Retinol dehydrogenase 10 is a feedback regulator of retinoic acid signalling during axis formation and patterning of the central nervous system

Regional specification is a fundamental process during early development of the central nervous system. In the amphibian gastrula, a group of mesodermal cells in the dorsal blastopore lip, called the Spemann's organizer, secrete signals that induce adjacent ectodermal cells to acquire a neural fate(De Robertis, 2006). An`activation/transformation' model has been proposed for the anteroposterior patterning of the CNS (Nieuwkoop,1952). First, the early neurectoderm acquires an anterior(forebrain) fate. In a second inductive wave, additional signals emanating from the organizer and/or the embryonic mesoderm transform neural tissue into more posterior midbrain, hindbrain, and spinal cord fates. Studies in Xenopus have identified several factors that mediate the first activation step, including soluble BMP antagonists, such as Chordin and Noggin(De Robertis and Kuroda, 2004;Khoka et al., 2005), Wnt antagonists(Niehrs, 2004), and active signals, including insulin-like growth factors(Pera et al., 2001; Richard-Parpaillon et al.,2002). These diverse signals are integrated at the level of Smad1 phosphorylation and turnover (Pera et al.,2003; Fuentealba et al.,2007). Three signalling pathways, including retinoic acid,fibroblast growth factor and Wnt, contribute to the transforming signal and interact in a complex manner to specify posterior neural structures(Durston et al., 1989; Cho and De Robertis, 1990; Kudoh et al., 2002; Shiotsugu et al., 2004; Olivera-Martinez and Storey,2007).

Retinoic acid (RA) is the most active naturally occurring member of a family of lipophilic molecules called retinoids, all of which are derived from vitamin A (Clagett-Dame and De Luca,2002). The RA signal is transduced through nuclear retinoic acid receptors, the RARs and RXRs, which control the expression of target genes involved in vertebrate pattern formation, organogenesis and tissue homeostasis(Mark et al., 2006). Maternal insufficiency of vitamin A or excess RA cause a wide range of teratologic effects - from limb malformations and organ defects to CNS abnormalities -indicating that the embryo requires a precisely regulated supply of retinoids(Ross et al., 2000). In Xenopus embryos, exogenously applied RA during gastrula stages produces a concentration-dependent truncation of anterior structures and an enhancement of posterior structures(Durston et al., 1989; Sive et al., 1990) through its influence on the embryonic mesoderm and ectoderm(Ruiz i Altaba and Jessell,1991; Papalopulu et al.,1991). RA regulates the expression of the homeotic Hox genes,which act in a combinatorial fashion (`Hox code') to specify axial identity in the trunk (Kessel and Gruss,1991; Kessel,1992) and the hindbrain(Marshall et al., 1992).

During embryonic development, the availability of RA is regulated by retinal dehydrogenases (RALDHs) that mediate the oxidation of retinal to RA,and by members of the cytochrome P450 family (CYP26s) that metabolize RA via oxidative inactivation (Niederreither and Dollé, 2008; Duester,2008). In several vertebrates, the RALDH2 gene exhibits tissue-specific expression (Niederreither et al., 1997; Swindell et al.,1999; Chen et al.,2001) at or adjacent to sites of RA signalling(Rossant et al., 1991; Mendelsohn et al., 1991; Balkan et al., 1992; Yelin et al., 2005). In Xenopus, overexpression of RALDH2 mimicked RA signalling(Chen et al., 2001). Loss-of-function studies in mice and zebrafish showed that RALDH2 is not only critical for development, but that it accounts for the majority of RA production in the embryo (Niederreither et al., 1999; Begemann et al.,2001; Grandel et al.,2002). Expression analysis in various species suggested that CYP26A1 is the major RA-degrading enzyme during gastrulation(Hollemann et al., 1998; de Roos et al., 1999; Swindell et al., 1999; Dobbs-McAuliffe et al., 2004). In Xenopus, overexpression of CYP26A1 mRNA caused phenotypes resembling RA deprivation (Hollemann et al., 1998). Functional studies in mouse and zebrafish embryos revealed a crucial role for CYP26A1 in axis specification, hindbrain patterning and tail formation (Abu-Abed et al., 2001; Sakai et al.,2001; Kudoh et al.,2002; Hernandez et al.,2007).

In embryos from placental species, vitamin A (retinol) is provided from the maternal circulation (Ward et al.,1997), whereas oviparous embryos use retinoid and carotinoid stores in the egg yolk (Lampert et al.,2003). A multitude of cytosolic alcohol dehydrogenases and microsomal short-chain dehydrogenases/reductases (SDRs)(Duester et al., 2003; Lidén and Eriksson,2006), as well as the recently identified CYP1B1 mono-oxygenase(Chambers et al., 2007), can mediate the first step of RA synthesis from vitamin A. Retinol dehydrogenase 10 (RDH10) is a member of the SDR family that oxidizes retinol into retinal(Wu et al., 2002; Wu et al., 2004) in an NAD⁺-dependent manner (Belyaeva et al., 2008), and exhibits tissue-specific expression at embryonic and foetal mouse stages (Sandell et al., 2007; Cammas et al.,2007; Romand et al.,2008). Analysis of an N-ethyl-N-nitrosourea-generated mutant called trex suggested an essential function of murine RDH10 for RA biosynthesis during limb, craniofacial and organ development(Sandell et al., 2007). However, RDH10 has not been studied in species other than mammals, its regulation is not understood, and the interplay with other RA metabolizing enzymes, as well as its functions in early aspects of embryonic development,remain to be addressed.

In a screen for secreted proteins, we have recently identified the Xenopus homologue of RDH10 (Pera et al., 2005). XRDH10 expression partially overlaps with that of XRALDH2 in the early embryo and is subject to negative-feedback regulation by endogenous RA. XRDH10 mimics RA signalling and modulates organizer-specific gene expression. We find that XRDH10 cooperates with XRALDH2, and that both enzymes are required to ensure proper RA signalling in the early embryo. Our data describe a novel role of RDH10 in axis formation and CNS development. We present a revised model for the generation of the RA morphogen gradient.

Full-length cDNA clones of XRDH10 and XRALDH2 in the expression vector pCS2 were obtained by secretion cloning(Pera et al., 2005). XRDH10 was fully sequenced (Gene Accession Number FJ213456). To generate the rescue construct pCS2-XRDH10^*, the wobbled nucleotides in codons 2-8 were exchanged via a PCR-based two-step mutagenesis from pCS2-XRDH10, using the primers XRDH10-wob-F-1st(ATGCATATAGTCCTCGAATTCTTTCTGGTC), XRDH10-wob-F-2nd(GCATCGATATGCATATCGTCGTGGAATTTTTTGTGGTC) XRDH10-R(GCCTCGAGTTAAAATTCCATTTTTTGTTTCATTG), and the final PCR products were inserted into the ClaI and XhoI restriction sites of pCS2. pCS2-mRALDH2 was generated from pCMV-Sport6-Aldh1a2(Imagenes GmbH, Germany; IMAGE ID, 30471325) by subcloning the insert into the EcoRI and XbaI sites of pCS2. Plasmid constructs were checked by sequencing and in vitro translation (TnT-Coupled Reticulocyte Lysate System, Promega).

To prepare sense RNA, pCS2 constructs of XRDH10, XRDH10^*,XRALDH2, mRALDH2 and XCYP26A1(Hollemann et al., 1998) (a kind gift of Tomas Pieler, Göttingen University, Germany) were linearized with NotI and transcribed with Sp6 RNA polymerase (mMessage Machine,Ambion). mRNA encoding nuclear β-galactosidase was synthesized from pXEXβgal (a kind gift of Richard Harland, UC Berkeley, CA, USA; XbaI digestion and T7 transcription). The XRDH10-MO(GGAAGAACTCGAGCACTATGTGCAT), XRALDH2-MO (GCATCTCTATTTTACTGGAAGTCAT)and standard control-MO were obtained from Gene Tools.

All-trans-retinoic acid (Sigma, R2625), all-trans-retinol (Fluka, 95144)and disulfiram (Sigma, T1132) were dissolved in DMSO as 10 mM, 50 mM and 250 mM stock solutions, respectively. All-trans-retinal (Sigma, R2500) and citral(Sigma, W230308) were dissolved in 70% ethanol as 5 mM and 40 mM stock solutions, respectively. The stock solutions were then diluted to the final concentrations either in 0.1×MBS (for treatment of whole embryos) or in 1×MBS (for treatment of animal cap explants).

Xenopus laevis embryos and explants were obtained, cultured,microinjected and subjected to whole-mount in situ hybridization and lineage tracing as described (Hou et al.,2007). Gelatine/albumin sections (40 μm) were done using a LeicaVT1200S vibratome.

Total RNA was extracted and the PCR reaction performed as reported(Hou et al., 2007), primers and cycle numbers are available on request. The PCR products were separated on 2% agarose gels.

We isolated a full-length cDNA clone of Xenopus laevis retinol dehydrogenase 10 (XRDH10) by secretion cloning from LiCl-dorsalized gastrula embryos (Pera et al., 2005). A 42-kDa protein was identified by SDS-PAGE in the supernatant of cDNA-transfected HEK293T cells after metabolic labelling with ³⁵S-methionine and ³⁵S-cysteine(Fig. 1A). XRDH10 encodes a 341 amino acid protein, containing an amino terminal cleavable signal peptide, an NAD⁺-cofactor binding site and a catalytic site(Fig. 1B). Two pseudoalleles of RDH10 in X. laevis were found, encoding XRDH10a and XRDH10b(95% amino acid identity). XRDH10a has considerable identity to human and mouse (88%), chick (87%) and zebrafish (77%) RDH10(Fig. 1C,D).

Analysis by RT-PCR indicated that XRDH10 is a maternal and zygotic gene with elevated expression levels at gastrula and neurula stages(Fig. 2A). Whole-mount in situ hybridization showed abundant transcripts in four-cell and blastula-stage embryos (Fig. 2B,C), and RT-PCR revealed equivalent levels of XRDH10 mRNA at the animal and vegetal pole (Fig. 2D). At gastrula stage, distinct expression of XRDH10 was observed in the invaginating mesoderm of the dorsal blastopore lip (Fig. 2E,F). The signals were embedded in the periblastoporal expression domain of XRALDH2 (Fig. 2G) (Chen et al.,2001), and juxtaposed to two distinct XCYP26A1 expression domains in the dorsal animal cap and the ventrolateral blastopore lip(Fig. 2H)(Hollemann et al., 1998). As gastrulation proceeded, XRDH10 transcripts were observed in the head process, anterior lateral plate, presomitic mesoderm, ventral blastopore lip and cardiac crescent (Fig. 2I-K). In neural plate stage embryos, XRDH10 and XRALDH2 genes displayed nested expression patterns in the paraxial trunk mesoderm, with XRDH10 transcripts localized more anteriorly than XRALDH2 signals (Fig. 2J,L) (Chen et al.,2001). These sites of expression were flanked by non-overlapping XCYP26A1 expression domains in the anterior and posterior parts of the neural plate (Fig. 2M)(Hollemann et al., 1998). In early tailbud stage embryos, XRDH10 mRNA overlapped with XRALDH2 expression in the eye field(Fig. 2N,O)(Chen et al., 2001), whereas distinct XCYP26A1 signals could be seen around the eye anlage(Fig. 2P)(Hollemann et al., 1998). Additional XRDH10 expression domains arose in the pronephros anlage,the trunk neural crest, and the posterior inner wall of the proctodeum(Fig. 2Q). In more advanced tailbud embryos, XRDH10 signals were seen in distinct territories of the neural tube (including in the telencephalon, midbrain, midbrain-hindbrain boundary and spinal cord), in the olfactory system, in the eyes and ears, and in the posterior branchial arch, the anterior lateral plate and the posterior notochord (Fig. 2R-X). Common XRALDH2 expression domains were seen in the telencephalon, spinal cord, eyes, ears, anterior lateral plate and pronephros(Fig. 2Y)(Chen et al., 2001). Adjacent,but non-overlapping XCYP26A1 expression appeared in the periocular region, in tissues that flank the pronephros, and in the tip of the tailbud(Fig. 2Z)(Hollemann et al., 1998). In conclusion, the gene expression of XRDH10 and XRALDH2overlapped at several sites, with XRDH10 expression domains being frequently embedded in those of XRALDH2. By contrast, XCYP26A1 displayed a complementary, non-overlapping expression pattern.

The regulation of RDH10 gene activity has not yet been studied. The overlap of RDH10 gene expression with sites of embryonic RA signalling in the frog and the mouse (Fig. 2) (Sandell et al.,2007; Cammas et al.,2007) raised the hypothesis that RA may regulate RDH10transcription. Treatment of Xenopus embryos with 5 μM RA induced a severe reduction of XRDH10 expression(Fig. 3A-D). Although microinjection of XRALDH2 mRNA alone had no effect, a combination of XRALDH2 mRNA and treatment with 5 μM retinal caused a robust downregulation of XRDH10 transcription(Fig. 3E,F). By contrast,exposure to the RA synthesis inhibitors disulfiram(Vermot and Pourquié,2005) or citral (3,7-dimethyl-2,6-octadienal)(Schuh et al., 1993) increased transcript levels of XRDH10 in the embryo(Fig. 3G-J). Similarly,microinjection of XCYP26A1 mRNA caused local upregulation of XRDH10 expression (Fig. 3K,L). We conclude that endogenous RA suppresses XRDH10gene expression and thereby controls the first enzymatic step of RA biosynthesis (Fig. 3M).

We investigated the activity of XRDH10 in Xenopus embryos(Fig. 4). Microinjection of XRDH10 mRNA into the animal pole at the four-cell stage caused a moderate reduction of head structures and shortening of the primary body axis(Fig. 4A,B). This phenotype is reminiscent of the microcephaly and shortened tails obtained by treating embryos with 0.1 μM retinoic acid (Fig. 4C) (Durston et al.,1989). Co-injection of XRDH10 and XCYP26A1 mRNA rescued head and tail structures (Fig. 4D), and treatment of XRDH10-injected embryos with citral restored axial development (Fig. 4E), suggesting that XRDH10 may elicit its activity via the RA pathway. To test whether XRDH10 affects RA signalling, we analyzed in animal cap explants a series of RA target genes, including Xgbx2, Xcad3,Meis3 and HoxD1 (von Bubnoff et al., 1995; Kolm et al.,1997; Dibner et al.,2004; Shiotsugu et al.,2004). RT-PCR analysis revealed that, similar to exogenous RA,injected XRDH10 mRNA induced an upregulation of these genes(Fig. 4F). The results show that XRDH10 and RA have common activities, and that XRDH10 activity is abrogated by the inhibition of RA signals, suggesting that XRDH10 activates RA signalling in Xenopus embryos.

The specific expression of XRDH10 in the dorsal blastopore lip(Fig. 2E,F) prompted us to analyze its effects on gene markers that demarcate the Spemann's organizer. To this end, we radially injected XRDH10 mRNA at the four-cell stage and analyzed embryos at stage 10.5 by whole-mount in situ hybridization. We found that XRDH10 overexpression led to an expansion of the Xlim1and Chordin expression domains(Fig. 4G,H,K,L), while Goosecoid and ADMP expression was reduced(Fig. 4O,P,S,T). In accordance with these results, treatment of embryos with 5 μM RA caused an upregulation of Xlim1 (Fig. 4I,J) (Taira et al.,1994) and Chordin(Fig. 4M,N) expression, but a downregulation of Goosecoid (Fig. 4Q,R) (Cho et al.,1991) and ADMP (Fig. 4U,V) expression. We note that Chordin expression in the dorsal ectoderm of earlier blastula embryos was moderately increased by XRDH10 mRNA injection and RA treatment (see Fig. S1 in the supplementary material). The organizer markers Noggin, Frzb1, sFRP2and Crescent were not obviously affected in gastrula embryos (see Fig. S2 in the supplementary material). We conclude that XRDH10overexpression mimics RA activity and differentially affects gene expression in the Spemann's organizer.

Next we analyzed the effects of XRDH10 on pattern formation at post-gastrulation stages (Fig. 5). At stage 12.5, HoxD1 is expressed in the trunk mesoderm and overlying ectoderm with the anterior boundary at the level of hindbrain rhombomere 4 (Fig. 5A). Unilateral injection of XRDH10 mRNA caused upregulation and anteriorward expansion of HoxD1 expression(Fig. 5B). XRALDH2alone or upon co-injection with XRDH10 mRNA had a similar effect(Fig. 5C,D). By contrast, XCYP26A1 reverted the effect of co-injected XRDH10 mRNA, as it reduced HoxD1 expression and shifted its anterior boundary posteriorly (Fig. 5E). At stage 14, Xlim1 labels two rows of neural expression in the trunk(arrowhead in Fig. 5F). Injected XRDH10 or XRALDH2 mRNA caused an anterior shift(Fig. 5G,H), and a combination of both mRNAs led to a robust expansion of these Xlim1-positive neural cells (Fig. 5I). XCYP26A1 overrode the effect of co-injected XRDH10 mRNA and suppressed Xlim1 expression (Fig. 5J).

Previous studies had shown that overexpression of XRALDH2posteriorized the neural tube (Chen et al., 2001), whereas XCYP26A1 had the opposite effect(Hollemann et al., 1998). At the tailbud stage, XRDH10 mRNA showed little effect when injected alone (Fig. 5L,Q). However, XRDH10 enhanced the posteriorizing effect of XRALDH2 mRNA and caused an anterior shift of the hindbrain rhombomeres 3 and 5(Krox20), and led to a distortion of the midbrain-hindbrain boundary(En2) and eye field (Rx2A) upon co-injection of both mRNAs(Fig. 5M,N,R,S). Conversely, a combination of XRDH10 and XCYP26A1 mRNA resulted in a pronounced posterior shift of these markers(Fig. 5O,T). The location of the telencephalon (FoxG1) was not affected by any of the injections(Fig. 5Q-T). The analysis of Krox20 expression showed that the frequency and extent of rhombomeric shifts induced by a combination of XRDH10 and XRALDH2exceeded the sum of effects induced by each mRNA alone(Fig. 5U). The data indicate that XRDH10 co-operates with XRALDH2 in stimulating RA signalling in the early embryo and that both enzymes exhibit synergistic effects on anteroposterior patterning of the CNS.

The relatively mild phenotype of XRDH10 mRNA-injected embryos raised the question of whether XRDH10 activity might be restricted by insufficient endogenous retinol concentrations. We therefore examined the effects of overexpressed XRDH10 in the presence of excessive retinol(Fig. 6). In accord with the observations of others (Durston et al.,1989), treatment of embryos between stages 9 and 12 with 50 μM retinol caused microcephaly (Fig. 6B). Although animal injection of XRDH10 mRNA alone had little effect (Fig. 4B), XRDH10 mRNA injection followed by treatment with retinol caused the complete loss of eye and head structures (anencephaly; Fig. 6C). The effects of XRDH10 and retinol were reverted by co-injection of XCYP26A1mRNA in a dose-dependent manner (Fig. 6D,E). At the tailbud stage, retinol treatment caused a mild reduction of the eye field marker Rx2A(Fig. 6F,G). As shown above,injected XRDH10 mRNA did not reduce the size of the eye field(Fig. 5L). However, a combination of XRDH10 mRNA injection and retinol administration led to a significant downsizing of the eye anlage(Fig. 6H), which was rescued by XCYP26A1 mRNA (Fig. 6I). Together, the results suggest that the supply of retinol may be limiting for XRDH10 activity during head development and eye formation.

To study the functional contribution of enzymes involved in RA biosynthesis, we downregulated endogenous XRDH10 and XRALDH2 proteins in Xenopus embryos (Fig. 7). Specific antisense morpholino oligonucleotides (MOs) directed against the translation initiation sites of the known pseudoalleles of XRDH10 (Fig. 7A) and XRALDH2 (Fig. 7B)reduced protein synthesis of their respective targets in an in vitro transcription-translation assay, whereas an unspecific control MO had no effect (Fig. 7C,D).

Microinjection of XRDH10-MO into the margin of two-cell-stage embryos caused a reduction of head structures and enlarged ventroposterior structures at the tailbud stage (Fig. 7F). In tadpole embryos, knockdown of XRDH10 led to smaller eyes and a significant shortening of the tail(Fig. 7I). Similar ventralized phenotypes were obtained with the XRALDH2-MO(Fig. 7G,J). To verify that the effects of the morpholino oligomers were specific, we generated a XRDH10 rescue construct, designated XRDH10^*, in which six nucleotides in the morpholino target sequence were mutagenized (see Materials and methods). Injection of XRDH10^* mRNA rescued the phenotype caused by XRDH10-MO (insets in Fig. 7F,I). Similarly,microinjection of mRNA for mouse RALDH2 (mRALDH2), which is not targeted by the morpholino oligo, neutralized the effect of XRALDH2-MO and restored normal axial development (insets in Fig. 7G,J).

In gastrula embryos, radially injected XRDH10-MO and XRALDH2-MO reduced Chordin gene expression in the dorsal blastopore lip (Fig. 7K-M). Concomitantly, depletion of XRDH10 and XRALDH2 caused a significant expansion of Goosecoid and ADMP expression(Fig. 7N-S). Despite the robust stimulation of Xlim1 expression in gain-of-function experiments(Fig. 4G-J), depletion of XRDH10 or XRALDH2 did not affect this marker at gastrula stage (data not shown). However, XRDH10-MO or XRALDH2-MO caused a posteriorward retraction of the Xlim1 expression domain in the pronephros anlage of neurula embryos (Fig. 7T-V).

We next investigated the effects of downregulating XRDH10 and XRALDH2 on anteroposterior patterning of the CNS (Fig. 8). At the advanced gastrula stage, unilaterally injected XRDH10-MO reduced transcript levels and shifted the anterior boundary of HoxD1 expression posteriorly(Fig. 8B). The XRALDH2-MO (Fig. 8C),or a combination of XRDH10-MO and XRALDH2-MO(Fig. 8D), caused a similar effect. The specificity of this phenotype was underscored by the findings that a control morpholino had no effect (Fig. 8A), and that co-injection of non-targeted XRDH10^* and mRALDH2 mRNAs with their respective MOs restored normal HoxD1 expression(Fig. 8E,F). In neurula embryos, XRDH10-MO and XRALDH2-MO caused a slight posterior distortion of the midbrain-hindbrain boundary (En2), a posterior shift of hindbrain rhombomeres (HoxB3, xCRABP), but no significant effect on HoxC6 expression in the spinal cord(Fig. 8G-L; see also Fig. S3 in the supplementary material). At the tail bud stage, XRDH10 and XRALDH2 morphant embryos exhibited a posterior shift of rhombomeres 3 and 5 (Krox20) relative to the unaffected eye field (Rx2A; Fig. 8M-R). Notably, the extent of the rhombomeric shift induced by 2.6 pmol XRDH10-MO was similar to that of an equimolar amount of XRALDH2-MO, and was not significantly increased when both MOs were injected together(Fig. 8S). Our results are consistent with those obtained from other loss-of-function experiments, using dominant-negative retinoid receptors (Kolm et al., 1997; Blumberg,1997; van der Wees et al.,1998) and the RA hydroxylase CYP26A1(Hollemann et al., 1998),supporting a contribution of XRDH10 and XRALDH2 in positioning hindbrain rhombomeres along the anteroposterior neuraxis in Xenopus.

To address whether XRDH10 is involved in vitamin A metabolism, we investigated the effects of downregulating the XRDH10 enzyme in the presence of exogenous retinol (Fig. 8T-W). To this end, we treated embryos with DMSO as a control or with 100 μM retinol. In advanced gastrula embryos, retinol led to an anterior expansion of HoxD1 expression(Fig. 8T,V). The retinol-mediated anterior expansion of HoxD1 expression was reverted by XRDH10-MO on the injected side (Fig. 8W), suggesting that the posteriorizing effect of retinol depends on XRDH10 activity. Together, the experiments demonstrate an involvement of XRDH10 in RA biosynthesis during axes formation and hindbrain patterning.

In this study, we investigated the role of RDH10 in the early Xenopus embryo. By gain- and loss-of-function assays, we have shown that XRDH10 upregulates retinoic acid (RA) signalling and is important for the correct specification of the dorsoventral and anteroposterior body axes. XRDH10 cooperates with XRALDH2 and participates in determining the position of the hindbrain rhombomeres. Our data suggest that XRDH10 contributes to building up the RA morphogen gradient in the embryo.

Xenopus RDH10 exhibits tissue-specific expression with common expression domains to mouse RDH10; for example, in the lateral trunk mesoderm, ventral neuroepithelium, at the midbrain-hindbrain boundary, and in sensory organs (Fig. 2)(Sandell et al., 2007; Cammas et al., 2007; Romand et al., 2008). However,there are differences in the timing of induction and distribution of gene transcripts in both species. The earliest expression of mouse RDH10was reported in head-fold-stage embryos just prior to somitogenesis(Sandell et al., 2007; Cammas et al., 2007). We detected abundant maternal XRDH10 gene products(Fig. 2A-D), which may contribute to the high levels of retinal observed in the egg and embryonic yolk (Azuma et al., 1990; Lampert et al., 2003), and robust expression levels during gastrulation(Fig. 2A,E,F,I,J), when the embryo is most sensitive to RA exposure(Durston et al., 1989; Sive et al., 1990). XRDH10 displayed distinct expression in the dorsal blastopore lip,head process, telencephalon anlage and neural crest, which have no apparent counterpart for mouse RDH10.

We found that RA downregulates transcript levels of RDH10 in Xenopus embryos (Fig. 3). Importantly, lowering of embryonic RA levels with the pharmacological inhibitors disulfiram and citral, or with the metabolic enzyme XCYP26A1, elevated XRDH10 expression, suggesting that endogenous RA suppresses XRDH10 gene activity. Previous studies have shown that RA downregulates RALDH2 expression(Niederreither et al., 1997; Chen et al., 2001; Dobbs-McAuliffe et al., 2004)and, conversely, RA upregulated CYP26A1 transcript levels in several species (White et al., 1996; Ray et al., 1997; Hollemann et al., 1998; White et al., 2007). Thus, the targeting of the XRDH10 gene by RA adds to an intricate regulatory network, in which RA suppresses anabolic (RA-synthesizing) and stimulates catabolic (RA-degrading) enzymes (Fig. 3M). This fine-tuned feedback control provides protection against exogenous retinoid fluctuations and allows the stabilization of local RA distributions in the embryo.

We observed novel expression domains of XRDH10, most strikingly in the dorsal blastopore lip (Fig. 2E,F). This group of cells, referred to as the Spemann's organizer, plays a prominent role in the specification of the embryonic body axes, and the induction and pattern formation of the developing CNS(De Robertis and Kuroda,2004). Interestingly, XRDH10 transcripts in the organizer not only overlap with XRALDH2 but are complementary to XCYP26A1 expression (Fig. 2E-H) (Hollemann et al.,1998; Chen et al.,2001). XRDH10 gene activity also coincides with active RA signalling in this region (Chen et al.,1994; Yelin et al.,2005). Our gain- and loss-of-function studies suggest a novel role for RA in positively regulating Chordin and negatively regulating ADMP expression (Figs 4, 7). Chordin is a soluble BMP antagonist and a key mediator of Spemann's organizer activity(Sasai et al., 1994; De Robertis and Kuroda, 2004). The anti-dorsalizing morphogenetic protein (ADMP) secreted from the dorsal gastrula organizer (Moos et al.,1995) induces BMP/Smad1 signalling via the ALK2 receptor(Reversade and De Robertis,2005). Knockdown of XRDH10 and XRALDH2 cause small head and enlarged ventroposterior structures (Fig. 7), a phenotype that is commonly seen upon elevated BMP/Smad1 activity (e.g. Pera et al.,2003; Fuentealba et al.,2007). The opposite transcriptional regulation of the secreted proteins Chordin and ADMP by RA (this study) suggests a possible mechanism of how XRDH10 and XRALDH2 could promote Spemann's organizer activity and dorsal development.

RA is a known morphogen that provides spatial information in various developmental contexts (Thaller and Eichele, 1987; Wolpert,1989; Kessel and Gruss,1991; White et al.,2007). Along the anteroposterior neuraxis, concentration-dependent gradients of RA specify positional values via the activation of Hox genes in the developing hindbrain (Gould et al.,1998; Dupé and Lumsden,2001) and spinal cord (Muhr et al., 1999; Liu et al.,2001). Previous models have suggested that these RA gradients arise mainly as a result of local RA supply by RALDH2 in the paraxial trunk mesoderm, diffusion of RA, and CYP26A1-mediated RA decay at the anterior and posterior ends of the neural plate (Maden,1999; Maden, 2002; Sirbu et al., 2005; Hernandez et al., 2007; White et al., 2007). During gastrulation of lower vertebrates, retinoid stores in the yolk translocate from vegetal cells mainly to the endoderm(Azuma et al., 1990; Lampert et al., 2003). However, RALDH2 activity in the dorsal blastopore lip and trunk mesoderm may lead to a local shortage of retinal that could cause a collapse of the RA gradient. Indeed, the phenotypes of XRDH10 morphants affecting organizer-specific gene expression and hindbrain patterning (Figs 7, 8) provide strong evidence for the need of ongoing retinol conversion to retinal in the developing embryo.

Our data suggest an alternative mode of RA gradient formation based on the specific expression characteristics and on cooperation between RDH10 and RALDH2 (Fig. 9). The microsomal RDH10 enzyme in the anterior trunk mesoderm produces and secretes retinal that diffuses into posterior cells, where RALDH2 converts it into RA. The combinatorial mRNA distribution of RDH10 and RALDH2 (with RDH10 being more anteriorly expressed than RALDH2) is crucial for the posteriorward flow of retinal. As a consequence, highest levels of RA are produced at the anterior front of the RALDH2expression domain, with decreasing concentrations towards its posterior end. As RDH10 ensures a continuous local supply of retinal, this enzyme also contributes to the stabilization of the RA gradient. At the neural plate stage, the peak of RA concentration is located at the hindbrain-spinal cord boundary, but as development progresses it moves concomitant with the translocation of the RDH10 and RALDH2 expression domains posteriorward. It is of interest that the RA-responsive Hox genes show a sharp anterior border of expression and posteriorly declining transcript levels(De Robertis et al., 1991),reflecting the gradual RA distribution generated by RDH10 and RALDH2. In conclusion, our data suggest that an initial RA gradient forms in the anterior trunk mesoderm already at the step of RA synthesis.

The combinatorial gene expression of two enzymes that act back-to-back to produce a signal, here referred to as a `biosynthetic enzyme code',constitutes a novel mechanism for forming and stabilizing a morphogen gradient. This mechanism may apply not only to the establishment of RA gradients along the embryonic axis, but also to other areas where RDH10 and RALDH2 overlap, such as in the dorsal blastopore lip, the pronephros anlage, the eye field and the ear placode. In the mouse,additional sites of overlapping RDH10 and RALDH2 expression have been reported for the limb anlage and the foetal brain(Niederreither et al., 1997; Sandell et al., 2007; Cammas et al., 2007; Romand, 2008). Future studies need to address the significance of an interaction between the two enzymes in these morphogenetic fields. It is noteworthy that the mechanism of nested gene expression and combinatorial action has initially been found in the homeotic Hox genes, which are the most prominent targets of RA signalling in vertebrates (Kessel and Gruss,1991). This suggests a common principle for the generation and downstream signalling events of this morphogen.