Retinoic acid controls proper head-to-trunk linkage in zebrafish by regulating an anteroposterior somitogenetic rate difference

During vertebrate development, the period of somite segmentation depends on a segmentation clock that is controlled by cyclically expressed genes such as Notch effectors, whereas the position of segmentation is determined by opposed gradients of fibroblast growth factor (FGF) and retinoic acid (RA) (Dubrulle et al., 2001; Holley, 2007; Pourquié, 2001; Sawada et al., 2001). Anterior somites, which give rise to part of the skull and the cervical vertebrae, are formed early, and posterior somites, from which the thoracic, lumbar and sacral vertebrae originate, then arise progressively as the axis elongates (Holley, 2006; Holley, 2007; Morin-Kensicki et al., 2002). Although it might be expected that, to generate the perfectly repetitive pattern of both somites and the resultant vertebrae, the periodicity of somite segmentation should be constant throughout somitogenesis, a substantial difference between anterior and posterior somitogenesis has been observed in several species, including amphioxus, mouse and zebrafish (Hanneman and Westerfield, 1989; Kimmel et al., 1995; Schubert et al., 2001; Tam, 1981). In zebrafish, for instance, the anterior somites form every 20 minutes, whereas the posterior somites form every 30 minutes (Hanneman and Westerfield, 1989; Kimmel et al., 1995). However, how the pace of somitogenesis changes and how this anteroposterior (AP) difference contributes to the later body plan remain unclear.

In this study, we observed somite segmentation in zebrafish embryos and showed that the AP difference is generated by a variable overlap in segmentation period for successive somites. We also found that a rate transition from anterior to posterior somitogenesis is regulated by RA signaling and is required for controlling somite/vertebra number and for correct linkage of the head to the trunk. Our results therefore answer crucial and longstanding questions about the formation of vertebrate segments.

To investigate what happens during the AP transition, we monitored the formation of the first eight somites by time-lapse imaging and found a difference between the first four and the later four somites: the first four, which are the anterior somites, formed quickly, whereas the posterior somites 5 to 8 formed more slowly (supplementary material Fig. S4 and Movie 1A,B). More posterior somites (for example, somites 9 to 12) also formed at a rate similar to that of somites 5 to 8 (supplementary material Movie 2). These observations confirmed that there is a rate/pace difference between anterior and posterior somitogenesis in zebrafish.

As the period of morphological somite formation is the same for the trunk somites (somites 6 to 18) (Schröter et al., 2008), we reasoned that the AP difference is generated by a distinction of segmentation between anterior and posterior somites. To test this possibility, we determined precisely when segmentation begins and ends (for details, see supplementary material Fig. S1 and Materials and methods), and estimated the segmentation period for each somite (Fig. 1A). Unexpectedly, we could not detect any statistically significant difference in segmentation period between somites 1 and 8 (28.9±2.1 minutes; Fig. 1B; supplementary material Table S1). Instead, we found a significant difference (P<0.05) between somites 1-4 and somites 5-8 in the interval between the segmentation start of one somite and the start of the next somite (termed ‘start to start’; Fig. 1C; supplementary material Table S3). Particularly for somites 1 to 4, because the next furrow was visible before the current boundary has completely formed, segmentation periods of anterior somites 1-4 highly overlapped by 13.6±2.5 minutes (47% of the segmentation period). By contrast, the overlap time for posterior somites 5 to 8 was only 2.7±2.5 minutes (9% of the segmentation period) (Fig. 1A-E). This difference between anterior and posterior somitogenesis was also seen in other zebrafish strains (Riken-wako, India and TL) (supplementary material Fig. S5A-C), indicating that the AP difference is not strain specific. These results therefore suggest that the AP difference is generated by variation in the overlap of segmentation period between one somite and the next somite.

Vertebrate somite segmentation is regulated by Notch, FGF and RA activities (Dubrulle et al., 2001; Holley, 2007; Pourquié, 2001; Sawada et al., 2001). To test whether these signals control the AP transition, we inhibited their activity using antisense morpholino oligonucleotides (MOs). Although knockdown of the Notch ligand deltad or the FGF ligand fgf8a, which is known to disrupt Notch or FGF signaling in the context of somitogenesis (Draper et al., 2003; Dubrulle et al., 2001; Duester, 2007; Mara et al., 2007; Wright et al., 2011), had no effect on the AP transition (supplementary material Fig. S5D,E), knockdown of raldh2 (aldh1a2 - Zebrafish Information Network) a major source of RA in the context of somitogenesis (Begemann et al., 2001; Duester, 2007), was detrimental. raldh2 morphants displayed an extension of the segmentation period at somite 4, leading to failure of the AP transition (Fig. 1F-H; supplementary material Tables S1, S2 and Movie 3A,B).

Time-lapse imaging of somitogenesis revealed that 8 and 7 somites were formed within 180 minutes in wild-type embryos and raldh2 morphants, respectively (Fig. 1A,F; supplementary material Movies 1, 3). These results suggest that failure of the AP transition affects the total number of somites. We therefore counted somite number in raldh2 morphants at later stages. raldh2 morphants displayed a persistent one-somite deficiency relative to control embryos throughout 12-16 hours post fertilization (hpf) (Fig. 2A,B; supplementary material Table S6), suggesting that a proper AP transition is required to control the number of somites.

Because a vertebra is formed from the caudal part of one somite and the rostral part of the next somite, somite and vertebra numbers are correlated (Morin-Kensicki et al., 2002). The first two somites do not contribute to the zebrafish vertebral column, whereas somites 3 to 34 give rise to 32 vertebrae: two cervical, ten rib-bearing, two rib- and hemal-arch-bearing, 14 hemal-arch-bearing and four tail fin set vertebrae are formed from anterior to posterior (Morin-Kensicki et al., 2002). As raldh2 morphants lose a somite owing to failure of the AP transition, it seems possible that raldh2 morphants should subsequently display a missing vertebra. Alternatively, as RA treatment is known to change vertebral subtypes through the regulation of Hox genes (Geelen, 1979; Kessel and Gruss, 1991), it is also possible that raldh2 knockdown may induce homeotic transformation. To test whether raldh2 knockdown leads to loss of vertebrae, change of vertebral subtypes or both, we enumerated the number of vertebrae in raldh2 morphants by staining the bones and cartilage. Although bones and cartilage within the vertebral column of raldh2 morphants looked normal, raldh2 knockdown resulted in loss of a specific vertebra, the second cervical vertebra (Fig. 2C,D; supplementary material Table S8). This phenotype was not likely to be induced by homeotic transformation. Rather, because the second cervical vertebra is derived from both the caudal part of somite 4 and the rostral part of somite 5, loss of the second cervical vertebra in raldh2 morphants is consistent with the earlier defect in the AP transition occurring during the segmentation of somite 4 (Fig. 1F-H). These results suggest that RA depletion fails to ensure the segmentation period of somite 4 and the AP transition, leading to the loss of a somite and eventually of the resultant neck vertebra. Our results therefore uncover a novel mechanism, in which RA controls proper head-to-trunk linkage in zebrafish by regulating the segmentation period of somite 4 and the AP transition.

Although segmentation of somite 4 occurs during 11-12 hpf (somitogenesis stages), the future somite boundary of this somite is determined earlier, during 9-10 hpf (gastrulation stages). Thus, we do not yet know whether a supply of RA during gastrulation, somitogenesis or both is required for a proper AP transition. To address this question, we treated embryos with diethylaminobenzaldehyde (DEAB), an RA synthesis enzyme inhibitor (Grandel and Brand, 2011; Hamade et al., 2006), in embryos during different time windows. DEAB treatment during 4-10 hpf (gastrulation stages), but not 10-14 hpf (somitogenesis stages), led to the loss of one somite (supplementary material Fig. S6A,B) and the failure of the AP transition (supplementary material Fig. S6C-E, Tables S1-S4 and Movie 4A,B), phenotypes that were reminiscent of raldh2 morphants (Fig. 2A,B; supplementary material Table S6). Importantly, transient RA administration in raldh2 morphants during 4-10 hpf could partially restore the loss of somite (Fig. 3A; supplementary material Table S7) and vertebra phenotypes (Fig. 3B). These results suggest that RA supply during gastrulation plays a crucial role in controlling somite number by adjusting the segmentation period of somite 4. To test this possibility, we performed time-lapse analyses for raldh2 morphants treated with RA during 4-10 hpf. As expected, RA could rescue the segmentation period of somite 4 and the AP transition in raldh2 morphants (Fig. 3C-E; supplementary material Tables S1-S4 and Movie 5A,B). More specifically, RA treatment during 7-10 hpf could also rescue the defect of the AP transition in raldh2 morphants (data not shown). These results therefore indicate that RA supply before somite segmentation stages (during gastrulation) is important for controlling the AP transition, which is a prerequisite step for determining the number of somites and resultant vertebrae.

We next investigated how RA controls the AP transition during gastrulation. As RA is implicated in the determination of somite size by antagonizing an opposed FGF gradient during somitogenesis (Duester, 2007), we tested the expression of fgf8a, mespb (mespba - Zebrafish Information Network) and papc (pcdh8 - Zebrafish Information Network), which is controlled by the RA/FGF gradient (Fior et al., 2012; Hamade et al., 2006; Holley, 2007; Kawakami et al., 2005; Lee et al., 2009; Yamamoto et al., 1998). However, we could not find a clear difference in expression between control and raldh2 morphants (supplementary material Fig. S7), suggesting that RA-dependent inhibition of FGF signaling is unlikely to regulate the AP transition. This is also supported by the observation that the AP transition occurred normally in fgf8a morphants (supplementary material Fig. S5E).

As RA depletion transiently affects the tempo of somite segmentation (Resende et al., 2010), another possibility is that RA spatially and temporally regulates the segmentation clock during the AP transition. To determine whether such a mechanism exists, we investigated the integrity of the somite segmentation clock in raldh2 morphants during late gastrulation by monitoring the expression of the cyclic gene her1, a major component of the zebrafish segmentation clock (Gajewski et al., 2003; Henry et al., 2002). raldh2 morphants at 9 hpf showed a wave-like propagation of her1 expression in a manner similar to control embryos, although the width of the her1 stripes was larger (Fig. 4A). The number of her1 stripes in control embryos increased from two to three by 10 hpf, whereas only two stripes could be seen in raldh2 morphants at 10 hpf owing to the absence of the most anterior her1 stripe (P<0.05; Fig. 4A). As the number of her1 stripes in raldh2 morphants increased to three by 11 hpf (Fig. 4A), the increase of her1 stripe number was therefore delayed in raldh2 morphants at late gastrulation. Proper transition between anterior and posterior somitogenesis may thus require an increase of her1 stripe number within the paraxial mesoderm at the time when the future boundary of somite 4 is determined, of which RA signaling may control this process.

When we compared her1 expression between control and raldh2 morphants at 10 hpf, we noticed that the upper stripes of her1 were thinner than those of control morphants (Fig. 4A), suggesting that RA signaling suppresses her1 expression at the anterior part of the paraxial mesoderm at late gastrulation through an unidentified effector(s). Ripply1 is known to repress the transcription of specific genes, which are targets of T-box transcription factor family members, by recruiting the co-repressor Groucho (Kawamura et al., 2005). As ripply1 knockdown results in the expansion of her1 expression into somite regions during somitogenesis, and as her1 expression is controlled by the three T-box family members spadetail (tbx16 - Zebrafish Information Network), no tail, fused somites (tbx6 - Zebrafish Information Network) (Garnett et al., 2009; Holley et al., 2000; Nikaido et al., 2002), it seemed possible that, by mediating RA signaling, Ripply1 suppresses her1 expression during the AP transition.

To test whether RA signaling affects ripply1 expression during late gastrulation, we investigated ripply1 expression in raldh2 morphants and found that an extra stripe of ripply1 could be seen at a location close to where the anterior stripe of her1 was lost (Fig. 4C). Importantly, two-color in situ hybridization of her1 and ripply1 revealed that the extra stripe of ripply1 was adjacent to the second stripe of her1 in raldh2 morphants (Fig. 4B). If this is the case, double knockdown of raldh2 and ripply1 should be able to rescue the effect of raldh2 knockdown for the number of her1 stripes. To test this possibility, we co-injected raldh2 MO and ripply1 MO and found a significant rescue phenotype for the number of her1 stripe (P<0.05; Fig. 4A,D). Therefore, our results suggest a possibility that the failure of AP transition is caused by RA/Ripply1-mediated spatial and temporal inhibition of her1 expression.

To investigate this further, we overexpressed either Ripply1 or Ripply1ΔWRPW, which cannot interact with Groucho (Kawamura et al., 2005), and performed in situ hybridization of her1 in the manipulated embryos. Although overexpression of Ripply1ΔWRPW did not affect her1 expression in the presomitic mesoderm (PSM) at 12 hpf, overexpression of Ripply1 inhibited her1 expression (Fig. 5A). Taken together, our results suggest that RA/Ripply1-mediated spatial and temporal regulation of her1 propagation leads to a proper transition from anterior to posterior somitogenesis (Fig. 5B).

RA signaling plays crucial roles in multiple developmental processes. In the case of somitogenesis, it has been reported that either complete loss of raldh2 (a major source of RA during development) or treatment of excess RA results in defects in the expression of Hox genes, differentiation of somite cells, wavefront formation and symmetric somite formation (Begemann et al., 2001; Geelen, 1979; Holley, 2007; Kawakami et al., 2005; Kessel and Gruss, 1991; Niederreither and Dollé, 2008). However, a defect in AP transition in manipulated embryos has not been observed. In the present study, we partially knocked raldh2 down in zebrafish embryos and showed a novel requirement of RA signaling for the AP transition.

Why does partial knockdown of raldh2 lead to the specific defect in AP transition? This may be explained by the differences in sensitivity towards RA between the processes of somitogenesis. In embryos injected with 1.7 ng raldh2 MO (partial knockdown situation), in which RA signaling is weakly activated (supplementary material Fig. S2), a small amount of RA generated by the rest of raldh2 and/or other sources of RA is not sufficient to regulate the AP transition, but sufficient for controlling other processes of somitogenesis (Fig. 1F-H; Fig. 2C,D).

A future somite boundary is determined at a position where her1 segmentation clock meets wavefront generated by FGF/RA gradients (Holley, 2007; Pourquié, 2001). Based on the positional information, mespb starts to express at the right position (rostral side of S-II), and then morphological somite segmentation takes place at B-1 (the boundary between S0 and S-I) (Holley, 2007; Pourquié, 2001). Because many processes such as gene expression, signal transduction and differentiation of PSM cells occur during a duration from the establishment of the future somite boundary to morphological segmentation of the resultant somite, it is easy to imagine that the duration is not short.

We realized that morphological segmentation between somite 4 and 5 occurring at around 11-12 hpf was delayed in raldh2 morphants (Fig. 1A,F). We also found a delay in the increase in her1 stripe during 10-11 hpf (Fig. 4A). Our results suggest a possibility that, in raldh2 morphants, the delay of morphological segmentation between somite 4 and 5 is caused by the delay in the increase of her1 stripe during late gastrulation. In other words, although the future boundary between somite 4 and 5 is established at the time (during 9-10 hpf) when the anteriormost her1 stripe appears and in the position (within the elongating paraxial mesoderm) where the anteriormost her1 stripe meets wavefront in wild-type embryos, only the timing of the formation of the future somite boundary is affected in raldh2 morphants. This possibility is supported by the data showing that the repetitive structures of somites looked normal in raldh2 morphants (supplementary material Movie 3A,B) and that the timing of morphological segmentation between somite 4 and 5 could be restored in raldh2 morphants treated with RA (4-10 or 7-10 hpf) (Fig. 3C-E; supplementary material Movie 5, 5-1).

During epiboly/gastrulation, the paraxial mesoderm regions including somite progenitor cells elongate quickly. During 90-100% epiboly (9-10 hpf), the number of her1 stripe increases from two to three within the elongating paraxial mesoderm in control embryos (Fig. 4A; Fig. 5B). However, the third strip of her1 did not appear by 10 hpf in raldh2 morphants although epiboly occurred normally (Fig. 4A). Thus, it seemed that an integration of clock and wavefront takes place in a different manner: The integration occurs at a position adjacent to the second and third stripe of her1 at 9 and 10 hpf, respectively. Our results therefore suggest that smooth increase of her1 stripe during 9-10 hpf is required for the proper transition of anterior- to posterior-somitogenesis, and that RA signaling represses ripply1 expression at an anterior region of the second stripe of her1, and regulates timing of the appearance of the third stripe of her1 to achieve the proper AP transition.

In raldh2 morphants, the period of morphological segmentation of somite 4 became longer (Fig. 1F,G). As RA is known to control the tempo of segmentation clock in chick (Resende et al., 2010), it is possible that RA signaling affects the period of her1 oscillation. However, this is not likely because the timing for morphological segmentation of somite 4 is different from that of the future somite boundary formation, which occurred at around 10 hpf. Although it is expected that either the activity of somite segmentation machineries (such as Ephs and Ephrins) (Holley, 2007) or cellular responses (or both) is altered in raldh2 morphants, how partial knockdown of raldh2 affects segmentation period of somite 4 remains unclear. Because these processes are dynamic, it would be of great interest to observe spatial and temporal dynamics of segmentation clock and wavefront coupled with morphological segmentation.

The difference between anterior and posterior somitogenesis was first observed in zebrafish about 30 years ago. However, how the AP difference occurs during development has remained unknown. In this study, we provide the first evidence that the AP difference is explained by variation in the duration of an overlap between successive somite segmentation periods. We also provide insight into the molecular mechanism by which transient repression of propagation of the her1 clock controls the proper transition from shorter to longer somitogenetic periodicity. Furthermore, our results reveal that the proper AP transition of somitogenesis is a prerequisite step for the correct linkage of head and trunk in the vertebrate body plan.

Wild-type zebrafish and zebrafish wild-type strains (Riken-wako, India and TL) were used in this study. Single- or double-color whole-mount in situ hybridization was performed as described previously (Matsui et al., 2005; Matsui et al., 2011). cDNA fragments of fgf8a, her1, mespb, papc, ripply1, tbx24 (tbx6 - Zebrafish Information Network), uncx4.1 and Venus were used as templates for antisense probes.

Embryos at 6-7 hpf were manually dechorionated and embedded in 1% low melting agarose. Time-lapse image acquisition was performed with an Olympus FV-1000-D confocal microscope and FLUOVIEW software. Z-stack images (ten planes and 17.15-μm intervals) were obtained every 2 minutes at 28.5°C. Using Metamorph software, images were combined into a time-stack, time-stamped and saved as a TIFF-stack and MOV movie.

From each somitogenesis movie, the time of segmentation start and end was visually annotated. Because somite boundary formation starts from the dorsal side in the lateral view, the start of segmentation (0 minutes) was defined as when a furrow appeared at the dorsal limit of the PSM (supplementary material Fig. S1A). Segmentation was considered to have ended when the boundary was clearly visible and the somite was completely separated from the PSM (supplementary material Fig. S1B).

Fish (1.5 months old) were fixed with 99.9% ethanol for at least 24 hours and their scales and internal organs were removed. They were treated with acetone for 24 hours, and then incubated in staining solution (0.015% Alcian Blue, 0.015% Alizarin Red in 70% ethanol) at 37°C for 2-3 days. After being washed with water, the fish were cleared using 1% potassium hydroxide (KOH) for 12-24 hours, and successively incubated in 1% KOH in 20% glycerol and 0.01% KOH in 20% glycerol. Stained fish were stored in 50% glycerol.

Embryos were treated with 10 μM DEAB (Tokyo Kasei) during either 4-10 or 10-14 hpf, and washed extensively with fish water. Embryos were then fixed and used for in situ hybridization.

The following antisense MO oligonucleotides against raldh2, deltad (deltaD - Zebrafish Information Network), fgf8a and a control MO were obtained from Gene Tools: control MO: 5′-CCTCTTACCTCAGTTACAATTTATA-3′; deltad MO: 5′-AACAGCTATCATTAGTCGTCCCATG-3′ (Kawamura et al., 2008) (a gift from Dr Shinji Takada); fgf8a MO: 5′-GAGTCTCATGTTTATAGCCTCAGTA-3′ (Matsui et al., 2011); raldh2 MO: 5′-GTTCAACTTCACTGGAGGTCATC-3′ (Begemann et al., 2001; Kawakami et al., 2005).

The plasmids pCS2-ripply1, pCS2-ripply1ΔWRPW (gifts from Dr Shinji Takada) and pCS2-monomeric red fluorescent protein (mRFP) were used in this study. ripply1, ripply1ΔWRPW and mRFP mRNA were synthesized using the SP6 mMessage mMachine System (Ambion).

For in situ hybridization and skeletal staining, raldh2 MO (1.7 ng) or control MO (1.7 ng) were injected into the yolk of one-cell-stage zebrafish embryos. Embryos were fixed at the indicated time points.

neckless (nls; aldh1a2 - Zebrafish Information Network) mutants display a variety of developmental defects, such as small hindbrain, lack of pectoral fins, asymmetric somite formation, small somites and short body (Begemann et al., 2001; Kawakami et al., 2005). However, injection of 1.7 ng raldh2 MO led to milder phenotypes compared with those of nls mutants (supplementary material Fig. S3). To test whether this amount of raldh2 MO could inhibit RA signaling in zebrafish embryos, we generated a transgenic line Tg[RARE:Venus] by using Tol2 system (Kawakami et al., 2004; Urasaki et al., 2006). Because Tg[RARE:Venus] carries 3 x RA response element (RARE, 5′-AGCTTGAAGGGTTCACCGAAAGTTCACTCGCA-3′) (Rossant et al., 1991), human β-globin minimal promoter, Venus full-length cDNA and SV40 poly A sequence, we can monitor RA signal activity within zebrafish embryo by observing Venus expression. As it has been reported that 8.5 or 15 ng raldh2 MO can phenocopy nls (Begemann et al., 2001), we injected 8.5 or 15 ng raldh2 MO into the Tg[RARE:Venus] embryos. As expected, Venus expression was strongly reduced, and the nls phenotypes (e.g. truncation of hindbrain, small and abnormal shape of somites, and short body) could be seen in the manipulated embryos (supplementary material Fig. S2). Although injection of 1.7 ng raldh2 MO led to mild defects in hindbrain and trunk, Venus signals were reduced in the manipulated embryos at 10 and 28 hpf (supplementary material Fig. S2). These results therefore suggest that injection of 1.7 ng raldh2 MO leads to a partial knockdown situation in zebrafish embryos.

For time-lapse imaging, raldh2 MO (1.7 ng), deltad MO (2 ng) or fgf8a MO (5 ng) were injected into the yolk of one-cell-stage zebrafish embryos as described previously (Matsui et al., 2011). After time-lapse imaging, embryos were recovered from 1% low melting agarose, left to develop until the indicated time points, and tested whether they show specific phenotypes induced by gene knockdown (supplementary material Fig. S3). Any embryo that did not display the specific gene knockdown phenotypes was excluded from the subsequent analyses.

Embryos injected with raldh2 MO (1.7 ng) were treated with 1 nM RA (Sigma) during at 4-10 or 7-10 hpf. For time-lapse imaging, embryos were immediately mounted in 1% low melting agarose and time-series data were collected. Embryos were then fixed at the indicated time points and used for in situ hybridization and skeletal staining.

Analysis of variance (ANOVA) method was used for multiple comparisons and significant differences were determined using Scheffe’s post-hoc test. Differences in gene expression pattern were analyzed by two-tailed Fisher’s exact tests. All statistical analyses were performed using R software. Results were considered significant when P<0.05.

We are grateful to Shinji Takada for sharing reagents, for helpful discussions and for critical reading of the manuscript. We also thank Ian Smith and Fiqri Dizar Khaidizar for help in preparing the manuscript; Koichi Kawakami, Takashi Akanuma and Jun Koyama for sharing reagents; and Maiko Yokouchi and Hiroko Shigesato for technical assistance.