Developmental analyses of mouse embryos and adults using a non-overlapping tracing system for all three germ layers

Specification of the cellular populations responsible for development or pathogenesis has attracted attention in recent decades. Various genetic techniques have been developed for analyses of gene functions and lineage fates. Among them, site-specific recombinases (SSRs), such as Cre, are powerful genetic tools for tracing the specific progeny of cells that once expressed a gene of interest in vitro and in vivo (Sternberg and Hamilton, 1981; Sternberg et al., 1981; Sauer and Henderson, 1988; Orban et al., 1992).

Recently, several additional SSR systems, such as Flp/FRT, Dre/rox, VCre/VloxP, phiC31/att, Vika/vox, and Nigri/nox, have come into use for the manipulation of two or more distinct genes in eukaryotes (McLeod et al., 1986; Dymecki, 1996; Raymond and Soriano, 2007; Anastassiadis et al., 2009; Suzuki and Nakayama, 2011; Groth et al., 2000; Belteki et al., 2003; Karimova et al., 2013, 2016; Liu et al., 2018). The context for induction of SSRs needs to be considered carefully to enable precise tracing. For example, Sox17-iCre;R26R mice, in which a codon-improved Cre (iCre) is inserted at the transcriptional start region (TSR) of the first exon in the Sox17 locus, show endothelial-specific expression of LacZ in arteries (Liao et al., 2009), whereas Sox17-2A-iCre;R26R mice, in which iCre is conjugated to the Sox17 gene with self-cleaving 2A peptide and the transcript preserves the intact Sox17 intragenic regions and 3′ untranslated region (UTR) sequence, show more precise and comprehensive expression of LacZ in endoderm-derived tissues (Engert et al., 2009). These results indicate that lineage-specific promoters should be introduced with their full context, including their intragenic regions, for precise lineage tracing. It is difficult to directly compare the dynamic contributions of distinct cell lineages using conventional single-gene tracing systems, which occasionally produce ectopic gene expression. Ectopic expression in single-gene tracing systems has been reported in T-Cre;R26R mice, which show an ectopic β-gal⁺ phenotype in ectodermal tissues, such as lens, caudal neural tube and posterior surface ectoderm (Perantoni et al., 2005), although brachyury (T) is well known as a pan-mesodermal marker.

Here, we sought to focus on the development of all three germ layers (ectoderm, mesoderm and endoderm); however, examining the contributions, interactions, trans-differentiation and gene expression orders among these during embryogenesis or in adults can be challenging without the use of invasive manipulations. For this purpose, we developed a new combinatorial lineage-tracing system for visualization of progenies derived from the three germ layers using self-cleaving 2A peptide and multiple SSRs to resolve the issues in the conventional systems outlined above. All three germ layers in the resultant quadruple gene-targeted mice were conditionally and simultaneously labelled with distinct fluorescent proteins. In addition to these technical advances, we provide significant new insights into lineage fate commitments in the current study.

To specify the three germ layers, we selected Sox1, brachyury and Sox17 as lineage markers for the neuroectoderm, mesoderm and endoderm, respectively (Takashima et al., 2007; Ying et al., 2003; Perantoni et al., 2005; Kumar et al., 2008; Liao et al., 2009; Engert et al., 2009, 2013; Choi et al., 2012; Burtscher et al., 2012). We designed the targeting constructs with a combinatorial system containing self-cleaving 2A peptides and multiple SSRs for labelling the three germ layer markers in an intact gene expression manner (Fig. 1, Fig. S1). Multiple genes fused with the 2A peptide sequence successfully result in bi-cistronic expression of proteins, as ribosomal skipping caused by the 2A peptide occurs at the conserved core sequence (GDVExNPGP) during translation (Luke et al., 2008). The 2A peptide-mediated expression of multiple proteins has a notable benefit in enabling higher expression levels of proteins compared with IRES (internal ribosome entry site)-mediated expression. Among the common 2A peptides, porcine teschovirus 1 2A (P2A) is reported to have a high cleavage efficiency (Kim et al., 2011; Liu et al., 2017); therefore, we exploited P2A for bi-cistronic expression of SSRs in equimolar amounts to lineage marker genes, and for retaining of their intragenic enhancers, 3′ UTR sequence and poly(A) tail. Next, we used Cre, Dre and a codon-optimized phiC31 (phiC31o; Raymond and Soriano, 2007) as lineage-labelling SSRs, given that all have high recombination efficiencies and do not show cross-reactivity (Fig. S2). As FLP or a codon-optimized FLP (FLPo) has been reported to produce efficient and independent recombination in wild FRT or mutant (F3 and F5) sets (Schlake and Bode, 1994), it was also used for simultaneous removal of puromycin-resistance cassettes in different lineage marker alleles in vivo. In the reporter construct, multiple fluorescent proteins are flanked by the corresponding recombination target sequences at the Gt(ROSA)26 locus, under the control of the CAGGS promoter, which enhances ubiquitous expression (Zong et al., 2005). However, their transcription is abrogated by a puromycin-resistance gene, 3-frame stop codons and poly(A) signal for complete prevention of ribosomal read-through before any recombinations (Fig. 1, Fig. S1). Thus, the reporter construct conditionally induces the distinct expression of fluorescent proteins in a lineage marker-dependent manner; i.e. Sox1 labels neuroectoderm as Venus⁺, brachyury labels mesoderm as mCherry⁺ and Sox17 labels endoderm as mCerulean⁺. Furthermore, this system can change colours from mCherry (brachyury⁺ mesendoderm) to mCerulean (Sox17⁺ endoderm) in response to further lineage specification.

Homologous recombination of the ROSA reporter allele (designated as ROSA26^MultiFPs) and three lineage-labelling alleles (designated as Sox1^2A-Cre, brachyury^2A-Dre and Sox17^2A-phiC31o) in the puromycin-resistant mES clones was confirmed by Southern blotting analysis or PCR-based genotyping methods (Fig. S1E,F). For chimeric mice production, we selected clones 1, 25 and 26 (all homozygous) as the ROSA reporter mES clones, and clones 49 (homozygous for brachyury and Sox17, and heterozygous for Sox1) and 2–3 (homozygous for Sox1 and Sox17, and heterozygous for brachyury) as the triple lineage-labelling mES clones. Chimera mice were generated by injection to eight-cell stage ICR embryos, and the chimeric contributions of the injected mES clones were estimated using the coat colours of chimeras. For both lines, high chimeric F0 males were crossed with CAG-FLPo (Yamazaki et al., 2016) females to remove puromycin-resistance cassettes in the targeted genes for restoration of their intact 3′ UTR sequences and poly(A) tails. Following the initial crossing, we genotyped the F1 mice to confirm germline transmissions. Among them, we chose clone 26 and 2-3 derivatives (designated as ROSA26^{MultiFPsΔPuro} and the triple lineage-labelling mice, respectively) owing to the stable production of the offspring. For ROSA26^{MultiFPsΔPuro}, the heterozygous mice were inbred, and a homozygous line was established. For the triple lineage-labelling mice, males heterozygous for all lineage markers were selected. Finally, we crossed the triple lineage-labelling males (Sox1^{2A-CreΔPuro/+}, brachyury^{2A-DreΔPuro/+} and Sox17^{2A-phiC31oΔPuro/+}) to ROSA26^{MultiFPsΔPuro/MultiFPsΔPuro} females to obtain the objective mice. We designated such quadruple gene-targeted mice as TRiCK (TRiple Coloured germ layer, Knock-in) mice.

We first validated the lineage-tracing phenotype of the TRiCK mice in developing embryos. To this end, we mated the two mouse lines as described above, expecting that TRiCK mice would be produced with a probability of one-eighth. As a preliminary experiment, we dissected a pregnant mouse and obtained ten E14.5 embryos; these displayed the different combinations of fluorescent protein expressions (Fig. 2A). Among them, one TRiCK mouse simultaneously showed non-overlapping and conditional labelling in all three germ layers (Fig. 2B), in which the brain, spinal cord and eyes were Venus⁺, the trunk was mCherry⁺, and mCerulean⁺ cells were scattered across the surface of the whole body. We next sought to examine the temporal patterning in the early development of TRiCK mouse embryos. The E8.5 embryo showed Venus⁺ neural plate surrounded with mCherry⁺ mesodermal cells (Fig. 2C). mCerulean⁺ endodermal cells were detected in the caudal part of the embryo. The TRiCK mouse embryos displayed the dynamic lineage specifications at E9.5 and E10.5, with the formation of Venus⁺ neural tube and optic vesicles, mCherry⁺ somites and organ buds, and definite commitment to mCerulean⁺ endoderm in the branchial arches (Fig. 2D,E). We also confirmed the specificity of the TRiCK system using a littermate embryo that did not show any fluorescent signals as a negative control (Fig. 2E). The TRiCK mouse embryos precisely maintained the fluorescent boundaries of the three germ layers until late development (Fig. 2B,F,G).

We next sought to analyse TRiCK embryos using recently developed 3D imaging techniques for detailed analyses of dynamic lineage specifications during embryogenesis. Many techniques for tissue clearing, such as CLARITY, PACT, 3DISCO, Scale, ScaleS, CUBIC, SeeDB and SeeDB2 (Chung et al., 2013; Treweek et al., 2015; Becker et al., 2012; Hama et al., 2011, 2015; Susaki et al., 2014; Ke et al., 2013, 2016), have been reported. These techniques are beneficial for spatial mapping of diverse cell populations and understanding their biological relationships. However, these techniques often induce a decline in fluorescent protein (XFP) signals during the clearing performance. One of the solutions to this issue is signal enhancement by XFP-specific antibodies; however, it is difficult to independently detect all XFPs used in the TRiCK system with antibodies, as Venus and mCerulean have highly preserved their protein sequences (230 of 239 amino acids are identical). Furthermore, the commercially available anti-GFP antibodies crossreact with both GFP/YFP (Venus) and CFP (mCerulean). Therefore, it is important to retain the intact XFP signals during the tissue clearing process.

As the first step for improving tissue clearing, we investigated the optimal fixation protocol, as the common fixation condition, 4% paraformaldehyde (PFA) in PBS, resulted in decreased XFP intensities and significantly increased background, especially at short wavelengths. We compared several concentrations of PFA, diluent buffers and additive combinations. Then, we observed the highest performance in preservation of XFP signals when we used 0.5-2% PFA diluted with PHEM buffer (pH adjusted to 8.0) instead of PBS. Furthermore, the addition of 5% D-sorbitol effectively suppressed the increase of auto-fluorescence during fixation (data not shown).

Next, we optimized the tissue clearing method. We started by evaluating the performance of Scale U2 and S4 solutions as promising previous techniques, because their aqueous-based methods are expected to be mild for XFPs (Hama et al., 2011, 2015). However, when we immersed the fixed E13.5 TRiCK embryo in Scale U2 or S4 solution, we observed a decrease in XFP signals and the accumulation of auto-fluorescence at short wavelengths that overlap with mCerulean detection (data not shown). To handle this issue, we focused on the SeeDB2 method (Ke et al., 2016), which is designed to preserve most of the XFP signals, and is based on the high refractive index (RI) of iohexol, usually used as a clinical contrast medium (e.g. Omunipaque350). The samples immersed in SeeDB2 medium showed partial transparency, which was insufficient for whole-embryo imaging (Fig. 3A). However, SeeDB2 controlled auto-fluorescence at low levels during the clearing process (data not shown). We also evaluated the Ce3D method (Li et al., 2017), which uses Histodenz, a brand of iohexol, as an alternative, but, unexpectedly, it caused extreme sample shrinkage in our hands (Fig. 3A). Therefore, we proceeded to screen the original additives in combination with iohexol for enhancement of tissue transparency with preservation of soft tissue structures. We compared several chemicals in general laboratory use, and found that the addition of monoethanolamine (MEA) delivered superior tissue transparency without obvious tissue deformation or decrease in XFP signals. Notably, triethanolamine did not show any transparency enhancement in combination to iohexol, although it was reported to be effective for tissue clearing in ScaleCUBIC-2 medium (Susaki et al., 2014). Further experiments indicated that D-sorbitol and EDTA can also contribute to the retention of XFP signals. We named the final formulation of tissue clearing solution IMES (iohexol, monoethanolamine, EDTA, sorbitol-based clearing) solution. The IMES method achieved high performance in tissue transparency with preservation of intact structures (Fig. 3A).

We immersed the fixed E13.5-14.5 TRiCK embryos in IMES solution and processed clearing, confirming that it is compatible with all XFPs in the system. The cleared TRiCK embryo showed almost completely complementary expression of XFPs throughout the entire body (Fig. 3B, Movie 1), but some cell populations were positive for both mCherry and mCerulean (Fig. 3C arrowheads). In the tongue, mCherry⁺ muscle cells were arranged in a line and lay beside the Venus⁺ cells, including gustatory nerves (Fig. 3D, Movie 2). We also confirmed that numerous Venus⁺ cells dispersed throughout the connective tissues of the head. Moreover, we could clearly distinguish the cell populations of different cell origins in the eye (Fig. 3E). Thus, we validated that TRiCK mice can potentially become a powerful tool in analyses of dynamic developmental relationships between the three germ layers when combined with tissue clearing methods, especially the IMES technique.

The TRiCK mice showed no abnormalities at birth, and grew to adults. We next examined contributions of three germ layers in different organs of over 8-week-old adults. The brain and spinal cord entirely expressed Venus; however, it was remarkable that mCherry⁺ cells and mCerulean⁺ cells were also conspicuously present throughout the brain (Fig. 4). As microglia originates from yolk-sac mesoderm (Kierdorf et al., 2013; Hoeffel et al., 2015) and brachyury is co-expressed with the classical microglial marker IBA-1 during in vitro differentiation to microglia (Ormel et al., 2018), we performed immunostaining for IBA-1 on floating brain slices and subsequently cleared using the IMES method. Nearly all IBA-1-positive cells were also labelled with mCherry (Fig. S3A,B), suggesting that microglia emerge from the brachyury-positive mesodermal cell population. Interestingly, mCherry⁺ microglia strongly accumulated at cerebellar sulci and resided along veins (Figs 4 and 5A,B). Although veins showed no detectable fluorescence (Fig. 5B), capillary arteries could be visualized with both mCerulean⁺ and mCherry⁺ endothelial cells, which might be the result of incomplete recombination by Sox17-driven phiC31o (Fig. 5B,C). Therefore, we further evaluated the recombination efficiency of phiC31o in brain capillaries that could be stained with the endothelial cell marker PECAM-1. By calculating the ratio of mCerulean⁺/PECAM-1⁺ cells to mCherry⁺/PECAM-1⁺ cells (Fig. S3C), we estimated phiC31o-mediated recombination in brain endothelial cells occurred at an average efficiency of 76% (Table 1).

Since it has been previously reported that the lenses and retina are Sox1 positive during embryogenesis (Kamachi et al., 1998; Nishiguchi et al., 1998), these were labelled with Sox1-dependent Venus (Fig. 4). Furthermore, connective tissues at eye rims, extraocular muscles and some corneal stromal cells were mCherry⁺ (Fig. 4, Fig. S4A). Consistent with their mesodermal origins, kidney and spleen uniformly displayed mCherry expression (Fig. 4). We observed a mosaic pattern of both mCherry⁺ and mCerulean⁺ cells in the lung (Fig. 4). Moreover, mCherry was intensely expressed in the trachea, and Venus⁺ peripheral nerves extended on its surface (Fig. 4, Fig. S4B). The peripheral nerves were also clearly visualized with Venus in the thoracic domain (Fig. S4C). In the gastrointestinal tract, nearly all cells were mCherry⁺, and the spotted clonal signals of mCerulean⁺ cells were also present at the greater curvature of the stomach and small parts of the intestine (Fig. 4). Notably, we also confirmed Venus⁺ labelling in the enteric nervous system (Fig. S4D).

The liver expressed both mCherry and mCerulean in a mosaic pattern with the different proportions in lobes, and we confirmed both mCherry⁺ and mCerulean⁺ cells mostly co-expressed a hepatocyte marker HNF4α (Figs 4 and 6A, Fig. S5, upper). To our surprise, the mCherry⁻/mCerulean⁻ double-negative hepatocytes were present in a considerable cell population of TRiCK mouse liver. To investigate this phenotype, we considered the possibility of XFP degradation using XFP-specific antibodies. While anti-RFP antibody staining showed complete colocalization with mCherry signals (Fig. S5, middle), anti-GFP antibody (it recognizes Venus and mCerulean in this system, but Venus expression was not detectable in the liver) staining resulted in some different profiles. Strong anti-GFP signals coincided with mCerulean expression (Fig. S5, bottom, arrows). In contrast, the cells with weak anti-GFP signals did not express definite mCerulean signals, or only expressed the minimum for detection (Fig. S5, bottom, arrowheads). These results indicate the hepatocytes, which were apparently negative for all lineage labels, partly resulted from the protein degradation of mCerulean.

We observed a more complex phenotype in the heart, in which mCherry was expressed in a marbled pattern throughout the entire organ, Venus labelled the cardiac nerves and numerous mCerulean⁺ cells were dispersed on the surface of the heart in an independent pattern to coronary artery routes (Figs 4 and 6B,C). We identified mCherry⁺ cell population as TNNI3⁺ cardiomyocytes (Fig. S6A). As for mCerulean⁺ cells, Sox17 is activated in the haematopoietic stem cells (HSCs) giving rise to foetal macrophages (Engert et al., 2009); HSC-derived cardiac macrophages reside in the adult heart (Epelman et al., 2014; Frantz and Nahrendorf, 2014). Therefore, we immunostained heart tissues with macrophage markers, such as CD11b and F4/80; however, none of these markers was co-expressed with mCerulean signals (data not shown). For further exploration, we tested colocalization of mCerulean and PECAM-1 with signal enhancement by anti-GFP antibody; mCerulean (anti-GFP signal)-positive cells mostly coincided with PECAM-1-positive capillaries in the heart (Fig. S6B). Consequently, we identified these cells as turned out endothelial cells. Moreover, we discovered there was an exceptional cell population, similar to that in the liver, that was negative for all lineage labels (Fig. 6B). However, these non-labelled cells were negative for anti-RFP antibody (Fig. S6C), which suggests that the protein degradation of mCherry was not involved in its marbled expression pattern.

We also examined the fluorescent phenotype in gonads. The testis expressed only mCherry, however, the ovary and uterus expressed punctate mCerulean signals, in addition to mCherry signal in the entire organs (Fig. 7). TRiCK mice were fertile and their offspring showed whole-body mCherry⁺ phenotype (data not shown), consistent with the initial involvement of brachyury to germ cell lineage commitment in mice, as previously described (Aramaki et al., 2013).

Here, we report a new combinatorial system using self-cleaving 2A peptides and multiple SSRs for triple lineage fate tracing, as well as a new tissue clearing method that preserves XFP intensities, with a particular focus on the development of the three germ layers. As conventional single-gene tracing systems have difficulties in comparing the contributions of different lineage populations, and in determining the sequential gene expression orders of different lineage markers, we designed a quadruple knock-in system to resolve these issues. In this system, three germ layers (neuroectoderm, mesoderm and endoderm) can be labelled conditionally and simultaneously by using distinct fluorescent proteins that label Sox1, brachyury and Sox17 gene expressions. To validate the system, we analysed the phenotype of the resultant quadruple gene-targeted (TRiCK) mice, as developing embryos and adults.

We confirmed that, in most instances, TRiCK mice precisely maintain the boundaries of multiple lineage progenies labelled with distinct fluorescent proteins in bona fide patterns that are consistent with the results from conventional single-gene tracing systems. As an exception, TRiCK embryos partly possessed a cell population double-positive for mCherry⁺ and mCerulean⁺ (Fig. 3C). However, because adult TRiCK mice over 8 weeks old did not show any overlap of XFP expression in any organ (Figs 4–7, Figs S3–S6), we concluded this phenotype in the developing embryo was due to the retention of mCherry protein, although its translation had been terminated by phiC31o-mediated recombination. Therefore, we considered that these cells were in transition from mCherry⁺ to mCerulean⁺ phenotype. In addition, there were some particular characteristics of the TRiCK mice that differed from the previous single-gene-tracing models. For example, the lenses and caudal neural tube were completely visualized with Sox1-driven Venus (Fig. 2B-G), although these tissues were reported to be labelled in T (brachyury)-Cre;R26R mice (Perantoni et al., 2005). These results point to the possibility that the previously reported ectopic expressions of lineage markers are the outcome of secondary gene expression. As the caudal neural tube was solely labelled with Sox1-driven Venus at E10.5 and later stages, in which the caudal axial stem cells have already shown their bi-potency to neural or mesodermal fates in Tbx6-KO mice (Takemoto et al., 2011), our results indicate that the lineage fate commitment of the caudal axial stem cells proceeds without lineage fluctuations. Thus, it is important to determine the order of gene expression for a better understanding of bona fide cellular origins or organ development.

It was also interesting that considerable cell populations, which were negative for all three lineage markers, could be observed in the liver and heart (Figs 4 and 6, Figs S5 and S6). While we observed that protein degradation of XFPs, especially with mCerulean, was partly related to this phenotype in the liver (Fig. S5), this would not affect mCherry detection in the liver and heart (Figs S5 and S6). These results are consistent with a previous study that referred to the possibility of incomplete recombination in the heart of T (brachyury)-Cre;R26R mice embryo (Perantoni et al., 2005). However, as mCherry⁺ mesendoderm differentiates to mCherry⁺ mesoderm or mCerulean⁺ endoderm in the TRiCK system, mesendoderm-derived tissues should express either mCherry or mCerulean. Thus, these cell populations appear never to have expressed brachyury at sufficient transcriptional levels for recombination throughout their lineage fate commitments from mesendoderm. Moreover, there are additional possible explanations, such as gene silencing of the reporter gene allele or that the cytotoxicity of SSRs or XFPs causes elimination of the recombined cells or the reporter gene allele. Notably, the marbled fluorescent pattern observed in the heart suggests that the non-labelled population was specified early in heart development, and that these cells clonally proliferated and maintained the quiescent mode for brachyury expression. We should further explore whether there are physiological differences between the labelled and non-labelled cell populations.

We detected incomplete recombination of Sox17-driven phiC31o in capillary endothelial cells (Fig. 5C, Fig. S3C, Table 1) and other endodermal tissues. As we also observed the relative low recombination efficiency of phiC31o in vitro compared with Cre and Dre (Fig. S2), this is a major issue that needs to be improved in the TRiCK system. In such circumstances, other efficient and incompatible SSR systems, e.g. the recently developed Vika/vox (Karimova et al., 2013) or Nigri/nox (Karimova et al., 2016; Liu et al., 2018), can be used to perform multiple genetic manipulations as replacements for conventional SSRs and/or the combinatorial genetic tool that enables labelling of a fourth target population, such as the neural crest lineage. In addition, Brachyury^2A-DreΔPuro homozygous mice exhibit the short tail phenotype and die before adulthood (data not shown), as is typical of brachyury-KO mice (Pennimpede et al., 2012). All lineage marker alleles targeted in the TRiCK system could be maintained as homozygous because 2A peptide-mediated protein cleavages support endogenous gene expression. However, our results suggest that 2A cleavage was not highly effective or that the cleaved fragment of 2A peptide influenced the function of the brachyury protein.

Moreover, the value of TRiCK mice is significantly increased when combined with recently developed tissue clearing methods for comprehensive analyses of diverse cell populations. To this end, we also developed a new aqueous-based clearing method, named IMES, in which MEA is one major component (20 v/v% in IMES) in the clearing step. Although the mechanism needs to be further clarified, MEA enhanced the decolouring process and improved the permeabilization of high RI-matched clearing solution into samples. It is noteworthy that such a high concentration of MEA did not cause tissue shrinkage, unlike N-methylacetamide used in the Ce3D method, but over 40 v/v% of MEA resulted in a dramatic loss of XFP signals (data not shown). The optimal formulation of IMES solution generates the high tissue transparency that is sufficient for three-dimensional lightsheet observations within 1 week, with preservation of intact XFP signals, including CFP, and with the lower background levels, which have not been achieved using previous methods.

The TRiCK mice described here can simultaneously visualize and distinguish cells derived from all three germ layers without the use of invasive manipulations. Notably, this system also sheds light on the mechanism for lineage fate commitment of caudal axial stem cells. In addition, we were able to readily distinguish intriguing cell types and tissues, such as microglia, capillary endothelium in the brain and heart, and the peripheral nervous network of each organ in the TRiCK mice. This may also be helpful in analysing temporal and spatial contributions of multiple cell lineages in pathogenesis, e.g. the bilateral characters of microglia and vascular remodelling after cerebral infarction (Mabuchi et al., 2000; Hu et al., 2012; Perego et al., 2011; Kanazawa et al., 2017). Moreover, the combination of TRiCK mice and the IMES clearing method enables the simultaneous multi-colour visualization of three germ layers and immunostained targets, which may facilitate the understanding of diverse cellular relationships. Thus, we expect that TRiCK mice will contribute to progress in the study of development, regeneration and pathogenesis.

All animal experiments were approved by the Institutional Animal Experiment Committee of Keio University (approval number 09091) and Osaka University (approval number Biken-AP-H25-02).

The C57B6/N and C57B6/J hybrid mouse embryonic stem (mES) cell line B6NJ-22^(UTR) (RIKEN BioResouce Research Center, AES0141; Tanimoto et al., 2008) were cultured at 37°C in 5% CO₂ under feeder-free conditions in N2B27 medium consisting of DMEM/F12 (Fujifilm Wako) and Neurobasal medium mixed at a 1:1 ratio supplemented with 1% N2 supplement, 2% B27 supplement, 2 mM L-glutamine, 0.1 mM NEAA, 1 mM sodium pyruvate (all from Thermo Fisher Scientific), 0.1 mM 2-ME (Sigma), 1000 U/ml hLIF (Merck Millipore), 1 μM PD0325901 (Cayman), 3 μM CHIR99021 (Cayman), 10 ng/ml hBMP4 (HumanZyme) and 10 mM SB431542 (Sigma) on 0.1% gelatinized dish. HEK293T cells were maintained in DMEM (Sigma) supplemented with 10% FBS (Sigma).

Targeting vectors were designed as shown in the figures (Fig. 1, Fig. S1). Briefly, a ROSA26 reporter plasmid was constructed according to standard cloning methods using the following elements: CAGGS promoter (pCAGGS-Cre; RIKEN BioResouce Research Center, RDB08998; Ogawa et al., 2006), Venus, a variant of yellow fluorescent protein with fast and efficient maturation for cell-biological applications (Nagai et al., 2002) derived from pTP1-Venus (Kohyama et al., 2005), mCherry (pG-PB-Zscan4c-mCherry-polyAfloxPGKneo, a gift from Minoru Ko, Keio University School of Medicine, Japan), mCerulean (Cerulean, a gift from Dave Piston, Washington University School of Medicine, USA; Addgene plasmid #15214; Rizzo et al., 2004), puromycin-resistance gene (puroR) and FRT-PGK/em7-puroR-FRT (pCAGGS-dre-ires-puro, a gift from A. Francis Stewart, Technische Universität Dresden, Germany; Anastassiadis et al., 2009) and pBTloxP (a gift from Takeshi Yagi, Osaka University, Japan; Yokota et al., 2011). The 5′ and 3′ homology arms for bacterial artificial chromosome (BAC) recombination, SSR target sequences, Kozak sequence and 3-frame stop codons were synthesized with primer extension techniques and then cloned. ROSA26 reporter plasmid was recombined to C57B6/N mouse BAC clone (RIKEN BioResouce Research Center, B6Ng01-316F09), and retrieved to a minimal vector using the Red/ET system (Gene Bridges). Sox1, brachyury and Sox17 lineage-labelling plasmids were constructed according to standard cloning methods using the following elements: Cre (pCAGGS-Cre; RIKEN BioResouce Research Center, RDB08998; Ogawa et al., 2006), Dre (pCAGGS-dre-ires-puro, a gift from A. Francis Stewart; Anastassiadis et al., 2009), phiC31o (pPGKPhiC31obpA, a gift from Philippe Soriano, Icahn School of Medicine at Mount Sinai, USA; Addgene plasmid #13795; Raymond and Soriano, 2007). The 5′ and 3′ homology arms for BAC recombination, in-frame GSG (Gly-Ser-Gly)-linker plus P2A and 2×SV40 nuclear localization signal (NLS) sequence were synthesized with primer extension techniques, and PGK/em7-puroR flanked by FRTs (FRT/F3/F5) was amplified by polymerase chain reaction (PCR) and then cloned. Sox1, brachyury and Sox17 lineage-labelling plasmids were recombined to B6N mouse BAC clones (RIKEN BioResouce Research Center, B6Ng01-307M06, B6Ng01-307A03, and B6Ng01-211A18, respectively), and retrieved to minimal vectors with Red/ET system. For constructing expression vectors of streptococcus pyogenes Cas9 D10A mutant (Cas9 nickase) and guide RNA (gRNA), a synthesized gRNA sequence designed to each target site was cloned to pSpCas9n(BB)-2A-Puro (a gift from Feng Zhang, Broad Institute of Massachusetts Institute of Technology and Harvard, USA; Addgene plasmid #62987; Ran et al., 2013). The gRNA sequences were generated using the CRISPR Design Tool (Ran et al., 2013; this tool is no longer available). The target sites of the recombination and gRNAs are listed in Table S1.

Circular targeting vectors were transfected to mES cells by electroporation using NEPA21 (NepaGene) under the following conditions: poring pulse [voltage: 125 V, pulse length: 5 ms, pulse interval: 50 ms, number of pulses: 2, fading rate: 10%, polarity: +] and transfer pulse [voltage: 20 V, pulse length: 50 ms, pulse interval: 50 ms, number of pulses: 5, fading rate: 40%, polarity: +/−]. Homologous recombination was enhanced by co-transfection of Cas9 nickase expression vectors, which were already cloned with the corresponding gRNAs. Notably, targeting of the Sox1, brachyury and Sox17 genes was performed at a single manipulation to maintain the quality of mES cells. After electroporation, cells recovered in medium without antibiotics for 48 h, and then puromycin selection (1 μg/ml) was started. Culture medium was changed every 2 days for 8-10 days, and the resulting puromycin-resistant colonies were picked and expanded.

Homologous recombination was confirmed using Southern blotting analysis and a PCR-based genotyping method. For Southern blotting analysis, genomic DNAs were extracted and purified from mES clones, and digested with NheI (ROSA26 and Sox1) or SacI (brachyury and Sox17). The digested genomic DNAs were electrophoresed in 0.8% agarose gel, and transferred to positive-charged nitro membranes. The detection of the DNA fragments was performed with DIG-labelled specific probes using PCR DIG Probe Synthetic kit (Roche), which were amplified with the primers listed in Table S2. For genotyping PCR, genomic DNAs were extracted and purified from mES clones or hind limb toes of the infant mice. The specific region of each targeted gene, as shown in Fig. S1, was amplified with PrimeSTAR MAX DNA polymerase (TAKARABIO) using the primers listed in Table S2. PCR amplifications were performed by two-step reaction with 2% DMSO under the following conditions: initial denaturing for 2 min at 98°C, 35 cycles of 10 s at 98°C and 1-5 min at 70°C (ROSA, Sox1, brachyury and FLPo) or 76°C (Sox17).

The homozygous recombinant mES clones were selected and injected into 8-cell embryos derived from ICR mice to produce ROSA reporter mice (ROSA26^MultiFPs) and triple lineage-labelling mice (Sox1^2A-Cre, Brachyury^2A-Dre and Sox17^2A-phiC31o). The resultant all F0 chimeric males were crossed with CAG-FLPo (Yamazaki et al., 2016) females to remove puromycin-resistance cassettes flanked by FRTs (FRT/F3/F5). ROSA26^{MultiFPsΔPuro} heterozygous F1 mice were inbred to homozygosity, and the resulting homozygous line was maintained. Finally, triple lineage-labelling heterozygous F1 males (Sox1^{2A-CreΔPuro/+}, Brachyury^{2A-DreΔPuro/+} and Sox17^{2A-phiC31oΔPuro/+}) were crossed with ROSA26^{MultiFPsΔPuro/MultiFPsΔPuro} homozygous females to generate the objective mice. The homozygous ROSA26^{MultiFPsΔPuro} mice and the triple lineage-labelling mice (heterozygous for Brachyury^2A-DreΔPuro, homozygous for Sox1^2A-CreΔPuro and Sox17^{2A-phiC31oΔPuro}) will be deposited in the RIKEN BioResource Research Center.

Embryos were dissected from the pregnant females at each time point after crossing. Adult mice were sacrificed at over 8 weeks of age, and all organs were dissected. All surgeries were performed under isoflurane-induced anaesthesia. Embryos and adult organs of mice were observed with fluorescent stereomicroscope system (Olympus, SZX16) with YFP/CFP/RFP2 filters. The contrast of fluorescent image was adjusted to the levels that negative control specimens or backgrounds show minimal auto-fluorescence, using levels tool in Adobe Photoshop CS5, and then the images were merged.

E13.5-E14.5 embryos were fixed with 2% PFA (Nacalai-tesque) in 5% D-sorbitol (Fujifilm Wako)–PHEM buffer [60 mM PIPES (Dojindo), 25 mM HEPES (Sigma), 10 mM EGTA (Dojindo) and 2 mM MgCl₂ (Nacalai-tesque), adjusted to pH 8.0] for 2 days at 4°C. After fixation, the samples were washed with PHEM buffer three times. For tissue clearing, the samples were immersed in the permeabilization buffer [PHEM buffer, 2 w/v% saponin (Nacalai-tesque), 5 w/v% D-sorbitol and 0.05 w/v% NaN₃ (Nacalai-tesque)] for over 8 h, and subsequently in a mixture of permeabilization buffer and IMES solution [Omunipaque350 (Daiichi-Sankyo), 2 w/v% saponin, 20 v/v% monoethanolamine (MEA) (Fujifilm Wako), 0.1 v/v% TritonX-100 (Fujifilm Wako), 5 w/v% D-sorbitol, 2.5 mM EDTA (Sigma) and 0.05 w/v% NaN₃] at ratios of 2:1 and 1:1 for 24-48 h, respectively, and then immersed in 100% IMES solution for 24-48 h. All clearing processes were performed at room temperature. Twenty-four hours before imaging, we moved the samples into the mounting solution (Omunipaque350, 0.1% TritonX-100, 5% D-sorbitol, 2.5 mM EDTA and 0.05% NaN₃, adjusted to pH 8.0 and passed through a 0.22 μm pore size filter). Tissue clearing protocols for Scale U2/S4, SeeDB2 and Ce3D methods have been previously published (Hama et al., 2011, 2015; Ke et al., 2016; Li et al., 2017, respectively). The spatial distributions of the labelled cells were observed on Lightsheet Z.1 microscope (Zeiss) using a 5× objective lens, and excited with 445, 514 and 561 nm lasers for mCerulean, Venus and mCherry, respectively. Images were processed using FIJI (Schindelin et al., 2012), Zen (Zeiss), Vision4D (Arivis) and Imaris (BITPLANE). We stitched the tiled input images and reconstructed the entire stack using Vision4D or Grid/Collection stitching plug-in (Preibisch et al., 2009) in FIJI.

Adult organs were fixed with the optimal fixative solutions. Brains were fixed with 2% PFA in 5% D-sorbitol–PHEM buffer (pH 8.0) for 2 days at 4°C. Livers and hearts were immersed in ice-cold 5 mM KCl-PHEM buffer (pH 8.0) for 30 min, and then fixed with 0.5% paraformaldehyde in PHEM buffer (pH 8.0) or with the addition of 5% D-sorbitol and 5 mM KCl for 2 days at 4°C. After fixation, the samples were washed with PHEM buffer three times. For vibratome sectioning, the samples were immersed in ice-cold PBS, and sectioned at 300 μm using a VT1200S (Leica) under the following conditions: amplitude 2 mm/s and speed 0.2 mm/s. Tissue slices were permeabilized and blocked in the blocking buffer [1% bovine serum albumin and 2% saponin (brain) or 1% bovine serum albumin, 2% saponin and 0.1-0.3% TritonX-100 (liver and heart) in PHEM buffer (pH 8.0) plus 0.05% NaN₃] over 3 h. Primary and secondary antibodies were diluted in the blocking buffer and incubated for 3 days at room temperature. All staining steps were carried out after 3 washes with the washing buffer [2% saponin (brain) or 2% saponin and 0.1-0.3% TritonX-100 (liver and heart) in PHEM buffer (pH 8.0) plus 0.05% NaN₃]. After secondary antibody washes, the slices were cleared in IMES solution for 20 min, transferred to the mounting solution as described above for 20 min and then mounted on the glass slides in the mounting solution with a 0.3 mm silicon spacer. For frozen sectioning, the heart samples were immersed in 30% sucrose-PHEM buffer (pH 8.0) for 2 days at 4°C, and then snap-frozen in OCT compound (Sakura-finetek). The frozen samples were cryosectioned at 10 μm using a CM3050S-IV (Leica). The sections were re-hydrated in PHEM buffer for 5 min, and permeabilized and blocked in the blocking buffer as described above, for 20 min. Primary and secondary antibodies were diluted with Can Get Signal solution (TOYOBO) and incubated for 30-60 min at room temperature, respectively. After three washes, the samples were directly mounted in the mounting solution. Confocal microscopy imaging was performed on an inverted LSM880 (Zeiss). Primary and secondary antibodies used in this study are listed in Table S3. Images were processed with FIJI.

Z-stack images of brain slices (over 200 μm thick) immunostained for PECAM1 were imported into FIJI. The antibody signals (Alexa647 for PECAM-1) and the fluorescent protein signals (mCherry and mCerulean) were processed to threshold images after contrast adjustment with the constant parameter settings. The double-positive (mCherry⁺/PECAM-1⁺ or mCerulean⁺/PECAM-1⁺) pixels were detected using the Calculator Plus plug-in, and the numbers and volumes of the double-positive objects were counted using 3D objects counter plug-in.

ROSA26 reporter plasmid was transfected to HEK293T cells with Lipofectamine 2000 Transfection Reagent (Thermo Fisher Scientific) following the manufacturer's instructions. At 24 h after transfection, the transfected cells were selected with 1 μg/ml puromycin. The resulting drug-resistant cells were passaged, and were advanced to the second transfection of each SSR-expressing plasmid, pCAGGS-Cre, pCAGGS-Dre-ires-puro or pPGKPhiC31obpA. We confirmed the fluorescent signals with the fluorescent microscope system IX81 (Olympus) 6 days after transfections.

We thank Doug Sipp for critical reading and revision of the manuscript; A. Francis Stewart for Dre; Philippe Soriano for phiC31o; Dave Piston for mCerulean; Minoru Ko for mCherry; Feng Zhang for CRISPR/Cas9n; Takeshi Yagi for pBTloxP2; Kana Namiki, Tetsushi Hoshida and Atsushi Miyawaki for technical advice; and members of Okano laboratory for discussions. Computational resources were provided by the Data Integration and Analysis Facility, National Institute for Basic Biology, Okazaki, Aichi, Japan.