Translating human genetics into mouse: The impact of ultra-rapid in vivo genome editing


  • Tomomi Aida,

    1. Laboratory of Molecular Neuroscience, Medical Research Institute, Tokyo Medical and Dental University, Yushima, Bunkyo-Ku, Tokyo, 113-8510, Japan
    Search for more papers by this author
  • Risa Imahashi,

    1. Laboratory of Molecular Neuroscience, Medical Research Institute, Tokyo Medical and Dental University, Yushima, Bunkyo-Ku, Tokyo, 113-8510, Japan
    Search for more papers by this author
  • Kohichi Tanaka

    Corresponding author
    1. Laboratory of Molecular Neuroscience, Medical Research Institute, Tokyo Medical and Dental University, Yushima, Bunkyo-Ku, Tokyo, 113-8510, Japan
    2. The Center for Brain Integration Research, Tokyo Medical and Dental University, Yushima, Bunkyo-Ku, Tokyo, 113-8510, Japan
    3. JST, CREST, Kawaguchi-shi, Saitama, Japan
    Search for more papers by this author


Gene-targeted mutant animals, such as knockout or knockin mice, have dramatically improved our understanding of the functions of genes in vivo and the genetic diversity that characterizes health and disease. However, the generation of targeted mice relies on gene targeting in embryonic stem (ES) cells, which is a time-consuming, laborious, and expensive process. The recent groundbreaking development of several genome editing technologies has enabled the targeted alteration of almost any sequence in any cell or organism. These technologies have now been applied to mouse zygotes (in vivo genome editing), thereby providing new avenues for simple, convenient, and ultra-rapid production of knockout or knockin mice without the need for ES cells. Here, we review recent achievements in the production of gene-targeted mice by in vivo genome editing.


The mouse has become the most commonly used animal model system in the biological and medical sciences because its genome can be specifically and precisely modified as desired (Capecchi 2005). The invaluable advantage of the mouse is the ability to perform homologous recombination in embryonic stem (ES) cells, an essential step in gene targeting and a technology that was unavailable in the majority of other mammalian species. Since the first success of gene targeting in mouse, thousands of mice, mainly knockout mice created by insertion of a selection marker or reporter into a target gene locus, have been created, unveiling the in vivo functions of the genes. As an extension of these efforts, large-scale international consortia were organized to provide knockout mice for all protein-coding genes and systematically analyze the resulting phenotypes (Sung et al. 2012; Menke 2013). The International Knockout Mouse Consortium (IKMC) released targeted ES cell lines, including knockout, conditional, and gene-trapped alleles, for more than 18 000 genes, in addition to mice targeted for over 2600 loci (Skarnes et al. 2011). Further, the Sanger Institute Mouse Genetics Project (MGP), a founding member of the International Mouse Phenotyping Consortium (IMPC), recently released pilot data from a large-scale systematic phenotype analysis of 489 knockout mouse strains, derived from more than 900 IKMC knockout ES cell lines (White et al. 2013). Surprisingly, unbiased screening by the MGP identified many previously unknown phenotypes in both new knockout mice and strains that were the subject of earlier reports. The IMPC will expand this phenotypic screening to cover 5000 knockout mouse lines over the next 4 years and to all 20 000 protein-coding genes in the future. These genome-wide and large-scale systematic knockout mouse resources are now publicly available; thus, researchers can focus their efforts on the detailed functional analysis of genes of interest, rather than on the construction of mouse lines.

Recent advances in genomic microarray and next generation sequencing technologies have revealed the landscape of human genetic diversity, which comprises tens of millions of common and rare variants associated with health and disease (Raychaudhuri 2011; 1000 Genomes Project Consortium et al. 2012). Although large-scale genome-wide association studies (GWAS), based on high-density genomic microarray technology, have identified hundreds of frequent variants associated with common and complex human diseases and traits, the majority of these are responsible for only a small amount of the disease risk (Manolio et al. 2009). In contrast, with recent advances in unbiased whole-genome and whole-exome sequencing approaches, inherited and rare de novo single nucleotide variants (SNVs) are now also believed to be important in common and complex diseases (Cirulli & Goldstein 2010; Veltman & Brunner 2012). The best examples are large-scale exome sequencing studies on hundreds of patient-parent trios or quartets for autism spectrum disorders (ASDs), which are neurodevelopmental conditions characterized by impairments in social interaction and stereotyped behaviors with a strong genetic component (Iossifov et al. 2012; Neale et al. 2012; O'Roak et al. 2012; Sanders et al. 2012). These studies consistently report higher rates of de novo, especially nonsense, splice site, or frameshift SNVs in patients with ASD than in their unaffected siblings (Veltman & Brunner 2012). Further, they also unveiled extreme genetic heterogeneity in ASD, as evidenced by the uncovering of de novo SNVs in hundreds of different genes in different individuals. This indicates the need for further efforts to investigate the biological consequences of these rare de novo SNVs (Veltman & Brunner 2012).

One possible approach to address the biological function of these SNVs, as well as determining whether they are causal for the human phenotype of interest, is the use of genetic mouse models incorporating the identified variants. Precisely modified knockin mouse models carrying such human SNVs provide a unique and direct approach for the investigation of the functional consequence of variants in vivo. Pioneer work in this field by Südhof and colleagues reported that a knockin mouse carrying a neuroligin-3 R451C SNV found in a subset of ASD patients exhibited abnormal behaviors that resembled those of human patients, in addition to abnormal synaptic transmission (Tabuchi et al. 2007). Importantly, in contrast to R451C knockin mouse, neuroligin-3 knockout mouse did not exhibit such abnormalities, suggesting that the R451C SNV represents a gain-of-function mutation (Tabuchi et al. 2007). Similarly, in Rett syndrome, an ASD caused by mutations in methyl-CpG-binding protein 2 (MeCP2), many SNVs throughout the MECP2 gene were identified and several knockin mouse lines carrying each SNV provided important insights into the biological and phenotypic significance of each variant, as well as identifying downstream targets of MeCP2 (Tao et al. 2009; Jentarra et al. 2010; Cohen et al. 2011; Goffin et al. 2011; Ebert et al. 2013; Lyst et al. 2013). Taken together these studies demonstrate that mouse models for human SNVs, rather than simple knockout mice, are essential and extremely valuable tools for the biological and phenotypic interpretation of human variants and the development of novel treatments.

Although the demand for precisely modified knockin mouse models carrying human SNVs is growing, a recent review by Menke reported that only 600 such mice could be found in the Mouse Genome Informatics database (Menke 2013). This is partially due to the difficulty of generating such mice by conventional gene targeting technology using homologous recombination in ES cells; a time-consuming, laborious, and expensive process (Capecchi 2005).

In vivo genome editing in mice

The recent emergence and drastic evolution of genome editing technologies is revolutionizing gene targeting in the mouse (Sung et al. 2012; Menke 2013). The methods are based on molecular tools, including zinc-finger nucleases (ZFNs), transcription activator-like effector (TALE) nucleases (TALENs), and clustered, regularly interspaced, short palindromic repeats (CRISPR) and the CRISPR-associated endonuclease (Cas), known as the CRISPR/Cas system. These methods provide exciting and groundbreaking opportunities, enabling direct and rapid gene targeting in fertilized mouse eggs, with no need for ES cells. The basic characteristics and various applications of these genome editing technologies will be discussed by others in this special issue of DGD. Here, we focus on gene targeting in the mouse by in vivo genome editing and review current achievements, issues to be solved, and future applications.

The principle of in vivo editing involves targeting of the genome by direct microinjection of plasmid DNA or mRNA encoding editing tools (ZFNs, TALENs, or CRISPR/Cas) into the cytoplasm or pronuclei of one-cell embryos to generate a DNA double-strand break (DSB) at a specific target locus (Fig. 1, Urnov et al. 2010; Joung & Sander 2013). The DSB is subsequently repaired by two major cellular endogenous DNA damage repair pathways; the error-prone, non-homologous end-joining (NHEJ) route, which results in small deletions or sequence insertions into the DSB site, and the homology-directed repair (HDR) pathway, which relies on a donor DNA template with homology to the DSB site to achieve precise homologous recombination. NHEJ occurs rapidly and preferentially, often leading to frameshift mutations and loss-of-function of the targeted genes, resulting in a knockout mouse when the protein-coding sequence is targeted (Fig. 1, Carbery et al. 2010; Sung et al. 2013; Shen et al. 2013; Wang et al. 2013a). HDR infrequently occurs and leads to precise and specific genome modifications, such as SNV substitutions, insertions, deletions, or gene replacement, when a targeting vector or synthetic single-strand oligonucleotide (ssOligo) is co-microinjected into mouse embryos, resulting in a knockin mouse (Fig. 1, Meyer et al. 2010; Cui et al. 2011; Meyer et al. 2012; Wefers et al. 2013; Wang et al. 2013a; Yang et al. 2013). A series of groundbreaking successes indicate that in vivo genome editing in the mouse is robust and has great potential as an alternative to the conventional gene targeting approach.

Figure 1.

Schematic diagram of in vivo genome editing in mouse. Transcription activator-like effector nucleases (TALENs) (blue), zinc-finger nucleases (ZFNs) (pink), or clustered, regularly interspaced, short palindromic repeats-associated endonuclease (CRISPR/Cas) (green) are microinjected into one-cell fertilized eggs (circle in the middle) derived from wildtype mice with or without ssOligo or targeting vector for knockin mouse production. Then, double-strand break (DSB) and subsequent homology-directed repair (HDR) (red) and/or non-homologous end-joining (NHEJ) (yellow) are induced within one-cell fertilized eggs, resulting in genetically-mosaic (striped), monoallelically-targeted heterozygous (gray-colored), or biallelically-targeted homozygous (black-colored) mutant mice at F0.

Knockout mice

After the first success of NHEJ-mediated gene knockout by in vivo genome editing in mammals was achieved in rat with ZFNs targeting three different genes (Geurts et al. 2009), it was rapidly applied to mouse (Table 1). Cui and colleagues from Sigma-Aldrich, the exclusive supplier of ZFNs, reported the first knockout mice by in vivo genome editing with ZFNs (Carbery et al. 2010). They targeted three endogenous genes in both FVB/N and C57BL/6J strains with targeting efficiencies from 20 to 75% of live newborns and no off-target effects at 20 potential sites. The founder mice were heterozygous or genetically mosaic, carried more than one mutant allele, and successfully produced F1 mutant progeny. Importantly, homozygous mice were generated from F1 mutants within 4 months (Fig. 2).

Table 1. Summary of non-homologous end-joining (NHEJ) in mice by in vivo genome editing
NucleasesTargetsGenesNHEJ (%)Off-targetF1Refs
  1. Including biallelic modification, mouse data only, §for Smurf1, 57.8–78.5% for biallelic modification for Tet1 and Tet2, ††RNA and mouse data only. n.d., not determined. F1 represents germline transmission. Percentages of NHEJ were calculated using the number of NHEJ positive pups as the numerator and the number of the total pups as the denominator. Fetus data were included in some studies.

ZFNs3 Mdr1a, Jag1, Notch3 19.5–77.3NoneYesCarbery et al. (2010)
ZFNs1 ROSA26 22.4n.d.n.d.Meyer et al. (2010)
ZFNs1 Mdr1a 7.5–75.0n.d.n.d.Cui et al. (2011)
ZFNs1 ROSA26 4.6–8.7n.d.n.d.Hermann et al. (2012)
TALENs2 Pibf1, Sepw1 48.7–76.9NoneYesSung et al. (2013)
TALENs1 Rab38 2.0–6.0n.d.YesWefers et al. (2013)
TALENs1 Zic2 10.0–46.7n.d.YesDavies et al. (2013)
TALENs10 Lepr, Pak1ip1, Gpr55, Rprm, Fbxo6, Smurf1, Tmem74, Wdr20a, Dcaf13, Fam73a 12.5–66.7None§YesQiu et al. (2013)
TALENs1 Mlkl 18.2n.d.YesWu et al. (2012)
TALENs2 C9orf72, Fus 7.5–41.2NoneYesPanda et al. (2013)
CRISPR/Cas2 Pouf5-EGFP, CAG-EGFP 14.3–20.0n.d.n.d.Shen et al. (2013)
CRISPR/Cas3 Tet1, Tet2, Tet3 66.7–100.0,Nonen.d.Wang et al. (2013a)
CRISPR/Cas3 Th, Rheb, Uhrf2 75.0–91.7††NoneYesLi et al. (2013)
Figure 2.

Time course for homozygous mutant mouse production. ES: embryonic stem (ES)-cell based traditional gene targeting method. Mosaic: genetically-mosaic founders are obtained at F0 by in vivo genome editing. Monoallele: monoallelically-targeted heterozygous founders are obtained at F0 by in vivo genome editing. Bialleles: biallelically-targeted homozygous founders are obtained at F0 by in vivo genome editing. Green: the term to obtain F0 adult mosaic founders from microinjection of genome editing tools, or F0 adult chimeric founders from ES cell microinjection. Orange: the term to obtain adult heterozygous mice from mating of F0 founders with wildtype mice or from microinjection of genome editing tools. Magenta: the term to obtain homozygous newborns from mating of heterozygous mice or from microinjection of genome editing tools. Best cases are shown.

Similar work was performed for the targeting of the ROSA26 locus, a safe harbor often used for gene targeting (Hermann et al. 2012). Although the efficiency in this study was <10%, one of the founders had a biallelic modification, resulting in an F0 biallelic knockout mouse that drastically reduced the time taken to produce homozygous knockouts (Fig. 2).

Although the simple modular DNA recognition code of TALENs and the existence of publicly available resources has resulted in the rapid expansion of TALENs as versatile genome editing tools, the first knockout mice by TALEN-mediated in vivo genome editing were reported in 2013 (Sung et al. 2013). Two endogenous genes were targeted with efficiencies from 49 to 77% in live newborns with no off-target effects. All alleles present in F0 founders were successfully transmitted to F1 progeny. Importantly, targeting and biallelic modification efficiencies were increased when microinjection was performed with a high dose of TALEN mRNA; when 50 ng/μL of TALEN mRNA targeting the Pibf1 gene was injected, six of eight F0 founders had biallelic modifications. Similar results were obtained for the Sepw1 gene, and the absence of Sepw1 protein was confirmed in the biallelically targeted F0 mutants. This work suggests that highly active TALENs are critical for efficient targeting and biallelic modification. Homozygous knockout mice can be generated within 1 month by TALEN-mediated in vivo genome editing (Fig. 2).

Since the first report, a flood of knockout mice generated by TALEN-mediated in vivo genome editing has been reported. Davies et al. (2013) targeted the Zic2 gene by TALENs in three different mouse strains, including CD1, C3H, and C57BL/6J. Targeting efficiencies producing live newborns or blastocysts varied, with 10%, 23%, and 46% for C57BL/6J, C3H, and CD1, respectively.

Li and colleagues generated a series of knockout mice for 10 genes, revealing the utility, convenience, and robustness of TALEN-mediated in vivo genome editing (Qiu et al. 2013). Targeting efficiencies varied from 13 to 67%, with an average of 40%, of live newborns. By using one TALEN for the Lepr gene, which encodes the Leptin receptor, they showed that there was no difference in the targeting efficiency between two different mouse strains (C57BL/6N and FVB/N). Among the F0 founders, one had biallelic modifications with different frame-shift deletions and exhibited an obese phenotype resembling that of Lepr mutant db/db mice. Further, there were no off-target effects, even at sites with only one mismatch to each TALEN. All the F0 founders tested transmitted the mutant alleles to F1 mice with high efficiency. These results suggest that this method of genome editing is highly accurate and efficient.

Further, Han and colleagues generated knockout mice for Mlkl gene (which encodes the mixed lineage kinase domain-like protein, essential for tumor necrosis factor induced necrosis) by TALEN-mediated in vivo genome editing (Wu et al. 2012). They injected TALEN mRNA into nearly 3000 embryos and obtained 71 mutants from 390 newborns (18% efficiency); of these, four were homozygous mutants.

The latest player, CRISPR/Cas system, is having a drastic impact on the field, due to its simplicity, incredibly high efficiency, and multiplexing capability. Huang and colleagues first generated a knockout mouse in which a green fluorescent protein (GFP) transgene was disrupted by CRISPR/Cas-mediated in vivo genome editing (Shen et al. 2013). They targeted GFP in two different mouse strains carrying transgenes that encoded GFP, and obtained GFP-deficient mice with targeting efficiencies from 14 to 20% of live newborns.

Next, Jaenisch and colleagues published a revolutionary paper describing the production of mice knocked out for multiple genes with extremely high efficiency (Wang et al. 2013a). In this study, they targeted three functionally redundant genes (Tet1, Tet2, and Tet3) encoding Ten-eleven translocation (Tet) enzymes that convert 5-methylcytosine to 5-hydroxymethylcytosine. They first investigated the optimal conditions for CRISPR/Cas-mediated in vivo genome editing by injecting various amounts (20–200 ng/μL) of Cas9 mRNA, with Tet1, Tet2 or Tet3 guide RNAs, into fertilized eggs, and found Cas9 dose-dependent increments in targeting efficiency, with low toxicity in newborn mice. Surprisingly, the vast majority (60–100%) of newborns carried biallelic modifications in each target gene. Further, they simultaneously targeted both Tet1 and Tet2, and obtained double mutants in which all four alleles of these genes were targeted in 80% of newborns without off-target effects. CRISPR/Cas-mediated in vivo genome editing can be completed within a month, since construction of CRISPR/Cas vector only takes a few days. This represents an incredible shortcut for the generation of single or double knockout mice, which often takes several years using conventional gene targeting methods (Fig. 2).

Liu and colleagues confirmed the high efficiency of CRISPR/Cas-mediated in vivo genome editing (Li et al. 2013). They generated three knockout mice targeting Th, Rheb, and Uhrf2, with targeting efficiencies from 75 to 92% and no off-target effects. They also targeted two adjacent sites, spanning 86 bp, in the Uhrf2 locus. Importantly, the mutations were successfully transmitted to the next generation, suggesting that the CRISPR/Cas system is the third tool after ZFNs and TALENs to allow heritable in vivo genome editing in mice.

Knockin mice

Although the success of in vivo genome editing has enabled the rapid generation of knockout mice, developing this technique for the production of knockin mouse models would fully exploit its capabilities (Table 2). In 2010, Kühn and colleagues reported pioneering work in the production of the first knockin mice using ZFN-mediated in vivo genome editing and targeting vectors (Meyer et al. 2010). They co-injected mRNAs encoding ZFNs for the ROSA26 locus and a targeting vector, containing a 4.2 kb LacZ reporter cassette, with homology arms flanking the target site of the locus, into one-cell mouse embryos. Fifty-eight embryos were analyzed and one was found to be a precisely modified knockin mouse and was functionally confirmed by X-gal staining. Next, the same ZFN mRNAs were co-injected with a targeting vector containing a 1.1 kb Venus reporter cassette, instead of LacZ. Among the 22 embryos obtained, they identified one as a precisely modified knockin mouse. The targeting efficiencies of the two experiments were 1.7 and 4.5%, respectively.

Table 2. Summary of homology-directed repair (HDR) in mice by in vivo genome editing
NucleasesTargetsGenesInsertsDonorsHDR (%)Off-targetF1Refs
  1. Mouse data only, including biallelic modification, §60% for double HDR of both Tet1 and Tet2, 61.3% for loxP site integration, ††16% for two loxP sites in one allele. n.d., not determined. F1 represents germline transmission. Percentages of HDR were calculated using the number of HDR positive pups as the numerator and the number of the total pups as the denominator. Fetus data were included in some studies.

ZFNs1 ROSA26 LacZ, Venus Plasmid1.7–4.5n.d.n.d.Meyer et al. (2010)
ZFNs1 Mdr1a NotI, GFP Plasmid5.0–25.0n.d.n.d.Cui et al. (2011)
ZFNs1 ROSA26 GFP Plasmid2.0n.d.YesHermann et al. (2012)
ZFNs1 Rab38 SNV Plasmid, ssOligo1.7–3.5n.d.YesMeyer et al. (2012)
TALENs1 Rab38 SNV Plasmid, ssOligo0.9–2.0n.d.YesWefers et al. (2013)
TALENs1 Fus SNV ssOligo4.0–8.4NoneYesPanda et al. (2013)
CRISPR/Cas2 Tet1, Tet2 EcoRI ssOligo70.0–80.0,§Nonen.d.Wang et al. (2013a)
CRISPR/Cas4 Sox2, Nanog, Oct4, MeCP2 V5, mCherry, GFP-Neo, LoxP Plasmid, ssOligo8.1–61.3,††Presentn.d.Yang et al. (2013)

Soon afterwards, Cui et al. (2011) also reported production of knockin mice and rats with relatively high efficiency. First they tested the targeted integration of a small 8 bp NotI site fragment flanked on each side by 800 bp of homology arms. Co-injection of a targeting vector with ZFNs for two genomic loci (Mdr1a and PXR) in both rats and mice resulted in knockin mutants with efficiencies of 6.7–25% of embryos. Next, they tested targeted integration of a long DNA fragment using the same donor vector, in which the NotI site was replaced with a 1.5 kb GFP cassette. They injected the GFP vector in the same way as the NotI sequence, and obtained knockin mice and rats for the two genomic loci with efficiencies from 2.4 to 8.3% of embryos or newborns. Further, they confirmed efficient germline transmission in both Mdr1a- and PXR-GFP knockin rats, with 50% of the F1 progeny corresponding to heterozygous mutants. Similar work was performed to target the ROSA26 locus to integrate a GFP fragment, and the efficiency reported was 2% (Hermann et al. 2012).

Although the donor vectors used in in vivo genome editing contain relatively short homology arms for the targeted integration of SNVs that require only a few nucleotide substitutions, their construction is still a disproportionately laborious and time-consuming task. The use of synthetic single stranded DNA oligonucleotides (ssOligos) as donors for HDR can bypass this process. Davis and colleagues reported ZFN-mediated targeted integration of point mutations with ssOligos in several human cell lines with efficiencies that were up to twice those achieved using conventional targeting vectors (Chen et al. 2011). Kühn and colleagues applied ssOligo donor to produce knockin mice carrying SNVs (Meyer et al. 2012). They first generated a knockin mouse carrying a G19V missense and several silent SNVs in the Rab38 gene, which encodes a small GTPase whose mutation results in a brown coat color, by co-injecting ZFNs with a conventional targeting vector. The targeting efficiency was 3.5% (three of 87 newborns), which is comparable to the efficiency they reported in their pioneering work (Meyer et al. 2010). All three founders exhibited efficient germline transmission, resulting in a brown coat color in F2 homozygous mutants (Meyer et al. 2012). Next, they co-injected the same ZFNs with a 144 nucleotide (nt) ssOligo containing seven substitutions into one-cell mouse embryos. They obtained one partially targeted mutant from 60 newborns with an efficiency of 1.7%, and the mutation was successfully transmitted to the F1 progeny. This work clearly reveals the enormous potential that ssOligos have for the replacement of conventional gene-targeting vectors in in vivo genome editing, which should greatly facilitate the rapid production of knockin mice.

In early 2013, Kühn and colleagues also generated Rab38 G19V knockin mice by TALENs using an ssOligo (Wefers et al. 2013). They first constructed TALENs targeting the same region of the Rab38 gene previously targeted by ZFNs, and found that the activity of the TALEN system was approximately twice that of ZFNs. Next, they co-injected TALENs and a ssOligo into one-cell mouse embryos and obtained one founder mouse carrying a partially targeted G19V allele from 117 newborns (an efficiency of 0.9%). The G19V allele was successfully transmitted to the F1 progeny. They also co-injected TALENs with a conventional targeting vector, rather than the ssOligo, and obtained one knockin founder carrying the G19V allele from 50 newborns (efficiency 2%). The G19V allele was also successfully transmitted to F1 progeny. As the construction of TALENs is much simpler than that of ZFNs, rapid production of knockin mice can be achieved using a combination of TALENs and ssOligos. However, the relatively low knockin efficiency of TALENs is a bottleneck that limits the dissemination of the method. To expand the applicability of TALEN-mediated in vivo genome editing, we developed highly active TALENs in collaboration with Dr Yamamoto's group at Hiroshima University. We focused on glutamate transporters, which are essential molecules that keep extracellular glutamate concentrations below neurotoxic levels (Tanaka et al. 1997; Watase et al. 1998; Matsugami et al. 2006; Aida et al. 2012). We previously reported GLAST, a glial glutamate transporter, knockout mouse as the first model for normal tension glaucoma (Harada et al. 1998, 2007; Bai et al. 2013a,b; Namekata et al. 2013). We also recently discovered deleterious missense mutations in EAAT1, a human orthologue of GLAST, in patients with glaucoma (Yanagisawa et al. unpubl. data, 2013). To generate knockin mice carrying these SNVs in the GLAST gene, we co-injected highly active TALENs targeting GLAST into one-cell mouse embryos with ssOligos carrying each SNV. We obtained several germline-competent knockin founders with targeting efficiency of approximately 20% (Aida et al. unpubl. data, 2013). This study yielded the highest reported efficiency, which was almost 25-fold higher than the efficiency in a previous report by Wefers et al. As a single microinjection is sufficient to obtain several knockin founders, our TALEN technology provides a fast and efficient approach for the production of genetic mouse models that reproduce the disease-associated SNVs of complex diseases. Recently, Kühn and colleagues reported improved knockin efficiencies that were up to 8%, using TALEN mRNAs transcribed from plasmids containing a poly A tail (Panda et al. 2013).

Later, Jaenisch and colleagues reported the groundbreaking production of mice carrying multiple knockin alleles in different genes using CRISPR/Cas-mediated in vivo genome editing with extremely high efficiency (Wang et al. 2013a). They co-injected fertilized eggs with Cas9 mRNA, Tet1 and Tet2 guide RNAs, and 126 nt ssOligos to substitute a SacI site in Tet1 and an EcoRV site in Tet2 with EcoRI sites. Surprisingly, the vast majority (70–80%) of newborns carried EcoRI sites at Tet1 or Tet2 loci and some were homozygous for the EcoRI sites. Further, 60% of the newborns had EcoRI sites at both Tet1 and Tet2 loci.

Soon after this major accomplishment, Jaenisch and colleagues also produced knockin mice carrying longer DNA insertions by CRISPR/Cas-mediated in vivo genome editing (Yang et al. 2013). They first targeted the last codon of the Sox2 gene with an ssOligo containing 42 nt short V5 epitope tag, and obtained targeted embryos and newborns with 34% efficiency. Next, they targeted the last codon of the Nanog gene with larger plasmid vector containing p2A-mCherry reporter cassette, and obtained targeted embryos and newborns with 8% efficiency. Further, they targeted the 3′ end of the Oct4 gene with a plasmid vector containing 3 kb sequence of IRES-EGFP-loxP-Neo-loxP reporter cassette, and obtained targeted newborns with an efficiency of 30%. Finally, they also successfully generated knockin mice carrying a conditional allele of Mecp2, by simultaneously targeting with two loxP-containing ssOligos, and obtained targeted embryos and newborns carrying two loxP sites in one allele with an efficiency of 16%. Thus, knockin mice carrying, not only a SNV, but also longer DNA fragments, can now be created within a month using in vivo genome editing with high efficiency (Fig. 2). Taken together, almost everything achieved by ES cell-based gene targeting can now be performed by the in vivo genome editing technologies. Further, the new techniques allow previously impossible achievements, such as ultra-rapid production, biallelic targeting in F0 mice and multiplexing, leading genome editing to be the method of first choice for gene targeting.

Off-target effects

The off-target effect, which involves non-specific recognition and digestion at non-targeted regions by ZFNs, TALENs, and the CRISPR/Cas system, has been extensively discussed in the field of genome editing. When compared to ZFNs, TALENs produce only minimal off-target effects (less than a tenth), even at highly similar non-specific target sites with only two mismatches in the TALEN recognition sequence in human cells (Mussolino et al. 2011). Consistent with in vitro data, three papers describing TALEN-mediated in vivo genome editing in mice reported no off-target effects at a total of 15 potential off-target sites, containing only one mismatch, for four TALEN pairs (Panda et al. 2013; Qiu et al. 2013; Sung et al. 2013). Thus, in addition to basic research, TALENs may be applicable to therapeutics, a field that demands high-specificity.

Although the CRISPR/Cas system is an easy, quick, and highly efficient genome editing tool, the small size of the sequence (20 nt) required for DNA-RNA hybridization may make off-target effects more frequent with the CRISPR/Cas system than with TALENs or ZFNs. Recent large-scale systematic reports revealed an unexpectedly high frequency off-target effects using the CRISPR/Cas system in several human cell lines (Fu et al. 2013; Hsu et al. 2013; Pattanayak et al. 2013). According to these reports, the CRISPR/Cas system can cleave off-target sites containing even up to five mismatches (Fu et al. 2013). Jaenisch and colleagues investigated potential off-target sites using their knockout and knockin mice, as well as newly established mouse ES cells, to examine the specificity of the CRISPR/Cas system in vivo (Wang et al. 2013a; Yang et al. 2013). Through analyses of 54 potential off-target sites for seven guide RNAs, they found several non-specific digestions at three sites containing one or two mismatches. These results indicate that off-target effects in the CRISPR/Cas system do exist in vivo, but may be lower than predicted from in vitro studies using human cell lines.

As the majority of recent studies have focused on selected candidates for potential off-target effects, unbiased and genome-wide characterization of off-target sites through whole-genome sequencing will be required to guide the more sophisticated and specific design of RNAs.

To reduce non-specific off-target effects in the CRISPR/Cas system, Cas9 nickase, a mutant form of Cas9 that cleaves single stranded DNA, may provide an alternative for the induction of HDR (Cong et al. 2013; Mali et al. 2013). Zhang and colleagues recently reported that off-target effects could be reduced by using nickase and a pair of guide RNAs, without affecting on-target cleavage activity (Ran et al. 2013). They also revealed that this double-nicking strategy could efficiently cleave on-target sites in mouse zygotes. In combination with future developments using mutant Cas9 variants or other more specific Cas9 orthologues, these methods could reduce off-target effects in the CRISPR/Cas system.

The impact of in vivo genome editing

In Figure 2, we summarize the time course for the production of gene-targeted mice by conventional ES cell-based methods and in vivo genome editing by ZFNs, TALENs, and the CRISPR/Cas system. At best, it takes approximately 1 year to obtain a homozygous mutant by the conventional ES cell-based method. Also, it is common to spend a year or more obtaining germline competent chimeric founders. However, in vivo genome editing is revolutionizing these complex processes and enables ultra-rapid production of gene-targeted mice. In many cases, a genetically mosaic F0 founder and F2 homozygous knockout or knockin mouse can be obtained within a month and approximately 7 months, respectively. Further, in the best cases, as reported by several groups (Meyer et al. 2010; Hermann et al. 2012; Qiu et al. 2013; Sung et al. 2013; Wang et al. 2013a; Wu et al. 2012), biallelically targeted homozygous knockout or knockin mice can be obtained within a month. Thus, the genome editing revolution provides practical and exciting opportunities for the research community to freely and rapidly generate gene-targeted mice.

We now have the means to functionally investigate the consequences of millions of rare SNVs in vivo using “humanized” mice carrying equivalent variants. The cutting-edge work of Gleeson and colleagues demonstrated an interdisciplinary, sequencing era approach that integrates human genetics and mouse models (Novarino et al. 2012). They performed exome sequencing in consanguineous families with ASD, epilepsy, and intellectual disability and identified homozygous gene-disrupting SNVs in the BCKDK gene, which inactivates an enzyme complex essential for the catabolism of branched-chain amino acids (BCAAs). Because the SNVs resulted in disruption of the BCKDK gene, instead of generating knockin mice carrying the SNVs, they were able to investigate BCKDK knockout mice that showed reduced BCAAs in various tissues and neurological abnormalities, similar to other mouse models for ASD. In addition to BCAAs, they discovered imbalanced amino acid levels in the mutant brain that may contribute to the defects in neurotransmitter synthesis and subsequent neurological abnormalities. Finally, they tried to treat the mutant mice and patients using dietary supplementation with BCAAs and successfully reversed neurological abnormalities in the mutant mice and normalized plasma BCAA levels in patients (Novarino et al. 2012). Because the vast majority of rare SNVs are missense or synonymous, instead of gene-disrupting nonsense, splice site, or frameshift variants (Veltman & Brunner 2012), the production of knockin mouse models carrying such variants will be essential. Thus, in vivo genome editing drastically accelerates functional investigation of rare SNVs.

In vivo genome editing also accelerates functional research of common SNVs in intronic or intergenic regions indicated by the ENCODE (Encyclopedia of DNA Elements) project or GWAS studies (ENCODE Project Consortium et al. 2012; Maurano et al. 2012), in addition to those in protein-coding sequences. Cutting-edge work by Taipale and colleagues demonstrated the utility of gene-targeted mouse models in investigating the function of a GWAS-identified SNV (Sur et al. 2012). They focused on a conserved 500 kb region upstream of the MYC oncogene, where multiple cancer-associated SNVs have been mapped. They generated mutant mice lacking the region containing the SNV strongly associated with cancer, and found that the mutant mice were resistant to tumorigenesis (Sur et al. 2012). Further, Sabeti and colleagues used precisely modified knockin mice carrying a V370A SNV in the ectodysplasin receptor, which resulted in the identification of GWAS as one of the strongest candidates for recent positive selection in human evolution (Kamberov et al. 2013). They found that the knockin mice not only recapitulated the human phenotype, but also had previously unknown traits which were, surprisingly, also confirmed in human. As most common SNVs identified by GWAS can only explain relatively small contributions to disease risk, and the functional interpretation of non-coding SNVs is difficult, the generation of knockin mice by the time-consuming, laborious, and expensive process of ES cell-based conventional gene targeting is now considered disproportionate. The advent of in vivo genome editing technology has transformed this situation, enabling the mouse as a useful animal model system for the functional analysis of common, non-coding SNVs.

Further, genome editing technologies allow previously impossible gene targeting in mice. First, the methods allow gene targeting at the locus where traditional homologous recombination cannot be applied, such as the Y chromosome. Because the Y chromosome has unique structure containing many palindromes, conventional gene targeting in ES cells has failed. Jaenisch and colleagues targeted Sry and Uty genes on Y chromosome in mouse ES cells by using TALENs, and successfully obtained knockout mice lacking Sry or Uty (Wang et al. 2013b). Thus, high sequence specificity of TALENs provides a novel approach for genetic manipulation of the Y chromosome. Second, the methods allow double gene targeting at the neighboring loci. When two genes are located next to each other on the same chromosome, it is almost impossible to obtain double knockout mice by crossing two single knockout mice. Thus, the researchers have generated double-targeted ES cells by sequential targeting, a process much more time-consuming, laborious, and expensive than single gene targeting (Kitajima et al. 2000). As Jaenisch and colleagues demonstrated (Wang et al. 2013a), now, multiple genes can be targeted simultaneously by in vivo genome editing, thus providing opportunities to investigate cooperative roles of functionally redundant, clustered genes. Third, the methods allow gene targeting in diverse genetic backgrounds of mouse strains. In traditional gene targeting, ES cells derived from 129 mouse strain are most often used due to high efficiency of gene targeting. However, it is preferred to perform subsequent analyses of targeted mice on C57BL/6 genetic background. Thus, time-consuming backcrossing which takes at least 1 year is essential. As several groups demonstrated (Davies et al. 2013; Qiu et al. 2013), in vivo genome editing can be applicable to any mouse strain and provide opportunities to analyze the targeted mice immediately without backcrossing.

Overall, in vivo genome editing technology drastically accelerates the translation of human genetics into the mouse, in addition to other higher species such as primates (Sasaki et al. 2009), and should revolutionize our understanding of the functional consequences of human genomic diversity in health and disease.


This work was supported by Strategic Research Program for Brain Sciences (SRPBS) from Ministry of Education, Culture, Sports, Science and Technology of Japan, and CREST from Japan Science Technology Agency to K.T. We thank T. Yamamoto and T. Sakuma (Hiroshima University) for technical support and useful discussions, H. Ishikubo and T. Usami (Tokyo Medical and Dental University) for technical support, and M. Abe (Niigata University) and K. Tanaka (Keio University) for useful discussions.