Background

Morphogenesis is a complex process in which appropriate cell types are differentiated and positioned at the right place and at the proper timing. The surprising reproducibility of developmental processes is underpinned by the robustness of the genetic program [1]. However, in spite of the high robustness under normal genetic conditions, the program can be easily collapsed by genetic abnormalities; for example, some genes require both alleles for proper function (i.e., haploinsufficiency) [2]. This type of fragility is frequently observed in mammalian heart development. In the heart, even a slight alteration of the program leads to congenital heart diseases (CHDs) and this fact is associated with the high frequency of CHDs, which is around one in one hundred births [3]. Genetic studies have shown that many of the transcription factor genes involved in the heart development are regulated in a highly spatiotemporal manner [4]. However, how such an intricate control of gene expression is achieved has not been well understood.

Comparative genomics have shown that the complexity of the body plan and the proportion of non-coding regions in the genome are positively correlated [5]. While most of the non-coding regions have previously been considered as “junk”, it is now accepted that some of them are necessary for the regulation of genes in a fine and complicated way [6]. Many evo-devo studies support this view, suggesting that the evolution of multicellular organisms was largely driven by the adjustments in transcriptional regulators, such as enhancer elements, rather than by functional evolution of protein-coding genes [7]. Recent advancements in genomics and transcriptomics have demonstrated that nearly half of the mammalian genome is actually transcribed into RNAs [8]. Long non-coding RNA (lncRNA) is an emerging class of RNA that is generally defined as RNAs longer than 200 nucleotides that lack the ability to produce functional proteins. Many of these molecules have been demonstrated to work as transcriptional or translational regulators [9]. Some lncRNAs are known to recruit epigenetic regulators to specific loci in the genome to modulate transcription. For example, a classical lncRNA, Xist, recruits Polycomb repressive complex 2 (PRC2) to the X chromosome in cis to inactivate one of the two X chromosomes to achieve dosage compensation [10]. Many lncRNAs studied thus far have been found to bind epigenetic factors and recruit them to defined genomic loci; however, not a small proportion of proposed lncRNA-PRC interactions have been suggested to be non-specific [11, 12]. Other lncRNAs function as post-transcriptional modulators of gene expression through the formation of duplexes with mRNA to inhibit translation by RNAi (i.e., antisense transcripts) [13], through the inhibition of miRNAs by working as so-called sponges [14] or by controlling splicing [15]. Although much attention has been paid to lncRNAs recently, the low conservation of sequences across species and the difficulty of determining their three-dimensional structures make it difficult to functionally and evolutionarily classify these molecules. Their biochemical characteristics (e.g., strong nonspecific binding to proteins) also make it difficult to dissect their precise molecular functions [12, 16]. Many lncRNAs show stage- and tissue-specific expression patterns, suggesting their roles in development [17].

Although several lncRNAs that function in mammalian heart development have been reported, the identification and characterization of lncRNAs in the mammalian heart are still insufficient [18,19,20,21,22,23]. Considering the regulatory nature of lncRNAs, they are thought to be key components in solving the aforementioned problems regarding the developmental fragility in mammalian hearts.

Here, we report that key cardiac transcription factors genes are located in close proximity to genes encoding lncRNAs. Interestingly, there are transcription factor and lncRNA pairs that are bidirectionally transcribed from the same promoter. We have focused on one lncRNA near Tbx5, which we call Tbx5ua, and showed that it is required for heart development. Tbx5ua-knockdown mice showed abnormally thin ventricular walls and were embryonic lethal.

Results

Identification of lncRNAs that are expressed during mouse heart development

To identify novel lncRNAs that are specifically expressed during heart development in mice, we extracted total RNA from the ventricles of embryonic day (E) 10.5 and E13.5 and 8 weeks-old mice and prepared cDNA libraries, that were subjected to paired-end 2 * 100 bp RNA-seq. The resulting read count was approximately 40 M reads for each sample. The obtained reads were mapped to the mouse genome (mm10) with Tophat2 [24], and the mapped reads were assembled using Cufflinks [25] with and without UCSC transcript annotations. Because many of the currently known functional lncRNAs are spliced and because it is difficult to confirm the existence of non-spliced transcripts unless they are expressed at very high levels, we focused on spliced lncRNA candidates in our analysis. We set the lower limit of expression at a fragments per kilobase of exon per million mapped fragments (fpkm) of 1, because above that level, the accuracy of the reconstruction of known transcripts without the transcript reference was sufficiently high (Additional file 1). We also checked if exons of known genes are mistaken as lncRNAs. We found that the direction of a majority of the lncRNAs that are located within 10,000 bp from known genes are in the opposite direction from them (225 vs 86), suggesting that such mis-annotations are rare.

From the assembled transcripts, already known mRNAs or functional RNAs that are not generally classified as lncRNAs (e.g., snoRNA and tRNA) were removed, and we also omitted RNAs that have CDS longer than 1/3 of their total length according to the standard of Ensembl, since they are potentially protein-coding transcripts. As a result, we were able to identify 787 candidates of spliced lncRNAs. To omit lncRNAs that are ubiquitously expressed without tissue specificity, we examined the expression of the obtained candidates in the mouse brain. Because the brain is an organ that diverges from the heart at a very early developmental stage and originates from the ectoderm, whereas the heart originates from the mesoderm, we used the brain as a reference organ. We here just wanted to exclude lncRNAs that are expressed with no tissue specificity and did not intended to find lncRNAs that are exclusively expressed in the heart since many genes are known to function differently according to the context of the tissues. The comparison revealed that 316 of the identified spliced lncRNA candidates were selectively expressed in the heart (Fig. 1a, Additional file 2: Table S1). We checked the expression of these genes in the kidney and the liver and found that only 34 were expressed in both of them, and 213 of them were expressed only in the heart. We found that some lncRNA candidates were expressed in a stage-specific manner, suggesting that they may have roles in heart development or maturation.

Fig. 1
figure 1

The screening procedure of lncRNAs. a The flowchart for the identification of lncRNAs expressed in the mouse heart. b The histograms of expression levels (fpkm) at E10.5. The number of genes expressed at E10.5 is shown in each parenthesis. The expression levels of lncRNAs are generally lower and lncRNAs with tissue-specificity rarely exceed 10 fpkm (red circle). c The distances from the nearest protein-coding genes were calculated and the distributions were plotted based on RNA-type. We require genes to be expressed in the ventricle at a minimum of one stage. The number of genes in each category is shown in the parenthesis. Heart-selective genes were generally located at greater distances

Full size image

Many of the cardiac transcription factor genes have neighboring lncRNAs

First, we plotted the distribution of the expression levels of the obtained lncRNAs at E10.5 along with that of mRNAs. Consistent with the previous reports, the expression levels of lncRNAs were much lower than those of mRNAs. Interestingly, almost no heart-selective lncRNAs had fpkm values higher than 10 (Fig. 1b). Since many lncRNAs are known to modulate the transcription of neighboring genes in cis, we tried to identify the neighboring genes of the identified lncRNAs. The distribution of the distances from the transcriptional start site (TSS) of lncRNAs to the nearest genes was examined (Fig. 1c). Overall, the distance distribution of all obtained lncRNAs seemed to be similar to that of mRNAs. However, heart-selective lncRNAs were unexpectedly found to be at greater distances to protein-coding genes. The median distances were 12,626, 12,024 and 22,522 for mRNAs, all lncRNAs and heart-selective lncRNAs, respectively (p ≈ 3.9 * 10− 1 for all lncRNAs vs. mRNAs; and p ≈ 8.4 * 10− 8 for heart-selective lncRNAs vs. mRNAs, Mann-Whitney U test). Next, we examined what types of genes were enriched among the genes closest to lncRNAs. To this end, we conducted a gene enrichment analysis on such protein coding genes using the DAVID bioinformatics tool (http://david.ncifcrf.gov/) [26] and found that transcription factor genes were enriched among genes near lncRNAs in the heart. We also found that the genes associated with heart development were more strongly enriched among the genes near heart-selective lncRNAs when compared to the genes near lncRNAs lacking tissue specificity (Tables 1, 2).

Table 1 Gene ontology analysis of the genes closest to all lncRNAs that are expressed in the heart
Full size table
Table 2 Gene ontology analysis of the genes closest to heart-selective lncRNAs
Full size table

We next tried to identify antisense lncRNAs and lncRNAs from bidirectional promoters. Bidirectional promoters produce two transcripts in a head-to-head divergent manner and attract a lot of attention as important sources of lncRNAs. Preceding studies have revealed that many of them regulate the genes with which they share promoters. We evaluated lncRNAs that had their TSS within 3000 bp from the promoter of protein coding genes as lncRNAs driven by bidirectional promoters. Consistent with the result that the distance between heart-selective lncRNAs and their neighboring genes is generally greater than the distance between all lncRNAs and their neighboring genes, both antisense lncRNA and bidirectional lncRNA were enriched among lncRNAs that are expressed both in the heart and the brain (Fig. 2a) (Additional file 3: Table S2 and Additional file 4: Table S3). Some of the lncRNAs and neighboring genes were judged to be both antisense and bidirectional because of alternative promoter isoforms.

Fig. 2
figure 2

lncRNAs are enriched near genes that are important for heart development. a Classification of the lncRNA candidates found in the screen. Heart-selective lncRNAs are less likely to be bidirectional or antisense lncRNAs. Some lncRNAs were judged to be both antisense and bidirectional due to alternative promoter isoforms. b Distribution of the Pearson correlation coefficients between the bidirectional promoter pairs over the course of development. The arrow indicates a negative correlation and the arrowhead indicates a positive correlation. c Hand1 shows a correlated expression pattern with its bidirectional lncRNA (left), while Sall4 exhibits the opposite trend (right). d The proportion of genes that possess bidirectional lncRNAs in the mouse were examined by referring to RefSeq and a paper that identified haploinsufficient genes. Bidirectional lncRNAs were significantly enriched among haploinsufficient genes

Full size image

Next, in order to clarify the relationship between mRNAs and their bidirectional lncRNAs, we calculated Pearson correlation coefficients between the log2-transformed expression levels of the bidirectional promoter pairs over the course of development. The distribution of the correlation coefficients is plotted in Fig. 2b. Many gene pairs clearly show positive or negative correlation, and the positive correlation appears to be dominant (Fig. 2c).

By searching the protein coding genes that are close to lncRNAs, we found many transcription factor genes that have critical functions for heart development (i.e., Tbx5, Tbx20, Nkx2–5, Gata4, Gata6, Sall4, Hand1, Hand2, Wt1, Nr2f1, Irx3 and Irx5). Notably many of these lncRNAs were bidirectional lncRNAs (i.e., Tbx5, Tbx20, Nkx2–5, Gata6, Sall4, Hand1, Hand2, Wt1, Nr2f1, Irx3 and Irx5). Some of these lncRNAs (e.g., those divergent to Irx5, Gata6 and Wt1) are expressed in the kidney or in the liver, and in such cases divergent genes are also expressed, suggesting that the expression of bidirectional pairs are correlated not only temporally but spatially. We examined the conservation of these lncRNAs near transcription factors by searching the RefSeq database and found that at least some lncRNAs were conserved in the human genome (Tbx5, Nkx2–5, Hand2, Gata6, Wt1 and Nr2f1) (Additional file 5) and that the bidirectional lncRNA to Tbx5 (Lnc125) was even conserved in chicken, which diverged from mammals 400 million years ago. Here, we judged bidirectional lncRNAs to be conserved solely based on the existence of transcripts at the corresponding loci, since the sequences of lncRNAs are known to evolve rapidly.

Because haploinsufficient transcription factor genes seem to be highly enriched among the genes that are in close proximity to divergent lncRNAs, we determined whether the enrichment was limited to the heart or whether it was more generally true [27, 28]. Using the mouse RefSeq transcript database (GRCm38.p3) and a paper that comprehensively identified haploinsufficient genes, we tried to determine the proportion of genes with bidirectional lncRNAs among all genes and among haploinsufficient genes [29] (Additional file 6: Table S4). We indeed found that haploinsufficient genes were significantly more enriched among genes with bidirectional lncRNAs (p = 3.4 * 10− 5 based on hypergeometric distribution) (Fig. 2d). To exclude the possibility that the tissue specificity of bidirectional lncRNAs and haploinsufficient genes generates pseudo-correlations, we calculated the proportion of housekeeping genes among all genes and among haploinsufficient genes and showed that the proportions were not significantly different (Additional file 7) [30].

Generally, the conservation of lncRNAs across species is very low compared to protein-coding transcripts. However, the Tbx5-divergent lncRNA is observed among a wide range of species. Tbx5 is also a dosage-sensitive gene [27]. These findings prompted us to examine the function of the Tbx5-divergent lncRNA.

Analysis of the Tbx5-divergent lncRNA

Tbx5 is a transcription factor that is known to be essential for the development of the heart and forelimb. Holt-Oram syndrome is a dominant disorder caused by a single-allele mutation of TBX5 and is characterized by hypoplasia of the forelimb, abnormalities in the thumb, and atrial and/or ventricular septal defects [31,32,33]. Importantly, the phenotypes of Holt-Oram syndrome show a high degree of variance, indicating that the dose of TBX5 is crucial in normal heart development [34].

Hereafter we will call this lncRNA as Tbx5 upstream antisense product (Tbx5ua). Tbx5ua homolog is present in human genome annotation and it is named TBX5-AS1 (Additional file 5). When compared with humans TBX5-AS1 the sequence of Tbx5ua is relatively well conserved at the 5′ region, although it is hard to judge if this conservation is the consequence of functional demand since both the promoter and enhancer elements also exhibit a high degree of conservation (Additional file 8). Tbx5ua is transcribed from one of the promoters of Tbx5 in the opposite direction and overlaps with the intron of one of the Tbx5 isoforms (Fig. 3a, RefSeq: XM_006530282.3, isoform 1). RNA-seq data and RefSeq genomic annotation suggest that Tbx5ua is alternatively spliced, producing several isoforms (Fig. 3a, Additional file 9). In Fig. 3a, we labeled isoforms that were identified in our RNA-seq experiment in at least one stage. Reanalysis of previously published intact/nuclear RNA-seq of cardiomyocytes revealed that Tbx5ua is not clearly localized (Fig. 3b) [35]. Previous study reports that quite a few lncRNAs actually exhibit this type of non-localized expression patterns [36].

Fig. 3
figure 3

Different expression patterns of Tbx5ua and Tbx5. a RefSeq genome annotation of Tbx5 locus. The isoforms of Tbx5 and Tbx5ua that were identified in our RNA-seq analysis were labeled with isoform numbers. The qPCR primers used to quantify Tbx5 and Tbx5ua are indicated as arrowheads. b The log10-ratios of intact/nuclear RNA abundances. Gapdh and Neat1 serve as controls for cytoplasmic and nuclear localized RNA, respectively. c The expression levels of Tbx5 and Tbx5ua during development as determined by qRT-PCR. In the ventricle, the expression level of Tbx5ua is increased with the progression of development while that of Tbx5 is decreased (n = 3). d The expression levels of Tbx5 and Tbx5ua in the left and right ventricles at E11.5 as determined by qRT-PCR. Unlike Tbx5, Tbx5ua was equally expressed in both ventricles (n = 3, *: p < 0.05, Welch's t-test). e Schematic diagram of the Tbx5ua knockdown experiment. Three tandem copies of bovine growth hormone polyadenylation signal were inserted along with the neomycin resistance gene (NeoR) or EGFP. The selection markers were subsequently removed with cell-permeable Cre recombinase

Full size image

We first quantified the expression level of the transcript in the heart ventricle, atrium and forelimb during normal development by quantitative RT-PCR (Fig. 3c). We found that the expression level of Tbx5ua was increased in the ventricle as development progressed, which was inconsistent with the expression pattern of Tbx5. We also examined the expression level of the Tbx5 isoform that is also transcribed from the bidirectional promoter (Isoform 2, RefSeq: XM_006530280.1). The expression level of that isoform was stable during the entire developmental process, which was also different from the expression pattern of Tbx5ua (Additional file 10). Next, we compared the expression level of the lncRNA in both of the ventricles at E11.5 because it is well-known that the expression level of Tbx5 is higher in the left ventricle than in the right ventricle and that the steep gradient is crucial for establishing a proper ventricular septum [37, 38]. We observed that Tbx5ua expression was almost the same between the left and right ventricles at E11.5, while we confirmed the differential expression level of Tbx5 (Fig. 3d). These results suggest that Tbx5ua is not just a byproduct of Tbx5 and is regulated separately as a different product.

Tbx5ua-knockdown (KD) mice were embryonic lethal with severe abnormalities in the heart

To determine the function of Tbx5ua, we knocked down both alleles of Tbx5ua by inserting three tandem copies of bovine growth hormone polyadenylation site (3xpA) at the second exon to prematurely stop transcription in C57BL/6 J-derived ES cells using the CRISPR/Cas9 system (Fig. 3e) [39]. By tetraploid complementation, we obtained completely ES cell-derived mouse embryos from two ES cell lines and their phenotypes were consistent between lines. The expression level of Tbx5ua in E9.5 KD mice was strongly repressed to approximately 1/10 of that in control embryos, showing successful knockdown (Fig. 4a). Although the expression levels of Tbx5 and Tbx5ua seemed to be anticorrelated in the heart during development (Fig. 3c), KD of Tbx5ua did not result in the increase of Tbx5 expression level. We also showed that the expression levels of the different Tbx5 isoforms that are transcribed from all three promoters were not significantly changed (Additional file 11).

Fig. 4
figure 4

The phenotype of Tbx5ua KD chimeric mice. a qRT-PCR of Tbx5ua and Tbx5 in E9.5 chimeric mice that were derived from WT and KD ES cells (n = 4, * p < 0.05, Welch's t-test). Tbx5ua was successfully knocked down. b, c, d The morphology of E9.5 chimeric embryos. KD embryos show right ventricular hypoplasia (c). The ventricular walls of KD embryos appeared abnormally thin (d). e, f The body size and forelimb of KD embryos appeared to be normal in E13.5 chimeric mice, whereas KD embryos showed severe heart defects, including a hypoplastic right ventricle

Full size image

Chimeric KD embryos showed right ventricular hypoplasia at E9.5 (Fig. 4b, c, Additional file 12A). Hematoxylin and eosin (HE) staining of the cryosections showed that the ventricular walls of E9.5 KD mice were irregular and lacked trabeculae at some parts in the ventricle (Fig. 4d, Additional files 12B and 13). None of the embryos showed a visible abnormality in the forelimbs, which is observed in Tbx5-deficient embryos. By E13.5, all of the KD embryos were dead with a pale body (Fig. 4e). The hearts showed severe ventricular hypoplasia (Fig. 4f), which was probably the cause of the lethality. The forelimbs seemed completely normal even at this stage, which was a significant difference between the phenotype of the Tbx5ua KD mice and that of the mouse model of Holt-Oram syndrome (i.e., Tbx5 heterozygous knockout) (Fig. 4e). The phenotypes among KD embryos were similar and heart-specific, suggesting that they are attributed to genomic modification. In situ hybridization of Tbx5 revealed normal mRNA expression in the KD ventricle (Additional file 14A). In situ hybridization of Nppa, which often shows altered expression pattern in embryos with abnormal morphogenesis, showed an expanded expression around the pre-ventricular septal region of KD embryos (Additional file 14B).

To comprehensively investigate the genes affected by Tbx5ua knockdown, we performed RNA-seq with the RNAs extracted from the ventricles of tetraploid chimeric embryos derived from either KD or WT ES cells. We used three embryos for each group and used the Smart-Seq2 protocol to generate libraries from the small amount of RNA. By gene ontology analysis, we found that the genes involved in heart development were significantly enriched among the genes that were determined to be significantly changed (False Discovery Rate; FDR < 0.10, Additional file 15A). However, none of the structural genes that are important for cardiomyocyte contraction were changed (Additional file 15C), suggesting the possibility that Tbx5ua has a critical role in morphogenesis rather than in cell differentiation. Finally, we conducted principal component analysis (PCA) on the RNA-seq data (Additional file 15D). The two groups were evidently distinguished only by considering the first principal component.

Discussion

In this study, we found that many cardiac transcription factor genes have neighboring spliced lncRNAs, especially bidirectional ones. The clear correlation of the expression levels of some bidirectional pairs suggests their regulatory roles. A typical example of such lncRNAs is Upperhand, which is divergent to Hand2 [18]. The transcription of Upperhand but not the mature transcripts were shown to be necessary for the transcription of Hand2 by altering the local epigenetic environment. Many lncRNAs are known to regulate local transcription through the recruitment of epigenetic-altering protein complexes. Since the expression level of transcription factors is generally low and many of them are haploinsufficient, even a relatively small fluctuation could lead to severe consequences. It is possible that divergent lncRNAs are enriched among dose-sensitive genes to stabilize the expression level of adjacent genes.

An alternative hypothesis is that these transcription factors could be setting up optimal transcriptional environments for lncRNAs to evolve. As some transcription factor genes can cause direct lineage reprogramming, they are thought to define cell types. Thus, the use of these preexisting transcriptional environments is a cost-efficient way to evolve cell type-specific lncRNAs. Some studies have demonstrated that bidirectional transcription is a general phenomenon and that a so-called transcription ripple effect exists [40, 41]. These findings also support our idea by showing that the preexisting transcriptional environment enables precursor transcripts to evolve into defined, functional ones. In summary, active transcription factor genes may have been good sources from which lncRNA genes could evolve due to the cell lineage-specific and active epigenetic environment.

We showed that Tbx5ua is conserved from mammals to birds. Comparison of the sequence of Tbx5ua between mouse and chicken showed less similarity, but it does not mean that the function is not conserved as the previous studies have shown that precise conservation at the sequence level is not necessarily required for the functional conservation of lncRNAs [42, 43]. Tbx5ua was not found in the NCBI genomic annotations of reptiles, amphibians or fish at the corresponding loci. In fact, by conducting the reanalysis on the publicly available RNA-seq data, including RNA-seq of the adult heart of chicken, anole and frog (GSE41338) [44], we could confirm that Tbx5ua is expressed only in chicken among these species at the adult stages (Additional file 16). It is interesting that Tbx5ua is conserved in two-ventricle animals, which possess a complete ventricular septum, but not in animals with non-septated hearts. There is a possibility that the acquisition of Tbx5ua might have contributed to the evolution of a complete ventricular septum.

We showed that Tbx5ua lncRNA is required for proper heart development. Since we knocked down Tbx5ua by prematurely terminating the transcription, the loading of transcription complex at the transcription start site is not inhibited. Thus, if the transcription of Tbx5ua itself is important for altering the local transcriptional environment, our KD scheme is not sufficient to assess the true function of Tbx5ua. Although preliminary, our data suggested that the expression pattern of Tbx5 protein is altered in the KD mice (Additional file 17). While we do not have any evidences supporting the direct roles of Tbx5ua on Tbx5, the function of Tbx5ua might be atypical for a divergent lncRNA since many of such lncRNAs like Upperhand are shown to alter the transcription of neighboring genes. How the left-sided expression of Tbx5 is regulated is an unsolved important issue to understand the molecular mechanism of heart development [45].

Conclusions

This study revealed that many genes involved in the heart development, particularly transcription factor genes, are associated with spliced lncRNAs that are derived from nearby genomic regions. Furthermore, many of these lncRNAs were divergently transcribed from the promoter of protein-coding genes. We find that bidirectional lncRNAs are enriched among haploinsufficient genes, suggesting that they have functional roles for the regulation of dose-sensitive genes.

Methods

RNA-seq

Total RNAs from embryonic and adult mice were extracted with Sepasol-RNA I Super G (Nacalai #09379–55). The cDNA libraries for paired-end RNA-seq for the screening of lncRNAs were prepared from 1 μg of RNAs with Truseq Stranded Total RNA Library Prep Kit (Illumina #RS-122-2201) according to Illumina’s instructions. The cDNA libraries for tetraploid chimeric mice were prepared by Smart-Seq2 protocol according to the original paper [46] with 12 cycles of preamplification and 9 cycles of enrichment PCR.

qRT-PCR

Total RNAs were extracted with Sepasol-RNA I Super G (Nacalai #09379–55). cDNA samples were prepared using RevaTra Ace qPCR RT Master Mix with gDNA remover (Toyobo #FSQ-301). Real-time PCR was performed with SYBR Premix EX Taq II (Takara #RR820). The PCR conditions were as follows: 95 °C for 30 s followed by 50 cycles of 95 °C for 5 s and 60 °C for 30 s, and a subsequent dissociation curve measurement. We used Gapdh as an internal control. Gene-specific primers are listed in the Additional file 18: Materials and Methods.

Mice

C57BL/6 J mice were purchased from CLEA Japan. For the first round of chimeric mice generation, we used mice form CLEA Japan. For the second experiment, we used tetraploid embryos from Ark Resource and recipient mice from Sankyo Labo Service. Mice are sacrificed by cervical dislocation.

Generation of genome edited ES cells

ES cells were cultured on the MEF feeder in ES culture medium (i.e., Knockout DMEM (Gibco #10829018), 20% Knockout Serum Replacement (Gibco #10828028), 1 * GlutaMAX (Gibco #10566016), 1 * NEAA (Sigma #M7145), 1 mM sodium pyruvate (Gibco # 11360070), 10− 4 M 2-Mercaptoethanol, 1000 U/ml LiF (Wako #198–15,781)).

The guide RNA (gRNA) target sequence to induce double strand break is 5’-GTCACTGCCGCTCCAATCCTCGG-3′. We designed the gRNA with Cas-OFFinder (http://www.rgenome.net/cas-offinder/), so that the number of off-target sites was as few as possible. Our gRNA has no potential off-target sites with 0, 1 or 2 mismatches and just 4 potential off-targets with 3 mismatches and proper PAM sequence, of them none is exonic. Homology directed repair donors were constructed so that the NeoR or EGFP expressing cassette was flanked by ~ 1,000 bp 5′ and 3′ homologous arms cloned from genomic DNA. ES cells were transfected with Cas9 expressing plasmid, gRNA expressing plasmid and the donor plasmid along with non-gRNA expressing negative control. After two days, the ES cells were passaged onto SNL feeder cells and cultured for 8 days with 250 μg/ml G418 and surviving EGFP-positive colonies were manually picked up. After one more cycle of single colony picking up to ensure that the ES cells are clonal, they were subjected to cell permeable Cre treatment [47] to remove the selection cassettes, and then EGFP-negative colonies were picked up to obtain cells without selection cassettes. Finally, the ES cells from each colony were genotyped and karyotyped.

Generation of tetraploid chimeric mice

Generation of chimeric mice was performed as described previously [39].

Histology

Immunohistochemistry for Tbx5 were performed as follows. Antigen retrieval was performed by microwaving the sections in 10 mM citrate acid pH 6.0. Then they were permeabilized for 10 min in 0.2% Triton X in PBS at RT. Blocking was performed with 10% Blocking One (Nacalai #03953–95) in PBST. Tbx5 antibody (Santa Cruz Biotechnology #sc-17,866) was diluted 1/100 in 5% Blocking One/PBT and second antibody (Invitrogen #A-11037) was diluted 1/200.

In situ hybridization was performed as follows. First, cryosections were permeabilized in 0.2 N HCl for 15 min. After washing with PBT three times, the sections were re-fixed with freshly made 4% PFA for 15 min. After washing, the sections were hybridized with DIG labeled probes at 70 °C for ON. The next day, the sections were washed with 0.2× SSC three times. After blocking the sections with 10% sheep serum for 1 h, 1/1000 diluted anti-DIG-AP Fab fragment (Roche #11093274910) were added and incubated for an hour at RT. After washing with TBST, the sections were washed with NTMT and colored with BM purple.