Active repression by RARγ signaling is required for vertebrate axial elongation

Repression mediated through unliganded retinoic acid receptors (RARs) is an important yet understudied function exhibited by nuclear receptors (reviewed by Weston et al., 2003). Although RA plays a major role in patterning the hindbrain, retina, placodes and somites, its absence is crucial for the development of structures found at the head and tail of the embryo. RARs exhibit basal repression in the absence of ligand, binding constitutively to their targets, recruiting co-repressors, and actively repressing the basal transcriptional machinery (Chen and Evans, 1995). When ligand is present, co-repressors are replaced by co-activators and target genes are transcribed (Chakravarti et al., 1996).

We previously demonstrated that repression mediated through unliganded RARs was important for anterior neural patterning, establishing a novel role for RAR as a repressor in vivo (Koide et al., 2001). Overexpression of a dominant-negative RARα expanded anterior and midbrain markers caudally and shifted somitomeres rostrally (Blumberg et al., 1997; Moreno and Kintner, 2004). Exogenous RA, constitutively active RARα or derepression of RARα produced the opposite effect: severe anterior truncations, diminished anterior markers, and anteriorly shifted midbrain and hindbrain markers. Stabilization of co-repressors resulted in enhanced anterior neural structures and posteriorly shifted mid/hindbrain markers (Koide et al., 2001).

Axial elongation requires continual replenishing of bipotential caudal progenitor cells (maintained by Wnt and FGF signaling, but inhibited by RA) that give rise to notochord, neural tube and somites (Cambray and Wilson, 2002; Davis and Kirschner, 2000). The most stem-like cells are located in the chordoneural hinge (CNH), where the posterior neural plate overlies the caudal notochord (Beck and Slack, 1998). Cells from the CNH contribute to presomitic mesoderm (PSM), which supplies committed somitic precursor cells to the rostral determination wavefront (reviewed by Dequeant and Pourquie, 2008). PSM is initially homogenous and unorganized [expressing Mesogenin1 (Msgn1) and Tbx6], then becomes patterned into somitomeres marked by Thylacine2 (Thyl2) and Ripply2 (reviewed by Dahmann et al., 2011). Epithelialization of presomitic domains results in mature somites (Nakaya et al., 2004).

RA is well known to function in the trunk, where it promotes differentiation of PSM into somitomeres (Moreno and Kintner, 2004). By contrast, RA is actively metabolized and cleared by CYP26A1 in the caudal region (Fujii et al., 1997). Treatment with RA leads to loss of posterior structures (Sive et al., 1990); Cyp26a1^−/− mice exhibit posterior truncations and homeotic vertebral transformations (Abu-Abed et al., 2001; Sakai et al., 2001). Exposing embryos to RA inhibits proliferation of axial progenitor cells in CNH and PSM, leading to axial truncation from premature exhaustion of the progenitor pool (Gomez and Pourquie, 2009). Therefore, RA is normally excluded from unsegmented mesenchyme in PSM and the CNH. RARγ is expressed at high levels throughout the entire caudal region, including CNH and PSM (Mollard et al., 2000; Pfeffer and De Robertis, 1994), yet, based on Cyp26a1 expression, RA is absent (de Roos et al., 1999). The physiological significance of RARγ expression in the embryonic posterior is uncertain. RARγ might function to terminate the body axis at late stages by inducing apoptosis (Olivera-Martinez et al., 2012), but that model would not explain the strong expression of RARγ observed at neurula, continuing through tailbud stages, despite the apparent absence of RA.

Rarγ2 skirts the posterior edge of the determination wavefront and is co-expressed with PSM, CNH and posterior Hox markers. We hypothesized that Rarγ2 serves a dual function: as an activator in somite differentiation but a repressor in the maintenance of PSM and the caudal progenitor pool. Loss of RARγ2 severely shortens the embryo body axis and inhibits somitogenesis. Loss of RARγ2 expands the anterior border of PSM expression near the wavefront (where activation is lost), but diminishes the expression domain of caudal PSM and posterior Hox genes (where repression is lost). Increasing RAR-mediated repression expands the expression of posterior Hox, PSM and CNH markers, creating smaller somitomere domains via an indirect, ‘repressing a repressor’ mechanism. Relief of repression results in a truncated body axis with decreased PSM and CNH markers. Axial extension and segmentation in vertebrates relies on the maintenance of unsegmented PSM mesenchyme and replenishing of caudal progenitor cells. Our data show that RARγ2 plays a crucial role in this process, repressing target genes to maintain PSM and caudal progenitors in the absence of RA, while activating others to promote somitogenesis in the presence of RA.

We showed previously that active repression of RAR target genes by unliganded RAR is required for head formation (Koide et al., 2001). Treatment with the pan-RAR inverse agonist AGN193109 increased the expression of genes involved in patterning anterior neural structures, whereas treatment with pan-RAR agonist TTNPB decreased the expression of anterior marker and cement gland-specific genes (Koide et al., 2001), revealing a set of genes specifically upregulated/downregulated by TTNPB (Arima et al., 2005). Validation studies identified a subset upregulated by AGN193109. We hypothesized that active repression by unliganded RARs is biologically important and designed an experiment to identify genes upregulated or downregulated by modulating repression. Percellome analysis (Kanno et al., 2006) quantified the copy number per embryo of all genes represented on Affymetrix Xenopus microarray v1.0. Among these we identified a collection of genes linked to the maintenance of caudal axial progenitors that were downregulated by TTNPB and upregulated by AGN193109 (Table 1). RAR-mediated repression upregulates the steady-state expression of posterior Hox paralogs 9-13 and genes found in both unsegmented PSM and CNH. Thus, we hypothesized that RAR is a repressor required for axial elongation.

The ability of unliganded RARs to behave as repressors is well documented, although not all human receptor subtypes can recruit co-repressors (e.g. SMRT) in the absence of ligand (Wong and Privalsky, 1998). We tested the ability of Xenopus RAR (xRAR) subtypes to repress basal activity of a luciferase-dependent reporter using the GAL4-RAR system (supplementary material Fig. S1D-F) (Blumberg et al., 1996). Xenopus RARα, RARβ and RARγ suppressed basal activity in vitro and in vivo (supplementary material Fig. S1A,C), whereas human RARβ and RARγ did not (supplementary material Fig. S1B). Thus, xRARs can function as repressors in the absence of ligand.

Whole-mount in situ hybridization (WISH) revealed that Rarγ2 is the predominant isoform expressed in the Xenopus embryonic posterior (supplementary material Fig. S2A). In late neurula and early tailbud stage embryos, Rarγ2 is strongly expressed in the anterior and posterior, but almost undetectable in the trunk. Rarγ2 expression later becomes pronounced in the tail and head, particularly in hyoid, branchial and mandibular neural crest. Rarγ1 is expressed similarly. QPCR analysis revealed that Rarγ2 is 1000- to 4000-fold more abundant than Rarγ1 at stages 10-22, and 100- to 400-fold more abundant at all other stages analyzed (supplementary material Fig. S2B). Subsequent experiments utilized Rarγ2-selective reagents. We conclude that Rarγ2 is the predominant isoform expressed in the posterior region of embryos.

Rarγ2 is expressed where RA is probably absent (owing to CYP26A1 expression). Key posterior genes were upregulated by AGN193109. We hypothesized that RARγ2 posterior to the wavefront is a repressor, maintaining unsegmented PSM and the progenitor cell pool required for axial elongation. We used double WISH to compare the expression of Rarγ2 with that of Hoxc10, an important member of the Abd-B Hox gene family promoting caudal development over thorax (Lamka et al., 1992). Rarγ2 expression completely overlaps caudal Hoxc10 expression (Fig. 1E,H) but not the anteriormost neural or lateral plate expression of Hoxc10 (Fig. 1E,H). These data position Rarγ2 as a potential regulator of posterior Hox genes and the caudal body plan.

We next defined the anterior limit of Rarγ2 expression relative to the determination wavefront. Myod is a general muscle marker abutting and partially overlapping Rarγ2 expression (Fig. 1A,B). Thyl2 and Ripply2 mark somitomeres, which are prepatterned PSM domains containing non-epithelialized, immature somites (Tam et al., 2000). Thyl2 and Ripply2 are only expressed in newly forming somitomeres and are assigned negative Roman numerals (S–I, S–II, etc.) versus mature somites (SI, SII, etc.) (Pourquie and Tam, 2001). Msgn1 (Buchberger et al., 2000) is expressed caudal to Thyl2 and Ripply2, marking non-patterned PSM-containing cells committed to the somitic fate (Nowotschin et al., 2012). Tbx6 is also expressed in PSM, but unlike Msgn1 its expression domain overlaps with somitomeres (Hitachi et al., 2008). Rarγ2 and Msgn1 are synexpressed at neurula (Fig. 1F,G) and tailbud (supplementary material Fig. S3) stages; Tbx6 expression overlaps Rarγ2 but extends rostrally beyond the Rarγ2 domain (Fig. 1C,D; supplementary material Fig. S3). Anterior expression of Rarγ2 mRNA ends at an RA-responsive region (supplementary material Fig. S4), coinciding with the most posterior somitomere domain (S–III) of Thyl2 or Ripply2 (Fig. 1I-M), thus skirting the posterior edge of the wavefront.

xNot, a notochord marker that regulates trunk and tail development, is concentrated in the extreme posterior notochord and floor plate by late neurula (von Dassow et al., 1993) and is often employed as a CNH marker in Xenopus (Beck and Slack, 1998) to reveal the location of bipotential stem cells (Cambray and Wilson, 2007; Takemoto et al., 2011). xNot is co-expressed with Rarγ2 (Fig. 1K), agreeing with data suggesting that Rarγ2 is present in CNH (Pfeffer and De Robertis, 1994). The double WISH data are consistent with Rarγ2 functioning as an activator near where RA is present at the wavefront, yet as a repressor where it coincides with Msgn1, xNot and Cyp26a1.

To separate the effects of RARγ in the posterior from RARα in the trunk, we characterized RARγ-selective agonist NRX204647 (4647) (Shimono et al., 2011; Thacher et al., 2000) and RARγ-selective inverse agonist NRX205099 (5099) (Tsang et al., 2003) in Xenopus embryos. Like AGN193109, 5099 is an inverse agonist, reducing RARγ signaling activity below basal levels by stabilizing the co-repressor complex bound to RARγ. Embryos treated with 1 µM agonist 4647 become primarily trunk (no head or tail structure), while 0.1 µM perturbs axial elongation (supplementary material Fig. S5), producing anterior truncations characteristic of RAR activators (Sive et al., 1990). Inverse agonist 5099 at 1 µM delayed development, producing enlarged heads and shortened trunks; half the dose elicited similar but weaker phenotypes, with effects absent at 0.1 µM (supplementary material Fig. S5). Treating neurula embryos significantly reduced severity but did not eliminate the phenotype (supplementary material Fig. S5).

To test the effects of these chemicals in vivo without interference from endogenous RARs, we mutated the DNA-binding specificity of a full-length RAR, RAR^{EGCKG→GSCKV}. The mutant receptor recognizes a mutant TK-luc reporter, (RXRE^1/2-GRE^1/2)×4 TK-luc, to which endogenous RARs do not bind (Klein et al., 1996). In transient transfection assays, 4647 selectively activated RARγ at doses below 0.1 µM (supplementary material Fig. S6A). Similarly, 5099 selectively antagonized 10 nM 9-cis RA activation of RARγ below 0.1 µM (supplementary material Fig. S6B). We conclude that 4647 and 5099 behave as subtype-selective ligands to activate or repress RARγ.

We hypothesized that 4647 treatment of embryos would decrease posterior Hox gene expression and markers of PSM, whereas 5099 would produce the opposite effect. Microarray analysis (Table 1) revealed that Hoxc13 and Hoxc10 expression was upregulated by inverse agonist AGN193109 and downregulated by agonist TTNPB. We infer that increased expression of Hoxc13 and Hoxc10 results from RAR repressing the expression of a repressor of their expression. The expression pattern of Hoxc13 (supplementary material Fig. S7) was not previously characterized.

We began soaking embryos in RARγ-selective doses of 4647, 5099 or vehicle control after gastrulation (stage 12.5) to focus on axial elongation. Treatment with 10 nM 4647 resulted in diminished caudal structures at stage 40 (supplementary material Fig. S5), reducing expression domains of Hoxc10, Hoxd10 and Hoxc13 (Fig. 2A-C). Conversely, treatment with 0.5 µM 5099 expanded their neural and lateral domains (Fig. 2A-C). To determine short-term effects of chemical treatments, we soaked embryos for 1 h at various stages and evaluated Hoxc10 expression (supplementary material Fig. S8) and that of Tbx6 (not shown) at stage 22. Repression by 5099 is required at early neurula, whereas activation by 4647 is required at mid- and late neurula stages for expected expansion and reduction, respectively, of Hoxc10 expression (supplementary material Fig. S8). Higher, non-receptor-selective doses exacerbated effects on posterior Hox genes (supplementary material Fig. S9), suggesting that RARγ2 is the primary mediator. Hoxc10 nearly abuts Krox20, demonstrating trunk shortening in 5099-treated embryos (supplementary material Fig. S9G,H). High doses of 4647 create embryos lacking anterior and posterior structures, as indicated by the absence of mid/hindbrain markers En2 and Krox20 and of posterior gene Hoxc10 (supplementary material Fig. S9C-F).

Msgn1 and Tbx6 were upregulated by inverse agonist and downregulated by agonist in the microarray analysis (Table 1). Msgn1 and Tbx6 domains were reduced at tailbud stages by post-gastrulation treatment of embryos with 4647, whereas expression was expanded in embryos treated with inverse agonist 5099 (Fig. 2D,E). However, in neurula stage embryos, 4647 reduced Msgn1 expression while Tbx6 expression was expanded (Fig. 3E,F,O,P). Expression of Tbx6 and Msgn1 was expanded by 5099 (Fig. 3I,J,Q,R), an effect that was more pronounced at higher doses (supplementary material Fig. S10I,J,Q,R). Somitomere markers Thyl2 and Ripply2 showed thicker domains; S–III expanded to the posteriormost edge of the embryo where somites are not found in controls (Fig. 3G,H). At non-receptor-selective doses, 4647 exacerbated the phenotypes of Msgn1, Tbx6 and Ripply2 (supplementary material Fig. S10E,F,H,O,P) and promoted ectopic expression of Thyl2 in the midline, with somitomeres occupying nearly the entire anteroposterior axis (supplementary material Fig. S10G). By contrast, 5099 treatment produced fewer, thinner somitomeres (Fig. 3K,L), an effect more pronounced at higher doses (supplementary material Fig. S10K,L).

Since Rarγ2 is co-expressed with Msgn1, we expected that 4647 would reduce and 5099 would expand Rarγ2 expression. Rarγ2 expression was expanded by inverse agonist and reduced by agonist (Fig. 2F) as verified by QPCR (supplementary material Fig. S11), which is surprising given that other receptor subtypes (RARα2 and RARβ2) are induced by agonist (Leroy et al., 1991; Sucov et al., 1990). The data indicate that 5099 enhances repression by RARγ, increasing caudal gene expression, whereas 4647 relieves repression by RARγ, diminishing caudal gene expression.

Treatment with 4647 activates RARγ and removes repressors from RARγ targets, creating posterior truncations. We hypothesized that loss of RARγ2 would phenocopy 4647 treatment once RARγ2-mediated repression was lost. We designed AUG MOs to capture both pseudoalleles of Rarγ2. Knockdown of RARγ2.1/2.2 resulted in loss of Hoxc10, Hoxd10, Hoxa11 and Hoxc13 expression, together with severe curvature and reduction of the injected side (Fig. 4A-D). Microinjection of splice-blocking MO capturing both pseudoalleles of Rarγ2 reduced the expression of Rarγ2 as measured by QPCR, phenocopying the AUG MOs (supplementary material Fig. S12). We demonstrated that axial truncation on the injected side was not due to developmental delay (supplementary material Fig. S13). To establish that RARγ2 is solely responsible for the axial truncations and reduction in posterior Hox and PSM domains, we showed that Rarγ2 MO can only be rescued with Rarγ2, but not Rarα2 or Rarβ2, mRNA (Fig. 5). RARγ2 knockdown reduced and shifted the expression of Msgn1 and Tbx6 anteriorly along the midline (Fig. 4E,F,I-J′) and caused an anterior shift in the paraxial domains of Thyl2 and Ripply2, while obliterating lateral expression (Fig. 4G,H). The complexity of the Rarγ2 MO phenotype is likely to be due to the fact that RARγ2 knockdown both disrupts its repressive function in the absence of ligand and its activation in the presence of ligand, particularly near the determination wavefront.

When the dominant-negative co-repressor c-SMRT is overexpressed, it binds RAR and blocks recruitment of co-repressors (Chen et al., 1996). We identified several c-SMRT isoforms from Xenopus, selecting that most similar to human c-SMRT that we used previously. Microinjection of Xenopus laevis (Xl) c-smrt mRNA relieved repression by GAL4-xRARγ in whole embryos (supplementary material Fig. S14). This effect was potentiated by addition of 1 µM TTNPB (supplementary material Fig. S14). Overexpression of Xl c-smrt mRNA caused significant reductions in the neural and lateral domains of Hoxc10 and Hoxd10 (Fig. 6B,D). Xl c-smrt also reduced Hoxc13, Tbx6, Msgn1 and xNot (Fig. 6F,H,H′,J,J′,L). Similar to Rarγ2 MO, moderate truncation of injected axes was observed in 70% of embryos, but the midline, rostral shifting of Tbx6 and Msgn1 (as in Rarγ2 MO embryos) was minimal. We conclude that Xl c-SMRT relieves repression of Rarγ2, causing loss of progenitor and PSM cells and posterior Hox gene expression.

Another method for relieving repression is overexpression of constitutively active VP16-RARγ2 (RARγ2 fused to the VP16 activation domain). Microinjection of VP16-Rarγ2 mRNA led to a truncated axis on the injected side in 100% of embryos and loss of Hoxc10, Hoxd10, Msgn1 and Tbx6 expression (Fig. 7). These embryos were less curved than Rarγ2 MO-injected or c-smrt-injected embryos, but rostral expansion of neural/midline and lateral domains was consistently observed, similar to Rarγ2 MO embryos.

Treatment with 4647 or microinjection of c-smrt or VP16-Rarγ2 mRNA relieved repression by RARγ, increasing RAR signaling, decreasing posterior Hox and PSM markers. Decreasing RAR signaling should produce the opposite effect. We microinjected mRNA overexpressing the RA catabolic enzyme CYP26A1 and observed rostral shifts in the lateral and neural expression domains of Hoxc10 and Hoxd10 (supplementary material Fig. S15). Microinjection of dominant-negative (DN)-RARγ2 should phenocopy 5099 treatment because co-repressors would be retained on RARγ2 targets, leading to repression. Overexpression of DN-RARγ2 increased the expression of Msgn1 and Tbx6 in both lateral and paraxial domains, and shifted xNot expression rostrally (Fig. 8B,D,F). DN-RARγ2 phenocopied the effects of Cyp26a1 mRNA (Moreno and Kintner, 2004) on somitomere markers Thyl2 and Ripply2; rostral shifting and knockdown of somitomere expression was the phenotype that we observed (Fig. 8H,J,K).

Microinjection of Rarγ2 MO alone resulted in knockdown of Hoxc10 and axial truncation (Fig. 9A,B,E). We hypothesized that this phenotype was due to loss of repression, reasoning that the phenotype should be rescued with DN-RARγ2. Axial defects and lateral knockdown of Hoxc10 expression were partially recovered with DN-Rarγ2 mRNA (Fig. 9C,D,E). The neural domain of Hoxc10 expression was rescued in nearly all embryos and a rostral shift often observed. We conclude that increasing repression with DN-RARγ2 or overexpressing CYP26A1 (removing ligand) promotes caudal gene expression, similar to chemical treatment with 5099. Moreover, loss of caudal structures and gene expression due to Rarγ2 MO are rescued by restoring repression with DN-RARγ2.

Most studies consider only one aspect of RAR signaling, namely its role as a ligand-activated transcription factor promoting the expression of target genes. In developmental biology, RA signaling has been studied extensively for its ability to promote differentiation and establish boundaries in somitogenesis, neurogenesis and rhombomere segmentation (reviewed by Rhinn and Dolle, 2012). Liganded RAR has been predicted to function passively in the caudal region until required to facilitate body axis cessation (Olivera-Martinez et al., 2012), when somitogenesis is nearing completion because the determination wavefront, moving the RA source caudally, has exhausted the progenitor cell pool (Gomez and Pourquie, 2009). Here, liganded RARγ would function as an activator promoting apoptosis (Shum et al., 1999) at terminal tailbud stage. However, this does not address why RARγ2 would be highly expressed where RA is presumed absent due to CYP26A1 expression. Here we show that RARγ is engaged in all stages of caudal development, not solely as a terminator of the body axis. RARγ functions as an unliganded repressor required for the maintenance of the posterior PSM and progenitor cell population that allows axial elongation (Fig. 10). RARγ acts as a liganded activator in the anterior, segmented PSM to facilitate somite differentiation (Fig. 10). Repression mediated by the unliganded receptor–co-repressor complex constitutes a novel mechanism by which posterior markers are upregulated during axial elongation in Xenopus embryos.

Our microarray results suggest that axial elongation is regulated by RAR-mediated repression. Enhancing repression with AGN193109 upregulated, and activation of RAR by TTNPB downregulated, many posterior Hox, PSM and CNH genes in neurula stage embryos. We identified AGN193109-upregulated genes expressed in PSM (Table 1) that are mostly absent from regions of somite maturation (Blewitt, 2009; Yoon et al., 2000). The CNH markers xBra3 and xNot were also upregulated by AGN193109, thus both PSM and CNH markers were upregulated by enhancing RAR repression and downregulated by increasing RAR activation. Current literature suggests the existence of a negative-feedback loop between these two populations of cells: Msgn1 is induced by Brachyury and Wnt8 in CNH but represses their expression to promote PSM fates (Fior et al., 2012; Yabe and Takada, 2012). Our results support a novel role of RAR repression in the maintenance of cells in both unsegmented PSM and stem-like CNH.

We showed that X. laevis RARα, RARβ and RARγ can repress basal transcriptional activity in the absence of RA and examined whether this repression is physiologically relevant in caudal development. Rarγ2 is expressed in embryonic regions where it might actively repress genes involved in axial elongation. Rarγ2 is synexpressed with the PSM marker Msgn1 and overlaps with Tbx6, Hoxc10, the S–III domains of Thyl2 and Ripply2, and the CNH marker xNot. By contrast, Rarγ2 is expressed at low levels in trunk (where Myod and Rarα are expressed) and in the anterior, segmented PSM expression domains of Thyl2 and Ripply2. Since absence of RA is required for the proliferation and/or survival of caudal PSM and CNH cells, the presence of RARγ in posterior tissue would be contradictory if it functioned as an activator. We infer that RARγ acts as a repressor throughout unsegmented PSM and CNH where RA is absent, but as an activator of somitomere markers near the differentiation wavefront where Rarγ2 overlaps with S–III and where Raldh2 expression indicates the presence of RA. It remains unknown what repressors RARγ targets to indirectly upregulate caudal genes. One possibility is that RARγ represses Ripply2, which functions to repress Tbx6 (reviewed by Dahmann et al., 2011), as supported by the observation that increasing activation with 4647 expands Ripply2 posteriorly. Hence, RARγ would normally function in the posterior to repress Ripply2, therefore promoting expression of Tbx6.

Since high doses of 4647 result in embryos consisting largely of trunk, it is predictable that nearly the entire embryo differentiated into somitomeres (with thicker boundaries). At lower, RARγ-selective 4647 doses, somitomeres were shifted posteriorly and thickened. This phenotype, which is also seen with RA treatment or FGF inhibition by SU5402, was attributed to increased numbers of cells allocated to somitomeres and a decreased progenitor pool (Dubrulle et al., 2001; Moreno and Kintner, 2004). 5099 upregulates both Tbx6 and Msgn1, indicating that unsegmented PSM is expanded by increased RAR repression. However, we note distinct differences in the effects of 4647 on Tbx6 versus Msgn1. Tbx6 is upregulated by 4647 at early stages but downregulated at later stages, as also observed for the T-box gene Tbx1 (Janesick et al., 2012). Unlike Msgn1, Tbx6 plays a dual role in the unsegmented PSM and the determination front where it controls the anteroposterior patterning of somitomeres via Ripply2 (Hitachi et al., 2008).

Msgn1 expression does not overlap somitomeres and functions to maintain unsegmented PSM by encouraging the differentiation of caudal stem cells. Loss of Msgn1 expression leads to smaller somitomeres owing to the accumulation of bipotential progenitor cells that have not received signals to commit to PSM fate (Fior et al., 2012; Yabe and Takada, 2012). Treatment with 4647 also leads to loss of Msgn1 and thus somitomeres should be smaller; however, they are larger. Despite such divergent early stage phenotypes, Msgn1^−/− embryos (Yoon and Wold, 2000) and 4647 embryos both display fewer somites and reduced caudal structures at late stages. Caudal progenitors cannot be instructed to become somites in Msgn1^−/− embryos. In 4647-treated embryos, the pool is expeditiously transformed into thickened somitomeres early, but the progenitor supply is exhausted before axial elongation is complete, reducing somitomere numbers. That 4647 can differentiate somitomeres at all without Msgn1 is intriguing. Either Tbx6 compensates for Msgn1 knockdown, or 4647 can induce uncommitted, non-PSM progenitor cells to differentiate into somitomeres.

If RARγ2 functions solely as a repressor, then RARγ2 knockdown should induce a loss of repression phenotype. Rarγ2 MO microinjection resulted in severely truncated body axes with caudal PSM and posterior Hox markers significantly reduced at tailbud stages, similar to 4647 treatment. This phenotype was attributed to axial defects, not merely developmental delay. We noted three differences between 4647-treated and Rarγ2 MO-injected embryos. First, axes of Rarγ2 MO embryos were significantly curved, which was attributed to imbalance/dominance of the uninjected side versus the truncated injected side. Second, caudal PSM markers, while qualitatively reduced with Rarγ2 MO, also expanded rostrally, even when accounting for shortened axes on injected sides. Third, thickened, posteriorly expanded somitomeres were not seen with Rarγ2 MO. RARγ2 acting as an activator near the somitogenesis front where RA is present would explain some discrepancies. RA functions in the determination wavefront to antagonize proliferating PSM and promote somitomere differentiation (Moreno and Kintner, 2004). If RA acts through RARγ2 in the wavefront, then loss of Rarγ2 should expand unsegmented PSM and reduce somitomere expression, exactly as observed.

Axial curvature and loss of Hoxc10 and Msgn1 expression in Rarγ2 MO-injected embryos could be rescued by Rarγ2, but not Rarα2 or Rarβ2 mRNA. Therefore, Rarγ2 is the sole receptor responsible for axial elongation, in agreement with Rarγ2 as the only RAR expressed in caudal domains. Rarβ2 is present only in trunk and pharyngeal arches (Escriva et al., 2006) and Rarα2 is completely absent from the blastopore and surrounding area (see figure S1A,B in the supplementary material of Janesick et al., 2013). Hoxc10 expression could be rescued in Rarγ2 MO-injected embryos by co-injecting DN-Rarγ2 mRNA, definitively establishing that RARγ2 functions as a repressor in the caudal domain. DN-RARγ2 restored Hoxc10 expression, especially in neural tube, where additional rostral expansion was often observed. DN-RARγ2 rescue restored curved axes only partially. We predict that axial curvature is a loss-of-activation effect inhibiting somitomere formation; therefore, the phenotype should not be rescued by DN-RARγ2, but rescued by wild-type RARγ2, as we observed.

Perhaps the most direct method for relieving repression of RARγ2 in caudal regions is overexpression of dominant-negative co-repressor c-SMRT, which binds RARγ2 preventing recruitment of co-repressors and thereby blocking repression. c-SMRT overexpression resulted in truncated axes with loss of posterior Hox, unsegmented PSM and CNH markers, but not rostral shifting of Msgn1 and Tbx6 as had been observed for Rarγ2 MO embryos. This indicates that rostral shifting in Rarγ2 MO embryos results from loss of activation rather than relief of repression. We previously showed that c-SMRT not only relieves repression of RAR but also potentiates ligand-mediated activation (Koide et al., 2001). Since c-SMRT was expressed ubiquitously, it could superactivate RARα or RARγ where RA is present. It should also be noted that c-SMRT can interact with other nuclear receptors and transcription factors. Therefore, we can only conclude that c-SMRT overexpression inhibits maintenance of the caudal PSM and progenitor pool (where RA is absent). We cannot draw conclusions about somitomere markers in c-SMRT overexpression embryos since their expression is controlled by RAR activation, which c-SMRT does not reduce.

We identified a novel function for RARγ as a transcriptional repressor in the regulation of posterior Hox genes. Posterior Hox genes pattern caudal embryonic regions, promote axial elongation (Young et al., 2009) and are linked to cell cycle progression (Gabellini et al., 2003) and therefore proliferation. Axial elongation involves the addition of tissue, as cells must proliferate to contribute segments. Normally, FGF and RA signaling are mutually antagonistic, but we provide evidence that RARγ can support proliferative mechanisms in the absence of RA.

Hox gene expression was altered by 4647 and 5099 treatment, even post-gastrulation. Hence, although Hox gene expression is initiated collinearly during gastrulation, this temporal pattern is not immutable. In support of this model, axial progenitor cells transplanted to anterior locations do not retain their previous Hox identity (McGrew et al., 2008). Furthermore, manipulation of anteroposterior locations of PSM and the determination wavefront resulted in corresponding changes in Hox gene expression (Iimura et al., 2009; Wellik, 2007). We showed that 4647 treatment pushes determination fronts caudally and observed posterior regressions of Hoxc10, Hoxd10 and Hoxc13 expression. Conversely, rostral expansion in PSM by increasing RAR repression was accompanied by anterior shifts in posterior Hox expression. Owing to posterior prevalence, rostral shifts of Hoxc10 or Hoxd10 expression could indicate that thoracic segments will develop caudal structures at later stages. Similarly, rostral shifts in Hoxc13 could drive lumbar segments to sacral morphology. Homeotic transformations from manipulating RAR repression deserve future study.

We conclude that the RAR-mediated repression of caudal genes is crucial for axial elongation, establishing another important role for active repression by nuclear receptors in body axis extension, as previously shown for head formation (Koide et al., 2001). RARγ2 is likely to function as an activator near the determination wavefront and a repressor to maintain axial progenitor pools in the PSM and CNH. As axial elongation nears completion, RARγ2 functions as an activator because the progenitor pool is exhausted and RA comes into close proximity to the caudal domain of RARγ2, where it can then promote apoptosis and terminate the body axis. This model is attractive because it utilizes the same protein to activate or repress target genes depending on the proximity to RA and explains the high levels of posterior RARγ2 expression. RARγ2 is likely to function in multiple steps of somitogenesis and axial elongation (Fig. 10): (1) preservation of undifferentiated states in the progenitor pools (marked by the CNH); (2) maintenance of PSM; (3) initiation of somitomere differentiation; and (4) axial termination. Future studies require RARγ target gene identification because very few ChIP studies have ascertained direct targets, and even fewer studies have explored subtype-selective RAR targets. In the case of inverse agonist-upregulated genes (the focal point of our study), identifying repressors of PSM and progenitors will be key, as these genes are likely to be targeted by unliganded RAR in a classic ‘repression of a repressor’ mechanism.

Xenopus laevis eggs from three different females were fertilized in vitro and embryos staged as described (Janesick et al., 2012). Stage 7 embryos were treated in groups of 25 in 60-mm Petri dishes with 10 ml 0.1× MBS containing 1 μM RAR agonist (TTNPB), 1 μM RAR inverse agonist (AGN193109) or vehicle control (0.1% ethanol). Three dishes per treatment per female were collected (27 dishes total: three technical replicates, three biological replicates per treatment). Each dish of embryos was harvested at stage 18 into 1.5 ml RNAlater (Invitrogen) and stored at 4°C. Samples were homogenized, RNA isolated and DNA quantitated (Kanno et al., 2006). Graded-dose spiked cocktail (GSC) made of five Bacillus subtilis RNA sequences present on Affymetrix GeneChip arrays (AFFX-ThrX-3_at, AFFX-LysX-3_at, AFFX-PheX-3_at, AFFX-DapX-3_at, AFFX-TrpnX-3_at) was added to the sample homogenates in proportion to their DNA concentration (Kanno et al., 2006). GSC-spiked sample homogenates were processed and probes synthesized using standard Affymetrix protocols, applied to Xenopus microarray v1.0 GeneChips and analyzed using Percellome software (Kanno et al., 2006). Absolutized mRNA levels were expressed as copy number per cell for each probe set.

Percellome microarray data were analyzed using CyberT (Kayala and Baldi, 2012). We did not use low value thresholding/offsetting or log/VSN normalizations. Bayesian analysis used a sliding window of 101 and confidence value of 10. The P-values reported are Bonferroni corrected and Benjamini and Hochber corrected. The full microarray dataset is available at GEO under accession number GSE57352. Genes included in Table 1 comprise a subset upregulated by AGN193109/downregulated by TTNPB based on their regional expression in the posterior.

Xenopus eggs were fertilized in vitro and embryos staged as described (Janesick et al., 2012). Embryos were injected bilaterally or unilaterally at the 2- or 4-cell stage with gene-specific morpholinos (MOs) (supplementary material Table S1) and/or mRNA together with 100 pg/embryo β-galactosidase (β-gal) mRNA. For all MO experiments, control embryos were injected with 10 ng standard control MO (GeneTools). Embryos were maintained in 0.1× MBS until appropriate stages. Embryos processed for WISH were fixed in MEMFA, stained with magenta-GAL (Biosynth), and then stored in 100% ethanol (Janesick et al., 2012).

pCDG1-DN-xRarγ2 was constructed by cloning amino acids 1-393 (lacking the AF-2 domain) into the NcoI-BamHI site of pCDG1 (Blumberg et al., 1998). pCDG1-VP16-xRarγ2 was constructed by cloning the VP16 activation domain upstream of xRarγ2 into pCDG1. pCDG1-xRarα2, pCMX-GAL4-Rarα and GAL4-Rarγ were from Blumberg et al. (Blumberg et al., 1996). X. laevis Rarβ1 and Rarβ2 sequences were found by aligning to the X. tropicalis sequences. pCDG1-xRarβ2 and pCMX-GAL4-xRarβ cloning primers are listed in supplementary material Table S2. pCDG1-xCyp26a1 and pCDG1-c-smrt were constructed by PCR amplification of xCyp26a1 coding regions (Hollemann et al., 1998) or Xl c-smrt (37b−, 41+) (Chen et al., 1996; Malartre et al., 2004) and cloning into pCDG1.

xRarα1^{EGCKG→GSCKV}, xRarα2^{EGCKG→GSCKV}, xRarβ2^{EGCKG→GSCKV}, xRarγ1^{EGCKG→GSCKV} and xRarγ2^{EGCKG→GSCKV} were designed according to Klein et al. (1996), constructed by two-fragment PCR, and cloned into pCDG1 (primer sequences are provided in supplementary material Table S3). Four copies of RXRE^1/2-GRE^1/2 (GGAAGGGTTCACCGAA-AGAACACTCGC) were cloned upstream of the TK-luciferase reporter. All pCDG1 plasmids were sequence verified, linearized with NotI, and mRNA transcribed using mMessage mMachine T7 (Ambion). pCS2-mCherry was linearized with NotI and transcribed from the SP6 promoter.

Microinjected embryos were treated at stage 8 with the following chemicals (in 0.1× MBS): TTNPB (RAR agonist), NRX204647 (RARγ-selective agonist), NRX205099 (RARγ-selective inverse agonist) or 0.1% ethanol vehicle. Twenty-five embryos were treated in each 60-mm Petri dish containing 10 ml chemical. Treated embryos were fixed in MEMFA and processed for WISH, or separated into five-embryo aliquots at stage 10.5 for luciferase assays, or separated into five-embryo aliquots at neurula or tailbud stage for QPCR as described (Janesick et al., 2012). Each group of five embryos was considered one biological replicate (n=1).

Embryos were microinjected or treated with chemicals after the completion of gastrulation (stage 12.5). WISH was performed as previously described (Janesick et al., 2012). Rarγ1, Rarγ2, Rarα (Blumberg et al., 1992), Hoxc10, Ripply2, Thyl2, Msgn1 (Klein et al., 2002), Hoxd10 (Lombardo and Slack, 2001), Tbx6 (Uchiyama et al., 2001), Raldh2 (Glinka et al., 1996) and Myod (Hopwood et al., 1989) probes were prepared by PCR amplification of coding regions from cDNA with T7 promoter at the 3′ end and in vitro transcribed. Hoxc13 sequence was derived from EST clone XL042b19. Relevant primers are listed in supplementary material Table S4. Krox20 (Bradley et al., 1993) and En2 (Bolce et al., 1992) probes were made using T7 and T3 polymerase from EcoRI and XbaI linearized plasmids, respectively. Probes were transcribed with MEGAscript T7 (Ambion) in the presence of digoxigenin-11-UTP (Roche). Double WISH was conducted as described (Janesick et al., 2012). DNP-Rarγ2 was transcribed in the presence of dinitrophenol-11-UTP (PerkinElmer). Hoxc10 expression was quantitated using MATLAB (MathWorks) (supplementary material Fig. S8). The number of purple pixels was calculated by thresholding individual RGB channels (R&B>170, G>120) and dividing by the total number of pixels occupied by the embryo.

1 µg CMX-Rar^{EGCKG→GSCKV} effector plasmid was co-transfected with 5 µg tk-(RXRE^1/2-GRE^1/2)×4 luciferase reporter and 5 µg pCMX-β-galactosidase transfection control plasmids as previously described (Chamorro-Garcia et al., 2012). For activation assays, NRX204647 was tested from 10⁻¹¹ M to 10⁻⁵ M. For antagonism assays, NRX205099 was tested from 10⁻¹⁰ M to 10⁻⁵ M against 10⁻⁸ M 9-cis RA. All transfections were performed in triplicate and reproduced in multiple experiments. Data are reported as normalized luciferase±s.e.m. or percentage reduction±s.e.m. using standard propagation of error (Bevington and Robinson, 2003).

Total RNA from five-embryo pools was DNase treated, LiCl precipitated, and reverse transcribed into cDNA (Janesick et al., 2012). First-strand cDNA was quantitated in a Light Cycler 480 System (Roche) using primer sets listed in supplementary material Table S5 and SYBR Green. Each primer set amplified a single band as determined by gel electrophoresis and melting curve analysis. QPCR data for supplementary material Figs S2 and S7 were analyzed by ΔCt relative to Histone H4, correcting for amplification efficiency between RARs (Pfaffl, 2001). QPCR data for supplementary material Figs S11 and S12 were analyzed by ΔΔCt relative to Histone H4, normalizing to control embryos. Error bars represent biological replicates calculated using standard propagation of error.

We thank Connie Chung for technical help during the early stages of this study and Dr Dennis Bittner for editorial assistance.