Retinoic acid and Cyp26b1 are critical regulators of osteogenesis in the axial skeleton

Retinoic acid (RA) has been demonstrated to play important roles in processes as diverse as cardiac specification(Keegan et al., 2005),patterning of both the central nervous system(Hernandez et al., 2007) and the tetrapod limb (Yashiro et al.,2004), as well as specification of germ cell fate(Bowles et al., 2006). RA is synthesized intracellularly from its precursor, vitamin A (or retinol), by retinaldehyde dehydrogenases (Raldhs) and transported to the nucleus where it initiates transcription of target genes(Ross et al., 2000; Maden, 2002). Levels of RA are determined by Raldhs and, in addition, by Cyp26 enzymes, members of the RA catabolizing cytochrome P450 family (White et al., 2000; MacLean et al.,2001). In vertebrates, three Cyp26 enzymes have been identified:Cyp26a1, Cyp26b1 and Cyp26c1 (White et al., 1996; Nelson 1999; Gu et al.,2005).

In mice, a Cyp26a1 knockout that exhibits spina bifida and tail truncation as a consequence of patterning defects, and also shows homeotic transformations of anterior vertebrae has been described(Abu-Abed et al., 2001; Sakai et al., 2001), whereas Cyp26b1 knockout mouse embryos display craniofacial defects and reduced limbs (Yashiro et al.,2004). Homozygous mutants die immediately after birth owing to respiratory distress before vertebral defects become obvious, and hence no other bone defects have been reported. In zebrafish, the effects of Cyp26 enzymes have been studied in the context of hindbrain(Emoto et al., 2005; Hernandez et al., 2007; White et al., 2007) and neural crest patterning (Reijntjes et al.,2007).

In this study, we have characterized a zebrafish mutant in cyp26b1, which exhibits severe over-ossification of the vertebral column. In both zebrafish and mouse, we report expression of cyp26b1in osteoblasts of the pre-vertebrae regions. We suggest that it is the tight regulation of RA within osteoblasts by Cyp26b1 activity that is crucial for axial osteogenesis.

In zebrafish and medaka, which are considered to have highly similar modes of axial skeleton ossification (Inohaya et al., 2007), vertebrae form by intramembranous ossification without passing through a cartilage stage(Ekanayake and Hall, 1987; Fleming et al., 2004). It has been suggested for both species that notochord cells are involved in the initial bone matrix formation of the centra during development of the vertebral column (Fleming et al.,2004; Inohaya et al.,2007). However, Inohaya et al.(Inohaya et al., 2007) could also show that sclerotome-derived progenitor cells differentiate into osteoblasts on the surface of the notochordal sheath, where they produce and mineralize bone matrix: this matrix is formed by cells distal to the notochord. Subsequent vertebral growth then proceeds from the anterior and posterior edges of the forming centra.

Until now, several in vitro studies have reported effects of RA exposure on mineralization, however, with contradicting results. Some studies describe an increase in mineralization (Skillington et al., 2002; Wang and Kirsch,2002; Song et al.,2005; Yamashita et al.,2005; Malladi et al.,2006; Wan et al.,2007), whereas others report a suppression of cell differentiation and therefore a decrease in mineralization(Cohen-Tanugi and Forest, 1998; Iba et al., 2001) upon RA treatment of cultured cells. Here, we establish an in vivo model to follow early osteoblasts. We show that, upon RA treatment, axial osteoblasts in osterix:nuGFP transgenic zebrafish retain their normal positions. Moreover, osteoblast numbers do not change. Therefore, over-ossification of the axial skeleton upon RA excess is most probably caused by increased activity of osteoblasts along the vertebral column.

A 4.1 kb upstream regulatory region of the medaka osterix gene was amplified using the primers 5′-TGAACATGTCAGTGCCATCA-3′ (forward)and 5′-CGGGACAGTTTGGAAGAAGT-3′ (reverse) and cloned in front of mCherry. Transgenic zebrafish were generated by injection into one-cell stage embryos using the I-SceI meganuclease approach(Rembold et al., 2006). The same upstream region was cloned in front of a nuclear localization signal followed by GFP (nuGFP), and transgenic zebrafish were generated using the Tol2 transposon system (Kawakami et al.,1998). Cloning details are available upon request.

YFP was recombined directly after the ATG site of the gene of interest on a BAC clone, containing the genomic information of the gene of interest(Kimura et al., 2006). The following BACs and primers were used: DKEY-53014 (containing 73.5 kb upstream of the cyp26b1 coding region, and >73.5 kb downstream of cyp26b1); Cyp26b1_GFP_fw(ttgctcatcactccaaagagatatttgagacaagtccccggacgttcacaACCATGGTGAGCAAGGGCGAGG AG);Cyp26b1_Neo_rev(cacgcagccagcgtcgccaacgccgagacaaggtcaaaactctcgaagagTCAGAAGAACTCGTCAAGAAGGCG);and CH211-51D23 (containing 111.6 kb upstream of the osx-coding region, and 66.3 kb downstream of osx); Osx_GFP_fw(cagctctcctctcccgcttttggattgaccctcactggactgcttcctccACCATGGTGAGCAAGGGCGAGGAG);and Osx_Neo_rev (gcagctgtgagatcgcagtgagttttccgtacctccagaatcgacgc ggcTCAGAAGAACTCGTCAAGAAGGCG).

The stocksteif mutation was mapped to linkage group 7 using standard simple sequence length polymorphism mapping. The following six primers were used for fine mapping: R3.2-fw (5′ ACCGTAATTGAAACCACGTC 3′), R3.2-rv (5′ GCCAAATATTTCGATCTGTG 3′), R1.11-fw(5′ GATGCTCAGACCTGTGTTTG 3′), R1.11-rv (5′TGAAGTCAATGCTGGTCAAC 3′), R2.22-fw (5′ TCACCCTTCCATGAACTTAAC 3′) and R2.22-rv (5′ AAACAGCCAGCGTAGATAAAC 3′).

Embryos were fixed in 3.5% formaldehyde/0.1 M sodium phosphate buffer for 1 hour and stored in 70% methanol. The protocol for simultaneous bone and cartilage staining was adapted from that previously described(Walker and Kimmel, 2007). Briefly, embryos were rinsed in 50% ethanol and subsequently stained with 0.2 mg/ml Alcian Blue 8 GX (Sigma) in 70% ethanol/80 mM MgCl₂. After washes in 0.02% Triton, embryos were bleached in 1%H₂O₂/1% KOH for 30 minutes, washed in a saturated sodium tetraborate solution, and digested for 1 hour in 1 mg/ml trypsin (Sigma) in 60% saturated sodiumtetraborate. Bones were stained with 0.04 mg/ml Alizarin red S (Sigma) in 1% KOH. Destaining was carried out in an increasing glycerol series (10%, 30%, 70%) and specimens were stored at 4°C in 70%glycerol.

For Alizarin Red staining only, the Alcian Blue step was omitted. Juvenile fish were fixed for 2 hours, incubated in acetone to remove fat (up to 24 hours) and digested with 10 mg/ml trypsin in 60% sodiumtetraborate overnight. Scales were removed manually. In vivo skeletal staining was carried out with 0.001% calcein (Sigma) in embryo water for at least 2 days.

Stock solutions of 10 mM all-trans retinoic acid (Sigma) and 10 mM R115866(Janssen Pharmaceutica) in DMSO were diluted in embryo medium. Sibling controls were incubated in corresponding dilutions of DMSO. RA treatments were carried out in the dark and solutions were changed every ∼12 hours.

Fixed samples were wrapped in parafilm and scanned with a Skyscan 1072 microCT system at 80 kV and 100 μA. The cubic voxelsize was 6.08 μm for the wild type and 2.73 μm for the stocksteif animals. The scans were reconstructed and subsequently segmented as previously described(Feldkamp et al., 1984; Waarsing et al., 2004). The segmented images were visualized as a surface using a custom-written Matlab 7.3 script.

For all sectioned zebrafish material, in situ hybridization was carried out first on whole-mount embryos(Schulte-Merker, 2002), and only subsequently were embryos embedded in 3% agarose and cut (100 μm sections) on a vibrotome (Microm HM650V), or embedded in plastic and cut (10μm sections) on a microtome (Leica RM2035). For sections of mouse material,paraffin-embedded embryos were sectioned then hybridized in situ, according to standard procedures (Moorman et al.,2001).

In a screen for genes that affect ossification in zebrafish, we identified a single allele of the stocksteif mutant, sst^t24295. Mutant embryos are characterized by an early onset over-ossification of the notochord, resulting in fused vertebral centra(Fig. 1A). In this original allele, sst mutants are comparable in size and external features with their wild-type siblings, and therefore the phenotype can be detected only by skeletal staining at 8 days post fertilization (dpf) or later. Skeletal staining of siblings shows an equidistant distribution of vertebrae along the notochord, whereas in mutants most of the vertebrae are fused. Examining mutants that were not completely synchronous enabled us to conclude that initially centra are placed in a proper pattern along the notochord, but that later the excess of developing bone causes fusion of the vertebrae (insets in Fig. 1A).

To assess whether the phenotypic effects might be due to early patterning defects, we performed in situ hybridization for somitic markers (myoD,smad1), which did not show any differences between mutants and siblings(data not shown).

In a few exceptional cases, mutants survived when separated early from their siblings, until a maximum age of 6 months. These mutants remain smaller(Fig. 1B), have problems swimming and exhibit a protruding jaw (Fig. 1C). In skeletal preparations, most centra appear fused, and only about five distinct intervertebral boundaries can be observed along the body axis (Fig. 1C). Neural and haemal arches both exhibit fan-shaped expansions(Fig. 1C). Furthermore, the angles of the arches with respect to the body axis are more irregular and obtuse in mutants than in siblings. The phenotype is not restricted to the vertebral column at this later stage: hypurals partially fuse in the tail, and the head skeleton shows abnormalitites as well(Fig. 1C). Phenotypic analysis of 3-, 6- and 9-week-old mutants showed that the severity of bone defects increases over time with the older mutants exhibiting the broadest arches(data not shown).

MicroCT scans demonstrate that in sst mutants excess bone can replace the notochord by proximal growth, leading to completely solid centra(Fig. 1D), whereas the notochord remains present as a continuous rod in the vertebral column of wild-type siblings (see Fig. S1 and S2 in the supplementary material).

In order to molecularly identify the stocksteif locus, we positionally cloned the gene. The mutation mapped to linkage group 7(Fig. 2A). Fine mapping identified two flanking markers on BAC clone zC197C14. In 839 embryos tested,R3.2 left two recombinants (<0.2 cM) and R1.11 left one recombinant(<0.1 cM). The single gene enclosed by both markers is cyp26b1. Sequence analysis of the coding region of cyp26b1, as well as exon-intron-boundaries revealed no mutations causing amino acid alterations. Indeed, the sequence of mutant cyp26b1 cDNA also showed no base changes.

To obtain additional alleles of sst, we performed a reverse genetic screen (Wienholds et al.,2002). One allele, a nonsense mutation in exon 1, is predicted to change a lysine to a stop codon at position 46 (K46STOP; sst^sa0002) out of 512 amino acids in the putative wild-type protein. Another allele, a splice donor mutation(sst^sa0003), changes the most 5′ nucleotide of intron 2 from G to A (Fig. 2B),probably resulting in deficient splicing of this intron and consequently a full knockout of the encoded protein. Fig. 2C,D show Alizarin Red stained progeny from a cross between the two original sst^sa0002 and sst^sa0003founders. Identical stainings were also obtained for transheterozygous embryos from crosses between sst^sa0002 and sst^t24295 carriers, and homozygous mutant embryos for both newly obtained alleles, sst^sa0002 and sst^sa0003 (data not shown).

As described above, we have been able to raise a few homozygous mutants from the original sst^t24295 allele. By contrast, all transheterozygous embryos (both sst^sa0002/sst^sa0003 and sst^sa0002/sst^t24295), as well as the homozygous embryos for sst^sa0002 and sst^sa0003 die around day 8 of development, without developing a swim bladder. In addition to the over-ossification of the vertebral column, these mutants also exhibit a protruding jaw and lack some tissue dorsal to the ethmoid plate (Fig. 2C′,D′,E-G). The cartilage derived ethmoid plate itself (asterisk in Fig. 2G) is reduced in size and shows a more narrow morphology in homozygous sst^sa0002 and sst^sa0003 mutants than in siblings.

However, mRNA expression of bone-specific markers such as type X collagen(Col10a1), shown to be expressed in intramembranous bone in zebrafish(Avaron et al., 2006) and osterix, a specific marker for mature osteoblasts(Nakashima et al., 2002) is not altered in mutants. In situ hybridizations for both markers at 4 dpf do not show any differences between siblings and mutants(Fig. 2H-K; see Fig. S3 in the supplementary material).

In situ hybridization for cyp26b1 showed that expression levels in homozygous mutants are upregulated in comparison with siblings (see Fig. S4H,I in the supplementary material), suggesting a positive-feedback response to increased RA levels.

In summary, this shows that cyp26b1 deficiency leads to cranial cartilage defects. By contrast, bone structures are not affected in the head at this stage, but show severe over-ossification in the trunk.

To further understand the role of Cyp26b1 in ossification, we examined its expression pattern both in zebrafish and mice. In zebrafish, we confirmed previously reported cyp26b1 expression up to 72 hpf(Zhao et al., 2005) in the head region, covering expression in the hindbrain, branchial arches and pectoral fins (Fig. 3A, see Fig. S4A-C in the supplementary material). In addition, we here show that previously unreported expression can also be found in bone elements [examples are shown in Fig. 3A(cleithrum), Fig. S4E (operculum) and Fig. S4F,G in the supplementary material(parasphenoid)]. Moreover, we report a segmented pattern in the area surrounding the notochord that is most clearly visible in sst mutants because of their upregulated cyp26b1 levels(Fig. 3B). To determine whether the latter staining represents expression in sclerotome-derived cells, we compared cyp26b1 expression with that of twist(Fig. 3C,D), which has been shown to be specific, within the somites, for sclerotome in zebrafish(Morin-Kensicki and Eisen,1997) as well as in medaka(Renn et al., 2006). Expression patterns of cyp26b1 and twist do correspond with one another, indicating that cyp26b1 in zebrafish is expressed in sclerotome-derived cells.

To investigate whether cyp26b1 is expressed in osteoblasts, we first generated a new transgenic zebrafish reporter line, in which mCherry expression is controlled by the medaka osterix promoter (Renn and Winkler, unpublished). Briefly, in transgenic larvae, osterix:mCherry-positive cells in the head are seen in all intramembranous bones, such as parasphenoid and cleithrum, and later in the operculum and branchiostegal rays, as well as in endochondral bones (data not shown).

In order to compare osterix expression with cyp26b1expression in vivo, we recombined YFP into a BAC clone that contained the genomic sequence of cyp26b1. This construct was injected into one-cell stage Tg(osx:mCherry) embryos. Because in these transient assays not all cells will incorporate the cyp26b1:YFP construct, we expected patchy cyp26b1:YFP expression. Indeed, we were able to detect single YFP-positive cells in osx:mCherry positive bone elements (Fig. 3F-K). In various embryos that we have analyzed, we have only seen YFP expression in places where we had previously observed cyp26b1 mRNA expression. Ectopic expression was never observed. We conclude that cyp26b1 is expressed in osteoblasts.

In mice, Cyp26b1 is expressed during early embryonic development in the hindbrain, neural tube, craniofacial structures, in outgrowing limb buds (MacLean et al., 2001)and in the pre-vertebrae regions at E12.5 and E14.5(Abu-Abed et al., 2002). We confirm and extend these findings by showing that also in mice it is the condensing sclerotome surrounding the notochord that expresses Cyp26b1 (Fig. 4B,C),as in zebrafish.

Furthermore, in situ hybridizations on consecutive sections using probes against Cyp26b1 and Osterix show expression of both genes in overlapping regions of the mouse axial skeleton(Fig. 4D-G). Vertebrae are formed in an anterior-posterior progression with more posterior vertebrae being less mature than anterior ones. Cyp26b1 is expressed in all forming vertebrae, including the most posterior ones (arrows in Fig. 4D), whereas Osterix is expressed only in the most mature vertebrae(Fig. 4F). Thus,Cyp26b1 expression precedes Osterix expression.

We conclude that, as in zebrafish, murine Cyp26b1 is expressed in osteoblasts and that expression of Cyp26b1 is conserved among vertebrates. The expression pattern of cyp26b1 is consistent with the axial skeletogenesis phenotype in stocksteif mutants.

In the course of these studies, we noticed that in zebrafish axial osterix expression could be detected only at points in time when neural and haemal arches were about to appear at day 17(Fleming et al., 2004) (K.M.S. and S.S.-M., unpublished). We have failed to detect osterix mRNA in the axis prior to this point in time. This suggests that cells that form the calcified material of the centra do not express osterix at these stages in zebrafish.

Further support for this notion stems from the analysis of a transgenic zebrafish line, in which mCherry is controlled by the medaka osterixpromoter. Here, we only detect mCherry-positive cells in the arches and at the anterior and posterior edges of the forming vertebrae, but not in the central part of the centra at this stage. Expression could be detected only shortly prior to the formation of neural and haemal arches (see Fig. 6A-D). To confirm that this is not a limitation of the transgene, we have recombined YFP into a BAC containing the zebrafish osterix gene. Again, after injections of this construct in one-cell stage embryos, we were able to detect expression only in cranial bone and, after day 17, in haemal and neural arches. The main bodies of the centra, however, were always devoid of YFP expression in these transient expression assays (data not shown). Both transgenic lines show identical results and exhibit expression patterns identical to those we observe with osterix mRNA in situ.

We conclude that zebrafish osterix is very likely not expressed in cells that are responsible for early ossification of the centra. osterix expression, in the axial skeleton, is restricted to the anterior and posterior edges of the forming vertebrae, as well as to neural and haemal arches.

As sst mutants lack Cyp26b1 activity and therefore should have at least a local excess of RA, we asked whether RA treatment of wild-type embryos can phenocopy the mutant phenotype. Treatments were started at day 4 or later to avoid defects associated with early RA treatments of embryos(Keegan et al., 2005; Hernandez et al., 2007; Reijntjes et al., 2007). Wild-type embryos were treated with 1 μM RA starting at day 4, 5, 6 or 7 of development. At 8 dpf, Alizarin Red staining was performed. Sibling controls received equivalent treatments of DMSO, with no detectable effect on ossification. RA treatments, however, resulted in three phenotypic classes,all exhibiting axial over-ossification, with the most severe one completely mimicking the sst mutant phenotype(Fig. 5A). As shown in Fig. 5B, all embryos for which treatment was started at 4 dpf showed a phenotype, with 35% exhibiting a completely over-ossified notochord. For treatments started at 5 dpf and 6 dpf,respectively, 90% and 70% showed a phenotype. Embryos treated for 24 hours with RA onwards from day 7 were indistinguishable from controls, presumably as 24 hours were insufficient to allow for mineralization to occur.

To investigate more specifically the effect of aberrant Cyp26b1 function,we treated wild-type embryos with 0.5 μM of the R115866 compound, a selective antagonist for the three Cyp26 enzymes(Hernandez et al., 2007). Treatments were identical, and results near-identical to the described RA treatments: all embryos for which treatment was started at day 4 showed a phenotype resembling the sst mutant(Fig. 5C).

Obvious explanations for the observed phenotype would be mislocalization of osteoblasts, an increased number of osteoblasts or increased osteoblast activity. In order to distinguish between these scenarios, and in order to quantify precisely osteoblasts in vivo, we established, next to the described osterix:mCherry line, an additional line with nuclear GFP(nuGFP) controlled by the medaka osterix promoter region. Unfortunately, because even the hypomorphic sst^t24295allele is near-lethal, we have not been able to obtain homozygous sst^t24295 larvae in this transgenic background. Therefore,we decided to visualize directly the response of osteoblasts to excess RA in the osterix:nuGFP zebrafish line.

When zebrafish embryos are treated with 0.1 μM RA from 17 dpf to 20d pf,i.e. after onset of endogenous and transgenic osterix expression,over-ossification of the axial skeleton can be observed as evidenced by fusion of vertebrae (Alizarin Red bone staining in Fig. 6E-H), identical to our RA treatments described above (compare with Fig. 5A). The localization of GFP-positive osteoblasts in RA-treated embryos, however, is not altered in comparison with DMSO controls (compare Fig. 6I with 6L). In most treated cases, the space between centra was easily recognizable, even though anterior and posterior edges of centra were not delineated as straight as in DMSO control cases, probably owing to the large amount of ectopic bone matrix that is laid down in the spaces between the centra.

Next, we performed cell counts. Although in projections of confocal scans it is not always clear whether a spot of GFP expression refers to one or more nuclei, nuclei of separate cells can be clearly distinguished from each other in single confocal scans [e.g. the encircled spot in one projection(Fig. 6J) was resolved to represent two cells in a single scan (Fig. 6K)]. Furthermore, cells that are seen in projections but belong to arches at the other side of the embryo can be discriminated in single scans[examples of such cells are indicated with arrows in projections(Fig. 6J,M)]. Therefore, all cell counts were carried out in single confocal scans. In at least five different larvae per condition, several segments were counted (schematic representation of counted area is seen in Fig. 6O). As shown in Fig. 6P, there is no significant difference in numbers of osterix:nuGFP-positive cells between DMSO controls and RA-treated larvae. In conclusion, RA treatment of embryos from 17 to 20 dpf does not alter localization of mature osterix-expressing osteoblasts or their numbers.

In this study, we characterized the zebrafish mutant stocksteif,which exhibits severe over-ossification of the vertebral column. Initially,centra are placed in a wild-type pattern along the notochord, but later the excess of developing bone causes fusion of the vertebrae. This suggests that over-ossification in sst mutants is not due to a defect in initial anterior-posterior patterning, and is therefore distinct from the fused somites mutant phenotype (van Eeden et al., 1996). Unchanged expression patterns of early somitic markers (myoD, smad1) confirmed this notion.

In a few exceptional cases, we have been able to raise homozygous mutants of the original sst^t24295 allele to a later stage in life. Phenotypic analysis of larvae at different ages showed that the severity of bone defects increases over time, indicating that the gene affected in sst mutants is crucial for controlling proper ossification not only during embryogenesis, but also in late larval and adult stages. We suggest that sst activity is required throughout life to control osteogenesis, at least in axial skeletogenesis. This notion is supported by the late RA treatments we performed (17-20 dpf), which also resulted in axial over-ossification, indicating sensitivity to RA at late larval stages.

Positional cloning of sst^t24295 revealed two flanking markers enclosing the retinoic acid metabolizing gene cyp26b1, but no causative mutation could be found in this allele.

In order to identify additional alleles, we undertook a reverse genetic screen. We obtained two additional alleles of sst, both of which fail to complement the original sst^t24295 allele. We focus in this study on sst^sa0002, which encodes a nonsense mutation in exon 1, predicted to change a lysine to a stop codon at position 46(K46STOP; sst^sa0002) of the protein. As this will truncate the protein even N-terminal to the cytochrome P450 domain (at amino acid 50),we conclude that this allele encodes a loss-of-function allele.

When comparing the sst^t24295 and the sst^sa0002 allele, there are a number of noteworthy differences. First, all sst^sa0002 mutants die shortly after hatching, whereas we have been able to raise very few sst^t24295 homozygotes to late larval and adult stages. Second, when examining the head cartilage, sst^t24295 does not show alterations when compared with wild-type siblings, whereas the sst^sa0002 allele shows a protruding jaw and a reduced and misshapen ethmoid plate. Both lines of evidence suggest that sst^t24295 is a hypomorph and that it represents a slightly weaker allele than sst^sa0002. The observed jaw phenotype is most likely the reason for early lethality in the strong alleles and correlates with the adult head skeleton phenotype that we find in the hypomorphic sst^t24295 allele(Fig. 1C).

In the original stocksteif allele, we also observe over-ossification of neural and haemal arches, whereas the other two sst alleles die too early to analyze arch formation. In mice, similar arch phenotypes can be observed by treatments with the Cyp26 inhibitor R115866, which results in over-ossification and fusion of arches of treated embryos (Laue et al., 2008). This suggests a conserved role for the function of Cyp26b1 among vertebrates.

Cyp26b1 knockout mice die before the onset of vertebral deformities, owing to respiratory distress. Mutant mouse embryos, however,manifest reduced limbs (Yashiro et al.,2004). In fish, we do not observe phenotypes related to fin outgrowth, most probably because fin rays are evolutionary completely different from the digits of a mammalian limb(Grandel and Schulte-Merker,1998).

Jaw phenotypes mimicking what we observe in both embryonic and adult cyp26b1 mutants were not described in Cyp26a1 mutant mice(Abu-Abed et al., 2001; Sakai et al., 2001). The Cyp26a1 mouse phenotype therefore is clearly distinct from what we observe in zebrafish cyp26b1 mutants. Moreover, expression patterns of both genes are also distinct from each other: compared with Cyp26b1,Cyp26a1 expression is restricted to the extremities of the vertebral arches and ribs, and has not been detected in the pre-vertebrae regions(Abu-Abed et al., 2002).

We conclude that stocksteif manifests a novel phenotype, which allows new insight into the function of cyp26b1 and retinoid signalling during vertebral ossification.

Pharmacological treatments with either RA itself or the Cyp26 inhibitor R115866 completely phenocopy the stocksteif over-ossification phenotype, demonstrating a role for RA in ossification of the vertebral column. We propose that it is the regulation of Cyp26b1 activity, which defines local concentrations of RA within the vertebral area. Disruption of the gene or altered RA levels result in over-ossification along the vertebral column, most probably due to a local excess of RA within osteoblasts.

To support this notion, we examined the axial expression of Cyp26b1 in zebrafish and mice. Localization of Cyp26b1transcripts in both species was found abutting the notochord, and comparison with zebrafish twist expression suggests that Cyp26b1 is indeed expressed in sclerotome-derived cells of both teleosts and mammals. Furthermore, in both zebrafish and mice, we could clearly establish that Cyp26b1 is expressed in cells that also express osterix(Fig. 3F-K; Fig. 4D-G). Therefore, we suggest that Cyp26b1 activity controls RA levels within osteoblasts, possibly setting up a boundary of low versus high RA levels between cells that express Cyp26b1 and those that do not.

In zebrafish, we have never been able to detect osterix expression in the axial skeleton before arches appear, neither by in situ hybridization nor in the described transgenic lines. However, there is mineralized matrix present, as visualized by Alizarin Red or calcein staining, raising the issue of which cells are responsible for initial ossification of the very earliest centrum material. We suggest the existence of two types of osteoblasts: both are osteoblasts in the functional sense of secreting matrix that is then mineralized. However, one group does not express osterix, whereas the other group does. The latter group includes the osteoblasts of the future head skeleton, the vertebral arches and late-appearing osteoblasts at the anterior and posterior edges of the centra (Fig. 6A,C). The osterix-negative cells provide the initial material of the centrum body.

Further evidence for this hypothesis stems from Fleming et al.(Fleming et al., 2004), who were never able to detect osteoblasts in centrum bone matrix, whereas they could detect them in developing skull bones. It remains to be seen whether the osterix-negative cells are identical to cells of the notochord, which are postulated by Fleming et al. (Fleming et al., 2004) to initiate centrum formation.

Finally, it is interesting to note that osterix (Sp7)-null mutant mice are born with mineralized centra, whereas they lack almost all craniofacial bone elements (Nakashima et al., 2002). Although it is possible that centrum osteoblasts in mice express osterix, but do not require it for osteoblast function, it does point to a difference between centrum osteoblasts and all other osteoblasts.

We considered three models that may explain why excess RA results in over-ossification: (1) RA causes ectopic bone formation by mis-positioning of osteoblasts; (2) RA causes an increase in osteoblast number; or (3) RA increases the activity of osteoblasts while not affecting their number.

To distinguish between these mechanisms, we made use of a zebrafish osterix:nuGFP transgenic reporter line, in order to directly visualize the response of mature osteoblasts to RA. We show that, upon RA treatment, osterix-positive osteoblasts do not change position, nor do they increase in number. We therefore favour the third model and suggest that osteoblasts increase their mineralization activity upon exposure to excess RA.

As at present there is no marker available for the putative osterix-negative osteoblasts in the centra, we are unable to assess whether RA also acts on this type of osteoblasts. It is noteworthy that skeletons of the adult hypomorphic sst mutants show over-ossification at precisely the sites where we do see osterix expression: anterior and posterior edges of the centra, as well as neural and haemal arches. Therefore, our results could be explained by postulating an increase in the activity of osterix-positive osteoblasts for the later aspects of the phenotype. However, we do consider it likely that RA also has an effect on osterix-negative osteoblasts, as the sst phenotype becomes apparent already at day 8 of development.

In the past, conflicting results have been reported about in vitro studies that have examined the effect of RA on mineralization. An increase in mineralization was reported (Skillington et al., 2002; Wang and Kirsch,2002; Song et al.,2005; Yamashita et al.,2005; Malladi et al.,2006; Wan et al.,2007), whereas different studies show suppression of cell differentiation with a concomitant decrease in mineralization(Cohen-Tanugi and Forest, 1998; Iba et al., 2001) upon RA treatment of cultured cells. Here, we show in an in vivo setting that RA increases bone formation and suggest this to occur by an increased activity of osteoblasts. We have demonstrated that zebrafish offer an attractive model for performing genetic and pharmacological studies on osteoblasts, allowing in vivo observations and histological read-outs at the same time. In a field where the interpretation of in vitro and ex vivo experiments is inherently difficult, the use of an in vivo model that allows monitoring osteoblasts in real time will be highly beneficial.

We have shown that systemic treatment of larvae with RA produces phenotypes very similar to those of cyp26b1 loss of function, even though one could speculate that mis-regulation of RA in axial osteoblasts causes a more spatially restricted increase in local RA levels, than exposing all cells to excess RA. We have incorporated all data presented in this manuscript in the following model.

In the wild-type situation, expression of Cyp26b1 protein within osteoblasts leads to lower RA levels within cyp26b1-positive cells(i.e. osteoblasts) than in neighbouring cells that do not express cyp26b1 (i.e. non-osteoblasts). We propose that the juxtaposition of areas with low versus high RA levels is significant in maintaining cyp26b1-positive cells in a state in which they produce tightly controlled amounts of calcified matrix. Only when cells are exposed to comparatively higher levels of RA, e.g. in cyp26b1 mutants or after exposure to exogenous RA, will they then begin to produce an excess of calcified tissue.

In RA-treated embryos, all cells within the embryo experience high RA levels, and consequently there is no boundary between `RA low' and `RA high'regions. Accordingly, both osterix-negative osteoblasts within the body of the centrum and osterix-positive osteoblasts in the anterior and posterior regions of the centrum produce more matrix, leading to the observed over-ossification phenotype.

In loss-of-function stocksteif mutant embryos, RA levels are not lowered in osteoblasts (or only mildly lowered in the hypomorphic allele),again leading to increased osteoblast activity and resulting in the same events as just described for systemic RA exposure.

In summary, we have shown that regulation of RA levels and the tight control of Cyp26b1 activity are essential for regulation of skeletogenesis in zebrafish. Our data demonstrate a previously unappreciated role for RA and cyp26b1 in osteogenesis of the vertebral column and provide novel insight into the regulation of bone formation.

The original sst^t24295 allele was identified in the Tübingen 2000 Screen (F. Van Bebber, E. Busch-Nentwich, R. Dahm, O. Frank, H.-G. Fronhöfer, H. Geiger, D. Gilmour, S. Holly, J. Hooge, D. Jülich, H. Knaut, F. Maderspacher, H.-M. Maischein, C. Neumann, C. Nüsslein-Volhard, H. Roehl, U. Schönberger, C. Seiler, S. Sidi, M. Sonawane, A. Wehner, P. Erker, H. Habeck, U. Hagner, C. E. Hennen Kaps, A. Kirchner, T. Koblizek, U. Langheinrich, C. Loeschke, C. Metzger, R. Nordin, J. Odenthal, M. Pezzuti, K. Schlombs, J. deSantana-Stamm, T. Trowe, G. Vacun, B. Walderich, A. Walker and C. Weiler). Janssen Pharmaceutica kindly provided R115866. Ansa Wasim performed initial genomic mappings. Jeroen Bussmann performed BAC recombineering experiments. Identification of additional sst alleles by TILLING at the Hubrecht Institute and the Sanger Center would not have been possible without the enthusiastic help of (in alphabetical order) Ewart de Bruijn, Elisabeth Busch, Edwin Cuppen, Ross Kettleborough, and Derek Stemple. Kathrin Laue and Matthias Hammerschmidt shared results before publication. All members of S.S.-M.'s laboratory are acknowledged for discussions. S.S.-M. gratefully recognizes the support of the Smart Mix Programme of the Netherlands Ministry of Economic Affairs and the Netherlands Ministry of Education, Culture and Science. In part, this work was funded by ESA (15452/01/NL/SH) and an A^*STAR BMRC grant(07/1/21/19/544) to C.W.