Efficient and rapid fluorescent protein knock-in with universal donors in mouse embryonic stem cells

Fluorescent protein (FP) knock-in enables endogenous tagging for protein visualization without overexpression artifacts (Gibson et al., 2013). A knock-in strategy allows an investigator to accurately observe and measure the dynamics of protein expression, localization and interactions in live cells. FP knock-in has been standard practice in yeast since the 1990s, because this organism can efficiently incorporate FP donors via homologous recombination (Huh et al., 2003; Wach et al., 1997). More recently, FP knock-in has become widely adopted in C. elegans (Dickinson et al., 2013, 2015; Dokshin et al., 2018; Paix et al., 2015) and Drosophila (Baena-López et al., 2013; Li-Kroeger et al., 2018), owing to the advent of CRISPR/Cas9 technology (Bukhari and Müller, 2019). When programmed by a single guide RNA (sgRNA), Cas9 introduces a targeted DNA double-strand break (DSB), which cells can repair by either homology-directed repair (HDR) or non-homologous end joining (NHEJ) (Chiruvella et al., 2013). HDR has been preferred due to its high fidelity (Haupt et al., 2018; Morrow et al., 2021; Ran et al., 2013; Yang et al., 2014). However, it is only active at certain cell cycle stages (Mao et al., 2008a) and requires homology arms that match the target. As a result, HDR-based tagging is much less efficient (He et al., 2016; Mao et al., 2008b) and requires laborious cloning for application in mammalian cells.

To circumvent these constraints, NHEJ has been recently introduced for FP knock-in in mammalian cells (Artegiani et al., 2020; Bachu et al., 2015; He et al., 2016; Lackner et al., 2015; Schmid-Burgk et al., 2016; Suzuki et al., 2016; Zeng et al., 2020; Zhong et al., 2021). One method, named CRISPR-assisted insertion tagging (CRISPaint) (Schmid-Burgk et al., 2016) is especially streamlined due to a minimal requirement for plasmid cloning. CRISPaint employs universal donor plasmids, such that the only necessary cloning involves the construction of gene-specific sgRNAs. The universal donor plasmid is introduced into cells via transfection, cleaved by Cas9 in parallel with the target gene and integrated into the target gene in a non-sequence-specific manner by NHEJ. To allow the use of any gene-specific sgRNA while maintaining the correct reading frame, CRISPaint uses generic ‘frame selectors’ to cleave the universal donor in one of the three possible reading frames (Schmid-Burgk et al., 2016).

Despite these advantages, to date CRISPaint has only been tested in a handful of cell lines. Additionally, only C-terminal insertion is feasible with the CRISPaint system in its present form, which constrains sgRNA options and limits its application to genes whose protein products are tolerant of a C-terminal tag. Many proteins require a specific C-terminal sequence for proper function, localization or post-translational modifications. In such cases, an N-terminal knock-in may be more appropriate. The choice between N-terminal and C-terminal knock-in depends on the specific protein of interest and experimental goals, and for some genes it may be necessary to tag both ends of a gene to determine which tag least perturbs protein function. Thus, there is a need for fast and efficient methods that allow flexibility in FP tagging at either end of a gene and that minimize the cost to generate and test multiple FP insertion locations in a particular gene.

Here, we have adapted the C-terminal insertion strategy based on the CRISPaint system for use in mouse embryonic stem cells (mESCs) and expanded its application to include N-terminal tagging through the introduction of innovative and universally applicable N-terminal donors. Our approach is both efficient and rapid, successfully tagging 5/5 targets on the first attempt and reducing the time from transfection to imaging to as little as 2 weeks. Moreover, we have developed a set of plasmids for multi-color tagging. These advances will facilitate cell biological studies in mESCs and potentially other mammalian cells, and potentially provide a faster and easier route to create knock-in mice.

CRISPaint is a modular three-plasmid system that consists of an FP donor vector, a frame selector Cas9/sgRNA vector and a target Cas9/sgRNA vector (Fig. S1A) (Schmid-Burgk et al., 2016). The donor vector contains an mNeonGreen (mNG) fluorescent protein-coding gene fused with a flexible linker, a self-cleaving T2A peptide and a Puromycin resistance (PuroR) gene. The T2A peptide induces ribosomal skipping during translation, such that the mNG-fused protein of interest and the PuroR are transcribed as a single mRNA but produce two separate protein products (Atkins et al., 2007; Doronina et al., 2008). A pre-made frame selector plasmid expresses Cas9 and an sgRNA that cleaves the linker region of the donor vector at one of three possible positions, chosen based on the target cleavage position to allow the linearized donor to incorporate into the cleaved gene in the correct reading frame. A second Cas9/sgRNA vector cleaves the gene of interest. Of note, the donor vector does not contain its own promoter, so the FP and PuroR sequences can be expressed only if the linearized donor is ligated to the cleaved gene in-frame and in the proper orientation (Fig. S1A).

To date, the CRISPaint system has been tested in a handful of human cells (Artegiani et al., 2020; Schmid-Burgk et al., 2016; Zeng et al., 2020). In this study, we sought to expand the application of CRISPaint into mESCs, which are a powerful and versatile model. mESCs originate from the inner cell mass of preimplantation mouse embryos (Evans and Kaufman, 1981; Martin, 1981). They are easy to maintain, can self-renew and can divide indefinitely in defined growth medium (Mulas et al., 2019). As they maintain naïve pluripotency, mESCs can develop into any type of somatic cells and can generate chimeric mice (Czechanski et al., 2014). Moreover, mESCs are compatible with high-resolution microscopy because of low cellular autofluorescence.

To adapt CRISPaint for use in mESCs, we first replaced the CMV promoters, which were used to drive Cas9 expression, with EF1α promoters. This change was necessary because the CMV promoter has been reported to lack transcriptional activity in mESCs (Chung et al., 2002; Wang et al., 2008) and is prone to transcriptional silencing or reduced expression in some stem cells or differentiated cells (Gray et al., 2011). Second, we observed that the CRISPaint Cas9/sgRNA vectors were prone to recombination in bacteria under standard cloning conditions. We identified a bacterial host strain and lower growth temperature that circumvents this issue and allows straightforward cloning of new gene targets into the Cas9/sgRNA vectors (see details in the Appendix).

As a first test, we attempted to tag the Myh9 gene, which encodes non-muscle Myosin IIA heavy chain, at the C terminus with mNG (Fig. 1A). Myh9 was an appealing first target because it has a single annotated isoform and a known spatiotemporal localization to the cleavage furrow during mitosis, allowing us to use microscopy to assess whether the tagged protein localizes and functions normally in cell division (Dean et al., 2005). We chose an sgRNA that cleaves near the C terminus of the gene using the CHOPCHOP sgRNA design tool (https://chopchop.cbu.uib.no/) (Labun et al., 2019). The chosen sgRNA cuts 26 nucleotides (3N+2 nt, where N is the number of codons) away from the stop codon (Fig. 1B, top). As two nucleotides were lost at the Myosin cleavage site relative to the reading frame, we used frame selector +2 to cleave the donor vector, which should favor in-frame insertions by providing two extra nucleotides in the linker region before the mNG-coding sequence (Fig. 1B, bottom). After transfecting and selecting puromycin-resistant cells, we pooled putatively edited cells and tested for insertions using PCR with a forward primer upstream of the cut site and a reverse primer inside the mNG-coding sequence (Fig. 1C). We detected a specific 0.8 kb band in the pooled knock-in cells, indicating successful FP insertion (Fig. 1C).

We next proceeded to isolate single knock-in clones from the pool of putatively edited cells. PCR genotyping showed that 15/20 clones had an integration of mNG, suggesting a high insertion efficiency (Figs 1D, S1B). Sanger sequencing revealed one clone with a single point mutation at the junction between Myh9 and mNG, and 14 clones with small indels at the junction. The most frequent deletion, removing six nucleotides (encoding two glycine residues) from the linker region of the universal donor vector, was observed in 10 independent clones (Fig. 1B,E). The presence of these indels was surprising because, in human cells, CRISPaint resulted in 61-86% seamless insertions when using the correct frame selector (Schmid-Burgk et al., 2016). This difference may be due to differences in NHEJ mechanisms in mESCs compared with human cells. There are at least two distinct NHEJ pathways, termed alternative NHEJ (aNHEJ) and classical NHEJ (cNHEJ), which produce indels at different frequencies and may have different relative activity levels in different cells (Ceccaldi et al., 2016; Deriano and Roth, 2013). Nevertheless, all the knock-in clones we isolated had deletions whose length was a multiple of three, resulting in an in-frame fusion of mNG to the C terminus of Myosin, as expected (Fig. 1B,E). The loss of six nucleotides (two amino acids) from the linker region of the donor vector is expected to be inconsequential for protein function. Indeed, live imaging confirmed that Myosin-mNG accumulated at the cleavage furrow in dividing cells, indicating the tagged protein is functional (Fig. 1F, Movie 1). In summary, we successfully modified CRISPaint for use in mESCs, and we achieved efficient and functional tagging at the C terminus of Myosin IIA despite small in-frame deletions at the junctions.

A potential concern when using this approach for gene tagging is that bacterial sequences, which are present in the donor plasmid backbone, could interfere with gene expression by disrupting the 3′UTR or other regulatory elements. This issue can be avoided by generating ‘minicircle’ donors from which the bacterial sequences have been removed (Schmid-Burgk et al., 2016). To confirm that the minicircle strategy is applicable in mESCs, we cloned a C-terminal parent vector in which attB-attP recombination sites flank the Linker–mNG–T2A–PuroR cassette (Fig. S1C). In this way, attB/attP recombination circularizes the donor sequences, and the bacterial backbone is eliminated due to cleavage at its 32 I-Scel restriction sites. We induced minicircle production in an engineered E. coli strain (Kay et al., 2010) and purified the minicircle donor (Fig. S1C-D). We note that, although we initially tested published protocols for minicircle production (Kay et al., 2010), we found that extensive further optimization was needed to eliminate bacterial DNA contamination and obtain sufficient yields for transfection (Y.S. and N.K., unpublished observations; see Appendix for details). We transfected the C-terminal mNG minicircle donor into mESCs along with the same Myh9 sgRNA and frame selector, as before (Fig. S1E). PCR genotyping of pooled edited cells demonstrated that the C-terminal minicircle donor was successfully inserted into the Myh9 locus (Fig. S1F).

The current version of CRISPaint is limited to C-terminal tagging, but some proteins have functional domains or motifs at their C termini that may be sterically hindered by FP fusion (Snapp, 2005). Therefore, we sought to develop a similar approach for introducing N-terminal tags. There are several important considerations for N-terminal tagging. First, to ensure gene expression, the insertion must be in-frame at both junctions. Second, no potential stop codons should exist inside the inserted sequence. This poses a challenge if the donor DNA contains the entire bacterial plasmid backbone; therefore, a minicircle donor is required. Third, the order of mNG and PuroR is important; the mNG needs to be adjacent to the gene of interest so that the mNG tag is not removed by T2A peptide cleavage.

To address these concerns, we created a novel N-terminal universal donor vector (Fig. S2A,B). To fuse mNG with the N terminus of a gene, we rearranged the position of the mNG and PuroR relative to the T2A peptide. To avoid putative stop codons, we flanked the attL sequence, which was left over from removal of the bacterial backbone during minicircle donor production, with splicing signals to form an artificial intron (Tikhonov et al., 2017). To facilitate in-frame insertion, we redesigned the linker to have a length that is a multiple of three. We produced new frame selector sgRNA vectors for the N-terminal minicircle donor to achieve a correct reading frame on both 5′ and 3′ junctions.

We tested our N-terminal minicircle donor strategy by tagging Myh9 at the N terminus (Fig. 2A). For N-terminal knock-in, we chose a sgRNA that cuts within the 5′ UTR, immediately before the start codon. As this cut site is 3N+2nt away from the stop codon, we used frame selector −2 to favor in-frame tagging (Fig. 2B). After co-transfection and drug selection, PCR genotyping indicated successful insertion in pooled edited cells (Fig. 2C). PCR genotyping of clonal cells showed that 9/12 clones had correct insertions with a specific 1.8 kb band (Fig. 2D, S2C). Two of the incorrect clones had multiple copies of the minicircle donor inserted head-to-tail into the cut site (Fig. S2C,F); these clones were successfully edited but are not usable for imaging due to the presence of an additional copy of mNG that is not fused to Myosin.

Sanger sequencing of the nine clonal cell lines that had correct insertions identified one insertion with a seamless 5′ junction, one with a seamless 3′ junction, and one that was seamless at both junctions (Figs 2B, bottom; 2E and S2D). Thus, it is possible to obtain clones with seamless insertion junctions in mESCs, perhaps with variable frequency at different loci. As expected, among the clones that had indels, all clones maintained the correct reading frame at the junction where the 3′ end of the donor was fused to Myosin (Fig. 2B, bottom, E). However, at the 5′ end of the donor, which is outside the coding region, 6/9 clones had deletions whose lengths were not a multiple of three. This suggests that the length of indels produced by NHEJ in mESCs is essentially random, but that drug selection can effectively enforce an in-frame fusion between the tag and the protein of interest. Timelapse live imaging suggested that the N-terminally tagged mNG-Myosin is functional and localized normally to cleavage furrows in cells undergoing cytokinesis (Fig. 2F, Movie 2). The fluorescence signal was consistent among different knock-in clones (Fig. S2E), suggesting that the small indels we observed did not affect Myosin expression or localization.

Because all the 12 mNG-Myosin clones we isolated were mono-allelic (i.e. only one copy of Myh9 was tagged), we wondered whether the sgRNA we chose was inefficient and had failed to cleave the other allele. Thus, we also sequenced the non-tagged allele in nine clones that had a correct insertion on one allele. Sanger sequencing revealed indels or mutations at the non-tagged allele in 9/9 clones (Fig. S2D). Thus, the sgRNA vector was efficient enough to cut both alleles, but in most cases only one allele was repaired via integration of the donor. In summary, we designed and produced novel N-terminal minicircle donors that could achieve efficient N-terminal endogenous tagging via NHEJ.

To demonstrate the generality of this approach, we tested its applicability in additional mES cell lines. We tagged Myc9 in CJ7 and E14 mESCs, and compared the knock-in results with those obtained in J1 mESCs above. PCR genotyping revealed a specific 0.6 kb band in all three knock-in mESCs (Fig. S2G, left), indicating successful insertions across cell lines. Furthermore, we assessed the frequency of GFP⁺ cells in the pooled edited cell populations (Fig. S2G, middle). We obtained knock-in cells in 15/15 independent transfection experiments across the three cell lines, confirming that our protocol is very robust and efficient. J1 and CJ7 cells showed varying insertion frequencies depending on selection conditions, with more stringent puromycin selection favoring higher GFP⁺ cell counts (Fig. S2G, right). Intriguingly, E14 consistently exhibited a high frequency of GFP⁺ cells in five independent experiments, regardless of puromycin concentration or treatment duration. We noted that E14 was more sensitive to puromycin, with a higher number of dead cells after selection compared with the other two ESCs. This suggests that the GFP⁺ cell frequency might be linked to the sensitivity of cells to puromycin treatment. By adjusting the drug concentration and treatment length through kill curve testing for each cell line, it is possible to easily enrich edited cells. Most importantly, these data show that our approach can produce knock-ins in multiple mES cell lines with similar ease and efficiency.

Next, to test whether our approach would be effective for other loci, we tagged additional genes. We targeted Tpx2, a key mitotic spindle assembly factor; β-catenin, an adherens junction protein; and Zo1, a tight junction protein. We had initially targeted Myh9 because this gene has only a single annotated isoform; however, these additional targets each exhibit multiple isoforms, resulting from alternative splicing (Fig. S3A-C). Therefore, a key consideration for FP knock-in is which isoform to tag. We chose to target the principal isoform, as identified by APPRIS (https://appris.bioinfo.cnio.es/) (Rodriguez et al., 2013), which flags the principal isoform based on protein structure, expression and conserved function (Fig. S3A-C). Based on the gene structure, we targeted the C terminus of each gene, and we used C-terminal minicircle donors (Fig. S1D) to avoid the possibility of disrupting the 3′UTR of each gene.

We readily obtained knock-in cells with tagged Tpx2, β-catenin and Zo1 on the first attempt. Insertion was verified by PCR using pooled edited cells after transfection and selection (Fig. S3D-F). Fluorescence imaging of all three sets of pooled cells showed that most cells were fluorescent, suggesting that knock-in efficiency was high (Fig. 3A-C). The pooled cells were immediately used for live imaging, which was possible within 7 to 14 days after transfection. All three tagged proteins exhibited the expected localization. Tpx2-mNG cells showed faint nuclear fluorescence in non-dividing cells, but bright mitotic spindle localization in dividing cells (Fig. 3A). Timelapse live imaging of Tpx2-mNG cells revealed dynamic localization of Tpx2 during mitosis (Fig. 3D, Movie 3). β-Catenin-mNG cells showed cortical accumulation and brighter signal at cell-cell junctions (Fig. 3D). Timelapse live imaging of β-catenin-mNG cells revealed de novo adherens junction assembly after cell division (Fig. 3E, Movie 4). Finally, Zo1-mNG cells displayed puncta on the cell cortex, which were concentrated at cell-cell contacts (Fig. 3C).

To further validate the expected localization of mNG-Myosin IIA, β-catenin-mNG and Zo1-mNG, we examined the localization of these proteins in 3D spheroid culture of the tagged mESCs (Fig. 4A), which can recapitulate morphogenesis during peri-implantation mouse embryonic development (Bedzhov and Zernicka-Goetz, 2014a). Naïve mESCs grow in 3D culture as disorganized clusters, but when mESCs exit naïve pluripotency in 3D culture, they establish epithelial apicobasal polarity, forming a sphere with an open lumen at the center (Bedzhov and Zernicka-Goetz, 2014a; Shahbazi et al., 2017). As expected, none of these proteins showed a polarized localization in naïve mESCs. In polarized mESCs, β-catenin-mNG localized at cell-cell contacts and was enriched at the apical adherens junction. Zo1-mNG localized to the apical edge of cell-cell contacts, consistent with the expected position of tight junctions in a polarized epithelium (Bryant and Mostov, 2008). Interestingly, mNG-Myosin showed both apical and basal accumulation in polarized mESCs. The apical enrichment was not unexpected as epithelia often enrich Myosin apically (Bryant and Mostov, 2008), but the basal localization has not been reported before, to our knowledge. To address potential tag artifacts, we conducted immunostaining on wild-type mESC spheroids. Native activated Myosin (revealed using a phospho-MLC antibody) exhibited a similar localization pattern to the tagged Myosin, confirming that this localization is not an artifact of introducing the mNG tag (Fig. 4B). The basal distribution of Myosin suggests that it may contribute to the organization of cells into a sphere. Of note, the imaging experiments using β-catenin-mNG and Zo1-mNG were performed using pooled puromycin-resistant cells 10 days after transfection, further demonstrating that our protocol can allow tagging and imaging of endogenous proteins in less than 2 weeks. In summary, live imaging of our knock-in cells in 3D spheroid culture further validates that these tagged proteins are functional, and shows that endogenous tagging can reveal changes in protein localization and cell organization that are a consequence of cell fate transitions.

We also picked monoclonal cells carrying tagged Tpx2, β-catenin and Zo1. As expected, the insertion efficiency was high: 15/18 of Tpx2 clones (Fig. S4A,B), 33/36 β-catenin clones (Fig. S5A,B) and 15/21 of Zo1 clones (Fig. S6A,B) had correct insertions. Sanger sequencing revealed a few seamless insertion junctions along with small insertions, deletions and mixtures of those (Figs S4C,D, S5C,D and S6C,D), but all tags were inserted in-frame. Additionally, we isolated two β-catenin-mNG clones and three Zo1-mNG clones that had both alleles tagged, demonstrating that bi-allelic tagging is possible (Figs S5A and S6A). Western blotting verified tagged β-catenin and Zo1 expression in cells carrying mono-allelic and bi-allelic tags: cells with bi-allelic tags lacked untagged β-catenin or Zo1 protein, whereas the cells with mono-allelic tags expressed both the tagged and untagged forms (Fig. 3F,G). In mono-allelic β-catenin-mNG clones, although the tagged copy of β-catenin-mNG was expressed and localized normally, the tagged protein was present at lower levels than the wild-type (untagged) protein (Fig. 3F). This was not due to incomplete cleavage of the T2A peptide, as western blotting showed no evidence for a β-catenin-mNG-PuroR fusion (Fig. S5E). In contrast, tagged Zo1 was expressed at equivalent levels to the untagged protein (Fig. S6E), demonstrating that reduced expression of tagged proteins is not a general feature of our knock-in approach. We suspect that the reduced expression of β-catenin-mNG is an idiosyncratic and protein-specific effect of a C-terminal tag on β-catenin stability, as the C-terminal tail of β-catenin has been shown to influence its binding affinity for the destruction complex components APC and Axin (Choi et al., 2006).

β-Catenin, a crucial transcriptional factor in canonical Wnt signaling, plays a vital role in stem cell renewal and differentiation (Sokol, 2011), as well as in mouse neuroectoderm differentiation (Haegel et al., 1995; Huelsken et al., 2000). We aimed to determine whether the reduced levels of tagged β-catenin compromise neuroectoderm differentiation potential. We used both wild-type and bi-allelic β-catenin-mNG mESCs to form embryoid bodies (EBs) (Fig. S7A,B), which are three-dimensional aggregates of mESCs that simulate early mouse embryonic development. The EBs were cultured for 6 days after removal of leukemia inhibitory factor (LIF) to allow spontaneous differentiation (Kibschull, 2017). Immunofluorescence staining of wild-type EBs suggested that, under our conditions, EBs differentiated towards anterior neuroectoderm fates marked by Otx2-positive staining. β-Catenin-mNG EBs exhibited similar Otx2 expression as the wild-type EBs, indicating that β-catenin tagging did not adversely affect their differentiation potential. Although we emphasize that proper maintenance of naïve pluripotency and differentiation potential will need to be assessed on a case-by-case basis for new tagged cell lines, in the case of β-catenin a modest reduction in levels of β-catenin-mNG relative to untagged β-catenin did not cause any evident phenotypes or prevent us from observing β-catenin-mNG localization (Figs 3 and 4, Fig. S7).

In our earlier experiments targeting Myh9, we exclusively isolated clones that had one tagged allele; the other allele was cleaved by the target sgRNA but not tagged. This suggested that ligation of the donor was a limiting factor, and we wondered whether using a higher concentration of donor would improve the efficiency of FP knock-in and bi-allelic tagging. Thus, we tested transfections with different [donor vector] : [frame selector] : [target sgRNA] plasmid moles ratios, ranging from 5:1:1 to 50:1:1. Regardless of the transfection ratio, PCR results confirmed an insertion in pooled cells in every experiment (Fig. S3D-F). Thus, all ratios can generate knock-ins with some efficiency. We quantified the frequency of FP⁺ cells in each experiment and found that the fraction of GFP⁺ cells ranged from 36% to 87%, which would be sufficient to allow imaging experiments in pooled cells without the need for clonal selection (Fig. S3G). When targeting Tpx2, we observed a trend of increasing efficiency with higher transfection ratio of donor, but no clear pattern was observed when targeting β-catenin (Fig. S3G). These data suggest that, in some cases, using more donors might favor integration, but tagging can succeed regardless of the amount of donor within the range we tested.

Overall, we conclude that by combining NHEJ and drug selection, endogenous tagging in mESCs is generally very efficient. Although most cell clones had only one allele edited, bi-allelic edits can be isolated by picking single clones. Owing to high insertion efficiency, pooled edited cells can be imaged less than 2 weeks after transfection.

To enable simultaneous labeling of two proteins with different tags in the same cell line, we constructed donor vectors for HaloTag insertion at either terminus of a protein (Fig. 5A). HaloTag is a self-labeling enzymatic tag that can accept a variety of substrates, including small-molecule fluorescent dyes that significantly outperform FPs for live imaging (Grimm et al., 2015, 2020; Los et al., 2008). We included a hygromycin-resistance marker in our HaloTag constructs, enabling their use in cells that already carry the puromycin resistance marker that was used to select mNG knock-ins. Our mNG and HaloTag knock-in donors use the same frame selector plasmids, so that a tagging strategy that has been demonstrated for one tag can be re-used to insert a different tag.

To demonstrate the utility of our Halo tagging vectors for two-color knock-in, we constructed dual-labeled cell lines carrying tags on both β-catenin and Zo1. We generated Zo1-HaloTag in a β-catenin-mNG background, and β-catenin-HaloTag in a Zo1-mNG background. After drug selection, PCR genotyping confirmed a correct insertion in both cell lines (Fig. 5B). We then incubated cells with the JFX₆₄₆ far-red fluorescent HaloTag ligand (Grimm et al., 2021) and performed live imaging in 3D culture. Both dual-labeled cell lines showed the expected distributions of β-catenin and Zo1 at the adherens and tight junctions, respectively (Fig. 5C). In summary, our method can label more than one protein with different FPs in the same cell line by sequential transfection, which makes it possible to evaluate colocalization of endogenous proteins in living cells. In the future, it should be possible to achieve multi-color labeling by adding additional FPs and selectable markers to our existing design.

We have demonstrated an efficient and rapid protocol for FP knock-in in mESCs. Five distinct loci have been edited with successful insertion on the first attempt, highlighting the generality and ease of the strategy. Our method requires minimal cloning because only the gene-specific sgRNA vector needs to be customized. Notably, our approach broadens the potential applications of CRISPaint, allowing for flexibility to tag either end of a gene to produce functional proteins suitable for live imaging.

Our workflow for FP knock-in in mESCs is simple and efficient (Fig. 6). Within 1 week, the sgRNA vector can be designed and cloned, and minicircle donors produced from premade parent donor plasmids. Transfection into mESCs takes under 1 h, and drug treatment begins the following day to enrich in-frame edited cells over 2-3 days. After drug selection, mESCs can be expanded for imaging, PCR genotyping and maintenance within 1-2 weeks. Upon PCR genotyping validation, monoclonal cell lines can be isolated by seeding low cell numbers in a 10 cm plate for 1 week. These cell lines can then be expanded for PCR genotyping and sequencing, which may take 2-3 weeks depending on cell proliferation rate. As a result, the total time required to isolate monoclonal knock-in cell lines is reasonably estimated as 5-6 weeks. A detailed protocol for streamlined FP knock-in is provided in the Appendix.

Combining NHEJ and drug selection gives rise to a high proportion (>70%) of functional knock-ins. The 2-week timetable from transfection to imaging is comparable with the time required for establishing a stable cell line expressing an exogenous transgene, but endogenous tagging can avoid overexpression artifacts and thus provides more reliable results. The high insertion efficiency facilitates isolation of monoclonal cell lines when needed, as fewer than 12 clones are sufficient to obtain multiple functional knock-in clones. It is feasible to isolate seamless and bi-allelic insertions, although this may require screening more clones.

Our approach, although focused on mESCs, is not limited to this cell type. It has the potential to be applied or adapted for use in various mammalian cells for several reasons. First, NHEJ is the major DNA repair pathway of DNA double-strand breaks during all cell cycle stages in mammalian cells (Guirouilh-Barbat et al., 2004; Mao et al., 2008a). Second, the original CRISPaint method was tested in HEK 293 and THP-1 cells (a human monocytic cell line derived from the blood) (Schmid-Burgk et al., 2016). Third, the EF1α promoter we used for expressing Cas9 is expected to be expressed in nearly all cells because it is derived from an essential translation factor.

A potential future application of this approach is to enable rapid generation of knock-in mice expressing FPs from endogenous loci. Although current approaches for gene tagging in mice via zygote or two-cell embryo injection can be highly efficient (Ge and Hunter, 2018; Gu et al., 2018), this varies based on the user as well as on the locus being targeted. In addition, one must then characterize the generated insertions in mice, which are often mosaic due to perdurance of the CRISPR reagents past the stage at which they are injected (Mehravar et al., 2018; Oliver et al., 2015; Yen et al., 2014). By rapidly editing mESCs with positive selection, characterizing them in culture and then injecting them into blastocysts to generate chimeras, it should be feasible to establish new mouse lines or to perform microscopy studies in the chimeric mice themselves.

It is important to note that, although we observed indels or mutations at the junctions between these targeted genes and the donor, these small indels do not significantly undermine the utility of this approach. The majority of indels in tagged alleles affected only the linker region of our targeting constructs. There were a minority of cases where one or two amino acids were lost from the tagged protein, but these small deletions would not be expected to compromise function in most cases. For cases where a tagged protein is very sensitive to such indels, we showed that it is straightforward (albeit more laborious) to isolate clones that have a seamless knock-in. Indels and seamless insertions most likely result from two distinct NHEJ pathways: aNHEJ and cNHEJ, respectively. aNHEJ is end resection dependent and thus produces small deletions, whereas cNHEJ is end resection independent and can produce seamless insertions (Ceccaldi et al., 2016; Chiruvella et al., 2013). In the future, it may be possible to increase the frequency of seamless insertions by inhibiting end resection and/or aNHEJ. Alternatively, by careful screening, it is feasible to achieve scarless insertion if necessary.

Compared with an overexpression strategy, endogenous gene tagging is expected to recapitulate the native expression pattern of the protein of interest more faithfully. However, a caveat is that endogenous tagging still might compromise the function of some proteins, either by influencing gene expression or by steric effects on protein function. For example, our mono-allelic mNG-Myh9 clones had indels at the 5′ end of the untagged allele, some of which may disrupt the expression of the untagged allele. For haploinsufficient genes, such an effect could cause unwanted phenotypes. Nevertheless, haploinsufficient genes are relatively infrequent, representing 3-5% of the genome (Deutschbauer et al., 2005; Huang et al., 2010). For these genes, there are two straightforward solutions. First, where feasible, one can simply choose to make a C-terminal tag, as C-terminal indels in the untagged copy are unlikely to compromise protein function. Second, when an N-terminal tag is required, one can readily isolate bi-allelic tags via clonal selection. We would expect bi-allelic N-terminally tagged clones to be more frequent for the small minority of genes that are haploinsufficient for cell viability because cells with a growth defect will be lost during selection. Additionally, the tagged alleles of β-catenin-mNG that we generated appear to be expressed at lower levels than the untagged protein, for reasons that are unclear. Although this did not prevent us from observing the localization of β-catenin-mNG in microscopy experiments, it could present an issue for experiments where a quantitatively normal level of tagged protein is crucial. Reduced expression of tagged proteins is not a general feature of our strategy or tagging constructs, as tagged ZO-1 expression was equivalent to that of the untagged protein. Determining in advance whether a tag will impact protein function can be challenging. Therefore, assessing the utility of cells carrying endogenously tagged proteins should be carried out on a case-by-case basis, depending on the specific goals of the experiment. Our rapid and versatile approach facilitates such characterization because it allows multiple cell lines, with different tag insertions, to be generated and tested for function in a given experiment.

Our strategy is expected to be less efficient when tagging silent, non-expressed genes compared with the actively transcribed genes we have tested here. Some genes relevant to stem cell biology and differentiation, such as Otx2, Sox11 and Oct6, are not expressed in naïve mESCs. Because the FP and drug resistance sequences rely on the native promoter for expression in our system, the drug resistance marker will not be expressed if the target gene is not expressed. However, NHEJ has been proven efficient at repairing DSBs in both active and silent genes (He et al., 2016). Thus, a possible workaround for a silent gene would be to target the gene in naïve pluripotent stem cells, pick clones and then generate duplicates of each clone in 96-well plates. One copy would be screened for insertions, either by PCR or by differentiating cells and performing drug selection, and then positive clones could be retrieved from the master plate. Alternatively, it might be possible to select for insertions under culture conditions that induce upregulation of some genes that are silent in naïve mESCs, but that are still permissive for reversion to the naïve state upon culturing in 2i/LIF (D′Aniello et al., 2017; Kinoshita et al., 2021; Neagu et al., 2020). A second limitation of our method is the inability of this approach to tag a gene internally, rather than at the N or C terminus. The CRISPR-mediated insertion of exon (CRISPIE) approach is probably a better choice for internal tags (Zhong et al., 2021). The CRISPIE system (Zhong et al., 2021) inserts a FP-coding sequence flanked by splicing signals into an intronic location of the desired gene so that the insertion junctions will be spliced out, leaving the protein coding sequence unaffected by indels caused by NHEJ. This approach is well-suited for internal FP insertion but requires extra cloning to produce N- or C-terminal tags. Despite these limitations, our protocol promises fast, easy and robust endogenous tagging in mammalian cells, allowing native protein visualization in living cells without overexpression.

A list of plasmids generated for this study is provided in Table S1. The key plasmids necessary to replicate this approach are available from Addgene (accession numbers are listed in Table S1). To preserve plasmid integrity and prevent random recombination, Cas9/sgRNA vectors were transformed in NEB stable competent cells (NEB C3040H). For plasmid production, NEB stable competent cells carrying Cas9/sgRNA were grown at a mild temperature (30°C) overnight (see Appendix).

pDD428, which contains the U6 promoter to express Myh9 C-terminal sgRNA and the EF1α promoter to express mCherry-Cas9, was used as a template for new target sgRNA vector construction. Two alternative cloning strategies were tested (see Appendix for details). First, the NEB site-directed mutagenesis kit, following the manufacturer's instructions, was used. The entire pDD428 vector was PCR amplified with a forward primer that included the new target protospacer in place of the existing protospacer, and a reverse primer that annealed to the U6 promoter. The PCR product was self-ligated and clones were verified by Sanger sequencing. In the second approach, a short PCR fragment was generated by amplifying pDD428 with a forward primer that included the new target protospacer and a universal reverse primer that annealed to the EF1α promoter. The vector backbone was prepared by digesting pDD428 with SalI-HF and EcoRI-HF, and NEB HiFi DNA assembly was performed to insert the PCR product containing the new protospacer into the digested vector. Table S2 provides primers used for each strategy and Table S3 lists the target protospacers used in our sgRNA vectors.

We have generated four parent donor vectors: mNG-PuroR and Halo-HygroR for C-terminal tagging, and PuroR-mNG and HygroR-Halo for N-terminal tagging (Fig. 5A). These constructs are based on an empty parent minicircle cloning vector (MN100A-1, System Biosciences). The premade parent donor vectors contain attB-attP recombination sites flanking the FP plus drug-resistant gene cassette. They have been transformed into an engineered E. coli called ZYCY10P3S2T (SBI MN900A-1), which enables minicircle donor production (Kay et al., 2010). To make minicircle donors, fresh ZYCY10P3S2T E. coli carrying a parent donor vector were grown in 50 ml of TB medium containing 50 µg/ ml kanamycin at 30°C with shaking at 250 rpm until the OD value was 4-6. When the required OD value was achieved, 50 ml of induction buffer was added and growth was continued at 30°C for 3 h to produce minicircles, then at 37°C for 1 h to degrade the vector backbone. The induction buffer was made by mixing 400 ml of fresh LB, 16 ml of 1 N NaOH and 0.4 ml 20% L-arabinose (Thermo Fisher, AC104980250). Finally, 100 ml induced E. coli were collected and aliquoted into 3 ml per tube. The induced E. coli was spun down at 4°C. After removing media, pellets of induced E. coli were stored at −20°C. To extract minicircle donors, the QIAprep Spin Miniprep Kit (Qiagen, 27104) was used. Before minicircle donor extraction, one aliquot of induced E. coli at 4°C was thawed and resuspended with 500 μl of P1 buffer. This was then lysed with 500 μl of P2 buffer for 5 min and mixed with 700 μl of N3 buffer, centrifuged and transferred supernatant to one QIAprep spin column. The spin column was washed with 500 μl of PB buffer and 750 μl of PE buffer. Finally, the minicircle donors were eluted with 50 μl of Nuclease-free water. 3 ml of induced E. coli yielded around 2.5 µg of minicircle donors (50 ng/μl in 50 μl), which is sufficient for six transfections. The quality of the minicircle donor was checked by linearizing the donor and performing gel electrophoresis (See Fig. S2B). See Appendix for details.

J1 (ATCC, SCRC-1010), CJ7 and E14 (a gift from Jonghwan Kim, University of Texas at Austin, USA) mESCs were maintained in 2i/LIF medium at 37°C in 7% CO₂ on gelatin-coated plates (Mulas et al., 2019). 2i/LIF was made from N2B27 with 3 µM of GSK3β inhibitor (Sigma-Aldrich, SML1046), 1 µM of MEK inhibitor (Sigma-Aldrich, PZ0162), 100 U/ml of leukemia inhibitory factor (Sigma-Aldrich, ESG1106) and 1× penicillin-streptomycin (Thermo Fisher Scientific, 15140122). N2B27 was made in a 1:1 mixture from 487 ml of DMEM/F12 (Sigma-Aldrich, D6421) and 487 ml of Neurobasal medium (Thermo Fisher Scientific, 21103049) supplemented with 0.5× B27 (Invitrogen, 17504044), 1× N2 (custom made, see below), 50 μM β-mercaptoethanol (Sigma-Aldrich, M3148-25ML) and 2 mM L-glutamine (Thermo Fisher Scientific, 25030081). Custom-made N2 contained DMEM/F12 (Sigma-Aldrich, D6421), 2.5 mg/ml insulin (Sigma-Aldrich, I9278), 10 mg/ml Apo-transferrin (Sigma-Aldrich, T1147), 0.75% bovine albumin fraction V (Thermo Fisher Scientific, 15260037), 1.98 μg/ml progesterone (Sigma-Aldrich, P8783-1G), 1.6 mg/ml putrescine dihydrochloride (Sigma-Aldrich, P5780-5G) and 0.518 μg/ml sodium selenite (Sigma-Aldrich, S5261-10G). The final density for cell seeding was 15×10⁴ cells/ml. 2i/LIF was renewed every day and mESCs were routinely passaged every 2-3 days. Mycoplasma testing (Southern Biotech, 13100-01) was performed periodically.

For routine passaging of mESCs in gelatin-coated 10 cm plates, the growth medium 2i/LIF was removed and 3.5 ml Accutase (Sigma-Aldrich, SCR005) was added to cells and incubated at 37°C for 5 min. To dissociate mESCs into single cell suspension, cells were pipetted up and down 10-15 times. The cell suspension was then added to 10.5 ml of wash medium (500 ml of DMEM/F12+8 ml of bovine albumin fraction V) in a 15 ml tube. Cells were spun down for 5 min at 100 g, and 1 ml of 2i/LIF was used to resuspend cells. 10 µl cells were mixed with 10 μl of Trypan blue (Thermo Fisher Scientific, 15250061) and counted using a cell counter (Invitrogen, AMQAF1000). Finally, 150×10⁶ cells with 10 ml of 2i/LIF were seeded in a new gelatin-coated 10 cm plate.

Growth factor reduced Matrigel (Corning 356230) was used in 3D spheroid cultures (Bedzhov and Zernicka-Goetz, 2014b; Shahbazi et al., 2017). One well of a 15-well imaging plate (ibidi 81506) was covered with 1 µl of ice-cold Matrigel and incubated for 15 min at 37°C to allow the solidification of Matrigel. N2B27 or 2i/LIF was used to resuspend mESCs, and cell density was adjusted to 10×10⁴ cell/ml. 30 µl of cells (0.3×10⁴ cell per well) were then seeded on each Matrigel-coated well. When the cells had settled down to the Matrigel (15 min after seeding), the media were removed and replaced with 50 µl of N2B27+M or 2i/LIF+M (N2B27 or 2i/LIF containing 5% Matrigel). It took 60-72 h for cells in N2B27+M to form polarized spheres. Cells in 2i/LIF remained unpolarized throughout 3D culture.

EB growth medium was made from DMEM/F12 (Sigma-Aldrich, D6421) with 20% KnockOut Serum Replacement (Thermo Fisher Scientific, 10828028), 2 mM L-glutamine (Thermo Fisher Scientific, 25030081), 1× EmbryoMax nucleosides (Sigma-Aldrich, ES-008-D), 1× MEM nonessential amino acids (Thermo Fisher Scientific, 11-140-050), 1× penicillin-streptomycin (Thermo Fisher Scientific, 15140122) and 100 μM β-mercaptoethanol (Sigma-Aldrich, M3148-25ML). AggreWell400 24-well plates (Stemcell Technologies, 34411) were used for EB formation. The plate was treated with 500 µl of anti-adherence rinsing solution (Stemcell Technologies, 07010). To spread the solution evenly and remove bubbles from the microwell, the plate was centrifuged at 1300 g for 5 min. The anti-adherence rinsing solution was aspirated after centrifugation. Each well was rinsed with 1 ml of warm EB growth medium then removed the medium and 1 ml of warm EB growth medium added. To generate EBs, a single-cell suspension of mESCs in EB growth medium was prepared with a cell density of 2.4×10⁶ cells/ml to achieve 2000 cells per microwell. The final volume of EB growth medium was added up to 2 ml/well. To ensure even distribution of cells throughout the well, pipetting and centrifugation at 100 g for 3 min were applied. Cells in the AggreWell plate were incubated at 37°C with 7% CO₂ and 95% humidity for 6 days. Warm EB growth medium was changed every day by slowly and gently removing 1 ml and replacing with 1 ml medium.

EBs were transferred and settled down in a tube, and spheroids were kept in the plates. Medium was aspirated and fixed with 4% PBS-paraformaldehyde (Thermo Fisher Scientific, J60401-AK) for 20 min at room temperature. PBS was then used to wash three times (5 min each).

For spheroids, the fixed samples were incubated with permeabilization buffer, composed of 0.3% Triton X-100 and 0.1 M glycine in PBS for 20 min at room temperature. PBST buffer with 0.1% Tween in PBS was used to wash three times (10 min each with rocking). To block spheroids, samples were incubated in blocking buffer (10% serum in PBST) for 2 h at room temperature. For myosin staining, spheroids were incubated with a primary antibody Phospho-MYL9 (Ser19) Polyclonal Antibody (1:50 dilution, Thermo Fisher Scientific, PA5-17726) overnight at 4°C on a rocker. PBST buffer was used for wash three times (1 h each). Secondary antibodies AlexaFluor 647 donkey anti-rabbit (1:500 dilution, Thermo Fisher Scientific, A-31573) and AlexaFluor 555 Phalloidin (1:400 dilution, Thermo Fisher Scientific, A34055) were applied for 1 h in the dark at room temperature. Spheroids were washed with PBST buffer three times (1 h each) and with PBS twice (10 min each). The PBS was aspirated and slides mounted with ProLong Mountant with DAPI (Thermo Fisher Scientific, P36941-2 ml) for imaging.

For EBs, the large size (>100 μm) requires cryostat sectioning. The fixed EBs were incubated in 30% sucrose overnight at 4°C. The EBs were embedded with OCT compound (Thermo Fisher Scientific, 23-730-571) at least three times and transferred into a 10×10×5 mm Cryomold (Thermo Fisher Scientific, NC9806558) at −80°C to solidify the EB molds. The EB molds were sectioned using an NX50 Cryostat at10 μm and EB samples were adhered to the positively charged slides (Thermo Fisher Scientific, 12-550-15) to eliminate sample loss during staining. The EB slides were allowed to dry, then washed with PBS twice (5 min each) in a coplin staining jar (Thermo Fisher Scientific, 08-817). The slides were permeabilized with PBST buffer for 15 min on the shaker. To save material, a circle was drawn around the EBs with a hydrophobic slide marker PAP pen (VWR 22005). EBs were incubated in blocking buffer with 10% serum in PBST for 1 h at room temperature. An Otx2 primary antibody was used for Otx2 staining (1:200 dilution, R&D Systems, AF1979) overnight at 4°C on a rocker. The slides were washed with PBST buffer twice (10 min each) and with PBS for 10 min. Secondary antibody AlexaFluor 647 donkey anti-goat (1:500 dilution, Thermo Fisher Scientific, A-21447) was applied for 1 h in the dark at room temperature. Slides were washed with PBST buffer twice (10 min each) and with PBS for 10 min, then moved from the PBS, mounted with ProLong Mountant with DAPI (Thermo Fisher Scientific, P36941-2 ml) and coverslipped. The slides were sealed with clear nail polish and store in dark slide box at 4°C until imaging.

Three plasmids (an FP donor vector, a frame selector Cas9/sgRNA vector and a target Cas9/sgRNA vector) were transfected into mESCs using Lipofectamine 2000 (Thermo Fisher Scientific, 11668027). The molar ratio of [donor vector]:[frame selector]:[target sgRNA] plasmid was typically 20:1:1, and was achieved using 364 ng minicircle donor, 130 ng target Cas9/sgRNA vector and 130 ng frame selector. The three plasmids were mixed and diluted to 50 µl in Opti-MEM Medium (Gibco, 31985062). 1.25 µl Lipofectamine 2000 was diluted to 50 µl in Opti-MEM Medium, then added to the 50 μl of DNA solution and incubated at room temperature for 5 min. After incubation, the 100 μl DNA/Lipofectamine 2000 mixture was transferred to a gelatin-coated 12-well plate. 5×10⁵ single cells in suspension per well were added to DNA/Lipofectamine 2000 mixture. 2i/LIF was added up to 1 ml per well. The same procedures were performed with 50 μl of pure Opti-MEM Medium and 50 μl of diluted Lipofectamine 2000 as a negative control. The transfection mixture was replaced with 2-3 ml of fresh 2i/LIF and mCherry-Cas9 signal observed in the transfected cells the following day. Triplicates of co-transfection were used for each gene tagging. Table S4 lists plasmids that were co-transfected in each experiment.

On day 2 after transfection, the transfected cells were treated with either 1-2 µg/ml puromycin or 100-200 µg/ml hygromycin, depending on the chosen donor vector. Each day, the plates were gently washed one or twice with wash medium to remove dead cells, and then fresh drug-containing medium was added. This process was repeated daily for 2-3 days of drug treatment. Most of the cells appeared floating and dead after 1 day of puromycin or 3 days of hygromycin treatment. After selection was complete, the medium was replaced with 2i/LIF to expand surviving cells. On day 6 or 7 after transfection, the drug-resistance cells started to form small colonies. The surviving cells were maintained in 2i/LIF and were ready for genomic DNA extraction or imaging within 14 days of transfection.

To generate monoclonal cell lines, 1×10⁴ pooled edited cells with 10 ml of 2i/LIF were seeded on gelatin-coated 10 cm plates. 2i/LIF was renewed every day until the individual cell formed a discernible colony that was visible with a naked eye. On day (d)7-d10, colonies were picked using a P10 pipet set at 5 µl under a dissection microscope. The picked colonies were incubated in U-shape-bottomed 96-well plates with 20 µl accutase in each well at 37°C for 10 min. After incubation, 150 µl of 2i/LIF was added to the accutase/cell mixture, pipetted up and down to dissociate colonies into single cells, then cell suspensions transferred to a gelatin-coated U-shape-bottomed 96-well plate. On the next day, most of the medium was gently removed but enough was left to cover the cells and fresh 150 µl of 2i/LIF was added. The 2i/LIF was changed every day until cells reached confluence. If cells did not attach to the bottom of the plate, 1% FBS in 2i/LIF was added to facilitate this.

After transfecting and enriching drug-resistant cells, putatively edited cells were split for genomic DNA extraction. To extract genomic DNA from mESCs, the PureLink Genomic DNA Mini Kit (Thermo Fisher Scientific, K182001) was used by following the manufacturer's instructions. Typically, 5×10⁵ mESCs yielded around 6.25 µg of genomic DNA (∼250 ng/μl in 25 μl). To verify insertion into pooled cells, genomic DNA of pooled cells was extracted for PCR genotyping. Meanwhile, genomic DNA of wild-type cells was collected as a negative control.

For monoclonal cells, genomic DNA Mini Kit can be used for PCR genotyping. However, it was expensive and time-consuming to genotype many clonal cell lines in this way. Quick lysis was therefore used in place of genomic DNA extraction for clonal cells. When clonal cells reached confluence in the 96-well plate, cells were split into two sets of plates, one in a 96-well plate for PCR genotyping, the other in a 24-well plate for subculture. 2i/LIF was changed every day until cells in the 96-well plate reached confluence. All the medium in the 96-well plate was gently removed and 100 µl of Quick lysis buffer was added. Quick lysis buffer comprised 10 mg/ml proteinase K powder (Sigma-Aldrich, P6556), 50 mM KCl, 10 mM Tris-HCl (pH 8), 2 mM MgCl₂, 1% (v/v) NP40 (Thermo Fisher Scientific, 85124), 0.45%(v/v) Tween 20 and milli-Q water. Colonies were dissociated by pipetting up and down, then transferred the cell lysates to PCR Tube Strips, the plate sealed and incubated at 65°C for 2 h followed by 95°C for 10 min. Finally, cell lysates were diluted 1:20 with nuclease-free water and 1 µl of the resulting solution was used for PCR genotyping.

To verify insertions, a forward primer upstream of the cutting site and a reverse primer inside the donor vector were used. For C-terminal insertions, a reverse primer inside the mNG or HaloTag coding sequence was used. For N-terminal insertion, the reverse primer was inside the PuroR-coding sequence. To distinguish whether an insertion is mono-allelic or bi-allelic (i.e. only one genomic copy or both genomic copies are tagged), a pair of primers upstream and downstream of the cutting site were used. Table S5 lists primer information for PCR genotyping (see Appendix for details).

As template for PCR genotyping, 50 ng of purified genomic DNA from pooled edited or wild-type cells or 1 µl diluted cell lysate from clonal cells was used. PCR was performed using LongAmp Taq 2× Master Mix (NEB M0287S) according to the manufacturer's instructions, and results were visualized by electrophoresis.

After verification by PCR and electrophoresis, specific bands were extracted using the Zymoclean Gel DNA Recovery Kits (Zymo D4007/D4008). Purified amplicons with a corresponding forward primer or reverse primer were submitted for Sanger sequencing (Table S5). Sanger sequencing results were aligned with the wild-type or knock-in allele to validate insertion and characterize the insertion sequence.

Cells were resuspended from 2D culture with PBS at various cell concentrations. After lysing with NuPAGE LDS Sample Buffer (Thermo Fisher Scientific, NP0007) containing 100 mM DTT, the reduced cell samples were incubated at 70°C for 10 min. Equal amounts of cell sample and 1× benzonase nuclease (Sigma-Aldrich, 70746-3) was added to degrade nucleic acids. Samples were sonicated to further reduce any contamination of nucleic acids. The samples were run on a 4-12% Bis-Tris NuPAGE gel (Thermo Fisher Scientific, NP0323) using MES Running buffer (Thermo Fisher Scientific, NP0002) at 160 V for 1 h with antioxidant. The gel was then transferred to a polyvinylidene fluoride (PVDF) membrane at 20 V for 1 h. A blocking buffer (5% non-fat dry milk dissolved in phosphate buffered saline-Tween (PBST) was used for 0.5 h at room temperature to prevent non-specific binding. The membrane was then incubated with primary antibodies overnight at 4°C: anti-α-tubulin (1:200, DSHB 12G10), anti-β-catenin (1:20,000 dilution, Sigma-Aldrich, ABE208) and anti-Zo1 (1:500 dilution, Thermo Fisher Scientific, 33-9100). The membrane was washed three times using PBST, blocked using 5% non-fat dry milk in PBST and then incubated with secondary antibodies for 45 min at room temperature: anti-mouse 680 (1:50,000 dilution, Thermo Fisher Scientific, A-21057) and anti-rabbit 800 (1:20,000 dilution, Thermo Fisher Scientific, SA5-10036). The membrane was washed three times with PBST followed by two washes with PBS and imaged using a LICOR-Odyssey CLx.

To image cells with HaloTag, HaloTag ligand dye JFX₆₄₆ (a gift from Luke Lavis, Janelia Research Campus, Howard Hughes Medical Institute, VA, USA) was used. It was firstly dissolved in acetonitrile to 1 mM and aliquoted into 2 µl in PCR tubes. The dye was dried with a vacuum centrifuge for storage at −20°C in a desiccator. Before imaging, one aliquot was dissolved with 2 µl of DMSO to reconstitute 1 mM dye then diluted to 0.037 µM with growth medium. Before imaging, cells were incubated with 0.037 µM JFX₆₄₆+2i/LIF for at least 30 min in a 15-well imaging plate at 37°C.

Cell tracing dyes including SPY555-Actin (Cytoskeleton CY-SC202), SPY650-Tubulin (Cytoskeleton CY-SC503) and SPY650-DNA (Cytoskeleton CY-SC501) were dissolved in 50 µl of fresh DMSO to reconstitute at 1000×. They were then aliquoted into 5 µl in PCR tubes, protected from light and stored at −20°C. Before imaging, cells were incubated with 1× cell tracing dye with growth medium for at least 15 min in a 15-well imaging plate at 37°C. There was no wash step after dye incubation because all dyes were fluorogenic and produced low background fluorescence.

Cells were seeded in 15-cell imaging slides and kept in the tissue culture incubator until ready for imaging. To observe epithelial polarity in 3D culture, cells were incubated for 60-72 h. To observe the frequency of GFP⁺ cells in 2D culture, cells were incubated for 12 h. Before imaging, a humidifier was filled with water and a gas mixer (Okolab 2GF-MIXER) with air pump and 100% CO₂ and temperature controller (Okolab H401-T-CONTROLLER) were turned on so that the imaging chamber (Okolab H201-NIKON-TI-S-ER) could pre-equilibrate to the correct humidity at 7% CO₂ and 37°C for live cells. Then images were acquired using one of two microscopes. The images shown in Fig. 1 were acquired using a Nikon Eclipse Ti-2 microscope equipped with 20×0.75 NA and 60×1.45 NA objectives, a Photometrics PrimeBSI camera, an OptoSpin filter wheel (CAIRN Research, Kent, England) and a vt-iSIM super-resolution confocal scan head (VisiTech international, Sunderland, UK). mNeonGreen fluorescence was excited using a 505 nm diode laser. All other images were acquired using a Nikon Eclipse Ti-2 microscope equipped with 20×0.75 NA and 60×1.4 NA objectives; an 89 North LDI-7 laser diode illuminator; a Photometrics Prime95B camera; and a Crest X-Light V3 spinning disk confocal head. mNG, SPY555-Actin, SPY650-Tubulin SPY650-DNA and HaloTag-JF₆₄₆ were excited using the appropriate lines of the LDI-7 illuminator. For fixed imaging, mESCs were imaged by multiple z-planes with a step of 1.5 µm. For Timelapse live imaging, cells were imaged every 5-30 min in multiple z-planes with a step of 1.5 µm.

To prepare figures, the midplane of confocal image stacks was selected, and images were cropped and rotated using FIJI. Some images were processed with a Gaussian Blur filter with a 0.7 pixel radius to reduce background noise, and the brightness and contrast were adjusted for visibility of the signals. No other image manipulations were performed.

After genotyping, at least five different monoclonal cells were pooled and frozen. To freeze edited cells for future use, cell number was counted and at least 5×10⁶ cells were collected in 2i/LIF per tube. Cells were then spun down for 5 min at 100 g and resuspended by using 500 μl of freezing medium containing 450 μl of 2i/LIF and 50 μl of DMSO per tube. Cells were transferred in a pre-labeled cryovial, put into a freezing chamber and placed in a −80°C freezer immediately. Finally, the cryovial was transferred from the freezing chamber to the −80°C freezer or a liquid nitrogen tank.

To recover edited cells in cryovial for imaging or sequential transfection, the frozen cells were thawed in a 37°C water bath for 1.5 min, then mixed with 10 ml wash medium and transferred into a 15 ml tube. Cells were spun down for 5 min at 100 g, resuspended with 10 ml of 2i/LIF and seeded onto a 10 cm gelatin-coated plate. Cells were passaged at least three times before any experiments.

We thank Ivy Chang and Qiuxia Zhao for help with cloning; Jonghwan Kim, Steve Vokes and members of the Dickinson lab for helpful discussions and comments on the manuscript; Luke Lavis for sharing JaneliaFluor dyes; and Xiangjun Zhao (University of Manchester) for sharing the quick lysis protocol. The authors gratefully acknowledge the support of the Microscopy and Imaging Facility of the Center for Biomedical Research Support at UT Austin.