Mosaic CRISPR-stop enables rapid phenotyping of nonsense mutations in essential genes

The emergence of CRISPR/Cas9-mediated gene editing has revolutionized the use of mouse genetics for functional studies (Cong et al., 2013). Base editor (BE) is generated by the fusion of Cas9 nickase (nCas9) with rat APOBEC1 or human APOBEC3, and uracil glycosylase inhibitor (UGI). Multiple versions of BE exist, but the key concepts underlying their function are the same: with the guidance of gene-specific small-guide RNAs (sgRNAs), BEs mediate the direct conversion of C to T (or G to A) in the mouse genome (Komor et al., 2016). When the C-to-T conversion occurs in some codons, it introduces a premature stop that, in the appropriate location, results in a loss-of-function effect, and is termed CRISPR-stop (Billon et al., 2017; Kuscu et al., 2017). In the developmental biology field, the use of LOF studies is considered essential for demonstrating the necessity of a gene (Housden et al., 2017). We previously used CRISPR-stop in one-cell stage mouse zygotes, and the mice directly derived from these manipulated zygotes were homozygous mutants and suitable for immediate phenotypic analysis (Zhang et al., 2018); using this approach, mouse breeding is bypassed and the time required for loss-of-function genetic screening is considerably shortened, particularly in the case of nonlethal genes where homozygous mutants survive.

Homozygous mutants of ∼25% of mammalian genes show embryonic or perinatal lethality, which precludes determination of the functions of the genes at older ages (White et al., 2013). A typical solution for overcoming this hurdle involves generating a conditional knockout model, where the key exon(s) of a gene of interest is first sandwiched between two loxp sequences, followed by crossing with mice in which Cre is expressed in a highly tissue-specific manner or CreER expression is temporally controlled using tamoxifen (Cox et al., 2012). Although this conditional knockout approach has contributed substantially towards the determination of the functions of several lethal genes at postnatal ages, the method is time-consuming and requires at least two additional rounds of breeding after obtaining the conditional mouse strain. Thus, it remains necessary to establish an alternative high-throughput genetic approach specific for screening lethal genes.

As illustrated by MARCM in Drosophila and MADM in mouse, mosaic (or chimeric) analysis represents a powerful approach for studying gene functions in animals that would otherwise die in the germline-null background (Lee and Luo, 1999; Zong et al., 2005). Therefore, our primary aim was to rapidly obtain homozygous mosaic Founder 0 (F0) mice that survive longer than regular knockout mice and are immediately ready for phenotypic analysis. Two-cell stage mouse embryos have advantages for gene targeting, and manipulating one of the two blastomeres is a successful approach to generating mosaic mice by methods other than CRISPR-stop (Gu et al., 2018; Wang et al., 2017; Wu et al., 2019). Here, to demonstrate the advantages of CRISPR-stop in generating homozygous mosaic F0 mice, we selected the mouse auditory organ, the cochlea, as the model, as in our previous study (Zhang et al., 2018).

The mammalian inner ear derives from the otocyst at approximately embryonic day 9 (∼E9), and the cochlea is located in the ventral region of the inner ear. The cochlear auditory epithelium, known as the organ of Corti, contains a single row of inner hair cells (IHCs), three rows of outer hair cells (OHCs) and multiple subtypes of supporting cells (SCs) (Driver and Kelley, 2020; Groves et al., 2013; Wu and Kelley, 2012). Atoh1, a bHLH transcriptional factor, plays a crucial role in HC fate specification, and no HCs are present in Atoh1^−/− mice, which die immediately after birth (Bermingham et al., 1999). Moreover, Sox (SRY-related HMG-box) family genes, such as Sox2 and Sox10, are necessary for normal cochlear and HC development. Sox2 expression is turned on in otocyst cells and is crucial for cochlear development, and Sox2 is widely used as a sensory progenitor marker in the embryonic cochlea (Dabdoub et al., 2008; Kiernan et al., 2005; Steevens et al., 2019). Sox10 expression, like that of Sox2, is also turned on in otocyst cells and promotes the survival of cochlear progenitors (Breuskin et al., 2009), and, notably, the cochlear duct is considerably shortened in Sox10-deficient mice (Breuskin et al., 2009) and human Sox10 mutation is associated with Waardenburg-Shah syndrome (Read and Newton, 1997). Furthermore, OHCs specifically express the motor protein prestin, whereas IHCs express the calcium sensor otoferlin (Liberman et al., 2002; Roux et al., 2006; Zheng et al., 2000). The high sensitivity of HCs to genetic mutation makes the cochlea a highly favorable model for studying the molecular mechanisms underlying the action of developmental genes, partly due to the ease of phenotypic analysis (Zhang et al., 2018).

In this study, by targeting only one of the two blastomeres in the mouse zygote, we first confirmed that the descendant cells derived from either of the two blastomeres can contribute to the inner cell mass (ICM) and to cochlear morphogenesis, which serves as the foundation for generating mosaic mice for cochlear analysis. Second, by using Atoh1 in a proof-of-concept study, we optimized our analysis pipeline, which we named mosaic CRISPR-stop; with this approach, one sgRNA used with human APOBEC3A-mediated CRISPR-stop was sufficient to generate mosaic mutants. Third, we used mosaic CRISPR-stop to generate mosaic Sox10 mutants that not only survived longer than regular Sox10 knockout mice but also exhibited the shortened cochlear phenotype. Most importantly, by using the mosaic CRISPR-stop strategy, we rapidly determined that cochlear OHCs were lost when Rbm24 was inactivated, and this OHC phenotype was further confirmed in our Rbm24 conditional knockout mice, which highlighted that mosaic CRISPR-stop was reliable and rapid.

Both blastomeres of two-cell stage mouse zygotes are totipotent and can differentiate into all cell types and integrate into all tissues of the future embryo (Edwards and Beard, 1997). Selectively manipulating one of the two blastomeres was successfully used to generate surviving mosaic Tet3 mutant mice that otherwise die at the perinatal stage in the germline Tet3^−/− background (Wang et al., 2017). However, the two blastomeres in the two-cell stage mouse zygote were recently reported to show asymmetric fate biases (Wang et al., 2018a). To confirm that, in our hands, the offspring cells of both blastomeres constituted the ICM that develops into the embryo, blastomeres from Rosa26-CAG-loxp-mTdtomato-loxp-mEGFP/+ (briefly, mTmG/+) mice (Muzumdar et al., 2007) were used for three different treatments in two-cell stage zygotes (Fig. 1A,B). In the first treatment, neither of the two blastomeres was injected with Cre mRNA, and, as expected, membrane Tdtomato expression was maintained and all examined blastocysts (n=19) were entirely Tdtomato⁺/EGFP⁻ (Fig. 1C-C″).

Second, Cre mRNA was injected into only one of the two blastomeres selected randomly, and this resulted in the turning off of Tdtomato expression but the turning on of membrane EGFP expression. Out of a total of 61 blastocysts analyzed, in 60 blastocysts (98.4%), both ICM and trophoblast (TB) cells comprised EGFP⁺ and Tdtomato⁺ cells, with each individual cell expressing either EGFP or Tdtomato (Fig. 1D-D″). In the one remaining blastocyst (1.6%), all the cells were Tdtomato⁺, which might be due to injection failure.

In the third case, both blastomeres were injected with Cre mRNA, which led to all blastocysts (n=22) being exclusively EGFP⁺ (Fig. 1E-E″). Together, our results showed that, although the two blastomeres potentially show fate bias (Wang et al., 2018a), both blastomeres contribute to the ICM. Because the entire embryo is derived from ICM descendants, we predicted that Tdtomato⁺ or EGFP⁺ cells would be distributed in all mouse tissues, including the auditory organ (the cochlea), in a mosaic manner after the second type of Cre treatment described above.

We next examined the mosaic distribution of offspring cells derived from each of the two blastomeres of mTmG/+ mice in cochlear development; for this purpose, we again subjected two-cell-stage blastomeres to the three different Cre treatments (Fig. 1). In the first case, where neither blastomere received Cre mRNA, 20 embryos were implanted and five mice were born. As expected, all cochlear cells in all five mice were Tdtomato⁺ at postnatal day 1 (P1) (Fig. 2A). In the second case, Cre mRNA was injected into one of the two blastomeres, and 144 embryos were implanted and 50 mice were born; here, at P1, 28 mice (56%) harbored both Tdtomato⁺ and EGFP⁺ cells in the cochlea (Fig. 2B), 12 mice (24%) harbored only Tdtomato⁺ cells and 10 mice (20%) harbored only EGFP⁺ cells. In the third case, both blastomeres were injected with Cre mRNA, and 19 embryos were implanted and five mice were born; as expected, all cochlear cells were EGFP⁺ at P1 (Fig. 2C). Notably, the distinct degrees of EGFP and Tdtomato mosaic distribution could be directly discerned from the tail skin cells under a fluorescence microscope (Fig. S1).

Focusing next on the 28 mice that expressed both EGFP and Tdtomato (Fig. 2B), we randomly selected 19 mice and quantified the mosaic expression specifically in HCs (Fig. 2D-G). Accordingly, the 19 mice were sorted into three categories wherein: (1) Tdtomato⁺ HCs were dominant (Fig. 2D-D″); (2) similar numbers of Tdtomato⁺ and EGFP⁺ HCs were present (Fig. 2E-E″); and (3) EGFP⁺ HCs were dominant (Fig. 2F-F″). Detailed information on these mice is summarized in Fig. 2G. Together, our results suggested that descendants of both blastomeres contributed to cochlear development in a stochastic manner. In approximately half the cases (56%), cochlear cells comprised a mixture of offspring cells derived from both blastomeres.

Considering the mosaic contribution of the two blastomeres to cochlear development, we hypothesized that, if homozygous mutations of lethal genes were induced in one of the two blastomeres, the mosaic F0 mice would survive longer than regular knockout mice. We further reasoned that, although different mosaic F0 mice might show milder phenotypes and larger phenotypic variations than regular knockout mice due to the aforementioned stochastic contribution, the key advantage would be that this rapid method would allow us to bypass the conditional knockout approach, which is time-consuming and laborious. To test our hypothesis, we selected a well-studied lethal gene, Atoh1, for the initial proof-of principle study.

Germ-line Atoh1^−/− mice die soon after birth and harbor no HCs at P0 (Bermingham et al., 1999). Here, we injected one of the two blastomeres of mTmG/+ mice with Atoh1 sgRNAs and Cre and BE3 mRNAs, and we used this as the experimental group (Fig. 3A,B). We used the same two efficient Atoh1 sgRNAs (sgRNA-1 and sgRNA-3) and BE3 as in our previous study (Zhang et al., 2018). The BE3 used here was the original rat APOBEC1 (rA1)-based BE3 (Komor et al., 2016). The control group was identical to the experimental group, except for the use of LacZ sgRNA; this control allowed us to rule out any potential developmental problem caused by BE components themselves. All the control mice in this study were injected with the same LacZ sgRNA (and this is not repeated again).

In the case of the control group, 21 blastomeres were implanted and nine mice were born; among three mice selected randomly, all were found to harbor normal HCs that were either EGFP⁺ or Tdtomato⁺ at P1 (Fig. 3C-C″). For the experimental group, 252 embryos were implanted and 75 mice were born. Based on the results of Sanger sequencing of tail DNA, 30 mice harboring potential mosaic homozygous Atoh1 mutation were selected and analyzed at P1 (n=15), P10 (n=11) and P44 (n=3). One mouse was selected for confirming germline transmission and, as expected, heterozygous Atoh1^+/− F1 (Founder 1) mice were born.

The 15 experimental group mice at P1 were further sorted into two categories according to the degree of the HC phenotype. The first category (n=4) was defined as ‘wild-type like' (Fig. 3D-D″), wherein the organ of Corti region contained no EGFP⁺ (Atoh1^−/−) cells and all IHCs and OHCs were Tdtomato⁺ (Atoh1^+/+). The second category (n=11) was defined as ‘mild', and mice in this category lost one or two rows of OHCs in a discrete manner (Fig. 3E-E″). Among the 11 mice at P10, 7 and 4 mice belonged to the wild-type like and mild categories, respectively. All three mice at P42 belonged to the mild category.

Collectively, our results suggested that mutation of Atoh1 in one of the two blastomeres successfully produced mosaic homozygous mutant F0 mice that presented the predicted phenotype of diminished HC numbers. More importantly, the mosaic Atoh1 mutant mice survived until adult ages. The HC phenotype varied among different F0 mice here, but even if Atoh1 functions had not already been recognized, our results would have allowed the importance of Atoh1 in HC fate specification to be appreciated.

Although we were using BE3 for the aforementioned Atoh1 mosaic analysis, BEs composed of human APOBEC3A, nCas9 and UGI (briefly, hA3A-BE3) were reported to show relatively higher editing efficiency in cell lines in vitro (Wang et al., 2018b; Zhou et al., 2019), which suggested the potential advantage of using hA3A-BE3 over using BE3. The cDNA sequence of hA3A-BE3 is listed in Table S1. However, whether this advantage would also hold when using one- or two-cell stage mouse zygotes was unknown. We hypothesized that the use of one efficient sgRNA combined with hA3A-BE3 might be sufficient to generate homozygous mosaic mutants; this procedure would make base editing easier to perform relative to the original one. Hereafter, we no longer used the mTmG/+ strain in order to facilitate co-staining of samples with different antibodies, and thus the injection of Cre mRNA was also no longer necessary.

We first tested our hypothesis by comparing base-editing efficiency between the most efficient Atoh1 sgRNA (sgRNA-3) injected alone and two Atoh1 sgRNAs (sgRNA-3 and sgRNA-1) injected together into one-cell-stage zygotes, with both treatments combined with the use of hA3A-BE3 (Fig. S2A,B). Sanger sequencing of blastocysts (n=24 for single sgRNA; n=25 for two sgRNAs) showed that single- and double-peak chromatographs of C/T were observed in both cases (Fig. S2C-L). Whereas the single peaks suggested efficient base editing in all blastocyst cells, the double peaks indicated a relatively lower editing efficiency or an injection time close to cell division, with base editing eventually occurring in only one of the two blastomeres. The general base-editing patterns were similar between the two scenarios (Fig. S2M). These results supported our hypothesis that one efficient sgRNA paired with hA3A-BE3 would be sufficient to induce base editing, which agreed with the higher base-editing efficiency of hA3A-BE3 than BE3 (Wang et al., 2018b).

Before switching to hA3A-BE3 for the mosaic mutant analysis of other lethal genes, we determined whether Atoh1-sgRNA-3 alone paired with hA3A-BE3 could efficiently generate mosaic Atoh1 mice. Atoh1-sgRNA-3 and hA3A-BE3 were injected into one blastomere of two-cell-stage zygotes (Fig. 4A,B); 115 embryos were implanted and 36 mice were born. Sanger sequencing of tail DNA showed that base editing failed in 3/36 mice (8.3%) but was successful in the remaining 33 mice (91.7%). The 33 mice were sorted into three categories, according to the tail-DNA sequencing patterns of C/T peaks (Fig. 4C-F) and according to the phenotype (remaining HC numbers) in mice analyzed at P1 (Fig. 4G-J, n=18) or P14 (Fig. 4K-N, n=12). Three mice died between P1 and P14. We aimed to determine the degree to which the tail-DNA sequencing pattern would match or correctly predict the HC phenotype in the cochlea.

According to the tail-DNA sequencing patterns, 7/33 (21.2%) mice were sorted into the ‘low' group, where the T-peak height was less than half the C-peak height (Fig. 4D); 8/33 (24.2%) mice were sorted into the ‘intermediate' group, where the T and C peaks were of similar heights (Fig. 4E); and 18/33 (54.6%) mice were sorted into the ‘high' group, in which the T-peak was at least twice as tall as the C peak (Fig. 4F). The higher the T peak was, the higher the number was of tail cells showing C-to-T conversion. According to the phenotypes, the mice were classified into three categories, wild-type like (Fig. 4H,L), mild (Fig. 4I,M) and severe (Fig. 4J,N), relative to the controls (Fig. 4G,K).

We focused on the 18 mice showing the ‘high' tail-DNA sequencing pattern. As noted above, three mice died before P14, and thus 10 and 5 mice were analyzed at P1 and P14, respectively. Among the 10 mice at P1, one mouse (10%) showed the wild-type like HC phenotype, whereas four (40%) and five (50%) mice showed the mild and severe phenotypes, respectively. Among the five mice at P14, four mice (80%) showed the mild phenotype and one mouse (20%) showed the severe phenotype. Detailed information on all mice produced using Atoh1-sgRNA-3 alone is summarized in Fig. 4O. Our results showed that when Atoh1-sgRNA-3 alone was used, if the tail-DNA sequencing pattern belonged to the ‘high' category, the chance of accurately predicting a severe HC phenotype was ∼40% (6/15), and the chance was increased to 93.3% (14/15) in terms of predicting a mild phenotype or higher (mild and severe phenotypes). Last, we also obtained 33 mice that were derived from 105 embryos produced by injecting both Atoh1-sgRNA-1 and Atoh1-sgRNA-3, and we performed similar analyses on these mice. The sequencing and immunostaining data are not presented here because they were similar to those shown in Fig. 4C-N, but the detailed information on all the mice is summarized in Fig. 4P. Notably, in the case of the mild phenotype (Fig. 4I,M), the second or third row (lateral side) OHCs were primarily affected, whereas IHCs (medial side) and the first row of OHCs were intact. This was likely due to the general medial-to-lateral differentiation wave that spreads among HCs (Groves et al., 2013; Tateya et al., 2019; Wu and Kelley, 2012). Thus, when the number of Atoh1⁺ cochlear sensory progenitors is decreased, they preferentially develop into HCs in the medial side. However, in the case of the severe phenotype, we could not readily determine whether the remaining HCs were IHCs or OHCs (Fig. 4J,N).

In summary, the mice that showed the ‘high' tail-DNA sequencing pattern displayed the severe HC phenotype. With a similar number of implanted embryos (115 versus 105), we found that the use of Atoh1-sgRNA-3 alone (Fig. 4O), relative to the use of two Atoh1 sgRNAs (Fig. 4P), generated more mice (18 versus 11) featuring the ‘high' sequencing pattern, as well as more mice (14 versus 7) showing mild or higher phenotypes. Thus, in the analyses detailed in the following sections, we used hA3A-BE3 paired with a single sgRNA per gene.

We also examined apoptosis in Atoh1 mosaic mutant mice produced by injecting hA3A-BE3 paired with Atoh1-sgRNA-3 alone into one of the two blastomeres. Mice showing the ‘high' tail-DNA sequencing pattern were selected for the analyses (Fig. S3A,B). Atoh1 not only specifies the development of cochlear sensory progenitor cells into HCs, but is also necessary for the survival of cochlear prosensory cells (Bermingham et al., 1999; Cai et al., 2013; Chen et al., 2002). Co-staining for cleaved (activated)-caspase 3 (a marker of apoptotic cells) and Sox2 revealed that no caspase3⁺/Sox2⁺ cells were present in the cochlear sensory region in control mice (n=3) (Fig. S3C-C″). By contrast, caspase3⁺/Sox2⁺ prosensory cells were primarily observed in apical turns in Atoh1 mosaic mutant mice (n=3) at E17.5 (Fig. S3D-D″). This agreed with the previous report that cochlear sensory cells die in a basal-to-apical gradient between E15.5 and E17.5 in Atoh1^−/− mice (Chen et al., 2002).

We next examined whether deficient SC development was observed in Atoh1 mosaic mutant mice, as in the case of Atoh1^−/− mice (Woods et al., 2004). Prox1 serves as a marker for two SC subtypes, pillar cells (PCs) and Deiters' cells (DCs), at neonatal ages (Bermingham-McDonogh et al., 2006). Double-staining for Prox1 and the pan-HC marker parvalbumin revealed that, whereas Prox1 was regularly expressed in two rows of PCs and three rows of DCs in control mice (n=3) at P0 (Fig. S3E,E′), the Prox1 expression pattern was aberrant in the mutant mice (dotted rectangle in Fig. S3F,F′). Generally, the higher the number of HCs present, the higher the number of Prox1+ SCs observed in a specific cochlear duct area. This agreed with the previous report that SC differentiation depends on the preceding differentiation of HCs (Woods et al., 2004). Overall, phenotypes of HC and SC development, and cell death could be reproduced in our Atoh1 mosaic mutant mice, which highlighted the power and reliability of the mosaic mutant model.

We next determined whether we could also reproduce the phenotypes of the lethal gene Sox10 in the cochlea by using the mosaic analysis method (Fig. 5A). Sox10^−/− mice die at perinatal ages (Breuskin et al., 2009; Britsch et al., 2001), and cochlear duct is markedly shorter in Sox10^−/− than in Sox10^+/+ mice partially due to the cell death of cochlear progenitors (Breuskin et al., 2009). Here, one pre-tested Sox10 sgRNA and hA3A-BE3 were injected into one of the two blastomeres, and 232 embryos were implanted and 77 mice were born (Fig. 5B). Sanger sequencing patterns of tail DNA showed that base editing failed in 21/77 mice (27.3%), but that 56/77 mice (72.7%) displayed successful C-to-T conversion (Fig. 5C-F).

Among the 56 mice, 14, 26 and 16 mice (25%, 46.4% and 28.6%) belonged respectively to the ‘low', ‘intermediate' and ‘high' categories (Fig. 5D-F) according to the C and T peak heights. We primarily focused on the 16 mice showing the ‘high' sequencing pattern (Fig. 5F), and we analyzed eight mice at P1 and five at P14; three mice died between P5 and P10. HC development was comparable between the control (Fig. 5G-H′) and experimental group (Fig. 5I-J′), and Sox10 was deleted in a mosaic pattern in the experimental group (Fig. 5J,J′) but not control group (Fig. 5H,H′). Cochlear duct length in the Sox10 mutant was 4684.0±101.4 μm at P1 (n=8) and 4943.8±334.8 μm at P14 (n=5), considerably shorter than 6192.4±60.3 μm at P1 (n=3) and 6041.7±23.9 μm at P14 (n=3) in the control group (Fig. 5K,L). Together, these results showed that, with the use of one sgRNA and hA3A-BE3, Sox10 expression was successfully inactivated, and the Sox10 mosaic mice not only survived until at least P14 but also displayed the phenotype of cochlear shortening.

Last, we tested whether mosaic CRISPR-stop could be used to rapidly determine the functions of Rbm24, which encodes an RNA-binding protein and was recently reported to be specifically expressed in cochlear IHCs and OHCs (Grifone et al., 2018). Rbm24^−/− mice die at early embryonic ages (Yang et al., 2014), and the roles played by Rbm24 in HCs in vivo remain unknown. We injected one of the two blastomeres with a single Rbm24 sgRNA (Rbm24-sgRNA-1) and hA3A-BE3 (Fig. 6A,B); only the Rbm24 sgRNA-1 was found to be suitable for base editing and it was functional but not as efficient as the Atoh1 and Sox10 sgRNAs (Fig. 6A). The Rbm24-sgRNA-1 was located in a GC-rich region (Fig. 6C). Accordingly, Rbm24 base editing was successfully induced by hA3A-BE3 but not by BE3.

For the experimental group, 166 embryos were implanted and 58 mice were born. Consistent with the low efficiency of Rbm24 sgRNA, a C-to-T conversion occurred in 18/58 mice (31%) (Fig. 6D-F) and failed in 40/58 mice (69%). According to the C and T peak heights in tail-DNA sequencing patterns, the 18 mice were sorted into the ‘low', ‘intermediate' and ‘high' categories, which respectively included 2 (11.1%), 12 (66.7%) and 4 (22.2%) mice (Fig. 6D-F). We focused on analyzing the four mice showing the ‘high' sequencing pattern (Fig. 6F). Co-immunostaining for Rbm24, prestin and otoferlin revealed that Rbm24 was expressed in all HCs in the control group (n=3) (Fig. 6G,G′), whereas Rbm24 expression was ablated in both IHCs and OHCs in all four experimental mice at P19 (Fig. 6H,H′). Moreover, in the absence of Rbm24, OHCs, but not IHCs, were lost (Fig. 6H,H′), and the presence of normal IHCs served as an optimal internal control and allowed us to assess the degree of OHC loss. The average ratio of OHCs to IHCs in the three cochlear turns was ∼3.25 in the control group, but this was markedly diminished, to ∼0.68, in Rbm24 mosaic mutants (Fig. 6I). Collectively, these results suggested that Rbm24 is necessary for OHC survival. This is the first report on the importance of Rbm24 in cochlear OHCs.

As a final step in our analyses, we used traditional conditional knockout approaches to validate the phenotype of OHC death following Rbm24 mutation. First, by using the CRISPR/Cas9 approach, we generated two Rbm24 mutant strains: (1) germline Rbm24^+/− knockout mice (Fig. S4A-C); and (2) conditional Rbm24 flox/+ mice (Fig. S4D-H). Furthermore, to bypass any potential haploinsufficiency due to Cre driver issues in the phenotype analysis (Atkinson et al., 2018; Matern et al., 2017), we also generated a new Atoh1-3*HA-P2A-Cre/+ (briefly, Atoh1-P2A-Cre/+) knock-in mouse strain, where Atoh1 was tagged with 3× HAs at its C-terminus and Cre was also completely under the control of the endogenous Atoh1 promoter and enhancers through the use of the 2A strategy (Li et al., 2018). Homozygous Atoh1-P2A-Cre/Atoh1-P2A-Cre mice were healthy and showed no phenotypes in the cochlea. Detailed information on the Atoh1-P2A-Cre strain will be reported elsewhere.

Double-staining for Rbm24 and the pan-HC marker parvalbumin revealed that Rbm24 was expressed in all HCs of Atoh1-P2A-Cre/+; Rbm24 flox/+ control mice (n=3) at P0 (Fig. S5A-B′). By contrast, Rbm24 was absent in almost all HCs in the experimental-group mice, Atoh1-P2A-Cre/+; Rbm24 flox/− (n=3), at P0 (Fig. S5C-D′). However, IHCs and OHCs were present in both control and experimental mice. These results suggested that Rbm24 is not necessary for early HC development (at least until P0).

Next, cochlear samples from both groups of mice were analyzed at P19. Triple staining for Rbm24, prestin, and otoferlin showed that Rbm24 expression was maintained in HCs in the control group (Fig. S5E,E′), but was almost completely ablated in all HCs in the experimental group (Fig. S5F,F′). More importantly, similar to the observations in the Rbm24 mosaic mutants (Fig. 6), ablation of Rbm24 caused cell death of OHCs, but not IHCs, in the experimental mice at P19 (Fig. S5F,F′). Again, the ratio of OHC to IHC numbers was notably decreased, from ∼3.17 (averaged number for the three turns) in control mice to ∼0.28 in experimental mice (Fig. S5G).

To uncover the detailed mechanisms underlying OHC loss in the absence of Rbm24, further investigation is necessary, and Rbm24 might also perform functions other than the maintenance of OHC survival. Nevertheless, despite potential phenotypic variations observed for lethal genes among different F0 mice at postnatal ages following the use of mosaic CRISPR-stop, this method was found to be rapid and reliable, as demonstrated by the results obtained for three lethal genes in this study: Atoh1, Sox10 and Rbm24.

In this study, we successfully established a rapid genetic approach for studying lethal genes; we named this method mosaic CRISPR-stop. To the best of our knowledge, this is the first report of the in vivo use of base editing in one of the two blastomeres of mouse zygotes for lethal genes. We first used our new method to rapidly reproduce the known phenotypes of two lethal genes, Atoh1 and Sox10, in cochlear development. Most importantly, we identified a previously unknown role of the lethal gene Rbm24 in OHC survival, and we further validated our finding by using a conditional Rbm24 knockout model.

During early embryogenesis, the first cell-fate divergence that occurs in the zygote is that of the TB versus the ICM (Wang et al., 2018a): whereas the TB becomes the placenta, the ICM develops into the entire embryo. Here, by exploiting the dual but incompatible colors of the Cre-dependent reporter mouse mTmG/+, we manipulated one of the two blastomeres by injecting Cre mRNA (Fig. 1); thus, the injected blastomere and all its offspring cells were EGFP⁺, whereas the un-injected blastomere and its progeny cells were Tdtomato⁺. The presence of both EGFP⁺ and Tdtomato⁺ cells in the ICM suggested that offspring cells derived from either of the two blastomeres can contribute to the development of various organs, including the cochlea.

Focusing on HCs within the cochlear sensory epithelium, we found three distribution patterns of EGFP⁺ and Tdtomato⁺ HCs (Fig. 2). Approximately half the mice harbored equal numbers of EGFP⁺ and Tdtomato⁺ HCs, and the distribution patterns of the EGFP⁺ or Tdtomato⁺ cells in each organ examined, or in different regions of a given organ, appeared independent. This suggested that the integration of EGFP⁺ and Tdtomato⁺ cells during organogenesis was stochastic. Notably, mosaic CRISPR-stop can be broadly used for studying organogenesis: other organs/tissues besides the inner ear were also targeted, because both blastomeres develop into all somatic cells and germ cells in the body. We confirmed that Atoh1 mosaic mutant F0 mice produced germ-line-stable offspring when bred with wild-type mice.

When using mosaic CRISPR-stop for screening lethal genes, several F0 mice are born concurrently; this raises the key question of which F0 mice should be focused on for the initial round of analysis? Sanger sequencing patterns of tail DNA clearly revealed the degree to which C-to-T (or G-to-A) conversion was achieved (Figs 4–6). The appearance of a higher T peak than C peak suggested that cells carrying the C-to-T conversion were dominant in the tail, and pattern was defined as ‘high' (Figs 4–6). Conversely, a higher C peak than T peak was defined as the ‘low' pattern, and in this case, tail cells lacking the C-to-T conversion were dominant (Figs 4–6). The ‘intermediate' sequencing pattern implied that the numbers of the two cell populations were similar (Figs 4–6).

We carefully compared the Sanger sequencing pattern of tail DNA and the phenotype of cochlear HCs in Atoh1 mosaic mice (Fig. 4). Severe HC phenotypes were detected only in mice in which the ‘high' sequencing pattern was found, and, accordingly, no aberrant phenotype was observed (i.e. wild-type like phenotype was detected) in the majority of the mice in which the ‘low' sequencing pattern was found. This supported the conclusion that the ‘high' tail-DNA sequencing pattern (higher proportion of cells harboring a C-to-T conversion) correlated positively with the severe phenotype (decreased HC numbers). Therefore, in initial phenotypic analyses, the focus should primarily be on mice featuring the ‘high' sequencing pattern.

However, mice featuring the ‘intermediate' sequencing pattern might also be suitable because some of these mice show mild phenotypes; we observed mild HC phenotypes in half the Atoh1 mosaic mice in which the ‘intermediate' sequencing pattern was found. Furthermore, 7 Atoh1 and 3 Sox10 mosaic mice featuring the ‘high', but not ‘intermediate' or ‘low', sequencing pattern died before the analysis, which suggests that mice displaying mild phenotypes could survive longer than mice displaying severe phenotypes; consequently, specific experimental purposes should form the basis for the balance sought between the degree of phenotype and the lifespan of mosaic mutants of lethal genes generated using mosaic CRISPR-stop. For example, the onset of hearing in the mouse is at P14 (Sonntag et al., 2009), and, therefore, generating mosaic mutants of lethal genes that can survive until P14 or longer would be favorable and sufficient for several types of studies on hearing. Last, mice featuring the ‘high' sequencing pattern occasionally displayed no aberrant phenotype (Fig. 4O); this could be due to the stochastic distribution of the progenies of the two blastomeres, and, in this case, the consequence was considerable integration of mutant cells into the tail but little or no integration in the cochlea.

The key advantage of mosaic CRISPR-stop over the traditional conditional knockout approach is that the F0 mosaic mice were homozygous and could immediately be used for phenotype analysis. Thus, in the inner-ear research field, for example, this method would enable rapid determination, within 6 weeks (3 weeks of pregnancy and 3 weeks after birth), of whether a lethal gene is crucial for cochlear development and sound detection, because the onset of hearing is at P14 and cochlear development is completed by P21 (Sonntag et al., 2009). Conversely, ∼6 months would be necessary to obtain a germline-stable conditional knockout strain and another ∼4 months for two rounds of breeding of this conditional strain with a tissue-specific Cre or CreER mouse strain (Cox et al., 2012). Accordingly, with the use of mosaic CRISPR-stop, only 2 months were required for us to ascertain that Rbm24 was crucial for OHC survival. However, we spent a considerably longer time (∼10 months) confirming the same phenotype by using the Rbm24 conditional knockout approach.

When mosaic CRISPR-stop was used here, the phenotypes among different F0 mosaic mice showed large variations, and we sorted these into three categories: wild-type like, mild and severe. In this regard, the conditional model offers advantages, particularly when an efficient Cre is available: because the floxed gene would be inactivated in the majority of cells, minimal phenotypic variation should be observed among different replicates. Therefore, when lethal genes of interest are identified using transcriptomic analysis, the most efficient approach would be to rapidly screen the genes by using mosaic CRISPR-stop, and upon obtaining positive feedback in terms of the results, conditional knockout models can be generated for relatively more comprehensive and mechanistic studies. Last, in terms of the BEs that should be used for mosaic CRISPR-stop, we would recommend hA3A-BE3 paired with one efficient sgRNA, owing to its high efficiency (particularly in the case of GC-rich genes). In this study, we obtained successful base editing of Rbm24 with hA3A-BE3 but not BE3.

The first mosaic analysis of lethal genes was performed using a two-step injection protocol: Cas9 mRNA was first injected into one-cell stage zygotes, and this was followed by a second injection of sgRNA into one of the two blastomeres on the next day (Wang et al., 2017). This method was further successfully simplified into a one-step injection in another study (Wu et al., 2019), with the sgRNAs and Cas9 being concurrently injected into one of the two blastomeres. As in the case of our mosaic CRISPR-stop analysis, the mosaic mutants of lethal genes that were obtained in these two studies survived longer than regular knockout mice and displayed previously recognized phenotypes (Wang et al., 2017; Wu et al., 2019).

The main difference between this study and the aforementioned two studies is in how the gene mutation is triggered. The BEs we used here induce premature emergence of a stop codon by converting C to G (or G to A) without causing a DNA double-strand break (DSB). By contrast, traditional Cas9 was used in the two previous studies, and this causes DSBs and produces gene mutations through a frame-shift or large deletion/insertion. In this regard, the advantage of mosaic CRISPR-stop is that the gene mutation is relatively cleaner and less DNA damage occurs in the cells. However, the disadvantage is that fewer appropriate sgRNAs are available for BEs than for Cas9, which is because BEs use a preferred editing window (4th-8th bases) and the C-to-G (or G-to-A) conversion must occur in-frame. Thus, our recommendation is that if appropriate and efficient sgRNAs are available for a BE, mosaic CRISPR-stop is the preferred method for mutating a lethal gene of interest; otherwise, one should consider using the method reported in the two aforementioned studies. This is also the method we currently follow for screening the functions of lethal genes in mouse cochlear development.

In summary, we have presented here a new method for loss-of-function analysis – mosaic CRISPR-stop. This method was designed to rapidly determine the functions of lethal genes without mouse breeding, in a mosaic genetic background in which the mice not only survive longer than germline homozygous mutants of the same gene, but also display the corresponding phenotypes. We believe that mosaic CRISPR-stop represents a powerful genetic tool that can markedly accelerate the discovery of the functions of lethal genes.

Two BEs were used in this study: BE3 (Komor et al., 2016) and hA3A-BE3 (Wang et al., 2018b; Zhou et al., 2019). BE3 is composed of rat APOBEC1, nCas9 and UGI, and the vector pCMV-BE3 (Addgene 73021) was used to amplify BE3 cDNA. The detailed protocol of BE3 in vitro transcription was described in our previous report (Zhang et al., 2018). The hA3A-BE3, which is generated from human APOBEC3, nCas9 and UGI, was kindly provided by the laboratory of Dr Hui Yang (Institute of Neuroscience, Chinese Academy of Sciences, Shanghai, China). The cDNA sequence of hA3A-BE3 is listed in Table S1. Two primers, hA3A-BE3-F and hA3A-BE3-R (Table S1), were used to amplify hA3A-BE3 containing the T7 promoter for subsequent in vitro transcription by using a KOD-plus-neo kit (TOYOBO, KOD-401). The PCR amplicon (5422 bp) was purified using a MinElute PCR purification kit (Qiagen, 28004), after which in vitro transcription was performed using the mMACHINE T7 Ultra kit (ThermoFisher Scientific, AM1345) and then further purification was performed with the MEGAclear Transcription Clean-Up Kit (Ambion, AM1908).

The sgRNA synthesis approach used for all genes was identical. The same vector pX330 (Addgene 42230) was used as the shared template to obtain the PCR amplicon (120 bp) by using the KOD-plus-neo kit. The forward primer (5′–3′) for each gene was distinct and featured the sequence TAATACGACTCACTATAGG+sgRNA (20 bp)+GTTTTAGAGCTAGAAATAG, where only the middle 20 bp region differed depending on the specific gene being targeted; conversely, the same reverse primer was used for all genes: AAAAGCACCGACTCGGTGCC. Detailed protocols for sgRNA purification are described in our previous report (Zhang et al., 2018).

Cre mRNA was only used when mTmG/+ embryos were manipulated; the Cre cDNA sequence is listed in Table S1. Cre-F and Cre-R primers (Table S1) were used to obtain the Cre amplicon (2046 bp), including the T7 promoter. All other Cre mRNA transcription and purification steps were identical to those used for BE3 and hA3A-BE3.

Wild-type spCas9 was used to generate germline Rbm24^+/− knockout mice and conditional Rbm24 flox/+ mice. The vector px260 (Addgene 42229) was used as the PCR template for obtaining the 4582-bp amplicon by using the primers spCas9-F and spCas9-R (Table S1), after which in vitro transcription and purification were performed as described above for BE3 and hA3A-BE3. All in vitro transcribed sgRNAs and the mRNAs encoding BE3, hA3A-BE3, Cre and spCas9 were saved as aliquots (1 μl/tube; 500 ng/μl) and stored at −80°C.

For each gene, four or five sgRNAs were designed and transcribed to screen for the most efficient sgRNAs. The two efficient Atoh1 sgRNAs were confirmed in our previous study (Zhang et al., 2018). The sgRNAs for Sox10 and Rbm24 were screened using hA3A-BE3. Each sgRNA and hA3A-BE3 mRNA were co-injected into one-cell-stage zygotes (six zygotes per gene) derived from superovulated B6D2F1 female mice (mixed background of C57BL/6 and DBA2J; 8 weeks old) that were pretreated with pregnant mare serum gonadotrophin (PMSG) and hCG, and crossed with B6D2F1 male mice. The injected zygotes were cultured until the blastocyte stage and then DNA was extracted for nested PCR because the amount of blastocyte DNA was limited. The occurrence of base editing (C-to-T conversion) confirmed through Sanger sequencing indicated the success of sgRNA guidance. The sgRNAs were defined as being efficient when base editing occurred in at least half of the injected zygotes.

For the manipulation of two-cell stage embryos, 8-week-old female mice were superovulated by first injecting 7.5 IU CARD HyperOva (Cosmo Bio, KYD-010-EX-X5) and then, 48 h later, injecting 7.5 IU hCG (San Sheng Biological Technology, 110041282). CARD HyperOva is similar to PMSG but is more efficient at inducing superovulation. Next, the superovulated females were crossed with male mice, and then one-cell stage zygotes were collected from oviducts at 24 h after hCG injection by using hyaluronidase (Sigma, H3884). The one-cell stage zygotes were cultured for an additional 23.5-25.5 h to allow them to develop into two-cell stage zygotes and for the two blastomeres to be completely separated.

For experiments described in Fig. 1B-E″, Cre mRNA (5 ng/μl; final solution for all following description) was injected into neither blastomere or into one or both of the two blastomeres obtained from breeding mTmG/mTmG homozygous male mice (The Jackson Laboratory, stock number 007676) with B6D2F1 wild-type female mice. Following injection, the embryos were cultured in KSOM+AA medium containing d-glucose until the blastocyst stage for confocal imaging.

For mosaic targeting of Atoh1 with BE3, Cre mRNA (5 ng/μl), BE3 mRNA (100 ng/μl), Atoh1-sgRNA-3 (50 ng/μl) and Atoh1-sgRNA-1 (50 ng/μl) were co-injected into one of the two blastomeres derived from breeding mTmG/mTmG homozygous male mice with B6D2F1 wild-type female mice. For hA3A-BE3-mediated mosaic targeting of Atoh1, Sox10 and Rbm24, hA3A-BE3 mRNA (100 ng/μl) was injected into one of the two blastomeres derived from breeding B6D2F1 wild-type male and female mice, together with the respective sgRNA (50 ng/μl) for each gene. The injected embryos were cultured in KSOM+AA medium containing d-glucose for 1 h before implantation into oviducts of pseudopregnant ICR females, which, 19 or 20 days later, delivered the F0 mice.

All mice were bred and raised in SPF-level animal rooms, and animal procedures were performed according to the guidelines (NA-032-2019) of the IACUC of Institute of Neuroscience (ION), Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences.

Rbm24^+/− germline knockout mice were generated using the CRISPR/Cas9 approach. Rbm24-sgRNA-2 and -3 (50 ng/μl each) were co-injected with Cas9 mRNA (100 ng/μl) into one-cell stage zygotes, which were then implanted into pseudopregnant female mice. F0 mice were screened using primers Rbm24-F1 and Rbm24-R1 (Fig. S4A), and the generation of a 404 bp amplicon indicated a large fragment deletion. F0 mice harboring large fragment deletions were further crossed with wild-type mice for germline transmission, and this was followed by confirmation through tail-DNA Sanger sequencing (Fig. S4B). Regular tail-DNA PCR genotyping was performed using a mixture of three primers, Rbm24-F1, Rbm24-F2 and Rbm24-R1, the sequences of which are listed in Table S1, and the PCR amplicon sizes (wild type, 520 bp; null allele, 404 bp) are indicated in Fig. S4C.

Rbm24 flox/+ conditional knockout mice were also generated using the CRISPR/Cas9 approach (Fig. S4D-F). Gene-targeting donor DNA (100 ng/μl) and Rbm24-sgRNA-4 and -5 (50 ng/μl each) were co-injected with Cas9 mRNA (100 ng/μl) into one-cell-stage zygotes, and the F0 mice with potentially accurate gene targeting were crossed with wild-type mice for germline transmission. Southern blotting was performed to confirm the lack of random insertion of donor DNA in the mouse genome (Fig. S4G). Regular tail-DNA PCR genotyping was performed using a mixture of two primers, Rbm24-F3 and Rbm24-R2, the sequences of which are listed in Table S1, and the PCR amplicon sizes (wild type, 502 bp; flox, 587 bp) are indicated in Fig. S4H. Detailed sequences of the donor DNA construct are available upon request.

Last, we also generated an Atoh1-P2A-Cre/+ knock-in mouse strain by using the CRISPR/Cas9 method. Homozygous Atoh1-P2A-Cre/Atoh1-P2A-Cre survived until adulthood and did not display any apparent phenotypes. The details of the construction and characterization of this strain will be reported elsewhere, but the Atoh1-P2A-Cre/+strain will be made available upon request. All mice generated in this study will be deposited into The Jackson Laboratory.

Tail genomic DNA was isolated through extraction with phenol:chloroform:isoamyl alcohol (24:24:1), and then 10 μg of the genomic-DNA samples from Rbm24 flox/+ and Rbm24+/+ mice were digested with AflII. A single 7.2 kb probe band was detected in Rbm24^+/+ mice, whereas 4.8 and 7.2 kb bands were detected in Rbm24 flox/+ mice (Fig. S4G). The probe sequences are listed in Table S1. Detailed Southern blotting procedures are described in our previous studies (Li et al., 2020, 2018).

Inner-ear tissues from P1 mice were dissected out and fixed in fresh 4% paraformaldehyde (PFA) overnight at 4°C. For mice older than P10, cardiac perfusion was performed with 1× PBS, which was followed by perfusion with fresh 4% PFA and by a second fixation in fresh 4% PFA overnight at 4°C. Inner-ear tissues were washed thrice with 1× PBS and treated with EDTA for decalcification before whole-mount analysis. The following primary antibodies were used: anti-myosin VI (rabbit, 1:200, 25-6791, Proteus Bioscience), anti-Sox10 (goat, 1:200, sc-17342, Santa Cruz Biotechnology), anti-Rbm24 (rabbit, 1:200, 18178-1-AP, Proteintech), anti-prestin (goat, 1:200, sc-22692, Santa Cruz Biotechnology), anti-Sox2 (goat, 1:1000, sc-17320, Santa Cruz Biotechnology), anti-Prox1 (rabbit, 1:500, AB5475, Millipore), anti-parvalbumin (mouse, 1:1000, P3088, Sigma), anti-cleaved-caspase 3 (rabbit, 1:200, 9664S, Cell Signaling Technology) and anti-otoferlin (mouse, 1:200, ab53233, Abcam). All secondary antibodies, which were compatible with different combinations of the primary antibodies, were purchased from Thermo Scientific. After immunostaining, samples were counterstained with Hoechst 33342 solution in PBST (1:1000; 62249, Thermo Scientific) to visualize cellular nuclei. Samples were mounted with Prolong Gold antifade mounting medium (P36930, Thermo Scientific). All immunofluorescence images were captured using a Nikon C2 confocal microscope.

For measuring cochlear length, whole-mount samples were prepared and divided into three turns (basal, middle and apical) and scanned using a confocal microscope (at 10×). ImageJ software was used to draw a line in the middle of IHCs and OHCs. For calculating the ratio of OHCs to IHCs, samples were scanned using a confocal microscope (at 60×). In the same scanned areas, the numbers of OHCs and IHCs were determined and normalized against each other. In control mice, the ratio of OHC to IHCs was ∼3.1-3.4. Statistical analyses were performed using one-way ANOVA, followed by Student's t-test with Bonferroni correction. GraphPad Prism 6.0 was used for all statistical analyses.

We thank Dr Qian Hu and the Optical Imaging Facility at the Institute of Neuroscience-Chinese Academy of Sciences for support with the image analysis, Dr Hui Yang (Principal Investigator at the Institute of Neuroscience) for sharing the zygote microinjection system used to generate the knock-in mice and for providing the hA3A-BE3 vector, and Ms Qian Liu (from the Embryology Department of the Animal Center at the Institute of Neuroscience) for helping us in transplanting zygotes into pseudopregnant female mice.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.196899.reviewer-comments.pdf