The second pharyngeal pouch is generated by dynamic remodeling of endodermal epithelium in zebrafish

Elaborate morphological features of organisms can often be explained by simple metameric motifs that are transiently established during vertebrate development (Carroll, 1995; Hannibal and Patel, 2013). For instance, conspicuous segmental structures called pharyngeal arches (PAs) are bilaterally arranged in the ventral region of the head (Fig. 1A) (Graham, 2001; Graham and Richardson, 2012). PAs give rise to the segmental organization of skeletons, muscles, nerves and vessels in the pharynx; therefore, specification and segmentation of PAs are crucial for development of the vertebrate head (Graham and Smith, 2001). The most anterior PA, referred to as PA1 or the mandibular arch (MA), forms the jaw. The second anterior PA, PA2, also called the hyoid arch (HA), generates the hyoid. In addition, the more posterior PAs, known as the branchial arches (BAs), give rise to many pharyngeal derivatives, including the thymus and parathyroid glands. Tripartite streams of cranial neural crest cells (CNCCs), referred to as the trigeminal, hyoid and branchial streams, contribute to PA1, PA2 and the BAs, respectively (Graham et al., 2004; Hunt et al., 1991; Lumsden et al., 1991; Schilling and Kimmel, 1994; Serbedzija et al., 1992). However, segmentation of PA units occurs independently of these CNCCs (Gavalas et al., 2001; Veitch et al., 1999). Rather, pharyngeal endoderm contributes to this segmentation by generating epithelial outpocketings called pharyngeal pouches (PPs), which physically define the anterior and posterior interfaces of each PA (Graham and Richardson, 2012; Graham and Smith, 2001). Each PP lies flat along the antero-posterior axis and consists of two layers of epithelial sheet.

The two most anterior PAs (MA and HA) are likely to have different characteristics from the posterior PAs (BAs). The anterior PAs express genes involved in myogenesis and chondrogenesis, whereas the posterior PAs do not undergo myogenesis and chondrogenesis in chick or mouse embryos (Poopalasundaram et al., 2019). Several other genes also exhibit anterior- and posterior-specific expression in pharyngeal ectoderm. For instance, the expression of sonic hedgehog (Shh) is excluded or highly reduced in posterior pouches in chicks and mice, and Shh signaling prevents Gcm2 expression in the first and second pharyngeal endoderm (Ankamreddy et al., 2019; Grevellec et al., 2011). In addition, expression of several Dlx genes is specific for anterior PAs in chick and mouse embryos (Poopalasundaram et al., 2019). Interestingly, the anterior PAs are segmented simultaneously, whereas the posterior PAs (BAs) are generated sequentially in anterior to posterior order (Crump et al., 2004a; Graham et al., 2014; Veitch et al., 1999), suggesting discrete regulation of anterior and posterior PA segmentation. Correspondingly, retinoic acid (RA) deficiency in zebrafish (Kopinke et al., 2006), quail (Quinlan et al., 2002), rat (White et al., 1998) and mouse (Wendling et al., 2000) embryos consistently results in abnormalities in segmentation of the third and more posterior PAs. Similarly, pax1 knockout medaka (Okada et al., 2016) and Ripply3 knockout mice (Okubo et al., 2011) show loss of BAs, but normal MAs and HAs. These studies suggest that distinct segmentation mechanisms for anterior and posterior PAs might cooperate to establish the entire series of PAs. However, development of this complex PA segmentation has been poorly understood. Especially puzzling has been the question of how the seamless array of PAs is generated from anterior and posterior PAs, which are formed by different mechanisms.

Recent studies of zebrafish PP development have revealed the dynamic cellular nature of the endoderm in forming PPs (Choe et al., 2013; Choe and Crump, 2014, 2015a,b), highlighting the advantages of the zebrafish model to dissect processes of PP development. In this study, we examined development of zebrafish pharyngeal endoderm, especially focusing on the formation of PP2, which is located at the boundary between the anterior and posterior PAs. Precise examination by live-imaging and cell-tracing experiments performed in zebrafish showed that PP2 is formed in an unexpected manner. Rostral and caudal edges of PP2 are initially formed separately, then subsequently establish contact through dynamic remodeling of endoderm epithelium, and finally form a mature PP bilayer.

To better understand the development of pharyngeal endoderm, we performed time-lapse imaging of endodermal cells in transgenic zebrafish, Tg(sox17:EGFP), in which EGFP expression was specifically driven in endodermal cells by the sox17 promoter (Mizoguchi et al., 2008). PP1 appeared at 16 h post-fertilization (hpf; Fig. 1B-D, Movie 1), whereas PP outpocketings, including PP3 and more posterior PPs, were generated sequentially from anterior to posterior after 16 hpf (Fig. 1E-K, Movie 2). Unexpectedly, we found that two endodermal bulges appeared at ∼16 hpf, with PP1 budding in the area where PP2 would be generated (Fig. 1D, Movie 1). Notably, these bulges were gradually remodeled and finally merged to form a bilayered PP2 (Fig. 1H-O, Movie 2). This remodeling occurred not prior to, but during, sequential generation of posterior PPs, and the two bulges were finally merged at the time when PP6 was formed (Fig. 1K). These findings strongly suggest that mature, bilayered PP2 is formed approximately after the formation of most posterior PPs (Fig. 1H-O, Movie 2).

To understand the dynamics of the endodermal bulges, we performed a lineage-tracing experiment by means of endoderm-specific photoconversion. To this end, we created Tg(sox17:Kaede), a transgenic line harboring Kaede expression under control of the sox17 promoter. Photoconversion of cells in the rostral bulge at 20 hpf revealed that these cells contributed to the rostral layer of PP2 at 48 hpf (Fig. 2A-E and Fig. S1). Descendants of these cells spread extensively to form the inner lining of the distal part of the HA, where an opercular flap would later expand (Fig. 2A-E and Fig. S1). By contrast, descendants from the caudal bulge became distributed in the caudal layer of PP2 at 48 hpf, especially its proximal region (Fig. 2F-J and Fig. S1). In addition, cells in the intermediate region between the rostral and caudal bulges contributed to the more distal and ventral regions of the caudal layer and the dorsal edge of PP2 (Fig. 2K-N and Fig. S1). Based on patterns of cell traces (n=29, Fig. 2 and Fig. S1), we outlined fates of endodermal cells in the future PP2 region at 20 hpf (Fig. 2O). In summary, PP2 was generated by dynamic remodeling of endodermal cells between the rostral and the caudal bulges, which directly contributed to respective rostral and caudal layers of PP2.

Our time-lapse observations and cell-tracing experiments revealed that PP2 arose from two distinct endodermal domains, which then coalesced into a pouch structure via dynamic remodeling of the endoderm. To understand development of PP2 more precisely, we examined expression of PP-specific genes in the future PP2 endoderm. First, we examined expression of nkx2.3, which is observed in both rostral and caudal compartments of all zebrafish PPs (Lee et al., 1996). Expression of nkx2.3 was initially evident at 20 hpf. At this stage, nkx2.3 expression was detected not only in PP1 but also in the rostral and caudal bulges of PP2 (Fig. 3A-D and Fig. S2). Whereas nkx2.3 expression in these areas appeared simultaneously (Fig. S2A), nkx2.3 expression corresponding to posterior PPs (PP3 and PP4) appeared gradually with sequential segmentation of the posterior PPs (Fig. S2B,C). Thus, nkx2.3 expression was detected in a temporal sequence similar to the progression of PA morphogenesis observed by time-lapse imaging (Fig. 1B-K, Movie 1). Consistent with PP2 maturation, nkx2.3-positive rostral and caudal bulges gradually converged to form PP2 (Fig. S2A-C). We also examined expression of pax1b, which is expressed in PPs of many vertebrates. Like nkx2.3, pax1b was expressed in all PPs. Furthermore, its expression was detected in the future PP2 ectoderm, including in the rostral and caudal bulges (Fig. S2D). These expression patterns of PP-specific genes suggest that at least some characteristics of PP2 were already present in the future PP2 ectoderm, including in the bulges. Furthermore, the separate expression of nkx2.3 in the two distinct bulges supports results obtained by time-lapse observations and cell-tracing experiments, showing that PP2 was formed by complex remodeling of the two distinct endodermal domains.

To further understand development of the future PP2 endoderm, we next examined rostral and caudal layer-specific molecular characteristics. We found that expression of tbx1 was specific to the rostral layer of each PP in normal embryos up to 30 hpf, whereas that of fgf3 was specific to the caudal layer of each PP (Fig. S2E-H). Interestingly, even at the beginning of PP2 development, expression of tbx1 was strongly detected in the rostral bulge, but not in the caudal bulge (Fig. 3E-H), and that of fgf3 was the opposite (Fig. 3E,F,I,J). This result strongly suggests that the rostral and caudal bulges have already acquired distinct rostrocaudal characteristics prior to epithelial remodeling to form PP2. In other words, immature PP2 first appeared as a wide and abnormally shaped epithelial sheet, including both rostral and caudal domains, and is subsequently remodeled into bilayered epithelium. We refer to these rostral and caudal bulges as R2 and C2, respectively, hereafter (Fig. 3K).

Because molecular characteristics of the rostral and caudal layers of PP2 were observed in the endodermal domains of R2 and C2, respectively, we next investigated whether rostral and caudal identities had actually been determined in R2 and C2. We specifically ablated endodermal cells in the R2 or C2 region using infrared laser-mediated heating (Kamei et al., 2009; Zeng et al., 2016). To visualize endodermal cells, we used Tg(sox17:EGFP) embryos in this laser-ablation experiment.

Consistent with results from cell-lineage tracing, ablation of R2 cells at 20 hpf (Fig. 4A,A′) impaired expansion of the rostral region of PP2 at 48 hpf (n=3/3, Fig. 4B,C). In later stages, R2 ablation resulted in loss of HA-derived dermal bones of branchiostegal rays (BR; n=12/16) and the opercular (OP; n=10/16), which together compose the operculum (Fig. 4D-F). This ablation occasionally reduced the size of other HA-derived skeletal structures, such as the hyomandibular (HM; n=3/16) and ceratohyal (CH; n=6/16) cartilages, but ceratobranchial (CB) cartilages, which are derived from BAs, were completely normal (n=16/16, Fig. 4D, Fig. S3C). For operculum development, Shh expression is required in the HA (Richardson et al., 2012). We found that shha was expressed in endoderm corresponding to R2-derived cells (Fig. 4G,H), and that this expression was reduced (n=5/12) or eliminated (n=5/12) by R2 ablation (Fig. 4I), suggesting that the R2 region gives rise to a signaling center of Shh for operculum formation. In contrast to R2, ablation of cells in the C2 region (Fig. 4J,J′) caused abnormalities in the proximo-caudal PP2 adjacent to BA1 (or PA3; n=3/3, Fig. 4K,L), resulting in specific loss of the first CB cartilage (CB1; n=8/8, Fig. 4M). However, ablation of cells between R2 and C2 (Fig. 4N,N′) did not cause any loss of the pharyngeal skeleton, although the position of CB1 on the ablation side shifted posteriorly and laterally (n=4/6, Fig. 4Q). Interestingly, this ablation caused a split of endoderm between HA and BA1 (n=10/12, Fig. 4O,P), which were almost normally formed, showing the requirement of intermediate endoderm for integration of HA and BA1. Loss of skeletal elements does not appear to result from deficits in CNCCs caused by infrared irradiation of the endodermal cells, because PA mesenchymal cells and expression of dlx2a, which is a credible marker of CNCCs in PAs, were not affected by ablation of adjacent endoderm (n=6, Fig. S3A,B). Therefore, we conclude that endodermal cells of R2, C2 and the intermediate region serve distinct functions in craniofacial development in zebrafish. Significantly, those functions were determined in endodermal domains prior to remodeling for morphological maturation of PP2.

Previous studies suggested that development of anterior and posterior PAs appears to occur via distinct molecular mechanisms (Kopinke et al., 2006; Okada et al., 2016; Okubo et al., 2011; Quinlan et al., 2002; Wendling et al., 2000; White et al., 1998). These studies suggest that a mechanistic boundary between them might exist around PP2. Because our findings enabled us to dissect developmental processes of PP2 into rostral and caudal events, we next addressed the issue of how separate molecular mechanisms could coordinate to produce PP2.

As RA signaling is specifically required for development of the posterior PPs (Kopinke et al., 2006; Quinlan et al., 2002; Wendling et al., 2000; White et al., 1998), we supposed that the R2–C2 boundary would correspond to the anterior border of RA function. Visualization of RA activity utilizing a transgenic fish Tg(RARE:Venus), which harbors a Venus reporter driven by RA-responsive elements (RARE) (Retnoaji et al., 2014), showed that Venus was expressed in C2 cells at 20 hpf and persisted in their descendants, as well as in posterior pharyngeal endoderm (Fig. 5A-F). By contrast, Venus expression was never detected in R2 and PP1 endoderm during our examination (Fig. 5A-F). Furthermore, treatment with N,N-diethylaminobenzaldehyde (DEAB), which inhibits RA biosynthesis, impaired C2 formation and fgf3 expression in C2 (Fig. 5K-N). By contrast, this treatment did not cause an obvious change in tbx1 expression in R2 (Fig. 5G-J). Thus, RA signaling was specifically activated in and required for pharyngeal endoderm posterior to R2, indicating that the anterior border of RA function actually corresponded to the R2–C2 border we had identified.

Our previous study also showed that pax1 is specifically required for the development of posterior PPs in medaka (Okada et al., 2016). Zebrafish has two pax1 homologs, pax1a and pax1b; therefore, we generated double-knockout (DKO) mutants of these genes (referred to as pax1 DKO) by performing CRISPR/Cas9-mediated mutagenesis (Fig. S4). As expected, pax1 DKO embryos showed definite abnormalities in development of their PPs posterior to C2, but not in R2 (Fig. 6A-D). Consistently, gill skeletons, but not opercular skeletons, were lost in pax1 DKO larvae (Fig. 6E-H). The anterior part of the HM (aHM), in which PP1 is required developmentally (Crump et al., 2004b), was lost in pax1 DKO larvae (Fig. 6E-H), although PP1 was normally formed in mutant embryos (Fig. 6A-D). Thus, pax1a/b genes appear to participate in aHM development at a later stage after PP1 formation, but this role apparently differs from their role in PP development posterior to C2.

The membrane protein, Alcam, accumulates in PP epithelium to stabilize bilayered PP morphology (Choe et al., 2013). Similar to RA and pax1a/b, we found that its accumulation was low in PP1 and R2, but high in C2 and more posterior PPs (Fig. 7). This result suggests that the R2–C2 border might also have separated the morphogenetic process of endodermal epithelial cells. Taken together, we conclude that two distinct developmental processes occur in pharyngeal endoderm either anterior or posterior to the R2–C2 border, and that these processes subsequently merge to form zebrafish PP2.

Generally, segmentation in animal development occurs in a simultaneous manner, as in Drosophila germ-band formation (Davis and Patel, 2002), or in a sequential manner, as represented by vertebrate somitogenesis (Pourquié, 2011). Interestingly, PA segmentation is thought to be achieved by a peculiar combination of these distinct styles. The anterior PAs, i.e. MA and HA, are segmented simultaneously, whereas the posterior PAs (BAs) are generated sequentially in anterior to posterior order (Crump et al., 2004a; Graham et al., 2014; Veitch et al., 1999). However, a developmental process employing a combined segmentation system has never been reported. Zebrafish PA development, in which dynamic epithelial remodeling of the endoderm occurs, is a good model to resolve development of the vertebrate head (Choe et al., 2013; Choe and Crump, 2014, 2015a,b). In this study, by using precise live imaging in zebrafish, we found that the bilayered morphology of PP2, the border between the anterior and posterior PAs, was established by dynamic remodeling of the endoderm after generation of both anterior and posterior PAs. Given that anterior and posterior PAs are formed independently, this endodermal remodeling appears to be essential to merge anterior and posterior PAs, resulting in formation of a seamless array of PAs in zebrafish.

Our cell-tracing experiments in endoderm clearly revealed that the two endoderm bulges R2 and C2, which appear in the future PP2 endoderm region, directly contribute to the rostral and caudal layers of PP2, respectively. The rostral or caudal identity of each bulge had already been determined prior to PP2 formation, as evidenced by gene expression and cell-ablation experiments. However, intermediate endodermal cells between R2 and C2 did not contribute to the formation of PA-derived skeletons, although they were required for a tight arrangement of anterior and posterior PAs. These results suggest that the identity of rostral or caudal layers of PP is specified prior to formation of bilayered pouch morphology, at least in the case of PP2.

Based on our results, we propose a novel mode of vertebrate PA development. That is, HA and BAs are established independently by distinct developmental mechanisms. Accordingly, posterior PA-specific defects and PP2 insufficiency, which are commonly reported as phenotypes of RA-deficient vertebrates (Kopinke et al., 2006; Quinlan et al., 2002; Wendling et al., 2000; White et al., 1998), are more reasonably explained as defects occurring posterior to C2. Similarly, PA development in pax1 mutants of teleost fish (Okada et al., 2016) and Ripply3 mutant mice (Okubo et al., 2011) appears to be impaired posterior to C2 endoderm. By contrast, tbx1 is required not only for posterior PA development, but also for HA development (Jerome and Papaioannou, 2001). However, tbx1 expression is independent of pax1 function in HA endoderm, but not in BA endoderm (Okada et al., 2016). Thus, the regulatory system of tbx1 expression is spatially consistent with our idea, and suggests that there are distinct gene regulatory networks between anterior and posterior PAs. It is also an interesting question as to whether Hox genes are involved in the anterior- or posterior-specific mechanism in endoderm development. In chick embryos, Hoxa3 expression transiently extends through PP2, whereas mouse Hoxa3 is expressed in PP3 and PP4, but not PP2 (Gordon, 2018). In zebrafish, hoxa3a and hoxb3a are expressed in posterior PAs (Hogan et al., 2004), but it is uncertain whether Hox paralogs are involved in the regulatory network specific to posterior PAs.

What is the significance of the specifically modified PP morphogenesis at PP2, in terms of morphological evolution of the vertebrate head? We suppose that this modification probably contributed to evolution of the opercular system, which is conserved in osteichthyans (Richardson et al., 2012). A recent study on a fossilized placoderm, Entelognathus, which has opercular and BRs, suggests that the osteichthyan-like pharyngeal system existed in stem gnathostomes (Zhu et al., 2013). Furthermore, it has also been suggested that the chondrichthyan affinity of acanthodians, which possess a hyoidean gill cover with BRs, implies unique evolution of the chondrichthyan pharyngeal system composed of septal gills (Brazeau and de Winter, 2015; Brazeau and Friedman, 2015; Janvier, 1996). Although skeletal elements of the operculum have been lost during tetrapod evolution, the embryonic opercular flap, which is derived from the Shh-expressing HA, encloses the posterior pharyngeal region during amniote development (Graham and Richardson, 2012; Richardson et al., 2012). In this study, we found that R2 endoderm contributes directly to the operculum, including Shh-expressing cells in the zebrafish HA. Thus, we propose that HA development, distinct from that of BAs, must have been acquired by stem gnathostomes as the crucial basis for the novel pharyngeal system of the hyoidean operculum, leading to extant osteichthyans.

It is still unclear how the development of PP1 and the rostral layer of PP2 is regulated. The independence of PP1 development from that of other PPs has been suggested by previous studies on zebrafish (Crump et al., 2004a,b; Talbot et al., 2012) and mouse (Jerome and Papaioannou, 2001; Xu et al., 2005) development. Interestingly, in amphioxus, formation of the first gill slit is less affected by pax1/9 knockdown than that of other gill slits, which exhibit severe defects (Liu et al., 2015). This finding implies evolutionarily conserved independence of the first endodermal bulge from others. Therefore, development of vertebrate PPs, especially in gnathostomes, could be considered as involving three sections: PP1, the rostral layer of PP2 and a more posterior region. Significantly, these endodermal sections correspond to the interfaces of the three streams of CNCCs comprising the MA, HA and BAs. Therefore, we propose that modifications in endodermal segmentation reinforced topological restrictions of neural crest cell streams in the PAs. Further studies on the development of PPs could answer one of the biggest questions regarding development of the vertebrate head, the logic for coordination between pre-patterned neural crest cells and endodermal segmentation.

Zebrafish with the TL2 background were used as the wild type, as described previously (Kishimoto et al., 2004). Collected embryos were incubated at 28°C. Embryos that would be fixed later than 25 hpf were treated with 0.003% 1-Phenyl-2-thiourea (PTU) from 10 hpf until fixation to inhibit melanin synthesis. This study was performed in accordance with Guidelines for Animal Experimentation of the National Institutes of Natural Sciences, with approval from its Institutional Animal Care and Use Committee.

Tg(sox17:EGFP) and Tg(RARE:Venus) were used in this study. For generation of Tg(sox17:Kaede), the Kaede complementary DNA fragment from pKaede-S1 (MBL) and the same promoter sequence as used for the Tg(sox17:EGFP) were combined and cloned into the pSK-tol2B vector (Yabe and Takada, 2012). Transgenesis was performed using the Tol2 system (Kawakami et al., 2004). For CRISPR/Cas9-mediated knockout, target sequences were determined using the ZiFiT Targeter (Sander et al., 2010). Construction of guide RNA vectors and preparation of single-guide RNA (sgRNA) and Cas9 mRNA were performed as described previously (Cong et al., 2013; Hwang et al., 2013). Mutation efficiency was assessed by performing a T7 endonuclease assay (Chen et al., 2012) with the following primers: 5′-TTGATTTAGGTCATGTGTGTTATATG-3′, 5′-TTTGTTTGTAGTCCCGTATGTTTTT-3′ for pax1a and 5′-GTTTTTCTGACAATGCAAAAAGTG-3′, 5′-CGTATTTCCCAAGCAAATATCC-3′ for pax1b. Details on pax1a and pax1b mutagenesis and sequences of sgRNAs are provided in Fig. S4. For microinjections, 1 nl of each injection solution (Tol2 transgenesis: 25 ng/μl plasmid, 50 ng/μl Tol2 mRNA, 0.2 M KCl and 0.05% Phenol Red; CRISPR/Cas9: 25 ng/μl sgRNA, 100 ng/μl Cas9 mRNA, 0.2 M KCl and 0.05% Phenol Red) were injected into one-cell-stage zebrafish embryos using an IM300 micro injector (Narishige).

Living embryos of Tg(sox17:EGFP), from which chorions had been manually removed, were anesthetized with 0.02% ethyl-3-aminobenzoate methanesulfonate (MS-222) in 1/3 zebrafish Ringer's solution. For observations with a Leica SP8, embryos were moved to mounting medium (0.15% low-melting-point agarose, 0.02% MS-222, 0.003% PTU in 1/3 zebrafish Ringer's solution) and individually set in the medium on a glass-bottomed dish. Embryos were manually positioned using a tungsten needle with an eyelash on its tip. Z-stack images were taken at 10-min intervals, and stack images were processed with an LAS X (Leica) to make 3D images, optical sections and movies. For imaging with a Zeiss Lightsheet Z.1, anesthetized embryos were mounted as previously described (Weber et al., 2014). Images taken at 10-min intervals were processed with ZEN Black (Zeiss) and subsequently with Imaris (Bitplane) to make movies.

Embryos of Tg(sox17:Kaede) for photoconversion or Tg(sox17:EGFP) for cell ablation were mounted as described above. Photoconversion was performed with a Leica SP8 using a 405-nm diode laser. Regions of interest (ROIs) in Kaede-expressing endoderm at 20 hpf were converted using the ROI tool in LAS X (Leica). Converted embryos were released from the gel and incubated at 28°C, and were observed again at 48 hpf. Cell ablation with an infrared laser-evoked gene operator (IR-LEGO) system (Sigmakoki IR-LEGO 1000, Sigma Koki) was performed as previously described (Zeng et al., 2016). High-power flash irradiation with an infrared laser (80 mW for 8 ms) was performed several times until EGFP signals of target regions had been eliminated. Ablated embryos were released from the gel and incubated at 28°C until subsequent experiments could be performed. For cell ablation controls, Tg(sox17:EGFP) embryos were injected with mRNA of histone H2A-mCherry at the one-cell stage to visualize the cell nucleus in the live condition. At 20 hpf, embryos were scanned with a Nikon A1 before ablation and moved to the IR-LEGO. After infrared irradiation on the IR-LEGO, embryos were immediately moved to the Nikon A1 and scanned again to evaluate off-target damage to cells of adjacent PAs. This procedure was repeated, and ablated embryos were fixed in 4% paraformaldehyde and stored in methanol at −20°C for in situ hybridization with a dlx2a probe, and for immunohistochemistry with anti-GFP antibody, to assess CNCCs in the PA.

Whole-mount in situ hybridization of zebrafish was performed as described previously (Jowett, 2001). For double-fluorescence in situ hybridization experiments, anti-DIG-POD (Roche) and anti-FITC-POD (Roche) were used to detect each hapten in RNA probes. Fluorescent signals were detected with a TSA Plus Cy3/fluorescein system (PerkinElmer). Plasmids for probes of dlx2a, nkx2.3 and shha (Krauss et al., 1993) were kindly donated by Drs M. Hibi, Y. Kikuchi and S. Krauss, respectively. Primers for cloning other probes of zebrafish genes were as follows: 5′-ATGCTTTCGTGTTTTGCAGAGCAAACATAC-3′, 5′-TTACGAGGATGAGGTAGAAAGGCTGAGTCC-3′ for pax1a; 5′-ATGCAAATGGATCAGACGTACGGGGAGGTG-3′, 5′-TTATGAGTCTGAGAGTCCATGAACAGCGCT-3′ for pax1b; 5′-ATGATTTCAGCAATATCAAGCCCGTGGCTG-3′, 5′-TTATCTGGGTCCGTAGTCATAATTAGTCGG-3′ for tbx1; and 5′-AAATCTCACGATAGGCTCCCTG-3′, 5′-AAAGTACTCCTGATTGCAGT-3′ for fgf3. Immunostaining was conducted as previously reported (Yabe et al., 2016) using primary antibodies anti-Alcam (1:500; Developmental Studies Hybridoma Bank, University of Iowa, USA), anti-Kaede (1:400; MBL, PM012), anti-GFP (1:400; Abcam, ab13970), followed by Alexa Fluor-conjugated secondary antibodies (Invitrogen). Bone and cartilage staining of zebrafish larvae was performed as previously described (Walker and Kimmel, 2007). Rehydrated embryos were treated with RNaseA and subsequently stained overnight at room temperature with YOYO1-Iodide in PBST (1:2000; Thermo Fisher Scientific). Stained embryos were rinsed with PBST several times, dehydrated with methanol and soaked in benzyl alcohol/benzyl benzoate (1:2 ratio) prior to confocal imaging.

DEAB stock, which was stored at −20°C, was prepared at 100 mM in dimethyl sulfoxide (DMSO). DEAB treatment was conducted at final concentrations of 10⁻⁴ M for zebrafish (from 10 hpf to 20 hpf), as previously described (Kopinke et al., 2006). Embryos were cultured in a dark incubator, at 28°C. Control embryos were treated under the same conditions, but with 0.1% DMSO only.

We thank Dr Hiroshi Wada for helpful suggestions on the manuscript. Many other individuals contributed to make this study possible and we gratefully acknowledge their support: Dr Yutaka Kikuchi provided Tg(sox17:EGFP) zebrafish, a plasmid of the sox17 promoter and the nkx2.3 probe; Dr Shin-ichi Higashijima provided technical support for CRISPR/Cas9 technology; Dr Stefan Krauss kindly provided the shha probe; Ms Hideko Utsumi offered technical support; and Mrs Kayoko Takashiro maintained the zebrafish. Staff members in the Spectrography and Bioimaging Facility, National Institute for Basic Biology Core Research Facilities provided technical support. All members of the Takada laboratory contributed helpful discussions. We thank Dr Steven D. Aird for editing the manuscript.