Restriction of retinoic acid activity by Cyp26b1 is required for proper timing and patterning of osteogenesis during zebrafish development

Skeletal development is highly conserved in vertebrates and involves two main processes: skeletal patterning to define the shape and location of the different skeletal elements within the developing body, and differentiation of skeletogenic cells (Karsenty and Wagner,2002; Mariani and Martin,2003). Cartilage-forming chondrocytes and bone-forming osteoblasts share a common mesenchymal progenitor that derives from neural crest,sclerotome or lateral plate mesoderm(Olsen et al., 2000). Skeletogenesis is initiated when mesenchymal cells aggregate to form mesenchymal condensations. In most parts of the bony skeleton, including the vertebral column of mammals, but not of teleosts(Bird and Mabee, 2003; Elizondo et al., 2005; Fleming et al., 2004; Inohaya et al., 2007), a cartilaginous anlage serves as a template to model the future bone(endochondral ossification). In this case, cells within the condensation become chondrocytes, whereas cells at the periphery of the skeletal element form a structure called the perichondrium(Karsenty and Wagner, 2002). During ossification, chondrocytes in the core of the condensate become hypertrophic, a transition reflected in the switch from Col2a1(encoding collagen type II) to Col10a1 (collagen type X) expression,while osteoblasts start to mature within the perichondrium (now also called the periosteum) and form ossification centers that eventually replace the cartilage (Colnot, 2005; Karsenty and Wagner, 2002). Alternatively, some skeletal elements, including parts of the craniofacial system, are generated by direct differentiation of mesenchymal cells into osteoblasts (intramembranous or dermal ossification). Maturing osteoblasts express the same marker genes as hypertrophic chondrocytes, including the transcription factor gene runx2 (also called cbfa1)(Flores et al., 2006; Flores et al., 2004), osteopontin (opn; also called spp1)(Kawasaki et al., 2004), which encodes a component of bone matrix (Alford and Hankenson, 2006), and, at least in zebrafish, col10a1(Avaron et al., 2006), whereas the transcription factor Osterix (Osx; also called Sp7) is a specific marker and regulator of the osteoblast lineage(Nakashima et al., 2002).

A known signal regulating skeletogenic cell development is all-trans retinoic acid (RA) (Adams et al.,2007; Weston et al.,2003), a derivative of vitamin A that is required for multiple processes of vertebrate development(Niederreither and Dolle,2008). RA is a diffusible lipophilic molecule that binds to nuclear receptors [retinoic acid receptors (RARs) and retinoid X receptors(RXRs)] to regulate the transcription of target genes. RA concentrations are determined by the balance between RA synthesis via retinaldehyde hydrogenases(Aldh1-3) and RA oxidation by cytochrome P450 enzymes of the Cyp26 class(Blomhoff and Blomhoff, 2006; Fujii et al., 1997; White et al., 1997). As in mammals, three different zebrafish cyp26 genes have been described: cyp26a1, cyp26b1 and cyp26c1 (formerly cyp26d1),which are expressed in distinct, but partially overlapping patterns(Abu-Abed et al., 2002; Emoto et al., 2005; Gu et al., 2005; Hernandez et al., 2007; Kudoh et al., 2002; MacLean et al., 2001; Tahayato et al., 2003; Zhao et al., 2005). The in vivo requirement for Cyp26 enzymes was revealed via Cyp26a1 and Cyp26b1 gene targeting in mouse(Abu-Abed et al., 2001; MacLean et al., 2007; Yashiro et al., 2004), and via cyp26a1 (giraffe) mutants(Emoto et al., 2005) and antisense-mediated knockdown of cyp26a1, cyp26b1 and cyp26c1in zebrafish (Echeverri and Oates,2007; Hernandez et al.,2007; Kudoh et al.,2002; Reijntjes et al.,2007; Shelton et al.,2006; White et al.,2007). Of the zebrafish reports, only one addressed the role of Cyp26 enzymes during skeletogenesis, claiming that Cyp26b1 is required for the patterning and migration of cranial neural crest(Reijntjes et al., 2007).

Knockout of Cyp26b1 in mouse causes severe limb defects that have been attributed to a combination of shifts in the proximodistal patterning of the limb bud and a retardation of chondrocyte maturation(Yashiro et al., 2004). This suggests that Cyp26b1 interferes with the reported role of RA in blocking chondrocyte specification from mesenchymal precursors(Weston et al., 2003). Other data suggest a later and seemingly opposing role for RA signaling in promoting hypertrophic maturation of chondrocytes and subsequent replacement by bone(Weston et al., 2003),although this has not yet been addressed genetically. Also, it has remained unclear to what extent this latter effect is due to interference with chondrocytes (Iwamoto et al.,1993; Weston et al.,2003) versus osteoblasts(Manji et al., 1998; Song et al., 2005) and with osteoblast maturation versus activity.

Here, we have studied the role of Cyp26b1 as an essential regulator of skeletal development in zebrafish. cyp26b1 is expressed in chondrogenic mesenchymal condensations as well as in osteoblast precursors of endochondral and intramembranous bones, including vertebrae. cyp26b1mutants display multiple defects during chondro- and osteogenesis, all of which can be mimicked by treatment with RA. This indicates that in contrast to a recent report (Reijntjes et al.,2007), zebrafish Cyp26b1 acts by restricting retinoid signaling. The hyperossification of craniofacial bones and vertebrae of mutant animals is anticipated by an increase in osteopontin expression in osteoblasts. Comparing the axial defects of cyp26b1 mutants with those caused by transgenic overexpression of the Bone morphogenetic protein Bmp2, a well-known positive regulator of osteoblast maturation upstream of Runx2 and Osx(Ulsamer et al., 2008; Wu et al., 2007), we noticed striking differences in the pattern of hyperossification and in the molecular signature of the supernumerary osteoblasts. Together, these data suggest that RA inhibition by Cyp26 enzymes is required to restrict ossification in vivo,most likely by attenuating osteoblast activity.

Unless noted otherwise, the dol allele ti230g was used(Piotrowski et al., 1996). For temporally controlled bmp2b overexpression, the transgenic line Tg(hsp70:bmp2b) was used as described(Chocron et al., 2007). According to RT-PCR analyses, the transgene is activated within 10 minutes after the heat shock, with transcripts remaining detectable for ∼24 hours.

Genetic rough mapping of dol^ti230g was carried out via bulk meiotic segregation analyses as described(Laue et al., 2008). For chromosomal walking, the BAC libraries CHORI-211 and DanioKey (RZPD) were screened by PCR. For identification of the mutant lesion and for genotyping,genomic cyp26b1 DNA or cDNA (GenBank accession number AY321366)fragments were amplified by PCR and sequenced. Primer sequences are available upon request.

An MO against the cyp26b1 splice-donor junction between exon 3 and intron 3 (Cyp26b1-SDEx3 MO, 5′-AGCTATTGACATTTTACCTTTCTGT-3′) was purchased from Gene Tools. For full knockdown, ∼0.5 pmol, and for hypomorphs (Fig. 6E), 0.2 pmol,of MO was injected per embryo at the one-cell stage as described(Nasevicius and Ekker, 2000). aldh1a MO was used as described(Begemann et al., 2001).

In situ hybridization and fluorescent double in situ hybridization were performed as described (Clay and Ramakrishnan, 2005; Hammerschmidt et al., 1996). For zebrafish opn and osx, cDNA fragments were amplified by RT-PCR with primer sequences obtained from the Ensembl database, and cloned into pCRII-TOPO (Invitrogen). For probe synthesis, plasmids were linearized with XhoI and transcribed with SP6 RNA polymerase. For cyp26b1, clone IMAGp998A0411194Q1 was obtained from RZPD, linearized with KpnI and transcribed with SP6 RNA polymerase. Probes for cyp26a1 and cyp26c1(Hernandez et al., 2007), col2a1 (Yan et al.,1995), col10a1 (Nica et al., 2006), dlx2a(Akimenko et al., 1994) and sox9a (Wada et al.,2005) were synthesized as described.

Immunostainings were carried out using the Vectastain ABC Kit (Axxora) with the following primary antibodies and dilutions: zn5 (1:500, ZIRC, University of Oregon), anti-GFP (1:200, Roche). For sectioning, embryos were embedded in paraffin.

Skeletal elements of zebrafish and mice were stained with Alcian Blue and/or Alizarin Red essentially as described(Kessel and Gruss, 1991; Walker and Kimmel, 2007).

Stock solutions (10 mM) of 4-(diethylamino)benzaldehyde (DEAB; Fluka),all-trans RA (Sigma) and R115866 (Johnson and Johnson Pharmaceutical Research and Development) were prepared in DMSO as described(Begemann et al., 2004; Begemann et al., 2001; Hernandez et al., 2007). Wild-type or cyp26b1 mutant zebrafish were incubated in the dark with final concentrations of 0.5 μM RA (before 50 hpf), 10 μM RA (after 48 hpf), 10 μM DEAB (before 50 hpf), 50 μM DEAB (after 48 hpf) or 10 μM R115866. For mouse experiments, R115866 was dissolved in PEG 200 (Sigma, 2.5 mg/ml) and administered to pregnant NMRI females by daily oral gavages,applying 10 mg per kilogram body weight, from E13-16.

The dolphin mutant dol^ti230g, which was previously identified in a large-scale ENU mutagenesis screen, is characterized by a beak-like appearance of the jaw(Fig. 1A,B)(Piotrowski et al., 1996) and,in some genetic backgrounds, by shorter and malformed pectoral fins(Fig. 1C,D). Using bulk segregation linkage analysis, the dolphin locus was mapped to a 0.38 cM interval on linkage group 7 (Fig. 1E); this was followed by a BAC walk to construct a contig covering the interval (Fig. 1F). Newly designed markers (KL1-7) along three overlapping BACs within the contig showed no recombination in 4500 meioses(Fig. 1F).

The three BACs contained several genes, all of which were sequenced. Only in the cyp26b1 gene of mutants was a mutation found, comprising a GT→AT transition in the splice-donor site of the exon 3-intron 3 junction(Fig. 1H,I). Sequencing of independent cDNA clones revealed the use of a downstream GT as a novel splice-donor in mutants (50/50), but not in wild-type embryos of the same strain (0/50) (Fig. 1I,J). The resulting transcript carries an insertion of seven nucleotides, leading to a frame shift and premature termination of the protein. This C-terminal truncation removes most of the highly conserved cytochrome P450 domain,including the oxygen-, steroid- and heme-binding sites(Fig. 1G). Subsequently, a second cyp26b1 allele, sa0002, was identified by TILLING(Wienholds et al., 2003), with an AAG→TAG nonsense mutation at nucleotide position 135 of the coding region, causing an even more severe truncation of the protein after 45 amino acid residues (Fig. 1G)(http://www.sanger.ac.uk/cgi-bin/Projects/D_rerio/mutres/tracking.pl). sa0002 failed to complement ti230g, and sa0002mutants showed craniofacial and axial defects indistinguishable from those of ti230g mutants (see Fig. S1A-C,E-G in the supplementary material). In contrast to full-length Cyp26b1, both truncated versions were completely inactive upon forced expression in early zebrafish embryos (see Fig. S1I-L in the supplementary material). Furthermore, the axial hyperossification of ti230g mutants could be rescued or converted to hypo-ossified phenotypes by temporally controlled reapplication of wild-type cyp26b1 (see Fig. S1H in the supplementary material). Finally, the defects of dol mutants could be phenocopied in wild-type fish by injecting an antisense MO targeting the splice site affected in the ti230g allele (Fig. 4G,N; Fig. 5C; Fig. 6E,F; see Fig. S2 in the supplementary material). Together, this indicates that the defects of zebrafish dol mutants are caused by null mutations in the cyp26b1 gene.

cyp26b1 has been shown to be expressed in various regions of zebrafish embryos (Hernandez et al.,2007; Zhao et al.,2005). We have extended these studies, focusing on developing skeletal elements. At 26 hours post-fertilization (hpf), cyp26b1 was expressed in multiple domains in close proximity to postmigratory cranial neural crest (CNC) cells, but not in CNC itself, as revealed by double stainings with cyp26b1 and the CNC markers dlx2a or fli1a (Fig. 2A-D). cyp26b1 was prominently expressed in the forehead close to the CNC that, according to cell-tracing experiments, gives rise to chondrocytes of the neurocranial ethmoid plate (Eberhart et al., 2006; Wada et al.,2005) (Fig. 2A,B). In addition, cyp26b1 was expressed adjacent to postmigratory CNC that gives rise to the pharyngeal arches (Fig. 2C,D).

At 48 hpf, cyp26b1 was coexpressed with sox9a(Fig. 2H) in mesenchymal condensations that give rise to pharyngeal arches(Fig. 2F) and neurocranial cartilages (Fig. 2G). However,concomitant with the onset of col2a1 expression, which is a marker for specified chondrocytes (Yan et al.,1995), cyp26b1 expression ceased within the cartilaginous elements but remained strong in perichondrial cells(Fig. 2I-K). A similar transient expression in mesenchymal condensations and persistent expression in perichondria of the craniofacial skeletal elements has also been described for mouse Cyp26b1 (Abu-Abed et al.,2002). The perichondrium is a supposed source of osteoblasts. However, in contrast to widespread perichondrial expression of cyp26b1, expression of the osteoblast markers osterix(osx) (Fig. 2L), osteopontin (opn) (Fig. 2M) and col10a1 (Fig. 2N) (Avaron et al.,2006) was confined to those perichondrial domains undergoing endochondral ossification (Fig. 2O). In such col10a1-positive cells, cyp26b1transcript levels were lower than in the adjacent col10a1-negative perichondrium (Fig. 2P,Q),suggesting that cyp26b1 expression decreases when osteoblasts differentiate and/or become active.

In addition to the perichondrium/periosteum, zebrafish cyp26b1 was expressed in various bone primordia that ossify in an intramembranous manner(Cubbage and Mabee, 1996; Elizondo et al., 2005). Examples include the opercle (Fig. 3A,D) and cleithrum (see Fig. S4G in the supplementary material). At 72 and 120 hpf, the opercular bone matrix(Fig. 3C,F) was surrounded by osteoblasts coexpressing col10a1, osx and opn(Fig. 3B,E; see S4A-F in the supplementary material). Double in situ hybridization for cyp26b1 and col10a1 further revealed that, similar to in the perichondrium, cyp26b1 levels in col10a1-positive osteoblasts of the opercle (Fig. 3A,D) and cleithrum (see Fig. S4G in the supplementary material) were considerably weaker than in adjacent cells, which are most likely immature and/or less active osteoblasts.

Osteoblast expression of cyp26b1 was also found in the elements of the axial skeleton. The anterior part of the notochord, which becomes uniformly ossified (basioccipital articulatory process) (see Fig. 6A), was lined by a continuous layer of cyp26b1-positive cells(Fig. 3G), whereas in trunk and tail, where ossification of vertebral primordia occurs in a segmented manner, cyp26b1-positive cells displayed a corresponding metameric distribution (Fig. 3H,I). The same metameric pattern was obtained for the osteoblast markers opnand col10a1 (Fig. 3L-N). The position of such osteoblasts at intersomitic borders coincided with the anterior borders of forming vertebral bodies stained with Alizarin Red (Fig. 3O),suggesting that cells were localized within the intervertebral zones. Comparative expression analyses at different developmental time points further revealed a continuous decline in the number of axial cyp26b1-positive cells from 96-156 hpf, while the number of col10a1- and opn-positive cells increased (Fig. 3P), with transient coexpression of cyp26b1 and opn in the same cells at 144 hpf(Fig. 3J).

Alcian Blue stainings of the cartilage of cyp26b1 mutants and morphants at 120 hpf revealed specific defects in midline elements of the visceral skeleton and the neurocranium, whereas most other cartilaginous elements appeared largely normal (Fig. 4E-G). The ceratohyals of the left and right second arches were fused in the midline (Fig. 4H,I), and the medial elements (basibranchial) of the posterior gill arches (p5, p6) were missing (Fig. 4J,K). In addition, the medial ethmoid plate of the neurocranium was missing or strongly reduced (Fig. 4L-N), very similar to the phenotype previously described for mutants in sonic hedgehog (Wada et al., 2005). To determine when this defect arises, we stained embryos for col2a1 and sox9a to label chondrocytes and their precursors. At 36 hpf, col2a1 is expressed in chondrocyte precursors of the trabeculae cranii (Fig. 4A), which are lateral structures of the neurocranium surrounding the ethomoid plate and extending further posteriorly(Schilling and Kimmel, 1997). At this stage, col2a1 expression in cyp26b1 mutants appeared normal (Fig. 4B). However, at 56 hpf, the col2a1 expression domains were shifted posteriorly and fused at the midline, anticipating the subsequent absence of the medial ethmoid (Fig. 4C,D).

As in higher vertebrates, the majority of the zebrafish craniofacial skeleton forms through endochondral ossification, starting at 6 days post-fertilization (dpf); for example, in restricted regions of the ceratohyal and hyomandibula of the second pharyngeal arch. Intramembranous bones form even earlier, with mineralization of the opercle starting at 3 dpf(Cubbage and Mabee, 1996).

Staining cyp26b1 mutants or morphants for mineralized bone matrix with Alizarin Red, we observed hyperossification of both endochondral and intramembranous bones. At 192 hpf, the mineralized domain in the opercle of mutants and morphants was larger than in wild-type animals(Fig. 5A-C). Also, mutants displayed significantly stronger and more advanced endochondral mineralization of the ceratohyal (Fig. 5D-F).

In addition to craniofacial defects, cyp26b1 mutants and morphants exhibited severe abnormalities in the axial skeleton, which in teleost larvae is formed through ossification of the sheath around the notochord (perichordal ossification). In zebrafish, unsegmented ossification around the anterior part of the notochord gives rise to the basioccipital articulatory process, while metameric mineralization more posteriorly forms the vertebral column. Vertebral ossification starts at the level of the fourth vertebral body(centrum), from where it proceeds anteriorly and posteriorly(Bird and Mabee, 2003; Gavaia et al., 2006; Stemple, 2005). At 180 hpf,wild-type larvae exhibited six to eight Alizarin Red-positive centra, with centra 3 and 4 correspondingly broader than the others(Fig. 6A). By contrast, cyp26b1 mutants and morphants showed a complete fusion of Alizarin Red-positive segments and an extension of staining into caudal regions, which in wild-type animals mineralize later (Du et al., 2001) (Fig. 6C,F). In cross-sections, the mineralized perichordal sheath of mutants appeared broader and more strongly stained than in wild-type siblings,whereas notochordal cells remained Alizarin Red-negative and normally vacuolated (Fig. 6G-J). Interestingly, cyp26b1 heterozygotes and wild-type embryos injected with lower amounts of cyp26b1 MO displayed an intermediate phenotype with distinct, but broader, centra in anterior regions and precocious centra mineralization in caudal regions of the notochord(Fig. 6, compare B,E with A,D). This suggests that Cyp26b1 is required to attenuate vertebral growth.

In mouse, there is evidence for a negative effect of Cyp26 enzymes on retinoid signaling (Fujii et al.,1997; Niederreither et al.,2002; White et al.,1997). However, according to a recent study, the phenotype of the cyp26b1 morphant zebrafish more closely resembles that of RA deficiency, suggesting that Cyp26 enzymes might generate, rather than metabolize, biologically active retinoids(Reijntjes et al., 2007). To test this, and to determine the crucial time window(s) of Cyp26b1 activity, we treated wild-type and cyp26b1 mutant embryos for various time intervals with all-trans RA (RA excess) or the competitive Aldh inhibitor 4-(diethylamino)benzaldehyde (DEAB; RA deficiency).

Strikingly, RA treatment of wild-type embryos from 24 to 50 hpf caused the same neurocranial phenotype as in cyp26b1 mutant embryos,characterized by the absence of the medial ethmoid plate at 120 hpf(Fig. 7A,B; compare with Fig. 4M) (n=25/25). By contrast, RA treatments commencing after 48 hpf did not alter anterior neurocranial morphology (data not shown). Conversely, DEAB treatment of cyp26b1 mutants and morphants from 24 to 50 hpf rescued the ethmoid phenotype (Fig. 7D)(n=9/10), whereas the same treatment of wild-type embryos left neurocranial morphology intact (Fig. 7C) (n=29/29) but caused a reduction of gill arches,reminiscent of the phenotype of aldh1a mutants(Begemann et al., 2001). Together, this indicates that the anterior neurocranial cartilage defects of cyp26b1 mutants are caused by RA excess during the second day of development.

Interference with bone development required significantly later treatments. Whereas RA treatment from 24 to 50 hpf had no effect on craniofacial ossification at 180 hpf (Fig. 7, compare F with E) (n=45/45), treatment from 72 to 96 hpf and onwards caused hyperossification of both endochondral and intramembranous bones (Fig. 7G)(n=25/26), comparable to that seen in cyp26b1 mutants(Fig. 5B,E). Consistently,craniofacial ossification of mutants was significantly reduced upon DEAB treatment starting at 72 or 96 hpf (Fig. 7H). Similarly, treatment of wild-type larvae with RA or the specific Cyp26 inhibitor R115866 (Njar et al., 2006; Stoppie et al.,2000) from 96 hpf resulted in axial hyperossification at 180 hpf(Fig. 7K,L) (n=32/33 and 27/27), as in cyp26b1 mutants(Fig. 7S), whereas earlier RA treatments (48-72 hpf) had no effect (Fig. 7J) (n=40/40). Conversely, ossification of centra was blocked or significantly reduced when wild-type or cyp26b1 mutant larvae were treated with DEAB from 96 hpf(Fig. 7Q,R) (n=33/33). Intriguingly, a significant reduction in the axial hyperossification of cyp26b1 mutants (Fig. 7M) to wild-type levels was obtained upon injection of aldh1a MO (Begemann et al.,2001) (Fig. 7S,T)(n=9/12). Together, this indicates that Cyp26b1 is required at different developmental stages to regulate skeletal patterning and ossification of skeletal elements by inactivating RA. This anti-RA effect is consistent with the data obtained in mouse, but in contrast to the conclusions by Reijntjes et al. described above(Reijntjes et al., 2007).

To gain insight into the cellular basis of RA-induced hyperossification, we compared the effects of RA application with those caused by excessive Bmp2, a well-known positive regulator of osteoblast maturation(Wu et al., 2007). For temporally controlled Bmp2 overexpression, we used transgenic fish that carry the bmp2b cDNA under the control of the heat-inducible hsp70promoter (Chocron et al.,2007). Applying the heat shock at 50 hpf, when RA applications had no effect on centra ossification (Fig. 7J), transgenic bmp2b fish displayed a fusion of centra 3-6 (Fig. 7O). These are the first ossifying centra during normal development (see above). By contrast,Bmp2 applied at 96 or 144 hpf left the early-ossifying centra unaffected,causing fusions of the later ossifying, more-anterior and more-posterior vertebral bodies (Fig. 7P)(data not shown), whereas centra 3-6 were fused upon RA application at these later stages (Fig. 7K; see Fig. S6A,B in the supplementary material). Strikingly, fusions of centra 3-6 were even obtained upon RA administration at 12 or 15 dpf(Fig. 7M,N; see Fig. S6C-F in the supplementary material). This indicates that Bmp2 affects osteoblast precursors at stages when they are not yet sensitive to RA, whereas RA can still affect osteoblasts long after they have become insensitive to Bmp2.

To directly compare the effects of Cyp26b1/RA and Bmp2b on osteoblasts, we stained for the osteoblast markers col10a1, osx, opn and cyp26b1 itself. In craniofacial skeletal elements of cyp26b1mutants, cyp26b1 expression was much stronger than in wild-type siblings (Fig. 8A). Similarly,osteoblasts of the opercle of mutants displayed stronger opnexpression. At 72 hpf, the number of opn-positive cells appeared normal, with higher expression levels per cell(Fig. 8B). However, at 120 hpf,there were supernumerary cells in normally opn-negative subdomains of the opercle (Fig. 8C). By contrast, expression levels and patterns of osx and col10a1appeared unaltered in mutants (Fig. 8D,E).

Similarly, in the axial skeleton, cyp26b1 mutants displayed a striking increase in cyp26b1- and opn-positive cells, with premature expression and an extension into more-posterior trunk regions(Fig. 8F,G; see Fig. S7 in the supplementary material), whereas the number of col10a1-positive cells was normal (Fig. 8H). Ectopic cyp26b1-positive cells were present ventral of the notochord, in contrast to their preferential perichordal localization in wild-type siblings(Fig. 8F). According to previous studies, axial osteoblasts stem from the sclerotome in ventral-most regions of the somites, from where they move dorsally towards the notochord(Inohaya et al., 2007; Morin-Kensicki and Eisen,1997), consistent with the ventral-to-dorsal progression of centra ossification (Fig. 6K). Thus,the ventral cyp26b1-positive cells are possibly immature osteoblasts that express cyp26b1 precociously in the mutant.

By contrast, overexpression of Bmp2b left the number of cyp26b1-positive cells unaffected(Fig. 8I), suggesting that their increase in cyp26b1 mutants might primarily reflect the loss of a negative RA→Cyp26 feedback loop, as described previously in other circumstances (Emoto et al.,2005). Strikingly, Bmp2b excess led to an increase not only of opn-positive, but also of col10a1-positive, cells(Fig. 8J,K; for quantification,see Fig. 8L). Thus, the specific increase of opn-positive cells in cyp26b1 mutants(Fig. 8G) might primarily reflect an increase in the activity of osteoblasts (see Discussion).

To study whether Cyp26 enzymes might have a similar role in restricting ossification during mammalian development, we treated mouse fetuses with the Cyp26 inhibitor R115866, starting at E13, shortly before the onset of vertebral ossification in untreated animals. At E18.5, treated mice often displayed fusions of neural arches of cervical vertebrae, particularly in C3-C5 (Fig. 9A,B)(n=4). However, no fusions were seen in thoracic, sacral or lumbar vertebrae, similar to the cervical restriction of fusions that is seen in several human vertebral disorders. Instead, in posterior regions, treated mice displayed ossification defects within vertebrae, including precocious fusions of neural arches with centra (Fig. 9C,D) (n=9). Also, the ribs were significantly thicker than in untreated embryos. In summary, this indicates that inhibition of Cyp26b1 causes similar shifts in the temporal and spatial pattern of ossification in mammals as in zebrafish.

Previous studies have shown that RA signaling plays multiple roles during skeletal patterning and the differentiation of skeletogenic cells. However,genetic evidence for the in vivo role of RA signaling and its inhibition during osteoblast development has thus far been missing. Here, we show that spatiotemporal restriction of RA signaling by Cyp26b1 is required to attenuate osteoblast maturation/activity and ossification during zebrafish and mouse development. These studies reveal a previously unrecognized effect of unrestricted RA signaling on vertebral column formation, which could also be relevant in human congenital disorders with vertebral fusions. Furthermore, we demonstrate an earlier role of Cyp26b1 in skeletal patterning of the neurocranium, consistent with the palatal clefting caused by fetal exposure to teratogenic retinoid doses in human(Lammer et al., 1985; Young et al., 2000).

Zebrafish cyp26b1 mutants and wild-type embryos treated with RA during the second day of development display very specific deficiencies of cartilaginous elements in the midline of the neurocranium and pharyngeal arches (Fig. 4). Spatial restriction of these defects could be due to functional redundancy between the three Cyp26 paralogs in other craniofacial regions. Consistently, the cyp26b1 expression domain in the forehead, adjacent to the reported location of precursors of the ethmoid plate, is one of the few cyp26b1-specific domains that lack cyp26a1 and cyp26c1 expression (Gu et al.,2005) (see Fig. S3A-C in the supplementary material).

What exactly excessive RA is doing to the midline cells remains elusive. In contrast to a recent report (Reijntjes et al., 2007), we found that the migration (see Fig. S5 in the supplementary material) and survival (our unpublished data) of CNC cells was unaffected in cyp26b1 mutants. In cell culture systems, RA can block chondrocyte specification (Weston et al.,2003). Accordingly, the limb malformations in Cyp26b1mutant mice are due to chondrocyte apoptosis, which might result from such failed specification (Yashiro et al.,2004). In cyp26b1 mutant zebrafish, however, such a mechanism seems unlikely because we could not detect any increase in the number of TUNEL- or Acridine Orange-positive cells in the affected craniofacial domain between 24 and 96 hpf (our unpublished observations). In addition, cell proliferation rates, as determined via anti-phosphohistone H3 immunostaining, appeared unaltered (our unpublished observations). In this light, we currently favor the possibility that Cyp26b1 might interfere with early skeletal patterning; for example, by modulating a morphogenetic effect of an RA gradient to determine differential cell fates, as has recently been reported for Cyp26a1 during hindbrain patterning(White et al., 2007). Indeed,both in the neurocranium and the branchial skeleton, cyp26b1 is expressed medial to the RA-synthesizing enzyme Aldh1a (see Fig. S3D-G in the supplementary material). This suggests that the craniofacial system might be patterned by a mediolateral RA gradient, with lowest levels in medial positions. This would explain why in cyp26b1 mutants, only the medial-most cells are lost (Fig. 4J-M). Similarly, the altered neurocranial col2a1expression pattern in cyp26b1 mutants at 50 hpf could indicate that medial positions (normally col2a1-negative) have acquired lateral(col2a1-positive) fates (Fig. 4C,D).

In Cyp26b1 mutant mice, possible defects during bone formation have only been marginally addressed(Yashiro et al., 2004). Also,expression of mouse Cyp26b1 in osteoblasts has not been described(Abu-Abed et al., 2002). Here,we show that zebrafish cyp26b1 transcripts colocalize at least transiently with osx, col10a1 and opn, even in developing intramembranous bones that lack chondrocytes, strongly suggesting that cyp26b1 is expressed in osteoblasts. Our analyses further indicate that cyp26b1 expression levels are particularly high in immature and/or less active osteoblasts, whereas expression in fully differentiated and/or highly active osteoblasts is much lower, in line with its proposed role in attenuating osteoblast maturation and/or activity. In cell culture systems,RA has also been shown to promote hypertrophic maturation/activity of chondrocytes (Weston et al.,2003). Whether a similar mechanism contributes to the hyperossification of endochondral bones in zebrafish cyp26b1 mutants remains unclear. However, this seems unlikely because cyp26b1 is only transiently expressed in chondrocytes and is switched off as they specify(Fig. 2).

In addition to endochondral bones, cyp26b1 mutants display hyperossification of intramembranous bones, which leads to overgrowth of the elements (Figs 5 and 6). In the vertebral column,this results in a complete fusion of centra(Fig. 6), whereas the opposite phenotype, complete loss of vertebral ossification, is obtained upon cyp26b1 overexpression (see Fig. S1H in the supplementary material). Vertebral fusions in cyp26b1 mutants manifest rather late, and cannot be due to shifts in vertebral identities because somitic expression of all tested Hox genes (hoxc6b, hoxc8a, hoxb8b, hoxb10a)(Prince et al., 1998) is unaffected at 24 hpf and later (our unpublished observations).

Our in situ hybridization analysis further revealed a precocious initiation of expression and a significant increase in the number of cyp26b1-and opn-positive cells in cyp26b1 mutants(Fig. 8; see Fig. S7 in the supplementary material). The precocious expression in ectopic positions could be interpreted as a consequence of premature osteoblast maturation. However,several lines of evidence suggest that in addition to maturation, or even instead of it, RA affects osteoblast activity. First, cyp26b1 mutants displayed a striking increase in opn transcript levels per cell,which, at least in the opercle, clearly preceded the increase in cell numbers. In osteoblast cell cultures, opn levels are often used to measure osteoblast activity because they increase in proportion to the amount of mineralized bone material, also in response to RA(Manji et al., 1998; Ohishi et al., 1995; Song et al., 2005). Second, cyp261 mutants displayed normal numbers of cells expressing other osteoblast markers, col10a1 and osx. This is in striking contrast to the effect of Bmp2 overexpression, a well-studied positive regulator of osteoblast maturation, which caused hyperossification accompanied by an increase of col10a1-positive cells (Figs 7 and 8). Finally, axial osteoblasts were still sensitive to RA many days after they had arrived at their final perichordal destination and after they had become insensitive to Bmp2 (15 versus 4 dpf) (compare Fig. 7Nwith Fig. 3H and Fig. 7P).

In cell culture studies, RA and Bmp2 have also been shown to stimulate osteoclasts, which are bone-resorbing cells of the hematopoietic lineage(Cowan et al., 2005; Kaji et al., 1995). However,for several reasons, this does not seem relevant for the ossification defects of cyp26b1 mutants. First, we would expect a stimulation of osteoclasts to result in loss, rather than the observed gain, of bone. Second,according to histological stainings of the osteoclast marker enzyme tartrate-resistant acid phosphatase (TRAP; Acp5 - ZFIN), osteoclasts only become active long after the bone phenotype of cyp26b1 mutants has become apparent (14 dpf) (Gavaia et al.,2006; Witten et al.,2001). Third, knockdown of pu.1 (Spi1 - ZFIN), which is required for specification of the entire myeloid lineage including osteoclasts(Rhodes et al., 2005; Zhao et al., 2007), did not affect axial ossification, although other myeloid derivatives, such as mpx-positive neutrophils, were completely absent (our unpublished results). This suggests that in contrast to Cyp26b1 and osteoblasts,osteoclasts are dispensable for bone formation during the larval stages we have investigated.

In higher vertebrates, such as birds and mammals, vertebra formation occurs via endochondral ossification, whereas in teleosts, vertebral bodies are formed through direct ossification in and around the notochordal sheath(perichordal centra). There has been some debate about the involvement of osteoblasts in fish axial skeleton development. According to one report,vertebral bodies in zebrafish arise by secretion of bone matrix from the notochord and without any involvement of osteoblasts(Fleming et al., 2004). By contrast, a more recent report claims that sclerotome-derived osteoblasts are present in intervertebral regions in Medaka(Inohaya et al., 2007). Our data are consistent with the latter report and with the situation in higher vertebrates, pointing to the presence and activity of osteoblasts during vertebral ossification in zebrafish. In addition, our data indicate that the role of Cyp26 enzymes in preventing hyperossification and vertebrae fusions has been largely conserved between fish and mammals. Mouse Cyp26b1displays metameric expression in the developing vertebral column(Abu-Abed et al., 2002), which could correspond to the cyp26b1 expression in intervertebral regions described here. Cyp26b1 mouse mutants have been reported to display a fusion of the two first cervical vertebrae, atlas and axis (G. A. MacLean, PhD thesis, Queen's University Kingston, Ontario, Canada, 2007; http://hdl.handle.net/1974/750). It had been proposed that this fusion is due to RA-induced homeotic transformations in vertebral anterior-posterior (AP) identity. However, we show here that inhibition of Cyp26 activity leads to hyperossifications and to fusions of cervical vertebrae when the drug is applied days after vertebral AP identity has been determined (E9-11)(Kessel, 1992; Kessel and Gruss, 1991). This suggests that, as in zebrafish, mouse Cyp26 enzymes are required to regulate ossification.

In humans, several disorders with cervical vertebral fusions have been described. A rather common (1:40,000) congenital disorder with such fusions is Klippel-Feil anomaly (KFA; OMIM 118100)(Kaplan et al., 2005; Tracy et al., 2004). KFA occurs sporadically, as well as in families with dominant or recessive inheritance. Its aetiology is unknown. Vertebral fusion is variably associated with craniofacial abnormalities, including frontonasal dysplasia, and various limb malformations. A similar association of vertebral and other developmental defects is observed in Goldenhar syndrome (OMIM 164210), MURCS association(OMIM 601076) and VATER association (OMIM 192350).

Zebrafish cyp26b1 mutants display a reduction in the anterior neurocranium and compromised pectoral fin development, consistent with frontonasal and limb abnormalities seen in some of the human syndromes. Homozygous null mutants die during late larval stages (10-15 dpf); however,hypomorphic alleles are sub-viable and characterized by progressive vertebral defects (Spoorendonk et al.,2008). Similar mutations could account for recessively inherited cases of human KFA, whereas the sporadic or dominantly inherited cases could be due to haploinsufficiencies of null mutations, as described here for the zebrafish ti230g allele, or to antimorphic mutations.

Interestingly, a sporadic case of KFA associated with craniofacial and ear defects has been correlated with an inversion on chromosome 2(p12q34)(Papagrigorakis et al., 2003). The human CYP26B1 gene is located at 2p13, close to this breakpoint. In collaboration with human geneticists, we are currently readdressing this case, and are sequencing CYP26B1 from other subjects with diagnosed KFA or Goldenhar syndrome (McGaughran et al., 2003).

We thank Elisabeth Busch and Derek Stemple for providing the sa0002 allele; Lieven Meerpoel (Johnson and Johnson Pharmaceutical Research and Development) for free samples of R115866; Cecilia Moens, Charles Kimmel, Vicky Prince and Tom Schilling for zebrafish reagents; Caroline Johner and Norbert Joswik for their help with the mouse experiments; Tom Schilling,Shannon Fisher, Henry Roehl and Jochen Holzschuh for helpful discussions; and Kirsten Spoorendonk and Stefan Schulte Merker for sharing unpublished data. Work in the laboratory of M.H. and the TILLING project leading to the isolation of the sa0002 allele were supported by the European Union(6th Framework Integrated Project `Zebrafish Models for Human Development and Disease').