RARβ2 is required for vertebrate somitogenesis

The vertebrate retinoic acid receptor (RAR) family comprises three genes encoding three RAR subtypes: RARα, RARβ and RARγ. These subtypes differ in their temporal and spatial expression, inducibility by retinoic acid (RA) (auto-regulation), post-translational modification, epigenetic regulation and basal repression (e.g. co-factor recruitment). It is posited that certain vertebrate innovations (neural crest, sensory placodes, segmentation of the brain, etc.) necessitated the evolution of the three RAR subtypes that subspecialized to modulate diverse developmental processes (Albalat et al., 2011). The ligand binding domain of an inferred ancestral RAR most closely resembles mammalian RARβ, which is considered to be the most primitive of RARs. Therefore, RARβ was likely to be the first RAR to evolve (Escriva et al., 2006).

One obstacle to studying the different roles of RAR subtypes is the lack of embryonic phenotypes observed in mouse RAR knockout studies. For example, disruption of murine RARβ2 led to mostly non-embryonic, adult phenotypes such as deficits in memory/spatial skills, premature alveolus formation (Massaro et al., 2000; Chiang et al., 1998), impaired growth, increased proliferation and pigmentation behind the lens, and occasional vertebral homeotic transformations (Ghyselinck et al., 1997). Rara^−/− embryos are viable up until 2 months after birth (Lufkin et al., 1993; Massaro et al., 2003). Only double subtype mutants (e.g. Rarb/Rarg^−/−, Rara/Rarg^−/−) yield overt embryonic phenotypes (Lohnes et al., 1994; Subbarayan et al., 1997, reviewed by Maden, 2010). This apparent functional redundancy may not represent the physiological condition, but rather an incomplete exploration of the possible phenotypes in the laboratory environment (Mark et al., 2006; Ghyselinck et al., 1997). Indeed, Rarb^−/− mice do not develop microphthalmia (Ghyselinck et al., 1997), as observed in humans with RARβ loss of function (Srour et al., 2013). Aside from oncogenic fusions (e.g. PML-RARα, RARγ/NUP98), there are no reported mutations for human RARα or RARγ associated with diseases (http://omim.org), suggesting that such mutations may be embryonic lethal. This provides evidence that human RARs generally cannot compensate for each other, and that mouse might not be an ideal model for studying the function of individual RAR subtypes.

In chick and zebrafish, loss of individual RARs causes specific phenotypes (Romeih et al., 2003; Garnaas et al., 2012; He et al., 2011; D'Aniello et al., 2013). RARα or RARγ knockdown in the frog, Xenopus laevis, produces specific defects in neuronal differentiation, pre-placodal ectoderm formation and axial elongation (Janesick et al., 2014, 2013, 2012; Koide et al., 2001). We have shown that RARγ is required for axial elongation and for maintenance of the caudal progenitor pool (Janesick et al., 2014). However, RARγ is mostly absent from the trunk and lateral plate mesoderm from which RA emanates. Therefore, RA-regulated processes such as neural tube patterning, hindbrain boundary setting, somite differentiation, limb development, and heart and lung morphogenesis (Maden, 2007; Niederreither and Dollé, 2008) likely rely on RARα or RARβ. Although RARβ is expressed in the somites and lateral plate mesoderm, positioning it to regulate somitogensis in chick and mouse (Cui et al., 2003; Ruberte et al., 1991; Romeih et al., 2003; Bayha et al., 2009), the functionality of RARβ in early somitic development had not been explored.

Somitogenesis is a process whereby cells from the caudal progenitor pool contribute to the presomitic mesoderm (PSM) that contains committed somite precursor cells to supply the rostral, determination wavefront (reviewed by Dequéant and Pourquié, 2008). The decision of cells within the caudal progenitor pool to become PSM is restricted by an alternative, mutually exclusive fate decision towards the elongating neural tube, indicated by Sox2 expression (Takemoto et al., 2011). The PSM is initially homogenous, marked by Mesogenin1 and Tbx6, then gradually organizes into a mesenchymal mass patterned as somitomeres (newly forming somites) (Dequéant and Pourquié, 2008) marked by genes such as Mespa and Ripply2. Tbx6 promotes somite maturation (Nikaido et al., 2002) and its targets are repressed by Ripply2 to facilitate establishment of somite boundaries (Dahmann et al., 2011). Epithelialization of the somitomeres produces mature somites (Nakaya et al., 2004), which are dorsoventrally segregated into epaxial and hypaxial territories (Cheng et al., 2004). Hypaxial dermomyotome cells ultimately delaminate and migrate to populate musculature in the limb, diaphragm and tongue (Dietrich et al., 1998; Martin and Harland, 2001).

Considering that knockdown of individual Xenopus RAR subtypes or isoforms yields distinct embryonic phenotypes, we characterized the role of RARβ2 in somitogenesis. Here, we show that expression of Rarβ2 is induced by the RAR agonist TTNPB, and diminished by the RAR antagonist AGN193109 due to the presence of two highly conserved RA response elements (RAREs) in the promoter of Rarβ2. Rarβ2 is the last RAR subtype to be expressed during Xenopus development. We hypothesize that other RAR subtypes are required to initiate or maintain Rarβ2 expression because knockdown of either RARα or RARγ ablates Rarβ2 expression. RARβ2 is spatially positioned to be the subtype most responsive to ligand emanating from the trunk, and is required to restrict PSM markers. Loss of RARβ2 yields a multifaceted phenotype on somitogenesis: somite number is decreased and domains are larger, chevron morphology is perturbed, and hypaxial myoblasts fail to migrate ventrally. We also explored the in vivo transcriptional relationships between Ripply2, an RAR-responsive gene that regulates boundary setting during somitogenesis, Tbx6 (a PSM marker) and Tbx3 (a hypaxial myoblast and notochord marker). We found that Ripply2 inhibits Tbx6 transcriptional activity, while the ability of Tbx3 to repress transcription does not involve Ripply2. Taken together, our results show for the first time that RARβ2 plays an early and important role in somitogenesis.

Using the latest genome assembly available for X. laevis (build 9.1) (Session et al., 2016), we identified Rarβ1 and Rarβ2 via comparison with X. tropicalis and submitted these sequences to NCBI (Rarβ1, KF547939; Rarβ2, KF547940). RARβ is found on chromosome 6, with homeolog copies Rarβ.L and Rarβ.S (Session et al., 2016). Fig. S1 demonstrates that Rarβ1 and Rarβ2 are isoforms, and differ by alternative promoter usage creating a distinct N-terminal structure, as is the case in mouse, chicken and other vertebrates (Zelent et al., 1991; Leroy et al., 1991; Brand et al., 1990). We mapped the 5′ UTRs of Rarβ1 and Rarβ2 from X. tropicalis to X. laevis, and designed antisense RNA probes that could distinguish between the two isoforms (Table S1). Whole-mount in situ hybridization determined the temporal and spatial expression of X. laevis Rarβ1 and Rarβ2 (Fig. 1). Rarβ1 is not detected by whole-mount in situ hybridization at any stage tested (tailbud stage shown in Fig. 1C). Rarβ2 first exhibits a specific expression pattern at mid-tailbud (approximately stage 26), when it is expressed in mature somites, eye, branchial arches, anterior neural tube and hatching gland (Fig. 1A,B). These data mostly agree with previous Xenopus expression data using a partial sequence (Escriva et al., 2006). A sense probe for Rarβ2 revealed no specific expression, confirming the specificity of our probe (not shown).

QPCR analysis was conducted with primers that amplify both homeologs of either Rarβ1 or Rarβ2 (Table S3). A comprehensive, quantitative comparison of all X. laevis RAR subtypes and isoforms is shown in Fig. S2. Rarβ1 is not robustly detected by QPCR until stage 40 (Fig. 1D), which is concordant with whole-mount in situ hybridization results. Rarβ2 is detected at stage 18 by QPCR (Fig. 1D), but does not exhibit distinct expression by whole-mount in situ hybridization at that stage. Although both Rarβ1 and Rarβ2 are maternal transcripts, Rarβ2 is the predominant Rarβ isoform expressed during early development. Rarβ2 mRNA is ∼1000-7000 times less abundant than Rarγ2 and ∼60-1000 times less abundant than Rarα1 and Rarα2 mRNAs at gastrula and neurula stages 10-18 (Fig. S2). At later stages, Rarβ2 is ∼10-100 times less abundant than Rarα1, Rarα2 and Rarγ2 (Fig. S2).

Vertebrate RARs possess RAREs in their regulatory regions and one or more isoforms are directly regulated by RA at the transcriptional level. The first characterized RAREs were identified in the human (de The et al., 1990) and mouse (Sucov et al., 1990) Rarβ2 promoters. We found that the archetypal ‘canonical’ RARE, a direct repeat separated by five nucleotides (DR5), is located ∼500 bp upstream of the RARβ2 start codon and is highly conserved in vertebrates (Fig. S3). In the ascidian, Ciona intestinalis, a DR2 RARE is found in the first intron of CiRAR. Inspection of the aligned promoter sequences revealed an additional conserved element in vertebrates composed of an upstream DR5 and an additional conserved half site (Fig. S3). Unlike other vertebrates, zebrafish lacks a recognizable Rarβ gene. Instead, the RARE is found in raraa, an ortholog of Rarα (Hale et al., 2006; Waxman and Yelon, 2007), which was reported to be the only RA-inducible zebrafish RAR (Linville et al., 2009). A putative RARE is found upstream of a Fugu rubripes gene model labeled ‘RARγ-A-like’, which a BLASTP search indicates is most closely related to RARβ (not shown).

The identification of an RARE in both homeologs of X. laevis Rarβ2, led us to hypothesize that Rarβ2 is RA inducible, as in other vertebrates. We asked which Xenopus RAR subtypes and isoforms responded to the RAR-selective agonist TTNPB, or to the RAR-selective antagonist AGN193109 (Koide et al., 2001). Rarβ2 is strongly upregulated by TTNPB and repressed by AGN193109 (Fig. 2A). Rarα isoforms are modestly induced by TTNPB and repressed by AGN193109, Rarγ2 is downregulated by TTNPB and Rarβ1 is not detected by QPCR (Fig. 2A). Furthermore, the expression domain of Rarβ2 is greatly expanded by TTNPB, particularly in the anterior neural tube and branchial arches (Fig. S4). Knockdown of either RARα or RARγ results in loss of Rarβ expression (Fig. 2B). Hence, Rarα2 and Rarγ2, which are expressed earlier than Rarβ2 (Fig. S2), are required to initiate or maintain Rarβ2 expression.

We hypothesized that elements of the Rarβ2 promoter are required for RA responsiveness and are occupied by RARs. Although the canonical RARE had previously been characterized (Fig. S3), the other conserved elements had not. We selectively mutated the canonical DR5, upstream DR5 and upstream half-site and cloned these putative promoters into a promoterless luciferase reporter to generate four distinct reporter constructs. The canonical DR5 and upstream DR5 are the most important for the TTNPB response in vivo (Fig. 3A), whereas mutating the upstream half-site does not affect TTNPB responsiveness (Fig. 3A). Next, we designed both wild-type and mutated oligonucleotide pairs containing each RARE with 5 bp flanking sequence. RARα, RARβ and RARγ are all capable of binding the upstream and canonical RAREs as heterodimers with Xenopus RXRα (Fig. S5).

As Rarβ1 is not expressed in the early embryo, we focused our analysis on Rarβ2 and targeted both homologs (L and S) for MO knockdown (Table S1) (Karpinka et al., 2015). We designed morpholinos selective for Rarβ2 and found that knockdown of S or L homeologs produces similar phenotypes on Myod expression; however, the L knockdown produced a subjectively stronger effect (Fig. S6). A combination of both MOs produced the strongest phenotypes; therefore, a mixture was used for subsequent experiments. Although Rarβ2 is specifically expressed in the hatching gland, we detected no difference in time to hatching or in the rate of hatching between Rarβ2 MO and control MO-injected embryos (data not shown).

Rarβ2 is expressed in mature trunk somites (Fig. 1) and RA is important for somite patterning (Moreno and Kintner, 2004; Janesick et al., 2014); therefore, we asked whether somite morphology in MO-injected embryos was disrupted. Analysis of the general muscle marker Myod at stage 40 revealed that somite number is reduced; somites appear thicker and migration of hypaxial muscle (red arrows) is abolished in bilaterally microinjected Rarβ2 MO embryos (Fig. 4A-C). Notably, the characteristic chevron-shaped somites seen in control embryos are transformed into U-shaped or straight somites in Rarβ2 MO embryos and some embryos show significant disorganization and blurring of somite boundaries (Fig. 4C). The same phenotype is observed at stage 45, indicating that loss of hypaxial migration is not simply a developmental delay. These tadpoles are also paralyzed (see Discussion). Unilaterally microinjected Rarβ2 MO embryos show thicker, fewer and disorganized somites, and loss of hypaxial migration (Fig. 4D-G). Melanophore migration also fails to occur on the injected side (compare with Fig. 4D,E). Expression of Tbx3, which marks hypaxial myoblasts (Martin et al., 2007), is knocked down and cells expressing Tbx3 fail to migrate ventrally from the anterior trunk somites compared with the uninjected side (compare with Fig. 4F,G).

Unilateral knockdown of RARβ2 produces similar phenotypes to Myod expression at tailbud stage 26 (Fig. 5A,B). The thicker, U-shaped morphology is predominant (Fig. 5B,C) but we also observed forked hypaxial regions (Fig. 5D), blurred somite domains (Fig. 5E) and criss-crossed somites (Fig. 5F), especially in the more anterior somites where Rarβ2 is strongly expressed. Coronal sections were used to measure somite size and number (Fig. 5G,H). We observed a reduced number of somites on the injected side of Rarβ2 MO embryos, and also a slight reduction in somite number on the uninjected side of Rarβ2 MO embryos compared with control MO (Fig. 5I). We did not detect significant changes in the unsegmented PSM length (Fig. S7); however, deciphering the exact rostral PSM boundary from Myod expression and morphology alone is challenging. Increased somite length (Fig. 5J) was visible and significant in Rarβ2 MO embryos, and is likely a strong contributing factor to decreased somite number, as more PSM cells are incorporated into each somite. We conclude that loss of Rarβ2 yields fewer and bigger, U-shaped somites with impaired boundary formation.

Segmented PSM markers Ripply2 and Mespa/Thyl2 are most readily viewed at neurula stages where microinjection of Rarβ2 MO shifts expression of these markers rostrally and each domain appears thicker (Fig. 6A,B) compared with control MO (Fig. S8A,B). The segmented boundaries of Ripply2 are mostly maintained compared with Rarγ-MO embryos where Ripply2 boundaries are lost (Janesick et al., 2014). Others have previously demonstrated that RA regulates laterality and coordinates left-right timing of the somitogenesis clock such that somites develop symmetrically across the midline (Kawakami et al., 2005; Vermot and Pourquié, 2005). Bilaterally injected Rarβ2 MO embryos exhibited noticeable left-right asymmetry in 32% of the embryos (Fig. S9). Unsegmented PSM markers Tbx6, Msgn1, Fgf8 and Esr5 are also shifted rostrally by Rarβ2 MO (Fig. 6C-F) compared with control MO (Fig. S8C-F), a phenotype distinct from Rarγ-MO embryos where PSM and caudal expression is diminished (Janesick et al., 2014). Double whole-mount in situ hybridization analysis verified that Rarβ2 and Ripply2 expression do not overlap (Fig. S10) and presumably this is the case for other PSM markers that are equal or posterior to Ripply2 (Hitachi et al., 2008). This suggests that RARβ2 is the receptor subtype that responds to RA and places Rarβ2 in the correct position to confine expression of PSM and caudal progenitor genes to the posterior territory.

Loss of somite chevron morphology is often attributed to deformities in the horizontal myoseptum and notochord (Rost et al., 2014). The myoseptum separates dorsal and ventral somite domains, and is required for aquatic locomotion. In zebrafish, the notochord is required for adaxial ‘muscle pioneers’, for myoseptum and for chevron somite morphology (Halpern et al., 1993; Brennan et al., 2002). We tested whether Rarβ2 MO embryos possessed a normal notochord histologically and by looking at expression of Xnot. Rarβ2 MO embryos form a notochord (not shown) and express Xnot, albeit the axis is shorter and crooked (Fig. S11B) compared with controls (Fig. S11A). Next, we investigated the expression of presumptive myoseptum or muscle pioneer markers, such as Cxcl12, Notum, Netrin, Wnt11 and Engrailed 1, that have been established in zebrafish. Most markers tested by us, or viewed in Xenbase (Karpinka et al., 2015) are not specific to the myoseptum or adaxial pioneers but rather are: (1) not expressed, (2) expressed in the neural tube floor plate or interneurons, or (3) expressed within the entire somite domain (Fig. S12). Therefore, either Xenopus does not possess a marker for these cells/structures, or the myotome is not compartmentalized as in zebrafish (see Discussion).

Despite the lack of molecular markers for the myoseptum and adaxial cells, Xenopus still possesses chevron-shaped somites, which become U-shaped and disorganized in Rarβ2 MO embryos. We hypothesized that the boundary-setting gene Ripply2 is involved in the phenotype. Ripply1/2 are spatially regulated by retinoids (Janesick et al., 2014; Moreno et al., 2008) and Rarβ2 MO causes rostral expansion and broadened domains of Ripply2 (Fig. 6A). We and others showed that developing organisms are exquisitely sensitive to misexpression of Ripply genes (Janesick et al., 2012; Li et al., 2013; Kawamura et al., 2005). Ripply1 overexpression eliminates notochord and myoseptum (Kawamura et al., 2005) and Ripply proteins commonly associate with T-BOX proteins, converting them to transcriptional repressors in the presence of Groucho (Hitachi et al., 2009; Janesick et al., 2012; Kawamura et al., 2005, 2008; Kondow et al., 2006, 2007; Windner et al., 2015). Tbx6 interacts with Ripply1 in zebrafish (Kawamura et al., 2008) and inhibits adaxial Myod expression, suppressing notochord formation (Goering et al., 2003). Tbx3 is a potential direct RAR target (Ballim et al., 2012) and is expressed in the notochord and hypaxial muscle (Takabatake et al., 2000; Martin et al., 2007). Thus, Ripply2, Tbx3 and Tbx6 are plausible candidates to regulate somite chevron morphology, presomitic mesoderm and hypaxial muscle migration.

There are two Ripply2 orthologs in Xenopus termed Bowline (Kondow et al., 2006) and Ledgerline (Chan et al., 2006), and two homeologs for each in X. laevis (Fig. S13A). Xenopus Ripply1 has apparently been lost during evolution (Janesick et al., 2012). Ledgerline and Bowline are both expressed in a Ripply2-like pattern. Neither Ledgerline nor Bowline is expressed in mature somites, in contrast to zebrafish ripply1 (Kawamura et al., 2005). Notochord, which is marked by Xnot expression, is completely obliterated in embryos microinjected with Bowline, and hyperdorsalization or double axes are induced (Fig. S11C), a phenomenon we and others observed for Ripply3 (Li et al., 2013). Embryos microinjected with Ledgerline diminished Xnot expression, but exhibited stronger axial defects (Fig. S11D). In embryos that still had body axes, we observed that Myod expression is completely abolished on the injected side (Fig. S13D).

Ripply2 has two conserved regions found in all Ripply family genes: a WRPW motif that facilitates interaction with Groucho (Fisher et al., 1996; Kondow et al., 2006) and an FPVQ motif that is predicted to mediate contacts with T-BOX proteins (Kawamura et al., 2008). We found that overexpression of Ripply2 double mutants (WRPW→AAAA; FPVQ→AAAA), which presumably cannot interact with Tbx or Groucho, are phenotypically normal in somite morphology, marked by Myod (Fig. S13D). Single-domain mutants (WRPW or FPVQ) are mostly normal except that Ledgerline WRPW mutants display an intermediate phenotype between wild type and double mutant overexpression (data not shown).

We have previously shown that Ripply3 inhibits Tbx1 transcriptional activity in vivo (Janesick et al., 2012). Similarly, zebrafish Ripply1 converts Tbx6 from a transcriptional activator to a repressor in vitro (Kawamura et al., 2008) and this protein interaction is essential for establishing the posterior somite boundary (Morimoto et al., 2007; Takahashi et al., 2010). We hypothesized that Xenopus Ripply2 would also convert Tbx6 into a transcriptional repressor. Microinjection experiments showed that Tbx6 is a transcriptional activator (Fig. 7). Xenopus Ripply2 (Bowline or Ledgerline) inhibits Tbx6 transcriptional activation in vivo, and this effect is blocked by mutation of the WRPW and FPVQ domains of Ripply2 (Fig. 7). By contrast, Tbx3 is a transcriptional repressor in vivo (Fig. 7), substantiating previous in vitro reports (He et al., 1999; Hoogaars et al., 2004). Competition with mutant Ripply2 mRNAs does not affect Tbx3 activity (Fig. 7); therefore, Tbx3 is unlikely to employ Ripply2 as a co-repressor. We conclude that Ripply2 is a retinoid-responsive gene modulating Tbx6 activity in the presomitic mesoderm, but not Tbx3 in the notochord and hypaxial muscle.

Rarβ2 is the predominant Rarβ isoform in X. laevis. Both Rarα and Rarγ are expressed earlier in developmental time and are required for the expression of Rarβ2. This is compatible with the observation that Rarβ2 expression is lost in chicken embryos injected in ovo with antisense oligonucleotides blocking Rarα2 (Cui et al., 2003). RARβ2 expression is RA-regulated: deletion of 24 bp of the Rarβ2 promoter, including the canonical RARE but sparing the TATA-box, abolished responsiveness to RA (Sucov et al., 1990). We identified a second RARE and additional half-site upstream of the canonical RARβ element, and showed that the second RARE also conferred activity in whole embryos. This second RARE might be redundant or could function as a shadow enhancer (Hong et al., 2008; Frankel et al., 2010; Perry et al., 2010) during different developmental times, or when vitamin A is less abundant. By contrast, we were unable to demonstrate activity in the conserved half-site. Although nuclear receptors can bind to half sites as monomers, this is not typically observed in RXR-dependent heterodimeric partners such as RAR (Mangelsdorf and Evans, 1995).

Our finding that Rarβ2 is the RAR subtype most strongly upregulated in response to the pan-RAR agonist TTNPB shows that Rarβ2 expression is sensitive and responsive to RA levels in the embryo. Rarβ2 is expressed in mature somites in X. laevis, is positioned closest to the presumed ALDH1A2 source of RA (Haselbeck et al., 1999) and is likely to be a primary responder to ligand. Previous studies have shown that RA upregulated the expression of Rarb2 in mouse P19 cells in the presence of cycloheximide (Dey et al., 1994) and expanded reporter expression in the limbs of Rarbβ2 promoter-lacZ transgenic mouse embryos (Mendelsohn et al., 1991). Rarβ2 is significantly affected in VAD embryos (Cui et al., 2003), suggesting that transcription of Rarβ2 requires RA. By contrast, we showed that Rarγ is inhibited by RAR agonist TTNPB, consistent with its role as an unliganded repressor that is required to maintain the presomitic mesoderm (PSM) and promote axial elongation (Janesick et al., 2014). Our working model is that Rarβ2 is activated in the trunk by RA to regulate somitogenesis, whereas Rarγ is present in the tail where RA is absent, sustaining the population of cells that will contribute to somites.

Somitogenesis is a process born out of the proliferation of caudal progenitors, typically marked by Wnt3 and Fgf8, which contribute to the unsegmented PSM (marked by Tbx6 and Msgn1). Newly forming somitomeres make up the segmented PSM (marked by Ripply2 and Mespa/Thyl2), and are subsequently epithelialized to become mature somites, marked by Myod and Rarβ2. Loss of RARβ2 leads to the expansion of Tbx6, Msgn1 and Fgf8, and corresponding rostral shifts and wider expression domains in Ripply2 and Mespa. The rostral expansion and shift of the PSM and somitomeres is anticipated considering that: (1) activation by RAR yields the opposite phenotype – caudal expansion of somitomeres and diminished PSM (Janesick et al., 2014; Moreno and Kintner, 2004); and (2) Rarβ2 expression in mature somites is spatially positioned to restrict the PSM. Other evidence supporting our results include the phenotypes of Raldh2^−/− embryos (Cunningham et al., 2015) and RA antagonist-treated embryos (Janesick et al., 2014), where Tbx6 expression is significantly expanded. Our results also reinforce the phenomenon of RAR-FGF mutual antagonism (Diez del Corral et al., 2003). Recently, murine RARβ2 was demonstrated to be recruited to the Fgf8 RARE in the trunk where it functioned as a transcriptional repressor (Kumar et al., 2016). If this mechanism is conserved in Xenopus, then loss of RARβ2 would relieve repression on the Fgf8 enhancer, and Fgf8 expression would be expanded rostrally, which is what we observe.

We hypothesized that the boundary-setting gene Ripply2 would play an important role downstream of RARβ2 to explain the disorganized and muddled somite boundaries in Rarβ2 MO embryos. Ripply1 and Ripply2 are spatially regulated by retinoids (Janesick et al., 2014; Moreno et al., 2008) and Ripply1 mutants lack both a notochord and myosepta (Kawamura et al., 2005), which could alter chevron morphology (Rost et al., 2014). The Ripply1 gene is absent in Xenopus (Janesick et al., 2012), but Ripply2 has been duplicated to generate two syntenic genes, Bowline and Ledgerline. We found some phenotypic differences when overexpressing Xenopus Ripply2 mRNAs: Ledgerline mRNA was a stronger inducer of axial defects, whereas Bowline mRNA was more effective at inhibiting notochord formation as marked by Xnot expression.

The correlation of rostral expansion and shifting of the unsegmented and segmented PSM in Rarβ2 MO embryos may be ascribed to their closely connected gene regulatory networks (reviewed by Dahmann et al., 2011). Tbx6 induces expression of Ripply1 (zebrafish) and Ripply2 (Xenopus) (Windner et al., 2015; Hitachi et al., 2008). Xenopus Tbx1 promotes expression of Ripply3 in the pre-placodal ectoderm (Janesick et al., 2012), and murine Tbx6 protein is responsible for setting the anterior boundary of Mesp2 expression (Oginuma et al., 2008). Therefore, Rarβ2 MO-induced rostral expansion of Tbx6 should alter somitomere position in the same anteroposterior direction. The standard model is that Tbx6 upregulates Mespa, setting the rostral somite border, which upregulates Ripply1 and Ripply2 to shut off Tbx6 expression, thus setting the caudal border (reviewed by Dahmann et al., 2011). We wanted to know whether Tbx6 transcriptional activity could be modulated by Ripply2. We found that Ripply2 converts Tbx6 to a transcriptional repressor, and that this is dependent on the WRPW (Groucho-interacting) and FPVQ (T-BOX interacting) domains of Ripply2. Hence, in areas where Tbx6 and Ripply2 are co-expressed, their interaction converts Tbx6 into a transcriptional repressor that restricts its targets (presumably Mespa) and defines the posterior somitomere boundary. RARβ2 loss of function results in the improper spatial positioning of Tbx6, Mespa and Ripply2, and widening of the Mespa and Ripply2 domains, thus impairing this intricate regulatory pathway and disrupting somite boundaries.

Anterior expansion of the PSM has been linked to the creation of smaller somites, which is attributable to increased proliferation, and a decreased number of cells differentiating into somites (Dubrulle et al., 2001; Hubaud and Pourquié, 2014). By a similar rationale, a larger PSM domain can cause somites to be displaced anteriorly, making fewer somites, as observed in Shisa2-MO embryos (Nagano et al., 2006). Although we observed reduced somite number in RARβ2 loss-of-function embryos, we found that the somites were larger, not smaller. We infer that the PSM is a limiting pool of cells; therefore, if more cells are being incorporated into each somite, then the final somite number will necessarily be reduced. One plausible explanation is that Rarβ2 MO extends or slows the segmentation clock period by modulating Notch signaling. Our previous microarray data have revealed that manipulation of RA signaling leads to significant changes in expression of oscillatory genes such as Hes9.1, Hes3.3, Hes7.1, Hes5.2, Hey1 and Hes2 (GEO Accession Number GSE57352). Increases in somite size have been observed in zebrafish and mouse, and are confusingly attributed to decreased and increased Notch signaling, respectively (reviewed by Oates et al., 2012). Notch-RAR crosstalk in somitogenesis has been poorly studied, although RA functions upstream of Notch signaling in primary neurogenesis (Franco et al., 1999). Importantly, as Delta-1 is a direct target of Tbx6 (White and Chapman, 2005), expansion of the Tbx6 expression domain or protein levels could manipulate the speed of the clock. The current model in the field holds that there is very tight coordination between the clock and wavefront position (reviewed by Wahi et al., 2016). Our data suggest that loss of RAR signaling might uncouple this connection. A future point of investigation will be to test whether manipulating Notch can rescue the Rarβ2 MO somitogenesis phenotype. The possibility that RAR signaling could potentially control both the somitogenesis wavefront position and the timing of somite differentiation is intriguing and worthy of further investigation in the future.

Although it was known that Rarβ is expressed in somites and lateral plate mesoderm (Cui et al., 2003; Ruberte et al., 1991; Romeih et al., 2003; Bayha et al., 2009), its role in somitogenesis was not addressed. A phenotype we observed with 100% penetrance in Rarβ2 MO-injected embryos is the loss of chevron-shaped somite morphology, the appearance of thicker and fewer, U-shaped somites and inhibition of hypaxial muscle migration. The signature chevron somite shape is found in all aquatic creatures, and is indispensable for locomotion (Rost et al., 2014). RAR loss-of-function-induced paralysis in Xenopus was previously attributed to loss of primary neurons (Janesick et al., 2013; Sharpe and Goldstone, 1997; Blumberg et al., 1997). However, other factors that control locomotion (e.g. notochord integrity, neuromuscular junctions, central pattern generators, myotome differentiation and specialization) might also be contributing to movement defects in RAR mutants or RAR antagonist-treated embryos. The loss of chevron morphology in the somites may also contribute to the paralysis phenotype commonly observed with RAR loss of function.

In zebrafish, somite chevron morphology is attributed to the proper development of notochord and horizontal myoseptum that separates the hypaxial and epaxial myoblast lineages. Myosepta are laminar tendons that foster the attachment of notochord to somite muscle (Bassett and Currie, 2003) and are eventually populated by slow heavy chain fibers (Brent and Tabin, 2004). Myoseptum and adaxial/pioneer cells are not anatomy terms in Xenopus (Karpinka et al., 2015), and probes designed against genes that mark these structures in zebrafish did not mark them in Xenopus. The early compartmentalization of muscle in teleosts (fast versus slow, hypaxial/epaxial, adaxial/lateral) is not necessarily conserved in other vertebrates. Amniotes have a somitic architecture characterized as ‘peppered’ with respect to the physical locations of myotome subtypes (Brennan et al., 2002). Lampreys lack a myoseptum and adaxial cells (Hammond et al., 2009) despite possessing chevron-shaped somites (Rost et al., 2014). Differences in Xenopus and zebrafish have already been recognized with regard to somite rotation, intersomitic boundary formation and other early morphogenic movements in somitogenesis (Afonin et al., 2006; Henry et al., 2005; Leal et al., 2014). This could contribute to the difficulty in making direct comparisons between zebrafish and Xenopus with regard to myosepta, adaxial cells and chevron morphology.

In Rarβ2 MO tadpole stage embryos, normal ventral migration of hypaxial myoblasts from the dermomyotome is inhibited. Rarβ2 is undetectable in newly developing somitomeres but predominantly expressed in the anteriormost 8-10 trunk somites, the same somites that contribute to hypaxial myoblast migration (Martin and Harland, 2001). Hypaxial delamination is RA dependent (Mic and Duester, 2003), and our results suggest that RARβ2 is the receptor subtype modulating this process. We demonstrated that expression of Tbx3, a RAR direct target (Ballim et al., 2012) and hypaxial marker (Martin et al., 2007), is diminished in RARβ2 morphants. Tbx3 is normally found in discrete patches of cells, ventral to and separate from the somites in stage 40 tadpoles. Rarβ2 loss of function disrupts this pattern, which was not attributable to developmental delay of migration.

In Rarβ2 MO tadpole embryos, melanophore and hypaxial myoblast migration are inhibited, similar to the observation that Pax3 ‘Splotch’ mutants lack both hypaxial muscle and melanocyte migration to the ventral belly (Auerbach, 1954; Brown et al., 2005). Melanophore and hypaxial migration are coordinated in Xenopus (Martin and Harland, 2001), although it has been proposed that melanophore and neural crest migration are not required for hypaxial migration (Martin and Harland, 2001). Both migrations are lost in RARβ2 loss of function, implying the existence of an upstream signal or developmental event connecting these processes. We conclude that Rarβ2 is required for migration of hypaxial muscle, which could have implications for the proper patterning of hypaxial-derived structures, such as the rectus abdominus, limbs and tongue.

Loss of chevron morphology and hypaxial muscle was often accompanied by blurred somite boundaries and/or disordered somites; the epaxial and hypaxial domains, normally joined at the chevron apex, appeared to be disconnected from each other. An intriguing area for future study is how confusion of rostral/caudal polarity, caused by dysregulation of the Ripply2/Tbx6/Mespa transcriptional network early during somitogenesis, affects later development of epaxial/hypaxial muscle. This concept has been partly explored in zebrafish (Hollway et al., 2007). Our results suggest that the two processes are connected, because somite boundaries are disorganized and Ripply2/Tbx6/Mespa are misexpressed at early stages, whereas hypaxial myoblasts are absent (as marked by Tbx3) and somites continue to be disordered at later stages in RARβ2 loss of function embryos.

The requirement for retinoic acid signaling in somitogenesis is an established principle in developmental biology. Most attention has focused on how RA establishes the determination wavefront by antagonizing FGF and Wnt, restricting the PSM and caudal progenitor pool. Less was known about how RA regulates somite patterning or specification of myotome subdomains. Fig. 8 summarizes the findings in this paper. We show that Rarβ2 is properly positioned to promote somitogenesis, and that it is the receptor subtype most likely to be responding to RA emanating from the trunk. The somitogenesis phenotype of RARβ2 loss of function is nuanced, but specific. In contrast to the RARγ loss of function phenotype, which reduces the PSM and the caudal progenitor pool, RARβ2 loss of function expands the PSM producing fewer and larger somites that lack chevron morphology and distinct boundaries. Migration of both melanophores and hypaxial myoblasts is completely inhibited in RARβ2 morphants. Ripply2, a recognized player in boundary setting, is rostrally shifted and broadened when RARβ2 is lost. Ripply2 controls the transcriptional activity of Tbx6 (PSM marker) but not Tbx3 (hypaxial marker). Hence, RARβ2 positively regulates Tbx3 to promote hypaxial muscle migration, but negatively regulates Tbx6 to restrict the PSM.

All experiments were approved by UC-Irvine IACUC. Xenopus eggs were fertilized in vitro and embryos staged as described previously (Janesick et al., 2012). Embryos were injected bilaterally or unilaterally at the two- or four-cell stage with gene-specific morpholinos (MO) (Table S1) and/or mRNA together with 100 pg/embryo β-galactosidase (β-gal) mRNA lineage tracer (LT). Embryos were maintained in 0.1× MBS until controls reached appropriate stages. Embryos processed for whole-mount in situ hybridization were fixed in MEMFA, stained with magenta-GAL (Biosynth) and stored in 100% ethanol (Janesick et al., 2012).

Whole-mount in situ hybridization was performed as previously described (Janesick et al., 2012). Rarβ1, Rarβ2 (Session et al., 2016), Xa-1 (Hemmati-Brivanlou et al., 1990), Ripply2/Bowline (Kondow et al., 2006), Mespa (Sparrow et al., 1998), Mesogenin1 (Joseph and Cassetta, 1999), Tbx3 (Li et al., 1997), Tbx6 (Uchiyama et al., 2001), Esr5 (Jen et al., 1999), Myod (Hopwood et al., 1989), Cxcl12a (Braun et al., 2002), En1 (Watanabe et al., 1993) and Mybpc1 (Session et al., 2016) probes were prepared via PCR amplification of coding regions incorporating a 3′ bacteriophage T7 promoter. pCS2-Fgf8 (a gift from Nancy Papalopulu) was linearized with BamHI. Relevant primers and restriction enzymes are listed in Table S2. Probes were transcribed with MEGAscript T7 (Life Technologies) and digoxigenin-11-UTP (Roche). Embryos stained with Myod by whole-mount in situ hybridization were cleared in Histoclear and embedded in Paraplast+ using Sakura Finetek disposable base molds (15 mm×15 mm×5 mm) and yellow embedding rings. Serial coronal sections (8 µm) were mounted onto Superfrost+ microscope slides, dried overnight, dewaxed with xylene and photographed under bright-field illumination.

Microinjected embryos were treated at stage 7/8 with TTNPB (a RAR agonist), AGN193109 a (RAR-selective antagonist) or 0.1% ethanol (vehicle) in 0.1× MBS as described previously (Janesick et al., 2012) and aged until control embryos reached tailbud stage. Embryos from each treatment were randomly separated into groups of five embryos (each group of five embryos was taken as one biological replicate, n=1) and homogenized in 200 μl TriPure (Roche). Total RNA was DNAse treated, LiCl precipitated, reverse transcribed into cDNA and quantitated in a Light Cycler 480 System (Roche) using the primer sets listed in Table S3 and SYBR green detection. Each primer set amplified a single band, as determined by gel electrophoresis and melting curve analysis. QPCR data for RAR staging were analyzed by ΔCt relative to Histone H4 and corrected for amplification efficiency between RARs (Pfaffl, 2001). QPCR data for fold induction by RAR agonist and antagonist were analyzed using the ΔΔCt method relative to Eef1a1, normalizing to control embryos (Schmittgen and Livak, 2008). Error bars represent biological replicates calculated using standard propagation of error (Bevington and Robinson, 2003).

pCDG1-Tbx3, pCDG1-Tbx6 and pCDG1-Ripply2 were constructed by PCR amplification of the Xenopus laevis Tbx6 (Uchiyama et al., 2001), Bowline (Kondow et al., 2006) or Ledgerline (Chan et al., 2006) protein-coding regions. pCDG1-Ripply2 mutant constructs were made by two-fragment PCR to generate the WRPW→AAAA and/or FPVQ→AAAA substitutions using the primers in Table S4. All constructs were cloned into the NcoI-BamHI site of pCDG1, sequence verified and linearized with NotI. The 5′-capped mRNA was transcribed using T7 mMESSAGE mMACHINE Kit (Thermo Fisher Scientific). pGL3-Basic-βRARE-Luciferase constructs were made by two-fragment PCR using primers in Table S5 and sequence verified. The (TBRE)₂-TK-luciferase reporter has been previously described (Janesick et al., 2012).

Embryos were microinjected unilaterally at the 2- or 4-cell stage with the βRARE-Luciferase reporter DNA, then treated as above at stage 7/8 with TTNPB or 0.1% ethanol (vehicle). When control embryos reached neurula stage, they were separated into groups of 10 embryos, homogenized and processed for luciferase assays as described previously (Janesick et al., 2014). Embryos were microinjected unilaterally at the 2- or 4-cell stage with the (TBRE)₂-TK-Luciferase reporter DNA, and different combinations of Tbx3, Tbx6 and Ripply2 (Bowline or Ledgerline) mRNA. When the embryos completed gastrulation, they were separated into groups of five embryos, homogenized and processed as described (Janesick et al., 2014).

βRARE oligonucleotides (Table S6) were diluted in 1× buffer M (Roche) to 2.5 μM, heated to 95°C and annealed slowly. Free 5′-OH termini (10 pmol) was labeled with [γ-³²P] ATP (6000 Ci/mmol, 10 mCi/ml) using T4 PNK enzyme (Roche). Unincorporated label was removed using ProbeQuant G-50 microcolumns (Amersham Biosciences). RAR and RXR mRNA was synthesized using mMessage Machine T7 (Ambion) from NotI-linearized templates. Trace-labeled ³⁵S-Met-RAR and ³⁵S-Met-RXR protein was synthesized from RNA using Retic Lysate IVT Kit (Ambion). Binding of DNA and protein was performed as described previously (Umesono et al., 1991) and analyzed on 6%, non-denaturing polyacrylamide (29:1) gel, 0.5× TBE. Each gel was exposed overnight in a phosphorimaging cassette and visualized by phosphorimaging with a Typhoon PhosphorImager (GE HealthCare).

We thank Kristin Ampig and Jingmin Zhou for technical assistance with paraffin wax sectioning and the cloning of Bowline and Ledgerline constructs.