Identification of maternal-effect genes in zebrafish using maternal crispants

In zebrafish, the functions of many genes during development have been determined by inducing loss-of-function conditions through reverse genetic approaches. The establishment and advancement of targeted gene-editing technology, such as zinc-finger nucleases (ZFNs) (Doyon et al., 2008; Meng et al., 2008), transcription activator-like effector nucleases (TALENs) (Huang et al., 2011) and clusters of regularly interspaced short palindromic repeats (CRISPR) (Hwang et al., 2013), have led to the ability to create targeted genetic lesions in candidate genes. Previous work has shown that embryos injected with CRISPR-Cas9 have the potential to contain biallelic editing, which allows for the screening of embryonic phenotypes in somatic tissue of injected (F0) embryos, known as a ‘crispant’ (Burger et al., 2016; Jao et al., 2013; Shah et al., 2015; Shankaran et al., 2017; Trubiroha et al., 2018; Wu et al., 2018).

Early embryogenesis is dependent on a pool of maternal products, such as RNAs, proteins and other biomolecules, which drive developmental processes until zygotic genome activation (ZGA). Depletion of these maternal products, when essential, results in embryonic lethality. In animal species, mutations in maternally expressed genes disrupt early cellular processes, such as egg polarity and activation, pronuclear fusion, cytokinesis, cell adhesion, germ cell specification, and axis induction (Abrams and Mullins, 2009; Dosch, 2015). Maternally expressed genes can be robustly targeted with CRISPR-Cas9 (Campbell et al., 2015; Eno et al., 2016; He et al., 2018). However, an extra generation is necessary to observe maternal-effect phenotypes, which requires additional resources.

Here, we use maternal crispants to promote biallelic germline editing to achieve phenotyping of maternally expressed gene function in the next generation. We first validated this technique by using maternal crispants to phenocopy three known maternal-effect genes: birc5b (motley), tmi and mid1ip1l (aura) (Eno et al., 2016; Nair et al., 2013, 2021 preprint). Then, we selected four candidates with a potential maternal-only gene function by using the zebrafish genome assembly to identify multiple-copy gene families and stage-specific RNA-sequencing databases to predict maternal-specific expression. These candidates were tested for maternal-effect phenotypes using maternal crispants. Out of these four candidate genes, we identified two zebrafish maternal-effect genes, kpna7 and fhdc3. Overall, these studies demonstrate the feasibility of using available databases coupled with a maternal crispant approach to identify and characterize, in a single generation, maternal-effect genes in zebrafish.

A traditional workflow for generating a mutant zebrafish with CRISPR-Cas9 uses guide RNAs targeting a specific gene of interest, co-injected in F0 early embryos with Cas9 mRNA or Cas9 protein. Successful targeting generates stable lines for the identification of a recessive zygotic mutation in F3 embryos, a process that requires three generations (in zebrafish, approximately 9-12 months) from the time of injection. The identification of a mutation in a maternal-effect gene requires an additional generation to assess the phenotype in the progeny from a germ line with the loss-of-function condition (Fig. 1A).

Recent work has shown that the biallelic editing ability of CRISPR-Cas9 allows for the screening of zygotic phenotypes in somatic cells of embryos in the F0 generation, also known as ‘crispants’ (Burger et al., 2016; Jao et al., 2013; Shah et al., 2015; Shankaran et al., 2017; Trubiroha et al., 2018; Wu et al., 2018). Depending on the developmental process of interest, crispant embryos allow screening for the phenotype of targeted candidate genes in a few hours to a few days. In zebrafish, multiplexing guide RNAs to a single gene target has been shown to generate biallelic lesions with high efficiency (Wu et al., 2018). This increased efficiency should, in principle, allow the observation of maternal-effect phenotypes in the F1 generation in what we refer to as ‘maternal crispant’ embryos. The generation of maternal crispants would allow the phenotype of mutations in maternally expressed genes to be tested directly in the progeny of injected individuals.

For maternal crispants to occur, there must be biallelic editing events in cells fated to develop into the germ line. To generate maternal crispants with high efficiency, we designed four guide RNAs for each target, which included known or candidate maternal-effect genes. Guide RNAs were typically designed to target the first conserved domain of the protein, as listed on Ensembl, to increase the probability that any insertions or deletions (indels), both in-frame and out-frame shifts, abolish protein function. To generate maternal crispants, a cocktail containing multiplexed guide RNAs targeting a single gene and Cas9 protein was injected into one-cell (F0) embryos. Most embryos injected with the multiple guide RNA:Cas9 cocktail (>80%, compared with uninjected controls) were viable up to 5 days post-fertilization (dpf; Fig. S1). Once the injected F0 embryos reached adulthood, the resulting females (hereafter referred to as ‘F0 females’) were screened for maternal-effect phenotypes by scoring their offspring for developmental abnormalities (Fig. 1B). The resultant clutches from F0 females, presumably depending on the fraction of the germ line containing biallelic lesions, exhibited a range of phenotypes, from 0 to 100% affected embryos.

Three known maternal-effect genes, birc5b, tmi and mid1ip1l, which when mutated result in defects in cytokinesis in the early embryonic cell divisions (Eno et al., 2016; Nair et al., 2013, 2021 preprint), were chosen to validate the maternal crispant approach. Guide RNAs targeted the conserved regions for birc5b (BIR domain; Fig. 2A) and tmi (start of the MAP65/Ase/PRC1 domain; Fig. 2B), and the single exon for mid1ip1l (Fig. 2C). F0 females were mated to wild-type males, and the resulting F1 clutches were examined for phenotypes. In nearly all cases (9 of 10 F0 females), F1 clutches contained embryos with defects indistinguishable from those of embryos derived from females homozygous for the corresponding target gene (birc5b: 78-100%; tmi; 27-47%; mid1ip1l: 0-60%) (Table 1, Fig. S2). Two of the F0 females (for birc5b) produced 100% maternal crispant embryos, indicating that in these females most of, if not the entire, germ line contained biallelic edits. The remaining F0 females produced a fraction of embryos exhibiting the target mutant phenotype as well as wild-type embryos, reflecting a mosaic germ line. One F0 female (for mid1ip1l) did not yield maternal crispant embryos.

Characterization of the maternal crispants confirmed cellular defects identical to those caused by the loss of function of the corresponding target genes. Like embryos from birc5b homozygous females (Nair et al., 2013), birc5b maternal crispant embryos failed to create a visible furrow (Fig. 2D) and, despite normal DNA segregation, failed to accumulate the cell adhesion component β-catenin at furrow sites (Fig. 2G). Similar to embryos from females homozygous for tmi mutations (Nair et al., 2021 preprint), tmi maternal crispant embryos showed incipient cleavage furrows that underwent regression (Fig. 2E), and exhibited reduced β-catenin deposition at the furrow and aberrant chromosome segregation (Fig. 2H). Corresponding to embryos from females homozygous for mid1ip1l mutations (Eno et al., 2016), mid1ip1l maternal crispants exhibited abnormal cell division and became partially syncytial (Fig. 2F) with β-catenin present in unreleased cortical granules and unavailable for recruitment to the furrow (Fig. 2I).

In summary, for all three targeted known maternal-effect genes, a maternal crispant approach reliably replicates loss-of-function phenotypes.

The analysis of genes required for maternal functions is made difficult because of the requirement for additional generations to assess phenotypes in the offspring. In the case of maternal genes also required zygotically, the requirement for zygotic function also precludes the possibility of obtaining viable individuals to test maternal function. Maternal-specific genes typically have high transcript levels in the early embryo with a severe reduction in mRNA levels as the embryo undergoes ZGA (approximately 900 genes) (Aanes et al., 2011). In addition, a number of characterized maternal-effect mutations in zebrafish occur in genes that arose through gene duplication events, with the gene affected by the maternal-effect mutation expressed largely or exclusively maternally, and other copies, presumably the ancestral ones, retaining a more widespread function (Campbell et al., 2015; Eno et al., 2016; Ge et al., 2014; Nair et al., 2013, 2021 preprint). This pattern is consistent with the functional divergence of duplicated genes (Sandve et al., 2018), leading to some genes acquiring a maternal-only function. Thus, in these gene families, the zygotically expressed paralog(s) will perform essential functions in the somatic cells, thus permitting somatic development. In contrast, the maternally expressed duplicated paralog will only affect developing oocytes and progeny, allowing for the determination of its role in early development.

To identify candidates to target using a maternal crispant approach, we used an RNA-seq database that provides a baseline transcriptional profile for multiple developmental stages of the zebrafish, from zygote to 5 dpf (White et al., 2017). We searched for uncharacterized genes that: (1) displayed an expected profile for a maternal-specific gene, i.e. expressed in the oocyte and early embryo and degraded at ZGA, and (2) are gene duplicates for which one or more additional copies are expressed both maternally and zygotically, or zygotically only, reflecting a possible role for this second gene duplicate throughout development. These criteria are expected to enrich for maternally expressed gene duplicates for which knock out would have no effect on zygotic development in F0 fish but could lead to maternal-effect phenotypes. Using these criteria, we selected four candidate maternal-effect genes: kpna7, fhdc3, zgc:114123 and kcns3b. These genes are predicted to carry out a diverse set of cellular functions in the zebrafish embryo, related to nuclear import (kpna7), regulation of actin dynamics (fhdc3 and zgc:114123), and maintenance of cell membrane potential (kcns3b). The application of the maternal crispant technique to these genes allowed for the identification of maternal-effect phenotypes for two of these genes, kpna7 and fhdc3. Targeting of zgc:114123 and kcns3b led to clutches from F0 injected females that contained only phenotypically wild-type embryos (Table S1, Fig. S3), despite the production of multiple indels in F0 embryo somatic tissues 24 h after injection (Fig. S4).

In eukaryotic cells, large proteins are actively transported from the cytoplasm into the nucleus via a transport system mediated by a heterodimeric Importin α/β (also known as Karyopherin α/β) complex, which recognizes cargo containing a nuclear localization signal (NLS) (Conti et al., 1998; Görlich et al., 1994; Imamoto et al., 1995; Moroianu et al., 1995; Yasuhara et al., 2004).

Eukaryotes contain a varying number of Importin α genes. For example, yeast has a single Kpna gene, whereas mammals have seven Kpna genes (Hu et al., 2010; Kelley et al., 2010; Tejomurtula et al., 2009). Among mammalian Kpna genes, only Kpna7 shows expression in the oocyte and the early embryo and RNA degradation at ZGA (Fig. 3A) (Boroviak et al., 2018; Hu et al., 2010; Tejomurtula et al., 2009; Wang et al., 2012). Similar to mammals, zebrafish also have seven Kpna genes (Fig. 3B), with kpna7 exhibiting maternal-specific expression and transcript degradation at ZGA (Fig. 3B). Therefore, we targeted zebrafish kpna7, which has a canonical structure for this family of proteins composed of an N-terminal Importin-β binding domain and multiple armadillo repeats at the C terminus (Fig. 3C) (Conti et al., 1998; Görlich et al., 1994).

For the zebrafish kpna7 maternal crispant, two guide RNAs were designed to target the Importin-β binding domain and two additional guide RNAs to target the first Armadillo repeat (Fig. 3C). Clutches from two different kpna7 F0 females exhibited a maternal phenotype in 100% of the embryos (Table 2). During the cleavage stages, the kpna7 maternal crispant embryos were morphologically indistinguishable from wild-type embryos. However, as they underwent development, the majority of these embryos stalled at sphere stage, with only a small fraction of embryos initiating epiboly and none completing this process (Fig. 4A). Labeling for β-catenin and DAPI at 6 h post-fertilization (hpf) showed that kpna7 maternal crispant embryos exhibit an uneven distribution of nuclei, with a subset of cells appearing to lack DNA entirely (Fig. 4B, Movie 1) (n=15) and others exhibiting nuclei of a range of sizes and shapes, in contrast to their even distribution and size in wild type (Fig. 4B, Movie 1) (n=16).

In contrast to kpna7 knockdown cancer cell lines, which exhibit supernumerary centrosomes (Vuorinen et al., 2018), all cells in kpna7 maternal crispants showed the normal centrosome pair component (Fig. 4C; wild type: n=44 cells in 6 embryos; kpna7 maternal crispants: n=39 cells in 7 embryos). However, kpna7 maternal crispants exhibited defective DNA morphology in a majority of cells, reflected in nuclear material that failed to segregate to the spindle poles (Fig. 4C; wild type: 0/22 abnormal DNA segregation events in 6 embryos; knpa7 maternal crispants: 17/21 abnormal DNA segregation events in 7 embryos). This defect, likely compounded through multiple rounds of cell division, presumably results in the uneven nuclear size and distribution observed at later stages. More comprehensive studies will be needed to study Kpna7 function in early development.

Proteins of the Formin family are key regulators of cytoskeletal reorganization through the promotion of actin nucleation in the cell, with a well-established role in processes such as cytokinesis, cell polarization and filopodia assembly (Breitsprecher and Goode, 2013; Goode and Eck, 2007; Hegsted et al., 2017). A subset of proteins in the Formin family is known as ‘inverted formins’, in which the regulatory N-terminal region has been lost, leaving the FH1 and FH2 domains at the N-terminal region of the protein (Breitsprecher and Goode, 2013; Hegsted et al., 2017; Higgs and Peterson, 2005). Inverted formins have been shown to be involved in the linking of F-actin and microtubules and have a role in membrane trafficking and lumen formation (Breitsprecher and Goode, 2013; Hegsted et al., 2017; Higgs and Peterson, 2005; Young et al., 2008).

In zebrafish, the Fhdc family consists of five different genes: fhdc1, fhdc2, fhdc3, fhdc4 and fhdc5; fhdc3 is the only member with maternal-only expression and fhdc5 is not expressed during the first 24 hpf (Fig. 5A). To generate fhdc3 maternal crispants, guide RNAs were designed with one guide RNA targeting the protein before the FH1 domain and the other three guide RNAs targeting the start of the FH2 domain (Fig. 5B). Three fhdc3 F0 females were screened, and all three generated clutches contained embryos exhibiting a consistent maternal-effect phenotype (Table 2). These embryos undergo normal cell cleavage during the first few cell cycles but, at approximately the 32-cell/64-cell stage, start to exhibit an aberrant constriction at the yolk-blastodisc boundary (Fig. 6A, black arrowhead). Immunolabeling of fhdc3 maternal crispants with β-catenin and DAPI showed that at the 8-cell stage, the cellular layout (a 2×4 cell matrix at this stage) is similar between the fhdc3 maternal crispants and stage-matched wild-type controls (Fig. 6C). However, at the 64-cell stage, when wild-type embryos exhibited a regular cellular layout (organized largely as two tiers of blastomeres, each in a 4×8 cell matrix), fhdc3 maternal crispant embryos exhibited a disorganized cellular layout (Fig. 6C). The blastoderm of early wild-type embryos has an oblong shape (Kimmel et al., 1995), possibly caused by the coupling of blastomere division to spindle lengthening (Wühr et al., 2010; our own observations). Using a score of circularity (with a perfect circle having a score of 1) to assess blastoderm shape broadly, we found that, starting at the 64-cell stage, fhdc3 maternal crispant embryos are significantly less oblong than wild type (Fig. 6D). Altogether, our observations indicate that embryos lacking fhdc3 function are unable to maintain the ordered cellular arrangement of the early wild-type blastoderm.

Despite earlier defects, by late-cleavage stages fhdc3 maternal crispants no longer showed an abnormal constriction of the yolk-blastoderm boundary phenotype, instead appearing at these stages, morphologically, relatively normal (Fig. 6A). Nevertheless, at these stages, fhdc3 maternal crispant embryos contained presumptive ectopic yolk inclusions in varying numbers and sizes embedded in the blastodisc, presumably from a defective cytoplasm-yolk boundary at earlier stages (Fig. 6A). In addition, fhdc3 embryos showed a consistent delay in the initiation epiboly (Fig. 6A), although embryos did not show yolk closure defects characteristic of previously described strong epiboly delayed mutant phenotypes (Kane et al., 1996) (Fig. 6B). Despite yolk inclusions and a delay in epiboly, fhdc3 maternal crispants were able to proceed through gastrulation without apparent gross morphological defects up to 5 dpf (Fig. 6B, Fig. S5).

A closer examination of fhdc3 maternal crispants at 24 hpf showed the continuing presence of presumptive ectopic yolk inclusions distributed in various regions of the embryo (Fig. 6B, black arrowheads, Fig. S5). fhdc3 maternal crispant embryos that contained inclusions were significantly less likely than wild-type controls to be viable at 5 dpf, as assessed by swim bladder inflation (wild type: 96%, n=72; fhdc3 maternal crispants: 39%, n=109; P=0.009) (Fig. S5).

The distinct aberrant constriction at the yolk-blastodisc boundary in fhdc3 maternal crispants, and a predicted role for Fhdc3 as an actin regulator, led us to analyze the cortical F-actin ring in fhdc3 maternal crispant embryos. Visualization of the yolk-blastodisc boundary of 16-cell embryos with rhodamine-conjugated phalloidin showed that the cortical F-actin ring in fhdc3 maternal crispant is wider compared with time-matched wild-type controls (fhdc3 maternal crispants: 24.8 μm, n=10; wild type: 14.1 μm, n=10; P=0.0003) (Fig. 6E).

Our observations suggest that fhdc3 plays an essential role in regulating the yolk-blastodisc boundary in the early embryo, and is required for segregation of the yolk away from the blastodisc and, eventually, embryonic survival.

The multiple populations of alleles in traditional crispants, which typically occur in diploid cells and involve a mixture of induced lesions, can lead to challenges in interpreting the sequences of targeted genes. To facilitate the analysis of lesions in maternal crispants, we took advantage of the ability to generate haploid embryos in zebrafish (Walker, 1998) and sequenced alleles present in single haploid maternal crispant embryos. This combined approach allowed for straightforward identification and characterization of genetic changes to the target sites by Sanger sequencing (Fig. 7).

We identified genetic lesions in single maternal crispant haploid embryos from birc5b, tmi, kpna7 and fhdc3 maternal crispants. mid1ip1l maternal crispant embryos had a fragile cortex, as previously shown in embryos from females homozygous for the mid1ip1l mutation (Eno et al., 2016), did not survive the manipulation and were excluded from the analysis. All generated maternal crispant haploid embryos were sequenced (3-13 embryos per female, see Materials and Methods). Every sequenced maternal crispant haploid contained genetic lesions in the expected targeted gene, consistent with the germ line containing biallelic mutations in those targets (Figs 8 and 9, Table 3).

In traditional zygotic crispants, it has been reported that the injected embryos are mosaic and contain multiple alleles, a majority of which are frameshifts leading to premature stop codons (Trubiroha et al., 2018; Wu et al., 2018). In these zygotic crispants, when multiple guide RNAs are used, mutations are observed in all four guide sites at varying frequencies, with some lesions being larger deletions spanning multiple guide sites (Wu et al., 2018). Similar to traditional crispants, edited lesions in maternal crispants included mutations in multiple guide sites as well as site-spanning deletions, with a majority of DNA lesions containing frameshifts and premature stop codons (Figs 8 and 9, Table 3). We found that maternal crispant embryos exhibit lesions in only a subset of the guide sites (Table 3), with the same guide sites being edited in different F0 females (Figs 8 and 9). For example, birc5b maternal crispants from three different F0 females contained mutations only in guide sites three and four, and fhdc3 maternal crispants from three different F0 females only in guide site four (Figs 8 and 9, Table 3). This is consistent with previous studies demonstrating that selective site usage and repair are influenced by the cellular environment (Allen et al., 2018).

Collectively, DNA lesions observed in the set of individually analyzed haploid maternal crispants represent the mixture of induced alleles present in the germ line, either in different germ cells or in either of the two different gene copies present in diploid, pre-meiotic germline cells. Thus, sampling of targeted lesions in maternal crispants is expected to reveal multiple lesions, reflecting both the mosaicism and diploid content of the maternal germ line. Consistent with this expected complexity, we observed up to four different alleles in clutches of maternal crispant embryos from single F0 females, with an average of approximately two alleles per clutch. Assuming random segregation, each of the two maternal alleles in germline cells containing biallelic lesions is expected to be transmitted to maternal crispants in an approximately 1:1 ratio. Interestingly, lesions identified in single maternal crispants haploids were not recovered in such equal ratios. For example, of ten haploid maternal crispant embryos sequenced from tmi crispant fish two, all had an identical allele (Fig. 9A). This unexpected uneven recovery of germline lesions could reflect a biased repair mechanism of CRISPR-Cas9-induced lesions in the germ line, as observed in other systems (Allen et al., 2018; Brunner et al., 2019).

In summary, we identified genetic lesions in targeted maternal genes in all analyzed haploid embryos from F0 females, consistent with maternal crispant progeny embryos being derived from germ cells with biallelic lesions in those gene targets.

Confirmation that observed maternal crispant phenotypes correspond to the targeted genes requires the analysis of the phenotype in stable CRISPR-Cas9 mutant lines, an approach that we carried out for kpna7. We outcrossed an F0 male sibling of an F0 female that yielded maternal crispant embryos to propagate a CRISPR-Cas9 edited allele, kpna7^uw107. This allele contains two mutations in the kpna7 gene, one in guide site three (6 bp deletion) and one in guide site four (11 bp deletion) (Fig. S6A), with the second mutation creating a premature stop codon (Fig. S6B). As expected, homozygous mutant females for kpna7^uw107 led to a 100% phenotype in progeny embryos similar to that of kpna7 maternal crispants: most embryos stalled at the sphere stage (Fig. S6C), with only a fraction of the embryos initiating epiboly and all embryos lysing prior to 24 hpf (n=365 embryos). The phenotype of embryos from non-mosaic kpna7^uw107 homozygous females confirms the target in the corresponding maternal crispants and provides proof of principle for the use of maternal crispants to identify maternal-effect genes.

In this study, we developed a reverse genetic approach to identify, in a single generation, maternal-effect genes required for early embryonic development of the zebrafish. The maternal crispant strategy described in this paper utilizes the biallelic editing abilities of CRISPR-Cas9 through multiplexing of guide RNAs (Burger et al., 2016; Jao et al., 2013; Shah et al., 2015; Shankaran et al., 2017; Trubiroha et al., 2018; Wu et al., 2018) to identify maternal-effect genes in a resource-efficient manner. As a proof of principle for the maternal crispant approach, we recovered maternal crispant embryos for five out of the seven genes tested. Furthermore, the combination of maternal haploids and Sanger sequencing allowed us to identify and characterize indels in maternal crispants, facilitating the association of observable phenotypes to the gene targets. We additionally validated the kpna7 maternal crispant phenotype using an established stable line. Thus, our studies show that the production of maternal crispants is an effective strategy for identification of maternal-effect genes.

Using a maternal crispant approach, we were able to identify maternal-effect phenotypes reliably. In experiments targeting the five genes for which maternal crispant phenotypes were observed, most (14/15) screened F0 females produced embryos that showed maternal-effect phenotypes. The clutches recovered from F0 females contained on average 69% maternal crispant embryos. The combination of these results suggests efficient biallelic editing of the germ line for multiple targets, thus allowing for the identification of maternal-effect phenotypes in the progeny of injected (F0) fish.

Other advantages of the maternal crispant screen are that the clutches can contain an internal control, and it is straightforward to establish stable mutant lines. The mosaic clutches that occur when phenotypically wild-type embryos are derived from oocytes without biallelic genetic edits in a mosaic F0 germ line can act as an internal control for variables such as fertilization time and developmental rate. This method also allows the simultaneous establishment of stable mutant lines, either from F0 sibling males, as demonstrated with the kpna7^uw107 allele, or from F1 offspring carrying induced lesions. These stable lines can be used to confirm phenotypes observed in the maternal crispants.

In these studies, we only identified maternal-effect phenotypes from gross morphological defects, and this approach may identify a large fraction of maternal-effect mutations. However, lethality might not be a necessary outcome for a maternal-effect phenotype. For example, depletion of Rest (NRSF) in zebrafish embryos causes the upregulation of target genes leading to changes in primary motor neuron architecture and behavior (Moravec et al., 2016). The use of transgenic lines that highlight specific structures or cell types, as has been done for heart and thyroid development in zygotic crispants (Trubiroha et al., 2018; Wu et al., 2018), would further increase the power of the approach. This maternal crispant technique also has the potential to determine the maternal function of genes required both maternally and zygotically, by specifically targeting Cas9 activity to the germ line (Moreno-Mateos et al., 2015).

A survey of haploid maternal crispants showed that genetic lesions tended to involve certain guide sites but not others. In some cases, multiple independently created lesions involving identical improperly repaired sequences were found in the progeny of different F0 females. This apparent guide site selectivity contributed to limited genetic variation in maternal crispants for specific genes.

Interestingly, we found that a majority of edited alleles are not recovered equally in maternal crispants of the same clutch. For example, in the case of tmi, ratios of recovery of two different alleles in haploid embryos were 7:1 (female 1) and 10:0 (female 2). Depending on the time of the edit, different numbers of germ cells will carry newly induced alleles, thus influencing the ratio of alleles present in the germ line. The uneven recovery of alleles could be caused by a repair confirmation bias for specific indels (Allen et al., 2018), resulting in identical independent edits in one or more germline cells. Another possibility is that a double-stranded break in one chromatid is corrected using an edited sister chromatid as a template, through homologous recombination/homology-directed repair (Brunner et al., 2019), creating identical mutations in both copies of the gene in the same germline cell. We cannot rule out an alternative explanation whereby the observed uneven allele recovery is caused by biased segregation during meiosis, possibly caused by damage to one chromatid.

We identified genes that are likely candidates to have a maternal-effect function using a combination of an RNA-sequencing database (White et al., 2017) and phylogenetic analysis. In eukaryotes, whole-gene duplications have allowed significant evolutionary events to occur because the functions of the duplicated genes are less constrained (Long et al., 2003; Lynch and Conery, 2000; Ohno, 2013). In these cases, one of the gene copies can provide an essential basic function, allowing a second gene copy to acquire new functions. The zebrafish genome has undergone multiple rounds of whole-genome duplications (Meyer and Schartl, 1999), with a portion of duplicated genes experiencing functional divergence to develop new roles, including a maternal-specific function. Bioinformatic analysis has determined that the zebrafish genome has 3991 duplicated sets of genes (Lu et al., 2012). A rough estimate suggests that out of the approximately 900 maternally expressed genes there may be over 100 genes that additionally belong to families of gene paralogs that can be tested with this approach. Our studies are consistent with this phenomenon and suggest that a combination of expression data coupled to phylogenetic analysis allows for the effective identification of gene candidates with maternal-specific functions.

Using the above criteria, we selected four candidate maternal genes and tested their function using a maternal crispant approach. Out of these four genes, we identified maternal-effect phenotypes in two of the genes, kpna7 and fhdc3.

Past research in mice has shown that loss-of-function conditions for kpna7, an Importin α involved in nuclear transport, results in imbalanced sex ratios, with mutant females producing embryos with an embryonic-lethal phenotype (Hu et al., 2010). Knockdown of this maternally expressed transcript in other mammalian embryos, i.e. cattle and pigs, also showed reduced cellular cleavages leading to arrested embryonic development, similar to the 100% maternal-effect lethal phenotype that is observed in zebrafish (Tejomurtula et al., 2009; Wang et al., 2012). This developmental stall in zebrafish is likely due at least partly to the elimination of nuclei from a subset of cells during development, which precludes zygotic genome activation and epibolic movements in those cells (Kane et al., 1996; Knaut et al., 2000; Lee et al., 2013; Lindeman and Pelegri, 2012). The kpna7 maternal crispant suggests a link between Kpna7 function in nuclear import and DNA segregation.

A second maternal crispant target in this study was fhdc3, a maternally expressed homolog of the inverted formin fhdc1. fhdc3 maternal crispants exhibit an exaggerated constriction of the cytokinetic ring. In wild type, this actin-based ring forms at the margin of the blastodisc during the early cleavage stages and is thought to drive ooplasmic flow and yolk segregation during cycles of contraction and relaxation associated with the cell cycle (Fuentes and Fernández, 2010; Leung et al., 2000). The exaggerated constriction phenotype in fhdc3 maternal crispants is no longer observed after the 6th cell cycle, a time coincident with the disappearance of the cytokinetic ring in wild type. However, fhdc3 crispant embryos continue to exhibit defects as reflected by presumptive ectopic yolk inclusions in the embryo proper and reduced embryonic viability. Thus, fhdc3 is a maternal gene with a temporarily restricted yet essential function in regulating the F-actin-based cytokinetic ring during the earliest embryonic cell cycles, which contributes to proper segregation of yolk from the cytoplasm.

We did not identify maternal-effect phenotypes for two targeted genes, the Spire family gene zgc:114123 and the S subfamily potassium channel kcns3b, despite efficient production of indels in somatic cells of injected F0 embryos. Considering the expression pattern of zgc:114123 and the role of Spire in oogenesis in other systems (Bradley et al., 2020; Leader et al., 2002; Manseau and Schüpbach, 1989; Pfender et al., 2011), it is possible that a loss of function for zgc:114123 affects oocyte development, precluding the production of maternal crispant embryos in F1 clutches. The number of maternal transcripts for kcns3b is significantly lower than other targets in our studies, so that the absence of maternal crispants for this gene may reflect a minor or non-essential role for this gene during development.

In summary, this study presents an effective approach for producing embryos with loss of function for maternal-effect genes, taking advantage of a high rate of CRISPR-Cas9-induced biallelic editing in the germ line. The resulting maternal crispants allow for the identification and characterization of maternal-effect genes in a single generation. In teleost fish, a subset of maternal genes has arisen from gene duplication events, and targeting maternal-specific duplicate genes could provide insight into the role of gene homologs in development and evolution. Thus, a maternal crispants approach could help to expedite our understanding of the role of maternal products during development.

Stocks of wild-type AB or crispant fish were raised and maintained under standard conditions at 28.5°C (Brand et al., 2002). Embryos were collected by natural matings and allowed to develop in E3 embryonic medium (Brand et al., 2002), with staging according to age and morphological standards (Kimmel et al., 1995). All zebrafish housing and experiments were approved by the University of Wisconsin-Madison Institutional Animal Care and Use Committee.

The CRISPR design web tool CHOPCHOP was used to identify guide RNA sites with high activity and minimal predicted off-target sites (Labun et al., 2016; Montague et al., 2014; Moravec and Pelegri, 2019). Four guide RNAs were selected for each target gene with the following criteria: predicted to have high cleaving activity; located in or near the first predicted protein domain (as identified by Ensembl; Aken et al., 2016); and minimal to no overlapping regions between guide RNAs.

Guide RNA templates were produced using an annealing/fill-in method as described (Gagnon et al., 2014). Briefly, guide RNA templates were generated with a gene-specific oligo containing an SP6 promoter sequence, a gene-specific target sequence, and an overlapping sequence complementary to the constant oligo (Table S2). Annealing/fill-in was performed using T4 DNA polymerase using 100 µM of each of the oligo fragments. Oligos for each of the four gene-specific guide RNAs were separately annealed to the constant oligo. After the T4 DNA polymerase fill-in, the four guide RNA templates were pooled together and concentrated by purification using a DNA Clean and Concentrator kit (Zymo Research).

A pooled guide RNA cocktail was synthesized using the MEGAshortscript T7 Transcription Kit (Invitrogen) using the concentrated pooled guide RNA template. After RNA synthesis, the guide RNA pool was purified using an ethanol/ammonium acetate protocol. The concentration of the guide RNA pool was measured by a NanoDrop ND-2000 spectrophotometer, and its integrity was checked by gel electrophoresis. Single-use aliquots of guide RNAs were stored at −80°C.

A mixture of the pooled guide RNA cocktail (200 pg/nl total final concentration) and Cas9 protein (PNA Bio; 400 pg/nl final concentration) was injected into one-cell-stage AB embryos. At 24 hpf, a subset of the injected embryos was collected, and DNA was extracted as previously described (Meeker et al., 2007). Guide activity was assayed by amplifying 120-100 bp fragment across the target site via PCR (Table S2) and then resolved on a 2.5% agarose gel. If Cas9-mediated genetic edits were created in the somatic cells, a population of DNA that contains a variety of indels would generate a smear, confirming Cas9 activity (Moravec and Pelegri, 2019).

Injected embryos were raised until sexual maturity, and then all F0 females were mated through natural crosses to AB males. F1 embryos were collected approximately every 15 min and were allowed to develop in 100-mm Petri dishes with 1× embryonic medium (E3); each 100 mm-diameter dish contained no more than 100 embryos to ensure proper medium oxygenation. Control synchronized wild-type embryos and phenotypically wild-type embryos present in the F1 clutches were used to determine time from fertilization. Development was monitored in live embryos every hour for the first 8-10 h, at 24 hpf, and twice a day thereafter until 5 dpf. Maternal crispants were scored for possible gross morphological defects, such as defective cellular division, deviations from the wild-type body plan, developmental delays, and failure to develop a swim bladder. For documentation of maternal-effect phenotypes, embryos younger than 24 hpf were left in their chorions and imaged on glass-bottom dishes containing 1× E3, and embryos older than 24 hpf were dechorionated and imaged on glass-bottom dishes containing 3% methylcellulose. All live images were acquired using a Leica MZ FLIII microscope equipped with a Leica EC3 camera and Leica Acquire version 1.0 RC4 software.

Embryos were fixed in 4% paraformaldehyde in PBS and immunolabeled as previously described (Lindeman and Pelegri, 2012), with antibody incubation overnight at 4°C. Primary antibodies were rabbit anti-β-catenin (1:1000, Sigma-Aldrich, C2206), mouse anti-β-catenin (1:1000, Sigma-Aldrich, C7207) and mouse anti-γ-tubulin (1:2000, Sigma-Aldrich, T6557). Secondary antibodies were goat anti-rabbit IgG, Alexa 488 (Molecular Probes, A-11008), goat anti-rabbit IgG, Cy3 (Jackson ImmunoResearch, 111-165-144), rabbit anti-mouse, Alexa 488 (Molecular Probes, A-11059) and goat anti-mouse IgM, Cy3 (Jackson ImmunoResearch, 115-165-062). Actin was labeled as previously described (Eno and Pelegri, 2018) with rhodamine-conjugated phalloidin. To detect DNA, embryos were labeled with a 0.5 μg/ml solution of 4′,6-diamidino-2-phenylindole-dihydrochloride (DAPI) for 10 min followed by three washes in PBS. After labeling, depending on imaging conditions, embryos were either directly mounted sideways on the slide or deyolked and mounted flat with the animal pole upwards, in both cases in 75% glycerol. Images were obtained using a Zeiss LSM780 confocal microscope and analyzed using ImageJ.

To examine the shape of the early fhdc3 and wild-type embryos, traces of the perimeters of embryos at 1.25 and 2 hpf were analyzed using ImageJ. Embryos with presumptive ectopic yolk inclusions were classified as having either small or large inclusions (less than or more than one-eighth of the blastodisc volume, respectively) by estimating the volume of the largest yolk inclusion in the blastodisc at sphere stage and, together with AB wild-type controls, were grown individually in 24-well tissue culture dishes containing 1-1.5 ml of E3 for 5 days. At 24 hpf, the location of the yolk inclusion was scored according to the location along the body plan as: head region (anterior to the posterior end of the otic vesicle), mid-section (from the otic vesicle to the posterior end of the yolk extension), or tail region (posterior to the end of the yolk extension). The presence of an inflated swim bladder was scored at 5 dpf as a sign of embryonic viability. Statistical analysis was performed using unpaired two-tailed Student's t-test.

Females that produced maternal crispant embryos were used for in vitro fertilization (IVF) modified for haploid production, using UV-irradiated sperm as previously described (Baars et al., 2016). During maternal crispant haploid production, to minimize in adult females side-effects associated with the modified IVF procedure, the F0 adult females were manipulated as little as possible (ideally only once), and all resulting maternal crispant haploid embryos, which varied in fraction within the clutch and number, were sequenced. Briefly, the night before the IVF procedure, single F0 adult females were placed into mating boxes together with single wild-type AB males, with a physical barrier separating males and females. On the morning of the IVF procedure, the barrier was removed, allowing natural mating. Tanks were monitored, and the F0 females were separated immediately after the first signs of egg release. In the same morning, a sperm solution was prepared using wild-type testes macerated in Hanks’ solution (corresponding to testes from one male per 100 µl of Hanks’), disseminating sperm into solution, with the solution then UV-treated for 90 s to inactivate its DNA. Separated F0 females were anesthetized with tricaine solution (2% buffered tricaine in 100 ml of conditioned fish water), and mature eggs were extruded by applying gentle pressure to the abdomen. The extruded eggs were fertilized by the addition of 100 µl of UV-treated sperm solution with gentle mixing, followed by water-activation through the addition of E3 embryonic medium (1 ml for the first minute and flooding of the plate after the first minute). The resulting maternal haploid embryos were allowed to develop for 6-8 h, with the time period augmenting DNA content through DNA replication in the developing embryos. After 6-8 h, DNA was extracted from embryos that displayed a maternal crispant phenotype as previously described (Meeker et al., 2007), except that we used half of the volume (50 µl of 20 mM NaOH and 5 µl of Tris-HCl pH 8.0) compared with the original protocol in order to concentrate the DNA.

To analyze genetic lesions in maternal haploids, DNA fragments corresponding to the targeted genes were PCR amplified using genomic DNA extracts from single haploid embryos. The PCR primers were designed to include multiple guide sites in order to capture potential site-spanning indels (Table S2). Duplicate PCR reactions of each sample were carried out, and the amplified products were pooled using a DNA Clean and Concentrator kit (Zymo Research). The resulting amplified fragments were analyzed by Sanger sequencing using forward and reverse sequencing primers (Table S2). Sequence results were analyzed using Sequencher.

We thank past and current members of the Pelegri lab for their contributions to our research, particularly our animal husbandry staff for the care of the aquatic facility. We are also grateful for the comments and insight on the manuscript by Christina Hansen and Ryan Trevena. Microscopy was performed at the Newcomb Imaging Center, Department of Botany at the University of Wisconsin-Madison.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.199536