles, from small indels to large deletions between targeted loci. It can be used to assay the efficiency of any gRNA, with no restrictions on target sequence that might be imposed by the use of restriction fragment length polymorphism (Jao et al., 2013), for example. The products of headloop PCR are also compatible with different sequencing methods, should further analysis of mutant haplotypes be required.

The second round of injections generates the F0 knockouts used for phenotyping. If phenotyping requires a transgenic line, for instance expressing GCaMP for brain imaging, the F0 approach has the additional advantage that it can be deployed directly in embryos from this line. We advise that control larvae are injected with a set of scrambled RNPs, as they control for any potential effect caused by the injection of Cas9 and exogenous RNA. This two-step approach assumes that the phenotyping requires substantial time or resources, for instance video tracking behaviour over multiple days. If phenotyping is rapid and/or largely automated (Eimon et al., 2018; Kokel et al., 2010), genotyping can be performed directly on a sample of the phenotyped animals. If a gRNA is found not to generate enough mutations, it can be replaced, and the experiment repeated.

In screening situations in which every phenotyped animal is not genotyped, the reliability of the F0 method depends on the reliability of the injections. For instance, if some eggs were missed during injections, the F0 population would include a proportion of wild-type animals, which would reduce the effect size between the control and the experimental group and make the phenotype less likely to be detected. To evaluate how resilient phenotyping would be in such conditions, we used bootstrapping to simulate distributions where a gradually larger proportion of the F0 population are in fact wild-type animals. Power calculations on simulations derived from the trpa1b and csnk1db F0 knockout experimental data show that a single 96-well plate, i.e. sample sizes of 48 larvae in each group, is more than sufficient to detect mutant phenotypes at a power of 0.8 and a significance level of 0.05, even with a relatively low proportion of knockout animals in the F0 population (28 and 59%, respectively; Figure 7B). Therefore, the high efficacy and throughput of the F0 method allows one to discover phenotypes with robust statistical power.

F0 knockouts vs stable knockout line—diversity of null alleles

A key characteristic of the F0 knockout approach is the diversity of null alleles. The F0 mutants do not have a unique, definable genotype. This can be a shortcoming, for instance in disease modelling applications where a specific mutation needs to be reproduced. However, frequently the experimental goal is to assess the consequences of the lack of a specific protein, not the consequences of a specific allele. In this context, the diversity of null alleles in F0 knockouts may have some advantages over stable mutant lines. With CRISPR-Cas9, stable mutant lines are often generated by introducing a single frameshift mutation. However, the assumption that this leads to a complete loss of protein function is not infallible. For example, in a survey of 193 knockout lines in HAP1 cells, around one third still produced residual levels of the target protein, thanks in part to genetic compensatory mechanisms such as skipping of the mutated exon or translation from alternative start codons (Smits et al., 2019). Such compensation can allow production of a partially functional truncated protein. Exon skipping has also been documented in stable zebrafish knockout lines (Anderson et al., 2017; Lalonde et al., 2017). By creating a diverse array of mutations at three sites per gene, each on a separate exon wherever possible, such compensatory mechanisms are not likely to allow the production of a functional protein in the F0 knockouts. Furthermore, a given phenotype may differ between different null alleles (Chiavacci et al., 2017; Schuermann et al., 2015) or between different genetic backgrounds (Garrity et al., 2002; Sanders and Whitlock, 2003). The F0 knockout method generates a variety of null alleles and can be deployed directly on the progeny of wild-type animals of different backgrounds. Accordingly, we propose that a knockout phenotype detected in this genetically diverse population of animals is likely to be a robust and reproducible description of the impact caused by the absence of the protein, akin to reaching a synthesised conclusion after comparing stable knockout lines of different alleles and from different founder animals.

Nevertheless, after screening, it is likely that stable knockout lines will need to be generated for more detailed and controlled studies. Directly raising the phenotyped F0 larvae may not be optimal as multi-locus targeting will result in complex genotypes. Instead, sequencing data, if available, can be used to select a gRNA that consistently generates high numbers of frameshift mutations. Furthermore, we have successfully used headloop PCR to detect mutations in tail clips from 48 to 72 hours post-fertilisation F1 embryos and sequenced the mutant haplotypes directly by Sanger sequencing. Embryos carrying mutant alleles could be identified within a single day, then grown directly into adults, thereby reducing drastically the number of fish that need to be raised and genotyped to generate a stable mutant line.

Other technical considerations for F0 knockouts

Although unviability of injected larvae was not a limitation in our experiments, we observed some unviable embryos in the populations of F0 knockouts, similar to previous studies (Wu et al., 2018). While unviability was highly variable, even between replicates of the same experiment (e.g. Figure 1D vs Figure 1—figure supplement 1B), it may broadly correlate with the number of generated double-strand breaks. Indeed, developmental defects slightly increased when adding more Cas9 (Figure 1—figure supplement 1) and were always more frequent when targeting more loci (Figure 1C,D and Figure 4A,B). Moreover, unviability remained lower in scrambled RNP-injected embryos compared to F0 knockout siblings, likely excluding chemical toxicity unrelated to Cas9-induced double-strand breaks. A sound strategy to reduce the number of double-strand breaks, while maintaining high proportions of knockout alleles, would be to reduce the number of loci targeted. Machine learning tools can predict editing outcomes and indel frequencies in cell cultures based on target sequence and genomic context (Shen et al., 2018). Hence, it may be feasible to systematically apply the frameshift model (Figure 1B,E,F and Figure 2D) directly at the gRNA design stage using predicted mutations as input. This would allow the user to select specific gRNAs that are predicted to produce a high number of frameshift mutations.

We sequenced off-target loci and found that off-target effects are unlikely to be a pervasive issue in F0 phenotypic screens. An off-target gene will typically be targeted by a single RNP. Therefore, even if off-target indels are generated sporadically, the build-up of frameshift probability and large deletions between loci cannot happen at the off-target gene, reducing the likelihood of generating a null allele. If a null allele arises at an off-target gene nevertheless, the lower mutagenesis makes it likely that this allele will neither be present biallelically nor in a large number of cells. The probability that an observed phenotype is a false positive is therefore likely to be low. Low penetrance of a given phenotype (i.e. present in only a small proportion of injected animals), despite evidence of highly mutagenic gRNAs at the targeted loci, may be an indicator of a false positive. In such cases, a solution is to replicate the finding with an independent set of gRNAs with different predicted off-targets.

While multi-locus strategies like ours achieve high proportions of null alleles in F0 knockouts, they admittedly inflate both the number of potential off-target loci and number of double-strand breaks. This cost-benefit balance may be specific to the phenotype under investigation. For example, for a phenotype whose spatial variation is visible in individual animals (Watson et al., 2020), the mutation of one or two loci per gene may be a valuable strategy. However, the study of continuous traits, particularly behavioural ones, likely require consistently high proportions of null alleles. In this case, the mutation of three loci with synthetic gRNAs, as we demonstrate, offers a reasonable compromise.

Conclusion

Building on published work (Hoshijima et al., 2019; Wu et al., 2018), we developed a simple and rapid zebrafish F0 knockout method using CRISPR-Cas9. By combining multi-locus targeting with high mutagenesis at each locus, the method converts the vast majority of wild-type or transgenic embryos directly into biallelic knockouts for any gene(s) of interest. We demonstrate that continuous traits, such as complex behavioural phenotypes, are accurately measured in populations of F0 knockouts. Cumulatively, methods like ours and pilot screens are establishing F0 knockouts as a revolutionary approach for large genetic screens in zebrafish.

Materials and methods

Key resources table
Reagent type (species)
or resource
DesignationSource or referenceIdentifiersAdditional information
Gene
(Danio rerio)
slc24a5EnsemblENSDARG00000024771
Gene
(D. rerio)
tyrEnsemblENSDARG00000039077
Gene
(D. rerio)
mab21l2EnsemblENSDARG00000015266
Gene
(D. rerio)
tbx16EnsemblENSDARG00000007329
Gene
(D. rerio)
tbx5aEnsemblENSDARG00000024894
Gene
(D. rerio)
ta (tbxta)EnsemblENSDARG00000101576
Gene
(D. rerio)
slc45a2EnsemblENSDARG00000002593
Gene
(D. rerio)
mitfaEnsemblENSDARG00000003732
Gene
(D. rerio)
mpv17EnsemblENSDARG00000032431
Gene
(D. rerio)
trpa1bEnsemblENSDARG00000031875
Gene
(D. rerio)
csnk1dbEnsemblENSDARG00000006125
Gene
(D. rerio)
scn1labEnsemblENSDARG00000062744
Genetic reagent
(D. rerio)
Tg(elavl3:GCaMP6s)a13203PMID:28892088ZFIN ID: ZDB-ALT-180502–2Kim et al., 2017
Genetic reagent
(D. rerio)
mitfaw2 (nacre)PMID:10433906ZFIN ID: ZDB-ALT-990423–22Lister et al., 1999
Genetic reagent
(D. rerio)
Tg(per3:luc)g1PMID:15685291ZFIN ID: ZDB-ALT-050225–2Kaneko and Cahill, 2005
Genetic reagent
(D. rerio)
Tg(elavl3:EGFP)knu3PMID:11071755ZFIN ID: ZDB-ALT-060301–2Park et al., 2000
Genetic reagent
(D. rerio)
mab21l2u517PMID:32930361mutant lineWycliffe et al., 2020
Genetic reagent
(D. rerio)
scn1labΔ44This studymutant lineAvailable from Hoffman lab
Sequence-based reagentAlt-R CRISPR-Cas9 crRNAsIDTsee Supplementary file 1
Sequence-based reagentAlt-R CRISPR-Cas9 Negative Control crRNA #1IDTCatalog #: 1072544see Supplementary file 1
Sequence-based reagentAlt-R CRISPR-Cas9 Negative Control crRNA #2IDTCatalog #: 1072545see Supplementary file 1
Sequence-based reagentAlt-R CRISPR-Cas9 Negative Control crRNA #3IDTCatalog #: 1072546see Supplementary file 1
Sequence-based reagentAlt-R CRISPR-Cas9 tracrRNAIDTCatalog #: 1072532
Sequence-based reagentPCR primersThermo Fishersee Supplementary file 1
Peptide, recombinant proteinAlt-R S.p. Cas9 Nuclease V3IDTCatalog #: 1081058
Chemical compound, drugMustard oil (allyl isothiocyanate)Sigma-AldrichCatalog #: W203408
Chemical compound, drugBeetle luciferinPromegaCatalog #: E1601
Chemical compound, drugPF-670462Sigma-AldrichCatalog #: SML0795
Software, algorithmampliCanPMID:30850374bioconductor.org/packages/release/bioc/html/amplican.html
Software, algorithmBioDare2PMID:24809473biodare2.ed.ac.uk
Software, algorithmZebraLabViewPoint Behavior Technologyviewpoint.fr/en/p/software/zebralab-zebrafish-behavior-screening
Software, algorithmMATLAB scripts for behaviour analysis: Vp_Extract.m and Vp_Analyse.mPMID:32241874Scripts included in the GitHub and Zenodo repositories (see Data/resource sharing)
Software, algorithmR packagesCRANsee Supplementary file 2
Software algorithmCommand line packagesCondasee Supplementary file 2
Software algorithmMATLAB toolboxesMathWorkssee Supplementary file 2
Software, algorithmR v3.6.2CRANr-project.org
Software, algorithmMATLAB R2018aMathWorksmathworks.com/products/matlab.html

Animals

Adult zebrafish were reared by University College London’s Fish Facility on a 14 hr:10 hr light:dark cycle. To obtain eggs, pairs of males and females were isolated in breeding boxes overnight, separated by a divider. Around 9 am the next day, the dividers were removed and eggs were collected 7–8 min later. The embryos were then raised in 10-cm Petri dishes filled with fish water (0.3 g/L Instant Ocean) in a 28.5°C incubator on a 14 hr:10 hr light:dark cycle. Debris and dead or dysmorphic embryos were removed every other day with a Pasteur pipette under a bright-field microscope and the fish water replaced. At the end of the experiments, larvae were euthanised with an overdose of 2-phenoxyethanol (ACROS Organics). Experimental procedures were in accordance with the Animals (Scientific Procedures) Act 1986 under Home Office project licences PA8D4D0E5 awarded to Jason Rihel and PAE2ECA7E awarded to Elena Dreosti. Adult zebrafish were kept according to FELASA guidelines (Aleström et al., 2020).

Wild types refer to AB × Tup LF fish. Throughout, F0 refers to embryos that were injected with gRNA/Cas9 RNPs at the single-cell stage. All experiments used wild-type progeny, except the crystal fish experiment, which used the progeny of an outcross of heterozygous Tg(elavl3:GCaMP6s)a13203/+ (Kim et al., 2017), mitfaw2/+ (nacre) (Lister et al., 1999) to wild type and the per3:luciferase (csnk1db) experiment, which used the progeny of a Tg(per3:luc)g1 (Kaneko and Cahill, 2005), Tg(elavl3:EGFP)knu3 (Park et al., 2000) homozygous incross.

Cas9/gRNA preparation

A protocol describing how to generate F0 knockout larvae for a single gene is available at dx.doi.org/10.17504/protocols.io.bfgyjjxw.

The protocol developed by Wu et al., 2018 served as a starting point. The key differences were: synthetic gRNAs were used here, as opposed to in vitro-transcribed; three loci per gene were targeted, as opposed to four; 28.5 fmol (1000 pg) total gRNA and 28.5 fmol (4700 pg) Cas9 (1 Cas9 to 1 gRNA) were injected, as opposed to 28.5 fmol (1000 pg) total gRNA and 4.75 fmol (800 pg) Cas9 (1 Cas9 to 6 gRNA) reported in Wu et al., 2018.

The synthetic gRNA was made of two components which were bought separately from Integrated DNA Technologies (IDT): the crRNA (Alt-R CRISPR-Cas9 crRNA) and tracrRNA (Alt-R CRISPR-Cas9 tracrRNA).

crRNA selection

View detailed protocol

The crRNA was the only component of the Cas9/gRNA ribonucleoprotein (RNP) which was specific to the target locus. IDT has a database of predesigned crRNAs for most annotated genes of the zebrafish genome (eu.idtdna.com). crRNAs for each target gene were ranked based on predicted on-target and off-target scores. Wherever possible, selected crRNAs targeted distinct exons, while proceeding down the list from the best predicted crRNA. RNPs were not tested for activity before experiments, with the exception of slc24a5 gRNA C which we identified as ineffective early during the development of the protocol.

Sequences of the crRNAs and information about the targeted loci are provided in Supplementary file 1.

crRNA/tracrRNA annealing

Request a detailed protocol

The crRNA and tracrRNA were received as pellets, which were individually resuspended in Duplex buffer (IDT) to form 200 μM stocks. Stocks were stored at −80°C before use.

The crRNA was annealed with the tracrRNA to form the gRNA by mixing each crRNA of the set separately with an equal molar amount of tracrRNA and diluting to 57 μM in Duplex buffer. This was usually: 1 μL crRNA 200 μM; 1 μL tracrRNA 200 μM; 1.51 μL Duplex buffer, heated to 95°C for 5 min, then cooled on ice.

gRNA/Cas9 assembly

Request a detailed protocol

Pre-assembled RNPs composed of Cas9 protein and gRNA are more effective in zebrafish than using a combination of Cas9 mRNA and gRNA (Burger et al., 2016).

Cas9 protein was bought from IDT (Alt-R S.p. Cas9 Nuclease V3, 61 µM) and diluted to 57 µM in Cas9 buffer: 20 mM Tris-HCl, 600 mM KCl, 20% glycerol (Wu et al., 2018). It was stored at −20°C before use. For each RNP, equal volumes of gRNA and Cas9 solutions were mixed (typically 1 µL gRNA; 1 µL Cas9), incubated at 37°C for 5 min then cooled on ice, generating a 28.5 µM RNP solution.

For the experiments testing different ratios of Cas9 to gRNA (Figure 1—figure supplement 1), before assembly with gRNA, Cas9 was further diluted to final concentrations of 28.5 μM for a 1:2 ratio; 19 μM for 1:3; 9.3 μM for 1:6. Assembly with gRNA was then performed as above.

RNP pooling

Request a detailed protocol

The three RNP solutions were pooled in equal amounts before injections. The concentration of each RNP was thus divided by three (9.5 µM each), leaving the total RNP concentration at 28.5 μM.

For the experiments testing different numbers of targeted loci (Figure 1C,D), the first RNP was injected alone, or the first two, three, four RNPs were pooled and injected. The order followed the IDT ranking when selecting a single crRNA per exon. The final total RNP concentration remained 28.5 μM, regardless of the number of unique RNPs.

When targeting two genes simultaneously (Figure 4A,B), both three-RNP pools were mixed in equal volumes. Preparation of the nine-RNP mix for the triple gene knockout (Figure 4C) was done in the same manner.

The RNPs were usually kept overnight at −20°C before injections the following day.

Injections

Request a detailed protocol

Approximately 1 nL of the three-RNP pool was injected into the yolk at the single-cell stage before cell inflation. This amounts to around 28.5 fmol of RNP (28.5 fmol [4700 pg] of Cas9 and 28.5 fmol [1000 pg] of total gRNA). Each unique RNP is present in equal amounts in the pool. Therefore, in the case of three RNPs, 9.5 fmol of each RNP were co-injected.

When targeting two genes simultaneously (Figure 4A,B), approximately 2 nL of the six-RNP mix were injected so the amount of RNP per gene would remain equal to when a single gene is targeted. Similarly, when targeting three genes for the crystal fish (Figure 4C), approximately 3 nL of the nine-RNP mix were injected.

scrambled RNPs

Request a detailed protocol

For the experiments targeting trpa1b, csnk1db, or scn1lab, three scrambled crRNAs (Alt-R CRISPR-Cas9 Negative Control crRNA #1, #2, #3) were prepared into RNPs and injected following the same steps as above. Sequences of the scrambled crRNAs are provided in Supplementary file 1.

Phenotype scores

Request a detailed protocol

In experiments targeting slc24a5 or tyr (Figure 1A,B and Figure 1—figure supplement 1), each animal was given a score from 1 to 5 based on its eye pigmentation at 2 dpf: score 5 if the eye was fully pigmented akin to wild types; 4 if it was mostly pigmented; 3 if approximately half the surface of the eye was pigmented; 2 if there were only one or two patches of pigmented cells; 1 if no pigmented cell could be detected. If the two eyes had substantially different pigmentation, the score of the darkest eye was recorded for that animal.

In the double gene knockout experiments (Figure 4A,B), only score 1 was counted as the expected slc24a5 or tyr knockout phenotype.

In the experiment targeting tbx5a (Figure 4A), both pectoral fins were inspected at 3 dpf. Only the absence of both pectoral fins was counted as the expected phenotype.

All scoring was done blinded to the condition.

Unviability

Request a detailed protocol

The percentage of unviable embryos was based on the total number of larvae that died or were dysmorphic (displaying developmental defects not associated with the expected phenotype) after 1 dpf. Unviable embryos at 0 dpf were excluded as they were likely either unfertilised eggs or eggs damaged by the needle. Common developmental defects included heart oedema, tail curvature, and absence of a swim bladder at 5 dpf. This death/dysmorphic count was divided by the total number of larvae at 1 dpf to get a percentage of unviable embryos. Percentage of unviable embryos in the uninjected or scrambled controls was usually zero or low (<9%). It was subtracted from the F0 unviability to account for only effects mediated by the mutagenic RNPs injection. For example, if 5% of the injected embryos died or were dysmorphic after 1 dpf, and 1% of the controls died, the unviability reported for the injected embryos was 4%.

Larvae in slc24a5, tyr, tbx5a, trpa1b, and scn1lab targeting experiments were followed until 5–6 dpf. Larvae in the crystal experiment were followed until 4 dpf. Larvae in the csnk1db experiment were followed until 9 dpf.

As homozygous ta knockouts are lethal early in development (Halpern et al., 1993), embryos in the tyr and ta double gene knockout experiment (Figure 4B) were followed until 2 dpf. Similarly, homozygous tbx16 knockouts have various trunk defects and are lethal early in development (Ho and Kane, 1990), so unviability in the tbx16 F0 knockouts was not quantified.

Unviable embryos were counted blinded to the condition.

Adult slc24a5 F0 fish

Request a detailed protocol

The slc24a5 F0 knockout larvae grown to adulthood (Figure 1G) were generated by injection of a three- set (slc24a5 gRNA A, B, D), and their lack of eye pigmentation was verified at 2 dpf.

crystal fish imaging

Request a detailed protocol

Progeny of an outcross of heterozygous Tg(elavl3:GCaMP6s)a13203/+, mitfaw2/+ (nacre) to wild type were injected with a pool of three sets of three RNPs, each set targeting one gene of mitfa, mpv17, and slc45a2. At 4 dpf, a GCaMP6s-positive crystal F0 fish and an uninjected control were mounted in 1% low melting point agarose (Sigma-Aldrich) in fish water. Pictures of the whole animal (Figure 4C left, pictures with fluorescence) were taken with an Olympus MVX10 microscope connected to a computer with the cellSens software (Olympus). A first picture was taken with white transillumination, then a second picture was taken with only 488 nm excitation light to visualise GCaMP6s fluorescence. Both pictures were then overlaid in ImageJ v1.51 (Schneider et al., 2012) with Image > Color > Merge channels. Pictures showing iridophores, or lack thereof (Figure 4C left, pictures without fluorescence), were taken with a Nikon SMZ1500 brightfield microscope with illumination from above the sample.

The crystal F0 fish was imaged with a custom-built two-photon microscope: Olympus XLUMPLFLN 20× 1.0 NA objective, 580 nm PMT dichroic, bandpass filters: 501/84 (green), 641/75 (red) (Semrock), Coherent Chameleon II ultrafast laser. Imaging was performed at 920 nm with a laser power at sample of 8–10 mW. Images were acquired by frame scanning (10-frame averaging) with a z-plane spacing of 2 µm. Images were 1300 × 1300 pixels for the head stack (Figure 4C right and Figure 4—video 1) and 800 × 800 for the eye stack (Figure 4—video 2), both 0.38 × 0.38 µm pixel size. The image included in Figure 4C (right) is a maximum intensity z-projection of 10 frames of the head stack (Figure 4—video 1). Contrast and brightness were adjusted in ImageJ (Schneider et al., 2012).

Illumina MiSeq

Throughout, deep sequencing refers to sequencing by Illumina MiSeq. For each gene, four injected larvae and one uninjected or scrambled RNPs-injected control larva were processed. For slc24a5, tyr, tbx16, tbx5a, ta, slc45a2, mitfa, mpv17, and scn1lab, injected larvae displaying the expected biallelic knockout phenotype were processed.

Genomic DNA extraction

Request a detailed protocol

The larvae were anaesthetised and their genomic DNA extracted by HotSHOT (Meeker et al., 2007), as follows. Individual larvae were transferred to a 96-well PCR plate. Excess liquid was removed from each well before adding 50 μl of 1× base solution (25 mM KOH, 0.2 mM EDTA in water). Plates were sealed and incubated at 95°C for 30 min then cooled to room temperature before the addition of 50 μl of 1× neutralisation solution (40 mM Tris-HCL in water). Genomic DNA was then stored at 4°C.

PCR

Request a detailed protocol

Each PCR well contained: 7.98 µL PCR mix (2 mM MgCl2, 14 mM pH 8.4 Tris-HCl, 68 mM KCl, 0.14% gelatine in water, autoclaved for 20 min, cooled to room temperature, chilled on ice, then added 1.8% 100 mg/ml BSA and 0.14% 100 mM d[A, C, G, T]TP), 3 µL 5× Phusion HF buffer (New England Biolabs), 2.7 µL dH2O, 0.3 µL forward primer (100 µM), 0.3 µL reverse primer (100 µM), 0.12 µL Phusion High-Fidelity DNA Polymerase (New England Biolabs), 1.0 µL genomic DNA; for a total of 15.4 µL. The PCR plate was sealed and placed into a thermocycler. The PCR program was: 95°C – 5 min, then 40 cycles of: 95°C – 30 s, 60°C – 30 s, 72°C – 30 s, then 72°C – 10 min then cooled to 10°C until collection. The PCR product’s concentration was quantified with Qubit (dsDNA High Sensitivity Assay) and its length was verified on a 2.5% agarose gel with GelRed (Biotium). Excess primers and dNTPs were removed by ExoSAP-IT (ThermoFisher) following the manufacturer’s instructions. The samples were then sent for Illumina MiSeq, which used MiSeq Reagent Nano Kit v2 (300 Cycles) (MS-103–1001).

Sequences and genomic positions of the PCR primers are provided in Supplementary file 1.

Illumina MiSeq data analysis

Illumina MiSeq data was received as two fastq files for each well, one forward and one reverse. The paired-end reads were aligned to the reference amplicon with the package bwa v0.7.17 and the resulted bam alignment file was sorted and indexed with samtools v1.9 (Li et al., 2009). To keep only high-quality reads, any read shorter than 140 bp, with a Phred quality score below 40, or with more than 20% of its length soft-clipped were discarded from the bam file before analysis. Whenever necessary, bam alignment files were visualised with IGV v2.4.10. The resulting filtered bam file was converted back to a forward and a reverse fastq file using bedtools v2.27.1 (Quinlan and Hall, 2010). The filtered fastq files were used as input to the R package ampliCan (Labun et al., 2019), together with a csv configuration file containing metadata information about the samples. AmpliCan was run with settings min_freq = 0.005 (any mutation at a frequency below this threshold was considered as a sequencing error) and average_quality = 25; other parameters were left as default. AmpliCan detected and quantified mutations in the reads and wrote results files that were used for subsequent analysis. Reads from uninjected or scrambled-injected controls were used to normalise the mutation counts, i.e. any mutation present in the control embryo was not counted as a Cas9-induced mutation in the injected ones. Downstream of ampliCan, any samples with less than 30× paired-end (60× single-read) coverage were excluded from further analysis.

Figure 2A plots the proportion of mutated reads and the proportion of reads with a frameshift mutation at each locus, as computed by ampliCan. If a read contained multiple indels, ampliCan summed them to conclude whether the read had a frameshift mutation or not. This frameshift prediction may be inaccurate in some rare cases where an indel was in an intron or disrupts an exon/intron boundary.

Probability of knockout by frameshift

Request a detailed protocol

There may be cases where an indel at a downstream locus restored the correct frame, which had been shifted at one of the upstream targets. The model of biallelic knockout by frameshift (Figure 1B) assumes this situation also lead to a knockout.

Proportion of frameshift alleles

Request a detailed protocol

The following refers to Figure 1E,F and Figure 2D. If a single locus is targeted, the proportion of frameshift alleles (pshift) was equal to the proportion of reads with a frameshift mutation, as counted by ampliCan in the MiSeq data. If only the first locus is targeted, the proportion of non-frameshift alleles is equal to the proportion of reads that did not have a frameshift mutation at this locus (either not mutated or total indel was of a length multiple of three), pnotshift1. If a second locus is targeted, proportion of frameshift alleles so far is pshift12=1pshift2×pnotshift1, and proportion of non-frameshift alleles so far is pnotshift12=1 pshift12, and so on. At locus l;

(1) pshiftl=1pshiftl1 ×pnotshift1l

This was done for each animal individually.

The order of the loci at each gene (locus 1, 2, , l in Equation 1) follows the ranking of crRNAs in IDT’s database, i.e. the alphabetical order of the locus names in Figure 2A.

Equation 1 assumes that genotypes at each locus of the allele were randomly assigned, i.e. that finding indel x at the first locus does not make it more or less likely to find indel y at the second locus of the same allele. While mutations at each locus may be independent events initially, some alleles might be disproportionately replicated across cell divisions, therefore it is an approximation. Equation 1 also assumes that reads at each locus were randomly sampled from the pool of alleles.

Comparisons of mutations between samples

Request a detailed protocol

This refers to Figure 2B. Reads from control larvae were not used in this analysis. From each sample, the top 10 most frequent indels were extracted. Any sample with less than 10 indels in total was discarded for this analysis. Pairwise intersections were then performed between each sample’s top 10, each time counting the number of common indels, defined as having the same start and end positions, the same type (deletion or insertion) and in the case of insertion, the same inserted sequence. Positions are given by ampliCan in relation to the protospacer adjacent motif (PAM) position, which is set at 0. The results were then grouped whether the intersection was performed between two samples from different loci (e.g. slc24a5, locus D, fish 1 vs csnk1db, locus A, fish 4; n = 6286 intersections) or two samples from the same locus but different fish (e.g. slc24a5, locus D, fish one vs slc24a5, locus D, fish 4; n = 155 intersections).

Probability of indel lengths

Request a detailed protocol

This refers to Figure 2C. Reads from control larvae were not used in this analysis. Only unique mutations in each sample were considered here. Duplicates were defined as any indel from the same sample, at the same positions, of the same length and type (insertion or deletion), and in the case of insertion with the same inserted sequence. This was to control as far as possible for coverage bias, i.e. the mutations from a sample with a particularly high coverage would be over-represented in the counts. Considering only unique mutations approximated the probability of each indel length after a double-strand break repair event. For example, a mutation occurring early, for instance at the two-cell stage, would then be replicated many times across cell divisions. The proportion of such a mutation in the final dataset would be high but would not necessarily reflect how likely this indel length was to occur during the repair of the Cas9-induced double-strand break. The counts of unique mutations from all samples were pooled then tallied by length. The frequencies in Figure 2C are the proportions of unique indels of these lengths in the final dataset. There is likely a modest bias against large indels, as they may be missed by the short MiSeq reads or may disrupt a PCR primer binding site and therefore not be amplified. Conversely, the frequency of the larger deletions may be slightly over-estimated due to PCR length bias (Dabney and Meyer, 2012).

Positions of deleted nucleotides

Request a detailed protocol

This refers to Figure 2—figure supplement 1. Reads from control larvae were not used in this analysis. Only unique deletions in each sample were considered here. For each indel, ampliCan provides the start and end positions in relation to the PAM position, i.e. the PAM nucleotide adjacent to the gRNA binding site is set at 0. If the gRNA binding site is on the positive strand, negative positions are on the 5'-side of the first PAM nucleotide and positive positions to the 3'-side of the first PAM nucleotide. If the gRNA binding site is on the negative strand, negative positions are on the 3'-side of the first PAM nucleotide and positive positions to the 5'-side of the first PAM nucleotide.

Sanger sequencing

Request a detailed protocol

Sanger sequencing was performed to detect large deletions between targeted loci of slc24a5 (Figure 2E). The same PCR primers as for MiSeq were used but were selected to amplify the whole region either between the first and second loci (B to D), or the second and third (D to A), or the first and third (B to A). Each PCR well contained: 9.4 µL PCR mix (as described above), 0.25 µL forward primer (100 µM), 0.25 µL reverse primer (100 µM), 0.1 µL Taq DNA polymerase (ThermoFisher), 1 µL genomic DNA (same lysates as used for Illumina MiSeq); for a total of 11 µL. PCR program was: 95°C – 5 min, then 40 cycles of: 95°C – 30 s, 60°C – 30 s, 72°C – 2 min, then 72°C – 10 min and cooled to 10°C until collection. The PCR product was verified on a 1% agarose gel by loading 2.5 µL of PCR product with 0.5 µL of loading dye (6×), with 2.5 µL of 100 bp DNA ladder (100 ng/µL, ThermoFisher) ran alongside. PCR products were then purified with the QIAquick PCR Purification Kit (Qiagen) and their concentrations were quantified with Qubit (dsDNA High Sensitivity Assay). Samples were sent to Source Bioscience for Sanger sequencing. Sanger traces in ab1 format were aligned to the reference amplicon by MAFFT v7 (Katoh and Standley, 2013) ran through Benchling (benchling.com). Traces included in Figure 2D were exported from Benchling.

Headloop PCR

View detailed protocol

Headloop PCR (Rand et al., 2005) was adapted to test for gRNA activity at target loci by suppressing wild-type haplotype amplification. This was achieved by adding a 5′ tag to a primer that contained the reverse complement of the target sequence. After second strand elongation, the headloop tag is able to bind to the target sequence in the same strand, directing elongation and formation of a stable hairpin. If the target sequence is mutated, the headloop tag cannot bind and the amplicon continues to be amplified exponentially. Assessment of the headloop PCR products on an agarose gel was sufficient to determine if a target locus had been efficiently mutated in F0 embryos.

Headloop assays were based on primer pairs used for MiSeq. The headloop tag, containing the reverse complement target sequence, replaced the MiSeq tag on one of the primers in a pair (Figure 3—figure supplement 3B). The tag sequence was selected so that: (1) the predicted Cas9 cut site would occur within the first 6 bp of the tag; (2) that it did not contain any SNPs; and (3) matched the GC-content and annealing temperature of the base primers as closely as possible. If the tagged primer and gRNA binding sequence were in the same direction, the reverse complement of the gRNA binding sequence was usually sufficient as headloop tag, with adjustments for GC-content and Tm, if necessary.

Sequences of the headloop PCR primers are provided in Supplementary file 1. A Python-based tool to help with the design of headloop primers is available at: https://github.com/GTPowell21/Headloop (Powell, 2020).

For headloop PCR, each well contained: 5 µL 5× Phusion HF buffer (ThermoFisher), 17.25 µL dH2O, 0.5 µL dNTPs (10 mM), 0.5 µL forward primer (10 µM), 0.5 µL reverse primer (10 µM), 0.25 µL Phusion Hot Start II polymerase (ThermoFisher), 1 µL genomic DNA (same lysates as used for Illumina MiSeq); for a total of 25 µL. PCR amplification was performed using an Eppendorf MasterCycler Pro S PCR machine. When REDTaq was used (Figure 3—figure supplement 3A), each PCR well contained: 10 µL REDTaq ReadyMix (Sigma-Aldrich), 8.2 µL dH2O, 0.4 µL forward primer (10 µM), 0.4 µL reverse primer (10 µM), 1 µL genomic DNA (same lysates as used for Illumina MiSeq); for a total of 20 µL. In all cases, PCR program was: 98°C – 90 s; then 30 cycles of: 98°C – 15 s, 60°C – 15 s, 72°C – 15 secs; then 72°C – 5 min. The number of PCR cycles was limited to 30 to identify poor performing gRNAs; this threshold could be adjusted as required. Amplification was assessed by agarose gel electrophoresis (Figure 3B, Figure 3—figure supplements 1A, 2B and 3A).

To calculate the headloop PCR score (Figure 3—figure supplement 1), the gels were imaged using a GelDoc Go Gel Imaging System (Bio-Rad). Band intensities were then quantified using the software Quantity One (Bio-Rad).

For detection of the small deletion in genes apoea and cd2ap (Figure 3—figure supplement 2), the headloop and standard PCR reactions were performed as above, except for the PCR amplifying apoea, which needed 35 cycles to obtain a robust signal on an agarose gel. PCR products were then sent to Source Bioscience for Sanger sequencing. Sanger traces were manually inspected for mixed peaks at the mutated locus. Traces included in Figure 3—figure supplement 2 were exported from Benchling.

Mustard oil assay

Request a detailed protocol

trpa1b F0 knockouts were generated as described above. At 4 dpf, 10 trpa1b F0 knockout and 10 scrambled-injected control larvae were placed into the lids of two 35-mm Petri dishes filled with 7.5 mL of fish water. After a few minutes, 5 mL of 1 µM mustard oil (allyl isothiocyanate, Sigma-Aldrich) were added to the dishes with a Pasteur pipette (final concentration 0.66 µM) and left for a few minutes to observe the response. Figure 5—video 1 was recorded with a custom-built behavioural setup described previously (Dreosti et al., 2015).

For quantification (Figure 5A), 24 trpa1b F0 knockout and 24 scrambled-injected control larvae were placed in individual wells of a mesh-bottom 96-well plate (Merck), with the receiver plate filled with fish water. After a few minutes, the mesh-bottom plate was transferred to a second receiver plate filled with 1 µM mustard oil and left for a few minutes to observe the response. Tracking was performed by a ZebraBox (ViewPoint Behavior Technology), as described below (see Behavioural video tracking). Upon inspection of the video, three larvae (2 trpa1b F0 and 1 scrambled-injected larvae) were excluded from subsequent analysis because a bubble had formed in the mesh-bottom of the wells. The activity trace (Figure 5A) was smoothed with a 60-second rolling average.

per3:luciferase assay

Request a detailed protocol

Progeny of a homozygous Tg(per3:luc)g1, Tg(elavl3:EGFP)knu3 incross (Kaneko and Cahill, 2005) were injected at the single-cell stage with RNPs targeting csnk1db. At 4 dpf, using a P1000 pipet set at 150 μL with a tip whose end was cut-off, individual larvae were transferred to a white 96-round well plate (Greiner Bio-One). No animals were added in the last two columns of wells to serve as blanks. 50 mM (100×) Beetle luciferin (Promega) in water was mixed with 0.1% DMSO in water or 0.1 mM PF-670462 (Sigma-Aldrich) in DMSO to obtain a 4× luciferin/4 μM PF-670462 or 4× luciferin/0.004% DMSO solution. 50 μL of this solution was added on top of each well. Blank wells were topped with the luciferin/DMSO solution. Final concentrations in the wells were: luciferin 0.5 mM; DMSO 0.001%; PF-670462 1 µM. The plate was sealed and transferred to a Packard NXT Topcount plate reader (Perkin Elmer). Recording was performed in constant dark during 123 hr, starting around 12 noon the first day, or CT3, i.e. 3 hr after the last Zeitgeber. Temperature in the room was 25–28°C.

Circadian data analysis

Request a detailed protocol

The light intensity emitted from each well was collected by the Topcount plate reader every 9.92 min in counts-per-second (cps). After formatting the raw Topcount data in R, the data were imported in BioDare2 (https://biodare2.ed.ac.uk/) (Zielinski et al., 2014). The average light level from the blank wells was used for background subtraction. Six larvae (1 scrambled + DMSO, 4 csnk1db F0 + DMSO, 1 csnk1db F0 + PF-670462) were excluded for subsequent analysis; five upon inspection of the timeseries because their traces showed sudden changes in amplitude or dampened cycling and one because no satisfactory fit could be found during period analysis (see below).

For period analysis, the timeseries was cropped to start 24 hr after the end of the last partial LD cycle (CT48 in Figure 5B) to analyse the circadian rhythm in free running conditions. Period analysis was performed on the cropped timeseries using the algorithm Fast Fourier Transform Non-Linear Least Squares (FFT NLLS) (Plautz et al., 1997) ran through BioDare2 (Zielinski et al., 2014).

The FFT NLLS algorithm fitted a cosine to the timeseries and extracted from the model the period that lies within a user-defined range of likely circadian periods. As the measure can be sensitive to this range, and as it was evident from the timeseries that the period was massively different between larvae treated with DMSO and larvae treated with PF-670462, the two groups were processed separately. The window of likely period lengths was first set to 18–32 hr (25 ± 7 hr) for all the fish treated with DMSO, that i.e. scrambled + DMSO and csnk1db F0 + DMSO. It was then set to 28–42 hr (35 ± 7 hr) for all the fish treated with PF-670462. Any equivocal period length was labelled by BioDare2 and the cosine fit was manually inspected. As mentioned above, one larva was excluded upon inspection as no fit could be found. The period length for each animal were exported from BioDare2 and plotted as a stripchart in R (Figure 5B bottom).

After period analysis, BioDare2’s amplification and baseline detrending and normalisation to the mean were applied to the timeseries. The detrended and normalised timeseries were exported from BioDare2 and plotted in R (Figure 5B top). Traces were smoothed with a 20-data point (~ 198 min) rolling average and were artificially spread over the Y axis so they would not overlap.

scn1lab stable knockout line

Request a detailed protocol

The scn1labΔ44 stable knockout line was generated using zinc-finger nucleases (ZFNs). A CompoZr ZFN was designed by Sigma-Aldrich to target exon 4 of scn1lab. CompoZr ZFN contained a DNA-binding domain and an obligate-heterodimer Fok1 nuclease domain, engineered for improved specificity (Miller et al., 2007). Activity of ZFN pairs as determined by the yeast MEL-1 reporter assay (Doyon et al., 2008) was 113.6%. ZFNs were prepared and used as previously described (Hoffman et al., 2016). We confirmed by Sanger sequencing the presence of a 44-nucleotide deletion in scn1lab exon 4 (chr6:10,299,906–10,299,949) and two SNPs: T>A at chr6:10,299,903 and T>C at chr6:10,299,904 (danRer11). The deletion includes the intron/exon four boundary.

ZFNs binding sequences and sequences of the PCR primers used for sequencing and genotyping are provided in Supplementary file 1.

Behavioural video tracking

Request a detailed protocol

For the F0 scn1lab knockout experiments (Figure 6 and Figure 6—figure supplement 1), wild-type embryos from two separate clutches were injected at the single-cell stage with RNPs targeting scn1lab. At 5 dpf, individual larvae were transferred to the wells of clear 96-square well plates (Whatman). To avoid any potential localisation bias during the tracking, conditions were alternated between columns of the 96-well plates. The plates were placed into two ZebraBoxes (ViewPoint Behavior Technology). From each well we recorded the number of pixels that changed intensity between successive frames. This metric, which we term Δ pixels, describes each animal’s behaviour over time as a sequence of zeros and positive values, denoting if the larva was still or moving. Tracking was performed at 25 frames per second on a 14 hr:10 hr light:dark cycle with the following ViewPoint parameters: detection sensitivity = 20, burst = 100, freezing = 3. Larvae were tracked for around 65 hr, generating sequences of roughly 5,850,000 Δ pixel values per animal. The day light level was calibrated at 125 μW with a Macam PM203 Optical Power Meter set at 555 nm. Evaporated water was replaced both mornings shortly after 9 am. At the end of the tracking, any larva unresponsive to a light touch with a P10 tip was excluded from subsequent analysis.

For the scn1lab stable knockout line experiment (Figure 6A), larvae were the progeny of a scn1lab+/Δ44 (heterozygous) incross. Behavioural tracking was performed as above, with the following amendments: the experiment started at 4 dpf; recording was performed at 15 frames per second; evaporated water was replaced both days around 2 pm, which created an artefactual drop followed by a peak in activity (Figure 6A).

Behavioural data analysis

Request a detailed protocol

Behavioural data were processed and analysed as previously described (Ghosh and Rihel, 2020). In brief, the raw file generated by the ZebraLab software (ViewPoint Behavior Technology) was exported into thousands of xls files each containing 50,000 rows of data. These files, together with a metadata file labelling each well with a condition, were input to the MatLab scripts Vp_Extract.m and Vp_Analyse.m (included in the GitHub and Zenodo repositories). To visualise larval activity over time, we summed Δ pixel changes into one-second bins and plotted the mean and standard error of the mean across larvae, smoothed with a 15-min rolling average. We considered the first day and night as a habituation period, and cropped these from all timeseries. For the scn1lab stable knockout line, the traces start at 9 am (lights on, ZT0) of 5 dpf (Figure 6A). For the scn1lab F0 knockout experiments, the traces start at 9 am of 6 dpf (Figure 6A and Figure 6—figure supplement 1). To quantify differences in behaviour between genotypes, we extracted 10 day and night behavioural parameters per larva: (1) active bout length (seconds); (2) active bout mean (Δ pixels); (3) active bout standard deviation (Δ pixels); (4) active bout total (Δ pixels); (5) active bout minimum (Δ pixels); (6) active bout maximum (Δ pixels); (7) number of active bouts; (8) total time active (%); (9) total activity (Δ pixels); and (10) inactive bout length (seconds). To compare F0 and stable line scn1lab mutant behaviour, we calculated the deviation (Z-score) of each mutant from their wild-type siblings across all parameters. We term these vectors behavioural fingerprints. We compared fingerprints across groups using both Pearson correlation and the Euclidean distance between each larva and its mean wild-type sibling fingerprint (Figure 6A). The statistical test used to compare effect sizes is a Z-test used for comparing two studies in meta-analysis (Borenstein et al., 2009). As a control, in the stable knockout experiment, the effect size between wild types and heterozygotes (Cohen’s d = 0.28) was significantly different (p = 0.04) than the effect size between wild types and homozygotes (Cohen’s d = 1.57).

Sample size simulations

Request a detailed protocol

This refers to Figure 7B. For trpa1b, each larva’s response to mustard oil was first summarised as the difference between the total activity (sum of Δ pixels/frame) during the first 3 min of exposure to mustard oil and the total activity during the 3 min just before switching the plates. Upon inspection of the density plots of these delta values, 3 trpa1b F0 animals that responded like scrambled controls and 1 scrambled control that did not respond to mustard oil (Figure 5A) were excluded to fit idealised normal distributions. For csnk1db, the values were the individual period lengths of the larvae treated with DMSO (scrambled + DMSO and csnk1db F0 + DMSO). The means and standard deviations of the resulting scrambled and knockout groups were used to fit normal distributions. At each simulation, 100 F0 knockout and 100 scrambled larvae were simulated. At the first simulation, the 100 values (delta activity or period lengths) of the F0 knockout group were randomly sampled from the scrambled normal distribution, simulating an experiment where all F0 eggs were missed during injections and are thus wild types (0% success rate). At the second simulation, 99 values of the F0 knockout group were randomly sampled from the scrambled normal distribution and 1 value was randomly sampled from the knockout normal distribution, simulating a 1% success rate at injections. As simulations progressed, more knockout larvae were gradually added to the F0 group, simulating improving success rates at injections. The simulations ended at the 101th iteration, where the 100 delta values of the F0 knockout group are sampled from the knockout distribution, simulating an ideal experiment where all the larvae in the F0 knockout group are biallelic knockouts (100% success rate). In all the 101 simulations, the data of the scrambled group were sampled every time from the scrambled distribution. At each simulation, Cohen’s d effect size was calculated then used to compute the minimum sample size for detection at 0.05 significance level and 0.8 statistical power. As simulations progressed, the F0 knockout and scrambled data gradually had less overlap, hence increasing effect size and decreasing the minimum sample size needed to detect the phenotype. The 101 simulations were iterated 10 times to produce error bars.

Pictures

Request a detailed protocol

Pictures of embryos in Figure 1C,D,H,I; Figure 4A,B; Figure 6A; Figure 1—figure supplement 1 were taken with an Olympus MVX10 microscope connected to a computer with the software cellSens (Olympus). A black outline was added around the embryos in Figure 4A,B.

Pictures of slc24a5 F0 adults (Figure 1G) were taken with a Canon 650D with a Sigma 30 mm f/1.4 DC HSM lens.

Statistics

Threshold for statistical significance was 0.05. In figures, ns refers to p > 0.05, * to p ≤ 0.05, ** to p ≤ 0.01, *** to p ≤ 0.001. In text, data distributions are reported as mean ± standard deviation, unless stated otherwise.

In Figure 2B, the numbers of indel lengths in common when intersecting the top 10 of two samples from different loci or two samples from the same locus but different fish were compared by Welch’s t-test.

In Figure 5A, each animal’s response to mustard oil was first summarised as the difference in total activities (as in Sample size simulations). The delta values from the scrambled controls and the trpa1b F0 knockout were then compared by Welch’s t-test.

In Figure 5B, the circadian periods were first compared by a one-way ANOVA, then the values from each group were compared to one another by pairwise Welch’s t-tests with Holm’s p-value adjustment method.

To compare the activity of scn1lab knockouts (F0 or stable) with their wild-type siblings, we statistically compared the total time active (%) parameter between genotypes within each experiment (F0 experiment 1, F0 experiment 2, stable knockout line) using a two-way ANOVA with condition (knockout and wild-type) and time (day and night) as interaction terms.

In Figure 6C, the Euclidean distances were first compared by a one-way ANOVA, then the values from each group were compared to one another by pairwise Welch’s t-tests with Holm’s p-value adjustment method.

Software

Data analysis was performed in R v3.6.2 ran through RStudio v1.2.5033 and MATLAB R2018a (MathWorks). Figures were prepared with Adobe Illustrator CC 2018 and assembled with Adobe InDesign CC 2018. Figure 4—videos 1 and 2 and Figure 5—video 1 were trimmed and annotated with Adobe Premiere Pro CC 2019.

R, MatLab, and command line packages used throughout this study are listed in Supplementary file 2.

Data/resource sharing

Request a detailed protocol

Data and code are available at https://github.com/francoiskroll/f0knockout (Kroll, 2020; copy archived at swh:1:rev:6d7db3aa702a5bad79fe36a800163e3f76705c4f) and on Zenodo at https://doi.org/10.5281/zenodo.3898915.

A protocol describing how to generate F0 knockout larvae for a single gene is available at dx.doi.org/10.17504/protocols.io.bfgyjjxw.

A Python-based tool to help with the design of headloop primers is available at https://github.com/GTPowell21/Headloop (Powell, 2020; copy archived at swh:1:rev:39ef51ce7b98b787f8efd54318561ac5ee44df65).

Data availability

Data and code can be found at https://github.com/francoiskroll/f0knockout (copy archived at https://archive.softwareheritage.org/swh:1:rev:6d7db3aa702a5bad79fe36a800163e3f76705c4f/) and as a repository on Zenodo: https://doi.org/10.5281/zenodo.3898915.

The following data sets were generated

References

  1. Book
    1. Borenstein M
    2. Hedges LV
    3. Higgins JPT
    (2009) Subgroup Analyses
    In: Rothstein H. R, editors. Introduction to Meta‐Analysis. Wiley Online Books. pp. 19–24.
    https://doi.org/10.1002/9780470743386.ch19
    1. Garrity DM
    2. Childs S
    3. Fishman MC
    (2002)
    The heartstrings mutation in zebrafish causes heart/fin Tbx5 deficiency syndrome
    Development 129:4635–4645.
    1. Howe K
    2. Clark MD
    3. Torroja CF
    4. Torrance J
    5. Berthelot C
    6. Muffato M
    7. Collins JE
    8. Humphray S
    9. McLaren K
    10. Matthews L
    11. McLaren S
    12. Sealy I
    13. Caccamo M
    14. Churcher C
    15. Scott C
    16. Barrett JC
    17. Koch R
    18. Rauch GJ
    19. White S
    20. Chow W
    21. Kilian B
    22. Quintais LT
    23. Guerra-Assunção JA
    24. Zhou Y
    25. Gu Y
    26. Yen J
    27. Vogel JH
    28. Eyre T
    29. Redmond S
    30. Banerjee R
    31. Chi J
    32. Fu B
    33. Langley E
    34. Maguire SF
    35. Laird GK
    36. Lloyd D
    37. Kenyon E
    38. Donaldson S
    39. Sehra H
    40. Almeida-King J
    41. Loveland J
    42. Trevanion S
    43. Jones M
    44. Quail M
    45. Willey D
    46. Hunt A
    47. Burton J
    48. Sims S
    49. McLay K
    50. Plumb B
    51. Davis J
    52. Clee C
    53. Oliver K
    54. Clark R
    55. Riddle C
    56. Elliot D
    57. Eliott D
    58. Threadgold G
    59. Harden G
    60. Ware D
    61. Begum S
    62. Mortimore B
    63. Mortimer B
    64. Kerry G
    65. Heath P
    66. Phillimore B
    67. Tracey A
    68. Corby N
    69. Dunn M
    70. Johnson C
    71. Wood J
    72. Clark S
    73. Pelan S
    74. Griffiths G
    75. Smith M
    76. Glithero R
    77. Howden P
    78. Barker N
    79. Lloyd C
    80. Stevens C
    81. Harley J
    82. Holt K
    83. Panagiotidis G
    84. Lovell J
    85. Beasley H
    86. Henderson C
    87. Gordon D
    88. Auger K
    89. Wright D
    90. Collins J
    91. Raisen C
    92. Dyer L
    93. Leung K
    94. Robertson L
    95. Ambridge K
    96. Leongamornlert D
    97. McGuire S
    98. Gilderthorp R
    99. Griffiths C
    100. Manthravadi D
    101. Nichol S
    102. Barker G
    103. Whitehead S
    104. Kay M
    105. Brown J
    106. Murnane C
    107. Gray E
    108. Humphries M
    109. Sycamore N
    110. Barker D
    111. Saunders D
    112. Wallis J
    113. Babbage A
    114. Hammond S
    115. Mashreghi-Mohammadi M
    116. Barr L
    117. Martin S
    118. Wray P
    119. Ellington A
    120. Matthews N
    121. Ellwood M
    122. Woodmansey R
    123. Clark G
    124. Cooper J
    125. Cooper J
    126. Tromans A
    127. Grafham D
    128. Skuce C
    129. Pandian R
    130. Andrews R
    131. Harrison E
    132. Kimberley A
    133. Garnett J
    134. Fosker N
    135. Hall R
    136. Garner P
    137. Kelly D
    138. Bird C
    139. Palmer S
    140. Gehring I
    141. Berger A
    142. Dooley CM
    143. Ersan-Ürün Z
    144. Eser C
    145. Geiger H
    146. Geisler M
    147. Karotki L
    148. Kirn A
    149. Konantz J
    150. Konantz M
    151. Oberländer M
    152. Rudolph-Geiger S
    153. Teucke M
    154. Lanz C
    155. Raddatz G
    156. Osoegawa K
    157. Zhu B
    158. Rapp A
    159. Widaa S
    160. Langford C
    161. Yang F
    162. Schuster SC
    163. Carter NP
    164. Harrow J
    165. Ning Z
    166. Herrero J
    167. Searle SM
    168. Enright A
    169. Geisler R
    170. Plasterk RH
    171. Lee C
    172. Westerfield M
    173. de Jong PJ
    174. Zon LI
    175. Postlethwait JH
    176. Nüsslein-Volhard C
    177. Hubbard TJ
    178. Roest Crollius H
    179. Rogers J
    180. Stemple DL
    (2013) The zebrafish reference genome sequence and its relationship to the human genome
    Nature 496:498–503.
    https://doi.org/10.1038/nature12111
    1. Kelsh RN
    2. Brand M
    3. Jiang YJ
    4. Heisenberg CP
    5. Lin S
    6. Haffter P
    7. Odenthal J
    8. Mullins MC
    9. van Eeden FJ
    10. Furutani-Seiki M
    11. Granato M
    12. Hammerschmidt M
    13. Kane DA
    14. Warga RM
    15. Beuchle D
    16. Vogelsang L
    17. Nüsslein-Volhard C
    (1996)
    Zebrafish pigmentation mutations and the processes of neural crest development
    Development 123:369–389.
    1. Lister JA
    2. Robertson CP
    3. Lepage T
    4. Johnson SL
    5. Raible DW
    (1999)
    Nacre encodes a zebrafish microphthalmia-related protein that regulates neural-crest-derived pigment cell fate
    Development 126:3757–3767.
    1. Streisinger G
    2. Singer F
    3. Walker C
    4. Knauber D
    5. Dower N
    (1986)
    Segregation analyses and gene-centromere distances in zebrafish
    Genetics 112:311–319.

Decision letter

  1. Stephen C Ekker
    Reviewing Editor; Mayo Clinic, United States
  2. Didier YR Stainier
    Senior Editor; Max Planck Institute for Heart and Lung Research, Germany
  3. Darius Balciunas
    Reviewer; Temple University, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

Your method is a nice next step in using gene editing technology for exploring functional genomics. Your application of this approach in the areas of behavioural science is especially noteworthy, especially for the adult zebrafish as a model system.

Decision letter after peer review:

Thank you for submitting your article "A simple and effective F0 knockout method for rapid screening of behaviour and other complex phenotypes" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Didier Stainier as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Darius Balciunas (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

As the editors have judged that your manuscript is of interest, but as described below that additional experiments are required before it is published, we would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option. (If your work is more suitable for medRxiv, you will need to post the preprint yourself, as the mechanisms for us to do so are still in development.)

Summary

Kroll and colleagues describe a new efficient strategy to reliably generate F0 zebrafish embryos with (multiple) genes knocked out using CRISPR/Cas9 RNPs. They showed that in addition to target single genes, this method could be successfully used to create double knockouts of slc24a5 and tbx5a gene pair, or tyr and ta gene pair, in F0 embryos. Strikingly, they also demonstrated direct generation of triple gene knockouts of mitfa, mpv17 and slc45a2 in F0 larvae, which fully recapitulated the pigmentation defects of the crystal mutant. As the authors point out, their methodology is extremely likely to be adapted for candidate genes for traits which display a range of phenotypes among wild type embryos or larvae.

The manuscript points out a rather obvious but somehow underreported feature of NHEJ-based mutagenesis: assuming random size of indels, when 100% of DNA is mutated fewer than 50% (.67x.67) of cells in an embryo will contain frameshift mutations in both alleles. Thus, successful recapitulation of a mutant phenotype in an F0 embryo relies on mutagenesis of an essential part of the protein (not always as straightforward as it seems), utilization of other repair pathways such as MMEJ (not always reliable), or fortuitous help from largely unknown factors which skew the distribution of indel sizes (multiple guide would RNAs need to be tested without guarantee of success). Simultaneously designing several guide RNAs against the gene and co-injecting them, as the authors propose, seems to be an excellent and straightforward strategy.

They established a rapid sequencing-free method to evaluate the activity of Cas9 RNP by using headloop PCR, facilitating the selection of target sites. This is a new tool for the zebrafish community.

Despite the presented data on several loci, it is not clear whether and how this method is better compared to a series of prior related F0 approaches. This question is the crux of this method manuscript.

Essential revisions

1) The authors need a specific direct comparison with prior reports, notably Wu et al. in 2018. Several genes were tested in both work, such as slc24a5, tyr, tbx16, and tbx5a, did you use or compare the same target sites in these genes as reported by Wu et al.?

2) The second major consideration in the field is validating F0 somatic mosaic results with non-mosaic outcomes in prospective loci. Replicating known prior phenotypes is an important first step. But clearly validating new loci where the outcome is unknown is the key challenge in the field and the bar by which these methods will be judged. N=1 locus data has many questions – was this gambler's luck (i.e. they were fortunate the first locus they tried worked)? Did they try others and not have them work?

Additional points

1) Successful multiplex targeting has already been achieved in zebrafish, including the Figure 6 in Jao et al., 2013 reference. This needs to be acknowledged and elaborated upon (different efficiencies, etc.).

2) The statement that "The common strategy is to inject ... RNP..." excludes a significant number of laboratories which prefer to inject Cas9 RNA. The proposed three-guide method should work just as well with Cas9 RNA.

3) The data in Figure 1—figure supplement 1 seems to show that relative concentration of functional Cas9 protein is rate-limiting, perhaps even at the highest 1:1 ratio. Statement that 1:1 ratio is "optimal" (page 7) implies that reduction in the amount of guide RNAs would lead to reduced penetrance of the phenotype, which may or may not be the case.

4) The observation that 41/41 adult slc24a5 fish displayed golden phenotype suggests that only pigmentation-negative embryos (perhaps the 63/67) were raised to adulthood. Please clarify.

5) I am not convinced that headloop PCR is sufficiently quantitative for assessment of guide RNAs for an F0 assay. What is the minimum mutagenesis rate needed to obtain a "positive" PCR result and does it vary between loci? For example, if a specific gRNA produces 30% indels, would it score as positive in Headloop PCR? Assuming 67% frameshift probability, such guide RNA would only produce about 4% (0.3 x 0.67 x 0.3 x 0.67) of biallelically mutated cells and be therefore quite useless in an F0 assay. This analysis can be performed by mixing wild type and mutant DNA in different ratios.

6) Is the dosage/amount of Cas9 or RNP used in this study different or comparable with Wu et al.? Does it account for the improvement of the method described in the study?

7) The authors propose to design the three target sites in distinct exon within each gene. Is it really important and/or necessary to achieve high efficient biallelic knockouts? Any evidence?

8) According to the section of 'Materials and methods', the synthetic gRNA was made of two components, i.e., crRNA and tracrRNA. Synthesis of gRNA as a single molecule by in vitro transcription is usually more popular and economic, is it really necessary to use crRNA and tracrRNA to achieve high efficient biallelic knockouts? Any evidence?

9) Could headloop PCR be used for the quantification of mutagenesis efficiency (indel-producing mutation rate) of Cas9/gRNA? How sensitive is this method? Could small indels (such as 1-bp insertion or deletion) be detected by the headloop PCR?

10) In addition to indels, deletions between two double strand breaks induced by two gRNAs are also important for the generation of biallelic knockouts of the target gene. The authors showed the analysis of mutations in each site (such as in Figure 2A), is it possible to quantify the distribution and contribution of all the different deletions?

11) Figure 1C and 1D: The authors compared the effects of the injection of 1, 2, 3, and 4 loci. How were the 1, 2, and 3 loci selected from the four target sites? Will each of the four loci give the same or different phenotypic ratio if tested individually? Will different combinations of 2 loci or 3 loci give the same or different phenotypic ratio? Or which combination of 2 loci or 3 loci will give the highest mutagenic effect? For example, in Figure 1C, the 3-loci showed comparable effect with 4-loci, while the 2-loci is less effective; is it possible to find other 2-loci combinations which could show higher mutagenic efficiency than the current 2-loci, such that the effect of the new 2-loci combination is as good as the 3-loci or 4-loci combination? Conversely, in Figure 1D, the 2-loci already showed the highest mutagenic effect, is it because of this particular 2-loci combination, or any 2-loci combination will show the same efficiency?

12) Figure 6: The phenotypes of scn1lab F0 knockouts are more severe than those of scn1lab-/- mutant. Any explanation?

13) Please provide the academic name of zebrafish in its first appearance.

https://doi.org/10.7554/eLife.59683.sa1

Author response

Essential revisions

1) The authors need a specific direct comparison with prior reports, notably Wu et al., 2018. Several genes were tested in both work, such as slc24a5, tyr, tbx16, and tbx5a, did you use or compare the same target sites in these genes as reported by Wu et al.?

We obtained strikingly similar results of phenotypic penetrance as Wu et al., 2018 for common genes tested (more details below), but targeting three loci per gene rather than four. This is favourable as it reduces potential off-target effects and may reduce unviability in the injected embryos (more details below). We did not target the same loci within each gene as Wu et al., 2018.

We added a paragraph in the Materials and methods (section Cas9/gRNA preparation) outlining the key differences in protocols with Wu et al., 2018. We used synthetic gRNAs (crRNA:tracrRNA duplexes), as opposed to in vitro-transcribed single-molecule gRNAs (sgRNAs); we targeted three loci per gene, as opposed to four; we injected 28.5 fmol (1000 pg) total gRNA and 28.5 fmol (4700 pg) Cas9 (1 Cas9 to 1 gRNA), as opposed to 28.5 fmol (1000 pg) total sgRNA and 4.75 fmol (800 pg) Cas9 (1 Cas9 to 6 gRNA) reported in Wu et al., 2018.

We obtained comparable results of phenotypic penetrance; Wu et al., 2018 versus present work:

  • slc24a5 (percentage of larvae without eye pigmentation): ~ 91% vs. 96%

  • tyr (percentage of larvae without eye pigmentation): ~ 94% vs. 99%

  • tbx16 (percentage of larvae displaying the spadetail phenotype): 100% vs. 100%

  • tbx5a (percentage of larvae lacking both pectoral fins): ~ 98% vs. 100%

  • slc24a5 and tbx5a (percentage of larvae lacking both pectoral fins and without eye pigmentation): ~ 93% vs. 93%.

Details about these comparisons are included in the GitHub and Zenodo repositories (wu_phenotypecomparisons.xlsx). We have also added a few sentences in the Results (sections Three synthetic gRNAs [...] and Multiple genes [...]) to highlight these comparable results.

It is more precarious to compare the adverse effects caused by injections between the present study and Wu et al., 2018 as we did not quantify them in the same manner. First, we understand the metric used by Wu et al. (percentage dysmorphic) as not including embryos found dead. The metric we used, which we termed unviability, included dead embryos found any time after 1 dpf (see Materials and methods, section Unviability). Second, Wu et al. tallied the number of dysmorphic embryos at the time of phenotyping, which seemed to be 2 dpf for the experiments targeting slc24a5, tyr, tbx16, tbx5a, or slc24a5 and tbx5a. We followed the larvae until 5–6 dpf. During our experiments, many unviable embryos were categorised as such because they failed to develop a swim bladder, but this can only be observed at 4–5 dpf. For both of these reasons, it is expected that the unviability levels we reported would be higher than the percentage dysmorphic reported in Wu et al., 2018, even if the experiments were identical.

Nonetheless, we can attempt a comparison by counting the number of dysmorphic embryos we found at 2 dpf during our experiments. Accordingly, it seems likely that we achieved lower percentages of dysmorphic embryos:

Wu et al., 2018 versus present work:

  • slc24a5: ~ 8% vs. ~ 2% (2/162)

  • tyr: ~ 10% vs. ~ 2% (4/215)

  • tbx16: as homozygous tbx16 knockouts have various trunk defects and are lethal early in development (Ho and Kane, 1990), we did not quantify unviability in the tbx16 F0 knockouts

  • tbx5a: ~ 8% vs. ~ 0% (0/43)

  • slc24a5 and tbx5a: ~ 11% vs. ~ 1% (1/29).

Details about these comparisons can be found in the GitHub and Zenodo repositories (wu_unviabilitycomparisons.xlsx).

We did not use or compare the same target sites as Wu et al., 2018. The gRNAs used in the present work were designed independently and by different algorithms. Wu et al. used CRISPR Design, a tool created by the Zhang Lab, which does not exist anymore (crispr.mit.edu). We used IDT’s tool, available on their website (eu.idtdna.com/site/order/designtool/index/CRISPR_CUSTOM). We searched for any fortuitous overlap. We used a total of 15 gRNAs targeting slc24a5, tyr, tbx16, or tbx5a; Wu et al. used 32. Four of the gRNAs we used (4/15) had some but incomplete overlap in their genomic binding sites with gRNAs from Wu et al. We also found that tbx5a locus D from the present work was the same as Wu et al.’s tbx5a locus 2 (Wu et al., 2018, Supplementary file 1). Details about the overlapping gRNAs are included in the GitHub and Zenodo repositories (wu_commonLoci.xlsx).

2) The second major consideration in the field is validating F0 somatic mosaic results with non-mosaic outcomes in prospective loci. Replicating known prior phenotypes is an important first step. But clearly validating new loci where the outcome is unknown is the key challenge in the field and the bar by which these methods will be judged. N=1 locus data has many questions – was this gambler's luck (i.e. they were fortunate the first locus they tried worked)? Did they try others and not have them work?

We targeted a total of 12 genes and demonstrated the corresponding 12 phenotypes, which all reproduced published mutants. Therefore, we are slightly unclear what is meant by “N = 1 locus data”. We also note that the 12 genes/phenotypes we presented in the study are all the genes we have targeted during the development and testing of the protocol. In other words, we do not have any example where we failed to replicate a stable mutant phenotype.

We understand “N = 1” may be referring to the experiment targeting scn1lab (Figure 6) being the only direct, side-by-side comparison between F0 knockouts and stable line mutants in our study, with the others referring to known, published mutants. The majority of the phenotypes we have tested are overt morphological differences which are well described in the literature; we therefore do not believe including stable line mutants for comparison is indispensable.

Nonetheless, to address this by providing another direct comparison between F0 knockouts and a stable mutant line, we targeted the gene mab21l2, which is required for eye development, and compared the phenotype of the F0 knockouts with homozygous mab21l2u517 mutants (Wycliffe et al., 2020). We included these new results as Figure 1I and in the text (Results, sectionThree synthetic gRNAs [...]).

We believe that robustly discovering new phenotypes directly in F0 knockouts is the key application of the method, and indeed its main challenge. In fact, we present an example where the phenotype (outcome) was unknown – csnk1db. The phenotype (circadian clock period lengthening) was predicted from pharmacological inhibition and genetic knockout in other species (e.g. Meng et al., 2010). The circadian period of csnk1db knockout was never measured in zebrafish to our knowledge. Therefore, it demonstrates the validation of a novel phenotype in zebrafish F0 knockouts.

It is precisely because discovering novel phenotypes is the main challenge of F0 knockout methods that an essential criterion for success was achieving high phenotypic penetrance and close to complete removal of wild-type alleles for test genes. When studying phenotypes that vary continuously in the population, such as locomotor activity (Figure 6) or circadian clock period (Figure 5B), it is not possible to readily identify animals in the F0 knockout population that display a phenotype, in contrast to situations in which animals exhibit overt morphological differences. Therefore, the majority of the animals in the F0 knockout population need to be bona fide knockouts. Continuous traits often have the added challenge that they are not phenotypes whose spatial variation is evident in individual animals (Watson et al., 2020), in contrast to developmental phenotypes like formation of electrical synapses (Shah et al., 2015) or distribution of microglia (Kuil et al., 2019). Therefore, most cells in each F0 knockout animal must carry biallelic loss-of-function mutations. With the pigmentation phenotypes golden (slc24a5 knockout) and sandy (tyr knockout), and the deep sequencing data at 32 targeted sites, we demonstrate that the F0 knockouts generated by the present method show very little mosaicism, in the sense that most wild-type alleles are removed, and hence can be used to discover novel phenotypes that require most or all the cells of most F0 knockout animals to be mutated.

Additional points

1) Successful multiplex targeting has already been achieved in zebrafish, including the Figure 6 in Jao et al., 2013 reference. This needs to be acknowledged and elaborated upon (different efficiencies, etc.).

This is correct, and we note there are other important studies: Hoshijima et al., 2019; Keatinge et al., 2020; Kim and Zhang, 2020; Shah et al., 2015. Our first example (simultaneous targeting of slc24a5 and tbx5a, Figure 4A) was directly inspired from Wu et al., 2018. We realise this was not acknowledged properly and have corrected it (Results, section Multiple genes […]).

What constitutes “successful” multiplex targeting depends on the applications for which the method is developed. In Figure 6 of Jao et al., 2013, five genes were targeted simultaneously: ddx19, egfp (as part of a transgene), slc24a5 (golden), mitfa, and tyr, with one gRNA targeting each gene. Jao et al. concluded that multiplex targeting was successful because some embryos showed a combination of three phenotypes. The goal of the experiment was not to demonstrate high penetrance of each phenotype or close to complete removal of wild-type alleles for each gene. For example, it is not possible to tell whether eye pigmentation defects were caused by disruption of slc24a5 or tyr (as acknowledged by the authors); the proportion of embryos displaying all three phenotypes together was not reported; phenotypic penetrance in individual animals is incomplete (severe pigmentation phenotype as in Jao et al., 2013, Figure 6C,D would be scored as 2 in the present work); and the proportion of mutated alleles is below 50% for both slc24a5 (golden) and mitfa. As highlighted appropriately in the previous reviewer’s question, a key challenge is the application of F0 knockout methods to discover new phenotypes (“when the outcome is unknown”). If a gene which would give rise to a behavioural phenotype in a homozygous knockout is disrupted incompletely in only some of the injected F0 larvae, it is unlikely that the phenotype be discovered in F0 knockouts, as we explain through Figure 7B. For example, if trpa1b and csnk1db were targeted simultaneously in n = 48 F0 embryos, but only 40% of larvae were true csnk1db F0 knockouts, it is unlikely that one would discover the circadian clock phenotype of csnk1db (Figure 7B).

To summarise, the method we present here is specifically designed to be applied to screens studying continuous traits (like many behavioural parameters) or phenotypes whose spatial variation is not visible in individual animals. In that context, multiplex targeting is only successful if phenotype penetrance is high (i.e. the majority of injected larvae display all phenotypes completely) and most wild-type alleles of all targeted genes are removed.

2) The statement that "The common strategy is to inject ... RNP..." excludes a significant number of laboratories which prefer to inject Cas9 RNA. The proposed three-guide method should work just as well with Cas9 RNA.

We think it is unlikely that the present method would work as well with Cas9 mRNA. Comparisons of Cas9 mRNA vs pre‑assembled Cas9 protein/gRNA RNP were performed by Burger et al., 2016 (Figure 2D). The authors targeted egfp present in two different transgenes by co-injecting Cas9 mRNA and an in vitro-transcribed sgRNA or a pre‑assembled RNP (Cas9 protein + in vitro-transcribed sgRNA), then quantified intensity of the EGFP signal from the F0 embryos. Using the assembled RNP led to a greater loss of the EGFP signal compared to using Cas9 mRNA (~ 28% greater loss of EGFP signal vs Cas9 mRNA for the first egfp transgene, ~ 7% for the second). Burger et al. hypothesised that the limiting factor when using Cas9 mRNA may be either the rate of Cas9 translation or degradation of the sgRNA before it is packaged in the RNP (Burger et al., 2016). Therefore, we would advise against using Cas9 mRNA, especially when studying continuous, quantitative traits where high phenotypic penetrance and close to complete removal of the wild-type alleles are crucial. We have highlighted this point in Results (section Three synthetic gRNAs […]) and in Materials and methods (section gRNA/Cas9 assembly).

3) The data in Figure 1—figure supplement 1 seems to show that relative concentration of functional Cas9 protein is rate-limiting, perhaps even at the highest 1:1 ratio. Statement that 1:1 ratio is "optimal" (page 7) implies that reduction in the amount of guide RNAs would lead to reduced penetrance of the phenotype, which may or may not be the case.

This may well be the case. We have edited the text to raise this possibility (Results, section Three synthetic gRNAs […]).

4) The observation that 41/41 adult slc24a5 fish displayed golden phenotype suggests that only pigmentation-negative embryos (perhaps the 63/67) were raised to adulthood. Please clarify.

This is correct—only slc24a5 F0 larvae displaying the golden phenotype (scored as 1 for eye pigmentation at 2 dpf) were raised to adulthood. This is mentioned in Materials and methods (section Adult slc24a5 F0 fish), but we have edited the main text to make this clearer for the readers (Results, section Three synthetic gRNAs […]). To explain in more detail the rationale behind this experiment: Wu et al. observed that targeting s1pr2 at a single locus generated a high proportion of larvae with the expected phenotype at 1 dpf, but that this proportion would then decline by 4 dpf (Wu et al., 2018, Figure 1A), suggesting the animals may ‘recover’ from the phenotype. As potential applications of the present method include the study of quantitative traits such as behavioural phenotypes (i.e. typically after 4–5 dpf), we deemed pertinent to confirm that the phenotypes we were observing persisted throughout the life of the animal.

5) I am not convinced that headloop PCR is sufficiently quantitative for assessment of guide RNAs for an F0 assay. What is the minimum mutagenesis rate needed to obtain a "positive" PCR result and does it vary between loci? For example, if a specific gRNA produces 30% indels, would it score as positive in Headloop PCR? Assuming 67% frameshift probability, such guide RNA would only produce about 4% (0.3 x 0.67 x 0.3 x 0.67) of biallelically mutated cells and be therefore quite useless in an F0 assay. This analysis can be performed by mixing wild type and mutant DNA in different ratios.

To test whether headloop PCR can also be used to exclude ‘mediocre’ gRNAs (< 60% mutated reads), we derived a score from the band intensities on agarose gel (Figure 3—figure supplement 1A). The headloop score is the ratio between the headloop PCR band intensity and the standard PCR band intensity. As it represents a proportion of the standard PCR band intensity, it should largely control for the variation in PCR efficiency between loci. We measured this score for the targeted loci of embryos injected with gRNAs targeting tbx16 and tyr, as both sets included a mediocre gRNA. To simulate more samples from mediocre gRNAs, we followed the reviewer’s suggestion and created new samples by mixing genomic DNA from injected embryos with genomic DNA from the uninjected control embryo, either 1:1 (mutated alleles diluted to ½) or 1:3 (mutated alleles diluted to ¼). We assumed the proportion of mutated reads for these samples to be ½ or ¼ of the proportion of mutated reads of the original injected sample. A headloop score threshold at 0.6 accurately excluded the two mediocre gRNAs (tbx16 gRNA B and tyr gRNA A) and all the diluted samples (Figure 3—figure supplement 1B). This threshold also included all the good gRNAs, with the exception of tbx16 gRNA D which was mistakenly excluded by headloop PCR although it generated a high proportion of mutated reads. We therefore obtained one false negative but no false positive. We conclude that headloop PCR can be used to exclude mediocre gRNAs without the need for deep sequencing, although it may at times be overly conservative. We have added these results in the text (Results, section Headloop PCR […]).

While headloop PCR can be used quantitatively to exclude mediocre gRNAs, multi-locus targeting buffers against rare mediocre gRNAs in terms of successful mutagenesis. For instance, the three-guide set targeting tyr (tyr gRNA A, B, C) or tbx16 (tbx16 gRNA A, B, D) both included a mediocre gRNA (tyr locus A: ~ 58% mutated reads; tbx16 locus B: ~ 41% mutated reads), yet the phenotype penetrance was high or maximal (Figure 1D and Figure 1—figure supplement 2). Furthermore, based on the deep sequencing results, the likelihood of selecting more than one mediocre gRNA in the same three-guide set is low. Only three out of 32 gRNAs generated less than 60% mutated reads (Figure 2A: tyr gRNA A, tbx16 gRNA B, scn1lab gRNA C). If the three-guide sets were randomly assigned from these 32 gRNAs, the probability of selecting more than one (i.e. two or three) mediocre gRNA in the same set is only 0.6%. Importantly, headloop PCR correctly excluded a gRNA that barely generated any mutation (Figure 3B).

Finally, headloop PCR being itself a novel method to detect the presence of mutations, we are still collecting data and plan to further improve the tool if possible. We have provided details of the headloop primers used in the manuscript (Supplementary file 1), as well as a link to a Python-based headloop primer design tool (Material and methods, section Headloop PCR […]) to allow the community to test, validate and adapt the method for other applications.

6) Is the dosage/amount of Cas9 or RNP used in this study different or comparable with Wu et al.? Does it account for the improvement of the method described in the study?

The amount of Cas9 protein was different in our study vs Wu et al., 2018. Wu et al. injected 1000 pg (28.5 fmol) sgRNA and 800 pg (4.75 fmol) Cas9, while we injected 1000 pg (28.5 fmol) gRNA and 4700 pg (28.5 fmol) Cas9, i.e. around 6 times more Cas9 in the present study. Indeed, Wu et al. obtained better results with a 6 Cas9 to 1 sgRNA ratio (Wu et al., 2018, Figure S2D), while we obtained better results with a 1 Cas9 to 1 gRNA ratio (Figure 1—figure supplement 1). This may be explained by the use of the synthetic gRNAs, as Hoshijima et al., 2019 reached a similar conclusion regarding the use of a 1 Cas9 to 1 gRNA ratio, also using synthetic gRNAs (crRNA:tracrRNA duplexes) from IDT (Hoshijima et al., 2019, Figure S1).

The injection of more Cas9 protein may have contributed to our ability to achieve similar results as Wu et al. while targeting fewer loci. Indeed, Wu et al. did not test if raising the amount of Cas9 above 800 pg (4.9 fmol; i.e. 200 pg or 1.2 fmol per locus) could raise the phenotype penetrance further, and both our experiments (Figure 1—figure supplement 1) and Hoshijima et al.’s (Figure S1) indicate that Cas9 is the limiting factor at concentrations below ~ 5 fmol (see Figure 1—figure supplement 1: 14.25 fmol Cas9, i.e. 4.75 fmol per target, was not sufficient).

7) The authors propose to design the three target sites in distinct exon within each gene. Is it really important and/or necessary to achieve high efficient biallelic knockouts? Any evidence?

It is not necessary for all genes. First, some genes only have a single or two exons (eg. mab21l2 is a single-exon gene) so the target loci cannot always be placed on three distinct exons. For mab21l2, the three target loci were placed on the unique exon, while avoiding overlap between the protospacer sequences (the 20-nucleotide genomic sequence the gRNA binds to) and the protospacer adjacent motifs (PAM). Second, the gRNA design algorithm (IDT) does not always offer target sites on three distinct exons within the top scoring gRNAs. This was the case for tyr: all three gRNAs targeted the same exon (exon 1/5), which evidently did not cause an issue.

However, whenever possible, we think targeting distinct exons should be preferred. The evidence for this recommendation is exon skipping (see Discussion, Design of F0 knockout screens […]). Skipping of an exon harbouring a frameshift mutation generated by CRISPR-Cas9 has been reported in zebrafish stable knockout lines (Lalonde et al., 2017). In HAP1 cell lines, exon skipping was sometimes sufficient to allow the production of a partially functional protein (Smits et al., 2019). Therefore, if the three target sites are in the same exon, there may be cases where the whole exon is skipped by alternative splicing, and the resulting protein be partially functional, hence not leading to a complete knockout. If distinct exons are mutated, it is unlikely that exon skipping is sufficient to produce a functional protein.

An additional way to counteract genetic compensation by exon skipping is to preferentially target asymmetrical exons, i.e. of length not a multiple of three. If an asymmetrical exon is skipped, the resulting transcript after splicing is frameshifted, and is therefore likely to contain a premature termination codon (Lalonde et al., 2017; Tuladhar et al., 2019).

If possible, introducing mutations only in the first few exons may not be recommended as it may allow the production of a protein with a truncation in the N‑terminal sequence thanks to an alternative translation initiation event (Makino et al., 2016; Smits et al., 2019). Around 40% of exons 2–10 are asymmetrical in the zebrafish genome (Anderson et al., 2017). Therefore, there should often be opportunities to place target sites on asymmetric exons.

In summary, our recommendation would be to target three asymmetrical exons spread across the gene, for instance exons 2, 4, 6 of a 10-exon gene. In practice, it might be rare that there is a set of three gRNAs that satisfy all the conditions within the top gRNAs suggested by the design algorithm, but these guidelines can serve as selection criteria.

8) According to the section of 'Materials and methods', the synthetic gRNA was made of two components, i.e., crRNA and tracrRNA. Synthesis of gRNA as a single molecule by in vitro transcription is usually more popular and economic, is it really necessary to use crRNA and tracrRNA to achieve high efficient biallelic knockouts? Any evidence?

It is highly likely that synthetic gRNAs (i.e. not in vitro-transcribed) are required for the success of the present method. The synthetic gRNA can be bought as a single-guide RNA (i.e. a single molecule) or in two components as a tracrRNA (constant whatever the target) and a target-specific crRNA, which are then annealed into a synthetic gRNA before injections as we described (also see online protocol at dx.doi.org/10.17504/protocols.io.bfgyjjxw). Both work well (Hoshijima et al., 2019, Figure 2), but crRNA:tracrRNA duplexes cost less.

In vitro-transcribed gRNAs are less mutagenic (Hoshijima et al., 2019, Figure 1). During in vitro transcription, RNA polymerase requires the presence of guanine nucleotides at the 5′ end of transcripts. If two guanine nucleotides are not present in the genomic sequence, it is typical to artificially add them. However, this creates one or two mismatches between the spacer (gRNA) and the target sequence (genome), which substantially affects mutagenesis (Hoshijima et al., 2019, Figure 2; Varshney et al., 2015, Supplementary Figure 7). The production of the synthetic crRNAs and tracrRNA we used (IDT) is proprietary, but likely involves solid-phase synthesis, which is not based on in vitro transcription and therefore does not require the presence of the guanine nucleotides.

To complete our response to Essential Revision 1, we believe the use of the synthetic gRNAs is likely to be the main reason why we were able to reach similar results as Wu et al., 2018 but targeting three loci instead of four. In Wu et al.’s work, there was a total of 32 gRNAs designed to target slc24a5, tyr, tbx16, or tbx5a. 14 had no mismatch with the genomic binding site; 9 had one mismatch; 9 had two mismatches. Wu et al., 2018 tested two combinations of three gRNAs on slc24a5, but phenotypic penetrance (proportion of 2-dpf larvae without eye pigmentation) did not raise above ~ 55% (Wu et al., 2018, Figure 2B). The two sets included a gRNA with one mismatch (slc24a5 gRNA 1). A detailed survey of Wu et al.’s gRNAs is available in the GitHub and Zenodo repositories (wu_gRNAs.xlsx). Overall, as we reached similar results of phenotypic penetrance for common genes with fewer targeted loci per gene, it seems likely that the extra target site in Wu et al., 2018 compensates for the shortcomings of the in vitro-transcribed sgRNAs.

The synthetic crRNAs and tracrRNA from IDT carry proprietary Alt-R modifications to improve nuclease resistance once in the cell. However, Hoshijima et al., 2019 tested in zebrafish embryos gRNAs (crRNA/tracrRNA duplexes) without these modifications and did not find strong differences with those carrying the Alt-R modifications.

As the use of synthetic gRNAs circumvents in vitro transcription, including quality control checks and steps that may fail and need to be repeated, it requires substantially less laboratory work. If this is accounted for in the cost analysis, it is likely that synthetic gRNAs are not, or only slightly, more expensive than in vitro-transcribed gRNAs.

To conclude, synthetic gRNAs should be preferred because they are more mutagenic (Hoshijima et al., 2019), hence fewer loci need to be mutated per gene. While the decision to use synthetic or in vitro-transcribed gRNAs can be based on financial considerations, targeting fewer loci also reduces potential off-target effects and possibly the number of unviable embryos (see Essential Revision 1). Use of synthetic gRNAs also helps standardising injections, therefore generating results that are more predictable and more reproducible between researchers and laboratories.

9) Could headloop PCR be used for the quantification of mutagenesis efficiency (indel-producing mutation rate) of Cas9/gRNA? How sensitive is this method? Could small indels (such as 1-bp insertion or deletion) be detected by the headloop PCR?

To test whether headloop PCR can be used to estimate mutagenesis, we derived a headloop score for each sample from the standard PCR and headloop PCR band intensities on an agarose gel (Figure 3—figure supplement 1A). The score correlated well with the percentage of mutated reads measured by deep sequencing (r = 0.66, n = 29 samples tested from n = 7 loci). The score was also a significant predictor of the proportion of mutated reads by linear regression (p < 0.001, R2 = 0.44) (Figure 3—figure supplement 1B). A notable deviation was tbx16 locus D, which repeatedly produced a low headloop PCR score (< 0.5) even though the percentage of mutated reads was high (Figure 2A). We could not find a clear explanation for this. Using the line of best fit to gauge the proportion of mutated reads solely from the headloop PCR score would give an estimate in the correct range (< 20% mutated reads away from the true measure) about three times out of four (18/23 samples, not including the diluted samples for which the proportion of mutated reads was not directly measured). In summary, headloop PCR can be used to approximate the proportion of mutated alleles in F0 knockout embryos. We have added a mention of these results in the text (Results, section Headloop PCR […]).

As to sensitivity towards small indels, we hypothesised that even small indels or point mutations may be sufficient to prevent formation of the hairpin as the use of a non-proofreading enzyme always gave a positive result (i.e. no headloop PCR band, Figure 3—figure supplement 3A). To confirm this, we tested whether we could detect two small indels in stable mutant lines we had previously genotyped by sequencing (Figure 3—figure supplement 2). The first is a 1-bp deletion followed by a point mutation (T>A) in gene apoea; the second is a 2-bp deletion in gene cd2ap. For each line, we genotyped by headloop PCR 32 embryos produced from an outcross of heterozygous to wild-type (i.e. embryos were either heterozygous or wild-type). All genotype calls by headloop PCR were then verified by Sanger sequencing. For both alleles, headloop PCR was 100% accurate at distinguishing heterozygous from wild-type embryos. We conclude that headloop PCR is sensitive to small deletions. We have added these results in the text (Results, section Headloop PCR […]). Sensitivity to small insertions remains to be tested, although it seems likely given the present results.

10) In addition to indels, deletions between two double strand breaks induced by two gRNAs are also important for the generation of biallelic knockouts of the target gene. The authors showed the analysis of mutations in each site (such as in Figure 2A), is it possible to quantify the distribution and contribution of all the different deletions?

We found by Sanger sequencing that large deletions spanning multiple targeted sites can occur (Figure 2E); it would indeed be interesting to quantify their abundance in the pool of alleles. Wu et al., 2018 made an attempt in that direction. They performed long-read sequencing (PacBio) on six embryos which had been injected with a four-RNP set targeting sox32. They found that 5.7 ± 1.8% of reads had deletions spanning multiple targeted sites (Wu et al., 2018, Figure S5A). As four sites were targeted within the gene, the average frequency for one large deletion would be 0.95% (there are six possible large deletions with four targeted sites). While it provides an estimate, we question how precise this measure is, which will also illustrate the potential hurdles to be considered for such an experiment.

Quantifying the distribution of indels around each targeted site together with the large deletions between targeted sites requires deep sequencing of a genomic region spanning all targeted sites in continuous reads. This would typically represent a region of at least multiple kilobases if each target site is on a distinct exon. Sequencing of long DNA molecules in continuous reads is not possible using Illumina MiSeq (set read length of maximum 300 bp) or Sanger sequencing (maximum read length of ~ 800 bp). It is achievable with long-read sequencing technologies, namely Oxford Nanopore Technologies (Nanopore) or Pacific Biosciences (PacBio). To increase the relative concentration of DNA molecules from the genomic region of interest and thereby achieve high sequencing coverage, deep sequencing is typically preceded by PCR amplification. To take slc24a5 as an example (Figure 2E), the shortest possible wild-type amplicon spanning all three target sites is ~ 3 kb. If deletion 3 (locus B–locus A) occurred, the amplicon would now be ~ 100 bp, i.e. ~ 30× shorter. PCR would therefore be performed on a mix of long (without large deletions, ~ 3 kb) and short (with large deletion, ~ 100 bp) alleles. However, DNA polymerases preferably amplify shorter products (Dabney and Meyer, 2012), hence biasing the original relative abundance. While preparing the samples for Sanger sequencing of slc24a5, we observed that while we could amplify the B–D and D–A wild-type amplicons (respectively 1.7 kb and 1.5 kb), the band was absent for the injected samples. For these, only strong bands at lower lengths were present, which likely represent a substantial PCR length bias. Therefore, for slc24a5, the 100-bp allele (deletion 3) would be amplified disproportionately over the 3-kb allele (without deletion). Accordingly, counting the number of sequencing reads from each allele as a proxy for their original abundance in the pool is likely to over-estimate the frequency of the large deletions. Wu et al., 2018 performed amplification prior to sequencing; therefore, we think their estimate might represent an upper limit.

Long-read sequencing protocols which do not involve amplification were developed to address such limitations. The Nanopore and PacBio amplification-free sequencing protocols follow similar a logic: CRISPR-Cas9 is used to cleave the genomic region of interest. Adapters are then attached to the fragments so they are preferentially sequenced (Oxford Nanopore Technologies, 2019; Tsai et al., 2017). Fragments of the original genomic DNA are sequenced, which prevents PCR bias. However, the number of DNA molecules available for sequencing is low as a result. Therefore, these protocols achieve substantially lower sequencing coverage of the region, even with longer sequencing times. This is likely to be an issue for detecting both the large deletions between targeted sites and many of the smaller indels around each site.

Sequencing coverage for an amplification-free protocol may be around 200× (eg. Hafford-Tear et al., 2019, Supplementary Figure S1; C. M. Watson et al., 2020, Table 2; Wieben et al., 2019, Table 1). In such a scenario, there would be more than 20% chance of missing a large deletion at a frequency below 1.6% in the pool of alleles. Therefore, many large deletions may be missed in such an experiment, even if their frequencies in the pool of alleles were pertinent.

In the case of the small indels around each targeted site, we discovered through deep sequencing that there was substantial diversity in alleles, including within individual animals. Most mutations at each locus are present at low frequencies in individual animals (> 95% mutations detected at each locus are present at frequencies below 10% in individual animals, Author response image 1). This may be because the mutation happened later in development or in a cell that did not contribute many daughter cells. High coverage is necessary to detect and quantify this diversity. Many of these alleles would likely be missed by the lower coverage of an amplification-free experiment.

Author response image 1
Most mutations are found at low frequencies in individual animals.

Frequency of each mutation in individual F0 knockout embryos (samples). Mutations are ranked from the most frequent in the embryo (a 3-bp deletion in gene ta, locus D, embryo 1 was present in 602/605 of the reads) to the least frequent (a 11-bp insertion in gene csnk1db, locus B, embryo 4 was present in 1/3189 of the reads). n = 7015 unique mutations.

Using Nanopore or PacBio, we estimate the cost of such an experiment to be around £5,000–6,000. This would be an attempt towards quantifying the frequency of large deletions within a single targeted gene (eg. slc24a5) and generated by a single three-RNP set (eg. slc24a5 gRNAs A, B, D). However, these frequencies are likely to be specific to each gene or three-RNP set. For example, a parameter which likely influences the rate at which a specific large deletion is generated is the distance between the two targeted sites (eg. Wu et al., 2018, Figure 6F). We also hypothesise that chromatin state of the genomic region may be a determinant. We would predict that two simultaneous double-strand breaks occurring in euchromatin are more likely to cause a large deletion than in heterochromatin. Incidentally, CRISPR-Cas9 activity may weakly correlate with gene expression and chromatin accessibility of the target site in zebrafish (Uusi-Mäkelä et al., 2018). Furthermore, multiple F0 knockout larvae would likely need to be pooled to gather sufficient genomic DNA. The result would thus be an average of multiple animals, while we have observed with deep sequencing that the most frequent alleles are often specific to each animal (see Figure 2B: two animals injected with the same RNP only share in average ~ 3 of their top 10 most frequent mutations).

In summary, PCR amplification followed by long-read sequencing is technically sound but would give somewhat over-estimated frequencies of large deletions. Protocols which do not rely on amplification exist, but the lower sequencing coverage they achieve is unlikely to be sufficient to detect these alleles which are presumably infrequent. The experiment is also costly, while the result it would generate is probably not generalisable to other RNP sets, genes, and animals.

While accurately quantifying the frequency of large deletions spanning multiple targeted sites remains challenging, we may be able to offer more estimates based on the literature.

Kim and Zhang, 2020 used the large deletions spanning multiple targeted sites to create stable zebrafish knockout lines where multiple exons are deleted. In a first experiment, they injected a mix of 7 sgRNAs targeting different exons of smarca2. They then grew 11 of these fish to adulthood as F0 founders and crossed them to wild types to generate 154 F1 animals, which were screened once adults for the presence of large deletions. A total of 8 animals issued from 3 F0 founders carried large deletions (Kim and Zhang, 2020, Table 1: 1/8, 4/8, 3/19). Assuming that the germline genotypes of the F0 founders was a random sampling from the F0 pool of alleles, the proportion of F1 adults with large deletions may be representative of the frequency of large deletions within each F0 parent. In the 3 positive F0 founders, this would amount to frequencies: 12.5% (1/8), 50% (4/8), and 15.8% (3/19). Of the 7 sgRNAs, one targeted exon 1, then two targeted each of exon 3, 15, 28 (Kim and Zhang, 2020, Figure 1). However, no mutations were detected in exon 3. There were therefore three possible large deletion (exon 1–exon 15, exon 1–exon 28, exon 15–exon 28). The average frequency of a site-spanning deletion within each positive F0 animal would thus be around 4.2%, 16.7%, 5.7%. However, most (3/11) F0 founders did not produce F1 embryos positive for large deletions. Assuming these F0 animals were negative for large deletions, and all 11 F0 animals were pooled together during a sequencing experiment, the average frequency of a site-spanning deletion would be around 2.4%. In a second, similar experiment, Kim and Zhang injected two pairs of sgRNAs, each targeting two exons of rnf185 or rnf215 (Kim and Zhang, 2020, Figure 2 and Figure 3). 9 F0 founders were grown to adulthood, then crossed to generate a total of 48 F1 adult fish that were screened for large deletions. For rnf185, 2 F1 animals generated from 2 of the F0 founders carried large deletions (Kim and Zhang, 2020, Table 2: 1/5, 1/5). This would amount to a frequency of this allele around 20% in the two F0 founders, but a probability of the large deletion occurring around 4.4% (large deletion occurred at 20% frequency in 2/9 animals). For rnf185, 3 F1 animals generated from 3 of the F0 founders carried large deletions (Kim and Zhang, 2020, Table 2: 1/5, 1/5, 1/5). This would amount to a frequency of this allele around 20% in the three F0 founders, but a probability of the large deletion occurring around 6.6% (large deletion occurred at 20% frequency in 2/9 animals).

Varshney et al., 2015 performed similar experiments. They injected embryos with a pair of sgRNAs targeting two sites within one of 15 genes. They grew F0 embryos to adulthood and crossed them to generate 769 F1 embryos. Among them, 24 had a large deletion spanning the two targeted sites. Similarly, if we assume that the germline genotype was a random sampling of the F0 pool of alleles, this would amount to an average frequency of large deletions in the F0 animals around 3% (24/769).

Overall, estimates for the probability of a large deletion to occur range between 0.95% (Wu et al., 2018) and 6.6% (Kim and Zhang, 2020), with variability between injected animals, targeted genes, and RNP sets. Importantly, the three above studies used in vitro-transcribed gRNAs. As mutagenesis is likely to be a strong determinant influencing the probability of large deletions, these may under-estimate what can be achieved with synthetic RNPs.

Finally, while quantifying the frequency of large deletions when using synthetic RNPs is certainly interesting for applications that specifically require them (eg. Kim and Zhang, 2020), it is important to highlight that the alleles carrying large deletions are extremely likely to represent loss-of-function mutations. Their contributions therefore add to frameshift mutations and mutations of key residues of the protein as all probable loss-of-function mutations, as we developed in the text (Results, section Sequencing of targeted loci […]).

11) Figure 1C,D: The authors compared the effects of the injection of 1, 2, 3, and 4 loci. How were the 1, 2, and 3 loci selected from the four target sites? Will each of the four loci give the same or different phenotypic ratio if tested individually? Will different combinations of 2 loci or 3 loci give the same or different phenotypic ratio? Or which combination of 2 loci or 3 loci will give the highest mutagenic effect? For example, in Figure 1C, the 3-loci showed comparable effect with 4-loci, while the 2-loci is less effective; is it possible to find other 2-loci combinations which could show higher mutagenic efficiency than the current 2-loci, such that the effect of the new 2-loci combination is as good as the 3-loci or 4-loci combination? Conversely, in Figure 1D, the 2-loci already showed the highest mutagenic effect, is it because of this particular 2-loci combination, or any 2-loci combination will show the same efficiency?

To answer specifically the first question raised, the order when targeting gradually more loci (Figure 1E,F) followed the IDT ranking when selecting a single crRNA per exon (see Material and methods, section RNP pooling). Therefore, when targeting a single locus, only the top predicted gRNA is injected; when targeting two loci, the top two predicted gRNAs are injected together; etc. Based on the deep sequencing data (Figure 2A), this amounts to essentially the same as picking the order randomly within the top 3 or 4 gRNAs, as the top predicted gRNA of the set is only rarely the most mutagenic one in practice. Indeed, when we sequenced 10 genes targeted by three or four-gRNA sets, there were only 3 sets in which the top predicted gRNA by IDT (gRNA A) was also the top performing gRNA (slc24a5, tbx5a, scn1lab). This is what is expected by chance alone (if the most mutagenic gRNA of each set was drawn randomly, 31% of gRNA sets would have the top predicted gRNA as most mutagenic gRNA). In other words, it is not possible to guess above chance the IDT ranking from the mutagenesis levels alone.

For many genes, it might be possible to find a better solution than the three-RNP strategy, i.e. to target only one or two loci while obtaining good phenotypic penetrance. The solution would be to test for instance ten gRNAs for the gene of interest, ideally that all target sequences coding for essential residues of the protein. Deep sequencing would then be performed to quantify the mutations they each generated and only the top one or two gRNAs that are highly mutagenic with a bias towards frameshift mutations would be selected. Examples of such ‘outstanding’ gRNAs can be found in our results. For example, the second gRNA of the tyr-targeting set (Figure 2A, tyr gRNA B) is clearly excellent: mutagenesis is high and it generates more frameshift mutations than expected by chance. This likely explains why injecting only tyr gRNA A and B was sufficient to achieve full phenotypic penetrance (Figure 1D, number of loci = 2). The disproportionate contribution of tyr gRNA B can also be seen in Figure 2F where the proportion of frameshift alleles rise sharply when tyr gRNA B is added to the mix (number of loci = 2). Another example is mpv17 gRNA B (see Figure 2A,B). However, testing numerous gRNAs until these excellent gRNAs are found is not a feasible solution if tens or hundreds of genes were to be tested in a screen. Simply buying one supplementary gRNA for each gene would greatly increase the total cost. An essential criterion for success of the method we developed was that it could be readily applied to any open-reading frame during a large genetic screen (see Results, section Three synthetic gRNAs […]). This is the reason why we did not want to test different combinations of gRNAs until we found one giving optimal results, nor did we target essential residues of the protein for each tested gene, as many genes selected for screening are likely to lack annotations. Using a combination of three synthetic gRNAs as we propose is a scalable solution to this challenge: as we demonstrate, it is applicable to any gene with little or no pre-screening of the gRNAs.

However, we developed headloop PCR exactly as an attempt to provide a solution to this challenge. It is a rapid and inexpensive way to test the gRNAs. Sufficient mutagenesis can be confirmed (Figure 3B and Figure 3—figure supplement 1), and even broadly estimated (Figure 3—figure supplement 1B). With this tool, we suggest that gRNAs could be tested during a validation round (see Discussion, Design of F0 knockout screens), which will ensure that the proposed three-gRNA solution gives high phenotypic penetrance for every gene tested.

Being able to systematically achieve high phenotypic penetrance while targeting fewer than three loci per gene would be a further improvement of the current method, as it would likely generate less unviable embryos and there would be less potential off-target effects. We offer suggestions (to be tested) on how to do this in the Discussion (section Other technical considerations […]). Meticulous design of the gRNAs, including using prediction of editing outcomes (Shen et al., 2018), followed by systematic quantification of mutagenesis by headloop PCR may allow for fewer loci to be targeted in each gene without costly and time-consuming screening of gRNAs. In any case, further improvements cannot be designed for specific genes as it would not be applicable to a large genetic screen.

Finally, we believe that this gene-specific search for an optimal combination of gRNAs may be useful in cases where 1) targeting three loci per gene seems to be too much, for instance because it generates many unviable embryos, 2) the experiment will be repeated many times. Such a case may be the F0 crystal fish (Figure 4C), if the approach was to become routine for imaging applications. Injecting three sets of three-RNP worked well but generated 50% unviable embryos. As the mutations induced by each gRNA of each set were quantified here, it may be possible to repeat the experiment but injecting only gRNAs that were highly mutagenic and generated a surplus of frameshift mutations (for instance mpv17 gRNA B). An additional approach would be to study the gene structure of slc45a2, mitfa, and mpv17 and select gRNAs that target essential residues of the protein.

12) Figure 6: The phenotypes of scn1lab F0 knockouts are more severe than those of scn1lab-/- mutant. Any explanation?

Indeed, the F0 knockout larvae sit on average at greater distances from their wild-type siblings than scn1labΔ44stable mutant larvae (Figure 6C).

However, when comparing statistically the effect sizes between experiments, the F0 knockout larvae do not have a significantly ‘stronger’ phenotype than stable mutant homozygotes. Indeed, the effect size in the stable mutant experiment (wild types vs stable mutant homozygotes: Cohen’s d = 1.57) is not significantly different (p = 0.18) than the effect size in the first F0 knockout experiment (F0 experiment 1, scrambled-injected controls vs F0 knockout larvae: Cohen’s d = 2.91). Similarly, the effect size in the stable mutant experiment is not significantly different (p = 0.27) than the effect size in the second F0 knockout experiment (F0 experiment 2, scrambled-injected controls vs F0 knockout larvae: Cohen’s d = 2.57). As a comparison, in the stable mutant experiment, the effect size between wild types and heterozygotes (Cohen’s d = 0.28) is significantly different (p = 0.04) from the effect size between wild types and homozygotes (Cohen’s d = 1.57). The statistical test is a Z-test used for comparing two studies in meta-analysis (Borenstein et al., 2009). In sum, effect size comparisons demonstrate that scn1lab F0 knockout phenotypes are of similar severity to the scn1labΔ44 stable mutant phenotype. We have added this analysis in the text (Results, section Continuous traits […]).

To reinforce this conclusion, we gathered and analysed data from a different scn1lab mutant line (Ellen Hoffman, unpublished), termed double indemnity (didys552) (Schoonheim et al., 2010). Unfortunately, the data was not collected in a way that allowed us to perform the same multi-parameter analysis as in Figure 6B,C. As metric for activity, we used the time the animal spent above an activity threshold (ViewPoint parameter freezing = 3) within each 1-minute epoch (eg. fish 1 was active for 3 seconds between minute 3 and 4). For each animal, these data were summed within each time window to obtain the total time spent active during that window (eg. fish 1 was active for a total of 3000 seconds during day 1 of tracking). The data for each animal were then normalised by calculating the Z-score from the mean of the wild-type or scrambled-injected siblings (eg. fish 1 was 2 standard deviations less active than its wild-type siblings during day 1 of tracking). scn1lab knockout animals, either F0 knockout larvae or stable mutant homozygotes, were consistently less active than their control siblings (wild-type or scrambled-injected) during the day (Author response image 2). This effect was within the same range in F0 knockout experiments and stable mutant experiments, either scn1labΔ44 or didy. During the night, scn1lab knockout animals tended to be more active than their control siblings, although this result was more variable between experiments. While the effect is clear in F0 knockout experiment 2 and the didy experiment, scn1labΔ44 homozygotes and knockouts in F0 knockout experiment 1 were not more active than their control siblings when comparing data summed across the entire night. However, their hyperactivity is unambiguous during the second half of the night (Figure 6A).

Author response image 2
scn1lab F0 knockouts do not have consistently more severe behavioural phenotypes than stable scn1lab mutant lines.

Total time active, represented as deviation from the paired wild-type (+/+) or scrambled-injected (scr) mean (Z-score), during the day (left) or the night (right). The tracking spanned two days and two nights (as in Figure 6A), so each animal contributed two data points to each plot. F0 exp1: scn1lab F0 experiment 1, same experiment as in Figure 6A right; F0 exp2: scn1lab F0 experiment 2, same experiment as in Figure 6—figure supplement 1; Δ44: stable scn1labΔ44 mutant line, same experiment as in Figure 6A left; didy: stable scn1lab didy mutant line. Black crosses mark the population means. (day, in order displayed) F0 exp1: *** p < 0.001; F0 exp2: *** p < 0.001; Δ44: ns p > 0.999, *** p < 0.001; didy: * p = 0.021, ** p = 0.002. (night, in order displayed) F0 exp1: ns p > 0.999; F0 exp2: ** p < 0.004; Δ44: ns p = 0.186, ns p > 0.999; didy: ns p > 0.999, *** p < 0.001. Statistics by pairwise Welch’s t-tests with Holm’s p-value adjustment.

In summary, scn1lab F0 knockout larvae do not have consistently more severe phenotypes than scn1lab stable mutant larvae. Indeed, this difference was not statistically significant, and was not replicated with a different behavioural parameter or with a different scn1lab mutant line (didy). Most likely, this apparent effect arose from the variability that is expected between experiments, clutches, genetic background, and mutant alleles.

In fact, it is remarkable that the multi-parameter behavioural phenotypes of the F0 knockout larvae and the stable mutant larvae correlate so strongly (r = 0.86 and r = 0.75 by Pearson correlation, Figure 6B). The scn1labΔ44 stable mutant experiment was performed five years before the F0 knockout experiments with some discrepancies in protocol (see Material and methods, section Behavioural video tracking). We believe it highlights again the robustness of the F0 knockout method for measurement of complex, behavioural phenotypes.

13) Please provide the academic name of zebrafish in its first appearance.

We have edited the text accordingly (see Introduction).

References

Anderson JL, Mulligan TS, Shen MC, Wang H, Scahill CM, Tan FJ, Du SJ, Busch-Nentwich EM, Farber SA. 2017. mRNA processing in mutant zebrafish lines generated by chemical and CRISPR-mediated mutagenesis produces unexpected transcripts that escape nonsense-mediated decay. PLOS Genetics 13:e1007105. . DOI: https://doi.org/10.1371/journal.pgen.1007105, PMID: 29161261

Borenstein M, Hedges LV, Higgins JPT, R RH. 2009. Subgroup Analyses. In: Rothstein H. R (Ed). Introduction to Meta‐Analysis. Wiley Online Books. p. 19–24. DOI: https://doi.org/10.1002/9780470743386.ch19

Burger A, Lindsay H, Felker A, Hess C, Anders C, Chiavacci E, Zaugg J, Weber LM, Catena R, Jinek M, Robinson MD, Mosimann C. 2016. Maximizing mutagenesis with solubilized CRISPR-Cas9 ribonucleoprotein complexes. Development 143:2025–2037. DOI: https://doi.org/10.1242/dev.134809, PMID: 27130213

Dabney J, Meyer M. 2012. Length and GC-biases during sequencing library amplification: A comparison of various polymerase-buffer systems with ancient and modern DNA sequencing libraries. BioTechniques 52:87–94. DOI: https://doi.org/10.2144/000113809, PMID: 22313406

Hafford-Tear NJ, Tsai Y-C, Sadan AN, Sanchez-Pintado B, Zarouchlioti C, Maher GJ, Liskova P, Tuft SJ, Hardcastle AJ, Clark TA, Davidson AE. 2019. CRISPR/Cas9-targeted enrichment and long-read sequencing of the Fuchs endothelial corneal dystrophy–associated TCF4 triplet repeat. Genetics in Medicine 21:2092–2102. DOI: https://doi.org/10.1038/s41436-019-0453-x, PMID: 30733599

Ho RK, Kane DA. 1990. Cell-autonomous action of zebrafish spt-1 mutation in specific mesodermal precursors. Nature 348:728–730. DOI: https://doi.org/10.1038/348728a0, PMID: 2259382

Hoshijima K, Jurynec MJ, Klatt Shaw D, Jacobi AM, Behlke MA, Grunwald DJ. 2019. Highly Efficient CRISPR-Cas9-Based Methods for Generating Deletion Mutations and F0 Embryos that Lack Gene Function in Zebrafish. Developmental Cell 51:645–657.e4. DOI: https://doi.org/10.1016/j.devcel.2019.10.004, PMID: 31708433

Jao LE, Wente SR, Chen W. 2013. Efficient multiplex biallelic zebrafish genome editing using a CRISPR nuclease system. PNAS 110:13904–13909. DOI: https://doi.org/10.1073/pnas.1308335110, PMID: 23918387

Keatinge M, Tsarouchas TM, Munir T, Larraz J, Gianni D, Tsai H-H, Becker CG, Lyons DA, Becker T. 2020. Phenotypic screening using synthetic CRISPR gRNAs reveals pro-regenerative genes in spinal cord injury. bioRxiv. DOI: https://doi.org/10.1101/2020.04.03.023119

Kim BH, Zhang G. 2020. Generating Stable Knockout Zebrafish Lines by Deleting Large Chromosomal Fragments Using Multiple gRNAs. G3: Genes, Genomes, Genetics 10:1029–1037. DOI: https://doi.org/10.1534/g3.119.401035

Kuil LE, Oosterhof N, Geurts SN, Van Der Linde HC, Meijering E, Van Ham TJ. 2019. Reverse genetic screen reveals that Il34 facilitates yolk sac macrophage distribution and seeding of the brain. Disease Models & Mechanims 12:dmm037762. DOI: https://doi.org/10.1242/dmm.037762, PMID: 30765415

Lalonde S, Stone OA, Lessard S, Lavertu A, Desjardins J, Beaudoin M, Rivas M, Stainier DiYR, Lettre G. 2017. Frameshift indels introduced by genome editing can lead to in-frame exon skipping. PLOoS ONE 12:e0178700. DOI: https://doi.org/10.1371/journal.pone.0178700

Makino S, Fukumura R, Gondo Y. 2016. Illegitimate translation causes unexpected gene expression from on-target out-of-frame alleles created by CRISPR-Cas9. Scientific Reports 6:39608. DOI: https://doi.org/10.1038/srep39608, PMID: 28000783

Meng Q-J, Maywood ES, Bechtold DA, Lu W-Q, Li J, Gibbs JE, Dupré SM, Chesham JE, Rajamohan F, Knafels J, Sneed B, Zawadzke LE, Ohren JF, Walton KM, Wager TT, Hastings MH, Loudon ASI. 2010. Entrainment of disrupted circadian behavior through inhibition of casein kinase 1 (CK1) enzymes. PNAS 107:15240–15245. DOI: https://doi.org/10.1073/pnas.1005101107, PMID: 20696890

Oxford Nanopore Technologies. 2019. A rapid CRISPR/Cas9-mediated, amplification-free target enrichment method for native-strand sequencing. https://nanoporetech.com/resource-centre/rapid-crisprcas9-mediated-amplification-free-target-enrichment-method-native-strand

Schoonheim PJ, Arrenberg AB, Del Bene F, Baier H. 2010. Optogenetic Localization and Genetic Perturbation of Saccade-Generating Neurons in Zebrafish. Journal of Neuroscience 30:7111–7120. DOI: https://doi.org/10.1523/JNEUROSCI.5193-09.2010, PMID: 20484654

Shah AN, Davey CF, Whitebirch AC, Miller AC, Moens CB. 2015. Rapid reverse genetic screening using CRISPR in zebrafish. Nature Methods 12:535–540. DOI: https://doi.org/10.1038/nmeth.3360, PMID: 25867848

Shen MW, Arbab M, Hsu JY, Worstell D, Culbertson SJ, Krabbe O, Cassa CA, Liu DR, Gifford DK, Sherwood RI. 2018. Predictable and precise template-free CRISPR editing of pathogenic variants. Nature 563:646–651. DOI: https://doi.org/10.1038/s41586-018-0686-x, PMID: 30405244

Smits AH, Ziebell F, Joberty G, Zinn N, Mueller WF, Clauder-Münster S, Eberhard D, Fälth Savitski M, Grandi P, Jakob P, Michon AM, Sun H, Tessmer K, Bürckstümmer T, Bantscheff M, Steinmetz LM, Drewes G, Huber W. 2019. Biological plasticity rescues target activity in CRISPR knock outs. Nature Methods 16:1087–1093. DOI: https://doi.org/10.1038/s41592-019-0614-5, PMID: 31659326

Tsai Y-C, Greenberg D, Powell J, Höijer I, Ameur A, Strahl M, Ellis E, Jonasson I, Mouro Pinto R, Wheeler V, Smith M, Gyllensten U, Sebra R, Korlach J, Clark T. 2017. Amplification-free, CRISPR-Cas9 Targeted Enrichment and SMRT Sequencing of Repeat-Expansion Disease Causative Genomic Regions. bioRxiv. DOI: https://doi.org/10.1101/203919

Tuladhar R, Yeu Y, Tyler Piazza J, Tan Z, Rene Clemenceau J, Wu X, Barrett Q, Herbert J, Mathews DH, Kim J, Hyun Hwang T, Lum L. 2019. CRISPR-Cas9-based mutagenesis frequently provokes on-target mRNA misregulation. Nat Commun 10:1–10. DOI: https://doi.org/10.1038/s41467-019-12028-5, PMID: 31492834

Uusi-Mäkelä MIE, Barker HR, Bäuerlein CA, Häkkinen T, Nykter M, Rämet M. 2018. Chromatin accessibility is associated with CRISPR-Cas9 efficiency in the zebrafish (Danio rerio). PLOS ONE 13:e0196238. DOI: https://doi.org/10.1371/journal.pone.0196238, PMID: 29684067

Varshney GK, Pei W, Lafave MC, Idol J, Xu L, Gallardo V, Carrington B, Bishop K, Jones M, Li M, Harper U, Huang SC, Prakash A, Chen W, Sood R, Ledin J, Burgess SM. 2015. High-throughput gene targeting and phenotyping in zebrafish using CRISPR/Cas9. Genome Research 25:1030–1042. DOI: https://doi.org/10.1101/gr.186379.114, PMID: 26048245

Watson CJ, Monstad-Rios AT, Bhimani RM, Gistelinck C, Willaert A, Coucke P, Hsu YH, Kwon RY. 2020. Phenomics-Based Quantification of CRISPR-Induced Mosaicism in Zebrafish. Cell Systems 10:275–286. DOI: https://doi.org/10.1016/j.cels.2020.02.007, PMID: 32191876

Watson CM, Crinnion LA, Hewitt S, Bates J, Robinson R, Carr IM, Sheridan E, Adlard J, Bonthron DT. 2020. Cas9-based enrichment and single-molecule sequencing for precise characterization of genomic duplications. Laboratory Investigation 100:135–146. /10.1038/s41374-019-0283-0, PMID: 31273287

Wieben ED, Aleff RA, Basu S, Sarangi V, Bowman B, McLaughlin IJ, Mills JR, Butz ML, Highsmith EW, Ida CM, Ekholm JM, Baratz KH, Fautsch MP. 2019. Amplification-free long-read sequencing of TCF4 expanded trinucleotide repeats in Fuchs Endothelial Corneal Dystrophy. PLOS ONE 14:e0219446. DOI: https://doi.org/10.1371/journal.pone.0219446, PMID: 31276570

Wu RS, Lam II, Clay H, Duong DN, Deo RC, Coughlin SR. 2018. A Rapid Method for Directed Gene Knockout for Screening in G0 Zebrafish. Dev Cell 46:112–125. DOI: https://doi.org/10.1016/j.devcel.2018.06.003, PMID: 29974860

Wycliffe R, Plaisancie J, Leaman S, Santis O, Tucker L, Cavieres D, Fernandez M, Weiss-Garrido C, Sobarzo C, Gestri G, Valdivia LE. 2020. Developmental delay during eye morphogenesis underlies optic cup and neurogenesis defects in mab21l2u517 zebrafish mutants. The International Journal of Developmental Biology52:e200173. DOI: https://doi.org/10.1387/ijdb.200173lv

https://doi.org/10.7554/eLife.59683.sa2

Article and author information

Author details

  1. François Kroll

    Department of Cell and Developmental Biology, University College London, London, United Kingdom
    Contribution
    Conceptualization, Resources, Data curation, Software, Formal analysis, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9908-2648
  2. Gareth T Powell

    Department of Cell and Developmental Biology, University College London, London, United Kingdom
    Contribution
    Conceptualization, Resources, Data curation, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing
    Competing interests
    No competing interests declared
  3. Marcus Ghosh

    Department of Cell and Developmental Biology, University College London, London, United Kingdom
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Validation, Visualization, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2428-4605
  4. Gaia Gestri

    Department of Cell and Developmental Biology, University College London, London, United Kingdom
    Contribution
    Conceptualization, Resources, Data curation, Formal analysis, Validation, Investigation, Visualization, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-8854-1546
  5. Paride Antinucci

    Department of Neuroscience, Physiology and Pharmacology, University College London, London, United Kingdom
    Contribution
    Conceptualization, Validation, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0573-5383
  6. Timothy J Hearn

    Department of Cell and Developmental Biology, University College London, London, United Kingdom
    Contribution
    Resources, Investigation, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  7. Hande Tunbak

    Wolfson Institute for Biomedical Research, University College London, London, United Kingdom
    Contribution
    Resources, Validation, Investigation, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3180-1401
  8. Sumi Lim

    Department of Cell and Developmental Biology, University College London, London, United Kingdom
    Contribution
    Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  9. Harvey W Dennis

    School of Biological Sciences, Faculty of Science, University of Bristol, Bristol, United Kingdom
    Contribution
    Investigation
    Competing interests
    No competing interests declared
  10. Joseph M Fernandez

    Child Study Center, Yale School of Medicine, New Haven, United States
    Contribution
    Methodology
    Competing interests
    No competing interests declared
  11. David Whitmore

    1. Department of Cell and Developmental Biology, University College London, London, United Kingdom
    2. Department of Molecular and Cell Biology, James Cook University, Townsville, Australia
    Contribution
    Resources, Funding acquisition, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  12. Elena Dreosti

    Wolfson Institute for Biomedical Research, University College London, London, United Kingdom
    Contribution
    Resources, Funding acquisition, Methodology
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6738-7057
  13. Stephen W Wilson

    Department of Cell and Developmental Biology, University College London, London, United Kingdom
    Contribution
    Conceptualization, Supervision, Funding acquisition, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8557-5940
  14. Ellen J Hoffman

    1. Child Study Center, Yale School of Medicine, New Haven, United States
    2. Department of Neuroscience, Yale School of Medicine, New Haven, United States
    Contribution
    Resources, Investigation, Methodology
    Competing interests
    No competing interests declared
  15. Jason Rihel

    Department of Cell and Developmental Biology, University College London, London, United Kingdom
    Contribution
    Conceptualization, Resources, Data curation, Supervision, Funding acquisition, Validation, Methodology, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    j.rihel@ucl.ac.uk
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-4067-2066

Funding

Wellcome Trust (Investigator Award 217150/Z/19/Z)

  • Jason Rihel

Biotechnology and Biological Sciences Research Council (BB/T001844/1)

  • Jason Rihel

ARUK (Interdisciplinary Research Grant)

  • Jason Rihel

Medical Research Council (Programme Grant MR/L003775/1)

  • Stephen W Wilson

Wolfson Foundation (Leonard Wolfson PhD Programme in Neurodegeneration)

  • François Kroll

Wellcome Trust (Investigator Award 095722/Z/11/Z)

  • Stephen W Wilson

Wellcome Trust (Sir Henry Wellcome Postdoctoral Fellowship 204708/Z/16/Z)

  • Paride Antinucci

Wellcome Trust (Biomedical Vacation Scholarship)

  • Harvey W Dennis

Medical Research Council (Doctoral Training Grant)

  • Marcus Ghosh

Medical Research Council (Programme Grant MR/T020164/1)

  • Gaia Gestri

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank the members of the Rihel and Wilson labs and other zebrafish groups at University College London (UCL) for helpful discussions. We thank Ana Faro for organising the Illumina MiSeq run and Isaac Bianco for use of the custom-built two-photon microscope. SW thanks Leonardo Valdivia for early work on the approach in his group. We thank all supporting staff at UCL including Fish Facility staff for fish care and husbandry. FK was supported and funded by the Leonard Wolfson PhD Programme in Neurodegeneration. The work was also funded by a BBSRC grant (BB/T001844/1) and an ARUK Interdisciplinary grant awarded to JR, Wellcome Trust Investigator Awards (217150/Z/19/Z and 095722/Z/11/Z) awarded to JR and SW, Medical Research Council Programme Grants (MR/L003775/1 and MR/T020164/1) awarded to SW and GG, a Medical Research Council Doctoral Training Grant awarded to MG, and a Sir Henry Wellcome Postdoctoral Fellowship (204708/Z/16/Z) awarded to PA. HD was funded by a Wellcome Biomedical Vacation Scholarship.

Ethics

Animal experimentation: Experimental procedures were in accordance with the Animals (Scientific Procedures) Act 1986 under Home Office project licences PA8D4D0E5 awarded to Jason Rihel and PAE2ECA7E awarded to Elena Dreosti. Adult zebrafish were kept according to FELASA guidelines (Aleström et al., 2019).

Senior Editor

  1. Didier YR Stainier, Max Planck Institute for Heart and Lung Research, Germany

Reviewing Editor

  1. Stephen C Ekker, Mayo Clinic, United States

Reviewer

  1. Darius Balciunas, Temple University, United States

Publication history

  1. Received: June 4, 2020
  2. Accepted: December 14, 2020
  3. Version of Record published: January 8, 2021 (version 1)

Copyright

© 2021, Kroll et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 4,796
    Page views
  • 431
    Downloads
  • 1
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Chromosomes and Gene Expression
    2. Genetics and Genomics
    Thomas J Etheridge et al.
    Research Article

    The essential Smc5/6 complex is required in response to replication stress and is best known for ensuring the fidelity of homologous recombination. Using single-molecule tracking in live fission yeast to investigate Smc5/6 chromatin association, we show that Smc5/6 is chromatin associated in unchallenged cells and this depends on the non-SMC protein Nse6. We define a minimum of two Nse6-dependent sub-pathways, one of which requires the BRCT-domain protein Brc1. Using defined mutants in genes encoding the core Smc5/6 complex subunits we show that the Nse3 double-stranded DNA binding activity and the arginine fingers of the two Smc5/6 ATPase binding sites are critical for chromatin association. Interestingly, disrupting the ssDNA binding activity at the hinge region does not prevent chromatin association but leads to elevated levels of gross chromosomal rearrangements during replication restart. This is consistent with a downstream function for ssDNA binding in regulating homologous recombination.

    1. Genetics and Genomics
    2. Immunology and Inflammation
    Terkild B Buus et al.
    Research Article

    Simultaneous measurement of surface proteins and gene expression within single cells using oligo-conjugated antibodies offers high-resolution snapshots of complex cell populations. Signal from oligo-conjugated antibodies is quantified by high-throughput sequencing and is highly scalable and sensitive. We investigated the response of oligo-conjugated antibodies towards four variables: concentration, staining volume, cell number at staining, and tissue. We find that staining with recommended antibody concentrations causes unnecessarily high background and amount of antibody used can be drastically reduced without loss of biological information. Reducing staining volume only affects antibodies targeting abundant epitopes used at low concentrations and is counteracted by reducing cell numbers. Adjusting concentrations increases signal, lowers background, and reduces costs. Background signal can account for a major fraction of total sequencing and is primarily derived from antibodies used at high concentrations. This study provides new insight into titration response and background of oligo-conjugated antibodies and offers concrete guidelines to improve such panels.